Revert "reduce number of stripe sizes tested and add a test for postgres version 17"

This reverts commit 4ab4510a84.

.github/actionlint.yml

@@ -25,4 +25,3 @@ config-variables:
   - PGREGRESS_PG17_PROJECT_ID
   - SLACK_ON_CALL_QA_STAGING_STREAM
   - DEV_AWS_OIDC_ROLE_MANAGE_BENCHMARK_EC2_VMS_ARN
-  - SLACK_ON_CALL_STORAGE_STAGING_STREAM

.github/actions/neon-project-create/action.yml

@@ -17,6 +17,31 @@ inputs:
   compute_units:
     description: '[Min, Max] compute units'
     default: '[1, 1]'
+  # settings below only needed if you want the project to be sharded from the beginning
+  shard_split_project:
+    description: 'by default new projects are not shard-split, specify true to shard-split'
+    required: false
+    default: 'false'
+  admin_api_key:
+    description: 'Admin API Key needed for shard-splitting. Must be specified if shard_split_project is true'
+    required: false
+  shard_count:
+    description: 'Number of shards to split the project into, only applies if shard_split_project is true'
+    required: false
+    default: '8'
+  stripe_size:
+    description: 'Stripe size, optional, in 8kiB pages.  e.g. set 2048 for 16MB stripes. Default is 128 MiB, only applies if shard_split_project is true'
+    required: false
+    default: '32768'
+  psql_path:
+    description: 'Path to psql binary - it is caller responsibility to provision the psql binary'
+    required: false
+    default: '/tmp/neon/pg_install/v16/bin/psql'
+  libpq_lib_path:
+    description: 'Path to directory containing libpq library - it is caller responsibility to provision the libpq library'
+    required: false
+    default: '/tmp/neon/pg_install/v16/lib'
+  
 
 outputs:
   dsn:
@@ -63,6 +88,23 @@ runs:
         echo "project_id=${project_id}" >> $GITHUB_OUTPUT
 
         echo "Project ${project_id} has been created"
+
+        if [ "${SHARD_SPLIT_PROJECT}" = "true" ]; then
+          # determine tenant ID
+          TENANT_ID=`${PSQL} ${dsn} -t -A -c "SHOW neon.tenant_id"`
+          
+          echo "Splitting project ${project_id} with tenant_id ${TENANT_ID} into $((SHARD_COUNT)) shards with stripe size $((STRIPE_SIZE))"
+
+          echo "Sending PUT request to https://${API_HOST}/regions/${REGION_ID}/api/v1/admin/storage/proxy/control/v1/tenant/${TENANT_ID}/shard_split"
+          echo "with body {\"new_shard_count\": $((SHARD_COUNT)), \"new_stripe_size\": $((STRIPE_SIZE))}"
+          
+          # we need an ADMIN API KEY to invoke storage controller API for shard splitting (bash -u above checks that the variable is set)
+          curl -X PUT \
+            "https://${API_HOST}/regions/${REGION_ID}/api/v1/admin/storage/proxy/control/v1/tenant/${TENANT_ID}/shard_split" \
+            -H "Accept: application/json" -H "Content-Type: application/json" -H "Authorization: Bearer ${ADMIN_API_KEY}" \
+            -d "{\"new_shard_count\": $SHARD_COUNT, \"new_stripe_size\": $STRIPE_SIZE}"
+        fi
+
       env:
         API_HOST: ${{ inputs.api_host }}
         API_KEY: ${{ inputs.api_key }}
@@ -70,3 +112,9 @@ runs:
         POSTGRES_VERSION: ${{ inputs.postgres_version }}
         MIN_CU: ${{ fromJSON(inputs.compute_units)[0] }}
         MAX_CU: ${{ fromJSON(inputs.compute_units)[1] }}
+        SHARD_SPLIT_PROJECT: ${{ inputs.shard_split_project }}
+        ADMIN_API_KEY: ${{ inputs.admin_api_key }}
+        SHARD_COUNT: ${{ inputs.shard_count }}
+        STRIPE_SIZE: ${{ inputs.stripe_size }}
+        PSQL: ${{ inputs.psql_path }}
+        LD_LIBRARY_PATH: ${{ inputs.libpq_lib_path }}

.github/workflows/_check-codestyle-rust.yml

@@ -1,91 +0,0 @@
-name: Check Codestyle Rust
-
-on:
-  workflow_call:
-    inputs:
-      build-tools-image:
-        description: "build-tools image"
-        required: true
-        type: string
-      archs:
-        description: "Json array of architectures to run on"
-        type: string
-
-
-defaults:
-  run:
-    shell: bash -euxo pipefail {0}
-
-jobs:
-  check-codestyle-rust:
-    strategy:
-      matrix:
-        arch: ${{ fromJson(inputs.archs) }}
-    runs-on: ${{ fromJson(format('["self-hosted", "{0}"]', matrix.arch == 'arm64' && 'small-arm64' || 'small')) }}
-
-    container:
-      image: ${{ inputs.build-tools-image }}
-      credentials:
-        username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
-        password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
-      options: --init
-
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-        with:
-          submodules: true
-
-      - name: Cache cargo deps
-        uses: actions/cache@v4
-        with:
-          path: |
-            ~/.cargo/registry
-            !~/.cargo/registry/src
-            ~/.cargo/git
-            target
-          key: v1-${{ runner.os }}-${{ runner.arch }}-cargo-${{ hashFiles('./Cargo.lock') }}-${{ hashFiles('./rust-toolchain.toml') }}-rust
-
-      # Some of our rust modules use FFI and need those to be checked
-      - name: Get postgres headers
-        run: make postgres-headers -j$(nproc)
-
-      # cargo hack runs the given cargo subcommand (clippy in this case) for all feature combinations.
-      # This will catch compiler & clippy warnings in all feature combinations.
-      # TODO: use cargo hack for build and test as well, but, that's quite expensive.
-      # NB: keep clippy args in sync with ./run_clippy.sh
-      #
-      # The only difference between "clippy --debug" and "clippy --release" is that in --release mode,
-      # #[cfg(debug_assertions)] blocks are not built. It's not worth building everything for second
-      # time just for that, so skip "clippy --release".
-      - run: |
-          CLIPPY_COMMON_ARGS="$( source .neon_clippy_args; echo "$CLIPPY_COMMON_ARGS")"
-          if [ "$CLIPPY_COMMON_ARGS" = "" ]; then
-            echo "No clippy args found in .neon_clippy_args"
-            exit 1
-          fi
-          echo "CLIPPY_COMMON_ARGS=${CLIPPY_COMMON_ARGS}" >> $GITHUB_ENV
-      - name: Run cargo clippy (debug)
-        run: cargo hack --features default --ignore-unknown-features --feature-powerset clippy $CLIPPY_COMMON_ARGS
-
-      - name: Check documentation generation
-        run: cargo doc --workspace --no-deps --document-private-items
-        env:
-          RUSTDOCFLAGS: "-Dwarnings -Arustdoc::private_intra_doc_links"
-
-      # Use `${{ !cancelled() }}` to run quck tests after the longer clippy run
-      - name: Check formatting
-        if: ${{ !cancelled() }}
-        run: cargo fmt --all -- --check
-
-      # https://github.com/facebookincubator/cargo-guppy/tree/bec4e0eb29dcd1faac70b1b5360267fc02bf830e/tools/cargo-hakari#2-keep-the-workspace-hack-up-to-date-in-ci
-      - name: Check rust dependencies
-        if: ${{ !cancelled() }}
-        run: |
-          cargo hakari generate --diff  # workspace-hack Cargo.toml is up-to-date
-          cargo hakari manage-deps --dry-run  # all workspace crates depend on workspace-hack
-
-      # https://github.com/EmbarkStudios/cargo-deny
-      - name: Check rust licenses/bans/advisories/sources
-        if: ${{ !cancelled() }}
-        run: cargo deny check --hide-inclusion-graph

.github/workflows/build_and_test.yml

@@ -164,11 +164,77 @@ jobs:
 
   check-codestyle-rust:
     needs: [ check-permissions, build-build-tools-image ]
-    uses: ./.github/workflows/_check-codestyle-rust.yml
-    with:
-      build-tools-image: ${{ needs.build-build-tools-image.outputs.image }}-bookworm
-      archs: '["x64", "arm64"]'
-    secrets: inherit
+    strategy:
+      matrix:
+        arch: [ x64, arm64 ]
+    runs-on: ${{ fromJson(format('["self-hosted", "{0}"]', matrix.arch == 'arm64' && 'small-arm64' || 'small')) }}
+
+    container:
+      image: ${{ needs.build-build-tools-image.outputs.image }}-bookworm
+      credentials:
+        username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
+        password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
+      options: --init
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+        with:
+          submodules: true
+
+      - name: Cache cargo deps
+        uses: actions/cache@v4
+        with:
+          path: |
+            ~/.cargo/registry
+            !~/.cargo/registry/src
+            ~/.cargo/git
+            target
+          key: v1-${{ runner.os }}-${{ runner.arch }}-cargo-${{ hashFiles('./Cargo.lock') }}-${{ hashFiles('./rust-toolchain.toml') }}-rust
+
+      # Some of our rust modules use FFI and need those to be checked
+      - name: Get postgres headers
+        run: make postgres-headers -j$(nproc)
+
+      # cargo hack runs the given cargo subcommand (clippy in this case) for all feature combinations.
+      # This will catch compiler & clippy warnings in all feature combinations.
+      # TODO: use cargo hack for build and test as well, but, that's quite expensive.
+      # NB: keep clippy args in sync with ./run_clippy.sh
+      #
+      # The only difference between "clippy --debug" and "clippy --release" is that in --release mode,
+      # #[cfg(debug_assertions)] blocks are not built. It's not worth building everything for second
+      # time just for that, so skip "clippy --release".
+      - run: |
+          CLIPPY_COMMON_ARGS="$( source .neon_clippy_args; echo "$CLIPPY_COMMON_ARGS")"
+          if [ "$CLIPPY_COMMON_ARGS" = "" ]; then
+            echo "No clippy args found in .neon_clippy_args"
+            exit 1
+          fi
+          echo "CLIPPY_COMMON_ARGS=${CLIPPY_COMMON_ARGS}" >> $GITHUB_ENV
+      - name: Run cargo clippy (debug)
+        run: cargo hack --features default --ignore-unknown-features --feature-powerset clippy $CLIPPY_COMMON_ARGS
+
+      - name: Check documentation generation
+        run: cargo doc --workspace --no-deps --document-private-items
+        env:
+            RUSTDOCFLAGS: "-Dwarnings -Arustdoc::private_intra_doc_links"
+
+      # Use `${{ !cancelled() }}` to run quck tests after the longer clippy run
+      - name: Check formatting
+        if: ${{ !cancelled() }}
+        run: cargo fmt --all -- --check
+
+      # https://github.com/facebookincubator/cargo-guppy/tree/bec4e0eb29dcd1faac70b1b5360267fc02bf830e/tools/cargo-hakari#2-keep-the-workspace-hack-up-to-date-in-ci
+      - name: Check rust dependencies
+        if: ${{ !cancelled() }}
+        run: |
+          cargo hakari generate --diff  # workspace-hack Cargo.toml is up-to-date
+          cargo hakari manage-deps --dry-run  # all workspace crates depend on workspace-hack
+
+      # https://github.com/EmbarkStudios/cargo-deny
+      - name: Check rust licenses/bans/advisories/sources
+        if: ${{ !cancelled() }}
+        run: cargo deny check --hide-inclusion-graph
 
   build-and-test-locally:
     needs: [ tag, build-build-tools-image ]
@@ -280,22 +346,25 @@ jobs:
       # XXX: no coverage data handling here, since benchmarks are run on release builds,
       # while coverage is currently collected for the debug ones
 
-  report-benchmarks-results-to-slack:
+  report-benchmarks-failures:
     needs: [ benchmarks, create-test-report ]
-    if: github.ref_name == 'main' && !cancelled() && contains(fromJSON('["success", "failure"]'), needs.benchmarks.result)
+    if: github.ref_name == 'main' && failure() && needs.benchmarks.result == 'failure'
+    permissions:
+      id-token: write # aws-actions/configure-aws-credentials
+      statuses: write
+      contents: write
+      pull-requests: write
     runs-on: ubuntu-22.04
 
     steps:
-    - uses: slackapi/slack-github-action@v2
+    - uses: slackapi/slack-github-action@v1
       with:
-        method: chat.postMessage
-        token: ${{ secrets.SLACK_BOT_TOKEN }}
-        payload: |
-          channel: "${{ vars.SLACK_ON_CALL_STORAGE_STAGING_STREAM }}"
-          text: |
-            Benchmarks on main: *${{ needs.benchmarks.result }}*
-            - <${{ needs.create-test-report.outputs.report-url }}|Allure report>
-            - <${{ github.event.head_commit.url }}|${{ github.sha }}>
+        channel-id: C060CNA47S9 # on-call-staging-storage-stream
+        slack-message: |
+          Benchmarks failed on main <${{ github.event.head_commit.url }}|${{ github.sha }}>
+          <${{ needs.create-test-report.outputs.report-url }}|Allure report>
+      env:
+        SLACK_BOT_TOKEN: ${{ secrets.SLACK_BOT_TOKEN }}
 
   create-test-report:
     needs: [ check-permissions, build-and-test-locally, coverage-report, build-build-tools-image, benchmarks ]

.github/workflows/ingest_benchmark.yml

@@ -28,7 +28,31 @@ jobs:
     strategy:
       fail-fast: false # allow other variants to continue even if one fails
       matrix:
-        target_project: [new_empty_project, large_existing_project]
+        include:
+          - target_project: new_empty_project
+            stripe_size: 128  # 1 MiB
+          - target_project: new_empty_project
+            stripe_size: 256 # 2 MiB
+          - target_project: new_empty_project
+            stripe_size: 512 # 4 MiB
+          - target_project: new_empty_project
+            stripe_size: 1024 # 8 MiB
+          - target_project: new_empty_project 
+            stripe_size: 2048 # 16 MiB
+          - target_project: new_empty_project
+            stripe_size: 4096 # 32 MiB
+          - target_project: new_empty_project 
+            stripe_size: 8192 # 64 MiB
+          - target_project: new_empty_project
+            stripe_size: 16384 # 128 MiB
+          - target_project: new_empty_project
+            stripe_size: 32768 # 256 MiB # note that this is different from null because using null will shard_split the project only if it reaches the threshold
+                               # while here it is sharded from the beginning with a shard size of 256 MiB
+          - target_project: new_empty_project
+            stripe_size: null # run with neon defaults which will shard split only when reaching the threshold
+          - target_project: large_existing_project
+            stripe_size: null # cannot re-shared or choose different stripe size for existing, already sharded project
+      max-parallel: 1 # we want to run each stripe size sequentially to be able to compare the results
     permissions:
       contents: write
       statuses: write
@@ -75,6 +99,10 @@ jobs:
         postgres_version: 16
         compute_units: '[7, 7]' # we want to test large compute here to avoid compute-side bottleneck
         api_key: ${{ secrets.NEON_STAGING_API_KEY }}
+        shard_split_project: ${{ matrix.stripe_size != null && 'true' || 'false' }}
+        admin_api_key: ${{ secrets.NEON_STAGING_ADMIN_API_KEY }} 
+        shard_count: 8
+        stripe_size: ${{ matrix.stripe_size }}
 
     - name: Initialize Neon project
       if: ${{ matrix.target_project == 'new_empty_project' }}

.github/workflows/pre-merge-checks.yml

@@ -1,12 +1,6 @@
 name: Pre-merge checks
 
 on:
-  pull_request:
-    paths:
-      - .github/workflows/_check-codestyle-python.yml
-      - .github/workflows/_check-codestyle-rust.yml
-      - .github/workflows/build-build-tools-image.yml
-      - .github/workflows/pre-merge-checks.yml
   merge_group:
     branches:
       - main
@@ -23,10 +17,8 @@ jobs:
     runs-on: ubuntu-22.04
     outputs:
       python-changed: ${{ steps.python-src.outputs.any_changed }}
-      rust-changed: ${{ steps.rust-src.outputs.any_changed }}
     steps:
       - uses: actions/checkout@v4
-
       - uses: tj-actions/changed-files@4edd678ac3f81e2dc578756871e4d00c19191daf # v45.0.4
         id: python-src
         with:
@@ -38,25 +30,11 @@ jobs:
             poetry.lock
             pyproject.toml
 
-      - uses: tj-actions/changed-files@4edd678ac3f81e2dc578756871e4d00c19191daf # v45.0.4
-        id: rust-src
-        with:
-          files: |
-            .github/workflows/_check-codestyle-rust.yml
-            .github/workflows/build-build-tools-image.yml
-            .github/workflows/pre-merge-checks.yml
-            **/**.rs
-            **/Cargo.toml
-            Cargo.toml
-            Cargo.lock
-
       - name: PRINT ALL CHANGED FILES FOR DEBUG PURPOSES
         env:
           PYTHON_CHANGED_FILES: ${{ steps.python-src.outputs.all_changed_files }}
-          RUST_CHANGED_FILES: ${{ steps.rust-src.outputs.all_changed_files }}
         run: |
           echo "${PYTHON_CHANGED_FILES}"
-          echo "${RUST_CHANGED_FILES}"
 
   build-build-tools-image:
     if: needs.get-changed-files.outputs.python-changed == 'true'
@@ -77,16 +55,6 @@ jobs:
       build-tools-image: ${{ needs.build-build-tools-image.outputs.image }}-bookworm-x64
     secrets: inherit
 
-  check-codestyle-rust:
-    if: needs.get-changed-files.outputs.rust-changed == 'true'
-    needs: [ get-changed-files, build-build-tools-image ]
-    uses: ./.github/workflows/_check-codestyle-rust.yml
-    with:
-      # `-bookworm-x64` suffix should match the combination in `build-build-tools-image`
-      build-tools-image: ${{ needs.build-build-tools-image.outputs.image }}-bookworm-x64
-      archs: '["x64"]'
-    secrets: inherit
-
   # To get items from the merge queue merged into main we need to satisfy "Status checks that are required".
   # Currently we require 2 jobs (checks with exact name):
   # - conclusion
@@ -99,7 +67,6 @@ jobs:
     needs:
       - get-changed-files
       - check-codestyle-python
-      - check-codestyle-rust
     runs-on: ubuntu-22.04
     steps:
       - name: Create fake `neon-cloud-e2e` check

Cargo.lock

@@ -1312,7 +1312,6 @@ dependencies = [
  "tracing-utils",
  "url",
  "utils",
- "uuid",
  "vm_monitor",
  "workspace_hack",
  "zstd",
@@ -3982,11 +3981,9 @@ name = "pagectl"
 version = "0.1.0"
 dependencies = [
  "anyhow",
- "bincode",
  "camino",
  "clap",
  "humantime",
- "itertools 0.10.5",
  "pageserver",
  "pageserver_api",
  "postgres_ffi",
@@ -4008,7 +4005,6 @@ dependencies = [
  "arc-swap",
  "async-compression",
  "async-stream",
- "bincode",
  "bit_field",
  "byteorder",
  "bytes",
@@ -4048,7 +4044,6 @@ dependencies = [
  "postgres_connection",
  "postgres_ffi",
  "postgres_initdb",
- "pprof",
  "pq_proto",
  "procfs",
  "rand 0.8.5",
@@ -5659,7 +5654,6 @@ dependencies = [
  "crc32c",
  "criterion",
  "desim",
- "env_logger 0.10.2",
  "fail",
  "futures",
  "hex",
@@ -5688,7 +5682,6 @@ dependencies = [
  "serde",
  "serde_json",
  "sha2",
- "smallvec",
  "storage_broker",
  "strum",
  "strum_macros",
@@ -5713,13 +5706,10 @@ dependencies = [
 name = "safekeeper_api"
 version = "0.1.0"
 dependencies = [
- "anyhow",
  "const_format",
- "pageserver_api",
  "postgres_ffi",
  "pq_proto",
  "serde",
- "serde_json",
  "tokio",
  "utils",
 ]
@@ -7567,21 +7557,12 @@ dependencies = [
  "anyhow",
  "async-compression",
  "bytes",
- "camino",
- "camino-tempfile",
- "criterion",
- "futures",
  "pageserver_api",
  "postgres_ffi",
- "pprof",
  "prost",
- "remote_storage",
  "serde",
- "serde_json",
  "thiserror",
- "tikv-jemallocator",
  "tokio",
- "tokio-util",
  "tonic",
  "tonic-build",
  "tracing",

build-tools.Dockerfile

@@ -115,7 +115,7 @@ RUN set -e \
 
 # Keep the version the same as in compute/compute-node.Dockerfile and
 # test_runner/regress/test_compute_metrics.py.
-ENV SQL_EXPORTER_VERSION=0.17.0
+ENV SQL_EXPORTER_VERSION=0.16.0
 RUN curl -fsSL \
     "https://github.com/burningalchemist/sql_exporter/releases/download/${SQL_EXPORTER_VERSION}/sql_exporter-${SQL_EXPORTER_VERSION}.linux-$(case "$(uname -m)" in x86_64) echo amd64;; aarch64) echo arm64;; esac).tar.gz" \
     --output sql_exporter.tar.gz \

compute/compute-node.Dockerfile

@@ -66,7 +66,6 @@ RUN cd postgres && \
     make MAKELEVEL=0 -j $(getconf _NPROCESSORS_ONLN) -s -C src/interfaces/libpq install && \
     # Enable some of contrib extensions
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/autoinc.control && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/dblink.control && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/bloom.control && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/earthdistance.control && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/insert_username.control && \
@@ -105,18 +104,16 @@ RUN cd postgres && \
         esac; \
     done;
 
-# Set PATH for all the subsequent build steps
-ENV PATH="/usr/local/pgsql/bin:$PATH"
-
 #########################################################################################
 #
 # Layer "postgis-build"
 # Build PostGIS from the upstream PostGIS mirror.
 #
 #########################################################################################
-FROM pg-build AS postgis-build
+FROM build-deps AS postgis-build
 ARG DEBIAN_VERSION
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 RUN apt update && \
     apt install --no-install-recommends --no-install-suggests -y \
     gdal-bin libboost-dev libboost-thread-dev libboost-filesystem-dev \
@@ -154,6 +151,8 @@ RUN case "${DEBIAN_VERSION}" in \
     DESTDIR=/sfcgal ninja install -j $(getconf _NPROCESSORS_ONLN) && \
     ninja clean && cp -R /sfcgal/* /
 
+ENV PATH="/usr/local/pgsql/bin:$PATH"
+
 # Postgis 3.5.0 supports v17
 RUN case "${PG_VERSION}" in \
     "v17") \
@@ -228,8 +227,9 @@ RUN case "${PG_VERSION}" in \
 # Build plv8
 #
 #########################################################################################
-FROM pg-build AS plv8-build
+FROM build-deps AS plv8-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 COPY compute/patches/plv8-3.1.10.patch /plv8-3.1.10.patch
 
@@ -264,6 +264,7 @@ RUN case "${PG_VERSION}" in \
     # generate and copy upgrade scripts
     mkdir -p upgrade && ./generate_upgrade.sh ${PLV8_TAG#v} && \
     cp upgrade/* /usr/local/pgsql/share/extension/ && \
+    export PATH="/usr/local/pgsql/bin:$PATH" && \
     make DOCKER=1 -j $(getconf _NPROCESSORS_ONLN) install && \
     rm -rf /plv8-* && \
     find /usr/local/pgsql/ -name "plv8-*.so" | xargs strip && \
@@ -290,8 +291,9 @@ RUN case "${PG_VERSION}" in \
 # Build h3_pg
 #
 #########################################################################################
-FROM pg-build AS h3-pg-build
+FROM build-deps AS h3-pg-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 # not version-specific
 # last release v4.1.0 - Jan 18, 2023
@@ -312,6 +314,7 @@ RUN mkdir -p /h3/usr/ && \
 RUN wget https://github.com/zachasme/h3-pg/archive/refs/tags/v4.1.3.tar.gz -O h3-pg.tar.gz && \
     echo "5c17f09a820859ffe949f847bebf1be98511fb8f1bd86f94932512c00479e324 h3-pg.tar.gz" | sha256sum --check && \
     mkdir h3-pg-src && cd h3-pg-src && tar xzf ../h3-pg.tar.gz --strip-components=1 -C . && \
+    export PATH="/usr/local/pgsql/bin:$PATH" && \
     make -j $(getconf _NPROCESSORS_ONLN) && \
     make -j $(getconf _NPROCESSORS_ONLN) install && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/h3.control && \
@@ -323,16 +326,17 @@ RUN wget https://github.com/zachasme/h3-pg/archive/refs/tags/v4.1.3.tar.gz -O h3
 # compile unit extension
 #
 #########################################################################################
-FROM pg-build AS unit-pg-build
+FROM build-deps AS unit-pg-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 # not version-specific
 # last release 7.9 - Sep 15, 2024
 RUN wget https://github.com/df7cb/postgresql-unit/archive/refs/tags/7.9.tar.gz -O postgresql-unit.tar.gz && \
     echo "e46de6245dcc8b2c2ecf29873dbd43b2b346773f31dd5ce4b8315895a052b456 postgresql-unit.tar.gz" | sha256sum --check && \
     mkdir postgresql-unit-src && cd postgresql-unit-src && tar xzf ../postgresql-unit.tar.gz --strip-components=1 -C . && \
-    make -j $(getconf _NPROCESSORS_ONLN) && \
-    make -j $(getconf _NPROCESSORS_ONLN) install && \
+    make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
+    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
     # unit extension's "create extension" script relies on absolute install path to fill some reference tables.
     # We move the extension from '/usr/local/pgsql/' to '/usr/local/'  after it is build. So we need to adjust the path.
     # This one-liner removes pgsql/ part of the path.
@@ -346,8 +350,9 @@ RUN wget https://github.com/df7cb/postgresql-unit/archive/refs/tags/7.9.tar.gz -
 # compile pgvector extension
 #
 #########################################################################################
-FROM pg-build AS vector-pg-build
+FROM build-deps AS vector-pg-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 COPY compute/patches/pgvector.patch /pgvector.patch
 
@@ -361,8 +366,8 @@ RUN wget https://github.com/pgvector/pgvector/archive/refs/tags/v0.8.0.tar.gz -O
     echo "867a2c328d4928a5a9d6f052cd3bc78c7d60228a9b914ad32aa3db88e9de27b0 pgvector.tar.gz" | sha256sum --check && \
     mkdir pgvector-src && cd pgvector-src && tar xzf ../pgvector.tar.gz --strip-components=1 -C . && \
     patch -p1 < /pgvector.patch && \
-    make -j $(getconf _NPROCESSORS_ONLN) OPTFLAGS="" && \
-    make -j $(getconf _NPROCESSORS_ONLN) OPTFLAGS="" install && \
+    make -j $(getconf _NPROCESSORS_ONLN) OPTFLAGS="" PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
+    make -j $(getconf _NPROCESSORS_ONLN) OPTFLAGS="" install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/vector.control
 
 #########################################################################################
@@ -371,15 +376,16 @@ RUN wget https://github.com/pgvector/pgvector/archive/refs/tags/v0.8.0.tar.gz -O
 # compile pgjwt extension
 #
 #########################################################################################
-FROM pg-build AS pgjwt-pg-build
+FROM build-deps AS pgjwt-pg-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 # not version-specific
 # doesn't use releases, last commit f3d82fd - Mar 2, 2023
 RUN wget https://github.com/michelp/pgjwt/archive/f3d82fd30151e754e19ce5d6a06c71c20689ce3d.tar.gz -O pgjwt.tar.gz && \
     echo "dae8ed99eebb7593b43013f6532d772b12dfecd55548d2673f2dfd0163f6d2b9 pgjwt.tar.gz" | sha256sum --check && \
     mkdir pgjwt-src && cd pgjwt-src && tar xzf ../pgjwt.tar.gz --strip-components=1 -C . && \
-    make -j $(getconf _NPROCESSORS_ONLN) install && \
+    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/pgjwt.control
 
 #########################################################################################
@@ -388,16 +394,17 @@ RUN wget https://github.com/michelp/pgjwt/archive/f3d82fd30151e754e19ce5d6a06c71
 # compile hypopg extension
 #
 #########################################################################################
-FROM pg-build AS hypopg-pg-build
+FROM build-deps AS hypopg-pg-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 # HypoPG 1.4.1 supports v17
 # last release 1.4.1 - Apr 28, 2024
 RUN wget https://github.com/HypoPG/hypopg/archive/refs/tags/1.4.1.tar.gz -O hypopg.tar.gz && \
     echo "9afe6357fd389d8d33fad81703038ce520b09275ec00153c6c89282bcdedd6bc hypopg.tar.gz" | sha256sum --check && \
     mkdir hypopg-src && cd hypopg-src && tar xzf ../hypopg.tar.gz --strip-components=1 -C . && \
-    make -j $(getconf _NPROCESSORS_ONLN) && \
-    make -j $(getconf _NPROCESSORS_ONLN) install && \
+    make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
+    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/hypopg.control
 
 #########################################################################################
@@ -406,16 +413,17 @@ RUN wget https://github.com/HypoPG/hypopg/archive/refs/tags/1.4.1.tar.gz -O hypo
 # compile pg_hashids extension
 #
 #########################################################################################
-FROM pg-build AS pg-hashids-pg-build
+FROM build-deps AS pg-hashids-pg-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 # not version-specific
 # last release v1.2.1 -Jan 12, 2018
 RUN wget https://github.com/iCyberon/pg_hashids/archive/refs/tags/v1.2.1.tar.gz -O pg_hashids.tar.gz && \
     echo "74576b992d9277c92196dd8d816baa2cc2d8046fe102f3dcd7f3c3febed6822a pg_hashids.tar.gz" | sha256sum --check && \
     mkdir pg_hashids-src && cd pg_hashids-src && tar xzf ../pg_hashids.tar.gz --strip-components=1 -C . && \
-    make -j $(getconf _NPROCESSORS_ONLN) USE_PGXS=1 && \
-    make -j $(getconf _NPROCESSORS_ONLN) install USE_PGXS=1 && \
+    make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config USE_PGXS=1 && \
+    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config USE_PGXS=1 && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/pg_hashids.control
 
 #########################################################################################
@@ -424,8 +432,9 @@ RUN wget https://github.com/iCyberon/pg_hashids/archive/refs/tags/v1.2.1.tar.gz
 # compile rum extension
 #
 #########################################################################################
-FROM pg-build AS rum-pg-build
+FROM build-deps AS rum-pg-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 COPY compute/patches/rum.patch /rum.patch
 
@@ -436,8 +445,8 @@ RUN wget https://github.com/postgrespro/rum/archive/cb1edffc57736cd2a4455f8d0fea
     echo "65e0a752e99f4c3226400c9b899f997049e93503db8bf5c8072efa136d32fd83 rum.tar.gz" | sha256sum --check && \
     mkdir rum-src && cd rum-src && tar xzf ../rum.tar.gz --strip-components=1 -C . && \
     patch -p1 < /rum.patch && \
-    make -j $(getconf _NPROCESSORS_ONLN) USE_PGXS=1 && \
-    make -j $(getconf _NPROCESSORS_ONLN) install USE_PGXS=1 && \
+    make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config USE_PGXS=1 && \
+    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config USE_PGXS=1 && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/rum.control
 
 #########################################################################################
@@ -446,16 +455,17 @@ RUN wget https://github.com/postgrespro/rum/archive/cb1edffc57736cd2a4455f8d0fea
 # compile pgTAP extension
 #
 #########################################################################################
-FROM pg-build AS pgtap-pg-build
+FROM build-deps AS pgtap-pg-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 # pgtap 1.3.3 supports v17
 # last release v1.3.3 - Apr 8, 2024
 RUN wget https://github.com/theory/pgtap/archive/refs/tags/v1.3.3.tar.gz -O pgtap.tar.gz && \
     echo "325ea79d0d2515bce96bce43f6823dcd3effbd6c54cb2a4d6c2384fffa3a14c7 pgtap.tar.gz" | sha256sum --check && \
     mkdir pgtap-src && cd pgtap-src && tar xzf ../pgtap.tar.gz --strip-components=1 -C . && \
-    make -j $(getconf _NPROCESSORS_ONLN) && \
-    make -j $(getconf _NPROCESSORS_ONLN) install && \
+    make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
+    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/pgtap.control
 
 #########################################################################################
@@ -464,16 +474,17 @@ RUN wget https://github.com/theory/pgtap/archive/refs/tags/v1.3.3.tar.gz -O pgta
 # compile ip4r extension
 #
 #########################################################################################
-FROM pg-build AS ip4r-pg-build
+FROM build-deps AS ip4r-pg-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 # not version-specific
 # last release v2.4.2 - Jul 29, 2023
 RUN wget https://github.com/RhodiumToad/ip4r/archive/refs/tags/2.4.2.tar.gz -O ip4r.tar.gz && \
     echo "0f7b1f159974f49a47842a8ab6751aecca1ed1142b6d5e38d81b064b2ead1b4b ip4r.tar.gz" | sha256sum --check && \
     mkdir ip4r-src && cd ip4r-src && tar xzf ../ip4r.tar.gz --strip-components=1 -C . && \
-    make -j $(getconf _NPROCESSORS_ONLN) && \
-    make -j $(getconf _NPROCESSORS_ONLN) install && \
+    make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
+    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/ip4r.control
 
 #########################################################################################
@@ -482,16 +493,17 @@ RUN wget https://github.com/RhodiumToad/ip4r/archive/refs/tags/2.4.2.tar.gz -O i
 # compile Prefix extension
 #
 #########################################################################################
-FROM pg-build AS prefix-pg-build
+FROM build-deps AS prefix-pg-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 # not version-specific
 # last release v1.2.10  - Jul 5, 2023
 RUN wget https://github.com/dimitri/prefix/archive/refs/tags/v1.2.10.tar.gz -O prefix.tar.gz && \
     echo "4342f251432a5f6fb05b8597139d3ccde8dcf87e8ca1498e7ee931ca057a8575 prefix.tar.gz" | sha256sum --check && \
     mkdir prefix-src && cd prefix-src && tar xzf ../prefix.tar.gz --strip-components=1 -C . && \
-    make -j $(getconf _NPROCESSORS_ONLN) && \
-    make -j $(getconf _NPROCESSORS_ONLN) install && \
+    make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
+    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/prefix.control
 
 #########################################################################################
@@ -500,16 +512,17 @@ RUN wget https://github.com/dimitri/prefix/archive/refs/tags/v1.2.10.tar.gz -O p
 # compile hll extension
 #
 #########################################################################################
-FROM pg-build AS hll-pg-build
+FROM build-deps AS hll-pg-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 # not version-specific
 # last release v2.18 - Aug 29, 2023
 RUN wget https://github.com/citusdata/postgresql-hll/archive/refs/tags/v2.18.tar.gz -O hll.tar.gz && \
     echo "e2f55a6f4c4ab95ee4f1b4a2b73280258c5136b161fe9d059559556079694f0e hll.tar.gz" | sha256sum --check && \
     mkdir hll-src && cd hll-src && tar xzf ../hll.tar.gz --strip-components=1 -C . && \
-    make -j $(getconf _NPROCESSORS_ONLN) && \
-    make -j $(getconf _NPROCESSORS_ONLN) install && \
+    make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
+    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/hll.control
 
 #########################################################################################
@@ -518,16 +531,17 @@ RUN wget https://github.com/citusdata/postgresql-hll/archive/refs/tags/v2.18.tar
 # compile plpgsql_check extension
 #
 #########################################################################################
-FROM pg-build AS plpgsql-check-pg-build
+FROM build-deps AS plpgsql-check-pg-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 # plpgsql_check v2.7.11 supports v17
 # last release v2.7.11 - Sep 16, 2024
 RUN wget https://github.com/okbob/plpgsql_check/archive/refs/tags/v2.7.11.tar.gz -O plpgsql_check.tar.gz && \
     echo "208933f8dbe8e0d2628eb3851e9f52e6892b8e280c63700c0f1ce7883625d172 plpgsql_check.tar.gz" | sha256sum --check && \
     mkdir plpgsql_check-src && cd plpgsql_check-src && tar xzf ../plpgsql_check.tar.gz --strip-components=1 -C . && \
-    make -j $(getconf _NPROCESSORS_ONLN) USE_PGXS=1 && \
-    make -j $(getconf _NPROCESSORS_ONLN) install USE_PGXS=1 && \
+    make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config USE_PGXS=1 && \
+    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config USE_PGXS=1 && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/plpgsql_check.control
 
 #########################################################################################
@@ -536,8 +550,11 @@ RUN wget https://github.com/okbob/plpgsql_check/archive/refs/tags/v2.7.11.tar.gz
 # compile timescaledb extension
 #
 #########################################################################################
-FROM pg-build AS timescaledb-pg-build
+FROM build-deps AS timescaledb-pg-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
 ARG PG_VERSION
+ENV PATH="/usr/local/pgsql/bin:$PATH"
 
 RUN case "${PG_VERSION}" in \
       "v14" | "v15") \
@@ -568,8 +585,11 @@ RUN case "${PG_VERSION}" in \
 # compile pg_hint_plan extension
 #
 #########################################################################################
-FROM pg-build AS pg-hint-plan-pg-build
+FROM build-deps AS pg-hint-plan-pg-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
 ARG PG_VERSION
+ENV PATH="/usr/local/pgsql/bin:$PATH"
 
 # version-specific, has separate releases for each version
 RUN case "${PG_VERSION}" in \
@@ -607,12 +627,14 @@ RUN case "${PG_VERSION}" in \
 # compile pg_cron extension
 #
 #########################################################################################
-FROM pg-build AS pg-cron-pg-build
+FROM build-deps AS pg-cron-pg-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 # This is an experimental extension that we do not support on prod yet.
 # !Do not remove!
 # We set it in shared_preload_libraries and computes will fail to start if library is not found.
+ENV PATH="/usr/local/pgsql/bin/:$PATH"
 RUN wget https://github.com/citusdata/pg_cron/archive/refs/tags/v1.6.4.tar.gz -O pg_cron.tar.gz && \
     echo "52d1850ee7beb85a4cb7185731ef4e5a90d1de216709d8988324b0d02e76af61 pg_cron.tar.gz" | sha256sum --check && \
     mkdir pg_cron-src && cd pg_cron-src && tar xzf ../pg_cron.tar.gz --strip-components=1 -C . && \
@@ -626,8 +648,9 @@ RUN wget https://github.com/citusdata/pg_cron/archive/refs/tags/v1.6.4.tar.gz -O
 # compile rdkit extension
 #
 #########################################################################################
-FROM pg-build AS rdkit-pg-build
+FROM build-deps AS rdkit-pg-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 RUN apt update && \
     apt install --no-install-recommends --no-install-suggests -y \
@@ -645,13 +668,7 @@ RUN apt update && \
 # Use new version only for v17
 # because Release_2024_09_1 has some backward incompatible changes
 # https://github.com/rdkit/rdkit/releases/tag/Release_2024_09_1
-
-# XXX: /usr/local/pgsql/bin is already in PATH, and that should be enough to find
-# pg_config. For some reason the rdkit cmake script doesn't work with just that,
-# however. By also adding /usr/local/pgsql, it works, which is weird because there
-# are no executables in that directory.
-ENV PATH="/usr/local/pgsql:$PATH"
-
+ENV PATH="/usr/local/pgsql/bin/:/usr/local/pgsql/:$PATH"
 RUN case "${PG_VERSION}" in \
     "v17") \
         export RDKIT_VERSION=Release_2024_09_1 \
@@ -704,11 +721,13 @@ RUN case "${PG_VERSION}" in \
 # compile pg_uuidv7 extension
 #
 #########################################################################################
-FROM pg-build AS pg-uuidv7-pg-build
+FROM build-deps AS pg-uuidv7-pg-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 # not version-specific
 # last release v1.6.0 - Oct 9, 2024
+ENV PATH="/usr/local/pgsql/bin/:$PATH"
 RUN wget https://github.com/fboulnois/pg_uuidv7/archive/refs/tags/v1.6.0.tar.gz -O pg_uuidv7.tar.gz && \
     echo "0fa6c710929d003f6ce276a7de7a864e9d1667b2d78be3dc2c07f2409eb55867 pg_uuidv7.tar.gz" | sha256sum --check && \
     mkdir pg_uuidv7-src && cd pg_uuidv7-src && tar xzf ../pg_uuidv7.tar.gz --strip-components=1 -C . && \
@@ -722,11 +741,13 @@ RUN wget https://github.com/fboulnois/pg_uuidv7/archive/refs/tags/v1.6.0.tar.gz
 # compile pg_roaringbitmap extension
 #
 #########################################################################################
-FROM pg-build AS pg-roaringbitmap-pg-build
+FROM build-deps AS pg-roaringbitmap-pg-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 # not version-specific
 # last release v0.5.4 - Jun 28, 2022
+ENV PATH="/usr/local/pgsql/bin/:$PATH"
 RUN wget https://github.com/ChenHuajun/pg_roaringbitmap/archive/refs/tags/v0.5.4.tar.gz -O pg_roaringbitmap.tar.gz && \
     echo "b75201efcb1c2d1b014ec4ae6a22769cc7a224e6e406a587f5784a37b6b5a2aa pg_roaringbitmap.tar.gz" | sha256sum --check && \
     mkdir pg_roaringbitmap-src && cd pg_roaringbitmap-src && tar xzf ../pg_roaringbitmap.tar.gz --strip-components=1 -C . && \
@@ -740,14 +761,16 @@ RUN wget https://github.com/ChenHuajun/pg_roaringbitmap/archive/refs/tags/v0.5.4
 # compile pg_semver extension
 #
 #########################################################################################
-FROM pg-build AS pg-semver-pg-build
+FROM build-deps AS pg-semver-pg-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 # Release 0.40.0 breaks backward compatibility with previous versions
 # see release note https://github.com/theory/pg-semver/releases/tag/v0.40.0
 # Use new version only for v17
 #
 # last release v0.40.0 - Jul 22, 2024
+ENV PATH="/usr/local/pgsql/bin/:$PATH"
 RUN case "${PG_VERSION}" in \
     "v17") \
         export SEMVER_VERSION=0.40.0 \
@@ -774,11 +797,13 @@ RUN case "${PG_VERSION}" in \
 # compile pg_embedding extension
 #
 #########################################################################################
-FROM pg-build AS pg-embedding-pg-build
+FROM build-deps AS pg-embedding-pg-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 # This is our extension, support stopped in favor of pgvector
 # TODO: deprecate it
 ARG PG_VERSION
+ENV PATH="/usr/local/pgsql/bin/:$PATH"
 RUN case "${PG_VERSION}" in \
       "v14" | "v15") \
         export PG_EMBEDDING_VERSION=0.3.5 \
@@ -799,18 +824,20 @@ RUN case "${PG_VERSION}" in \
 # compile anon extension
 #
 #########################################################################################
-FROM pg-build AS pg-anon-pg-build
+FROM build-deps AS pg-anon-pg-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 # This is an experimental extension, never got to real production.
 # !Do not remove! It can be present in shared_preload_libraries and compute will fail to start if library is not found.
+ENV PATH="/usr/local/pgsql/bin/:$PATH"
 RUN case "${PG_VERSION}" in "v17") \
     echo "postgresql_anonymizer does not yet support PG17" && exit 0;; \
     esac && \
     wget  https://github.com/neondatabase/postgresql_anonymizer/archive/refs/tags/neon_1.1.1.tar.gz -O pg_anon.tar.gz && \
     echo "321ea8d5c1648880aafde850a2c576e4a9e7b9933a34ce272efc839328999fa9  pg_anon.tar.gz" | sha256sum --check && \
     mkdir pg_anon-src && cd pg_anon-src && tar xzf ../pg_anon.tar.gz --strip-components=1 -C . && \
-    make -j $(getconf _NPROCESSORS_ONLN) install && \
+    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/anon.control
 
 #########################################################################################
@@ -819,8 +846,9 @@ RUN case "${PG_VERSION}" in "v17") \
 # This layer is used to build `pgrx` deps
 #
 #########################################################################################
-FROM pg-build AS rust-extensions-build
+FROM build-deps AS rust-extensions-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 RUN apt update && \
     apt install --no-install-recommends --no-install-suggests -y curl libclang-dev && \
@@ -828,7 +856,7 @@ RUN apt update && \
     useradd -ms /bin/bash nonroot -b /home
 
 ENV HOME=/home/nonroot
-ENV PATH="/home/nonroot/.cargo/bin:$PATH"
+ENV PATH="/home/nonroot/.cargo/bin:/usr/local/pgsql/bin/:$PATH"
 USER nonroot
 WORKDIR /home/nonroot
 
@@ -855,8 +883,9 @@ USER root
 # and eventually get merged with `rust-extensions-build`
 #
 #########################################################################################
-FROM pg-build AS rust-extensions-build-pgrx12
+FROM build-deps AS rust-extensions-build-pgrx12
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 RUN apt update && \
     apt install --no-install-recommends --no-install-suggests -y curl libclang-dev && \
@@ -864,7 +893,7 @@ RUN apt update && \
     useradd -ms /bin/bash nonroot -b /home
 
 ENV HOME=/home/nonroot
-ENV PATH="/home/nonroot/.cargo/bin:$PATH"
+ENV PATH="/home/nonroot/.cargo/bin:/usr/local/pgsql/bin/:$PATH"
 USER nonroot
 WORKDIR /home/nonroot
 
@@ -872,7 +901,7 @@ RUN curl -sSO https://static.rust-lang.org/rustup/dist/$(uname -m)-unknown-linux
     chmod +x rustup-init && \
     ./rustup-init -y --no-modify-path --profile minimal --default-toolchain stable && \
     rm rustup-init && \
-    cargo install --locked --version 0.12.9 cargo-pgrx && \
+    cargo install --locked --version 0.12.6 cargo-pgrx && \
     /bin/bash -c 'cargo pgrx init --pg${PG_VERSION:1}=/usr/local/pgsql/bin/pg_config'
 
 USER root
@@ -909,19 +938,19 @@ RUN apt update && apt install --no-install-recommends --no-install-suggests -y p
     mkdir pgrag-src && cd pgrag-src && tar xzf ../pgrag.tar.gz --strip-components=1 -C . && \
     \
     cd exts/rag && \
-    sed -i 's/pgrx = "0.12.6"/pgrx = { version = "0.12.9", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    sed -i 's/pgrx = "0.12.6"/pgrx = { version = "0.12.6", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
     cargo pgrx install --release && \
     echo "trusted = true" >> /usr/local/pgsql/share/extension/rag.control && \
     \
     cd ../rag_bge_small_en_v15 && \
-    sed -i 's/pgrx = "0.12.6"/pgrx = { version = "0.12.9", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    sed -i 's/pgrx = "0.12.6"/pgrx = { version = "0.12.6", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
     ORT_LIB_LOCATION=/home/nonroot/onnxruntime-src/build/Linux \
         REMOTE_ONNX_URL=http://pg-ext-s3-gateway/pgrag-data/bge_small_en_v15.onnx \
         cargo pgrx install --release --features remote_onnx && \
     echo "trusted = true" >> /usr/local/pgsql/share/extension/rag_bge_small_en_v15.control && \
     \
     cd ../rag_jina_reranker_v1_tiny_en && \
-    sed -i 's/pgrx = "0.12.6"/pgrx = { version = "0.12.9", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    sed -i 's/pgrx = "0.12.6"/pgrx = { version = "0.12.6", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
     ORT_LIB_LOCATION=/home/nonroot/onnxruntime-src/build/Linux \
         REMOTE_ONNX_URL=http://pg-ext-s3-gateway/pgrag-data/jina_reranker_v1_tiny_en.onnx \
         cargo pgrx install --release --features remote_onnx && \
@@ -946,8 +975,7 @@ RUN wget https://github.com/supabase/pg_jsonschema/archive/refs/tags/v0.3.3.tar.
     # against postgres forks that decided to change their ABI name (like us).
     # With that we can build extensions without forking them and using stock
     # pgx. As this feature is new few manual version bumps were required.
-    sed -i 's/pgrx = "0.12.6"/pgrx = { version = "0.12.9", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
-    sed -i 's/pgrx-tests = "0.12.6"/pgrx-tests = "0.12.9"/g' Cargo.toml && \
+    sed -i 's/pgrx = "0.12.6"/pgrx = { version = "0.12.6", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
     cargo pgrx install --release && \
     echo "trusted = true" >> /usr/local/pgsql/share/extension/pg_jsonschema.control
 
@@ -965,8 +993,7 @@ ARG PG_VERSION
 RUN wget https://github.com/supabase/pg_graphql/archive/refs/tags/v1.5.9.tar.gz -O pg_graphql.tar.gz && \
     echo "cf768385a41278be1333472204fc0328118644ae443182cf52f7b9b23277e497 pg_graphql.tar.gz" | sha256sum --check && \
     mkdir pg_graphql-src && cd pg_graphql-src && tar xzf ../pg_graphql.tar.gz --strip-components=1 -C . && \
-    sed -i 's/pgrx = "=0.12.6"/pgrx = { version = "=0.12.9", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
-    sed -i 's/pgrx-tests = "=0.12.6"/pgrx-tests = "=0.12.9"/g' Cargo.toml && \
+    sed -i 's/pgrx = "=0.12.6"/pgrx = { version = "0.12.6", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
     cargo pgrx install --release && \
     # it's needed to enable extension because it uses untrusted C language
     sed -i 's/superuser = false/superuser = true/g' /usr/local/pgsql/share/extension/pg_graphql.control && \
@@ -987,58 +1014,33 @@ ARG PG_VERSION
 RUN wget https://github.com/kelvich/pg_tiktoken/archive/9118dd4549b7d8c0bbc98e04322499f7bf2fa6f7.tar.gz -O pg_tiktoken.tar.gz && \
     echo "a5bc447e7920ee149d3c064b8b9f0086c0e83939499753178f7d35788416f628 pg_tiktoken.tar.gz" | sha256sum --check && \
     mkdir pg_tiktoken-src && cd pg_tiktoken-src && tar xzf ../pg_tiktoken.tar.gz --strip-components=1 -C . && \
-    sed -i 's/pgrx = { version = "=0.12.6",/pgrx = { version = "0.12.9",/g' Cargo.toml && \
-    sed -i 's/pgrx-tests = "=0.12.6"/pgrx-tests = "0.12.9"/g' Cargo.toml && \
+    # TODO update pgrx version in the pg_tiktoken repo and remove this line
+    sed -i 's/pgrx = { version = "=0.10.2",/pgrx = { version = "0.11.3",/g' Cargo.toml && \
+    sed -i 's/pgrx-tests = "=0.10.2"/pgrx-tests = "0.11.3"/g' Cargo.toml && \
     cargo pgrx install --release && \
     echo "trusted = true" >> /usr/local/pgsql/share/extension/pg_tiktoken.control
 
 #########################################################################################
 #
 # Layer "pg-pgx-ulid-build"
-# Compile "pgx_ulid" extension for v16 and below
+# Compile "pgx_ulid" extension
 #
 #########################################################################################
 
 FROM rust-extensions-build AS pg-pgx-ulid-build
 ARG PG_VERSION
 
-RUN case "${PG_VERSION}" in \
-    "v14" | "v15" | "v16") \
-        ;; \
-    *) \
-        echo "skipping the version of pgx_ulid for $PG_VERSION" && exit 0 \
-        ;; \
+# doesn't support v17 yet
+# https://github.com/pksunkara/pgx_ulid/pull/52
+RUN case "${PG_VERSION}" in "v17") \
+    echo "pgx_ulid does not support pg17 as of the latest version (0.1.5)" && exit 0;; \
     esac && \
     wget https://github.com/pksunkara/pgx_ulid/archive/refs/tags/v0.1.5.tar.gz -O pgx_ulid.tar.gz && \
-    echo "9d1659a2da65af0133d5451c454de31b37364e3502087dadf579f790bc8bef17  pgx_ulid.tar.gz" | sha256sum --check && \
+    echo "9d1659a2da65af0133d5451c454de31b37364e3502087dadf579f790bc8bef17 pgx_ulid.tar.gz" | sha256sum --check && \
     mkdir pgx_ulid-src && cd pgx_ulid-src && tar xzf ../pgx_ulid.tar.gz --strip-components=1 -C . && \
-    sed -i 's/pgrx       = "^0.11.2"/pgrx       = { version = "0.11.3", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    sed -i 's/pgrx       = "^0.11.2"/pgrx = { version = "=0.11.3", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
     cargo pgrx install --release && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/ulid.control
-
-#########################################################################################
-#
-# Layer "pg-pgx-ulid-pgrx12-build"
-# Compile "pgx_ulid" extension for v17 and up
-#
-#########################################################################################
-
-FROM rust-extensions-build-pgrx12 AS pg-pgx-ulid-pgrx12-build
-ARG PG_VERSION
-
-RUN case "${PG_VERSION}" in \
-    "v17") \
-        ;; \
-    *) \
-        echo "skipping the version of pgx_ulid for $PG_VERSION" && exit 0 \
-        ;; \
-    esac && \
-    wget https://github.com/pksunkara/pgx_ulid/archive/refs/tags/v0.2.0.tar.gz -O pgx_ulid.tar.gz && \
-    echo "cef6a9a2e5e7bd1a10a18989286586ee9e6c1c06005a4055cff190de41bf3e9f pgx_ulid.tar.gz" | sha256sum --check && \
-    mkdir pgx_ulid-src && cd pgx_ulid-src && tar xzf ../pgx_ulid.tar.gz --strip-components=1 -C . && \
-    sed -i 's/pgrx       = "^0.12.7"/pgrx       = { version = "0.12.9", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
-    cargo pgrx install --release && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/pgx_ulid.control
+    echo "trusted = true" >> /usr/local/pgsql/share/extension/ulid.control
 
 #########################################################################################
 #
@@ -1056,11 +1058,7 @@ ARG PG_VERSION
 RUN wget https://github.com/neondatabase/pg_session_jwt/archive/refs/tags/v0.2.0.tar.gz -O pg_session_jwt.tar.gz && \
     echo "5ace028e591f2e000ca10afa5b1ca62203ebff014c2907c0ec3b29c36f28a1bb pg_session_jwt.tar.gz" | sha256sum --check && \
     mkdir pg_session_jwt-src && cd pg_session_jwt-src && tar xzf ../pg_session_jwt.tar.gz --strip-components=1 -C . && \
-    sed -i 's/pgrx = "0.12.6"/pgrx = { version = "0.12.9", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
-    sed -i 's/version = "0.12.6"/version = "0.12.9"/g' pgrx-tests/Cargo.toml && \
-    sed -i 's/pgrx = "=0.12.6"/pgrx = { version = "=0.12.9", features = [ "unsafe-postgres" ] }/g' pgrx-tests/Cargo.toml && \
-    sed -i 's/pgrx-macros = "=0.12.6"/pgrx-macros = "=0.12.9"/g' pgrx-tests/Cargo.toml && \
-    sed -i 's/pgrx-pg-config = "=0.12.6"/pgrx-pg-config = "=0.12.9"/g' pgrx-tests/Cargo.toml && \
+    sed -i 's/pgrx = "0.12.6"/pgrx = { version = "=0.12.6", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
     cargo pgrx install --release
 
 #########################################################################################
@@ -1070,11 +1068,13 @@ RUN wget https://github.com/neondatabase/pg_session_jwt/archive/refs/tags/v0.2.0
 #
 #########################################################################################
 
-FROM pg-build AS wal2json-pg-build
+FROM build-deps AS wal2json-pg-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 # wal2json wal2json_2_6 supports v17
 # last release wal2json_2_6 - Apr 25, 2024
+ENV PATH="/usr/local/pgsql/bin/:$PATH"
 RUN wget https://github.com/eulerto/wal2json/archive/refs/tags/wal2json_2_6.tar.gz -O wal2json.tar.gz && \
     echo "18b4bdec28c74a8fc98a11c72de38378a760327ef8e5e42e975b0029eb96ba0d wal2json.tar.gz" | sha256sum --check && \
     mkdir wal2json-src && cd wal2json-src && tar xzf ../wal2json.tar.gz --strip-components=1 -C . && \
@@ -1087,11 +1087,13 @@ RUN wget https://github.com/eulerto/wal2json/archive/refs/tags/wal2json_2_6.tar.
 # compile pg_ivm extension
 #
 #########################################################################################
-FROM pg-build AS pg-ivm-build
+FROM build-deps AS pg-ivm-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 # pg_ivm v1.9 supports v17
 # last release v1.9 - Jul 31
+ENV PATH="/usr/local/pgsql/bin/:$PATH"
 RUN wget https://github.com/sraoss/pg_ivm/archive/refs/tags/v1.9.tar.gz -O pg_ivm.tar.gz && \
     echo "59e15722939f274650abf637f315dd723c87073496ca77236b044cb205270d8b pg_ivm.tar.gz" | sha256sum --check && \
     mkdir pg_ivm-src && cd pg_ivm-src && tar xzf ../pg_ivm.tar.gz --strip-components=1 -C . && \
@@ -1105,11 +1107,13 @@ RUN wget https://github.com/sraoss/pg_ivm/archive/refs/tags/v1.9.tar.gz -O pg_iv
 # compile pg_partman extension
 #
 #########################################################################################
-FROM pg-build AS pg-partman-build
+FROM build-deps AS pg-partman-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 # should support v17 https://github.com/pgpartman/pg_partman/discussions/693
 # last release 5.1.0  Apr 2, 2024
+ENV PATH="/usr/local/pgsql/bin/:$PATH"
 RUN wget https://github.com/pgpartman/pg_partman/archive/refs/tags/v5.1.0.tar.gz -O pg_partman.tar.gz && \
     echo "3e3a27d7ff827295d5c55ef72f07a49062d6204b3cb0b9a048645d6db9f3cb9f pg_partman.tar.gz" | sha256sum --check && \
     mkdir pg_partman-src && cd pg_partman-src && tar xzf ../pg_partman.tar.gz --strip-components=1 -C . && \
@@ -1125,6 +1129,9 @@ RUN wget https://github.com/pgpartman/pg_partman/archive/refs/tags/v5.1.0.tar.gz
 #########################################################################################
 FROM rust-extensions-build AS pg-mooncake-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+ENV PATH="/usr/local/pgsql/bin/:$PATH"
 
 RUN wget https://github.com/Mooncake-Labs/pg_mooncake/releases/download/v0.1.0/pg_mooncake-0.1.0.tar.gz -O pg_mooncake.tar.gz && \
     echo "eafd059b77f541f11525eb8affcd66a176968cbd8fe7c0d436e733f2aa4da59f pg_mooncake.tar.gz" | sha256sum --check && \
@@ -1140,8 +1147,11 @@ RUN wget https://github.com/Mooncake-Labs/pg_mooncake/releases/download/v0.1.0/p
 #
 #########################################################################################
 
-FROM pg-build AS pg-repack-build
+FROM build-deps AS pg-repack-build
 ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+ENV PATH="/usr/local/pgsql/bin/:$PATH"
 
 RUN wget https://github.com/reorg/pg_repack/archive/refs/tags/ver_1.5.2.tar.gz -O pg_repack.tar.gz && \
     echo '4516cad42251ed3ad53ff619733004db47d5755acac83f75924cd94d1c4fb681 pg_repack.tar.gz' | sha256sum --check && \
@@ -1183,7 +1193,6 @@ COPY --from=timescaledb-pg-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg-hint-plan-pg-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg-cron-pg-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg-pgx-ulid-build /usr/local/pgsql/ /usr/local/pgsql/
-COPY --from=pg-pgx-ulid-pgrx12-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg-session-jwt-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=rdkit-pg-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg-uuidv7-pg-build /usr/local/pgsql/ /usr/local/pgsql/
@@ -1265,11 +1274,11 @@ RUN set -e \
 #
 #########################################################################################
 
-FROM quay.io/prometheuscommunity/postgres-exporter:v0.16.0 AS postgres-exporter
+FROM quay.io/prometheuscommunity/postgres-exporter:v0.12.1 AS postgres-exporter
 
 # Keep the version the same as in build-tools.Dockerfile and
 # test_runner/regress/test_compute_metrics.py.
-FROM burningalchemist/sql_exporter:0.17.0 AS sql-exporter
+FROM burningalchemist/sql_exporter:0.16.0 AS sql-exporter
 
 #########################################################################################
 #

compute_tools/Cargo.toml

@@ -51,7 +51,6 @@ tracing-subscriber.workspace = true
 tracing-utils.workspace = true
 thiserror.workspace = true
 url.workspace = true
-uuid.workspace = true
 prometheus.workspace = true
 
 postgres_initdb.workspace = true

compute_tools/src/bin/fast_import.rs

@@ -31,7 +31,7 @@ use camino::{Utf8Path, Utf8PathBuf};
 use clap::Parser;
 use compute_tools::extension_server::{get_pg_version, PostgresMajorVersion};
 use nix::unistd::Pid;
-use tracing::{error, info, info_span, warn, Instrument};
+use tracing::{info, info_span, warn, Instrument};
 use utils::fs_ext::is_directory_empty;
 
 #[path = "fast_import/aws_s3_sync.rs"]
@@ -41,19 +41,12 @@ mod child_stdio_to_log;
 #[path = "fast_import/s3_uri.rs"]
 mod s3_uri;
 
-const PG_WAIT_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(600);
-const PG_WAIT_RETRY_INTERVAL: std::time::Duration = std::time::Duration::from_millis(300);
-
 #[derive(clap::Parser)]
 struct Args {
     #[clap(long)]
     working_directory: Utf8PathBuf,
     #[clap(long, env = "NEON_IMPORTER_S3_PREFIX")]
-    s3_prefix: Option<s3_uri::S3Uri>,
-    #[clap(long)]
-    source_connection_string: Option<String>,
-    #[clap(short, long)]
-    interactive: bool,
+    s3_prefix: s3_uri::S3Uri,
     #[clap(long)]
     pg_bin_dir: Utf8PathBuf,
     #[clap(long)]
@@ -84,70 +77,30 @@ pub(crate) async fn main() -> anyhow::Result<()> {
 
     info!("starting");
 
-    let args = Args::parse();
+    let Args {
+        working_directory,
+        s3_prefix,
+        pg_bin_dir,
+        pg_lib_dir,
+    } = Args::parse();
 
-    // Validate arguments
-    if args.s3_prefix.is_none() && args.source_connection_string.is_none() {
-        anyhow::bail!("either s3_prefix or source_connection_string must be specified");
-    }
-    if args.s3_prefix.is_some() && args.source_connection_string.is_some() {
-        anyhow::bail!("only one of s3_prefix or source_connection_string can be specified");
-    }
+    let aws_config = aws_config::load_defaults(BehaviorVersion::v2024_03_28()).await;
 
-    let working_directory = args.working_directory;
-    let pg_bin_dir = args.pg_bin_dir;
-    let pg_lib_dir = args.pg_lib_dir;
-
-    // Initialize AWS clients only if s3_prefix is specified
-    let (aws_config, kms_client) = if args.s3_prefix.is_some() {
-        let config = aws_config::load_defaults(BehaviorVersion::v2024_03_28()).await;
-        let kms = aws_sdk_kms::Client::new(&config);
-        (Some(config), Some(kms))
-    } else {
-        (None, None)
-    };
-
-    // Get source connection string either from S3 spec or direct argument
-    let source_connection_string = if let Some(s3_prefix) = &args.s3_prefix {
-        let spec: Spec = {
-            let spec_key = s3_prefix.append("/spec.json");
-            let s3_client = aws_sdk_s3::Client::new(aws_config.as_ref().unwrap());
-            let object = s3_client
-                .get_object()
-                .bucket(&spec_key.bucket)
-                .key(spec_key.key)
-                .send()
-                .await
-                .context("get spec from s3")?
-                .body
-                .collect()
-                .await
-                .context("download spec body")?;
-            serde_json::from_slice(&object.into_bytes()).context("parse spec as json")?
-        };
-
-        match spec.encryption_secret {
-            EncryptionSecret::KMS { key_id } => {
-                let mut output = kms_client
-                    .unwrap()
-                    .decrypt()
-                    .key_id(key_id)
-                    .ciphertext_blob(aws_sdk_s3::primitives::Blob::new(
-                        spec.source_connstring_ciphertext_base64,
-                    ))
-                    .send()
-                    .await
-                    .context("decrypt source connection string")?;
-                let plaintext = output
-                    .plaintext
-                    .take()
-                    .context("get plaintext source connection string")?;
-                String::from_utf8(plaintext.into_inner())
-                    .context("parse source connection string as utf8")?
-            }
-        }
-    } else {
-        args.source_connection_string.unwrap()
+    let spec: Spec = {
+        let spec_key = s3_prefix.append("/spec.json");
+        let s3_client = aws_sdk_s3::Client::new(&aws_config);
+        let object = s3_client
+            .get_object()
+            .bucket(&spec_key.bucket)
+            .key(spec_key.key)
+            .send()
+            .await
+            .context("get spec from s3")?
+            .body
+            .collect()
+            .await
+            .context("download spec body")?;
+        serde_json::from_slice(&object.into_bytes()).context("parse spec as json")?
     };
 
     match tokio::fs::create_dir(&working_directory).await {
@@ -170,6 +123,15 @@ pub(crate) async fn main() -> anyhow::Result<()> {
         .await
         .context("create pgdata directory")?;
 
+    //
+    // Setup clients
+    //
+    let aws_config = aws_config::load_defaults(BehaviorVersion::v2024_03_28()).await;
+    let kms_client = aws_sdk_kms::Client::new(&aws_config);
+
+    //
+    //  Initialize pgdata
+    //
     let pgbin = pg_bin_dir.join("postgres");
     let pg_version = match get_pg_version(pgbin.as_ref()) {
         PostgresMajorVersion::V14 => 14,
@@ -208,13 +170,7 @@ pub(crate) async fn main() -> anyhow::Result<()> {
         .args(["-c", &format!("max_parallel_workers={nproc}")])
         .args(["-c", &format!("max_parallel_workers_per_gather={nproc}")])
         .args(["-c", &format!("max_worker_processes={nproc}")])
-        .args([
-            "-c",
-            &format!(
-                "effective_io_concurrency={}",
-                if cfg!(target_os = "macos") { 0 } else { 100 }
-            ),
-        ])
+        .args(["-c", "effective_io_concurrency=100"])
         .env_clear()
         .stdout(std::process::Stdio::piped())
         .stderr(std::process::Stdio::piped())
@@ -229,58 +185,44 @@ pub(crate) async fn main() -> anyhow::Result<()> {
         )
         .instrument(info_span!("postgres")),
     );
-
-    // Create neondb database in the running postgres
     let restore_pg_connstring =
         format!("host=localhost port=5432 user={superuser} dbname=postgres");
-
-    let start_time = std::time::Instant::now();
-
     loop {
-        if start_time.elapsed() > PG_WAIT_TIMEOUT {
-            error!(
-                "timeout exceeded: failed to poll postgres and create database within 10 minutes"
-            );
-            std::process::exit(1);
-        }
-
-        match tokio_postgres::connect(&restore_pg_connstring, tokio_postgres::NoTls).await {
-            Ok((client, connection)) => {
-                // Spawn the connection handling task to maintain the connection
-                tokio::spawn(async move {
-                    if let Err(e) = connection.await {
-                        warn!("connection error: {}", e);
-                    }
-                });
-
-                match client.simple_query("CREATE DATABASE neondb;").await {
-                    Ok(_) => {
-                        info!("created neondb database");
-                        break;
-                    }
-                    Err(e) => {
-                        warn!(
-                            "failed to create database: {}, retying in {}s",
-                            e,
-                            PG_WAIT_RETRY_INTERVAL.as_secs_f32()
-                        );
-                        tokio::time::sleep(PG_WAIT_RETRY_INTERVAL).await;
-                        continue;
-                    }
-                }
-            }
-            Err(_) => {
-                info!(
-                    "postgres not ready yet, retrying in {}s",
-                    PG_WAIT_RETRY_INTERVAL.as_secs_f32()
-                );
-                tokio::time::sleep(PG_WAIT_RETRY_INTERVAL).await;
-                continue;
-            }
+        let res = tokio_postgres::connect(&restore_pg_connstring, tokio_postgres::NoTls).await;
+        if res.is_ok() {
+            info!("postgres is ready, could connect to it");
+            break;
         }
     }
 
