fix test failure

Signed-off-by: Alex Chi Z <chi@neon.tech>

.github/actionlint.yml

@@ -25,3 +25,4 @@ 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/workflows/_build-and-test-locally.yml

@@ -229,8 +229,13 @@ jobs:
           ${cov_prefix} cargo nextest run $CARGO_FLAGS $CARGO_FEATURES -E '!package(pageserver)'
 
           # run pageserver tests with different settings
-          for io_engine in std-fs tokio-epoll-uring ; do
-            NEON_PAGESERVER_UNIT_TEST_VIRTUAL_FILE_IOENGINE=$io_engine ${cov_prefix} cargo nextest run $CARGO_FLAGS $CARGO_FEATURES  -E 'package(pageserver)'
+          for get_vectored_concurrent_io in sequential sidecar-task; do
+            for io_engine in std-fs tokio-epoll-uring ; do
+              NEON_PAGESERVER_UNIT_TEST_GET_VECTORED_CONCURRENT_IO=$get_vectored_concurrent_io \
+                NEON_PAGESERVER_UNIT_TEST_VIRTUAL_FILE_IOENGINE=$io_engine \
+                ${cov_prefix} \
+                cargo nextest run $CARGO_FLAGS $CARGO_FEATURES  -E 'package(pageserver)'
+            done
           done
 
           # Run separate tests for real S3
@@ -314,6 +319,7 @@ jobs:
           CHECK_ONDISK_DATA_COMPATIBILITY: nonempty
           BUILD_TAG: ${{ inputs.build-tag }}
           PAGESERVER_VIRTUAL_FILE_IO_ENGINE: tokio-epoll-uring
+          PAGESERVER_GET_VECTORED_CONCURRENT_IO: sidecar-task
           USE_LFC: ${{ matrix.lfc_state == 'with-lfc' && 'true' || 'false' }}
 
       # Temporary disable this step until we figure out why it's so flaky

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

@@ -0,0 +1,91 @@
+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

@@ -164,77 +164,11 @@ jobs:
 
   check-codestyle-rust:
     needs: [ check-permissions, build-build-tools-image ]
-    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
+    uses: ./.github/workflows/_check-codestyle-rust.yml
+    with:
+      build-tools-image: ${{ needs.build-build-tools-image.outputs.image }}-bookworm
+      archs: '["x64", "arm64"]'
+    secrets: inherit
 
   build-and-test-locally:
     needs: [ tag, build-build-tools-image ]
@@ -346,25 +280,22 @@ 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-failures:
+  report-benchmarks-results-to-slack:
     needs: [ benchmarks, create-test-report ]
-    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
+    if: github.ref_name == 'main' && !cancelled() && contains(fromJSON('["success", "failure"]'), needs.benchmarks.result)
     runs-on: ubuntu-22.04
 
     steps:
-    - uses: slackapi/slack-github-action@v1
+    - uses: slackapi/slack-github-action@v2
       with:
-        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 }}
+        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 }}>
 
   create-test-report:
     needs: [ check-permissions, build-and-test-locally, coverage-report, build-build-tools-image, benchmarks ]
@@ -889,11 +820,11 @@ jobs:
       - name: Print logs and clean up
         if: always()
         run: |
-          docker compose -f ./docker-compose/docker-compose.yml logs || 0
-          docker compose -f ./docker-compose/docker-compose.yml down
+          docker compose --profile test-extensions -f ./docker-compose/docker-compose.yml logs || true
+          docker compose --profile test-extensions -f ./docker-compose/docker-compose.yml down
 
   promote-images-dev:
-    needs: [ check-permissions, tag, vm-compute-node-image ]
+    needs: [ check-permissions, tag, vm-compute-node-image, neon-image ]
     runs-on: ubuntu-22.04
 
     permissions:
@@ -928,7 +859,7 @@ jobs:
           done
 
   promote-images-prod:
-    needs: [ check-permissions, tag, test-images, vm-compute-node-image ]
+    needs: [ check-permissions, tag, test-images, promote-images-dev ]
     runs-on: ubuntu-22.04
     if: github.ref_name == 'main' || github.ref_name == 'release' || github.ref_name == 'release-proxy' || github.ref_name == 'release-compute'
 
@@ -1147,12 +1078,6 @@ jobs:
               console.log(`Tag ${tag} created successfully.`);
             }
 
-            // TODO: check how GitHub releases looks for proxy/compute releases and enable them if they're ok
-            if (context.ref !== 'refs/heads/release') {
-              console.log(`GitHub release skipped for ${context.ref}.`);
-              return;
-            }
-
             try {
               const existingRelease = await github.rest.repos.getReleaseByTag({
                 owner: context.repo.owner,
@@ -1171,7 +1096,8 @@ jobs:
                 owner: context.repo.owner,
                 repo: context.repo.repo,
                 tag_name: tag,
-                generate_release_notes: true,
+                // TODO: Automate release notes properly
+                generate_release_notes: false,
               });
               console.log(`Release for tag ${tag} created successfully.`);
             }

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

@@ -1,6 +1,12 @@
 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
@@ -17,8 +23,10 @@ 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:
@@ -30,11 +38,25 @@ 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'
@@ -55,6 +77,16 @@ 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
@@ -67,6 +99,7 @@ jobs:
     needs:
       - get-changed-files
       - check-codestyle-python
+      - check-codestyle-rust
     runs-on: ubuntu-22.04
     steps:
       - name: Create fake `neon-cloud-e2e` check

.github/workflows/release.yml

@@ -3,8 +3,9 @@ name: Create Release Branch
 on:
   schedule:
     # It should be kept in sync with if-condition in jobs
-    - cron: '0 6 * * FRI' # Storage release
     - cron: '0 6 * * THU' # Proxy release
+    - cron: '0 6 * * FRI' # Storage release
+    - cron: '0 7 * * FRI' # Compute release
   workflow_dispatch:
     inputs:
       create-storage-release-branch:
@@ -55,7 +56,7 @@ jobs:
       ci-access-token: ${{ secrets.CI_ACCESS_TOKEN }}
 
   create-compute-release-branch:
-    if: inputs.create-compute-release-branch
+    if: ${{ github.event.schedule == '0 7 * * FRI' || inputs.create-compute-release-branch }}
 
     permissions:
       contents: write

Cargo.lock

@@ -942,6 +942,18 @@ version = "1.6.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "8c3c1a368f70d6cf7302d78f8f7093da241fb8e8807c05cc9e51a125895a6d5b"
 
+[[package]]
+name = "bb8"
+version = "0.8.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d89aabfae550a5c44b43ab941844ffcd2e993cb6900b342debf59e9ea74acdb8"
+dependencies = [
+ "async-trait",
+ "futures-util",
+ "parking_lot 0.12.1",
+ "tokio",
+]
+
 [[package]]
 name = "bcder"
 version = "0.7.4"
@@ -1301,7 +1313,7 @@ dependencies = [
  "tar",
  "thiserror",
  "tokio",
- "tokio-postgres",
+ "tokio-postgres 0.7.7",
  "tokio-stream",
  "tokio-util",
  "tower 0.5.2",
@@ -1312,6 +1324,7 @@ dependencies = [
  "tracing-utils",
  "url",
  "utils",
+ "uuid",
  "vm_monitor",
  "workspace_hack",
  "zstd",
@@ -1409,7 +1422,7 @@ dependencies = [
  "storage_broker",
  "thiserror",
  "tokio",
- "tokio-postgres",
+ "tokio-postgres 0.7.7",
  "tokio-util",
  "toml",
  "toml_edit",
@@ -1605,6 +1618,32 @@ dependencies = [
  "typenum",
 ]
 
+[[package]]
+name = "curve25519-dalek"
+version = "4.1.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "97fb8b7c4503de7d6ae7b42ab72a5a59857b4c937ec27a3d4539dba95b5ab2be"
+dependencies = [
+ "cfg-if",
+ "cpufeatures",
+ "curve25519-dalek-derive",
+ "digest",
+ "fiat-crypto",
+ "rustc_version",
+ "subtle",
+]
+
+[[package]]
+name = "curve25519-dalek-derive"
+version = "0.1.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f46882e17999c6cc590af592290432be3bce0428cb0d5f8b6715e4dc7b383eb3"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.90",
+]
+
 [[package]]
 name = "darling"
 version = "0.20.1"
@@ -1759,11 +1798,24 @@ dependencies = [
  "chrono",
  "diesel_derives",
  "itoa",
- "pq-sys",
- "r2d2",
  "serde_json",
 ]
 
+[[package]]
+name = "diesel-async"
+version = "0.5.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "51a307ac00f7c23f526a04a77761a0519b9f0eb2838ebf5b905a58580095bdcb"
+dependencies = [
+ "async-trait",
+ "bb8",
+ "diesel",
+ "futures-util",
+ "scoped-futures",
+ "tokio",
+ "tokio-postgres 0.7.12",
+]
+
 [[package]]
 name = "diesel_derives"
 version = "2.2.1"
@@ -1875,6 +1927,28 @@ dependencies = [
  "spki 0.7.3",
 ]
 
+[[package]]
+name = "ed25519"
+version = "2.2.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "115531babc129696a58c64a4fef0a8bf9e9698629fb97e9e40767d235cfbcd53"
+dependencies = [
+ "signature 2.2.0",
+]
+
+[[package]]
+name = "ed25519-dalek"
+version = "2.1.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4a3daa8e81a3963a60642bcc1f90a670680bd4a77535faa384e9d1c79d620871"
+dependencies = [
+ "curve25519-dalek",
+ "ed25519",
+ "rand_core 0.6.4",
+ "sha2",
+ "subtle",
+]
+
 [[package]]
 name = "either"
 version = "1.8.1"
@@ -2113,6 +2187,12 @@ dependencies = [
  "subtle",
 ]
 
+[[package]]
+name = "fiat-crypto"
+version = "0.2.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "28dea519a9695b9977216879a3ebfddf92f1c08c05d984f8996aecd6ecdc811d"
+
 [[package]]
 name = "filetime"
 version = "0.2.22"
@@ -3927,9 +4007,11 @@ name = "pagectl"
 version = "0.1.0"
 dependencies = [
  "anyhow",
+ "bincode",
  "camino",
  "clap",
  "humantime",
+ "itertools 0.10.5",
  "pageserver",
  "pageserver_api",
  "postgres_ffi",
@@ -3951,6 +4033,7 @@ dependencies = [
  "arc-swap",
  "async-compression",
  "async-stream",
+ "bincode",
  "bit_field",
  "byteorder",
  "bytes",
@@ -3984,12 +4067,13 @@ dependencies = [
  "pageserver_compaction",
  "pin-project-lite",
  "postgres",
- "postgres-protocol",
- "postgres-types",
+ "postgres-protocol 0.6.4",
+ "postgres-types 0.2.4",
  "postgres_backend",
  "postgres_connection",
  "postgres_ffi",
  "postgres_initdb",
+ "pprof",
  "pq_proto",
  "procfs",
  "rand 0.8.5",
@@ -4015,7 +4099,7 @@ dependencies = [
  "tokio",
  "tokio-epoll-uring",
  "tokio-io-timeout",
- "tokio-postgres",
+ "tokio-postgres 0.7.7",
  "tokio-stream",
  "tokio-tar",
  "tokio-util",
@@ -4073,7 +4157,7 @@ dependencies = [
  "serde",
  "thiserror",
  "tokio",
- "tokio-postgres",
+ "tokio-postgres 0.7.7",
  "tokio-stream",
  "tokio-util",
  "utils",
@@ -4379,7 +4463,7 @@ dependencies = [
  "futures-util",
  "log",
  "tokio",
- "tokio-postgres",
+ "tokio-postgres 0.7.7",
 ]
 
 [[package]]
@@ -4400,6 +4484,24 @@ dependencies = [
  "stringprep",
 ]
 
+[[package]]
+name = "postgres-protocol"
+version = "0.6.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "acda0ebdebc28befa84bee35e651e4c5f09073d668c7aed4cf7e23c3cda84b23"
+dependencies = [
+ "base64 0.22.1",
+ "byteorder",
+ "bytes",
+ "fallible-iterator",
+ "hmac",
+ "md-5",
+ "memchr",
+ "rand 0.8.5",
+ "sha2",
+ "stringprep",
+]
+
 [[package]]
 name = "postgres-protocol2"
 version = "0.1.0"
@@ -4423,7 +4525,18 @@ source = "git+https://github.com/neondatabase/rust-postgres.git?branch=neon#511f
 dependencies = [
  "bytes",
  "fallible-iterator",
- "postgres-protocol",
+ "postgres-protocol 0.6.4",
+]
+
+[[package]]
+name = "postgres-types"
+version = "0.2.8"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f66ea23a2d0e5734297357705193335e0a957696f34bed2f2faefacb2fec336f"
+dependencies = [
+ "bytes",
+ "fallible-iterator",
+ "postgres-protocol 0.6.7",
 ]
 
 [[package]]
@@ -4448,7 +4561,7 @@ dependencies = [
  "serde",
  "thiserror",
  "tokio",
- "tokio-postgres",
+ "tokio-postgres 0.7.7",
  "tokio-postgres-rustls",
  "tokio-rustls 0.26.0",
  "tokio-util",
@@ -4463,7 +4576,7 @@ dependencies = [
  "itertools 0.10.5",
  "once_cell",
  "postgres",
- "tokio-postgres",
+ "tokio-postgres 0.7.7",
  "url",
 ]
 
@@ -4550,15 +4663,6 @@ version = "0.2.17"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "5b40af805b3121feab8a3c29f04d8ad262fa8e0561883e7653e024ae4479e6de"
 
-[[package]]
-name = "pq-sys"
-version = "0.6.3"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "f6cc05d7ea95200187117196eee9edd0644424911821aeb28a18ce60ea0b8793"
-dependencies = [
- "vcpkg",
-]
-
 [[package]]
 name = "pq_proto"
 version = "0.1.0"
@@ -4566,7 +4670,7 @@ dependencies = [
  "byteorder",
  "bytes",
  "itertools 0.10.5",
- "postgres-protocol",
+ "postgres-protocol 0.6.4",
  "rand 0.8.5",
  "serde",
  "thiserror",
@@ -4745,6 +4849,7 @@ dependencies = [
  "consumption_metrics",
  "dashmap 5.5.0",
  "ecdsa 0.16.9",
+ "ed25519-dalek",
  "env_logger 0.10.2",
  "fallible-iterator",
  "flate2",
@@ -4813,7 +4918,7 @@ dependencies = [
  "tikv-jemalloc-ctl",
  "tikv-jemallocator",
  "tokio",
- "tokio-postgres",
+ "tokio-postgres 0.7.7",
  "tokio-postgres2",
  "tokio-rustls 0.26.0",
  "tokio-tungstenite 0.21.0",
@@ -4870,17 +4975,6 @@ dependencies = [
  "proc-macro2",
 ]
 
-[[package]]
-name = "r2d2"
-version = "0.8.10"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "51de85fb3fb6524929c8a2eb85e6b6d363de4e8c48f9e2c2eac4944abc181c93"
-dependencies = [
- "log",
- "parking_lot 0.12.1",
- "scheduled-thread-pool",
-]
-
 [[package]]
 name = "rand"
 version = "0.7.3"
@@ -5599,6 +5693,7 @@ dependencies = [
  "crc32c",
  "criterion",
  "desim",
+ "env_logger 0.10.2",
  "fail",
  "futures",
  "hex",
@@ -5611,7 +5706,7 @@ dependencies = [
  "pageserver_api",
  "parking_lot 0.12.1",
  "postgres",
- "postgres-protocol",
+ "postgres-protocol 0.6.4",
  "postgres_backend",
  "postgres_ffi",
  "pprof",
@@ -5627,6 +5722,7 @@ dependencies = [
  "serde",
  "serde_json",
  "sha2",
+ "smallvec",
  "storage_broker",
  "strum",
  "strum_macros",
@@ -5634,7 +5730,7 @@ dependencies = [
  "tikv-jemallocator",
  "tokio",
  "tokio-io-timeout",
- "tokio-postgres",
+ "tokio-postgres 0.7.7",
  "tokio-stream",
  "tokio-tar",
  "tokio-util",
@@ -5651,10 +5747,13 @@ dependencies = [
 name = "safekeeper_api"
 version = "0.1.0"
 dependencies = [
+ "anyhow",
  "const_format",
+ "pageserver_api",
  "postgres_ffi",
  "pq_proto",
  "serde",
+ "serde_json",
  "tokio",
  "utils",
 ]
@@ -5690,12 +5789,12 @@ dependencies = [
 ]
 
 [[package]]
-name = "scheduled-thread-pool"
-version = "0.2.7"
+name = "scoped-futures"
+version = "0.1.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "3cbc66816425a074528352f5789333ecff06ca41b36b0b0efdfbb29edc391a19"
+checksum = "1b24aae2d0636530f359e9d5ef0c04669d11c5e756699b27a6a6d845d8329091"
 dependencies = [
- "parking_lot 0.12.1",
+ "pin-project-lite",
 ]
 
 [[package]]
@@ -6230,6 +6329,7 @@ dependencies = [
  "clap",
  "control_plane",
  "diesel",
+ "diesel-async",
  "diesel_migrations",
  "fail",
  "futures",
@@ -6244,10 +6344,10 @@ dependencies = [
  "pageserver_api",
  "pageserver_client",
  "postgres_connection",
- "r2d2",
  "rand 0.8.5",
  "reqwest",
  "routerify",
+ "scoped-futures",
  "scopeguard",
  "serde",
  "serde_json",
@@ -6300,7 +6400,7 @@ dependencies = [
  "serde_json",
  "storage_controller_client",
  "tokio",
- "tokio-postgres",
+ "tokio-postgres 0.7.7",
  "tokio-postgres-rustls",
  "tokio-stream",
  "tokio-util",
@@ -6709,7 +6809,7 @@ dependencies = [
 [[package]]
 name = "tokio-epoll-uring"
 version = "0.1.0"
-source = "git+https://github.com/neondatabase/tokio-epoll-uring.git?branch=main#33e00106a268644d02ba0461bbd64476073b0ee1"
+source = "git+https://github.com/neondatabase/tokio-epoll-uring.git?branch=main#781989bb540a1408b0b93daa1e9d1fa452195497"
 dependencies = [
  "futures",
  "nix 0.26.4",
@@ -6759,13 +6859,39 @@ dependencies = [
  "percent-encoding",
  "phf",
  "pin-project-lite",
- "postgres-protocol",
- "postgres-types",
+ "postgres-protocol 0.6.4",
+ "postgres-types 0.2.4",
  "socket2",
  "tokio",
  "tokio-util",
 ]
 
+[[package]]
+name = "tokio-postgres"
+version = "0.7.12"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3b5d3742945bc7d7f210693b0c58ae542c6fd47b17adbbda0885f3dcb34a6bdb"
+dependencies = [
+ "async-trait",
+ "byteorder",
+ "bytes",
+ "fallible-iterator",
+ "futures-channel",
+ "futures-util",
+ "log",
+ "parking_lot 0.12.1",
+ "percent-encoding",
+ "phf",
+ "pin-project-lite",
+ "postgres-protocol 0.6.7",
+ "postgres-types 0.2.8",
+ "rand 0.8.5",
+ "socket2",
+ "tokio",
+ "tokio-util",
+ "whoami",
+]
+
 [[package]]
 name = "tokio-postgres-rustls"
 version = "0.12.0"
@@ -6775,7 +6901,7 @@ dependencies = [
  "ring",
  "rustls 0.23.18",
  "tokio",
- "tokio-postgres",
+ "tokio-postgres 0.7.7",
  "tokio-rustls 0.26.0",
  "x509-certificate",
 ]
@@ -6939,12 +7065,9 @@ version = "0.12.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "877c5b330756d856ffcc4553ab34a5684481ade925ecc54bcd1bf02b1d0d4d52"
 dependencies = [
- "async-stream",
  "async-trait",
- "axum",
  "base64 0.22.1",
  "bytes",
- "h2 0.4.4",
  "http 1.1.0",
  "http-body 1.0.0",
  "http-body-util",
@@ -6956,7 +7079,6 @@ dependencies = [
  "prost",
  "rustls-native-certs 0.8.0",
  "rustls-pemfile 2.1.1",
- "socket2",
  "tokio",
  "tokio-rustls 0.26.0",
  "tokio-stream",
@@ -7308,7 +7430,7 @@ dependencies = [
 [[package]]
 name = "uring-common"
 version = "0.1.0"
-source = "git+https://github.com/neondatabase/tokio-epoll-uring.git?branch=main#33e00106a268644d02ba0461bbd64476073b0ee1"
+source = "git+https://github.com/neondatabase/tokio-epoll-uring.git?branch=main#781989bb540a1408b0b93daa1e9d1fa452195497"
 dependencies = [
  "bytes",
  "io-uring",
@@ -7437,12 +7559,6 @@ version = "0.1.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "830b7e5d4d90034032940e4ace0d9a9a057e7a45cd94e6c007832e39edb82f6d"
 
-[[package]]
-name = "vcpkg"
-version = "0.2.15"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "accd4ea62f7bb7a82fe23066fb0957d48ef677f6eeb8215f372f52e48bb32426"
-
 [[package]]
 name = "version_check"
 version = "0.9.4"
@@ -7462,7 +7578,7 @@ dependencies = [
  "serde_json",
  "sysinfo",
  "tokio",
- "tokio-postgres",
+ "tokio-postgres 0.7.7",
  "tokio-util",
  "tracing",
  "tracing-subscriber",
@@ -7502,13 +7618,21 @@ 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",
- "tonic",
+ "tokio-util",
  "tonic-build",
  "tracing",
  "utils",
@@ -7917,8 +8041,6 @@ version = "0.1.0"
 dependencies = [
  "ahash",
  "anyhow",
- "axum",
- "axum-core",
  "base64 0.13.1",
  "base64 0.21.1",
  "base64ct",
@@ -7999,7 +8121,6 @@ dependencies = [
  "toml_edit",
  "tonic",
  "tower 0.4.13",
- "tower 0.5.2",
  "tracing",
  "tracing-core",
  "url",

Cargo.toml

@@ -187,7 +187,7 @@ tokio-tar = "0.3"
 tokio-util = { version = "0.7.10", features = ["io", "rt"] }
 toml = "0.8"
 toml_edit = "0.22"
-tonic = {version = "0.12.3", features = ["tls", "tls-roots"]}
+tonic = {version = "0.12.3", default-features = false, features = ["channel", "tls", "tls-roots"]}
 tower = { version = "0.5.2", default-features = false }
 tower-http = { version = "0.6.2", features = ["request-id", "trace"] }
 tower-service = "0.3.3"

Dockerfile

@@ -71,6 +71,7 @@ RUN set -e \
         ca-certificates \
 	# System postgres for use with client libraries (e.g. in storage controller)
         postgresql-15 \
+        openssl \
     && rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/* \
     && useradd -d /data neon \
     && chown -R neon:neon /data

Makefile

@@ -3,7 +3,6 @@ ROOT_PROJECT_DIR := $(dir $(abspath $(lastword $(MAKEFILE_LIST))))
 # Where to install Postgres, default is ./pg_install, maybe useful for package managers
 POSTGRES_INSTALL_DIR ?= $(ROOT_PROJECT_DIR)/pg_install/
 
-OPENSSL_PREFIX_DIR := /usr/local/openssl
 ICU_PREFIX_DIR := /usr/local/icu
 
 #
@@ -26,11 +25,9 @@ endif
 ifeq ($(shell test -e /home/nonroot/.docker_build && echo -n yes),yes)
 	# Exclude static build openssl, icu for local build (MacOS, Linux)
 	# Only keep for build type release and debug
-	PG_CFLAGS += -I$(OPENSSL_PREFIX_DIR)/include
 	PG_CONFIGURE_OPTS += --with-icu
 	PG_CONFIGURE_OPTS += ICU_CFLAGS='-I/$(ICU_PREFIX_DIR)/include -DU_STATIC_IMPLEMENTATION'
 	PG_CONFIGURE_OPTS += ICU_LIBS='-L$(ICU_PREFIX_DIR)/lib -L$(ICU_PREFIX_DIR)/lib64 -licui18n -licuuc -licudata -lstdc++ -Wl,-Bdynamic -lm'
-	PG_CONFIGURE_OPTS += LDFLAGS='-L$(OPENSSL_PREFIX_DIR)/lib -L$(OPENSSL_PREFIX_DIR)/lib64 -L$(ICU_PREFIX_DIR)/lib -L$(ICU_PREFIX_DIR)/lib64 -Wl,-Bstatic -lssl -lcrypto -Wl,-Bdynamic -lrt -lm -ldl -lpthread'
 endif
 
 UNAME_S := $(shell uname -s)
@@ -67,8 +64,6 @@ CARGO_BUILD_FLAGS += $(filter -j1,$(MAKEFLAGS))
 CARGO_CMD_PREFIX += $(if $(filter n,$(MAKEFLAGS)),,+)
 # Force cargo not to print progress bar
 CARGO_CMD_PREFIX += CARGO_TERM_PROGRESS_WHEN=never CI=1
-# Set PQ_LIB_DIR to make sure `storage_controller` get linked with bundled libpq (through diesel)
-CARGO_CMD_PREFIX += PQ_LIB_DIR=$(POSTGRES_INSTALL_DIR)/v16/lib
 
 CACHEDIR_TAG_CONTENTS := "Signature: 8a477f597d28d172789f06886806bc55"
 

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=0.16.0
+ENV SQL_EXPORTER_VERSION=0.17.0
 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 \
@@ -190,21 +190,6 @@ RUN for package in Capture::Tiny DateTime Devel::Cover Digest::MD5 File::Spec JS
     && make install \
     && rm -rf ../lcov.tar.gz
 
-# Compile and install the static OpenSSL library
-ENV OPENSSL_VERSION=1.1.1w
-ENV OPENSSL_PREFIX=/usr/local/openssl
-RUN wget -O /tmp/openssl-${OPENSSL_VERSION}.tar.gz https://www.openssl.org/source/openssl-${OPENSSL_VERSION}.tar.gz && \
-    echo "cf3098950cb4d853ad95c0841f1f9c6d3dc102dccfcacd521d93925208b76ac8 /tmp/openssl-${OPENSSL_VERSION}.tar.gz" | sha256sum --check && \
-    cd /tmp && \
-    tar xzvf /tmp/openssl-${OPENSSL_VERSION}.tar.gz && \
-    rm /tmp/openssl-${OPENSSL_VERSION}.tar.gz && \
-    cd /tmp/openssl-${OPENSSL_VERSION} && \
-    ./config --prefix=${OPENSSL_PREFIX}  -static --static no-shared -fPIC && \
-    make -j "$(nproc)" && \
-    make install && \
-    cd /tmp && \
-    rm -rf /tmp/openssl-${OPENSSL_VERSION}
-
 # Use the same version of libicu as the compute nodes so that
 # clusters created using inidb on pageserver can be used by computes.
 #

compute/compute-node.Dockerfile

@@ -66,6 +66,10 @@ 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/postgres_fdw.control && \
+    file=/usr/local/pgsql/share/extension/postgres_fdw--1.0.sql && [ -e $file ] && \
+    echo 'GRANT USAGE ON FOREIGN DATA WRAPPER postgres_fdw TO neon_superuser;' >> $file && \
     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 && \
@@ -104,16 +108,18 @@ RUN cd postgres && \
         esac; \
     done;
 
+# Set PATH for all the subsequent build steps
+ENV PATH="/usr/local/pgsql/bin:$PATH"
+
 #########################################################################################
 #
 # Layer "postgis-build"
 # Build PostGIS from the upstream PostGIS mirror.
 #
 #########################################################################################
-FROM build-deps AS postgis-build
+FROM pg-build AS postgis-build
 ARG DEBIAN_VERSION
 ARG PG_VERSION
-COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 RUN apt update && \
     apt install --no-install-recommends --no-install-suggests -y \
     gdal-bin libboost-dev libboost-thread-dev libboost-filesystem-dev \
@@ -151,8 +157,6 @@ RUN case "${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") \
@@ -170,7 +174,6 @@ RUN case "${PG_VERSION}" in \
     wget https://download.osgeo.org/postgis/source/postgis-${POSTGIS_VERSION}.tar.gz -O postgis.tar.gz && \
     echo "${POSTGIS_CHECKSUM} postgis.tar.gz" | sha256sum --check && \
     mkdir postgis-src && cd postgis-src && tar xzf ../postgis.tar.gz --strip-components=1 -C . && \
-    find /usr/local/pgsql -type f | sed 's|^/usr/local/pgsql/||' > /before.txt &&\
     ./autogen.sh && \
     ./configure --with-sfcgal=/usr/local/bin/sfcgal-config && \
     make -j $(getconf _NPROCESSORS_ONLN) && \
@@ -220,11 +223,7 @@ RUN case "${PG_VERSION}" in \
     cmake -GNinja -DCMAKE_BUILD_TYPE=Release .. && \
     ninja -j $(getconf _NPROCESSORS_ONLN) && \
     ninja -j $(getconf _NPROCESSORS_ONLN) install && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/pgrouting.control && \
-    find /usr/local/pgsql -type f | sed 's|^/usr/local/pgsql/||' > /after.txt &&\
-    cp /usr/local/pgsql/share/extension/pgrouting.control /extensions/postgis && \
-    sort -o /before.txt /before.txt && sort -o /after.txt /after.txt && \
-    comm -13 /before.txt /after.txt | tar --directory=/usr/local/pgsql --zstd -cf /extensions/postgis.tar.zst -T -
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/pgrouting.control
 
 #########################################################################################
 #
@@ -232,9 +231,8 @@ RUN case "${PG_VERSION}" in \
 # Build plv8
 #
 #########################################################################################
-FROM build-deps AS plv8-build
+FROM pg-build 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
 
@@ -269,7 +267,6 @@ 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 $(getconf _NPROCESSORS_ONLN) install && \
     rm -rf /plv8-* && \
     find /usr/local/pgsql/ -name "plv8-*.so" | xargs strip && \
@@ -296,9 +293,8 @@ RUN case "${PG_VERSION}" in \
 # Build h3_pg
 #
 #########################################################################################
-FROM build-deps AS h3-pg-build
+FROM pg-build 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
@@ -319,7 +315,6 @@ 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 && \
@@ -331,17 +326,16 @@ RUN wget https://github.com/zachasme/h3-pg/archive/refs/tags/v4.1.3.tar.gz -O h3
 # compile unit extension
 #
 #########################################################################################
-FROM build-deps AS unit-pg-build
+FROM pg-build 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) PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
-    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
+    make -j $(getconf _NPROCESSORS_ONLN) && \
+    make -j $(getconf _NPROCESSORS_ONLN) install && \
     # 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.
@@ -355,9 +349,8 @@ RUN wget https://github.com/df7cb/postgresql-unit/archive/refs/tags/7.9.tar.gz -
 # compile pgvector extension
 #
 #########################################################################################
-FROM build-deps AS vector-pg-build
+FROM pg-build AS vector-pg-build
 ARG PG_VERSION
-COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 COPY compute/patches/pgvector.patch /pgvector.patch
 
@@ -370,9 +363,11 @@ COPY compute/patches/pgvector.patch /pgvector.patch
 RUN wget https://github.com/pgvector/pgvector/archive/refs/tags/v0.8.0.tar.gz -O pgvector.tar.gz && \
     echo "867a2c328d4928a5a9d6f052cd3bc78c7d60228a9b914ad32aa3db88e9de27b0 pgvector.tar.gz" | sha256sum --check && \
     mkdir pgvector-src && cd pgvector-src && tar xzf ../pgvector.tar.gz --strip-components=1 -C . && \
+    wget https://github.com/pgvector/pgvector/raw/refs/tags/v0.7.4/sql/vector.sql -O ./sql/vector--0.7.4.sql && \
+    echo "10218d05dc02299562252a9484775178b14a1d8edb92a2d1672ef488530f7778 ./sql/vector--0.7.4.sql" | sha256sum --check && \
     patch -p1 < /pgvector.patch && \
-    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 && \
+    make -j $(getconf _NPROCESSORS_ONLN) OPTFLAGS="" && \
+    make -j $(getconf _NPROCESSORS_ONLN) OPTFLAGS="" install && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/vector.control
 
 #########################################################################################
@@ -381,16 +376,15 @@ RUN wget https://github.com/pgvector/pgvector/archive/refs/tags/v0.8.0.tar.gz -O
 # compile pgjwt extension
 #
 #########################################################################################
-FROM build-deps AS pgjwt-pg-build
+FROM pg-build 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 PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
+    make -j $(getconf _NPROCESSORS_ONLN) install && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/pgjwt.control
 
 #########################################################################################
@@ -399,17 +393,16 @@ RUN wget https://github.com/michelp/pgjwt/archive/f3d82fd30151e754e19ce5d6a06c71
 # compile hypopg extension
 #
 #########################################################################################
-FROM build-deps AS hypopg-pg-build
+FROM pg-build 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) PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
-    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
+    make -j $(getconf _NPROCESSORS_ONLN) && \
+    make -j $(getconf _NPROCESSORS_ONLN) install && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/hypopg.control
 
 #########################################################################################
@@ -418,17 +411,16 @@ RUN wget https://github.com/HypoPG/hypopg/archive/refs/tags/1.4.1.tar.gz -O hypo
 # compile pg_hashids extension
 #
 #########################################################################################
-FROM build-deps AS pg-hashids-pg-build
+FROM pg-build 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) 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 && \
+    make -j $(getconf _NPROCESSORS_ONLN) USE_PGXS=1 && \
+    make -j $(getconf _NPROCESSORS_ONLN) install USE_PGXS=1 && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/pg_hashids.control
 
 #########################################################################################
@@ -437,9 +429,8 @@ RUN wget https://github.com/iCyberon/pg_hashids/archive/refs/tags/v1.2.1.tar.gz
 # compile rum extension
 #
 #########################################################################################
-FROM build-deps AS rum-pg-build
+FROM pg-build AS rum-pg-build
 ARG PG_VERSION
-COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 COPY compute/patches/rum.patch /rum.patch
 
@@ -450,8 +441,8 @@ 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) 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 && \
+    make -j $(getconf _NPROCESSORS_ONLN) USE_PGXS=1 && \
+    make -j $(getconf _NPROCESSORS_ONLN) install USE_PGXS=1 && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/rum.control
 
 #########################################################################################
@@ -460,17 +451,16 @@ RUN wget https://github.com/postgrespro/rum/archive/cb1edffc57736cd2a4455f8d0fea
 # compile pgTAP extension
 #
 #########################################################################################
-FROM build-deps AS pgtap-pg-build
+FROM pg-build 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) PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
-    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
+    make -j $(getconf _NPROCESSORS_ONLN) && \
+    make -j $(getconf _NPROCESSORS_ONLN) install && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/pgtap.control
 
 #########################################################################################
@@ -479,17 +469,16 @@ RUN wget https://github.com/theory/pgtap/archive/refs/tags/v1.3.3.tar.gz -O pgta
 # compile ip4r extension
 #
 #########################################################################################
-FROM build-deps AS ip4r-pg-build
+FROM pg-build 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) PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
-    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
+    make -j $(getconf _NPROCESSORS_ONLN) && \
+    make -j $(getconf _NPROCESSORS_ONLN) install && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/ip4r.control
 
 #########################################################################################
@@ -498,17 +487,16 @@ RUN wget https://github.com/RhodiumToad/ip4r/archive/refs/tags/2.4.2.tar.gz -O i
 # compile Prefix extension
 #
 #########################################################################################
-FROM build-deps AS prefix-pg-build
+FROM pg-build 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) PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
-    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
+    make -j $(getconf _NPROCESSORS_ONLN) && \
+    make -j $(getconf _NPROCESSORS_ONLN) install && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/prefix.control
 
 #########################################################################################
@@ -517,17 +505,16 @@ RUN wget https://github.com/dimitri/prefix/archive/refs/tags/v1.2.10.tar.gz -O p
 # compile hll extension
 #
 #########################################################################################
-FROM build-deps AS hll-pg-build
+FROM pg-build 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) PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
-    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
+    make -j $(getconf _NPROCESSORS_ONLN) && \
+    make -j $(getconf _NPROCESSORS_ONLN) install && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/hll.control
 
 #########################################################################################
@@ -536,17 +523,16 @@ RUN wget https://github.com/citusdata/postgresql-hll/archive/refs/tags/v2.18.tar
 # compile plpgsql_check extension
 #
 #########################################################################################
-FROM build-deps AS plpgsql-check-pg-build
+FROM pg-build 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) 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 && \
+    make -j $(getconf _NPROCESSORS_ONLN) USE_PGXS=1 && \
+    make -j $(getconf _NPROCESSORS_ONLN) install USE_PGXS=1 && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/plpgsql_check.control
 
 #########################################################################################
@@ -555,11 +541,8 @@ RUN wget https://github.com/okbob/plpgsql_check/archive/refs/tags/v2.7.11.tar.gz
 # compile timescaledb extension
 #
 #########################################################################################
-FROM build-deps AS timescaledb-pg-build
-COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
-
+FROM pg-build AS timescaledb-pg-build
 ARG PG_VERSION
-ENV PATH="/usr/local/pgsql/bin:$PATH"
 
 RUN case "${PG_VERSION}" in \
       "v14" | "v15") \
@@ -590,11 +573,8 @@ RUN case "${PG_VERSION}" in \
 # compile pg_hint_plan extension
 #
 #########################################################################################
-FROM build-deps AS pg-hint-plan-pg-build
-COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
-
+FROM pg-build AS pg-hint-plan-pg-build
 ARG PG_VERSION
-ENV PATH="/usr/local/pgsql/bin:$PATH"
 
 # version-specific, has separate releases for each version
 RUN case "${PG_VERSION}" in \
@@ -632,14 +612,12 @@ RUN case "${PG_VERSION}" in \
 # compile pg_cron extension
 #
 #########################################################################################
-FROM build-deps AS pg-cron-pg-build
+FROM pg-build 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 . && \
@@ -653,9 +631,8 @@ RUN wget https://github.com/citusdata/pg_cron/archive/refs/tags/v1.6.4.tar.gz -O
 # compile rdkit extension
 #
 #########################################################################################
-FROM build-deps AS rdkit-pg-build
+FROM pg-build 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 \
@@ -673,7 +650,13 @@ 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
-ENV PATH="/usr/local/pgsql/bin/:/usr/local/pgsql/:$PATH"
+
+# XXX: /usr/local/pgsql/bin is already in PATH, and that should be enough to find
+# pg_config. For some reason the rdkit cmake script doesn't work with just that,
+# however. By also adding /usr/local/pgsql, it works, which is weird because there
+# are no executables in that directory.
+ENV PATH="/usr/local/pgsql:$PATH"
+
 RUN case "${PG_VERSION}" in \
     "v17") \
         export RDKIT_VERSION=Release_2024_09_1 \
@@ -726,13 +709,11 @@ RUN case "${PG_VERSION}" in \
 # compile pg_uuidv7 extension
 #
 #########################################################################################
-FROM build-deps AS pg-uuidv7-pg-build
+FROM pg-build 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 . && \
@@ -746,13 +727,11 @@ RUN wget https://github.com/fboulnois/pg_uuidv7/archive/refs/tags/v1.6.0.tar.gz
 # compile pg_roaringbitmap extension
 #
 #########################################################################################
-FROM build-deps AS pg-roaringbitmap-pg-build
+FROM pg-build 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 . && \
@@ -766,16 +745,14 @@ RUN wget https://github.com/ChenHuajun/pg_roaringbitmap/archive/refs/tags/v0.5.4
 # compile pg_semver extension
 #
 #########################################################################################
-FROM build-deps AS pg-semver-pg-build
+FROM pg-build 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 \
@@ -802,13 +779,11 @@ RUN case "${PG_VERSION}" in \
 # compile pg_embedding extension
 #
 #########################################################################################
-FROM build-deps AS pg-embedding-pg-build
-COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+FROM pg-build AS pg-embedding-pg-build
 
 # 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 \
@@ -829,26 +804,19 @@ RUN case "${PG_VERSION}" in \
 # compile anon extension
 #
 #########################################################################################
-FROM build-deps AS pg-anon-pg-build
+FROM pg-build 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 . && \
-    find /usr/local/pgsql -type f | sed 's|^/usr/local/pgsql/||' > /before.txt &&\
-    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
-    echo 'trusted = true' >> /usr/local/pgsql/share/extension/anon.control && \
-    find /usr/local/pgsql -type f | sed 's|^/usr/local/pgsql/||' > /after.txt &&\
-    mkdir -p /extensions/anon && cp /usr/local/pgsql/share/extension/anon.control /extensions/anon && \
-    sort -o /before.txt /before.txt && sort -o /after.txt /after.txt && \
-    comm -13 /before.txt /after.txt | tar --directory=/usr/local/pgsql --zstd -cf /extensions/anon.tar.zst -T -
+    make -j $(getconf _NPROCESSORS_ONLN) install && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/anon.control
 
