DNM: rip out throttling code (the recorder's borrowing of ctx obviously breaks &mut RequestContext passing)

## Problem
When creating a new shard the storage controller schedules via
Scheduler::schedule_shard. This does not take into account the number of
attached shards. What it does take into account is the node affinity:
when a shard is scheduled, all its nodes (primaries and secondaries) get
their affinity incremented.

For two node clusters and shards with one secondary we have a
pathological case where all primaries are scheduled on the same node.
Now that we track the count of attached shards per node, this is trivial
to fix. Still, the "proper" fix is to use the pageserver's utilization
score.

Closes https://github.com/neondatabase/neon/issues/8041

## Summary of changes
Use attached shard count when deciding which node to schedule a fresh
shard on.

.dockerignore

@@ -8,6 +8,7 @@
 !scripts/combine_control_files.py
 !scripts/ninstall.sh
 !vm-cgconfig.conf
+!docker-compose/run-tests.sh
 
 # Directories
 !.cargo/
@@ -20,7 +21,7 @@
 !patches/
 !pgxn/
 !proxy/
-!s3_scrubber/
+!storage_scrubber/
 !safekeeper/
 !storage_broker/
 !storage_controller/

.github/actions/run-python-test-set/action.yml

@@ -183,8 +183,7 @@ runs:
 
         # Run the tests.
         #
-        # The junit.xml file allows CI tools to display more fine-grained test information
-        # in its "Tests" tab in the results page.
+        # --alluredir saves test results in Allure format (in a specified directory)
         # --verbose prints name of each test (helpful when there are
         # multiple tests in one file)
         # -rA prints summary in the end
@@ -193,7 +192,6 @@ runs:
         #
         mkdir -p $TEST_OUTPUT/allure/results
         "${cov_prefix[@]}" ./scripts/pytest \
-          --junitxml=$TEST_OUTPUT/junit.xml \
           --alluredir=$TEST_OUTPUT/allure/results \
           --tb=short \
           --verbose \

.github/workflows/actionlint.yml

@@ -36,15 +36,16 @@ jobs:
           fail_on_error: true
           filter_mode: nofilter
           level: error
-      - run: |
+
+      - name: Disallow 'ubuntu-latest' runners
+        run: |
           PAT='^\s*runs-on:.*-latest'
-          if grep -ERq $PAT .github/workflows
-          then
+          if grep -ERq $PAT .github/workflows; then
             grep -ERl $PAT .github/workflows |\
             while read -r f
             do
               l=$(grep -nE $PAT .github/workflows/release.yml | awk -F: '{print $1}' | head -1)
-              echo "::error file=$f,line=$l::Please, do not use ubuntu-latest images to run on, use LTS instead."
+              echo "::error file=$f,line=$l::Please use 'ubuntu-22.04' instead of 'ubuntu-latest'"
             done
             exit 1
           fi

.github/workflows/benchmarking.yml

@@ -99,7 +99,7 @@ jobs:
         # Set --sparse-ordering option of pytest-order plugin
         # to ensure tests are running in order of appears in the file.
         # It's important for test_perf_pgbench.py::test_pgbench_remote_* tests
-        extra_params: -m remote_cluster --sparse-ordering --timeout 5400 --ignore test_runner/performance/test_perf_olap.py
+        extra_params: -m remote_cluster --sparse-ordering --timeout 5400 --ignore test_runner/performance/test_perf_olap.py --ignore test_runner/performance/test_perf_pgvector_queries.py
       env:
         BENCHMARK_CONNSTR: ${{ steps.create-neon-project.outputs.dsn }}
         VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
@@ -410,14 +410,14 @@ jobs:
         PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
         BENCHMARK_CONNSTR: ${{ steps.set-up-connstr.outputs.connstr }}
 
-    - name: Benchmark pgvector hnsw queries
+    - name: Benchmark pgvector queries
       uses: ./.github/actions/run-python-test-set
       with:
         build_type: ${{ env.BUILD_TYPE }}
-        test_selection: performance
+        test_selection: performance/test_perf_pgvector_queries.py
         run_in_parallel: false
         save_perf_report: ${{ env.SAVE_PERF_REPORT }}
-        extra_params: -m remote_cluster --timeout 21600 -k test_pgbench_remote_pgvector
+        extra_params: -m remote_cluster --timeout 21600 
       env:
         BENCHMARK_CONNSTR: ${{ steps.set-up-connstr.outputs.connstr }}
         VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"

.github/workflows/build-build-tools-image.yml

@@ -30,7 +30,6 @@ jobs:
   check-image:
     uses: ./.github/workflows/check-build-tools-image.yml
 
-  # This job uses older version of GitHub Actions because it's run on gen2 runners, which don't support node 20 (for newer versions)
   build-image:
     needs: [ check-image ]
     if: needs.check-image.outputs.found == 'false'
@@ -55,7 +54,7 @@ jobs:
             exit 1
           fi
 
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@v4
 
       # Use custom DOCKER_CONFIG directory to avoid conflicts with default settings
       # The default value is ~/.docker

.github/workflows/build_and_test.yml

@@ -299,21 +299,21 @@ jobs:
         uses: actions/cache@v4
         with:
           path: pg_install/v14
-          key: v1-${{ runner.os }}-${{ matrix.build_type }}-pg-${{ steps.pg_v14_rev.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
+          key: v1-${{ runner.os }}-${{ matrix.build_type }}-pg-${{ steps.pg_v14_rev.outputs.pg_rev }}-${{ hashFiles('Makefile', 'Dockerfile.build-tools') }}
 
       - name: Cache postgres v15 build
         id: cache_pg_15
         uses: actions/cache@v4
         with:
           path: pg_install/v15
-          key: v1-${{ runner.os }}-${{ matrix.build_type }}-pg-${{ steps.pg_v15_rev.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
+          key: v1-${{ runner.os }}-${{ matrix.build_type }}-pg-${{ steps.pg_v15_rev.outputs.pg_rev }}-${{ hashFiles('Makefile', 'Dockerfile.build-tools') }}
 
       - name: Cache postgres v16 build
         id: cache_pg_16
         uses: actions/cache@v4
         with:
           path: pg_install/v16
-          key: v1-${{ runner.os }}-${{ matrix.build_type }}-pg-${{ steps.pg_v16_rev.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
+          key: v1-${{ runner.os }}-${{ matrix.build_type }}-pg-${{ steps.pg_v16_rev.outputs.pg_rev }}-${{ hashFiles('Makefile', 'Dockerfile.build-tools') }}
 
       - name: Build postgres v14
         if: steps.cache_pg_14.outputs.cache-hit != 'true'
@@ -337,34 +337,8 @@ jobs:
         run: |
           ${cov_prefix} mold -run cargo build $CARGO_FLAGS $CARGO_FEATURES --bins --tests
 
-      - name: Run rust tests
-        env:
-          NEXTEST_RETRIES: 3
-        run: |
-          #nextest does not yet support running doctests
-          cargo test --doc $CARGO_FLAGS $CARGO_FEATURES
-
-          for io_engine in std-fs tokio-epoll-uring ; do
-            NEON_PAGESERVER_UNIT_TEST_VIRTUAL_FILE_IOENGINE=$io_engine ${cov_prefix} cargo nextest run $CARGO_FLAGS $CARGO_FEATURES
-          done
-
-          # Run separate tests for real S3
-          export ENABLE_REAL_S3_REMOTE_STORAGE=nonempty
-          export REMOTE_STORAGE_S3_BUCKET=neon-github-ci-tests
-          export REMOTE_STORAGE_S3_REGION=eu-central-1
-          # Avoid `$CARGO_FEATURES` since there's no `testing` feature in the e2e tests now
-          ${cov_prefix} cargo nextest run $CARGO_FLAGS -E 'package(remote_storage)' -E 'test(test_real_s3)'
-
-          # Run separate tests for real Azure Blob Storage
-          # XXX: replace region with `eu-central-1`-like region
-          export ENABLE_REAL_AZURE_REMOTE_STORAGE=y
-          export AZURE_STORAGE_ACCOUNT="${{ secrets.AZURE_STORAGE_ACCOUNT_DEV }}"
-          export AZURE_STORAGE_ACCESS_KEY="${{ secrets.AZURE_STORAGE_ACCESS_KEY_DEV }}"
-          export REMOTE_STORAGE_AZURE_CONTAINER="${{ vars.REMOTE_STORAGE_AZURE_CONTAINER }}"
-          export REMOTE_STORAGE_AZURE_REGION="${{ vars.REMOTE_STORAGE_AZURE_REGION }}"
-          # Avoid `$CARGO_FEATURES` since there's no `testing` feature in the e2e tests now
-          ${cov_prefix} cargo nextest run $CARGO_FLAGS -E 'package(remote_storage)' -E 'test(test_real_azure)'
-
+      # Do install *before* running rust tests because they might recompile the
+      # binaries with different features/flags.
       - name: Install rust binaries
         run: |
           # Install target binaries
@@ -405,6 +379,32 @@ jobs:
             done
           fi
 
+      - name: Run rust tests
+        env:
+          NEXTEST_RETRIES: 3
+        run: |
+          #nextest does not yet support running doctests
+          cargo test --doc $CARGO_FLAGS $CARGO_FEATURES
+
+          for io_engine in std-fs tokio-epoll-uring ; do
+            NEON_PAGESERVER_UNIT_TEST_VIRTUAL_FILE_IOENGINE=$io_engine ${cov_prefix} cargo nextest run $CARGO_FLAGS $CARGO_FEATURES
+          done
+
+          # Run separate tests for real S3
+          export ENABLE_REAL_S3_REMOTE_STORAGE=nonempty
+          export REMOTE_STORAGE_S3_BUCKET=neon-github-ci-tests
+          export REMOTE_STORAGE_S3_REGION=eu-central-1
+          ${cov_prefix} cargo nextest run $CARGO_FLAGS $CARGO_FEATURES -E 'package(remote_storage)' -E 'test(test_real_s3)'
+
+          # Run separate tests for real Azure Blob Storage
+          # XXX: replace region with `eu-central-1`-like region
+          export ENABLE_REAL_AZURE_REMOTE_STORAGE=y
+          export AZURE_STORAGE_ACCOUNT="${{ secrets.AZURE_STORAGE_ACCOUNT_DEV }}"
+          export AZURE_STORAGE_ACCESS_KEY="${{ secrets.AZURE_STORAGE_ACCESS_KEY_DEV }}"
+          export REMOTE_STORAGE_AZURE_CONTAINER="${{ vars.REMOTE_STORAGE_AZURE_CONTAINER }}"
+          export REMOTE_STORAGE_AZURE_REGION="${{ vars.REMOTE_STORAGE_AZURE_REGION }}"
+          ${cov_prefix} cargo nextest run $CARGO_FLAGS $CARGO_FEATURES -E 'package(remote_storage)' -E 'test(test_real_azure)'
+
       - name: Install postgres binaries
         run: cp -a pg_install /tmp/neon/pg_install
 
@@ -859,6 +859,26 @@ jobs:
           tags: |
             neondatabase/compute-node-${{ matrix.version }}:${{ needs.tag.outputs.build-tag }}-${{ matrix.arch }}
 
+      - name: Build neon extensions test image
+        if: matrix.version == 'v16'
+        uses: docker/build-push-action@v5
+        with:
+          context: .
+          build-args: |
+            GIT_VERSION=${{ github.event.pull_request.head.sha || github.sha }}
+            PG_VERSION=${{ matrix.version }}
+            BUILD_TAG=${{ needs.tag.outputs.build-tag }}
+            TAG=${{ needs.build-build-tools-image.outputs.image-tag }}
+          provenance: false
+          push: true
+          pull: true
+          file: Dockerfile.compute-node
+          target: neon-pg-ext-test
+          cache-from: type=registry,ref=neondatabase/neon-test-extensions-${{ matrix.version }}:cache-${{ matrix.arch }}
+          cache-to: type=registry,ref=neondatabase/neon-test-extensions-${{ matrix.version }}:cache-${{ matrix.arch }},mode=max
+          tags: |
+            neondatabase/neon-test-extensions-${{ matrix.version }}:${{needs.tag.outputs.build-tag}}-${{ matrix.arch }}
+
       - name: Build compute-tools image
         # compute-tools are Postgres independent, so build it only once
         if: matrix.version == 'v16'
@@ -902,6 +922,13 @@ jobs:
                                              neondatabase/compute-node-${{ matrix.version }}:${{ needs.tag.outputs.build-tag }}-x64 \
                                              neondatabase/compute-node-${{ matrix.version }}:${{ needs.tag.outputs.build-tag }}-arm64
 
+      - name: Create multi-arch neon-test-extensions image
+        if: matrix.version == 'v16'
+        run: |
+          docker buildx imagetools create -t neondatabase/neon-test-extensions-${{ matrix.version }}:${{ needs.tag.outputs.build-tag }} \
+                                             neondatabase/neon-test-extensions-${{ matrix.version }}:${{ needs.tag.outputs.build-tag }}-x64 \
+                                             neondatabase/neon-test-extensions-${{ matrix.version }}:${{ needs.tag.outputs.build-tag }}-arm64
+
       - name: Create multi-arch compute-tools image
         if: matrix.version == 'v16'
         run: |
@@ -938,7 +965,7 @@ jobs:
 
     steps:
       - name: Checkout
-        uses: actions/checkout@v1
+        uses: actions/checkout@v4
         with:
           fetch-depth: 0
 
@@ -996,6 +1023,18 @@ jobs:
         with:
           fetch-depth: 0
 
+      # Use custom DOCKER_CONFIG directory to avoid conflicts with default settings
+      # The default value is ~/.docker
+      - name: Set custom docker config directory
+        run: |
+          mkdir -p .docker-custom
+          echo DOCKER_CONFIG=$(pwd)/.docker-custom >> $GITHUB_ENV
+
+      - uses: docker/login-action@v3
+        with:
+          username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
+          password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
+
       # `neondatabase/neon` contains multiple binaries, all of them use the same input for the version into the same version formatting library.
       # Pick pageserver as currently the only binary with extra "version" features printed in the string to verify.
       # Regular pageserver version string looks like
@@ -1020,7 +1059,7 @@ jobs:
             exit 1
           fi
 
-      - name: Verify docker-compose example
+      - name: Verify docker-compose example and test extensions
         timeout-minutes: 20
         run: env TAG=${{needs.tag.outputs.build-tag}} ./docker-compose/docker_compose_test.sh
 
@@ -1030,6 +1069,11 @@ jobs:
           docker compose -f ./docker-compose/docker-compose.yml logs || 0
           docker compose -f ./docker-compose/docker-compose.yml down
 
+      - name: Remove custom docker config directory
+        if: always()
+        run: |
+          rm -rf .docker-custom
+
   promote-images:
     needs: [ check-permissions, tag, test-images, vm-compute-node-image ]
     runs-on: ubuntu-22.04
@@ -1043,7 +1087,8 @@ jobs:
           username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
           password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
 