-    let restore_pg_connstring = restore_pg_connstring.replace("dbname=postgres", "dbname=neondb");
+    //
+    // Decrypt connection string
+    //
+    let source_connection_string = {
+        match spec.encryption_secret {
+            EncryptionSecret::KMS { key_id } => {
+                let mut output = kms_client
+                    .decrypt()
+                    .key_id(key_id)
+                    .ciphertext_blob(aws_sdk_s3::primitives::Blob::new(
+                        spec.source_connstring_ciphertext_base64,
+                    ))
+                    .send()
+                    .await
+                    .context("decrypt source connection string")?;
+                let plaintext = output
+                    .plaintext
+                    .take()
+                    .context("get plaintext source connection string")?;
+                String::from_utf8(plaintext.into_inner())
+                    .context("parse source connection string as utf8")?
+            }
+        }
+    };
+
+    //
+    // Start the work
+    //
 
     let dumpdir = working_directory.join("dumpdir");
 
@@ -368,12 +310,6 @@ pub(crate) async fn main() -> anyhow::Result<()> {
         }
     }
 
-    // If interactive mode, wait for Ctrl+C
-    if args.interactive {
-        info!("Running in interactive mode. Press Ctrl+C to shut down.");
-        tokio::signal::ctrl_c().await.context("wait for ctrl-c")?;
-    }
-
     info!("shutdown postgres");
     {
         nix::sys::signal::kill(
@@ -389,24 +325,21 @@ pub(crate) async fn main() -> anyhow::Result<()> {
             .context("wait for postgres to shut down")?;
     }
 
-    // Only sync if s3_prefix was specified
-    if let Some(s3_prefix) = args.s3_prefix {
-        info!("upload pgdata");
-        aws_s3_sync::sync(Utf8Path::new(&pgdata_dir), &s3_prefix.append("/pgdata/"))
-            .await
-            .context("sync dump directory to destination")?;
+    info!("upload pgdata");
+    aws_s3_sync::sync(Utf8Path::new(&pgdata_dir), &s3_prefix.append("/pgdata/"))
+        .await
+        .context("sync dump directory to destination")?;
 
-        info!("write status");
-        {
-            let status_dir = working_directory.join("status");
-            std::fs::create_dir(&status_dir).context("create status directory")?;
-            let status_file = status_dir.join("pgdata");
-            std::fs::write(&status_file, serde_json::json!({"done": true}).to_string())
-                .context("write status file")?;
-            aws_s3_sync::sync(&status_dir, &s3_prefix.append("/status/"))
-                .await
-                .context("sync status directory to destination")?;
-        }
+    info!("write status");
+    {
+        let status_dir = working_directory.join("status");
+        std::fs::create_dir(&status_dir).context("create status directory")?;
+        let status_file = status_dir.join("pgdata");
+        std::fs::write(&status_file, serde_json::json!({"done": true}).to_string())
+            .context("write status file")?;
+        aws_s3_sync::sync(&status_dir, &s3_prefix.append("/status/"))
+            .await
+            .context("sync status directory to destination")?;
     }
 
     Ok(())

compute_tools/src/http/routes/extension_server.rs

@@ -17,8 +17,7 @@ use crate::{
 
 #[derive(Debug, Clone, Deserialize)]
 pub(in crate::http) struct ExtensionServerParams {
-    #[serde(default)]
-    is_library: bool,
+    is_library: Option<bool>,
 }
 
 /// Download a remote extension.
@@ -52,7 +51,7 @@ pub(in crate::http) async fn download_extension(
 
         remote_extensions.get_ext(
             &filename,
-            params.is_library,
+            params.is_library.unwrap_or(false),
             &compute.build_tag,
             &compute.pgversion,
         )

compute_tools/src/http/server.rs

@@ -1,14 +1,15 @@
 use std::{
     net::{IpAddr, Ipv6Addr, SocketAddr},
-    sync::Arc,
+    sync::{
+        atomic::{AtomicU64, Ordering},
+        Arc,
+    },
     thread,
     time::Duration,
 };
 
 use anyhow::Result;
 use axum::{
-    extract::Request,
-    middleware::{self, Next},
     response::{IntoResponse, Response},
     routing::{get, post},
     Router,
@@ -16,9 +17,11 @@ use axum::{
 use http::StatusCode;
 use tokio::net::TcpListener;
 use tower::ServiceBuilder;
-use tower_http::{request_id::PropagateRequestIdLayer, trace::TraceLayer};
+use tower_http::{
+    request_id::{MakeRequestId, PropagateRequestIdLayer, RequestId, SetRequestIdLayer},
+    trace::TraceLayer,
+};
 use tracing::{debug, error, info, Span};
-use uuid::Uuid;
 
 use super::routes::{
     check_writability, configure, database_schema, dbs_and_roles, extension_server, extensions,
@@ -31,24 +34,30 @@ async fn handle_404() -> Response {
     StatusCode::NOT_FOUND.into_response()
 }
 
-const X_REQUEST_ID: &str = "x-request-id";
+#[derive(Clone, Default)]
+struct ComputeMakeRequestId(Arc<AtomicU64>);
 
-/// This middleware function allows compute_ctl to generate its own request ID
-/// if one isn't supplied. The control plane will always send one as a UUID. The
-/// neon Postgres extension on the other hand does not send one.
-async fn maybe_add_request_id_header(mut request: Request, next: Next) -> Response {
-    let headers = request.headers_mut();
+impl MakeRequestId for ComputeMakeRequestId {
+    fn make_request_id<B>(
+        &mut self,
+        _request: &http::Request<B>,
+    ) -> Option<tower_http::request_id::RequestId> {
+        let request_id = self
+            .0
+            .fetch_add(1, Ordering::SeqCst)
+            .to_string()
+            .parse()
+            .unwrap();
 
-    if headers.get(X_REQUEST_ID).is_none() {
-        headers.append(X_REQUEST_ID, Uuid::new_v4().to_string().parse().unwrap());
+        Some(RequestId::new(request_id))
     }
-
-    next.run(request).await
 }
 
 /// Run the HTTP server and wait on it forever.
 #[tokio::main]
 async fn serve(port: u16, compute: Arc<ComputeNode>) {
+    const X_REQUEST_ID: &str = "x-request-id";
+
     let mut app = Router::new()
         .route("/check_writability", post(check_writability::is_writable))
         .route("/configure", post(configure::configure))
@@ -73,8 +82,9 @@ async fn serve(port: u16, compute: Arc<ComputeNode>) {
         .fallback(handle_404)
         .layer(
             ServiceBuilder::new()
-                // Add this middleware since we assume the request ID exists
-                .layer(middleware::from_fn(maybe_add_request_id_header))
+                .layer(SetRequestIdLayer::x_request_id(
+                    ComputeMakeRequestId::default(),
+                ))
                 .layer(
                     TraceLayer::new_for_http()
                         .on_request(|request: &http::Request<_>, _span: &Span| {

control_plane/storcon_cli/src/main.rs

@@ -1,17 +1,12 @@
 use futures::StreamExt;
-use std::{
-    collections::{HashMap, HashSet},
-    str::FromStr,
-    time::Duration,
-};
+use std::{str::FromStr, time::Duration};
 
 use clap::{Parser, Subcommand};
 use pageserver_api::{
     controller_api::{
         AvailabilityZone, NodeAvailabilityWrapper, NodeDescribeResponse, NodeShardResponse,
-        SafekeeperDescribeResponse, SafekeeperSchedulingPolicyRequest, ShardSchedulingPolicy,
-        ShardsPreferredAzsRequest, SkSchedulingPolicy, TenantCreateRequest, TenantDescribeResponse,
-        TenantPolicyRequest,
+        SafekeeperDescribeResponse, ShardSchedulingPolicy, TenantCreateRequest,
+        TenantDescribeResponse, TenantPolicyRequest,
     },
     models::{
         EvictionPolicy, EvictionPolicyLayerAccessThreshold, LocationConfigSecondary,
@@ -158,12 +153,6 @@ enum Command {
         #[arg(long)]
         tenant_id: TenantId,
     },
-    TenantSetPreferredAz {
-        #[arg(long)]
-        tenant_id: TenantId,
-        #[arg(long)]
-        preferred_az: Option<String>,
-    },
     /// Uncleanly drop a tenant from the storage controller: this doesn't delete anything from pageservers. Appropriate
     /// if you e.g. used `tenant-warmup` by mistake on a tenant ID that doesn't really exist, or is in some other region.
     TenantDrop {
@@ -232,13 +221,6 @@ enum Command {
     },
     /// List safekeepers known to the storage controller
     Safekeepers {},
-    /// Set the scheduling policy of the specified safekeeper
-    SafekeeperScheduling {
-        #[arg(long)]
-        node_id: NodeId,
-        #[arg(long)]
-        scheduling_policy: SkSchedulingPolicyArg,
-    },
 }
 
 #[derive(Parser)]
@@ -291,17 +273,6 @@ impl FromStr for PlacementPolicyArg {
     }
 }
 
-#[derive(Debug, Clone)]
-struct SkSchedulingPolicyArg(SkSchedulingPolicy);
-
-impl FromStr for SkSchedulingPolicyArg {
-    type Err = anyhow::Error;
-
-    fn from_str(s: &str) -> Result<Self, Self::Err> {
-        SkSchedulingPolicy::from_str(s).map(Self)
-    }
-}
-
 #[derive(Debug, Clone)]
 struct ShardSchedulingPolicyArg(ShardSchedulingPolicy);
 
@@ -431,12 +402,11 @@ async fn main() -> anyhow::Result<()> {
             resp.sort_by(|a, b| a.listen_http_addr.cmp(&b.listen_http_addr));
 
             let mut table = comfy_table::Table::new();
-            table.set_header(["Id", "Hostname", "AZ", "Scheduling", "Availability"]);
+            table.set_header(["Id", "Hostname", "Scheduling", "Availability"]);
             for node in resp {
                 table.add_row([
                     format!("{}", node.id),
                     node.listen_http_addr,
-                    node.availability_zone_id,
                     format!("{:?}", node.scheduling),
                     format!("{:?}", node.availability),
                 ]);
@@ -496,65 +466,33 @@ async fn main() -> anyhow::Result<()> {
             println!("{table}");
         }
         Command::Tenants { node_id: None } => {
-            // Set up output formatting
+            let mut resp = storcon_client
+                .dispatch::<(), Vec<TenantDescribeResponse>>(
+                    Method::GET,
+                    "control/v1/tenant".to_string(),
+                    None,
+                )
+                .await?;
+
+            resp.sort_by(|a, b| a.tenant_id.cmp(&b.tenant_id));
+
             let mut table = comfy_table::Table::new();
             table.set_header([
                 "TenantId",
-                "Preferred AZ",
                 "ShardCount",
                 "StripeSize",
                 "Placement",
                 "Scheduling",
             ]);
-
-            // Pagination loop over listing API
-            let mut start_after = None;
-            const LIMIT: usize = 1000;
-            loop {
-                let path = match start_after {
-                    None => format!("control/v1/tenant?limit={LIMIT}"),
-                    Some(start_after) => {
-                        format!("control/v1/tenant?limit={LIMIT}&start_after={start_after}")
-                    }
-                };
-
-                let resp = storcon_client
-                    .dispatch::<(), Vec<TenantDescribeResponse>>(Method::GET, path, None)
-                    .await?;
-
-                if resp.is_empty() {
-                    // End of data reached
-                    break;
-                }
-
-                // Give some visual feedback while we're building up the table (comfy_table doesn't have
-                // streaming output)
-                if resp.len() >= LIMIT {
-                    eprint!(".");
-                }
-
-                start_after = Some(resp.last().unwrap().tenant_id);
-
-                for tenant in resp {
-                    let shard_zero = tenant.shards.into_iter().next().unwrap();
-                    table.add_row([
-                        format!("{}", tenant.tenant_id),
-                        shard_zero
-                            .preferred_az_id
-                            .as_ref()
-                            .cloned()
-                            .unwrap_or("".to_string()),
-                        format!("{}", shard_zero.tenant_shard_id.shard_count.literal()),
-                        format!("{:?}", tenant.stripe_size),
-                        format!("{:?}", tenant.policy),
-                        format!("{:?}", shard_zero.scheduling_policy),
-                    ]);
-                }
-            }
-
-            // Terminate progress dots
-            if table.row_count() > LIMIT {
-                eprint!("");
+            for tenant in resp {
+                let shard_zero = tenant.shards.into_iter().next().unwrap();
+                table.add_row([
+                    format!("{}", tenant.tenant_id),
+                    format!("{}", shard_zero.tenant_shard_id.shard_count.literal()),
+                    format!("{:?}", tenant.stripe_size),
+                    format!("{:?}", tenant.policy),
+                    format!("{:?}", shard_zero.scheduling_policy),
+                ]);
             }
 
             println!("{table}");
@@ -676,19 +614,6 @@ async fn main() -> anyhow::Result<()> {
                     None,
                 )
                 .await?;
-
-            let nodes = storcon_client
-                .dispatch::<(), Vec<NodeDescribeResponse>>(
-                    Method::GET,
-                    "control/v1/node".to_string(),
-                    None,
-                )
-                .await?;
-            let nodes = nodes
-                .into_iter()
-                .map(|n| (n.id, n))
-                .collect::<HashMap<_, _>>();
-
             println!("Tenant {tenant_id}");
             let mut table = comfy_table::Table::new();
             table.add_row(["Policy", &format!("{:?}", policy)]);
@@ -697,14 +622,7 @@ async fn main() -> anyhow::Result<()> {
             println!("{table}");
             println!("Shards:");
             let mut table = comfy_table::Table::new();
-            table.set_header([
-                "Shard",
-                "Attached",
-                "Attached AZ",
-                "Secondary",
-                "Last error",
-                "status",
-            ]);
+            table.set_header(["Shard", "Attached", "Secondary", "Last error", "status"]);
             for shard in shards {
                 let secondary = shard
                     .node_secondary
@@ -727,18 +645,11 @@ async fn main() -> anyhow::Result<()> {
                 }
                 let status = status_parts.join(",");
 
-                let attached_node = shard
-                    .node_attached
-                    .as_ref()
-                    .map(|id| nodes.get(id).expect("Shard references nonexistent node"));
-
                 table.add_row([
                     format!("{}", shard.tenant_shard_id),
-                    attached_node
-                        .map(|n| format!("{} ({})", n.listen_http_addr, n.id))
-                        .unwrap_or(String::new()),
-                    attached_node
-                        .map(|n| n.availability_zone_id.clone())
+                    shard
+                        .node_attached
+                        .map(|n| format!("{}", n))
                         .unwrap_or(String::new()),
                     secondary,
                     shard.last_error,
@@ -747,66 +658,6 @@ async fn main() -> anyhow::Result<()> {
             }
             println!("{table}");
         }
-        Command::TenantSetPreferredAz {
-            tenant_id,
-            preferred_az,
-        } => {
-            // First learn about the tenant's shards
-            let describe_response = storcon_client
-                .dispatch::<(), TenantDescribeResponse>(
-                    Method::GET,
-                    format!("control/v1/tenant/{tenant_id}"),
-                    None,
-                )
-                .await?;
-
-            // Learn about nodes to validate the AZ ID
-            let nodes = storcon_client
-                .dispatch::<(), Vec<NodeDescribeResponse>>(
-                    Method::GET,
-                    "control/v1/node".to_string(),
-                    None,
-                )
-                .await?;
-
-            if let Some(preferred_az) = &preferred_az {
-                let azs = nodes
-                    .into_iter()
-                    .map(|n| (n.availability_zone_id))
-                    .collect::<HashSet<_>>();
-                if !azs.contains(preferred_az) {
-                    anyhow::bail!(
-                        "AZ {} not found on any node: known AZs are: {:?}",
-                        preferred_az,
-                        azs
-                    );
-                }
-            } else {
-                // Make it obvious to the user that since they've omitted an AZ, we're clearing it
-                eprintln!("Clearing preferred AZ for tenant {}", tenant_id);
-            }
-
-            // Construct a request that modifies all the tenant's shards
-            let req = ShardsPreferredAzsRequest {
-                preferred_az_ids: describe_response
-                    .shards
-                    .into_iter()
-                    .map(|s| {
-                        (
-                            s.tenant_shard_id,
-                            preferred_az.clone().map(AvailabilityZone),
-                        )
-                    })
-                    .collect(),
-            };
-            storcon_client
-                .dispatch::<ShardsPreferredAzsRequest, ()>(
-                    Method::PUT,
-                    "control/v1/preferred_azs".to_string(),
-                    Some(req),
-                )
-                .await?;
-        }
         Command::TenantWarmup { tenant_id } => {
             let describe_response = storcon_client
                 .dispatch::<(), TenantDescribeResponse>(
@@ -1221,23 +1072,6 @@ async fn main() -> anyhow::Result<()> {
             }
             println!("{table}");
         }
-        Command::SafekeeperScheduling {
-            node_id,
-            scheduling_policy,
-        } => {
-            let scheduling_policy = scheduling_policy.0;
-            storcon_client
-                .dispatch::<SafekeeperSchedulingPolicyRequest, ()>(
-                    Method::POST,
-                    format!("control/v1/safekeeper/{node_id}/scheduling_policy"),
-                    Some(SafekeeperSchedulingPolicyRequest { scheduling_policy }),
-                )
-                .await?;
-            println!(
-                "Scheduling policy of {node_id} set to {}",
-                String::from(scheduling_policy)
-            );
-        }
     }
 
     Ok(())

docs/rfcs/035-safekeeper-dynamic-membership-change.md

@@ -81,7 +81,7 @@ configuration generation in them is less than its current one. Namely, it
 refuses to vote, to truncate WAL in `handle_elected` and to accept WAL. In
 response it sends its current configuration generation to let walproposer know.
 