 #########################################################################################
 #
@@ -856,9 +824,8 @@ RUN case "${PG_VERSION}" in "v17") \
 # This layer is used to build `pgrx` deps
 #
 #########################################################################################
-FROM build-deps AS rust-extensions-build
+FROM pg-build 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 && \
@@ -866,7 +833,7 @@ RUN apt update && \
     useradd -ms /bin/bash nonroot -b /home
 
 ENV HOME=/home/nonroot
-ENV PATH="/home/nonroot/.cargo/bin:/usr/local/pgsql/bin/:$PATH"
+ENV PATH="/home/nonroot/.cargo/bin:$PATH"
 USER nonroot
 WORKDIR /home/nonroot
 
@@ -893,9 +860,8 @@ USER root
 # and eventually get merged with `rust-extensions-build`
 #
 #########################################################################################
-FROM build-deps AS rust-extensions-build-pgrx12
+FROM pg-build 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 && \
@@ -903,7 +869,7 @@ RUN apt update && \
     useradd -ms /bin/bash nonroot -b /home
 
 ENV HOME=/home/nonroot
-ENV PATH="/home/nonroot/.cargo/bin:/usr/local/pgsql/bin/:$PATH"
+ENV PATH="/home/nonroot/.cargo/bin:$PATH"
 USER nonroot
 WORKDIR /home/nonroot
 
@@ -911,7 +877,7 @@ RUN curl -sSO https://static.rust-lang.org/rustup/dist/$(uname -m)-unknown-linux
     chmod +x rustup-init && \
     ./rustup-init -y --no-modify-path --profile minimal --default-toolchain stable && \
     rm rustup-init && \
-    cargo install --locked --version 0.12.6 cargo-pgrx && \
+    cargo install --locked --version 0.12.9 cargo-pgrx && \
     /bin/bash -c 'cargo pgrx init --pg${PG_VERSION:1}=/usr/local/pgsql/bin/pg_config'
 
 USER root
@@ -948,19 +914,19 @@ 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 = "0.12.6", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    sed -i 's/pgrx = "0.12.6"/pgrx = { version = "0.12.9", 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 = "0.12.6", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    sed -i 's/pgrx = "0.12.6"/pgrx = { version = "0.12.9", 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 = "0.12.6", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    sed -i 's/pgrx = "0.12.6"/pgrx = { version = "0.12.9", 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 && \
@@ -985,7 +951,8 @@ 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 = "0.12.6", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    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 && \
     cargo pgrx install --release && \
     echo "trusted = true" >> /usr/local/pgsql/share/extension/pg_jsonschema.control
 
@@ -1003,7 +970,8 @@ 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 = "0.12.6", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    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 && \
     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 && \
@@ -1024,33 +992,58 @@ 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 . && \
-    # 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 && \
+    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 && \
     cargo pgrx install --release && \
     echo "trusted = true" >> /usr/local/pgsql/share/extension/pg_tiktoken.control
 
 #########################################################################################
 #
 # Layer "pg-pgx-ulid-build"
-# Compile "pgx_ulid" extension
+# Compile "pgx_ulid" extension for v16 and below
 #
 #########################################################################################
 
 FROM rust-extensions-build AS pg-pgx-ulid-build
 ARG PG_VERSION
 
-# doesn't support v17 yet
-# https://github.com/pksunkara/pgx_ulid/pull/52
-RUN case "${PG_VERSION}" in "v17") \
-    echo "pgx_ulid does not support pg17 as of the latest version (0.1.5)" && exit 0;; \
+RUN case "${PG_VERSION}" in \
+    "v14" | "v15" | "v16") \
+        ;; \
+    *) \
+        echo "skipping the version of pgx_ulid for $PG_VERSION" && exit 0 \
+        ;; \
     esac && \
     wget https://github.com/pksunkara/pgx_ulid/archive/refs/tags/v0.1.5.tar.gz -O pgx_ulid.tar.gz && \
-    echo "9d1659a2da65af0133d5451c454de31b37364e3502087dadf579f790bc8bef17 pgx_ulid.tar.gz" | sha256sum --check && \
+    echo "9d1659a2da65af0133d5451c454de31b37364e3502087dadf579f790bc8bef17  pgx_ulid.tar.gz" | sha256sum --check && \
     mkdir pgx_ulid-src && cd pgx_ulid-src && tar xzf ../pgx_ulid.tar.gz --strip-components=1 -C . && \
-    sed -i 's/pgrx       = "^0.11.2"/pgrx = { version = "=0.11.3", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    sed -i 's/pgrx       = "^0.11.2"/pgrx       = { version = "0.11.3", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
     cargo pgrx install --release && \
-    echo "trusted = true" >> /usr/local/pgsql/share/extension/ulid.control
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/ulid.control
+
+#########################################################################################
+#
+# Layer "pg-pgx-ulid-pgrx12-build"
+# Compile "pgx_ulid" extension for v17 and up
+#
+#########################################################################################
+
+FROM rust-extensions-build-pgrx12 AS pg-pgx-ulid-pgrx12-build
+ARG PG_VERSION
+
+RUN case "${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
 
 #########################################################################################
 #
@@ -1065,10 +1058,14 @@ ARG PG_VERSION
 # NOTE: local_proxy depends on the version of pg_session_jwt
 # Do not update without approve from proxy team
 # Make sure the version is reflected in proxy/src/serverless/local_conn_pool.rs
-RUN wget https://github.com/neondatabase/pg_session_jwt/archive/refs/tags/v0.1.2-v17.tar.gz -O pg_session_jwt.tar.gz && \
-    echo "c8ecbed9cb8c6441bce5134a176002b043018adf9d05a08e457dda233090a86e pg_session_jwt.tar.gz" | sha256sum --check && \
+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 = "=0.12.6", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    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 && \
     cargo pgrx install --release
 
 #########################################################################################
@@ -1078,13 +1075,11 @@ RUN wget https://github.com/neondatabase/pg_session_jwt/archive/refs/tags/v0.1.2
 #
 #########################################################################################
 
-FROM build-deps AS wal2json-pg-build
+FROM pg-build 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 . && \
@@ -1097,13 +1092,11 @@ RUN wget https://github.com/eulerto/wal2json/archive/refs/tags/wal2json_2_6.tar.
 # compile pg_ivm extension
 #
 #########################################################################################
-FROM build-deps AS pg-ivm-build
+FROM pg-build 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 . && \
@@ -1117,13 +1110,11 @@ RUN wget https://github.com/sraoss/pg_ivm/archive/refs/tags/v1.9.tar.gz -O pg_iv
 # compile pg_partman extension
 #
 #########################################################################################
-FROM build-deps AS pg-partman-build
+FROM pg-build 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/:$PATH"
 RUN wget https://github.com/pgpartman/pg_partman/archive/refs/tags/v5.1.0.tar.gz -O pg_partman.tar.gz && \
     echo "3e3a27d7ff827295d5c55ef72f07a49062d6204b3cb0b9a048645d6db9f3cb9f pg_partman.tar.gz" | sha256sum --check && \
     mkdir pg_partman-src && cd pg_partman-src && tar xzf ../pg_partman.tar.gz --strip-components=1 -C . && \
@@ -1139,9 +1130,6 @@ RUN wget https://github.com/pgpartman/pg_partman/archive/refs/tags/v5.1.0.tar.gz
 #########################################################################################
 FROM rust-extensions-build AS pg-mooncake-build
 ARG PG_VERSION
-COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
-
-ENV PATH="/usr/local/pgsql/bin/:$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 && \
@@ -1157,11 +1145,8 @@ RUN wget https://github.com/Mooncake-Labs/pg_mooncake/releases/download/v0.1.0/p
 #
 #########################################################################################
 
-FROM build-deps AS pg-repack-build
+FROM pg-build 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 && \
@@ -1203,6 +1188,7 @@ 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/
@@ -1280,15 +1266,16 @@ RUN set -e \
 
 #########################################################################################
 #
-# Layers "postgres-exporter" and "sql-exporter"
+# Layers "postgres-exporter", "pgbouncer-exporter", and "sql-exporter"
 #
 #########################################################################################
 
-FROM quay.io/prometheuscommunity/postgres-exporter:v0.12.1 AS postgres-exporter
+FROM quay.io/prometheuscommunity/postgres-exporter:v0.16.0 AS postgres-exporter
+FROM quay.io/prometheuscommunity/pgbouncer-exporter:v0.10.2 AS pgbouncer-exporter
 
 # Keep the version the same as in build-tools.Dockerfile and
 # test_runner/regress/test_compute_metrics.py.
-FROM burningalchemist/sql_exporter:0.16.0 AS sql-exporter
+FROM burningalchemist/sql_exporter:0.17.0 AS sql-exporter
 
 #########################################################################################
 #
@@ -1366,9 +1353,6 @@ COPY --from=pg-roaringbitmap-pg-build /pg_roaringbitmap.tar.gz /ext-src
 COPY --from=pg-semver-pg-build /pg_semver.tar.gz /ext-src
 #COPY --from=pg-embedding-pg-build /home/nonroot/pg_embedding-src/ /ext-src
 #COPY --from=wal2json-pg-build /wal2json_2_5.tar.gz /ext-src
-#pg_anon is not supported yet for pg v17 so, don't fail if nothing found
-COPY --from=pg-anon-pg-build /pg_anon.tar.g? /ext-src
-COPY compute/patches/pg_anon.patch /ext-src
 COPY --from=pg-ivm-build /pg_ivm.tar.gz /ext-src
 COPY --from=pg-partman-build /pg_partman.tar.gz /ext-src
 RUN cd /ext-src/ && for f in *.tar.gz; \
@@ -1379,9 +1363,6 @@ RUN cd /ext-src/rum-src && patch -p1 <../rum.patch
 RUN cd /ext-src/pgvector-src && patch -p1 <../pgvector.patch
 RUN cd /ext-src/pg_hint_plan-src && patch -p1 < /ext-src/pg_hint_plan_${PG_VERSION}.patch
 COPY --chmod=755 docker-compose/run-tests.sh /run-tests.sh
-RUN case "${PG_VERSION}" in "v17") \
-    echo "postgresql_anonymizer does not yet support PG17" && exit 0;; \
-    esac && patch -p1 </ext-src/pg_anon.patch
 RUN patch -p1 </ext-src/pg_cron.patch
 ENV PATH=/usr/local/pgsql/bin:$PATH
 ENV PGHOST=compute
@@ -1422,6 +1403,7 @@ RUN mkdir -p /etc/local_proxy && chown postgres:postgres /etc/local_proxy
 
 # Metrics exporter binaries and  configuration files
 COPY --from=postgres-exporter /bin/postgres_exporter /bin/postgres_exporter
+COPY --from=pgbouncer-exporter /bin/pgbouncer_exporter /bin/pgbouncer_exporter
 COPY --from=sql-exporter      /bin/sql_exporter      /bin/sql_exporter
 
 COPY --chown=postgres compute/etc/postgres_exporter.yml /etc/postgres_exporter.yml

compute/etc/pgbouncer.ini

@@ -19,6 +19,8 @@ max_prepared_statements=0
 admin_users=postgres
 unix_socket_dir=/tmp/
 unix_socket_mode=0777
+; required for pgbouncer_exporter
+ignore_startup_parameters=extra_float_digits
 
 ;; Disable connection logging. It produces a lot of logs that no one looks at,
 ;; and we can get similar log entries from the proxy too. We had incidents in

compute/patches/pgvector.patch

@@ -1,8 +1,24 @@
+diff --git a/Makefile b/Makefile
+index 7a4b88c..56678af 100644
+--- a/Makefile
++++ b/Makefile
+@@ -3,7 +3,10 @@ EXTVERSION = 0.8.0
+ 
+ MODULE_big = vector
+ DATA = $(wildcard sql/*--*--*.sql)
+-DATA_built = sql/$(EXTENSION)--$(EXTVERSION).sql
++# This change is needed to install different per-version SQL files
++# like pgvector--0.8.0.sql and pgvector--0.7.4.sql
++# The corresponding file is downloaded during the Docker image build process
++DATA_built = sql/$(EXTENSION)--$(EXTVERSION).sql sql/vector--0.7.4.sql
+ OBJS = src/bitutils.o src/bitvec.o src/halfutils.o src/halfvec.o src/hnsw.o src/hnswbuild.o src/hnswinsert.o src/hnswscan.o src/hnswutils.o src/hnswvacuum.o src/ivfbuild.o src/ivfflat.o src/ivfinsert.o src/ivfkmeans.o src/ivfscan.o src/ivfutils.o src/ivfvacuum.o src/sparsevec.o src/vector.o
+ HEADERS = src/halfvec.h src/sparsevec.h src/vector.h
+ 
 diff --git a/src/hnswbuild.c b/src/hnswbuild.c
-index dcfb2bd..d5189ee 100644
+index b667478..fc1897c 100644
 --- a/src/hnswbuild.c
 +++ b/src/hnswbuild.c
-@@ -860,9 +860,17 @@ HnswParallelBuildMain(dsm_segment *seg, shm_toc *toc)
+@@ -843,9 +843,17 @@ HnswParallelBuildMain(dsm_segment *seg, shm_toc *toc)
  
  	hnswarea = shm_toc_lookup(toc, PARALLEL_KEY_HNSW_AREA, false);
  
@@ -20,7 +36,7 @@ index dcfb2bd..d5189ee 100644
  	/* Close relations within worker */
  	index_close(indexRel, indexLockmode);
  	table_close(heapRel, heapLockmode);
-@@ -1117,12 +1125,38 @@ BuildIndex(Relation heap, Relation index, IndexInfo *indexInfo,
+@@ -1100,12 +1108,38 @@ BuildIndex(Relation heap, Relation index, IndexInfo *indexInfo,
  	SeedRandom(42);
  #endif
  

compute/vm-image-spec-bookworm.yaml

@@ -27,6 +27,10 @@ commands:
     user: nobody
     sysvInitAction: respawn
     shell: 'DATA_SOURCE_NAME="user=cloud_admin sslmode=disable dbname=postgres application_name=postgres-exporter" /bin/postgres_exporter --config.file=/etc/postgres_exporter.yml'
+  - name: pgbouncer-exporter
+    user: postgres
+    sysvInitAction: respawn
+    shell: '/bin/pgbouncer_exporter --pgBouncer.connectionString="postgres:///pgbouncer?host=/tmp&port=6432&dbname=pgbouncer&user=pgbouncer"'
   - name: sql-exporter
     user: nobody
     sysvInitAction: respawn

compute/vm-image-spec-bullseye.yaml

@@ -27,6 +27,10 @@ commands:
     user: nobody
     sysvInitAction: respawn
     shell: 'DATA_SOURCE_NAME="user=cloud_admin sslmode=disable dbname=postgres application_name=postgres-exporter" /bin/postgres_exporter --config.file=/etc/postgres_exporter.yml'
+  - name: pgbouncer-exporter
+    user: postgres
+    sysvInitAction: respawn
+    shell: '/bin/pgbouncer_exporter --pgBouncer.connectionString="postgres:///pgbouncer?host=/tmp&port=6432&dbname=pgbouncer&user=pgbouncer"'
   - name: sql-exporter
     user: nobody
     sysvInitAction: respawn

compute_tools/Cargo.toml

@@ -51,6 +51,7 @@ 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::{info, info_span, warn, Instrument};
+use tracing::{error, info, info_span, warn, Instrument};
 use utils::fs_ext::is_directory_empty;
 
 #[path = "fast_import/aws_s3_sync.rs"]
@@ -41,16 +41,25 @@ 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: s3_uri::S3Uri,
+    s3_prefix: Option<s3_uri::S3Uri>,
+    #[clap(long)]
+    source_connection_string: Option<String>,
+    #[clap(short, long)]
+    interactive: bool,
     #[clap(long)]
     pg_bin_dir: Utf8PathBuf,
     #[clap(long)]
     pg_lib_dir: Utf8PathBuf,
+    #[clap(long)]
+    pg_port: Option<u16>, // port to run postgres on, 5432 is default
 }
 
 #[serde_with::serde_as]
@@ -67,6 +76,13 @@ enum EncryptionSecret {
     KMS { key_id: String },
 }
 
+// copied from pageserver_api::config::defaults::DEFAULT_LOCALE to avoid dependency just for a constant
+const DEFAULT_LOCALE: &str = if cfg!(target_os = "macos") {
+    "C"
+} else {
+    "C.UTF-8"
+};
+
 #[tokio::main]
 pub(crate) async fn main() -> anyhow::Result<()> {
     utils::logging::init(
@@ -77,30 +93,74 @@ pub(crate) async fn main() -> anyhow::Result<()> {
 
     info!("starting");
 
-    let Args {
-        working_directory,
-        s3_prefix,
-        pg_bin_dir,
-        pg_lib_dir,
-    } = Args::parse();
+    let args = Args::parse();
 
-    let aws_config = aws_config::load_defaults(BehaviorVersion::v2024_03_28()).await;
+    // 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 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")?
+    let working_directory = args.working_directory;
+    let pg_bin_dir = args.pg_bin_dir;
+    let pg_lib_dir = args.pg_lib_dir;
+    let pg_port = args.pg_port.unwrap_or_else(|| {
+        info!("pg_port not specified, using default 5432");
+        5432
+    });
+
+    // 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()
     };
 
     match tokio::fs::create_dir(&working_directory).await {
@@ -123,15 +183,6 @@ 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,
@@ -142,7 +193,7 @@ pub(crate) async fn main() -> anyhow::Result<()> {
     let superuser = "cloud_admin"; // XXX: this shouldn't be hard-coded
     postgres_initdb::do_run_initdb(postgres_initdb::RunInitdbArgs {
         superuser,
-        locale: "en_US.UTF-8", // XXX: this shouldn't be hard-coded,
+        locale: DEFAULT_LOCALE, // XXX: this shouldn't be hard-coded,
         pg_version,
         initdb_bin: pg_bin_dir.join("initdb").as_ref(),
         library_search_path: &pg_lib_dir, // TODO: is this right? Prob works in compute image, not sure about neon_local.
@@ -159,6 +210,7 @@ pub(crate) async fn main() -> anyhow::Result<()> {
     let mut postgres_proc = tokio::process::Command::new(pgbin)
         .arg("-D")
         .arg(&pgdata_dir)
+        .args(["-p", &format!("{pg_port}")])
         .args(["-c", "wal_level=minimal"])
         .args(["-c", "shared_buffers=10GB"])
         .args(["-c", "max_wal_senders=0"])
@@ -170,8 +222,15 @@ 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", "effective_io_concurrency=100"])
+        .args([
+            "-c",
+            &format!(
+                "effective_io_concurrency={}",
+                if cfg!(target_os = "macos") { 0 } else { 100 }
+            ),
+        ])
         .env_clear()
+        .env("LD_LIBRARY_PATH", &pg_lib_dir)
         .stdout(std::process::Stdio::piped())
         .stderr(std::process::Stdio::piped())
         .spawn()
@@ -185,44 +244,58 @@ 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");
+        format!("host=localhost port={pg_port} user={superuser} dbname=postgres");
+
+    let start_time = std::time::Instant::now();
+
     loop {
-        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;
+        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;
+            }
         }
     }
 
-    //
-    // 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 restore_pg_connstring = restore_pg_connstring.replace("dbname=postgres", "dbname=neondb");
 
     let dumpdir = working_directory.join("dumpdir");
 
@@ -256,6 +329,7 @@ pub(crate) async fn main() -> anyhow::Result<()> {
             .arg(&source_connection_string)
             // how we run it
             .env_clear()
+            .env("LD_LIBRARY_PATH", &pg_lib_dir)
             .kill_on_drop(true)
             .stdout(std::process::Stdio::piped())
             .stderr(std::process::Stdio::piped())
@@ -289,6 +363,7 @@ pub(crate) async fn main() -> anyhow::Result<()> {
             .arg(&dumpdir)
             // how we run it
             .env_clear()
+            .env("LD_LIBRARY_PATH", &pg_lib_dir)
             .kill_on_drop(true)
             .stdout(std::process::Stdio::piped())
             .stderr(std::process::Stdio::piped())
@@ -310,6 +385,12 @@ 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(
@@ -325,21 +406,24 @@ pub(crate) async fn main() -> anyhow::Result<()> {
             .context("wait for postgres to shut down")?;
     }
 
-    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/"))
+    // 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 status directory to destination")?;
+            .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")?;
+        }
     }
 
     Ok(())

compute_tools/src/compute.rs

@@ -41,14 +41,14 @@ use crate::local_proxy;
 use crate::pg_helpers::*;
 use crate::spec::*;
 use crate::spec_apply::ApplySpecPhase::{
-    CreateAndAlterDatabases, CreateAndAlterRoles, CreateAvailabilityCheck, CreateSuperUser,
-    DropInvalidDatabases, DropRoles, HandleNeonExtension, HandleOtherExtensions,
-    RenameAndDeleteDatabases, RenameRoles, RunInEachDatabase,
+    CreateAndAlterDatabases, CreateAndAlterRoles, CreateAvailabilityCheck, CreateSchemaNeon,
+    CreateSuperUser, DropInvalidDatabases, DropRoles, FinalizeDropLogicalSubscriptions,
+    HandleNeonExtension, HandleOtherExtensions, RenameAndDeleteDatabases, RenameRoles,
+    RunInEachDatabase,
 };
 use crate::spec_apply::PerDatabasePhase;
 use crate::spec_apply::PerDatabasePhase::{
-    ChangeSchemaPerms, DeleteDBRoleReferences, DropSubscriptionsForDeletedDatabases,
-    HandleAnonExtension,
+    ChangeSchemaPerms, DeleteDBRoleReferences, DropLogicalSubscriptions, HandleAnonExtension,
 };
 use crate::spec_apply::{apply_operations, MutableApplyContext, DB};
 use crate::sync_sk::{check_if_synced, ping_safekeeper};
@@ -340,6 +340,15 @@ impl ComputeNode {
         self.state.lock().unwrap().status
     }
 
+    pub fn get_timeline_id(&self) -> Option<TimelineId> {
+        self.state
+            .lock()
+            .unwrap()
+            .pspec
+            .as_ref()
+            .map(|s| s.timeline_id)
+    }
+
     // Remove `pgdata` directory and create it again with right permissions.
     fn create_pgdata(&self) -> Result<()> {
         // Ignore removal error, likely it is a 'No such file or directory (os error 2)'.
@@ -929,6 +938,48 @@ impl ComputeNode {
                 .map(|role| (role.name.clone(), role))
                 .collect::<HashMap<String, Role>>();
 
+            // Check if we need to drop subscriptions before starting the endpoint.
+            //
+            // It is important to do this operation exactly once when endpoint starts on a new branch.
+            // Otherwise, we may drop not inherited, but newly created subscriptions.
+            //
+            // We cannot rely only on spec.drop_subscriptions_before_start flag,
+            // because if for some reason compute restarts inside VM,
+            // it will start again with the same spec and flag value.
+            //
+            // To handle this, we save the fact of the operation in the database
+            // in the neon.drop_subscriptions_done table.
+            // If the table does not exist, we assume that the operation was never performed, so we must do it.
+            // If table exists, we check if the operation was performed on the current timelilne.
+            //
+            let mut drop_subscriptions_done = false;
+
+            if spec.drop_subscriptions_before_start {
+                let timeline_id = self.get_timeline_id().context("timeline_id must be set")?;
+                let query = format!("select 1 from neon.drop_subscriptions_done where timeline_id = '{}'", timeline_id);
+
+                info!("Checking if drop subscription operation was already performed for timeline_id: {}", timeline_id);
+
+                drop_subscriptions_done =  match
+                    client.simple_query(&query).await {
+                    Ok(result) => {
+                        matches!(&result[0], postgres::SimpleQueryMessage::Row(_))
+                    },
+                    Err(e) =>
+                    {
+                        match e.code() {
+                            Some(&SqlState::UNDEFINED_TABLE) => false,
+                            _ => {
+                                // We don't expect any other error here, except for the schema/table not existing
+                                error!("Error checking if drop subscription operation was already performed: {}", e);
+                                return Err(e.into());
+                            }
+                        }
+                    }
+                }
+            };
+
+
             let jwks_roles = Arc::new(
                 spec.as_ref()
                     .local_proxy_config
@@ -996,7 +1047,7 @@ impl ComputeNode {
                         jwks_roles.clone(),
                         concurrency_token.clone(),
                         db,
-                        [DropSubscriptionsForDeletedDatabases].to_vec(),
+                        [DropLogicalSubscriptions].to_vec(),
                     );
 
                     Ok(spawn(fut))
@@ -1024,6 +1075,7 @@ impl ComputeNode {
                 CreateAndAlterRoles,
                 RenameAndDeleteDatabases,
                 CreateAndAlterDatabases,
+                CreateSchemaNeon,
             ] {
                 info!("Applying phase {:?}", &phase);
                 apply_operations(
@@ -1064,6 +1116,17 @@ impl ComputeNode {
                     }
 
                     let conf = Arc::new(conf);
+                    let mut phases = vec![
+                        DeleteDBRoleReferences,
+                        ChangeSchemaPerms,
+                        HandleAnonExtension,
+                    ];
+
+                    if spec.drop_subscriptions_before_start && !drop_subscriptions_done {
+                        info!("Adding DropLogicalSubscriptions phase because drop_subscriptions_before_start is set");
+                        phases.push(DropLogicalSubscriptions);
+                    }
+
                     let fut = Self::apply_spec_sql_db(
                         spec.clone(),
                         conf,
@@ -1071,12 +1134,7 @@ impl ComputeNode {
                         jwks_roles.clone(),
                         concurrency_token.clone(),
                         db,
-                        [
-                            DeleteDBRoleReferences,
-                            ChangeSchemaPerms,
-                            HandleAnonExtension,
-                        ]
-                        .to_vec(),
+                        phases,
                     );
 
                     Ok(spawn(fut))
@@ -1088,12 +1146,20 @@ impl ComputeNode {
                 handle.await??;
             }
 
-            for phase in vec![
+            let mut phases = vec![
                 HandleOtherExtensions,
-                HandleNeonExtension,
+                HandleNeonExtension, // This step depends on CreateSchemaNeon
                 CreateAvailabilityCheck,
                 DropRoles,
-            ] {
+            ];
+
+            // This step depends on CreateSchemaNeon
+            if spec.drop_subscriptions_before_start && !drop_subscriptions_done {
+                info!("Adding FinalizeDropLogicalSubscriptions phase because drop_subscriptions_before_start is set");
+                phases.push(FinalizeDropLogicalSubscriptions);
+            }
+
+            for phase in phases {
                 debug!("Applying phase {:?}", &phase);
                 apply_operations(
                     spec.clone(),
@@ -1463,6 +1529,14 @@ impl ComputeNode {
                         Ok(())
                     },
                 )?;
+
+                let postgresql_conf_path = pgdata_path.join("postgresql.conf");
+                if config::line_in_file(
+                    &postgresql_conf_path,
+                    "neon.disable_logical_replication_subscribers=false",
+                )? {
+                    info!("updated postgresql.conf to set neon.disable_logical_replication_subscribers=false");
+                }
                 self.pg_reload_conf()?;
             }
             self.post_apply_config()?;

compute_tools/src/config.rs

@@ -129,6 +129,13 @@ pub fn write_postgres_conf(
 
     writeln!(file, "neon.extension_server_port={}", extension_server_port)?;
 
+    if spec.drop_subscriptions_before_start {
+        writeln!(file, "neon.disable_logical_replication_subscribers=true")?;
+    } else {
+        // be explicit about the default value
+        writeln!(file, "neon.disable_logical_replication_subscribers=false")?;
+    }
+
     // This is essential to keep this line at the end of the file,
     // because it is intended to override any settings above.
     writeln!(file, "include_if_exists = 'compute_ctl_temp_override.conf'")?;

compute_tools/src/http/routes/extension_server.rs

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

compute_tools/src/http/server.rs

@@ -1,15 +1,14 @@
 use std::{
     net::{IpAddr, Ipv6Addr, SocketAddr},
-    sync::{
-        atomic::{AtomicU64, Ordering},
-        Arc,
-    },
+    sync::Arc,
     thread,
     time::Duration,
 };
 
 use anyhow::Result;
 use axum::{
+    extract::Request,
+    middleware::{self, Next},
     response::{IntoResponse, Response},
     routing::{get, post},
     Router,
@@ -17,11 +16,9 @@ use axum::{
 use http::StatusCode;
 use tokio::net::TcpListener;
 use tower::ServiceBuilder;
-use tower_http::{
-    request_id::{MakeRequestId, PropagateRequestIdLayer, RequestId, SetRequestIdLayer},
-    trace::TraceLayer,
-};
+use tower_http::{request_id::PropagateRequestIdLayer, 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,
@@ -34,30 +31,24 @@ async fn handle_404() -> Response {
     StatusCode::NOT_FOUND.into_response()
 }
 
-#[derive(Clone, Default)]
-struct ComputeMakeRequestId(Arc<AtomicU64>);
+const X_REQUEST_ID: &str = "x-request-id";
 
-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();
+/// 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();
 
-        Some(RequestId::new(request_id))
+    if headers.get(X_REQUEST_ID).is_none() {
+        headers.append(X_REQUEST_ID, Uuid::new_v4().to_string().parse().unwrap());
     }
+
+    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))
@@ -82,9 +73,8 @@ async fn serve(port: u16, compute: Arc<ComputeNode>) {
         .fallback(handle_404)
         .layer(
             ServiceBuilder::new()
-                .layer(SetRequestIdLayer::x_request_id(
-                    ComputeMakeRequestId::default(),
-                ))
+                // Add this middleware since we assume the request ID exists
+                .layer(middleware::from_fn(maybe_add_request_id_header))
                 .layer(
                     TraceLayer::new_for_http()
                         .on_request(|request: &http::Request<_>, _span: &Span| {

compute_tools/src/spec_apply.rs

@@ -47,7 +47,7 @@ pub enum PerDatabasePhase {
     DeleteDBRoleReferences,
     ChangeSchemaPerms,
     HandleAnonExtension,
-    DropSubscriptionsForDeletedDatabases,
+    DropLogicalSubscriptions,
 }
 
 #[derive(Clone, Debug)]
@@ -58,11 +58,13 @@ pub enum ApplySpecPhase {
     CreateAndAlterRoles,
     RenameAndDeleteDatabases,
     CreateAndAlterDatabases,
+    CreateSchemaNeon,
     RunInEachDatabase { db: DB, subphase: PerDatabasePhase },
     HandleOtherExtensions,
     HandleNeonExtension,
     CreateAvailabilityCheck,
     DropRoles,
+    FinalizeDropLogicalSubscriptions,
 }
 
 pub struct Operation {
@@ -331,7 +333,7 @@ async fn get_operations<'a>(
                             // NB: there could be other db states, which prevent us from dropping
                             // the database. For example, if db is used by any active subscription
                             // or replication slot.
-                            // Such cases are handled in the DropSubscriptionsForDeletedDatabases
+                            // Such cases are handled in the DropLogicalSubscriptions
                             // phase. We do all the cleanup before actually dropping the database.
                             let drop_db_query: String = format!(
                                 "DROP DATABASE IF EXISTS {} WITH (FORCE)",
@@ -442,13 +444,19 @@ async fn get_operations<'a>(
 
             Ok(Box::new(operations))
         }
+        ApplySpecPhase::CreateSchemaNeon => Ok(Box::new(once(Operation {
+            query: String::from("CREATE SCHEMA IF NOT EXISTS neon"),
+            comment: Some(String::from(
+                "create schema for neon extension and utils tables",
+            )),
+        }))),
         ApplySpecPhase::RunInEachDatabase { db, subphase } => {
             match subphase {
-                PerDatabasePhase::DropSubscriptionsForDeletedDatabases => {
+                PerDatabasePhase::DropLogicalSubscriptions => {
                     match &db {
                         DB::UserDB(db) => {
                             let drop_subscription_query: String = format!(
-                                include_str!("sql/drop_subscription_for_drop_dbs.sql"),
+                                include_str!("sql/drop_subscriptions.sql"),
                                 datname_str = escape_literal(&db.name),
                             );
 
@@ -666,10 +674,6 @@ async fn get_operations<'a>(
         }
         ApplySpecPhase::HandleNeonExtension => {
             let operations = vec![
-                Operation {
-                    query: String::from("CREATE SCHEMA IF NOT EXISTS neon"),
-                    comment: Some(String::from("init: add schema for extension")),
-                },
                 Operation {
                     query: String::from("CREATE EXTENSION IF NOT EXISTS neon WITH SCHEMA neon"),
                     comment: Some(String::from(
@@ -712,5 +716,9 @@ async fn get_operations<'a>(
 
             Ok(Box::new(operations))
         }
+        ApplySpecPhase::FinalizeDropLogicalSubscriptions => Ok(Box::new(once(Operation {
+            query: String::from(include_str!("sql/finalize_drop_subscriptions.sql")),
+            comment: None,
+        }))),
     }
 }

compute_tools/src/sql/finalize_drop_subscriptions.sql

@@ -0,0 +1,21 @@
+DO $$
+BEGIN
+    IF NOT EXISTS(
+        SELECT 1
+        FROM pg_catalog.pg_tables
+        WHERE tablename = 'drop_subscriptions_done'
+        AND schemaname = 'neon'
+    )
+    THEN
+        CREATE TABLE neon.drop_subscriptions_done
+        (id serial primary key, timeline_id text);
+    END IF;
+
+    -- preserve the timeline_id of the last drop_subscriptions run
+    -- to ensure that the cleanup of a timeline is executed only once.
+    -- use upsert to avoid the table bloat in case of cascade branching (branch of a branch)
+    INSERT INTO neon.drop_subscriptions_done VALUES (1, current_setting('neon.timeline_id'))
+    ON CONFLICT (id) DO UPDATE
+    SET timeline_id = current_setting('neon.timeline_id');
+END
+$$

control_plane/src/bin/neon_local.rs

@@ -1357,6 +1357,7 @@ async fn handle_endpoint(subcmd: &EndpointCmd, env: &local_env::LocalEnv) -> Res
                 args.pg_version,
                 mode,
                 !args.update_catalog,
+                false,
             )?;
         }
         EndpointCmd::Start(args) => {

control_plane/src/endpoint.rs

@@ -76,6 +76,7 @@ pub struct EndpointConf {
     http_port: u16,
     pg_version: u32,
     skip_pg_catalog_updates: bool,
+    drop_subscriptions_before_start: bool,
     features: Vec<ComputeFeature>,
 }
 
@@ -143,6 +144,7 @@ impl ComputeControlPlane {
         pg_version: u32,
         mode: ComputeMode,
         skip_pg_catalog_updates: bool,
+        drop_subscriptions_before_start: bool,
     ) -> Result<Arc<Endpoint>> {
         let pg_port = pg_port.unwrap_or_else(|| self.get_port());
         let http_port = http_port.unwrap_or_else(|| self.get_port() + 1);
@@ -162,6 +164,7 @@ impl ComputeControlPlane {
             // with this we basically test a case of waking up an idle compute, where
             // we also skip catalog updates in the cloud.
             skip_pg_catalog_updates,
+            drop_subscriptions_before_start,
             features: vec![],
         });
 
@@ -177,6 +180,7 @@ impl ComputeControlPlane {
                 pg_port,
                 pg_version,
                 skip_pg_catalog_updates,
+                drop_subscriptions_before_start,
                 features: vec![],
             })?,
         )?;
@@ -240,6 +244,7 @@ pub struct Endpoint {
     // Optimizations
     skip_pg_catalog_updates: bool,
 
+    drop_subscriptions_before_start: bool,
     // Feature flags
     features: Vec<ComputeFeature>,
 }
@@ -291,6 +296,7 @@ impl Endpoint {
             tenant_id: conf.tenant_id,
             pg_version: conf.pg_version,
             skip_pg_catalog_updates: conf.skip_pg_catalog_updates,
+            drop_subscriptions_before_start: conf.drop_subscriptions_before_start,
             features: conf.features,
         })
     }
@@ -625,6 +631,7 @@ impl Endpoint {
             shard_stripe_size: Some(shard_stripe_size),
             local_proxy_config: None,
             reconfigure_concurrency: 1,
+            drop_subscriptions_before_start: self.drop_subscriptions_before_start,
         };
         let spec_path = self.endpoint_path().join("spec.json");
         std::fs::write(spec_path, serde_json::to_string_pretty(&spec)?)?;

control_plane/src/local_env.rs

@@ -483,7 +483,6 @@ impl LocalEnv {
             .iter()
             .find(|(mapped_tenant_id, _)| mapped_tenant_id == &tenant_id)
             .map(|&(_, timeline_id)| timeline_id)
-            .map(TimelineId::from)
     }
 
     pub fn timeline_name_mappings(&self) -> HashMap<TenantTimelineId, String> {

control_plane/src/pageserver.rs

@@ -352,6 +352,16 @@ impl PageServerNode {
                 .map(serde_json::from_str)
                 .transpose()
                 .context("Failed to parse 'compaction_algorithm' json")?,
+            l0_flush_delay_threshold: settings
+                .remove("l0_flush_delay_threshold")
+                .map(|x| x.parse::<usize>())
+                .transpose()
+                .context("Failed to parse 'l0_flush_delay_threshold' as an integer")?,
+            l0_flush_stall_threshold: settings
+                .remove("l0_flush_stall_threshold")
+                .map(|x| x.parse::<usize>())
+                .transpose()
+                .context("Failed to parse 'l0_flush_stall_threshold' as an integer")?,
             gc_horizon: settings
                 .remove("gc_horizon")
                 .map(|x| x.parse::<u64>())
@@ -418,6 +428,26 @@ impl PageServerNode {
                 .map(serde_json::from_str)
                 .transpose()
                 .context("parse `wal_receiver_protocol_override` from json")?,
+            rel_size_v2_enabled: settings
+                .remove("rel_size_v2_enabled")
+                .map(|x| x.parse::<bool>())
+                .transpose()
+                .context("Failed to parse 'rel_size_v2_enabled' as bool")?,
+            gc_compaction_enabled: settings
+                .remove("gc_compaction_enabled")
+                .map(|x| x.parse::<bool>())
+                .transpose()
+                .context("Failed to parse 'gc_compaction_enabled' as bool")?,
+            gc_compaction_initial_threshold_kb: settings
+                .remove("gc_compaction_initial_threshold_kb")
+                .map(|x| x.parse::<u64>())
+                .transpose()
+                .context("Failed to parse 'gc_compaction_initial_threshold_kb' as integer")?,
+            gc_compaction_ratio_percent: settings
+                .remove("gc_compaction_ratio_percent")
+                .map(|x| x.parse::<u64>())
+                .transpose()
+                .context("Failed to parse 'gc_compaction_ratio_percent' as integer")?,
         };
         if !settings.is_empty() {
             bail!("Unrecognized tenant settings: {settings:?}")

control_plane/src/storage_controller.rs

@@ -822,10 +822,7 @@ impl StorageController {
         self.dispatch(
             Method::PUT,
             format!("control/v1/tenant/{tenant_shard_id}/migrate"),
-            Some(TenantShardMigrateRequest {
-                tenant_shard_id,
-                node_id,
-            }),
+            Some(TenantShardMigrateRequest { node_id }),
         )
         .await
     }

control_plane/storcon_cli/src/main.rs

@@ -1,12 +1,17 @@
 use futures::StreamExt;
-use std::{str::FromStr, time::Duration};
+use std::{
+    collections::{HashMap, HashSet},
+    str::FromStr,
+    time::Duration,
+};
 
 use clap::{Parser, Subcommand};
 use pageserver_api::{
     controller_api::{
         AvailabilityZone, NodeAvailabilityWrapper, NodeDescribeResponse, NodeShardResponse,
-        SafekeeperDescribeResponse, ShardSchedulingPolicy, TenantCreateRequest,
-        TenantDescribeResponse, TenantPolicyRequest,
+        SafekeeperDescribeResponse, SafekeeperSchedulingPolicyRequest, ShardSchedulingPolicy,
+        ShardsPreferredAzsRequest, SkSchedulingPolicy, TenantCreateRequest, TenantDescribeResponse,
+        TenantPolicyRequest,
     },
     models::{
         EvictionPolicy, EvictionPolicyLayerAccessThreshold, LocationConfigSecondary,
@@ -112,6 +117,13 @@ enum Command {
         #[arg(long)]
         node: NodeId,
     },
+    /// Migrate the secondary location for a tenant shard to a specific pageserver.
+    TenantShardMigrateSecondary {
+        #[arg(long)]
+        tenant_shard_id: TenantShardId,
+        #[arg(long)]
+        node: NodeId,
+    },
     /// Cancel any ongoing reconciliation for this shard
     TenantShardCancelReconcile {
         #[arg(long)]
@@ -146,6 +158,12 @@ 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 {
@@ -214,6 +232,13 @@ 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)]
@@ -266,6 +291,17 @@ 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);
 