-      - uses: docker/login-action@v3
+      - name: Login to dev ECR
+        uses: docker/login-action@v3
         with:
           registry: 369495373322.dkr.ecr.eu-central-1.amazonaws.com
           username: ${{ secrets.AWS_ACCESS_KEY_DEV }}
@@ -1074,6 +1119,24 @@ jobs:
                                                  $repo/vm-compute-node-${version}:${{ needs.tag.outputs.build-tag }}
             done
           done
+          docker buildx imagetools create -t neondatabase/neon-test-extensions-v16:latest \
+                                             neondatabase/neon-test-extensions-v16:${{ needs.tag.outputs.build-tag }}
+
+      - name: Login to prod ECR
+        uses: docker/login-action@v3
+        if: github.ref_name == 'release'|| github.ref_name == 'release-proxy'
+        with:
+          registry: 093970136003.dkr.ecr.eu-central-1.amazonaws.com
+          username: ${{ secrets.PROD_GHA_RUNNER_LIMITED_AWS_ACCESS_KEY_ID }}
+          password: ${{ secrets.PROD_GHA_RUNNER_LIMITED_AWS_SECRET_ACCESS_KEY }}
+
+      - name: Copy all images to prod ECR
+        if: github.ref_name == 'release'|| github.ref_name == 'release-proxy'
+        run: |
+          for image in neon compute-tools {vm-,}compute-node-{v14,v15,v16}; do
+            docker buildx imagetools create -t 093970136003.dkr.ecr.eu-central-1.amazonaws.com/${image}:${{ needs.tag.outputs.build-tag }} \
+                                               369495373322.dkr.ecr.eu-central-1.amazonaws.com/${image}:${{ needs.tag.outputs.build-tag }}
+          done
 
   trigger-custom-extensions-build-and-wait:
     needs: [ check-permissions, tag ]

.github/workflows/check-build-tools-image.yml

@@ -25,26 +25,17 @@ jobs:
       found: ${{ steps.check-image.outputs.found }}
 
     steps:
+      - uses: actions/checkout@v4
+
       - name: Get build-tools image tag for the current commit
         id: get-build-tools-tag
         env:
-          # Usually, for COMMIT_SHA, we use `github.event.pull_request.head.sha || github.sha`, but here, even for PRs,
-          # we want to use `github.sha` i.e. point to a phantom merge commit to determine the image tag correctly.
-          COMMIT_SHA: ${{ github.sha }}
-          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          IMAGE_TAG: |
+            ${{ hashFiles('Dockerfile.build-tools',
+                          '.github/workflows/check-build-tools-image.yml',
+                          '.github/workflows/build-build-tools-image.yml') }}
         run: |
-          LAST_BUILD_TOOLS_SHA=$(
-            gh api \
-              -H "Accept: application/vnd.github+json" \
-              -H "X-GitHub-Api-Version: 2022-11-28" \
-              --method GET \
-              --field path=Dockerfile.build-tools \
-              --field sha=${COMMIT_SHA} \
-              --field per_page=1 \
-              --jq ".[0].sha" \
-              "/repos/${GITHUB_REPOSITORY}/commits"
-          )
-          echo "image-tag=${LAST_BUILD_TOOLS_SHA}" | tee -a $GITHUB_OUTPUT
+          echo "image-tag=${IMAGE_TAG}" | tee -a $GITHUB_OUTPUT
 
       - name: Check if such tag found in the registry
         id: check-image

.github/workflows/release.yml

@@ -52,13 +52,15 @@ jobs:
       env:
         GH_TOKEN: ${{ secrets.CI_ACCESS_TOKEN }}
       run: |
+        TITLE="Storage & Compute release ${RELEASE_DATE}"
+
         cat << EOF > body.md
-          ## Storage & Compute release ${RELEASE_DATE}
+          ## ${TITLE}
 
           **Please merge this Pull Request using 'Create a merge commit' button**
         EOF
 
-        gh pr create --title "Release ${RELEASE_DATE}" \
+        gh pr create --title "${TITLE}" \
                      --body-file "body.md" \
                      --head "${RELEASE_BRANCH}" \
                      --base "release"
@@ -91,13 +93,15 @@ jobs:
       env:
         GH_TOKEN: ${{ secrets.CI_ACCESS_TOKEN }}
       run: |
+        TITLE="Proxy release ${RELEASE_DATE}"
+
         cat << EOF > body.md
-          ## Proxy release ${RELEASE_DATE}
+          ## ${TITLE}
 
           **Please merge this Pull Request using 'Create a merge commit' button**
         EOF
 
-        gh pr create --title "Proxy release ${RELEASE_DATE}" \
+        gh pr create --title "${TITLE}" \
                      --body-file "body.md" \
                      --head "${RELEASE_BRANCH}" \
                      --base "release-proxy"

Cargo.lock

@@ -5109,54 +5109,6 @@ version = "1.0.13"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "f91339c0467de62360649f8d3e185ca8de4224ff281f66000de5eb2a77a79041"
 
-[[package]]
-name = "s3_scrubber"
-version = "0.1.0"
-dependencies = [
- "anyhow",
- "async-stream",
- "aws-config",
- "aws-sdk-s3",
- "aws-smithy-async",
- "bincode",
- "bytes",
- "camino",
- "chrono",
- "clap",
- "crc32c",
- "either",
- "futures",
- "futures-util",
- "hex",
- "histogram",
- "humantime",
- "itertools",
- "once_cell",
- "pageserver",
- "pageserver_api",
- "postgres_ffi",
- "rand 0.8.5",
- "remote_storage",
- "reqwest 0.12.4",
- "rustls 0.22.4",
- "rustls-native-certs 0.7.0",
- "serde",
- "serde_json",
- "serde_with",
- "thiserror",
- "tokio",
- "tokio-postgres",
- "tokio-postgres-rustls",
- "tokio-rustls 0.25.0",
- "tokio-stream",
- "tokio-util",
- "tracing",
- "tracing-appender",
- "tracing-subscriber",
- "utils",
- "workspace_hack",
-]
-
 [[package]]
 name = "safekeeper"
 version = "0.1.0"
@@ -5206,6 +5158,7 @@ dependencies = [
  "tokio-io-timeout",
  "tokio-postgres",
  "tokio-stream",
+ "tokio-tar",
  "tokio-util",
  "toml_edit",
  "tracing",
@@ -5801,6 +5754,7 @@ dependencies = [
  "r2d2",
  "reqwest 0.12.4",
  "routerify",
+ "scopeguard",
  "serde",
  "serde_json",
  "strum",
@@ -5813,6 +5767,54 @@ dependencies = [
  "workspace_hack",
 ]
 
+[[package]]
+name = "storage_scrubber"
+version = "0.1.0"
+dependencies = [
+ "anyhow",
+ "async-stream",
+ "aws-config",
+ "aws-sdk-s3",
+ "aws-smithy-async",
+ "bincode",
+ "bytes",
+ "camino",
+ "chrono",
+ "clap",
+ "crc32c",
+ "either",
+ "futures",
+ "futures-util",
+ "hex",
+ "histogram",
+ "humantime",
+ "itertools",
+ "once_cell",
+ "pageserver",
+ "pageserver_api",
+ "postgres_ffi",
+ "rand 0.8.5",
+ "remote_storage",
+ "reqwest 0.12.4",
+ "rustls 0.22.4",
+ "rustls-native-certs 0.7.0",
+ "serde",
+ "serde_json",
+ "serde_with",
+ "thiserror",
+ "tokio",
+ "tokio-postgres",
+ "tokio-postgres-rustls",
+ "tokio-rustls 0.25.0",
+ "tokio-stream",
+ "tokio-util",
+ "tracing",
+ "tracing-appender",
+ "tracing-subscriber",
+ "utils",
+ "workspace_hack",
+]
+
 [[package]]
 name = "storcon_cli"
 version = "0.1.0"
@@ -5820,6 +5822,7 @@ dependencies = [
  "anyhow",
  "clap",
  "comfy-table",
+ "futures",
  "humantime",
  "hyper 0.14.26",
  "pageserver_api",

Cargo.toml

@@ -13,7 +13,7 @@ members = [
     "safekeeper",
     "storage_broker",
     "storage_controller",
-    "s3_scrubber",
+    "storage_scrubber",
     "workspace_hack",
     "trace",
     "libs/compute_api",
@@ -120,7 +120,7 @@ num_cpus = "1.15"
 num-traits = "0.2.15"
 once_cell = "1.13"
 opentelemetry = "0.20.0"
-opentelemetry-otlp = { version = "0.13.0", default_features=false, features = ["http-proto", "trace", "http", "reqwest-client"] }
+opentelemetry-otlp = { version = "0.13.0", default-features=false, features = ["http-proto", "trace", "http", "reqwest-client"] }
 opentelemetry-semantic-conventions = "0.12.0"
 parking_lot = "0.12"
 parquet = { version = "51.0.0", default-features = false, features = ["zstd"] }
@@ -128,7 +128,7 @@ parquet_derive = "51.0.0"
 pbkdf2 = { version = "0.12.1", features = ["simple", "std"] }
 pin-project-lite = "0.2"
 procfs = "0.14"
-prometheus = {version = "0.13", default_features=false, features = ["process"]} # removes protobuf dependency
+prometheus = {version = "0.13", default-features=false, features = ["process"]} # removes protobuf dependency
 prost = "0.11"
 rand = "0.8"
 redis = { version = "0.25.2", features = ["tokio-rustls-comp", "keep-alive"] }
@@ -184,7 +184,7 @@ tower-service = "0.3.2"
 tracing = "0.1"
 tracing-error = "0.2.0"
 tracing-opentelemetry = "0.21.0"
-tracing-subscriber = { version = "0.3", default_features = false, features = ["smallvec", "fmt", "tracing-log", "std", "env-filter", "json", "ansi"] }
+tracing-subscriber = { version = "0.3", default-features = false, features = ["smallvec", "fmt", "tracing-log", "std", "env-filter", "json", "ansi"] }
 twox-hash = { version = "1.6.3", default-features = false }
 url = "2.2"
 urlencoding = "2.1"

Dockerfile

@@ -69,8 +69,6 @@ RUN set -e \
     && apt install -y \
         libreadline-dev \
         libseccomp-dev \
-        libicu67 \
-        openssl \
         ca-certificates \
     && rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/* \
     && useradd -d /data neon \

Dockerfile.build-tools

@@ -112,6 +112,45 @@ 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=3.2.2
+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 "197149c18d9e9f292c43f0400acaba12e5f52cacfe050f3d199277ea738ec2e7 /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-${OPENSSL_VERSION}
+
+# Use the same version of libicu as the compute nodes so that
+# clusters created using inidb on pageserver can be used by computes.
+#
+# TODO: at this time, Dockerfile.compute-node uses the debian bullseye libicu
+# package, which is 67.1. We're duplicating that knowledge here, and also, technically,
+# Debian has a few patches on top of 67.1 that we're not adding here.
+ENV ICU_VERSION=67.1
+ENV ICU_PREFIX=/usr/local/icu
+
+# Download and build static ICU
+RUN wget -O /tmp/libicu-${ICU_VERSION}.tgz https://github.com/unicode-org/icu/releases/download/release-${ICU_VERSION//./-}/icu4c-${ICU_VERSION//./_}-src.tgz && \
+    echo "94a80cd6f251a53bd2a997f6f1b5ac6653fe791dfab66e1eb0227740fb86d5dc /tmp/libicu-${ICU_VERSION}.tgz" | sha256sum --check && \
+    mkdir /tmp/icu && \
+    pushd /tmp/icu && \
+    tar -xzf /tmp/libicu-${ICU_VERSION}.tgz && \
+    pushd icu/source && \
+    ./configure --prefix=${ICU_PREFIX}  --enable-static --enable-shared=no CXXFLAGS="-fPIC" CFLAGS="-fPIC" && \
+    make -j "$(nproc)" && \
+    make install && \
+    popd && \
+    rm -rf icu && \
+    rm -f /tmp/libicu-${ICU_VERSION}.tgz && \
+    popd
+
 # Switch to nonroot user
 USER nonroot:nonroot
 WORKDIR /home/nonroot
@@ -141,7 +180,7 @@ WORKDIR /home/nonroot
 
 # Rust
 # Please keep the version of llvm (installed above) in sync with rust llvm (`rustc --version --verbose | grep LLVM`)
-ENV RUSTC_VERSION=1.78.0
+ENV RUSTC_VERSION=1.79.0
 ENV RUSTUP_HOME="/home/nonroot/.rustup"
 ENV PATH="/home/nonroot/.cargo/bin:${PATH}"
 RUN curl -sSO https://static.rust-lang.org/rustup/dist/$(uname -m)-unknown-linux-gnu/rustup-init && whoami && \
@@ -170,3 +209,6 @@ RUN whoami \
     && rustup --version --verbose \
     && rustc --version --verbose \
     && clang --version
+
+# Set following flag to check in Makefile if its running in Docker
+RUN touch /home/nonroot/.docker_build

Dockerfile.compute-node

@@ -246,8 +246,8 @@ COPY patches/pgvector.patch /pgvector.patch
 # By default, pgvector Makefile uses `-march=native`. We don't want that,
 # because we build the images on different machines than where we run them.
 # Pass OPTFLAGS="" to remove it.
-RUN wget https://github.com/pgvector/pgvector/archive/refs/tags/v0.7.1.tar.gz -O pgvector.tar.gz && \
-    echo "fe6c8cb4e0cd1a8cb60f5badf9e1701e0fcabcfc260931c26d01e155c4dd21d1 pgvector.tar.gz" | sha256sum --check && \
+RUN wget https://github.com/pgvector/pgvector/archive/refs/tags/v0.7.2.tar.gz -O pgvector.tar.gz && \
+    echo "617fba855c9bcb41a2a9bc78a78567fd2e147c72afd5bf9d37b31b9591632b30 pgvector.tar.gz" | sha256sum --check && \
     mkdir pgvector-src && cd pgvector-src && tar xzf ../pgvector.tar.gz --strip-components=1 -C . && \
     patch -p1 < /pgvector.patch && \
     make -j $(getconf _NPROCESSORS_ONLN) OPTFLAGS="" PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
@@ -928,6 +928,69 @@ RUN rm -r /usr/local/pgsql/include
 # if they were to be used by other libraries.
 RUN rm /usr/local/pgsql/lib/lib*.a
 