-Safekeeper gets `PUT /v1/tenants/{tenant_id}/timelines/{timeline_id}/configuration`
+Safekeeper gets `PUT /v1/tenants/{tenant_id}/timelines/{timeline_id}/configuration` 
 accepting `Configuration`. Safekeeper switches to the given conf it is higher than its
 current one and ignores it otherwise. In any case it replies with
 ```
@@ -103,7 +103,7 @@ currently and tries to communicate with all of them. However, the list does not
 define consensus members. Instead, on start walproposer tracks highest
 configuration it receives from `AcceptorGreeting`s. Once it assembles greetings
 from majority of `sk_set` and majority of `new_sk_set` (if it is present), it
-establishes this configuration as its own and moves to voting.
+establishes this configuration as its own and moves to voting. 
 
 It should stop talking to safekeepers not listed in the configuration at this
 point, though it is not unsafe to continue doing so.
@@ -119,7 +119,7 @@ refusal to accept due to configuration change) it simply restarts.
 The following algorithm can be executed anywhere having access to configuration
 storage and safekeepers. It is safe to interrupt / restart it and run multiple
 instances of it concurrently, though likely one of them won't make
-progress then. It accepts `desired_set: Vec<NodeId>` as input.
+progress then. It accepts `desired_set: Vec<NodeId>` as input. 
 
 Algorithm will refuse to make the change if it encounters previous interrupted
 change attempt, but in this case it will try to finish it.
@@ -140,7 +140,7 @@ storage are reachable.
    safe. Failed CAS aborts the procedure.
 4) Call `PUT` `configuration` on safekeepers from the current set,
    delivering them `joint_conf`. Collecting responses from majority is required
-   to proceed. If any response returned generation higher than
+   to proceed. If any response returned generation higher than 
    `joint_conf.generation`, abort (another switch raced us). Otherwise, choose
    max `<last_log_term, flush_lsn>` among responses and establish it as
    (in memory) `sync_position`. Also choose max `term` and establish it as (in
@@ -149,49 +149,49 @@ storage are reachable.
    without ack from the new set. Similarly, we'll bump term on new majority
    to `sync_term` so that two computes with the same term are never elected.
 4) Initialize timeline on safekeeper(s) from `new_sk_set` where it
-   doesn't exist yet by doing `pull_timeline` from the majority of the
+   doesn't exist yet by doing `pull_timeline` from the majority of the 
    current set. Doing that on majority of `new_sk_set` is enough to
    proceed, but it is reasonable to ensure that all `new_sk_set` members
    are initialized -- if some of them are down why are we migrating there?
-5) Call `POST` `bump_term(sync_term)` on safekeepers from the new set.
+5) Call `POST` `bump_term(sync_term)` on safekeepers from the new set. 
    Success on majority is enough.
 6) Repeatedly call `PUT` `configuration` on safekeepers from the new set,
    delivering them `joint_conf` and collecting their positions. This will
-   switch them to the `joint_conf` which generally won't be needed
+   switch them to the `joint_conf` which generally won't be needed 
    because `pull_timeline` already includes it and plus additionally would be
    broadcast by compute. More importantly, we may proceed to the next step
-   only when `<last_log_term, flush_lsn>` on the majority of the new set reached
-   `sync_position`. Similarly, on the happy path no waiting is not needed because
+   only when `<last_log_term, flush_lsn>` on the majority of the new set reached 
+   `sync_position`. Similarly, on the happy path no waiting is not needed because 
    `pull_timeline` already includes it. However, we should double
     check to be safe. For example, timeline could have been created earlier e.g.
-    manually or after try-to-migrate, abort, try-to-migrate-again sequence.
-7) Create `new_conf: Configuration` incrementing `join_conf` generation and having new
-   safekeeper set as `sk_set` and None `new_sk_set`. Write it to configuration
+    manually or after try-to-migrate, abort, try-to-migrate-again sequence. 
+7) Create `new_conf: Configuration` incrementing `join_conf` generation and having new 
+   safekeeper set as `sk_set` and None `new_sk_set`. Write it to configuration 
    storage under one more CAS.
 8) Call `PUT` `configuration` on safekeepers from the new set,
-   delivering them `new_conf`. It is enough to deliver it to the majority
+   delivering them `new_conf`. It is enough to deliver it to the majority 
    of the new set; the rest can be updated by compute.
 
 I haven't put huge effort to make the description above very precise, because it
 is natural language prone to interpretations anyway. Instead I'd like to make TLA+
 spec of it.
 
-Description above focuses on safety. To make the flow practical and live, here a few more
+Description above focuses on safety. To make the flow practical and live, here a few more 
 considerations.
-1) It makes sense to ping new set to ensure it we are migrating to live node(s) before
+1) It makes sense to ping new set to ensure it we are migrating to live node(s) before 
   step 3.
-2) If e.g. accidentally wrong new sk set has been specified, before CAS in step `6` is completed
+2) If e.g. accidentally wrong new sk set has been specified, before CAS in step `6` is completed 
    it is safe to rollback to the old conf with one more CAS.
-3) On step 4 timeline might be already created on members of the new set for various reasons;
+3) On step 4 timeline might be already created on members of the new set for various reasons; 
    the simplest is the procedure restart. There are more complicated scenarious like mentioned
-   in step 5. Deleting and re-doing `pull_timeline` is generally unsafe without involving
-   generations, so seems simpler to treat existing timeline as success. However, this also
+   in step 5. Deleting and re-doing `pull_timeline` is generally unsafe without involving 
+   generations, so seems simpler to treat existing timeline as success. However, this also 
    has a disadvantage: you might imagine an surpassingly unlikely schedule where condition in
    the step 5 is never reached until compute is (re)awaken up to synchronize new member(s).
    I don't think we'll observe this in practice, but can add waking up compute if needed.
 4) In the end timeline should be locally deleted on the safekeeper(s) which are
    in the old set but not in the new one, unless they are unreachable. To be
-   safe this also should be done under generation number (deletion proceeds only if
+   safe this also should be done under generation number (deletion proceeds only if 
    current configuration is <= than one in request and safekeeper is not memeber of it).
 5) If current conf fetched on step 1 is already not joint and members equal to `desired_set`,
    jump to step 7, using it as `new_conf`.
@@ -202,87 +202,47 @@ The procedure ought to be driven from somewhere. Obvious candidates are control
 plane and storage_controller; and as each of them already has db we don't want
 yet another storage. I propose to manage safekeepers in storage_controller
 because 1) since it is in rust it simplifies simulation testing (more on this
-below) 2) it already manages pageservers.
+below) 2) it already manages pageservers. 
 
 This assumes that migration will be fully usable only after we migrate all
 tenants/timelines to storage_controller. It is discussible whether we want also
 to manage pageserver attachments for all of these, but likely we do.
 
-This requires us to define storcon <-> cplane interface and changes.
+This requires us to define storcon <-> cplane interface.
 
-### storage_controller <-> control plane interface and changes
+### storage_controller <-> control plane interface
 
 First of all, control plane should
 [change](https://neondb.slack.com/archives/C03438W3FLZ/p1719226543199829)
 storing safekeepers per timeline instead of per tenant because we can't migrate
-tenants atomically.
+tenants atomically. 
 
 The important question is how updated configuration is delivered from
 storage_controller to control plane to provide it to computes. As always, there
 are two options, pull and push. Let's do it the same push as with pageserver
 `/notify-attach` because 1) it keeps storage_controller out of critical compute
-start path 2) uniformity. It makes storage_controller responsible for retrying
-notifying control plane until it succeeds.
+start path 2) provides easier upgrade: there won't be such a thing as 'timeline
+managed by control plane / storcon', cplane just takes the value out of its db
+when needed 3) uniformity. It makes storage_controller responsible for retrying notifying
+control plane until it succeeds.
 
-It is not needed for the control plane to fully know the `Configuration`. It is
-enough for it to only to be aware of the list of safekeepers in the latest
-configuration to supply it to compute, plus associated generation number to
-protect from stale update requests and to also pass it to compute.
-
-So, cplane `/notify-safekeepers` for the timeline can accept JSON like
-```
-{
-   tenant_id: String,
-   timeline_id: String,
-   generation: u32,
-   safekeepers: Vec<SafekeeperId>,
-}
-```
-where `SafekeeperId` is
-```
-{
-   node_id: u64,
-   host: String
-}
-```
-In principle `host` is redundant, but may be useful for observability.
-
-The request updates list of safekeepers in the db if the provided conf
-generation is higher (the cplane db should also store generations for this).
-Similarly to
-[`/notify-attach`](https://www.notion.so/neondatabase/Storage-Controller-Control-Plane-interface-6de56dd310a043bfa5c2f5564fa98365),
-it should update db which makes the call successful, and then try to schedule
-`apply_config` if possible, it is ok if not. storage_controller should rate
-limit calling the endpoint, but likely this won't be needed, as migration
+So, cplane `/notify-safekeepers` for the timeline accepts `Configuration` and
+updates it in the db if the provided conf generation is higher (the cplane db
+should also store generations for this). Similarly to [`/notify-attach`](https://www.notion.so/neondatabase/Storage-Controller-Control-Plane-interface-6de56dd310a043bfa5c2f5564fa98365), it
+should update db which makes the call successful, and then try to schedule
+`apply_config` if possible, it is ok if not. storage_controller 
+should rate limit calling the endpoint, but likely this won't be needed, as migration
 throughput is limited by `pull_timeline`.
 
 Timeline (branch) creation in cplane should call storage_controller POST
 `tenant/:tenant_id/timeline` like it currently does for sharded tenants.
-Response should be augmented with `safekeepers_generation` and `safekeepers`
-fields like described in `/notify-safekeepers` above. Initially (currently)
-these fields may be absent; in this case cplane chooses safekeepers on its own
-like it currently does. The call should be retried until succeeds.
+Response should be augmented with `safekeeper_conf: Configuration`. The call
+should be retried until succeeds.
 
 Timeline deletion and tenant deletion in cplane should call appropriate
 storage_controller endpoints like it currently does for sharded tenants. The
 calls should be retried until they succeed.
 
-When compute receives safekeepers list from control plane it needs to know the
-generation to checked whether it should be updated (note that compute may get
-safekeeper list from either cplane or safekeepers). Currently `neon.safekeepers`
-GUC is just a comma separates list of `host:port`. Let's prefix it with
-`g#<generation>:` to this end, so it will look like
-```
-g#42:safekeeper-0.eu-central-1.aws.neon.tech:6401,safekeeper-2.eu-central-1.aws.neon.tech:6401,safekeeper-1.eu-central-1.aws.neon.tech:6401
-```
-
-To summarize, list of cplane changes:
-- per tenant -> per timeline safekeepers management and addition of int `safekeeper_generation` field.
-- `/notify-safekeepers` endpoint.
-- Branch creation call may return list of safekeepers and when it is
-  present cplane should adopt it instead of choosing on its own like it does currently.
-- `neon.safekeepers` GUC should be prefixed with `g#<generation>:`.
-
 ### storage_controller implementation
 
 Current 'load everything on startup and keep in memory' easy design is fine.
@@ -400,10 +360,10 @@ source safekeeper might fail, which is not a problem if we are going to
 decomission the node but leaves garbage otherwise. I'd propose in the first version
 1) Don't attempt deletion at all if node status is `offline`.
 2) If it failed, just issue warning.
-And add PUT `/control/v1/safekeepers/:node_id/scrub` endpoint which would find and
-remove garbage timelines for manual use. It will 1) list all timelines on the
-safekeeper 2) compare each one against configuration storage: if timeline
-doesn't exist at all (had been deleted), it can be deleted. Otherwise, it can
+And add PUT `/control/v1/safekeepers/:node_id/scrub` endpoint which would find and 
+remove garbage timelines for manual use. It will 1) list all timelines on the 
+safekeeper 2) compare each one against configuration storage: if timeline 
+doesn't exist at all (had been deleted), it can be deleted. Otherwise, it can 
 be deleted under generation number if node is not member of current generation.
 
 Automating this is untrivial; we'd need to register all potential missing
@@ -452,8 +412,8 @@ There should be following layers of tests:
 3) Since simulation testing injects at relatively high level points (not
    syscalls), it omits some code, in particular `pull_timeline`. Thus it is
    better to have basic tests covering whole system as well. Extended version of
-   `test_restarts_under_load` would do: start background load and do migration
-   under it, then restart endpoint and check that no reported commits
+   `test_restarts_under_load` would do: start background load and do migration 
+   under it, then restart endpoint and check that no reported commits 
    had been lost. I'd also add one more creating classic network split scenario, with
    one compute talking to AC and another to BD while migration from nodes ABC to ABD
    happens.
@@ -462,51 +422,35 @@ There should be following layers of tests:
 
 ## Order of implementation and rollout
 
-Note that
+Note that 
 - Control plane parts and integration with it is fully independent from everything else
   (tests would use simulation and neon_local).
-- It is reasonable to make compute <-> safekeepers protocol change
-  independent of enabling generations.
 - There is a lot of infra work making storage_controller aware of timelines and safekeepers
   and its impl/rollout should be separate from migration itself.
-- Initially walproposer can just stop working while it observes joint configuration.
+- Initially walproposer can just stop working while it observers joint configuration.
   Such window would be typically very short anyway.
-- Obviously we want to test the whole thing thoroughly on staging and only then
-  gradually enable in prod.
 
-Let's have the following implementation bits for gradual rollout:
-- compute gets `neon.safekeepers_proto_version` flag.
-  Initially both compute and safekeepers will be able to talk both
-  versions so that we can delay force restart of them and for
-  simplicity of rollback in case it is needed.
-- storcon gets `-set-safekeepers` config option disabled by
-  default. Timeline creation request chooses safekeepers
-  (and returns them in response to cplane) only when it is set to
-  true.
-- control_plane [see above](storage_controller-<->-control-plane interface-and-changes)
-  prefixes `neon.safekeepers` GUC with generation number. When it is 0
-  (or prefix not present at all), walproposer behaves as currently, committing on
-  the provided safekeeper list -- generations are disabled.
-  If it is non 0 it follows this RFC rules.
-- We provide a script for manual migration to storage controller.
-  It selects timeline(s) from control plane (specified or all of them) db
-  and calls special import endpoint on storage controller which is very
-  similar to timeline creation: it inserts into the db, sets
-  configuration to initial on the safekeepers, calls cplane
-  `notify-safekeepers`.
+To rollout smoothly, both walproposer and safekeeper should have flag
+`configurations_enabled`; when set to false, they would work as currently, i.e.
+walproposer is able to commit on whatever safekeeper set it is provided. Until
+all timelines are managed by storcon we'd need to use current script to migrate
+and update/drop entries in the storage_controller database if it has any.
 
-Then the rollout for a region would be:
-- Current situation: safekeepers are choosen by control_plane.
-- We manually migrate some timelines, test moving them around.
-- Then we enable `--set-safekeepers` so that all new timelines
-  are on storage controller.
-- Finally migrate all existing timelines using the script (no
-  compute should be speaking old proto version at this point).
+Safekeepers would need to be able to talk both current and new protocol version
+with compute to reduce number of computes restarted in prod once v2 protocol is
+deployed (though before completely switching we'd need to force this).
+
+Let's have the following rollout order:
+- storage_controller becomes aware of safekeepers;
+- storage_controller gets timeline creation for new timelines and deletion requests, but
+  doesn't manage all timelines yet. Migration can be tested on these new timelines.
+  To keep control plane and storage_controller databases in sync while control 
+  plane still chooses the safekeepers initially (until all timelines are imported
+  it can choose better), `TimelineCreateRequest` can get optional safekeepers
+  field with safekeepers chosen by cplane.
+- Then we can import all existing timelines from control plane to
+  storage_controller and gradually enable configurations region by region.
 
-Until all timelines are managed by storcon we'd need to use current ad hoc
-script to migrate if needed. To keep state clean, all storage controller managed
-timelines must be migrated before that, or controller db and configurations
-state of safekeepers dropped manually.
 
 Very rough implementation order:
 - Add concept of configurations to safekeepers (including control file),
@@ -514,10 +458,10 @@ Very rough implementation order:
 - Implement walproposer changes, including protocol.
 - Implement storconn part. Use it in neon_local (and pytest).
 - Make cplane store safekeepers per timeline instead of per tenant.
-- Implement cplane/storcon integration. Route branch creation/deletion
+- Implement cplane/storcon integration. Route branch creation/deletion 
   through storcon. Then we can test migration of new branches.
-- Finally import existing branches. Then we can drop cplane
-  safekeeper selection code. Gradually enable configurations at
+- Finally import existing branches. Then we can drop cplane 
+  safekeeper selection code. Gradually enable configurations at 
   computes and safekeepers. Before that, all computes must talk only
   v3 protocol version.
 

docs/rfcs/040-profiling.md

@@ -1,247 +0,0 @@
-# CPU and Memory Profiling
-
-Created 2025-01-12 by Erik Grinaker.
-
-See also [internal user guide](https://www.notion.so/neondatabase/Storage-CPU-Memory-Profiling-14bf189e004780228ec7d04442742324?pvs=4).
-
-## Summary
-
-This document proposes a standard cross-team pattern for CPU and memory profiling across
-applications and languages, using the [pprof](https://github.com/google/pprof) profile format.
-
-It enables both ad hoc profiles via HTTP endpoints, and continuous profiling across the fleet via
-[Grafana Cloud Profiles](https://grafana.com/docs/grafana-cloud/monitor-applications/profiles/).
-Continuous profiling incurs an overhead of about 0.1% CPU usage and 3% slower heap allocations.
-
-## Motivation
-
-CPU and memory profiles are crucial observability tools for understanding performance issues,
-resource exhaustion, and resource costs. They allow answering questions like:
-
-* Why is this process using 100% CPU?
-* How do I make this go faster?
-* Why did this process run out of memory?
-* Why are we paying for all these CPU cores and memory chips?
-
-Go has [first-class support](https://pkg.go.dev/net/http/pprof) for profiling included in its
-standard library, using the [pprof](https://github.com/google/pprof) profile format and associated
-tooling.
-
-This is not the case for Rust and C, where obtaining profiles can be rather cumbersome. It requires
-installing and running additional tools like `perf` as root on production nodes, with analysis tools
-that can be hard to use and often don't give good results. This is not only annoying, but can also
-significantly affect the resolution time of production incidents.
-
-This proposal will:
-
-* Provide CPU and heap profiles in pprof format via HTTP API.
-* Record continuous profiles in Grafana for aggregate historical analysis.
-* Make it easy for anyone to see a flamegraph in less than one minute.
-* Be reasonably consistent across teams and services (Rust, Go, C).
-
-## Non Goals (For Now)
-
-* [Additional profile types](https://grafana.com/docs/pyroscope/next/configure-client/profile-types/)
-  like mutexes, locks, goroutines, etc.
-* [Runtime trace integration](https://grafana.com/docs/pyroscope/next/configure-client/trace-span-profiles/).
-* [Profile-guided optimization](https://en.wikipedia.org/wiki/Profile-guided_optimization).
-
-## Using Profiles
-
-Ready-to-use profiles can be obtained using e.g. `curl`. For Rust services:
-
-```
-$ curl localhost:9898/profile/cpu >profile.pb.gz
-```
-
-pprof profiles can be explored using the [`pprof`](https://github.com/google/pprof) web UI, which
-provides flamegraphs, call graphs, plain text listings, and more:
-
-```
-$ pprof -http :6060 <profile>
-```
-
-Some endpoints (e.g. Rust-based ones) can also generate flamegraph SVGs directly:
-
-```
-$ curl localhost:9898/profile/cpu?format=svg >profile.svg
-$ open profile.svg
-```
-
-Continuous profiles are available in Grafana under Explore → Profiles → Explore Profiles
-(currently only in [staging](https://neonstaging.grafana.net/a/grafana-pyroscope-app/profiles-explorer)).
-
-## API Requirements
-
-* HTTP endpoints that return a profile in pprof format (with symbols).
-  * CPU: records a profile over the request time interval (`seconds` query parameter).
-  * Memory: returns the current in-use heap allocations.
-* Unauthenticated, as it should not expose user data or pose a denial-of-service risk.
-* Default sample frequency should not impact service (maximum 5% CPU overhead).
-* Linux-compatibility.
-
-Nice to have:
-
-* Return flamegraph SVG directly from the HTTP endpoint if requested.
-* Configurable sample frequency for CPU profiles.
-* Historical heap allocations, by count and bytes.
-* macOS-compatiblity.
-
-## Rust Profiling
-
-[`libs/utils/src/http/endpoint.rs`](https://github.com/neondatabase/neon/blob/8327f68043e692c77f70d6a6dafa463636c01578/libs/utils/src/http/endpoint.rs)
-contains ready-to-use HTTP endpoints for CPU and memory profiling:
-[`profile_cpu_handler`](https://github.com/neondatabase/neon/blob/8327f68043e692c77f70d6a6dafa463636c01578/libs/utils/src/http/endpoint.rs#L338) and [`profile_heap_handler`](https://github.com/neondatabase/neon/blob/8327f68043e692c77f70d6a6dafa463636c01578/libs/utils/src/http/endpoint.rs#L416).
-
-### CPU
-
-CPU profiles are provided by [pprof-rs](https://github.com/tikv/pprof-rs) via
-[`profile_cpu_handler`](https://github.com/neondatabase/neon/blob/8327f68043e692c77f70d6a6dafa463636c01578/libs/utils/src/http/endpoint.rs#L338).
-Expose it unauthenticated at `/profile/cpu`.
-
-Parameters:
-
-* `format`: profile output format (`pprof` or `svg`; default `pprof`).
-* `seconds`: duration to collect profile over, in seconds (default `5`).
-* `frequency`: how often to sample thread stacks, in Hz (default `99`).
-* `force`: if `true`, cancel a running profile and start a new one (default `false`).
-
-Works on Linux and macOS.
-
-### Memory
-
-Use the jemalloc allocator via [`tikv-jemallocator`](https://github.com/tikv/jemallocator),
-and enable profiling with samples every 2 MB allocated:
-
-```rust
-#[global_allocator]
-static GLOBAL: tikv_jemallocator::Jemalloc = tikv_jemallocator::Jemalloc;
-
-#[allow(non_upper_case_globals)]
-#[export_name = "malloc_conf"]
-pub static malloc_conf: &[u8] = b"prof:true,prof_active:true,lg_prof_sample:21\0";
-```
-
-pprof profiles are generated by
-[`jemalloc-pprof`](https://github.com/polarsignals/rust-jemalloc-pprof) via
-[`profile_heap_handler`](https://github.com/neondatabase/neon/blob/8327f68043e692c77f70d6a6dafa463636c01578/libs/utils/src/http/endpoint.rs#L416).
-Expose it unauthenticated at `/profile/heap`.
-
-Parameters:
-
-* `format`: profile output format (`pprof`, `svg`, or `jemalloc`; default `pprof`).
-
-Works on Linux only, due to [jemalloc limitations](https://github.com/jemalloc/jemalloc/issues/26).
-
-## Go Profiling
-
-The Go standard library includes pprof profiling via HTTP API in
-[`net/http/pprof`](https://pkg.go.dev/net/http/pprof). Expose it unauthenticated at
-`/debug/pprof`.
-
-Works on Linux and macOS.
-
-### CPU 
-
-Via `/debug/pprof/profile`. Parameters:
-
-* `debug`: profile output format (`0` is pprof, `1` or above is plaintext; default `0`).
-* `seconds`: duration to collect profile over, in seconds (default `30`).
-
-Does not support a frequency parameter (see [#57488](https://github.com/golang/go/issues/57488)),
-and defaults to 100 Hz. A lower frequency can be hardcoded via `SetCPUProfileRate`, but the default
-is likely ok (estimated 1% overhead).
-
-### Memory
-
-Via `/debug/pprof/heap`. Parameters:
-
-* `seconds`: take a delta profile over the given duration, in seconds (default `0`).
-* `gc`: if `1`, garbage collect before taking profile.
-
-## C Profiling
-
-[gperftools](https://github.com/gperftools/gperftools) provides in-process CPU and heap profiling
-with pprof output.
-
-However, continuous profiling of PostgreSQL is expensive (many computes), and has limited value
-since we don't own the internals anyway.
-
-Ad hoc profiling might still be useful, but the compute team considers existing tooling sufficient,
-so this is not a priority at the moment.
-
-## Grafana Continuous Profiling
-
-[Grafana Alloy](https://grafana.com/docs/alloy/latest/) continually scrapes CPU and memory profiles
-across the fleet, and archives them as time series. This can be used to analyze resource usage over
-time, either in aggregate or zoomed in to specific events and nodes.
-
-Profiles are retained for 30 days. Profile ingestion volume for CPU+heap at 60-second intervals
-is about 0.5 GB/node/day, or about $0.25/node/day = $7.5/node/month ($0.50/GB).
-
-It is currently enabled in [staging](https://neonstaging.grafana.net/a/grafana-pyroscope-app/profiles-explorer)
-for Pageserver and Safekeeper.
-
-### Scraping
-
-* CPU profiling: 59 seconds at 19 Hz every 60 seconds.
-* Heap profiling: heap snapshot with 2 MB frequency every 60 seconds.
-
-There are two main approaches that can be taken for CPU profiles:
-
-* Continuous low-frequency profiles (e.g. 19 Hz for 60 seconds every 60 seconds).
-* Occasional high-frequency profiles (e.g. 99 Hz for 5 seconds every 60 seconds).
-
-We choose continuous low-frequency profiles where possible. This has a fixed low overhead, instead
-of a spiky high overhead. It likely also gives a more representative view of resource usage.
-However, a 19 Hz rate gives a minimum resolution of 52.6 ms per sample, which may be larger than the
-actual runtime of small functions. Note that Go does not support a frequency parameter, so we must
-use a fixed frequency for all profiles via `SetCPUProfileRate()` (default 100 Hz).
-
-Only one CPU profile can be taken at a time. With continuous profiling, one will always be running.
-To allow also taking an ad hoc CPU profile, the Rust endpoint supports a `force` query parameter to
-cancel a running profile and start a new one.
-
-### Overhead
-
-With Rust:
-
-* CPU profiles at 19 Hz frequency: 0.1% overhead.
-* Heap profiles at 2 MB frequency: 3% allocation overhead.
-* Profile call/encoding/symbolization: 20 ms every 60 seconds, or 0.03% of 1 CPU (for Pageserver).
-* Profile symbolization caches: 125 MB memory, or 0.4% of 32 GB (for Pageserver).
-
-Benchmarks with pprof-rs showed that the CPU time for taking a stack trace of a 40-frame stack was
-11 µs using the `frame-pointer` feature, and 1.4 µs using `libunwind` with DWARF. `libunwind` saw
-frequent seg faults, so we use `frame-pointer` and build binaries with frame pointers (negligible
-overhead).
-
-CPU profiles work by installing an `ITIMER_PROF` for the process, which triggers a `SIGPROF` signal
-after a given amount of cumulative CPU time across all CPUs. The signal handler will run for one
-of the currently executing threads and take a stack trace. Thus, a 19 Hz profile will take 1 stack
-trace every 52.6 ms CPU time -- assuming 11 µs for a stack trace, this is 0.02% overhead, but
-likely 0.1% in practice (given e.g. context switches).
-
-Heap profiles work by probabilistically taking a stack trace on allocations, adjusted for the
-allocation size. A 1 MB allocation takes about 15 µs in benchmarks, and a stack trace about 1 µs,
-so we can estimate that a 2 MB sampling frequency has about 3% allocation overhead -- this is 
-consistent with benchmarks. This is significantly larger than CPU profiles, but mitigated by the
-fact that performance-sensitive code will avoid allocations as far as possible.
-
-Profile symbolization uses in-memory caches for symbol lookups. These take about 125 MB for
-Pageserver.
-
-## Alternatives Considered
-
-* eBPF profiles.
-  * Don't require instrumenting the binary.
-  * Use less resources.
-  * Can profile in kernel space too.
-  * Supported by Grafana.
-  * Less information about stack frames and spans.
-  * Limited tooling for local analysis.
-  * Does not support heap profiles.
-  * Does not work on macOS.
-
-* [Polar Signals](https://www.polarsignals.com) instead of Grafana.
-  * We already use Grafana for everything else. Appears good enough.

libs/pageserver_api/src/controller_api.rs

@@ -87,7 +87,7 @@ impl Display for AvailabilityZone {
 #[derive(Serialize, Deserialize)]
 pub struct ShardsPreferredAzsRequest {
     #[serde(flatten)]
-    pub preferred_az_ids: HashMap<TenantShardId, Option<AvailabilityZone>>,
+    pub preferred_az_ids: HashMap<TenantShardId, AvailabilityZone>,
 }
 
 #[derive(Serialize, Deserialize)]
@@ -144,8 +144,6 @@ pub struct NodeDescribeResponse {
     pub availability: NodeAvailabilityWrapper,
     pub scheduling: NodeSchedulingPolicy,
 
-    pub availability_zone_id: String,
-
     pub listen_http_addr: String,
     pub listen_http_port: u16,
 
@@ -324,7 +322,7 @@ impl From<NodeSchedulingPolicy> for String {
 #[derive(Serialize, Deserialize, Clone, Copy, Eq, PartialEq, Debug)]
 pub enum SkSchedulingPolicy {
     Active,
-    Pause,
+    Disabled,
     Decomissioned,
 }
 
@@ -334,13 +332,9 @@ impl FromStr for SkSchedulingPolicy {
     fn from_str(s: &str) -> Result<Self, Self::Err> {
         Ok(match s {
             "active" => Self::Active,
-            "pause" => Self::Pause,
+            "disabled" => Self::Disabled,
             "decomissioned" => Self::Decomissioned,
-            _ => {
-                return Err(anyhow::anyhow!(
-                    "Unknown scheduling policy '{s}', try active,pause,decomissioned"
-                ))
-            }
+            _ => return Err(anyhow::anyhow!("Unknown scheduling state '{s}'")),
         })
     }
 }
@@ -350,7 +344,7 @@ impl From<SkSchedulingPolicy> for String {
         use SkSchedulingPolicy::*;
         match value {
             Active => "active",
-            Pause => "pause",
+            Disabled => "disabled",
             Decomissioned => "decomissioned",
         }
         .to_string()
@@ -373,16 +367,6 @@ pub enum PlacementPolicy {
     Detached,
 }
 
-impl PlacementPolicy {
-    pub fn want_secondaries(&self) -> usize {
-        match self {
-            PlacementPolicy::Attached(secondary_count) => *secondary_count,
-            PlacementPolicy::Secondary => 1,
-            PlacementPolicy::Detached => 0,
-        }
-    }
-}
-
 #[derive(Serialize, Deserialize, Debug)]
 pub struct TenantShardMigrateResponse {}
 
@@ -420,6 +404,8 @@ pub struct MetadataHealthListOutdatedResponse {
 }
 
 /// Publicly exposed safekeeper description
+///
+/// The `active` flag which we have in the DB is not included on purpose: it is deprecated.
 #[derive(Serialize, Deserialize, Clone)]
 pub struct SafekeeperDescribeResponse {
     pub id: NodeId,
@@ -435,11 +421,6 @@ pub struct SafekeeperDescribeResponse {
     pub scheduling_policy: SkSchedulingPolicy,
 }
 
-#[derive(Serialize, Deserialize, Clone)]
-pub struct SafekeeperSchedulingPolicyRequest {
-    pub scheduling_policy: SkSchedulingPolicy,
-}
-
 #[cfg(test)]
 mod test {
     use super::*;

libs/pageserver_api/src/key.rs

@@ -24,9 +24,7 @@ pub struct Key {
 
 /// When working with large numbers of Keys in-memory, it is more efficient to handle them as i128 than as
 /// a struct of fields.
-#[derive(
-    Clone, Copy, Default, Hash, PartialEq, Eq, Ord, PartialOrd, Serialize, Deserialize, Debug,
-)]
+#[derive(Clone, Copy, Hash, PartialEq, Eq, Ord, PartialOrd, Serialize, Deserialize, Debug)]
 pub struct CompactKey(i128);
 
 /// The storage key size.

libs/pageserver_api/src/models.rs

@@ -29,10 +29,11 @@ use utils::{
 };
 
 use crate::{
-    key::{CompactKey, Key},
+    key::Key,
     reltag::RelTag,
     shard::{ShardCount, ShardStripeSize, TenantShardId},
 };
+use anyhow::bail;
 use bytes::{Buf, BufMut, Bytes, BytesMut};
 
 /// The state of a tenant in this pageserver.
@@ -1399,8 +1400,6 @@ pub enum PagestreamFeMessage {
     GetPage(PagestreamGetPageRequest),
     DbSize(PagestreamDbSizeRequest),
     GetSlruSegment(PagestreamGetSlruSegmentRequest),
-    #[cfg(feature = "testing")]
-    Test(PagestreamTestRequest),
 }
 
 // Wrapped in libpq CopyData
@@ -1412,22 +1411,6 @@ pub enum PagestreamBeMessage {
     Error(PagestreamErrorResponse),
     DbSize(PagestreamDbSizeResponse),
     GetSlruSegment(PagestreamGetSlruSegmentResponse),
-    #[cfg(feature = "testing")]
-    Test(PagestreamTestResponse),
-}
-
-// Keep in sync with `pagestore_client.h`
-#[repr(u8)]
-enum PagestreamFeMessageTag {
-    Exists = 0,
-    Nblocks = 1,
-    GetPage = 2,
-    DbSize = 3,
-    GetSlruSegment = 4,
-    /* future tags above this line */
-    /// For testing purposes, not available in production.
-    #[cfg(feature = "testing")]
-    Test = 99,
 }
 
 // Keep in sync with `pagestore_client.h`
@@ -1439,28 +1422,7 @@ enum PagestreamBeMessageTag {
     Error = 103,
     DbSize = 104,
     GetSlruSegment = 105,
-    /* future tags above this line */
-    /// For testing purposes, not available in production.
-    #[cfg(feature = "testing")]
-    Test = 199,
 }
-
-impl TryFrom<u8> for PagestreamFeMessageTag {
-    type Error = u8;
-    fn try_from(value: u8) -> Result<Self, u8> {
-        match value {
-            0 => Ok(PagestreamFeMessageTag::Exists),
-            1 => Ok(PagestreamFeMessageTag::Nblocks),
-            2 => Ok(PagestreamFeMessageTag::GetPage),
-            3 => Ok(PagestreamFeMessageTag::DbSize),
-            4 => Ok(PagestreamFeMessageTag::GetSlruSegment),
-            #[cfg(feature = "testing")]
-            99 => Ok(PagestreamFeMessageTag::Test),
-            _ => Err(value),
-        }
-    }
-}
-
 impl TryFrom<u8> for PagestreamBeMessageTag {
     type Error = u8;
     fn try_from(value: u8) -> Result<Self, u8> {
@@ -1471,8 +1433,6 @@ impl TryFrom<u8> for PagestreamBeMessageTag {
             103 => Ok(PagestreamBeMessageTag::Error),
             104 => Ok(PagestreamBeMessageTag::DbSize),
             105 => Ok(PagestreamBeMessageTag::GetSlruSegment),
-            #[cfg(feature = "testing")]
-            199 => Ok(PagestreamBeMessageTag::Test),
             _ => Err(value),
         }
     }
@@ -1590,20 +1550,6 @@ pub struct PagestreamDbSizeResponse {
     pub db_size: i64,
 }
 
-#[cfg(feature = "testing")]
-#[derive(Debug, PartialEq, Eq, Clone)]
-pub struct PagestreamTestRequest {
-    pub hdr: PagestreamRequest,
-    pub batch_key: u64,
-    pub message: String,
-}
-
-#[cfg(feature = "testing")]
-#[derive(Debug)]
-pub struct PagestreamTestResponse {
-    pub req: PagestreamTestRequest,
-}
-
 // This is a cut-down version of TenantHistorySize from the pageserver crate, omitting fields
 // that require pageserver-internal types.  It is sufficient to get the total size.
 #[derive(Serialize, Deserialize, Debug)]
@@ -1623,7 +1569,7 @@ impl PagestreamFeMessage {
 
         match self {
             Self::Exists(req) => {
-                bytes.put_u8(PagestreamFeMessageTag::Exists as u8);
+                bytes.put_u8(0);
                 bytes.put_u64(req.hdr.reqid);
                 bytes.put_u64(req.hdr.request_lsn.0);
                 bytes.put_u64(req.hdr.not_modified_since.0);
@@ -1634,7 +1580,7 @@ impl PagestreamFeMessage {
             }
 
             Self::Nblocks(req) => {
-                bytes.put_u8(PagestreamFeMessageTag::Nblocks as u8);
+                bytes.put_u8(1);
                 bytes.put_u64(req.hdr.reqid);
                 bytes.put_u64(req.hdr.request_lsn.0);
                 bytes.put_u64(req.hdr.not_modified_since.0);
@@ -1645,7 +1591,7 @@ impl PagestreamFeMessage {
             }
 
             Self::GetPage(req) => {
-                bytes.put_u8(PagestreamFeMessageTag::GetPage as u8);
+                bytes.put_u8(2);
                 bytes.put_u64(req.hdr.reqid);
                 bytes.put_u64(req.hdr.request_lsn.0);
                 bytes.put_u64(req.hdr.not_modified_since.0);
@@ -1657,7 +1603,7 @@ impl PagestreamFeMessage {
             }
 
             Self::DbSize(req) => {
-                bytes.put_u8(PagestreamFeMessageTag::DbSize as u8);
+                bytes.put_u8(3);
                 bytes.put_u64(req.hdr.reqid);
                 bytes.put_u64(req.hdr.request_lsn.0);
                 bytes.put_u64(req.hdr.not_modified_since.0);
@@ -1665,24 +1611,13 @@ impl PagestreamFeMessage {
             }
 
             Self::GetSlruSegment(req) => {
-                bytes.put_u8(PagestreamFeMessageTag::GetSlruSegment as u8);
+                bytes.put_u8(4);
                 bytes.put_u64(req.hdr.reqid);
                 bytes.put_u64(req.hdr.request_lsn.0);
                 bytes.put_u64(req.hdr.not_modified_since.0);
                 bytes.put_u8(req.kind);
                 bytes.put_u32(req.segno);
             }
-            #[cfg(feature = "testing")]
-            Self::Test(req) => {
-                bytes.put_u8(PagestreamFeMessageTag::Test as u8);
-                bytes.put_u64(req.hdr.reqid);
-                bytes.put_u64(req.hdr.request_lsn.0);
-                bytes.put_u64(req.hdr.not_modified_since.0);
-                bytes.put_u64(req.batch_key);
-                let message = req.message.as_bytes();
-                bytes.put_u64(message.len() as u64);
-                bytes.put_slice(message);
-            }
         }
 
         bytes.into()
@@ -1710,66 +1645,56 @@ impl PagestreamFeMessage {
             ),
         };
 
-        match PagestreamFeMessageTag::try_from(msg_tag)
-            .map_err(|tag: u8| anyhow::anyhow!("invalid tag {tag}"))?
-        {
-            PagestreamFeMessageTag::Exists => {
-                Ok(PagestreamFeMessage::Exists(PagestreamExistsRequest {
-                    hdr: PagestreamRequest {
-                        reqid,
-                        request_lsn,
-                        not_modified_since,
-                    },
-                    rel: RelTag {
-                        spcnode: body.read_u32::<BigEndian>()?,
-                        dbnode: body.read_u32::<BigEndian>()?,
-                        relnode: body.read_u32::<BigEndian>()?,
-                        forknum: body.read_u8()?,
-                    },
-                }))
-            }
-            PagestreamFeMessageTag::Nblocks => {
-                Ok(PagestreamFeMessage::Nblocks(PagestreamNblocksRequest {
-                    hdr: PagestreamRequest {
-                        reqid,
-                        request_lsn,
-                        not_modified_since,
-                    },
-                    rel: RelTag {
-                        spcnode: body.read_u32::<BigEndian>()?,
-                        dbnode: body.read_u32::<BigEndian>()?,
-                        relnode: body.read_u32::<BigEndian>()?,
-                        forknum: body.read_u8()?,
-                    },
-                }))
-            }
-            PagestreamFeMessageTag::GetPage => {
-                Ok(PagestreamFeMessage::GetPage(PagestreamGetPageRequest {
-                    hdr: PagestreamRequest {
-                        reqid,
-                        request_lsn,
-                        not_modified_since,
-                    },
-                    rel: RelTag {
-                        spcnode: body.read_u32::<BigEndian>()?,
-                        dbnode: body.read_u32::<BigEndian>()?,
-                        relnode: body.read_u32::<BigEndian>()?,
-                        forknum: body.read_u8()?,
-                    },
-                    blkno: body.read_u32::<BigEndian>()?,
-                }))
-            }
-            PagestreamFeMessageTag::DbSize => {
-                Ok(PagestreamFeMessage::DbSize(PagestreamDbSizeRequest {
-                    hdr: PagestreamRequest {
-                        reqid,
-                        request_lsn,
-                        not_modified_since,
-                    },
+        match msg_tag {
+            0 => Ok(PagestreamFeMessage::Exists(PagestreamExistsRequest {
+                hdr: PagestreamRequest {
+                    reqid,
+                    request_lsn,
+                    not_modified_since,
+                },
+                rel: RelTag {
+                    spcnode: body.read_u32::<BigEndian>()?,
                     dbnode: body.read_u32::<BigEndian>()?,
-                }))
-            }
-            PagestreamFeMessageTag::GetSlruSegment => Ok(PagestreamFeMessage::GetSlruSegment(
+                    relnode: body.read_u32::<BigEndian>()?,
+                    forknum: body.read_u8()?,
+                },
+            })),
+            1 => Ok(PagestreamFeMessage::Nblocks(PagestreamNblocksRequest {
+                hdr: PagestreamRequest {
+                    reqid,
+                    request_lsn,
+                    not_modified_since,
+                },
+                rel: RelTag {
+                    spcnode: body.read_u32::<BigEndian>()?,
+                    dbnode: body.read_u32::<BigEndian>()?,
+                    relnode: body.read_u32::<BigEndian>()?,
+                    forknum: body.read_u8()?,
+                },
+            })),
+            2 => Ok(PagestreamFeMessage::GetPage(PagestreamGetPageRequest {
+                hdr: PagestreamRequest {
+                    reqid,
+                    request_lsn,
+                    not_modified_since,
+                },
+                rel: RelTag {
+                    spcnode: body.read_u32::<BigEndian>()?,
+                    dbnode: body.read_u32::<BigEndian>()?,
+                    relnode: body.read_u32::<BigEndian>()?,
+                    forknum: body.read_u8()?,
+                },
+                blkno: body.read_u32::<BigEndian>()?,
+            })),
+            3 => Ok(PagestreamFeMessage::DbSize(PagestreamDbSizeRequest {
+                hdr: PagestreamRequest {
+                    reqid,
+                    request_lsn,
+                    not_modified_since,
+                },
+                dbnode: body.read_u32::<BigEndian>()?,
+            })),
+            4 => Ok(PagestreamFeMessage::GetSlruSegment(
                 PagestreamGetSlruSegmentRequest {
                     hdr: PagestreamRequest {
                         reqid,
@@ -1780,21 +1705,7 @@ impl PagestreamFeMessage {
                     segno: body.read_u32::<BigEndian>()?,
                 },
             )),
-            #[cfg(feature = "testing")]
-            PagestreamFeMessageTag::Test => Ok(PagestreamFeMessage::Test(PagestreamTestRequest {
-                hdr: PagestreamRequest {
-                    reqid,
-                    request_lsn,
-                    not_modified_since,
-                },
-                batch_key: body.read_u64::<BigEndian>()?,
-                message: {
-                    let len = body.read_u64::<BigEndian>()?;
-                    let mut buf = vec![0; len as usize];
-                    body.read_exact(&mut buf)?;
-                    String::from_utf8(buf)?
-                },
-            })),
+            _ => bail!("unknown smgr message tag: {:?}", msg_tag),
         }
     }
 }
@@ -1837,15 +1748,6 @@ impl PagestreamBeMessage {
                         bytes.put_u32((resp.segment.len() / BLCKSZ as usize) as u32);
                         bytes.put(&resp.segment[..]);
                     }
-
-                    #[cfg(feature = "testing")]
-                    Self::Test(resp) => {
-                        bytes.put_u8(Tag::Test as u8);
-                        bytes.put_u64(resp.req.batch_key);
-                        let message = resp.req.message.as_bytes();
-                        bytes.put_u64(message.len() as u64);
-                        bytes.put_slice(message);
-                    }
                 }
             }
             PagestreamProtocolVersion::V3 => {
@@ -1914,18 +1816,6 @@ impl PagestreamBeMessage {
                         bytes.put_u32((resp.segment.len() / BLCKSZ as usize) as u32);
                         bytes.put(&resp.segment[..]);
                     }
-
-                    #[cfg(feature = "testing")]
-                    Self::Test(resp) => {
-                        bytes.put_u8(Tag::Test as u8);
-                        bytes.put_u64(resp.req.hdr.reqid);
-                        bytes.put_u64(resp.req.hdr.request_lsn.0);
-                        bytes.put_u64(resp.req.hdr.not_modified_since.0);
-                        bytes.put_u64(resp.req.batch_key);
-                        let message = resp.req.message.as_bytes();
-                        bytes.put_u64(message.len() as u64);
-                        bytes.put_slice(message);
-                    }
                 }
             }
         }
@@ -2068,28 +1958,6 @@ impl PagestreamBeMessage {
                         segment: segment.into(),
                     })
                 }
-                #[cfg(feature = "testing")]
-                Tag::Test => {
-                    let reqid = buf.read_u64::<BigEndian>()?;
-                    let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
-                    let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
-                    let batch_key = buf.read_u64::<BigEndian>()?;
-                    let len = buf.read_u64::<BigEndian>()?;
-                    let mut msg = vec![0; len as usize];
-                    buf.read_exact(&mut msg)?;
-                    let message = String::from_utf8(msg)?;
-                    Self::Test(PagestreamTestResponse {
-                        req: PagestreamTestRequest {
-                            hdr: PagestreamRequest {
-                                reqid,
-                                request_lsn,
-                                not_modified_since,
-                            },
-                            batch_key,
-                            message,
-                        },
-                    })
-                }
             };
         let remaining = buf.into_inner();
         if !remaining.is_empty() {
@@ -2109,25 +1977,6 @@ impl PagestreamBeMessage {
             Self::Error(_) => "Error",
             Self::DbSize(_) => "DbSize",
             Self::GetSlruSegment(_) => "GetSlruSegment",
-            #[cfg(feature = "testing")]
-            Self::Test(_) => "Test",
-        }
-    }
-}
-
-#[derive(Debug, Serialize, Deserialize)]
-pub struct PageTraceEvent {
-    pub key: CompactKey,
-    pub effective_lsn: Lsn,
-    pub time: SystemTime,
-}
-
-impl Default for PageTraceEvent {
-    fn default() -> Self {
-        Self {
-            key: Default::default(),
-            effective_lsn: Default::default(),
-            time: std::time::UNIX_EPOCH,
         }
     }
 }

libs/pageserver_api/src/shard.rs

@@ -31,8 +31,6 @@
 //! - In a tenant with 4 shards, each shard has ShardCount(N), ShardNumber(i) where i in 0..N-1 (inclusive),
 //!   and their slugs are 0004, 0104, 0204, and 0304.
 
-use std::hash::{Hash, Hasher};
-
 use crate::{key::Key, models::ShardParameters};
 use postgres_ffi::relfile_utils::INIT_FORKNUM;
 use serde::{Deserialize, Serialize};
@@ -50,23 +48,6 @@ pub struct ShardIdentity {
     layout: ShardLayout,
 }
 
-/// Hash implementation
-///
-/// The stripe size cannot change dynamically, so it can be ignored for efficiency reasons.
-impl Hash for ShardIdentity {
-    fn hash<H: Hasher>(&self, state: &mut H) {
-        let ShardIdentity {
-            number,
-            count,
-            stripe_size: _,
-            layout: _,
-        } = self;
-
-        number.0.hash(state);
-        count.0.hash(state);
-    }
-}
-
 /// Stripe size in number of pages
 #[derive(Clone, Copy, Serialize, Deserialize, Eq, PartialEq, Debug)]
 pub struct ShardStripeSize(pub u32);
@@ -78,7 +59,7 @@ impl Default for ShardStripeSize {
 }
 
 /// Layout version: for future upgrades where we might change how the key->shard mapping works
-#[derive(Clone, Copy, Serialize, Deserialize, Eq, PartialEq, Hash, Debug)]
+#[derive(Clone, Copy, Serialize, Deserialize, Eq, PartialEq, Debug)]
 pub struct ShardLayout(u8);
 
 const LAYOUT_V1: ShardLayout = ShardLayout(1);

libs/postgres_ffi/src/walrecord.rs

@@ -16,7 +16,7 @@ use utils::bin_ser::DeserializeError;
 use utils::lsn::Lsn;
 
 #[repr(C)]
-#[derive(Clone, Debug, Serialize, Deserialize)]
+#[derive(Debug, Serialize, Deserialize)]
 pub struct XlMultiXactCreate {
     pub mid: MultiXactId,
     /* new MultiXact's ID */
@@ -46,7 +46,7 @@ impl XlMultiXactCreate {
 }
 
 #[repr(C)]
-#[derive(Clone, Debug, Serialize, Deserialize)]
+#[derive(Debug, Serialize, Deserialize)]
 pub struct XlMultiXactTruncate {
     pub oldest_multi_db: Oid,
     /* to-be-truncated range of multixact offsets */
@@ -72,7 +72,7 @@ impl XlMultiXactTruncate {
 }
 
 #[repr(C)]
-#[derive(Clone, Debug, Serialize, Deserialize)]
+#[derive(Debug, Serialize, Deserialize)]
 pub struct XlRelmapUpdate {
     pub dbid: Oid,   /* database ID, or 0 for shared map */
     pub tsid: Oid,   /* database's tablespace, or pg_global */
@@ -90,7 +90,7 @@ impl XlRelmapUpdate {
 }
 
 #[repr(C)]
-#[derive(Clone, Debug, Serialize, Deserialize)]
+#[derive(Debug, Serialize, Deserialize)]
 pub struct XlReploriginDrop {
     pub node_id: RepOriginId,
 }
@@ -104,7 +104,7 @@ impl XlReploriginDrop {
 }
 
 #[repr(C)]
-#[derive(Clone, Debug, Serialize, Deserialize)]
+#[derive(Debug, Serialize, Deserialize)]
 pub struct XlReploriginSet {
     pub remote_lsn: Lsn,
     pub node_id: RepOriginId,
@@ -911,7 +911,7 @@ impl XlSmgrCreate {
 }
 
 #[repr(C)]
-#[derive(Clone, Debug, Serialize, Deserialize)]
+#[derive(Debug, Serialize, Deserialize)]
 pub struct XlSmgrTruncate {
     pub blkno: BlockNumber,
     pub rnode: RelFileNode,
@@ -984,7 +984,7 @@ impl XlDropDatabase {
 /// xl_xact_parsed_abort structs in PostgreSQL, but we use the same
 /// struct for commits and aborts.
 ///
-#[derive(Clone, Debug, Serialize, Deserialize)]
+#[derive(Debug, Serialize, Deserialize)]
 pub struct XlXactParsedRecord {
     pub xid: TransactionId,
     pub info: u8,

libs/remote_storage/src/config.rs

@@ -43,17 +43,6 @@ impl RemoteStorageKind {
     }
 }
 
-impl RemoteStorageConfig {
-    /// Helper to fetch the configured concurrency limit.
-    pub fn concurrency_limit(&self) -> Option<usize> {
-        match &self.storage {
-            RemoteStorageKind::LocalFs { .. } => None,
-            RemoteStorageKind::AwsS3(c) => Some(c.concurrency_limit.into()),
-            RemoteStorageKind::AzureContainer(c) => Some(c.concurrency_limit.into()),
-        }
-    }
-}
-
 fn default_timeout() -> Duration {
     RemoteStorageConfig::DEFAULT_TIMEOUT
 }

libs/safekeeper_api/Cargo.toml

@@ -5,12 +5,9 @@ edition.workspace = true
 license.workspace = true
 
 [dependencies]
-anyhow.workspace = true
 const_format.workspace = true
 serde.workspace = true
-serde_json.workspace = true
 postgres_ffi.workspace = true
 pq_proto.workspace = true
 tokio.workspace = true
 utils.workspace = true
-pageserver_api.workspace = true

libs/safekeeper_api/src/lib.rs

@@ -4,15 +4,12 @@ use const_format::formatcp;
 use pq_proto::SystemId;
 use serde::{Deserialize, Serialize};
 
-pub mod membership;
 /// Public API types
 pub mod models;
 
 /// Consensus logical timestamp. Note: it is a part of sk control file.
 pub type Term = u64;
-/// With this term timeline is created initially. It
-/// is a normal term except wp is never elected with it.
-pub const INITIAL_TERM: Term = 0;
+pub const INVALID_TERM: Term = 0;
 
 /// Information about Postgres. Safekeeper gets it once and then verifies all
 /// further connections from computes match. Note: it is a part of sk control

libs/safekeeper_api/src/membership.rs

@@ -1,166 +0,0 @@
-//! Types defining safekeeper membership, see
-//! rfcs/035-safekeeper-dynamic-membership-change.md
-//! for details.
-
-use std::{collections::HashSet, fmt::Display};
-
-use anyhow;
-use anyhow::bail;
-use serde::{Deserialize, Serialize};
-use utils::id::NodeId;
-
-/// Number uniquely identifying safekeeper configuration.
-/// Note: it is a part of sk control file.
-pub type Generation = u32;
-/// 1 is the first valid generation, 0 is used as
-/// a placeholder before we fully migrate to generations.
-pub const INVALID_GENERATION: Generation = 0;
-pub const INITIAL_GENERATION: Generation = 1;
-
-/// Membership is defined by ids so e.g. walproposer uses them to figure out
-/// quorums, but we also carry host and port to give wp idea where to connect.
-#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
-pub struct SafekeeperId {
-    pub id: NodeId,
-    pub host: String,
-    /// We include here only port for computes -- that is, pg protocol tenant
-    /// only port, or wide pg protocol port if the former is not configured.
-    pub pg_port: u16,
-}
-
-impl Display for SafekeeperId {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        write!(f, "[id={}, ep={}:{}]", self.id, self.host, self.pg_port)
-    }
-}
-
-/// Set of safekeepers.
-#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
-#[serde(transparent)]
-pub struct MemberSet {
-    pub members: Vec<SafekeeperId>,
-}
-
-impl MemberSet {
-    pub fn empty() -> Self {
-        MemberSet {
-            members: Vec::new(),
-        }
-    }
-
-    pub fn new(members: Vec<SafekeeperId>) -> anyhow::Result<Self> {
-        let hs: HashSet<NodeId> = HashSet::from_iter(members.iter().map(|sk| sk.id));
-        if hs.len() != members.len() {
-            bail!("duplicate safekeeper id in the set {:?}", members);
-        }
-        Ok(MemberSet { members })
-    }
-
-    pub fn contains(&self, sk: &SafekeeperId) -> bool {
-        self.members.iter().any(|m| m.id == sk.id)
-    }
-
-    pub fn add(&mut self, sk: SafekeeperId) -> anyhow::Result<()> {
-        if self.contains(&sk) {
-            bail!(format!(
-                "sk {} is already member of the set {}",
-                sk.id, self
-            ));
-        }
-        self.members.push(sk);
-        Ok(())
-    }
-}
-
-impl Display for MemberSet {
-    /// Display as a comma separated list of members.
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        let sks_str = self
-            .members
-            .iter()
-            .map(|m| m.to_string())
-            .collect::<Vec<_>>();
-        write!(f, "({})", sks_str.join(", "))
-    }
-}
-
-/// Safekeeper membership configuration.
-/// Note: it is a part of both control file and http API.
-#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
-pub struct Configuration {
-    /// Unique id.
-    pub generation: Generation,
-    /// Current members of the configuration.
-    pub members: MemberSet,
-    /// Some means it is a joint conf.
-    pub new_members: Option<MemberSet>,
-}
-
-impl Configuration {
-    /// Used for pre-generations timelines, will be removed eventually.
-    pub fn empty() -> Self {
-        Configuration {
-            generation: INVALID_GENERATION,
-            members: MemberSet::empty(),
-            new_members: None,
-        }
-    }
-}
-
-impl Display for Configuration {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        write!(
-            f,
-            "gen={}, members={}, new_members={}",
-            self.generation,
-            self.members,
-            self.new_members
-                .as_ref()
-                .map(ToString::to_string)
-                .unwrap_or(String::from("none"))
-        )
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::{MemberSet, SafekeeperId};
-    use utils::id::NodeId;
-
-    #[test]
-    fn test_member_set() {
-        let mut members = MemberSet::empty();
-        members
-            .add(SafekeeperId {
-                id: NodeId(42),
-                host: String::from("lala.org"),
-                pg_port: 5432,
-            })
-            .unwrap();
-
-        members
-            .add(SafekeeperId {
-                id: NodeId(42),
-                host: String::from("lala.org"),
-                pg_port: 5432,
-            })
-            .expect_err("duplicate must not be allowed");
-
-        members
-            .add(SafekeeperId {
-                id: NodeId(43),
-                host: String::from("bubu.org"),
-                pg_port: 5432,
-            })
-            .unwrap();
-
-        println!("members: {}", members);
-
-        let j = serde_json::to_string(&members).expect("failed to serialize");
-        println!("members json: {}", j);
-        assert_eq!(
-            j,
-            r#"[{"id":42,"host":"lala.org","pg_port":5432},{"id":43,"host":"bubu.org","pg_port":5432}]"#
-        );
-    }
-}

libs/safekeeper_api/src/models.rs

@@ -1,6 +1,5 @@
 //! Types used in safekeeper http API. Many of them are also reused internally.
 
-use pageserver_api::shard::ShardIdentity;
 use postgres_ffi::TimestampTz;
 use serde::{Deserialize, Serialize};
 use std::net::SocketAddr;
@@ -12,7 +11,7 @@ use utils::{
     pageserver_feedback::PageserverFeedback,
 };
 
-use crate::{membership::Configuration, ServerInfo, Term};
+use crate::{ServerInfo, Term};
 
 #[derive(Debug, Serialize)]
 pub struct SafekeeperStatus {
@@ -23,16 +22,13 @@ pub struct SafekeeperStatus {
 pub struct TimelineCreateRequest {
     pub tenant_id: TenantId,
     pub timeline_id: TimelineId,
-    pub mconf: Configuration,
+    pub peer_ids: Option<Vec<NodeId>>,
     pub pg_version: u32,
     pub system_id: Option<u64>,
-    // By default WAL_SEGMENT_SIZE
     pub wal_seg_size: Option<u32>,
-    pub start_lsn: Lsn,
-    // Normal creation should omit this field (start_lsn initializes all LSNs).
-    // However, we allow specifying custom value higher than start_lsn for
-    // manual recovery case, see test_s3_wal_replay.
-    pub commit_lsn: Option<Lsn>,
+    pub commit_lsn: Lsn,
+    // If not passed, it is assigned to the beginning of commit_lsn segment.
+    pub local_start_lsn: Option<Lsn>,
 }
 
 /// Same as TermLsn, but serializes LSN using display serializer
@@ -147,13 +143,7 @@ pub type ConnectionId = u32;
 
 /// Serialize is used only for json'ing in API response. Also used internally.
 #[derive(Debug, Clone, Serialize, Deserialize)]
-pub enum WalSenderState {
-    Vanilla(VanillaWalSenderState),
-    Interpreted(InterpretedWalSenderState),
-}
-
-#[derive(Debug, Clone, Serialize, Deserialize)]
-pub struct VanillaWalSenderState {
+pub struct WalSenderState {
     pub ttid: TenantTimelineId,
     pub addr: SocketAddr,
     pub conn_id: ConnectionId,
@@ -162,17 +152,6 @@ pub struct VanillaWalSenderState {
     pub feedback: ReplicationFeedback,
 }
 
-#[derive(Debug, Clone, Serialize, Deserialize)]
-pub struct InterpretedWalSenderState {
-    pub ttid: TenantTimelineId,
-    pub shard: ShardIdentity,
-    pub addr: SocketAddr,
-    pub conn_id: ConnectionId,
-    // postgres application_name
-    pub appname: Option<String>,
-    pub feedback: ReplicationFeedback,
-}
-
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct WalReceiverState {
     /// None means it is recovery initiated by us (this safekeeper).
@@ -193,7 +172,6 @@ pub enum WalReceiverStatus {
 pub struct TimelineStatus {
     pub tenant_id: TenantId,
     pub timeline_id: TimelineId,
-    pub mconf: Configuration,
     pub acceptor_state: AcceptorStateStatus,
     pub pg_info: ServerInfo,
     pub flush_lsn: Lsn,
@@ -208,20 +186,6 @@ pub struct TimelineStatus {
     pub walreceivers: Vec<WalReceiverState>,
 }
 
-/// Request to switch membership configuration.
-#[derive(Serialize, Deserialize)]
-#[serde(transparent)]
-pub struct TimelineMembershipSwitchRequest {
-    pub mconf: Configuration,
-}
-
-/// In response both previous and current configuration are sent.
-#[derive(Serialize, Deserialize)]
-pub struct TimelineMembershipSwitchResponse {
-    pub previous_conf: Configuration,
-    pub current_conf: Configuration,
-}
-
 fn lsn_invalid() -> Lsn {
     Lsn::INVALID
 }

libs/utils/src/guard_arc_swap.rs

@@ -1,54 +0,0 @@
-//! A wrapper around `ArcSwap` that ensures there is only one writer at a time and writes
-//! don't block reads.
-
-use arc_swap::ArcSwap;
-use std::sync::Arc;
-use tokio::sync::TryLockError;
-
-pub struct GuardArcSwap<T> {
-    inner: ArcSwap<T>,
-    guard: tokio::sync::Mutex<()>,
-}
-
-pub struct Guard<'a, T> {
-    _guard: tokio::sync::MutexGuard<'a, ()>,
-    inner: &'a ArcSwap<T>,
-}
-
-impl<T> GuardArcSwap<T> {
-    pub fn new(inner: T) -> Self {
-        Self {
-            inner: ArcSwap::new(Arc::new(inner)),
-            guard: tokio::sync::Mutex::new(()),
-        }
-    }
-
-    pub fn read(&self) -> Arc<T> {
-        self.inner.load_full()
-    }
-
-    pub async fn write_guard(&self) -> Guard<'_, T> {
-        Guard {
-            _guard: self.guard.lock().await,
-            inner: &self.inner,
-        }
-    }
-
-    pub fn try_write_guard(&self) -> Result<Guard<'_, T>, TryLockError> {
-        let guard = self.guard.try_lock()?;
-        Ok(Guard {
-            _guard: guard,
-            inner: &self.inner,
-        })
-    }
-}
-
-impl<T> Guard<'_, T> {
-    pub fn read(&self) -> Arc<T> {
-        self.inner.load_full()
-    }
-
-    pub fn write(&mut self, value: T) {
-        self.inner.store(Arc::new(value));
-    }
-}

libs/utils/src/lib.rs

@@ -98,8 +98,6 @@ pub mod try_rcu;
 
 pub mod pprof;
 
-pub mod guard_arc_swap;
-
 // Re-export used in macro. Avoids adding git-version as dep in target crates.
 #[doc(hidden)]
 pub use git_version;

libs/wal_decoder/Cargo.toml

@@ -24,18 +24,3 @@ workspace_hack = { version = "0.1", path = "../../workspace_hack" }
 
 [build-dependencies]
 tonic-build.workspace = true
-
-[dev-dependencies]
-criterion.workspace = true
-camino.workspace = true
-camino-tempfile.workspace = true
-remote_storage.workspace = true
-tokio-util.workspace = true
-serde_json.workspace = true
-futures.workspace = true
-tikv-jemallocator.workspace = true
-pprof.workspace = true
-
-[[bench]]
-name = "bench_interpret_wal"
-harness = false

libs/wal_decoder/benches/README.md

@@ -1,34 +0,0 @@
-## WAL Decoding and Interpretation Benchmarks
-
-Note that these benchmarks pull WAL from a public bucket in S3
-as a preparation step. Hence, you need a way to auth with AWS.
-You can achieve this by copying the `~/.aws/config` file from
-the AWS SSO notion page and exporting `AWS_PROFILE=dev` when invoking
-the benchmarks.
-
-To run benchmarks:
-
-```sh
-aws sso login --profile dev
-
-# All benchmarks.
-AWS_PROFILE=dev cargo bench --package wal_decoder
-
-# Specific file.
-AWS_PROFILE=dev cargo bench --package wal_decoder --bench bench_interpret_wal
-
-# Specific benchmark.
-AWS_PROFILE=dev cargo bench --package wal_decoder --bench bench_interpret_wal unsharded
-
-# List available benchmarks.
-cargo bench --package wal_decoder --benches -- --list
-
-# Generate flamegraph profiles using pprof-rs, profiling for 10 seconds.
-# Output in target/criterion/*/profile/flamegraph.svg.
-AWS_PROFILE=dev cargo bench --package wal_decoder --bench bench_interpret_wal unsharded -- --profile-time 10
-```
-
-Additional charts and statistics are available in `target/criterion/report/index.html`.
-
-Benchmarks are automatically compared against the previous run. To compare against other runs, see
-`--baseline` and `--save-baseline`.

libs/wal_decoder/benches/bench_interpret_wal.rs

@@ -1,250 +0,0 @@
-use anyhow::Context;
-use criterion::{criterion_group, criterion_main, Criterion};
-use futures::{stream::FuturesUnordered, StreamExt};
-use pageserver_api::shard::{ShardIdentity, ShardStripeSize};
-use postgres_ffi::{waldecoder::WalStreamDecoder, MAX_SEND_SIZE, WAL_SEGMENT_SIZE};
-use pprof::criterion::{Output, PProfProfiler};
-use serde::Deserialize;
-use std::{env, num::NonZeroUsize, sync::Arc};
-
-use camino::{Utf8Path, Utf8PathBuf};
-use camino_tempfile::Utf8TempDir;
-use remote_storage::{
-    DownloadOpts, GenericRemoteStorage, ListingMode, RemoteStorageConfig, RemoteStorageKind,
-    S3Config,
-};
-use tokio_util::sync::CancellationToken;
-use utils::{
-    lsn::Lsn,
-    shard::{ShardCount, ShardNumber},
-};
-use wal_decoder::models::InterpretedWalRecord;
-
-const S3_BUCKET: &str = "neon-github-public-dev";
-const S3_REGION: &str = "eu-central-1";
-const BUCKET_PREFIX: &str = "wal-snapshots/bulk-insert/";
-const METADATA_FILENAME: &str = "metadata.json";
-
-/// Use jemalloc, and configure it to sample allocations for profiles every 1 MB.
-/// This mirrors the configuration in bin/safekeeper.rs.
-#[global_allocator]
-static GLOBAL: tikv_jemallocator::Jemalloc = tikv_jemallocator::Jemalloc;
-
-#[allow(non_upper_case_globals)]
-#[export_name = "malloc_conf"]
-pub static malloc_conf: &[u8] = b"prof:true,prof_active:true,lg_prof_sample:20\0";
-
-async fn create_s3_client() -> anyhow::Result<Arc<GenericRemoteStorage>> {
-    let remote_storage_config = RemoteStorageConfig {
-        storage: RemoteStorageKind::AwsS3(S3Config {
-            bucket_name: S3_BUCKET.to_string(),
-            bucket_region: S3_REGION.to_string(),
-            prefix_in_bucket: Some(BUCKET_PREFIX.to_string()),
-            endpoint: None,
-            concurrency_limit: NonZeroUsize::new(100).unwrap(),
-            max_keys_per_list_response: None,
-            upload_storage_class: None,
-        }),
-        timeout: RemoteStorageConfig::DEFAULT_TIMEOUT,
-        small_timeout: RemoteStorageConfig::DEFAULT_SMALL_TIMEOUT,
-    };
-    Ok(Arc::new(
-        GenericRemoteStorage::from_config(&remote_storage_config)
-            .await
-            .context("remote storage init")?,
-    ))
-}
-
-async fn download_bench_data(
-    client: Arc<GenericRemoteStorage>,
-    cancel: &CancellationToken,
-) -> anyhow::Result<Utf8TempDir> {
-    let temp_dir_parent: Utf8PathBuf = env::current_dir().unwrap().try_into()?;
-    let temp_dir = camino_tempfile::tempdir_in(temp_dir_parent)?;
-
-    eprintln!("Downloading benchmark data to {:?}", temp_dir);
-
-    let listing = client
-        .list(None, ListingMode::NoDelimiter, None, cancel)
-        .await?;
-
-    let mut downloads = listing
-        .keys
-        .into_iter()
-        .map(|obj| {
-            let client = client.clone();
-            let temp_dir_path = temp_dir.path().to_owned();
-
-            async move {
-                let remote_path = obj.key;
-                let download = client
-                    .download(&remote_path, &DownloadOpts::default(), cancel)
-                    .await?;
-                let mut body = tokio_util::io::StreamReader::new(download.download_stream);
-
-                let file_name = remote_path.object_name().unwrap();
-                let file_path = temp_dir_path.join(file_name);
-                let file = tokio::fs::OpenOptions::new()
-                    .create(true)
-                    .truncate(true)
-                    .write(true)
-                    .open(&file_path)
-                    .await?;
-
-                let mut writer = tokio::io::BufWriter::new(file);
-                tokio::io::copy_buf(&mut body, &mut writer).await?;
-
-                Ok::<(), anyhow::Error>(())
-            }
-        })
-        .collect::<FuturesUnordered<_>>();
-
-    while let Some(download) = downloads.next().await {
-        download?;
-    }
-
-    Ok(temp_dir)
-}
-
-struct BenchmarkData {
-    wal: Vec<u8>,
-    meta: BenchmarkMetadata,
-}
-
-#[derive(Deserialize)]
-struct BenchmarkMetadata {
-    pg_version: u32,
-    start_lsn: Lsn,
-}
-
-async fn load_bench_data(path: &Utf8Path, input_size: usize) -> anyhow::Result<BenchmarkData> {
-    eprintln!("Loading benchmark data from {:?}", path);
-
-    let mut entries = tokio::fs::read_dir(path).await?;
-    let mut ordered_segment_paths = Vec::new();
-    let mut metadata = None;
-
-    while let Some(entry) = entries.next_entry().await? {
-        if entry.file_name() == METADATA_FILENAME {
-            let bytes = tokio::fs::read(entry.path()).await?;
-            metadata = Some(
-                serde_json::from_slice::<BenchmarkMetadata>(&bytes)
-                    .context("failed to deserialize metadata.json")?,
-            );
-        } else {
-            ordered_segment_paths.push(entry.path());
-        }
-    }
-
-    ordered_segment_paths.sort();
-
-    let mut buffer = Vec::new();
-    for path in ordered_segment_paths {
-        if buffer.len() >= input_size {
-            break;
-        }
-
-        use async_compression::tokio::bufread::ZstdDecoder;
-        let file = tokio::fs::File::open(path).await?;
-        let reader = tokio::io::BufReader::new(file);
-        let decoder = ZstdDecoder::new(reader);
-        let mut reader = tokio::io::BufReader::new(decoder);
-        tokio::io::copy_buf(&mut reader, &mut buffer).await?;
-    }
-
-    buffer.truncate(input_size);
-
-    Ok(BenchmarkData {
-        wal: buffer,
-        meta: metadata.unwrap(),
-    })
-}
-
-fn criterion_benchmark(c: &mut Criterion) {
-    const INPUT_SIZE: usize = 128 * 1024 * 1024;
-
-    let setup_runtime = tokio::runtime::Builder::new_current_thread()
-        .enable_all()
-        .build()
-        .unwrap();
-
-    let (_temp_dir, bench_data) = setup_runtime.block_on(async move {
-        let cancel = CancellationToken::new();
-        let client = create_s3_client().await.unwrap();
-        let temp_dir = download_bench_data(client, &cancel).await.unwrap();
-        let bench_data = load_bench_data(temp_dir.path(), INPUT_SIZE).await.unwrap();
-
-        (temp_dir, bench_data)
-    });
-
-    eprintln!(
-        "Benchmarking against {} MiB of WAL",
-        INPUT_SIZE / 1024 / 1024
-    );
-
-    let mut group = c.benchmark_group("decode-interpret-wal");
-    group.throughput(criterion::Throughput::Bytes(bench_data.wal.len() as u64));
-    group.sample_size(10);
-
-    group.bench_function("unsharded", |b| {
-        b.iter(|| decode_interpret_main(&bench_data, &[ShardIdentity::unsharded()]))
-    });
-
-    let eight_shards = (0..8)
-        .map(|i| ShardIdentity::new(ShardNumber(i), ShardCount(8), ShardStripeSize(8)).unwrap())
-        .collect::<Vec<_>>();
-
-    group.bench_function("8/8-shards", |b| {
-        b.iter(|| decode_interpret_main(&bench_data, &eight_shards))
-    });
-
-    let four_shards = eight_shards
-        .into_iter()
-        .filter(|s| s.number.0 % 2 == 0)
-        .collect::<Vec<_>>();
-    group.bench_function("4/8-shards", |b| {
-        b.iter(|| decode_interpret_main(&bench_data, &four_shards))
-    });
-
-    let two_shards = four_shards
-        .into_iter()
-        .filter(|s| s.number.0 % 4 == 0)
-        .collect::<Vec<_>>();
-    group.bench_function("2/8-shards", |b| {
-        b.iter(|| decode_interpret_main(&bench_data, &two_shards))
-    });
-}
-
-fn decode_interpret_main(bench: &BenchmarkData, shards: &[ShardIdentity]) {
-    let r = decode_interpret(bench, shards);
-    if let Err(e) = r {
-        panic!("{e:?}");
-    }
-}
-
-fn decode_interpret(bench: &BenchmarkData, shard: &[ShardIdentity]) -> anyhow::Result<()> {
-    let mut decoder = WalStreamDecoder::new(bench.meta.start_lsn, bench.meta.pg_version);
-    let xlogoff: usize = bench.meta.start_lsn.segment_offset(WAL_SEGMENT_SIZE);
-
-    for chunk in bench.wal[xlogoff..].chunks(MAX_SEND_SIZE) {
-        decoder.feed_bytes(chunk);
-        while let Some((lsn, recdata)) = decoder.poll_decode().unwrap() {
-            assert!(lsn.is_aligned());
-            let _ = InterpretedWalRecord::from_bytes_filtered(
-                recdata,
-                shard,
-                lsn,
-                bench.meta.pg_version,
-            )
-            .unwrap();
-        }
-    }
-
-    Ok(())
-}
-criterion_group!(
-    name=benches;
-    config=Criterion::default().with_profiler(PProfProfiler::new(100, Output::Flamegraph(None)));
-    targets=criterion_benchmark
-);
-criterion_main!(benches);

libs/wal_decoder/src/decoder.rs

@@ -1,8 +1,6 @@
 //! This module contains logic for decoding and interpreting
 //! raw bytes which represent a raw Postgres WAL record.
 
-use std::collections::HashMap;
-
 use crate::models::*;
 use crate::serialized_batch::SerializedValueBatch;
 use bytes::{Buf, Bytes};
@@ -16,15 +14,15 @@ use utils::lsn::Lsn;
 
 impl InterpretedWalRecord {
     /// Decode and interpreted raw bytes which represent one Postgres WAL record.
-    /// Data blocks which do not match any of the provided shard identities are filtered out.
+    /// Data blocks which do not match the provided shard identity are filtered out.
     /// Shard 0 is a special case since it tracks all relation sizes. We only give it
     /// the keys that are being written as that is enough for updating relation sizes.
     pub fn from_bytes_filtered(
         buf: Bytes,
-        shards: &[ShardIdentity],
+        shard: &ShardIdentity,
         next_record_lsn: Lsn,
         pg_version: u32,
-    ) -> anyhow::Result<HashMap<ShardIdentity, InterpretedWalRecord>> {
+    ) -> anyhow::Result<InterpretedWalRecord> {
         let mut decoded = DecodedWALRecord::default();
         decode_wal_record(buf, &mut decoded, pg_version)?;
         let xid = decoded.xl_xid;
@@ -35,57 +33,43 @@ impl InterpretedWalRecord {
             FlushUncommittedRecords::No
         };
 
-        let mut shard_records: HashMap<ShardIdentity, InterpretedWalRecord> =
-            HashMap::with_capacity(shards.len());
-        for shard in shards {
-            shard_records.insert(
-                *shard,
-                InterpretedWalRecord {
-                    metadata_record: None,
-                    batch: SerializedValueBatch::default(),
-                    next_record_lsn,
-                    flush_uncommitted,
-                    xid,
-                },
-            );
-        }
-
-        MetadataRecord::from_decoded_filtered(
-            &decoded,
-            &mut shard_records,
-            next_record_lsn,
-            pg_version,
-        )?;
-        SerializedValueBatch::from_decoded_filtered(
+        let metadata_record =
+            MetadataRecord::from_decoded_filtered(&decoded, shard, next_record_lsn, pg_version)?;
+        let batch = SerializedValueBatch::from_decoded_filtered(
             decoded,
-            &mut shard_records,
+            shard,
             next_record_lsn,
             pg_version,
         )?;
 
-        Ok(shard_records)
+        Ok(InterpretedWalRecord {
+            metadata_record,
+            batch,
+            next_record_lsn,
+            flush_uncommitted,
+            xid,
+        })
     }
 }
 
 impl MetadataRecord {
-    /// Populates the given `shard_records` with metadata records from this WAL record, if any,
-    /// discarding those belonging to other shards.
+    /// Builds a metadata record for this WAL record, if any.
     ///
-    /// Only metadata records relevant for the given shards is emitted. Currently, most metadata
+    /// Only metadata records relevant for the given shard are emitted. Currently, most metadata
     /// records are broadcast to all shards for simplicity, but this should be improved.
     fn from_decoded_filtered(
         decoded: &DecodedWALRecord,
-        shard_records: &mut HashMap<ShardIdentity, InterpretedWalRecord>,
+        shard: &ShardIdentity,
         next_record_lsn: Lsn,
         pg_version: u32,
-    ) -> anyhow::Result<()> {
+    ) -> anyhow::Result<Option<MetadataRecord>> {
         // Note: this doesn't actually copy the bytes since
         // the [`Bytes`] type implements it via a level of indirection.
         let mut buf = decoded.record.clone();
         buf.advance(decoded.main_data_offset);
 
         // First, generate metadata records from the decoded WAL record.
-        let metadata_record = match decoded.xl_rmid {
+        let mut metadata_record = match decoded.xl_rmid {
             pg_constants::RM_HEAP_ID | pg_constants::RM_HEAP2_ID => {
                 Self::decode_heapam_record(&mut buf, decoded, pg_version)?
             }
@@ -128,65 +112,41 @@ impl MetadataRecord {
         };
 
         // Next, filter the metadata record by shard.
-        for (shard, record) in shard_records.iter_mut() {
-            match metadata_record {
-                Some(
-                    MetadataRecord::Heapam(HeapamRecord::ClearVmBits(ref clear_vm_bits))
-                    | MetadataRecord::Neonrmgr(NeonrmgrRecord::ClearVmBits(ref clear_vm_bits)),
-                ) => {
-                    // Route VM page updates to the shards that own them. VM pages are stored in the VM fork
-                    // of the main relation. These are sharded and managed just like regular relation pages.
-                    // See: https://github.com/neondatabase/neon/issues/9855