@@ -395,11 +431,12 @@ 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", "Scheduling", "Availability"]);
+            table.set_header(["Id", "Hostname", "AZ", "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),
                 ]);
@@ -459,33 +496,65 @@ async fn main() -> anyhow::Result<()> {
             println!("{table}");
         }
         Command::Tenants { node_id: None } => {
-            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));
-
+            // Set up output formatting
             let mut table = comfy_table::Table::new();
             table.set_header([
                 "TenantId",
+                "Preferred AZ",
                 "ShardCount",
                 "StripeSize",
                 "Placement",
                 "Scheduling",
             ]);
-            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),
-                ]);
+
+            // 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!("");
             }
 
             println!("{table}");
@@ -540,10 +609,7 @@ async fn main() -> anyhow::Result<()> {
             tenant_shard_id,
             node,
         } => {
-            let req = TenantShardMigrateRequest {
-                tenant_shard_id,
-                node_id: node,
-            };
+            let req = TenantShardMigrateRequest { node_id: node };
 
             storcon_client
                 .dispatch::<TenantShardMigrateRequest, TenantShardMigrateResponse>(
@@ -553,6 +619,20 @@ async fn main() -> anyhow::Result<()> {
                 )
                 .await?;
         }
+        Command::TenantShardMigrateSecondary {
+            tenant_shard_id,
+            node,
+        } => {
+            let req = TenantShardMigrateRequest { node_id: node };
+
+            storcon_client
+                .dispatch::<TenantShardMigrateRequest, TenantShardMigrateResponse>(
+                    Method::PUT,
+                    format!("control/v1/tenant/{tenant_shard_id}/migrate_secondary"),
+                    Some(req),
+                )
+                .await?;
+        }
         Command::TenantShardCancelReconcile { tenant_shard_id } => {
             storcon_client
                 .dispatch::<(), ()>(
@@ -596,6 +676,19 @@ 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)]);
@@ -604,7 +697,14 @@ async fn main() -> anyhow::Result<()> {
             println!("{table}");
             println!("Shards:");
             let mut table = comfy_table::Table::new();
-            table.set_header(["Shard", "Attached", "Secondary", "Last error", "status"]);
+            table.set_header([
+                "Shard",
+                "Attached",
+                "Attached AZ",
+                "Secondary",
+                "Last error",
+                "status",
+            ]);
             for shard in shards {
                 let secondary = shard
                     .node_secondary
@@ -627,11 +727,18 @@ 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),
-                    shard
-                        .node_attached
-                        .map(|n| format!("{}", n))
+                    attached_node
+                        .map(|n| format!("{} ({})", n.listen_http_addr, n.id))
+                        .unwrap_or(String::new()),
+                    attached_node
+                        .map(|n| n.availability_zone_id.clone())
                         .unwrap_or(String::new()),
                     secondary,
                     shard.last_error,
@@ -640,6 +747,66 @@ 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>(
@@ -915,10 +1082,7 @@ async fn main() -> anyhow::Result<()> {
                             .dispatch::<TenantShardMigrateRequest, TenantShardMigrateResponse>(
                                 Method::PUT,
                                 format!("control/v1/tenant/{}/migrate", mv.tenant_shard_id),
-                                Some(TenantShardMigrateRequest {
-                                    tenant_shard_id: mv.tenant_shard_id,
-                                    node_id: mv.to,
-                                }),
+                                Some(TenantShardMigrateRequest { node_id: mv.to }),
                             )
                             .await
                             .map_err(|e| (mv.tenant_shard_id, mv.from, mv.to, e))
@@ -1057,6 +1221,23 @@ 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(())

docker-compose/compute_wrapper/Dockerfile

@@ -10,10 +10,7 @@ USER root
 RUN apt-get update &&       \
     apt-get install -y curl \
                        jq   \
-                       python3-pip \
                        netcat-openbsd
-#Faker is required for the pg_anon test
-RUN case $COMPUTE_IMAGE in compute-node-v17) OPT="--break-system-packages";; *) OPT= ;; esac && pip3 install $OPT Faker
 #This is required for the pg_hintplan test
 RUN mkdir -p /ext-src/pg_hint_plan-src && chown postgres /ext-src/pg_hint_plan-src 
 

docker-compose/docker-compose.yml

@@ -150,8 +150,8 @@ services:
         - REPOSITORY=${REPOSITORY:-neondatabase}
         - COMPUTE_IMAGE=compute-node-v${PG_VERSION:-16}
         - TAG=${TAG:-latest}
-        - http_proxy=$http_proxy
-        - https_proxy=$https_proxy
+        - http_proxy=${http_proxy:-}
+        - https_proxy=${https_proxy:-}
     environment:
       - PG_VERSION=${PG_VERSION:-16}
       #- RUST_BACKTRACE=1
@@ -185,6 +185,8 @@ services:
   neon-test-extensions:
     profiles: ["test-extensions"]
     image: ${REPOSITORY:-neondatabase}/neon-test-extensions-v${PG_TEST_VERSION:-16}:${TAG:-latest}
+    environment:
+      - PGPASSWORD=cloud_admin
     entrypoint:
       - "/bin/bash"
       - "-c"

docker-compose/docker_compose_test.sh

@@ -18,14 +18,10 @@ cd $(dirname $0)
 COMPUTE_CONTAINER_NAME=docker-compose-compute-1
 TEST_CONTAINER_NAME=docker-compose-neon-test-extensions-1
 PSQL_OPTION="-h localhost -U cloud_admin -p 55433 -d postgres"
-: ${http_proxy:=}
-: ${https_proxy:=}
-export http_proxy https_proxy
 
 cleanup() {
     echo "show container information"
     docker ps
-    docker compose --profile test-extensions -f $COMPOSE_FILE logs
     echo "stop containers..."
     docker compose --profile test-extensions -f $COMPOSE_FILE down
 }
@@ -35,12 +31,6 @@ for pg_version in ${TEST_VERSION_ONLY-14 15 16 17}; do
     echo "clean up containers if exists"
     cleanup
     PG_TEST_VERSION=$((pg_version < 16 ? 16 : pg_version))
-    # The support of pg_anon not yet added to PG17, so we have to add the corresponding option for other PG versions
-    if [ "${pg_version}" -ne 17 ]; then
-      SPEC_PATH="compute_wrapper/var/db/postgres/specs"
-      mv $SPEC_PATH/spec.json $SPEC_PATH/spec.bak
-      jq '.cluster.settings += [{"name": "session_preload_libraries","value": "anon","vartype": "string"}]' "${SPEC_PATH}/spec.bak" > "${SPEC_PATH}/spec.json"
-    fi
     PG_VERSION=$pg_version PG_TEST_VERSION=$PG_TEST_VERSION docker compose --profile test-extensions -f $COMPOSE_FILE up --build -d
 
     echo "wait until the compute is ready. timeout after 60s. "
@@ -50,7 +40,6 @@ for pg_version in ${TEST_VERSION_ONLY-14 15 16 17}; do
         cnt=`expr $cnt + 3`
         if [ $cnt -gt 60 ]; then
             echo "timeout before the compute is ready."
-            cleanup
             exit 1
         fi
         if docker compose --profile test-extensions -f $COMPOSE_FILE logs "compute_is_ready" | grep -q "accepting connections"; then
@@ -62,36 +51,19 @@ for pg_version in ${TEST_VERSION_ONLY-14 15 16 17}; do
     done
 
     if [ $pg_version -ge 16 ]; then
-        echo Enabling trust connection
-        docker exec $COMPUTE_CONTAINER_NAME bash -c "sed -i '\$d' /var/db/postgres/compute/pg_hba.conf && echo -e 'host\t all\t all\t all\t trust' >> /var/db/postgres/compute/pg_hba.conf && psql $PSQL_OPTION -c 'select pg_reload_conf()' "
-        echo Adding postgres role
-        docker exec $COMPUTE_CONTAINER_NAME psql $PSQL_OPTION -c "CREATE ROLE postgres SUPERUSER LOGIN"
         # This is required for the pg_hint_plan test, to prevent flaky log message causing the test to fail
         # It cannot be moved to Dockerfile now because the database directory is created after the start of the container
         echo Adding dummy config
         docker exec $COMPUTE_CONTAINER_NAME touch /var/db/postgres/compute/compute_ctl_temp_override.conf
-        # This block is required for the pg_anon extension test.
-        # The test assumes that it is running on the same host with the postgres engine.
-        # In our case it's not true, that's why we are copying files to the compute node
+        # The following block copies the files for the pg_hintplan test to the compute node for the extension test in an isolated docker-compose environment
         TMPDIR=$(mktemp -d)
-        # Add support for pg_anon for pg_v16
-        if [ $pg_version -ne 17 ]; then
-          docker cp $TEST_CONTAINER_NAME:/ext-src/pg_anon-src/data $TMPDIR/data
-          echo -e '1\t too \t many \t tabs' > $TMPDIR/data/bad.csv
-          docker cp $TMPDIR/data $COMPUTE_CONTAINER_NAME:/tmp/tmp_anon_alternate_data
-        rm -rf $TMPDIR
-        fi
-        TMPDIR=$(mktemp -d)
-        # The following block does the same for the pg_hintplan test
         docker cp $TEST_CONTAINER_NAME:/ext-src/pg_hint_plan-src/data $TMPDIR/data
         docker cp $TMPDIR/data $COMPUTE_CONTAINER_NAME:/ext-src/pg_hint_plan-src/
         rm -rf $TMPDIR
         # We are running tests now
-        if docker exec -e SKIP=timescaledb-src,rdkit-src,postgis-src,pgx_ulid-src,pgtap-src,pg_tiktoken-src,pg_jsonschema-src,pg_graphql-src,kq_imcx-src,wal2json_2_5-src \
+        if ! docker exec -e SKIP=timescaledb-src,rdkit-src,postgis-src,pgx_ulid-src,pgtap-src,pg_tiktoken-src,pg_jsonschema-src,pg_graphql-src,kq_imcx-src,wal2json_2_5-src \
             $TEST_CONTAINER_NAME /run-tests.sh | tee testout.txt
         then
-            cleanup
-        else
             FAILED=$(tail -1 testout.txt)
             for d in $FAILED
             do
@@ -101,13 +73,7 @@ for pg_version in ${TEST_VERSION_ONLY-14 15 16 17}; do
                 cat $d/regression.out $d/regression.diffs || true
             done
         rm -rf $FAILED
-        cleanup
         exit 1
         fi
     fi
-    cleanup
-    # Restore the original spec.json
-    if [ "$pg_version" -ne 17 ]; then
-      mv "$SPEC_PATH/spec.bak" "$SPEC_PATH/spec.json"
-    fi
 done

docker-compose/run-tests.sh

@@ -7,7 +7,10 @@ LIST=$( (echo -e "${SKIP//","/"\n"}"; ls -d -- *-src) | sort | uniq -u)
 for d in ${LIST}
 do
        [ -d "${d}" ] || continue
-    psql -c "select 1" >/dev/null || break
+       if ! psql -w -c "select 1" >/dev/null; then
+          FAILED="${d} ${FAILED}"
+          break
+       fi
        USE_PGXS=1 make -C "${d}" installcheck || FAILED="${d} ${FAILED}"
 done
 [ -z "${FAILED}" ] && exit 0

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`.
@@ -202,47 +202,87 @@ 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.
+This requires us to define storcon <-> cplane interface and changes.
 
-### storage_controller <-> control plane interface
+### storage_controller <-> control plane interface and changes
 
 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) 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.
+start path 2) uniformity. It makes storage_controller responsible for retrying
+notifying control plane until it succeeds.
 
-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
+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
 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 `safekeeper_conf: Configuration`. The call
-should be retried until succeeds.
+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.
 
 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.
@@ -360,10 +400,10 @@ 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
@@ -412,8 +452,8 @@ 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.
@@ -422,35 +462,51 @@ 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 observers joint configuration.
+- Initially walproposer can just stop working while it observes 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.
 
-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.
+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`.
 
-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.
+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).
 
+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),
@@ -458,10 +514,10 @@ 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

@@ -0,0 +1,247 @@
+# 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.

docs/rfcs/041-sharded-ingest.md

@@ -0,0 +1,255 @@
+# 
+Created on Aug 2024
+Implemented on Jan 2025
+
+## Summary
+
+Data in large tenants is split up between multiple pageservers according to key hashes, as
+introduced in the [sharding RFC](031-sharding-static.md) and [shard splitting RFC](032-shard-splitting.md).
+
+Whereas currently we send all WAL to all pageserver shards, and each shard filters out the data that it needs,
+in this RFC we add a mechanism to filter the WAL on the safekeeper, so that each shard receives
+only the data it needs.
+
+This will place some extra CPU load on the safekeepers, in exchange for reducing the network bandwidth
+for ingesting WAL back to scaling as O(1) with shard count, rather than O(N_shards).
+
+## Motivation
+
+1. Large databases require higher shard counts.  Whereas currently we run with up to 8 shards for tenants
+with a few TB of storage, the next order of magnitude capacity increase will require tens of shards, such
+that sending all WAL to all shards is impractical in terms of bandwidth.
+2. For contemporary database sizes (~2TB), the pageserver is the bottleneck for ingest: since each
+   shard has to decode and process the whole WAL, sharding doesn't fully relieve this bottleneck.  To achieve significantly higher ingest speeds, we need to filter the WAL earlier so that each pageserver
+   only has to process relevant parts.
+
+## Non Goals (if relevant)
+
+We do not seek to introduce multiple WALs per timeline, or to share the work of handling a timeline's
+WAL across safekeepers (beyond simple 3x replication).  This RFC may be thought of as an incremental
+move of the ingestion bottleneck up the stack: instead of high write rates bottlenecking on the
+pageserver, they will bottleneck on the safekeeper.
+
+## Impacted components (e.g. pageserver, safekeeper, console, etc)
+
+Safekeeper, pageserver.
+
+There will be no control plane or storage controller coordination needed, as pageservers will directly
+indicate their sharding parameters to the safekeeper when subscribing for WAL.
+
+## Proposed implementation
+
+Terminology:
+- "Data pages" refers to postgres relation blocks, and SLRU blocks.
+- "Metadata pages" refers to everything else the pageserver stores, such as relation sizes and
+  directories of relations.
+
+### Phase 1: Refactor ingest
+
+Currently, pageserver ingest code is structured approximately as follows:
+1. `handle_walreceiver_connection` reads a stream of binary WAL records off a network
+   socket
+2. `WalIngest::ingest_record` to translate the record into a series of page-level modifications
+3. `DatadirModification` accumulates page updates from several `ingest_record` calls, and when
+   its `commit()` method is called, flushes these into a Timeline's open `InMemoryLayer`.
+
+This process currently assumes access to a pageserver `Timeline` throughout `ingest_record` and
+from `DatadirModification`, which is used to do read-modify-write cycles on metadata pages
+such as relation sizes and the master DBDIR page.  It also assumes that records are ingested
+strictly one after the other: they cannot be ingested in parallel because each record assumes
+that earlier records' changes have already been applied to `Timeline`.
+
+This code will be refactored to disentangle the simple, fast decode of relation page writes
+from the more complex logic for updating internal metadata.  An intermediate representation
+called `InterpretedWalRecords` will be introduced.  This is similar to the internal state of
+a `DatadirModification`, but does not require access to a Timeline.  Instead of storing
+metadata updates as materialized writes to pages, it will accumulate these as abstract operations,
+for example rather than including a write to a relation size key, this structure will include
+an operation that indicates "Update relation _foo_'s size to the max of its current value and
+_bar_", such that these may be applied later to a real Timeline.
+
+The `DatadirModification` will be aware of the `EphemeralFile` format, so that as it accumulates
+simple page writes of relation blocks, it can write them directly into a buffer in the serialized
+format.  This will avoid the need to later deserialize/reserialize this data when passing the
+structure between safekeeper and pageserver.
+
+The new pipeline will be:
+1. `handle_walreceiver_connection` reads a stream of binary WAL records off a network
+2. A `InterpretedWalRecords` is generated from the incoming WAL records.  This does not
+   require a reference to a Timeline.
+3. The logic that is current spread between `WalIngest` and `DatadirModification` for updating
+   metadata will be refactored to consume the metadata operations from the `InterpretedWalRecords`
+   and turn them into literal writes to metadata pages.  This part must be done sequentially.
+4. The resulting buffer of metadata page writes is combined with the buffer of relation block
+   writes, and written into the `InMemoryLayer`.
+
+Implemented in:
+1. https://github.com/neondatabase/neon/pull/9472
+2. https://github.com/neondatabase/neon/pull/9504
+3. https://github.com/neondatabase/neon/pull/9524
+
+### Phase 2: Decode & filter on safekeeper
+
+In the previous phase, the ingest code was modified to be able to do most of its work without access to
+a Timeline: this first stage of ingest simply converts a series of binary wal records into
+a buffer of relation/SLRU page writes, and a buffer of abstract metadata writes.
+
+The modified ingest code may be transplanted from pageserver to safekeeper (probably via a
+shared crate).  The safekeeper->pageserver network protocol is modified to:
+ - in subscription requests, send the `ShardIdentity` from the pageserver to the safekeeper
+ - in responses, transmit a `InterpretedWalRecords` instead of a raw `WalRecord`.
+ - use the `ShardIdentity` to filter the `ProcessedWalIngest` to relevant content for
+   the subscribing shard before transmitting it.
+
+The overall behavior of the pageserver->safekeeper interaction remains the same, in terms of
+consistent LSN feedback, and connection management.  Only the payload of the subscriptions
+changes, to express an LSN range of WAL as a filtered `ProcessedWalIngest` instead of the
+raw data.
+
+The ingest code on the pageserver can now skip the part where it does the first phase of
+processing, as it will receive pre-processed, compressed data off the wire.
+
+Note that `InterpretedWalRecord` batches multiple `InterpretedWalRecord(s)` in the same network
+message. Safekeeper reads WAL in chunks of 16 blocks and then decodes as many Postgres WAL records
+as possible. Each Postgres WAL record maps to one `InterpretedWalRecord` for potentially multiple shards.
+Hence, the size of the batch is given by the number of Postgres WAL records that fit in 16 blocks.
+
+The protocol needs to support evolution. Protobuf was chosen here with the view that, in the future,
+we may migrate it to GRPC altogether
+
+Implemented in:
+1. https://github.com/neondatabase/neon/pull/9746
+2. https://github.com/neondatabase/neon/pull/9821
+
+### Phase 3: Fan out interpreted WAL
+
+In the previous phase, the initial processing of WAL was moved to the safekeeper, but it is still
+done once for each shard: this will generate O(N_shards) CPU work on the safekeeper (especially
+when considering converting to Protobuf format and compression).
+
+To avoid this, we fan-out WAL from one (tenant, timeline, shard) to all other shards subscribed on
+the same safekeeper. Under normal operation, the WAL will be read from disk, decoded and interpreted
+_only_ once per (safekeeper, timeline).
+
+When the first shard of a sharded timeline subscribes to a given safekeeper a task is spawned
+for the WAL reader (`InterpretedWalReader`). This task reads WAL, decodes, interprets it and sends
+it to the sender (`InterpretedWalSender`). The sender is a future that is polled from the connection
+task. When further shards subscribe on the safekeeper they will attach themselves to the existing WAL reader.
+There's two cases to consider:
+1. The shard's requested `start_lsn` is ahead of the current position of the WAL reader. In this case, the shard
+will start receiving data when the reader reaches that LSN. The intuition here is that there's little to gain
+by letting shards "front-run" since compute backpressure is based on the laggard LSN.
+2. The shard's requested `start_lsn` is below the current position of the WAL reader. In this case, the WAL reader
+gets reset to this requested position (same intuition). Special care is taken such that advanced shards do not receive
+interpreted WAL records below their current position.
+
+The approach above implies that there is at most one WAL reader per (tenant, timeline) on a given safekeeper at any point in time.
+If this turns out to be operationally problematic, there's a trick we can deploy: `--max-delta-for-fanout` is an optional safekeeper
+argument that controls the max absolute delta between a new shard and the current WAL position of the WAL reader. If the absolute
+delta is above that value, a new reader is spawned. Note that there's currently no concurrency control on the number of WAL readers,
+so it's recommended to use large values to avoid pushing CPU utilisation too high.
+
+Unsharded tenants do not spawn a separate task for the interpreted WAL reader since there's no benefit to it. Instead they poll
+the reader and sender concurrently from the connection task.
+
+Shard splits are interesting here because it is the only case when the same shard might have two subscriptions at the same time.
+This is handled by giving readers a unique identifier. Both shards will receive the same data while respecting their requested start
+position.
+
+Implemented in:
+1. https://github.com/neondatabase/neon/pull/10190
+
+## Deployment
+
+Each phase shall be deployed independently. Special care should be taken around protocol changes.
+
+## Observability Tips
+
+* The safekeeper logs the protocol requested by the pageserver
+along with the pageserver ID, tenant, timeline and shard: `starting streaming from`.
+* There's metrics for the number of wal readers:
+  * `safekeeper_wal_readers{kind="task", target=~"pageserver.*"}` gives the number of wal reader tasks for each SK
+  * `safekeeper_wal_readers{kind="future", target=~"pageserver.*"}` gives the numer of wal readers polled inline by each SK
+  * `safekeeper_interpreted_wal_reader_tasks` gives the number of wal reader tasks per tenant, timeline
+* Interesting log lines for the fan-out reader:
+  * `Spawning interpreted`: first shard creates the interpreted wal reader
+  * `Fanning out`: a subsequent shard attaches itself to an interpreted wal reader
+  * `Aborting interpreted`: all senders have finished and the reader task is being aborted
+
+## Future Optimizations
+
+This sections describes some improvement areas which may be revisited in the future.
+
+### Buffering of Interpreted WAL
+
+The interpreted WAL reader may buffer interpreted WAL records in user space to help with serving
+subscriptions that are lagging behind the current position of the reader.
+
+Counterpoints:
+* Safekeepers serve many thousands of timelines and allocating a buffer for each might be wasteful,
+especially given that it would go unused on the happy path.
+* WAL is buffered in the kernel page cache. Usually we'd only pay the CPU cost of decoding and interpreting.
+
+### Tweaking the Pagserver Safekeeper Selection Algorithm
+
+We could make the pageserver aware of which safekeeper's already host shards for the timeline along
+with their current WAL positions. The pageserver should then prefer safkeepers that are in the same
+AZ _and_ already have a shard with a position close to the desired start position.
+
+We currently run one safekeeper per AZ, so the point is mute until that changes.
+
+### Pipelining first ingest phase
+
+The first ingest phase is a stateless transformation of a binary WAL record into a pre-processed
+output per shard.  To put multiple CPUs to work, we may pipeline this processing up to some defined buffer
+depth.
+
+## Alternatives considered
+
+### Give safekeepers enough state to fully decode WAL
+
+In this RFC, we only do the first phase of ingest on the safekeeper, because this is
+the phase that is stateless.  Subsequent changes then happen on the pageserver, with
+access to the `Timeline` state.
+
+We could do more work on the safekeeper if we transmitted metadata state to the safekeeper
+when subscribing to the WAL: for example, by telling the safekeeper all the relation sizes,
+so that it could then generate all the metadata writes for relation sizes.
+
+We avoid doing this for several reasons:
+1. Complexity: it's a more invasive protocol change
+2. Decoupling: having the safekeeper understand the `ProcessedWalIngest` already somewhat
+   infects it with knowledge of the pageserver, but this is mainly an abstract structure
+   that describes postgres writes.  However, if we taught the safekeeper about the exact
+   way that pageserver deals with metadata keys, this would be a much tighter coupling.
+3. Load: once the WAL has been processed to the point that it can be split between shards,
+   it is preferable to share out work on the remaining shards rather than adding extra CPU
+   load to the safekeeper.
+
+### Do pre-processing on the compute instead of the safekeeper
+
+Since our first stage of ingest is stateless, it could be done at any stage in the pipeline,
+all the way up to the compute.
+
+We choose not to do this, because it is useful for the safekeeper to store the raw WAL rather
+than just the preprocessed WAL:
+- The safekeeper still needs to be able to serve raw WAL back to postgres for e.g. physical replication
+- It simplifies our paxos implementation to have the offset in the write log be literally
+  the same as the LSN
+- Raw WAL must have a stable protocol since we might have to re-ingest it at arbitrary points in the future.
+  Storing raw WAL give us more flexibility to evolve the pageserver, safekeeper protocol.
+
+### Do wal pre-processing on shard 0 or a separate service, send it to other shards from there
+
+If we wanted to keep the safekeepers as entirely pure stores of raw WAL bytes, then
+we could do the initial decode and shard-splitting in some other location:
+- Shard 0 could subscribe to the full WAL and then send writes to other shards
+- A new intermediate service between the safekeeper and pageserver could do the splitting.
+
+So why not?
+- Extra network hop from shard 0 to the final destination shard
+- Clearly there is more infrastructure involved here compared with doing it inline on the safekeeper.
+- Safekeepers already have very light CPU load: typical cloud instances shapes with appropriate
+  disks for the safekeepers effectively have "free" CPU resources.
+- Doing extra work on shard 0 would complicate scheduling of shards on pageservers, because
+  shard 0 would have significantly higher CPU load under write workloads than other shards.

docs/storage_broker.md

@@ -1,6 +1,6 @@
 # Storage broker
 
-Storage broker targets two issues
+Storage broker targets two issues:
 - Allowing safekeepers and pageservers learn which nodes also hold their
   timelines, and timeline statuses there.
 - Avoiding O(n^2) connections between storage nodes while doing so.
@@ -19,7 +19,7 @@ Currently, the only message is `SafekeeperTimelineInfo`. Each safekeeper, for
 each active timeline, once in a while pushes timeline status to the broker.
 Other nodes subscribe and receive this info, using it per above.
 
-Broker serves /metrics on the same port as grpc service.
+Broker serves /metrics on the same port as grpc service. 
 