+
+#########################################################################################
+#
+# Layer neon-pg-ext-test
+#
+#########################################################################################
+
+FROM neon-pg-ext-build AS neon-pg-ext-test
+ARG PG_VERSION
+RUN mkdir /ext-src
+
+#COPY --from=postgis-build /postgis.tar.gz /ext-src/
+#COPY --from=postgis-build /sfcgal/* /usr
+COPY --from=plv8-build /plv8.tar.gz /ext-src/
+COPY --from=h3-pg-build /h3-pg.tar.gz /ext-src/
+COPY --from=unit-pg-build /postgresql-unit.tar.gz /ext-src/
+COPY --from=vector-pg-build /pgvector.tar.gz /ext-src/
+COPY --from=vector-pg-build /pgvector.patch /ext-src/
+COPY --from=pgjwt-pg-build /pgjwt.tar.gz /ext-src
+#COPY --from=pg-jsonschema-pg-build /home/nonroot/pg_jsonschema.tar.gz /ext-src
+#COPY --from=pg-graphql-pg-build /home/nonroot/pg_graphql.tar.gz /ext-src
+#COPY --from=pg-tiktoken-pg-build /home/nonroot/pg_tiktoken.tar.gz /ext-src
+COPY --from=hypopg-pg-build /hypopg.tar.gz /ext-src
+COPY --from=pg-hashids-pg-build /pg_hashids.tar.gz /ext-src
+#COPY --from=rum-pg-build /rum.tar.gz /ext-src
+#COPY --from=pgtap-pg-build /pgtap.tar.gz /ext-src
+COPY --from=ip4r-pg-build /ip4r.tar.gz /ext-src
+COPY --from=prefix-pg-build /prefix.tar.gz /ext-src
+COPY --from=hll-pg-build /hll.tar.gz /ext-src
+COPY --from=plpgsql-check-pg-build /plpgsql_check.tar.gz /ext-src
+#COPY --from=timescaledb-pg-build /timescaledb.tar.gz /ext-src
+COPY --from=pg-hint-plan-pg-build /pg_hint_plan.tar.gz /ext-src
+COPY patches/pg_hintplan.patch /ext-src
+#COPY --from=kq-imcx-pg-build /kq_imcx.tar.gz /ext-src
+COPY --from=pg-cron-pg-build /pg_cron.tar.gz /ext-src
+COPY patches/pg_cron.patch /ext-src
+#COPY --from=pg-pgx-ulid-build /home/nonroot/pgx_ulid.tar.gz /ext-src
+COPY --from=rdkit-pg-build /rdkit.tar.gz /ext-src
+COPY --from=pg-uuidv7-pg-build /pg_uuidv7.tar.gz /ext-src
+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
+COPY --from=pg-anon-pg-build /pg_anon.tar.gz /ext-src
+COPY 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; \
+    do echo $f; dname=$(echo $f | sed 's/\.tar.*//')-src; \
+    rm -rf $dname; mkdir $dname; tar xzf $f --strip-components=1 -C $dname \
+    || exit 1; rm -f $f; done
+RUN cd /ext-src/pgvector-src && patch -p1 <../pgvector.patch
+# cmake is required for the h3 test
+RUN apt-get update && apt-get install -y cmake
+RUN patch -p1 < /ext-src/pg_hintplan.patch
+COPY --chmod=755 docker-compose/run-tests.sh /run-tests.sh
+RUN patch -p1 </ext-src/pg_anon.patch
+RUN patch -p1 </ext-src/pg_cron.patch
+ENV PATH=/usr/local/pgsql/bin:$PATH
+ENV PGHOST=compute
+ENV PGPORT=55433
+ENV PGUSER=cloud_admin
+ENV PGDATABASE=postgres
 #########################################################################################
 #
 # Final layer

Makefile

@@ -3,6 +3,9 @@ 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
+
 #
 # We differentiate between release / debug build types using the BUILD_TYPE
 # environment variable.
@@ -20,6 +23,16 @@ else
 	$(error Bad build type '$(BUILD_TYPE)', see Makefile for options)
 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 := $(shell uname -s)
 ifeq ($(UNAME_S),Linux)
 	# Seccomp BPF is only available for Linux
@@ -28,7 +41,7 @@ else ifeq ($(UNAME_S),Darwin)
 	ifndef DISABLE_HOMEBREW
 		# macOS with brew-installed openssl requires explicit paths
 		# It can be configured with OPENSSL_PREFIX variable
-		OPENSSL_PREFIX ?= $(shell brew --prefix openssl@3)
+		OPENSSL_PREFIX := $(shell brew --prefix openssl@3)
 		PG_CONFIGURE_OPTS += --with-includes=$(OPENSSL_PREFIX)/include --with-libraries=$(OPENSSL_PREFIX)/lib
 		PG_CONFIGURE_OPTS += PKG_CONFIG_PATH=$(shell brew --prefix icu4c)/lib/pkgconfig
 		# macOS already has bison and flex in the system, but they are old and result in postgres-v14 target failure
@@ -124,6 +137,8 @@ postgres-%: postgres-configure-% \
 	$(MAKE) -C $(POSTGRES_INSTALL_DIR)/build/$*/contrib/pageinspect install
 	+@echo "Compiling amcheck $*"
 	$(MAKE) -C $(POSTGRES_INSTALL_DIR)/build/$*/contrib/amcheck install
+	+@echo "Compiling test_decoding $*"
+	$(MAKE) -C $(POSTGRES_INSTALL_DIR)/build/$*/contrib/test_decoding install
 
 .PHONY: postgres-clean-%
 postgres-clean-%:

compute_tools/src/bin/compute_ctl.rs

@@ -735,7 +735,7 @@ fn cli() -> clap::Command {
             Arg::new("filecache-connstr")
                 .long("filecache-connstr")
                 .default_value(
-                    "host=localhost port=5432 dbname=postgres user=cloud_admin sslmode=disable",
+                    "host=localhost port=5432 dbname=postgres user=cloud_admin sslmode=disable application_name=vm-monitor",
                 )
                 .value_name("FILECACHE_CONNSTR"),
         )

compute_tools/src/compute.rs

@@ -918,38 +918,39 @@ impl ComputeNode {
         // temporarily reset max_cluster_size in config
         // to avoid the possibility of hitting the limit, while we are reconfiguring:
         // creating new extensions, roles, etc...
-        config::compute_ctl_temp_override_create(pgdata_path, "neon.max_cluster_size=-1")?;
-        self.pg_reload_conf()?;
+        config::with_compute_ctl_tmp_override(pgdata_path, "neon.max_cluster_size=-1", || {
+            self.pg_reload_conf()?;
 
-        let mut client = Client::connect(self.connstr.as_str(), NoTls)?;
+            let mut client = Client::connect(self.connstr.as_str(), NoTls)?;
 
-        // Proceed with post-startup configuration. Note, that order of operations is important.
-        // Disable DDL forwarding because control plane already knows about these roles/databases.
-        if spec.mode == ComputeMode::Primary {
-            client.simple_query("SET neon.forward_ddl = false")?;
-            cleanup_instance(&mut client)?;
-            handle_roles(&spec, &mut client)?;
-            handle_databases(&spec, &mut client)?;
-            handle_role_deletions(&spec, self.connstr.as_str(), &mut client)?;
-            handle_grants(
-                &spec,
-                &mut client,
-                self.connstr.as_str(),
-                self.has_feature(ComputeFeature::AnonExtension),
-            )?;
-            handle_extensions(&spec, &mut client)?;
-            handle_extension_neon(&mut client)?;
-            // We can skip handle_migrations here because a new migration can only appear
-            // if we have a new version of the compute_ctl binary, which can only happen
-            // if compute got restarted, in which case we'll end up inside of apply_config
-            // instead of reconfigure.
-        }
+            // Proceed with post-startup configuration. Note, that order of operations is important.
+            // Disable DDL forwarding because control plane already knows about these roles/databases.
+            if spec.mode == ComputeMode::Primary {
+                client.simple_query("SET neon.forward_ddl = false")?;
+                cleanup_instance(&mut client)?;
+                handle_roles(&spec, &mut client)?;
+                handle_databases(&spec, &mut client)?;
+                handle_role_deletions(&spec, self.connstr.as_str(), &mut client)?;
+                handle_grants(
+                    &spec,
+                    &mut client,
+                    self.connstr.as_str(),
+                    self.has_feature(ComputeFeature::AnonExtension),
+                )?;
+                handle_extensions(&spec, &mut client)?;
+                handle_extension_neon(&mut client)?;
+                // We can skip handle_migrations here because a new migration can only appear
+                // if we have a new version of the compute_ctl binary, which can only happen
+                // if compute got restarted, in which case we'll end up inside of apply_config
+                // instead of reconfigure.
+            }
 
-        // 'Close' connection
-        drop(client);
+            // 'Close' connection
+            drop(client);
+
+            Ok(())
+        })?;
 
-        // reset max_cluster_size in config back to original value and reload config
-        config::compute_ctl_temp_override_remove(pgdata_path)?;
         self.pg_reload_conf()?;
 
         let unknown_op = "unknown".to_string();
@@ -1040,12 +1041,17 @@ impl ComputeNode {
                 // temporarily reset max_cluster_size in config
                 // to avoid the possibility of hitting the limit, while we are applying config:
                 // creating new extensions, roles, etc...
-                config::compute_ctl_temp_override_create(pgdata_path, "neon.max_cluster_size=-1")?;
-                self.pg_reload_conf()?;
+                config::with_compute_ctl_tmp_override(
+                    pgdata_path,
+                    "neon.max_cluster_size=-1",
+                    || {
+                        self.pg_reload_conf()?;
 
-                self.apply_config(&compute_state)?;
+                        self.apply_config(&compute_state)?;
 
-                config::compute_ctl_temp_override_remove(pgdata_path)?;
+                        Ok(())
+                    },
+                )?;
                 self.pg_reload_conf()?;
             }
             self.post_apply_config()?;

compute_tools/src/config.rs

@@ -131,18 +131,17 @@ pub fn write_postgres_conf(
     Ok(())
 }
 
-/// create file compute_ctl_temp_override.conf in pgdata_dir
-/// add provided options to this file
-pub fn compute_ctl_temp_override_create(pgdata_path: &Path, options: &str) -> Result<()> {
+pub fn with_compute_ctl_tmp_override<F>(pgdata_path: &Path, options: &str, exec: F) -> Result<()>
+where
+    F: FnOnce() -> Result<()>,
+{
     let path = pgdata_path.join("compute_ctl_temp_override.conf");
     let mut file = File::create(path)?;
     write!(file, "{}", options)?;
-    Ok(())
-}
 
-/// remove file compute_ctl_temp_override.conf in pgdata_dir
-pub fn compute_ctl_temp_override_remove(pgdata_path: &Path) -> Result<()> {
-    let path = pgdata_path.join("compute_ctl_temp_override.conf");
-    std::fs::remove_file(path)?;
-    Ok(())
+    let res = exec();
+
+    file.set_len(0)?;
+
+    res
 }

compute_tools/src/http/api.rs

@@ -17,7 +17,7 @@ use hyper::header::CONTENT_TYPE;
 use hyper::service::{make_service_fn, service_fn};
 use hyper::{Body, Method, Request, Response, Server, StatusCode};
 use tokio::task;
-use tracing::{error, info, warn};
+use tracing::{debug, error, info, warn};
 use tracing_utils::http::OtelName;
 use utils::http::request::must_get_query_param;
 
@@ -48,7 +48,7 @@ async fn routes(req: Request<Body>, compute: &Arc<ComputeNode>) -> Response<Body
     match (req.method(), req.uri().path()) {
         // Serialized compute state.
         (&Method::GET, "/status") => {
-            info!("serving /status GET request");
+            debug!("serving /status GET request");
             let state = compute.state.lock().unwrap();
             let status_response = status_response_from_state(&state);
             Response::new(Body::from(serde_json::to_string(&status_response).unwrap()))

compute_tools/src/migrations/0000-neon_superuser_bypass_rls.sql

@@ -0,0 +1 @@
+ALTER ROLE neon_superuser BYPASSRLS;

compute_tools/src/migrations/0001-alter_roles.sql

@@ -0,0 +1,18 @@
+DO $$
+DECLARE
+    role_name text;
+BEGIN
+    FOR role_name IN SELECT rolname FROM pg_roles WHERE pg_has_role(rolname, 'neon_superuser', 'member')
+    LOOP
+        RAISE NOTICE 'EXECUTING ALTER ROLE % INHERIT', quote_ident(role_name);
+        EXECUTE 'ALTER ROLE ' || quote_ident(role_name) || ' INHERIT';
+    END LOOP;
+
+    FOR role_name IN SELECT rolname FROM pg_roles
+        WHERE
+            NOT pg_has_role(rolname, 'neon_superuser', 'member') AND NOT starts_with(rolname, 'pg_')
+    LOOP
+        RAISE NOTICE 'EXECUTING ALTER ROLE % NOBYPASSRLS', quote_ident(role_name);
+        EXECUTE 'ALTER ROLE ' || quote_ident(role_name) || ' NOBYPASSRLS';
+    END LOOP;
+END $$;

compute_tools/src/migrations/0002-grant_pg_create_subscription_to_neon_superuser.sql

@@ -0,0 +1,6 @@
+DO $$
+BEGIN
+    IF (SELECT setting::numeric >= 160000 FROM pg_settings WHERE name = 'server_version_num') THEN
+        EXECUTE 'GRANT pg_create_subscription TO neon_superuser';
+    END IF;
+END $$;

compute_tools/src/migrations/0003-grant_pg_monitor_to_neon_superuser.sql

@@ -0,0 +1 @@
+GRANT pg_monitor TO neon_superuser WITH ADMIN OPTION;

compute_tools/src/migrations/0004-grant_all_on_tables_to_neon_superuser.sql

@@ -0,0 +1,4 @@
+-- SKIP: Deemed insufficient for allowing relations created by extensions to be
+--       interacted with by neon_superuser without permission issues.
+
+ALTER DEFAULT PRIVILEGES IN SCHEMA public GRANT ALL ON TABLES TO neon_superuser;

compute_tools/src/migrations/0005-grant_all_on_sequences_to_neon_superuser.sql

@@ -0,0 +1,4 @@
+-- SKIP: Deemed insufficient for allowing relations created by extensions to be
+--       interacted with by neon_superuser without permission issues.
+
+ALTER DEFAULT PRIVILEGES IN SCHEMA public GRANT ALL ON SEQUENCES TO neon_superuser;

compute_tools/src/migrations/0006-grant_all_on_tables_to_neon_superuser_with_grant_option.sql

@@ -0,0 +1,3 @@
+-- SKIP: Moved inline to the handle_grants() functions.
+
+ALTER DEFAULT PRIVILEGES IN SCHEMA public GRANT ALL ON TABLES TO neon_superuser WITH GRANT OPTION;

compute_tools/src/migrations/0007-grant_all_on_sequences_to_neon_superuser_with_grant_option.sql

@@ -0,0 +1,3 @@
+-- SKIP: Moved inline to the handle_grants() functions.
+
+ALTER DEFAULT PRIVILEGES IN SCHEMA public GRANT ALL ON SEQUENCES TO neon_superuser WITH GRANT OPTION;

compute_tools/src/migrations/0008-revoke_replication_for_previously_allowed_roles.sql