-                    let is_local_vm_page = |heap_blk| {
-                        let vm_blk = pg_constants::HEAPBLK_TO_MAPBLOCK(heap_blk);
-                        shard.is_key_local(&rel_block_to_key(clear_vm_bits.vm_rel, vm_blk))
-                    };
-                    // Send the old and new VM page updates to their respective shards.
-                    let updated_old_heap_blkno = clear_vm_bits
-                        .old_heap_blkno
-                        .filter(|&blkno| is_local_vm_page(blkno));
-                    let updated_new_heap_blkno = clear_vm_bits
-                        .new_heap_blkno
-                        .filter(|&blkno| is_local_vm_page(blkno));
-                    // If neither VM page belongs to this shard, discard the record.
-                    if updated_old_heap_blkno.is_some() || updated_new_heap_blkno.is_some() {
-                        // Clone the record and update it for the current shard.
-                        let mut for_shard = metadata_record.clone();
-                        match for_shard {
-                            Some(
-                                MetadataRecord::Heapam(HeapamRecord::ClearVmBits(
-                                    ref mut clear_vm_bits,
-                                ))
-                                | MetadataRecord::Neonrmgr(NeonrmgrRecord::ClearVmBits(
-                                    ref mut clear_vm_bits,
-                                )),
-                            ) => {
-                                clear_vm_bits.old_heap_blkno = updated_old_heap_blkno;
-                                clear_vm_bits.new_heap_blkno = updated_new_heap_blkno;
-                                record.metadata_record = for_shard;
-                            }
-                            _ => {
-                                unreachable!("for_shard is a clone of what we checked above")
-                            }
-                        }
-                    }
-                }
-                Some(MetadataRecord::LogicalMessage(LogicalMessageRecord::Put(_))) => {
-                    // Filter LogicalMessage records (AUX files) to only be stored on shard zero
-                    if shard.is_shard_zero() {
-                        record.metadata_record = metadata_record;
-                        // No other shards should receive this record, so we stop traversing shards early.
-                        break;
-                    }
-                }
-                _ => {
-                    // All other metadata records are sent to all shards.
-                    record.metadata_record = metadata_record.clone();
+        match metadata_record {
+            Some(
+                MetadataRecord::Heapam(HeapamRecord::ClearVmBits(ref mut clear_vm_bits))
+                | MetadataRecord::Neonrmgr(NeonrmgrRecord::ClearVmBits(ref mut clear_vm_bits)),
+            ) => {
+                // Route VM page updates to the shards that own them. VM pages are stored in the VM fork
+                // of the main relation. These are sharded and managed just like regular relation pages.
+                // See: https://github.com/neondatabase/neon/issues/9855
+                let is_local_vm_page = |heap_blk| {
+                    let vm_blk = pg_constants::HEAPBLK_TO_MAPBLOCK(heap_blk);
+                    shard.is_key_local(&rel_block_to_key(clear_vm_bits.vm_rel, vm_blk))
+                };
+                // Send the old and new VM page updates to their respective shards.
+                clear_vm_bits.old_heap_blkno = clear_vm_bits
+                    .old_heap_blkno
+                    .filter(|&blkno| is_local_vm_page(blkno));
+                clear_vm_bits.new_heap_blkno = clear_vm_bits
+                    .new_heap_blkno
+                    .filter(|&blkno| is_local_vm_page(blkno));
+                // If neither VM page belongs to this shard, discard the record.
+                if clear_vm_bits.old_heap_blkno.is_none() && clear_vm_bits.new_heap_blkno.is_none()
+                {
+                    metadata_record = None
                 }
             }
+            Some(MetadataRecord::LogicalMessage(LogicalMessageRecord::Put(_))) => {
+                // Filter LogicalMessage records (AUX files) to only be stored on shard zero
+                if !shard.is_shard_zero() {
+                    metadata_record = None;
+                }
+            }
+            _ => {}
         }
 
-        Ok(())
+        Ok(metadata_record)
     }
 
     fn decode_heapam_record(

libs/wal_decoder/src/models.rs

@@ -48,7 +48,7 @@ pub mod proto {
     tonic::include_proto!("interpreted_wal");
 }
 
-#[derive(Copy, Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub enum FlushUncommittedRecords {
     Yes,
     No,
@@ -64,7 +64,7 @@ pub struct InterpretedWalRecords {
 }
 
 /// An interpreted Postgres WAL record, ready to be handled by the pageserver
-#[derive(Serialize, Deserialize, Clone)]
+#[derive(Serialize, Deserialize)]
 pub struct InterpretedWalRecord {
     /// Optional metadata record - may cause writes to metadata keys
     /// in the storage engine
@@ -107,7 +107,7 @@ impl InterpretedWalRecord {
 
 /// The interpreted part of the Postgres WAL record which requires metadata
 /// writes to the underlying storage engine.
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub enum MetadataRecord {
     Heapam(HeapamRecord),
     Neonrmgr(NeonrmgrRecord),
@@ -123,12 +123,12 @@ pub enum MetadataRecord {
     Replorigin(ReploriginRecord),
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub enum HeapamRecord {
     ClearVmBits(ClearVmBits),
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub struct ClearVmBits {
     pub new_heap_blkno: Option<u32>,
     pub old_heap_blkno: Option<u32>,
@@ -136,29 +136,29 @@ pub struct ClearVmBits {
     pub flags: u8,
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub enum NeonrmgrRecord {
     ClearVmBits(ClearVmBits),
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub enum SmgrRecord {
     Create(SmgrCreate),
     Truncate(XlSmgrTruncate),
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub struct SmgrCreate {
     pub rel: RelTag,
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub enum DbaseRecord {
     Create(DbaseCreate),
     Drop(DbaseDrop),
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub struct DbaseCreate {
     pub db_id: Oid,
     pub tablespace_id: Oid,
@@ -166,32 +166,32 @@ pub struct DbaseCreate {
     pub src_tablespace_id: Oid,
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub struct DbaseDrop {
     pub db_id: Oid,
     pub tablespace_ids: Vec<Oid>,
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub enum ClogRecord {
     ZeroPage(ClogZeroPage),
     Truncate(ClogTruncate),
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub struct ClogZeroPage {
     pub segno: u32,
     pub rpageno: u32,
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub struct ClogTruncate {
     pub pageno: u32,
     pub oldest_xid: TransactionId,
     pub oldest_xid_db: Oid,
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub enum XactRecord {
     Commit(XactCommon),
     Abort(XactCommon),
@@ -200,7 +200,7 @@ pub enum XactRecord {
     Prepare(XactPrepare),
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub struct XactCommon {
     pub parsed: XlXactParsedRecord,
     pub origin_id: u16,
@@ -209,73 +209,73 @@ pub struct XactCommon {
     pub lsn: Lsn,
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub struct XactPrepare {
     pub xl_xid: TransactionId,
     pub data: Bytes,
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub enum MultiXactRecord {
     ZeroPage(MultiXactZeroPage),
     Create(XlMultiXactCreate),
     Truncate(XlMultiXactTruncate),
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub struct MultiXactZeroPage {
     pub slru_kind: SlruKind,
     pub segno: u32,
     pub rpageno: u32,
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub enum RelmapRecord {
     Update(RelmapUpdate),
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub struct RelmapUpdate {
     pub update: XlRelmapUpdate,
     pub buf: Bytes,
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub enum XlogRecord {
     Raw(RawXlogRecord),
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub struct RawXlogRecord {
     pub info: u8,
     pub lsn: Lsn,
     pub buf: Bytes,
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub enum LogicalMessageRecord {
     Put(PutLogicalMessage),
     #[cfg(feature = "testing")]
     Failpoint,
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub struct PutLogicalMessage {
     pub path: String,
     pub buf: Bytes,
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub enum StandbyRecord {
     RunningXacts(StandbyRunningXacts),
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub struct StandbyRunningXacts {
     pub oldest_running_xid: TransactionId,
 }
 
-#[derive(Clone, Serialize, Deserialize)]
+#[derive(Serialize, Deserialize)]
 pub enum ReploriginRecord {
     Set(XlReploriginSet),
     Drop(XlReploriginDrop),

libs/wal_decoder/src/serialized_batch.rs

@@ -5,7 +5,7 @@
 //! Such batches are created from decoded PG wal records and ingested
 //! by the pageserver by writing directly to the ephemeral file.
 
-use std::collections::{BTreeSet, HashMap};
+use std::collections::BTreeSet;
 
 use bytes::{Bytes, BytesMut};
 use pageserver_api::key::rel_block_to_key;
@@ -22,8 +22,6 @@ use utils::lsn::Lsn;
 
 use pageserver_api::key::Key;
 
-use crate::models::InterpretedWalRecord;
-
 static ZERO_PAGE: Bytes = Bytes::from_static(&[0u8; BLCKSZ as usize]);
 
 /// Accompanying metadata for the batch
@@ -32,7 +30,7 @@ static ZERO_PAGE: Bytes = Bytes::from_static(&[0u8; BLCKSZ as usize]);
 /// relation sizes. In the case of "observed" values, we only need to know
 /// the key and LSN, so two types of metadata are supported to save on network
 /// bandwidth.
-#[derive(Serialize, Deserialize, Clone)]
+#[derive(Serialize, Deserialize)]
 pub enum ValueMeta {
     Serialized(SerializedValueMeta),
     Observed(ObservedValueMeta),
@@ -79,7 +77,7 @@ impl PartialEq for OrderedValueMeta {
 impl Eq for OrderedValueMeta {}
 
 /// Metadata for a [`Value`] serialized into the batch.
-#[derive(Serialize, Deserialize, Clone)]
+#[derive(Serialize, Deserialize)]
 pub struct SerializedValueMeta {
     pub key: CompactKey,
     pub lsn: Lsn,
@@ -91,14 +89,14 @@ pub struct SerializedValueMeta {
 }
 
 /// Metadata for a [`Value`] observed by the batch
-#[derive(Serialize, Deserialize, Clone)]
+#[derive(Serialize, Deserialize)]
 pub struct ObservedValueMeta {
     pub key: CompactKey,
     pub lsn: Lsn,
 }
 
 /// Batch of serialized [`Value`]s.
-#[derive(Serialize, Deserialize, Clone)]
+#[derive(Serialize, Deserialize)]
 pub struct SerializedValueBatch {
     /// [`Value`]s serialized in EphemeralFile's native format,
     /// ready for disk write by the pageserver
@@ -130,8 +128,7 @@ impl Default for SerializedValueBatch {
 }
 
 impl SerializedValueBatch {
-    /// Populates the given `shard_records` with value batches from this WAL record, if any,
-    /// discarding those belonging to other shards.
+    /// Build a batch of serialized values from a decoded PG WAL record
     ///
     /// The batch will only contain values for keys targeting the specifiec
     /// shard. Shard 0 is a special case, where any keys that don't belong to
@@ -139,20 +136,21 @@ impl SerializedValueBatch {
     /// but absent from the raw buffer [`SerializedValueBatch::raw`]).
     pub(crate) fn from_decoded_filtered(
         decoded: DecodedWALRecord,
-        shard_records: &mut HashMap<ShardIdentity, InterpretedWalRecord>,
+        shard: &ShardIdentity,
         next_record_lsn: Lsn,
         pg_version: u32,
-    ) -> anyhow::Result<()> {
-        // First determine how big the buffers need to be and allocate it up-front.
+    ) -> anyhow::Result<SerializedValueBatch> {
+        // First determine how big the buffer needs to be and allocate it up-front.
         // This duplicates some of the work below, but it's empirically much faster.
-        for (shard, record) in shard_records.iter_mut() {
-            assert!(record.batch.is_empty());
-
-            let estimate = Self::estimate_buffer_size(&decoded, shard, pg_version);
-            record.batch.raw = Vec::with_capacity(estimate);
-        }
+        let estimated_buffer_size = Self::estimate_buffer_size(&decoded, shard, pg_version);
+        let mut buf = Vec::<u8>::with_capacity(estimated_buffer_size);
 
+        let mut metadata: Vec<ValueMeta> = Vec::with_capacity(decoded.blocks.len());
+        let mut max_lsn: Lsn = Lsn(0);
+        let mut len: usize = 0;
         for blk in decoded.blocks.iter() {
+            let relative_off = buf.len() as u64;
+
             let rel = RelTag {
                 spcnode: blk.rnode_spcnode,
                 dbnode: blk.rnode_dbnode,
@@ -170,98 +168,99 @@ impl SerializedValueBatch {
                 );
             }
 
-            for (shard, record) in shard_records.iter_mut() {
-                let key_is_local = shard.is_key_local(&key);
+            let key_is_local = shard.is_key_local(&key);
 
-                tracing::debug!(
-                    lsn=%next_record_lsn,
-                    key=%key,
-                    "ingest: shard decision {}",
-                    if !key_is_local { "drop" } else { "keep" },
-                );
+            tracing::debug!(
+                lsn=%next_record_lsn,
+                key=%key,
+                "ingest: shard decision {}",
+                if !key_is_local { "drop" } else { "keep" },
+            );
 
-                if !key_is_local {
-                    if shard.is_shard_zero() {
-                        // Shard 0 tracks relation sizes.  Although we will not store this block, we will observe
-                        // its blkno in case it implicitly extends a relation.
-                        record
-                            .batch
-                            .metadata
-                            .push(ValueMeta::Observed(ObservedValueMeta {
-                                key: key.to_compact(),
-                                lsn: next_record_lsn,
-                            }))
-                    }
-
-                    continue;
-                }
-
-                // Instead of storing full-page-image WAL record,
-                // it is better to store extracted image: we can skip wal-redo
-                // in this case. Also some FPI records may contain multiple (up to 32) pages,
-                // so them have to be copied multiple times.
-                //
-                let val = if Self::block_is_image(&decoded, blk, pg_version) {
-                    // Extract page image from FPI record
-                    let img_len = blk.bimg_len as usize;
-                    let img_offs = blk.bimg_offset as usize;
-                    let mut image = BytesMut::with_capacity(BLCKSZ as usize);
-                    // TODO(vlad): skip the copy
-                    image.extend_from_slice(&decoded.record[img_offs..img_offs + img_len]);
-
-                    if blk.hole_length != 0 {
-                        let tail = image.split_off(blk.hole_offset as usize);
-                        image.resize(image.len() + blk.hole_length as usize, 0u8);
-                        image.unsplit(tail);
-                    }
-                    //
-                    // Match the logic of XLogReadBufferForRedoExtended:
-                    // The page may be uninitialized. If so, we can't set the LSN because
-                    // that would corrupt the page.
-                    //
-                    if !page_is_new(&image) {
-                        page_set_lsn(&mut image, next_record_lsn)
-                    }
-                    assert_eq!(image.len(), BLCKSZ as usize);
-
-                    Value::Image(image.freeze())
-                } else {
-                    Value::WalRecord(NeonWalRecord::Postgres {
-                        will_init: blk.will_init || blk.apply_image,
-                        rec: decoded.record.clone(),
-                    })
-                };
-
-                let relative_off = record.batch.raw.len() as u64;
-
-                val.ser_into(&mut record.batch.raw)
-                    .expect("Writing into in-memory buffer is infallible");
-
-                let val_ser_size = record.batch.raw.len() - relative_off as usize;
-
-                record
-                    .batch
-                    .metadata
-                    .push(ValueMeta::Serialized(SerializedValueMeta {
+            if !key_is_local {
+                if shard.is_shard_zero() {
+                    // Shard 0 tracks relation sizes.  Although we will not store this block, we will observe
+                    // its blkno in case it implicitly extends a relation.
+                    metadata.push(ValueMeta::Observed(ObservedValueMeta {
                         key: key.to_compact(),
                         lsn: next_record_lsn,
-                        batch_offset: relative_off,
-                        len: val_ser_size,
-                        will_init: val.will_init(),
-                    }));
-                record.batch.max_lsn = std::cmp::max(record.batch.max_lsn, next_record_lsn);
-                record.batch.len += 1;
+                    }))
+                }
+
+                continue;
             }
+
+            // Instead of storing full-page-image WAL record,
+            // it is better to store extracted image: we can skip wal-redo
+            // in this case. Also some FPI records may contain multiple (up to 32) pages,
+            // so them have to be copied multiple times.
+            //
+            let val = if Self::block_is_image(&decoded, blk, pg_version) {
+                // Extract page image from FPI record
+                let img_len = blk.bimg_len as usize;
+                let img_offs = blk.bimg_offset as usize;
+                let mut image = BytesMut::with_capacity(BLCKSZ as usize);
+                // TODO(vlad): skip the copy
+                image.extend_from_slice(&decoded.record[img_offs..img_offs + img_len]);
+
+                if blk.hole_length != 0 {
+                    let tail = image.split_off(blk.hole_offset as usize);
+                    image.resize(image.len() + blk.hole_length as usize, 0u8);
+                    image.unsplit(tail);
+                }
+                //
+                // Match the logic of XLogReadBufferForRedoExtended:
+                // The page may be uninitialized. If so, we can't set the LSN because
+                // that would corrupt the page.
+                //
+                if !page_is_new(&image) {
+                    page_set_lsn(&mut image, next_record_lsn)
+                }
+                assert_eq!(image.len(), BLCKSZ as usize);
+
+                Value::Image(image.freeze())
+            } else {
+                Value::WalRecord(NeonWalRecord::Postgres {
+                    will_init: blk.will_init || blk.apply_image,
+                    rec: decoded.record.clone(),
+                })
+            };
+
+            val.ser_into(&mut buf)
+                .expect("Writing into in-memory buffer is infallible");
+
+            let val_ser_size = buf.len() - relative_off as usize;
+
+            metadata.push(ValueMeta::Serialized(SerializedValueMeta {
+                key: key.to_compact(),
+                lsn: next_record_lsn,
+                batch_offset: relative_off,
+                len: val_ser_size,
+                will_init: val.will_init(),
+            }));
+            max_lsn = std::cmp::max(max_lsn, next_record_lsn);
+            len += 1;
         }
 
         if cfg!(any(debug_assertions, test)) {
-            // Validate that the batches are correct
-            for record in shard_records.values() {
-                record.batch.validate_lsn_order();
-            }
+            let batch = Self {
+                raw: buf,
+                metadata,
+                max_lsn,
+                len,
+            };
+
+            batch.validate_lsn_order();
+
+            return Ok(batch);
         }
 
-        Ok(())
+        Ok(Self {
+            raw: buf,
+            metadata,
+            max_lsn,
+            len,
+        })
     }
 
     /// Look into the decoded PG WAL record and determine

pageserver/Cargo.toml

@@ -8,7 +8,7 @@ license.workspace = true
 default = []
 # Enables test-only APIs, incuding failpoints. In particular, enables the `fail_point!` macro,
 # which adds some runtime cost to run tests on outage conditions
-testing = ["fail/failpoints", "pageserver_api/testing", "wal_decoder/testing", "pageserver_client/testing"]
+testing = ["fail/failpoints", "pageserver_api/testing", "wal_decoder/testing"]
 
 [dependencies]
 anyhow.workspace = true
@@ -16,7 +16,6 @@ arc-swap.workspace = true
 async-compression.workspace = true
 async-stream.workspace = true
 bit_field.workspace = true
-bincode.workspace = true
 byteorder.workspace = true
 bytes.workspace = true
 camino.workspace = true
@@ -45,7 +44,6 @@ postgres_backend.workspace = true
 postgres-protocol.workspace = true
 postgres-types.workspace = true
 postgres_initdb.workspace = true
-pprof.workspace = true
 rand.workspace = true
 range-set-blaze = { version = "0.1.16", features = ["alloc"] }
 regex.workspace = true
@@ -110,11 +108,3 @@ harness = false
 [[bench]]
 name = "bench_ingest"
 harness = false
-
-[[bench]]
-name = "upload_queue"
-harness = false
-
-[[bin]]
-name = "test_helper_slow_client_reads"
-required-features = [ "testing" ]

pageserver/benches/upload_queue.rs

@@ -1,87 +0,0 @@
-//! Upload queue benchmarks.
-
-use std::str::FromStr as _;
-use std::sync::atomic::AtomicU32;
-use std::sync::Arc;
-
-use criterion::{criterion_group, criterion_main, Bencher, Criterion};
-use pageserver::tenant::metadata::TimelineMetadata;
-use pageserver::tenant::remote_timeline_client::index::LayerFileMetadata;
-use pageserver::tenant::storage_layer::LayerName;
-use pageserver::tenant::upload_queue::{Delete, UploadOp, UploadQueue, UploadTask};
-use pageserver::tenant::IndexPart;
-use pprof::criterion::{Output, PProfProfiler};
-use utils::generation::Generation;
-use utils::shard::{ShardCount, ShardIndex, ShardNumber};
-
-// Register benchmarks with Criterion.
-criterion_group!(
-    name = benches;
-    config = Criterion::default().with_profiler(PProfProfiler::new(100, Output::Flamegraph(None)));
-    targets = bench_upload_queue_next_ready,
-);
-criterion_main!(benches);
-
-/// Benchmarks the cost of UploadQueue::next_ready() with the given number of in-progress tasks
-/// (which is equivalent to tasks ahead of it in the queue). This has linear cost, and the upload
-/// queue as a whole is thus quadratic.
-///
-/// UploadOp::UploadLayer requires an entire tenant and timeline to construct, so we just test
-/// Delete and UploadMetadata instead. This is incidentally the most expensive case.
-fn bench_upload_queue_next_ready(c: &mut Criterion) {
-    let mut g = c.benchmark_group("upload_queue_next_ready");
-    for inprogress in [0, 1, 10, 100, 1_000, 10_000, 100_000, 1_000_000] {
-        g.bench_function(format!("inprogress={inprogress}"), |b| {
-            run_bench(b, inprogress).unwrap()
-        });
-    }
-
-    fn run_bench(b: &mut Bencher, inprogress: usize) -> anyhow::Result<()> {
-        // Construct two layers. layer0 is in the indexes, layer1 will be deleted.
-        let layer0 = LayerName::from_str("000000000000000000000000000000000000-100000000000000000000000000000000000__00000000016B59D8-00000000016B5A51").expect("invalid name");
-        let layer1 = LayerName::from_str("100000000000000000000000000000000001-200000000000000000000000000000000000__00000000016B59D8-00000000016B5A51").expect("invalid name");
-
-        let metadata = LayerFileMetadata {
-            shard: ShardIndex::new(ShardNumber(1), ShardCount(2)),
-            generation: Generation::Valid(1),
-            file_size: 0,
-        };
-
-        // Construct the (initial and uploaded) index with layer0.
-        let mut index = IndexPart::empty(TimelineMetadata::example());
-        index.layer_metadata.insert(layer0, metadata.clone());
-
-        // Construct the queue.
-        let mut queue = UploadQueue::Uninitialized;
-        let queue = queue.initialize_with_current_remote_index_part(&index, 0)?;
-
-        // Populate inprogress_tasks with a bunch of layer1 deletions.
-        let delete = UploadOp::Delete(Delete {
-            layers: vec![(layer1, metadata)],
-        });
-
-        for task_id in 0..(inprogress as u64) {
-            queue.inprogress_tasks.insert(
-                task_id,
-                Arc::new(UploadTask {
-                    task_id,
-                    retries: AtomicU32::new(0),
-                    op: delete.clone(),
-                    coalesced_ops: Vec::new(),
-                }),
-            );
-        }
-
-        // Benchmark index upload scheduling.
-        let index_upload = UploadOp::UploadMetadata {
-            uploaded: Box::new(index),
-        };
-
-        b.iter(|| {
-            queue.queued_operations.push_front(index_upload.clone());
-            assert!(queue.next_ready().is_some());
-        });
-
-        Ok(())
-    }
-}

pageserver/client/Cargo.toml

@@ -4,9 +4,6 @@ version = "0.1.0"
 edition.workspace = true
 license.workspace = true
 
-[features]
-testing = [ "pageserver_api/testing" ]
-
 [dependencies]
 pageserver_api.workspace = true
 thiserror.workspace = true

pageserver/client/src/page_service.rs

@@ -1,9 +1,6 @@
-use std::sync::{Arc, Mutex};
+use std::pin::Pin;
 
-use futures::{
-    stream::{SplitSink, SplitStream},
-    SinkExt, StreamExt,
-};
+use futures::SinkExt;
 use pageserver_api::{
     models::{
         PagestreamBeMessage, PagestreamFeMessage, PagestreamGetPageRequest,
@@ -13,6 +10,7 @@ use pageserver_api::{
 };
 use tokio::task::JoinHandle;
 use tokio_postgres::CopyOutStream;
+use tokio_stream::StreamExt;
 use tokio_util::sync::CancellationToken;
 use utils::{
     id::{TenantId, TimelineId},
@@ -64,28 +62,15 @@ impl Client {
             .client
             .copy_both_simple(&format!("pagestream_v3 {tenant_id} {timeline_id}"))
             .await?;
-        let (sink, stream) = copy_both.split(); // TODO: actually support splitting of the CopyBothDuplex so the lock inside this split adaptor goes away.
         let Client {
             cancel_on_client_drop,
             conn_task,
             client: _,
         } = self;
-        let shared = Arc::new(Mutex::new(PagestreamShared::ConnTaskRunning(
-            ConnTaskRunning {
-                cancel_on_client_drop,
-                conn_task,
-            },
-        )));
         Ok(PagestreamClient {
-            sink: PagestreamSender {
-                shared: shared.clone(),
-                sink,
-            },
-            stream: PagestreamReceiver {
-                shared: shared.clone(),
-                stream,
-            },
-            shared,
+            copy_both: Box::pin(copy_both),
+            conn_task,
+            cancel_on_client_drop,
         })
     }
 
@@ -112,28 +97,7 @@ impl Client {
 
 /// Create using [`Client::pagestream`].
 pub struct PagestreamClient {
-    shared: Arc<Mutex<PagestreamShared>>,
-    sink: PagestreamSender,
-    stream: PagestreamReceiver,
-}
-
-pub struct PagestreamSender {
-    #[allow(dead_code)]
-    shared: Arc<Mutex<PagestreamShared>>,
-    sink: SplitSink<tokio_postgres::CopyBothDuplex<bytes::Bytes>, bytes::Bytes>,
-}
-
-pub struct PagestreamReceiver {
-    #[allow(dead_code)]
-    shared: Arc<Mutex<PagestreamShared>>,
-    stream: SplitStream<tokio_postgres::CopyBothDuplex<bytes::Bytes>>,
-}
-
-enum PagestreamShared {
-    ConnTaskRunning(ConnTaskRunning),
-    ConnTaskCancelledJoinHandleReturnedOrDropped,
-}
-struct ConnTaskRunning {
+    copy_both: Pin<Box<tokio_postgres::CopyBothDuplex<bytes::Bytes>>>,
     cancel_on_client_drop: Option<tokio_util::sync::DropGuard>,
     conn_task: JoinHandle<()>,
 }
@@ -146,11 +110,11 @@ pub struct RelTagBlockNo {
 impl PagestreamClient {
     pub async fn shutdown(self) {
         let Self {
-            shared,
-            sink,
-            stream,
-        } = { self };
-        // The `copy_both` split into `sink` and `stream` contains internal channel sender, the receiver of which is polled by `conn_task`.
+            copy_both,
+            cancel_on_client_drop: cancel_conn_task,
+            conn_task,
+        } = self;
+        // The `copy_both` contains internal channel sender, the receiver of which is polled by `conn_task`.
         // When `conn_task` observes the sender has been dropped, it sends a `FeMessage::CopyFail` into the connection.
         // (see https://github.com/neondatabase/rust-postgres/blob/2005bf79573b8add5cf205b52a2b208e356cc8b0/tokio-postgres/src/copy_both.rs#L56).
         //
@@ -167,77 +131,27 @@ impl PagestreamClient {
         //
         // NB: page_service doesn't have a use case to exit the `pagestream` mode currently.
         // => https://github.com/neondatabase/neon/issues/6390
-        let ConnTaskRunning {
-            cancel_on_client_drop,
-            conn_task,
-        } = {
-            let mut guard = shared.lock().unwrap();
-            match std::mem::replace(
-                &mut *guard,
-                PagestreamShared::ConnTaskCancelledJoinHandleReturnedOrDropped,
-            ) {
-                PagestreamShared::ConnTaskRunning(conn_task_running) => conn_task_running,
-                PagestreamShared::ConnTaskCancelledJoinHandleReturnedOrDropped => unreachable!(),
-            }
-        };
-        let _ = cancel_on_client_drop.unwrap();
+        let _ = cancel_conn_task.unwrap();
         conn_task.await.unwrap();
-
-        // Now drop the split copy_both.
-        drop(sink);
-        drop(stream);
-    }
-
-    pub fn split(self) -> (PagestreamSender, PagestreamReceiver) {
-        let Self {
-            shared: _,
-            sink,
-            stream,
-        } = self;
-        (sink, stream)
+        drop(copy_both);
     }
 
     pub async fn getpage(
         &mut self,
         req: PagestreamGetPageRequest,
     ) -> anyhow::Result<PagestreamGetPageResponse> {
-        self.getpage_send(req).await?;
-        self.getpage_recv().await
-    }
+        let req = PagestreamFeMessage::GetPage(req);
+        let req: bytes::Bytes = req.serialize();
+        // let mut req = tokio_util::io::ReaderStream::new(&req);
+        let mut req = tokio_stream::once(Ok(req));
 
-    pub async fn getpage_send(&mut self, req: PagestreamGetPageRequest) -> anyhow::Result<()> {
-        self.sink.getpage_send(req).await
-    }
+        self.copy_both.send_all(&mut req).await?;
 
-    pub async fn getpage_recv(&mut self) -> anyhow::Result<PagestreamGetPageResponse> {
-        self.stream.getpage_recv().await
-    }
-}
-
-impl PagestreamSender {
-    // TODO: maybe make this impl Sink instead for better composability?
-    pub async fn send(&mut self, msg: PagestreamFeMessage) -> anyhow::Result<()> {
-        let msg = msg.serialize();
-        self.sink.send_all(&mut tokio_stream::once(Ok(msg))).await?;
-        Ok(())
-    }
-
-    pub async fn getpage_send(&mut self, req: PagestreamGetPageRequest) -> anyhow::Result<()> {
-        self.send(PagestreamFeMessage::GetPage(req)).await
-    }
-}
-
-impl PagestreamReceiver {
-    // TODO: maybe make this impl Stream instead for better composability?
-    pub async fn recv(&mut self) -> anyhow::Result<PagestreamBeMessage> {
-        let next: Option<Result<bytes::Bytes, _>> = self.stream.next().await;
+        let next: Option<Result<bytes::Bytes, _>> = self.copy_both.next().await;
         let next: bytes::Bytes = next.unwrap()?;
-        PagestreamBeMessage::deserialize(next)
-    }
 
-    pub async fn getpage_recv(&mut self) -> anyhow::Result<PagestreamGetPageResponse> {
-        let next: PagestreamBeMessage = self.recv().await?;
-        match next {
+        let msg = PagestreamBeMessage::deserialize(next)?;
+        match msg {
             PagestreamBeMessage::GetPage(p) => Ok(p),
             PagestreamBeMessage::Error(e) => anyhow::bail!("Error: {:?}", e),
             PagestreamBeMessage::Exists(_)
@@ -246,14 +160,7 @@ impl PagestreamReceiver {
             | PagestreamBeMessage::GetSlruSegment(_) => {
                 anyhow::bail!(
                     "unexpected be message kind in response to getpage request: {}",
-                    next.kind()
-                )
-            }
-            #[cfg(feature = "testing")]
-            PagestreamBeMessage::Test(_) => {
-                anyhow::bail!(
-                    "unexpected be message kind in response to getpage request: {}",
-                    next.kind()
+                    msg.kind()
                 )
             }
         }

pageserver/ctl/Cargo.toml

@@ -8,11 +8,9 @@ license.workspace = true
 
 [dependencies]
 anyhow.workspace = true
-bincode.workspace = true
 camino.workspace = true
 clap = { workspace = true, features = ["string"] }
 humantime.workspace = true
-itertools.workspace = true
 pageserver = { path = ".." }
 pageserver_api.workspace = true
 remote_storage = { path = "../../libs/remote_storage" }

pageserver/ctl/src/main.rs

@@ -9,9 +9,7 @@ mod index_part;
 mod key;
 mod layer_map_analyzer;
 mod layers;
-mod page_trace;
 
-use page_trace::PageTraceCmd;
 use std::{
     str::FromStr,
     time::{Duration, SystemTime},
@@ -66,7 +64,6 @@ enum Commands {
     Layer(LayerCmd),
     /// Debug print a hex key found from logs
     Key(key::DescribeKeyCommand),
-    PageTrace(PageTraceCmd),
 }
 
 /// Read and update pageserver metadata file
@@ -186,7 +183,6 @@ async fn main() -> anyhow::Result<()> {
                 .await?;
         }
         Commands::Key(dkc) => dkc.execute(),
-        Commands::PageTrace(cmd) => page_trace::main(&cmd)?,
     };
     Ok(())
 }

pageserver/ctl/src/page_trace.rs

@@ -1,73 +0,0 @@
-use std::collections::HashMap;
-use std::io::BufReader;
-
-use camino::Utf8PathBuf;
-use clap::Parser;
-use itertools::Itertools as _;
-use pageserver_api::key::{CompactKey, Key};
-use pageserver_api::models::PageTraceEvent;
-use pageserver_api::reltag::RelTag;
-
-/// Parses a page trace (as emitted by the `page_trace` timeline API), and outputs stats.
-#[derive(Parser)]
-pub(crate) struct PageTraceCmd {
-    /// Trace input file.
-    path: Utf8PathBuf,
-}
-
-pub(crate) fn main(cmd: &PageTraceCmd) -> anyhow::Result<()> {
-    let mut file = BufReader::new(std::fs::OpenOptions::new().read(true).open(&cmd.path)?);
-    let mut events: Vec<PageTraceEvent> = Vec::new();
-    loop {
-        match bincode::deserialize_from(&mut file) {
-            Ok(event) => events.push(event),
-            Err(err) => {
-                if let bincode::ErrorKind::Io(ref err) = *err {
-                    if err.kind() == std::io::ErrorKind::UnexpectedEof {
-                        break;
-                    }
-                }
-                return Err(err.into());
-            }
-        }
-    }
-
-    let mut reads_by_relation: HashMap<RelTag, i64> = HashMap::new();
-    let mut reads_by_key: HashMap<CompactKey, i64> = HashMap::new();
-
-    for event in events {
-        let key = Key::from_compact(event.key);
-        let reltag = RelTag {
-            spcnode: key.field2,
-            dbnode: key.field3,
-            relnode: key.field4,
-            forknum: key.field5,
-        };
-
-        *reads_by_relation.entry(reltag).or_default() += 1;
-        *reads_by_key.entry(event.key).or_default() += 1;
-    }
-
-    let multi_read_keys = reads_by_key
-        .into_iter()
-        .filter(|(_, count)| *count > 1)
-        .sorted_by_key(|(key, count)| (-*count, *key))
-        .collect_vec();
-
-    println!("Multi-read keys: {}", multi_read_keys.len());
-    for (key, count) in multi_read_keys {
-        println!("  {key}: {count}");
-    }
-
-    let reads_by_relation = reads_by_relation
-        .into_iter()
-        .sorted_by_key(|(rel, count)| (-*count, *rel))
-        .collect_vec();
-
-    println!("Reads by relation:");
-    for (reltag, count) in reads_by_relation {
-        println!("  {reltag}: {count}");
-    }
-
-    Ok(())
-}

pageserver/src/bin/test_helper_slow_client_reads.rs

@@ -1,65 +0,0 @@
-use std::{
-    io::{stdin, stdout, Read, Write},
-    time::Duration,
-};
-
-use clap::Parser;
-use pageserver_api::models::{PagestreamRequest, PagestreamTestRequest};
-use utils::{
-    id::{TenantId, TimelineId},
-    lsn::Lsn,
-};
-
-#[derive(clap::Parser)]
-struct Args {
-    connstr: String,
-    tenant_id: TenantId,
-    timeline_id: TimelineId,
-}
-
-#[tokio::main]
-async fn main() -> anyhow::Result<()> {
-    let Args {
-        connstr,
-        tenant_id,
-        timeline_id,
-    } = Args::parse();
-    let client = pageserver_client::page_service::Client::new(connstr).await?;
-    let client = client.pagestream(tenant_id, timeline_id).await?;
-    let (mut sender, _receiver) = client.split();
-
-    eprintln!("filling the pipe");
-    let mut msg = 0;
-    loop {
-        msg += 1;
-        let fut = sender.send(pageserver_api::models::PagestreamFeMessage::Test(
-            PagestreamTestRequest {
-                hdr: PagestreamRequest {
-                    reqid: 0,
-                    request_lsn: Lsn(23),
-                    not_modified_since: Lsn(23),
-                },
-                batch_key: 42,
-                message: format!("message {}", msg),
-            },
-        ));
-        let Ok(res) = tokio::time::timeout(Duration::from_secs(10), fut).await else {
-            eprintln!("pipe seems full");
-            break;
-        };
-        let _: () = res?;
-    }
-
-    let n = stdout().write(b"R")?;
-    assert_eq!(n, 1);
-    stdout().flush()?;
-
-    eprintln!("waiting for signal to tell us to exit");
-
-    let mut buf = [0u8; 1];
-    stdin().read_exact(&mut buf)?;
-
-    eprintln!("termination signal received, exiting");
-
-    anyhow::Ok(())
-}

pageserver/src/http/routes.rs

@@ -27,7 +27,6 @@ use pageserver_api::models::LocationConfigMode;
 use pageserver_api::models::LsnLease;
 use pageserver_api::models::LsnLeaseRequest;
 use pageserver_api::models::OffloadedTimelineInfo;
-use pageserver_api::models::PageTraceEvent;
 use pageserver_api::models::ShardParameters;
 use pageserver_api::models::TenantConfigPatchRequest;
 use pageserver_api::models::TenantDetails;
@@ -52,9 +51,7 @@ use pageserver_api::shard::TenantShardId;
 use remote_storage::DownloadError;
 use remote_storage::GenericRemoteStorage;
 use remote_storage::TimeTravelError;
-use scopeguard::defer;
 use tenant_size_model::{svg::SvgBranchKind, SizeResult, StorageModel};
-use tokio::time::Instant;
 use tokio_util::io::StreamReader;
 use tokio_util::sync::CancellationToken;
 use tracing::*;
@@ -1524,71 +1521,6 @@ async fn timeline_gc_unblocking_handler(
     block_or_unblock_gc(request, false).await
 }
 
-/// Traces GetPage@LSN requests for a timeline, and emits metadata in an efficient binary encoding.
-/// Use the `pagectl page-trace` command to decode and analyze the output.
-async fn timeline_page_trace_handler(
-    request: Request<Body>,
-    cancel: CancellationToken,
-) -> Result<Response<Body>, ApiError> {
-    let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
-    let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
-    let state = get_state(&request);
-    check_permission(&request, None)?;
-
-    let size_limit: usize = parse_query_param(&request, "size_limit_bytes")?.unwrap_or(1024 * 1024);
-    let time_limit_secs: u64 = parse_query_param(&request, "time_limit_secs")?.unwrap_or(5);
-
-    // Convert size limit to event limit based on the serialized size of an event. The event size is
-    // fixed, as the default bincode serializer uses fixed-width integer encoding.
-    let event_size = bincode::serialize(&PageTraceEvent::default())
-        .map_err(|err| ApiError::InternalServerError(err.into()))?
-        .len();
-    let event_limit = size_limit / event_size;
-
-    let timeline =
-        active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
-            .await?;
-
-    // Install a page trace, unless one is already in progress. We just use a buffered channel,
-    // which may 2x the memory usage in the worst case, but it's still bounded.
-    let (trace_tx, mut trace_rx) = tokio::sync::mpsc::channel(event_limit);
-    let cur = timeline.page_trace.load();
-    let installed = cur.is_none()
-        && timeline
-            .page_trace
-            .compare_and_swap(cur, Some(Arc::new(trace_tx)))
-            .is_none();
-    if !installed {
-        return Err(ApiError::Conflict("page trace already active".to_string()));
-    }
-    defer!(timeline.page_trace.store(None)); // uninstall on return
-
-    // Collect the trace and return it to the client. We could stream the response, but this is
-    // simple and fine.
-    let mut body = Vec::with_capacity(size_limit);
-    let deadline = Instant::now() + Duration::from_secs(time_limit_secs);
-
-    while body.len() < size_limit {
-        tokio::select! {
-            event = trace_rx.recv() => {
-                let Some(event) = event else {
-                    break; // shouldn't happen (sender doesn't close, unless timeline dropped)
-                };
-                bincode::serialize_into(&mut body, &event)
-                    .map_err(|err| ApiError::InternalServerError(err.into()))?;
-            }
-            _ = tokio::time::sleep_until(deadline) => break, // time limit reached
-            _ = cancel.cancelled() => return Err(ApiError::Cancelled),
-        }
-    }
-
-    Ok(Response::builder()
-        .status(StatusCode::OK)
-        .header(header::CONTENT_TYPE, "application/octet-stream")
-        .body(hyper::Body::from(body))
-        .unwrap())
-}
-
 /// Adding a block is `POST ../block_gc`, removing a block is `POST ../unblock_gc`.
 ///
 /// Both are technically unsafe because they might fire off index uploads, thus they are POST.
@@ -3547,10 +3479,6 @@ pub fn make_router(
             "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/unblock_gc",
             |r| api_handler(r, timeline_gc_unblocking_handler),
         )
-        .get(
-            "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/page_trace",
-            |r| api_handler(r, timeline_page_trace_handler),
-        )
         .post("/v1/tenant/:tenant_shard_id/heatmap_upload", |r| {
             api_handler(r, secondary_upload_handler)
         })

pageserver/src/import_datadir.rs

@@ -278,8 +278,6 @@ async fn import_wal(
 
     let mut walingest = WalIngest::new(tline, startpoint, ctx).await?;
 
-    let shard = vec![*tline.get_shard_identity()];
-
     while last_lsn <= endpoint {
         // FIXME: assume postgresql tli 1 for now
         let filename = XLogFileName(1, segno, WAL_SEGMENT_SIZE);
@@ -316,12 +314,10 @@ async fn import_wal(
             if let Some((lsn, recdata)) = waldecoder.poll_decode()? {
                 let interpreted = InterpretedWalRecord::from_bytes_filtered(
                     recdata,
-                    &shard,
+                    tline.get_shard_identity(),
                     lsn,
                     tline.pg_version,
-                )?
-                .remove(tline.get_shard_identity())
-                .unwrap();
+                )?;
 
                 walingest
                     .ingest_record(interpreted, &mut modification, ctx)
@@ -415,7 +411,6 @@ pub async fn import_wal_from_tar(
     let mut offset = start_lsn.segment_offset(WAL_SEGMENT_SIZE);
     let mut last_lsn = start_lsn;
     let mut walingest = WalIngest::new(tline, start_lsn, ctx).await?;
-    let shard = vec![*tline.get_shard_identity()];
 
     // Ingest wal until end_lsn
     info!("importing wal until {}", end_lsn);
@@ -464,12 +459,10 @@ pub async fn import_wal_from_tar(
             if let Some((lsn, recdata)) = waldecoder.poll_decode()? {
                 let interpreted = InterpretedWalRecord::from_bytes_filtered(
                     recdata,
-                    &shard,
+                    tline.get_shard_identity(),
                     lsn,
                     tline.pg_version,
-                )?
-                .remove(tline.get_shard_identity())
-                .unwrap();
+                )?;
 
                 walingest
                     .ingest_record(interpreted, &mut modification, ctx)

pageserver/src/metrics.rs

@@ -1224,189 +1224,117 @@ pub(crate) struct SmgrOpTimerInner {
     global_flush_in_progress_micros: IntCounter,
     per_timeline_flush_in_progress_micros: IntCounter,
 
-    throttling: Arc<tenant_throttling::Pagestream>,
-
     timings: SmgrOpTimerState,
 }
 
-/// The stages of request processing are represented by the enum variants.
-/// Used as part of [`SmgrOpTimerInner::timings`].
-///
-/// Request processing calls into the `SmgrOpTimer::observe_*` methods at the
-/// transition points.
-/// These methods bump relevant counters and then update [`SmgrOpTimerInner::timings`]
-/// to the next state.
-///
-/// Each request goes through every stage, in all configurations.
-///
 #[derive(Debug)]
 enum SmgrOpTimerState {
     Received {
-        // In the future, we may want to track the full time the request spent
-        // inside pageserver process (time spent in kernel buffers can't be tracked).
-        // `received_at` would be used for that.
-        #[allow(dead_code)]
         received_at: Instant,
     },
-    Throttling {
+    ThrottleDoneExecutionStarting {
+        received_at: Instant,
         throttle_started_at: Instant,
+        started_execution_at: Instant,
     },
-    Batching {
-        throttle_done_at: Instant,
-    },
-    Executing {
-        execution_started_at: Instant,
-    },
-    Flushing,
-    // NB: when adding observation points, remember to update the Drop impl.
 }
 
-// NB: when adding observation points, remember to update the Drop impl.
-impl SmgrOpTimer {
-    /// See [`SmgrOpTimerState`] for more context.
-    pub(crate) fn observe_throttle_start(&mut self, at: Instant) {
-        let Some(inner) = self.0.as_mut() else {
-            return;
-        };
-        let SmgrOpTimerState::Received { received_at: _ } = &mut inner.timings else {
-            return;
-        };
-        inner.throttling.count_accounted_start.inc();
-        inner.timings = SmgrOpTimerState::Throttling {
-            throttle_started_at: at,
-        };
-    }
-
-    /// See [`SmgrOpTimerState`] for more context.
-    pub(crate) fn observe_throttle_done(&mut self, throttle: ThrottleResult) {
-        let Some(inner) = self.0.as_mut() else {
-            return;
-        };
-        let SmgrOpTimerState::Throttling {
-            throttle_started_at,
-        } = &inner.timings
-        else {
-            return;
-        };
-        inner.throttling.count_accounted_finish.inc();
-        match throttle {
-            ThrottleResult::NotThrottled { end } => {
-                inner.timings = SmgrOpTimerState::Batching {
-                    throttle_done_at: end,
-                };
-            }
-            ThrottleResult::Throttled { end } => {
-                // update metrics
-                inner.throttling.count_throttled.inc();
-                inner
-                    .throttling
-                    .wait_time
-                    .inc_by((end - *throttle_started_at).as_micros().try_into().unwrap());
-                // state transition
-                inner.timings = SmgrOpTimerState::Batching {
-                    throttle_done_at: end,
-                };
-            }
-        }
-    }
-
-    /// See [`SmgrOpTimerState`] for more context.
-    pub(crate) fn observe_execution_start(&mut self, at: Instant) {
-        let Some(inner) = self.0.as_mut() else {
-            return;
-        };
-        let SmgrOpTimerState::Batching { throttle_done_at } = &inner.timings else {
-            return;
-        };
-        // update metrics
-        let batch = at - *throttle_done_at;
-        inner.global_batch_wait_time.observe(batch.as_secs_f64());
-        inner
-            .per_timeline_batch_wait_time
-            .observe(batch.as_secs_f64());
-        // state transition
-        inner.timings = SmgrOpTimerState::Executing {
-            execution_started_at: at,
-        }
-    }
-
-    /// For all but the first caller, this is a no-op.
-    /// The first callers receives Some, subsequent ones None.
-    ///
-    /// See [`SmgrOpTimerState`] for more context.
-    pub(crate) fn observe_execution_end_flush_start(
-        &mut self,
-        at: Instant,
-    ) -> Option<SmgrOpFlushInProgress> {
-        // NB: unlike the other observe_* methods, this one take()s.
-        #[allow(clippy::question_mark)] // maintain similar code pattern.
-        let Some(mut inner) = self.0.take() else {
-            return None;
-        };
-        let SmgrOpTimerState::Executing {
-            execution_started_at,
-        } = &inner.timings
-        else {
-            return None;
-        };
-        // update metrics
-        let execution = at - *execution_started_at;
-        inner
-            .global_execution_latency_histo
-            .observe(execution.as_secs_f64());
-        if let Some(per_timeline_execution_latency_histo) =
-            &inner.per_timeline_execution_latency_histo
-        {
-            per_timeline_execution_latency_histo.observe(execution.as_secs_f64());
-        }
-
-        // state transition
-        inner.timings = SmgrOpTimerState::Flushing;
-
-        // return the flush in progress object which
-        // will do the remaining metrics updates
-        let SmgrOpTimerInner {
-            global_flush_in_progress_micros,
-            per_timeline_flush_in_progress_micros,
-            ..
-        } = inner;
-        Some(SmgrOpFlushInProgress {
-            flush_started_at: at,
-            global_micros: global_flush_in_progress_micros,
-            per_timeline_micros: per_timeline_flush_in_progress_micros,
-        })
-    }
-}
-
-/// The last stage of request processing is serializing and flushing the request
-/// into the TCP connection. We want to make slow flushes observable
-/// _while they are occuring_, so this struct provides a wrapper method [`Self::measure`]
-/// to periodically bump the metric.
-///
-/// If in the future we decide that we're not interested in live updates, we can
-/// add another `observe_*` method to [`SmgrOpTimer`], follow the existing pattern there,
-/// and remove this struct from the code base.
 pub(crate) struct SmgrOpFlushInProgress {
     flush_started_at: Instant,
     global_micros: IntCounter,
     per_timeline_micros: IntCounter,
 }
 
+impl SmgrOpTimer {
+    pub(crate) fn observe_throttle_done_execution_starting(&mut self, throttle: &ThrottleResult) {
+        let inner = self.0.as_mut().expect("other public methods consume self");
+        match (&mut inner.timings, throttle) {
+            (SmgrOpTimerState::Received { received_at }, throttle) => match throttle {
+                ThrottleResult::NotThrottled { start } => {
+                    inner.timings = SmgrOpTimerState::ThrottleDoneExecutionStarting {
+                        received_at: *received_at,
+                        throttle_started_at: *start,
+                        started_execution_at: *start,
+                    };
+                }
+                ThrottleResult::Throttled { start, end } => {
+                    inner.timings = SmgrOpTimerState::ThrottleDoneExecutionStarting {
+                        received_at: *start,
+                        throttle_started_at: *start,
+                        started_execution_at: *end,
+                    };
+                }
+            },
+            (x, _) => panic!("called in unexpected state: {x:?}"),
+        }
+    }
+
+    pub(crate) fn observe_smgr_op_completion_and_start_flushing(mut self) -> SmgrOpFlushInProgress {
+        let (flush_start, inner) = self
+            .smgr_op_end()
+            .expect("this method consume self, and the only other caller is drop handler");
+        let SmgrOpTimerInner {
+            global_flush_in_progress_micros,
+            per_timeline_flush_in_progress_micros,
+            ..
+        } = inner;
+        SmgrOpFlushInProgress {
+            flush_started_at: flush_start,
+            global_micros: global_flush_in_progress_micros,
+            per_timeline_micros: per_timeline_flush_in_progress_micros,
+        }
+    }
+
+    /// Returns `None`` if this method has already been called, `Some` otherwise.
+    fn smgr_op_end(&mut self) -> Option<(Instant, SmgrOpTimerInner)> {
+        let inner = self.0.take()?;
+
+        let now = Instant::now();
+
+        let batch;
+        let execution;
+        let throttle;
+        match inner.timings {
+            SmgrOpTimerState::Received { received_at } => {
+                batch = (now - received_at).as_secs_f64();
+                // TODO: use label for dropped requests.
+                // This is quite rare in practice, only during tenant/pageservers shutdown.
+                throttle = Duration::ZERO;
+                execution = Duration::ZERO.as_secs_f64();
+            }
+            SmgrOpTimerState::ThrottleDoneExecutionStarting {
+                received_at,
+                throttle_started_at,
+                started_execution_at,
+            } => {
+                batch = (throttle_started_at - received_at).as_secs_f64();
+                throttle = started_execution_at - throttle_started_at;
+                execution = (now - started_execution_at).as_secs_f64();
+            }
+        }
+
+        // update time spent in batching
+        inner.global_batch_wait_time.observe(batch);
+        inner.per_timeline_batch_wait_time.observe(batch);
+
+        // time spent in throttle metric is updated by throttle impl
+        let _ = throttle;
+
+        // update metrics for execution latency
+        inner.global_execution_latency_histo.observe(execution);
+        if let Some(per_timeline_execution_latency_histo) =
+            &inner.per_timeline_execution_latency_histo
+        {
+            per_timeline_execution_latency_histo.observe(execution);
+        }
+
+        Some((now, inner))
+    }
+}
+
 impl Drop for SmgrOpTimer {
     fn drop(&mut self) {
-        // In case of early drop, update any of the remaining metrics with
-        // observations so that (started,finished) counter pairs balance out
-        // and all counters on the latency path have the the same number of
-        // observations.
-        // It's technically lying and it would be better if each metric had
-        // a separate label or similar for cancelled requests.
-        // But we don't have that right now and counter pairs balancing
-        // out is useful when using the metrics in panels and whatnot.
-        let now = Instant::now();
-        self.observe_throttle_start(now);
-        self.observe_throttle_done(ThrottleResult::NotThrottled { end: now });
-        self.observe_execution_start(now);
-        self.observe_execution_end_flush_start(now);
+        self.smgr_op_end();
     }
 }
 
@@ -1417,12 +1345,12 @@ impl SmgrOpFlushInProgress {
     {
         let mut fut = std::pin::pin!(fut);
 
+        let now = Instant::now();
         // Whenever observe_guard gets called, or dropped,
         // it adds the time elapsed since its last call to metrics.
         // Last call is tracked in `now`.
         let mut observe_guard = scopeguard::guard(
             || {
-                let now = Instant::now();
                 let elapsed = now - self.flush_started_at;
                 self.global_micros
                     .inc_by(u64::try_from(elapsed.as_micros()).unwrap());
@@ -1463,10 +1391,9 @@ pub enum SmgrQueryType {
     GetPageAtLsn,
     GetDbSize,
     GetSlruSegment,
-    #[cfg(feature = "testing")]
-    Test,
 }
 
+#[derive(Debug)]
 pub(crate) struct SmgrQueryTimePerTimeline {
     global_started: [IntCounter; SmgrQueryType::COUNT],
     global_latency: [Histogram; SmgrQueryType::COUNT],
@@ -1478,7 +1405,6 @@ pub(crate) struct SmgrQueryTimePerTimeline {
     per_timeline_flush_in_progress_micros: IntCounter,
     global_batch_wait_time: Histogram,
     per_timeline_batch_wait_time: Histogram,
-    throttling: Arc<tenant_throttling::Pagestream>,
 }
 
 static SMGR_QUERY_STARTED_GLOBAL: Lazy<IntCounterVec> = Lazy::new(|| {
@@ -1684,11 +1610,7 @@ static PAGE_SERVICE_SMGR_BATCH_WAIT_TIME_GLOBAL: Lazy<Histogram> = Lazy::new(||
 });
 
 impl SmgrQueryTimePerTimeline {
-    pub(crate) fn new(
-        tenant_shard_id: &TenantShardId,
-        timeline_id: &TimelineId,
-        pagestream_throttle_metrics: Arc<tenant_throttling::Pagestream>,
-    ) -> Self {
+    pub(crate) fn new(tenant_shard_id: &TenantShardId, timeline_id: &TimelineId) -> Self {
         let tenant_id = tenant_shard_id.tenant_id.to_string();
         let shard_slug = format!("{}", tenant_shard_id.shard_slug());
         let timeline_id = timeline_id.to_string();
@@ -1749,7 +1671,6 @@ impl SmgrQueryTimePerTimeline {
             per_timeline_flush_in_progress_micros,
             global_batch_wait_time,
             per_timeline_batch_wait_time,
-            throttling: pagestream_throttle_metrics,
         }
     }
     pub(crate) fn start_smgr_op(&self, op: SmgrQueryType, received_at: Instant) -> SmgrOpTimer {
@@ -1765,24 +1686,88 @@ impl SmgrQueryTimePerTimeline {
         SmgrOpTimer(Some(SmgrOpTimerInner {
             global_execution_latency_histo: self.global_latency[op as usize].clone(),
             per_timeline_execution_latency_histo: per_timeline_latency_histo,
+            timings: SmgrOpTimerState::Received { received_at },
             global_flush_in_progress_micros: self.global_flush_in_progress_micros.clone(),
             per_timeline_flush_in_progress_micros: self
                 .per_timeline_flush_in_progress_micros
                 .clone(),
             global_batch_wait_time: self.global_batch_wait_time.clone(),
             per_timeline_batch_wait_time: self.per_timeline_batch_wait_time.clone(),
-            throttling: self.throttling.clone(),
-            timings: SmgrOpTimerState::Received { received_at },
         }))
     }
 
-    /// TODO: do something about this? seems odd, we have a similar call on SmgrOpTimer
     pub(crate) fn observe_getpage_batch_start(&self, batch_size: usize) {
         self.global_batch_size.observe(batch_size as f64);
         self.per_timeline_batch_size.observe(batch_size as f64);
     }
 }
 
+#[cfg(test)]
+mod smgr_query_time_tests {
+    use std::time::Instant;
+
+    use pageserver_api::shard::TenantShardId;
+    use strum::IntoEnumIterator;
+    use utils::id::{TenantId, TimelineId};
+
+    // Regression test, we used hard-coded string constants before using an enum.
+    #[test]
+    fn op_label_name() {
+        use super::SmgrQueryType::*;
+        let expect: [(super::SmgrQueryType, &'static str); 5] = [
+            (GetRelExists, "get_rel_exists"),
+            (GetRelSize, "get_rel_size"),
+            (GetPageAtLsn, "get_page_at_lsn"),
+            (GetDbSize, "get_db_size"),
+            (GetSlruSegment, "get_slru_segment"),
+        ];
+        for (op, expect) in expect {
+            let actual: &'static str = op.into();
+            assert_eq!(actual, expect);
+        }
+    }
+
+    #[test]
+    fn basic() {
+        let ops: Vec<_> = super::SmgrQueryType::iter().collect();
+
+        for op in &ops {
+            let tenant_id = TenantId::generate();
+            let timeline_id = TimelineId::generate();
+            let metrics = super::SmgrQueryTimePerTimeline::new(
+                &TenantShardId::unsharded(tenant_id),
+                &timeline_id,
+            );
+
+            let get_counts = || {
+                let global: u64 = ops
+                    .iter()
+                    .map(|op| metrics.global_latency[*op as usize].get_sample_count())
+                    .sum();
+                (
+                    global,
+                    metrics.per_timeline_getpage_latency.get_sample_count(),
+                )
+            };
+
+            let (pre_global, pre_per_tenant_timeline) = get_counts();
+            assert_eq!(pre_per_tenant_timeline, 0);
+
+            let timer = metrics.start_smgr_op(*op, Instant::now());
+            drop(timer);
+
+            let (post_global, post_per_tenant_timeline) = get_counts();
+            if matches!(op, super::SmgrQueryType::GetPageAtLsn) {
+                // getpage ops are tracked per-timeline, others aren't
+                assert_eq!(post_per_tenant_timeline, 1);
+            } else {
+                assert_eq!(post_per_tenant_timeline, 0);
+            }
+            assert!(post_global > pre_global);
+        }
+    }
+}
+
 // keep in sync with control plane Go code so that we can validate
 // compute's basebackup_ms metric with our perspective in the context of SLI/SLO.
 static COMPUTE_STARTUP_BUCKETS: Lazy<[f64; 28]> = Lazy::new(|| {
@@ -3578,7 +3563,9 @@ pub(crate) mod tenant_throttling {
     use once_cell::sync::Lazy;
     use utils::shard::TenantShardId;
 
-    pub(crate) struct GlobalAndPerTenantIntCounter {
+    use crate::tenant::{self};
+
+    struct GlobalAndPerTenantIntCounter {
         global: IntCounter,
         per_tenant: IntCounter,
     }
@@ -3596,10 +3583,10 @@ pub(crate) mod tenant_throttling {
     }
 
     pub(crate) struct Metrics<const KIND: usize> {
-        pub(super) count_accounted_start: GlobalAndPerTenantIntCounter,
-        pub(super) count_accounted_finish: GlobalAndPerTenantIntCounter,
-        pub(super) wait_time: GlobalAndPerTenantIntCounter,
-        pub(super) count_throttled: GlobalAndPerTenantIntCounter,
+        count_accounted_start: GlobalAndPerTenantIntCounter,
+        count_accounted_finish: GlobalAndPerTenantIntCounter,
+        wait_time: GlobalAndPerTenantIntCounter,
+        count_throttled: GlobalAndPerTenantIntCounter,
     }
 
     static COUNT_ACCOUNTED_START: Lazy<metrics::IntCounterVec> = Lazy::new(|| {
@@ -3734,6 +3721,26 @@ pub(crate) mod tenant_throttling {
             }
         }
     }
+
+    impl<const KIND: usize> tenant::throttle::Metric for Metrics<KIND> {
+        #[inline(always)]
+        fn accounting_start(&self) {
+            self.count_accounted_start.inc();
+        }
+        #[inline(always)]
+        fn accounting_finish(&self) {
+            self.count_accounted_finish.inc();
+        }
+        #[inline(always)]
+        fn observe_throttling(
+            &self,
+            tenant::throttle::Observation { wait_time }: &tenant::throttle::Observation,
+        ) {
+            let val = u64::try_from(wait_time.as_micros()).unwrap();
+            self.wait_time.inc_by(val);
+            self.count_throttled.inc();
+        }
+    }
 }
 
 pub(crate) mod disk_usage_based_eviction {

pageserver/src/page_service.rs

@@ -67,7 +67,6 @@ use crate::tenant::PageReconstructError;
 use crate::tenant::Timeline;
 use crate::{basebackup, timed_after_cancellation};
 use pageserver_api::key::rel_block_to_key;
-use pageserver_api::models::PageTraceEvent;
 use pageserver_api::reltag::SlruKind;
 use postgres_ffi::pg_constants::DEFAULTTABLESPACE_OID;
 use postgres_ffi::BLCKSZ;
@@ -555,52 +554,37 @@ struct BatchedGetPageRequest {
     timer: SmgrOpTimer,
 }
 
-#[cfg(feature = "testing")]
-struct BatchedTestRequest {
-    req: models::PagestreamTestRequest,
-    timer: SmgrOpTimer,
-}
-
-/// NB: we only hold [`timeline::handle::WeakHandle`] inside this enum,
-/// so that we don't keep the [`Timeline::gate`] open while the batch
-/// is being built up inside the [`spsc_fold`] (pagestream pipelining).
 enum BatchedFeMessage {
     Exists {
         span: Span,
         timer: SmgrOpTimer,
-        shard: timeline::handle::WeakHandle<TenantManagerTypes>,
+        shard: timeline::handle::Handle<TenantManagerTypes>,
         req: models::PagestreamExistsRequest,
     },
     Nblocks {
         span: Span,
         timer: SmgrOpTimer,
-        shard: timeline::handle::WeakHandle<TenantManagerTypes>,
+        shard: timeline::handle::Handle<TenantManagerTypes>,
         req: models::PagestreamNblocksRequest,
     },
     GetPage {
         span: Span,
-        shard: timeline::handle::WeakHandle<TenantManagerTypes>,
+        shard: timeline::handle::Handle<TenantManagerTypes>,
         effective_request_lsn: Lsn,
         pages: smallvec::SmallVec<[BatchedGetPageRequest; 1]>,
     },
     DbSize {
         span: Span,
         timer: SmgrOpTimer,
-        shard: timeline::handle::WeakHandle<TenantManagerTypes>,
+        shard: timeline::handle::Handle<TenantManagerTypes>,
         req: models::PagestreamDbSizeRequest,
     },
     GetSlruSegment {
         span: Span,
         timer: SmgrOpTimer,
-        shard: timeline::handle::WeakHandle<TenantManagerTypes>,
+        shard: timeline::handle::Handle<TenantManagerTypes>,
         req: models::PagestreamGetSlruSegmentRequest,
     },
-    #[cfg(feature = "testing")]
-    Test {
-        span: Span,
-        shard: timeline::handle::WeakHandle<TenantManagerTypes>,
-        requests: Vec<BatchedTestRequest>,
-    },
     RespondError {
         span: Span,
         error: BatchedPageStreamError,
@@ -608,27 +592,43 @@ enum BatchedFeMessage {
 }
 
 impl BatchedFeMessage {
-    fn observe_execution_start(&mut self, at: Instant) {
-        match self {
-            BatchedFeMessage::Exists { timer, .. }
-            | BatchedFeMessage::Nblocks { timer, .. }
-            | BatchedFeMessage::DbSize { timer, .. }
-            | BatchedFeMessage::GetSlruSegment { timer, .. } => {
-                timer.observe_execution_start(at);
+    async fn throttle_and_record_start_processing(
+        &mut self,
+        cancel: &CancellationToken,
+    ) -> Result<(), QueryError> {
+        let (shard, tokens, timers) = match self {
+            BatchedFeMessage::Exists { shard, timer, .. }
+            | BatchedFeMessage::Nblocks { shard, timer, .. }
+            | BatchedFeMessage::DbSize { shard, timer, .. }
+            | BatchedFeMessage::GetSlruSegment { shard, timer, .. } => {
+                (
+                    shard,
+                    // 1 token is probably under-estimating because these
+                    // request handlers typically do several Timeline::get calls.
+                    1,
+                    itertools::Either::Left(std::iter::once(timer)),
+                )
             }
-            BatchedFeMessage::GetPage { pages, .. } => {
-                for page in pages {
-                    page.timer.observe_execution_start(at);
-                }
+            BatchedFeMessage::GetPage { shard, pages, .. } => (
+                shard,
+                pages.len(),
+                itertools::Either::Right(pages.iter_mut().map(|p| &mut p.timer)),
+            ),
+            BatchedFeMessage::RespondError { .. } => return Ok(()),
+        };
+        let throttled = tokio::select! {
+            throttled = shard.pagestream_throttle.throttle(tokens) => { throttled }
+            _ = shard.cancel.cancelled() => {
+                return Err(QueryError::Shutdown);
             }
-            #[cfg(feature = "testing")]
-            BatchedFeMessage::Test { requests, .. } => {
-                for req in requests {
-                    req.timer.observe_execution_start(at);
-                }
+            _ = cancel.cancelled() => {
+                return Err(QueryError::Shutdown);
             }
-            BatchedFeMessage::RespondError { .. } => {}
+        };
+        for timer in timers {
+            timer.observe_throttle_done_execution_starting(&throttled);
         }
+        Ok(())
     }
 }
 
@@ -720,26 +720,6 @@ impl PageServerHandler {
         let neon_fe_msg =
             PagestreamFeMessage::parse(&mut copy_data_bytes.reader(), protocol_version)?;
 
-        // TODO: turn in to async closure once available to avoid repeating received_at
-        async fn record_op_start_and_throttle(
-            shard: &timeline::handle::Handle<TenantManagerTypes>,
-            op: metrics::SmgrQueryType,
-            received_at: Instant,
-        ) -> Result<SmgrOpTimer, QueryError> {
-            // It's important to start the smgr op metric recorder as early as possible
-            // so that the _started counters are incremented before we do
-            // any serious waiting, e.g., for throttle, batching, or actual request handling.
-            let mut timer = shard.query_metrics.start_smgr_op(op, received_at);
-            let now = Instant::now();
-            timer.observe_throttle_start(now);
-            let throttled = tokio::select! {
-                res = shard.pagestream_throttle.throttle(1, now) => res,
-                _ = shard.cancel.cancelled() => return Err(QueryError::Shutdown),
-            };
-            timer.observe_throttle_done(throttled);
-            Ok(timer)
-        }
-
         let batched_msg = match neon_fe_msg {
             PagestreamFeMessage::Exists(req) => {
                 let span = tracing::info_span!(parent: parent_span, "handle_get_rel_exists_request", rel = %req.rel, req_lsn = %req.hdr.request_lsn);
@@ -747,16 +727,13 @@ impl PageServerHandler {
                     .get(tenant_id, timeline_id, ShardSelector::Zero)
                     .instrument(span.clone()) // sets `shard_id` field
                     .await?;
-                let timer = record_op_start_and_throttle(
-                    &shard,
-                    metrics::SmgrQueryType::GetRelExists,
-                    received_at,
-                )
-                .await?;
+                let timer = shard
+                    .query_metrics
+                    .start_smgr_op(metrics::SmgrQueryType::GetRelExists, received_at);
                 BatchedFeMessage::Exists {
                     span,
                     timer,
-                    shard: shard.downgrade(),
+                    shard,
                     req,
                 }
             }
@@ -766,16 +743,13 @@ impl PageServerHandler {
                     .get(tenant_id, timeline_id, ShardSelector::Zero)
                     .instrument(span.clone()) // sets `shard_id` field
                     .await?;
-                let timer = record_op_start_and_throttle(
-                    &shard,
-                    metrics::SmgrQueryType::GetRelSize,
-                    received_at,
-                )
-                .await?;
+                let timer = shard
+                    .query_metrics
+                    .start_smgr_op(metrics::SmgrQueryType::GetRelSize, received_at);
                 BatchedFeMessage::Nblocks {
                     span,
                     timer,
-                    shard: shard.downgrade(),
+                    shard,
                     req,
                 }
             }
@@ -785,16 +759,13 @@ impl PageServerHandler {
                     .get(tenant_id, timeline_id, ShardSelector::Zero)
                     .instrument(span.clone()) // sets `shard_id` field
                     .await?;
-                let timer = record_op_start_and_throttle(
-                    &shard,
-                    metrics::SmgrQueryType::GetDbSize,
-                    received_at,
-                )
-                .await?;
+                let timer = shard
+                    .query_metrics
+                    .start_smgr_op(metrics::SmgrQueryType::GetDbSize, received_at);
                 BatchedFeMessage::DbSize {
                     span,
                     timer,
-                    shard: shard.downgrade(),
+                    shard,
                     req,
                 }
             }
@@ -804,16 +775,13 @@ impl PageServerHandler {
                     .get(tenant_id, timeline_id, ShardSelector::Zero)
                     .instrument(span.clone()) // sets `shard_id` field
                     .await?;
-                let timer = record_op_start_and_throttle(
-                    &shard,
-                    metrics::SmgrQueryType::GetSlruSegment,
-                    received_at,
-                )
-                .await?;
+                let timer = shard
+                    .query_metrics
+                    .start_smgr_op(metrics::SmgrQueryType::GetSlruSegment, received_at);
                 BatchedFeMessage::GetSlruSegment {
                     span,
                     timer,
-                    shard: shard.downgrade(),
+                    shard,
                     req,
                 }
             }
@@ -858,14 +826,13 @@ impl PageServerHandler {
                     }
                 };
 
-                let timer = record_op_start_and_throttle(
-                    &shard,
-                    metrics::SmgrQueryType::GetPageAtLsn,
-                    received_at,
-                )
-                .await?;
+                // It's important to start the timer before waiting for the LSN
+                // so that the _started counters are incremented before we do
+                // any serious waiting, e.g., for LSNs.
+                let timer = shard
+                    .query_metrics
+                    .start_smgr_op(metrics::SmgrQueryType::GetPageAtLsn, received_at);
 
-                // We're holding the Handle
                 let effective_request_lsn = match Self::wait_or_get_last_lsn(
                     &shard,
                     req.hdr.request_lsn,
@@ -883,27 +850,11 @@ impl PageServerHandler {
                 };
                 BatchedFeMessage::GetPage {
                     span,
-                    shard: shard.downgrade(),
+                    shard,
                     effective_request_lsn,
                     pages: smallvec::smallvec![BatchedGetPageRequest { req, timer }],
                 }
             }
-            #[cfg(feature = "testing")]
-            PagestreamFeMessage::Test(req) => {
-                let span = tracing::info_span!(parent: parent_span, "handle_test_request");
-                let shard = timeline_handles
-                    .get(tenant_id, timeline_id, ShardSelector::Zero)
-                    .instrument(span.clone()) // sets `shard_id` field
-                    .await?;
-                let timer =
-                    record_op_start_and_throttle(&shard, metrics::SmgrQueryType::Test, received_at)
-                        .await?;
-                BatchedFeMessage::Test {
-                    span,
-                    shard: shard.downgrade(),
-                    requests: vec![BatchedTestRequest { req, timer }],
-                }
-            }
         };
         Ok(Some(batched_msg))
     }
@@ -945,7 +896,9 @@ impl PageServerHandler {
                     assert_eq!(accum_pages.len(), max_batch_size.get());
                     return false;
                 }
-                if !accum_shard.is_same_handle_as(&this_shard) {
+                if (accum_shard.tenant_shard_id, accum_shard.timeline_id)
+                    != (this_shard.tenant_shard_id, this_shard.timeline_id)
+                {
                     trace!(%accum_lsn, %this_lsn, "stopping batching because timeline object mismatch");
                     // TODO: we _could_ batch & execute each shard seperately (and in parallel).
                     // But the current logic for keeping responses in order does not support that.
@@ -964,44 +917,6 @@ impl PageServerHandler {
                 accum_pages.extend(this_pages);
                 Ok(())
             }
-            #[cfg(feature = "testing")]
-            (
-                Ok(BatchedFeMessage::Test {
-                    shard: accum_shard,
-                    requests: accum_requests,
-                    ..
-                }),
-                BatchedFeMessage::Test {
-                    shard: this_shard,
-                    requests: this_requests,
-                    ..
-                },
-            ) if (|| {
-                assert!(this_requests.len() == 1);
-                if accum_requests.len() >= max_batch_size.get() {
-                    trace!(%max_batch_size, "stopping batching because of batch size");
-                    assert_eq!(accum_requests.len(), max_batch_size.get());
-                    return false;
-                }
-                if !accum_shard.is_same_handle_as(&this_shard) {
-                    trace!("stopping batching because timeline object mismatch");
-                    // TODO: we _could_ batch & execute each shard seperately (and in parallel).
-                    // But the current logic for keeping responses in order does not support that.
-                    return false;
-                }
-                let this_batch_key = this_requests[0].req.batch_key;
-                let accum_batch_key = accum_requests[0].req.batch_key;
-                if this_requests[0].req.batch_key != accum_requests[0].req.batch_key {
-                    trace!(%accum_batch_key, %this_batch_key, "stopping batching because batch key changed");
-                    return false;
-                }
-                true
-            })() =>
-            {
-                // ok to batch
-                accum_requests.extend(this_requests);
-                Ok(())
-            }
             // something batched already but this message is unbatchable
             (_, this_msg) => {
                 // by default, don't continue batching
@@ -1022,13 +937,6 @@ impl PageServerHandler {
     where
         IO: AsyncRead + AsyncWrite + Send + Sync + Unpin,
     {
-        let started_at = Instant::now();
-        let batch = {
-            let mut batch = batch;
-            batch.observe_execution_start(started_at);
-            batch
-        };
-
         // invoke handler function
         let (handler_results, span): (
             Vec<Result<(PagestreamBeMessage, SmgrOpTimer), BatchedPageStreamError>>,
@@ -1043,7 +951,7 @@ impl PageServerHandler {
                 fail::fail_point!("ps::handle-pagerequest-message::exists");
                 (
                     vec![self
-                        .handle_get_rel_exists_request(&*shard.upgrade()?, &req, ctx)
+                        .handle_get_rel_exists_request(&shard, &req, ctx)
                         .instrument(span.clone())
                         .await