 grpcurl can be used to check which values are currently being pushed:
 ```

libs/compute_api/src/spec.rs

@@ -138,6 +138,13 @@ pub struct ComputeSpec {
     /// enough spare connections for reconfiguration process to succeed.
     #[serde(default = "default_reconfigure_concurrency")]
     pub reconfigure_concurrency: usize,
+
+    /// If set to true, the compute_ctl will drop all subscriptions before starting the
+    /// compute. This is needed when we start an endpoint on a branch, so that child
+    /// would not compete with parent branch subscriptions
+    /// over the same replication content from publisher.
+    #[serde(default)] // Default false
+    pub drop_subscriptions_before_start: bool,
 }
 
 /// Feature flag to signal `compute_ctl` to enable certain experimental functionality.

libs/pageserver_api/src/config.rs

@@ -120,6 +120,7 @@ pub struct ConfigToml {
     pub no_sync: Option<bool>,
     pub wal_receiver_protocol: PostgresClientProtocol,
     pub page_service_pipelining: PageServicePipeliningConfig,
+    pub get_vectored_concurrent_io: GetVectoredConcurrentIo,
 }
 
 #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
@@ -158,6 +159,25 @@ pub enum PageServiceProtocolPipelinedExecutionStrategy {
     Tasks,
 }
 
+#[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
+#[serde(tag = "mode", rename_all = "kebab-case")]
+#[serde(deny_unknown_fields)]
+pub enum GetVectoredConcurrentIo {
+    /// The read path is fully sequential: layers are visited
+    /// one after the other and IOs are issued and waited upon
+    /// from the same task that traverses the layers.
+    Sequential,
+    /// The read path still traverses layers sequentially, and
+    /// index blocks will be read into the PS PageCache from
+    /// that task, with waiting.
+    /// But data IOs are dispatched and waited upon from a sidecar
+    /// task so that the traversing task can continue to traverse
+    /// layers while the IOs are in flight.
+    /// If the PS PageCache miss rate is low, this improves
+    /// throughput dramatically.
+    SidecarTask,
+}
+
 pub mod statvfs {
     pub mod mock {
         #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
@@ -234,9 +254,18 @@ pub struct TenantConfigToml {
     // Duration::ZERO means automatic compaction is disabled.
     #[serde(with = "humantime_serde")]
     pub compaction_period: Duration,
-    // Level0 delta layer threshold for compaction.
+    /// Level0 delta layer threshold for compaction.
     pub compaction_threshold: usize,
     pub compaction_algorithm: crate::models::CompactionAlgorithmSettings,
+    /// Level0 delta layer threshold at which to delay layer flushes for compaction backpressure,
+    /// such that they take 2x as long, and start waiting for layer flushes during ephemeral layer
+    /// rolls. This helps compaction keep up with WAL ingestion, and avoids read amplification
+    /// blowing up. Should be >compaction_threshold. If None, defaults to 2 * compaction_threshold.
+    /// 0 to disable.
+    pub l0_flush_delay_threshold: Option<usize>,
+    /// Level0 delta layer threshold at which to stall layer flushes. 0 to disable. If None,
+    /// defaults to 4 * compaction_threshold. Must be >compaction_threshold to avoid deadlock.
+    pub l0_flush_stall_threshold: Option<usize>,
     // Determines how much history is retained, to allow
     // branching and read replicas at an older point in time.
     // The unit is #of bytes of WAL.
@@ -301,6 +330,20 @@ pub struct TenantConfigToml {
     pub timeline_offloading: bool,
 
     pub wal_receiver_protocol_override: Option<PostgresClientProtocol>,
+
+    /// Enable rel_size_v2 for this tenant. Once enabled, the tenant will persist this information into
+    /// `index_part.json`, and it cannot be reversed.
+    pub rel_size_v2_enabled: Option<bool>,
+
+    // gc-compaction related configs
+    /// Enable automatic gc-compaction trigger on this tenant.
+    pub gc_compaction_enabled: bool,
+    /// The initial threshold for gc-compaction in KB. Once the total size of layers below the gc-horizon is above this threshold,
+    /// gc-compaction will be triggered.
+    pub gc_compaction_initial_threshold_kb: u64,
+    /// The ratio that triggers the auto gc-compaction. If (the total size of layers between L2 LSN and gc-horizon) / (size below the L2 LSN)
+    /// is above this ratio, gc-compaction will be triggered.
+    pub gc_compaction_ratio_percent: u64,
 }
 
 pub mod defaults {
@@ -450,6 +493,11 @@ impl Default for ConfigToml {
                     execution: PageServiceProtocolPipelinedExecutionStrategy::ConcurrentFutures,
                 })
             },
+            get_vectored_concurrent_io: if !cfg!(test) {
+                GetVectoredConcurrentIo::Sequential
+            } else {
+                GetVectoredConcurrentIo::SidecarTask
+            },
         }
     }
 }
@@ -494,6 +542,9 @@ pub mod tenant_conf_defaults {
     // By default ingest enough WAL for two new L0 layers before checking if new image
     // image layers should be created.
     pub const DEFAULT_IMAGE_LAYER_CREATION_CHECK_THRESHOLD: u8 = 2;
+    pub const DEFAULT_GC_COMPACTION_ENABLED: bool = false;
+    pub const DEFAULT_GC_COMPACTION_INITIAL_THRESHOLD_KB: u64 = 10240000;
+    pub const DEFAULT_GC_COMPACTION_RATIO_PERCENT: u64 = 100;
 }
 
 impl Default for TenantConfigToml {
@@ -510,6 +561,8 @@ impl Default for TenantConfigToml {
             compaction_algorithm: crate::models::CompactionAlgorithmSettings {
                 kind: DEFAULT_COMPACTION_ALGORITHM,
             },
+            l0_flush_delay_threshold: None,
+            l0_flush_stall_threshold: None,
             gc_horizon: DEFAULT_GC_HORIZON,
             gc_period: humantime::parse_duration(DEFAULT_GC_PERIOD)
                 .expect("cannot parse default gc period"),
@@ -538,6 +591,10 @@ impl Default for TenantConfigToml {
             lsn_lease_length_for_ts: LsnLease::DEFAULT_LENGTH_FOR_TS,
             timeline_offloading: false,
             wal_receiver_protocol_override: None,
+            rel_size_v2_enabled: None,
+            gc_compaction_enabled: DEFAULT_GC_COMPACTION_ENABLED,
+            gc_compaction_initial_threshold_kb: DEFAULT_GC_COMPACTION_INITIAL_THRESHOLD_KB,
+            gc_compaction_ratio_percent: DEFAULT_GC_COMPACTION_RATIO_PERCENT,
         }
     }
 }

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, AvailabilityZone>,
+    pub preferred_az_ids: HashMap<TenantShardId, Option<AvailabilityZone>>,
 }
 
 #[derive(Serialize, Deserialize)]
@@ -144,6 +144,8 @@ pub struct NodeDescribeResponse {
     pub availability: NodeAvailabilityWrapper,
     pub scheduling: NodeSchedulingPolicy,
 
+    pub availability_zone_id: String,
+
     pub listen_http_addr: String,
     pub listen_http_port: u16,
 
@@ -179,7 +181,6 @@ pub struct TenantDescribeResponseShard {
 /// specifies some constraints, e.g. asking it to get off particular node(s)
 #[derive(Serialize, Deserialize, Debug)]
 pub struct TenantShardMigrateRequest {
-    pub tenant_shard_id: TenantShardId,
     pub node_id: NodeId,
 }
 
@@ -323,7 +324,7 @@ impl From<NodeSchedulingPolicy> for String {
 #[derive(Serialize, Deserialize, Clone, Copy, Eq, PartialEq, Debug)]
 pub enum SkSchedulingPolicy {
     Active,
-    Disabled,
+    Pause,
     Decomissioned,
 }
 
@@ -333,9 +334,13 @@ impl FromStr for SkSchedulingPolicy {
     fn from_str(s: &str) -> Result<Self, Self::Err> {
         Ok(match s {
             "active" => Self::Active,
-            "disabled" => Self::Disabled,
+            "pause" => Self::Pause,
             "decomissioned" => Self::Decomissioned,
-            _ => return Err(anyhow::anyhow!("Unknown scheduling state '{s}'")),
+            _ => {
+                return Err(anyhow::anyhow!(
+                    "Unknown scheduling policy '{s}', try active,pause,decomissioned"
+                ))
+            }
         })
     }
 }
@@ -345,7 +350,7 @@ impl From<SkSchedulingPolicy> for String {
         use SkSchedulingPolicy::*;
         match value {
             Active => "active",
-            Disabled => "disabled",
+            Pause => "pause",
             Decomissioned => "decomissioned",
         }
         .to_string()
@@ -368,6 +373,16 @@ 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 {}
 
@@ -405,8 +420,6 @@ 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,
@@ -422,6 +435,11 @@ 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

@@ -24,7 +24,9 @@ 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, Hash, PartialEq, Eq, Ord, PartialOrd, Serialize, Deserialize, Debug)]
+#[derive(
+    Clone, Copy, Default, Hash, PartialEq, Eq, Ord, PartialOrd, Serialize, Deserialize, Debug,
+)]
 pub struct CompactKey(i128);
 
 /// The storage key size.
@@ -706,7 +708,7 @@ pub fn repl_origin_key_range() -> Range<Key> {
 /// Non inherited range for vectored get.
 pub const NON_INHERITED_RANGE: Range<Key> = AUX_FILES_KEY..AUX_FILES_KEY.next();
 /// Sparse keyspace range for vectored get. Missing key error will be ignored for this range.
-pub const NON_INHERITED_SPARSE_RANGE: Range<Key> = Key::metadata_key_range();
+pub const SPARSE_RANGE: Range<Key> = Key::metadata_key_range();
 
 impl Key {
     // AUX_FILES currently stores only data for logical replication (slots etc), and
@@ -714,7 +716,42 @@ impl Key {
     // switch (and generally it likely should be optional), so ignore these.
     #[inline(always)]
     pub fn is_inherited_key(self) -> bool {
-        !NON_INHERITED_RANGE.contains(&self) && !NON_INHERITED_SPARSE_RANGE.contains(&self)
+        if self.is_sparse() {
+            self.is_inherited_sparse_key()
+        } else {
+            !NON_INHERITED_RANGE.contains(&self)
+        }
+    }
+
+    #[inline(always)]
+    pub fn is_sparse(self) -> bool {
+        self.field1 >= METADATA_KEY_BEGIN_PREFIX && self.field1 < METADATA_KEY_END_PREFIX
+    }
+
+    /// Check if the key belongs to the inherited keyspace.
+    fn is_inherited_sparse_key(self) -> bool {
+        debug_assert!(self.is_sparse());
+        self.field1 == RELATION_SIZE_PREFIX
+    }
+
+    pub fn sparse_non_inherited_keyspace() -> Range<Key> {
+        // The two keys are adjacent; if we will have non-adjancent keys in the future, we should return a keyspace
+        debug_assert_eq!(AUX_KEY_PREFIX + 1, REPL_ORIGIN_KEY_PREFIX);
+        Key {
+            field1: AUX_KEY_PREFIX,
+            field2: 0,
+            field3: 0,
+            field4: 0,
+            field5: 0,
+            field6: 0,
+        }..Key {
+            field1: REPL_ORIGIN_KEY_PREFIX + 1,
+            field2: 0,
+            field3: 0,
+            field4: 0,
+            field5: 0,
+            field6: 0,
+        }
     }
 
     #[inline(always)]

libs/pageserver_api/src/models.rs

@@ -29,11 +29,10 @@ use utils::{
 };
 
 use crate::{
-    key::Key,
+    key::{CompactKey, Key},
     reltag::RelTag,
     shard::{ShardCount, ShardStripeSize, TenantShardId},
 };
-use anyhow::bail;
 use bytes::{Buf, BufMut, Bytes, BytesMut};
 
 /// The state of a tenant in this pageserver.
@@ -463,6 +462,10 @@ pub struct TenantConfigPatch {
     #[serde(skip_serializing_if = "FieldPatch::is_noop")]
     pub compaction_algorithm: FieldPatch<CompactionAlgorithmSettings>,
     #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub l0_flush_delay_threshold: FieldPatch<usize>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub l0_flush_stall_threshold: FieldPatch<usize>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
     pub gc_horizon: FieldPatch<u64>,
     #[serde(skip_serializing_if = "FieldPatch::is_noop")]
     pub gc_period: FieldPatch<String>,
@@ -498,6 +501,14 @@ pub struct TenantConfigPatch {
     pub timeline_offloading: FieldPatch<bool>,
     #[serde(skip_serializing_if = "FieldPatch::is_noop")]
     pub wal_receiver_protocol_override: FieldPatch<PostgresClientProtocol>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub rel_size_v2_enabled: FieldPatch<bool>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub gc_compaction_enabled: FieldPatch<bool>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub gc_compaction_initial_threshold_kb: FieldPatch<u64>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub gc_compaction_ratio_percent: FieldPatch<u64>,
 }
 
 /// An alternative representation of `pageserver::tenant::TenantConf` with
@@ -511,6 +522,8 @@ pub struct TenantConfig {
     pub compaction_threshold: Option<usize>,
     // defer parsing compaction_algorithm, like eviction_policy
     pub compaction_algorithm: Option<CompactionAlgorithmSettings>,
+    pub l0_flush_delay_threshold: Option<usize>,
+    pub l0_flush_stall_threshold: Option<usize>,
     pub gc_horizon: Option<u64>,
     pub gc_period: Option<String>,
     pub image_creation_threshold: Option<usize>,
@@ -529,6 +542,10 @@ pub struct TenantConfig {
     pub lsn_lease_length_for_ts: Option<String>,
     pub timeline_offloading: Option<bool>,
     pub wal_receiver_protocol_override: Option<PostgresClientProtocol>,
+    pub rel_size_v2_enabled: Option<bool>,
+    pub gc_compaction_enabled: Option<bool>,
+    pub gc_compaction_initial_threshold_kb: Option<u64>,
+    pub gc_compaction_ratio_percent: Option<u64>,
 }
 
 impl TenantConfig {
@@ -540,6 +557,8 @@ impl TenantConfig {
             mut compaction_period,
             mut compaction_threshold,
             mut compaction_algorithm,
+            mut l0_flush_delay_threshold,
+            mut l0_flush_stall_threshold,
             mut gc_horizon,
             mut gc_period,
             mut image_creation_threshold,
@@ -558,6 +577,10 @@ impl TenantConfig {
             mut lsn_lease_length_for_ts,
             mut timeline_offloading,
             mut wal_receiver_protocol_override,
+            mut rel_size_v2_enabled,
+            mut gc_compaction_enabled,
+            mut gc_compaction_initial_threshold_kb,
+            mut gc_compaction_ratio_percent,
         } = self;
 
         patch.checkpoint_distance.apply(&mut checkpoint_distance);
@@ -568,6 +591,12 @@ impl TenantConfig {
         patch.compaction_period.apply(&mut compaction_period);
         patch.compaction_threshold.apply(&mut compaction_threshold);
         patch.compaction_algorithm.apply(&mut compaction_algorithm);
+        patch
+            .l0_flush_delay_threshold
+            .apply(&mut l0_flush_delay_threshold);
+        patch
+            .l0_flush_stall_threshold
+            .apply(&mut l0_flush_stall_threshold);
         patch.gc_horizon.apply(&mut gc_horizon);
         patch.gc_period.apply(&mut gc_period);
         patch
@@ -602,6 +631,16 @@ impl TenantConfig {
         patch
             .wal_receiver_protocol_override
             .apply(&mut wal_receiver_protocol_override);
+        patch.rel_size_v2_enabled.apply(&mut rel_size_v2_enabled);
+        patch
+            .gc_compaction_enabled
+            .apply(&mut gc_compaction_enabled);
+        patch
+            .gc_compaction_initial_threshold_kb
+            .apply(&mut gc_compaction_initial_threshold_kb);
+        patch
+            .gc_compaction_ratio_percent
+            .apply(&mut gc_compaction_ratio_percent);
 
         Self {
             checkpoint_distance,
@@ -610,6 +649,8 @@ impl TenantConfig {
             compaction_period,
             compaction_threshold,
             compaction_algorithm,
+            l0_flush_delay_threshold,
+            l0_flush_stall_threshold,
             gc_horizon,
             gc_period,
             image_creation_threshold,
@@ -628,6 +669,10 @@ impl TenantConfig {
             lsn_lease_length_for_ts,
             timeline_offloading,
             wal_receiver_protocol_override,
+            rel_size_v2_enabled,
+            gc_compaction_enabled,
+            gc_compaction_initial_threshold_kb,
+            gc_compaction_ratio_percent,
         }
     }
 }
@@ -1400,6 +1445,8 @@ pub enum PagestreamFeMessage {
     GetPage(PagestreamGetPageRequest),
     DbSize(PagestreamDbSizeRequest),
     GetSlruSegment(PagestreamGetSlruSegmentRequest),
+    #[cfg(feature = "testing")]
+    Test(PagestreamTestRequest),
 }
 
 // Wrapped in libpq CopyData
@@ -1411,6 +1458,22 @@ 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`
@@ -1422,7 +1485,28 @@ 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> {
@@ -1433,6 +1517,8 @@ 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),
         }
     }
@@ -1550,6 +1636,20 @@ 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)]
@@ -1569,7 +1669,7 @@ impl PagestreamFeMessage {
 
         match self {
             Self::Exists(req) => {
-                bytes.put_u8(0);
+                bytes.put_u8(PagestreamFeMessageTag::Exists 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);
@@ -1580,7 +1680,7 @@ impl PagestreamFeMessage {
             }
 
             Self::Nblocks(req) => {
-                bytes.put_u8(1);
+                bytes.put_u8(PagestreamFeMessageTag::Nblocks 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);
@@ -1591,7 +1691,7 @@ impl PagestreamFeMessage {
             }
 
             Self::GetPage(req) => {
-                bytes.put_u8(2);
+                bytes.put_u8(PagestreamFeMessageTag::GetPage 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);
@@ -1603,7 +1703,7 @@ impl PagestreamFeMessage {
             }
 
             Self::DbSize(req) => {
-                bytes.put_u8(3);
+                bytes.put_u8(PagestreamFeMessageTag::DbSize 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);
@@ -1611,13 +1711,24 @@ impl PagestreamFeMessage {
             }
 
             Self::GetSlruSegment(req) => {
-                bytes.put_u8(4);
+                bytes.put_u8(PagestreamFeMessageTag::GetSlruSegment 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_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()
@@ -1645,56 +1756,66 @@ impl PagestreamFeMessage {
             ),
         };
 
-        match msg_tag {
-            0 => Ok(PagestreamFeMessage::Exists(PagestreamExistsRequest {
-                hdr: PagestreamRequest {
-                    reqid,
-                    request_lsn,
-                    not_modified_since,
-                },
-                rel: RelTag {
-                    spcnode: body.read_u32::<BigEndian>()?,
+        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,
+                    },
                     dbnode: body.read_u32::<BigEndian>()?,
-                    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(
+                }))
+            }
+            PagestreamFeMessageTag::GetSlruSegment => Ok(PagestreamFeMessage::GetSlruSegment(
                 PagestreamGetSlruSegmentRequest {
                     hdr: PagestreamRequest {
                         reqid,
@@ -1705,7 +1826,21 @@ impl PagestreamFeMessage {
                     segno: body.read_u32::<BigEndian>()?,
                 },
             )),
-            _ => bail!("unknown smgr message tag: {:?}", msg_tag),
+            #[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)?
+                },
+            })),
         }
     }
 }
@@ -1748,6 +1883,15 @@ 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 => {
@@ -1816,6 +1960,18 @@ 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);
+                    }
                 }
             }
         }
@@ -1958,6 +2114,28 @@ 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() {
@@ -1977,6 +2155,25 @@ 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

@@ -31,6 +31,8 @@
 //! - 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};
@@ -48,6 +50,23 @@ 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);
@@ -59,7 +78,7 @@ 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, Debug)]
+#[derive(Clone, Copy, Serialize, Deserialize, Eq, PartialEq, Hash, 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(Debug, Serialize, Deserialize)]
+#[derive(Clone, Debug, Serialize, Deserialize)]
 pub struct XlMultiXactCreate {
     pub mid: MultiXactId,
     /* new MultiXact's ID */
@@ -46,7 +46,7 @@ impl XlMultiXactCreate {
 }
 
 #[repr(C)]
-#[derive(Debug, Serialize, Deserialize)]
+#[derive(Clone, 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(Debug, Serialize, Deserialize)]
+#[derive(Clone, 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(Debug, Serialize, Deserialize)]
+#[derive(Clone, Debug, Serialize, Deserialize)]
 pub struct XlReploriginDrop {
     pub node_id: RepOriginId,
 }
@@ -104,7 +104,7 @@ impl XlReploriginDrop {
 }
 
 #[repr(C)]
-#[derive(Debug, Serialize, Deserialize)]
+#[derive(Clone, Debug, Serialize, Deserialize)]
 pub struct XlReploriginSet {
     pub remote_lsn: Lsn,
     pub node_id: RepOriginId,
@@ -911,7 +911,7 @@ impl XlSmgrCreate {
 }
 
 #[repr(C)]
-#[derive(Debug, Serialize, Deserialize)]
+#[derive(Clone, 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(Debug, Serialize, Deserialize)]
+#[derive(Clone, Debug, Serialize, Deserialize)]
 pub struct XlXactParsedRecord {
     pub xid: TransactionId,
     pub info: u8,

libs/pq_proto/src/lib.rs

@@ -44,7 +44,7 @@ pub struct ProtocolVersion(u32);
 
 impl ProtocolVersion {
     pub const fn new(major: u16, minor: u16) -> Self {
-        Self((major as u32) << 16 | minor as u32)
+        Self(((major as u32) << 16) | minor as u32)
     }
     pub const fn minor(self) -> u16 {
         self.0 as u16

libs/remote_storage/src/config.rs

@@ -43,6 +43,17 @@ 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

@@ -5,9 +5,12 @@ 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

@@ -4,12 +4,15 @@ 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;
-pub const INVALID_TERM: Term = 0;
+/// 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;
 
 /// 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

@@ -0,0 +1,166 @@
+//! 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

@@ -1,5 +1,6 @@
 //! 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;
@@ -11,7 +12,7 @@ use utils::{
     pageserver_feedback::PageserverFeedback,
 };
 
-use crate::{ServerInfo, Term};
+use crate::{membership::Configuration, ServerInfo, Term};
 
 #[derive(Debug, Serialize)]
 pub struct SafekeeperStatus {
@@ -22,13 +23,16 @@ pub struct SafekeeperStatus {
 pub struct TimelineCreateRequest {
     pub tenant_id: TenantId,
     pub timeline_id: TimelineId,
-    pub peer_ids: Option<Vec<NodeId>>,
+    pub mconf: Configuration,
     pub pg_version: u32,
     pub system_id: Option<u64>,
+    // By default WAL_SEGMENT_SIZE
     pub wal_seg_size: Option<u32>,
-    pub commit_lsn: Lsn,
-    // If not passed, it is assigned to the beginning of commit_lsn segment.
-    pub local_start_lsn: Option<Lsn>,
+    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>,
 }
 
 /// Same as TermLsn, but serializes LSN using display serializer
@@ -143,7 +147,13 @@ pub type ConnectionId = u32;
 
 /// Serialize is used only for json'ing in API response. Also used internally.
 #[derive(Debug, Clone, Serialize, Deserialize)]
-pub struct WalSenderState {
+pub enum WalSenderState {
+    Vanilla(VanillaWalSenderState),
+    Interpreted(InterpretedWalSenderState),
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct VanillaWalSenderState {
     pub ttid: TenantTimelineId,
     pub addr: SocketAddr,
     pub conn_id: ConnectionId,
@@ -152,6 +162,17 @@ pub struct WalSenderState {
     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).
@@ -172,6 +193,7 @@ 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,
@@ -186,6 +208,20 @@ 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
 }
@@ -241,3 +277,8 @@ pub struct TimelineTermBumpResponse {
     pub previous_term: u64,
     pub current_term: u64,
 }
+
+#[derive(Debug, Clone, Deserialize, Serialize)]
+pub struct SafekeeperUtilization {
+    pub timeline_count: u64,
+}

libs/utils/src/env.rs

@@ -2,6 +2,7 @@
 
 use std::{fmt::Display, str::FromStr};
 
+/// For types `V` that implement [`FromStr`].
 pub fn var<V, E>(varname: &str) -> Option<V>
 where
     V: FromStr<Err = E>,
@@ -10,7 +11,9 @@ where
     match std::env::var(varname) {
         Ok(s) => Some(
             s.parse()
-                .map_err(|e| format!("failed to parse env var {varname}: {e:#}"))
+                .map_err(|e| {
+                    format!("failed to parse env var {varname} using FromStr::parse: {e:#}")
+                })
                 .unwrap(),
         ),
         Err(std::env::VarError::NotPresent) => None,
@@ -19,3 +22,24 @@ where
         }
     }
 }
+
+/// For types `V` that implement [`serde::de::DeserializeOwned`].
+pub fn var_serde_json_string<V>(varname: &str) -> Option<V>
+where
+    V: serde::de::DeserializeOwned,
+{
+    match std::env::var(varname) {
+        Ok(s) => Some({
+            let value = serde_json::Value::String(s);
+            serde_json::from_value(value)
+                .map_err(|e| {
+                    format!("failed to parse env var {varname} as a serde_json json string: {e:#}")
+                })
+                .unwrap()
+        }),
+        Err(std::env::VarError::NotPresent) => None,
+        Err(std::env::VarError::NotUnicode(_)) => {
+            panic!("env var {varname} is not unicode")
+        }
+    }
+}

libs/utils/src/failpoint_support.rs

@@ -11,31 +11,55 @@ use tracing::*;
 
 /// Declare a failpoint that can use to `pause` failpoint action.
 /// We don't want to block the executor thread, hence, spawn_blocking + await.
+///
+/// Optionally pass a cancellation token, and this failpoint will drop out of
+/// its pause when the cancellation token fires. This is useful for testing
+/// cases where we would like to block something, but test its clean shutdown behavior.
+/// The macro evaluates to a Result in that case, where Ok(()) is the case
+/// where the failpoint was not paused, and Err() is the case where cancellation
+/// token fired while evaluating the failpoint.
+///
+/// Remember to unpause the failpoint in the test; until that happens, one of the
+/// limited number of spawn_blocking thread pool threads is leaked.
 #[macro_export]
 macro_rules! pausable_failpoint {
-    ($name:literal) => {
+    ($name:literal) => {{
         if cfg!(feature = "testing") {
-            tokio::task::spawn_blocking({
-                let current = tracing::Span::current();
+            let cancel = ::tokio_util::sync::CancellationToken::new();
+            let _ = $crate::pausable_failpoint!($name, &cancel);
+        }
+    }};
+    ($name:literal, $cancel:expr) => {{
+        if cfg!(feature = "testing") {
+            let failpoint_fut = ::tokio::task::spawn_blocking({
+                let current = ::tracing::Span::current();
                 move || {
                     let _entered = current.entered();
-                    tracing::info!("at failpoint {}", $name);
-                    fail::fail_point!($name);
+                    ::tracing::info!("at failpoint {}", $name);
+                    ::fail::fail_point!($name);
+                }
+            });
+            let cancel_fut = async move {
+                $cancel.cancelled().await;
+            };
+            ::tokio::select! {
+                res = failpoint_fut => {
+                    res.expect("spawn_blocking");
+                    // continue with execution
+                    Ok(())
+                },
+                _ = cancel_fut => {
+                    Err(())
                 }
-            })
-            .await
-            .expect("spawn_blocking");
-        }
-    };
-    ($name:literal, $cond:expr) => {
-        if cfg!(feature = "testing") {
-            if $cond {
-                pausable_failpoint!($name)
             }
+        } else {
+            Ok(())
         }
-    };
+    }};
 }
 
+pub use pausable_failpoint;
+
 /// use with fail::cfg("$name", "return(2000)")
 ///
 /// The effect is similar to a "sleep(2000)" action, i.e. we sleep for the

libs/utils/src/generation.rs

@@ -112,9 +112,9 @@ impl Serialize for Generation {
             // We should never be asked to serialize a None. Structures
             // that include an optional generation should convert None to an
             // Option<Generation>::None
-            Err(serde::ser::Error::custom(
-                "Tried to serialize invalid generation ({self})",
-            ))
+            Err(serde::ser::Error::custom(format!(
+                "Tried to serialize invalid generation ({self:?})"
+            )))
         }
     }
 }

libs/utils/src/guard_arc_swap.rs

@@ -0,0 +1,54 @@
+//! 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/http/endpoint.rs

@@ -15,7 +15,7 @@ use once_cell::sync::Lazy;
 use regex::Regex;
 use routerify::ext::RequestExt;
 use routerify::{Middleware, RequestInfo, Router, RouterBuilder};
-use tokio::sync::{mpsc, Mutex};
+use tokio::sync::{mpsc, Mutex, Notify};
 use tokio_stream::wrappers::ReceiverStream;
 use tokio_util::io::ReaderStream;
 use tracing::{debug, info, info_span, warn, Instrument};
@@ -350,33 +350,53 @@ pub async fn profile_cpu_handler(req: Request<Body>) -> Result<Response<Body>, A
     };
     let seconds = match parse_query_param(&req, "seconds")? {
         None => 5,
-        Some(seconds @ 1..=30) => seconds,
-        Some(_) => return Err(ApiError::BadRequest(anyhow!("duration must be 1-30 secs"))),
+        Some(seconds @ 1..=60) => seconds,
+        Some(_) => return Err(ApiError::BadRequest(anyhow!("duration must be 1-60 secs"))),
     };
     let frequency_hz = match parse_query_param(&req, "frequency")? {
         None => 99,
         Some(1001..) => return Err(ApiError::BadRequest(anyhow!("frequency must be <=1000 Hz"))),
         Some(frequency) => frequency,
     };
-
-    // Only allow one profiler at a time.
-    static PROFILE_LOCK: Lazy<Mutex<()>> = Lazy::new(|| Mutex::new(()));
-    let _lock = PROFILE_LOCK
-        .try_lock()
-        .map_err(|_| ApiError::Conflict("profiler already running".into()))?;
+    let force: bool = parse_query_param(&req, "force")?.unwrap_or_default();
 
     // Take the profile.
-    let report = tokio::task::spawn_blocking(move || {
+    static PROFILE_LOCK: Lazy<Mutex<()>> = Lazy::new(|| Mutex::new(()));
+    static PROFILE_CANCEL: Lazy<Notify> = Lazy::new(Notify::new);
+
+    let report = {
+        // Only allow one profiler at a time. If force is true, cancel a running profile (e.g. a
+        // Grafana continuous profile). We use a try_lock() loop when cancelling instead of waiting
+        // for a lock(), to avoid races where the notify isn't currently awaited.
+        let _lock = loop {
+            match PROFILE_LOCK.try_lock() {
+                Ok(lock) => break lock,
+                Err(_) if force => PROFILE_CANCEL.notify_waiters(),
+                Err(_) => {
+                    return Err(ApiError::Conflict(
+                        "profiler already running (use ?force=true to cancel it)".into(),
+                    ))
+                }
+            }
+            tokio::time::sleep(Duration::from_millis(1)).await; // don't busy-wait
+        };
+
         let guard = ProfilerGuardBuilder::default()
             .frequency(frequency_hz)
             .blocklist(&["libc", "libgcc", "pthread", "vdso"])
-            .build()?;
-        std::thread::sleep(Duration::from_secs(seconds));
-        guard.report().build()
-    })
-    .await
-    .map_err(|join_err| ApiError::InternalServerError(join_err.into()))?
-    .map_err(|pprof_err| ApiError::InternalServerError(pprof_err.into()))?;
+            .build()
+            .map_err(|err| ApiError::InternalServerError(err.into()))?;
+
+        tokio::select! {
+            _ = tokio::time::sleep(Duration::from_secs(seconds)) => {},
+            _ = PROFILE_CANCEL.notified() => {},
+        };
+
+        guard
+            .report()
+            .build()
+            .map_err(|err| ApiError::InternalServerError(err.into()))?
+    };
 
     // Return the report in the requested format.
     match format {

libs/utils/src/lib.rs

@@ -98,6 +98,8 @@ 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/utils/src/lsn.rs

@@ -260,7 +260,7 @@ impl FromStr for Lsn {
         {
             let left_num = u32::from_str_radix(left, 16).map_err(|_| LsnParseError)?;
             let right_num = u32::from_str_radix(right, 16).map_err(|_| LsnParseError)?;
-            Ok(Lsn((left_num as u64) << 32 | right_num as u64))
+            Ok(Lsn(((left_num as u64) << 32) | right_num as u64))
         } else {
             Err(LsnParseError)
         }

libs/utils/src/sync/gate.rs

@@ -64,6 +64,12 @@ pub struct GateGuard {
     gate: Arc<GateInner>,
 }
 
+impl GateGuard {
+    pub fn try_clone(&self) -> Result<Self, GateError> {
+        Gate::enter_impl(self.gate.clone())
+    }
+}
+
 impl Drop for GateGuard {
     fn drop(&mut self) {
         if self.gate.closing.load(Ordering::Relaxed) {
@@ -107,11 +113,11 @@ impl Gate {
     /// to avoid blocking close() indefinitely: typically types that contain a Gate will
     /// also contain a CancellationToken.
     pub fn enter(&self) -> Result<GateGuard, GateError> {
-        let permit = self
-            .inner
-            .sem
-            .try_acquire()
-            .map_err(|_| GateError::GateClosed)?;
+        Self::enter_impl(self.inner.clone())
+    }
+
+    fn enter_impl(gate: Arc<GateInner>) -> Result<GateGuard, GateError> {
+        let permit = gate.sem.try_acquire().map_err(|_| GateError::GateClosed)?;
 
         // we now have the permit, let's disable the normal raii functionality and leave
         // "returning" the permit to our GateGuard::drop.
@@ -122,7 +128,7 @@ impl Gate {
 
         Ok(GateGuard {
             span_at_enter: tracing::Span::current(),
-            gate: self.inner.clone(),
+            gate,
         })
     }
 
@@ -252,4 +258,39 @@ mod tests {
         // Attempting to enter() is still forbidden
         gate.enter().expect_err("enter should fail finishing close");
     }
+
+    #[tokio::test(start_paused = true)]
+    async fn clone_gate_guard() {
+        let gate = Gate::default();
+        let forever = Duration::from_secs(24 * 7 * 365);
+
+        let guard1 = gate.enter().expect("gate isn't closed");
+
+        let guard2 = guard1.try_clone().expect("gate isn't clsoed");
+
+        let mut close_fut = std::pin::pin!(gate.close());
+
+        tokio::time::timeout(forever, &mut close_fut)
+            .await
+            .unwrap_err();
+
+        // we polled close_fut once, that should prevent all later enters and clones
+        gate.enter().unwrap_err();
+        guard1.try_clone().unwrap_err();
+        guard2.try_clone().unwrap_err();
+
+        // guard2 keeps gate open even if guard1 is closed
+        drop(guard1);
+        tokio::time::timeout(forever, &mut close_fut)
+            .await
+            .unwrap_err();
+
+        drop(guard2);
+
+        // now that the last guard is dropped, closing should complete
+        close_fut.await;
+
+        // entering is still forbidden
+        gate.enter().expect_err("enter should stilll fail");
+    }
 }

libs/wal_decoder/Cargo.toml

@@ -17,10 +17,24 @@ postgres_ffi.workspace = true
 serde.workspace = true
 thiserror.workspace = true
 tokio = { workspace = true, features = ["io-util"] }
-tonic.workspace = true
 tracing.workspace = true
 utils.workspace = true
 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

@@ -0,0 +1,34 @@
+## 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

@@ -0,0 +1,250 @@
+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

@@ -1,6 +1,8 @@
 //! 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};
@@ -14,15 +16,15 @@ use utils::lsn::Lsn;
 
 impl InterpretedWalRecord {
     /// Decode and interpreted raw bytes which represent one Postgres WAL record.
-    /// Data blocks which do not match the provided shard identity are filtered out.
+    /// Data blocks which do not match any of the provided shard identities 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,
-        shard: &ShardIdentity,
+        shards: &[ShardIdentity],
         next_record_lsn: Lsn,
         pg_version: u32,
-    ) -> anyhow::Result<InterpretedWalRecord> {
+    ) -> anyhow::Result<HashMap<ShardIdentity, InterpretedWalRecord>> {
         let mut decoded = DecodedWALRecord::default();
         decode_wal_record(buf, &mut decoded, pg_version)?;
         let xid = decoded.xl_xid;
@@ -33,43 +35,57 @@ impl InterpretedWalRecord {
             FlushUncommittedRecords::No
         };
 
-        let metadata_record =
-            MetadataRecord::from_decoded_filtered(&decoded, shard, next_record_lsn, pg_version)?;
-        let batch = SerializedValueBatch::from_decoded_filtered(
+        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(
             decoded,
-            shard,
+            &mut shard_records,
             next_record_lsn,
             pg_version,
         )?;
 
-        Ok(InterpretedWalRecord {
-            metadata_record,
-            batch,
-            next_record_lsn,
-            flush_uncommitted,
-            xid,
-        })
+        Ok(shard_records)
     }
 }
 
 impl MetadataRecord {
-    /// Builds a metadata record for this WAL record, if any.
+    /// Populates the given `shard_records` with metadata records from this WAL record, if any,
+    /// discarding those belonging to other shards.
     ///
-    /// Only metadata records relevant for the given shard are emitted. Currently, most metadata
+    /// Only metadata records relevant for the given shards is emitted. Currently, most metadata
     /// records are broadcast to all shards for simplicity, but this should be improved.
     fn from_decoded_filtered(
         decoded: &DecodedWALRecord,
-        shard: &ShardIdentity,
+        shard_records: &mut HashMap<ShardIdentity, InterpretedWalRecord>,
         next_record_lsn: Lsn,
         pg_version: u32,
-    ) -> anyhow::Result<Option<MetadataRecord>> {
+    ) -> anyhow::Result<()> {
         // 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 mut metadata_record = match decoded.xl_rmid {
+        let 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)?
             }
@@ -112,41 +128,65 @@ impl MetadataRecord {
         };
 
         // Next, filter the metadata record by shard.
-        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
+        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();
                 }
             }
-            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(metadata_record)
+        Ok(())
     }
 
     fn decode_heapam_record(

libs/wal_decoder/src/models.rs

@@ -45,10 +45,10 @@ pub mod proto {
     #![allow(clippy::derive_partial_eq_without_eq)]
     // The generated ValueMeta has a `len` method generate for its `len` field.
     #![allow(clippy::len_without_is_empty)]
-    tonic::include_proto!("interpreted_wal");
+    include!(concat!(env!("OUT_DIR"), concat!("/interpreted_wal.rs")));
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Copy, Clone, 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)]
+#[derive(Serialize, Deserialize, Clone)]
 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(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 pub enum MetadataRecord {
     Heapam(HeapamRecord),
     Neonrmgr(NeonrmgrRecord),
@@ -123,12 +123,12 @@ pub enum MetadataRecord {
     Replorigin(ReploriginRecord),
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 pub enum HeapamRecord {
     ClearVmBits(ClearVmBits),
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Clone, 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(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 pub enum NeonrmgrRecord {
     ClearVmBits(ClearVmBits),
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 pub enum SmgrRecord {
     Create(SmgrCreate),
     Truncate(XlSmgrTruncate),
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 pub struct SmgrCreate {
     pub rel: RelTag,
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 pub enum DbaseRecord {
     Create(DbaseCreate),
     Drop(DbaseDrop),
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Clone, 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(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 pub struct DbaseDrop {
     pub db_id: Oid,
     pub tablespace_ids: Vec<Oid>,
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 pub enum ClogRecord {
     ZeroPage(ClogZeroPage),
     Truncate(ClogTruncate),
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 pub struct ClogZeroPage {
     pub segno: u32,
     pub rpageno: u32,
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 pub struct ClogTruncate {
     pub pageno: u32,
     pub oldest_xid: TransactionId,
     pub oldest_xid_db: Oid,
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 pub enum XactRecord {
     Commit(XactCommon),
     Abort(XactCommon),
@@ -200,7 +200,7 @@ pub enum XactRecord {
     Prepare(XactPrepare),
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 pub struct XactCommon {
     pub parsed: XlXactParsedRecord,
     pub origin_id: u16,
@@ -209,73 +209,73 @@ pub struct XactCommon {
     pub lsn: Lsn,
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 pub struct XactPrepare {
     pub xl_xid: TransactionId,
     pub data: Bytes,
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 pub enum MultiXactRecord {
     ZeroPage(MultiXactZeroPage),
     Create(XlMultiXactCreate),
     Truncate(XlMultiXactTruncate),
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 pub struct MultiXactZeroPage {
     pub slru_kind: SlruKind,
     pub segno: u32,
     pub rpageno: u32,
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 pub enum RelmapRecord {
     Update(RelmapUpdate),
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 pub struct RelmapUpdate {
     pub update: XlRelmapUpdate,
     pub buf: Bytes,
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 pub enum XlogRecord {
     Raw(RawXlogRecord),
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 pub struct RawXlogRecord {
     pub info: u8,
     pub lsn: Lsn,
     pub buf: Bytes,
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 pub enum LogicalMessageRecord {
     Put(PutLogicalMessage),
     #[cfg(feature = "testing")]
     Failpoint,
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 pub struct PutLogicalMessage {
     pub path: String,
     pub buf: Bytes,
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 pub enum StandbyRecord {
     RunningXacts(StandbyRunningXacts),
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 pub struct StandbyRunningXacts {
     pub oldest_running_xid: TransactionId,
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Clone, 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;
+use std::collections::{BTreeSet, HashMap};
 
 use bytes::{Bytes, BytesMut};
 use pageserver_api::key::rel_block_to_key;
@@ -22,6 +22,8 @@ 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
@@ -30,7 +32,7 @@ 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)]
+#[derive(Serialize, Deserialize, Clone)]
 pub enum ValueMeta {
     Serialized(SerializedValueMeta),
     Observed(ObservedValueMeta),
@@ -77,7 +79,7 @@ impl PartialEq for OrderedValueMeta {
 impl Eq for OrderedValueMeta {}
 
 /// Metadata for a [`Value`] serialized into the batch.
-#[derive(Serialize, Deserialize)]
+#[derive(Serialize, Deserialize, Clone)]
 pub struct SerializedValueMeta {
     pub key: CompactKey,
     pub lsn: Lsn,
@@ -89,14 +91,14 @@ pub struct SerializedValueMeta {
 }
 
 /// Metadata for a [`Value`] observed by the batch
-#[derive(Serialize, Deserialize)]
+#[derive(Serialize, Deserialize, Clone)]
 pub struct ObservedValueMeta {
     pub key: CompactKey,
     pub lsn: Lsn,
 }
 
 /// Batch of serialized [`Value`]s.
-#[derive(Serialize, Deserialize)]
+#[derive(Serialize, Deserialize, Clone)]
 pub struct SerializedValueBatch {
     /// [`Value`]s serialized in EphemeralFile's native format,
     /// ready for disk write by the pageserver
@@ -128,7 +130,8 @@ impl Default for SerializedValueBatch {
 }
 
 impl SerializedValueBatch {
-    /// Build a batch of serialized values from a decoded PG WAL record
+    /// Populates the given `shard_records` with value batches from this WAL record, if any,
+    /// discarding those belonging to other shards.
     ///
     /// 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
@@ -136,21 +139,20 @@ impl SerializedValueBatch {
     /// but absent from the raw buffer [`SerializedValueBatch::raw`]).
     pub(crate) fn from_decoded_filtered(
         decoded: DecodedWALRecord,
-        shard: &ShardIdentity,
+        shard_records: &mut HashMap<ShardIdentity, InterpretedWalRecord>,
         next_record_lsn: Lsn,
         pg_version: u32,
-    ) -> anyhow::Result<SerializedValueBatch> {
-        // First determine how big the buffer needs to be and allocate it up-front.
+    ) -> anyhow::Result<()> {
+        // First determine how big the buffers need to be and allocate it up-front.
         // This duplicates some of the work below, but it's empirically much faster.
-        let estimated_buffer_size = Self::estimate_buffer_size(&decoded, shard, pg_version);
-        let mut buf = Vec::<u8>::with_capacity(estimated_buffer_size);
+        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 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,
@@ -168,99 +170,98 @@ impl SerializedValueBatch {
                 );
             }
 
-            let key_is_local = shard.is_key_local(&key);
+            for (shard, record) in shard_records.iter_mut() {
+                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.
-                    metadata.push(ValueMeta::Observed(ObservedValueMeta {
+                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 {