@@ -0,0 +1,13 @@
+-- SKIP: The original goal of this migration was to prevent creating
+--       subscriptions, but this migration was insufficient.
+
+DO $$
+DECLARE
+    role_name TEXT;
+BEGIN
+    FOR role_name IN SELECT rolname FROM pg_roles WHERE rolreplication IS TRUE
+    LOOP
+        RAISE NOTICE 'EXECUTING ALTER ROLE % NOREPLICATION', quote_ident(role_name);
+        EXECUTE 'ALTER ROLE ' || quote_ident(role_name) || ' NOREPLICATION';
+    END LOOP;
+END $$;

compute_tools/src/spec.rs

@@ -774,44 +774,21 @@ pub fn handle_migrations(client: &mut Client) -> Result<()> {
     // !BE SURE TO ONLY ADD MIGRATIONS TO THE END OF THIS ARRAY. IF YOU DO NOT, VERY VERY BAD THINGS MAY HAPPEN!
     // !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
 
+    // Add new migrations in numerical order.
     let migrations = [
-        "ALTER ROLE neon_superuser BYPASSRLS",
-        r#"
-DO $$
-DECLARE
-    role_name text;
-BEGIN
-    FOR role_name IN SELECT rolname FROM pg_roles WHERE pg_has_role(rolname, 'neon_superuser', 'member')
-    LOOP
-        RAISE NOTICE 'EXECUTING ALTER ROLE % INHERIT', quote_ident(role_name);
-        EXECUTE 'ALTER ROLE ' || quote_ident(role_name) || ' INHERIT';
-    END LOOP;
-
-    FOR role_name IN SELECT rolname FROM pg_roles
-        WHERE
-            NOT pg_has_role(rolname, 'neon_superuser', 'member') AND NOT starts_with(rolname, 'pg_')
-    LOOP
-        RAISE NOTICE 'EXECUTING ALTER ROLE % NOBYPASSRLS', quote_ident(role_name);
-        EXECUTE 'ALTER ROLE ' || quote_ident(role_name) || ' NOBYPASSRLS';
-    END LOOP;
-END $$;
-"#,
-        r#"
-DO $$
-BEGIN
-    IF (SELECT setting::numeric >= 160000 FROM pg_settings WHERE name = 'server_version_num') THEN
-        EXECUTE 'GRANT pg_create_subscription TO neon_superuser';
-    END IF;
-END
-$$;"#,
-        "GRANT pg_monitor TO neon_superuser WITH ADMIN OPTION",
-        // Don't remove: these are some SQLs that we originally applied in migrations but turned out to execute somewhere else.
-        "",
-        "",
-        "",
-        "",
-        "",
-        // Add new migrations below.
+        include_str!("./migrations/0000-neon_superuser_bypass_rls.sql"),
+        include_str!("./migrations/0001-alter_roles.sql"),
+        include_str!("./migrations/0002-grant_pg_create_subscription_to_neon_superuser.sql"),
+        include_str!("./migrations/0003-grant_pg_monitor_to_neon_superuser.sql"),
+        include_str!("./migrations/0004-grant_all_on_tables_to_neon_superuser.sql"),
+        include_str!("./migrations/0005-grant_all_on_sequences_to_neon_superuser.sql"),
+        include_str!(
+            "./migrations/0006-grant_all_on_tables_to_neon_superuser_with_grant_option.sql"
+        ),
+        include_str!(
+            "./migrations/0007-grant_all_on_sequences_to_neon_superuser_with_grant_option.sql"
+        ),
+        include_str!("./migrations/0008-revoke_replication_for_previously_allowed_roles.sql"),
     ];
 
     let mut func = || {
@@ -847,10 +824,13 @@ $$;"#,
 
     while current_migration < migrations.len() {
         let migration = &migrations[current_migration];
-        if migration.is_empty() {
-            info!("Skip migration id={}", current_migration);
+        if migration.starts_with("-- SKIP") {
+            info!("Skipping migration id={}", current_migration);
         } else {
-            info!("Running migration:\n{}\n", migration);
+            info!(
+                "Running migration id={}:\n{}\n",
+                current_migration, migration
+            );
             client.simple_query(migration).with_context(|| {
                 format!("handle_migrations current_migration={}", current_migration)
             })?;

control_plane/src/background_process.rs

@@ -69,6 +69,9 @@ where
     // Not generic AsRef<OsStr>, otherwise empty `envs` prevents type inference
     EI: IntoIterator<Item = (String, String)>,
 {
+    if !datadir.metadata().context("stat datadir")?.is_dir() {
+        anyhow::bail!("`datadir` must be a directory when calling this function: {datadir:?}");
+    }
     let log_path = datadir.join(format!("{process_name}.log"));
     let process_log_file = fs::OpenOptions::new()
         .create(true)
@@ -85,7 +88,13 @@ where
     let background_command = command
         .stdout(process_log_file)
         .stderr(same_file_for_stderr)
-        .args(args);
+        .args(args)
+        // spawn all child processes in their datadir, useful for all kinds of things,
+        // not least cleaning up child processes e.g. after an unclean exit from the test suite:
+        // ```
+        // lsof  -d cwd -a +D  Users/cs/src/neon/test_output
+        // ```
+        .current_dir(datadir);
 
     let filled_cmd = fill_env_vars_prefixed_neon(fill_remote_storage_secrets_vars(
         fill_rust_env_vars(background_command),

control_plane/src/bin/neon_local.rs

@@ -87,7 +87,8 @@ fn main() -> Result<()> {
         handle_init(sub_args).map(Some)
     } else {
         // all other commands need an existing config
-        let mut env = LocalEnv::load_config().context("Error loading config")?;
+        let mut env =
+            LocalEnv::load_config(&local_env::base_path()).context("Error loading config")?;
         let original_env = env.clone();
 
         let rt = tokio::runtime::Builder::new_current_thread()
@@ -364,7 +365,8 @@ fn handle_init(init_match: &ArgMatches) -> anyhow::Result<LocalEnv> {
 
     LocalEnv::init(init_conf, force)
         .context("materialize initial neon_local environment on disk")?;
-    Ok(LocalEnv::load_config().expect("freshly written config should be loadable"))
+    Ok(LocalEnv::load_config(&local_env::base_path())
+        .expect("freshly written config should be loadable"))
 }
 
 /// The default pageserver is the one where CLI tenant/timeline operations are sent by default.

control_plane/src/local_env.rs

@@ -42,8 +42,8 @@ pub struct LocalEnv {
     // compute endpoints).
     //
     // This is not stored in the config file. Rather, this is the path where the
-    // config file itself is. It is read from the NEON_REPO_DIR env variable or
-    // '.neon' if not given.
+    // config file itself is. It is read from the NEON_REPO_DIR env variable which
+    // must be an absolute path. If the env var is not set, $PWD/.neon is used.
     pub base_data_dir: PathBuf,
 
     // Path to postgres distribution. It's expected that "bin", "include",
@@ -431,9 +431,7 @@ impl LocalEnv {
     }
 
     ///  Construct `Self` from on-disk state.
-    pub fn load_config() -> anyhow::Result<Self> {
-        let repopath = base_path();
-
+    pub fn load_config(repopath: &Path) -> anyhow::Result<Self> {
         if !repopath.exists() {
             bail!(
                 "Neon config is not found in {}. You need to run 'neon_local init' first",
@@ -461,7 +459,7 @@ impl LocalEnv {
                 branch_name_mappings,
             } = on_disk_config;
             LocalEnv {
-                base_data_dir: repopath.clone(),
+                base_data_dir: repopath.to_owned(),
                 pg_distrib_dir,
                 neon_distrib_dir,
                 default_tenant_id,
@@ -482,7 +480,7 @@ impl LocalEnv {
             "we ensure this during deserialization"
         );
         env.pageservers = {
-            let iter = std::fs::read_dir(&repopath).context("open dir")?;
+            let iter = std::fs::read_dir(repopath).context("open dir")?;
             let mut pageservers = Vec::new();
             for res in iter {
                 let dentry = res?;
@@ -719,10 +717,25 @@ impl LocalEnv {
 }
 
 pub fn base_path() -> PathBuf {
-    match std::env::var_os("NEON_REPO_DIR") {
-        Some(val) => PathBuf::from(val),
-        None => PathBuf::from(".neon"),
-    }
+    let path = match std::env::var_os("NEON_REPO_DIR") {
+        Some(val) => {
+            let path = PathBuf::from(val);
+            if !path.is_absolute() {
+                // repeat the env var in the error because our default is always absolute
+                panic!("NEON_REPO_DIR must be an absolute path, got {path:?}");
+            }
+            path
+        }
+        None => {
+            let pwd = std::env::current_dir()
+                // technically this can fail but it's quite unlikeley
+                .expect("determine current directory");
+            let pwd_abs = pwd.canonicalize().expect("canonicalize current directory");
+            pwd_abs.join(".neon")
+        }
+    };
+    assert!(path.is_absolute());
+    path
 }
 
 /// Generate a public/private key pair for JWT authentication

control_plane/src/pageserver.rs

@@ -383,6 +383,10 @@ impl PageServerNode {
                 .map(|x| x.parse::<AuxFilePolicy>())
                 .transpose()
                 .context("Failed to parse 'switch_aux_file_policy'")?,
+            lsn_lease_length: settings.remove("lsn_lease_length").map(|x| x.to_string()),
+            lsn_lease_length_for_ts: settings
+                .remove("lsn_lease_length_for_ts")
+                .map(|x| x.to_string()),
         };
         if !settings.is_empty() {
             bail!("Unrecognized tenant settings: {settings:?}")
@@ -506,6 +510,10 @@ impl PageServerNode {
                     .map(|x| x.parse::<AuxFilePolicy>())
                     .transpose()
                     .context("Failed to parse 'switch_aux_file_policy'")?,
+                lsn_lease_length: settings.remove("lsn_lease_length").map(|x| x.to_string()),
+                lsn_lease_length_for_ts: settings
+                    .remove("lsn_lease_length_for_ts")
+                    .map(|x| x.to_string()),
             }
         };
 

control_plane/src/safekeeper.rs

@@ -14,6 +14,7 @@ use camino::Utf8PathBuf;
 use postgres_connection::PgConnectionConfig;
 use reqwest::{IntoUrl, Method};
 use thiserror::Error;
+use utils::auth::{Claims, Scope};
 use utils::{http::error::HttpErrorBody, id::NodeId};
 
 use crate::{
@@ -197,7 +198,7 @@ impl SafekeeperNode {
             &datadir,
             &self.env.safekeeper_bin(),
             &args,
-            [],
+            self.safekeeper_env_variables()?,
             background_process::InitialPidFile::Expect(self.pid_file()),
             || async {
                 match self.check_status().await {
@@ -210,6 +211,18 @@ impl SafekeeperNode {
         .await
     }
 
+    fn safekeeper_env_variables(&self) -> anyhow::Result<Vec<(String, String)>> {
+        // Generate a token to connect from safekeeper to peers
+        if self.conf.auth_enabled {
+            let token = self
+                .env
+                .generate_auth_token(&Claims::new(None, Scope::SafekeeperData))?;
+            Ok(vec![("SAFEKEEPER_AUTH_TOKEN".to_owned(), token)])
+        } else {
+            Ok(Vec::new())
+        }
+    }
+
     ///
     /// Stop the server.
     ///

control_plane/src/storage_controller.rs

@@ -46,6 +46,7 @@ const STORAGE_CONTROLLER_POSTGRES_VERSION: u32 = 16;
 pub struct AttachHookRequest {
     pub tenant_shard_id: TenantShardId,
     pub node_id: Option<NodeId>,
+    pub generation_override: Option<i32>,
 }
 
 #[derive(Serialize, Deserialize)]
@@ -313,15 +314,17 @@ impl StorageController {
             args.push(format!("--split-threshold={split_threshold}"))
         }
 
+        args.push(format!(
+            "--neon-local-repo-dir={}",
+            self.env.base_data_dir.display()
+        ));
+
         background_process::start_process(
             COMMAND,
             &self.env.base_data_dir,
             &self.env.storage_controller_bin(),
             args,
-            [(
-                "NEON_REPO_DIR".to_string(),
-                self.env.base_data_dir.to_string_lossy().to_string(),
-            )],
+            [],
             background_process::InitialPidFile::Create(self.pid_file()),
             || async {
                 match self.ready().await {
@@ -440,6 +443,7 @@ impl StorageController {
         let request = AttachHookRequest {
             tenant_shard_id,
             node_id: Some(pageserver_id),
+            generation_override: None,
         };
 
         let response = self

control_plane/storcon_cli/Cargo.toml

@@ -9,6 +9,7 @@ license.workspace = true
 anyhow.workspace = true
 clap.workspace = true
 comfy-table.workspace = true
+futures.workspace = true
 humantime.workspace = true
 hyper.workspace = true
 pageserver_api.workspace = true

control_plane/storcon_cli/src/main.rs

@@ -1,3 +1,4 @@
+use futures::StreamExt;
 use std::{collections::HashMap, str::FromStr, time::Duration};
 
 use clap::{Parser, Subcommand};
@@ -148,6 +149,22 @@ enum Command {
         #[arg(long)]
         threshold: humantime::Duration,
     },
+    // Drain a set of specified pageservers by moving the primary attachments to pageservers
+    // outside of the specified set.
+    Drain {
+        // Set of pageserver node ids to drain.
+        #[arg(long)]
+        nodes: Vec<NodeId>,
+        // Optional: migration concurrency (default is 8)
+        #[arg(long)]
+        concurrency: Option<usize>,
+        // Optional: maximum number of shards to migrate
+        #[arg(long)]
+        max_shards: Option<usize>,
+        // Optional: when set to true, nothing is migrated, but the plan is printed to stdout
+        #[arg(long)]
+        dry_run: Option<bool>,
+    },
 }
 