                         .map(|msg| (msg, timer))
@@ -1060,7 +968,7 @@ impl PageServerHandler {
                 fail::fail_point!("ps::handle-pagerequest-message::nblocks");
                 (
                     vec![self
-                        .handle_get_nblocks_request(&*shard.upgrade()?, &req, ctx)
+                        .handle_get_nblocks_request(&shard, &req, ctx)
                         .instrument(span.clone())
                         .await
                         .map(|msg| (msg, timer))
@@ -1081,7 +989,7 @@ impl PageServerHandler {
                         trace!(npages, "handling getpage request");
                         let res = self
                             .handle_get_page_at_lsn_request_batched(
-                                &*shard.upgrade()?,
+                                &shard,
                                 effective_request_lsn,
                                 pages,
                                 ctx,
@@ -1103,7 +1011,7 @@ impl PageServerHandler {
                 fail::fail_point!("ps::handle-pagerequest-message::dbsize");
                 (
                     vec![self
-                        .handle_db_size_request(&*shard.upgrade()?, &req, ctx)
+                        .handle_db_size_request(&shard, &req, ctx)
                         .instrument(span.clone())
                         .await
                         .map(|msg| (msg, timer))
@@ -1120,7 +1028,7 @@ impl PageServerHandler {
                 fail::fail_point!("ps::handle-pagerequest-message::slrusegment");
                 (
                     vec![self
-                        .handle_get_slru_segment_request(&*shard.upgrade()?, &req, ctx)
+                        .handle_get_slru_segment_request(&shard, &req, ctx)
                         .instrument(span.clone())
                         .await
                         .map(|msg| (msg, timer))
@@ -1128,27 +1036,6 @@ impl PageServerHandler {
                     span,
                 )
             }
-            #[cfg(feature = "testing")]
-            BatchedFeMessage::Test {
-                span,
-                shard,
-                requests,
-            } => {
-                fail::fail_point!("ps::handle-pagerequest-message::test");
-                (
-                    {
-                        let npages = requests.len();
-                        trace!(npages, "handling getpage request");
-                        let res = self
-                            .handle_test_request_batch(&*shard.upgrade()?, requests, ctx)
-                            .instrument(span.clone())
-                            .await;
-                        assert_eq!(res.len(), npages);
-                        res
-                    },
-                    span,
-                )
-            }
             BatchedFeMessage::RespondError { span, error } => {
                 // We've already decided to respond with an error, so we don't need to
                 // call the handler.
@@ -1216,11 +1103,8 @@ impl PageServerHandler {
             // The timer's underlying metric is used for a storage-internal latency SLO and
             // we don't want to include latency in it that we can't control.
             // And as pointed out above, in this case, we don't control the time that flush will take.
-            let flushing_timer = timer.map(|mut timer| {
-                timer
-                    .observe_execution_end_flush_start(Instant::now())
-                    .expect("we are the first caller")
-            });
+            let flushing_timer =
+                timer.map(|timer| timer.observe_smgr_op_completion_and_start_flushing());
 
             // what we want to do
             let flush_fut = pgb_writer.flush();
@@ -1374,7 +1258,7 @@ impl PageServerHandler {
                 Ok(msg) => msg,
                 Err(e) => break e,
             };
-            let msg = match msg {
+            let mut msg = match msg {
                 Some(msg) => msg,
                 None => {
                     debug!("pagestream subprotocol end observed");
@@ -1382,6 +1266,10 @@ impl PageServerHandler {
                 }
             };
 
+            if let Err(cancelled) = msg.throttle_and_record_start_processing(&self.cancel).await {
+                break cancelled;
+            }
+
             let err = self
                 .pagesteam_handle_batched_message(pgb_writer, msg, &cancel, protocol_version, ctx)
                 .await;
@@ -1541,12 +1429,15 @@ impl PageServerHandler {
                             return Ok(());
                         }
                     };
-                    let batch = match batch {
+                    let mut batch = match batch {
                         Ok(batch) => batch,
                         Err(e) => {
                             return Err(e);
                         }
                     };
+                    batch
+                        .throttle_and_record_start_processing(&self.cancel)
+                        .await?;
                     self.pagesteam_handle_batched_message(
                         pgb_writer,
                         batch,
@@ -1814,20 +1705,6 @@ impl PageServerHandler {
             .query_metrics
             .observe_getpage_batch_start(requests.len());
 
-        // If a page trace is running, submit an event for this request.
-        if let Some(page_trace) = timeline.page_trace.load().as_ref() {
-            let time = SystemTime::now();
-            for batch in &requests {
-                let key = rel_block_to_key(batch.req.rel, batch.req.blkno).to_compact();
-                // Ignore error (trace buffer may be full or tracer may have disconnected).
-                _ = page_trace.try_send(PageTraceEvent {
-                    key,
-                    effective_lsn,
-                    time,
-                });
-            }
-        }
-
         let results = timeline
             .get_rel_page_at_lsn_batched(
                 requests.iter().map(|p| (&p.req.rel, &p.req.blkno)),
@@ -1886,51 +1763,6 @@ impl PageServerHandler {
         ))
     }
 
-    // NB: this impl mimics what we do for batched getpage requests.
-    #[cfg(feature = "testing")]
-    #[instrument(skip_all, fields(shard_id))]
-    async fn handle_test_request_batch(
-        &mut self,
-        timeline: &Timeline,
-        requests: Vec<BatchedTestRequest>,
-        _ctx: &RequestContext,
-    ) -> Vec<Result<(PagestreamBeMessage, SmgrOpTimer), BatchedPageStreamError>> {
-        // real requests would do something with the timeline
-        let mut results = Vec::with_capacity(requests.len());
-        for _req in requests.iter() {
-            tokio::task::yield_now().await;
-
-            results.push({
-                if timeline.cancel.is_cancelled() {
-                    Err(PageReconstructError::Cancelled)
-                } else {
-                    Ok(())
-                }
-            });
-        }
-
-        // TODO: avoid creating the new Vec here
-        Vec::from_iter(
-            requests
-                .into_iter()
-                .zip(results.into_iter())
-                .map(|(req, res)| {
-                    res.map(|()| {
-                        (
-                            PagestreamBeMessage::Test(models::PagestreamTestResponse {
-                                req: req.req.clone(),
-                            }),
-                            req.timer,
-                        )
-                    })
-                    .map_err(|e| BatchedPageStreamError {
-                        err: PageStreamError::from(e),
-                        req: req.req.hdr,
-                    })
-                }),
-        )
-    }
-
     /// Note on "fullbackup":
     /// Full basebackups should only be used for debugging purposes.
     /// Originally, it was introduced to enable breaking storage format changes,
@@ -2546,14 +2378,6 @@ impl From<GetActiveTimelineError> for QueryError {
     }
 }
 
-impl From<crate::tenant::timeline::handle::HandleUpgradeError> for QueryError {
-    fn from(e: crate::tenant::timeline::handle::HandleUpgradeError) -> Self {
-        match e {
-            crate::tenant::timeline::handle::HandleUpgradeError::ShutDown => QueryError::Shutdown,
-        }
-    }
-}
-
 fn set_tracing_field_shard_id(timeline: &Timeline) {
     debug_assert_current_span_has_tenant_and_timeline_id_no_shard_id();
     tracing::Span::current().record(

pageserver/src/tenant.rs

@@ -365,9 +365,8 @@ pub struct Tenant {
 
     /// Throttle applied at the top of [`Timeline::get`].
     /// All [`Tenant::timelines`] of a given [`Tenant`] instance share the same [`throttle::Throttle`] instance.
-    pub(crate) pagestream_throttle: Arc<throttle::Throttle>,
-
-    pub(crate) pagestream_throttle_metrics: Arc<crate::metrics::tenant_throttling::Pagestream>,
+    pub(crate) pagestream_throttle:
+        Arc<throttle::Throttle<crate::metrics::tenant_throttling::Pagestream>>,
 
     /// An ongoing timeline detach concurrency limiter.
     ///
@@ -1688,7 +1687,6 @@ impl Tenant {
                     TimelineResources {
                         remote_client,
                         pagestream_throttle: self.pagestream_throttle.clone(),
-                        pagestream_throttle_metrics: self.pagestream_throttle_metrics.clone(),
                         l0_flush_global_state: self.l0_flush_global_state.clone(),
                     },
                     LoadTimelineCause::Attach,
@@ -3994,9 +3992,6 @@ impl Tenant {
         Ok(timeline)
     }
 
-    /// [`Tenant::shutdown`] must be called before dropping the returned [`Tenant`] object
-    /// to ensure proper cleanup of background tasks and metrics.
-    //
     // Allow too_many_arguments because a constructor's argument list naturally grows with the
     // number of attributes in the struct: breaking these out into a builder wouldn't be helpful.
     #[allow(clippy::too_many_arguments)]
@@ -4105,10 +4100,8 @@ impl Tenant {
             gate: Gate::default(),
             pagestream_throttle: Arc::new(throttle::Throttle::new(
                 Tenant::get_pagestream_throttle_config(conf, &attached_conf.tenant_conf),
+                crate::metrics::tenant_throttling::Metrics::new(&tenant_shard_id),
             )),
-            pagestream_throttle_metrics: Arc::new(
-                crate::metrics::tenant_throttling::Pagestream::new(&tenant_shard_id),
-            ),
             tenant_conf: Arc::new(ArcSwap::from_pointee(attached_conf)),
             ongoing_timeline_detach: std::sync::Mutex::default(),
             gc_block: Default::default(),
@@ -5015,7 +5008,6 @@ impl Tenant {
         TimelineResources {
             remote_client: self.build_timeline_remote_client(timeline_id),
             pagestream_throttle: self.pagestream_throttle.clone(),
-            pagestream_throttle_metrics: self.pagestream_throttle_metrics.clone(),
             l0_flush_global_state: self.l0_flush_global_state.clone(),
         }
     }

pageserver/src/tenant/metadata.rs

@@ -320,6 +320,7 @@ impl TimelineMetadata {
 
     // Checksums make it awkward to build a valid instance by hand.  This helper
     // provides a TimelineMetadata with a valid checksum in its header.
+    #[cfg(test)]
     pub fn example() -> Self {
         let instance = Self::new(
             "0/16960E8".parse::<Lsn>().unwrap(),

pageserver/src/tenant/remote_timeline_client.rs

@@ -63,18 +63,22 @@
 //! The contract between client and its user is that the user is responsible of
 //! scheduling operations in an order that keeps the remote consistent as
 //! described above.
-//!
 //! From the user's perspective, the operations are executed sequentially.
 //! Internally, the client knows which operations can be performed in parallel,
 //! and which operations act like a "barrier" that require preceding operations
 //! to finish. The calling code just needs to call the schedule-functions in the
 //! correct order, and the client will parallelize the operations in a way that
-//! is safe. For more details, see `UploadOp::can_bypass`.
+//! is safe.
+//!
+//! The caller should be careful with deletion, though. They should not delete
+//! local files that have been scheduled for upload but not yet finished uploading.
+//! Otherwise the upload will fail. To wait for an upload to finish, use
+//! the 'wait_completion' function (more on that later.)
 //!
 //! All of this relies on the following invariants:
 //!
 //! - We rely on read-after write consistency in the remote storage.
-//! - Layer files are immutable.
+//! - Layer files are immutable
 //!
 //! NB: Pageserver assumes that it has exclusive write access to the tenant in remote
 //! storage. Different tenants can be attached to different pageservers, but if the
@@ -382,12 +386,6 @@ pub(crate) struct RemoteTimelineClient {
     cancel: CancellationToken,
 }
 
-impl Drop for RemoteTimelineClient {
-    fn drop(&mut self) {
-        debug!("dropping RemoteTimelineClient");
-    }
-}
-
 impl RemoteTimelineClient {
     ///
     /// Create a remote storage client for given timeline
@@ -431,16 +429,8 @@ impl RemoteTimelineClient {
     /// an index file upload, i.e., it's not empty.
     /// The given `index_part` must be the one on the remote.
     pub fn init_upload_queue(&self, index_part: &IndexPart) -> anyhow::Result<()> {
-        // Set the maximum number of inprogress tasks to the remote storage concurrency. There's
-        // certainly no point in starting more upload tasks than this.
-        let inprogress_limit = self
-            .conf
-            .remote_storage_config
-            .as_ref()
-            .and_then(|r| r.concurrency_limit())
-            .unwrap_or(0);
         let mut upload_queue = self.upload_queue.lock().unwrap();
-        upload_queue.initialize_with_current_remote_index_part(index_part, inprogress_limit)?;
+        upload_queue.initialize_with_current_remote_index_part(index_part)?;
         self.update_remote_physical_size_gauge(Some(index_part));
         info!(
             "initialized upload queue from remote index with {} layer files",
@@ -455,16 +445,8 @@ impl RemoteTimelineClient {
         &self,
         local_metadata: &TimelineMetadata,
     ) -> anyhow::Result<()> {
-        // Set the maximum number of inprogress tasks to the remote storage concurrency. There's
-        // certainly no point in starting more upload tasks than this.
-        let inprogress_limit = self
-            .conf
-            .remote_storage_config
-            .as_ref()
-            .and_then(|r| r.concurrency_limit())
-            .unwrap_or(0);
         let mut upload_queue = self.upload_queue.lock().unwrap();
-        upload_queue.initialize_empty_remote(local_metadata, inprogress_limit)?;
+        upload_queue.initialize_empty_remote(local_metadata)?;
         self.update_remote_physical_size_gauge(None);
         info!("initialized upload queue as empty");
         Ok(())
@@ -480,15 +462,9 @@ impl RemoteTimelineClient {
         let deleted_at = index_part.deleted_at.ok_or(anyhow::anyhow!(
             "bug: it is responsibility of the caller to provide index part from MaybeDeletedIndexPart::Deleted"
         ))?;
-        let inprogress_limit = self
-            .conf
-            .remote_storage_config
-            .as_ref()
-            .and_then(|r| r.concurrency_limit())
-            .unwrap_or(0);
 
         let mut upload_queue = self.upload_queue.lock().unwrap();
-        upload_queue.initialize_with_current_remote_index_part(index_part, inprogress_limit)?;
+        upload_queue.initialize_with_current_remote_index_part(index_part)?;
         self.update_remote_physical_size_gauge(Some(index_part));
         self.stop_impl(&mut upload_queue);
 
@@ -1879,17 +1855,57 @@ impl RemoteTimelineClient {
         Ok(())
     }
 
+    ///
     /// Pick next tasks from the queue, and start as many of them as possible without violating
     /// the ordering constraints.
     ///
-    /// TODO: consider limiting the number of in-progress tasks, beyond what remote_storage does.
-    /// This can launch an unbounded number of queued tasks. `UploadQueue::next_ready()` also has
-    /// worst-case quadratic cost in the number of tasks, and may struggle beyond 10,000 tasks.
+    /// The caller needs to already hold the `upload_queue` lock.
     fn launch_queued_tasks(self: &Arc<Self>, upload_queue: &mut UploadQueueInitialized) {
-        while let Some((mut next_op, coalesced_ops)) = upload_queue.next_ready() {
-            debug!("starting op: {next_op}");
+        while let Some(next_op) = upload_queue.queued_operations.front() {
+            // Can we run this task now?
+            let can_run_now = match next_op {
+                UploadOp::UploadLayer(..) => {
+                    // Can always be scheduled.
+                    true
+                }
+                UploadOp::UploadMetadata { .. } => {
+                    // These can only be performed after all the preceding operations
+                    // have finished.
+                    upload_queue.inprogress_tasks.is_empty()
+                }
+                UploadOp::Delete(..) => {
+                    // Wait for preceding uploads to finish. Concurrent deletions are OK, though.
+                    upload_queue.num_inprogress_deletions == upload_queue.inprogress_tasks.len()
+                }
 
-            // Prepare upload.
+                UploadOp::Barrier(_) | UploadOp::Shutdown => {
+                    upload_queue.inprogress_tasks.is_empty()
+                }
+            };
+
+            // If we cannot launch this task, don't look any further.
+            //
+            // In some cases, we could let some non-frontmost tasks to "jump the queue" and launch
+            // them now, but we don't try to do that currently.  For example, if the frontmost task
+            // is an index-file upload that cannot proceed until preceding uploads have finished, we
+            // could still start layer uploads that were scheduled later.
+            if !can_run_now {
+                break;
+            }
+
+            if let UploadOp::Shutdown = next_op {
+                // leave the op in the queue but do not start more tasks; it will be dropped when
+                // the stop is called.
+                upload_queue.shutdown_ready.close();
+                break;
+            }
+
+            // We can launch this task. Remove it from the queue first.
+            let mut next_op = upload_queue.queued_operations.pop_front().unwrap();
+
+            debug!("starting op: {}", next_op);
+
+            // Update the counters and prepare
             match &mut next_op {
                 UploadOp::UploadLayer(layer, meta, mode) => {
                     if upload_queue
@@ -1900,14 +1916,18 @@ impl RemoteTimelineClient {
                     } else {
                         *mode = Some(OpType::MayReorder)
                     }
+                    upload_queue.num_inprogress_layer_uploads += 1;
+                }
+                UploadOp::UploadMetadata { .. } => {
+                    upload_queue.num_inprogress_metadata_uploads += 1;
                 }
-                UploadOp::UploadMetadata { .. } => {}
                 UploadOp::Delete(Delete { layers }) => {
                     for (name, meta) in layers {
                         upload_queue
                             .recently_deleted
                             .insert((name.clone(), meta.generation));
                     }
+                    upload_queue.num_inprogress_deletions += 1;
                 }
                 UploadOp::Barrier(sender) => {
                     sender.send_replace(());
@@ -1924,7 +1944,6 @@ impl RemoteTimelineClient {
             let task = Arc::new(UploadTask {
                 task_id: upload_task_id,
                 op: next_op,
-                coalesced_ops,
                 retries: AtomicU32::new(0),
             });
             upload_queue
@@ -2008,8 +2027,6 @@ impl RemoteTimelineClient {
 
             let upload_result: anyhow::Result<()> = match &task.op {
                 UploadOp::UploadLayer(ref layer, ref layer_metadata, mode) => {
-                    // TODO: check if this mechanism can be removed now that can_bypass() performs
-                    // conflict checks during scheduling.
                     if let Some(OpType::FlushDeletion) = mode {
                         if self.config.read().unwrap().block_deletions {
                             // Of course, this is not efficient... but usually the queue should be empty.
@@ -2232,8 +2249,13 @@ impl RemoteTimelineClient {
             upload_queue.inprogress_tasks.remove(&task.task_id);
 
             let lsn_update = match task.op {
-                UploadOp::UploadLayer(_, _, _) => None,
+                UploadOp::UploadLayer(_, _, _) => {
+                    upload_queue.num_inprogress_layer_uploads -= 1;
+                    None
+                }
                 UploadOp::UploadMetadata { ref uploaded } => {
+                    upload_queue.num_inprogress_metadata_uploads -= 1;
+
                     // the task id is reused as a monotonicity check for storing the "clean"
                     // IndexPart.
                     let last_updater = upload_queue.clean.1;
@@ -2267,7 +2289,10 @@ impl RemoteTimelineClient {
                         None
                     }
                 }
-                UploadOp::Delete(_) => None,
+                UploadOp::Delete(_) => {
+                    upload_queue.num_inprogress_deletions -= 1;
+                    None
+                }
                 UploadOp::Barrier(..) | UploadOp::Shutdown => unreachable!(),
             };
 
@@ -2292,9 +2317,6 @@ impl RemoteTimelineClient {
         }
 
         self.metric_end(&task.op);
-        for coalesced_op in &task.coalesced_ops {
-            self.metric_end(coalesced_op);
-        }
     }
 
     fn metric_impl(
@@ -2387,7 +2409,6 @@ impl RemoteTimelineClient {
                     // but for this use case it doesnt really makes sense to bring unsafe code only for this usage point.
                     // Deletion is not really perf sensitive so there shouldnt be any problems with cloning a fraction of it.
                     let upload_queue_for_deletion = UploadQueueInitialized {
-                        inprogress_limit: initialized.inprogress_limit,
                         task_counter: 0,
                         dirty: initialized.dirty.clone(),
                         clean: initialized.clean.clone(),
@@ -2395,6 +2416,9 @@ impl RemoteTimelineClient {
                         visible_remote_consistent_lsn: initialized
                             .visible_remote_consistent_lsn
                             .clone(),
+                        num_inprogress_layer_uploads: 0,
+                        num_inprogress_metadata_uploads: 0,
+                        num_inprogress_deletions: 0,
                         inprogress_tasks: HashMap::default(),
                         queued_operations: VecDeque::default(),
                         #[cfg(feature = "testing")]
@@ -2421,6 +2445,14 @@ impl RemoteTimelineClient {
                     }
                 };
 
+                // consistency check
+                assert_eq!(
+                    qi.num_inprogress_layer_uploads
+                        + qi.num_inprogress_metadata_uploads
+                        + qi.num_inprogress_deletions,
+                    qi.inprogress_tasks.len()
+                );
+
                 // We don't need to do anything here for in-progress tasks. They will finish
                 // on their own, decrement the unfinished-task counter themselves, and observe
                 // that the queue is Stopped.
@@ -2867,8 +2899,8 @@ mod tests {
             let mut guard = client.upload_queue.lock().unwrap();
             let upload_queue = guard.initialized_mut().unwrap();
             assert!(upload_queue.queued_operations.is_empty());
-            assert_eq!(upload_queue.inprogress_tasks.len(), 2);
-            assert_eq!(upload_queue.num_inprogress_layer_uploads(), 2);
+            assert!(upload_queue.inprogress_tasks.len() == 2);
+            assert!(upload_queue.num_inprogress_layer_uploads == 2);
 
             // also check that `latest_file_changes` was updated
             assert!(upload_queue.latest_files_changes_since_metadata_upload_scheduled == 2);
@@ -2938,8 +2970,8 @@ mod tests {
             // Deletion schedules upload of the index file, and the file deletion itself
             assert_eq!(upload_queue.queued_operations.len(), 2);
             assert_eq!(upload_queue.inprogress_tasks.len(), 1);
-            assert_eq!(upload_queue.num_inprogress_layer_uploads(), 1);
-            assert_eq!(upload_queue.num_inprogress_deletions(), 0);
+            assert_eq!(upload_queue.num_inprogress_layer_uploads, 1);
+            assert_eq!(upload_queue.num_inprogress_deletions, 0);
             assert_eq!(
                 upload_queue.latest_files_changes_since_metadata_upload_scheduled,
                 0

pageserver/src/tenant/remote_timeline_client/index.rs

@@ -104,7 +104,7 @@ impl IndexPart {
 
     pub const FILE_NAME: &'static str = "index_part.json";
 
-    pub fn empty(metadata: TimelineMetadata) -> Self {
+    pub(crate) fn empty(metadata: TimelineMetadata) -> Self {
         IndexPart {
             version: Self::LATEST_VERSION,
             layer_metadata: Default::default(),

pageserver/src/tenant/storage_layer/layer.rs

@@ -1812,7 +1812,7 @@ enum LayerKind {
 
 /// Guard for forcing a layer be resident while it exists.
 #[derive(Clone)]
-pub struct ResidentLayer {
+pub(crate) struct ResidentLayer {
     owner: Layer,
     downloaded: Arc<DownloadedLayer>,
 }

pageserver/src/tenant/throttle.rs

@@ -3,7 +3,7 @@ use std::{
         atomic::{AtomicU64, Ordering},
         Arc,
     },
-    time::Instant,
+    time::{Duration, Instant},
 };
 
 use arc_swap::ArcSwap;
@@ -16,8 +16,9 @@ use utils::leaky_bucket::{LeakyBucketConfig, RateLimiter};
 /// To share a throttle among multiple entities, wrap it in an [`Arc`].
 ///
 /// The intial use case for this is tenant-wide throttling of getpage@lsn requests.
-pub struct Throttle {
+pub struct Throttle<M: Metric> {
     inner: ArcSwap<Inner>,
+    metric: M,
     /// will be turned into [`Stats::count_accounted_start`]
     count_accounted_start: AtomicU64,
     /// will be turned into [`Stats::count_accounted_finish`]
@@ -35,6 +36,15 @@ pub struct Inner {
 
 pub type Config = pageserver_api::models::ThrottleConfig;
 
+pub struct Observation {
+    pub wait_time: Duration,
+}
+pub trait Metric {
+    fn accounting_start(&self);
+    fn accounting_finish(&self);
+    fn observe_throttling(&self, observation: &Observation);
+}
+
 /// See [`Throttle::reset_stats`].
 pub struct Stats {
     /// Number of requests that started [`Throttle::throttle`] calls.
@@ -49,14 +59,18 @@ pub struct Stats {
 }
 
 pub enum ThrottleResult {
-    NotThrottled { end: Instant },
-    Throttled { end: Instant },
+    NotThrottled { start: Instant },
+    Throttled { start: Instant, end: Instant },
 }
 
-impl Throttle {
-    pub fn new(config: Config) -> Self {
+impl<M> Throttle<M>
+where
+    M: Metric,
+{
+    pub fn new(config: Config, metric: M) -> Self {
         Self {
             inner: ArcSwap::new(Arc::new(Self::new_inner(config))),
+            metric,
             count_accounted_start: AtomicU64::new(0),
             count_accounted_finish: AtomicU64::new(0),
             count_throttled: AtomicU64::new(0),
@@ -113,27 +127,32 @@ impl Throttle {
         self.inner.load().rate_limiter.steady_rps()
     }
 
-    /// `start` must be [`Instant::now`] or earlier.
-    pub async fn throttle(&self, key_count: usize, start: Instant) -> ThrottleResult {
+    pub async fn throttle(&self, key_count: usize) -> ThrottleResult {
         let inner = self.inner.load_full(); // clones the `Inner` Arc
 
+        let start = std::time::Instant::now();
+
         if !inner.enabled {
-            return ThrottleResult::NotThrottled { end: start };
+            return ThrottleResult::NotThrottled { start };
         }
 
+        self.metric.accounting_start();
         self.count_accounted_start.fetch_add(1, Ordering::Relaxed);
         let did_throttle = inner.rate_limiter.acquire(key_count).await;
         self.count_accounted_finish.fetch_add(1, Ordering::Relaxed);
+        self.metric.accounting_finish();
 
         if did_throttle {
             self.count_throttled.fetch_add(1, Ordering::Relaxed);
-            let end = Instant::now();
-            let wait_time = end - start;
+            let now = Instant::now();
+            let wait_time = now - start;
             self.sum_throttled_usecs
                 .fetch_add(wait_time.as_micros() as u64, Ordering::Relaxed);
-            ThrottleResult::Throttled { end }
+            let observation = Observation { wait_time };
+            self.metric.observe_throttling(&observation);
+            ThrottleResult::Throttled { start, end: now }
         } else {
-            ThrottleResult::NotThrottled { end: start }
+            ThrottleResult::NotThrottled { start }
         }
     }
 }

pageserver/src/tenant/timeline.rs

@@ -14,7 +14,7 @@ pub mod uninit;
 mod walreceiver;
 
 use anyhow::{anyhow, bail, ensure, Context, Result};
-use arc_swap::{ArcSwap, ArcSwapOption};
+use arc_swap::ArcSwap;
 use bytes::Bytes;
 use camino::Utf8Path;
 use chrono::{DateTime, Utc};
@@ -23,7 +23,6 @@ use fail::fail_point;
 use handle::ShardTimelineId;
 use offload::OffloadError;
 use once_cell::sync::Lazy;
-use pageserver_api::models::PageTraceEvent;
 use pageserver_api::{
     config::tenant_conf_defaults::DEFAULT_COMPACTION_THRESHOLD,
     key::{
@@ -43,7 +42,6 @@ use rand::Rng;
 use remote_storage::DownloadError;
 use serde_with::serde_as;
 use storage_broker::BrokerClientChannel;
-use tokio::sync::mpsc::Sender;
 use tokio::{
     runtime::Handle,
     sync::{oneshot, watch},
@@ -51,9 +49,7 @@ use tokio::{
 use tokio_util::sync::CancellationToken;
 use tracing::*;
 use utils::{
-    fs_ext,
-    guard_arc_swap::GuardArcSwap,
-    pausable_failpoint,
+    fs_ext, pausable_failpoint,
     postgres_client::PostgresClientProtocol,
     sync::gate::{Gate, GateGuard},
 };
@@ -76,7 +72,6 @@ use std::{pin::pin, sync::OnceLock};
 
 use crate::{
     aux_file::AuxFileSizeEstimator,
-    page_service::TenantManagerTypes,
     tenant::{
         config::AttachmentMode,
         layer_map::{LayerMap, SearchResult},
@@ -213,8 +208,8 @@ fn drop_wlock<T>(rlock: tokio::sync::RwLockWriteGuard<'_, T>) {
 /// The outward-facing resources required to build a Timeline
 pub struct TimelineResources {
     pub remote_client: RemoteTimelineClient,
-    pub pagestream_throttle: Arc<crate::tenant::throttle::Throttle>,
-    pub pagestream_throttle_metrics: Arc<crate::metrics::tenant_throttling::Pagestream>,
+    pub pagestream_throttle:
+        Arc<crate::tenant::throttle::Throttle<crate::metrics::tenant_throttling::Pagestream>>,
     pub l0_flush_global_state: l0_flush::L0FlushGlobalState,
 }
 
@@ -356,8 +351,8 @@ pub struct Timeline {
     // though let's keep them both for better error visibility.
     pub initdb_lsn: Lsn,
 
-    /// The repartitioning result. Allows a single writer and multiple readers.
-    pub(crate) partitioning: GuardArcSwap<((KeyPartitioning, SparseKeyPartitioning), Lsn)>,
+    /// When did we last calculate the partitioning? Make it pub to test cases.
+    pub(super) partitioning: tokio::sync::Mutex<((KeyPartitioning, SparseKeyPartitioning), Lsn)>,
 
     /// Configuration: how often should the partitioning be recalculated.
     repartition_threshold: u64,
@@ -417,7 +412,8 @@ pub struct Timeline {
     gc_lock: tokio::sync::Mutex<()>,
 
     /// Cloned from [`super::Tenant::pagestream_throttle`] on construction.
-    pub(crate) pagestream_throttle: Arc<crate::tenant::throttle::Throttle>,
+    pub(crate) pagestream_throttle:
+        Arc<crate::tenant::throttle::Throttle<crate::metrics::tenant_throttling::Pagestream>>,
 
     /// Size estimator for aux file v2
     pub(crate) aux_file_size_estimator: AuxFileSizeEstimator,
@@ -432,15 +428,12 @@ pub struct Timeline {
 
     pub(crate) l0_flush_global_state: L0FlushGlobalState,
 
-    pub(crate) handles: handle::PerTimelineState<TenantManagerTypes>,
+    pub(crate) handles: handle::PerTimelineState<crate::page_service::TenantManagerTypes>,
 
     pub(crate) attach_wal_lag_cooldown: Arc<OnceLock<WalLagCooldown>>,
 
     /// Cf. [`crate::tenant::CreateTimelineIdempotency`].
     pub(crate) create_idempotency: crate::tenant::CreateTimelineIdempotency,
-
-    /// If Some, collects GetPage metadata for an ongoing PageTrace.
-    pub(crate) page_trace: ArcSwapOption<Sender<PageTraceEvent>>,
 }
 
 pub type TimelineDeleteProgress = Arc<tokio::sync::Mutex<DeleteTimelineFlow>>;
@@ -2317,7 +2310,6 @@ impl Timeline {
                 query_metrics: crate::metrics::SmgrQueryTimePerTimeline::new(
                     &tenant_shard_id,
                     &timeline_id,
-                    resources.pagestream_throttle_metrics,
                 ),
 
                 directory_metrics: array::from_fn(|_| AtomicU64::new(0)),
@@ -2343,8 +2335,7 @@ impl Timeline {
                     // initial logical size is 0.
                     LogicalSize::empty_initial()
                 },
-
-                partitioning: GuardArcSwap::new((
+                partitioning: tokio::sync::Mutex::new((
                     (KeyPartitioning::new(), KeyPartitioning::new().into_sparse()),
                     Lsn(0),
                 )),
@@ -2389,8 +2380,6 @@ impl Timeline {
                 attach_wal_lag_cooldown,
 
                 create_idempotency,
-
-                page_trace: Default::default(),
             };
 
             result.repartition_threshold =
@@ -3792,34 +3781,35 @@ impl Timeline {
                 return Err(FlushLayerError::Cancelled);
             }
 
-            // Ensure that we have a single call to `create_image_layers` with a combined dense keyspace.
-            // So that the key ranges don't overlap.
-            let mut partitions = KeyPartitioning::default();
-            partitions.parts.extend(rel_partition.parts);
-            if !metadata_partition.parts.is_empty() {
-                assert_eq!(
-                    metadata_partition.parts.len(),
-                    1,
-                    "currently sparse keyspace should only contain a single metadata keyspace"
-                );
-                // Safety: create_image_layers treat sparse keyspaces differently that it does not scan
-                // every single key within the keyspace, and therefore, it's safe to force converting it
-                // into a dense keyspace before calling this function.
-                partitions
-                    .parts
-                    .extend(metadata_partition.into_dense().parts);
-            }
-
             let mut layers_to_upload = Vec::new();
             layers_to_upload.extend(
                 self.create_image_layers(
-                    &partitions,
+                    &rel_partition,
                     self.initdb_lsn,
                     ImageLayerCreationMode::Initial,
                     ctx,
                 )
                 .await?,
             );
+            if !metadata_partition.parts.is_empty() {
+                assert_eq!(
+                    metadata_partition.parts.len(),
+                    1,
+                    "currently sparse keyspace should only contain a single metadata keyspace"
+                );
+                layers_to_upload.extend(
+                    self.create_image_layers(
+                        // Safety: create_image_layers treat sparse keyspaces differently that it does not scan
+                        // every single key within the keyspace, and therefore, it's safe to force converting it
+                        // into a dense keyspace before calling this function.
+                        &metadata_partition.into_dense(),
+                        self.initdb_lsn,
+                        ImageLayerCreationMode::Initial,
+                        ctx,
+                    )
+                    .await?,
+                );
+            }
 
             (layers_to_upload, None)
         } else {
@@ -4032,15 +4022,18 @@ impl Timeline {
         flags: EnumSet<CompactFlags>,
         ctx: &RequestContext,
     ) -> Result<((KeyPartitioning, SparseKeyPartitioning), Lsn), CompactionError> {
-        let Ok(mut guard) = self.partitioning.try_write_guard() else {
+        let Ok(mut partitioning_guard) = self.partitioning.try_lock() else {
             // NB: there are two callers, one is the compaction task, of which there is only one per struct Tenant and hence Timeline.
             // The other is the initdb optimization in flush_frozen_layer, used by `boostrap_timeline`, which runs before `.activate()`
             // and hence before the compaction task starts.
+            // Note that there are a third "caller" that will take the `partitioning` lock. It is `gc_compaction_split_jobs` for
+            // gc-compaction where it uses the repartition data to determine the split jobs. In the future, it might use its own
+            // heuristics, but for now, we should allow concurrent access to it and let the caller retry compaction.
             return Err(CompactionError::Other(anyhow!(
-                "repartition() called concurrently"
+                "repartition() called concurrently, this is rare and a retry should be fine"
             )));
         };
-        let ((dense_partition, sparse_partition), partition_lsn) = &*guard.read();
+        let ((dense_partition, sparse_partition), partition_lsn) = &*partitioning_guard;
         if lsn < *partition_lsn {
             return Err(CompactionError::Other(anyhow!(
                 "repartition() called with LSN going backwards, this should not happen"
@@ -4068,9 +4061,9 @@ impl Timeline {
         let sparse_partitioning = SparseKeyPartitioning {
             parts: vec![sparse_ks],
         }; // no partitioning for metadata keys for now
-        let result = ((dense_partitioning, sparse_partitioning), lsn);
-        guard.write(result.clone());
-        Ok(result)
+        *partitioning_guard = ((dense_partitioning, sparse_partitioning), lsn);
+
+        Ok((partitioning_guard.0.clone(), partitioning_guard.1))
     }
 
     // Is it time to create a new image layer for the given partition?
@@ -4626,10 +4619,6 @@ impl Drop for Timeline {
                 }
             }
         }
-        info!(
-            "Timeline {} for tenant {} is being dropped",
-            self.timeline_id, self.tenant_shard_id.tenant_id
-        );
     }
 }
 

pageserver/src/tenant/timeline/compaction.rs

@@ -1776,10 +1776,7 @@ impl Timeline {
         base_img_from_ancestor: Option<(Key, Lsn, Bytes)>,
     ) -> anyhow::Result<KeyHistoryRetention> {
         // Pre-checks for the invariants
-
-        let debug_mode = cfg!(debug_assertions) || cfg!(feature = "testing");
-
-        if debug_mode {
+        if cfg!(debug_assertions) {
             for (log_key, _, _) in full_history {
                 assert_eq!(log_key, &key, "mismatched key");
             }
@@ -1925,19 +1922,15 @@ impl Timeline {
             output
         }
 
-        let mut key_exists = false;
         for (i, split_for_lsn) in split_history.into_iter().enumerate() {
             // TODO: there could be image keys inside the splits, and we can compute records_since_last_image accordingly.
             records_since_last_image += split_for_lsn.len();
-            // Whether to produce an image into the final layer files
-            let produce_image = if i == 0 && !has_ancestor {
+            let generate_image = if i == 0 && !has_ancestor {
                 // We always generate images for the first batch (below horizon / lowest retain_lsn)
                 true
             } else if i == batch_cnt - 1 {
                 // Do not generate images for the last batch (above horizon)
                 false
-            } else if records_since_last_image == 0 {
-                false
             } else if records_since_last_image >= delta_threshold_cnt {
                 // Generate images when there are too many records
                 true
@@ -1952,45 +1945,29 @@ impl Timeline {
                     break;
                 }
             }
-            if replay_history.is_empty() && !key_exists {
-                // The key does not exist at earlier LSN, we can skip this iteration.
-                retention.push(Vec::new());
-                continue;
-            } else {
-                key_exists = true;
+            if let Some((_, _, val)) = replay_history.first() {
+                if !val.will_init() {
+                    return Err(anyhow::anyhow!("invalid history, no base image")).with_context(
+                        || {
+                            generate_debug_trace(
+                                Some(&replay_history),
+                                full_history,
+                                retain_lsn_below_horizon,
+                                horizon,
+                            )
+                        },
+                    );
+                }
             }
-            let Some((_, _, val)) = replay_history.first() else {
-                unreachable!("replay history should not be empty once it exists")
-            };
-            if !val.will_init() {
-                return Err(anyhow::anyhow!("invalid history, no base image")).with_context(|| {
-                    generate_debug_trace(
-                        Some(&replay_history),
-                        full_history,
-                        retain_lsn_below_horizon,
-                        horizon,
-                    )
-                });
-            }
-            // Whether to reconstruct the image. In debug mode, we will generate an image
-            // at every retain_lsn to ensure data is not corrupted, but we won't put the
-            // image into the final layer.
-            let generate_image = produce_image || debug_mode;
-            if produce_image {
+            if generate_image && records_since_last_image > 0 {
                 records_since_last_image = 0;
-            }
-            let img_and_lsn = if generate_image {
-                let replay_history_for_debug = if debug_mode {
+                let replay_history_for_debug = if cfg!(debug_assertions) {
                     Some(replay_history.clone())
                 } else {
                     None
                 };
                 let replay_history_for_debug_ref = replay_history_for_debug.as_deref();
-                let history = if produce_image {
-                    std::mem::take(&mut replay_history)
-                } else {
-                    replay_history.clone()
-                };
+                let history = std::mem::take(&mut replay_history);
                 let mut img = None;
                 let mut records = Vec::with_capacity(history.len());
                 if let (_, lsn, Value::Image(val)) = history.first().as_ref().unwrap() {
@@ -2027,20 +2004,8 @@ impl Timeline {
                 }
                 records.reverse();
                 let state = ValueReconstructState { img, records };
-                // last batch does not generate image so i is always in range, unless we force generate
-                // an image during testing
-                let request_lsn = if i >= lsn_split_points.len() {
-                    Lsn::MAX
-                } else {
-                    lsn_split_points[i]
-                };
+                let request_lsn = lsn_split_points[i]; // last batch does not generate image so i is always in range
                 let img = self.reconstruct_value(key, request_lsn, state).await?;
-                Some((request_lsn, img))
-            } else {
-                None
-            };
-            if produce_image {
-                let (request_lsn, img) = img_and_lsn.unwrap();
                 replay_history.push((key, request_lsn, Value::Image(img.clone())));
                 retention.push(vec![(request_lsn, Value::Image(img))]);
             } else {
@@ -2146,7 +2111,12 @@ impl Timeline {
         let mut compact_jobs = Vec::new();
         // For now, we simply use the key partitioning information; we should do a more fine-grained partitioning
         // by estimating the amount of files read for a compaction job. We should also partition on LSN.
-        let ((dense_ks, sparse_ks), _) = self.partitioning.read().as_ref().clone();
+        let ((dense_ks, sparse_ks), _) = {
+            let Ok(partition) = self.partitioning.try_lock() else {
+                bail!("failed to acquire partition lock during gc-compaction");
+            };
+            partition.clone()
+        };
         // Truncate the key range to be within user specified compaction range.
         fn truncate_to(
             source_start: &Key,
@@ -2303,8 +2273,6 @@ impl Timeline {
         let compact_key_range = job.compact_key_range;
         let compact_lsn_range = job.compact_lsn_range;
 
-        let debug_mode = cfg!(debug_assertions) || cfg!(feature = "testing");
-
         info!("running enhanced gc bottom-most compaction, dry_run={dry_run}, compact_key_range={}..{}, compact_lsn_range={}..{}", compact_key_range.start, compact_key_range.end, compact_lsn_range.start, compact_lsn_range.end);
 
         scopeguard::defer! {
@@ -2430,7 +2398,7 @@ impl Timeline {
                 .first()
                 .copied()
                 .unwrap_or(job_desc.gc_cutoff);
-            if debug_mode {
+            if cfg!(debug_assertions) {
                 assert_eq!(
                     res,
                     job_desc

pageserver/src/tenant/timeline/delete.rs

@@ -301,7 +301,6 @@ impl DeleteTimelineFlow {
                 TimelineResources {
                     remote_client,
                     pagestream_throttle: tenant.pagestream_throttle.clone(),
-                    pagestream_throttle_metrics: tenant.pagestream_throttle_metrics.clone(),
                     l0_flush_global_state: tenant.l0_flush_global_state.clone(),
                 },
                 // Important. We dont pass ancestor above because it can be missing.

pageserver/src/tenant/timeline/handle.rs

@@ -32,151 +32,54 @@
 //!
 //! # Design
 //!
-//! ## Data Structures
-//!
 //! There are three user-facing data structures:
 //! - `PerTimelineState`: a struct embedded into each Timeline struct. Lifetime == Timeline lifetime.
 //! - `Cache`: a struct private to each connection handler; Lifetime == connection lifetime.
 //! - `Handle`: a smart pointer that holds the Timeline gate open and derefs to `&Timeline`.
-//! - `WeakHandle`: downgrade of a `Handle` that does not keep the gate open, but allows
-//!   trying to ugprade back to a `Handle`, guaranteeing it's the same `Timeline` *object*.
+//!   Lifetime: for a single request dispatch on the Timeline (i.e., one getpage request)
 //!
-//! Internally, there is 0 or 1 `HandleInner` per `(Cache,Timeline)`.
-//! Since Cache:Connection is 1:1, there is 0 or 1 `HandleInner` per `(Connection,Timeline)`.
+//! The `Handle` is just a wrapper around an `Arc<HandleInner>`.
 //!
-//! The `HandleInner`  is allocated as a `Arc<Mutex<HandleInner>>` and
-//! referenced weakly and strongly from various places which we are now illustrating.
-//! For brevity, we will omit the `Arc<Mutex<>>` part in the following and instead
-//! use `strong ref` and `weak ref` when referring to the `Arc<Mutex<HandleInner>>`
-//! or `Weak<Mutex<HandleInner>>`, respectively.
-//!
-//! - The `Handle` is a strong ref.
-//! - The `WeakHandle` is a weak ref.
-//! - The `PerTimelineState` contains a `HashMap<CacheId, strong ref>`.
-//! - The `Cache` is a `HashMap<unique identifier for the shard, weak ref>`.
-//!
-//! Lifetimes:
-//! - `WeakHandle` and `Handle`: single pagestream request.
-//! - `Cache`: single page service connection.
-//! - `PerTimelineState`:  lifetime of the Timeline object (i.e., i.e., till `Timeline::shutdown`).
-//!
-//! ## Request Handling Flow (= filling and using the `Cache``)
+//! There is one long-lived `Arc<HandleInner>`, which is stored in the `PerTimelineState`.
+//! The `Cache` stores a `Weak<HandleInner>` for each cached Timeline.
 //!
 //! To dispatch a request, the page service connection calls `Cache::get`.
 //!
 //! A cache miss means we consult the tenant manager for shard routing,
-//! resulting in an `Arc<Timeline>`. We enter its gate _once_ and store it in the the
-//! `Arc<Mutex<HandleInner>>>`. A weak ref is stored in the `Cache`
-//! and a strong ref in the `PerTimelineState`.
-//! A strong ref is returned wrapped in a `Handle`.
+//! resulting in an `Arc<Timeline>`. We enter its gate _once_ and construct an
+//! `Arc<HandleInner>`. We store a `Weak<HandleInner>` in the cache
+//! and the `Arc<HandleInner>` in the `PerTimelineState`.
 //!
 //! For subsequent requests, `Cache::get` will perform a "fast path" shard routing
-//! and find the weak ref in the cache.
-//! We upgrade the weak ref to a strong ref and return it wrapped in a `Handle`.
+//! and find the `Weak<HandleInner>` in the cache.
+//! We upgrade the `Weak<HandleInner>` to an `Arc<HandleInner>` and wrap it in the user-facing `Handle` type.
 //!
-//! The pagestream processing is pipelined and involves a batching step.
-//! While a request is batching, the `Handle` is downgraded to a `WeakHandle`.
-//! When the batch is ready to be executed, the `WeakHandle` is upgraded back to a `Handle`
-//! and the request handler dispatches the request to the right `<Handle as Deref<Target = Timeline>>::$request_method`.
+//! The request handler dispatches the request to the right `<Handle as Deref<Target = Timeline>>::$request_method`.
 //! It then drops the `Handle`, which drops the `Arc<HandleInner>`.
 //!
-//! # Performance
+//! # Memory Management / How The Reference Cycle Is Broken
 //!
-//! Remember from the introductory section:
+//! The attentive reader may have noticed the strong reference cycle
+//! from `Arc<HandleInner>` to `PerTimelineState` to `Arc<Timeline>`.
 //!
-//! > However, we want to avoid the overhead of entering the gate for every
-//! > method invocation.
-//!
-//! Why do we want to avoid that?
-//! Because the gate is a shared location in memory and entering it involves
-//! bumping refcounts, which leads to cache contention if done frequently
-//! from multiple cores in parallel.
-//!
-//! So, we only acquire the `GateGuard` once on `Cache` miss, and wrap it in an `Arc`.
-//! That `Arc` is private to the `HandleInner` and hence to the connection.
-//! (Review the "Data Structures" section if that is unclear to you.)
-//!
-//! A `WeakHandle` is a weak ref to the `HandleInner`.
-//! When upgrading a `WeakHandle`, we upgrade to a strong ref to the `HandleInner` and
-//! further acquire an additional strong ref to the `Arc<GateGuard>` inside it.
-//! Again, this manipulation of ref counts is is cheap because `Arc` is private to the connection.
-//!
-//! When downgrading a `Handle` to a `WeakHandle`, we drop the `Arc<GateGuard>`.
-//! Again, this is cheap because the `Arc` is private to the connection.
-//!
-//! In addition to the GateGuard, we need to provide `Deref<Target=Timeline>` impl.
-//! For this, both `Handle` need infallible access to an `Arc<Timeline>`.
-//! We could clone the `Arc<Timeline>` when upgrading a `WeakHandle`, but that would cause contention
-//! on the shared memory location that trakcs the refcount of the `Arc<Timeline>`.
-//! Instead, we wrap the `Arc<Timeline>` into another `Arc`.
-//! so that we can clone it cheaply when upgrading a `WeakHandle`.
-//!
-//! # Shutdown
-//!
-//! The attentive reader may have noticed the following reference cycle around the `Arc<Timeline>`:
-//!
-//! ```text
-//! Timeline --owns--> PerTimelineState --strong--> HandleInner --strong--> Timeline
-//! ```
-//!
-//! Further, there is this cycle:
-//!
-//! ```text
-//! Timeline --owns--> PerTimelineState --strong--> HandleInner --strong--> GateGuard --keepalive--> Timeline
-//! ```
-//!
-//! The former cycle is a memory leak if not broken.
-//! The latter cycle further prevents the Timeline from shutting down
-//! because we certainly won't drop the Timeline while the GateGuard is alive.
-//! Preventing shutdown is the whole point of this handle/cache system,
-//! but when the Timeline needs to shut down, we need to break the cycle.
+//! This cycle is intentional: while it exists, the `Cache` can upgrade its
+//! `Weak<HandleInner>` to an `Arc<HandleInner>` in a single atomic operation.
 //!
 //! The cycle is broken by either
-//! - Timeline shutdown (=> `PerTimelineState::shutdown`)
-//! - Connection shutdown (=> dropping the `Cache`).
+//! - `PerTimelineState::shutdown` or
+//! - dropping the `Cache`.
 //!
-//! Both transition the `HandleInner` from [`HandleInner::KeepingTimelineGateOpen`] to
-//! [`HandleInner::ShutDown`], which drops the only long-lived strong ref to the
-//! `Arc<GateGuard>`.
-//!
-//! `PerTimelineState::shutdown` drops all the `HandleInners` it contains,
-//! thereby breaking the cycle.
-//! It also initiates draining of already existing `Handle`s by
-//! poisoning things so that no new `HandleInner`'s can be added
-//! to the `PerTimelineState`, which will make subsequent `Cache::get` fail.
-//!
-//! Concurrently existing / already upgraded `Handle`s will extend the
-//! lifetime of the `Arc<Mutex<HandleInner>>` and hence cycles.
+//! Concurrently existing `Handle`s will extend the existence of the cycle.
 //! However, since `Handle`s are short-lived and new `Handle`s are not
-//! handed out from `Cache::get` or `WeakHandle::upgrade` after
-//! `PerTimelineState::shutdown`, that extension of the cycle is bounded.
-//!
-//! Concurrently existing `WeakHandle`s will fail to `upgrade()`:
-//! while they will succeed in upgrading `Weak<Mutex<HandleInner>>`,
-//! they will find the inner in state `HandleInner::ShutDown` state where the
-//! `Arc<GateGuard>` and Timeline has already been dropped.
-//!
-//! Dropping the `Cache` undoes the registration of this `Cache`'s
-//! `HandleInner`s from all the `PerTimelineState`s, i.e., it
-//! removes the strong ref to each of its `HandleInner`s
-//! from all the `PerTimelineState`.
-//!
-//! # Locking Rules
-//!
-//! To prevent deadlocks we:
-//!
-//! 1. Only ever hold one of the locks at a time.
-//! 2. Don't add more than one Drop impl that locks on the
-//!    cycles above.
-//!
-//! As per (2), that impl is in `Drop for Cache`.
+//! handed out after either `PerTimelineState::shutdown` or `Cache` drop,
+//! that extension of the cycle is bounded.
 //!
 //! # Fast Path for Shard Routing
 //!
 //! The `Cache` has a fast path for shard routing to avoid calling into
 //! the tenant manager for every request.
 //!
-//! The `Cache` maintains a hash map of `ShardTimelineId` to `WeakHandle`s.
+//! The `Cache` maintains a hash map of `ShardTimelineId` to `Weak<HandleInner>`.
 //!
 //! The current implementation uses the first entry in the hash map
 //! to determine the `ShardParameters` and derive the correct
@@ -184,18 +87,18 @@
 //!
 //! It then looks up the hash map for that `ShardTimelineId := {ShardIndex,TimelineId}`.
 //!
-//! If the lookup is successful and the `WeakHandle` can be upgraded,
+//! If the lookup is successful and the `Weak<HandleInner>` can be upgraded,
 //! it's a hit.
 //!
 //! ## Cache invalidation
 //!
-//! The insight is that cache invalidation is sufficient and most efficiently if done lazily.
+//! The insight is that cache invalidation is sufficient and most efficiently done lazily.
 //! The only reasons why an entry in the cache can become stale are:
 //! 1. The `PerTimelineState` / Timeline is shutting down e.g. because the shard is
 //!    being detached, timeline or shard deleted, or pageserver is shutting down.
 //! 2. We're doing a shard split and new traffic should be routed to the child shards.
 //!
-//! Regarding (1), we will eventually fail to upgrade the `WeakHandle` once the
+//! Regarding (1), we will eventually fail to upgrade the `Weak<HandleInner>` once the
 //! timeline has shut down, and when that happens, we remove the entry from the cache.
 //!
 //! Regarding (2), the insight is that it is toally fine to keep dispatching requests
@@ -204,6 +107,8 @@
 
 use std::collections::hash_map;
 use std::collections::HashMap;
+use std::sync::atomic::AtomicBool;
+use std::sync::atomic::Ordering;
 use std::sync::Arc;
 use std::sync::Mutex;
 use std::sync::Weak;
@@ -247,7 +152,7 @@ pub(crate) struct Cache<T: Types> {
     map: Map<T>,
 }
 
-type Map<T> = HashMap<ShardTimelineId, WeakHandle<T>>;
+type Map<T> = HashMap<ShardTimelineId, Weak<HandleInner<T>>>;
 
 impl<T: Types> Default for Cache<T> {
     fn default() -> Self {
@@ -265,22 +170,12 @@ pub(crate) struct ShardTimelineId {
 }
 
 /// See module-level comment.
-pub(crate) struct Handle<T: Types> {
-    timeline: Arc<T::Timeline>,
-    #[allow(dead_code)] // the field exists to keep the gate open
-    gate_guard: Arc<utils::sync::gate::GateGuard>,
-    inner: Arc<Mutex<HandleInner<T>>>,
-}
-pub(crate) struct WeakHandle<T: Types> {
-    inner: Weak<Mutex<HandleInner<T>>>,
-}
-enum HandleInner<T: Types> {
-    KeepingTimelineGateOpen {
-        #[allow(dead_code)]
-        gate_guard: Arc<utils::sync::gate::GateGuard>,
-        timeline: Arc<T::Timeline>,
-    },
-    ShutDown,
+pub(crate) struct Handle<T: Types>(Arc<HandleInner<T>>);
+struct HandleInner<T: Types> {
+    shut_down: AtomicBool,
+    timeline: T::Timeline,
+    // The timeline's gate held open.
+    _gate_guard: utils::sync::gate::GateGuard,
 }
 
 /// Embedded in each [`Types::Timeline`] as the anchor for the only long-lived strong ref to `HandleInner`.
@@ -288,8 +183,7 @@ enum HandleInner<T: Types> {
 /// See module-level comment for details.
 pub struct PerTimelineState<T: Types> {
     // None = shutting down
-    #[allow(clippy::type_complexity)]
-    handles: Mutex<Option<HashMap<CacheId, Arc<Mutex<HandleInner<T>>>>>>,
+    handles: Mutex<Option<HashMap<CacheId, Arc<HandleInner<T>>>>>,
 }
 
 impl<T: Types> Default for PerTimelineState<T> {
@@ -349,24 +243,49 @@ impl<T: Types> Cache<T> {
         shard_selector: ShardSelector,
         tenant_manager: &T::TenantManager,
     ) -> Result<Handle<T>, GetError<T>> {
-        // terminates because when every iteration we remove an element from the map
-        let miss: ShardSelector = loop {
+        // terminates because each iteration removes an element from the map
+        loop {
+            let handle = self
+                .get_impl(timeline_id, shard_selector, tenant_manager)
+                .await?;
+            if handle.0.shut_down.load(Ordering::Relaxed) {
+                let removed = self
+                    .map
+                    .remove(&handle.0.timeline.shard_timeline_id())
+                    .expect("invariant of get_impl is that the returned handle is in the map");
+                assert!(
+                    Weak::ptr_eq(&removed, &Arc::downgrade(&handle.0)),
+                    "shard_timeline_id() incorrect?"
+                );
+            } else {
+                return Ok(handle);
+            }
+        }
+    }
+
+    #[instrument(level = "trace", skip_all)]
+    async fn get_impl(
+        &mut self,
+        timeline_id: TimelineId,
+        shard_selector: ShardSelector,
+        tenant_manager: &T::TenantManager,
+    ) -> Result<Handle<T>, GetError<T>> {
+        let miss: ShardSelector = {
             let routing_state = self.shard_routing(timeline_id, shard_selector);
             match routing_state {
                 RoutingResult::FastPath(handle) => return Ok(handle),
                 RoutingResult::SlowPath(key) => match self.map.get(&key) {
                     Some(cached) => match cached.upgrade() {
-                        Ok(upgraded) => return Ok(upgraded),
-                        Err(HandleUpgradeError::ShutDown) => {
-                            // TODO: dedup with shard_routing()
+                        Some(upgraded) => return Ok(Handle(upgraded)),
+                        None => {
                             trace!("handle cache stale");
                             self.map.remove(&key).unwrap();
-                            continue;
+                            ShardSelector::Known(key.shard_index)
                         }
                     },
-                    None => break ShardSelector::Known(key.shard_index),
+                    None => ShardSelector::Known(key.shard_index),
                 },
-                RoutingResult::NeedConsultTenantManager => break shard_selector,
+                RoutingResult::NeedConsultTenantManager => shard_selector,
             }
         };
         self.get_miss(timeline_id, miss, tenant_manager).await
@@ -383,7 +302,7 @@ impl<T: Types> Cache<T> {
             let Some((first_key, first_handle)) = self.map.iter().next() else {
                 return RoutingResult::NeedConsultTenantManager;
             };
-            let Ok(first_handle) = first_handle.upgrade() else {
+            let Some(first_handle) = first_handle.upgrade() else {
                 // TODO: dedup with get()
                 trace!("handle cache stale");
                 let first_key_owned = *first_key;
@@ -391,7 +310,7 @@ impl<T: Types> Cache<T> {
                 continue;
             };
 
-            let first_handle_shard_identity = first_handle.get_shard_identity();
+            let first_handle_shard_identity = first_handle.timeline.get_shard_identity();
             let make_shard_index = |shard_num: ShardNumber| ShardIndex {
                 shard_number: shard_num,
                 shard_count: first_handle_shard_identity.count,
@@ -410,11 +329,11 @@ impl<T: Types> Cache<T> {
             };
             let first_handle_shard_timeline_id = ShardTimelineId {
                 shard_index: first_handle_shard_identity.shard_index(),
-                timeline_id: first_handle.shard_timeline_id().timeline_id,
+                timeline_id: first_handle.timeline.shard_timeline_id().timeline_id,
             };
 
             if need_shard_timeline_id == first_handle_shard_timeline_id {
-                return RoutingResult::FastPath(first_handle);
+                return RoutingResult::FastPath(Handle(first_handle));
             } else {
                 return RoutingResult::SlowPath(need_shard_timeline_id);
             }
@@ -438,30 +357,23 @@ impl<T: Types> Cache<T> {
                     ShardSelector::Known(idx) => assert_eq!(idx, &key.shard_index),
                 }
 
-                trace!("creating new HandleInner");
-                let handle_inner_arc = Arc::new(Mutex::new(HandleInner::KeepingTimelineGateOpen {
-                    gate_guard: Arc::new(
-                        // this enter() is expensive in production code because
-                        // it hits the global Arc<Timeline>::gate refcounts
-                        match timeline.gate().enter() {
-                            Ok(guard) => guard,
-                            Err(_) => {
-                                return Err(GetError::TimelineGateClosed);
-                            }
-                        },
-                    ),
-                    // this clone is expensive in production code because
-                    // it hits the global Arc<Timeline>::clone refcounts
-                    timeline: Arc::new(timeline.clone()),
-                }));
-                let handle_weak = WeakHandle {
-                    inner: Arc::downgrade(&handle_inner_arc),
+                let gate_guard = match timeline.gate().enter() {
+                    Ok(guard) => guard,
+                    Err(_) => {
+                        return Err(GetError::TimelineGateClosed);
+                    }
                 };
-                let handle = handle_weak
-                    .upgrade()
-                    .ok()
-                    .expect("we just created it and it's not linked anywhere yet");
-                {
+                trace!("creating new HandleInner");
+                let handle = Arc::new(
+                    // TODO: global metric that keeps track of the number of live HandlerTimeline instances
+                    // so we can identify reference cycle bugs.
+                    HandleInner {
+                        shut_down: AtomicBool::new(false),
+                        _gate_guard: gate_guard,
+                        timeline: timeline.clone(),
+                    },
+                );
+                let handle = {
                     let mut lock_guard = timeline
                         .per_timeline_state()
                         .handles
@@ -469,8 +381,7 @@ impl<T: Types> Cache<T> {
                         .expect("mutex poisoned");
                     match &mut *lock_guard {
                         Some(per_timeline_state) => {
-                            let replaced =
-                                per_timeline_state.insert(self.id, Arc::clone(&handle_inner_arc));
+                            let replaced = per_timeline_state.insert(self.id, Arc::clone(&handle));
                             assert!(replaced.is_none(), "some earlier code left a stale handle");
                             match self.map.entry(key) {
                                 hash_map::Entry::Occupied(_o) => {
@@ -481,7 +392,8 @@ impl<T: Types> Cache<T> {
                                     unreachable!()
                                 }
                                 hash_map::Entry::Vacant(v) => {
-                                    v.insert(handle_weak);
+                                    v.insert(Arc::downgrade(&handle));
+                                    handle
                                 }
                             }
                         }
@@ -489,62 +401,14 @@ impl<T: Types> Cache<T> {
                             return Err(GetError::PerTimelineStateShutDown);
                         }
                     }
-                }
-                Ok(handle)
+                };
+                Ok(Handle(handle))
             }
             Err(e) => Err(GetError::TenantManager(e)),
         }
     }
 }
 
-pub(crate) enum HandleUpgradeError {
-    ShutDown,
-}
-
-impl<T: Types> WeakHandle<T> {
-    pub(crate) fn upgrade(&self) -> Result<Handle<T>, HandleUpgradeError> {
-        let Some(inner) = Weak::upgrade(&self.inner) else {
-            return Err(HandleUpgradeError::ShutDown);
-        };
-        let lock_guard = inner.lock().expect("poisoned");
-        match &*lock_guard {
-            HandleInner::KeepingTimelineGateOpen {
-                timeline,
-                gate_guard,
-            } => {
-                let gate_guard = Arc::clone(gate_guard);
-                let timeline = Arc::clone(timeline);
-                drop(lock_guard);
-                Ok(Handle {
-                    timeline,
-                    gate_guard,
-                    inner,
-                })
-            }
-            HandleInner::ShutDown => Err(HandleUpgradeError::ShutDown),
-        }
-    }
-
-    pub(crate) fn is_same_handle_as(&self, other: &WeakHandle<T>) -> bool {
-        Weak::ptr_eq(&self.inner, &other.inner)
-    }
-}
-
-impl<T: Types> std::ops::Deref for Handle<T> {
-    type Target = T::Timeline;
-    fn deref(&self) -> &Self::Target {
-        &self.timeline
-    }
-}
-
-impl<T: Types> Handle<T> {
-    pub(crate) fn downgrade(&self) -> WeakHandle<T> {
-        WeakHandle {
-            inner: Arc::downgrade(&self.inner),
-        }
-    }
-}
-
 impl<T: Types> PerTimelineState<T> {
     /// After this method returns, [`Cache::get`] will never again return a [`Handle`]
     /// to the [`Types::Timeline`] that embeds this per-timeline state.
@@ -566,62 +430,43 @@ impl<T: Types> PerTimelineState<T> {
             trace!("already shut down");
             return;
         };
-        for handle_inner_arc in handles.values() {
+        for handle in handles.values() {
             // Make hits fail.
-            let mut lock_guard = handle_inner_arc.lock().expect("poisoned");
-            lock_guard.shutdown();
+            handle.shut_down.store(true, Ordering::Relaxed);
         }
         drop(handles);
     }
 }
 
+impl<T: Types> std::ops::Deref for Handle<T> {
+    type Target = T::Timeline;
+    fn deref(&self) -> &Self::Target {
+        &self.0.timeline
+    }
+}
+
+#[cfg(test)]
+impl<T: Types> Drop for HandleInner<T> {
+    fn drop(&mut self) {
+        trace!("HandleInner dropped");
+    }
+}
+
 // When dropping a [`Cache`], prune its handles in the [`PerTimelineState`] to break the reference cycle.
 impl<T: Types> Drop for Cache<T> {
     fn drop(&mut self) {
-        for (
-            _,
-            WeakHandle {
-                inner: handle_inner_weak,
-            },
-        ) in self.map.drain()
-        {
-            let Some(handle_inner_arc) = handle_inner_weak.upgrade() else {
-                continue;
-            };
-            let Some(handle_timeline) = handle_inner_arc
-                // locking rules: drop lock before acquiring other lock below
-                .lock()
-                .expect("poisoned")
-                .shutdown()
-            else {
-                // Concurrent PerTimelineState::shutdown.
-                continue;
-            };
-            // Clean up per_timeline_state so the HandleInner allocation can be dropped.
-            let per_timeline_state = handle_timeline.per_timeline_state();
-            let mut handles_lock_guard = per_timeline_state.handles.lock().expect("mutex poisoned");
-            let Some(handles) = &mut *handles_lock_guard else {
-                continue;
-            };
-            let Some(removed_handle_inner_arc) = handles.remove(&self.id) else {
-                // Concurrent PerTimelineState::shutdown.
-                continue;
-            };
-            drop(handles_lock_guard); // locking rules!
-            assert!(Arc::ptr_eq(&removed_handle_inner_arc, &handle_inner_arc));
-        }
-    }
-}
-
-impl<T: Types> HandleInner<T> {
-    fn shutdown(&mut self) -> Option<Arc<T::Timeline>> {
-        match std::mem::replace(self, HandleInner::ShutDown) {
-            HandleInner::KeepingTimelineGateOpen { timeline, .. } => Some(timeline),
-            HandleInner::ShutDown => {
-                // Duplicate shutdowns are possible because both Cache::drop and PerTimelineState::shutdown
-                // may do it concurrently, but locking rules disallow holding per-timeline-state lock and
-                // the handle lock at the same time.
-                None
+        for (_, weak) in self.map.drain() {
+            if let Some(strong) = weak.upgrade() {
+                // handle is still being kept alive in PerTimelineState
+                let timeline = strong.timeline.per_timeline_state();
+                let mut handles = timeline.handles.lock().expect("mutex poisoned");
+                if let Some(handles) = &mut *handles {
+                    let Some(removed) = handles.remove(&self.id) else {
+                        // There could have been a shutdown inbetween us upgrading the weak and locking the mutex.
+                        continue;
+                    };
+                    assert!(Arc::ptr_eq(&removed, &strong));
+                }
             }
         }
     }
@@ -629,8 +474,6 @@ impl<T: Types> HandleInner<T> {
 
 #[cfg(test)]
 mod tests {
-    use std::sync::Weak;
-
     use pageserver_api::{
         key::{rel_block_to_key, Key, DBDIR_KEY},
         models::ShardParameters,
@@ -740,13 +583,39 @@ mod tests {
         //
         // fill the cache
         //
+        assert_eq!(
+            (Arc::strong_count(&shard0), Arc::weak_count(&shard0)),
+            (2, 1),
+            "strong: shard0, mgr; weak: myself"
+        );
+
         let handle: Handle<_> = cache
             .get(timeline_id, ShardSelector::Page(key), &mgr)
             .await
             .expect("we have the timeline");
+        let handle_inner_weak = Arc::downgrade(&handle.0);
         assert!(Weak::ptr_eq(&handle.myself, &shard0.myself));
+        assert_eq!(
+            (
+                Weak::strong_count(&handle_inner_weak),
+                Weak::weak_count(&handle_inner_weak)
+            ),
+            (2, 2),
+            "strong: handle, per_timeline_state, weak: handle_inner_weak, cache"
+        );
         assert_eq!(cache.map.len(), 1);
+
+        assert_eq!(
+            (Arc::strong_count(&shard0), Arc::weak_count(&shard0)),
+            (3, 1),
+            "strong: handleinner(per_timeline_state), shard0, mgr; weak: myself"
+        );
         drop(handle);
+        assert_eq!(
+            (Arc::strong_count(&shard0), Arc::weak_count(&shard0)),
+            (3, 1),
+            "strong: handleinner(per_timeline_state), shard0, mgr; weak: myself"
+        );
 
         //
         // demonstrate that Handle holds up gate closure
@@ -771,11 +640,21 @@ mod tests {
         // SHUTDOWN
         shard0.per_timeline_state.shutdown(); // keeping handle alive across shutdown
 
+        assert_eq!(
+            1,
+            Weak::strong_count(&handle_inner_weak),
+            "through local var handle"
+        );
         assert_eq!(
             cache.map.len(),
             1,
             "this is an implementation detail but worth pointing out: we can't clear the cache from shutdown(), it's cleared on first access after"
         );
+        assert_eq!(
+            (Arc::strong_count(&shard0), Arc::weak_count(&shard0)),
+            (3, 1),
+            "strong: handleinner(via handle), shard0, mgr; weak: myself"
+        );
 
         // this handle is perfectly usable
         handle.getpage();
@@ -799,6 +678,16 @@ mod tests {
         }
 
         drop(handle);
+        assert_eq!(
+            0,
+            Weak::strong_count(&handle_inner_weak),
+            "the HandleInner destructor already ran"
+        );
+        assert_eq!(
+            (Arc::strong_count(&shard0), Arc::weak_count(&shard0)),
+            (2, 1),
+            "strong: shard0, mgr; weak: myself"
+        );
 
         // closing gate succeeds after dropping handle
         tokio::select! {
@@ -817,8 +706,10 @@ mod tests {
         assert_eq!(cache.map.len(), 0);
 
         // ensure all refs to shard0 are gone and we're not leaking anything
+        let myself = Weak::clone(&shard0.myself);
         drop(shard0);
         drop(mgr);
+        assert_eq!(Weak::strong_count(&myself), 0);
     }
 
     #[tokio::test]
@@ -1057,11 +948,15 @@ mod tests {
                 handle
             };
             handle.getpage();
-            used_handles.push(Arc::downgrade(&handle.timeline));
+            used_handles.push(Arc::downgrade(&handle.0));
         }
 
-        // No handles exist, thus gates are closed and don't require shutdown.
-        // Thus the gate should close immediately, even without shutdown.
+        // No handles exist, thus gates are closed and don't require shutdown
+        assert!(used_handles
+            .iter()
+            .all(|weak| Weak::strong_count(weak) == 0));
+
+        // ... thus the gate should close immediately, even without shutdown
         tokio::select! {
             _ = shard0.gate.close() => { }
             _ = tokio::time::sleep(FOREVER) => {
@@ -1069,75 +964,4 @@ mod tests {
             }
         }
     }
-
-    #[tokio::test(start_paused = true)]
-    async fn test_weak_handles() {
-        crate::tenant::harness::setup_logging();
-        let timeline_id = TimelineId::generate();
-        let shard0 = Arc::new_cyclic(|myself| StubTimeline {
-            gate: Default::default(),
-            id: timeline_id,
-            shard: ShardIdentity::unsharded(),
-            per_timeline_state: PerTimelineState::default(),
-            myself: myself.clone(),
-        });
-        let mgr = StubManager {
-            shards: vec![shard0.clone()],
-        };
-
-        let refcount_start = Arc::strong_count(&shard0);
-
-        let key = DBDIR_KEY;
-
-        let mut cache = Cache::<TestTypes>::default();
-
-        let handle = cache
-            .get(timeline_id, ShardSelector::Page(key), &mgr)
-            .await
-            .expect("we have the timeline");
-        assert!(Weak::ptr_eq(&handle.myself, &shard0.myself));
-
-        let weak_handle = handle.downgrade();
-
-        drop(handle);
-
-        let upgraded_handle = weak_handle.upgrade().ok().expect("we can upgrade it");
-
-        // Start shutdown
-        shard0.per_timeline_state.shutdown();
-
-        // Upgrades during shutdown don't work, even if upgraded_handle exists.
-        weak_handle
-            .upgrade()
-            .err()
-            .expect("can't upgrade weak handle as soon as shutdown started");
-
-        // But upgraded_handle is still alive, so the gate won't close.
-        tokio::select! {
-            _ = shard0.gate.close() => {
-                panic!("handle is keeping gate open");
-            }
-            _ = tokio::time::sleep(FOREVER) => { }
-        }
-
-        // Drop the last handle.
-        drop(upgraded_handle);
-
-        // The gate should close now, despite there still being a weak_handle.
-        tokio::select! {
-            _ = shard0.gate.close() => { }
-            _ = tokio::time::sleep(FOREVER) => {
-                panic!("only strong handle is dropped and we shut down per-timeline-state")
-            }
-        }
-
-        // The weak handle still can't be upgraded.
-        weak_handle
-            .upgrade()
-            .err()
-            .expect("still shouldn't be able to upgrade the weak handle");
-
-        // There should be no strong references to the timeline object except the one on "stack".
-        assert_eq!(Arc::strong_count(&shard0), refcount_start);
-    }
 }

pageserver/src/tenant/timeline/walreceiver/walreceiver_connection.rs

@@ -140,7 +140,7 @@ pub(super) async fn handle_walreceiver_connection(
 
     let (replication_client, connection) = {
         let mut config = wal_source_connconf.to_tokio_postgres_config();
-        config.application_name(format!("pageserver-{}", node.0).as_str());