                         key: key.to_compact(),
                         lsn: next_record_lsn,
-                    }))
-                }
-
-                continue;
+                        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;
             }
-
-            // 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)) {
-            let batch = Self {
-                raw: buf,
-                metadata,
-                max_lsn,
-                len,
-            };
-
-            batch.validate_lsn_order();
-
-            return Ok(batch);
+            // Validate that the batches are correct
+            for record in shard_records.values() {
+                record.batch.validate_lsn_order();
+            }
         }
 
-        Ok(Self {
-            raw: buf,
-            metadata,
-            max_lsn,
-            len,
-        })
+        Ok(())
     }
 
     /// 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"]
+testing = ["fail/failpoints", "pageserver_api/testing", "wal_decoder/testing", "pageserver_client/testing"]
 
 [dependencies]
 anyhow.workspace = true
@@ -16,6 +16,7 @@ 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
@@ -44,6 +45,7 @@ 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
@@ -108,3 +110,11 @@ 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

@@ -0,0 +1,87 @@
+//! 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

@@ -4,6 +4,9 @@ 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

@@ -1,6 +1,9 @@
-use std::pin::Pin;
+use std::sync::{Arc, Mutex};
 
-use futures::SinkExt;
+use futures::{
+    stream::{SplitSink, SplitStream},
+    SinkExt, StreamExt,
+};
 use pageserver_api::{
     models::{
         PagestreamBeMessage, PagestreamFeMessage, PagestreamGetPageRequest,
@@ -10,7 +13,6 @@ 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},
@@ -62,15 +64,28 @@ 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 {
-            copy_both: Box::pin(copy_both),
-            conn_task,
-            cancel_on_client_drop,
+            sink: PagestreamSender {
+                shared: shared.clone(),
+                sink,
+            },
+            stream: PagestreamReceiver {
+                shared: shared.clone(),
+                stream,
+            },
+            shared,
         })
     }
 
@@ -97,7 +112,28 @@ impl Client {
 
 /// Create using [`Client::pagestream`].
 pub struct PagestreamClient {
-    copy_both: Pin<Box<tokio_postgres::CopyBothDuplex<bytes::Bytes>>>,
+    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 {
     cancel_on_client_drop: Option<tokio_util::sync::DropGuard>,
     conn_task: JoinHandle<()>,
 }
@@ -110,11 +146,11 @@ pub struct RelTagBlockNo {
 impl PagestreamClient {
     pub async fn shutdown(self) {
         let Self {
-            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`.
+            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`.
         // 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).
         //
@@ -131,27 +167,77 @@ 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 _ = cancel_conn_task.unwrap();
+        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();
         conn_task.await.unwrap();
-        drop(copy_both);
+
+        // Now drop the split copy_both.
+        drop(sink);
+        drop(stream);
+    }
+
+    pub fn split(self) -> (PagestreamSender, PagestreamReceiver) {
+        let Self {
+            shared: _,
+            sink,
+            stream,
+        } = self;
+        (sink, stream)
     }
 
     pub async fn getpage(
         &mut self,
         req: PagestreamGetPageRequest,
     ) -> anyhow::Result<PagestreamGetPageResponse> {
-        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));
+        self.getpage_send(req).await?;
+        self.getpage_recv().await
+    }
 
-        self.copy_both.send_all(&mut req).await?;
+    pub async fn getpage_send(&mut self, req: PagestreamGetPageRequest) -> anyhow::Result<()> {
+        self.sink.getpage_send(req).await
+    }
 
-        let next: Option<Result<bytes::Bytes, _>> = self.copy_both.next().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: bytes::Bytes = next.unwrap()?;
+        PagestreamBeMessage::deserialize(next)
+    }
 
-        let msg = PagestreamBeMessage::deserialize(next)?;
-        match msg {
+    pub async fn getpage_recv(&mut self) -> anyhow::Result<PagestreamGetPageResponse> {
+        let next: PagestreamBeMessage = self.recv().await?;
+        match next {
             PagestreamBeMessage::GetPage(p) => Ok(p),
             PagestreamBeMessage::Error(e) => anyhow::bail!("Error: {:?}", e),
             PagestreamBeMessage::Exists(_)
@@ -160,7 +246,14 @@ impl PagestreamClient {
             | PagestreamBeMessage::GetSlruSegment(_) => {
                 anyhow::bail!(
                     "unexpected be message kind in response to getpage request: {}",
-                    msg.kind()
+                    next.kind()
+                )
+            }
+            #[cfg(feature = "testing")]
+            PagestreamBeMessage::Test(_) => {
+                anyhow::bail!(
+                    "unexpected be message kind in response to getpage request: {}",
+                    next.kind()
                 )
             }
         }

pageserver/ctl/Cargo.toml

@@ -8,9 +8,11 @@ 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

@@ -9,7 +9,9 @@ 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},
@@ -64,6 +66,7 @@ enum Commands {
     Layer(LayerCmd),
     /// Debug print a hex key found from logs
     Key(key::DescribeKeyCommand),
+    PageTrace(PageTraceCmd),
 }
 
 /// Read and update pageserver metadata file
@@ -183,6 +186,7 @@ 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

@@ -0,0 +1,73 @@
+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/pagebench/src/cmd/getpage_latest_lsn.rs

@@ -13,7 +13,7 @@ use rand::prelude::*;
 use tokio::task::JoinSet;
 use tracing::info;
 
-use std::collections::HashSet;
+use std::collections::{HashSet, VecDeque};
 use std::future::Future;
 use std::num::NonZeroUsize;
 use std::pin::Pin;
@@ -63,6 +63,10 @@ pub(crate) struct Args {
     #[clap(long)]
     set_io_mode: Option<pageserver_api::models::virtual_file::IoMode>,
 
+    /// Queue depth generated in each client.
+    #[clap(long, default_value = "1")]
+    queue_depth: NonZeroUsize,
+
     targets: Option<Vec<TenantTimelineId>>,
 }
 
@@ -298,6 +302,7 @@ async fn main_impl(
             start_work_barrier.wait().await;
             let client_start = Instant::now();
             let mut ticks_processed = 0;
+            let mut inflight = VecDeque::new();
             while !cancel.is_cancelled() {
                 // Detect if a request took longer than the RPS rate
                 if let Some(period) = &rps_period {
@@ -311,31 +316,37 @@ async fn main_impl(
                     ticks_processed = periods_passed_until_now;
                 }
 
-                let start = Instant::now();
-                let req = {
-                    let mut rng = rand::thread_rng();
-                    let r = &ranges[weights.sample(&mut rng)];
-                    let key: i128 = rng.gen_range(r.start..r.end);
-                    let key = Key::from_i128(key);
-                    assert!(key.is_rel_block_key());
-                    let (rel_tag, block_no) = key
-                        .to_rel_block()
-                        .expect("we filter non-rel-block keys out above");
-                    PagestreamGetPageRequest {
-                        hdr: PagestreamRequest {
-                            reqid: 0,
-                            request_lsn: if rng.gen_bool(args.req_latest_probability) {
-                                Lsn::MAX
-                            } else {
-                                r.timeline_lsn
+                while inflight.len() < args.queue_depth.get() {
+                    let start = Instant::now();
+                    let req = {
+                        let mut rng = rand::thread_rng();
+                        let r = &ranges[weights.sample(&mut rng)];
+                        let key: i128 = rng.gen_range(r.start..r.end);
+                        let key = Key::from_i128(key);
+                        assert!(key.is_rel_block_key());
+                        let (rel_tag, block_no) = key
+                            .to_rel_block()
+                            .expect("we filter non-rel-block keys out above");
+                        PagestreamGetPageRequest {
+                            hdr: PagestreamRequest {
+                                reqid: 0,
+                                request_lsn: if rng.gen_bool(args.req_latest_probability) {
+                                    Lsn::MAX
+                                } else {
+                                    r.timeline_lsn
+                                },
+                                not_modified_since: r.timeline_lsn,
                             },
-                            not_modified_since: r.timeline_lsn,
-                        },
-                        rel: rel_tag,
-                        blkno: block_no,
-                    }
-                };
-                client.getpage(req).await.unwrap();
+                            rel: rel_tag,
+                            blkno: block_no,
+                        }
+                    };
+                    client.getpage_send(req).await.unwrap();
+                    inflight.push_back(start);
+                }
+
+                let start = inflight.pop_front().unwrap();
+                client.getpage_recv().await.unwrap();
                 let end = Instant::now();
                 live_stats.request_done();
                 ticks_processed += 1;

pageserver/src/basebackup.rs

@@ -25,6 +25,7 @@ use tokio_tar::{Builder, EntryType, Header};
 
 use crate::context::RequestContext;
 use crate::pgdatadir_mapping::Version;
+use crate::tenant::storage_layer::IoConcurrency;
 use crate::tenant::Timeline;
 use pageserver_api::reltag::{RelTag, SlruKind};
 
@@ -123,6 +124,13 @@ where
         full_backup,
         replica,
         ctx,
+        io_concurrency: IoConcurrency::spawn_from_conf(
+            timeline.conf,
+            timeline
+                .gate
+                .enter()
+                .map_err(|e| BasebackupError::Server(e.into()))?,
+        ),
     };
     basebackup
         .send_tarball()
@@ -144,6 +152,7 @@ where
     full_backup: bool,
     replica: bool,
     ctx: &'a RequestContext,
+    io_concurrency: IoConcurrency,
 }
 
 /// A sink that accepts SLRU blocks ordered by key and forwards
@@ -303,7 +312,7 @@ where
             for part in slru_partitions.parts {
                 let blocks = self
                     .timeline
-                    .get_vectored(part, self.lsn, self.ctx)
+                    .get_vectored(part, self.lsn, self.io_concurrency.clone(), self.ctx)
                     .await
                     .map_err(|e| BasebackupError::Server(e.into()))?;
 
@@ -358,7 +367,7 @@ where
         let start_time = Instant::now();
         let aux_files = self
             .timeline
-            .list_aux_files(self.lsn, self.ctx)
+            .list_aux_files(self.lsn, self.ctx, self.io_concurrency.clone())
             .await
             .map_err(|e| BasebackupError::Server(e.into()))?;
         let aux_scan_time = start_time.elapsed();
@@ -422,7 +431,7 @@ where
         }
         let repl_origins = self
             .timeline
-            .get_replorigins(self.lsn, self.ctx)
+            .get_replorigins(self.lsn, self.ctx, self.io_concurrency.clone())
             .await
             .map_err(|e| BasebackupError::Server(e.into()))?;
         let n_origins = repl_origins.len();
@@ -489,7 +498,13 @@ where
             for blknum in startblk..endblk {
                 let img = self
                     .timeline
-                    .get_rel_page_at_lsn(src, blknum, Version::Lsn(self.lsn), self.ctx)
+                    .get_rel_page_at_lsn(
+                        src,
+                        blknum,
+                        Version::Lsn(self.lsn),
+                        self.ctx,
+                        self.io_concurrency.clone(),
+                    )
                     .await
                     .map_err(|e| BasebackupError::Server(e.into()))?;
                 segment_data.extend_from_slice(&img[..]);

pageserver/src/bin/pageserver.rs

@@ -53,10 +53,12 @@ project_build_tag!(BUILD_TAG);
 #[global_allocator]
 static GLOBAL: tikv_jemallocator::Jemalloc = tikv_jemallocator::Jemalloc;
 
-/// Configure jemalloc to sample allocations for profiles every 1 MB (1 << 20).
+/// Configure jemalloc to profile heap allocations by sampling stack traces every 2 MB (1 << 21).
+/// This adds roughly 3% overhead for allocations on average, which is acceptable considering
+/// performance-sensitive code will avoid allocations as far as possible anyway.
 #[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";
+pub static malloc_conf: &[u8] = b"prof:true,prof_active:true,lg_prof_sample:21\0";
 
 const PID_FILE_NAME: &str = "pageserver.pid";
 
@@ -133,6 +135,7 @@ fn main() -> anyhow::Result<()> {
     info!(?conf.virtual_file_io_mode, "starting with virtual_file IO mode");
     info!(?conf.wal_receiver_protocol, "starting with WAL receiver protocol");
     info!(?conf.page_service_pipelining, "starting with page service pipelining config");
+    info!(?conf.get_vectored_concurrent_io, "starting with get_vectored IO concurrency config");
 
     // The tenants directory contains all the pageserver local disk state.
     // Create if not exists and make sure all the contents are durable before proceeding.

pageserver/src/bin/test_helper_slow_client_reads.rs

@@ -0,0 +1,65 @@
+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/config.rs

@@ -191,6 +191,8 @@ pub struct PageServerConf {
     pub wal_receiver_protocol: PostgresClientProtocol,
 
     pub page_service_pipelining: pageserver_api::config::PageServicePipeliningConfig,
+
+    pub get_vectored_concurrent_io: pageserver_api::config::GetVectoredConcurrentIo,
 }
 
 /// Token for authentication to safekeepers
@@ -352,6 +354,7 @@ impl PageServerConf {
             no_sync,
             wal_receiver_protocol,
             page_service_pipelining,
+            get_vectored_concurrent_io,
         } = config_toml;
 
         let mut conf = PageServerConf {
@@ -396,6 +399,7 @@ impl PageServerConf {
             import_pgdata_aws_endpoint_url,
             wal_receiver_protocol,
             page_service_pipelining,
+            get_vectored_concurrent_io,
 
             // ------------------------------------------------------------
             // fields that require additional validation or custom handling

pageserver/src/http/routes.rs

@@ -27,6 +27,7 @@ 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;
@@ -51,7 +52,9 @@ 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::*;
@@ -81,6 +84,7 @@ use crate::tenant::remote_timeline_client::list_remote_tenant_shards;
 use crate::tenant::remote_timeline_client::list_remote_timelines;
 use crate::tenant::secondary::SecondaryController;
 use crate::tenant::size::ModelInputs;
+use crate::tenant::storage_layer::IoConcurrency;
 use crate::tenant::storage_layer::LayerAccessStatsReset;
 use crate::tenant::storage_layer::LayerName;
 use crate::tenant::timeline::import_pgdata;
@@ -1521,6 +1525,71 @@ 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.
@@ -2870,8 +2939,15 @@ async fn list_aux_files(
         active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
             .await?;
 
+    let io_concurrency = IoConcurrency::spawn_from_conf(
+        state.conf,
+        timeline.gate.enter().map_err(|_| ApiError::Cancelled)?,
+    );
+
     let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
-    let files = timeline.list_aux_files(body.lsn, &ctx).await?;
+    let files = timeline
+        .list_aux_files(body.lsn, &ctx, io_concurrency)
+        .await?;
     json_response(StatusCode::OK, files)
 }
 
@@ -3479,6 +3555,10 @@ 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

@@ -278,6 +278,8 @@ 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);
@@ -314,10 +316,12 @@ async fn import_wal(
             if let Some((lsn, recdata)) = waldecoder.poll_decode()? {
                 let interpreted = InterpretedWalRecord::from_bytes_filtered(
                     recdata,
-                    tline.get_shard_identity(),
+                    &shard,
                     lsn,
                     tline.pg_version,
-                )?;
+                )?
+                .remove(tline.get_shard_identity())
+                .unwrap();
 
                 walingest
                     .ingest_record(interpreted, &mut modification, ctx)
@@ -411,6 +415,7 @@ 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);
@@ -459,10 +464,12 @@ pub async fn import_wal_from_tar(
             if let Some((lsn, recdata)) = waldecoder.poll_decode()? {
                 let interpreted = InterpretedWalRecord::from_bytes_filtered(
                     recdata,
-                    tline.get_shard_identity(),
+                    &shard,
                     lsn,
                     tline.pg_version,
-                )?;
+                )?
+                .remove(tline.get_shard_identity())
+                .unwrap();
 
                 walingest
                     .ingest_record(interpreted, &mut modification, ctx)

pageserver/src/metrics.rs

@@ -3,7 +3,7 @@ use metrics::{
     register_counter_vec, register_gauge_vec, register_histogram, register_histogram_vec,
     register_int_counter, register_int_counter_pair_vec, register_int_counter_vec,
     register_int_gauge, register_int_gauge_vec, register_uint_gauge, register_uint_gauge_vec,
-    Counter, CounterVec, Gauge, GaugeVec, Histogram, HistogramVec, IntCounter, IntCounterPair,
+    Counter, CounterVec, GaugeVec, Histogram, HistogramVec, IntCounter, IntCounterPair,
     IntCounterPairVec, IntCounterVec, IntGauge, IntGaugeVec, UIntGauge, UIntGaugeVec,
 };
 use once_cell::sync::Lazy;
@@ -38,6 +38,9 @@ pub(crate) enum StorageTimeOperation {
     #[strum(serialize = "layer flush")]
     LayerFlush,
 
+    #[strum(serialize = "layer flush delay")]
+    LayerFlushDelay,
+
     #[strum(serialize = "compact")]
     Compact,
 
@@ -100,71 +103,30 @@ pub(crate) static VEC_READ_NUM_LAYERS_VISITED: Lazy<Histogram> = Lazy::new(|| {
     .expect("failed to define a metric")
 });
 
-// Metrics collected on operations on the storage repository.
-#[derive(
-    Clone, Copy, enum_map::Enum, strum_macros::EnumString, strum_macros::Display, IntoStaticStr,
-)]
-pub(crate) enum GetKind {
-    Singular,
-    Vectored,
-}
-
-pub(crate) struct ReconstructTimeMetrics {
-    singular: Histogram,
-    vectored: Histogram,
-}
-
-pub(crate) static RECONSTRUCT_TIME: Lazy<ReconstructTimeMetrics> = Lazy::new(|| {
-    let inner = register_histogram_vec!(
-        "pageserver_getpage_reconstruct_seconds",
-        "Time spent in reconstruct_value (reconstruct a page from deltas)",
-        &["get_kind"],
-        CRITICAL_OP_BUCKETS.into(),
+pub(crate) static CONCURRENT_INITDBS: Lazy<UIntGauge> = Lazy::new(|| {
+    register_uint_gauge!(
+        "pageserver_concurrent_initdb",
+        "Number of initdb processes running"
     )
-    .expect("failed to define a metric");
-
-    ReconstructTimeMetrics {
-        singular: inner.with_label_values(&[GetKind::Singular.into()]),
-        vectored: inner.with_label_values(&[GetKind::Vectored.into()]),
-    }
+    .expect("failed to define a metric")
 });
 
-impl ReconstructTimeMetrics {
-    pub(crate) fn for_get_kind(&self, get_kind: GetKind) -> &Histogram {
-        match get_kind {
-            GetKind::Singular => &self.singular,
-            GetKind::Vectored => &self.vectored,
-        }
-    }
-}
-
-pub(crate) struct ReconstructDataTimeMetrics {
-    singular: Histogram,
-    vectored: Histogram,
-}
-
-impl ReconstructDataTimeMetrics {
-    pub(crate) fn for_get_kind(&self, get_kind: GetKind) -> &Histogram {
-        match get_kind {
-            GetKind::Singular => &self.singular,
-            GetKind::Vectored => &self.vectored,
-        }
-    }
-}
-
-pub(crate) static GET_RECONSTRUCT_DATA_TIME: Lazy<ReconstructDataTimeMetrics> = Lazy::new(|| {
-    let inner = register_histogram_vec!(
-        "pageserver_getpage_get_reconstruct_data_seconds",
-        "Time spent in get_reconstruct_value_data",
-        &["get_kind"],
-        CRITICAL_OP_BUCKETS.into(),
+pub(crate) static INITDB_SEMAPHORE_ACQUISITION_TIME: Lazy<Histogram> = Lazy::new(|| {
+    register_histogram!(
+        "pageserver_initdb_semaphore_seconds_global",
+        "Time spent getting a permit from the global initdb semaphore",
+        STORAGE_OP_BUCKETS.into()
     )
-    .expect("failed to define a metric");
+    .expect("failed to define metric")
+});
 
-    ReconstructDataTimeMetrics {
-        singular: inner.with_label_values(&[GetKind::Singular.into()]),
-        vectored: inner.with_label_values(&[GetKind::Vectored.into()]),
-    }
+pub(crate) static INITDB_RUN_TIME: Lazy<Histogram> = Lazy::new(|| {
+    register_histogram!(
+        "pageserver_initdb_seconds_global",
+        "Time spent performing initdb",
+        STORAGE_OP_BUCKETS.into()
+    )
+    .expect("failed to define metric")
 });
 
 pub(crate) struct GetVectoredLatency {
@@ -436,15 +398,6 @@ pub(crate) static WAIT_LSN_TIME: Lazy<Histogram> = Lazy::new(|| {
     .expect("failed to define a metric")
 });
 
-static FLUSH_WAIT_UPLOAD_TIME: Lazy<GaugeVec> = Lazy::new(|| {
-    register_gauge_vec!(
-        "pageserver_flush_wait_upload_seconds",
-        "Time spent waiting for preceding uploads during layer flush",
-        &["tenant_id", "shard_id", "timeline_id"]
-    )
-    .expect("failed to define a metric")
-});
-
 static LAST_RECORD_LSN: Lazy<IntGaugeVec> = Lazy::new(|| {
     register_int_gauge_vec!(
         "pageserver_last_record_lsn",
@@ -1224,117 +1177,189 @@ 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,
     },
-    ThrottleDoneExecutionStarting {
-        received_at: Instant,
+    Throttling {
         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) {
-        self.smgr_op_end();
+        // 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);
     }
 }
 
@@ -1345,12 +1370,12 @@ 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());
@@ -1391,9 +1416,10 @@ 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],
@@ -1405,6 +1431,7 @@ 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(|| {
@@ -1610,7 +1637,11 @@ 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) -> Self {
+    pub(crate) fn new(
+        tenant_shard_id: &TenantShardId,
+        timeline_id: &TimelineId,
+        pagestream_throttle_metrics: Arc<tenant_throttling::Pagestream>,
+    ) -> 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();
@@ -1671,6 +1702,7 @@ 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 {
@@ -1686,88 +1718,24 @@ 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(|| {
@@ -2509,12 +2477,19 @@ impl StorageTimeMetricsTimer {
         }
     }
 
-    /// Record the time from creation to now.
-    pub fn stop_and_record(self) {
-        let duration = self.start.elapsed().as_secs_f64();
-        self.metrics.timeline_sum.inc_by(duration);
+    /// Returns the elapsed duration of the timer.
+    pub fn elapsed(&self) -> Duration {
+        self.start.elapsed()
+    }
+
+    /// Record the time from creation to now and return it.
+    pub fn stop_and_record(self) -> Duration {
+        let duration = self.elapsed();
+        let seconds = duration.as_secs_f64();
+        self.metrics.timeline_sum.inc_by(seconds);
         self.metrics.timeline_count.inc();
-        self.metrics.global_histogram.observe(duration);
+        self.metrics.global_histogram.observe(seconds);
+        duration
     }
 
     /// Turns this timer into a timer, which will always record -- usually this means recording
@@ -2534,6 +2509,13 @@ impl Drop for AlwaysRecordingStorageTimeMetricsTimer {
     }
 }
 
+impl AlwaysRecordingStorageTimeMetricsTimer {
+    /// Returns the elapsed duration of the timer.
+    pub fn elapsed(&self) -> Duration {
+        self.0.as_ref().expect("not dropped yet").elapsed()
+    }
+}
+
 /// Timing facilities for an globally histogrammed metric, which is supported by per tenant and
 /// timeline total sum and count.
 #[derive(Clone, Debug)]
@@ -2586,7 +2568,7 @@ pub(crate) struct TimelineMetrics {
     shard_id: String,
     timeline_id: String,
     pub flush_time_histo: StorageTimeMetrics,
-    pub flush_wait_upload_time_gauge: Gauge,
+    pub flush_delay_histo: StorageTimeMetrics,
     pub compact_time_histo: StorageTimeMetrics,
     pub create_images_time_histo: StorageTimeMetrics,
     pub logical_size_histo: StorageTimeMetrics,
@@ -2632,9 +2614,12 @@ impl TimelineMetrics {
             &shard_id,
             &timeline_id,
         );
-        let flush_wait_upload_time_gauge = FLUSH_WAIT_UPLOAD_TIME
-            .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
-            .unwrap();
+        let flush_delay_histo = StorageTimeMetrics::new(
+            StorageTimeOperation::LayerFlushDelay,
+            &tenant_id,
+            &shard_id,
+            &timeline_id,
+        );
         let compact_time_histo = StorageTimeMetrics::new(
             StorageTimeOperation::Compact,
             &tenant_id,
@@ -2780,7 +2765,7 @@ impl TimelineMetrics {
             shard_id,
             timeline_id,
             flush_time_histo,
-            flush_wait_upload_time_gauge,
+            flush_delay_histo,
             compact_time_histo,
             create_images_time_histo,
             logical_size_histo,
@@ -2830,14 +2815,6 @@ impl TimelineMetrics {
         self.resident_physical_size_gauge.get()
     }
 
-    pub(crate) fn flush_wait_upload_time_gauge_add(&self, duration: f64) {
-        self.flush_wait_upload_time_gauge.add(duration);
-        crate::metrics::FLUSH_WAIT_UPLOAD_TIME
-            .get_metric_with_label_values(&[&self.tenant_id, &self.shard_id, &self.timeline_id])
-            .unwrap()
-            .add(duration);
-    }
-
     pub(crate) fn shutdown(&self) {
         let was_shutdown = self
             .shutdown
@@ -2855,7 +2832,6 @@ impl TimelineMetrics {
         let shard_id = &self.shard_id;
         let _ = LAST_RECORD_LSN.remove_label_values(&[tenant_id, shard_id, timeline_id]);
         let _ = DISK_CONSISTENT_LSN.remove_label_values(&[tenant_id, shard_id, timeline_id]);
-        let _ = FLUSH_WAIT_UPLOAD_TIME.remove_label_values(&[tenant_id, shard_id, timeline_id]);
         let _ = STANDBY_HORIZON.remove_label_values(&[tenant_id, shard_id, timeline_id]);
         {
             RESIDENT_PHYSICAL_SIZE_GLOBAL.sub(self.resident_physical_size_get());
@@ -3563,9 +3539,7 @@ pub(crate) mod tenant_throttling {
     use once_cell::sync::Lazy;
     use utils::shard::TenantShardId;
 
-    use crate::tenant::{self};
-
-    struct GlobalAndPerTenantIntCounter {
+    pub(crate) struct GlobalAndPerTenantIntCounter {
         global: IntCounter,
         per_tenant: IntCounter,
     }
@@ -3583,10 +3557,10 @@ pub(crate) mod tenant_throttling {
     }
 
     pub(crate) struct Metrics<const KIND: usize> {
-        count_accounted_start: GlobalAndPerTenantIntCounter,
-        count_accounted_finish: GlobalAndPerTenantIntCounter,
-        wait_time: GlobalAndPerTenantIntCounter,
-        count_throttled: GlobalAndPerTenantIntCounter,
+        pub(super) count_accounted_start: GlobalAndPerTenantIntCounter,
+        pub(super) count_accounted_finish: GlobalAndPerTenantIntCounter,
+        pub(super) wait_time: GlobalAndPerTenantIntCounter,
+        pub(super) count_throttled: GlobalAndPerTenantIntCounter,
     }
 
     static COUNT_ACCOUNTED_START: Lazy<metrics::IntCounterVec> = Lazy::new(|| {
@@ -3721,26 +3695,6 @@ 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 {
@@ -3900,7 +3854,6 @@ pub fn preinitialize_metrics(conf: &'static PageServerConf) {
     });
 
     // Custom
-    Lazy::force(&RECONSTRUCT_TIME);
     Lazy::force(&BASEBACKUP_QUERY_TIME);
     Lazy::force(&COMPUTE_COMMANDS_COUNTERS);
     Lazy::force(&tokio_epoll_uring::THREAD_LOCAL_METRICS_STORAGE);

pageserver/src/page_service.rs

@@ -39,6 +39,7 @@ use tokio::io::{AsyncWriteExt, BufWriter};
 use tokio::task::JoinHandle;
 use tokio_util::sync::CancellationToken;
 use tracing::*;
+use utils::sync::gate::{Gate, GateGuard};
 use utils::sync::spsc_fold;
 use utils::{
     auth::{Claims, Scope, SwappableJwtAuth},
@@ -61,12 +62,14 @@ use crate::task_mgr::{self, COMPUTE_REQUEST_RUNTIME};
 use crate::tenant::mgr::ShardSelector;
 use crate::tenant::mgr::TenantManager;
 use crate::tenant::mgr::{GetActiveTenantError, GetTenantError, ShardResolveResult};
+use crate::tenant::storage_layer::IoConcurrency;
 use crate::tenant::timeline::{self, WaitLsnError};
 use crate::tenant::GetTimelineError;
 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;
@@ -89,6 +92,7 @@ pub struct Listener {
 pub struct Connections {
     cancel: CancellationToken,
     tasks: tokio::task::JoinSet<ConnectionHandlerResult>,
+    gate: Gate,
 }
 
 pub fn spawn(
@@ -109,6 +113,7 @@ pub fn spawn(
     let task = COMPUTE_REQUEST_RUNTIME.spawn(task_mgr::exit_on_panic_or_error(
         "libpq listener",
         libpq_listener_main(
+            conf,
             tenant_manager,
             pg_auth,
             tcp_listener,
@@ -133,11 +138,16 @@ impl Listener {
 }
 impl Connections {
     pub(crate) async fn shutdown(self) {
-        let Self { cancel, mut tasks } = self;
+        let Self {
+            cancel,
+            mut tasks,
+            gate,
+        } = self;
         cancel.cancel();
         while let Some(res) = tasks.join_next().await {
             Self::handle_connection_completion(res);
         }
+        gate.close().await;
     }
 
     fn handle_connection_completion(res: Result<anyhow::Result<()>, tokio::task::JoinError>) {
@@ -157,7 +167,9 @@ impl Connections {
 /// Returns Ok(()) upon cancellation via `cancel`, returning the set of
 /// open connections.
 ///
+#[allow(clippy::too_many_arguments)]
 pub async fn libpq_listener_main(
+    conf: &'static PageServerConf,
     tenant_manager: Arc<TenantManager>,
     auth: Option<Arc<SwappableJwtAuth>>,
     listener: tokio::net::TcpListener,
@@ -167,9 +179,15 @@ pub async fn libpq_listener_main(
     listener_cancel: CancellationToken,
 ) -> Connections {
     let connections_cancel = CancellationToken::new();
+    let connections_gate = Gate::default();
     let mut connection_handler_tasks = tokio::task::JoinSet::default();
 
     loop {
+        let gate_guard = match connections_gate.enter() {
+            Ok(guard) => guard,
+            Err(_) => break,
+        };
+
         let accepted = tokio::select! {
             biased;
             _ = listener_cancel.cancelled() => break,
@@ -189,6 +207,7 @@ pub async fn libpq_listener_main(
                 let connection_ctx = listener_ctx
                     .detached_child(TaskKind::PageRequestHandler, DownloadBehavior::Download);
                 connection_handler_tasks.spawn(page_service_conn_main(
+                    conf,
                     tenant_manager.clone(),
                     local_auth,
                     socket,
@@ -196,6 +215,7 @@ pub async fn libpq_listener_main(
                     pipelining_config.clone(),
                     connection_ctx,
                     connections_cancel.child_token(),
+                    gate_guard,
                 ));
             }
             Err(err) => {
@@ -210,13 +230,16 @@ pub async fn libpq_listener_main(
     Connections {
         cancel: connections_cancel,
         tasks: connection_handler_tasks,
+        gate: connections_gate,
     }
 }
 
 type ConnectionHandlerResult = anyhow::Result<()>;
 
 #[instrument(skip_all, fields(peer_addr))]
+#[allow(clippy::too_many_arguments)]
 async fn page_service_conn_main(
+    conf: &'static PageServerConf,
     tenant_manager: Arc<TenantManager>,
     auth: Option<Arc<SwappableJwtAuth>>,
     socket: tokio::net::TcpStream,
@@ -224,6 +247,7 @@ async fn page_service_conn_main(
     pipelining_config: PageServicePipeliningConfig,
     connection_ctx: RequestContext,
     cancel: CancellationToken,
+    gate_guard: GateGuard,
 ) -> ConnectionHandlerResult {
     let _guard = LIVE_CONNECTIONS
         .with_label_values(&["page_service"])
@@ -273,11 +297,13 @@ async fn page_service_conn_main(
     // But it's in a shared crate, so, we store connection_ctx inside PageServerHandler
     // and create the per-query context in process_query ourselves.
     let mut conn_handler = PageServerHandler::new(
+        conf,
         tenant_manager,
         auth,
         pipelining_config,
         connection_ctx,
         cancel.clone(),
+        gate_guard,
     );
     let pgbackend = PostgresBackend::new_from_io(socket, peer_addr, auth_type, None)?;
 
@@ -309,6 +335,7 @@ async fn page_service_conn_main(
 }
 
 struct PageServerHandler {
+    conf: &'static PageServerConf,
     auth: Option<Arc<SwappableJwtAuth>>,
     claims: Option<Claims>,
 
@@ -324,6 +351,8 @@ struct PageServerHandler {
     timeline_handles: Option<TimelineHandles>,
 
     pipelining_config: PageServicePipeliningConfig,
+
+    gate_guard: GateGuard,
 }
 
 struct TimelineHandles {
@@ -554,37 +583,52 @@ 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::Handle<TenantManagerTypes>,
+        shard: timeline::handle::WeakHandle<TenantManagerTypes>,
         req: models::PagestreamExistsRequest,
     },
     Nblocks {
         span: Span,
         timer: SmgrOpTimer,
-        shard: timeline::handle::Handle<TenantManagerTypes>,
+        shard: timeline::handle::WeakHandle<TenantManagerTypes>,
         req: models::PagestreamNblocksRequest,
     },
     GetPage {
         span: Span,
-        shard: timeline::handle::Handle<TenantManagerTypes>,
+        shard: timeline::handle::WeakHandle<TenantManagerTypes>,
         effective_request_lsn: Lsn,
         pages: smallvec::SmallVec<[BatchedGetPageRequest; 1]>,
     },
     DbSize {
         span: Span,
         timer: SmgrOpTimer,
-        shard: timeline::handle::Handle<TenantManagerTypes>,
+        shard: timeline::handle::WeakHandle<TenantManagerTypes>,
         req: models::PagestreamDbSizeRequest,
     },
     GetSlruSegment {
         span: Span,
         timer: SmgrOpTimer,
-        shard: timeline::handle::Handle<TenantManagerTypes>,
+        shard: timeline::handle::WeakHandle<TenantManagerTypes>,
         req: models::PagestreamGetSlruSegmentRequest,
     },
+    #[cfg(feature = "testing")]
+    Test {
+        span: Span,
+        shard: timeline::handle::WeakHandle<TenantManagerTypes>,
+        requests: Vec<BatchedTestRequest>,
+    },
     RespondError {
         span: Span,
         error: BatchedPageStreamError,
@@ -592,61 +636,49 @@ enum BatchedFeMessage {
 }
 
 impl BatchedFeMessage {
-    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)),
-                )
+    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);
             }
-            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);
+            BatchedFeMessage::GetPage { pages, .. } => {
+                for page in pages {
+                    page.timer.observe_execution_start(at);
+                }
             }
-            _ = cancel.cancelled() => {
-                return Err(QueryError::Shutdown);
+            #[cfg(feature = "testing")]
+            BatchedFeMessage::Test { requests, .. } => {
+                for req in requests {
+                    req.timer.observe_execution_start(at);
+                }
             }
-        };
-        for timer in timers {
-            timer.observe_throttle_done_execution_starting(&throttled);
+            BatchedFeMessage::RespondError { .. } => {}
         }
-        Ok(())
     }
 }
 
 impl PageServerHandler {
     pub fn new(
+        conf: &'static PageServerConf,
         tenant_manager: Arc<TenantManager>,
         auth: Option<Arc<SwappableJwtAuth>>,
         pipelining_config: PageServicePipeliningConfig,
         connection_ctx: RequestContext,
         cancel: CancellationToken,
+        gate_guard: GateGuard,
     ) -> Self {
         PageServerHandler {
+            conf,
             auth,
             claims: None,
             connection_ctx,
             timeline_handles: Some(TimelineHandles::new(tenant_manager)),
             cancel,
             pipelining_config,
+            gate_guard,
         }
     }
 
@@ -720,6 +752,26 @@ 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);
@@ -727,13 +779,16 @@ impl PageServerHandler {
                     .get(tenant_id, timeline_id, ShardSelector::Zero)
                     .instrument(span.clone()) // sets `shard_id` field
                     .await?;
-                let timer = shard
-                    .query_metrics
-                    .start_smgr_op(metrics::SmgrQueryType::GetRelExists, received_at);
+                let timer = record_op_start_and_throttle(
+                    &shard,
+                    metrics::SmgrQueryType::GetRelExists,
+                    received_at,
+                )
+                .await?;
                 BatchedFeMessage::Exists {
                     span,
                     timer,
-                    shard,
+                    shard: shard.downgrade(),
                     req,
                 }
             }
@@ -743,13 +798,16 @@ impl PageServerHandler {
                     .get(tenant_id, timeline_id, ShardSelector::Zero)
                     .instrument(span.clone()) // sets `shard_id` field
                     .await?;
-                let timer = shard
-                    .query_metrics
-                    .start_smgr_op(metrics::SmgrQueryType::GetRelSize, received_at);
+                let timer = record_op_start_and_throttle(
+                    &shard,
+                    metrics::SmgrQueryType::GetRelSize,
+                    received_at,
+                )
+                .await?;
                 BatchedFeMessage::Nblocks {
                     span,
                     timer,
-                    shard,
+                    shard: shard.downgrade(),
                     req,
                 }
             }
@@ -759,13 +817,16 @@ impl PageServerHandler {
                     .get(tenant_id, timeline_id, ShardSelector::Zero)
                     .instrument(span.clone()) // sets `shard_id` field
                     .await?;
-                let timer = shard
-                    .query_metrics
-                    .start_smgr_op(metrics::SmgrQueryType::GetDbSize, received_at);
+                let timer = record_op_start_and_throttle(
+                    &shard,
+                    metrics::SmgrQueryType::GetDbSize,
+                    received_at,
+                )
+                .await?;
                 BatchedFeMessage::DbSize {
                     span,
                     timer,
-                    shard,
+                    shard: shard.downgrade(),
                     req,
                 }
             }
@@ -775,13 +836,16 @@ impl PageServerHandler {
                     .get(tenant_id, timeline_id, ShardSelector::Zero)
                     .instrument(span.clone()) // sets `shard_id` field
                     .await?;
-                let timer = shard
-                    .query_metrics
-                    .start_smgr_op(metrics::SmgrQueryType::GetSlruSegment, received_at);
+                let timer = record_op_start_and_throttle(
+                    &shard,
+                    metrics::SmgrQueryType::GetSlruSegment,
+                    received_at,
+                )
+                .await?;
                 BatchedFeMessage::GetSlruSegment {
                     span,
                     timer,
-                    shard,
+                    shard: shard.downgrade(),
                     req,
                 }
             }
@@ -826,13 +890,14 @@ impl PageServerHandler {
                     }
                 };
 
-                // 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);
+                let timer = record_op_start_and_throttle(
+                    &shard,
+                    metrics::SmgrQueryType::GetPageAtLsn,
+                    received_at,
+                )
+                .await?;
 
+                // We're holding the Handle
                 let effective_request_lsn = match Self::wait_or_get_last_lsn(
                     &shard,
                     req.hdr.request_lsn,
@@ -850,11 +915,27 @@ impl PageServerHandler {
                 };
                 BatchedFeMessage::GetPage {
                     span,
-                    shard,
+                    shard: shard.downgrade(),
                     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))
     }
@@ -896,9 +977,7 @@ impl PageServerHandler {
                     assert_eq!(accum_pages.len(), max_batch_size.get());
                     return false;
                 }
-                if (accum_shard.tenant_shard_id, accum_shard.timeline_id)
-                    != (this_shard.tenant_shard_id, this_shard.timeline_id)
-                {
+                if !accum_shard.is_same_handle_as(&this_shard) {
                     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.
@@ -917,6 +996,44 @@ 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
@@ -930,6 +1047,7 @@ impl PageServerHandler {
         &mut self,
         pgb_writer: &mut PostgresBackend<IO>,
         batch: BatchedFeMessage,
+        io_concurrency: IoConcurrency,
         cancel: &CancellationToken,
         protocol_version: PagestreamProtocolVersion,
         ctx: &RequestContext,
@@ -937,6 +1055,13 @@ 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>>,
@@ -951,7 +1076,7 @@ impl PageServerHandler {
                 fail::fail_point!("ps::handle-pagerequest-message::exists");
                 (
                     vec![self
-                        .handle_get_rel_exists_request(&shard, &req, ctx)
+                        .handle_get_rel_exists_request(&*shard.upgrade()?, &req, ctx)
                         .instrument(span.clone())
                         .await
                         .map(|msg| (msg, timer))
@@ -968,7 +1093,7 @@ impl PageServerHandler {
                 fail::fail_point!("ps::handle-pagerequest-message::nblocks");
                 (
                     vec![self
-                        .handle_get_nblocks_request(&shard, &req, ctx)
+                        .handle_get_nblocks_request(&*shard.upgrade()?, &req, ctx)
                         .instrument(span.clone())
                         .await
                         .map(|msg| (msg, timer))
@@ -989,9 +1114,10 @@ impl PageServerHandler {
                         trace!(npages, "handling getpage request");
                         let res = self
                             .handle_get_page_at_lsn_request_batched(
-                                &shard,
+                                &*shard.upgrade()?,
                                 effective_request_lsn,
                                 pages,
+                                io_concurrency,
                                 ctx,
                             )
                             .instrument(span.clone())
@@ -1011,7 +1137,7 @@ impl PageServerHandler {
                 fail::fail_point!("ps::handle-pagerequest-message::dbsize");
                 (
                     vec![self
-                        .handle_db_size_request(&shard, &req, ctx)
+                        .handle_db_size_request(&*shard.upgrade()?, &req, ctx)
                         .instrument(span.clone())
                         .await
                         .map(|msg| (msg, timer))
@@ -1028,7 +1154,7 @@ impl PageServerHandler {
                 fail::fail_point!("ps::handle-pagerequest-message::slrusegment");
                 (
                     vec![self
-                        .handle_get_slru_segment_request(&shard, &req, ctx)
+                        .handle_get_slru_segment_request(&*shard.upgrade()?, &req, ctx)
                         .instrument(span.clone())
                         .await
                         .map(|msg| (msg, timer))
@@ -1036,6 +1162,27 @@ 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.
@@ -1103,8 +1250,11 @@ 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(|timer| timer.observe_smgr_op_completion_and_start_flushing());
+            let flushing_timer = timer.map(|mut timer| {
+                timer
+                    .observe_execution_end_flush_start(Instant::now())
+                    .expect("we are the first caller")
+            });
 
             // what we want to do
             let flush_fut = pgb_writer.flush();
@@ -1172,6 +1322,17 @@ impl PageServerHandler {
             }
         }
 
+        let io_concurrency = IoConcurrency::spawn_from_conf(
+            self.conf,
+            match self.gate_guard.try_clone() {
+                Ok(guard) => guard,
+                Err(_) => {
+                    info!("shutdown request received in page handler");
+                    return Err(QueryError::Shutdown);
+                }
+            },
+        );
+
         let pgb_reader = pgb
             .split()
             .context("implementation error: split pgb into reader and writer")?;
@@ -1193,6 +1354,7 @@ impl PageServerHandler {
                     request_span,
                     pipelining_config,
                     protocol_version,
+                    io_concurrency,
                     &ctx,