 #[derive(Parser)]
@@ -737,6 +754,194 @@ async fn main() -> anyhow::Result<()> {
                 })
                 .await?;
         }
+        Command::Drain {
+            nodes,
+            concurrency,
+            max_shards,
+            dry_run,
+        } => {
+            // Load the list of nodes, split them up into the drained and filled sets,
+            // and validate that draining is possible.
+            let node_descs = storcon_client
+                .dispatch::<(), Vec<NodeDescribeResponse>>(
+                    Method::GET,
+                    "control/v1/node".to_string(),
+                    None,
+                )
+                .await?;
+
+            let mut node_to_drain_descs = Vec::new();
+            let mut node_to_fill_descs = Vec::new();
+
+            for desc in node_descs {
+                let to_drain = nodes.iter().any(|id| *id == desc.id);
+                if to_drain {
+                    node_to_drain_descs.push(desc);
+                } else {
+                    node_to_fill_descs.push(desc);
+                }
+            }
+
+            if nodes.len() != node_to_drain_descs.len() {
+                anyhow::bail!("Drain requested for node which doesn't exist.")
+            }
+
+            node_to_fill_descs.retain(|desc| {
+                matches!(desc.availability, NodeAvailabilityWrapper::Active)
+                    && matches!(
+                        desc.scheduling,
+                        NodeSchedulingPolicy::Active | NodeSchedulingPolicy::Filling
+                    )
+            });
+
+            if node_to_fill_descs.is_empty() {
+                anyhow::bail!("There are no nodes to drain to")
+            }
+
+            // Set the node scheduling policy to draining for the nodes which
+            // we plan to drain.
+            for node_desc in node_to_drain_descs.iter() {
+                let req = NodeConfigureRequest {
+                    node_id: node_desc.id,
+                    availability: None,
+                    scheduling: Some(NodeSchedulingPolicy::Draining),
+                };
+
+                storcon_client
+                    .dispatch::<_, ()>(
+                        Method::PUT,
+                        format!("control/v1/node/{}/config", node_desc.id),
+                        Some(req),
+                    )
+                    .await?;
+            }
+
+            // Perform the drain: move each tenant shard scheduled on a node to
+            // be drained to a node which is being filled. A simple round robin
+            // strategy is used to pick the new node.
+            let tenants = storcon_client
+                .dispatch::<(), Vec<TenantDescribeResponse>>(
+                    Method::GET,
+                    "control/v1/tenant".to_string(),
+                    None,
+                )
+                .await?;
+
+            let mut selected_node_idx = 0;
+
+            struct DrainMove {
+                tenant_shard_id: TenantShardId,
+                from: NodeId,
+                to: NodeId,
+            }
+
+            let mut moves: Vec<DrainMove> = Vec::new();
+
+            let shards = tenants
+                .into_iter()
+                .flat_map(|tenant| tenant.shards.into_iter());
+            for shard in shards {
+                if let Some(max_shards) = max_shards {
+                    if moves.len() >= max_shards {
+                        println!(
+                            "Stop planning shard moves since the requested maximum was reached"
+                        );
+                        break;
+                    }
+                }
+
+                let should_migrate = {
+                    if let Some(attached_to) = shard.node_attached {
+                        node_to_drain_descs
+                            .iter()
+                            .map(|desc| desc.id)
+                            .any(|id| id == attached_to)
+                    } else {
+                        false
+                    }
+                };
+
+                if !should_migrate {
+                    continue;
+                }
+
+                moves.push(DrainMove {
+                    tenant_shard_id: shard.tenant_shard_id,
+                    from: shard
+                        .node_attached
+                        .expect("We only migrate attached tenant shards"),
+                    to: node_to_fill_descs[selected_node_idx].id,
+                });
+                selected_node_idx = (selected_node_idx + 1) % node_to_fill_descs.len();
+            }
+
+            let total_moves = moves.len();
+
+            if dry_run == Some(true) {
+                println!("Dryrun requested. Planned {total_moves} moves:");
+                for mv in &moves {
+                    println!("{}: {} -> {}", mv.tenant_shard_id, mv.from, mv.to)
+                }
+
+                return Ok(());
+            }
+
+            const DEFAULT_MIGRATE_CONCURRENCY: usize = 8;
+            let mut stream = futures::stream::iter(moves)
+                .map(|mv| {
+                    let client = Client::new(cli.api.clone(), cli.jwt.clone());
+                    async move {
+                        client
+                            .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,
+                                }),
+                            )
+                            .await
+                            .map_err(|e| (mv.tenant_shard_id, mv.from, mv.to, e))
+                    }
+                })
+                .buffered(concurrency.unwrap_or(DEFAULT_MIGRATE_CONCURRENCY));
+
+            let mut success = 0;
+            let mut failure = 0;
+
+            while let Some(res) = stream.next().await {
+                match res {
+                    Ok(_) => {
+                        success += 1;
+                    }
+                    Err((tenant_shard_id, from, to, error)) => {
+                        failure += 1;
+                        println!(
+                            "Failed to migrate {} from node {} to node {}: {}",
+                            tenant_shard_id, from, to, error
+                        );
+                    }
+                }
+
+                if (success + failure) % 20 == 0 {
+                    println!(
+                        "Processed {}/{} shards: {} succeeded, {} failed",
+                        success + failure,
+                        total_moves,
+                        success,
+                        failure
+                    );
+                }
+            }
+
+            println!(
+                "Processed {}/{} shards: {} succeeded, {} failed",
+                success + failure,
+                total_moves,
+                success,
+                failure
+            );
+        }
     }
 
     Ok(())

docker-compose/compute_wrapper/Dockerfile

@@ -8,6 +8,11 @@ USER root
 RUN apt-get update &&       \
     apt-get install -y curl \
                        jq   \
+                       python3-pip \
                        netcat
+#Faker is required for the pg_anon test
+RUN pip3 install 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 
 
-USER postgres
+USER postgres

docker-compose/compute_wrapper/var/db/postgres/specs/spec.json

@@ -95,7 +95,7 @@
             },
             {
                 "name": "shared_preload_libraries",
-                "value": "neon",
+                "value": "neon,pg_cron,timescaledb,pg_stat_statements",
                 "vartype": "string"
             },
             {
@@ -127,6 +127,16 @@
                 "name": "max_replication_flush_lag",
                 "value": "10GB",
                 "vartype": "string"
+            },
+            {
+                "name": "cron.database",
+                "value": "postgres",
+                "vartype": "string"
+            },
+            {
+                "name": "session_preload_libraries",
+                "value": "anon",
+                "vartype": "string"
             }
         ]
     },

docker-compose/docker-compose.yml

@@ -1,5 +1,3 @@
-version: '3'
-
 services:
   minio:
     restart: always
@@ -161,12 +159,12 @@ services:
       context: ./compute_wrapper/
       args:
         - REPOSITORY=${REPOSITORY:-neondatabase}
-        - COMPUTE_IMAGE=compute-node-v${PG_VERSION:-14}
+        - COMPUTE_IMAGE=compute-node-v${PG_VERSION:-16}
         - TAG=${TAG:-latest}
         - http_proxy=$http_proxy
         - https_proxy=$https_proxy
     environment:
-      - PG_VERSION=${PG_VERSION:-14}
+      - PG_VERSION=${PG_VERSION:-16}
       #- RUST_BACKTRACE=1
     # Mount the test files directly, for faster editing cycle.
     volumes:
@@ -194,3 +192,14 @@ services:
          done"
     depends_on:
       - compute
+
+  neon-test-extensions:
+    profiles: ["test-extensions"]
+    image: ${REPOSITORY:-neondatabase}/neon-test-extensions-v${PG_TEST_VERSION:-16}:${TAG:-latest}
+    entrypoint:
+      - "/bin/bash"
+      - "-c"
+    command:
+      - sleep 1800
+    depends_on:
+      - compute

docker-compose/docker_compose_test.sh

@@ -7,52 +7,94 @@
 # Implicitly accepts `REPOSITORY` and `TAG` env vars that are passed into the compose file
 # Their defaults point at DockerHub `neondatabase/neon:latest` image.`,
 # to verify custom image builds (e.g pre-published ones).
-
+#
+# A test script for postgres extensions
+# Currently supports only v16
+#
 set -eux -o pipefail
 
-SCRIPT_DIR="$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )"
-COMPOSE_FILE=$SCRIPT_DIR/docker-compose.yml
-
+COMPOSE_FILE='docker-compose.yml'
+cd $(dirname $0)
 COMPUTE_CONTAINER_NAME=docker-compose-compute-1
-SQL="CREATE TABLE t(key int primary key, value text); insert into t values(1,1); select * from t;"
-PSQL_OPTION="-h localhost -U cloud_admin -p 55433 -c '$SQL' postgres"
+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 -f $COMPOSE_FILE logs
+    docker compose --profile test-extensions -f $COMPOSE_FILE logs
     echo "stop containers..."
-    docker compose -f $COMPOSE_FILE down
+    docker compose --profile test-extensions -f $COMPOSE_FILE down
 }
 
-echo "clean up containers if exists"
-cleanup
-
 for pg_version in 14 15 16; do
-    echo "start containers (pg_version=$pg_version)."
-    PG_VERSION=$pg_version docker compose -f $COMPOSE_FILE up --build -d
+    echo "clean up containers if exists"
+    cleanup
+    PG_TEST_VERSION=$(($pg_version < 16 ? 16 : $pg_version))
+    PG_VERSION=$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. "
     cnt=0
-    while sleep 1; do
+    while sleep 3; do
         # check timeout
-        cnt=`expr $cnt + 1`
+        cnt=`expr $cnt + 3`
         if [ $cnt -gt 60 ]; then
             echo "timeout before the compute is ready."
             cleanup
             exit 1
         fi
-
-        # check if the compute is ready
-        set +o pipefail
-        result=`docker compose -f $COMPOSE_FILE logs "compute_is_ready" | grep "accepting connections" | wc -l`
-        set -o pipefail
-        if [ $result -eq 1 ]; then
+        if docker compose --profile test-extensions -f $COMPOSE_FILE logs "compute_is_ready" | grep -q "accepting connections"; then
             echo "OK. The compute is ready to connect."
             echo "execute simple queries."
             docker exec $COMPUTE_CONTAINER_NAME /bin/bash -c "psql $PSQL_OPTION"
-            cleanup
             break
         fi
     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 $COMPUTE_CONTAINER_NAME touch /var/db/postgres/compute/compute_ctl_temp_override.conf
+        # This block is required for the pg_anon extension test.
+        # The test assumes that it is running on the same host with the postgres engine.
+        # In our case it's not true, that's why we are copying files to the compute node
+        TMPDIR=$(mktemp -d)
+        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 $TMPDIR
+        TMPDIR=$(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=rum-src,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
+                mkdir $d
+                docker cp $TEST_CONTAINER_NAME:/ext-src/$d/regression.diffs $d || true
+                docker cp $TEST_CONTAINER_NAME:/ext-src/$d/regression.out $d || true
+                cat $d/regression.out $d/regression.diffs || true
+            done
+        rm -rf $FAILED
+        cleanup
+        exit 1
+        fi
+    fi
+    cleanup
 done

docker-compose/run-tests.sh

@@ -0,0 +1,15 @@
+#!/bin/bash
+set -x
+
+cd /ext-src
+FAILED=
+LIST=$((echo ${SKIP} | sed 's/,/\n/g'; ls -d *-src) | sort | uniq -u)
+for d in ${LIST}
+do
+       [ -d ${d} ] || continue
+    psql -c "select 1" >/dev/null || break
+       make -C ${d} installcheck || FAILED="${d} ${FAILED}"
+done
+[ -z "${FAILED}" ] && exit 0
+echo ${FAILED}
+exit 1

docs/core_changes.md

@@ -11,15 +11,28 @@ page server. We currently use the same binary for both, with --wal-redo runtime
 the WAL redo mode. Some PostgreSQL changes are needed in the compute node, while others are just for
 the WAL redo process.
 
-In addition to core PostgreSQL changes, there is a Neon extension in contrib/neon, to hook into the
-smgr interface. Once all the core changes have been submitted to upstream or eliminated some other
-way, the extension could live outside the postgres repository and build against vanilla PostgreSQL.
+In addition to core PostgreSQL changes, there is a Neon extension in the pgxn/neon directory that
+hooks into the smgr interface, and rmgr extension in pgxn/neon_rmgr. The extensions are loaded into
+the Postgres processes with shared_preload_libraries. Most of the Neon-specific code is in the
+extensions, and for any new features, that is preferred over modifying core PostgreSQL code.
 
 Below is a list of all the PostgreSQL source code changes, categorized into changes needed for
 compute, and changes needed for the WAL redo process:
 
 # Changes for Compute node
 
+## Prefetching
+
+There are changes in many places to perform prefetching, for example for sequential scans. Neon
+doesn't benefit from OS readahead, and the latency to pageservers is quite high compared to local
+disk, so prefetching is critical for performance, also for sequential scans.
+
+### How to get rid of the patch
+
+Upcoming "streaming read" work in v17 might simplify this. And async I/O work in v18 will hopefully
+do more.
+
+
 ## Add t_cid to heap WAL records
 
 ```
@@ -37,54 +50,11 @@ The problem is that the XLOG_HEAP_INSERT record does not include the command id
 
 Bite the bullet and submit the patch to PostgreSQL, to add the t_cid to the WAL records. It makes the WAL records larger, which could make this unpopular in the PostgreSQL community. However, it might simplify some logical decoding code; Andres Freund briefly mentioned in PGCon 2022 discussion on Heikki's Neon presentation that logical decoding currently needs to jump through some hoops to reconstruct the same information.
 
+Update from Heikki (2024-04-17): I tried to write an upstream patch for that, to use the t_cid field for logical decoding, but it was not as straightforward as it first sounded.
 
 ### Alternatives
 Perhaps we could write an extra WAL record with the t_cid information, when a page is evicted that contains rows that were touched a transaction that's still running. However, that seems very complicated.
 
-## ginfast.c
-
-```
-diff --git a/src/backend/access/gin/ginfast.c b/src/backend/access/gin/ginfast.c
-index e0d9940946..2d964c02e9 100644
---- a/src/backend/access/gin/ginfast.c
-+++ b/src/backend/access/gin/ginfast.c
-@@ -285,6 +285,17 @@ ginHeapTupleFastInsert(GinState *ginstate, GinTupleCollector *collector)
-                memset(&sublist, 0, sizeof(GinMetaPageData));
-                makeSublist(index, collector->tuples, collector->ntuples, &sublist);
- 
-+               if (metadata->head != InvalidBlockNumber)
-+               {
-+                       /*
-+                        * ZENITH: Get buffer before XLogBeginInsert() to avoid recursive call
-+                        * of XLogBeginInsert(). Reading a new buffer might evict a dirty page from
-+                        * the buffer cache, and if that page happens to be an FSM or VM page, zenith_write()
-+                        * will try to WAL-log an image of the page.
-+                        */
-+                       buffer = ReadBuffer(index, metadata->tail);
-+               }
-+
-                if (needWal)
-                        XLogBeginInsert();
- 
-@@ -316,7 +327,6 @@ ginHeapTupleFastInsert(GinState *ginstate, GinTupleCollector *collector)
-                        data.prevTail = metadata->tail;
-                        data.newRightlink = sublist.head;
- 
--                       buffer = ReadBuffer(index, metadata->tail);
-                        LockBuffer(buffer, GIN_EXCLUSIVE);
-                        page = BufferGetPage(buffer);
-```
-
-The problem is explained in the comment above
-
-### How to get rid of the patch
-
-Can we stop WAL-logging FSM or VM pages? Or delay the WAL logging until we're out of the critical
-section or something.
-
-Maybe some bigger rewrite of FSM and VM would help to avoid WAL-logging FSM and VM page images?
-
-
 ## Mark index builds that use buffer manager without logging explicitly
 