+        config.application_name("pageserver");
         config.replication_mode(tokio_postgres::config::ReplicationMode::Physical);
         match time::timeout(connect_timeout, config.connect(postgres::NoTls)).await {
             Ok(client_and_conn) => client_and_conn?,
@@ -264,8 +264,6 @@ pub(super) async fn handle_walreceiver_connection(
 
     let mut walingest = WalIngest::new(timeline.as_ref(), startpoint, &ctx).await?;
 
-    let shard = vec![*timeline.get_shard_identity()];
-
     let interpreted_proto_config = match protocol {
         PostgresClientProtocol::Vanilla => None,
         PostgresClientProtocol::Interpreted {
@@ -478,12 +476,10 @@ pub(super) async fn handle_walreceiver_connection(
                         // Deserialize and interpret WAL record
                         let interpreted = InterpretedWalRecord::from_bytes_filtered(
                             recdata,
-                            &shard,
+                            modification.tline.get_shard_identity(),
                             next_record_lsn,
                             modification.tline.pg_version,
-                        )?
-                        .remove(timeline.get_shard_identity())
-                        .unwrap();
+                        )?;
 
                         if matches!(interpreted.flush_uncommitted, FlushUncommittedRecords::Yes)
                             && uncommitted_records > 0

pageserver/src/walingest.rs

@@ -2163,12 +2163,10 @@ mod tests {
             while let Some((lsn, recdata)) = decoder.poll_decode().unwrap() {
                 let interpreted = InterpretedWalRecord::from_bytes_filtered(
                     recdata,
-                    &[*modification.tline.get_shard_identity()],
+                    modification.tline.get_shard_identity(),
                     lsn,
                     modification.tline.pg_version,
                 )
-                .unwrap()
-                .remove(modification.tline.get_shard_identity())
                 .unwrap();
 
                 walingest

pgxn/neon/libpagestore.c

@@ -911,74 +911,7 @@ pageserver_receive(shardno_t shard_no)
 		}
 		PG_CATCH();
 		{
-			neon_shard_log(shard_no, LOG, "pageserver_receive: disconnect due to failure while parsing response");
-			pageserver_disconnect(shard_no);
-			PG_RE_THROW();
-		}
-		PG_END_TRY();
-
-		if (message_level_is_interesting(PageStoreTrace))
-		{
-			char	   *msg = nm_to_string((NeonMessage *) resp);
-
-			neon_shard_log(shard_no, PageStoreTrace, "got response: %s", msg);
-			pfree(msg);
-		}
-	}
-	else if (rc == -1)
-	{
-		neon_shard_log(shard_no, LOG, "pageserver_receive disconnect: psql end of copy data: %s", pchomp(PQerrorMessage(pageserver_conn)));
-		pageserver_disconnect(shard_no);
-		resp = NULL;
-	}
-	else if (rc == -2)
-	{
-		char	   *msg = pchomp(PQerrorMessage(pageserver_conn));
-
-		pageserver_disconnect(shard_no);
-		neon_shard_log(shard_no, ERROR, "pageserver_receive disconnect: could not read COPY data: %s", msg);
-	}
-	else
-	{
-		pageserver_disconnect(shard_no);
-		neon_shard_log(shard_no, ERROR, "pageserver_receive disconnect: unexpected PQgetCopyData return value: %d", rc);
-	}
-
-	shard->nresponses_received++;
-	return (NeonResponse *) resp;
-}
-
-static NeonResponse *
-pageserver_try_receive(shardno_t shard_no)
-{
-	StringInfoData resp_buff;
-	NeonResponse *resp;
-	PageServer *shard = &page_servers[shard_no];
-	PGconn	   *pageserver_conn = shard->conn;
-	/* read response */
-	int			rc;
-
-	if (shard->state != PS_Connected)
-		return NULL;
-
-	Assert(pageserver_conn);
-
-	rc = PQgetCopyData(shard->conn, &resp_buff.data, 1 /* async = true */);
-
-	if (rc == 0)
-		return NULL;
-	else if (rc > 0)
-	{
-		PG_TRY();
-		{
-			resp_buff.len = rc;
-			resp_buff.cursor = 0;
-			resp = nm_unpack_response(&resp_buff);
-			PQfreemem(resp_buff.data);
-		}
-		PG_CATCH();
-		{
-			neon_shard_log(shard_no, LOG, "pageserver_receive: disconnect due to failure while parsing response");
+			neon_shard_log(shard_no, LOG, "pageserver_receive: disconnect due malformatted response");
 			pageserver_disconnect(shard_no);
 			PG_RE_THROW();
 		}
@@ -1047,7 +980,6 @@ page_server_api api =
 	.send = pageserver_send,
 	.flush = pageserver_flush,
 	.receive = pageserver_receive,
-	.try_receive = pageserver_try_receive,
 	.disconnect = pageserver_disconnect_shard
 };
 

pgxn/neon/pagestore_client.h

@@ -34,8 +34,6 @@ typedef enum
 	T_NeonGetPageRequest,
 	T_NeonDbSizeRequest,
 	T_NeonGetSlruSegmentRequest,
-	/* future tags above this line */
-	T_NeonTestRequest = 99, /* only in cfg(feature = "testing") */
 
 	/* pagestore -> pagestore_client */
 	T_NeonExistsResponse = 100,
@@ -44,8 +42,6 @@ typedef enum
 	T_NeonErrorResponse,
 	T_NeonDbSizeResponse,
 	T_NeonGetSlruSegmentResponse,
-	/* future tags above this line */
-	T_NeonTestResponse = 199, /* only in cfg(feature = "testing") */
 } NeonMessageTag;
 
 typedef uint64 NeonRequestId;
@@ -196,29 +192,9 @@ typedef uint16 shardno_t;
 
 typedef struct
 {
-	/*
-	 * Send this request to the PageServer associated with this shard.
-	 */
 	bool		(*send) (shardno_t  shard_no, NeonRequest * request);
-	/*
-	 * Blocking read for the next response of this shard.
-	 *
-	 * When a CANCEL signal is handled, the connection state will be
-	 * unmodified.
-	 */
 	NeonResponse *(*receive) (shardno_t shard_no);
-	/*
-	 * Try get the next response from the TCP buffers, if any.
-	 * Returns NULL when the data is not yet available. 
-	 */
-	NeonResponse *(*try_receive) (shardno_t shard_no);
-	/*
-	 * Make sure all requests are sent to PageServer.
-	 */
 	bool		(*flush) (shardno_t shard_no);
-	/*
-	 * Disconnect from this pageserver shard.
-	 */
 	void        (*disconnect) (shardno_t shard_no);
 } page_server_api;
 

pgxn/neon/pagestore_smgr.c

@@ -405,56 +405,6 @@ compact_prefetch_buffers(void)
 	return false;
 }
 
-/*
- * If there might be responses still in the TCP buffer, then
- * we should try to use those, so as to reduce any TCP backpressure
- * on the OS/PS side.
- *
- * This procedure handles that.
- *
- * Note that this is only valid as long as the only pipelined
- * operations in the TCP buffer are getPage@Lsn requests.
- */
-static void
-prefetch_pump_state(void)
-{
-	while (MyPState->ring_receive != MyPState->ring_flush)
-	{
-		NeonResponse   *response;
-		PrefetchRequest *slot;
-		MemoryContext	old;
-
-		slot = GetPrfSlot(MyPState->ring_receive);
-
-		old = MemoryContextSwitchTo(MyPState->errctx);
-		response = page_server->try_receive(slot->shard_no);
-		MemoryContextSwitchTo(old);
-
-		if (response == NULL)
-			break;
-
-		/* The slot should still be valid */
-		if (slot->status != PRFS_REQUESTED ||
-			slot->response != NULL ||
-			slot->my_ring_index != MyPState->ring_receive)
-			neon_shard_log(slot->shard_no, ERROR,
-						   "Incorrect prefetch slot state after receive: status=%d response=%p my=%lu receive=%lu",
-						   slot->status, slot->response,
-						   (long) slot->my_ring_index, (long) MyPState->ring_receive);
-
-		/* update prefetch state */
-		MyPState->n_responses_buffered += 1;
-		MyPState->n_requests_inflight -= 1;
-		MyPState->ring_receive += 1;
-		MyNeonCounters->getpage_prefetches_buffered =
-			MyPState->n_responses_buffered;
-
-		/* update slot state */
-		slot->status = PRFS_RECEIVED;
-		slot->response = response;
-	}
-}
-
 void
 readahead_buffer_resize(int newsize, void *extra)
 {
@@ -2858,8 +2808,6 @@ neon_prefetch(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum,
 			   MyPState->ring_last <= ring_index);
 	}
 
-	prefetch_pump_state();
-
 	return false;
 }
 
@@ -2901,8 +2849,6 @@ neon_prefetch(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum)
 	Assert(ring_index < MyPState->ring_unused &&
 		   MyPState->ring_last <= ring_index);
 
-	prefetch_pump_state();
-
 	return false;
 }
 #endif /* PG_MAJORVERSION_NUM < 17 */
@@ -2945,8 +2891,6 @@ neon_writeback(SMgrRelation reln, ForkNumber forknum,
 	 */
 	neon_log(SmgrTrace, "writeback noop");
 
-	prefetch_pump_state();
-
 #ifdef DEBUG_COMPARE_LOCAL
 	if (IS_LOCAL_REL(reln))
 		mdwriteback(reln, forknum, blocknum, nblocks);
@@ -3201,8 +3145,6 @@ neon_read(SMgrRelation reln, ForkNumber forkNum, BlockNumber blkno, void *buffer
 	neon_get_request_lsns(InfoFromSMgrRel(reln), forkNum, blkno, &request_lsns, 1, NULL);
 	neon_read_at_lsn(InfoFromSMgrRel(reln), forkNum, blkno, request_lsns, buffer);
 
-	prefetch_pump_state();
-
 #ifdef DEBUG_COMPARE_LOCAL
 	if (forkNum == MAIN_FORKNUM && IS_LOCAL_REL(reln))
 	{
@@ -3340,8 +3282,6 @@ neon_readv(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum,
 	neon_read_at_lsnv(InfoFromSMgrRel(reln), forknum, blocknum, request_lsns,
 					  buffers, nblocks, read);
 
-	prefetch_pump_state();
-
 #ifdef DEBUG_COMPARE_LOCAL
 	if (forkNum == MAIN_FORKNUM && IS_LOCAL_REL(reln))
 	{
@@ -3510,8 +3450,6 @@ neon_write(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, const vo
 
 	lfc_write(InfoFromSMgrRel(reln), forknum, blocknum, buffer);
 
-	prefetch_pump_state();
-
 #ifdef DEBUG_COMPARE_LOCAL
 	if (IS_LOCAL_REL(reln))
 		#if PG_MAJORVERSION_NUM >= 17
@@ -3565,8 +3503,6 @@ neon_writev(SMgrRelation reln, ForkNumber forknum, BlockNumber blkno,
 
 	lfc_writev(InfoFromSMgrRel(reln), forknum, blkno, buffers, nblocks);
 
-	prefetch_pump_state();
-
 #ifdef DEBUG_COMPARE_LOCAL
 	if (IS_LOCAL_REL(reln))
 		mdwritev(reln, forknum, blocknum, &buffer, 1, skipFsync);
@@ -3856,8 +3792,6 @@ neon_immedsync(SMgrRelation reln, ForkNumber forknum)
 
 	neon_log(SmgrTrace, "[NEON_SMGR] immedsync noop");
 
-	prefetch_pump_state();
-
 #ifdef DEBUG_COMPARE_LOCAL
 	if (IS_LOCAL_REL(reln))
 		mdimmedsync(reln, forknum);

proxy/src/context/parquet.rs

@@ -187,6 +187,10 @@ pub async fn worker(
     let rx = futures::stream::poll_fn(move |cx| rx.poll_recv(cx));
     let rx = rx.map(RequestData::from);
 
+    let storage = GenericRemoteStorage::from_config(&remote_storage_config)
+        .await
+        .context("remote storage init")?;
+
     let properties = WriterProperties::builder()
         .set_data_page_size_limit(config.parquet_upload_page_size)
         .set_compression(config.parquet_upload_compression);
@@ -220,18 +224,18 @@ pub async fn worker(
         let rx_disconnect = futures::stream::poll_fn(move |cx| rx_disconnect.poll_recv(cx));
         let rx_disconnect = rx_disconnect.map(RequestData::from);
 
+        let storage_disconnect =
+            GenericRemoteStorage::from_config(&disconnect_events_storage_config)
+                .await
+                .context("remote storage for disconnect events init")?;
         let parquet_config_disconnect = parquet_config.clone();
         tokio::try_join!(
-            worker_inner(remote_storage_config, rx, parquet_config),
-            worker_inner(
-                disconnect_events_storage_config,
-                rx_disconnect,
-                parquet_config_disconnect
-            )
+            worker_inner(storage, rx, parquet_config),
+            worker_inner(storage_disconnect, rx_disconnect, parquet_config_disconnect)
         )
         .map(|_| ())
     } else {
-        worker_inner(remote_storage_config, rx, parquet_config).await
+        worker_inner(storage, rx, parquet_config).await
     }
 }
 
@@ -247,32 +251,18 @@ struct ParquetConfig {
     test_remote_failures: u64,
 }
 
-impl ParquetConfig {
-    async fn storage(
-        &self,
-        storage_config: &RemoteStorageConfig,
-    ) -> anyhow::Result<GenericRemoteStorage> {
-        let storage = GenericRemoteStorage::from_config(storage_config)
-            .await
-            .context("remote storage init")?;
-
-        #[cfg(any(test, feature = "testing"))]
-        if self.test_remote_failures > 0 {
-            return Ok(GenericRemoteStorage::unreliable_wrapper(
-                storage,
-                self.test_remote_failures,
-            ));
-        }
-
-        Ok(storage)
-    }
-}
-
 async fn worker_inner(
-    storage_config: RemoteStorageConfig,
+    storage: GenericRemoteStorage,
     rx: impl Stream<Item = RequestData>,
     config: ParquetConfig,
 ) -> anyhow::Result<()> {
+    #[cfg(any(test, feature = "testing"))]
+    let storage = if config.test_remote_failures > 0 {
+        GenericRemoteStorage::unreliable_wrapper(storage, config.test_remote_failures)
+    } else {
+        storage
+    };
+
     let mut rx = std::pin::pin!(rx);
 
     let mut rows = Vec::with_capacity(config.rows_per_group);
@@ -295,7 +285,7 @@ async fn worker_inner(
         }
         if len > config.file_size || force {
             last_upload = time::Instant::now();
-            let file = upload_parquet(w, len, &storage_config, &config).await?;
+            let file = upload_parquet(w, len, &storage).await?;
             w = SerializedFileWriter::new(file, schema.clone(), config.propeties.clone())?;
             len = 0;
         }
@@ -308,7 +298,7 @@ async fn worker_inner(
     }
 
     if !w.flushed_row_groups().is_empty() {
-        let _rtchk: Writer<BytesMut> = upload_parquet(w, len, &storage_config, &config).await?;
+        let _rtchk: Writer<BytesMut> = upload_parquet(w, len, &storage).await?;
     }
 
     Ok(())
@@ -350,8 +340,7 @@ where
 async fn upload_parquet(
     mut w: SerializedFileWriter<Writer<BytesMut>>,
     len: i64,
-    storage_config: &RemoteStorageConfig,
-    config: &ParquetConfig,
+    storage: &GenericRemoteStorage,
 ) -> anyhow::Result<Writer<BytesMut>> {
     let len_uncompressed = w
         .flushed_row_groups()
@@ -388,15 +377,6 @@ async fn upload_parquet(
         size, compression, "uploading request parquet file"
     );
 
-    // A bug in azure-sdk means that the identity-token-file that expires after
-    // 1 hour is not refreshed. This identity-token is used to fetch the actual azure storage
-    // tokens that last for 24 hours. After this 24 hour period, azure-sdk tries to refresh
-    // the storage token, but the identity token has now expired.
-    // <https://github.com/Azure/azure-sdk-for-rust/issues/1739>
-    //
-    // To work around this, we recreate the storage every time.
-    let storage = config.storage(storage_config).await?;
-
     let year = now.year();
     let month = now.month();
     let day = now.day();
@@ -451,8 +431,8 @@ mod tests {
     use rand::rngs::StdRng;
     use rand::{Rng, SeedableRng};
     use remote_storage::{
-        RemoteStorageConfig, RemoteStorageKind, S3Config, DEFAULT_MAX_KEYS_PER_LIST_RESPONSE,
-        DEFAULT_REMOTE_STORAGE_S3_CONCURRENCY_LIMIT,
+        GenericRemoteStorage, RemoteStorageConfig, RemoteStorageKind, S3Config,
+        DEFAULT_MAX_KEYS_PER_LIST_RESPONSE, DEFAULT_REMOTE_STORAGE_S3_CONCURRENCY_LIMIT,
     };
     use tokio::sync::mpsc;
     use tokio::time;
@@ -579,11 +559,12 @@ mod tests {
             timeout: std::time::Duration::from_secs(120),
             small_timeout: std::time::Duration::from_secs(30),
         };
-
-        worker_inner(remote_storage_config, rx, config)
+        let storage = GenericRemoteStorage::from_config(&remote_storage_config)
             .await
             .unwrap();
 
+        worker_inner(storage, rx, config).await.unwrap();
+
         let mut files = WalkDir::new(tmpdir.as_std_path())
             .into_iter()
             .filter_map(|entry| entry.ok())

safekeeper/Cargo.toml

@@ -26,7 +26,6 @@ hex.workspace = true
 humantime.workspace = true
 http.workspace = true
 hyper0.workspace = true
-itertools.workspace = true
 futures.workspace = true
 once_cell.workspace = true
 parking_lot.workspace = true
@@ -40,7 +39,6 @@ scopeguard.workspace = true
 reqwest = { workspace = true, features = ["json"] }
 serde.workspace = true
 serde_json.workspace = true
-smallvec.workspace = true
 strum.workspace = true
 strum_macros.workspace = true
 thiserror.workspace = true
@@ -65,7 +63,6 @@ storage_broker.workspace = true
 tokio-stream.workspace = true
 utils.workspace = true
 wal_decoder.workspace = true
-env_logger.workspace = true
 
 workspace_hack.workspace = true
 

safekeeper/src/bin/safekeeper.rs

@@ -207,13 +207,6 @@ struct Args {
     /// Also defines interval for eviction retries.
     #[arg(long, value_parser = humantime::parse_duration, default_value = DEFAULT_EVICTION_MIN_RESIDENT)]
     eviction_min_resident: Duration,
-    /// Enable fanning out WAL to different shards from the same reader
-    #[arg(long)]
-    wal_reader_fanout: bool,
-    /// Only fan out the WAL reader if the absoulte delta between the new requested position
-    /// and the current position of the reader is smaller than this value.
-    #[arg(long)]
-    max_delta_for_fanout: Option<u64>,
 }
 
 // Like PathBufValueParser, but allows empty string.
@@ -377,8 +370,6 @@ async fn main() -> anyhow::Result<()> {
         control_file_save_interval: args.control_file_save_interval,
         partial_backup_concurrency: args.partial_backup_concurrency,
         eviction_min_resident: args.eviction_min_resident,
-        wal_reader_fanout: args.wal_reader_fanout,
-        max_delta_for_fanout: args.max_delta_for_fanout,
     });
 
     // initialize sentry if SENTRY_DSN is provided

safekeeper/src/control_file.rs

@@ -3,7 +3,6 @@
 use anyhow::{bail, ensure, Context, Result};
 use byteorder::{LittleEndian, ReadBytesExt, WriteBytesExt};
 use camino::{Utf8Path, Utf8PathBuf};
-use safekeeper_api::membership::INVALID_GENERATION;
 use tokio::fs::File;
 use tokio::io::AsyncWriteExt;
 use utils::crashsafe::durable_rename;
@@ -14,14 +13,14 @@ use std::ops::Deref;
 use std::path::Path;
 use std::time::Instant;
 
-use crate::control_file_upgrade::downgrade_v10_to_v9;
+use crate::control_file_upgrade::downgrade_v9_to_v8;
 use crate::control_file_upgrade::upgrade_control_file;
 use crate::metrics::PERSIST_CONTROL_FILE_SECONDS;
 use crate::state::{EvictionState, TimelinePersistentState};
 use utils::bin_ser::LeSer;
 
 pub const SK_MAGIC: u32 = 0xcafeceefu32;
-pub const SK_FORMAT_VERSION: u32 = 10;
+pub const SK_FORMAT_VERSION: u32 = 9;
 
 // contains persistent metadata for safekeeper
 pub const CONTROL_FILE_NAME: &str = "safekeeper.control";
@@ -170,11 +169,10 @@ impl TimelinePersistentState {
         let mut buf: Vec<u8> = Vec::new();
         WriteBytesExt::write_u32::<LittleEndian>(&mut buf, SK_MAGIC)?;
 
-        if self.mconf.generation == INVALID_GENERATION {
-            // Temp hack for forward compatibility test: in case of none
-            // configuration save cfile in previous v9 format.
-            const PREV_FORMAT_VERSION: u32 = 9;
-            let prev = downgrade_v10_to_v9(self);
+        if self.eviction_state == EvictionState::Present {
+            // temp hack for forward compatibility
+            const PREV_FORMAT_VERSION: u32 = 8;
+            let prev = downgrade_v9_to_v8(self);
             WriteBytesExt::write_u32::<LittleEndian>(&mut buf, PREV_FORMAT_VERSION)?;
             prev.ser_into(&mut buf)?;
         } else {
@@ -235,7 +233,6 @@ impl Storage for FileStorage {
 #[cfg(test)]
 mod test {
     use super::*;
-    use safekeeper_api::membership::{Configuration, MemberSet};
     use tokio::fs;
     use utils::lsn::Lsn;
 
@@ -245,11 +242,6 @@ mod test {
     async fn test_read_write_safekeeper_state() -> anyhow::Result<()> {
         let tempdir = camino_tempfile::tempdir()?;
         let mut state = TimelinePersistentState::empty();
-        state.mconf = Configuration {
-            generation: 42,
-            members: MemberSet::empty(),
-            new_members: None,
-        };
         let mut storage = FileStorage::create_new(tempdir.path(), state.clone(), NO_SYNC).await?;
 
         // Make a change.

safekeeper/src/control_file_upgrade.rs

@@ -1,22 +1,17 @@
 //! Code to deal with safekeeper control file upgrades
-use std::vec;
-
 use crate::{
     safekeeper::{AcceptorState, PgUuid, TermHistory, TermLsn},
-    state::{EvictionState, TimelinePersistentState},
+    state::{EvictionState, PersistedPeers, TimelinePersistentState},
     wal_backup_partial,
 };
 use anyhow::{bail, Result};
 use pq_proto::SystemId;
-use safekeeper_api::{
-    membership::{Configuration, INVALID_GENERATION},
-    ServerInfo, Term,
-};
+use safekeeper_api::{ServerInfo, Term};
 use serde::{Deserialize, Serialize};
 use tracing::*;
 use utils::{
     bin_ser::LeSer,
-    id::{NodeId, TenantId, TimelineId},
+    id::{TenantId, TimelineId},
     lsn::Lsn,
 };
 
@@ -238,90 +233,6 @@ pub struct SafeKeeperStateV8 {
     pub partial_backup: wal_backup_partial::State,
 }
 
-#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
-pub struct PersistedPeers(pub Vec<(NodeId, PersistedPeerInfo)>);
-
-#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
-pub struct PersistedPeerInfo {
-    /// LSN up to which safekeeper offloaded WAL to s3.
-    pub backup_lsn: Lsn,
-    /// Term of the last entry.
-    pub term: Term,
-    /// LSN of the last record.
-    pub flush_lsn: Lsn,
-    /// Up to which LSN safekeeper regards its WAL as committed.
-    pub commit_lsn: Lsn,
-}
-
-impl PersistedPeerInfo {
-    pub fn new() -> Self {
-        Self {
-            backup_lsn: Lsn::INVALID,
-            term: safekeeper_api::INITIAL_TERM,
-            flush_lsn: Lsn(0),
-            commit_lsn: Lsn(0),
-        }
-    }
-}
-
-// make clippy happy
-impl Default for PersistedPeerInfo {
-    fn default() -> Self {
-        Self::new()
-    }
-}
-
-/// Note: SafekeeperStateVn is old name for TimelinePersistentStateVn.
-#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
-pub struct TimelinePersistentStateV9 {
-    #[serde(with = "hex")]
-    pub tenant_id: TenantId,
-    #[serde(with = "hex")]
-    pub timeline_id: TimelineId,
-    /// persistent acceptor state
-    pub acceptor_state: AcceptorState,
-    /// information about server
-    pub server: ServerInfo,
-    /// Unique id of the last *elected* proposer we dealt with. Not needed
-    /// for correctness, exists for monitoring purposes.
-    #[serde(with = "hex")]
-    pub proposer_uuid: PgUuid,
-    /// Since which LSN this timeline generally starts. Safekeeper might have
-    /// joined later.
-    pub timeline_start_lsn: Lsn,
-    /// Since which LSN safekeeper has (had) WAL for this timeline.
-    /// All WAL segments next to one containing local_start_lsn are
-    /// filled with data from the beginning.
-    pub local_start_lsn: Lsn,
-    /// Part of WAL acknowledged by quorum *and available locally*. Always points
-    /// to record boundary.
-    pub commit_lsn: Lsn,
-    /// LSN that points to the end of the last backed up segment. Useful to
-    /// persist to avoid finding out offloading progress on boot.
-    pub backup_lsn: Lsn,
-    /// Minimal LSN which may be needed for recovery of some safekeeper (end_lsn
-    /// of last record streamed to everyone). Persisting it helps skipping
-    /// recovery in walproposer, generally we compute it from peers. In
-    /// walproposer proto called 'truncate_lsn'. Updates are currently drived
-    /// only by walproposer.
-    pub peer_horizon_lsn: Lsn,
-    /// LSN of the oldest known checkpoint made by pageserver and successfully
-    /// pushed to s3. We don't remove WAL beyond it. Persisted only for
-    /// informational purposes, we receive it from pageserver (or broker).
-    pub remote_consistent_lsn: Lsn,
-    /// Peers and their state as we remember it. Knowing peers themselves is
-    /// fundamental; but state is saved here only for informational purposes and
-    /// obviously can be stale. (Currently not saved at all, but let's provision
-    /// place to have less file version upgrades).
-    pub peers: PersistedPeers,
-    /// Holds names of partial segments uploaded to remote storage. Used to
-    /// clean up old objects without leaving garbage in remote storage.
-    pub partial_backup: wal_backup_partial::State,
-    /// Eviction state of the timeline. If it's Offloaded, we should download
-    /// WAL files from remote storage to serve the timeline.
-    pub eviction_state: EvictionState,
-}
-
 pub fn upgrade_control_file(buf: &[u8], version: u32) -> Result<TimelinePersistentState> {
     // migrate to storing full term history
     if version == 1 {
@@ -337,7 +248,6 @@ pub fn upgrade_control_file(buf: &[u8], version: u32) -> Result<TimelinePersiste
         return Ok(TimelinePersistentState {
             tenant_id: oldstate.server.tenant_id,
             timeline_id: oldstate.server.timeline_id,
-            mconf: Configuration::empty(),
             acceptor_state: ac,
             server: ServerInfo {
                 pg_version: oldstate.server.pg_version,
@@ -351,9 +261,9 @@ pub fn upgrade_control_file(buf: &[u8], version: u32) -> Result<TimelinePersiste
             backup_lsn: Lsn(0),
             peer_horizon_lsn: oldstate.truncate_lsn,
             remote_consistent_lsn: Lsn(0),
+            peers: PersistedPeers(vec![]),
             partial_backup: wal_backup_partial::State::default(),
             eviction_state: EvictionState::Present,
-            creation_ts: std::time::SystemTime::UNIX_EPOCH,
         });
     // migrate to hexing some ids
     } else if version == 2 {
@@ -367,7 +277,6 @@ pub fn upgrade_control_file(buf: &[u8], version: u32) -> Result<TimelinePersiste
         return Ok(TimelinePersistentState {
             tenant_id: oldstate.server.tenant_id,
             timeline_id: oldstate.server.timeline_id,
-            mconf: Configuration::empty(),
             acceptor_state: oldstate.acceptor_state,
             server,
             proposer_uuid: oldstate.proposer_uuid,
@@ -377,9 +286,9 @@ pub fn upgrade_control_file(buf: &[u8], version: u32) -> Result<TimelinePersiste
             backup_lsn: Lsn(0),
             peer_horizon_lsn: oldstate.truncate_lsn,
             remote_consistent_lsn: Lsn(0),
+            peers: PersistedPeers(vec![]),
             partial_backup: wal_backup_partial::State::default(),
             eviction_state: EvictionState::Present,
-            creation_ts: std::time::SystemTime::UNIX_EPOCH,
         });
     // migrate to moving tenant_id/timeline_id to the top and adding some lsns
     } else if version == 3 {
@@ -393,7 +302,6 @@ pub fn upgrade_control_file(buf: &[u8], version: u32) -> Result<TimelinePersiste
         return Ok(TimelinePersistentState {
             tenant_id: oldstate.server.tenant_id,
             timeline_id: oldstate.server.timeline_id,
-            mconf: Configuration::empty(),
             acceptor_state: oldstate.acceptor_state,
             server,
             proposer_uuid: oldstate.proposer_uuid,
@@ -403,9 +311,9 @@ pub fn upgrade_control_file(buf: &[u8], version: u32) -> Result<TimelinePersiste
             backup_lsn: Lsn(0),
             peer_horizon_lsn: oldstate.truncate_lsn,
             remote_consistent_lsn: Lsn(0),
+            peers: PersistedPeers(vec![]),
             partial_backup: wal_backup_partial::State::default(),
             eviction_state: EvictionState::Present,
-            creation_ts: std::time::SystemTime::UNIX_EPOCH,
         });
     // migrate to having timeline_start_lsn
     } else if version == 4 {
@@ -419,7 +327,6 @@ pub fn upgrade_control_file(buf: &[u8], version: u32) -> Result<TimelinePersiste
         return Ok(TimelinePersistentState {
             tenant_id: oldstate.tenant_id,
             timeline_id: oldstate.timeline_id,
-            mconf: Configuration::empty(),
             acceptor_state: oldstate.acceptor_state,
             server,
             proposer_uuid: oldstate.proposer_uuid,
@@ -429,9 +336,9 @@ pub fn upgrade_control_file(buf: &[u8], version: u32) -> Result<TimelinePersiste
             backup_lsn: Lsn::INVALID,
             peer_horizon_lsn: oldstate.peer_horizon_lsn,
             remote_consistent_lsn: Lsn(0),
+            peers: PersistedPeers(vec![]),
             partial_backup: wal_backup_partial::State::default(),
             eviction_state: EvictionState::Present,
-            creation_ts: std::time::SystemTime::UNIX_EPOCH,
         });
     } else if version == 5 {
         info!("reading safekeeper control file version {}", version);
@@ -465,7 +372,6 @@ pub fn upgrade_control_file(buf: &[u8], version: u32) -> Result<TimelinePersiste
         return Ok(TimelinePersistentState {
             tenant_id: oldstate.tenant_id,
             timeline_id: oldstate.timeline_id,
-            mconf: Configuration::empty(),
             acceptor_state: oldstate.acceptor_state,
             server: oldstate.server,
             proposer_uuid: oldstate.proposer_uuid,
@@ -475,9 +381,9 @@ pub fn upgrade_control_file(buf: &[u8], version: u32) -> Result<TimelinePersiste
             backup_lsn: oldstate.backup_lsn,
             peer_horizon_lsn: oldstate.peer_horizon_lsn,
             remote_consistent_lsn: oldstate.remote_consistent_lsn,
+            peers: oldstate.peers,
             partial_backup: wal_backup_partial::State::default(),
             eviction_state: EvictionState::Present,
-            creation_ts: std::time::SystemTime::UNIX_EPOCH,
         });
     } else if version == 8 {
         let oldstate = SafeKeeperStateV8::des(&buf[..buf.len()])?;
@@ -485,7 +391,6 @@ pub fn upgrade_control_file(buf: &[u8], version: u32) -> Result<TimelinePersiste
         return Ok(TimelinePersistentState {
             tenant_id: oldstate.tenant_id,
             timeline_id: oldstate.timeline_id,
-            mconf: Configuration::empty(),
             acceptor_state: oldstate.acceptor_state,
             server: oldstate.server,
             proposer_uuid: oldstate.proposer_uuid,
@@ -495,28 +400,9 @@ pub fn upgrade_control_file(buf: &[u8], version: u32) -> Result<TimelinePersiste
             backup_lsn: oldstate.backup_lsn,
             peer_horizon_lsn: oldstate.peer_horizon_lsn,
             remote_consistent_lsn: oldstate.remote_consistent_lsn,
+            peers: oldstate.peers,
             partial_backup: oldstate.partial_backup,
             eviction_state: EvictionState::Present,
-            creation_ts: std::time::SystemTime::UNIX_EPOCH,
-        });
-    } else if version == 9 {
-        let oldstate = TimelinePersistentStateV9::des(&buf[..buf.len()])?;
-        return Ok(TimelinePersistentState {
-            tenant_id: oldstate.tenant_id,
-            timeline_id: oldstate.timeline_id,
-            mconf: Configuration::empty(),
-            acceptor_state: oldstate.acceptor_state,
-            server: oldstate.server,
-            proposer_uuid: oldstate.proposer_uuid,
-            timeline_start_lsn: oldstate.timeline_start_lsn,
-            local_start_lsn: oldstate.local_start_lsn,
-            commit_lsn: oldstate.commit_lsn,
-            backup_lsn: oldstate.backup_lsn,
-            peer_horizon_lsn: oldstate.peer_horizon_lsn,
-            remote_consistent_lsn: oldstate.remote_consistent_lsn,
-            partial_backup: oldstate.partial_backup,
-            eviction_state: oldstate.eviction_state,
-            creation_ts: std::time::SystemTime::UNIX_EPOCH,
         });
     }
 
@@ -526,11 +412,9 @@ pub fn upgrade_control_file(buf: &[u8], version: u32) -> Result<TimelinePersiste
     bail!("unsupported safekeeper control file version {}", version)
 }
 
-// Used as a temp hack to make forward compatibility test work. Should be
-// removed after PR adding v10 is merged.
-pub fn downgrade_v10_to_v9(state: &TimelinePersistentState) -> TimelinePersistentStateV9 {
-    assert!(state.mconf.generation == INVALID_GENERATION);
-    TimelinePersistentStateV9 {
+pub fn downgrade_v9_to_v8(state: &TimelinePersistentState) -> SafeKeeperStateV8 {
+    assert!(state.eviction_state == EvictionState::Present);
+    SafeKeeperStateV8 {
         tenant_id: state.tenant_id,
         timeline_id: state.timeline_id,
         acceptor_state: state.acceptor_state.clone(),
@@ -542,9 +426,8 @@ pub fn downgrade_v10_to_v9(state: &TimelinePersistentState) -> TimelinePersisten
         backup_lsn: state.backup_lsn,
         peer_horizon_lsn: state.peer_horizon_lsn,
         remote_consistent_lsn: state.remote_consistent_lsn,
-        peers: PersistedPeers(vec![]),
+        peers: state.peers.clone(),
         partial_backup: state.partial_backup.clone(),
-        eviction_state: state.eviction_state,
     }
 }
 
@@ -554,7 +437,7 @@ mod tests {
 
     use utils::{id::NodeId, Hex};
 
-    use crate::control_file_upgrade::PersistedPeerInfo;
+    use crate::safekeeper::PersistedPeerInfo;
 
     use super::*;
 

safekeeper/src/copy_timeline.rs

@@ -1,7 +1,6 @@
 use anyhow::{bail, Result};
 use camino::Utf8PathBuf;
 use postgres_ffi::{MAX_SEND_SIZE, WAL_SEGMENT_SIZE};
-use safekeeper_api::membership::Configuration;
 use std::sync::Arc;
 use tokio::{
     fs::OpenOptions,
@@ -148,10 +147,10 @@ pub async fn handle_request(
 
     let mut new_state = TimelinePersistentState::new(
         &request.destination_ttid,
-        Configuration::empty(),
         state.server.clone(),
-        start_lsn,
+        vec![],
         request.until_lsn,
+        start_lsn,
     )?;
     new_state.timeline_start_lsn = start_lsn;
     new_state.peer_horizon_lsn = request.until_lsn;

safekeeper/src/handler.rs

@@ -52,70 +52,16 @@ pub struct SafekeeperPostgresHandler {
 
 /// Parsed Postgres command.
 enum SafekeeperPostgresCommand {
-    StartWalPush {
-        proto_version: u32,
-        // Eventually timelines will be always created explicitly by storcon.
-        // This option allows legacy behaviour for compute to do that until we
-        // fully migrate.
-        allow_timeline_creation: bool,
-    },
-    StartReplication {
-        start_lsn: Lsn,
-        term: Option<Term>,
-    },
+    StartWalPush,
+    StartReplication { start_lsn: Lsn, term: Option<Term> },
     IdentifySystem,
     TimelineStatus,
-    JSONCtrl {
-        cmd: AppendLogicalMessage,
-    },
+    JSONCtrl { cmd: AppendLogicalMessage },
 }
 
 fn parse_cmd(cmd: &str) -> anyhow::Result<SafekeeperPostgresCommand> {
     if cmd.starts_with("START_WAL_PUSH") {
-        // Allow additional options in postgres START_REPLICATION style like
-        //   START_WAL_PUSH (proto_version '3', allow_timeline_creation 'false').
-        // Parsing here is very naive and breaks in case of commas or
-        // whitespaces in values, but enough for our purposes.
-        let re = Regex::new(r"START_WAL_PUSH(\s+?\((.*)\))?").unwrap();
-        let caps = re
-            .captures(cmd)
-            .context(format!("failed to parse START_WAL_PUSH command {}", cmd))?;
-        // capture () content
-        let options = caps.get(2).map(|m| m.as_str()).unwrap_or("");
-        // default values
-        let mut proto_version = 2;
-        let mut allow_timeline_creation = true;
-        for kvstr in options.split(",") {
-            if kvstr.is_empty() {
-                continue;
-            }
-            let mut kvit = kvstr.split_whitespace();
-            let key = kvit.next().context(format!(
-                "failed to parse key in kv {} in command {}",
-                kvstr, cmd
-            ))?;
-            let value = kvit.next().context(format!(
-                "failed to parse value in kv {} in command {}",
-                kvstr, cmd
-            ))?;
-            let value_trimmed = value.trim_matches('\'');
-            if key == "proto_version" {
-                proto_version = value_trimmed.parse::<u32>().context(format!(
-                    "failed to parse proto_version value {} in command {}",
-                    value, cmd
-                ))?;
-            }
-            if key == "allow_timeline_creation" {
-                allow_timeline_creation = value_trimmed.parse::<bool>().context(format!(
-                    "failed to parse allow_timeline_creation value {} in command {}",
-                    value, cmd
-                ))?;
-            }
-        }
-        Ok(SafekeeperPostgresCommand::StartWalPush {
-            proto_version,
-            allow_timeline_creation,
-        })
+        Ok(SafekeeperPostgresCommand::StartWalPush)
     } else if cmd.starts_with("START_REPLICATION") {
         let re = Regex::new(
             // We follow postgres START_REPLICATION LOGICAL options to pass term.
@@ -149,7 +95,7 @@ fn parse_cmd(cmd: &str) -> anyhow::Result<SafekeeperPostgresCommand> {
 
 fn cmd_to_string(cmd: &SafekeeperPostgresCommand) -> &str {
     match cmd {
-        SafekeeperPostgresCommand::StartWalPush { .. } => "START_WAL_PUSH",
+        SafekeeperPostgresCommand::StartWalPush => "START_WAL_PUSH",
         SafekeeperPostgresCommand::StartReplication { .. } => "START_REPLICATION",
         SafekeeperPostgresCommand::TimelineStatus => "TIMELINE_STATUS",
         SafekeeperPostgresCommand::IdentifySystem => "IDENTIFY_SYSTEM",
@@ -347,11 +293,8 @@ impl<IO: AsyncRead + AsyncWrite + Unpin + Send> postgres_backend::Handler<IO>
             self.ttid = TenantTimelineId::new(tenant_id, timeline_id);
 
             match cmd {
-                SafekeeperPostgresCommand::StartWalPush {
-                    proto_version,
-                    allow_timeline_creation,
-                } => {
-                    self.handle_start_wal_push(pgb, proto_version, allow_timeline_creation)
+                SafekeeperPostgresCommand::StartWalPush => {
+                    self.handle_start_wal_push(pgb)
                         .instrument(info_span!("WAL receiver"))
                         .await
                 }
@@ -524,39 +467,3 @@ impl SafekeeperPostgresHandler {
         }
     }
 }
-
-#[cfg(test)]
-mod tests {
-    use super::SafekeeperPostgresCommand;
-
-    /// Test parsing of START_WAL_PUSH command
-    #[test]
-    fn test_start_wal_push_parse() {
-        let cmd = "START_WAL_PUSH";
-        let parsed = super::parse_cmd(cmd).expect("failed to parse");
-        match parsed {
-            SafekeeperPostgresCommand::StartWalPush {
-                proto_version,
-                allow_timeline_creation,
-            } => {
-                assert_eq!(proto_version, 2);
-                assert!(allow_timeline_creation);
-            }
-            _ => panic!("unexpected command"),
-        }
-
-        let cmd =
-            "START_WAL_PUSH (proto_version '3', allow_timeline_creation 'false', unknown 'hoho')";
-        let parsed = super::parse_cmd(cmd).expect("failed to parse");
-        match parsed {
-            SafekeeperPostgresCommand::StartWalPush {
-                proto_version,
-                allow_timeline_creation,
-            } => {
-                assert_eq!(proto_version, 3);
-                assert!(!allow_timeline_creation);
-            }
-            _ => panic!("unexpected command"),
-        }
-    }
-}

safekeeper/src/http/routes.rs

@@ -1,5 +1,4 @@
 use hyper::{Body, Request, Response, StatusCode};
-use safekeeper_api::models;
 use safekeeper_api::models::AcceptorStateStatus;
 use safekeeper_api::models::SafekeeperStatus;
 use safekeeper_api::models::TermSwitchApiEntry;
@@ -112,15 +111,14 @@ async fn timeline_create_handler(mut request: Request<Body>) -> Result<Response<
         system_id: request_data.system_id.unwrap_or(0),
         wal_seg_size: request_data.wal_seg_size.unwrap_or(WAL_SEGMENT_SIZE as u32),
     };
+    let local_start_lsn = request_data.local_start_lsn.unwrap_or_else(|| {
+        request_data
+            .commit_lsn
+            .segment_lsn(server_info.wal_seg_size as usize)
+    });
     let global_timelines = get_global_timelines(&request);
     global_timelines
-        .create(
-            ttid,
-            request_data.mconf,
-            server_info,
-            request_data.start_lsn,
-            request_data.commit_lsn.unwrap_or(request_data.start_lsn),
-        )
+        .create(ttid, server_info, request_data.commit_lsn, local_start_lsn)
         .await
         .map_err(ApiError::InternalServerError)?;
 
@@ -184,7 +182,6 @@ async fn timeline_status_handler(request: Request<Body>) -> Result<Response<Body
     let status = TimelineStatus {
         tenant_id: ttid.tenant_id,
         timeline_id: ttid.timeline_id,
-        mconf: state.mconf,
         acceptor_state: acc_state,
         pg_info: state.server,
         flush_lsn,
@@ -195,7 +192,7 @@ async fn timeline_status_handler(request: Request<Body>) -> Result<Response<Body
         peer_horizon_lsn: inmem.peer_horizon_lsn,
         remote_consistent_lsn: inmem.remote_consistent_lsn,
         peers: tli.get_peers(conf).await,
-        walsenders: tli.get_walsenders().get_all_public(),
+        walsenders: tli.get_walsenders().get_all(),
         walreceivers: tli.get_walreceivers().get_all(),
     };
     json_response(StatusCode::OK, status)
@@ -270,28 +267,6 @@ async fn timeline_snapshot_handler(request: Request<Body>) -> Result<Response<Bo
     Ok(response)
 }
 
-/// Consider switching timeline membership configuration to the provided one.
-async fn timeline_membership_handler(
-    mut request: Request<Body>,
-) -> Result<Response<Body>, ApiError> {
-    let ttid = TenantTimelineId::new(
-        parse_request_param(&request, "tenant_id")?,
-        parse_request_param(&request, "timeline_id")?,
-    );
-    check_permission(&request, Some(ttid.tenant_id))?;
-
-    let global_timelines = get_global_timelines(&request);
-    let tli = global_timelines.get(ttid).map_err(ApiError::from)?;
-
-    let data: models::TimelineMembershipSwitchRequest = json_request(&mut request).await?;
-    let response = tli
-        .membership_switch(data.mconf)
-        .await
-        .map_err(ApiError::InternalServerError)?;
-
-    json_response(StatusCode::OK, response)
-}
-
 async fn timeline_copy_handler(mut request: Request<Body>) -> Result<Response<Body>, ApiError> {
     check_permission(&request, None)?;
 
@@ -643,10 +618,6 @@ pub fn make_router(
             "/v1/tenant/:tenant_id/timeline/:timeline_id/snapshot/:destination_id",
             |r| request_span(r, timeline_snapshot_handler),
         )
-        .post(
-            "/v1/tenant/:tenant_id/timeline/:timeline_id/membership",
-            |r| request_span(r, timeline_membership_handler),
-        )
         .post(
             "/v1/tenant/:tenant_id/timeline/:source_timeline_id/copy",
             |r| request_span(r, timeline_copy_handler),

safekeeper/src/json_ctrl.rs

@@ -8,7 +8,6 @@
 
 use anyhow::Context;
 use postgres_backend::QueryError;
-use safekeeper_api::membership::Configuration;
 use safekeeper_api::{ServerInfo, Term};
 use serde::{Deserialize, Serialize};
 use tokio::io::{AsyncRead, AsyncWrite};
@@ -106,7 +105,6 @@ async fn prepare_safekeeper(
         .global_timelines
         .create(
             spg.ttid,
-            Configuration::empty(),
             ServerInfo {
                 pg_version,
                 wal_seg_size: WAL_SEGMENT_SIZE as u32,

safekeeper/src/lib.rs

@@ -108,8 +108,6 @@ pub struct SafeKeeperConf {
     pub control_file_save_interval: Duration,
     pub partial_backup_concurrency: usize,
     pub eviction_min_resident: Duration,
-    pub wal_reader_fanout: bool,
-    pub max_delta_for_fanout: Option<u64>,
 }
 
 impl SafeKeeperConf {
@@ -152,8 +150,6 @@ impl SafeKeeperConf {
             control_file_save_interval: Duration::from_secs(1),
             partial_backup_concurrency: 1,
             eviction_min_resident: Duration::ZERO,
-            wal_reader_fanout: false,
-            max_delta_for_fanout: None,
         }
     }
 }

safekeeper/src/metrics.rs

@@ -12,9 +12,9 @@ use metrics::{
     pow2_buckets,
     proto::MetricFamily,
     register_histogram, register_histogram_vec, register_int_counter, register_int_counter_pair,
-    register_int_counter_pair_vec, register_int_counter_vec, register_int_gauge,
-    register_int_gauge_vec, Gauge, GaugeVec, Histogram, HistogramVec, IntCounter, IntCounterPair,
-    IntCounterPairVec, IntCounterVec, IntGauge, IntGaugeVec, DISK_FSYNC_SECONDS_BUCKETS,
+    register_int_counter_pair_vec, register_int_counter_vec, register_int_gauge, Gauge, GaugeVec,
+    Histogram, HistogramVec, IntCounter, IntCounterPair, IntCounterPairVec, IntCounterVec,
+    IntGauge, IntGaugeVec, DISK_FSYNC_SECONDS_BUCKETS,
 };
 use once_cell::sync::Lazy;
 use postgres_ffi::XLogSegNo;
@@ -211,14 +211,6 @@ pub static WAL_RECEIVERS: Lazy<IntGauge> = Lazy::new(|| {
     )
     .expect("Failed to register safekeeper_wal_receivers")
 });
-pub static WAL_READERS: Lazy<IntGaugeVec> = Lazy::new(|| {
-    register_int_gauge_vec!(
-        "safekeeper_wal_readers",
-        "Number of active WAL readers (may serve pageservers or other safekeepers)",
-        &["kind", "target"]
-    )
-    .expect("Failed to register safekeeper_wal_receivers")
-});
 pub static WAL_RECEIVER_QUEUE_DEPTH: Lazy<Histogram> = Lazy::new(|| {
     // Use powers of two buckets, but add a bucket at 0 and the max queue size to track empty and
     // full queues respectively.
@@ -451,7 +443,6 @@ pub struct FullTimelineInfo {
     pub timeline_is_active: bool,
     pub num_computes: u32,
     pub last_removed_segno: XLogSegNo,
-    pub interpreted_wal_reader_tasks: usize,
 
     pub epoch_start_lsn: Lsn,
     pub mem_state: TimelineMemState,
@@ -481,7 +472,6 @@ pub struct TimelineCollector {
     disk_usage: GenericGaugeVec<AtomicU64>,
     acceptor_term: GenericGaugeVec<AtomicU64>,
     written_wal_bytes: GenericGaugeVec<AtomicU64>,
-    interpreted_wal_reader_tasks: GenericGaugeVec<AtomicU64>,
     written_wal_seconds: GaugeVec,
     flushed_wal_seconds: GaugeVec,
     collect_timeline_metrics: Gauge,
@@ -680,16 +670,6 @@ impl TimelineCollector {
         .unwrap();
         descs.extend(active_timelines_count.desc().into_iter().cloned());
 
-        let interpreted_wal_reader_tasks = GenericGaugeVec::new(
-            Opts::new(
-                "safekeeper_interpreted_wal_reader_tasks",
-                "Number of active interpreted wal reader tasks, grouped by timeline",
-            ),
-            &["tenant_id", "timeline_id"],
-        )
-        .unwrap();
-        descs.extend(interpreted_wal_reader_tasks.desc().into_iter().cloned());
-
         TimelineCollector {
             global_timelines,
             descs,
@@ -713,7 +693,6 @@ impl TimelineCollector {
             collect_timeline_metrics,
             timelines_count,
             active_timelines_count,
-            interpreted_wal_reader_tasks,
         }
     }
 }
@@ -742,7 +721,6 @@ impl Collector for TimelineCollector {
         self.disk_usage.reset();
         self.acceptor_term.reset();
         self.written_wal_bytes.reset();
-        self.interpreted_wal_reader_tasks.reset();
         self.written_wal_seconds.reset();
         self.flushed_wal_seconds.reset();
 
@@ -804,9 +782,6 @@ impl Collector for TimelineCollector {
             self.written_wal_bytes
                 .with_label_values(labels)
                 .set(tli.wal_storage.write_wal_bytes);
-            self.interpreted_wal_reader_tasks
-                .with_label_values(labels)
-                .set(tli.interpreted_wal_reader_tasks as u64);
             self.written_wal_seconds
                 .with_label_values(labels)
                 .set(tli.wal_storage.write_wal_seconds);
@@ -859,7 +834,6 @@ impl Collector for TimelineCollector {
         mfs.extend(self.disk_usage.collect());
         mfs.extend(self.acceptor_term.collect());
         mfs.extend(self.written_wal_bytes.collect());
-        mfs.extend(self.interpreted_wal_reader_tasks.collect());
         mfs.extend(self.written_wal_seconds.collect());
         mfs.extend(self.flushed_wal_seconds.collect());
 

safekeeper/src/receive_wal.rs

@@ -21,7 +21,6 @@ use postgres_backend::PostgresBackend;
 use postgres_backend::PostgresBackendReader;
 use postgres_backend::QueryError;
 use pq_proto::BeMessage;
-use safekeeper_api::membership::Configuration;
 use safekeeper_api::models::{ConnectionId, WalReceiverState, WalReceiverStatus};
 use safekeeper_api::ServerInfo;
 use std::future;
@@ -200,14 +199,9 @@ impl SafekeeperPostgresHandler {
     pub async fn handle_start_wal_push<IO: AsyncRead + AsyncWrite + Unpin>(
         &mut self,
         pgb: &mut PostgresBackend<IO>,
-        proto_version: u32,
-        allow_timeline_creation: bool,
     ) -> Result<(), QueryError> {
         let mut tli: Option<WalResidentTimeline> = None;
-        if let Err(end) = self
-            .handle_start_wal_push_guts(pgb, &mut tli, proto_version, allow_timeline_creation)
-            .await
-        {
+        if let Err(end) = self.handle_start_wal_push_guts(pgb, &mut tli).await {
             // Log the result and probably send it to the client, closing the stream.
             let handle_end_fut = pgb.handle_copy_stream_end(end);
             // If we managed to create the timeline, augment logging with current LSNs etc.
@@ -227,8 +221,6 @@ impl SafekeeperPostgresHandler {
         &mut self,
         pgb: &mut PostgresBackend<IO>,
         tli: &mut Option<WalResidentTimeline>,
-        proto_version: u32,
-        allow_timeline_creation: bool,
     ) -> Result<(), CopyStreamHandlerEnd> {
         // The `tli` parameter is only used for passing _out_ a timeline, one should
         // not have been passed in.
@@ -257,17 +249,12 @@ impl SafekeeperPostgresHandler {
             conn_id: self.conn_id,
             pgb_reader: &mut pgb_reader,
             peer_addr,
-            proto_version,
             acceptor_handle: &mut acceptor_handle,
             global_timelines: self.global_timelines.clone(),
         };
 
-        // Read first message and create timeline if needed and allowed. This
-        // won't be when timelines will be always created by storcon and
-        // allow_timeline_creation becomes false.
-        let res = network_reader
-            .read_first_message(allow_timeline_creation)
-            .await;
+        // Read first message and create timeline if needed.
+        let res = network_reader.read_first_message().await;
 
         let network_res = if let Ok((timeline, next_msg)) = res {
             let pageserver_feedback_rx: tokio::sync::broadcast::Receiver<PageserverFeedback> =
@@ -325,7 +312,6 @@ struct NetworkReader<'a, IO> {
     conn_id: ConnectionId,
     pgb_reader: &'a mut PostgresBackendReader<IO>,
     peer_addr: SocketAddr,
-    proto_version: u32,
     // WalAcceptor is spawned when we learn server info from walproposer and
     // create timeline; handle is put here.
     acceptor_handle: &'a mut Option<JoinHandle<anyhow::Result<()>>>,
@@ -335,10 +321,9 @@ struct NetworkReader<'a, IO> {
 impl<IO: AsyncRead + AsyncWrite + Unpin> NetworkReader<'_, IO> {
     async fn read_first_message(
         &mut self,
-        allow_timeline_creation: bool,
     ) -> Result<(WalResidentTimeline, ProposerAcceptorMessage), CopyStreamHandlerEnd> {
         // Receive information about server to create timeline, if not yet.
-        let next_msg = read_message(self.pgb_reader, self.proto_version).await?;
+        let next_msg = read_message(self.pgb_reader).await?;
         let tli = match next_msg {
             ProposerAcceptorMessage::Greeting(ref greeting) => {
                 info!(
@@ -350,22 +335,11 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> NetworkReader<'_, IO> {
                     system_id: greeting.system_id,
                     wal_seg_size: greeting.wal_seg_size,
                 };
-                let tli = if allow_timeline_creation {
-                    self.global_timelines
-                        .create(
-                            self.ttid,
-                            Configuration::empty(),
-                            server_info,
-                            Lsn::INVALID,
-                            Lsn::INVALID,
-                        )
-                        .await
-                        .context("create timeline")?
-                } else {
-                    self.global_timelines
-                        .get(self.ttid)
-                        .context("get timeline")?
-                };
+                let tli = self
+                    .global_timelines
+                    .create(self.ttid, server_info, Lsn::INVALID, Lsn::INVALID)
+                    .await
+                    .context("create timeline")?;
                 tli.wal_residence_guard().await?
             }
             _ => {
@@ -394,7 +368,7 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> NetworkReader<'_, IO> {
         ));
 
         // Forward all messages to WalAcceptor
-        read_network_loop(self.pgb_reader, msg_tx, next_msg, self.proto_version).await
+        read_network_loop(self.pgb_reader, msg_tx, next_msg).await
     }
 }
 
@@ -402,10 +376,9 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> NetworkReader<'_, IO> {
 /// TODO: Return Ok(None) on graceful termination.
 async fn read_message<IO: AsyncRead + AsyncWrite + Unpin>(
     pgb_reader: &mut PostgresBackendReader<IO>,
-    proto_version: u32,
 ) -> Result<ProposerAcceptorMessage, CopyStreamHandlerEnd> {
     let copy_data = pgb_reader.read_copy_message().await?;
-    let msg = ProposerAcceptorMessage::parse(copy_data, proto_version)?;
+    let msg = ProposerAcceptorMessage::parse(copy_data)?;
     Ok(msg)
 }
 
@@ -413,7 +386,6 @@ async fn read_network_loop<IO: AsyncRead + AsyncWrite + Unpin>(
     pgb_reader: &mut PostgresBackendReader<IO>,
     msg_tx: Sender<ProposerAcceptorMessage>,
     mut next_msg: ProposerAcceptorMessage,
-    proto_version: u32,
 ) -> Result<(), CopyStreamHandlerEnd> {
     /// Threshold for logging slow WalAcceptor sends.
     const SLOW_THRESHOLD: Duration = Duration::from_secs(5);
@@ -446,7 +418,7 @@ async fn read_network_loop<IO: AsyncRead + AsyncWrite + Unpin>(
         WAL_RECEIVER_QUEUE_DEPTH_TOTAL.inc();
         WAL_RECEIVER_QUEUE_SIZE_TOTAL.add(size as i64);
 
-        next_msg = read_message(pgb_reader, proto_version).await?;
+        next_msg = read_message(pgb_reader).await?;
     }
 }
 

safekeeper/src/safekeeper.rs

@@ -7,6 +7,7 @@ use bytes::{Buf, BufMut, Bytes, BytesMut};
 use postgres_ffi::{TimeLineID, MAX_SEND_SIZE};
 use safekeeper_api::models::HotStandbyFeedback;
 use safekeeper_api::Term;
+use safekeeper_api::INVALID_TERM;
 use serde::{Deserialize, Serialize};
 use std::cmp::max;
 use std::cmp::min;
@@ -29,7 +30,7 @@ use utils::{
     lsn::Lsn,
 };
 
-pub const SK_PROTOCOL_VERSION: u32 = 2;
+const SK_PROTOCOL_VERSION: u32 = 2;
 pub const UNKNOWN_SERVER_VERSION: u32 = 0;
 
 #[derive(Debug, Clone, Copy, Serialize, Deserialize, PartialEq, Eq, PartialOrd, Ord)]
@@ -192,6 +193,36 @@ impl AcceptorState {
     }
 }
 
+#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
+pub struct PersistedPeerInfo {
+    /// LSN up to which safekeeper offloaded WAL to s3.
+    pub backup_lsn: Lsn,
+    /// Term of the last entry.
+    pub term: Term,
+    /// LSN of the last record.
+    pub flush_lsn: Lsn,
+    /// Up to which LSN safekeeper regards its WAL as committed.
+    pub commit_lsn: Lsn,
+}
+
+impl PersistedPeerInfo {
+    pub fn new() -> Self {
+        Self {
+            backup_lsn: Lsn::INVALID,
+            term: INVALID_TERM,
+            flush_lsn: Lsn(0),
+            commit_lsn: Lsn(0),
+        }
+    }
+}
+
+// make clippy happy
+impl Default for PersistedPeerInfo {
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
 // protocol messages
 
 /// Initial Proposer -> Acceptor message
@@ -317,14 +348,7 @@ pub enum ProposerAcceptorMessage {
 
 impl ProposerAcceptorMessage {
     /// Parse proposer message.
-    pub fn parse(msg_bytes: Bytes, proto_version: u32) -> Result<ProposerAcceptorMessage> {
-        if proto_version != SK_PROTOCOL_VERSION {
-            bail!(
-                "incompatible protocol version {}, expected {}",
-                proto_version,
-                SK_PROTOCOL_VERSION
-            );
-        }
+    pub fn parse(msg_bytes: Bytes) -> Result<ProposerAcceptorMessage> {
         // xxx using Reader is inefficient but easy to work with bincode
         let mut stream = msg_bytes.reader();
         // u64 is here to avoid padding; it will be removed once we stop packing C structs into the wire as is
@@ -986,7 +1010,7 @@ where
 
     /// Update commit_lsn from peer safekeeper data.
     pub async fn record_safekeeper_info(&mut self, sk_info: &SafekeeperTimelineInfo) -> Result<()> {
-        if Lsn(sk_info.commit_lsn) != Lsn::INVALID {
+        if (Lsn(sk_info.commit_lsn) != Lsn::INVALID) && (sk_info.last_log_term != INVALID_TERM) {
             // Note: the check is too restrictive, generally we can update local
             // commit_lsn if our history matches (is part of) history of advanced
             // commit_lsn provider.
@@ -1001,20 +1025,12 @@ where
 #[cfg(test)]
 mod tests {
     use futures::future::BoxFuture;
-
     use postgres_ffi::{XLogSegNo, WAL_SEGMENT_SIZE};
-    use safekeeper_api::{
-        membership::{Configuration, MemberSet, SafekeeperId},
-        ServerInfo,
-    };
+    use safekeeper_api::ServerInfo;
 
     use super::*;
-    use crate::state::{EvictionState, TimelinePersistentState};
-    use std::{
-        ops::Deref,
-        str::FromStr,
-        time::{Instant, UNIX_EPOCH},
-    };
+    use crate::state::{EvictionState, PersistedPeers, TimelinePersistentState};
+    use std::{ops::Deref, str::FromStr, time::Instant};
 
     // fake storage for tests
     struct InMemoryState {
@@ -1297,21 +1313,12 @@ mod tests {
 
     #[test]
     fn test_sk_state_bincode_serde_roundtrip() {
+        use utils::Hex;
         let tenant_id = TenantId::from_str("cf0480929707ee75372337efaa5ecf96").unwrap();
         let timeline_id = TimelineId::from_str("112ded66422aa5e953e5440fa5427ac4").unwrap();
         let state = TimelinePersistentState {
             tenant_id,
             timeline_id,
-            mconf: Configuration {
-                generation: 42,
-                members: MemberSet::new(vec![SafekeeperId {
-                    id: NodeId(1),
-                    host: "hehe.org".to_owned(),
-                    pg_port: 5432,
-                }])
-                .expect("duplicate member"),
-                new_members: None,
-            },
             acceptor_state: AcceptorState {
                 term: 42,
                 term_history: TermHistory(vec![TermLsn {
@@ -1335,13 +1342,70 @@ mod tests {
             backup_lsn: Lsn(1234567300),
             peer_horizon_lsn: Lsn(9999999),
             remote_consistent_lsn: Lsn(1234560000),
+            peers: PersistedPeers(vec![(
+                NodeId(1),
+                PersistedPeerInfo {
+                    backup_lsn: Lsn(1234567000),
+                    term: 42,
+                    flush_lsn: Lsn(1234567800 - 8),
+                    commit_lsn: Lsn(1234567600),
+                },
+            )]),
             partial_backup: crate::wal_backup_partial::State::default(),
             eviction_state: EvictionState::Present,
-            creation_ts: UNIX_EPOCH,
         };
 
         let ser = state.ser().unwrap();
 
+        #[rustfmt::skip]
+        let expected = [
+            // tenant_id as length prefixed hex
+            0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x63, 0x66, 0x30, 0x34, 0x38, 0x30, 0x39, 0x32, 0x39, 0x37, 0x30, 0x37, 0x65, 0x65, 0x37, 0x35, 0x33, 0x37, 0x32, 0x33, 0x33, 0x37, 0x65, 0x66, 0x61, 0x61, 0x35, 0x65, 0x63, 0x66, 0x39, 0x36,
+            // timeline_id as length prefixed hex
+            0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x31, 0x31, 0x32, 0x64, 0x65, 0x64, 0x36, 0x36, 0x34, 0x32, 0x32, 0x61, 0x61, 0x35, 0x65, 0x39, 0x35, 0x33, 0x65, 0x35, 0x34, 0x34, 0x30, 0x66, 0x61, 0x35, 0x34, 0x32, 0x37, 0x61, 0x63, 0x34,
+            // term
+            0x2a, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            // length prefix
+            0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            // unsure why this order is swapped
+            0x29, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            // pg_version
+            0x0e, 0x00, 0x00, 0x00,
+            // systemid
+            0x21, 0x43, 0x65, 0x87, 0x78, 0x56, 0x34, 0x12,
+            // wal_seg_size
+            0x78, 0x56, 0x34, 0x12,
+            // pguuid as length prefixed hex
+            0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x63, 0x34, 0x37, 0x61, 0x34, 0x32, 0x61, 0x35, 0x30, 0x66, 0x34, 0x34, 0x65, 0x35, 0x35, 0x33, 0x65, 0x39, 0x61, 0x35, 0x32, 0x61, 0x34, 0x32, 0x36, 0x36, 0x65, 0x64, 0x32, 0x64, 0x31, 0x31,
+
+            // timeline_start_lsn
+            0x00, 0x56, 0x34, 0x12, 0x00, 0x00, 0x00, 0x00,
+            0x12, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x78, 0x02, 0x96, 0x49, 0x00, 0x00, 0x00, 0x00,
+            0x84, 0x00, 0x96, 0x49, 0x00, 0x00, 0x00, 0x00,
+            0x7f, 0x96, 0x98, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x00, 0xe4, 0x95, 0x49, 0x00, 0x00, 0x00, 0x00,
+            // length prefix for persistentpeers
+            0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            // nodeid
+            0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            // backuplsn
+            0x58, 0xff, 0x95, 0x49, 0x00, 0x00, 0x00, 0x00,
+            0x2a, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x70, 0x02, 0x96, 0x49, 0x00, 0x00, 0x00, 0x00,
+            0xb0, 0x01, 0x96, 0x49, 0x00, 0x00, 0x00, 0x00,
+            // partial_backup
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            // eviction_state
+            0x00, 0x00, 0x00, 0x00,
+        ];
+
+        assert_eq!(Hex(&ser), Hex(&expected));
+
         let deser = TimelinePersistentState::des(&ser).unwrap();
 
         assert_eq!(deser, state);

safekeeper/src/send_interpreted_wal.rs

@@ -1,330 +1,96 @@
-use std::collections::HashMap;
-use std::fmt::Display;
-use std::sync::Arc;
 use std::time::Duration;
 
-use anyhow::{anyhow, Context};
-use futures::future::Either;
+use anyhow::Context;
 use futures::StreamExt;
 use pageserver_api::shard::ShardIdentity;
 use postgres_backend::{CopyStreamHandlerEnd, PostgresBackend};
-use postgres_ffi::waldecoder::WalDecodeError;
+use postgres_ffi::MAX_SEND_SIZE;
 use postgres_ffi::{get_current_timestamp, waldecoder::WalStreamDecoder};
 use pq_proto::{BeMessage, InterpretedWalRecordsBody, WalSndKeepAlive};
 use tokio::io::{AsyncRead, AsyncWrite};
-use tokio::sync::mpsc::error::SendError;
-use tokio::task::JoinHandle;
 use tokio::time::MissedTickBehavior;
-use tracing::{info_span, Instrument};
 use utils::lsn::Lsn;
 use utils::postgres_client::Compression;
 use utils::postgres_client::InterpretedFormat;
 use wal_decoder::models::{InterpretedWalRecord, InterpretedWalRecords};
 use wal_decoder::wire_format::ToWireFormat;
 
-use crate::metrics::WAL_READERS;
-use crate::send_wal::{EndWatchView, WalSenderGuard};
-use crate::timeline::WalResidentTimeline;
-use crate::wal_reader_stream::{StreamingWalReader, WalBytes};
+use crate::send_wal::EndWatchView;
+use crate::wal_reader_stream::{WalBytes, WalReaderStreamBuilder};
 
-/// Identifier used to differentiate between senders of the same
-/// shard.
-///
-/// In the steady state there's only one, but two pageservers may
-/// temporarily have the same shard attached and attempt to ingest
-/// WAL for it. See also [`ShardSenderId`].
-#[derive(Hash, Eq, PartialEq, Copy, Clone)]
-struct SenderId(u8);
-
-impl SenderId {
-    fn first() -> Self {
-        SenderId(0)
-    }
-
-    fn next(&self) -> Self {
-        SenderId(self.0.checked_add(1).expect("few senders"))
-    }
+/// Shard-aware interpreted record sender.
+/// This is used for sending WAL to the pageserver. Said WAL
+/// is pre-interpreted and filtered for the shard.
+pub(crate) struct InterpretedWalSender<'a, IO> {
+    pub(crate) format: InterpretedFormat,
+    pub(crate) compression: Option<Compression>,
+    pub(crate) pgb: &'a mut PostgresBackend<IO>,
+    pub(crate) wal_stream_builder: WalReaderStreamBuilder,
+    pub(crate) end_watch_view: EndWatchView,
+    pub(crate) shard: ShardIdentity,
+    pub(crate) pg_version: u32,
+    pub(crate) appname: Option<String>,
 }
 
-#[derive(Hash, Eq, PartialEq)]
-struct ShardSenderId {
-    shard: ShardIdentity,
-    sender_id: SenderId,
-}
-
-impl Display for ShardSenderId {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        write!(f, "{}{}", self.sender_id.0, self.shard.shard_slug())
-    }
-}
-
-impl ShardSenderId {
-    fn new(shard: ShardIdentity, sender_id: SenderId) -> Self {
-        ShardSenderId { shard, sender_id }
-    }
-
-    fn shard(&self) -> ShardIdentity {
-        self.shard
-    }
-}
-
-/// Shard-aware fan-out interpreted record reader.
-/// Reads WAL from disk, decodes it, intepretets it, and sends
-/// it to any [`InterpretedWalSender`] connected to it.
-/// Each [`InterpretedWalSender`] corresponds to one shard
-/// and gets interpreted records concerning that shard only.
-pub(crate) struct InterpretedWalReader {
-    wal_stream: StreamingWalReader,
-    shard_senders: HashMap<ShardIdentity, smallvec::SmallVec<[ShardSenderState; 1]>>,
-    shard_notification_rx: Option<tokio::sync::mpsc::UnboundedReceiver<AttachShardNotification>>,
-    state: Arc<std::sync::RwLock<InterpretedWalReaderState>>,
-    pg_version: u32,
-}
-
-/// A handle for [`InterpretedWalReader`] which allows for interacting with it
-/// when it runs as a separate tokio task.
-#[derive(Debug)]
-pub(crate) struct InterpretedWalReaderHandle {
-    join_handle: JoinHandle<Result<(), InterpretedWalReaderError>>,
-    state: Arc<std::sync::RwLock<InterpretedWalReaderState>>,
-    shard_notification_tx: tokio::sync::mpsc::UnboundedSender<AttachShardNotification>,
-}
-
-struct ShardSenderState {
-    sender_id: SenderId,
-    tx: tokio::sync::mpsc::Sender<Batch>,
-    next_record_lsn: Lsn,
-}
-
-/// State of [`InterpretedWalReader`] visible outside of the task running it.
-#[derive(Debug)]
-pub(crate) enum InterpretedWalReaderState {
-    Running { current_position: Lsn },
-    Done,
-}
-
-pub(crate) struct Batch {
+struct Batch {
     wal_end_lsn: Lsn,
     available_wal_end_lsn: Lsn,
     records: InterpretedWalRecords,
 }
 
-#[derive(thiserror::Error, Debug)]
-pub enum InterpretedWalReaderError {
-    /// Handler initiates the end of streaming.
-    #[error("decode error: {0}")]
-    Decode(#[from] WalDecodeError),
-    #[error("read or interpret error: {0}")]
-    ReadOrInterpret(#[from] anyhow::Error),
-    #[error("wal stream closed")]
-    WalStreamClosed,
-}
+impl<IO: AsyncRead + AsyncWrite + Unpin> InterpretedWalSender<'_, IO> {
+    /// Send interpreted WAL to a receiver.
+    /// Stops when an error occurs or the receiver is caught up and there's no active compute.
+    ///
+    /// Err(CopyStreamHandlerEnd) is always returned; Result is used only for ?
+    /// convenience.
+    pub(crate) async fn run(self) -> Result<(), CopyStreamHandlerEnd> {
+        let mut wal_position = self.wal_stream_builder.start_pos();
+        let mut wal_decoder =
+            WalStreamDecoder::new(self.wal_stream_builder.start_pos(), self.pg_version);
 
-impl InterpretedWalReaderState {
-    fn current_position(&self) -> Option<Lsn> {
-        match self {
-            InterpretedWalReaderState::Running {
-                current_position, ..
-            } => Some(*current_position),
-            InterpretedWalReaderState::Done => None,
-        }
-    }
-}
+        let stream = self.wal_stream_builder.build(MAX_SEND_SIZE).await?;
+        let mut stream = std::pin::pin!(stream);
 
-pub(crate) struct AttachShardNotification {
-    shard_id: ShardIdentity,
-    sender: tokio::sync::mpsc::Sender<Batch>,
-    start_pos: Lsn,
-}
+        let mut keepalive_ticker = tokio::time::interval(Duration::from_secs(1));
+        keepalive_ticker.set_missed_tick_behavior(MissedTickBehavior::Skip);
+        keepalive_ticker.reset();
 
-impl InterpretedWalReader {
-    /// Spawn the reader in a separate tokio task and return a handle
-    pub(crate) fn spawn(
-        wal_stream: StreamingWalReader,
-        start_pos: Lsn,
-        tx: tokio::sync::mpsc::Sender<Batch>,
-        shard: ShardIdentity,
-        pg_version: u32,
-        appname: &Option<String>,
-    ) -> InterpretedWalReaderHandle {
-        let state = Arc::new(std::sync::RwLock::new(InterpretedWalReaderState::Running {
-            current_position: start_pos,
-        }));
-
-        let (shard_notification_tx, shard_notification_rx) = tokio::sync::mpsc::unbounded_channel();
-
-        let reader = InterpretedWalReader {
-            wal_stream,
-            shard_senders: HashMap::from([(
-                shard,
-                smallvec::smallvec![ShardSenderState {
-                    sender_id: SenderId::first(),
-                    tx,
-                    next_record_lsn: start_pos,
-                }],
-            )]),
-            shard_notification_rx: Some(shard_notification_rx),
-            state: state.clone(),
-            pg_version,
-        };
-
-        let metric = WAL_READERS
-            .get_metric_with_label_values(&["task", appname.as_deref().unwrap_or("safekeeper")])
-            .unwrap();
-
-        let join_handle = tokio::task::spawn(
-            async move {
-                metric.inc();
-                scopeguard::defer! {
-                    metric.dec();
-                }
-
-                let res = reader.run_impl(start_pos).await;
-                if let Err(ref err) = res {
-                    tracing::error!("Task finished with error: {err}");
-                }
-                res
-            }
-            .instrument(info_span!("interpreted wal reader")),
-        );
-
-        InterpretedWalReaderHandle {
-            join_handle,
-            state,
-            shard_notification_tx,
-        }
-    }
-
-    /// Construct the reader without spawning anything
-    /// Callers should drive the future returned by [`Self::run`].
-    pub(crate) fn new(
-        wal_stream: StreamingWalReader,
-        start_pos: Lsn,
-        tx: tokio::sync::mpsc::Sender<Batch>,
-        shard: ShardIdentity,
-        pg_version: u32,
-    ) -> InterpretedWalReader {
-        let state = Arc::new(std::sync::RwLock::new(InterpretedWalReaderState::Running {
-            current_position: start_pos,
-        }));
-
-        InterpretedWalReader {
-            wal_stream,
-            shard_senders: HashMap::from([(
-                shard,
-                smallvec::smallvec![ShardSenderState {
-                    sender_id: SenderId::first(),
-                    tx,
-                    next_record_lsn: start_pos,
-                }],
-            )]),
-            shard_notification_rx: None,
-            state: state.clone(),
-            pg_version,
-        }
-    }
-
-    /// Entry point for future (polling) based wal reader.
-    pub(crate) async fn run(
-        self,
-        start_pos: Lsn,
-        appname: &Option<String>,
-    ) -> Result<(), CopyStreamHandlerEnd> {
-        let metric = WAL_READERS
-            .get_metric_with_label_values(&["future", appname.as_deref().unwrap_or("safekeeper")])
-            .unwrap();
-
-        metric.inc();
-        scopeguard::defer! {
-            metric.dec();
-        }
-
-        let res = self.run_impl(start_pos).await;
-        if let Err(err) = res {
-            tracing::error!("Interpreted wal reader encountered error: {err}");
-        } else {
-            tracing::info!("Interpreted wal reader exiting");
-        }
-
-        Err(CopyStreamHandlerEnd::Other(anyhow!(
-            "interpreted wal reader finished"
-        )))
-    }
-
-    /// Send interpreted WAL to one or more [`InterpretedWalSender`]s
-    /// Stops when an error is encountered or when the [`InterpretedWalReaderHandle`]
-    /// goes out of scope.
-    async fn run_impl(mut self, start_pos: Lsn) -> Result<(), InterpretedWalReaderError> {
-        let defer_state = self.state.clone();
-        scopeguard::defer! {
-            *defer_state.write().unwrap() = InterpretedWalReaderState::Done;
-        }
-
-        let mut wal_decoder = WalStreamDecoder::new(start_pos, self.pg_version);
+        let (tx, mut rx) = tokio::sync::mpsc::channel::<Batch>(2);
 
         loop {
             tokio::select! {
-                // Main branch for reading WAL and forwarding it
-                wal_or_reset = self.wal_stream.next() => {
-                    let wal = wal_or_reset.map(|wor| wor.get_wal().expect("reset handled in select branch below"));
-                    let WalBytes {
-                        wal,
-                        wal_start_lsn: _,
-                        wal_end_lsn,
-                        available_wal_end_lsn,
-                    } = match wal {
-                        Some(some) => some.map_err(InterpretedWalReaderError::ReadOrInterpret)?,
-                        None => {
-                            // [`StreamingWalReader::next`] is an endless stream of WAL.
-                            // It shouldn't ever finish unless it panicked or became internally
-                            // inconsistent.
-                            return Result::Err(InterpretedWalReaderError::WalStreamClosed);
-                        }
+                // Get some WAL from the stream and then: decode, interpret and push it down the
+                // pipeline.
+                wal = stream.next(), if tx.capacity() > 0 => {
+                    let WalBytes { wal, wal_start_lsn: _, wal_end_lsn, available_wal_end_lsn } = match wal {
+                        Some(some) => some?,
+                        None => { break; }
                     };
 
+                    wal_position = wal_end_lsn;
                     wal_decoder.feed_bytes(&wal);
 
-                    // Deserialize and interpret WAL records from this batch of WAL.
-                    // Interpreted records for each shard are collected separately.
-                    let shard_ids = self.shard_senders.keys().copied().collect::<Vec<_>>();
-                    let mut records_by_sender: HashMap<ShardSenderId, Vec<InterpretedWalRecord>> = HashMap::new();
+                    let mut records = Vec::new();
                     let mut max_next_record_lsn = None;
-                    while let Some((next_record_lsn, recdata)) = wal_decoder.poll_decode()?
+                    while let Some((next_record_lsn, recdata)) = wal_decoder
+                        .poll_decode()
+                        .with_context(|| "Failed to decode WAL")?
                     {
                         assert!(next_record_lsn.is_aligned());
                         max_next_record_lsn = Some(next_record_lsn);
 
+                        // Deserialize and interpret WAL record
                         let interpreted = InterpretedWalRecord::from_bytes_filtered(
                             recdata,
-                            &shard_ids,
+                            &self.shard,
                             next_record_lsn,
                             self.pg_version,
                         )
                         .with_context(|| "Failed to interpret WAL")?;
 
-                        for (shard, record) in interpreted {
-                            if record.is_empty() {
-                                continue;
-                            }
-
-                            let mut states_iter = self.shard_senders
-                                .get(&shard)
-                                .expect("keys collected above")
-                                .iter()
-                                .filter(|state| record.next_record_lsn > state.next_record_lsn)
-                                .peekable();
-                            while let Some(state) = states_iter.next() {
-                                let shard_sender_id = ShardSenderId::new(shard, state.sender_id);
-
-                                // The most commont case is one sender per shard. Peek and break to avoid the
-                                // clone in that situation.
-                                if states_iter.peek().is_none() {
-                                    records_by_sender.entry(shard_sender_id).or_default().push(record);
-                                    break;
-                                } else {
-                                    records_by_sender.entry(shard_sender_id).or_default().push(record.clone());
-                                }
-                            }
+                        if !interpreted.is_empty() {
+                            records.push(interpreted);
                         }
                     }
 
@@ -333,169 +99,19 @@ impl InterpretedWalReader {
                         None => { continue; }
                     };
 
-                    // Update the current position such that new receivers can decide
-                    // whether to attach to us or spawn a new WAL reader.
-                    match &mut *self.state.write().unwrap() {
-                        InterpretedWalReaderState::Running { current_position, .. } => {
-                            *current_position = max_next_record_lsn;
-                        },
-                        InterpretedWalReaderState::Done => {
-                            unreachable!()
-                        }
-                    }
-
-                    // Send interpreted records downstream. Anything that has already been seen
-                    // by a shard is filtered out.
-                    let mut shard_senders_to_remove = Vec::new();
-                    for (shard, states) in &mut self.shard_senders {
-                        for state in states {
-                            if max_next_record_lsn <= state.next_record_lsn {
-                                continue;
-                            }
-
-                            let shard_sender_id = ShardSenderId::new(*shard, state.sender_id);
-                            let records = records_by_sender.remove(&shard_sender_id).unwrap_or_default();
-
-                            let batch = InterpretedWalRecords {
-                                records,
-                                next_record_lsn: Some(max_next_record_lsn),
-                            };
-
-                            let res = state.tx.send(Batch {
-                                wal_end_lsn,
-                                available_wal_end_lsn,
-                                records: batch,
-                            }).await;
-
-                            if res.is_err() {
-                                shard_senders_to_remove.push(shard_sender_id);
-                            } else {
-                                state.next_record_lsn = max_next_record_lsn;
-                            }
-                        }
-                    }
-