                 )
                 .await
@@ -1206,6 +1368,7 @@ impl PageServerHandler {
                     timeline_handles,
                     request_span,
                     protocol_version,
+                    io_concurrency,
                     &ctx,
                 )
                 .await
@@ -1233,6 +1396,7 @@ impl PageServerHandler {
         mut timeline_handles: TimelineHandles,
         request_span: Span,
         protocol_version: PagestreamProtocolVersion,
+        io_concurrency: IoConcurrency,
         ctx: &RequestContext,
     ) -> (
         (PostgresBackendReader<IO>, TimelineHandles),
@@ -1258,7 +1422,7 @@ impl PageServerHandler {
                 Ok(msg) => msg,
                 Err(e) => break e,
             };
-            let mut msg = match msg {
+            let msg = match msg {
                 Some(msg) => msg,
                 None => {
                     debug!("pagestream subprotocol end observed");
@@ -1266,12 +1430,15 @@ 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)
+                .pagesteam_handle_batched_message(
+                    pgb_writer,
+                    msg,
+                    io_concurrency.clone(),
+                    &cancel,
+                    protocol_version,
+                    ctx,
+                )
                 .await;
             match err {
                 Ok(()) => {}
@@ -1295,6 +1462,7 @@ impl PageServerHandler {
         request_span: Span,
         pipelining_config: PageServicePipeliningConfigPipelined,
         protocol_version: PagestreamProtocolVersion,
+        io_concurrency: IoConcurrency,
         ctx: &RequestContext,
     ) -> (
         (PostgresBackendReader<IO>, TimelineHandles),
@@ -1429,18 +1597,16 @@ impl PageServerHandler {
                             return Ok(());
                         }
                     };
-                    let mut batch = match batch {
+                    let 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,
+                        io_concurrency.clone(),
                         &cancel,
                         protocol_version,
                         &ctx,
@@ -1697,6 +1863,7 @@ impl PageServerHandler {
         timeline: &Timeline,
         effective_lsn: Lsn,
         requests: smallvec::SmallVec<[BatchedGetPageRequest; 1]>,
+        io_concurrency: IoConcurrency,
         ctx: &RequestContext,
     ) -> Vec<Result<(PagestreamBeMessage, SmgrOpTimer), BatchedPageStreamError>> {
         debug_assert_current_span_has_tenant_and_timeline_id();
@@ -1705,10 +1872,25 @@ 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)),
                 effective_lsn,
+                io_concurrency,
                 ctx,
             )
             .await;
@@ -1763,6 +1945,51 @@ 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,
@@ -2378,6 +2605,14 @@ 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/pgdatadir_mapping.rs

@@ -17,6 +17,7 @@ use crate::span::{
     debug_assert_current_span_has_tenant_and_timeline_id,
     debug_assert_current_span_has_tenant_and_timeline_id_no_shard_id,
 };
+use crate::tenant::storage_layer::IoConcurrency;
 use crate::tenant::timeline::GetVectoredError;
 use anyhow::{ensure, Context};
 use bytes::{Buf, Bytes, BytesMut};
@@ -200,6 +201,7 @@ impl Timeline {
         blknum: BlockNumber,
         version: Version<'_>,
         ctx: &RequestContext,
+        io_concurrency: IoConcurrency,
     ) -> Result<Bytes, PageReconstructError> {
         match version {
             Version::Lsn(effective_lsn) => {
@@ -208,6 +210,7 @@ impl Timeline {
                     .get_rel_page_at_lsn_batched(
                         pages.iter().map(|(tag, blknum)| (tag, blknum)),
                         effective_lsn,
+                        io_concurrency.clone(),
                         ctx,
                     )
                     .await;
@@ -246,6 +249,7 @@ impl Timeline {
         &self,
         pages: impl ExactSizeIterator<Item = (&RelTag, &BlockNumber)>,
         effective_lsn: Lsn,
+        io_concurrency: IoConcurrency,
         ctx: &RequestContext,
     ) -> Vec<Result<Bytes, PageReconstructError>> {
         debug_assert_current_span_has_tenant_and_timeline_id();
@@ -309,7 +313,10 @@ impl Timeline {
             acc.to_keyspace()
         };
 
-        match self.get_vectored(keyspace, effective_lsn, ctx).await {
+        match self
+            .get_vectored(keyspace, effective_lsn, io_concurrency, ctx)
+            .await
+        {
             Ok(results) => {
                 for (key, res) in results {
                     let mut key_slots = keys_slots.remove(&key).unwrap().into_iter();
@@ -889,9 +896,15 @@ impl Timeline {
         &self,
         lsn: Lsn,
         ctx: &RequestContext,
+        io_concurrency: IoConcurrency,
     ) -> Result<HashMap<String, Bytes>, PageReconstructError> {
         let kv = self
-            .scan(KeySpace::single(Key::metadata_aux_key_range()), lsn, ctx)
+            .scan(
+                KeySpace::single(Key::metadata_aux_key_range()),
+                lsn,
+                ctx,
+                io_concurrency,
+            )
             .await?;
         let mut result = HashMap::new();
         let mut sz = 0;
@@ -914,8 +927,9 @@ impl Timeline {
         &self,
         lsn: Lsn,
         ctx: &RequestContext,
+        io_concurrency: IoConcurrency,
     ) -> Result<(), PageReconstructError> {
-        self.list_aux_files_v2(lsn, ctx).await?;
+        self.list_aux_files_v2(lsn, ctx, io_concurrency).await?;
         Ok(())
     }
 
@@ -923,17 +937,24 @@ impl Timeline {
         &self,
         lsn: Lsn,
         ctx: &RequestContext,
+        io_concurrency: IoConcurrency,
     ) -> Result<HashMap<String, Bytes>, PageReconstructError> {
-        self.list_aux_files_v2(lsn, ctx).await
+        self.list_aux_files_v2(lsn, ctx, io_concurrency).await
     }
 
     pub(crate) async fn get_replorigins(
         &self,
         lsn: Lsn,
         ctx: &RequestContext,
+        io_concurrency: IoConcurrency,
     ) -> Result<HashMap<RepOriginId, Lsn>, PageReconstructError> {
         let kv = self
-            .scan(KeySpace::single(repl_origin_key_range()), lsn, ctx)
+            .scan(
+                KeySpace::single(repl_origin_key_range()),
+                lsn,
+                ctx,
+                io_concurrency,
+            )
             .await?;
         let mut result = HashMap::new();
         for (k, v) in kv {
@@ -2432,7 +2453,11 @@ mod tests {
             ("foo/bar2".to_string(), Bytes::from_static(b"content2")),
         ]);
 
-        let readback = tline.list_aux_files(Lsn(0x1008), &ctx).await?;
+        let io_concurrency = IoConcurrency::spawn_for_test();
+
+        let readback = tline
+            .list_aux_files(Lsn(0x1008), &ctx, io_concurrency.clone())
+            .await?;
         assert_eq!(readback, expect_1008);
 
         // Second modification: update one key, remove the other
@@ -2444,11 +2469,15 @@ mod tests {
         let expect_2008 =
             HashMap::from([("foo/bar1".to_string(), Bytes::from_static(b"content3"))]);
 
-        let readback = tline.list_aux_files(Lsn(0x2008), &ctx).await?;
+        let readback = tline
+            .list_aux_files(Lsn(0x2008), &ctx, io_concurrency.clone())
+            .await?;
         assert_eq!(readback, expect_2008);
 
         // Reading back in time works
-        let readback = tline.list_aux_files(Lsn(0x1008), &ctx).await?;
+        let readback = tline
+            .list_aux_files(Lsn(0x1008), &ctx, io_concurrency.clone())
+            .await?;
         assert_eq!(readback, expect_1008);
 
         Ok(())

pageserver/src/tenant.rs

@@ -95,6 +95,9 @@ use crate::deletion_queue::DeletionQueueError;
 use crate::import_datadir;
 use crate::is_uninit_mark;
 use crate::l0_flush::L0FlushGlobalState;
+use crate::metrics::CONCURRENT_INITDBS;
+use crate::metrics::INITDB_RUN_TIME;
+use crate::metrics::INITDB_SEMAPHORE_ACQUISITION_TIME;
 use crate::metrics::TENANT;
 use crate::metrics::{
     remove_tenant_metrics, BROKEN_TENANTS_SET, CIRCUIT_BREAKERS_BROKEN, CIRCUIT_BREAKERS_UNBROKEN,
@@ -365,8 +368,9 @@ 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<crate::metrics::tenant_throttling::Pagestream>>,
+    pub(crate) pagestream_throttle: Arc<throttle::Throttle>,
+
+    pub(crate) pagestream_throttle_metrics: Arc<crate::metrics::tenant_throttling::Pagestream>,
 
     /// An ongoing timeline detach concurrency limiter.
     ///
@@ -1687,6 +1691,7 @@ 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,
@@ -3992,6 +3997,9 @@ 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)]
@@ -4100,8 +4108,10 @@ 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(),
@@ -5008,6 +5018,7 @@ 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(),
         }
     }
@@ -5339,8 +5350,17 @@ async fn run_initdb(
         initdb_bin_path, initdb_target_dir, initdb_lib_dir,
     );
 
-    let _permit = INIT_DB_SEMAPHORE.acquire().await;
+    let _permit = {
+        let _timer = INITDB_SEMAPHORE_ACQUISITION_TIME.start_timer();
+        INIT_DB_SEMAPHORE.acquire().await
+    };
 
+    CONCURRENT_INITDBS.inc();
+    scopeguard::defer! {
+        CONCURRENT_INITDBS.dec();
+    }
+
+    let _timer = INITDB_RUN_TIME.start_timer();
     let res = postgres_initdb::do_run_initdb(postgres_initdb::RunInitdbArgs {
         superuser: &conf.superuser,
         locale: &conf.locale,
@@ -5433,6 +5453,8 @@ pub(crate) mod harness {
                 compaction_period: Some(tenant_conf.compaction_period),
                 compaction_threshold: Some(tenant_conf.compaction_threshold),
                 compaction_algorithm: Some(tenant_conf.compaction_algorithm),
+                l0_flush_delay_threshold: tenant_conf.l0_flush_delay_threshold,
+                l0_flush_stall_threshold: tenant_conf.l0_flush_stall_threshold,
                 gc_horizon: Some(tenant_conf.gc_horizon),
                 gc_period: Some(tenant_conf.gc_period),
                 image_creation_threshold: Some(tenant_conf.image_creation_threshold),
@@ -5455,6 +5477,12 @@ pub(crate) mod harness {
                 lsn_lease_length_for_ts: Some(tenant_conf.lsn_lease_length_for_ts),
                 timeline_offloading: Some(tenant_conf.timeline_offloading),
                 wal_receiver_protocol_override: tenant_conf.wal_receiver_protocol_override,
+                rel_size_v2_enabled: tenant_conf.rel_size_v2_enabled,
+                gc_compaction_enabled: Some(tenant_conf.gc_compaction_enabled),
+                gc_compaction_initial_threshold_kb: Some(
+                    tenant_conf.gc_compaction_initial_threshold_kb,
+                ),
+                gc_compaction_ratio_percent: Some(tenant_conf.gc_compaction_ratio_percent),
             }
         }
     }
@@ -5682,13 +5710,13 @@ mod tests {
     use bytes::{Bytes, BytesMut};
     use hex_literal::hex;
     use itertools::Itertools;
-    use pageserver_api::key::{Key, AUX_KEY_PREFIX, NON_INHERITED_RANGE};
+    use pageserver_api::key::{Key, AUX_KEY_PREFIX, NON_INHERITED_RANGE, RELATION_SIZE_PREFIX};
     use pageserver_api::keyspace::KeySpace;
     use pageserver_api::models::{CompactionAlgorithm, CompactionAlgorithmSettings};
     use pageserver_api::value::Value;
     use pageserver_compaction::helpers::overlaps_with;
     use rand::{thread_rng, Rng};
-    use storage_layer::PersistentLayerKey;
+    use storage_layer::{IoConcurrency, PersistentLayerKey};
     use tests::storage_layer::ValuesReconstructState;
     use tests::timeline::{GetVectoredError, ShutdownMode};
     use timeline::{CompactOptions, DeltaLayerTestDesc};
@@ -6469,6 +6497,7 @@ mod tests {
     async fn test_get_vectored() -> anyhow::Result<()> {
         let harness = TenantHarness::create("test_get_vectored").await?;
         let (tenant, ctx) = harness.load().await;
+        let io_concurrency = IoConcurrency::spawn_for_test();
         let tline = tenant
             .create_test_timeline(TIMELINE_ID, Lsn(0x08), DEFAULT_PG_VERSION, &ctx)
             .await?;
@@ -6533,7 +6562,7 @@ mod tests {
                 .get_vectored_impl(
                     read.clone(),
                     reads_lsn,
-                    &mut ValuesReconstructState::new(),
+                    &mut ValuesReconstructState::new(io_concurrency.clone()),
                     &ctx,
                 )
                 .await;
@@ -6580,6 +6609,7 @@ mod tests {
         let harness = TenantHarness::create("test_get_vectored_aux_files").await?;
 
         let (tenant, ctx) = harness.load().await;
+        let io_concurrency = IoConcurrency::spawn_for_test();
         let tline = tenant
             .create_empty_timeline(TIMELINE_ID, Lsn(0), DEFAULT_PG_VERSION, &ctx)
             .await?;
@@ -6614,7 +6644,7 @@ mod tests {
             .get_vectored_impl(
                 aux_keyspace.clone(),
                 read_lsn,
-                &mut ValuesReconstructState::new(),
+                &mut ValuesReconstructState::new(io_concurrency.clone()),
                 &ctx,
             )
             .await;
@@ -6662,6 +6692,7 @@ mod tests {
         )
         .await?;
         let (tenant, ctx) = harness.load().await;
+        let io_concurrency = IoConcurrency::spawn_for_test();
 
         let mut current_key = Key::from_hex("010000000033333333444444445500000000").unwrap();
         let gap_at_key = current_key.add(100);
@@ -6762,7 +6793,7 @@ mod tests {
             .get_vectored_impl(
                 read.clone(),
                 current_lsn,
-                &mut ValuesReconstructState::new(),
+                &mut ValuesReconstructState::new(io_concurrency.clone()),
                 &ctx,
             )
             .await?;
@@ -6805,6 +6836,7 @@ mod tests {
     async fn test_get_vectored_ancestor_descent() -> anyhow::Result<()> {
         let harness = TenantHarness::create("test_get_vectored_on_lsn_axis").await?;
         let (tenant, ctx) = harness.load().await;
+        let io_concurrency = IoConcurrency::spawn_for_test();
 
         let start_key = Key::from_hex("010000000033333333444444445500000000").unwrap();
         let end_key = start_key.add(1000);
@@ -6897,7 +6929,7 @@ mod tests {
                         ranges: vec![child_gap_at_key..child_gap_at_key.next()],
                     },
                     query_lsn,
-                    &mut ValuesReconstructState::new(),
+                    &mut ValuesReconstructState::new(io_concurrency.clone()),
                     &ctx,
                 )
                 .await;
@@ -7343,6 +7375,7 @@ mod tests {
     async fn test_metadata_scan() -> anyhow::Result<()> {
         let harness = TenantHarness::create("test_metadata_scan").await?;
         let (tenant, ctx) = harness.load().await;
+        let io_concurrency = IoConcurrency::spawn_for_test();
         let tline = tenant
             .create_test_timeline(TIMELINE_ID, Lsn(0x10), DEFAULT_PG_VERSION, &ctx)
             .await?;
@@ -7396,7 +7429,7 @@ mod tests {
                 .get_vectored_impl(
                     keyspace.clone(),
                     lsn,
-                    &mut ValuesReconstructState::default(),
+                    &mut ValuesReconstructState::new(io_concurrency.clone()),
                     &ctx,
                 )
                 .await?
@@ -7511,6 +7544,7 @@ mod tests {
         let harness = TenantHarness::create("test_aux_file_e2e").await.unwrap();
 
         let (tenant, ctx) = harness.load().await;
+        let io_concurrency = IoConcurrency::spawn_for_test();
 
         let mut lsn = Lsn(0x08);
 
@@ -7530,7 +7564,10 @@ mod tests {
         }
 
         // we can read everything from the storage
-        let files = tline.list_aux_files(lsn, &ctx).await.unwrap();
+        let files = tline
+            .list_aux_files(lsn, &ctx, io_concurrency.clone())
+            .await
+            .unwrap();
         assert_eq!(
             files.get("pg_logical/mappings/test1"),
             Some(&bytes::Bytes::from_static(b"first"))
@@ -7546,7 +7583,10 @@ mod tests {
             modification.commit(&ctx).await.unwrap();
         }
 
-        let files = tline.list_aux_files(lsn, &ctx).await.unwrap();
+        let files = tline
+            .list_aux_files(lsn, &ctx, io_concurrency.clone())
+            .await
+            .unwrap();
         assert_eq!(
             files.get("pg_logical/mappings/test2"),
             Some(&bytes::Bytes::from_static(b"second"))
@@ -7557,7 +7597,10 @@ mod tests {
             .await
             .unwrap();
 
-        let files = child.list_aux_files(lsn, &ctx).await.unwrap();
+        let files = child
+            .list_aux_files(lsn, &ctx, io_concurrency.clone())
+            .await
+            .unwrap();
         assert_eq!(files.get("pg_logical/mappings/test1"), None);
         assert_eq!(files.get("pg_logical/mappings/test2"), None);
     }
@@ -7566,6 +7609,7 @@ mod tests {
     async fn test_metadata_image_creation() -> anyhow::Result<()> {
         let harness = TenantHarness::create("test_metadata_image_creation").await?;
         let (tenant, ctx) = harness.load().await;
+        let io_concurrency = IoConcurrency::spawn_for_test();
         let tline = tenant
             .create_test_timeline(TIMELINE_ID, Lsn(0x10), DEFAULT_PG_VERSION, &ctx)
             .await?;
@@ -7585,8 +7629,9 @@ mod tests {
             keyspace: &KeySpace,
             lsn: Lsn,
             ctx: &RequestContext,
+            io_concurrency: IoConcurrency,
         ) -> anyhow::Result<(BTreeMap<Key, Result<Bytes, PageReconstructError>>, usize)> {
-            let mut reconstruct_state = ValuesReconstructState::default();
+            let mut reconstruct_state = ValuesReconstructState::new(io_concurrency);
             let res = tline
                 .get_vectored_impl(keyspace.clone(), lsn, &mut reconstruct_state, ctx)
                 .await?;
@@ -7634,7 +7679,8 @@ mod tests {
 
             if iter % 5 == 0 {
                 let (_, before_delta_file_accessed) =
-                    scan_with_statistics(&tline, &keyspace, lsn, &ctx).await?;
+                    scan_with_statistics(&tline, &keyspace, lsn, &ctx, io_concurrency.clone())
+                        .await?;
                 tline
                     .compact(
                         &cancel,
@@ -7648,7 +7694,8 @@ mod tests {
                     )
                     .await?;
                 let (_, after_delta_file_accessed) =
-                    scan_with_statistics(&tline, &keyspace, lsn, &ctx).await?;
+                    scan_with_statistics(&tline, &keyspace, lsn, &ctx, io_concurrency.clone())
+                        .await?;
                 assert!(after_delta_file_accessed < before_delta_file_accessed, "after_delta_file_accessed={after_delta_file_accessed}, before_delta_file_accessed={before_delta_file_accessed}");
                 // Given that we already produced an image layer, there should be no delta layer needed for the scan, but still setting a low threshold there for unforeseen circumstances.
                 assert!(
@@ -7737,11 +7784,23 @@ mod tests {
     async fn test_vectored_missing_metadata_key_reads() -> anyhow::Result<()> {
         let harness = TenantHarness::create("test_vectored_missing_metadata_key_reads").await?;
         let (tenant, ctx) = harness.load().await;
+        let io_concurrency = IoConcurrency::spawn_for_test();
 
         let base_key = Key::from_hex("620000000033333333444444445500000000").unwrap();
         let base_key_child = Key::from_hex("620000000033333333444444445500000001").unwrap();
         let base_key_nonexist = Key::from_hex("620000000033333333444444445500000002").unwrap();
+        let base_key_overwrite = Key::from_hex("620000000033333333444444445500000003").unwrap();
+
+        let base_inherited_key = Key::from_hex("610000000033333333444444445500000000").unwrap();
+        let base_inherited_key_child =
+            Key::from_hex("610000000033333333444444445500000001").unwrap();
+        let base_inherited_key_nonexist =
+            Key::from_hex("610000000033333333444444445500000002").unwrap();
+        let base_inherited_key_overwrite =
+            Key::from_hex("610000000033333333444444445500000003").unwrap();
+
         assert_eq!(base_key.field1, AUX_KEY_PREFIX); // in case someone accidentally changed the prefix...
+        assert_eq!(base_inherited_key.field1, RELATION_SIZE_PREFIX);
 
         let tline = tenant
             .create_test_timeline_with_layers(
@@ -7750,7 +7809,18 @@ mod tests {
                 DEFAULT_PG_VERSION,
                 &ctx,
                 Vec::new(), // delta layers
-                vec![(Lsn(0x20), vec![(base_key, test_img("metadata key 1"))])], // image layers
+                vec![(
+                    Lsn(0x20),
+                    vec![
+                        (base_inherited_key, test_img("metadata inherited key 1")),
+                        (
+                            base_inherited_key_overwrite,
+                            test_img("metadata key overwrite 1a"),
+                        ),
+                        (base_key, test_img("metadata key 1")),
+                        (base_key_overwrite, test_img("metadata key overwrite 1b")),
+                    ],
+                )], // image layers
                 Lsn(0x20), // it's fine to not advance LSN to 0x30 while using 0x30 to get below because `get_vectored_impl` does not wait for LSN
             )
             .await?;
@@ -7764,7 +7834,18 @@ mod tests {
                 Vec::new(), // delta layers
                 vec![(
                     Lsn(0x30),
-                    vec![(base_key_child, test_img("metadata key 2"))],
+                    vec![
+                        (
+                            base_inherited_key_child,
+                            test_img("metadata inherited key 2"),
+                        ),
+                        (
+                            base_inherited_key_overwrite,
+                            test_img("metadata key overwrite 2a"),
+                        ),
+                        (base_key_child, test_img("metadata key 2")),
+                        (base_key_overwrite, test_img("metadata key overwrite 2b")),
+                    ],
                 )], // image layers
                 Lsn(0x30),
             )
@@ -7786,6 +7867,26 @@ mod tests {
             get_vectored_impl_wrapper(&tline, base_key_nonexist, lsn, &ctx).await?,
             None
         );
+        assert_eq!(
+            get_vectored_impl_wrapper(&tline, base_key_overwrite, lsn, &ctx).await?,
+            Some(test_img("metadata key overwrite 1b"))
+        );
+        assert_eq!(
+            get_vectored_impl_wrapper(&tline, base_inherited_key, lsn, &ctx).await?,
+            Some(test_img("metadata inherited key 1"))
+        );
+        assert_eq!(
+            get_vectored_impl_wrapper(&tline, base_inherited_key_child, lsn, &ctx).await?,
+            None
+        );
+        assert_eq!(
+            get_vectored_impl_wrapper(&tline, base_inherited_key_nonexist, lsn, &ctx).await?,
+            None
+        );
+        assert_eq!(
+            get_vectored_impl_wrapper(&tline, base_inherited_key_overwrite, lsn, &ctx).await?,
+            Some(test_img("metadata key overwrite 1a"))
+        );
 
         // test vectored get on child timeline
         assert_eq!(
@@ -7800,6 +7901,82 @@ mod tests {
             get_vectored_impl_wrapper(&child, base_key_nonexist, lsn, &ctx).await?,
             None
         );
+        assert_eq!(
+            get_vectored_impl_wrapper(&child, base_inherited_key, lsn, &ctx).await?,
+            Some(test_img("metadata inherited key 1"))
+        );
+        assert_eq!(
+            get_vectored_impl_wrapper(&child, base_inherited_key_child, lsn, &ctx).await?,
+            Some(test_img("metadata inherited key 2"))
+        );
+        assert_eq!(
+            get_vectored_impl_wrapper(&child, base_inherited_key_nonexist, lsn, &ctx).await?,
+            None
+        );
+        assert_eq!(
+            get_vectored_impl_wrapper(&child, base_key_overwrite, lsn, &ctx).await?,
+            Some(test_img("metadata key overwrite 2b"))
+        );
+        assert_eq!(
+            get_vectored_impl_wrapper(&child, base_inherited_key_overwrite, lsn, &ctx).await?,
+            Some(test_img("metadata key overwrite 2a"))
+        );
+
+        // test vectored scan on parent timeline
+        let mut reconstruct_state = ValuesReconstructState::new(io_concurrency.clone());
+        let res = tline
+            .get_vectored_impl(
+                KeySpace::single(Key::metadata_key_range()),
+                lsn,
+                &mut reconstruct_state,
+                &ctx,
+            )
+            .await?;
+
+        assert_eq!(
+            res.into_iter()
+                .map(|(k, v)| (k, v.unwrap()))
+                .collect::<Vec<_>>(),
+            vec![
+                (base_inherited_key, test_img("metadata inherited key 1")),
+                (
+                    base_inherited_key_overwrite,
+                    test_img("metadata key overwrite 1a")
+                ),
+                (base_key, test_img("metadata key 1")),
+                (base_key_overwrite, test_img("metadata key overwrite 1b")),
+            ]
+        );
+
+        // test vectored scan on child timeline
+        let mut reconstruct_state = ValuesReconstructState::new(io_concurrency.clone());
+        let res = child
+            .get_vectored_impl(
+                KeySpace::single(Key::metadata_key_range()),
+                lsn,
+                &mut reconstruct_state,
+                &ctx,
+            )
+            .await?;
+
+        assert_eq!(
+            res.into_iter()
+                .map(|(k, v)| (k, v.unwrap()))
+                .collect::<Vec<_>>(),
+            vec![
+                (base_inherited_key, test_img("metadata inherited key 1")),
+                (
+                    base_inherited_key_child,
+                    test_img("metadata inherited key 2")
+                ),
+                (
+                    base_inherited_key_overwrite,
+                    test_img("metadata key overwrite 2a")
+                ),
+                (base_key_child, test_img("metadata key 2")),
+                (base_key_overwrite, test_img("metadata key overwrite 2b")),
+            ]
+        );
 
         Ok(())
     }
@@ -7810,7 +7987,9 @@ mod tests {
         lsn: Lsn,
         ctx: &RequestContext,
     ) -> Result<Option<Bytes>, GetVectoredError> {
-        let mut reconstruct_state = ValuesReconstructState::new();
+        let io_concurrency =
+            IoConcurrency::spawn_from_conf(tline.conf, tline.gate.enter().unwrap());
+        let mut reconstruct_state = ValuesReconstructState::new(io_concurrency);
         let mut res = tline
             .get_vectored_impl(
                 KeySpace::single(key..key.next()),
@@ -7911,6 +8090,7 @@ mod tests {
             .await
             .unwrap();
         let (tenant, ctx) = harness.load().await;
+        let io_concurrency = IoConcurrency::spawn_for_test();
 
         let key0 = Key::from_hex("620000000033333333444444445500000000").unwrap();
         let key1 = Key::from_hex("620000000033333333444444445500000001").unwrap();
@@ -7970,7 +8150,7 @@ mod tests {
 
         // Image layers are created at last_record_lsn
         let images = tline
-            .inspect_image_layers(Lsn(0x40), &ctx)
+            .inspect_image_layers(Lsn(0x40), &ctx, io_concurrency.clone())
             .await
             .unwrap()
             .into_iter()
@@ -7985,6 +8165,7 @@ mod tests {
             .await
             .unwrap();
         let (tenant, ctx) = harness.load().await;
+        let io_concurrency = IoConcurrency::spawn_for_test();
 
         let key1 = Key::from_hex("620000000033333333444444445500000001").unwrap();
         let key2 = Key::from_hex("620000000033333333444444445500000002").unwrap();
@@ -8035,7 +8216,7 @@ mod tests {
 
         // Image layers are created at last_record_lsn
         let images = tline
-            .inspect_image_layers(Lsn(0x30), &ctx)
+            .inspect_image_layers(Lsn(0x30), &ctx, io_concurrency.clone())
             .await
             .unwrap()
             .into_iter()
@@ -8044,10 +8225,18 @@ mod tests {
         assert_eq!(images.len(), 0); // the image layer should not contain tombstones, or it is not created
     }
 
+    fn key_max_minus_1() -> Key {
+        Key {
+            field6: u32::MAX - 1,
+            ..Key::MAX
+        }
+    }
+
     #[tokio::test]
     async fn test_simple_bottom_most_compaction_images() -> anyhow::Result<()> {
         let harness = TenantHarness::create("test_simple_bottom_most_compaction_images").await?;
         let (tenant, ctx) = harness.load().await;
+        let io_concurrency = IoConcurrency::spawn_for_test();
 
         fn get_key(id: u32) -> Key {
             // using aux key here b/c they are guaranteed to be inside `collect_keyspace`.
@@ -8189,7 +8378,7 @@ mod tests {
 
         // Check if the image layer at the GC horizon contains exactly what we want
         let image_at_gc_horizon = tline
-            .inspect_image_layers(Lsn(0x30), &ctx)
+            .inspect_image_layers(Lsn(0x30), &ctx, io_concurrency.clone())
             .await
             .unwrap()
             .into_iter()
@@ -8223,7 +8412,7 @@ mod tests {
             vec![
                 // Image layer at GC horizon
                 PersistentLayerKey {
-                    key_range: Key::MIN..Key::MAX,
+                    key_range: Key::MIN..key_max_minus_1(),
                     lsn_range: Lsn(0x30)..Lsn(0x31),
                     is_delta: false
                 },
@@ -9902,7 +10091,12 @@ mod tests {
 
         let keyspace = KeySpace::single(get_key(0)..get_key(10));
         let results = tline
-            .get_vectored(keyspace, delta_layer_end_lsn, &ctx)
+            .get_vectored(
+                keyspace,
+                delta_layer_end_lsn,
+                IoConcurrency::sequential(),
+                &ctx,
+            )
             .await
             .expect("No vectored errors");
         for (key, res) in results {
@@ -10590,7 +10784,7 @@ mod tests {
             vec![
                 // The compacted image layer (full key range)
                 PersistentLayerKey {
-                    key_range: Key::MIN..Key::MAX,
+                    key_range: Key::MIN..key_max_minus_1(),
                     lsn_range: Lsn(0x10)..Lsn(0x11),
                     is_delta: false,
                 },
@@ -10937,7 +11131,7 @@ mod tests {
             vec![
                 // The compacted image layer (full key range)
                 PersistentLayerKey {
-                    key_range: Key::MIN..Key::MAX,
+                    key_range: Key::MIN..key_max_minus_1(),
                     lsn_range: Lsn(0x10)..Lsn(0x11),
                     is_delta: false,
                 },

pageserver/src/tenant/config.rs

@@ -11,7 +11,7 @@
 pub(crate) use pageserver_api::config::TenantConfigToml as TenantConf;
 use pageserver_api::models::CompactionAlgorithmSettings;
 use pageserver_api::models::EvictionPolicy;
-use pageserver_api::models::{self, TenantConfigPatch, ThrottleConfig};
+use pageserver_api::models::{self, TenantConfigPatch};
 use pageserver_api::shard::{ShardCount, ShardIdentity, ShardNumber, ShardStripeSize};
 use serde::de::IntoDeserializer;
 use serde::{Deserialize, Serialize};
@@ -281,6 +281,14 @@ pub struct TenantConfOpt {
     #[serde(default)]
     pub compaction_algorithm: Option<CompactionAlgorithmSettings>,
 
+    #[serde(skip_serializing_if = "Option::is_none")]
+    #[serde(default)]
+    pub l0_flush_delay_threshold: Option<usize>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
+    #[serde(default)]
+    pub l0_flush_stall_threshold: Option<usize>,
+
     #[serde(skip_serializing_if = "Option::is_none")]
     #[serde(default)]
     pub gc_horizon: Option<u64>,
@@ -357,6 +365,18 @@ pub struct TenantConfOpt {
 
     #[serde(skip_serializing_if = "Option::is_none")]
     pub wal_receiver_protocol_override: Option<PostgresClientProtocol>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub rel_size_v2_enabled: Option<bool>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub gc_compaction_enabled: Option<bool>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub gc_compaction_initial_threshold_kb: Option<u64>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub gc_compaction_ratio_percent: Option<u64>,
 }
 
 impl TenantConfOpt {
@@ -382,6 +402,12 @@ impl TenantConfOpt {
                 .as_ref()
                 .unwrap_or(&global_conf.compaction_algorithm)
                 .clone(),
+            l0_flush_delay_threshold: self
+                .l0_flush_delay_threshold
+                .or(global_conf.l0_flush_delay_threshold),
+            l0_flush_stall_threshold: self
+                .l0_flush_stall_threshold
+                .or(global_conf.l0_flush_stall_threshold),
             gc_horizon: self.gc_horizon.unwrap_or(global_conf.gc_horizon),
             gc_period: self.gc_period.unwrap_or(global_conf.gc_period),
             image_creation_threshold: self
@@ -425,6 +451,16 @@ impl TenantConfOpt {
             wal_receiver_protocol_override: self
                 .wal_receiver_protocol_override
                 .or(global_conf.wal_receiver_protocol_override),
+            rel_size_v2_enabled: self.rel_size_v2_enabled.or(global_conf.rel_size_v2_enabled),
+            gc_compaction_enabled: self
+                .gc_compaction_enabled
+                .unwrap_or(global_conf.gc_compaction_enabled),
+            gc_compaction_initial_threshold_kb: self
+                .gc_compaction_initial_threshold_kb
+                .unwrap_or(global_conf.gc_compaction_initial_threshold_kb),
+            gc_compaction_ratio_percent: self
+                .gc_compaction_ratio_percent
+                .unwrap_or(global_conf.gc_compaction_ratio_percent),
         }
     }
 
@@ -436,6 +472,8 @@ impl TenantConfOpt {
             mut compaction_period,
             mut compaction_threshold,
             mut compaction_algorithm,
+            mut l0_flush_delay_threshold,
+            mut l0_flush_stall_threshold,
             mut gc_horizon,
             mut gc_period,
             mut image_creation_threshold,
@@ -454,6 +492,10 @@ impl TenantConfOpt {
             mut lsn_lease_length_for_ts,
             mut timeline_offloading,
             mut wal_receiver_protocol_override,
+            mut rel_size_v2_enabled,
+            mut gc_compaction_enabled,
+            mut gc_compaction_initial_threshold_kb,
+            mut gc_compaction_ratio_percent,
         } = self;
 
         patch.checkpoint_distance.apply(&mut checkpoint_distance);
@@ -470,6 +512,12 @@ impl TenantConfOpt {
             .apply(&mut compaction_period);
         patch.compaction_threshold.apply(&mut compaction_threshold);
         patch.compaction_algorithm.apply(&mut compaction_algorithm);
+        patch
+            .l0_flush_delay_threshold
+            .apply(&mut l0_flush_delay_threshold);
+        patch
+            .l0_flush_stall_threshold
+            .apply(&mut l0_flush_stall_threshold);
         patch.gc_horizon.apply(&mut gc_horizon);
         patch
             .gc_period
@@ -522,6 +570,16 @@ impl TenantConfOpt {
         patch
             .wal_receiver_protocol_override
             .apply(&mut wal_receiver_protocol_override);
+        patch.rel_size_v2_enabled.apply(&mut rel_size_v2_enabled);
+        patch
+            .gc_compaction_enabled
+            .apply(&mut gc_compaction_enabled);
+        patch
+            .gc_compaction_initial_threshold_kb
+            .apply(&mut gc_compaction_initial_threshold_kb);
+        patch
+            .gc_compaction_ratio_percent
+            .apply(&mut gc_compaction_ratio_percent);
 
         Ok(Self {
             checkpoint_distance,
@@ -530,6 +588,8 @@ impl TenantConfOpt {
             compaction_period,
             compaction_threshold,
             compaction_algorithm,
+            l0_flush_delay_threshold,
+            l0_flush_stall_threshold,
             gc_horizon,
             gc_period,
             image_creation_threshold,
@@ -548,6 +608,10 @@ impl TenantConfOpt {
             lsn_lease_length_for_ts,
             timeline_offloading,
             wal_receiver_protocol_override,
+            rel_size_v2_enabled,
+            gc_compaction_enabled,
+            gc_compaction_initial_threshold_kb,
+            gc_compaction_ratio_percent,
         })
     }
 }
@@ -583,6 +647,8 @@ impl From<TenantConfOpt> for models::TenantConfig {
             compaction_target_size: value.compaction_target_size,
             compaction_period: value.compaction_period.map(humantime),
             compaction_threshold: value.compaction_threshold,
+            l0_flush_delay_threshold: value.l0_flush_delay_threshold,
+            l0_flush_stall_threshold: value.l0_flush_stall_threshold,
             gc_horizon: value.gc_horizon,
             gc_period: value.gc_period.map(humantime),
             image_creation_threshold: value.image_creation_threshold,
@@ -597,12 +663,16 @@ impl From<TenantConfOpt> for models::TenantConfig {
                 .map(humantime),
             heatmap_period: value.heatmap_period.map(humantime),
             lazy_slru_download: value.lazy_slru_download,
-            timeline_get_throttle: value.timeline_get_throttle.map(ThrottleConfig::from),
+            timeline_get_throttle: value.timeline_get_throttle,
             image_layer_creation_check_threshold: value.image_layer_creation_check_threshold,
             lsn_lease_length: value.lsn_lease_length.map(humantime),
             lsn_lease_length_for_ts: value.lsn_lease_length_for_ts.map(humantime),
             timeline_offloading: value.timeline_offloading,
             wal_receiver_protocol_override: value.wal_receiver_protocol_override,
+            rel_size_v2_enabled: value.rel_size_v2_enabled,
+            gc_compaction_enabled: value.gc_compaction_enabled,
+            gc_compaction_initial_threshold_kb: value.gc_compaction_initial_threshold_kb,
+            gc_compaction_ratio_percent: value.gc_compaction_ratio_percent,
         }
     }
 }

pageserver/src/tenant/disk_btree.rs

@@ -84,17 +84,17 @@ impl Value {
 
     fn to_u64(self) -> u64 {
         let b = &self.0;
-        (b[0] as u64) << 32
-            | (b[1] as u64) << 24
-            | (b[2] as u64) << 16
-            | (b[3] as u64) << 8
+        ((b[0] as u64) << 32)
+            | ((b[1] as u64) << 24)
+            | ((b[2] as u64) << 16)
+            | ((b[3] as u64) << 8)
             | b[4] as u64
     }
 
     fn to_blknum(self) -> u32 {
         let b = &self.0;
         assert!(b[0] == 0x80);
-        (b[1] as u32) << 24 | (b[2] as u32) << 16 | (b[3] as u32) << 8 | b[4] as u32
+        ((b[1] as u32) << 24) | ((b[2] as u32) << 16) | ((b[3] as u32) << 8) | b[4] as u32
     }
 }
 

pageserver/src/tenant/layer_map.rs

@@ -57,6 +57,7 @@ use std::collections::{HashMap, VecDeque};
 use std::iter::Peekable;
 use std::ops::Range;
 use std::sync::Arc;
+use tokio::sync::watch;
 use utils::lsn::Lsn;
 
 use historic_layer_coverage::BufferedHistoricLayerCoverage;
@@ -67,7 +68,6 @@ use super::storage_layer::{LayerVisibilityHint, PersistentLayerDesc};
 ///
 /// LayerMap tracks what layers exist on a timeline.
 ///
-#[derive(Default)]
 pub struct LayerMap {
     //
     // 'open_layer' holds the current InMemoryLayer that is accepting new
@@ -93,7 +93,25 @@ pub struct LayerMap {
 
     /// L0 layers have key range Key::MIN..Key::MAX, and locating them using R-Tree search is very inefficient.
     /// So L0 layers are held in l0_delta_layers vector, in addition to the R-tree.
+    ///
+    /// NB: make sure to notify `watch_l0_deltas` on changes.
     l0_delta_layers: Vec<Arc<PersistentLayerDesc>>,
+
+    /// Notifies about L0 delta layer changes, sending the current number of L0 layers.
+    watch_l0_deltas: watch::Sender<usize>,
+}
+
+impl Default for LayerMap {
+    fn default() -> Self {
+        Self {
+            open_layer: Default::default(),
+            next_open_layer_at: Default::default(),
+            frozen_layers: Default::default(),
+            historic: Default::default(),
+            l0_delta_layers: Default::default(),
+            watch_l0_deltas: watch::channel(0).0,
+        }
+    }
 }
 
 /// The primary update API for the layer map.
@@ -466,6 +484,8 @@ impl LayerMap {
 
         if Self::is_l0(&layer_desc.key_range, layer_desc.is_delta) {
             self.l0_delta_layers.push(layer_desc.clone().into());
+            self.watch_l0_deltas
+                .send_replace(self.l0_delta_layers.len());
         }
 
         self.historic.insert(
@@ -488,6 +508,8 @@ impl LayerMap {
             let mut l0_delta_layers = std::mem::take(&mut self.l0_delta_layers);
             l0_delta_layers.retain(|other| other.key() != layer_key);
             self.l0_delta_layers = l0_delta_layers;
+            self.watch_l0_deltas
+                .send_replace(self.l0_delta_layers.len());
             // this assertion is related to use of Arc::ptr_eq in Self::compare_arced_layers,
             // there's a chance that the comparison fails at runtime due to it comparing (pointer,
             // vtable) pairs.
@@ -850,6 +872,11 @@ impl LayerMap {
         &self.l0_delta_layers
     }
 
+    /// Subscribes to L0 delta layer changes, sending the current number of L0 delta layers.
+    pub fn watch_level0_deltas(&self) -> watch::Receiver<usize> {
+        self.watch_l0_deltas.subscribe()
+    }
+
     /// debugging function to print out the contents of the layer map
     #[allow(unused)]
     pub async fn dump(&self, verbose: bool, ctx: &RequestContext) -> Result<()> {

pageserver/src/tenant/metadata.rs

@@ -320,7 +320,6 @@ 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

@@ -63,22 +63,18 @@
 //! 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.
-//!
-//! 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.)
+//! is safe. For more details, see `UploadOp::can_bypass`.
 //!
 //! 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
@@ -386,6 +382,12 @@ 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
@@ -429,8 +431,16 @@ 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)?;
+        upload_queue.initialize_with_current_remote_index_part(index_part, inprogress_limit)?;
         self.update_remote_physical_size_gauge(Some(index_part));
         info!(
             "initialized upload queue from remote index with {} layer files",
@@ -445,8 +455,16 @@ 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)?;
+        upload_queue.initialize_empty_remote(local_metadata, inprogress_limit)?;
         self.update_remote_physical_size_gauge(None);
         info!("initialized upload queue as empty");
         Ok(())
@@ -462,9 +480,15 @@ 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)?;
+        upload_queue.initialize_with_current_remote_index_part(index_part, inprogress_limit)?;
         self.update_remote_physical_size_gauge(Some(index_part));
         self.stop_impl(&mut upload_queue);
 
@@ -779,6 +803,12 @@ impl RemoteTimelineClient {
 
         upload_queue.dirty.metadata.apply(update);
 
+        // Defense in depth: if we somehow generated invalid metadata, do not persist it.
+        upload_queue
+            .dirty
+            .validate()
+            .map_err(|e| anyhow::anyhow!(e))?;
+
         self.schedule_index_upload(upload_queue);
 
         Ok(())
@@ -1855,57 +1885,17 @@ impl RemoteTimelineClient {
         Ok(())
     }
 
-    ///
     /// Pick next tasks from the queue, and start as many of them as possible without violating
     /// the ordering constraints.
     ///
-    /// The caller needs to already hold the `upload_queue` lock.
+    /// 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.
     fn launch_queued_tasks(self: &Arc<Self>, upload_queue: &mut UploadQueueInitialized) {
-        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()
-                }
+        while let Some((mut next_op, coalesced_ops)) = upload_queue.next_ready() {
+            debug!("starting op: {next_op}");
 
-                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
+            // Prepare upload.
             match &mut next_op {
                 UploadOp::UploadLayer(layer, meta, mode) => {
                     if upload_queue
@@ -1916,18 +1906,14 @@ 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(());
@@ -1944,6 +1930,7 @@ impl RemoteTimelineClient {
             let task = Arc::new(UploadTask {
                 task_id: upload_task_id,
                 op: next_op,
+                coalesced_ops,
                 retries: AtomicU32::new(0),
             });
             upload_queue
@@ -2027,6 +2014,8 @@ 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.
@@ -2249,13 +2238,8 @@ impl RemoteTimelineClient {
             upload_queue.inprogress_tasks.remove(&task.task_id);
 
             let lsn_update = match task.op {
-                UploadOp::UploadLayer(_, _, _) => {
-                    upload_queue.num_inprogress_layer_uploads -= 1;
-                    None
-                }
+                UploadOp::UploadLayer(_, _, _) => 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;
@@ -2289,10 +2273,7 @@ impl RemoteTimelineClient {
                         None
                     }
                 }
-                UploadOp::Delete(_) => {
-                    upload_queue.num_inprogress_deletions -= 1;
-                    None
-                }
+                UploadOp::Delete(_) => None,