 ```
@@ -95,6 +65,8 @@ Maybe some bigger rewrite of FSM and VM would help to avoid WAL-logging FSM and
 also some changes in src/backend/storage/smgr/smgr.c
 ```
 
+pgvector 0.6.0 also needs a similar change, which would be very nice to get rid of too.
+
 When a GIN index is built, for example, it is built by inserting the entries into the index more or
 less normally, but without WAL-logging anything. After the index has been built, we iterate through
 all pages and write them to the WAL. That doesn't work for Neon, because if a page is not WAL-logged
@@ -109,6 +81,10 @@ an operation: `smgr_start_unlogged_build`, `smgr_finish_unlogged_build_phase_1`
 I think it would make sense to be more explicit about that in PostgreSQL too. So extract these
 changes to a patch and post to pgsql-hackers.
 
+Perhaps we could deduce that an unlogged index build has started when we see a page being evicted
+with zero LSN. How to be sure it's an unlogged index build rather than a bug? Currently we have a
+check for that and PANIC if we see page with zero LSN being evicted. And how do we detect when the
+index build has finished? See https://github.com/neondatabase/neon/pull/7440 for an attempt at that.
 
 ## Track last-written page LSN
 
@@ -140,57 +116,6 @@ The old method is still available, though.
 Wait until v15?
 
 
-## Cache relation sizes
-
-The Neon extension contains a little cache for smgrnblocks() and smgrexists() calls, to avoid going
-to the page server every time. It might be useful to cache those in PostgreSQL, maybe in the
-relcache? (I think we do cache nblocks in relcache already, check why that's not good enough for
-Neon)
-
-
-## Use buffer manager when extending VM or FSM
-
-```
- src/backend/storage/freespace/freespace.c                   |   14 +-
- src/backend/access/heap/visibilitymap.c                     |   15 +-
-
-diff --git a/src/backend/access/heap/visibilitymap.c b/src/backend/access/heap/visibilitymap.c
-index e198df65d8..addfe93eac 100644
---- a/src/backend/access/heap/visibilitymap.c
-+++ b/src/backend/access/heap/visibilitymap.c
-@@ -652,10 +652,19 @@ vm_extend(Relation rel, BlockNumber vm_nblocks)
-        /* Now extend the file */
-        while (vm_nblocks_now < vm_nblocks)
-        {
--               PageSetChecksumInplace((Page) pg.data, vm_nblocks_now);
-+               /*
-+                * ZENITH: Initialize VM pages through buffer cache to prevent loading
-+                * them from pageserver.
-+                */
-+               Buffer  buffer = ReadBufferExtended(rel, VISIBILITYMAP_FORKNUM, P_NEW,
-+                                                                                       RBM_ZERO_AND_LOCK, NULL);
-+               Page    page = BufferGetPage(buffer);
-+
-+               PageInit((Page) page, BLCKSZ, 0);
-+               PageSetChecksumInplace(page, vm_nblocks_now);
-+               MarkBufferDirty(buffer);
-+               UnlockReleaseBuffer(buffer);
- 
--               smgrextend(rel->rd_smgr, VISIBILITYMAP_FORKNUM, vm_nblocks_now,
--                                  pg.data, false);
-                vm_nblocks_now++;
-        }
-```
-
-### Problem we're trying to solve
-
-???
-
-### How to get rid of the patch
-
-Maybe this would be a reasonable change in PostgreSQL too?
-
-
 ## Allow startup without reading checkpoint record
 
 In Neon, the compute node is stateless. So when we are launching compute node, we need to provide
@@ -231,7 +156,7 @@ index 0415df9ccb..9f9db3c8bc 100644
   * crash we can lose (skip over) as many values as we pre-logged.
   */
 -#define SEQ_LOG_VALS   32
-+/* Zenith XXX: to ensure sequence order of sequence in Zenith we need to WAL log each sequence update. */
++/* Neon XXX: to ensure sequence order of sequence in Zenith we need to WAL log each sequence update. */
 +/* #define SEQ_LOG_VALS        32 */
 +#define SEQ_LOG_VALS   0
 ```
@@ -250,66 +175,6 @@ would be weird if the sequence moved backwards though, think of PITR.
 Or add a GUC for the amount to prefix to PostgreSQL, and force it to 1 in Neon.
 
 
-## Walproposer
-
-```
- src/Makefile                                                |    1 +
- src/backend/replication/libpqwalproposer/Makefile           |   37 +
- src/backend/replication/libpqwalproposer/libpqwalproposer.c |  416 ++++++++++++
- src/backend/postmaster/bgworker.c                           |    4 +
- src/backend/postmaster/postmaster.c                         |    6 +
- src/backend/replication/Makefile                            |    4 +-
- src/backend/replication/walproposer.c                       | 2350 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
- src/backend/replication/walproposer_utils.c                 |  402 +++++++++++
- src/backend/replication/walreceiver.c                       |    7 +
- src/backend/replication/walsender.c                         |  320 ++++++---
- src/backend/storage/ipc/ipci.c                              |    6 +
- src/include/replication/walproposer.h                       |  565 ++++++++++++++++
-```
-
-WAL proposer is communicating with safekeeper and ensures WAL durability by quorum writes.  It is
-currently implemented as patch to standard WAL sender.
-
-### How to get rid of the patch
-
-Refactor into an extension. Submit hooks or APIs into upstream if necessary.
-
-@MMeent did some work on this already: https://github.com/neondatabase/postgres/pull/96
-
-## Ignore unexpected data beyond EOF in bufmgr.c
-
-```
-@@ -922,11 +928,14 @@ ReadBuffer_common(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
-                 */
-                bufBlock = isLocalBuf ? LocalBufHdrGetBlock(bufHdr) : BufHdrGetBlock(bufHdr);
-                if (!PageIsNew((Page) bufBlock))
--                       ereport(ERROR,
-+               {
-+                        // XXX-ZENITH
-+                        MemSet((char *) bufBlock, 0, BLCKSZ);
-+                        ereport(DEBUG1,
-                                        (errmsg("unexpected data beyond EOF in block %u of relation %s",
-                                                        blockNum, relpath(smgr->smgr_rnode, forkNum)),
-                                         errhint("This has been seen to occur with buggy kernels; consider updating your system.")));
--
-+               }
-                /*
-                 * We *must* do smgrextend before succeeding, else the page will not
-                 * be reserved by the kernel, and the next P_NEW call will decide to
-```
-
-PostgreSQL is a bit sloppy with extending relations. Usually, the relation is extended with zeros
-first, then the page is filled, and finally the new page WAL-logged. But if multiple backends extend
-a relation at the same time, the pages can be WAL-logged in different order.
-
-I'm not sure what scenario exactly required this change in Neon, though.
-
-### How to get rid of the patch
-
-Submit patches to pgsql-hackers, to tighten up the WAL-logging around relation extension. It's a bit
-confusing even in PostgreSQL. Maybe WAL log the intention to extend first, then extend the relation,
-and finally WAL-log that the extension succeeded.
-
 ## Make smgr interface available to extensions
 
 ```
@@ -321,6 +186,8 @@ and finally WAL-log that the extension succeeded.
 
 Submit to upstream. This could be useful for the Disk Encryption patches too, or for compression.
 
+We have submitted this to upstream, but it's moving at glacial a speed.
+https://commitfest.postgresql.org/47/4428/
 
 ## Added relpersistence argument to smgropen()
 
@@ -444,6 +311,148 @@ Ignore it. This is only needed for disaster recovery, so once we've eliminated a
 patches, we can just keep it around as a patch or as separate branch in a repo.
 
 
+## pg_waldump flags to ignore errors
+
+After creating a new project or branch in Neon, the first timeline can begin in the middle of a WAL segment. pg_waldump chokes on that, so we added some flags to make it possible to ignore errors.
+
+### How to get rid of the patch
+
+Like previous one, ignore it.
+
+
+
+## Backpressure if pageserver doesn't ingest WAL fast enough
+
+```
+@@ -3200,6 +3202,7 @@ ProcessInterrupts(void)
+                return;
+        InterruptPending = false;
+ 
++retry:
+        if (ProcDiePending)
+        {
+                ProcDiePending = false;
+@@ -3447,6 +3450,13 @@ ProcessInterrupts(void)
+ 
+        if (ParallelApplyMessagePending)
+                HandleParallelApplyMessages();
++
++       /* Call registered callback if any */
++       if (ProcessInterruptsCallback)
++       {
++               if (ProcessInterruptsCallback())
++                       goto retry;
++       }
+ }
+```
+
+
+### How to get rid of the patch
+
+Submit a patch to upstream, for a hook in ProcessInterrupts. Could be useful for other extensions
+too.
+
+
+## SLRU on-demand download
+
+```
+ src/backend/access/transam/slru.c | 105 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++-------------
+ 1 file changed, 92 insertions(+), 13 deletions(-)
+```
+
+### Problem we're trying to solve
+
+Previously, SLRU files were included in the basebackup, but the total size of them can be large,
+several GB, and downloading them all made the startup time too long.
+
+### Alternatives
+
+FUSE hook or LD_PRELOAD trick to intercept the reads on SLRU files
+
+
+## WAL-log an all-zeros page as one large hole
+
+- In XLogRecordAssemble()
+
+### Problem we're trying to solve
+
+This change was made in v16. Starting with v16, when PostgreSQL extends a relation, it first extends
+it with zeros, and it can extend the relation more than one block at a time. The all-zeros page is WAL-ogged, but it's very wasteful to include 8 kB of zeros in the WAL for that. This hack was made so that we WAL logged a compact record with a whole-page "hole". However, PostgreSQL has assertions that prevent that such WAL records from being replayed, so this breaks compatibility such that unmodified PostreSQL cannot process Neon-generated WAL.
+
+### How to get rid of the patch
+
+Find another compact representation for a full-page image of an all-zeros page. A compressed image perhaps.
+
+
+## Shut down walproposer after checkpointer
+
+```
++                       /* Neon: Also allow walproposer background worker to be treated like a WAL sender, so that it's shut down last */
++                       if ((bp->bkend_type == BACKEND_TYPE_NORMAL || bp->bkend_type == BACKEND_TYPE_BGWORKER) &&
+```
+
+This changes was needed so that postmaster shuts down the walproposer process only after the shutdown checkpoint record is written. Otherwise, the shutdown record will never make it to the safekeepers.
+
+### How to get rid of the patch
+
+Do a bigger refactoring of the postmaster state machine, such that a background worker can specify
+the shutdown ordering by itself. The postmaster state machine has grown pretty complicated, and
+would benefit from a refactoring for the sake of readability anyway.
+
+
+## EXPLAIN changes for prefetch and LFC
+
+### How to get rid of the patch
+
+Konstantin submitted a patch to -hackers already: https://commitfest.postgresql.org/47/4643/. Get that into a committable state.
+
+
+## On-demand download of extensions
+
+### How to get rid of the patch
+
+FUSE or LD_PRELOAD trickery to intercept reads?
+
+
+## Publication superuser checks
+
+We have hacked CreatePublication so that also neon_superuser can create them.
+
+### How to get rid of the patch
+
+Create an upstream patch with more fine-grained privileges for publications CREATE/DROP that can be GRANTed to users.
+
+
+## WAL log replication slots
+
+### How to get rid of the patch
+
+Utilize the upcoming v17 "slot sync worker", or a similar neon-specific background worker process, to periodically WAL-log the slots, or to export them somewhere else.
+
+
+## WAL-log replication snapshots
+
+### How to get rid of the patch
+
+WAL-log them periodically, from a backgound worker.
+
+
+## WAL-log relmapper files
+
+Similarly to replications snapshot files, the CID mapping files generated during VACUUM FULL of a catalog table are WAL-logged
+
+### How to get rid of the patch
+
+WAL-log them periodically, from a backgound worker.
+
+
+## XLogWaitForReplayOf()
+
+??
+
+
+
+
 # Not currently committed but proposed
 
 ## Disable ring buffer buffer manager strategies
@@ -472,23 +481,10 @@ hint bits are set. Wal logging hint bits updates requires FPI which significantl
 
 Add special WAL record for setting page hints.
 
-## Prefetching
-
-### Why?
-
-As far as pages in Neon are loaded on demand, to reduce node startup time
-and also speedup some massive queries we need some mechanism for bulk loading to
-reduce page request round-trip overhead.
-
-Currently Postgres is supporting prefetching only for bitmap scan.
-In Neon we should also use prefetch for sequential and index scans, because the OS is not doing it for us.
-For sequential scan we could prefetch some number of following pages. For index scan we could prefetch pages
-of heap relation addressed by TIDs.
-
 ## Prewarming
 
 ### Why?
 
-Short downtime (or, in other words, fast compute node restart time) is one of the key feature of Zenith.
+Short downtime (or, in other words, fast compute node restart time) is one of the key feature of Neon.
 But overhead of request-response round-trip for loading pages on demand can make started node warm-up quite slow.
 We can capture state of compute node buffer cache and send bulk request for this pages at startup.

docs/docker.md

@@ -4,18 +4,18 @@
 
 Currently we build two main images:
 
-- [neondatabase/neon](https://hub.docker.com/repository/docker/zenithdb/zenith) — image with pre-built `pageserver`, `safekeeper` and `proxy` binaries and all the required runtime dependencies. Built from [/Dockerfile](/Dockerfile).
-- [neondatabase/compute-node](https://hub.docker.com/repository/docker/zenithdb/compute-node) — compute node image with pre-built Postgres binaries from [neondatabase/postgres](https://github.com/neondatabase/postgres).
+- [neondatabase/neon](https://hub.docker.com/repository/docker/neondatabase/neon) — image with pre-built `pageserver`, `safekeeper` and `proxy` binaries and all the required runtime dependencies. Built from [/Dockerfile](/Dockerfile).
+- [neondatabase/compute-node-v16](https://hub.docker.com/repository/docker/neondatabase/compute-node-v16) — compute node image with pre-built Postgres binaries from [neondatabase/postgres](https://github.com/neondatabase/postgres). Similar images exist for v15 and v14.
 
 And additional intermediate image:
 
 - [neondatabase/compute-tools](https://hub.docker.com/repository/docker/neondatabase/compute-tools) — compute node configuration management tools.
 
-## Building pipeline
+## Build pipeline
 
 We build all images after a successful `release` tests run and push automatically to Docker Hub with two parallel CI jobs
 
-1. `neondatabase/compute-tools` and `neondatabase/compute-node`
+1. `neondatabase/compute-tools` and `neondatabase/compute-node-v16` (and -v15 and -v14)
 
 2. `neondatabase/neon`
 
@@ -34,12 +34,12 @@ You can see a [docker compose](https://docs.docker.com/compose/) example to crea
 1. create containers
 
 You can specify version of neon cluster using following environment values.
-- PG_VERSION: postgres version for compute (default is 14)
-- TAG: the tag version of [docker image](https://registry.hub.docker.com/r/neondatabase/neon/tags) (default is latest), which is tagged in [CI test](/.github/workflows/build_and_test.yml)
+- PG_VERSION: postgres version for compute (default is 16 as of this writing)
+- TAG: the tag version of [docker image](https://registry.hub.docker.com/r/neondatabase/neon/tags), which is tagged in [CI test](/.github/workflows/build_and_test.yml). Default is 'latest'
 ```
 $ cd docker-compose/
 $ docker-compose down   # remove the containers if exists
-$ PG_VERSION=15 TAG=2937 docker-compose up --build -d  # You can specify the postgres and image version
+$ PG_VERSION=16 TAG=latest docker-compose up --build -d  # You can specify the postgres and image version
 Creating network "dockercompose_default" with the default driver
 Creating docker-compose_storage_broker_1       ... done
 (...omit...)
@@ -47,29 +47,31 @@ Creating docker-compose_storage_broker_1       ... done
 