-                    // Clean up any shard senders that have dropped out.
-                    // This is inefficient, but such events are rare (connection to PS termination)
-                    // and the number of subscriptions on the same shards very small (only one
-                    // for the steady state).
-                    for to_remove in shard_senders_to_remove {
-                        let shard_senders = self.shard_senders.get_mut(&to_remove.shard()).expect("saw it above");
-                        if let Some(idx) = shard_senders.iter().position(|s| s.sender_id == to_remove.sender_id) {
-                            shard_senders.remove(idx);
-                            tracing::info!("Removed shard sender {}", to_remove);
-                        }
-
-                        if shard_senders.is_empty() {
-                            self.shard_senders.remove(&to_remove.shard());
-                        }
-                    }
-                },
-                // Listen for new shards that want to attach to this reader.
-                // If the reader is not running as a task, then this is not supported
-                // (see the pending branch below).
-                notification = match self.shard_notification_rx.as_mut() {
-                        Some(rx) => Either::Left(rx.recv()),
-                        None => Either::Right(std::future::pending())
-                    } => {
-                    if let Some(n) = notification {
-                        let AttachShardNotification { shard_id, sender, start_pos } = n;
-
-                        // Update internal and external state, then reset the WAL stream
-                        // if required.
-                        let senders = self.shard_senders.entry(shard_id).or_default();
-                        let new_sender_id = match senders.last() {
-                            Some(sender) => sender.sender_id.next(),
-                            None => SenderId::first()
-                        };
-
-                        senders.push(ShardSenderState { sender_id: new_sender_id, tx: sender, next_record_lsn: start_pos});
-                        let current_pos = self.state.read().unwrap().current_position().unwrap();
-                        if start_pos < current_pos {
-                            self.wal_stream.reset(start_pos).await;
-                            wal_decoder = WalStreamDecoder::new(start_pos, self.pg_version);
-                        }
-
-                        tracing::info!(
-                            "Added shard sender {} with start_pos={} current_pos={}",
-                            ShardSenderId::new(shard_id, new_sender_id), start_pos, current_pos
-                        );
-                    }
-                }
-            }
-        }
-    }
-}
-
-impl InterpretedWalReaderHandle {
-    /// Fan-out the reader by attaching a new shard to it
-    pub(crate) fn fanout(
-        &self,
-        shard_id: ShardIdentity,
-        sender: tokio::sync::mpsc::Sender<Batch>,
-        start_pos: Lsn,
-    ) -> Result<(), SendError<AttachShardNotification>> {
-        self.shard_notification_tx.send(AttachShardNotification {
-            shard_id,
-            sender,
-            start_pos,
-        })
-    }
-
-    /// Get the current WAL position of the reader
-    pub(crate) fn current_position(&self) -> Option<Lsn> {
-        self.state.read().unwrap().current_position()
-    }
-
-    pub(crate) fn abort(&self) {
-        self.join_handle.abort()
-    }
-}
-
-impl Drop for InterpretedWalReaderHandle {
-    fn drop(&mut self) {
-        tracing::info!("Aborting interpreted wal reader");
-        self.abort()
-    }
-}
-
-pub(crate) struct InterpretedWalSender<'a, IO> {
-    pub(crate) format: InterpretedFormat,
-    pub(crate) compression: Option<Compression>,
-    pub(crate) appname: Option<String>,
-
-    pub(crate) tli: WalResidentTimeline,
-    pub(crate) start_lsn: Lsn,
-
-    pub(crate) pgb: &'a mut PostgresBackend<IO>,
-    pub(crate) end_watch_view: EndWatchView,
-    pub(crate) wal_sender_guard: Arc<WalSenderGuard>,
-    pub(crate) rx: tokio::sync::mpsc::Receiver<Batch>,
-}
-
-impl<IO: AsyncRead + AsyncWrite + Unpin> InterpretedWalSender<'_, IO> {
-    /// Send interpreted WAL records over the network.
-    /// Also manages keep-alives if nothing was sent for a while.
-    pub(crate) async fn run(mut self) -> Result<(), CopyStreamHandlerEnd> {
-        let mut keepalive_ticker = tokio::time::interval(Duration::from_secs(1));
-        keepalive_ticker.set_missed_tick_behavior(MissedTickBehavior::Skip);
-        keepalive_ticker.reset();
-
-        let mut wal_position = self.start_lsn;
-
-        loop {
-            tokio::select! {
-                batch = self.rx.recv() => {
-                    let batch = match batch {
-                        Some(b) => b,
-                        None => {
-                            return Result::Err(
-                                CopyStreamHandlerEnd::Other(anyhow!("Interpreted WAL reader exited early"))
-                            );
-                        }
+                    let batch = InterpretedWalRecords {
+                        records,
+                        next_record_lsn: Some(max_next_record_lsn),
                     };
 
-                    wal_position = batch.wal_end_lsn;
+                    tx.send(Batch {wal_end_lsn, available_wal_end_lsn, records: batch}).await.unwrap();
+                },
+                // For a previously interpreted batch, serialize it and push it down the wire.
+                batch = rx.recv() => {
+                    let batch = match batch {
+                        Some(b) => b,
+                        None => { break; }
+                    };
 
                     let buf = batch
                         .records
@@ -516,21 +132,7 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> InterpretedWalSender<'_, IO> {
                         })).await?;
                 }
                 // Send a periodic keep alive when the connection has been idle for a while.
-                // Since we've been idle, also check if we can stop streaming.
                 _ = keepalive_ticker.tick() => {
-                    if let Some(remote_consistent_lsn) = self.wal_sender_guard
-                        .walsenders()
-                        .get_ws_remote_consistent_lsn(self.wal_sender_guard.id())
-                    {
-                        if self.tli.should_walsender_stop(remote_consistent_lsn).await {
-                            // Stop streaming if the receivers are caught up and
-                            // there's no active compute. This causes the loop in
-                            // [`crate::send_interpreted_wal::InterpretedWalSender::run`]
-                            // to exit and terminate the WAL stream.
-                            break;
-                        }
-                    }
-
                     self.pgb
                         .write_message(&BeMessage::KeepAlive(WalSndKeepAlive {
                             wal_end: self.end_watch_view.get().0,
@@ -538,259 +140,14 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> InterpretedWalSender<'_, IO> {
                             request_reply: true,
                         }))
                         .await?;
-                },
+                }
             }
         }
 
+        // The loop above ends when the receiver is caught up and there's no more WAL to send.
         Err(CopyStreamHandlerEnd::ServerInitiated(format!(
             "ending streaming to {:?} at {}, receiver is caughtup and there is no computes",
             self.appname, wal_position,
         )))
     }
 }
-#[cfg(test)]
-mod tests {
-    use std::{collections::HashMap, str::FromStr, time::Duration};
-
-    use pageserver_api::shard::{ShardIdentity, ShardStripeSize};
-    use postgres_ffi::MAX_SEND_SIZE;
-    use tokio::sync::mpsc::error::TryRecvError;
-    use utils::{
-        id::{NodeId, TenantTimelineId},
-        lsn::Lsn,
-        shard::{ShardCount, ShardNumber},
-    };
-
-    use crate::{
-        send_interpreted_wal::{Batch, InterpretedWalReader},
-        test_utils::Env,
-        wal_reader_stream::StreamingWalReader,
-    };
-
-    #[tokio::test]
-    async fn test_interpreted_wal_reader_fanout() {
-        let _ = env_logger::builder().is_test(true).try_init();
-
-        const SIZE: usize = 8 * 1024;
-        const MSG_COUNT: usize = 200;
-        const PG_VERSION: u32 = 17;
-        const SHARD_COUNT: u8 = 2;
-
-        let start_lsn = Lsn::from_str("0/149FD18").unwrap();
-        let env = Env::new(true).unwrap();
-        let tli = env
-            .make_timeline(NodeId(1), TenantTimelineId::generate(), start_lsn)
-            .await
-            .unwrap();
-
-        let resident_tli = tli.wal_residence_guard().await.unwrap();
-        let end_watch = Env::write_wal(tli, start_lsn, SIZE, MSG_COUNT)
-            .await
-            .unwrap();
-        let end_pos = end_watch.get();
-
-        tracing::info!("Doing first round of reads ...");
-
-        let streaming_wal_reader = StreamingWalReader::new(
-            resident_tli,
-            None,
-            start_lsn,
-            end_pos,
-            end_watch,
-            MAX_SEND_SIZE,
-        );
-
-        let shard_0 = ShardIdentity::new(
-            ShardNumber(0),
-            ShardCount(SHARD_COUNT),
-            ShardStripeSize::default(),
-        )
-        .unwrap();
-
-        let shard_1 = ShardIdentity::new(
-            ShardNumber(1),
-            ShardCount(SHARD_COUNT),
-            ShardStripeSize::default(),
-        )
-        .unwrap();
-
-        let mut shards = HashMap::new();
-
-        for shard_number in 0..SHARD_COUNT {
-            let shard_id = ShardIdentity::new(
-                ShardNumber(shard_number),
-                ShardCount(SHARD_COUNT),
-                ShardStripeSize::default(),
-            )
-            .unwrap();
-            let (tx, rx) = tokio::sync::mpsc::channel::<Batch>(MSG_COUNT * 2);
-            shards.insert(shard_id, (Some(tx), Some(rx)));
-        }
-
-        let shard_0_tx = shards.get_mut(&shard_0).unwrap().0.take().unwrap();
-        let mut shard_0_rx = shards.get_mut(&shard_0).unwrap().1.take().unwrap();
-
-        let handle = InterpretedWalReader::spawn(
-            streaming_wal_reader,
-            start_lsn,
-            shard_0_tx,
-            shard_0,
-            PG_VERSION,
-            &Some("pageserver".to_string()),
-        );
-
-        tracing::info!("Reading all WAL with only shard 0 attached ...");
-
-        let mut shard_0_interpreted_records = Vec::new();
-        while let Some(batch) = shard_0_rx.recv().await {
-            shard_0_interpreted_records.push(batch.records);
-            if batch.wal_end_lsn == batch.available_wal_end_lsn {
-                break;
-            }
-        }
-
-        let shard_1_tx = shards.get_mut(&shard_1).unwrap().0.take().unwrap();
-        let mut shard_1_rx = shards.get_mut(&shard_1).unwrap().1.take().unwrap();
-
-        tracing::info!("Attaching shard 1 to the reader at start of WAL");
-        handle.fanout(shard_1, shard_1_tx, start_lsn).unwrap();
-
-        tracing::info!("Reading all WAL with shard 0 and shard 1 attached ...");
-
-        let mut shard_1_interpreted_records = Vec::new();
-        while let Some(batch) = shard_1_rx.recv().await {
-            shard_1_interpreted_records.push(batch.records);
-            if batch.wal_end_lsn == batch.available_wal_end_lsn {
-                break;
-            }
-        }
-
-        // This test uses logical messages. Those only go to shard 0. Check that the
-        // filtering worked and shard 1 did not get any.
-        assert!(shard_1_interpreted_records
-            .iter()
-            .all(|recs| recs.records.is_empty()));
-
-        // Shard 0 should not receive anything more since the reader is
-        // going through wal that it has already processed.
-        let res = shard_0_rx.try_recv();
-        if let Ok(ref ok) = res {
-            tracing::error!(
-                "Shard 0 received batch: wal_end_lsn={} available_wal_end_lsn={}",
-                ok.wal_end_lsn,
-                ok.available_wal_end_lsn
-            );
-        }
-        assert!(matches!(res, Err(TryRecvError::Empty)));
-
-        // Check that the next records lsns received by the two shards match up.
-        let shard_0_next_lsns = shard_0_interpreted_records
-            .iter()
-            .map(|recs| recs.next_record_lsn)
-            .collect::<Vec<_>>();
-        let shard_1_next_lsns = shard_1_interpreted_records
-            .iter()
-            .map(|recs| recs.next_record_lsn)
-            .collect::<Vec<_>>();
-        assert_eq!(shard_0_next_lsns, shard_1_next_lsns);
-
-        handle.abort();
-        let mut done = false;
-        for _ in 0..5 {
-            if handle.current_position().is_none() {
-                done = true;
-                break;
-            }
-            tokio::time::sleep(Duration::from_millis(1)).await;
-        }
-
-        assert!(done);
-    }
-
-    #[tokio::test]
-    async fn test_interpreted_wal_reader_same_shard_fanout() {
-        let _ = env_logger::builder().is_test(true).try_init();
-
-        const SIZE: usize = 8 * 1024;
-        const MSG_COUNT: usize = 200;
-        const PG_VERSION: u32 = 17;
-        const SHARD_COUNT: u8 = 2;
-        const ATTACHED_SHARDS: u8 = 4;
-
-        let start_lsn = Lsn::from_str("0/149FD18").unwrap();
-        let env = Env::new(true).unwrap();
-        let tli = env
-            .make_timeline(NodeId(1), TenantTimelineId::generate(), start_lsn)
-            .await
-            .unwrap();
-
-        let resident_tli = tli.wal_residence_guard().await.unwrap();
-        let end_watch = Env::write_wal(tli, start_lsn, SIZE, MSG_COUNT)
-            .await
-            .unwrap();
-        let end_pos = end_watch.get();
-
-        let streaming_wal_reader = StreamingWalReader::new(
-            resident_tli,
-            None,
-            start_lsn,
-            end_pos,
-            end_watch,
-            MAX_SEND_SIZE,
-        );
-
-        let shard_0 = ShardIdentity::new(
-            ShardNumber(0),
-            ShardCount(SHARD_COUNT),
-            ShardStripeSize::default(),
-        )
-        .unwrap();
-
-        let (tx, rx) = tokio::sync::mpsc::channel::<Batch>(MSG_COUNT * 2);
-        let mut batch_receivers = vec![rx];
-
-        let handle = InterpretedWalReader::spawn(
-            streaming_wal_reader,
-            start_lsn,
-            tx,
-            shard_0,
-            PG_VERSION,
-            &Some("pageserver".to_string()),
-        );
-
-        for _ in 0..(ATTACHED_SHARDS - 1) {
-            let (tx, rx) = tokio::sync::mpsc::channel::<Batch>(MSG_COUNT * 2);
-            handle.fanout(shard_0, tx, start_lsn).unwrap();
-            batch_receivers.push(rx);
-        }
-
-        loop {
-            let batch = batch_receivers.first_mut().unwrap().recv().await.unwrap();
-            for rx in batch_receivers.iter_mut().skip(1) {
-                let other_batch = rx.recv().await.unwrap();
-
-                assert_eq!(batch.wal_end_lsn, other_batch.wal_end_lsn);
-                assert_eq!(
-                    batch.available_wal_end_lsn,
-                    other_batch.available_wal_end_lsn
-                );
-            }
-
-            if batch.wal_end_lsn == batch.available_wal_end_lsn {
-                break;
-            }
-        }
-
-        handle.abort();
-        let mut done = false;
-        for _ in 0..5 {
-            if handle.current_position().is_none() {
-                done = true;
-                break;
-            }
-            tokio::time::sleep(Duration::from_millis(1)).await;
-        }
-
-        assert!(done);
-    }
-}

safekeeper/src/send_wal.rs

@@ -2,18 +2,16 @@
 //! with the "START_REPLICATION" message, and registry of walsenders.
 
 use crate::handler::SafekeeperPostgresHandler;
-use crate::metrics::{RECEIVED_PS_FEEDBACKS, WAL_READERS};
+use crate::metrics::RECEIVED_PS_FEEDBACKS;
 use crate::receive_wal::WalReceivers;
 use crate::safekeeper::TermLsn;
-use crate::send_interpreted_wal::{
-    Batch, InterpretedWalReader, InterpretedWalReaderHandle, InterpretedWalSender,
-};
+use crate::send_interpreted_wal::InterpretedWalSender;
 use crate::timeline::WalResidentTimeline;
-use crate::wal_reader_stream::StreamingWalReader;
+use crate::wal_reader_stream::WalReaderStreamBuilder;
 use crate::wal_storage::WalReader;
 use anyhow::{bail, Context as AnyhowContext};
 use bytes::Bytes;
-use futures::FutureExt;
+use futures::future::Either;
 use parking_lot::Mutex;
 use postgres_backend::PostgresBackend;
 use postgres_backend::{CopyStreamHandlerEnd, PostgresBackendReader, QueryError};
@@ -21,16 +19,16 @@ use postgres_ffi::get_current_timestamp;
 use postgres_ffi::{TimestampTz, MAX_SEND_SIZE};
 use pq_proto::{BeMessage, WalSndKeepAlive, XLogDataBody};
 use safekeeper_api::models::{
-    HotStandbyFeedback, ReplicationFeedback, StandbyFeedback, StandbyReply,
-    INVALID_FULL_TRANSACTION_ID,
+    ConnectionId, HotStandbyFeedback, ReplicationFeedback, StandbyFeedback, StandbyReply,
+    WalSenderState, INVALID_FULL_TRANSACTION_ID,
 };
 use safekeeper_api::Term;
 use tokio::io::{AsyncRead, AsyncWrite};
 use utils::failpoint_support;
+use utils::id::TenantTimelineId;
 use utils::pageserver_feedback::PageserverFeedback;
 use utils::postgres_client::PostgresClientProtocol;
 
-use itertools::Itertools;
 use std::cmp::{max, min};
 use std::net::SocketAddr;
 use std::sync::Arc;
@@ -52,12 +50,6 @@ pub struct WalSenders {
     walreceivers: Arc<WalReceivers>,
 }
 
-pub struct WalSendersTimelineMetricValues {
-    pub ps_feedback_counter: u64,
-    pub last_ps_feedback: PageserverFeedback,
-    pub interpreted_wal_reader_tasks: usize,
-}
-
 impl WalSenders {
     pub fn new(walreceivers: Arc<WalReceivers>) -> Arc<WalSenders> {
         Arc::new(WalSenders {
@@ -68,8 +60,21 @@ impl WalSenders {
 
     /// Register new walsender. Returned guard provides access to the slot and
     /// automatically deregisters in Drop.
-    fn register(self: &Arc<WalSenders>, walsender_state: WalSenderState) -> WalSenderGuard {
+    fn register(
+        self: &Arc<WalSenders>,
+        ttid: TenantTimelineId,
+        addr: SocketAddr,
+        conn_id: ConnectionId,
+        appname: Option<String>,
+    ) -> WalSenderGuard {
         let slots = &mut self.mutex.lock().slots;
+        let walsender_state = WalSenderState {
+            ttid,
+            addr,
+            conn_id,
+            appname,
+            feedback: ReplicationFeedback::Pageserver(PageserverFeedback::empty()),
+        };
         // find empty slot or create new one
         let pos = if let Some(pos) = slots.iter().position(|s| s.is_none()) {
             slots[pos] = Some(walsender_state);
@@ -85,79 +90,9 @@ impl WalSenders {
         }
     }
 
-    fn create_or_update_interpreted_reader<
-        FUp: FnOnce(&Arc<InterpretedWalReaderHandle>) -> anyhow::Result<()>,
-        FNew: FnOnce() -> InterpretedWalReaderHandle,
-    >(
-        self: &Arc<WalSenders>,
-        id: WalSenderId,
-        start_pos: Lsn,
-        max_delta_for_fanout: Option<u64>,
-        update: FUp,
-        create: FNew,
-    ) -> anyhow::Result<()> {
-        let state = &mut self.mutex.lock();
-
-        let mut selected_interpreted_reader = None;
-        for slot in state.slots.iter().flatten() {
-            if let WalSenderState::Interpreted(slot_state) = slot {
-                if let Some(ref interpreted_reader) = slot_state.interpreted_wal_reader {
-                    let select = match (interpreted_reader.current_position(), max_delta_for_fanout)
-                    {
-                        (Some(pos), Some(max_delta)) => {
-                            let delta = pos.0.abs_diff(start_pos.0);
-                            delta <= max_delta
-                        }
-                        // Reader is not active
-                        (None, _) => false,
-                        // Gating fanout by max delta is disabled.
-                        // Attach to any active reader.
-                        (_, None) => true,
-                    };
-
-                    if select {
-                        selected_interpreted_reader = Some(interpreted_reader.clone());
-                        break;
-                    }
-                }
-            }
-        }
-
-        let slot = state.get_slot_mut(id);
-        let slot_state = match slot {
-            WalSenderState::Interpreted(s) => s,
-            WalSenderState::Vanilla(_) => unreachable!(),
-        };
-
-        let selected_or_new = match selected_interpreted_reader {
-            Some(selected) => {
-                update(&selected)?;
-                selected
-            }
-            None => Arc::new(create()),
-        };
-
-        slot_state.interpreted_wal_reader = Some(selected_or_new);
-
-        Ok(())
-    }
-
     /// Get state of all walsenders.
-    pub fn get_all_public(self: &Arc<WalSenders>) -> Vec<safekeeper_api::models::WalSenderState> {
-        self.mutex
-            .lock()
-            .slots
-            .iter()
-            .flatten()
-            .map(|state| match state {
-                WalSenderState::Vanilla(s) => {
-                    safekeeper_api::models::WalSenderState::Vanilla(s.clone())
-                }
-                WalSenderState::Interpreted(s) => {
-                    safekeeper_api::models::WalSenderState::Interpreted(s.public_state.clone())
-                }
-            })
-            .collect()
+    pub fn get_all(self: &Arc<WalSenders>) -> Vec<WalSenderState> {
+        self.mutex.lock().slots.iter().flatten().cloned().collect()
     }
 
     /// Get LSN of the most lagging pageserver receiver. Return None if there are no
@@ -168,7 +103,7 @@ impl WalSenders {
             .slots
             .iter()
             .flatten()
-            .filter_map(|s| match s.get_feedback() {
+            .filter_map(|s| match s.feedback {
                 ReplicationFeedback::Pageserver(feedback) => Some(feedback.last_received_lsn),
                 ReplicationFeedback::Standby(_) => None,
             })
@@ -176,25 +111,9 @@ impl WalSenders {
     }
 
     /// Returns total counter of pageserver feedbacks received and last feedback.
-    pub fn info_for_metrics(self: &Arc<WalSenders>) -> WalSendersTimelineMetricValues {
+    pub fn get_ps_feedback_stats(self: &Arc<WalSenders>) -> (u64, PageserverFeedback) {
         let shared = self.mutex.lock();
-
-        let interpreted_wal_reader_tasks = shared
-            .slots
-            .iter()
-            .filter_map(|ss| match ss {
-                Some(WalSenderState::Interpreted(int)) => int.interpreted_wal_reader.as_ref(),
-                Some(WalSenderState::Vanilla(_)) => None,
-                None => None,
-            })
-            .unique_by(|reader| Arc::as_ptr(reader))
-            .count();
-
-        WalSendersTimelineMetricValues {
-            ps_feedback_counter: shared.ps_feedback_counter,
-            last_ps_feedback: shared.last_ps_feedback,
-            interpreted_wal_reader_tasks,
-        }
+        (shared.ps_feedback_counter, shared.last_ps_feedback)
     }
 
     /// Get aggregated hot standby feedback (we send it to compute).
@@ -205,7 +124,7 @@ impl WalSenders {
     /// Record new pageserver feedback, update aggregated values.
     fn record_ps_feedback(self: &Arc<WalSenders>, id: WalSenderId, feedback: &PageserverFeedback) {
         let mut shared = self.mutex.lock();
-        *shared.get_slot_mut(id).get_mut_feedback() = ReplicationFeedback::Pageserver(*feedback);
+        shared.get_slot_mut(id).feedback = ReplicationFeedback::Pageserver(*feedback);
         shared.last_ps_feedback = *feedback;
         shared.ps_feedback_counter += 1;
         drop(shared);
@@ -224,10 +143,10 @@ impl WalSenders {
             "Record standby reply: ts={} apply_lsn={}",
             reply.reply_ts, reply.apply_lsn
         );
-        match &mut slot.get_mut_feedback() {
+        match &mut slot.feedback {
             ReplicationFeedback::Standby(sf) => sf.reply = *reply,
             ReplicationFeedback::Pageserver(_) => {
-                *slot.get_mut_feedback() = ReplicationFeedback::Standby(StandbyFeedback {
+                slot.feedback = ReplicationFeedback::Standby(StandbyFeedback {
                     reply: *reply,
                     hs_feedback: HotStandbyFeedback::empty(),
                 })
@@ -239,10 +158,10 @@ impl WalSenders {
     fn record_hs_feedback(self: &Arc<WalSenders>, id: WalSenderId, feedback: &HotStandbyFeedback) {
         let mut shared = self.mutex.lock();
         let slot = shared.get_slot_mut(id);
-        match &mut slot.get_mut_feedback() {
+        match &mut slot.feedback {
             ReplicationFeedback::Standby(sf) => sf.hs_feedback = *feedback,
             ReplicationFeedback::Pageserver(_) => {
-                *slot.get_mut_feedback() = ReplicationFeedback::Standby(StandbyFeedback {
+                slot.feedback = ReplicationFeedback::Standby(StandbyFeedback {
                     reply: StandbyReply::empty(),
                     hs_feedback: *feedback,
                 })
@@ -256,7 +175,7 @@ impl WalSenders {
     pub fn get_ws_remote_consistent_lsn(self: &Arc<WalSenders>, id: WalSenderId) -> Option<Lsn> {
         let shared = self.mutex.lock();
         let slot = shared.get_slot(id);
-        match slot.get_feedback() {
+        match slot.feedback {
             ReplicationFeedback::Pageserver(feedback) => Some(feedback.remote_consistent_lsn),
             _ => None,
         }
@@ -280,47 +199,6 @@ struct WalSendersShared {
     slots: Vec<Option<WalSenderState>>,
 }
 
-/// Safekeeper internal definitions of wal sender state
-///
-/// As opposed to [`safekeeper_api::models::WalSenderState`] these struct may
-/// include state that we don not wish to expose to the public api.
-#[derive(Debug, Clone)]
-pub(crate) enum WalSenderState {
-    Vanilla(VanillaWalSenderInternalState),
-    Interpreted(InterpretedWalSenderInternalState),
-}
-
-type VanillaWalSenderInternalState = safekeeper_api::models::VanillaWalSenderState;
-
-#[derive(Debug, Clone)]
-pub(crate) struct InterpretedWalSenderInternalState {
-    public_state: safekeeper_api::models::InterpretedWalSenderState,
-    interpreted_wal_reader: Option<Arc<InterpretedWalReaderHandle>>,
-}
-
-impl WalSenderState {
-    fn get_addr(&self) -> &SocketAddr {
-        match self {
-            WalSenderState::Vanilla(state) => &state.addr,
-            WalSenderState::Interpreted(state) => &state.public_state.addr,
-        }
-    }
-
-    fn get_feedback(&self) -> &ReplicationFeedback {
-        match self {
-            WalSenderState::Vanilla(state) => &state.feedback,
-            WalSenderState::Interpreted(state) => &state.public_state.feedback,
-        }
-    }
-
-    fn get_mut_feedback(&mut self) -> &mut ReplicationFeedback {
-        match self {
-            WalSenderState::Vanilla(state) => &mut state.feedback,
-            WalSenderState::Interpreted(state) => &mut state.public_state.feedback,
-        }
-    }
-}
-
 impl WalSendersShared {
     fn new() -> Self {
         WalSendersShared {
@@ -347,7 +225,7 @@ impl WalSendersShared {
         let mut agg = HotStandbyFeedback::empty();
         let mut reply_agg = StandbyReply::empty();
         for ws_state in self.slots.iter().flatten() {
-            if let ReplicationFeedback::Standby(standby_feedback) = ws_state.get_feedback() {
+            if let ReplicationFeedback::Standby(standby_feedback) = ws_state.feedback {
                 let hs_feedback = standby_feedback.hs_feedback;
                 // doing Option math like op1.iter().chain(op2.iter()).min()
                 // would be nicer, but we serialize/deserialize this struct
@@ -439,7 +317,7 @@ impl SafekeeperPostgresHandler {
     /// Wrapper around handle_start_replication_guts handling result. Error is
     /// handled here while we're still in walsender ttid span; with API
     /// extension, this can probably be moved into postgres_backend.
-    pub async fn handle_start_replication<IO: AsyncRead + AsyncWrite + Unpin + Send>(
+    pub async fn handle_start_replication<IO: AsyncRead + AsyncWrite + Unpin>(
         &mut self,
         pgb: &mut PostgresBackend<IO>,
         start_pos: Lsn,
@@ -464,7 +342,7 @@ impl SafekeeperPostgresHandler {
         Ok(())
     }
 
-    pub async fn handle_start_replication_guts<IO: AsyncRead + AsyncWrite + Unpin + Send>(
+    pub async fn handle_start_replication_guts<IO: AsyncRead + AsyncWrite + Unpin>(
         &mut self,
         pgb: &mut PostgresBackend<IO>,
         start_pos: Lsn,
@@ -474,30 +352,12 @@ impl SafekeeperPostgresHandler {
         let appname = self.appname.clone();
 
         // Use a guard object to remove our entry from the timeline when we are done.
-        let ws_guard = match self.protocol() {
-            PostgresClientProtocol::Vanilla => Arc::new(tli.get_walsenders().register(
-                WalSenderState::Vanilla(VanillaWalSenderInternalState {
-                    ttid: self.ttid,
-                    addr: *pgb.get_peer_addr(),
-                    conn_id: self.conn_id,
-                    appname: self.appname.clone(),
-                    feedback: ReplicationFeedback::Pageserver(PageserverFeedback::empty()),
-                }),
-            )),
-            PostgresClientProtocol::Interpreted { .. } => Arc::new(tli.get_walsenders().register(
-                WalSenderState::Interpreted(InterpretedWalSenderInternalState {
-                    public_state: safekeeper_api::models::InterpretedWalSenderState {
-                        ttid: self.ttid,
-                        shard: self.shard.unwrap(),
-                        addr: *pgb.get_peer_addr(),
-                        conn_id: self.conn_id,
-                        appname: self.appname.clone(),
-                        feedback: ReplicationFeedback::Pageserver(PageserverFeedback::empty()),
-                    },
-                    interpreted_wal_reader: None,
-                }),
-            )),
-        };
+        let ws_guard = Arc::new(tli.get_walsenders().register(
+            self.ttid,
+            *pgb.get_peer_addr(),
+            self.conn_id,
+            self.appname.clone(),
+        ));
 
         // Walsender can operate in one of two modes which we select by
         // application_name: give only committed WAL (used by pageserver) or all
@@ -543,7 +403,7 @@ impl SafekeeperPostgresHandler {
                     pgb,
                     // should succeed since we're already holding another guard
                     tli: tli.wal_residence_guard().await?,
-                    appname: appname.clone(),
+                    appname,
                     start_pos,
                     end_pos,
                     term,
@@ -553,7 +413,7 @@ impl SafekeeperPostgresHandler {
                     send_buf: vec![0u8; MAX_SEND_SIZE],
                 };
 
-                FutureExt::boxed(sender.run())
+                Either::Left(sender.run())
             }
             PostgresClientProtocol::Interpreted {
                 format,
@@ -561,96 +421,27 @@ impl SafekeeperPostgresHandler {
             } => {
                 let pg_version = tli.tli.get_state().await.1.server.pg_version / 10000;
                 let end_watch_view = end_watch.view();
-                let wal_residence_guard = tli.wal_residence_guard().await?;
-                let (tx, rx) = tokio::sync::mpsc::channel::<Batch>(2);
-                let shard = self.shard.unwrap();
+                let wal_stream_builder = WalReaderStreamBuilder {
+                    tli: tli.wal_residence_guard().await?,
+                    start_pos,
+                    end_pos,
+                    term,
+                    end_watch,
+                    wal_sender_guard: ws_guard.clone(),
+                };
 
-                if self.conf.wal_reader_fanout && !shard.is_unsharded() {
-                    let ws_id = ws_guard.id();
-                    ws_guard.walsenders().create_or_update_interpreted_reader(
-                        ws_id,
-                        start_pos,
-                        self.conf.max_delta_for_fanout,
-                        {
-                            let tx = tx.clone();
-                            |reader| {
-                                tracing::info!(
-                                    "Fanning out interpreted wal reader at {}",
-                                    start_pos
-                                );
-                                reader
-                                    .fanout(shard, tx, start_pos)
-                                    .with_context(|| "Failed to fan out reader")
-                            }
-                        },
-                        || {
-                            tracing::info!("Spawning interpreted wal reader at {}", start_pos);
+                let sender = InterpretedWalSender {
+                    format,
+                    compression,
+                    pgb,
+                    wal_stream_builder,
+                    end_watch_view,
+                    shard: self.shard.unwrap(),
+                    pg_version,
+                    appname,
+                };
 
-                            let wal_stream = StreamingWalReader::new(
-                                wal_residence_guard,
-                                term,
-                                start_pos,
-                                end_pos,
-                                end_watch,
-                                MAX_SEND_SIZE,
-                            );
-
-                            InterpretedWalReader::spawn(
-                                wal_stream, start_pos, tx, shard, pg_version, &appname,
-                            )
-                        },
-                    )?;
-
-                    let sender = InterpretedWalSender {
-                        format,
-                        compression,
-                        appname,
-                        tli: tli.wal_residence_guard().await?,
-                        start_lsn: start_pos,
-                        pgb,
-                        end_watch_view,
-                        wal_sender_guard: ws_guard.clone(),
-                        rx,
-                    };
-
-                    FutureExt::boxed(sender.run())
-                } else {
-                    let wal_reader = StreamingWalReader::new(
-                        wal_residence_guard,
-                        term,
-                        start_pos,
-                        end_pos,
-                        end_watch,
-                        MAX_SEND_SIZE,
-                    );
-
-                    let reader =
-                        InterpretedWalReader::new(wal_reader, start_pos, tx, shard, pg_version);
-
-                    let sender = InterpretedWalSender {
-                        format,
-                        compression,
-                        appname: appname.clone(),
-                        tli: tli.wal_residence_guard().await?,
-                        start_lsn: start_pos,
-                        pgb,
-                        end_watch_view,
-                        wal_sender_guard: ws_guard.clone(),
-                        rx,
-                    };
-
-                    FutureExt::boxed(async move {
-                        // Sender returns an Err on all code paths.
-                        // If the sender finishes first, we will drop the reader future.
-                        // If the reader finishes first, the sender will finish too since
-                        // the wal sender has dropped.
-                        let res = tokio::try_join!(sender.run(), reader.run(start_pos, &appname));
-                        match res.map(|_| ()) {
-                            Ok(_) => unreachable!("sender finishes with Err by convention"),
-                            err_res => err_res,
-                        }
-                    })
-                }
+                Either::Right(sender.run())
             }
         };
 
@@ -679,8 +470,7 @@ impl SafekeeperPostgresHandler {
             .clone();
         info!(
             "finished streaming to {}, feedback={:?}",
-            ws_state.get_addr(),
-            ws_state.get_feedback(),
+            ws_state.addr, ws_state.feedback,
         );
 
         // Join pg backend back.
@@ -788,18 +578,6 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> WalSender<'_, IO> {
     /// Err(CopyStreamHandlerEnd) is always returned; Result is used only for ?
     /// convenience.
     async fn run(mut self) -> Result<(), CopyStreamHandlerEnd> {
-        let metric = WAL_READERS
-            .get_metric_with_label_values(&[
-                "future",
-                self.appname.as_deref().unwrap_or("safekeeper"),
-            ])
-            .unwrap();
-
-        metric.inc();
-        scopeguard::defer! {
-            metric.dec();
-        }
-
         loop {
             // Wait for the next portion if it is not there yet, or just
             // update our end of WAL available for sending value, we
@@ -1035,7 +813,7 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> ReplyReader<IO> {
 #[cfg(test)]
 mod tests {
     use safekeeper_api::models::FullTransactionId;
-    use utils::id::{TenantId, TenantTimelineId, TimelineId};
+    use utils::id::{TenantId, TimelineId};
 
     use super::*;
 
@@ -1052,13 +830,13 @@ mod tests {
 
     // add to wss specified feedback setting other fields to dummy values
     fn push_feedback(wss: &mut WalSendersShared, feedback: ReplicationFeedback) {
-        let walsender_state = WalSenderState::Vanilla(VanillaWalSenderInternalState {
+        let walsender_state = WalSenderState {
             ttid: mock_ttid(),
             addr: mock_addr(),
             conn_id: 1,
             appname: None,
             feedback,
-        });
+        };
         wss.slots.push(Some(walsender_state))
     }
 

safekeeper/src/state.rs

@@ -1,25 +1,20 @@
 //! Defines per timeline data stored persistently (SafeKeeperPersistentState)
 //! and its wrapper with in memory layer (SafekeeperState).
 
-use std::{cmp::max, ops::Deref, time::SystemTime};
+use std::{cmp::max, ops::Deref};
 
 use anyhow::{bail, Result};
 use postgres_ffi::WAL_SEGMENT_SIZE;
-use safekeeper_api::{
-    membership::Configuration,
-    models::{TimelineMembershipSwitchResponse, TimelineTermBumpResponse},
-    ServerInfo, Term, INITIAL_TERM,
-};
+use safekeeper_api::{models::TimelineTermBumpResponse, ServerInfo, Term};
 use serde::{Deserialize, Serialize};
-use tracing::info;
 use utils::{
-    id::{TenantId, TenantTimelineId, TimelineId},
+    id::{NodeId, TenantId, TenantTimelineId, TimelineId},
     lsn::Lsn,
 };
 
 use crate::{
     control_file,
-    safekeeper::{AcceptorState, PgUuid, TermHistory, TermLsn, UNKNOWN_SERVER_VERSION},
+    safekeeper::{AcceptorState, PersistedPeerInfo, PgUuid, TermHistory, UNKNOWN_SERVER_VERSION},
     timeline::TimelineError,
     wal_backup_partial::{self},
 };
@@ -32,8 +27,6 @@ pub struct TimelinePersistentState {
     pub tenant_id: TenantId,
     #[serde(with = "hex")]
     pub timeline_id: TimelineId,
-    /// Membership configuration.
-    pub mconf: Configuration,
     /// persistent acceptor state
     pub acceptor_state: AcceptorState,
     /// information about server
@@ -65,15 +58,22 @@ pub struct TimelinePersistentState {
     /// pushed to s3. We don't remove WAL beyond it. Persisted only for
     /// informational purposes, we receive it from pageserver (or broker).
     pub remote_consistent_lsn: Lsn,
+    /// Peers and their state as we remember it. Knowing peers themselves is
+    /// fundamental; but state is saved here only for informational purposes and
+    /// obviously can be stale. (Currently not saved at all, but let's provision
+    /// place to have less file version upgrades).
+    pub peers: PersistedPeers,
     /// Holds names of partial segments uploaded to remote storage. Used to
     /// clean up old objects without leaving garbage in remote storage.
     pub partial_backup: wal_backup_partial::State,
     /// Eviction state of the timeline. If it's Offloaded, we should download
     /// WAL files from remote storage to serve the timeline.
     pub eviction_state: EvictionState,
-    pub creation_ts: SystemTime,
 }
 
+#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
+pub struct PersistedPeers(pub Vec<(NodeId, PersistedPeerInfo)>);
+
 /// State of the local WAL files. Used to track current timeline state,
 /// that can be either WAL files are present on disk or last partial segment
 /// is offloaded to remote storage.
@@ -87,14 +87,12 @@ pub enum EvictionState {
 }
 
 impl TimelinePersistentState {
-    /// commit_lsn is the same as start_lsn in the normal creaiton; see
-    /// `TimelineCreateRequest` comments.`
     pub fn new(
         ttid: &TenantTimelineId,
-        mconf: Configuration,
         server_info: ServerInfo,
-        start_lsn: Lsn,
+        peers: Vec<NodeId>,
         commit_lsn: Lsn,
+        local_start_lsn: Lsn,
     ) -> anyhow::Result<TimelinePersistentState> {
         if server_info.wal_seg_size == 0 {
             bail!(TimelineError::UninitializedWalSegSize(*ttid));
@@ -104,59 +102,49 @@ impl TimelinePersistentState {
             bail!(TimelineError::UninitialinzedPgVersion(*ttid));
         }
 
-        if commit_lsn < start_lsn {
+        if commit_lsn < local_start_lsn {
             bail!(
-                "commit_lsn {} is smaller than start_lsn {}",
+                "commit_lsn {} is smaller than local_start_lsn {}",
                 commit_lsn,
-                start_lsn
+                local_start_lsn
             );
         }
 
-        // If we are given with init LSN, initialize term history with it. It
-        // ensures that walproposer always must be able to find a common point
-        // in histories; if it can't something is corrupted. Not having LSN here
-        // is so far left for legacy case where timeline is created by compute
-        // and LSN during creation is not known yet.
-        let term_history = if commit_lsn != Lsn::INVALID {
-            TermHistory(vec![TermLsn {
-                term: INITIAL_TERM,
-                lsn: start_lsn,
-            }])
-        } else {
-            TermHistory::empty()
-        };
-
         Ok(TimelinePersistentState {
             tenant_id: ttid.tenant_id,
             timeline_id: ttid.timeline_id,
-            mconf,
             acceptor_state: AcceptorState {
-                term: INITIAL_TERM,
-                term_history,
+                term: 0,
+                term_history: TermHistory::empty(),
             },
             server: server_info,
             proposer_uuid: [0; 16],
-            timeline_start_lsn: start_lsn,
-            local_start_lsn: start_lsn,
+            timeline_start_lsn: Lsn(0),
+            local_start_lsn,
             commit_lsn,
-            backup_lsn: start_lsn,
-            peer_horizon_lsn: start_lsn,
+            backup_lsn: local_start_lsn,
+            peer_horizon_lsn: local_start_lsn,
             remote_consistent_lsn: Lsn(0),
+            peers: PersistedPeers(
+                peers
+                    .iter()
+                    .map(|p| (*p, PersistedPeerInfo::new()))
+                    .collect(),
+            ),
             partial_backup: wal_backup_partial::State::default(),
             eviction_state: EvictionState::Present,
-            creation_ts: SystemTime::now(),
         })
     }
 
     pub fn empty() -> Self {
         TimelinePersistentState::new(
             &TenantTimelineId::empty(),
-            Configuration::empty(),
             ServerInfo {
                 pg_version: 170000, /* Postgres server version (major * 10000) */
                 system_id: 0,       /* Postgres system identifier */
                 wal_seg_size: WAL_SEGMENT_SIZE as u32,
             },
+            vec![],
             Lsn::INVALID,
             Lsn::INVALID,
         )
@@ -261,31 +249,6 @@ where
             current_term: after,
         })
     }
-
-    /// Switch into membership configuration `to` if it is higher than the
-    /// current one.
-    pub async fn membership_switch(
-        &mut self,
-        to: Configuration,
-    ) -> Result<TimelineMembershipSwitchResponse> {
-        let before = self.mconf.clone();
-        // Is switch allowed?
-        if to.generation <= self.mconf.generation {
-            info!(
-                "ignoring request to switch membership conf to lower {}, current conf {}",
-                to, self.mconf
-            );
-        } else {
-            let mut state = self.start_change();
-            state.mconf = to.clone();
-            self.finish_change(&state).await?;
-            info!("switched membership conf to {} from {}", to, before);
-        }
-        Ok(TimelineMembershipSwitchResponse {
-            previous_conf: before,
-            current_conf: self.mconf.clone(),
-        })
-    }
 }
 
 impl<CTRL> Deref for TimelineState<CTRL>

safekeeper/src/test_utils.rs

@@ -1,19 +1,13 @@
 use std::sync::Arc;
 
 use crate::rate_limit::RateLimiter;
-use crate::receive_wal::WalAcceptor;
-use crate::safekeeper::{
-    AcceptorProposerMessage, AppendRequest, AppendRequestHeader, ProposerAcceptorMessage,
-    ProposerElected, SafeKeeper, TermHistory,
-};
-use crate::send_wal::EndWatch;
+use crate::safekeeper::{ProposerAcceptorMessage, ProposerElected, SafeKeeper, TermHistory};
 use crate::state::{TimelinePersistentState, TimelineState};
 use crate::timeline::{get_timeline_dir, SharedState, StateSK, Timeline};
 use crate::timelines_set::TimelinesSet;
 use crate::wal_backup::remote_timeline_path;
-use crate::{control_file, receive_wal, wal_storage, SafeKeeperConf};
+use crate::{control_file, wal_storage, SafeKeeperConf};
 use camino_tempfile::Utf8TempDir;
-use postgres_ffi::v17::wal_generator::{LogicalMessageGenerator, WalGenerator};
 use tokio::fs::create_dir_all;
 use utils::id::{NodeId, TenantTimelineId};
 use utils::lsn::Lsn;
@@ -113,59 +107,4 @@ impl Env {
         );
         Ok(timeline)
     }
-
-    // This will be dead code when building a non-benchmark target with the
-    // benchmarking feature enabled.
-    #[allow(dead_code)]
-    pub(crate) async fn write_wal(
-        tli: Arc<Timeline>,
-        start_lsn: Lsn,
-        msg_size: usize,
-        msg_count: usize,
-    ) -> anyhow::Result<EndWatch> {
-        let (msg_tx, msg_rx) = tokio::sync::mpsc::channel(receive_wal::MSG_QUEUE_SIZE);
-        let (reply_tx, mut reply_rx) = tokio::sync::mpsc::channel(receive_wal::REPLY_QUEUE_SIZE);
-
-        let end_watch = EndWatch::Commit(tli.get_commit_lsn_watch_rx());
-
-        WalAcceptor::spawn(tli.wal_residence_guard().await?, msg_rx, reply_tx, Some(0));
-
-        let prefix = c"p";
-        let prefixlen = prefix.to_bytes_with_nul().len();
-        assert!(msg_size >= prefixlen);
-        let message = vec![0; msg_size - prefixlen];
-
-        let walgen =
-            &mut WalGenerator::new(LogicalMessageGenerator::new(prefix, &message), start_lsn);
-        for _ in 0..msg_count {
-            let (lsn, record) = walgen.next().unwrap();
-
-            let req = AppendRequest {
-                h: AppendRequestHeader {
-                    term: 1,
-                    term_start_lsn: start_lsn,
-                    begin_lsn: lsn,
-                    end_lsn: lsn + record.len() as u64,
-                    commit_lsn: lsn,
-                    truncate_lsn: Lsn(0),
-                    proposer_uuid: [0; 16],
-                },
-                wal_data: record,
-            };
-
-            let end_lsn = req.h.end_lsn;
-
-            let msg = ProposerAcceptorMessage::AppendRequest(req);
-            msg_tx.send(msg).await?;
-            while let Some(reply) = reply_rx.recv().await {
-                if let AcceptorProposerMessage::AppendResponse(resp) = reply {
-                    if resp.flush_lsn >= end_lsn {
-                        break;
-                    }
-                }
-            }
-        }
-
-        Ok(end_watch)
-    }
 }

safekeeper/src/timeline.rs

@@ -4,10 +4,7 @@
 use anyhow::{anyhow, bail, Result};
 use camino::{Utf8Path, Utf8PathBuf};
 use remote_storage::RemotePath;
-use safekeeper_api::membership::Configuration;
-use safekeeper_api::models::{
-    PeerInfo, TimelineMembershipSwitchResponse, TimelineTermBumpResponse,
-};
+use safekeeper_api::models::{PeerInfo, TimelineTermBumpResponse};
 use safekeeper_api::Term;
 use tokio::fs::{self};
 use tokio_util::sync::CancellationToken;
@@ -35,7 +32,7 @@ use crate::control_file;
 use crate::rate_limit::RateLimiter;
 use crate::receive_wal::WalReceivers;
 use crate::safekeeper::{AcceptorProposerMessage, ProposerAcceptorMessage, SafeKeeper, TermLsn};
-use crate::send_wal::{WalSenders, WalSendersTimelineMetricValues};
+use crate::send_wal::WalSenders;
 use crate::state::{EvictionState, TimelineMemState, TimelinePersistentState, TimelineState};
 use crate::timeline_guard::ResidenceGuard;
 use crate::timeline_manager::{AtomicStatus, ManagerCtl};
@@ -191,13 +188,6 @@ impl StateSK {
         self.state_mut().term_bump(to).await
     }
 
-    pub async fn membership_switch(
-        &mut self,
-        to: Configuration,
-    ) -> Result<TimelineMembershipSwitchResponse> {
-        self.state_mut().membership_switch(to).await
-    }
-
     /// Close open WAL files to release FDs.
     fn close_wal_store(&mut self) {
         if let StateSK::Loaded(sk) = self {
@@ -712,22 +702,16 @@ impl Timeline {
             return None;
         }
 
-        let WalSendersTimelineMetricValues {
-            ps_feedback_counter,
-            last_ps_feedback,
-            interpreted_wal_reader_tasks,
-        } = self.walsenders.info_for_metrics();
-
+        let (ps_feedback_count, last_ps_feedback) = self.walsenders.get_ps_feedback_stats();
         let state = self.read_shared_state().await;
         Some(FullTimelineInfo {
             ttid: self.ttid,
-            ps_feedback_count: ps_feedback_counter,
+            ps_feedback_count,
             last_ps_feedback,
             wal_backup_active: self.wal_backup_active.load(Ordering::Relaxed),
             timeline_is_active: self.broker_active.load(Ordering::Relaxed),
             num_computes: self.walreceivers.get_num() as u32,
             last_removed_segno: self.last_removed_segno.load(Ordering::Relaxed),
-            interpreted_wal_reader_tasks,
             epoch_start_lsn: state.sk.term_start_lsn(),
             mem_state: state.sk.state().inmem.clone(),
             persisted_state: TimelinePersistentState::clone(state.sk.state()),
@@ -746,7 +730,7 @@ impl Timeline {
         debug_dump::Memory {
             is_cancelled: self.is_cancelled(),
             peers_info_len: state.peers_info.0.len(),
-            walsenders: self.walsenders.get_all_public(),
+            walsenders: self.walsenders.get_all(),
             wal_backup_active: self.wal_backup_active.load(Ordering::Relaxed),
             active: self.broker_active.load(Ordering::Relaxed),
             num_computes: self.walreceivers.get_num() as u32,
@@ -784,14 +768,6 @@ impl Timeline {
         state.sk.term_bump(to).await
     }
 
-    pub async fn membership_switch(
-        self: &Arc<Self>,
-        to: Configuration,
-    ) -> Result<TimelineMembershipSwitchResponse> {
-        let mut state = self.write_shared_state().await;
-        state.sk.membership_switch(to).await
-    }
-
     /// Guts of [`Self::wal_residence_guard`] and [`Self::try_wal_residence_guard`]
     async fn do_wal_residence_guard(
         self: &Arc<Self>,

safekeeper/src/timelines_global_map.rs

@@ -12,7 +12,6 @@ use crate::{control_file, wal_storage, SafeKeeperConf};
 use anyhow::{bail, Context, Result};
 use camino::Utf8PathBuf;
 use camino_tempfile::Utf8TempDir;
-use safekeeper_api::membership::Configuration;
 use safekeeper_api::ServerInfo;
 use serde::Serialize;
 use std::collections::HashMap;
@@ -215,10 +214,9 @@ impl GlobalTimelines {
     pub(crate) async fn create(
         &self,
         ttid: TenantTimelineId,
-        mconf: Configuration,
         server_info: ServerInfo,
-        start_lsn: Lsn,
         commit_lsn: Lsn,
+        local_start_lsn: Lsn,
     ) -> Result<Arc<Timeline>> {
         let (conf, _, _) = {
             let state = self.state.lock().unwrap();
@@ -241,7 +239,8 @@ impl GlobalTimelines {
 
         // TODO: currently we create only cfile. It would be reasonable to
         // immediately initialize first WAL segment as well.
-        let state = TimelinePersistentState::new(&ttid, mconf, server_info, start_lsn, commit_lsn)?;
+        let state =
+            TimelinePersistentState::new(&ttid, server_info, vec![], commit_lsn, local_start_lsn)?;
         control_file::FileStorage::create_new(&tmp_dir_path, state, conf.no_sync).await?;
         let timeline = self.load_temp_timeline(ttid, &tmp_dir_path, true).await?;
         Ok(timeline)

safekeeper/src/wal_reader_stream.rs

@@ -1,16 +1,34 @@
-use std::{
-    pin::Pin,
-    task::{Context, Poll},
-};
+use std::sync::Arc;
 
+use async_stream::try_stream;
 use bytes::Bytes;
-use futures::{stream::BoxStream, Stream, StreamExt};
+use futures::Stream;
+use postgres_backend::CopyStreamHandlerEnd;
+use safekeeper_api::Term;
+use std::time::Duration;
+use tokio::time::timeout;
 use utils::lsn::Lsn;
 
-use crate::{send_wal::EndWatch, timeline::WalResidentTimeline, wal_storage::WalReader};
-use safekeeper_api::Term;
+use crate::{
+    send_wal::{EndWatch, WalSenderGuard},
+    timeline::WalResidentTimeline,
+};
+
+pub(crate) struct WalReaderStreamBuilder {
+    pub(crate) tli: WalResidentTimeline,
+    pub(crate) start_pos: Lsn,
+    pub(crate) end_pos: Lsn,
+    pub(crate) term: Option<Term>,
+    pub(crate) end_watch: EndWatch,
+    pub(crate) wal_sender_guard: Arc<WalSenderGuard>,
+}
+
+impl WalReaderStreamBuilder {
+    pub(crate) fn start_pos(&self) -> Lsn {
+        self.start_pos
+    }
+}
 
-#[derive(PartialEq, Eq, Debug)]
 pub(crate) struct WalBytes {
     /// Raw PG WAL
     pub(crate) wal: Bytes,
@@ -26,270 +44,106 @@ pub(crate) struct WalBytes {
     pub(crate) available_wal_end_lsn: Lsn,
 }
 
-struct PositionedWalReader {
-    start: Lsn,
-    end: Lsn,
-    reader: Option<WalReader>,
-}
-
-/// A streaming WAL reader wrapper which can be reset while running
-pub(crate) struct StreamingWalReader {
-    stream: BoxStream<'static, WalOrReset>,
-    start_changed_tx: tokio::sync::watch::Sender<Lsn>,
-}
-
-pub(crate) enum WalOrReset {
-    Wal(anyhow::Result<WalBytes>),
-    Reset(Lsn),
-}
-
-impl WalOrReset {
-    pub(crate) fn get_wal(self) -> Option<anyhow::Result<WalBytes>> {
-        match self {
-            WalOrReset::Wal(wal) => Some(wal),
-            WalOrReset::Reset(_) => None,
-        }
-    }
-}
-
-impl StreamingWalReader {
-    pub(crate) fn new(
-        tli: WalResidentTimeline,
-        term: Option<Term>,
-        start: Lsn,
-        end: Lsn,
-        end_watch: EndWatch,
+impl WalReaderStreamBuilder {
+    /// Builds a stream of Postgres WAL starting from [`Self::start_pos`].
+    /// The stream terminates when the receiver (pageserver) is fully caught up
+    /// and there's no active computes.
+    pub(crate) async fn build(
+        self,
         buffer_size: usize,
-    ) -> Self {
-        let (start_changed_tx, start_changed_rx) = tokio::sync::watch::channel(start);
-
-        let state = WalReaderStreamState {
+    ) -> anyhow::Result<impl Stream<Item = Result<WalBytes, CopyStreamHandlerEnd>>> {
+        // TODO(vlad): The code below duplicates functionality from [`crate::send_wal`].
+        // We can make the raw WAL sender use this stream too and remove the duplication.
+        let Self {
             tli,
-            wal_reader: PositionedWalReader {
-                start,
-                end,
-                reader: None,
-            },
+            mut start_pos,
+            mut end_pos,
             term,
-            end_watch,
-            buffer: vec![0; buffer_size],
-            buffer_size,
-        };
+            mut end_watch,
+            wal_sender_guard,
+        } = self;
+        let mut wal_reader = tli.get_walreader(start_pos).await?;
+        let mut buffer = vec![0; buffer_size];
 
-        // When a change notification is received while polling the internal
-        // reader, stop polling the read future and service the change.
-        let stream = futures::stream::unfold(
-            (state, start_changed_rx),
-            |(mut state, mut rx)| async move {
-                let wal_or_reset = tokio::select! {
-                    read_res = state.read() => { WalOrReset::Wal(read_res) },
-                    changed_res = rx.changed() => {
-                        if changed_res.is_err() {
-                            return None;
+        const POLL_STATE_TIMEOUT: Duration = Duration::from_secs(1);
+
+        Ok(try_stream! {
+            loop {
+                let have_something_to_send = end_pos > start_pos;
+
+                if !have_something_to_send {
+                    // wait for lsn
+                    let res = timeout(POLL_STATE_TIMEOUT, end_watch.wait_for_lsn(start_pos, term)).await;
+                    match res {
+                        Ok(ok) => {
+                            end_pos = ok?;
+                        },
+                        Err(_) => {
+                            if let EndWatch::Commit(_) = end_watch {
+                                if let Some(remote_consistent_lsn) = wal_sender_guard
+                                    .walsenders()
+                                    .get_ws_remote_consistent_lsn(wal_sender_guard.id())
+                                {
+                                    if tli.should_walsender_stop(remote_consistent_lsn).await {
+                                        // Stop streaming if the receivers are caught up and
+                                        // there's no active compute. This causes the loop in
+                                        // [`crate::send_interpreted_wal::InterpretedWalSender::run`]
+                                        // to exit and terminate the WAL stream.
+                                        return;
+                                    }
+                                }
+                            }
+
+                            continue;
                         }
-
-                        let new_start_pos = rx.borrow_and_update();
-                        WalOrReset::Reset(*new_start_pos)
                     }
+                }
+
+
+                assert!(
+                    end_pos > start_pos,
+                    "nothing to send after waiting for WAL"
+                );
+
+                // try to send as much as available, capped by the buffer size
+                let mut chunk_end_pos = start_pos + buffer_size as u64;
+                // if we went behind available WAL, back off
+                if chunk_end_pos >= end_pos {
+                    chunk_end_pos = end_pos;
+                } else {
+                    // If sending not up to end pos, round down to page boundary to
+                    // avoid breaking WAL record not at page boundary, as protocol
+                    // demands. See walsender.c (XLogSendPhysical).
+                    chunk_end_pos = chunk_end_pos
+                        .checked_sub(chunk_end_pos.block_offset())
+                        .unwrap();
+                }
+                let send_size = (chunk_end_pos.0 - start_pos.0) as usize;
+                let buffer = &mut buffer[..send_size];
+                let send_size: usize;
+                {
+                    // If uncommitted part is being pulled, check that the term is
+                    // still the expected one.
+                    let _term_guard = if let Some(t) = term {
+                        Some(tli.acquire_term(t).await?)
+                    } else {
+                        None
+                    };
+                    // Read WAL into buffer. send_size can be additionally capped to
+                    // segment boundary here.
+                    send_size = wal_reader.read(buffer).await?
+                };
+                let wal = Bytes::copy_from_slice(&buffer[..send_size]);
+
+                yield WalBytes {
+                    wal,
+                    wal_start_lsn: start_pos,
+                    wal_end_lsn: start_pos + send_size as u64,
+                    available_wal_end_lsn: end_pos
                 };
 
-                if let WalOrReset::Reset(lsn) = wal_or_reset {
-                    state.wal_reader.start = lsn;
-                    state.wal_reader.reader = None;
-                }
-
-                Some((wal_or_reset, (state, rx)))
-            },
-        )
-        .boxed();
-
-        Self {
-            stream,
-            start_changed_tx,
-        }
-    }
-
-    /// Reset the stream to a given position.
-    pub(crate) async fn reset(&mut self, start: Lsn) {
-        self.start_changed_tx.send(start).unwrap();
-        while let Some(wal_or_reset) = self.stream.next().await {
-            match wal_or_reset {
-                WalOrReset::Reset(at) => {
-                    // Stream confirmed the reset.
-                    // There may only one ongoing reset at any given time,
-                    // hence the assertion.
-                    assert_eq!(at, start);
-                    break;
-                }
-                WalOrReset::Wal(_) => {
-                    // Ignore wal generated before reset was handled
-                }
+                start_pos += send_size as u64;
             }
-        }
-    }
-}
-
-impl Stream for StreamingWalReader {
-    type Item = WalOrReset;
-
-    fn poll_next(mut self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Option<Self::Item>> {
-        Pin::new(&mut self.stream).poll_next(cx)
-    }
-}
-
-struct WalReaderStreamState {
-    tli: WalResidentTimeline,
-    wal_reader: PositionedWalReader,
-    term: Option<Term>,
-    end_watch: EndWatch,
-    buffer: Vec<u8>,
-    buffer_size: usize,
-}
-
-impl WalReaderStreamState {
-    async fn read(&mut self) -> anyhow::Result<WalBytes> {
-        // Create reader if needed
-        if self.wal_reader.reader.is_none() {
-            self.wal_reader.reader = Some(self.tli.get_walreader(self.wal_reader.start).await?);
-        }
-
-        let have_something_to_send = self.wal_reader.end > self.wal_reader.start;
-        if !have_something_to_send {
-            tracing::debug!(
-                "Waiting for wal: start={}, end={}",
-                self.wal_reader.end,
-                self.wal_reader.start
-            );
-            self.wal_reader.end = self
-                .end_watch
-                .wait_for_lsn(self.wal_reader.start, self.term)
-                .await?;
-            tracing::debug!(
-                "Done waiting for wal: start={}, end={}",
-                self.wal_reader.end,
-                self.wal_reader.start
-            );
-        }
-
-        assert!(
-            self.wal_reader.end > self.wal_reader.start,
-            "nothing to send after waiting for WAL"
-        );
-
-        // Calculate chunk size
-        let mut chunk_end_pos = self.wal_reader.start + self.buffer_size as u64;
-        if chunk_end_pos >= self.wal_reader.end {
-            chunk_end_pos = self.wal_reader.end;
-        } else {
-            chunk_end_pos = chunk_end_pos
-                .checked_sub(chunk_end_pos.block_offset())
-                .unwrap();
-        }
-
-        let send_size = (chunk_end_pos.0 - self.wal_reader.start.0) as usize;
-        let buffer = &mut self.buffer[..send_size];
-
-        // Read WAL
-        let send_size = {
-            let _term_guard = if let Some(t) = self.term {
-                Some(self.tli.acquire_term(t).await?)
-            } else {
-                None
-            };
-            self.wal_reader
-                .reader
-                .as_mut()
-                .unwrap()
-                .read(buffer)
-                .await?
-        };
-
-        let wal = Bytes::copy_from_slice(&buffer[..send_size]);
-        let result = WalBytes {
-            wal,
-            wal_start_lsn: self.wal_reader.start,
-            wal_end_lsn: self.wal_reader.start + send_size as u64,
-            available_wal_end_lsn: self.wal_reader.end,
-        };
-
-        self.wal_reader.start += send_size as u64;
-
-        Ok(result)
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use std::str::FromStr;
-
-    use futures::StreamExt;
-    use postgres_ffi::MAX_SEND_SIZE;
-    use utils::{
-        id::{NodeId, TenantTimelineId},
-        lsn::Lsn,
-    };
-
-    use crate::{test_utils::Env, wal_reader_stream::StreamingWalReader};
-
-    #[tokio::test]
-    async fn test_streaming_wal_reader_reset() {
-        let _ = env_logger::builder().is_test(true).try_init();
-
-        const SIZE: usize = 8 * 1024;
-        const MSG_COUNT: usize = 200;
-
-        let start_lsn = Lsn::from_str("0/149FD18").unwrap();
-        let env = Env::new(true).unwrap();
-        let tli = env
-            .make_timeline(NodeId(1), TenantTimelineId::generate(), start_lsn)
-            .await
-            .unwrap();
-
-        let resident_tli = tli.wal_residence_guard().await.unwrap();
-        let end_watch = Env::write_wal(tli, start_lsn, SIZE, MSG_COUNT)
-            .await
-            .unwrap();
-        let end_pos = end_watch.get();
-
-        tracing::info!("Doing first round of reads ...");
-
-        let mut streaming_wal_reader = StreamingWalReader::new(
-            resident_tli,
-            None,
-            start_lsn,
-            end_pos,
-            end_watch,
-            MAX_SEND_SIZE,
-        );
-
-        let mut before_reset = Vec::new();
-        while let Some(wor) = streaming_wal_reader.next().await {
-            let wal = wor.get_wal().unwrap().unwrap();
-            let stop = wal.available_wal_end_lsn == wal.wal_end_lsn;
-            before_reset.push(wal);
-
-            if stop {
-                break;
-            }
-        }
-
-        tracing::info!("Resetting the WAL stream ...");
-
-        streaming_wal_reader.reset(start_lsn).await;
-
-        tracing::info!("Doing second round of reads ...");
-
-        let mut after_reset = Vec::new();
-        while let Some(wor) = streaming_wal_reader.next().await {
-            let wal = wor.get_wal().unwrap().unwrap();
-            let stop = wal.available_wal_end_lsn == wal.wal_end_lsn;
-            after_reset.push(wal);
-
-            if stop {
-                break;
-            }
-        }
-
-        assert_eq!(before_reset, after_reset);
+        })
     }
 }

safekeeper/tests/walproposer_sim/safekeeper.rs

@@ -15,15 +15,13 @@ use desim::{
 };
 use http::Uri;
 use safekeeper::{
-    safekeeper::{
-        ProposerAcceptorMessage, SafeKeeper, SK_PROTOCOL_VERSION, UNKNOWN_SERVER_VERSION,
-    },
+    safekeeper::{ProposerAcceptorMessage, SafeKeeper, UNKNOWN_SERVER_VERSION},
     state::{TimelinePersistentState, TimelineState},
     timeline::TimelineError,
     wal_storage::Storage,
     SafeKeeperConf,
 };
-use safekeeper_api::{membership::Configuration, ServerInfo};
+use safekeeper_api::ServerInfo;
 use tracing::{debug, info_span, warn};
 use utils::{
     id::{NodeId, TenantId, TenantTimelineId, TimelineId},
@@ -98,13 +96,8 @@ impl GlobalMap {
         let commit_lsn = Lsn::INVALID;
         let local_start_lsn = Lsn::INVALID;
 
-        let state = TimelinePersistentState::new(
-            &ttid,
-            Configuration::empty(),
-            server_info,
-            commit_lsn,
-            local_start_lsn,
-        )?;
+        let state =
+            TimelinePersistentState::new(&ttid, server_info, vec![], commit_lsn, local_start_lsn)?;
 
         let disk_timeline = self.disk.put_state(&ttid, state);
         let control_store = DiskStateStorage::new(disk_timeline.clone());
@@ -180,8 +173,6 @@ pub fn run_server(os: NodeOs, disk: Arc<SafekeeperDisk>) -> Result<()> {
         control_file_save_interval: Duration::from_secs(1),
         partial_backup_concurrency: 1,
         eviction_min_resident: Duration::ZERO,
-        wal_reader_fanout: false,
-        max_delta_for_fanout: None,
     };
 
     let mut global = GlobalMap::new(disk, conf.clone())?;
@@ -287,7 +278,7 @@ impl ConnState {
                 bail!("finished processing START_REPLICATION")
             }
 
-            let msg = ProposerAcceptorMessage::parse(copy_data, SK_PROTOCOL_VERSION)?;
+            let msg = ProposerAcceptorMessage::parse(copy_data)?;
             debug!("got msg: {:?}", msg);
             self.process(msg, global)
         } else {

storage_controller/Cargo.toml

@@ -55,4 +55,4 @@ r2d2 = { version = "0.8.10" }
 utils = { path = "../libs/utils/" }
 metrics = { path = "../libs/metrics/" }
 control_plane = { path = "../control_plane" }
-workspace_hack = { version = "0.1", path = "../workspace_hack" }
+workspace_hack = { version = "0.1", path = "../workspace_hack" }

storage_controller/migrations/2025-01-15-181207_safekeepers_disabled_to_pause/down.sql

@@ -1,2 +0,0 @@
-ALTER TABLE safekeepers ALTER COLUMN scheduling_policy SET DEFAULT 'disabled';
-UPDATE safekeepers SET scheduling_policy = 'disabled' WHERE scheduling_policy = 'pause';

storage_controller/migrations/2025-01-15-181207_safekeepers_disabled_to_pause/up.sql

@@ -1,2 +0,0 @@
-ALTER TABLE safekeepers ALTER COLUMN scheduling_policy SET DEFAULT 'pause';
-UPDATE safekeepers SET scheduling_policy = 'pause' WHERE scheduling_policy = 'disabled';

storage_controller/src/drain_utils.rs

@@ -112,7 +112,7 @@ impl TenantShardDrain {
             }
         }
 
-        match tenant_shard.preferred_secondary(scheduler) {
+        match scheduler.node_preferred(tenant_shard.intent.get_secondary()) {
             Some(node) => Some(node),
             None => {
                 tracing::warn!(

storage_controller/src/http.rs

@@ -15,7 +15,7 @@ use metrics::{BuildInfo, NeonMetrics};
 use pageserver_api::controller_api::{
     MetadataHealthListOutdatedRequest, MetadataHealthListOutdatedResponse,
     MetadataHealthListUnhealthyResponse, MetadataHealthUpdateRequest, MetadataHealthUpdateResponse,
-    SafekeeperSchedulingPolicyRequest, ShardsPreferredAzsRequest, TenantCreateRequest,
+    ShardsPreferredAzsRequest, TenantCreateRequest,
 };
 use pageserver_api::models::{
     TenantConfigPatchRequest, TenantConfigRequest, TenantLocationConfigRequest,
@@ -653,10 +653,6 @@ async fn handle_tenant_list(
 ) -> Result<Response<Body>, ApiError> {
     check_permissions(&req, Scope::Admin)?;
 
-    let limit: Option<usize> = parse_query_param(&req, "limit")?;
-    let start_after: Option<TenantId> = parse_query_param(&req, "start_after")?;
-    tracing::info!("start_after: {:?}", start_after);
-
     match maybe_forward(req).await {
         ForwardOutcome::Forwarded(res) => {
             return res;
@@ -664,7 +660,7 @@ async fn handle_tenant_list(
         ForwardOutcome::NotForwarded(_req) => {}
     };
 
-    json_response(StatusCode::OK, service.tenant_list(limit, start_after))
+    json_response(StatusCode::OK, service.tenant_list())
 }
 
 async fn handle_node_register(req: Request<Body>) -> Result<Response<Body>, ApiError> {
@@ -1305,35 +1301,6 @@ async fn handle_upsert_safekeeper(mut req: Request<Body>) -> Result<Response<Bod
         .unwrap())
 }
 
-/// Sets the scheduling policy of the specified safekeeper
-async fn handle_safekeeper_scheduling_policy(
-    mut req: Request<Body>,
-) -> Result<Response<Body>, ApiError> {
-    check_permissions(&req, Scope::Admin)?;
-
-    let body = json_request::<SafekeeperSchedulingPolicyRequest>(&mut req).await?;
-    let id = parse_request_param::<i64>(&req, "id")?;
-
-    let req = match maybe_forward(req).await {
-        ForwardOutcome::Forwarded(res) => {
-            return res;
-        }
-        ForwardOutcome::NotForwarded(req) => req,
-    };
-
-    let state = get_state(&req);
-
-    state
-        .service
-        .set_safekeeper_scheduling_policy(id, body.scheduling_policy)
-        .await?;
-
-    Ok(Response::builder()
-        .status(StatusCode::NO_CONTENT)
-        .body(Body::empty())
-        .unwrap())
-}
-
 /// Common wrapper for request handlers that call into Service and will operate on tenants: they must only
 /// be allowed to run if Service has finished its initial reconciliation.
 async fn tenant_service_handler<R, H>(
@@ -1902,18 +1869,7 @@ pub fn make_router(
         })
         .post("/control/v1/safekeeper/:id", |r| {
             // id is in the body
-            named_request_span(
-                r,
-                handle_upsert_safekeeper,
-                RequestName("v1_safekeeper_post"),
-            )
-        })
-        .post("/control/v1/safekeeper/:id/scheduling_policy", |r| {
-            named_request_span(
-                r,
-                handle_safekeeper_scheduling_policy,
-                RequestName("v1_safekeeper_status"),
-            )
+            named_request_span(r, handle_upsert_safekeeper, RequestName("v1_safekeeper"))
         })
         // Tenant Shard operations
         .put("/control/v1/tenant/:tenant_shard_id/migrate", |r| {

storage_controller/src/metrics.rs

@@ -53,16 +53,6 @@ pub(crate) struct StorageControllerMetricGroup {
     /// How many shards are not scheduled into their preferred AZ
     pub(crate) storage_controller_schedule_az_violation: measured::Gauge,
 
-    /// How many shard locations (secondary or attached) on each node
-    pub(crate) storage_controller_node_shards: measured::GaugeVec<NodeLabelGroupSet>,
-
-    /// How many _attached_ shard locations on each node
-    pub(crate) storage_controller_node_attached_shards: measured::GaugeVec<NodeLabelGroupSet>,
-
-    /// How many _home_ shard locations on each node (i.e. the node's AZ matches the shard's
-    /// preferred AZ)
-    pub(crate) storage_controller_node_home_shards: measured::GaugeVec<NodeLabelGroupSet>,
-
     /// How many shards would like to reconcile but were blocked by concurrency limits
     pub(crate) storage_controller_pending_reconciles: measured::Gauge,
 
@@ -142,15 +132,6 @@ impl Default for StorageControllerMetrics {
     }
 }
 
-#[derive(measured::LabelGroup, Clone)]
-#[label(set = NodeLabelGroupSet)]
-pub(crate) struct NodeLabelGroup<'a> {
-    #[label(dynamic_with = lasso::ThreadedRodeo, default)]
-    pub(crate) az: &'a str,
-    #[label(dynamic_with = lasso::ThreadedRodeo, default)]
-    pub(crate) node_id: &'a str,
-}
-
 #[derive(measured::LabelGroup)]
 #[label(set = ReconcileCompleteLabelGroupSet)]
 pub(crate) struct ReconcileCompleteLabelGroup {

storage_controller/src/node.rs

@@ -299,7 +299,6 @@ impl Node {
             id: self.id,
             availability: self.availability.clone().into(),
             scheduling: self.scheduling,
-            availability_zone_id: self.availability_zone_id.0.clone(),
             listen_http_addr: self.listen_http_addr.clone(),
             listen_http_port: self.listen_http_port,
             listen_pg_addr: self.listen_pg_addr.clone(),

storage_controller/src/persistence.rs

@@ -708,11 +708,10 @@ impl Persistence {
         Ok(())
     }
 
-    /// Note that passing None for a shard clears the preferred AZ (rather than leaving it unmodified)
     pub(crate) async fn set_tenant_shard_preferred_azs(
         &self,
-        preferred_azs: Vec<(TenantShardId, Option<AvailabilityZone>)>,
-    ) -> DatabaseResult<Vec<(TenantShardId, Option<AvailabilityZone>)>> {
+        preferred_azs: Vec<(TenantShardId, AvailabilityZone)>,
+    ) -> DatabaseResult<Vec<(TenantShardId, AvailabilityZone)>> {
         use crate::schema::tenant_shards::dsl::*;
 
         self.with_measured_conn(DatabaseOperation::SetPreferredAzs, move |conn| {
@@ -723,7 +722,7 @@ impl Persistence {
                     .filter(tenant_id.eq(tenant_shard_id.tenant_id.to_string()))
                     .filter(shard_number.eq(tenant_shard_id.shard_number.0 as i32))
                     .filter(shard_count.eq(tenant_shard_id.shard_count.literal() as i32))
-                    .set(preferred_az_id.eq(preferred_az.as_ref().map(|az| az.0.clone())))
+                    .set(preferred_az_id.eq(preferred_az.0.clone()))
                     .execute(conn)?;
 
                 if updated == 1 {
@@ -1104,37 +1103,6 @@ impl Persistence {
         })
         .await
     }
-
-    pub(crate) async fn set_safekeeper_scheduling_policy(
-        &self,
-        id_: i64,
-        scheduling_policy_: SkSchedulingPolicy,
-    ) -> Result<(), DatabaseError> {
-        use crate::schema::safekeepers::dsl::*;
-
-        self.with_conn(move |conn| -> DatabaseResult<()> {
-            #[derive(Insertable, AsChangeset)]
-            #[diesel(table_name = crate::schema::safekeepers)]
-            struct UpdateSkSchedulingPolicy<'a> {
-                id: i64,
-                scheduling_policy: &'a str,
-            }
-            let scheduling_policy_ = String::from(scheduling_policy_);
-
-            let rows_affected = diesel::update(safekeepers.filter(id.eq(id_)))
-                .set(scheduling_policy.eq(scheduling_policy_))
-                .execute(conn)?;
-
-            if rows_affected != 1 {
-                return Err(DatabaseError::Logical(format!(
-                    "unexpected number of rows ({rows_affected})",
-                )));
-            }
-
-            Ok(())
-        })
-        .await
-    }
 }
 
 /// Parts of [`crate::tenant_shard::TenantShard`] that are stored durably

storage_controller/src/reconciler.rs

@@ -826,21 +826,7 @@ impl Reconciler {
             if self.cancel.is_cancelled() {
                 return Err(ReconcileError::Cancel);
             }
-            // We only try to configure secondary locations if the node is available.  This does
-            // not stop us succeeding with the reconcile, because our core goal is to make the
-            // shard _available_ (the attached location), and configuring secondary locations
-            // can be done lazily when the node becomes available (via background reconciliation).