                 UploadOp::Barrier(..) | UploadOp::Shutdown => unreachable!(),
             };
 
@@ -2317,6 +2298,9 @@ impl RemoteTimelineClient {
         }
 
         self.metric_end(&task.op);
+        for coalesced_op in &task.coalesced_ops {
+            self.metric_end(coalesced_op);
+        }
     }
 
     fn metric_impl(
@@ -2409,6 +2393,7 @@ 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(),
@@ -2416,9 +2401,6 @@ 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")]
@@ -2445,14 +2427,6 @@ 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.
@@ -2899,8 +2873,8 @@ 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!(upload_queue.inprogress_tasks.len() == 2);
-            assert!(upload_queue.num_inprogress_layer_uploads == 2);
+            assert_eq!(upload_queue.inprogress_tasks.len(), 2);
+            assert_eq!(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);
@@ -2970,8 +2944,8 @@ 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

@@ -77,8 +77,32 @@ pub struct IndexPart {
     ///
     /// None means no aux files have been written to the storage before the point
     /// when this flag is introduced.
+    ///
+    /// This flag is not used any more as all tenants have been transitioned to the new aux file policy.
     #[serde(skip_serializing_if = "Option::is_none", default)]
     pub(crate) last_aux_file_policy: Option<AuxFilePolicy>,
+
+    #[serde(skip_serializing_if = "Option::is_none", default)]
+    pub(crate) rel_size_migration: Option<RelSizeMigration>,
+
+    /// The LSN of gc-compaction horizon. Once gc-compaction is finished for all layer files below an LSN, this LSN will be updated.
+    #[serde(skip_serializing_if = "Option::is_none", default)]
+    pub(crate) l2_lsn: Option<Lsn>,
+}
+
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub enum RelSizeMigration {
+    /// The tenant is using the old rel_size format.
+    /// Note that this enum is persisted as `Option<RelSizeMigration>` in the index part, so
+    /// `None` is the same as `Some(RelSizeMigration::Legacy)`.
+    Legacy,
+    /// The tenant is migrating to the new rel_size format. Both old and new rel_size format are
+    /// persisted in the index part. The read path will read both formats and merge them.
+    Migrating,
+    /// The tenant has migrated to the new rel_size format. Only the new rel_size format is persisted
+    /// in the index part, and the read path will not read the old format.
+    Migrated,
 }
 
 impl IndexPart {
@@ -97,14 +121,16 @@ impl IndexPart {
     /// - 8: added `archived_at`
     /// - 9: +gc_blocking
     /// - 10: +import_pgdata
-    const LATEST_VERSION: usize = 10;
+    /// - 11: +rel_size_migration
+    /// - 12: +l2_lsn
+    const LATEST_VERSION: usize = 12;
 
     // Versions we may see when reading from a bucket.
-    pub const KNOWN_VERSIONS: &'static [usize] = &[1, 2, 3, 4, 5, 6, 7, 8, 9, 10];
+    pub const KNOWN_VERSIONS: &'static [usize] = &[1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12];
 
     pub const FILE_NAME: &'static str = "index_part.json";
 
-    pub(crate) fn empty(metadata: TimelineMetadata) -> Self {
+    pub fn empty(metadata: TimelineMetadata) -> Self {
         IndexPart {
             version: Self::LATEST_VERSION,
             layer_metadata: Default::default(),
@@ -116,6 +142,8 @@ impl IndexPart {
             gc_blocking: None,
             last_aux_file_policy: None,
             import_pgdata: None,
+            rel_size_migration: None,
+            l2_lsn: None,
         }
     }
 
@@ -152,6 +180,21 @@ impl IndexPart {
         };
         is_same_remote_layer_path(name, metadata, name, index_metadata)
     }
+
+    /// Check for invariants in the index: this is useful when uploading an index to ensure that if
+    /// we encounter a bug, we do not persist buggy metadata.
+    pub(crate) fn validate(&self) -> Result<(), String> {
+        if self.import_pgdata.is_none()
+            && self.metadata.ancestor_timeline().is_none()
+            && self.layer_metadata.is_empty()
+        {
+            // Unless we're in the middle of a raw pgdata import, or this is a child timeline,the index must
+            // always have at least one layer.
+            return Err("Index has no ancestor and no layers".to_string());
+        }
+
+        Ok(())
+    }
 }
 
 /// Metadata gathered for each of the layer files.
@@ -401,6 +444,8 @@ mod tests {
             gc_blocking: None,
             last_aux_file_policy: None,
             import_pgdata: None,
+            rel_size_migration: None,
+            l2_lsn: None,
         };
 
         let part = IndexPart::from_json_bytes(example.as_bytes()).unwrap();
@@ -446,6 +491,8 @@ mod tests {
             gc_blocking: None,
             last_aux_file_policy: None,
             import_pgdata: None,
+            rel_size_migration: None,
+            l2_lsn: None,
         };
 
         let part = IndexPart::from_json_bytes(example.as_bytes()).unwrap();
@@ -492,6 +539,8 @@ mod tests {
             gc_blocking: None,
             last_aux_file_policy: None,
             import_pgdata: None,
+            rel_size_migration: None,
+            l2_lsn: None,
         };
 
         let part = IndexPart::from_json_bytes(example.as_bytes()).unwrap();
@@ -541,6 +590,8 @@ mod tests {
             gc_blocking: None,
             last_aux_file_policy: None,
             import_pgdata: None,
+            rel_size_migration: None,
+            l2_lsn: None,
         };
 
         let empty_layers_parsed = IndexPart::from_json_bytes(empty_layers_json.as_bytes()).unwrap();
@@ -585,6 +636,8 @@ mod tests {
             gc_blocking: None,
             last_aux_file_policy: None,
             import_pgdata: None,
+            rel_size_migration: None,
+            l2_lsn: None,
         };
 
         let part = IndexPart::from_json_bytes(example.as_bytes()).unwrap();
@@ -632,6 +685,8 @@ mod tests {
             gc_blocking: None,
             last_aux_file_policy: None,
             import_pgdata: None,
+            rel_size_migration: None,
+            l2_lsn: None,
         };
 
         let part = IndexPart::from_json_bytes(example.as_bytes()).unwrap();
@@ -684,6 +739,8 @@ mod tests {
             gc_blocking: None,
             last_aux_file_policy: Some(AuxFilePolicy::V2),
             import_pgdata: None,
+            rel_size_migration: None,
+            l2_lsn: None,
         };
 
         let part = IndexPart::from_json_bytes(example.as_bytes()).unwrap();
@@ -741,6 +798,8 @@ mod tests {
             gc_blocking: None,
             last_aux_file_policy: Default::default(),
             import_pgdata: None,
+            rel_size_migration: None,
+            l2_lsn: None,
         };
 
         let part = IndexPart::from_json_bytes(example.as_bytes()).unwrap();
@@ -799,6 +858,8 @@ mod tests {
             gc_blocking: None,
             last_aux_file_policy: Default::default(),
             import_pgdata: None,
+            rel_size_migration: None,
+            l2_lsn: None,
         };
 
         let part = IndexPart::from_json_bytes(example.as_bytes()).unwrap();
@@ -862,6 +923,8 @@ mod tests {
             last_aux_file_policy: Default::default(),
             archived_at: None,
             import_pgdata: None,
+            rel_size_migration: None,
+            l2_lsn: None,
         };
 
         let part = IndexPart::from_json_bytes(example.as_bytes()).unwrap();
@@ -937,7 +1000,168 @@ mod tests {
                 started_at: parse_naive_datetime("2024-11-13T09:23:42.123000000"),
                 finished_at: parse_naive_datetime("2024-11-13T09:42:23.123000000"),
                 idempotency_key: import_pgdata::index_part_format::IdempotencyKey::new("specified-by-client-218a5213-5044-4562-a28d-d024c5f057f5".to_string()),
-            })))
+            }))),
+            rel_size_migration: None,
+            l2_lsn: None,
+        };
+
+        let part = IndexPart::from_json_bytes(example.as_bytes()).unwrap();
+        assert_eq!(part, expected);
+    }
+
+    #[test]
+    fn v11_rel_size_migration_is_parsed() {
+        let example = r#"{
+            "version": 11,
+            "layer_metadata":{
+                "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__0000000001696070-00000000016960E9": { "file_size": 25600000 },
+                "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51": { "file_size": 9007199254741001 }
+            },
+            "disk_consistent_lsn":"0/16960E8",
+            "metadata": {
+                "disk_consistent_lsn": "0/16960E8",
+                "prev_record_lsn": "0/1696070",
+                "ancestor_timeline": "e45a7f37d3ee2ff17dc14bf4f4e3f52e",
+                "ancestor_lsn": "0/0",
+                "latest_gc_cutoff_lsn": "0/1696070",
+                "initdb_lsn": "0/1696070",
+                "pg_version": 14
+            },
+            "gc_blocking": {
+                "started_at": "2024-07-19T09:00:00.123",
+                "reasons": ["DetachAncestor"]
+            },
+            "import_pgdata": {
+                "V1": {
+                    "Done": {
+                        "idempotency_key": "specified-by-client-218a5213-5044-4562-a28d-d024c5f057f5",
+                        "started_at": "2024-11-13T09:23:42.123",
+                        "finished_at": "2024-11-13T09:42:23.123"
+                    }
+                }
+            },
+            "rel_size_migration": "legacy"
+        }"#;
+
+        let expected = IndexPart {
+            version: 11,
+            layer_metadata: HashMap::from([
+                ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__0000000001696070-00000000016960E9".parse().unwrap(), LayerFileMetadata {
+                    file_size: 25600000,
+                    generation: Generation::none(),
+                    shard: ShardIndex::unsharded()
+                }),
+                ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap(), LayerFileMetadata {
+                    file_size: 9007199254741001,
+                    generation: Generation::none(),
+                    shard: ShardIndex::unsharded()
+                })
+            ]),
+            disk_consistent_lsn: "0/16960E8".parse::<Lsn>().unwrap(),
+            metadata: TimelineMetadata::new(
+                Lsn::from_str("0/16960E8").unwrap(),
+                Some(Lsn::from_str("0/1696070").unwrap()),
+                Some(TimelineId::from_str("e45a7f37d3ee2ff17dc14bf4f4e3f52e").unwrap()),
+                Lsn::INVALID,
+                Lsn::from_str("0/1696070").unwrap(),
+                Lsn::from_str("0/1696070").unwrap(),
+                14,
+            ).with_recalculated_checksum().unwrap(),
+            deleted_at: None,
+            lineage: Default::default(),
+            gc_blocking: Some(GcBlocking {
+                started_at: parse_naive_datetime("2024-07-19T09:00:00.123000000"),
+                reasons: enumset::EnumSet::from_iter([GcBlockingReason::DetachAncestor]),
+            }),
+            last_aux_file_policy: Default::default(),
+            archived_at: None,
+            import_pgdata: Some(import_pgdata::index_part_format::Root::V1(import_pgdata::index_part_format::V1::Done(import_pgdata::index_part_format::Done{
+                started_at: parse_naive_datetime("2024-11-13T09:23:42.123000000"),
+                finished_at: parse_naive_datetime("2024-11-13T09:42:23.123000000"),
+                idempotency_key: import_pgdata::index_part_format::IdempotencyKey::new("specified-by-client-218a5213-5044-4562-a28d-d024c5f057f5".to_string()),
+            }))),
+            rel_size_migration: Some(RelSizeMigration::Legacy),
+            l2_lsn: None,
+        };
+
+        let part = IndexPart::from_json_bytes(example.as_bytes()).unwrap();
+        assert_eq!(part, expected);
+    }
+
+    #[test]
+    fn v12_l2_lsn_is_parsed() {
+        let example = r#"{
+            "version": 12,
+            "layer_metadata":{
+                "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__0000000001696070-00000000016960E9": { "file_size": 25600000 },
+                "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51": { "file_size": 9007199254741001 }
+            },
+            "disk_consistent_lsn":"0/16960E8",
+            "metadata": {
+                "disk_consistent_lsn": "0/16960E8",
+                "prev_record_lsn": "0/1696070",
+                "ancestor_timeline": "e45a7f37d3ee2ff17dc14bf4f4e3f52e",
+                "ancestor_lsn": "0/0",
+                "latest_gc_cutoff_lsn": "0/1696070",
+                "initdb_lsn": "0/1696070",
+                "pg_version": 14
+            },
+            "gc_blocking": {
+                "started_at": "2024-07-19T09:00:00.123",
+                "reasons": ["DetachAncestor"]
+            },
+            "import_pgdata": {
+                "V1": {
+                    "Done": {
+                        "idempotency_key": "specified-by-client-218a5213-5044-4562-a28d-d024c5f057f5",
+                        "started_at": "2024-11-13T09:23:42.123",
+                        "finished_at": "2024-11-13T09:42:23.123"
+                    }
+                }
+            },
+            "rel_size_migration": "legacy",
+            "l2_lsn": "0/16960E8"
+        }"#;
+
+        let expected = IndexPart {
+            version: 12,
+            layer_metadata: HashMap::from([
+                ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__0000000001696070-00000000016960E9".parse().unwrap(), LayerFileMetadata {
+                    file_size: 25600000,
+                    generation: Generation::none(),
+                    shard: ShardIndex::unsharded()
+                }),
+                ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap(), LayerFileMetadata {
+                    file_size: 9007199254741001,
+                    generation: Generation::none(),
+                    shard: ShardIndex::unsharded()
+                })
+            ]),
+            disk_consistent_lsn: "0/16960E8".parse::<Lsn>().unwrap(),
+            metadata: TimelineMetadata::new(
+                Lsn::from_str("0/16960E8").unwrap(),
+                Some(Lsn::from_str("0/1696070").unwrap()),
+                Some(TimelineId::from_str("e45a7f37d3ee2ff17dc14bf4f4e3f52e").unwrap()),
+                Lsn::INVALID,
+                Lsn::from_str("0/1696070").unwrap(),
+                Lsn::from_str("0/1696070").unwrap(),
+                14,
+            ).with_recalculated_checksum().unwrap(),
+            deleted_at: None,
+            lineage: Default::default(),
+            gc_blocking: Some(GcBlocking {
+                started_at: parse_naive_datetime("2024-07-19T09:00:00.123000000"),
+                reasons: enumset::EnumSet::from_iter([GcBlockingReason::DetachAncestor]),
+            }),
+            last_aux_file_policy: Default::default(),
+            archived_at: None,
+            import_pgdata: Some(import_pgdata::index_part_format::Root::V1(import_pgdata::index_part_format::V1::Done(import_pgdata::index_part_format::Done{
+                started_at: parse_naive_datetime("2024-11-13T09:23:42.123000000"),
+                finished_at: parse_naive_datetime("2024-11-13T09:42:23.123000000"),
+                idempotency_key: import_pgdata::index_part_format::IdempotencyKey::new("specified-by-client-218a5213-5044-4562-a28d-d024c5f057f5".to_string()),
+            }))),
+            rel_size_migration: Some(RelSizeMigration::Legacy),
+            l2_lsn: Some("0/16960E8".parse::<Lsn>().unwrap()),
         };
 
         let part = IndexPart::from_json_bytes(example.as_bytes()).unwrap();

pageserver/src/tenant/remote_timeline_client/upload.rs

@@ -40,6 +40,10 @@ pub(crate) async fn upload_index_part(
     });
     pausable_failpoint!("before-upload-index-pausable");
 
+    // Safety: refuse to persist invalid index metadata, to mitigate the impact of any bug that produces this
+    // (this should never happen)
+    index_part.validate().map_err(|e| anyhow::anyhow!(e))?;
+
     // FIXME: this error comes too late
     let serialized = index_part.to_json_bytes()?;
     let serialized = Bytes::from(serialized);

pageserver/src/tenant/storage_layer.rs

@@ -10,18 +10,26 @@ mod layer_desc;
 mod layer_name;
 pub mod merge_iterator;
 
+use crate::config::PageServerConf;
 use crate::context::{AccessStatsBehavior, RequestContext};
 use bytes::Bytes;
-use pageserver_api::key::{Key, NON_INHERITED_SPARSE_RANGE};
+use futures::stream::FuturesUnordered;
+use futures::StreamExt;
+use pageserver_api::key::Key;
 use pageserver_api::keyspace::{KeySpace, KeySpaceRandomAccum};
 use pageserver_api::record::NeonWalRecord;
 use pageserver_api::value::Value;
-use std::cmp::{Ordering, Reverse};
+use std::cmp::Ordering;
 use std::collections::hash_map::Entry;
 use std::collections::{BinaryHeap, HashMap};
+use std::future::Future;
 use std::ops::Range;
+use std::pin::Pin;
+use std::sync::atomic::AtomicUsize;
 use std::sync::Arc;
 use std::time::{Duration, SystemTime, UNIX_EPOCH};
+use tracing::{trace, Instrument};
+use utils::sync::gate::GateGuard;
 
 use utils::lsn::Lsn;
 
@@ -78,30 +86,151 @@ pub(crate) enum ValueReconstructSituation {
     Continue,
 }
 
-/// Reconstruct data accumulated for a single key during a vectored get
-#[derive(Debug, Default, Clone)]
-pub(crate) struct VectoredValueReconstructState {
-    pub(crate) records: Vec<(Lsn, NeonWalRecord)>,
-    pub(crate) img: Option<(Lsn, Bytes)>,
-
-    situation: ValueReconstructSituation,
+/// On disk representation of a value loaded in a buffer
+#[derive(Debug)]
+pub(crate) enum OnDiskValue {
+    /// Unencoded [`Value::Image`]
+    RawImage(Bytes),
+    /// Encoded [`Value`]. Can deserialize into an image or a WAL record
+    WalRecordOrImage(Bytes),
 }
 
-impl VectoredValueReconstructState {
-    fn get_cached_lsn(&self) -> Option<Lsn> {
-        self.img.as_ref().map(|img| img.0)
+/// Reconstruct data accumulated for a single key during a vectored get
+#[derive(Debug, Default)]
+pub(crate) struct VectoredValueReconstructState {
+    pub(crate) on_disk_values: Vec<(Lsn, OnDiskValueIoWaiter)>,
+
+    pub(crate) situation: ValueReconstructSituation,
+}
+
+#[derive(Debug)]
+pub(crate) struct OnDiskValueIoWaiter {
+    rx: tokio::sync::oneshot::Receiver<OnDiskValueIoResult>,
+}
+
+#[derive(Debug)]
+#[must_use]
+pub(crate) enum OnDiskValueIo {
+    /// Traversal identified this IO as required to complete the vectored get.
+    Required {
+        num_active_ios: Arc<AtomicUsize>,
+        tx: tokio::sync::oneshot::Sender<OnDiskValueIoResult>,
+    },
+    /// Sparse keyspace reads always read all the values for a given key,
+    /// even though only the first value is needed.
+    ///
+    /// This variant represents the unnecessary IOs for those values at lower LSNs
+    /// that aren't needed, but are currently still being done.
+    ///
+    /// The execution of unnecessary IOs was a pre-existing behavior before concurrent IO.
+    /// We added this explicit representation here so that we can drop
+    /// unnecessary IO results immediately, instead of buffering them in
+    /// `oneshot` channels inside [`VectoredValueReconstructState`] until
+    /// [`VectoredValueReconstructState::collect_pending_ios`] gets called.
+    Unnecessary,
+}
+
+type OnDiskValueIoResult = Result<OnDiskValue, std::io::Error>;
+
+impl OnDiskValueIo {
+    pub(crate) fn complete(self, res: OnDiskValueIoResult) {
+        match self {
+            OnDiskValueIo::Required { num_active_ios, tx } => {
+                num_active_ios.fetch_sub(1, std::sync::atomic::Ordering::Release);
+                let _ = tx.send(res);
+            }
+            OnDiskValueIo::Unnecessary => {
+                // Nobody cared, see variant doc comment.
+            }
+        }
     }
 }
 
-impl From<VectoredValueReconstructState> for ValueReconstructState {
-    fn from(mut state: VectoredValueReconstructState) -> Self {
-        // walredo expects the records to be descending in terms of Lsn
-        state.records.sort_by_key(|(lsn, _)| Reverse(*lsn));
+#[derive(Debug, thiserror::Error)]
+pub(crate) enum WaitCompletionError {
+    #[error("OnDiskValueIo was dropped without completing, likely the sidecar task panicked")]
+    IoDropped,
+}
 
-        ValueReconstructState {
-            records: state.records,
-            img: state.img,
+impl OnDiskValueIoWaiter {
+    pub(crate) async fn wait_completion(self) -> Result<OnDiskValueIoResult, WaitCompletionError> {
+        // NB: for Unnecessary IOs, this method never gets called because we don't add them to `on_disk_values`.
+        self.rx.await.map_err(|_| WaitCompletionError::IoDropped)
+    }
+}
+
+impl VectoredValueReconstructState {
+    /// # Cancel-Safety
+    ///
+    /// Technically fine to stop polling this future, but, the IOs will still
+    /// be executed to completion by the sidecar task and hold on to / consume resources.
+    /// Better not do it to make reasonsing about the system easier.
+    pub(crate) async fn collect_pending_ios(
+        self,
+    ) -> Result<ValueReconstructState, PageReconstructError> {
+        use utils::bin_ser::BeSer;
+
+        let mut res = Ok(ValueReconstructState::default());
+
+        // We should try hard not to bail early, so that by the time we return from this
+        // function, all IO for this value is done. It's not required -- we could totally
+        // stop polling the IO futures in the sidecar task, they need to support that,
+        // but just stopping to poll doesn't reduce the IO load on the disk. It's easier
+        // to reason about the system if we just wait for all IO to complete, even if
+        // we're no longer interested in the result.
+        //
+        // Revisit this when IO futures are replaced with a more sophisticated IO system
+        // and an IO scheduler, where we know which IOs were submitted and which ones
+        // just queued. Cf the comment on IoConcurrency::spawn_io.
+        for (lsn, waiter) in self.on_disk_values {
+            let value_recv_res = waiter
+                .wait_completion()
+                // we rely on the caller to poll us to completion, so this is not a bail point
+                .await;
+            // Force not bailing early by wrapping the code into a closure.
+            #[allow(clippy::redundant_closure_call)]
+            let _: () = (|| {
+                match (&mut res, value_recv_res) {
+                    (Err(_), _) => {
+                        // We've already failed, no need to process more.
+                    }
+                    (Ok(_), Err(wait_err)) => {
+                        // This shouldn't happen - likely the sidecar task panicked.
+                        res = Err(PageReconstructError::Other(wait_err.into()));
+                    }
+                    (Ok(_), Ok(Err(err))) => {
+                        let err: std::io::Error = err;
+                        // TODO: returning IO error here will fail a compute query.
+                        // Probably not what we want, we're not doing `maybe_fatal_err`
+                        // in the IO futures.
+                        // But it's been like that for a long time, not changing it
+                        // as part of concurrent IO.
+                        // => https://github.com/neondatabase/neon/issues/10454
+                        res = Err(PageReconstructError::Other(err.into()));
+                    }
+                    (Ok(ok), Ok(Ok(OnDiskValue::RawImage(img)))) => {
+                        assert!(ok.img.is_none());
+                        ok.img = Some((lsn, img));
+                    }
+                    (Ok(ok), Ok(Ok(OnDiskValue::WalRecordOrImage(buf)))) => {
+                        match Value::des(&buf) {
+                            Ok(Value::WalRecord(rec)) => {
+                                ok.records.push((lsn, rec));
+                            }
+                            Ok(Value::Image(img)) => {
+                                assert!(ok.img.is_none());
+                                ok.img = Some((lsn, img));
+                            }
+                            Err(err) => {
+                                res = Err(PageReconstructError::Other(err.into()));
+                            }
+                        }
+                    }
+                }
+            })();
         }
+
+        res
     }
 }
 
@@ -109,7 +238,7 @@ impl From<VectoredValueReconstructState> for ValueReconstructState {
 pub(crate) struct ValuesReconstructState {
     /// The keys will be removed after `get_vectored` completes. The caller outside `Timeline`
     /// should not expect to get anything from this hashmap.
-    pub(crate) keys: HashMap<Key, Result<VectoredValueReconstructState, PageReconstructError>>,
+    pub(crate) keys: HashMap<Key, VectoredValueReconstructState>,
     /// The keys which are already retrieved
     keys_done: KeySpaceRandomAccum,
 
@@ -119,27 +248,365 @@ pub(crate) struct ValuesReconstructState {
     // Statistics that are still accessible as a caller of `get_vectored_impl`.
     layers_visited: u32,
     delta_layers_visited: u32,
+
+    pub(crate) io_concurrency: IoConcurrency,
+    num_active_ios: Arc<AtomicUsize>,
+}
+
+/// The level of IO concurrency to be used on the read path
+///
+/// The desired end state is that we always do parallel IO.
+/// This struct and the dispatching in the impl will be removed once
+/// we've built enough confidence.
+pub(crate) enum IoConcurrency {
+    Sequential,
+    SidecarTask {
+        task_id: usize,
+        ios_tx: tokio::sync::mpsc::UnboundedSender<IoFuture>,
+    },
+}
+
+type IoFuture = Pin<Box<dyn Send + Future<Output = ()>>>;
+
+pub(crate) enum SelectedIoConcurrency {
+    Sequential,
+    SidecarTask(GateGuard),
+}
+
+impl std::fmt::Debug for IoConcurrency {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        match self {
+            IoConcurrency::Sequential => write!(f, "Sequential"),
+            IoConcurrency::SidecarTask { .. } => write!(f, "SidecarTask"),
+        }
+    }
+}
+
+impl std::fmt::Debug for SelectedIoConcurrency {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        match self {
+            SelectedIoConcurrency::Sequential => write!(f, "Sequential"),
+            SelectedIoConcurrency::SidecarTask(_) => write!(f, "SidecarTask"),
+        }
+    }
+}
+
+impl IoConcurrency {
+    /// Force sequential IO. This is a temporary workaround until we have
+    /// moved plumbing-through-the-call-stack
+    /// of IoConcurrency into `RequestContextq.
+    ///
+    /// DO NOT USE for new code.
+    ///
+    /// Tracking issue: <https://github.com/neondatabase/neon/issues/10460>.
+    pub(crate) fn sequential() -> Self {
+        Self::spawn(SelectedIoConcurrency::Sequential)
+    }
+
+    pub(crate) fn spawn_from_conf(
+        conf: &'static PageServerConf,
+        gate_guard: GateGuard,
+    ) -> IoConcurrency {
+        use pageserver_api::config::GetVectoredConcurrentIo;
+        let selected = match conf.get_vectored_concurrent_io {
+            GetVectoredConcurrentIo::Sequential => SelectedIoConcurrency::Sequential,
+            GetVectoredConcurrentIo::SidecarTask => SelectedIoConcurrency::SidecarTask(gate_guard),
+        };
+        Self::spawn(selected)
+    }
+
+    pub(crate) fn spawn(io_concurrency: SelectedIoConcurrency) -> Self {
+        match io_concurrency {
+            SelectedIoConcurrency::Sequential => IoConcurrency::Sequential,
+            SelectedIoConcurrency::SidecarTask(gate_guard) => {
+                let (ios_tx, ios_rx) = tokio::sync::mpsc::unbounded_channel();
+                static TASK_ID: AtomicUsize = AtomicUsize::new(0);
+                let task_id = TASK_ID.fetch_add(1, std::sync::atomic::Ordering::Relaxed);
+                // TODO: enrich the span with more context (tenant,shard,timeline) + (basebackup|pagestream|...)
+                let span =
+                    tracing::info_span!(parent: None, "IoConcurrency_sidecar", task_id = task_id);
+                trace!(task_id, "spawning sidecar task");
+                tokio::spawn(async move {
+                    trace!("start");
+                    scopeguard::defer!{ trace!("end") };
+                    type IosRx = tokio::sync::mpsc::UnboundedReceiver<IoFuture>;
+                    enum State {
+                        Waiting {
+                            // invariant: is_empty(), but we recycle the allocation
+                            empty_futures: FuturesUnordered<IoFuture>,
+                            ios_rx: IosRx,
+                        },
+                        Executing {
+                            futures: FuturesUnordered<IoFuture>,
+                            ios_rx: IosRx,
+                        },
+                        ShuttingDown {
+                            futures: FuturesUnordered<IoFuture>,
+                        },
+                    }
+                    let mut state = State::Waiting {
+                        empty_futures: FuturesUnordered::new(),
+                        ios_rx,
+                    };
+                    loop {
+                        match state {
+                            State::Waiting {
+                                empty_futures,
+                                mut ios_rx,
+                            } => {
+                                assert!(empty_futures.is_empty());
+                                tokio::select! {
+                                    fut = ios_rx.recv() => {
+                                        if let Some(fut) = fut {
+                                            trace!("received new io future");
+                                            empty_futures.push(fut);
+                                            state = State::Executing { futures: empty_futures, ios_rx };
+                                        } else {
+                                            state = State::ShuttingDown { futures: empty_futures }
+                                        }
+                                    }
+                                }
+                            }
+                            State::Executing {
+                                mut futures,
+                                mut ios_rx,
+                            } => {
+                                tokio::select! {
+                                    res = futures.next() => {
+                                        trace!("io future completed");
+                                        assert!(res.is_some());
+                                        if futures.is_empty() {
+                                            state = State::Waiting { empty_futures: futures, ios_rx};
+                                        } else {
+                                            state = State::Executing { futures, ios_rx };
+                                        }
+                                    }
+                                    fut = ios_rx.recv() => {
+                                        if let Some(fut) = fut {
+                                            trace!("received new io future");
+                                            futures.push(fut);
+                                            state =  State::Executing { futures, ios_rx};
+                                        } else {
+                                            state = State::ShuttingDown { futures };
+                                        }
+                                    }
+                                }
+                            }
+                            State::ShuttingDown {
+                                mut futures,
+                            } => {
+                                trace!("shutting down");
+                                while let Some(()) = futures.next().await {
+                                    trace!("io future completed (shutdown)");
+                                    // drain
+                                }
+                                trace!("shutdown complete");
+                                break;
+                            }
+                        }
+                    }
+                    drop(gate_guard); // drop it right before we exit
+                }.instrument(span));
+                IoConcurrency::SidecarTask { task_id, ios_tx }
+            }
+        }
+    }
+
+    pub(crate) fn clone(&self) -> Self {
+        match self {
+            IoConcurrency::Sequential => IoConcurrency::Sequential,
+            IoConcurrency::SidecarTask { task_id, ios_tx } => IoConcurrency::SidecarTask {
+                task_id: *task_id,
+                ios_tx: ios_tx.clone(),
+            },
+        }
+    }
+
+    /// Submit an IO to be executed in the background. DEADLOCK RISK, read the full doc string.
+    ///
+    /// The IO is represented as an opaque future.
+    /// IO completion must be handled inside the future, e.g., through a oneshot channel.
+    ///
+    /// The API seems simple but there are multiple **pitfalls** involving
+    /// DEADLOCK RISK.
+    ///
+    /// First, there are no guarantees about the exexecution of the IO.
+    /// It may be `await`ed in-place before this function returns.
+    /// It may be polled partially by this task and handed off to another task to be finished.
+    /// It may be polled and then dropped before returning ready.
+    ///
+    /// This means that submitted IOs must not be interedependent.
+    /// Interdependence may be through shared limited resources, e.g.,
+    /// - VirtualFile file descriptor cache slot acquisition
+    /// - tokio-epoll-uring slot
+    ///
+    /// # Why current usage is safe from deadlocks
+    ///
+    /// Textbook condition for a deadlock is that _all_ of the following be given
+    /// - Mutual exclusion
+    /// - Hold and wait
+    /// - No preemption
+    /// - Circular wait
+    ///
+    /// The current usage is safe because:
+    /// - Mutual exclusion: IO futures definitely use mutexes, no way around that for now
+    /// - Hold and wait: IO futures currently hold two kinds of locks/resources while waiting
+    ///   for acquisition of other resources:
+    ///    - VirtualFile file descriptor cache slot tokio mutex
+    ///    - tokio-epoll-uring slot (uses tokio notify => wait queue, much like mutex)
+    /// - No preemption: there's no taking-away of acquired locks/resources => given
+    /// - Circular wait: this is the part of the condition that isn't met: all IO futures
+    ///   first acquire VirtualFile mutex, then tokio-epoll-uring slot.
+    ///   There is no IO future that acquires slot before VirtualFile.
+    ///   Hence there can be no circular waiting.
+    ///   Hence there cannot be a deadlock.
+    ///
+    /// This is a very fragile situation and must be revisited whenver any code called from
+    /// inside the IO futures is changed.
+    ///
+    /// We will move away from opaque IO futures towards well-defined IOs at some point in
+    /// the future when we have shipped this first version of concurrent IO to production
+    /// and are ready to retire the Sequential mode which runs the futures in place.
+    /// Right now, while brittle, the opaque IO approach allows us to ship the feature
+    /// with minimal changes to the code and minimal changes to existing behavior in Sequential mode.
+    ///
+    /// Also read the comment in `collect_pending_ios`.
+    pub(crate) async fn spawn_io<F>(&mut self, fut: F)
+    where
+        F: std::future::Future<Output = ()> + Send + 'static,
+    {
+        match self {
+            IoConcurrency::Sequential => fut.await,
+            IoConcurrency::SidecarTask { ios_tx, .. } => {
+                let fut = Box::pin(fut);
+                // NB: experiments showed that doing an opportunistic poll of `fut` here was bad for throughput
+                // while insignificant for latency.
+                // It would make sense to revisit the tokio-epoll-uring API in the future such that we can try
+                // a submission here, but never poll the future. That way, io_uring can make proccess while
+                // the future sits in the ios_tx queue.
+                match ios_tx.send(fut) {
+                    Ok(()) => {}
+                    Err(_) => {
+                        unreachable!("the io task must have exited, likely it panicked")
+                    }
+                }
+            }
+        }
+    }
+
+    #[cfg(test)]
+    pub(crate) fn spawn_for_test() -> impl std::ops::DerefMut<Target = Self> {
+        use std::ops::{Deref, DerefMut};
+        use tracing::info;
+        use utils::sync::gate::Gate;
+
+        // Spawn needs a Gate, give it one.
+        struct Wrapper {
+            inner: IoConcurrency,
+            #[allow(dead_code)]
+            gate: Box<Gate>,
+        }
+        impl Deref for Wrapper {
+            type Target = IoConcurrency;
+
+            fn deref(&self) -> &Self::Target {
+                &self.inner
+            }
+        }
+        impl DerefMut for Wrapper {
+            fn deref_mut(&mut self) -> &mut Self::Target {
+                &mut self.inner
+            }
+        }
+        let gate = Box::new(Gate::default());
+
+        // The default behavior when running Rust unit tests without any further
+        // flags is to use the new behavior.
+        // The CI uses the following environment variable to unit test both old
+        // and new behavior.
+        // NB: the Python regression & perf tests take the `else` branch
+        // below and have their own defaults management.
+        let selected = {
+            // The pageserver_api::config type is unsuitable because it's internally tagged.
+            #[derive(serde::Deserialize)]
+            #[serde(rename_all = "kebab-case")]
+            enum TestOverride {
+                Sequential,
+                SidecarTask,
+            }
+            use once_cell::sync::Lazy;
+            static TEST_OVERRIDE: Lazy<TestOverride> = Lazy::new(|| {
+                utils::env::var_serde_json_string(
+                    "NEON_PAGESERVER_UNIT_TEST_GET_VECTORED_CONCURRENT_IO",
+                )
+                .unwrap_or(TestOverride::SidecarTask)
+            });
+
+            match *TEST_OVERRIDE {
+                TestOverride::Sequential => SelectedIoConcurrency::Sequential,
+                TestOverride::SidecarTask => {
+                    SelectedIoConcurrency::SidecarTask(gate.enter().expect("just created it"))
+                }
+            }
+        };
+
+        info!(?selected, "get_vectored_concurrent_io test");
+
+        Wrapper {
+            inner: Self::spawn(selected),
+            gate,
+        }
+    }
+}
+
+/// Make noise in case the [`ValuesReconstructState`] gets dropped while
+/// there are still IOs in flight.
+/// Refer to `collect_pending_ios` for why we prefer not to do that.
+//
+/// We log from here instead of from the sidecar task because the [`ValuesReconstructState`]
+/// gets dropped in a tracing span with more context.
+/// We repeat the sidecar tasks's `task_id` so we can correlate what we emit here with
+/// the logs / panic handler logs from the sidecar task, which also logs the `task_id`.
+impl Drop for ValuesReconstructState {
+    fn drop(&mut self) {
+        let num_active_ios = self
+            .num_active_ios
+            .load(std::sync::atomic::Ordering::Acquire);
+        if num_active_ios == 0 {
+            return;
+        }
+        let sidecar_task_id = match &self.io_concurrency {
+            IoConcurrency::Sequential => None,
+            IoConcurrency::SidecarTask { task_id, .. } => Some(*task_id),
+        };
+        tracing::warn!(
+            num_active_ios,
+            ?sidecar_task_id,
+            backtrace=%std::backtrace::Backtrace::force_capture(),
+            "dropping ValuesReconstructState while some IOs have not been completed",
+        );
+    }
 }
 
 impl ValuesReconstructState {
-    pub(crate) fn new() -> Self {
+    pub(crate) fn new(io_concurrency: IoConcurrency) -> Self {
         Self {
             keys: HashMap::new(),
             keys_done: KeySpaceRandomAccum::new(),
             keys_with_image_coverage: None,
             layers_visited: 0,
             delta_layers_visited: 0,
+            io_concurrency,
+            num_active_ios: Arc::new(AtomicUsize::new(0)),
         }
     }
 
-    /// Associate a key with the error which it encountered and mark it as done
-    pub(crate) fn on_key_error(&mut self, key: Key, err: PageReconstructError) {
-        let previous = self.keys.insert(key, Err(err));
-        if let Some(Ok(state)) = previous {
-            if state.situation == ValueReconstructSituation::Continue {
-                self.keys_done.add_key(key);
-            }
-        }
+    /// Absolutely read [`IoConcurrency::spawn_io`] to learn about assumptions & pitfalls.
+    pub(crate) async fn spawn_io<F>(&mut self, fut: F)
+    where
+        F: std::future::Future<Output = ()> + Send + 'static,
+    {
+        self.io_concurrency.spawn_io(fut).await;
     }
 
     pub(crate) fn on_layer_visited(&mut self, layer: &ReadableLayer) {
@@ -159,29 +626,6 @@ impl ValuesReconstructState {
         self.layers_visited
     }
 
-    /// This function is called after reading a keyspace from a layer.
-    /// It checks if the read path has now moved past the cached Lsn for any keys.
-    ///
-    /// Implementation note: We intentionally iterate over the keys for which we've
-    /// already collected some reconstruct data. This avoids scaling complexity with
-    /// the size of the search space.
-    pub(crate) fn on_lsn_advanced(&mut self, keyspace: &KeySpace, advanced_to: Lsn) {
-        for (key, value) in self.keys.iter_mut() {
-            if !keyspace.contains(key) {
-                continue;
-            }
-
-            if let Ok(state) = value {
-                if state.situation != ValueReconstructSituation::Complete
-                    && state.get_cached_lsn() >= Some(advanced_to)
-                {
-                    state.situation = ValueReconstructSituation::Complete;
-                    self.keys_done.add_key(*key);
-                }
-            }
-        }
-    }
-
     /// On hitting image layer, we can mark all keys in this range as done, because
     /// if the image layer does not contain a key, it is deleted/never added.
     pub(crate) fn on_image_layer_visited(&mut self, key_range: &Range<Key>) {
@@ -199,70 +643,42 @@ impl ValuesReconstructState {
     ///
     /// If the key is in the sparse keyspace (i.e., aux files), we do not track them in
     /// `key_done`.
-    pub(crate) fn update_key(
-        &mut self,
-        key: &Key,
-        lsn: Lsn,
-        value: Value,
-    ) -> ValueReconstructSituation {
-        let state = self
-            .keys
-            .entry(*key)
-            .or_insert(Ok(VectoredValueReconstructState::default()));
-        let is_sparse_key = NON_INHERITED_SPARSE_RANGE.contains(key);
-        if let Ok(state) = state {
-            let key_done = match state.situation {
-                ValueReconstructSituation::Complete => {
-                    if is_sparse_key {
-                        // Sparse keyspace might be visited multiple times because
-                        // we don't track unmapped keyspaces.
-                        return ValueReconstructSituation::Complete;
-                    } else {
-                        unreachable!()
-                    }
-                }
-                ValueReconstructSituation::Continue => match value {
-                    Value::Image(img) => {
-                        state.img = Some((lsn, img));
-                        true
-                    }
-                    Value::WalRecord(rec) => {
-                        debug_assert!(
-                            Some(lsn) > state.get_cached_lsn(),
-                            "Attempt to collect a record below cached LSN for walredo: {} < {}",
-                            lsn,
-                            state
-                                .get_cached_lsn()
-                                .expect("Assertion can only fire if a cached lsn is present")
-                        );
+    // TODO: rename this method & update description.
+    pub(crate) fn update_key(&mut self, key: &Key, lsn: Lsn, completes: bool) -> OnDiskValueIo {
+        let state = self.keys.entry(*key).or_default();
 
-                        let will_init = rec.will_init();
-                        state.records.push((lsn, rec));
-                        will_init
-                    }
-                },
-            };
+        let is_sparse_key = key.is_sparse();
 
-            if key_done && state.situation == ValueReconstructSituation::Continue {
-                state.situation = ValueReconstructSituation::Complete;
-                if !is_sparse_key {
-                    self.keys_done.add_key(*key);
+        let required_io = match state.situation {
+            ValueReconstructSituation::Complete => {
+                if is_sparse_key {
+                    // Sparse keyspace might be visited multiple times because
+                    // we don't track unmapped keyspaces.
+                    return OnDiskValueIo::Unnecessary;
+                } else {
+                    unreachable!()
                 }
             }
+            ValueReconstructSituation::Continue => {
+                self.num_active_ios
+                    .fetch_add(1, std::sync::atomic::Ordering::Release);