 2. connect compute node
 ```
-$ echo "localhost:55433:postgres:cloud_admin:cloud_admin" >> ~/.pgpass
-$ chmod 600 ~/.pgpass
-$ psql -h localhost -p 55433 -U cloud_admin
+$ psql postgresql://cloud_admin:cloud_admin@localhost:55433/postgres
+psql (16.3)
+Type "help" for help.
+
 postgres=# CREATE TABLE t(key int primary key, value text);
 CREATE TABLE
-postgres=# insert into t values(1,1);
+postgres=# insert into t values(1, 1);
 INSERT 0 1
 postgres=# select * from t;
- key | value
+ key | value 
 -----+-------
    1 | 1
 (1 row)
+
 ```
 
 3. If you want to see the log, you can use `docker-compose logs` command.
 ```
 # check the container name you want to see
 $ docker ps
-CONTAINER ID   IMAGE                                              COMMAND                  CREATED         STATUS         PORTS                                                                                                                                  NAMES
-d6968a5ae912   dockercompose_compute                              "/shell/compute.sh"      5 minutes ago   Up 5 minutes   0.0.0.0:3080->3080/tcp, 0.0.0.0:55433->55433/tcp                                                                                       dockercompose_compute_1
+CONTAINER ID   IMAGE                                              COMMAND                  CREATED         STATUS         PORTS                                                                                      NAMES
+3582f6d76227   docker-compose_compute                             "/shell/compute.sh"      2 minutes ago   Up 2 minutes   0.0.0.0:3080->3080/tcp, :::3080->3080/tcp, 0.0.0.0:55433->55433/tcp, :::55433->55433/tcp   docker-compose_compute_1
 (...omit...)
 
-$ docker logs -f dockercompose_compute_1
+$ docker logs -f docker-compose_compute_1
 2022-10-21 06:15:48.757 GMT [56] LOG:  connection authorized: user=cloud_admin database=postgres application_name=psql
 2022-10-21 06:17:00.307 GMT [56] LOG:  [NEON_SMGR] libpagestore: connected to 'host=pageserver port=6400'
 (...omit...)

docs/pageserver-pagecache.md

@@ -5,4 +5,3 @@ TODO:
 - shared across tenants
 - store pages from layer files
 - store pages from "in-memory layer"
-- store materialized pages

docs/pageserver-services.md

@@ -101,11 +101,12 @@ or
 ```toml
 [remote_storage]
 container_name = 'some-container-name'
+storage_account = 'somestorageaccnt'
 container_region = 'us-east'
 prefix_in_container = '/test-prefix/'
 ```
 
-`AZURE_STORAGE_ACCOUNT` and `AZURE_STORAGE_ACCESS_KEY` env variables can be used to specify the azure credentials if needed.
+The `AZURE_STORAGE_ACCESS_KEY` env variable can be used to specify the azure credentials if needed.
 
 ## Repository background tasks
 

docs/settings.md

@@ -134,7 +134,7 @@ depends on that, so if you change it, bad things will happen.
 
 #### page_cache_size
 
-Size of the page cache, to hold materialized page versions. Unit is
+Size of the page cache. Unit is
 number of 8 kB blocks. The default is 8192, which means 64 MB.
 
 #### max_file_descriptors

libs/pageserver_api/src/controller_api.rs

@@ -209,6 +209,7 @@ pub enum NodeSchedulingPolicy {
     Active,
     Filling,
     Pause,
+    PauseForRestart,
     Draining,
 }
 
@@ -220,6 +221,7 @@ impl FromStr for NodeSchedulingPolicy {
             "active" => Ok(Self::Active),
             "filling" => Ok(Self::Filling),
             "pause" => Ok(Self::Pause),
+            "pause_for_restart" => Ok(Self::PauseForRestart),
             "draining" => Ok(Self::Draining),
             _ => Err(anyhow::anyhow!("Unknown scheduling state '{s}'")),
         }
@@ -233,6 +235,7 @@ impl From<NodeSchedulingPolicy> for String {
             Active => "active",
             Filling => "filling",
             Pause => "pause",
+            PauseForRestart => "pause_for_restart",
             Draining => "draining",
         }
         .to_string()

libs/pageserver_api/src/keyspace.rs

@@ -558,6 +558,12 @@ impl KeySpaceRandomAccum {
         self.ranges.push(range);
     }
 
+    pub fn add_keyspace(&mut self, keyspace: KeySpace) {
+        for range in keyspace.ranges {
+            self.add_range(range);
+        }
+    }
+
     pub fn to_keyspace(mut self) -> KeySpace {
         let mut ranges = Vec::new();
         if !self.ranges.is_empty() {

libs/pageserver_api/src/models.rs

@@ -177,6 +177,20 @@ serde_with::serde_conv!(
     |value: String| -> Result<_, humantime::TimestampError> { humantime::parse_rfc3339(&value) }
 );
 
+impl LsnLease {
+    /// The default length for an explicit LSN lease request (10 minutes).
+    pub const DEFAULT_LENGTH: Duration = Duration::from_secs(10 * 60);
+
+    /// The default length for an implicit LSN lease granted during
+    /// `get_lsn_by_timestamp` request (1 minutes).
+    pub const DEFAULT_LENGTH_FOR_TS: Duration = Duration::from_secs(60);
+
+    /// Checks whether the lease is expired.
+    pub fn is_expired(&self, now: &SystemTime) -> bool {
+        now > &self.valid_until
+    }
+}
+
 /// The only [`TenantState`] variants we could be `TenantState::Activating` from.
 #[derive(Clone, Copy, Debug, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
 pub enum ActivatingFrom {
@@ -322,6 +336,8 @@ pub struct TenantConfig {
     pub timeline_get_throttle: Option<ThrottleConfig>,
     pub image_layer_creation_check_threshold: Option<u8>,
     pub switch_aux_file_policy: Option<AuxFilePolicy>,
+    pub lsn_lease_length: Option<String>,
+    pub lsn_lease_length_for_ts: Option<String>,
 }
 
 /// The policy for the aux file storage. It can be switched through `switch_aux_file_policy`

libs/remote_storage/src/azure_blob.rs

@@ -54,7 +54,10 @@ impl AzureBlobStorage {
             azure_config.container_name
         );
 
-        let account = env::var("AZURE_STORAGE_ACCOUNT").expect("missing AZURE_STORAGE_ACCOUNT");
+        // Use the storage account from the config by default, fall back to env var if not present.
+        let account = azure_config.storage_account.clone().unwrap_or_else(|| {
+            env::var("AZURE_STORAGE_ACCOUNT").expect("missing AZURE_STORAGE_ACCOUNT")
+        });
 
         // If the `AZURE_STORAGE_ACCESS_KEY` env var has an access key, use that,
         // otherwise try the token based credentials.

libs/remote_storage/src/lib.rs

@@ -466,7 +466,11 @@ impl GenericRemoteStorage {
                 Self::AwsS3(Arc::new(S3Bucket::new(s3_config, timeout)?))
             }
             RemoteStorageKind::AzureContainer(azure_config) => {
-                info!("Using azure container '{}' in region '{}' as a remote storage, prefix in container: '{:?}'",
+                let storage_account = azure_config
+                    .storage_account
+                    .as_deref()
+                    .unwrap_or("<AZURE_STORAGE_ACCOUNT>");
+                info!("Using azure container '{}' in account '{storage_account}' in region '{}' as a remote storage, prefix in container: '{:?}'",
                       azure_config.container_name, azure_config.container_region, azure_config.prefix_in_container);
                 Self::AzureBlob(Arc::new(AzureBlobStorage::new(azure_config, timeout)?))
             }
@@ -589,6 +593,8 @@ impl Debug for S3Config {
 pub struct AzureConfig {
     /// Name of the container to connect to.
     pub container_name: String,
+    /// Name of the storage account the container is inside of
+    pub storage_account: Option<String>,
     /// The region where the bucket is located at.
     pub container_region: String,
     /// A "subfolder" in the container, to use the same container separately by multiple remote storage users at once.
@@ -603,8 +609,9 @@ impl Debug for AzureConfig {
     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
         f.debug_struct("AzureConfig")
             .field("bucket_name", &self.container_name)
+            .field("storage_account", &self.storage_account)
             .field("bucket_region", &self.container_region)
-            .field("prefix_in_bucket", &self.prefix_in_container)
+            .field("prefix_in_container", &self.prefix_in_container)
             .field("concurrency_limit", &self.concurrency_limit)
             .field(
                 "max_keys_per_list_response",
@@ -718,6 +725,12 @@ impl RemoteStorageConfig {
             (None, None, None, Some(container_name), Some(container_region)) => {
                 RemoteStorageKind::AzureContainer(AzureConfig {
                     container_name: parse_toml_string("container_name", container_name)?,
+                    storage_account: toml
+                        .get("storage_account")
+                        .map(|storage_account| {
+                            parse_toml_string("storage_account", storage_account)
+                        })
+                        .transpose()?,
                     container_region: parse_toml_string("container_region", container_region)?,
                     prefix_in_container: toml
                         .get("prefix_in_container")

libs/remote_storage/tests/test_real_azure.rs

@@ -212,6 +212,7 @@ fn create_azure_client(
     let remote_storage_config = RemoteStorageConfig {
         storage: RemoteStorageKind::AzureContainer(AzureConfig {
             container_name: remote_storage_azure_container,
+            storage_account: None,
             container_region: remote_storage_azure_region,
             prefix_in_container: Some(format!("test_{millis}_{random:08x}/")),
             concurrency_limit: NonZeroUsize::new(100).unwrap(),

libs/tracing-utils/Cargo.toml

@@ -7,7 +7,7 @@ license.workspace = true
 [dependencies]
 hyper.workspace = true
 opentelemetry = { workspace = true, features=["rt-tokio"] }
-opentelemetry-otlp = { workspace = true, default_features=false, features = ["http-proto", "trace", "http", "reqwest-client"] }
+opentelemetry-otlp = { workspace = true, default-features=false, features = ["http-proto", "trace", "http", "reqwest-client"] }
 opentelemetry-semantic-conventions.workspace = true
 reqwest = { workspace = true, default-features = false, features = ["rustls-tls"] }
 tokio = { workspace = true, features = ["rt", "rt-multi-thread"] }

libs/utils/src/backoff.rs

@@ -37,6 +37,20 @@ pub fn exponential_backoff_duration_seconds(n: u32, base_increment: f64, max_sec
     }
 }
 
+pub trait Op<T, E> {
+    async fn call(&mut self) -> Result<T, E>;
+}
+
+impl<T, E, F, Fut> Op<T, E> for F
+where
+    F: FnMut() -> Fut,
+    Fut: Future<Output = Result<T, E>>,
+{
+    async fn call(&mut self) -> Result<T, E> {
+        (&mut *self)().await
+    }
+}
+
 /// Retries passed operation until one of the following conditions are met:
 /// - encountered error is considered as permanent (non-retryable)
 /// - retries have been exhausted
@@ -51,8 +65,8 @@ pub fn exponential_backoff_duration_seconds(n: u32, base_increment: f64, max_sec
 /// for any other error type. Final failed attempt is logged with `{:?}`.
 ///
 /// Returns `None` if cancellation was noticed during backoff or the terminal result.
-pub async fn retry<T, O, F, E>(
-    mut op: O,
+pub async fn retry<T, E>(
+    mut op: impl Op<T, E>,
     is_permanent: impl Fn(&E) -> bool,
     warn_threshold: u32,
     max_retries: u32,
@@ -63,8 +77,6 @@ where
     // Not std::error::Error because anyhow::Error doesnt implement it.
     // For context see https://github.com/dtolnay/anyhow/issues/63
     E: Display + Debug + 'static,
-    O: FnMut() -> F,
-    F: Future<Output = Result<T, E>>,
 {
     let mut attempts = 0;
     loop {
@@ -72,7 +84,7 @@ where
             return None;
         }
 
-        let result = op().await;
+        let result = op.call().await;
         match &result {
             Ok(_) => {
                 if attempts > 0 {

libs/vm_monitor/src/cgroup.rs

@@ -25,6 +25,8 @@ pub struct Config {
     ///
     /// For simplicity, this value must be greater than or equal to `memory_history_len`.
     memory_history_log_interval: usize,
+    /// The max number of iterations to skip before logging the next iteration
+    memory_history_log_noskip_interval: Duration,
 }
 
 impl Default for Config {
@@ -33,6 +35,7 @@ impl Default for Config {
             memory_poll_interval: Duration::from_millis(100),
             memory_history_len: 5, // use 500ms of history for decision-making
             memory_history_log_interval: 20, // but only log every ~2s (otherwise it's spammy)
+            memory_history_log_noskip_interval: Duration::from_secs(15), // but only if it's changed, or 60 seconds have passed
         }
     }
 }
@@ -85,7 +88,12 @@ impl CgroupWatcher {
 
         // buffer for samples that will be logged. once full, it remains so.
         let history_log_len = self.config.memory_history_log_interval;
+        let max_skip = self.config.memory_history_log_noskip_interval;
         let mut history_log_buf = vec![MemoryStatus::zeroed(); history_log_len];
+        let mut last_logged_memusage = MemoryStatus::zeroed();
+
+        // Ensure that we're tracking a value that's definitely in the past, as Instant::now is only guaranteed to be non-decreasing on Rust's T1-supported systems.
+        let mut can_skip_logs_until = Instant::now() - max_skip;
 
         for t in 0_u64.. {
             ticker.tick().await;
@@ -115,12 +123,24 @@ impl CgroupWatcher {
             // equal to the logging interval, we can just log the entire buffer every time we set
             // the last entry, which also means that for this log line, we can ignore that it's a
             // ring buffer (because all the entries are in order of increasing time).
-            if i == history_log_len - 1 {
+            //
+            // We skip logging the data if data hasn't meaningfully changed in a while, unless
+            // we've already ignored previous iterations for the last max_skip period.
+            if i == history_log_len - 1
+                && (now > can_skip_logs_until
+                    || !history_log_buf
+                        .iter()
+                        .all(|usage| last_logged_memusage.status_is_close_or_similar(usage)))
+            {
                 info!(
                     history = ?MemoryStatus::debug_slice(&history_log_buf),
                     summary = ?summary,
                     "Recent cgroup memory statistics history"
                 );
+
+                can_skip_logs_until = now + max_skip;
+
+                last_logged_memusage = *history_log_buf.last().unwrap();
             }
 
             updates
@@ -232,6 +252,24 @@ impl MemoryStatus {
 
         DS(slice)
     }
+
+    /// Check if the other memory status is a close or similar result.
+    /// Returns true if the larger value is not larger than the smaller value
+    /// by 1/8 of the smaller value, and within 128MiB.
+    /// See tests::check_similarity_behaviour for examples of behaviour
+    fn status_is_close_or_similar(&self, other: &MemoryStatus) -> bool {
+        let margin;
+        let diff;
+        if self.non_reclaimable >= other.non_reclaimable {