-            if node.is_available() {
-                self.location_config(&node, conf, None, false).await?;
-            } else {
-                // If the node is unavailable, we skip and consider the reconciliation successful: this
-                // is a common case where a pageserver is marked unavailable: we demote a location on
-                // that unavailable pageserver to secondary.
-                tracing::info!("Skipping configuring secondary location {node}, it is unavailable");
-                self.observed
-                    .locations
-                    .insert(node.get_id(), ObservedStateLocation { conf: None });
-            }
+            self.location_config(&node, conf, None, false).await?;
         }
 
         // The condition below identifies a detach. We must have no attached intent and

storage_controller/src/scheduler.rs

@@ -1,4 +1,4 @@
-use crate::{metrics::NodeLabelGroup, node::Node, tenant_shard::TenantShard};
+use crate::{node::Node, tenant_shard::TenantShard};
 use itertools::Itertools;
 use pageserver_api::{controller_api::AvailabilityZone, models::PageserverUtilization};
 use serde::Serialize;
@@ -32,9 +32,6 @@ pub(crate) struct SchedulerNode {
     shard_count: usize,
     /// How many shards are currently attached on this node, via their [`crate::tenant_shard::IntentState`].
     attached_shard_count: usize,
-    /// How many shards have a location on this node (via [`crate::tenant_shard::IntentState`]) _and_ this node
-    /// is in their preferred AZ (i.e. this is their 'home' location)
-    home_shard_count: usize,
     /// Availability zone id in which the node resides
     az: AvailabilityZone,
 
@@ -50,12 +47,6 @@ pub(crate) trait NodeSchedulingScore: Debug + Ord + Copy + Sized {
         preferred_az: &Option<AvailabilityZone>,
         context: &ScheduleContext,
     ) -> Option<Self>;
-
-    /// Return a score that drops any components based on node utilization: this is useful
-    /// for finding scores for scheduling optimisation, when we want to avoid rescheduling
-    /// shards due to e.g. disk usage, to avoid flapping.
-    fn for_optimization(&self) -> Self;
-
     fn is_overloaded(&self) -> bool;
     fn node_id(&self) -> NodeId;
 }
@@ -145,13 +136,17 @@ impl PartialOrd for SecondaryAzMatch {
 /// Ordering is given by member declaration order (top to bottom).
 #[derive(Debug, PartialEq, Eq, PartialOrd, Ord, Clone, Copy)]
 pub(crate) struct NodeAttachmentSchedulingScore {
+    /// The number of shards belonging to the tenant currently being
+    /// scheduled that are attached to this node.
+    affinity_score: AffinityScore,
     /// Flag indicating whether this node matches the preferred AZ
     /// of the shard. For equal affinity scores, nodes in the matching AZ
     /// are considered first.
     az_match: AttachmentAzMatch,
-    /// The number of shards belonging to the tenant currently being
-    /// scheduled that are attached to this node.
-    affinity_score: AffinityScore,
+    /// Size of [`ScheduleContext::attached_nodes`] for the current node.
+    /// This normally tracks the number of attached shards belonging to the
+    /// tenant being scheduled that are already on this node.
+    attached_shards_in_context: usize,
     /// Utilisation score that combines shard count and disk utilisation
     utilization_score: u64,
     /// Total number of shards attached to this node. When nodes have identical utilisation, this
@@ -182,25 +177,13 @@ impl NodeSchedulingScore for NodeAttachmentSchedulingScore {
                 .copied()
                 .unwrap_or(AffinityScore::FREE),
             az_match: AttachmentAzMatch(AzMatch::new(&node.az, preferred_az.as_ref())),
+            attached_shards_in_context: context.attached_nodes.get(node_id).copied().unwrap_or(0),
             utilization_score: utilization.cached_score(),
             total_attached_shard_count: node.attached_shard_count,
             node_id: *node_id,
         })
     }
 
-    /// For use in scheduling optimisation, where we only want to consider the aspects
-    /// of the score that can only be resolved by moving things (such as inter-shard affinity
-    /// and AZ affinity), and ignore aspects that reflect the total utilization of a node (which
-    /// can fluctuate for other reasons)
-    fn for_optimization(&self) -> Self {
-        Self {
-            utilization_score: 0,
-            total_attached_shard_count: 0,
-            node_id: NodeId(0),
-            ..*self
-        }
-    }
-
     fn is_overloaded(&self) -> bool {
         PageserverUtilization::is_overloaded(self.utilization_score)
     }
@@ -225,9 +208,9 @@ pub(crate) struct NodeSecondarySchedulingScore {
     affinity_score: AffinityScore,
     /// Utilisation score that combines shard count and disk utilisation
     utilization_score: u64,
-    /// Anti-affinity with other non-home locations: this gives the behavior that secondaries
-    /// will spread out across the nodes in an AZ.
-    total_non_home_shard_count: usize,
+    /// Total number of shards attached to this node. When nodes have identical utilisation, this
+    /// acts as an anti-affinity between attached shards.
+    total_attached_shard_count: usize,
     /// Convenience to make selection deterministic in tests and empty systems
     node_id: NodeId,
 }
@@ -254,20 +237,11 @@ impl NodeSchedulingScore for NodeSecondarySchedulingScore {
                 .copied()
                 .unwrap_or(AffinityScore::FREE),
             utilization_score: utilization.cached_score(),
-            total_non_home_shard_count: (node.shard_count - node.home_shard_count),
+            total_attached_shard_count: node.attached_shard_count,
             node_id: *node_id,
         })
     }
 
-    fn for_optimization(&self) -> Self {
-        Self {
-            utilization_score: 0,
-            total_non_home_shard_count: 0,
-            node_id: NodeId(0),
-            ..*self
-        }
-    }
-
     fn is_overloaded(&self) -> bool {
         PageserverUtilization::is_overloaded(self.utilization_score)
     }
@@ -319,10 +293,6 @@ impl AffinityScore {
     pub(crate) fn inc(&mut self) {
         self.0 += 1;
     }
-
-    pub(crate) fn dec(&mut self) {
-        self.0 -= 1;
-    }
 }
 
 impl std::ops::Add for AffinityScore {
@@ -354,6 +324,9 @@ pub(crate) struct ScheduleContext {
     /// Sparse map of nodes: omitting a node implicitly makes its affinity [`AffinityScore::FREE`]
     pub(crate) nodes: HashMap<NodeId, AffinityScore>,
 
+    /// Specifically how many _attached_ locations are on each node
+    pub(crate) attached_nodes: HashMap<NodeId, usize>,
+
     pub(crate) mode: ScheduleMode,
 }
 
@@ -361,6 +334,7 @@ impl ScheduleContext {
     pub(crate) fn new(mode: ScheduleMode) -> Self {
         Self {
             nodes: HashMap::new(),
+            attached_nodes: HashMap::new(),
             mode,
         }
     }
@@ -374,31 +348,25 @@ impl ScheduleContext {
         }
     }
 
-    /// Remove `shard`'s contributions to this context.  This is useful when considering scheduling
-    /// this shard afresh, where we don't want it to e.g. experience anti-affinity to its current location.
-    pub(crate) fn project_detach(&self, shard: &TenantShard) -> Self {
-        let mut new_context = self.clone();
-
-        if let Some(attached) = shard.intent.get_attached() {
-            if let Some(score) = new_context.nodes.get_mut(attached) {
-                score.dec();
-            }
-        }
-
-        for secondary in shard.intent.get_secondary() {
-            if let Some(score) = new_context.nodes.get_mut(secondary) {
-                score.dec();
-            }
-        }
-
-        new_context
+    pub(crate) fn push_attached(&mut self, node_id: NodeId) {
+        let entry = self.attached_nodes.entry(node_id).or_default();
+        *entry += 1;
+    }
+
+    pub(crate) fn get_node_affinity(&self, node_id: NodeId) -> AffinityScore {
+        self.nodes
+            .get(&node_id)
+            .copied()
+            .unwrap_or(AffinityScore::FREE)
+    }
+
+    pub(crate) fn get_node_attachments(&self, node_id: NodeId) -> usize {
+        self.attached_nodes.get(&node_id).copied().unwrap_or(0)
     }
 
-    /// For test, track the sum of AffinityScore values, which is effectively how many
-    /// attached or secondary locations have been registered with this context.
     #[cfg(test)]
-    pub(crate) fn location_count(&self) -> usize {
-        self.nodes.values().map(|i| i.0).sum()
+    pub(crate) fn attach_count(&self) -> usize {
+        self.attached_nodes.values().sum()
     }
 }
 
@@ -420,7 +388,6 @@ impl Scheduler {
                 SchedulerNode {
                     shard_count: 0,
                     attached_shard_count: 0,
-                    home_shard_count: 0,
                     may_schedule: node.may_schedule(),
                     az: node.get_availability_zone_id().clone(),
                 },
@@ -448,7 +415,6 @@ impl Scheduler {
                 SchedulerNode {
                     shard_count: 0,
                     attached_shard_count: 0,
-                    home_shard_count: 0,
                     may_schedule: node.may_schedule(),
                     az: node.get_availability_zone_id().clone(),
                 },
@@ -461,9 +427,6 @@ impl Scheduler {
                     Some(node) => {
                         node.shard_count += 1;
                         node.attached_shard_count += 1;
-                        if Some(&node.az) == shard.preferred_az() {
-                            node.home_shard_count += 1;
-                        }
                     }
                     None => anyhow::bail!(
                         "Tenant {} references nonexistent node {}",
@@ -475,12 +438,7 @@ impl Scheduler {
 
             for node_id in shard.intent.get_secondary() {
                 match expect_nodes.get_mut(node_id) {
-                    Some(node) => {
-                        node.shard_count += 1;
-                        if Some(&node.az) == shard.preferred_az() {
-                            node.home_shard_count += 1;
-                        }
-                    }
+                    Some(node) => node.shard_count += 1,
                     None => anyhow::bail!(
                         "Tenant {} references nonexistent node {}",
                         shard.tenant_shard_id,
@@ -524,20 +482,13 @@ impl Scheduler {
     ///
     /// It is an error to call this for a node that is not known to the scheduler (i.e. passed into
     /// [`Self::new`] or [`Self::node_upsert`])
-    pub(crate) fn update_node_ref_counts(
-        &mut self,
-        node_id: NodeId,
-        preferred_az: Option<&AvailabilityZone>,
-        update: RefCountUpdate,
-    ) {
+    pub(crate) fn update_node_ref_counts(&mut self, node_id: NodeId, update: RefCountUpdate) {
         let Some(node) = self.nodes.get_mut(&node_id) else {
             debug_assert!(false);
             tracing::error!("Scheduler missing node {node_id}");
             return;
         };
 
-        let is_home_az = Some(&node.az) == preferred_az;
-
         match update {
             RefCountUpdate::PromoteSecondary => {
                 node.attached_shard_count += 1;
@@ -545,31 +496,19 @@ impl Scheduler {
             RefCountUpdate::Attach => {
                 node.shard_count += 1;
                 node.attached_shard_count += 1;
-                if is_home_az {
-                    node.home_shard_count += 1;
-                }
             }
             RefCountUpdate::Detach => {
                 node.shard_count -= 1;
                 node.attached_shard_count -= 1;
-                if is_home_az {
-                    node.home_shard_count -= 1;
-                }
             }
             RefCountUpdate::DemoteAttached => {
                 node.attached_shard_count -= 1;
             }
             RefCountUpdate::AddSecondary => {
                 node.shard_count += 1;
-                if is_home_az {
-                    node.home_shard_count += 1;
-                }
             }
             RefCountUpdate::RemoveSecondary => {
                 node.shard_count -= 1;
-                if is_home_az {
-                    node.home_shard_count -= 1;
-                }
             }
         }
 
@@ -655,7 +594,6 @@ impl Scheduler {
                 entry.insert(SchedulerNode {
                     shard_count: 0,
                     attached_shard_count: 0,
-                    home_shard_count: 0,
                     may_schedule: node.may_schedule(),
                     az: node.get_availability_zone_id().clone(),
                 });
@@ -669,20 +607,33 @@ impl Scheduler {
         }
     }
 
-    /// Calculate a single node's score, used in optimizer logic to compare specific
-    /// nodes' scores.
-    pub(crate) fn compute_node_score<Score>(
-        &mut self,
-        node_id: NodeId,
-        preferred_az: &Option<AvailabilityZone>,
-        context: &ScheduleContext,
-    ) -> Option<Score>
-    where
-        Score: NodeSchedulingScore,
-    {
-        self.nodes
-            .get_mut(&node_id)
-            .and_then(|node| Score::generate(&node_id, node, preferred_az, context))
+    /// Where we have several nodes to choose from, for example when picking a secondary location
+    /// to promote to an attached location, this method may be used to pick the best choice based
+    /// on the scheduler's knowledge of utilization and availability.
+    ///
+    /// If the input is empty, or all the nodes are not elegible for scheduling, return None: the
+    /// caller can pick a node some other way.
+    pub(crate) fn node_preferred(&self, nodes: &[NodeId]) -> Option<NodeId> {
+        if nodes.is_empty() {
+            return None;
+        }
+
+        // TODO: When the utilization score returned by the pageserver becomes meaningful,
+        // schedule based on that instead of the shard count.
+        let node = nodes
+            .iter()
+            .map(|node_id| {
+                let may_schedule = self
+                    .nodes
+                    .get(node_id)
+                    .map(|n| !matches!(n.may_schedule, MaySchedule::No))
+                    .unwrap_or(false);
+                (*node_id, may_schedule)
+            })
+            .max_by_key(|(_n, may_schedule)| *may_schedule);
+
+        // If even the preferred node has may_schedule==false, return None
+        node.and_then(|(node_id, may_schedule)| if may_schedule { Some(node_id) } else { None })
     }
 
     /// Compute a schedulling score for each node that the scheduler knows of
@@ -776,7 +727,7 @@ impl Scheduler {
             tracing::info!(
             "scheduler selected node {node_id} (elegible nodes {:?}, hard exclude: {hard_exclude:?}, soft exclude: {context:?})",
             scores.iter().map(|i| i.node_id().0).collect::<Vec<_>>()
-       );
+        );
         }
 
         // Note that we do not update shard count here to reflect the scheduling: that
@@ -792,74 +743,47 @@ impl Scheduler {
     }
 
     /// For choosing which AZ to schedule a new shard into, use this.  It will return the
-    /// AZ with the the lowest number of shards currently scheduled in this AZ as their home
-    /// location.
+    /// AZ with the lowest median utilization.
     ///
     /// We use an AZ-wide measure rather than simply selecting the AZ of the least-loaded
     /// node, because while tenants start out single sharded, when they grow and undergo
-    /// shard-split, they will occupy space on many nodes within an AZ.  It is important
-    /// that we pick the AZ in a way that balances this _future_ load.
+    /// shard-split, they will occupy space on many nodes within an AZ.
     ///
-    /// Once we've picked an AZ, subsequent scheduling within that AZ will be driven by
-    /// nodes' utilization scores.
+    /// We use median rather than total free space or mean utilization, because
+    /// we wish to avoid preferring AZs that have low-load nodes resulting from
+    /// recent replacements.
+    ///
+    /// The practical result is that we will pick an AZ based on its median node, and
+    /// then actually _schedule_ the new shard onto the lowest-loaded node in that AZ.
     pub(crate) fn get_az_for_new_tenant(&self) -> Option<AvailabilityZone> {
         if self.nodes.is_empty() {
             return None;
         }
 
-        #[derive(Default)]
-        struct AzScore {
-            home_shard_count: usize,
-            scheduleable: bool,
+        let mut scores_by_az = HashMap::new();
+        for (node_id, node) in &self.nodes {
+            let az_scores = scores_by_az.entry(&node.az).or_insert_with(Vec::new);
+            let score = match &node.may_schedule {
+                MaySchedule::Yes(utilization) => utilization.score(),
+                MaySchedule::No => PageserverUtilization::full().score(),
+            };
+            az_scores.push((node_id, node, score));
         }
 
-        let mut azs: HashMap<&AvailabilityZone, AzScore> = HashMap::new();
-        for node in self.nodes.values() {
-            let az = azs.entry(&node.az).or_default();
-            az.home_shard_count += node.home_shard_count;
-            az.scheduleable |= matches!(node.may_schedule, MaySchedule::Yes(_));
+        // Sort by utilization.  Also include the node ID to break ties.
+        for scores in scores_by_az.values_mut() {
+            scores.sort_by_key(|i| (i.2, i.0));
         }
 
-        // If any AZs are schedulable, then filter out the non-schedulable ones (i.e. AZs where
-        // all nodes are overloaded or otherwise unschedulable).
-        if azs.values().any(|i| i.scheduleable) {
-            azs.retain(|_, i| i.scheduleable);
-        }
-
-        // Find the AZ with the lowest number of shards currently allocated
-        Some(
-            azs.into_iter()
-                .min_by_key(|i| (i.1.home_shard_count, i.0))
-                .unwrap()
-                .0
-                .clone(),
-        )
-    }
-
-    pub(crate) fn get_node_az(&self, node_id: &NodeId) -> Option<AvailabilityZone> {
-        self.nodes.get(node_id).map(|n| n.az.clone())
-    }
-
-    /// For use when choosing a preferred secondary location: filter out nodes that are not
-    /// available, and gather their AZs.
-    pub(crate) fn filter_usable_nodes(
-        &self,
-        nodes: &[NodeId],
-    ) -> Vec<(NodeId, Option<AvailabilityZone>)> {
-        nodes
+        let mut median_by_az = scores_by_az
             .iter()
-            .filter_map(|node_id| {
-                let node = self
-                    .nodes
-                    .get(node_id)
-                    .expect("Referenced nodes always exist");
-                if matches!(node.may_schedule, MaySchedule::Yes(_)) {
-                    Some((*node_id, Some(node.az.clone())))
-                } else {
-                    None
-                }
-            })
-            .collect()
+            .map(|(az, nodes)| (*az, nodes.get(nodes.len() / 2).unwrap().2))
+            .collect::<Vec<_>>();
+        // Sort by utilization.  Also include the AZ to break ties.
+        median_by_az.sort_by_key(|i| (i.1, i.0));
+
+        // Return the AZ with the lowest median utilization
+        Some(median_by_az.first().unwrap().0.clone())
     }
 
     /// Unit test access to internal state
@@ -872,33 +796,6 @@ impl Scheduler {
     pub(crate) fn get_node_attached_shard_count(&self, node_id: NodeId) -> usize {
         self.nodes.get(&node_id).unwrap().attached_shard_count
     }
-
-    /// Some metrics that we only calculate periodically: this is simpler than
-    /// rigorously updating them on every change.
-    pub(crate) fn update_metrics(&self) {
-        for (node_id, node) in &self.nodes {
-            let node_id_str = format!("{}", node_id);
-            let label_group = NodeLabelGroup {
-                az: &node.az.0,
-                node_id: &node_id_str,
-            };
-
-            crate::metrics::METRICS_REGISTRY
-                .metrics_group
-                .storage_controller_node_shards
-                .set(label_group.clone(), node.shard_count as i64);
-
-            crate::metrics::METRICS_REGISTRY
-                .metrics_group
-                .storage_controller_node_attached_shards
-                .set(label_group.clone(), node.attached_shard_count as i64);
-
-            crate::metrics::METRICS_REGISTRY
-                .metrics_group
-                .storage_controller_node_home_shards
-                .set(label_group.clone(), node.home_shard_count as i64);
-        }
-    }
 }
 
 #[cfg(test)]
@@ -946,14 +843,7 @@ pub(crate) mod test_utils {
 
 #[cfg(test)]
 mod tests {
-    use pageserver_api::{
-        controller_api::NodeAvailability, models::utilization::test_utilization,
-        shard::ShardIdentity,
-    };
-    use utils::{
-        id::TenantId,
-        shard::{ShardCount, ShardNumber, TenantShardId},
-    };
+    use pageserver_api::{controller_api::NodeAvailability, models::utilization::test_utilization};
 
     use super::*;
 
@@ -963,8 +853,8 @@ mod tests {
         let nodes = test_utils::make_test_nodes(2, &[]);
 
         let mut scheduler = Scheduler::new(nodes.values());
-        let mut t1_intent = IntentState::new(None);
-        let mut t2_intent = IntentState::new(None);
+        let mut t1_intent = IntentState::new();
+        let mut t2_intent = IntentState::new();
 
         let context = ScheduleContext::default();
 
@@ -1040,7 +930,7 @@ mod tests {
             let scheduled = scheduler
                 .schedule_shard::<AttachedShardTag>(&[], &None, context)
                 .unwrap();
-            let mut intent = IntentState::new(None);
+            let mut intent = IntentState::new();
             intent.set_attached(scheduler, Some(scheduled));
             scheduled_intents.push(intent);
             assert_eq!(scheduled, expect_node);
@@ -1173,7 +1063,7 @@ mod tests {
             let scheduled = scheduler
                 .schedule_shard::<Tag>(&[], &preferred_az, context)
                 .unwrap();
-            let mut intent = IntentState::new(preferred_az.clone());
+            let mut intent = IntentState::new();
             intent.set_attached(scheduler, Some(scheduled));
             scheduled_intents.push(intent);
             assert_eq!(scheduled, expect_node);
@@ -1199,9 +1089,9 @@ mod tests {
             &mut context,
         );
 
-        // Node 1 and 3 (az-a) have same affinity score, so prefer the lowest node id.
+        // Node 2 is not in "az-a", but it has the lowest affinity so we prefer that.
         assert_scheduler_chooses::<AttachedShardTag>(
-            NodeId(1),
+            NodeId(2),
             Some(az_a_tag.clone()),
             &mut scheduled_intents,
             &mut scheduler,
@@ -1217,6 +1107,26 @@ mod tests {
             &mut context,
         );
 
+        // Avoid nodes in "az-b" for the secondary location.
+        // Nodes 1 and 3 are identically loaded, so prefer the lowest node id.
+        assert_scheduler_chooses::<SecondaryShardTag>(
+            NodeId(1),
+            Some(az_b_tag.clone()),
+            &mut scheduled_intents,
+            &mut scheduler,
+            &mut context,
+        );
+
+        // Avoid nodes in "az-b" for the secondary location.
+        // Node 3 has lower affinity score than 1, so prefer that.
+        assert_scheduler_chooses::<SecondaryShardTag>(
+            NodeId(3),
+            Some(az_b_tag.clone()),
+            &mut scheduled_intents,
+            &mut scheduler,
+            &mut context,
+        );
+
         for mut intent in scheduled_intents {
             intent.clear(&mut scheduler);
         }
@@ -1240,292 +1150,34 @@ mod tests {
 
         let mut scheduler = Scheduler::new(nodes.values());
 
-        /// Force the `home_shard_count` of a node directly: this is the metric used
-        /// by the scheduler when picking AZs.
-        fn set_shard_count(scheduler: &mut Scheduler, node_id: NodeId, shard_count: usize) {
-            let node = scheduler.nodes.get_mut(&node_id).unwrap();
-            node.home_shard_count = shard_count;
+        /// Force the utilization of a node in Scheduler's state to a particular
+        /// number of bytes used.
+        fn set_utilization(scheduler: &mut Scheduler, node_id: NodeId, shard_count: u32) {
+            let mut node = Node::new(
+                node_id,
+                "".to_string(),
+                0,
+                "".to_string(),
+                0,
+                scheduler.nodes.get(&node_id).unwrap().az.clone(),
+            );
+            node.set_availability(NodeAvailability::Active(test_utilization::simple(
+                shard_count,
+                0,
+            )));
+            scheduler.node_upsert(&node);
         }
 
         // Initial empty state.  Scores are tied, scheduler prefers lower AZ ID.
         assert_eq!(scheduler.get_az_for_new_tenant(), Some(az_a_tag.clone()));
 
-        // Home shard count is higher in AZ A, so AZ B will be preferred
-        set_shard_count(&mut scheduler, NodeId(1), 10);
-        assert_eq!(scheduler.get_az_for_new_tenant(), Some(az_b_tag.clone()));
-
-        // Total home shard count is higher in AZ B, so we revert to preferring AZ A
-        set_shard_count(&mut scheduler, NodeId(4), 6);
-        set_shard_count(&mut scheduler, NodeId(5), 6);
+        // Put some utilization on one node in AZ A: this should change nothing, as the median hasn't changed
+        set_utilization(&mut scheduler, NodeId(1), 1000000);
         assert_eq!(scheduler.get_az_for_new_tenant(), Some(az_a_tag.clone()));
-    }
 
-    /// Test that when selecting AZs for many new tenants, we get the expected balance across nodes
-    #[test]
-    fn az_selection_many() {
-        let az_a_tag = AvailabilityZone("az-a".to_string());
-        let az_b_tag = AvailabilityZone("az-b".to_string());
-        let az_c_tag = AvailabilityZone("az-c".to_string());
-        let nodes = test_utils::make_test_nodes(
-            6,
-            &[
-                az_a_tag.clone(),
-                az_b_tag.clone(),
-                az_c_tag.clone(),
-                az_a_tag.clone(),
-                az_b_tag.clone(),
-                az_c_tag.clone(),
-            ],
-        );
-
-        let mut scheduler = Scheduler::new(nodes.values());
-
-        // We should get 1/6th of these on each node, give or take a few...
-        let total_tenants = 300;
-
-        // ...where the 'few' is the number of AZs, because the scheduling will sometimes overshoot
-        // on one AZ before correcting itself.  This is because we select the 'home' AZ based on
-        // an AZ-wide metric, but we select the location for secondaries on a purely node-based
-        // metric (while excluding the home AZ).
-        let grace = 3;
-
-        let mut scheduled_shards = Vec::new();
-        for _i in 0..total_tenants {
-            let preferred_az = scheduler.get_az_for_new_tenant().unwrap();
-
-            let mut node_home_counts = scheduler
-                .nodes
-                .iter()
-                .map(|(node_id, node)| (node_id, node.home_shard_count))
-                .collect::<Vec<_>>();
-            node_home_counts.sort_by_key(|i| i.0);
-            eprintln!("Selected {}, vs nodes {:?}", preferred_az, node_home_counts);
-
-            let tenant_shard_id = TenantShardId {
-                tenant_id: TenantId::generate(),
-                shard_number: ShardNumber(0),
-                shard_count: ShardCount(1),
-            };
-
-            let shard_identity = ShardIdentity::new(
-                tenant_shard_id.shard_number,
-                tenant_shard_id.shard_count,
-                pageserver_api::shard::ShardStripeSize(1),
-            )
-            .unwrap();
-            let mut shard = TenantShard::new(
-                tenant_shard_id,
-                shard_identity,
-                pageserver_api::controller_api::PlacementPolicy::Attached(1),
-                Some(preferred_az),
-            );
-
-            let mut context = ScheduleContext::default();
-            shard.schedule(&mut scheduler, &mut context).unwrap();
-            eprintln!("Scheduled shard at {:?}", shard.intent);
-
-            scheduled_shards.push(shard);
-        }
-
-        for (node_id, node) in &scheduler.nodes {
-            eprintln!(
-                "Node {}: {} {} {}",
-                node_id, node.shard_count, node.attached_shard_count, node.home_shard_count
-            );
-        }
-
-        for node in scheduler.nodes.values() {
-            assert!((node.home_shard_count as i64 - total_tenants as i64 / 6).abs() < grace);
-        }
-
-        for mut shard in scheduled_shards {
-            shard.intent.clear(&mut scheduler);
-        }
-    }
-
-    #[test]
-    /// Make sure that when we have an odd number of nodes and an even number of shards, we still
-    /// get scheduling stability.
-    fn odd_nodes_stability() {
-        let az_a = AvailabilityZone("az-a".to_string());
-        let az_b = AvailabilityZone("az-b".to_string());
-
-        let nodes = test_utils::make_test_nodes(
-            10,
-            &[
-                az_a.clone(),
-                az_a.clone(),
-                az_a.clone(),
-                az_a.clone(),
-                az_a.clone(),
-                az_b.clone(),
-                az_b.clone(),
-                az_b.clone(),
-                az_b.clone(),
-                az_b.clone(),
-            ],
-        );
-        let mut scheduler = Scheduler::new(nodes.values());
-
-        // Need to keep these alive because they contribute to shard counts via RAII
-        let mut scheduled_shards = Vec::new();
-
-        let mut context = ScheduleContext::default();
-
-        fn schedule_shard(
-            tenant_shard_id: TenantShardId,
-            expect_attached: NodeId,
-            expect_secondary: NodeId,
-            scheduled_shards: &mut Vec<TenantShard>,
-            scheduler: &mut Scheduler,
-            preferred_az: Option<AvailabilityZone>,
-            context: &mut ScheduleContext,
-        ) {
-            let shard_identity = ShardIdentity::new(
-                tenant_shard_id.shard_number,
-                tenant_shard_id.shard_count,
-                pageserver_api::shard::ShardStripeSize(1),
-            )
-            .unwrap();
-            let mut shard = TenantShard::new(
-                tenant_shard_id,
-                shard_identity,
-                pageserver_api::controller_api::PlacementPolicy::Attached(1),
-                preferred_az,
-            );
-
-            shard.schedule(scheduler, context).unwrap();
-
-            assert_eq!(shard.intent.get_attached().unwrap(), expect_attached);
-            assert_eq!(
-                shard.intent.get_secondary().first().unwrap(),
-                &expect_secondary
-            );
-
-            scheduled_shards.push(shard);
-        }
-
-        let tenant_id = TenantId::generate();
-
-        schedule_shard(
-            TenantShardId {
-                tenant_id,
-                shard_number: ShardNumber(0),
-                shard_count: ShardCount(8),
-            },
-            NodeId(1),
-            NodeId(6),
-            &mut scheduled_shards,
-            &mut scheduler,
-            Some(az_a.clone()),
-            &mut context,
-        );
-
-        schedule_shard(
-            TenantShardId {
-                tenant_id,
-                shard_number: ShardNumber(1),
-                shard_count: ShardCount(8),
-            },
-            NodeId(2),
-            NodeId(7),
-            &mut scheduled_shards,
-            &mut scheduler,
-            Some(az_a.clone()),
-            &mut context,
-        );
-
-        schedule_shard(
-            TenantShardId {
-                tenant_id,
-                shard_number: ShardNumber(2),
-                shard_count: ShardCount(8),
-            },
-            NodeId(3),
-            NodeId(8),
-            &mut scheduled_shards,
-            &mut scheduler,
-            Some(az_a.clone()),
-            &mut context,
-        );
-
-        schedule_shard(
-            TenantShardId {
-                tenant_id,
-                shard_number: ShardNumber(3),
-                shard_count: ShardCount(8),
-            },
-            NodeId(4),
-            NodeId(9),
-            &mut scheduled_shards,
-            &mut scheduler,
-            Some(az_a.clone()),
-            &mut context,
-        );
-
-        schedule_shard(
-            TenantShardId {
-                tenant_id,
-                shard_number: ShardNumber(4),
-                shard_count: ShardCount(8),
-            },
-            NodeId(5),
-            NodeId(10),
-            &mut scheduled_shards,
-            &mut scheduler,
-            Some(az_a.clone()),
-            &mut context,
-        );
-
-        schedule_shard(
-            TenantShardId {
-                tenant_id,
-                shard_number: ShardNumber(5),
-                shard_count: ShardCount(8),
-            },
-            NodeId(1),
-            NodeId(6),
-            &mut scheduled_shards,
-            &mut scheduler,
-            Some(az_a.clone()),
-            &mut context,
-        );
-
-        schedule_shard(
-            TenantShardId {
-                tenant_id,
-                shard_number: ShardNumber(6),
-                shard_count: ShardCount(8),
-            },
-            NodeId(2),
-            NodeId(7),
-            &mut scheduled_shards,
-            &mut scheduler,
-            Some(az_a.clone()),
-            &mut context,
-        );
-
-        schedule_shard(
-            TenantShardId {
-                tenant_id,
-                shard_number: ShardNumber(7),
-                shard_count: ShardCount(8),
-            },
-            NodeId(3),
-            NodeId(8),
-            &mut scheduled_shards,
-            &mut scheduler,
-            Some(az_a.clone()),
-            &mut context,
-        );
-
-        // Assert that the optimizer suggests nochanges, i.e. our initial scheduling was stable.
-        for shard in &scheduled_shards {
-            assert_eq!(shard.optimize_attachment(&mut scheduler, &context), None);
-        }
-
-        for mut shard in scheduled_shards {
-            shard.intent.clear(&mut scheduler);
-        }
+        // Put some utilization on a second node in AZ A: now the median has changed, so the scheduler
+        // should prefer the other AZ.
+        set_utilization(&mut scheduler, NodeId(2), 1000000);
+        assert_eq!(scheduler.get_az_for_new_tenant(), Some(az_b_tag.clone()));
     }
 }

storage_controller/src/service.rs

@@ -47,7 +47,7 @@ use pageserver_api::{
         AvailabilityZone, MetadataHealthRecord, MetadataHealthUpdateRequest, NodeAvailability,
         NodeRegisterRequest, NodeSchedulingPolicy, NodeShard, NodeShardResponse, PlacementPolicy,
         SafekeeperDescribeResponse, ShardSchedulingPolicy, ShardsPreferredAzsRequest,
-        ShardsPreferredAzsResponse, SkSchedulingPolicy, TenantCreateRequest, TenantCreateResponse,
+        ShardsPreferredAzsResponse, TenantCreateRequest, TenantCreateResponse,
         TenantCreateResponseShard, TenantDescribeResponse, TenantDescribeResponseShard,
         TenantLocateResponse, TenantPolicyRequest, TenantShardMigrateRequest,
         TenantShardMigrateResponse,
@@ -1404,11 +1404,7 @@ impl Service {
 
             // We will populate intent properly later in [`Self::startup_reconcile`], initially populate
             // it with what we can infer: the node for which a generation was most recently issued.
-            let mut intent = IntentState::new(
-                tsp.preferred_az_id
-                    .as_ref()
-                    .map(|az| AvailabilityZone(az.clone())),
-            );
+            let mut intent = IntentState::new();
             if let Some(generation_pageserver) = tsp.generation_pageserver.map(|n| NodeId(n as u64))
             {
                 if nodes.contains_key(&generation_pageserver) {
@@ -2478,30 +2474,19 @@ impl Service {
         tenant_id: TenantId,
         _guard: &TracingExclusiveGuard<TenantOperations>,
     ) -> Result<(), ApiError> {
-        // Check if the tenant is present in memory, and select an AZ to use when loading
-        // if we will load it.
-        let load_in_az = {
+        let present_in_memory = {
             let locked = self.inner.read().unwrap();
-            let existing = locked
+            locked
                 .tenants
                 .range(TenantShardId::tenant_range(tenant_id))
-                .next();
-
-            // If the tenant is not present in memory, we expect to load it from database,
-            // so let's figure out what AZ to load it into while we have self.inner locked.
-            if existing.is_none() {
-                locked
-                    .scheduler
-                    .get_az_for_new_tenant()
-                    .ok_or(ApiError::BadRequest(anyhow::anyhow!(
-                        "No AZ with nodes found to load tenant"
-                    )))?
-            } else {
-                // We already have this tenant in memory
-                return Ok(());
-            }
+                .next()
+                .is_some()
         };
 
+        if present_in_memory {
+            return Ok(());
+        }
+
         let tenant_shards = self.persistence.load_tenant(tenant_id).await?;
         if tenant_shards.is_empty() {
             return Err(ApiError::NotFound(
@@ -2509,20 +2494,8 @@ impl Service {
             ));
         }
 
-        // Update the persistent shards with the AZ that we are about to apply to in-memory state
-        self.persistence
-            .set_tenant_shard_preferred_azs(
-                tenant_shards
-                    .iter()
-                    .map(|t| {
-                        (
-                            t.get_tenant_shard_id().expect("Corrupt shard in database"),
-                            Some(load_in_az.clone()),
-                        )
-                    })
-                    .collect(),
-            )
-            .await?;
+        // TODO: choose a fresh AZ to use for this tenant when un-detaching: there definitely isn't a running
+        // compute, so no benefit to making AZ sticky across detaches.
 
         let mut locked = self.inner.write().unwrap();
         tracing::info!(
@@ -2532,7 +2505,7 @@ impl Service {
         );
 
         locked.tenants.extend(tenant_shards.into_iter().map(|p| {
-            let intent = IntentState::new(Some(load_in_az.clone()));
+            let intent = IntentState::new();
             let shard =
                 TenantShard::from_persistent(p, intent).expect("Corrupt shard row in database");
 
@@ -4158,42 +4131,17 @@ impl Service {
         .ok_or_else(|| ApiError::NotFound(anyhow::anyhow!("Tenant {tenant_id} not found").into()))
     }
 
-    /// limit & offset are pagination parameters. Since we are walking an in-memory HashMap, `offset` does not
-    /// avoid traversing data, it just avoid returning it. This is suitable for our purposes, since our in memory
-    /// maps are small enough to traverse fast, our pagination is just to avoid serializing huge JSON responses
-    /// in our external API.
-    pub(crate) fn tenant_list(
-        &self,
-        limit: Option<usize>,
-        start_after: Option<TenantId>,
-    ) -> Vec<TenantDescribeResponse> {
+    pub(crate) fn tenant_list(&self) -> Vec<TenantDescribeResponse> {
         let locked = self.inner.read().unwrap();
 
-        // Apply start_from parameter
-        let shard_range = match start_after {
-            None => locked.tenants.range(..),
-            Some(tenant_id) => locked.tenants.range(
-                TenantShardId {
-                    tenant_id,
-                    shard_number: ShardNumber(u8::MAX),
-                    shard_count: ShardCount(u8::MAX),
-                }..,
-            ),
-        };
-
         let mut result = Vec::new();
-        for (_tenant_id, tenant_shards) in &shard_range.group_by(|(id, _shard)| id.tenant_id) {
+        for (_tenant_id, tenant_shards) in
+            &locked.tenants.iter().group_by(|(id, _shard)| id.tenant_id)
+        {
             result.push(
                 self.tenant_describe_impl(tenant_shards.map(|(_k, v)| v))
                     .expect("Groups are always non-empty"),
             );
-
-            // Enforce `limit` parameter
-            if let Some(limit) = limit {
-                if result.len() >= limit {
-                    break;
-                }
-            }
         }
 
         result
@@ -4288,22 +4236,6 @@ impl Service {
                 }
 
                 tracing::info!("Restoring parent shard {tenant_shard_id}");
-
-                // Drop any intents that refer to unavailable nodes, to enable this abort to proceed even
-                // if the original attachment location is offline.
-                if let Some(node_id) = shard.intent.get_attached() {
-                    if !nodes.get(node_id).unwrap().is_available() {
-                        tracing::info!("Demoting attached intent for {tenant_shard_id} on unavailable node {node_id}");
-                        shard.intent.demote_attached(scheduler, *node_id);
-                    }
-                }
-                for node_id in shard.intent.get_secondary().clone() {
-                    if !nodes.get(&node_id).unwrap().is_available() {
-                        tracing::info!("Dropping secondary intent for {tenant_shard_id} on unavailable node {node_id}");
-                        shard.intent.remove_secondary(scheduler, node_id);
-                    }
-                }
-
                 shard.splitting = SplitState::Idle;
                 if let Err(e) = shard.schedule(scheduler, &mut ScheduleContext::default()) {
                     // If this shard can't be scheduled now (perhaps due to offline nodes or
@@ -4457,13 +4389,15 @@ impl Service {
 
                     let mut child_state =
                         TenantShard::new(child, child_shard, policy.clone(), preferred_az.clone());
-                    child_state.intent =
-                        IntentState::single(scheduler, Some(pageserver), preferred_az.clone());
+                    child_state.intent = IntentState::single(scheduler, Some(pageserver));
                     child_state.observed = ObservedState {
                         locations: child_observed,
                     };
                     child_state.generation = Some(generation);
                     child_state.config = config.clone();
+                    if let Some(preferred_az) = &preferred_az {
+                        child_state.set_preferred_az(preferred_az.clone());
+                    }
 
                     // The child's TenantShard::splitting is intentionally left at the default value of Idle,
                     // as at this point in the split process we have succeeded and this part is infallible:
@@ -5080,8 +5014,6 @@ impl Service {
                         // If our new attached node was a secondary, it no longer should be.
                         shard.intent.remove_secondary(scheduler, migrate_req.node_id);
 
-                        shard.intent.set_attached(scheduler, Some(migrate_req.node_id));
-
                         // If we were already attached to something, demote that to a secondary
                         if let Some(old_attached) = old_attached {
                             if n > 0 {
@@ -5093,6 +5025,8 @@ impl Service {
                                 shard.intent.push_secondary(scheduler, old_attached);
                             }
                         }
+
+                        shard.intent.set_attached(scheduler, Some(migrate_req.node_id));
                     }
                     PlacementPolicy::Secondary => {
                         shard.intent.clear(scheduler);
@@ -5411,15 +5345,6 @@ impl Service {
 
         expect_shards.sort_by_key(|tsp| (tsp.tenant_id.clone(), tsp.shard_number, tsp.shard_count));
 
-        // Because JSON contents of persistent tenants might disagree with the fields in current `TenantConfig`
-        // definition, we will do an encode/decode cycle to ensure any legacy fields are dropped and any new
-        // fields are added, before doing a comparison.
-        for tsp in &mut persistent_shards {
-            let config: TenantConfig = serde_json::from_str(&tsp.config)
-                .map_err(|e| ApiError::InternalServerError(e.into()))?;
-            tsp.config = serde_json::to_string(&config).expect("Encoding config is infallible");
-        }
-
         if persistent_shards != expect_shards {
             tracing::error!("Consistency check failed on shards.");
 
@@ -5787,7 +5712,7 @@ impl Service {
             register_req.listen_http_port,
             register_req.listen_pg_addr,
             register_req.listen_pg_port,
-            register_req.availability_zone_id.clone(),
+            register_req.availability_zone_id,
         );
 
         // TODO: idempotency if the node already exists in the database
@@ -5807,9 +5732,8 @@ impl Service {
             .set(locked.nodes.len() as i64);
 
         tracing::info!(
-            "Registered pageserver {} ({}), now have {} pageservers",
+            "Registered pageserver {}, now have {} pageservers",
             register_req.node_id,
-            register_req.availability_zone_id,
             locked.nodes.len()
         );
         Ok(())
@@ -6424,7 +6348,7 @@ impl Service {
     /// available.  A return value of 0 indicates that everything is fully reconciled already.
     fn reconcile_all(&self) -> usize {
         let mut locked = self.inner.write().unwrap();
-        let (nodes, tenants, scheduler) = locked.parts_mut();
+        let (nodes, tenants, _scheduler) = locked.parts_mut();
         let pageservers = nodes.clone();
 
         // This function is an efficient place to update lazy statistics, since we are walking
@@ -6485,9 +6409,6 @@ impl Service {
             }
         }
 
-        // Some metrics are calculated from SchedulerNode state, update these periodically
-        scheduler.update_metrics();
-
         // Process any deferred tenant drops
         for (tenant_id, guard) in drop_detached_tenants {
             self.maybe_drop_tenant(tenant_id, &mut locked, &guard);
@@ -6546,7 +6467,6 @@ impl Service {
                 // Shard was dropped between planning and execution;
                 continue;
             };
-            tracing::info!(tenant_shard_id=%tenant_shard_id, "Applying optimization: {optimization:?}");
             if shard.apply_optimization(scheduler, optimization) {
                 optimizations_applied += 1;
                 if self.maybe_reconcile_shard(shard, nodes).is_some() {
@@ -6577,13 +6497,7 @@ impl Service {
 
         let mut work = Vec::new();
         let mut locked = self.inner.write().unwrap();
-        let (_nodes, tenants, scheduler) = locked.parts_mut();
-
-        // We are going to plan a bunch of optimisations before applying any of them, so the
-        // utilisation stats on nodes will be effectively stale for the >1st optimisation we
-        // generate.  To avoid this causing unstable migrations/flapping, it's important that the
-        // code in TenantShard for finding optimisations uses [`NodeAttachmentSchedulingScore::disregard_utilization`]
-        // to ignore the utilisation component of the score.
+        let (nodes, tenants, scheduler) = locked.parts_mut();
 
         for (_tenant_id, schedule_context, shards) in
             TenantShardContextIterator::new(tenants, ScheduleMode::Speculative)
@@ -6614,28 +6528,13 @@ impl Service {
                     continue;
                 }
 
-                // Fast path: we may quickly identify shards that don't have any possible optimisations
-                if !shard.maybe_optimizable(scheduler, &schedule_context) {
-                    if cfg!(feature = "testing") {
-                        // Check that maybe_optimizable doesn't disagree with the actual optimization functions.
-                        // Only do this in testing builds because it is not a correctness-critical check, so we shouldn't
-                        // panic in prod if we hit this, or spend cycles on it in prod.
-                        assert!(shard
-                            .optimize_attachment(scheduler, &schedule_context)
-                            .is_none());
-                        assert!(shard
-                            .optimize_secondary(scheduler, &schedule_context)
-                            .is_none());
-                    }
-                    continue;
-                }
-
+                // TODO: optimization calculations are relatively expensive: create some fast-path for
+                // the common idle case (avoiding the search on tenants that we have recently checked)
                 if let Some(optimization) =
-                    // If idle, maybe optimize attachments: if a shard has a secondary location that is preferable to
+                    // If idle, maybe ptimize attachments: if a shard has a secondary location that is preferable to
                     // its primary location based on soft constraints, cut it over.
-                    shard.optimize_attachment(scheduler, &schedule_context)
+                    shard.optimize_attachment(nodes, &schedule_context)
                 {
-                    tracing::info!(tenant_shard_id=%shard.tenant_shard_id, "Identified optimization for attachment: {optimization:?}");
                     work.push((shard.tenant_shard_id, optimization));
                     break;
                 } else if let Some(optimization) =
@@ -6645,7 +6544,6 @@ impl Service {
                     // in the same tenant with secondary locations on the node where they originally split.
                     shard.optimize_secondary(scheduler, &schedule_context)
                 {
-                    tracing::info!(tenant_shard_id=%shard.tenant_shard_id, "Identified optimization for secondary: {optimization:?}");
                     work.push((shard.tenant_shard_id, optimization));
                     break;
                 }
@@ -6694,10 +6592,8 @@ impl Service {
                         }
                     }
                 }
-                ScheduleOptimizationAction::ReplaceSecondary(_)
-                | ScheduleOptimizationAction::CreateSecondary(_)
-                | ScheduleOptimizationAction::RemoveSecondary(_) => {
-                    // No extra checks needed to manage secondaries: this does not interrupt client access
+                ScheduleOptimizationAction::ReplaceSecondary(_) => {
+                    // No extra checks needed to replace a secondary: this does not interrupt client access
                     validated_work.push((tenant_shard_id, optimization))
                 }
             };
@@ -6769,35 +6665,26 @@ impl Service {
     /// we have this helper to move things along faster.
     #[cfg(feature = "testing")]
     async fn kick_secondary_download(&self, tenant_shard_id: TenantShardId) {
-        let (attached_node, secondaries) = {
+        let (attached_node, secondary_node) = {
             let locked = self.inner.read().unwrap();
             let Some(shard) = locked.tenants.get(&tenant_shard_id) else {
-                tracing::warn!(
-                    "Skipping kick of secondary download for {tenant_shard_id}: not found"
-                );
                 return;
             };
-
-            let Some(attached) = shard.intent.get_attached() else {
-                tracing::warn!(
-                    "Skipping kick of secondary download for {tenant_shard_id}: no attached"
-                );
+            let (Some(attached), Some(secondary)) = (
+                shard.intent.get_attached(),
+                shard.intent.get_secondary().first(),
+            ) else {
                 return;
             };
-
-            let secondaries = shard
-                .intent
-                .get_secondary()
-                .iter()
-                .map(|n| locked.nodes.get(n).unwrap().clone())
-                .collect::<Vec<_>>();
-
-            (locked.nodes.get(attached).unwrap().clone(), secondaries)
+            (
+                locked.nodes.get(attached).unwrap().clone(),
+                locked.nodes.get(secondary).unwrap().clone(),
+            )
         };
 
         // Make remote API calls to upload + download heatmaps: we ignore errors because this is just
         // a 'kick' to let scheduling optimisation run more promptly.
-        match attached_node
+        attached_node
             .with_client_retries(
                 |client| async move { client.tenant_heatmap_upload(tenant_shard_id).await },
                 &self.config.jwt_token,
@@ -6806,57 +6693,22 @@ impl Service {
                 SHORT_RECONCILE_TIMEOUT,
                 &self.cancel,
             )
-            .await
-        {
-            Some(Err(e)) => {
-                tracing::info!(
-                    "Failed to upload heatmap from {attached_node} for {tenant_shard_id}: {e}"
-                );
-            }
-            None => {
-                tracing::info!(
-                    "Cancelled while uploading heatmap from {attached_node} for {tenant_shard_id}"
-                );
-            }
-            Some(Ok(_)) => {
-                tracing::info!(
-                    "Successfully uploaded heatmap from {attached_node} for {tenant_shard_id}"
-                );
-            }
-        }
+            .await;
 
-        for secondary_node in secondaries {
-            match secondary_node
-                .with_client_retries(
-                    |client| async move {
-                        client
-                            .tenant_secondary_download(
-                                tenant_shard_id,
-                                Some(Duration::from_secs(1)),
-                            )
-                            .await
-                    },
-                    &self.config.jwt_token,
-                    3,
-                    10,
-                    SHORT_RECONCILE_TIMEOUT,
-                    &self.cancel,
-                )
-                .await
-            {
-                Some(Err(e)) => {
-                    tracing::info!(
-                "Failed to download heatmap from {secondary_node} for {tenant_shard_id}: {e}"
-            );
-                }
-                None => {
-                    tracing::info!("Cancelled while downloading heatmap from {secondary_node} for {tenant_shard_id}");
-                }
-                Some(Ok(progress)) => {
-                    tracing::info!("Successfully downloaded heatmap from {secondary_node} for {tenant_shard_id}: {progress:?}");
-                }
-            }
-        }
+        secondary_node
+            .with_client_retries(
+                |client| async move {
+                    client
+                        .tenant_secondary_download(tenant_shard_id, Some(Duration::from_secs(1)))
+                        .await
+                },
+                &self.config.jwt_token,
+                3,
+                10,
+                SHORT_RECONCILE_TIMEOUT,
+                &self.cancel,
+            )
+            .await;
     }
 
     /// Look for shards which are oversized and in need of splitting
@@ -7279,95 +7131,49 @@ impl Service {
         Ok(())
     }
 
-    /// Create a node fill plan (pick secondaries to promote), based on:
-    /// 1. Shards which have a secondary on this node, and this node is in their home AZ, and are currently attached to a node
-    ///    outside their home AZ, should be migrated back here.
-    /// 2. If after step 1 we have not migrated enough shards for this node to have its fair share of
-    ///    attached shards, we will promote more shards from the nodes with the most attached shards, unless
-    ///    those shards have a home AZ that doesn't match the node we're filling.
+    /// Create a node fill plan (pick secondaries to promote) that meets the following requirements:
+    /// 1. The node should be filled until it reaches the expected cluster average of
+    ///    attached shards. If there are not enough secondaries on the node, the plan stops early.
+    /// 2. Select tenant shards to promote such that the number of attached shards is balanced
+    ///    throughout the cluster. We achieve this by picking tenant shards from each node,
+    ///    starting from the ones with the largest number of attached shards, until the node
+    ///    reaches the expected cluster average.
+    /// 3. Avoid promoting more shards of the same tenant than required. The upper bound
+    ///    for the number of tenants from the same shard promoted to the node being filled is:
+    ///    shard count for the tenant divided by the number of nodes in the cluster.
     fn fill_node_plan(&self, node_id: NodeId) -> Vec<TenantShardId> {
         let mut locked = self.inner.write().unwrap();
-        let (nodes, tenants, _scheduler) = locked.parts_mut();
-
-        let node_az = nodes
-            .get(&node_id)
-            .expect("Node must exist")
-            .get_availability_zone_id()
-            .clone();
-
-        // The tenant shard IDs that we plan to promote from secondary to attached on this node
-        let mut plan = Vec::new();
-
-        // Collect shards which do not have a preferred AZ & are elegible for moving in stage 2
-        let mut free_tids_by_node: HashMap<NodeId, Vec<TenantShardId>> = HashMap::new();
-
-        // Don't respect AZ preferences if there is only one AZ.  This comes up in tests, but it could
-        // conceivably come up in real life if deploying a single-AZ region intentionally.
-        let respect_azs = nodes
-            .values()
-            .map(|n| n.get_availability_zone_id())
-            .unique()
-            .count()
-            > 1;
-
-        // Step 1: collect all shards that we are required to migrate back to this node because their AZ preference
-        // requires it.
-        for (tsid, tenant_shard) in tenants {
-            if !tenant_shard.intent.get_secondary().contains(&node_id) {
-                // Shard doesn't have a secondary on this node, ignore it.
-                continue;
-            }
-
-            // AZ check: when filling nodes after a restart, our intent is to move _back_ the
-            // shards which belong on this node, not to promote shards whose scheduling preference
-            // would be on their currently attached node.  So will avoid promoting shards whose
-            // home AZ doesn't match the AZ of the node we're filling.
-            match tenant_shard.preferred_az() {
-                _ if !respect_azs => {
-                    if let Some(primary) = tenant_shard.intent.get_attached() {
-                        free_tids_by_node.entry(*primary).or_default().push(*tsid);
-                    }
-                }
-                None => {
-                    // Shard doesn't have an AZ preference: it is elegible to be moved, but we
-                    // will only do so if our target shard count requires it.
-                    if let Some(primary) = tenant_shard.intent.get_attached() {
-                        free_tids_by_node.entry(*primary).or_default().push(*tsid);
-                    }
-                }
-                Some(az) if az == &node_az => {
-                    // This shard's home AZ is equal to the node we're filling: it should
-                    // be moved back to this node as part of filling, unless its currently
-                    // attached location is also in its home AZ.
-                    if let Some(primary) = tenant_shard.intent.get_attached() {
-                        if nodes
-                            .get(primary)
-                            .expect("referenced node must exist")
-                            .get_availability_zone_id()
-                            != tenant_shard
-                                .preferred_az()
-                                .expect("tenant must have an AZ preference")
-                        {
-                            plan.push(*tsid)
-                        }
-                    } else {
-                        plan.push(*tsid)
-                    }
-                }
-                Some(_) => {
-                    // This shard's home AZ is somewhere other than the node we're filling,
-                    // it may not be moved back to this node as part of filling.  Ignore it
-                }
-            }
-        }
-
-        // Step 2: also promote any AZ-agnostic shards as required to achieve the target number of attachments
         let fill_requirement = locked.scheduler.compute_fill_requirement(node_id);
 
+        let mut tids_by_node = locked
+            .tenants
+            .iter_mut()
+            .filter_map(|(tid, tenant_shard)| {
+                if !matches!(
+                    tenant_shard.get_scheduling_policy(),
+                    ShardSchedulingPolicy::Active
+                ) {
+                    // Only include tenants in fills if they have a normal (Active) scheduling policy.  We
+                    // even exclude Essential, because moving to fill a node is not essential to keeping this
+                    // tenant available.
+                    return None;
+                }
+
+                if tenant_shard.intent.get_secondary().contains(&node_id) {
+                    if let Some(primary) = tenant_shard.intent.get_attached() {
+                        return Some((*primary, *tid));
+                    }
+                }
+
+                None
+            })
+            .into_group_map();
+
         let expected_attached = locked.scheduler.expected_attached_shard_count();
         let nodes_by_load = locked.scheduler.nodes_by_attached_shard_count();
 
         let mut promoted_per_tenant: HashMap<TenantId, usize> = HashMap::new();
+        let mut plan = Vec::new();
 
         for (node_id, attached) in nodes_by_load {
             let available = locked.nodes.get(&node_id).is_some_and(|n| n.is_available());
@@ -7376,7 +7182,7 @@ impl Service {
             }
 
             if plan.len() >= fill_requirement
-                || free_tids_by_node.is_empty()
+                || tids_by_node.is_empty()
                 || attached <= expected_attached
             {
                 break;
@@ -7388,7 +7194,7 @@ impl Service {
 
             let mut remove_node = false;
             while take > 0 {
-                match free_tids_by_node.get_mut(&node_id) {
+                match tids_by_node.get_mut(&node_id) {
                     Some(tids) => match tids.pop() {
                         Some(tid) => {
                             let max_promote_for_tenant = std::cmp::max(
@@ -7414,7 +7220,7 @@ impl Service {
             }
 
             if remove_node {
-                free_tids_by_node.remove(&node_id);
+                tids_by_node.remove(&node_id);
             }
         }
 
@@ -7681,16 +7487,6 @@ impl Service {
         self.persistence.safekeeper_upsert(record).await
     }
 
-    pub(crate) async fn set_safekeeper_scheduling_policy(
-        &self,
-        id: i64,
-        scheduling_policy: SkSchedulingPolicy,
-    ) -> Result<(), DatabaseError> {
-        self.persistence
-            .set_safekeeper_scheduling_policy(id, scheduling_policy)
-            .await
-    }
-
     pub(crate) async fn update_shards_preferred_azs(
         &self,
         req: ShardsPreferredAzsRequest,

storage_controller/src/service/context_iterator.rs

@@ -43,6 +43,9 @@ impl<'a> Iterator for TenantShardContextIterator<'a> {
 
             // Accumulate the schedule context for all the shards in a tenant
             schedule_context.avoid(&shard.intent.all_pageservers());
+            if let Some(attached) = shard.intent.get_attached() {
+                schedule_context.push_attached(*attached);
+            }
             tenant_shards.push(shard);
 
             if tenant_shard_id.shard_number.0 == tenant_shard_id.shard_count.count() - 1 {
@@ -112,7 +115,7 @@ mod tests {
         assert_eq!(tenant_id, t1_id);
         assert_eq!(shards[0].tenant_shard_id.shard_number, ShardNumber(0));
         assert_eq!(shards.len(), 1);
-        assert_eq!(context.location_count(), 2);
+        assert_eq!(context.attach_count(), 1);
 
         let (tenant_id, context, shards) = iter.next().unwrap();
         assert_eq!(tenant_id, t2_id);
@@ -121,13 +124,13 @@ mod tests {
         assert_eq!(shards[2].tenant_shard_id.shard_number, ShardNumber(2));
         assert_eq!(shards[3].tenant_shard_id.shard_number, ShardNumber(3));
         assert_eq!(shards.len(), 4);
-        assert_eq!(context.location_count(), 8);
+        assert_eq!(context.attach_count(), 4);
 
         let (tenant_id, context, shards) = iter.next().unwrap();
         assert_eq!(tenant_id, t3_id);
         assert_eq!(shards[0].tenant_shard_id.shard_number, ShardNumber(0));
         assert_eq!(shards.len(), 1);
-        assert_eq!(context.location_count(), 2);
+        assert_eq!(context.attach_count(), 1);
 
         for shard in tenants.values_mut() {
             shard.intent.clear(&mut scheduler);

test_runner/fixtures/neon_fixtures.py

@@ -370,7 +370,6 @@ class NeonEnvBuilder:
         pageserver_config_override: str | Callable[[dict[str, Any]], None] | None = None,
         num_safekeepers: int = 1,
         num_pageservers: int = 1,
-        num_azs: int = 1,
         # Use non-standard SK ids to check for various parsing bugs
         safekeepers_id_start: int = 0,
         # fsync is disabled by default to make the tests go faster
@@ -402,7 +401,6 @@ class NeonEnvBuilder:
         self.pageserver_config_override = pageserver_config_override
         self.num_safekeepers = num_safekeepers
         self.num_pageservers = num_pageservers
-        self.num_azs = num_azs
         self.safekeepers_id_start = safekeepers_id_start
         self.safekeepers_enable_fsync = safekeepers_enable_fsync
         self.auth_enabled = auth_enabled
@@ -992,7 +990,6 @@ class NeonEnv:
         self.endpoints = EndpointFactory(self)
         self.safekeepers: list[Safekeeper] = []
         self.pageservers: list[NeonPageserver] = []
-        self.num_azs = config.num_azs
         self.broker = NeonBroker(self)
         self.pageserver_remote_storage = config.pageserver_remote_storage
         self.safekeepers_remote_storage = config.safekeepers_remote_storage
@@ -1093,21 +1090,14 @@ class NeonEnv:
                 http=self.port_distributor.get_port(),
             )
 
-            # Availabilty zones may also be configured manually with `NeonEnvBuilder.pageserver_config_override`
-            if self.num_azs > 1:
-                # Round-robin assignment of AZ names like us-east-2a, us-east-2b, etc.
-                az_prefix = DEFAULT_AZ_ID[:-1]
-                availability_zone = f"{az_prefix}{chr(ord('a') + (ps_id - 1) % self.num_azs)}"
-            else:
-                availability_zone = DEFAULT_AZ_ID
-
             ps_cfg: dict[str, Any] = {
                 "id": ps_id,
                 "listen_pg_addr": f"localhost:{pageserver_port.pg}",
                 "listen_http_addr": f"localhost:{pageserver_port.http}",
                 "pg_auth_type": pg_auth_type,
                 "http_auth_type": http_auth_type,
-                "availability_zone": availability_zone,
+                # Default which can be overriden with `NeonEnvBuilder.pageserver_config_override`
+                "availability_zone": DEFAULT_AZ_ID,
                 # Disable pageserver disk syncs in tests: when running tests concurrently, this avoids
                 # the pageserver taking a long time to start up due to syncfs flushing other tests' data
                 "no_sync": True,
@@ -1894,10 +1884,7 @@ class NeonStorageController(MetricsGetter, LogUtils):
         )
         return response.json()
 
-    def tenant_shard_dump(self):
-        """
-        Debug listing API: dumps the internal map of tenant shards
-        """
+    def tenant_list(self):
         response = self.request(
             "GET",
             f"{self.api}/debug/v1/tenant",
@@ -1905,18 +1892,6 @@ class NeonStorageController(MetricsGetter, LogUtils):
         )
         return response.json()
 
-    def tenant_list(self, **kwargs):
-        """
-        Control API tenant listing: a vector of the same content returned by tenant_describe
-        """
-        response = self.request(
-            "GET",
-            f"{self.api}/control/v1/tenant",
-            headers=self.headers(TokenScope.ADMIN),
-            params=kwargs,
-        )
-        return response.json()
-
     def node_configure(self, node_id, body: dict[str, Any]):
         log.info(f"node_configure({node_id}, {body})")
         body["node_id"] = node_id
@@ -2263,7 +2238,7 @@ class NeonStorageController(MetricsGetter, LogUtils):
         """
         Get the intent and observed placements of all tenants known to the storage controller.
         """
-        tenants = self.tenant_shard_dump()
+        tenants = self.tenant_list()
 
         tenant_placement: defaultdict[str, dict[str, Any]] = defaultdict(
             lambda: {
@@ -2346,14 +2321,6 @@ class NeonStorageController(MetricsGetter, LogUtils):
             json=body,
         )
 
-    def safekeeper_scheduling_policy(self, id: int, scheduling_policy: str):
-        self.request(
-            "POST",
-            f"{self.api}/control/v1/safekeeper/{id}/scheduling_policy",
-            headers=self.headers(TokenScope.ADMIN),
-            json={"id": id, "scheduling_policy": scheduling_policy},
-        )
-
     def get_safekeeper(self, id: int) -> dict[str, Any] | None:
         try:
             response = self.request(
@@ -4153,7 +4120,7 @@ class Endpoint(PgProtocol, LogUtils):
 
     # Checkpoints running endpoint and returns pg_wal size in MB.
     def get_pg_wal_size(self):
-        log.info(f"checkpointing at LSN {self.safe_psql('select pg_current_wal_lsn()')[0][0]}")
+        log.info(f'checkpointing at LSN {self.safe_psql("select pg_current_wal_lsn()")[0][0]}')
         self.safe_psql("checkpoint")
         assert self.pgdata_dir is not None  # please mypy
         return get_dir_size(self.pgdata_dir / "pg_wal") / 1024 / 1024
@@ -4993,7 +4960,7 @@ def logical_replication_sync(
         if res:
             log.info(f"subscriber_lsn={res}")
             subscriber_lsn = Lsn(res)
-            log.info(f"Subscriber LSN={subscriber_lsn}, publisher LSN={publisher_lsn}")
+            log.info(f"Subscriber LSN={subscriber_lsn}, publisher LSN={ publisher_lsn}")
             if subscriber_lsn >= publisher_lsn:
                 return subscriber_lsn
         time.sleep(0.5)

test_runner/fixtures/pageserver/http.py

@@ -15,6 +15,7 @@ from requests.adapters import HTTPAdapter
 from urllib3.util.retry import Retry
 
 from fixtures.common_types import (
+    Id,
     Lsn,
     TenantId,
     TenantShardId,
@@ -24,7 +25,7 @@ from fixtures.common_types import (
 from fixtures.log_helper import log
 from fixtures.metrics import Metrics, MetricsGetter, parse_metrics
 from fixtures.pg_version import PgVersion
-from fixtures.utils import EnhancedJSONEncoder, Fn
+from fixtures.utils import Fn
 
 
 class PageserverApiException(Exception):
@@ -82,6 +83,14 @@ class TimelineCreateRequest:
     mode: TimelineCreateRequestMode
 
     def to_json(self) -> str:
+        class EnhancedJSONEncoder(json.JSONEncoder):
+            def default(self, o):
+                if dataclasses.is_dataclass(o) and not isinstance(o, type):
+                    return dataclasses.asdict(o)
+                elif isinstance(o, Id):
+                    return o.id.hex()
+                return super().default(o)
+
         # mode is flattened
         this = dataclasses.asdict(self)
         mode = this.pop("mode")

test_runner/fixtures/safekeeper/http.py

@@ -10,7 +10,7 @@ import requests
 from fixtures.common_types import Lsn, TenantId, TenantTimelineId, TimelineId
 from fixtures.log_helper import log
 from fixtures.metrics import Metrics, MetricsGetter, parse_metrics
-from fixtures.utils import EnhancedJSONEncoder, wait_until
+from fixtures.utils import wait_until
 
 if TYPE_CHECKING:
     from typing import Any
@@ -25,7 +25,6 @@ class Walreceiver:
 
 @dataclass
 class SafekeeperTimelineStatus:
-    mconf: Configuration | None
     term: int
     last_log_term: int
     pg_version: int  # Not exactly a PgVersion, safekeeper returns version as int, for example 150002 for 15.2
@@ -70,56 +69,6 @@ class TermBumpResponse:
         )
 
 
-@dataclass
-class SafekeeperId:
-    id: int
-    host: str
-    pg_port: int
-
-
-@dataclass
-class Configuration:
-    generation: int
-    members: list[SafekeeperId]
-    new_members: list[SafekeeperId] | None
-
-    @classmethod
-    def from_json(cls, d: dict[str, Any]) -> Configuration:
-        generation = d["generation"]
-        members = d["members"]
-        new_members = d.get("new_members")
-        return Configuration(generation, members, new_members)
-
-    def to_json(self) -> str:
-        return json.dumps(self, cls=EnhancedJSONEncoder)
-
-
-@dataclass
-class TimelineCreateRequest:
-    tenant_id: TenantId
-    timeline_id: TimelineId
-    mconf: Configuration
-    # not exactly PgVersion, for example 150002 for 15.2
-    pg_version: int
-    start_lsn: Lsn
-    commit_lsn: Lsn | None
-
-    def to_json(self) -> str:
-        return json.dumps(self, cls=EnhancedJSONEncoder)
-
-
-@dataclass
-class TimelineMembershipSwitchResponse:
-    previous_conf: Configuration
-    current_conf: Configuration
-
-    @classmethod
-    def from_json(cls, d: dict[str, Any]) -> TimelineMembershipSwitchResponse:
-        previous_conf = Configuration.from_json(d["previous_conf"])
-        current_conf = Configuration.from_json(d["current_conf"])
-        return TimelineMembershipSwitchResponse(previous_conf, current_conf)
-
-
 class SafekeeperHttpClient(requests.Session, MetricsGetter):
     HTTPError = requests.HTTPError
 
@@ -182,8 +131,20 @@ class SafekeeperHttpClient(requests.Session, MetricsGetter):
         resj = res.json()
         return [TenantTimelineId.from_json(ttidj) for ttidj in resj]
 
-    def timeline_create(self, r: TimelineCreateRequest):
-        res = self.post(f"http://localhost:{self.port}/v1/tenant/timeline", data=r.to_json())
+    def timeline_create(
+        self,
+        tenant_id: TenantId,
+        timeline_id: TimelineId,
+        pg_version: int,  # Not exactly a PgVersion, safekeeper returns version as int, for example 150002 for 15.2
+        commit_lsn: Lsn,
+    ):
+        body = {
+            "tenant_id": str(tenant_id),
+            "timeline_id": str(timeline_id),
+            "pg_version": pg_version,
+            "commit_lsn": str(commit_lsn),
+        }
+        res = self.post(f"http://localhost:{self.port}/v1/tenant/timeline", json=body)
         res.raise_for_status()
 
     def timeline_status(
@@ -193,10 +154,7 @@ class SafekeeperHttpClient(requests.Session, MetricsGetter):
         res.raise_for_status()
         resj = res.json()
         walreceivers = [Walreceiver(wr["conn_id"], wr["status"]) for wr in resj["walreceivers"]]
-        # It is always normally not None, it is allowed only to make forward compat tests happy.
-        mconf = Configuration.from_json(resj["mconf"]) if "mconf" in resj else None
         return SafekeeperTimelineStatus(
-            mconf=mconf,
             term=resj["acceptor_state"]["term"],
             last_log_term=resj["acceptor_state"]["epoch"],
             pg_version=resj["pg_info"]["pg_version"],
@@ -222,11 +180,6 @@ class SafekeeperHttpClient(requests.Session, MetricsGetter):
     def get_commit_lsn(self, tenant_id: TenantId, timeline_id: TimelineId) -> Lsn:
         return self.timeline_status(tenant_id, timeline_id).commit_lsn
 
-    # Get timeline membership configuration.
-    def get_membership(self, tenant_id: TenantId, timeline_id: TimelineId) -> Configuration:
-        # make mypy happy
-        return self.timeline_status(tenant_id, timeline_id).mconf  # type: ignore
-
     # only_local doesn't remove segments in the remote storage.
     def timeline_delete(
         self, tenant_id: TenantId, timeline_id: TimelineId, only_local: bool = False
@@ -273,16 +226,6 @@ class SafekeeperHttpClient(requests.Session, MetricsGetter):
         assert isinstance(res_json, dict)
         return res_json
 
-    def membership_switch(
-        self, tenant_id: TenantId, timeline_id: TimelineId, to: Configuration
-    ) -> TimelineMembershipSwitchResponse:
-        res = self.post(
-            f"http://localhost:{self.port}/v1/tenant/{tenant_id}/timeline/{timeline_id}/membership",
-            data=to.to_json(),
-        )
-        res.raise_for_status()
-        return TimelineMembershipSwitchResponse.from_json(res.json())
-
     def copy_timeline(self, tenant_id: TenantId, timeline_id: TimelineId, body: dict[str, Any]):
         res = self.post(
             f"http://localhost:{self.port}/v1/tenant/{tenant_id}/timeline/{timeline_id}/copy",

test_runner/fixtures/utils.py

@@ -1,7 +1,6 @@
 from __future__ import annotations
 
 import contextlib
-import dataclasses
 import json
 import os
 import re
@@ -22,7 +21,6 @@ import zstandard
 from psycopg2.extensions import cursor
 from typing_extensions import override
 
-from fixtures.common_types import Id, Lsn
 from fixtures.log_helper import log
 from fixtures.pageserver.common_types import (
     parse_delta_layer,
@@ -607,22 +605,6 @@ class PropagatingThread(threading.Thread):
         return self.ret
 
 
-class EnhancedJSONEncoder(json.JSONEncoder):
-    """
-    Default json.JSONEncoder works only on primitive builtins. Extend it to any
-    dataclass plus our custom types.
-    """
-
-    def default(self, o):
-        if dataclasses.is_dataclass(o) and not isinstance(o, type):
-            return dataclasses.asdict(o)
-        elif isinstance(o, Id):
-            return o.id.hex()
-        elif isinstance(o, Lsn):
-            return str(o)  # standard hex notation
-        return super().default(o)
-
-
 def human_bytes(amt: float) -> str:
     """
     Render a bytes amount into nice IEC bytes string.