+                let (tx, rx) = tokio::sync::oneshot::channel();
+                state.on_disk_values.push((lsn, OnDiskValueIoWaiter { rx }));
+                OnDiskValueIo::Required {
+                    tx,
+                    num_active_ios: Arc::clone(&self.num_active_ios),
+                }
+            }
+        };
 
-            state.situation
-        } else {
-            ValueReconstructSituation::Complete
+        if completes && state.situation == ValueReconstructSituation::Continue {
+            state.situation = ValueReconstructSituation::Complete;
+            if !is_sparse_key {
+                self.keys_done.add_key(*key);
+            }
         }
-    }
 
-    /// Returns the Lsn at which this key is cached if one exists.
-    /// The read path should go no further than this Lsn for the given key.
-    pub(crate) fn get_cached_lsn(&self, key: &Key) -> Option<Lsn> {
-        self.keys
-            .get(key)
-            .and_then(|k| k.as_ref().ok())
-            .and_then(|state| state.get_cached_lsn())
+        required_io
     }
 
     /// Returns the key space describing the keys that have
@@ -276,12 +692,6 @@ impl ValuesReconstructState {
     }
 }
 
-impl Default for ValuesReconstructState {
-    fn default() -> Self {
-        Self::new()
-    }
-}
-
 /// A key that uniquely identifies a layer in a timeline
 #[derive(Debug, PartialEq, Eq, Clone, Hash)]
 pub(crate) enum LayerId {
@@ -720,3 +1130,78 @@ impl<T: std::fmt::Display> std::fmt::Debug for RangeDisplayDebug<'_, T> {
         write!(f, "{}..{}", self.0.start, self.0.end)
     }
 }
+
+#[cfg(test)]
+mod tests2 {
+    use pageserver_api::key::DBDIR_KEY;
+    use tracing::info;
+
+    use super::*;
+    use crate::tenant::storage_layer::IoConcurrency;
+
+    /// TODO: currently this test relies on manual visual inspection of the --no-capture output.
+    /// Should look like so:
+    /// ```text
+    /// RUST_LOG=trace cargo nextest run  --features testing  --no-capture test_io_concurrency_noise
+    /// running 1 test
+    /// 2025-01-21T17:42:01.335679Z  INFO get_vectored_concurrent_io test selected=SidecarTask
+    /// 2025-01-21T17:42:01.335680Z TRACE spawning sidecar task task_id=0
+    /// 2025-01-21T17:42:01.335937Z TRACE IoConcurrency_sidecar{task_id=0}: start
+    /// 2025-01-21T17:42:01.335972Z TRACE IoConcurrency_sidecar{task_id=0}: received new io future
+    /// 2025-01-21T17:42:01.335999Z  INFO IoConcurrency_sidecar{task_id=0}: waiting for signal to complete IO
+    /// 2025-01-21T17:42:01.336229Z  WARN dropping ValuesReconstructState while some IOs have not been completed num_active_ios=1 sidecar_task_id=Some(0) backtrace=   0: <pageserver::tenant::storage_layer::ValuesReconstructState as core::ops::drop::Drop>::drop
+    ///              at ./src/tenant/storage_layer.rs:553:24
+    ///    1: core::ptr::drop_in_place<pageserver::tenant::storage_layer::ValuesReconstructState>
+    ///              at /home/christian/.rustup/toolchains/1.84.0-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/ptr/mod.rs:521:1
+    ///    2: core::mem::drop
+    ///              at /home/christian/.rustup/toolchains/1.84.0-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/mem/mod.rs:942:24
+    ///    3: pageserver::tenant::storage_layer::tests2::test_io_concurrency_noise::{{closure}}
+    ///              at ./src/tenant/storage_layer.rs:1159:9
+    ///   ...
+    ///   49: <unknown>
+    /// 2025-01-21T17:42:01.452293Z  INFO IoConcurrency_sidecar{task_id=0}: completing IO
+    /// 2025-01-21T17:42:01.452357Z TRACE IoConcurrency_sidecar{task_id=0}: io future completed
+    /// 2025-01-21T17:42:01.452473Z TRACE IoConcurrency_sidecar{task_id=0}: end
+    /// test tenant::storage_layer::tests2::test_io_concurrency_noise ... ok
+    ///
+    /// ```
+    #[tokio::test]
+    async fn test_io_concurrency_noise() {
+        crate::tenant::harness::setup_logging();
+
+        let io_concurrency = IoConcurrency::spawn_for_test();
+        match *io_concurrency {
+            IoConcurrency::Sequential => {
+                // This test asserts behavior in sidecar mode, doesn't make sense in sequential mode.
+                return;
+            }
+            IoConcurrency::SidecarTask { .. } => {}
+        }
+        let mut reconstruct_state = ValuesReconstructState::new(io_concurrency.clone());
+
+        let (io_fut_is_waiting_tx, io_fut_is_waiting) = tokio::sync::oneshot::channel();
+        let (do_complete_io, should_complete_io) = tokio::sync::oneshot::channel();
+        let (io_fut_exiting_tx, io_fut_exiting) = tokio::sync::oneshot::channel();
+
+        let io = reconstruct_state.update_key(&DBDIR_KEY, Lsn(8), true);
+        reconstruct_state
+            .spawn_io(async move {
+                info!("waiting for signal to complete IO");
+                io_fut_is_waiting_tx.send(()).unwrap();
+                should_complete_io.await.unwrap();
+                info!("completing IO");
+                io.complete(Ok(OnDiskValue::RawImage(Bytes::new())));
+                io_fut_exiting_tx.send(()).unwrap();
+            })
+            .await;
+
+        io_fut_is_waiting.await.unwrap();
+
+        // this is what makes the noise
+        drop(reconstruct_state);
+
+        do_complete_io.send(()).unwrap();
+
+        io_fut_exiting.await.unwrap();
+    }
+}

pageserver/src/tenant/storage_layer/delta_layer.rs

@@ -41,13 +41,12 @@ use crate::tenant::vectored_blob_io::{
     BlobFlag, BufView, StreamingVectoredReadPlanner, VectoredBlobReader, VectoredRead,
     VectoredReadPlanner,
 };
-use crate::tenant::PageReconstructError;
 use crate::virtual_file::owned_buffers_io::io_buf_ext::{FullSlice, IoBufExt};
 use crate::virtual_file::IoBufferMut;
 use crate::virtual_file::{self, MaybeFatalIo, VirtualFile};
 use crate::TEMP_FILE_SUFFIX;
 use crate::{DELTA_FILE_MAGIC, STORAGE_FORMAT_VERSION};
-use anyhow::{anyhow, bail, ensure, Context, Result};
+use anyhow::{bail, ensure, Context, Result};
 use camino::{Utf8Path, Utf8PathBuf};
 use futures::StreamExt;
 use itertools::Itertools;
@@ -60,7 +59,7 @@ use pageserver_api::shard::TenantShardId;
 use pageserver_api::value::Value;
 use rand::{distributions::Alphanumeric, Rng};
 use serde::{Deserialize, Serialize};
-use std::collections::VecDeque;
+use std::collections::{HashMap, VecDeque};
 use std::fs::File;
 use std::io::SeekFrom;
 use std::ops::Range;
@@ -77,7 +76,10 @@ use utils::{
     lsn::Lsn,
 };
 
-use super::{AsLayerDesc, LayerName, PersistentLayerDesc, ValuesReconstructState};
+use super::{
+    AsLayerDesc, LayerName, OnDiskValue, OnDiskValueIo, PersistentLayerDesc, ResidentLayer,
+    ValuesReconstructState,
+};
 
 ///
 /// Header stored in the beginning of the file
@@ -226,7 +228,7 @@ pub struct DeltaLayerInner {
     index_start_blk: u32,
     index_root_blk: u32,
 
-    file: VirtualFile,
+    file: Arc<VirtualFile>,
     file_id: FileId,
 
     layer_key_range: Range<Key>,
@@ -795,9 +797,11 @@ impl DeltaLayerInner {
         max_vectored_read_bytes: Option<MaxVectoredReadBytes>,
         ctx: &RequestContext,
     ) -> anyhow::Result<Self> {
-        let file = VirtualFile::open_v2(path, ctx)
-            .await
-            .context("open layer file")?;
+        let file = Arc::new(
+            VirtualFile::open_v2(path, ctx)
+                .await
+                .context("open layer file")?,
+        );
 
         let file_id = page_cache::next_file_id();
 
@@ -842,12 +846,11 @@ impl DeltaLayerInner {
     // Look up the keys in the provided keyspace and update
     // the reconstruct state with whatever is found.
     //
-    // If the key is cached, go no further than the cached Lsn.
-    //
     // Currently, the index is visited for each range, but this
     // can be further optimised to visit the index only once.
     pub(super) async fn get_values_reconstruct_data(
         &self,
+        this: ResidentLayer,
         keyspace: KeySpace,
         lsn_range: Range<Lsn>,
         reconstruct_state: &mut ValuesReconstructState,
@@ -875,17 +878,14 @@ impl DeltaLayerInner {
             data_end_offset,
             index_reader,
             planner,
-            reconstruct_state,
             ctx,
         )
         .await
         .map_err(GetVectoredError::Other)?;
 
-        self.do_reads_and_update_state(reads, reconstruct_state, ctx)
+        self.do_reads_and_update_state(this, reads, reconstruct_state, ctx)
             .await;
 
-        reconstruct_state.on_lsn_advanced(&keyspace, lsn_range.start);
-
         Ok(())
     }
 
@@ -895,7 +895,6 @@ impl DeltaLayerInner {
         data_end_offset: u64,
         index_reader: DiskBtreeReader<Reader, DELTA_KEY_SIZE>,
         mut planner: VectoredReadPlanner,
-        reconstruct_state: &mut ValuesReconstructState,
         ctx: &RequestContext,
     ) -> anyhow::Result<Vec<VectoredRead>>
     where
@@ -922,10 +921,9 @@ impl DeltaLayerInner {
                 assert!(key >= range.start);
 
                 let outside_lsn_range = !lsn_range.contains(&lsn);
-                let below_cached_lsn = reconstruct_state.get_cached_lsn(&key) >= Some(lsn);
 
                 let flag = {
-                    if outside_lsn_range || below_cached_lsn {
+                    if outside_lsn_range {
                         BlobFlag::Ignore
                     } else if blob_ref.will_init() {
                         BlobFlag::ReplaceAll
@@ -994,98 +992,78 @@ impl DeltaLayerInner {
 
     async fn do_reads_and_update_state(
         &self,
+        this: ResidentLayer,
         reads: Vec<VectoredRead>,
         reconstruct_state: &mut ValuesReconstructState,
         ctx: &RequestContext,
     ) {
-        let vectored_blob_reader = VectoredBlobReader::new(&self.file);
-        let mut ignore_key_with_err = None;
-
         let max_vectored_read_bytes = self
             .max_vectored_read_bytes
             .expect("Layer is loaded with max vectored bytes config")
             .0
             .into();
         let buf_size = Self::get_min_read_buffer_size(&reads, max_vectored_read_bytes);
-        let mut buf = Some(IoBufferMut::with_capacity(buf_size));
 
         // Note that reads are processed in reverse order (from highest key+lsn).
         // This is the order that `ReconstructState` requires such that it can
         // track when a key is done.
         for read in reads.into_iter().rev() {
-            let res = vectored_blob_reader
-                .read_blobs(&read, buf.take().expect("Should have a buffer"), ctx)
-                .await;
-
-            let blobs_buf = match res {
-                Ok(blobs_buf) => blobs_buf,
-                Err(err) => {
-                    let kind = err.kind();
-                    for (_, blob_meta) in read.blobs_at.as_slice() {
-                        reconstruct_state.on_key_error(
-                            blob_meta.key,
-                            PageReconstructError::Other(anyhow!(
-                                "Failed to read blobs from virtual file {}: {}",
-                                self.file.path(),
-                                kind
-                            )),
-                        );
-                    }
-
-                    // We have "lost" the buffer since the lower level IO api
-                    // doesn't return the buffer on error. Allocate a new one.
-                    buf = Some(IoBufferMut::with_capacity(buf_size));
-
-                    continue;
-                }
-            };
-            let view = BufView::new_slice(&blobs_buf.buf);
-            for meta in blobs_buf.blobs.iter().rev() {
-                if Some(meta.meta.key) == ignore_key_with_err {
-                    continue;
-                }
-                let blob_read = meta.read(&view).await;
-                let blob_read = match blob_read {
-                    Ok(buf) => buf,
-                    Err(e) => {
-                        reconstruct_state.on_key_error(
-                            meta.meta.key,
-                            PageReconstructError::Other(anyhow!(e).context(format!(
-                                "Failed to decompress blob from virtual file {}",
-                                self.file.path(),
-                            ))),
-                        );
-
-                        ignore_key_with_err = Some(meta.meta.key);
-                        continue;
-                    }
-                };
-
-                let value = Value::des(&blob_read);
-
-                let value = match value {
-                    Ok(v) => v,
-                    Err(e) => {
-                        reconstruct_state.on_key_error(
-                            meta.meta.key,
-                            PageReconstructError::Other(anyhow!(e).context(format!(
-                                "Failed to deserialize blob from virtual file {}",
-                                self.file.path(),
-                            ))),
-                        );
-
-                        ignore_key_with_err = Some(meta.meta.key);
-                        continue;
-                    }
-                };
-
-                // Invariant: once a key reaches [`ValueReconstructSituation::Complete`]
-                // state, no further updates shall be made to it. The call below will
-                // panic if the invariant is violated.
-                reconstruct_state.update_key(&meta.meta.key, meta.meta.lsn, value);
+            let mut ios: HashMap<(Key, Lsn), OnDiskValueIo> = Default::default();
+            for (_, blob_meta) in read.blobs_at.as_slice().iter().rev() {
+                let io = reconstruct_state.update_key(
+                    &blob_meta.key,
+                    blob_meta.lsn,
+                    blob_meta.will_init,
+                );
+                ios.insert((blob_meta.key, blob_meta.lsn), io);
             }
 
-            buf = Some(blobs_buf.buf);
+            let read_extend_residency = this.clone();
+            let read_from = self.file.clone();
+            let read_ctx = ctx.attached_child();
+            reconstruct_state
+                .spawn_io(async move {
+                    let vectored_blob_reader = VectoredBlobReader::new(&read_from);
+                    let buf = IoBufferMut::with_capacity(buf_size);
+
+                    let res = vectored_blob_reader.read_blobs(&read, buf, &read_ctx).await;
+                    match res {
+                        Ok(blobs_buf) => {
+                            let view = BufView::new_slice(&blobs_buf.buf);
+                            for meta in blobs_buf.blobs.iter().rev() {
+                                let io = ios.remove(&(meta.meta.key, meta.meta.lsn)).unwrap();
+
+                                let blob_read = meta.read(&view).await;
+                                let blob_read = match blob_read {
+                                    Ok(buf) => buf,
+                                    Err(e) => {
+                                        io.complete(Err(e));
+                                        continue;
+                                    }
+                                };
+
+                                io.complete(Ok(OnDiskValue::WalRecordOrImage(
+                                    blob_read.into_bytes(),
+                                )));
+                            }
+
+                            assert!(ios.is_empty());
+                        }
+                        Err(err) => {
+                            for (_, sender) in ios {
+                                sender.complete(Err(std::io::Error::new(
+                                    err.kind(),
+                                    "vec read failed",
+                                )));
+                            }
+                        }
+                    }
+
+                    // keep layer resident until this IO is done; this spawned IO future generally outlives the
+                    // call to `self` / the `Arc<DownloadedLayer>` / the `ResidentLayer` that guarantees residency
+                    drop(read_extend_residency);
+                })
+                .await;
         }
     }
 
@@ -1224,7 +1202,14 @@ impl DeltaLayerInner {
             let actionable = if let Some((key, lsn, start_offset)) = prev.take() {
                 let end_offset = offset;
 
-                Some((BlobMeta { key, lsn }, start_offset..end_offset))
+                Some((
+                    BlobMeta {
+                        key,
+                        lsn,
+                        will_init: false,
+                    },
+                    start_offset..end_offset,
+                ))
             } else {
                 None
             };
@@ -1560,7 +1545,9 @@ impl DeltaLayerIterator<'_> {
                 let lsn = DeltaKey::extract_lsn_from_buf(&raw_key);
                 let blob_ref = BlobRef(value);
                 let offset = blob_ref.pos();
-                if let Some(batch_plan) = self.planner.handle(key, lsn, offset) {
+                if let Some(batch_plan) =
+                    self.planner.handle(key, lsn, offset, blob_ref.will_init())
+                {
                     break batch_plan;
                 }
             } else {
@@ -1673,7 +1660,6 @@ pub(crate) mod test {
             .expect("In memory disk finish should never fail");
         let reader = DiskBtreeReader::<_, DELTA_KEY_SIZE>::new(0, root_offset, disk);
         let planner = VectoredReadPlanner::new(100);
-        let mut reconstruct_state = ValuesReconstructState::new();
         let ctx = RequestContext::new(TaskKind::UnitTest, DownloadBehavior::Error);
 
         let keyspace = KeySpace {
@@ -1691,7 +1677,6 @@ pub(crate) mod test {
             disk_offset,
             reader,
             planner,
-            &mut reconstruct_state,
             &ctx,
         )
         .await
@@ -1935,7 +1920,6 @@ pub(crate) mod test {
             );
 
             let planner = VectoredReadPlanner::new(constants::MAX_VECTORED_READ_BYTES);
-            let mut reconstruct_state = ValuesReconstructState::new();
             let keyspace = pick_random_keyspace(rng, &entries_meta.key_range);
             let data_end_offset = inner.index_start_blk as u64 * PAGE_SZ as u64;
 
@@ -1945,7 +1929,6 @@ pub(crate) mod test {
                 data_end_offset,
                 index_reader,
                 planner,
-                &mut reconstruct_state,
                 &ctx,
             )
             .await?;

pageserver/src/tenant/storage_layer/image_layer.rs

@@ -38,12 +38,11 @@ use crate::tenant::vectored_blob_io::{
     BlobFlag, BufView, StreamingVectoredReadPlanner, VectoredBlobReader, VectoredRead,
     VectoredReadPlanner,
 };
-use crate::tenant::PageReconstructError;
 use crate::virtual_file::owned_buffers_io::io_buf_ext::IoBufExt;
 use crate::virtual_file::IoBufferMut;
 use crate::virtual_file::{self, MaybeFatalIo, VirtualFile};
 use crate::{IMAGE_FILE_MAGIC, STORAGE_FORMAT_VERSION, TEMP_FILE_SUFFIX};
-use anyhow::{anyhow, bail, ensure, Context, Result};
+use anyhow::{bail, ensure, Context, Result};
 use bytes::Bytes;
 use camino::{Utf8Path, Utf8PathBuf};
 use hex;
@@ -56,12 +55,13 @@ use pageserver_api::shard::{ShardIdentity, TenantShardId};
 use pageserver_api::value::Value;
 use rand::{distributions::Alphanumeric, Rng};
 use serde::{Deserialize, Serialize};
-use std::collections::VecDeque;
+use std::collections::{HashMap, VecDeque};
 use std::fs::File;
 use std::io::SeekFrom;
 use std::ops::Range;
 use std::os::unix::prelude::FileExt;
 use std::str::FromStr;
+use std::sync::Arc;
 use tokio::sync::OnceCell;
 use tokio_stream::StreamExt;
 use tracing::*;
@@ -73,7 +73,10 @@ use utils::{
 };
 
 use super::layer_name::ImageLayerName;
-use super::{AsLayerDesc, LayerName, PersistentLayerDesc, ValuesReconstructState};
+use super::{
+    AsLayerDesc, LayerName, OnDiskValue, OnDiskValueIo, PersistentLayerDesc, ResidentLayer,
+    ValuesReconstructState,
+};
 
 ///
 /// Header stored in the beginning of the file
@@ -164,7 +167,7 @@ pub struct ImageLayerInner {
     key_range: Range<Key>,
     lsn: Lsn,
 
-    file: VirtualFile,
+    file: Arc<VirtualFile>,
     file_id: FileId,
 
     max_vectored_read_bytes: Option<MaxVectoredReadBytes>,
@@ -391,9 +394,11 @@ impl ImageLayerInner {
         max_vectored_read_bytes: Option<MaxVectoredReadBytes>,
         ctx: &RequestContext,
     ) -> anyhow::Result<Self> {
-        let file = VirtualFile::open_v2(path, ctx)
-            .await
-            .context("open layer file")?;
+        let file = Arc::new(
+            VirtualFile::open_v2(path, ctx)
+                .await
+                .context("open layer file")?,
+        );
         let file_id = page_cache::next_file_id();
         let block_reader = FileBlockReader::new(&file, file_id);
         let summary_blk = block_reader
@@ -439,6 +444,7 @@ impl ImageLayerInner {
     // the reconstruct state with whatever is found.
     pub(super) async fn get_values_reconstruct_data(
         &self,
+        this: ResidentLayer,
         keyspace: KeySpace,
         reconstruct_state: &mut ValuesReconstructState,
         ctx: &RequestContext,
@@ -448,7 +454,7 @@ impl ImageLayerInner {
             .await
             .map_err(GetVectoredError::Other)?;
 
-        self.do_reads_and_update_state(reads, reconstruct_state, ctx)
+        self.do_reads_and_update_state(this, reads, reconstruct_state, ctx)
             .await;
 
         reconstruct_state.on_image_layer_visited(&self.key_range);
@@ -570,6 +576,7 @@ impl ImageLayerInner {
 
     async fn do_reads_and_update_state(
         &self,
+        this: ResidentLayer,
         reads: Vec<VectoredRead>,
         reconstruct_state: &mut ValuesReconstructState,
         ctx: &RequestContext,
@@ -580,8 +587,13 @@ impl ImageLayerInner {
             .0
             .into();
 
-        let vectored_blob_reader = VectoredBlobReader::new(&self.file);
         for read in reads.into_iter() {
+            let mut ios: HashMap<(Key, Lsn), OnDiskValueIo> = Default::default();
+            for (_, blob_meta) in read.blobs_at.as_slice() {
+                let io = reconstruct_state.update_key(&blob_meta.key, blob_meta.lsn, true);
+                ios.insert((blob_meta.key, blob_meta.lsn), io);
+            }
+
             let buf_size = read.size();
 
             if buf_size > max_vectored_read_bytes {
@@ -611,50 +623,51 @@ impl ImageLayerInner {
                 }
             }
 
-            let buf = IoBufferMut::with_capacity(buf_size);
-            let res = vectored_blob_reader.read_blobs(&read, buf, ctx).await;
+            let read_extend_residency = this.clone();
+            let read_from = self.file.clone();
+            let read_ctx = ctx.attached_child();
+            reconstruct_state
+                .spawn_io(async move {
+                    let buf = IoBufferMut::with_capacity(buf_size);
+                    let vectored_blob_reader = VectoredBlobReader::new(&read_from);
+                    let res = vectored_blob_reader.read_blobs(&read, buf, &read_ctx).await;
 
-            match res {
-                Ok(blobs_buf) => {
-                    let view = BufView::new_slice(&blobs_buf.buf);
-                    for meta in blobs_buf.blobs.iter() {
-                        let img_buf = meta.read(&view).await;
+                    match res {
+                        Ok(blobs_buf) => {
+                            let view = BufView::new_slice(&blobs_buf.buf);
+                            for meta in blobs_buf.blobs.iter() {
+                                let io: OnDiskValueIo =
+                                    ios.remove(&(meta.meta.key, meta.meta.lsn)).unwrap();
+                                let img_buf = meta.read(&view).await;
 
-                        let img_buf = match img_buf {
-                            Ok(img_buf) => img_buf,
-                            Err(e) => {
-                                reconstruct_state.on_key_error(
-                                    meta.meta.key,
-                                    PageReconstructError::Other(anyhow!(e).context(format!(
-                                        "Failed to decompress blob from virtual file {}",
-                                        self.file.path(),
-                                    ))),
-                                );
+                                let img_buf = match img_buf {
+                                    Ok(img_buf) => img_buf,
+                                    Err(e) => {
+                                        io.complete(Err(e));
+                                        continue;
+                                    }
+                                };
 
-                                continue;
+                                io.complete(Ok(OnDiskValue::RawImage(img_buf.into_bytes())));
                             }
-                        };
-                        reconstruct_state.update_key(
-                            &meta.meta.key,
-                            self.lsn,
-                            Value::Image(img_buf.into_bytes()),
-                        );
+
+                            assert!(ios.is_empty());
+                        }
+                        Err(err) => {
+                            for (_, io) in ios {
+                                io.complete(Err(std::io::Error::new(
+                                    err.kind(),
+                                    "vec read failed",
+                                )));
+                            }
+                        }
                     }
-                }
-                Err(err) => {
-                    let kind = err.kind();
-                    for (_, blob_meta) in read.blobs_at.as_slice() {
-                        reconstruct_state.on_key_error(
-                            blob_meta.key,
-                            PageReconstructError::from(anyhow!(
-                                "Failed to read blobs from virtual file {}: {}",
-                                self.file.path(),
-                                kind
-                            )),
-                        );
-                    }
-                }
-            };
+
+                    // keep layer resident until this IO is done; this spawned IO future generally outlives the
+                    // call to `self` / the `Arc<DownloadedLayer>` / the `ResidentLayer` that guarantees residency
+                    drop(read_extend_residency);
+                })
+                .await;
         }
     }
 
@@ -1069,6 +1082,7 @@ impl ImageLayerIterator<'_> {
                     Key::from_slice(&raw_key[..KEY_SIZE]),
                     self.image_layer.lsn,
                     offset,
+                    true,
                 ) {
                     break batch_plan;
                 }

pageserver/src/tenant/storage_layer/inmemory_layer.rs

@@ -8,23 +8,22 @@ use crate::assert_u64_eq_usize::{u64_to_usize, U64IsUsize, UsizeIsU64};
 use crate::config::PageServerConf;
 use crate::context::{PageContentKind, RequestContext, RequestContextBuilder};
 use crate::tenant::ephemeral_file::EphemeralFile;
+use crate::tenant::storage_layer::{OnDiskValue, OnDiskValueIo};
 use crate::tenant::timeline::GetVectoredError;
-use crate::tenant::PageReconstructError;
 use crate::virtual_file::owned_buffers_io::io_buf_ext::IoBufExt;
 use crate::{l0_flush, page_cache};
-use anyhow::{anyhow, Result};
+use anyhow::Result;
 use camino::Utf8PathBuf;
 use pageserver_api::key::CompactKey;
 use pageserver_api::key::Key;
 use pageserver_api::keyspace::KeySpace;
 use pageserver_api::models::InMemoryLayerInfo;
 use pageserver_api::shard::TenantShardId;
-use pageserver_api::value::Value;
 use std::collections::{BTreeMap, HashMap};
 use std::sync::{Arc, OnceLock};
 use std::time::Instant;
 use tracing::*;
-use utils::{bin_ser::BeSer, id::TimelineId, lsn::Lsn, vec_map::VecMap};
+use utils::{id::TimelineId, lsn::Lsn, vec_map::VecMap};
 use wal_decoder::serialized_batch::{SerializedValueBatch, SerializedValueMeta, ValueMeta};
 // avoid binding to Write (conflicts with std::io::Write)
 // while being able to use std::fmt::Write's methods
@@ -36,9 +35,7 @@ use std::sync::atomic::Ordering as AtomicOrdering;
 use std::sync::atomic::{AtomicU64, AtomicUsize};
 use tokio::sync::RwLock;
 
-use super::{
-    DeltaLayerWriter, PersistentLayerDesc, ValueReconstructSituation, ValuesReconstructState,
-};
+use super::{DeltaLayerWriter, PersistentLayerDesc, ValuesReconstructState};
 
 pub(crate) mod vectored_dio_read;
 
@@ -112,8 +109,8 @@ const MAX_SUPPORTED_BLOB_LEN_BITS: usize = {
 ///
 /// Layout:
 /// - 1 bit: `will_init`
-/// - [`MAX_SUPPORTED_BLOB_LEN_BITS`]: `len`
-/// - [`MAX_SUPPORTED_POS_BITS`]: `pos`
+/// - [`MAX_SUPPORTED_BLOB_LEN_BITS`][]: `len`
+/// - [`MAX_SUPPORTED_POS_BITS`](IndexEntry::MAX_SUPPORTED_POS_BITS): `pos`
 #[derive(Debug, Clone, Copy, PartialEq, Eq)]
 pub struct IndexEntry(u64);
 
@@ -415,10 +412,8 @@ impl InMemoryLayer {
 
     // Look up the keys in the provided keyspace and update
     // the reconstruct state with whatever is found.
-    //
-    // If the key is cached, go no further than the cached Lsn.
     pub(crate) async fn get_values_reconstruct_data(
-        &self,
+        self: &Arc<InMemoryLayer>,
         keyspace: KeySpace,
         end_lsn: Lsn,
         reconstruct_state: &mut ValuesReconstructState,
@@ -435,6 +430,9 @@ impl InMemoryLayer {
             read: vectored_dio_read::LogicalRead<Vec<u8>>,
         }
         let mut reads: HashMap<Key, Vec<ValueRead>> = HashMap::new();
+        let mut ios: HashMap<(Key, Lsn), OnDiskValueIo> = Default::default();
+
+        let lsn_range = self.start_lsn..end_lsn;
 
         for range in keyspace.ranges.iter() {
             for (key, vec_map) in inner
@@ -442,12 +440,7 @@ impl InMemoryLayer {
                 .range(range.start.to_compact()..range.end.to_compact())
             {
                 let key = Key::from_compact(*key);
-                let lsn_range = match reconstruct_state.get_cached_lsn(&key) {
-                    Some(cached_lsn) => (cached_lsn + 1)..end_lsn,
-                    None => self.start_lsn..end_lsn,
-                };
-
-                let slice = vec_map.slice_range(lsn_range);
+                let slice = vec_map.slice_range(lsn_range.clone());
 
                 for (entry_lsn, index_entry) in slice.iter().rev() {
                     let IndexEntryUnpacked {
@@ -463,55 +456,59 @@ impl InMemoryLayer {
                             Vec::with_capacity(len as usize),
                         ),
                     });
+
+                    let io = reconstruct_state.update_key(&key, *entry_lsn, will_init);
+                    ios.insert((key, *entry_lsn), io);
+
                     if will_init {
                         break;
                     }
                 }
             }
         }
+        drop(inner); // release the lock before we spawn the IO; if it's serial-mode IO we will deadlock on the read().await below
+        let read_from = Arc::clone(self);
+        let read_ctx = ctx.attached_child();
+        reconstruct_state
+            .spawn_io(async move {
+                let inner = read_from.inner.read().await;
+                let f = vectored_dio_read::execute(
+                    &inner.file,
+                    reads
+                        .iter()
+                        .flat_map(|(_, value_reads)| value_reads.iter().map(|v| &v.read)),
+                    &read_ctx,
+                );
+                send_future::SendFuture::send(f) // https://github.com/rust-lang/rust/issues/96865
+                    .await;
 
-        // Execute the reads.
-
-        let f = vectored_dio_read::execute(
-            &inner.file,
-            reads
-                .iter()
-                .flat_map(|(_, value_reads)| value_reads.iter().map(|v| &v.read)),
-            &ctx,
-        );
-        send_future::SendFuture::send(f) // https://github.com/rust-lang/rust/issues/96865
-            .await;
-
-        // Process results into the reconstruct state
-        'next_key: for (key, value_reads) in reads {
-            for ValueRead { entry_lsn, read } in value_reads {
-                match read.into_result().expect("we run execute() above") {
-                    Err(e) => {
-                        reconstruct_state.on_key_error(key, PageReconstructError::from(anyhow!(e)));
-                        continue 'next_key;
-                    }
-                    Ok(value_buf) => {
-                        let value = Value::des(&value_buf);
-                        if let Err(e) = value {
-                            reconstruct_state
-                                .on_key_error(key, PageReconstructError::from(anyhow!(e)));
-                            continue 'next_key;
+                for (key, value_reads) in reads {
+                    for ValueRead { entry_lsn, read } in value_reads {
+                        let io = ios.remove(&(key, entry_lsn)).expect("sender must exist");
+                        match read.into_result().expect("we run execute() above") {
+                            Err(e) => {
+                                io.complete(Err(std::io::Error::new(
+                                    e.kind(),
+                                    "dio vec read failed",
+                                )));
+                            }
+                            Ok(value_buf) => {
+                                io.complete(Ok(OnDiskValue::WalRecordOrImage(value_buf.into())));
+                            }
                         }
-
-                        let key_situation =
-                            reconstruct_state.update_key(&key, entry_lsn, value.unwrap());
-                        if key_situation == ValueReconstructSituation::Complete {
-                            // TODO: metric to see if we fetched more values than necessary
-                            continue 'next_key;
-                        }
-
-                        // process the next value in the next iteration of the loop
                     }
                 }
-            }
-        }
 
-        reconstruct_state.on_lsn_advanced(&keyspace, self.start_lsn);
+                assert!(ios.is_empty());
+
+                // Keep layer existent until this IO is done;
+                // This is kinda forced for InMemoryLayer because we need to inner.read() anyway,
+                // but it's less obvious for DeltaLayer and ImageLayer. So, keep this explicit
+                // drop for consistency among all three layer types.
+                drop(inner);
+                drop(read_from);
+            })
+            .await;
 
         Ok(())
     }
@@ -606,6 +603,7 @@ impl InMemoryLayer {
         // Write the batch to the file
         inner.file.write_raw(&raw, ctx).await?;
         let new_size = inner.file.len();
+
         let expected_new_len = base_offset
             .checked_add(raw.len().into_u64())
             // write_raw would error if we were to overflow u64.

pageserver/src/tenant/storage_layer/layer.rs

@@ -308,7 +308,7 @@ impl Layer {
         reconstruct_data: &mut ValuesReconstructState,
         ctx: &RequestContext,
     ) -> Result<(), GetVectoredError> {
-        let layer = self
+        let downloaded = self
             .0
             .get_or_maybe_download(true, Some(ctx))
             .await
@@ -318,11 +318,15 @@ impl Layer {
                 }
                 other => GetVectoredError::Other(anyhow::anyhow!(other)),
             })?;
+        let this = ResidentLayer {
+            downloaded: downloaded.clone(),
+            owner: self.clone(),
+        };
 
         self.record_access(ctx);
 
-        layer
-            .get_values_reconstruct_data(keyspace, lsn_range, reconstruct_data, &self.0, ctx)
+        downloaded
+            .get_values_reconstruct_data(this, keyspace, lsn_range, reconstruct_data, ctx)
             .instrument(tracing::debug_span!("get_values_reconstruct_data", layer=%self))
             .await
             .map_err(|err| match err {
@@ -1768,25 +1772,25 @@ impl DownloadedLayer {
 
     async fn get_values_reconstruct_data(
         &self,
+        this: ResidentLayer,
         keyspace: KeySpace,
         lsn_range: Range<Lsn>,
         reconstruct_data: &mut ValuesReconstructState,
-        owner: &Arc<LayerInner>,
         ctx: &RequestContext,
     ) -> Result<(), GetVectoredError> {
         use LayerKind::*;
 
         match self
-            .get(owner, ctx)
+            .get(&this.owner.0, ctx)
             .await
             .map_err(GetVectoredError::Other)?
         {
             Delta(d) => {
-                d.get_values_reconstruct_data(keyspace, lsn_range, reconstruct_data, ctx)
+                d.get_values_reconstruct_data(this, keyspace, lsn_range, reconstruct_data, ctx)
                     .await
             }
             Image(i) => {
-                i.get_values_reconstruct_data(keyspace, reconstruct_data, ctx)
+                i.get_values_reconstruct_data(this, keyspace, reconstruct_data, ctx)
                     .await
             }
         }
@@ -1812,7 +1816,7 @@ enum LayerKind {
 
 /// Guard for forcing a layer be resident while it exists.
 #[derive(Clone)]
-pub(crate) struct ResidentLayer {
+pub struct ResidentLayer {
     owner: Layer,
     downloaded: Arc<DownloadedLayer>,
 }

pageserver/src/tenant/storage_layer/layer/tests.rs

@@ -1,6 +1,6 @@
 use std::time::UNIX_EPOCH;
 
-use pageserver_api::key::CONTROLFILE_KEY;
+use pageserver_api::key::{Key, CONTROLFILE_KEY};
 use tokio::task::JoinSet;
 use utils::{
     completion::{self, Completion},
@@ -9,7 +9,13 @@ use utils::{
 
 use super::failpoints::{Failpoint, FailpointKind};
 use super::*;
-use crate::{context::DownloadBehavior, tenant::storage_layer::LayerVisibilityHint};
+use crate::{
+    context::DownloadBehavior,
+    tenant::{
+        harness::test_img,
+        storage_layer::{IoConcurrency, LayerVisibilityHint},
+    },
+};
 use crate::{task_mgr::TaskKind, tenant::harness::TenantHarness};
 
 /// Used in tests to advance a future to wanted await point, and not futher.
@@ -28,23 +34,55 @@ async fn smoke_test() {
     let span = h.span();
     let download_span = span.in_scope(|| tracing::info_span!("downloading", timeline_id = 1));
     let (tenant, _) = h.load().await;
+    let io_concurrency = IoConcurrency::spawn_for_test();
 
     let ctx = RequestContext::new(TaskKind::UnitTest, DownloadBehavior::Download);
 
+    let image_layers = vec![(
+        Lsn(0x40),
+        vec![(
+            Key::from_hex("620000000033333333444444445500000000").unwrap(),
+            test_img("foo"),
+        )],
+    )];
+
+    // Create a test timeline with one real layer, and one synthetic test layer.  The synthetic
+    // one is only there so that we can GC the real one without leaving the timeline's metadata
+    // empty, which is an illegal state (see [`IndexPart::validate`]).
     let timeline = tenant
-        .create_test_timeline(TimelineId::generate(), Lsn(0x10), 14, &ctx)
+        .create_test_timeline_with_layers(
+            TimelineId::generate(),
+            Lsn(0x10),
+            14,
+            &ctx,
+            Default::default(),
+            image_layers,
+            Lsn(0x100),
+        )
         .await
         .unwrap();
 
-    let layer = {
+    // Grab one of the timeline's layers to exercise in the test, and the other layer that is just
+    // there to avoid the timeline being illegally empty
+    let (layer, dummy_layer) = {
         let mut layers = {
             let layers = timeline.layers.read().await;
             layers.likely_resident_layers().cloned().collect::<Vec<_>>()
         };
 
-        assert_eq!(layers.len(), 1);
+        assert_eq!(layers.len(), 2);
 
-        layers.swap_remove(0)
+        layers.sort_by_key(|l| l.layer_desc().get_key_range().start);
+        let synthetic_layer = layers.pop().unwrap();
+        let real_layer = layers.pop().unwrap();
+        tracing::info!(
+            "real_layer={:?} ({}), synthetic_layer={:?} ({})",
+            real_layer,
+            real_layer.layer_desc().file_size,
+            synthetic_layer,
+            synthetic_layer.layer_desc().file_size
+        );
+        (real_layer, synthetic_layer)
     };
 
     // all layers created at pageserver are like `layer`, initialized with strong
@@ -55,7 +93,7 @@ async fn smoke_test() {
     };
 
     let img_before = {
-        let mut data = ValuesReconstructState::default();
+        let mut data = ValuesReconstructState::new(io_concurrency.clone());
         layer
             .get_values_reconstruct_data(
                 controlfile_keyspace.clone(),
@@ -65,10 +103,13 @@ async fn smoke_test() {
             )
             .await
             .unwrap();
+
         data.keys
             .remove(&CONTROLFILE_KEY)
             .expect("must be present")
-            .expect("should not error")
+            .collect_pending_ios()
+            .await
+            .expect("must not error")
             .img
             .take()
             .expect("tenant harness writes the control file")
@@ -87,7 +128,7 @@ async fn smoke_test() {
 
     // on accesses when the layer is evicted, it will automatically be downloaded.
     let img_after = {
-        let mut data = ValuesReconstructState::default();
+        let mut data = ValuesReconstructState::new(io_concurrency.clone());
         layer
             .get_values_reconstruct_data(
                 controlfile_keyspace.clone(),
@@ -101,7 +142,9 @@ async fn smoke_test() {
         data.keys
             .remove(&CONTROLFILE_KEY)
             .expect("must be present")
-            .expect("should not error")
+            .collect_pending_ios()
+            .await
+            .expect("must not error")
             .img
             .take()
             .expect("tenant harness writes the control file")
@@ -173,10 +216,13 @@ async fn smoke_test() {
 
     let rtc = &timeline.remote_client;
 
+    // Simulate GC removing our test layer.
     {
-        let layers = &[layer];
         let mut g = timeline.layers.write().await;
+
+        let layers = &[layer];
         g.open_mut().unwrap().finish_gc_timeline(layers);
+
         // this just updates the remote_physical_size for demonstration purposes
         rtc.schedule_gc_update(layers).unwrap();
     }
@@ -191,7 +237,10 @@ async fn smoke_test() {
 
     rtc.wait_completion().await.unwrap();
 
-    assert_eq!(rtc.get_remote_physical_size(), 0);
+    assert_eq!(
+        rtc.get_remote_physical_size(),
+        dummy_layer.metadata().file_size
+    );
     assert_eq!(0, LAYER_IMPL_METRICS.inits_cancelled.get())
 }
 

pageserver/src/tenant/tasks.rs

@@ -67,10 +67,9 @@ pub(crate) async fn concurrent_background_tasks_rate_limit_permit(
 ) -> tokio::sync::SemaphorePermit<'static> {
     let _guard = crate::metrics::BACKGROUND_LOOP_SEMAPHORE.measure_acquisition(loop_kind);
 
-    pausable_failpoint!(
-        "initial-size-calculation-permit-pause",
-        loop_kind == BackgroundLoopKind::InitialLogicalSizeCalculation
-    );
+    if loop_kind == BackgroundLoopKind::InitialLogicalSizeCalculation {
+        pausable_failpoint!("initial-size-calculation-permit-pause");
+    }
 
     // TODO: assert that we run on BACKGROUND_RUNTIME; requires tokio_unstable Handle::id();
     match CONCURRENT_BACKGROUND_TASKS.acquire().await {

pageserver/src/tenant/throttle.rs

@@ -3,7 +3,7 @@ use std::{
         atomic::{AtomicU64, Ordering},
         Arc,
     },
-    time::{Duration, Instant},
+    time::Instant,
 };
 
 use arc_swap::ArcSwap;
@@ -16,9 +16,8 @@ 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<M: Metric> {
+pub struct Throttle {
     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`]
@@ -36,15 +35,6 @@ 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.
@@ -59,18 +49,14 @@ pub struct Stats {
 }
 
 pub enum ThrottleResult {
-    NotThrottled { start: Instant },
-    Throttled { start: Instant, end: Instant },
+    NotThrottled { end: Instant },
+    Throttled { end: Instant },
 }
 
-impl<M> Throttle<M>
-where
-    M: Metric,
-{
-    pub fn new(config: Config, metric: M) -> Self {
+impl Throttle {
+    pub fn new(config: Config) -> 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),
@@ -127,32 +113,27 @@ where
         self.inner.load().rate_limiter.steady_rps()
     }
 
-    pub async fn throttle(&self, key_count: usize) -> ThrottleResult {
+    /// `start` must be [`Instant::now`] or earlier.
+    pub async fn throttle(&self, key_count: usize, start: Instant) -> ThrottleResult {
         let inner = self.inner.load_full(); // clones the `Inner` Arc
 
-        let start = std::time::Instant::now();
-
         if !inner.enabled {
-            return ThrottleResult::NotThrottled { start };
+            return ThrottleResult::NotThrottled { end: 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 now = Instant::now();
-            let wait_time = now - start;
+            let end = Instant::now();
+            let wait_time = end - start;
             self.sum_throttled_usecs
                 .fetch_add(wait_time.as_micros() as u64, Ordering::Relaxed);
-            let observation = Observation { wait_time };
-            self.metric.observe_throttling(&observation);
-            ThrottleResult::Throttled { start, end: now }
+            ThrottleResult::Throttled { end }
         } else {
-            ThrottleResult::NotThrottled { start }
+            ThrottleResult::NotThrottled { end: start }
         }
     }
 }