+            margin = other.non_reclaimable / 8;
+            diff = self.non_reclaimable - other.non_reclaimable;
+        } else {
+            margin = self.non_reclaimable / 8;
+            diff = other.non_reclaimable - self.non_reclaimable;
+        }
+
+        diff < margin && diff < 128 * 1024 * 1024
+    }
 }
 
 #[cfg(test)]
@@ -261,4 +299,65 @@ mod tests {
         assert_eq!(values(2, 4), [9, 0, 1, 2]);
         assert_eq!(values(2, 10), [3, 4, 5, 6, 7, 8, 9, 0, 1, 2]);
     }
+
+    #[test]
+    fn check_similarity_behaviour() {
+        // This all accesses private methods, so we can't actually run this
+        // as doctests, because doctests run as an external crate.
+        let mut small = super::MemoryStatus {
+            non_reclaimable: 1024,
+        };
+        let mut large = super::MemoryStatus {
+            non_reclaimable: 1024 * 1024 * 1024 * 1024,
+        };
+
+        // objects are self-similar, no matter the size
+        assert!(small.status_is_close_or_similar(&small));
+        assert!(large.status_is_close_or_similar(&large));
+
+        // inequality is symmetric
+        assert!(!small.status_is_close_or_similar(&large));
+        assert!(!large.status_is_close_or_similar(&small));
+
+        small.non_reclaimable = 64;
+        large.non_reclaimable = (small.non_reclaimable / 8) * 9;
+
+        // objects are self-similar, no matter the size
+        assert!(small.status_is_close_or_similar(&small));
+        assert!(large.status_is_close_or_similar(&large));
+
+        // values are similar if the larger value is larger by less than
+        // 12.5%, i.e. 1/8 of the smaller value.
+        // In the example above, large is exactly 12.5% larger, so this doesn't
+        // match.
+        assert!(!small.status_is_close_or_similar(&large));
+        assert!(!large.status_is_close_or_similar(&small));
+
+        large.non_reclaimable -= 1;
+        assert!(large.status_is_close_or_similar(&large));
+
+        assert!(small.status_is_close_or_similar(&large));
+        assert!(large.status_is_close_or_similar(&small));
+
+        // The 1/8 rule only applies up to 128MiB of difference
+        small.non_reclaimable = 1024 * 1024 * 1024 * 1024;
+        large.non_reclaimable = small.non_reclaimable / 8 * 9;
+        assert!(small.status_is_close_or_similar(&small));
+        assert!(large.status_is_close_or_similar(&large));
+
+        assert!(!small.status_is_close_or_similar(&large));
+        assert!(!large.status_is_close_or_similar(&small));
+        // the large value is put just above the threshold
+        large.non_reclaimable = small.non_reclaimable + 128 * 1024 * 1024;
+        assert!(large.status_is_close_or_similar(&large));
+
+        assert!(!small.status_is_close_or_similar(&large));
+        assert!(!large.status_is_close_or_similar(&small));
+        // now below
+        large.non_reclaimable -= 1;
+        assert!(large.status_is_close_or_similar(&large));
+
+        assert!(small.status_is_close_or_similar(&large));
+        assert!(large.status_is_close_or_similar(&small));
+    }
 }

libs/vm_monitor/src/dispatcher.rs

@@ -12,11 +12,11 @@ use futures::{
     stream::{SplitSink, SplitStream},
     SinkExt, StreamExt,
 };
-use tracing::info;
+use tracing::{debug, info};
 
 use crate::protocol::{
-    OutboundMsg, ProtocolRange, ProtocolResponse, ProtocolVersion, PROTOCOL_MAX_VERSION,
-    PROTOCOL_MIN_VERSION,
+    OutboundMsg, OutboundMsgKind, ProtocolRange, ProtocolResponse, ProtocolVersion,
+    PROTOCOL_MAX_VERSION, PROTOCOL_MIN_VERSION,
 };
 
 /// The central handler for all communications in the monitor.
@@ -118,7 +118,12 @@ impl Dispatcher {
     /// serialize the wrong thing and send it, since `self.sink.send` will take
     /// any string.
     pub async fn send(&mut self, message: OutboundMsg) -> anyhow::Result<()> {
-        info!(?message, "sending message");
+        if matches!(&message.inner, OutboundMsgKind::HealthCheck { .. }) {
+            debug!(?message, "sending message");
+        } else {
+            info!(?message, "sending message");
+        }
+
         let json = serde_json::to_string(&message).context("failed to serialize message")?;
         self.sink
             .send(Message::Text(json))

libs/vm_monitor/src/runner.rs

@@ -12,7 +12,7 @@ use axum::extract::ws::{Message, WebSocket};
 use futures::StreamExt;
 use tokio::sync::{broadcast, watch};
 use tokio_util::sync::CancellationToken;
-use tracing::{error, info, warn};
+use tracing::{debug, error, info, warn};
 
 use crate::cgroup::{self, CgroupWatcher};
 use crate::dispatcher::Dispatcher;
@@ -474,26 +474,29 @@ impl Runner {
                 // there is a message from the agent
                 msg = self.dispatcher.source.next() => {
                     if let Some(msg) = msg {
-                        // Don't use 'message' as a key as the string also uses
-                        // that for its key
-                        info!(?msg, "received message");
-                        match msg {
+                        match &msg {
                             Ok(msg) => {
                                 let message: InboundMsg = match msg {
                                     Message::Text(text) => {
-                                        serde_json::from_str(&text).context("failed to deserialize text message")?
+                                        serde_json::from_str(text).context("failed to deserialize text message")?
                                     }
                                     other => {
                                         warn!(
                                             // Don't use 'message' as a key as the
                                             // string also uses that for its key
                                             msg = ?other,
-                                            "agent should only send text messages but received different type"
+                                            "problem processing incoming message: agent should only send text messages but received different type"
                                         );
                                         continue
                                     },
                                 };
 
+                                if matches!(&message.inner, InboundMsgKind::HealthCheck { .. }) {
+                                    debug!(?msg, "received message");
+                                } else {
+                                    info!(?msg, "received message");
+                                }
+
                                 let out = match self.process_message(message.clone()).await {
                                     Ok(Some(out)) => out,
                                     Ok(None) => continue,
@@ -517,7 +520,11 @@ impl Runner {
                                     .await
                                     .context("failed to send message")?;
                             }
-                            Err(e) => warn!("{e}"),
+                            Err(e) => warn!(
+                                error = format!("{e}"),
+                                msg = ?msg,
+                                "received error message"
+                            ),
                         }
                     } else {
                         anyhow::bail!("dispatcher connection closed")

pageserver/ctl/src/index_part.rs

@@ -1,11 +1,6 @@
-use std::collections::HashMap;
-
 use anyhow::Context;
 use camino::Utf8PathBuf;
-use pageserver::tenant::remote_timeline_client::index::LayerFileMetadata;
-use pageserver::tenant::storage_layer::LayerName;
-use pageserver::tenant::{metadata::TimelineMetadata, IndexPart};
-use utils::lsn::Lsn;
+use pageserver::tenant::IndexPart;
 
 #[derive(clap::Subcommand)]
 pub(crate) enum IndexPartCmd {
@@ -17,20 +12,7 @@ pub(crate) async fn main(cmd: &IndexPartCmd) -> anyhow::Result<()> {
         IndexPartCmd::Dump { path } => {
             let bytes = tokio::fs::read(path).await.context("read file")?;
             let des: IndexPart = IndexPart::from_s3_bytes(&bytes).context("deserialize")?;
-            #[derive(serde::Serialize)]
-            struct Output<'a> {
-                layer_metadata: &'a HashMap<LayerName, LayerFileMetadata>,
-                disk_consistent_lsn: Lsn,
-                timeline_metadata: &'a TimelineMetadata,
-            }
-
-            let output = Output {
-                layer_metadata: &des.layer_metadata,
-                disk_consistent_lsn: des.metadata.disk_consistent_lsn(),
-                timeline_metadata: &des.metadata,
-            };
-
-            let output = serde_json::to_string_pretty(&output).context("serialize output")?;
+            let output = serde_json::to_string_pretty(&des).context("serialize output")?;
             println!("{output}");
             Ok(())
         }

pageserver/ctl/src/layer_map_analyzer.rs

@@ -99,7 +99,11 @@ pub(crate) fn parse_filename(name: &str) -> Option<LayerFile> {
 }
 
 // Finds the max_holes largest holes, ignoring any that are smaller than MIN_HOLE_LENGTH"
-async fn get_holes(path: &Utf8Path, max_holes: usize, ctx: &RequestContext) -> Result<Vec<Hole>> {
+async fn get_holes(
+    path: &Utf8Path,
+    max_holes: usize,
+    ctx: &mut RequestContext,
+) -> Result<Vec<Hole>> {
     let file = VirtualFile::open(path, ctx).await?;
     let file_id = page_cache::next_file_id();
     let block_reader = FileBlockReader::new(&file, file_id);

pageserver/ctl/src/layers.rs

@@ -57,7 +57,7 @@ pub(crate) enum LayerCmd {
     },
 }
 
-async fn read_delta_file(path: impl AsRef<Path>, ctx: &RequestContext) -> Result<()> {
+async fn read_delta_file(path: impl AsRef<Path>, ctx: &mut RequestContext) -> Result<()> {
     let path = Utf8Path::from_path(path.as_ref()).expect("non-Unicode path");
     virtual_file::init(10, virtual_file::api::IoEngineKind::StdFs);
     page_cache::init(100);

pageserver/src/basebackup.rs

@@ -60,7 +60,7 @@ pub async fn send_basebackup_tarball<'a, W>(
     req_lsn: Option<Lsn>,
     prev_lsn: Option<Lsn>,
     full_backup: bool,
-    ctx: &'a RequestContext,
+    ctx: &'a mut RequestContext,
 ) -> Result<(), BasebackupError>
 where
     W: AsyncWrite + Send + Sync + Unpin,
@@ -141,7 +141,7 @@ where
     lsn: Lsn,
     prev_record_lsn: Lsn,
     full_backup: bool,
-    ctx: &'a RequestContext,
+    ctx: &'a mut RequestContext,
 }
 
 /// A sink that accepts SLRU blocks ordered by key and forwards

pageserver/src/consumption_metrics.rs

@@ -2,10 +2,9 @@
 //! and push them to a HTTP endpoint.
 use crate::context::{DownloadBehavior, RequestContext};
 use crate::task_mgr::{self, TaskKind, BACKGROUND_RUNTIME};
+use crate::tenant::size::CalculateSyntheticSizeError;
 use crate::tenant::tasks::BackgroundLoopKind;
-use crate::tenant::{
-    mgr::TenantManager, LogicalSizeCalculationCause, PageReconstructError, Tenant,
-};
+use crate::tenant::{mgr::TenantManager, LogicalSizeCalculationCause, Tenant};
 use camino::Utf8PathBuf;
 use consumption_metrics::EventType;
 use pageserver_api::models::TenantState;
@@ -52,7 +51,7 @@ pub async fn collect_metrics(
     node_id: NodeId,
     local_disk_storage: Utf8PathBuf,
     cancel: CancellationToken,
-    ctx: RequestContext,
+    mut ctx: RequestContext,
 ) -> anyhow::Result<()> {
     if _cached_metric_collection_interval != Duration::ZERO {
         tracing::warn!(
@@ -61,7 +60,7 @@ pub async fn collect_metrics(
     }
 
     // spin up background worker that caclulates tenant sizes
-    let worker_ctx =
+    let mut worker_ctx =
         ctx.detached_child(TaskKind::CalculateSyntheticSize, DownloadBehavior::Download);
     task_mgr::spawn(
         BACKGROUND_RUNTIME.handle(),
@@ -77,7 +76,7 @@ pub async fn collect_metrics(
                     tenant_manager,
                     synthetic_size_calculation_interval,
                     &cancel,
-                    &worker_ctx,
+                    &mut worker_ctx,
                 )
                 .instrument(info_span!("synthetic_size_worker"))
                 .await?;
@@ -123,7 +122,8 @@ pub async fn collect_metrics(
         let started_at = Instant::now();
 
         // these are point in time, with variable "now"
-        let metrics = metrics::collect_all_metrics(&tenant_manager, &cached_metrics, &ctx).await;
+        let metrics =
+            metrics::collect_all_metrics(&tenant_manager, &cached_metrics, &mut ctx).await;
 
         let metrics = Arc::new(metrics);
 
@@ -281,7 +281,7 @@ async fn calculate_synthetic_size_worker(
     tenant_manager: Arc<TenantManager>,
     synthetic_size_calculation_interval: Duration,
     cancel: &CancellationToken,
-    ctx: &RequestContext,
+    ctx: &mut RequestContext,
 ) -> anyhow::Result<()> {
     info!("starting calculate_synthetic_size_worker");
     scopeguard::defer! {
@@ -341,7 +341,7 @@ async fn calculate_synthetic_size_worker(
     }
 }
 
-async fn calculate_and_log(tenant: &Tenant, cancel: &CancellationToken, ctx: &RequestContext) {
+async fn calculate_and_log(tenant: &Tenant, cancel: &CancellationToken, ctx: &mut RequestContext) {
     const CAUSE: LogicalSizeCalculationCause =
         LogicalSizeCalculationCause::ConsumptionMetricsSyntheticSize;
 
@@ -350,19 +350,12 @@ async fn calculate_and_log(tenant: &Tenant, cancel: &CancellationToken, ctx: &Re
     // Same for the loop that fetches computed metrics.
     // By using the same limiter, we centralize metrics collection for "start" and "finished" counters,
     // which turns out is really handy to understand the system.
-    let Err(e) = tenant.calculate_synthetic_size(CAUSE, cancel, ctx).await else {
-        return;
-    };
-
-    // this error can be returned if timeline is shutting down, but it does not
-    // mean the synthetic size worker should terminate.
-    let shutting_down = matches!(
-        e.downcast_ref::<PageReconstructError>(),
-        Some(PageReconstructError::Cancelled)
-    );
-
-    if !shutting_down {
-        let tenant_shard_id = tenant.tenant_shard_id();
-        error!("failed to calculate synthetic size for tenant {tenant_shard_id}: {e:#}");
+    match tenant.calculate_synthetic_size(CAUSE, cancel, ctx).await {
+        Ok(_) => {}
+        Err(CalculateSyntheticSizeError::Cancelled) => {}
+        Err(e) => {
+            let tenant_shard_id = tenant.tenant_shard_id();
+            error!("failed to calculate synthetic size for tenant {tenant_shard_id}: {e:#}");
+        }
     }
 }

pageserver/src/consumption_metrics/metrics.rs

@@ -184,7 +184,7 @@ impl MetricsKey {
 pub(super) async fn collect_all_metrics(
     tenant_manager: &Arc<TenantManager>,
     cached_metrics: &Cache,
-    ctx: &RequestContext,
+    ctx: &mut RequestContext,
 ) -> Vec<RawMetric> {
     use pageserver_api::models::TenantState;
 
@@ -220,7 +220,7 @@ pub(super) async fn collect_all_metrics(
     res
 }
 
-async fn collect<S>(tenants: S, cache: &Cache, ctx: &RequestContext) -> Vec<RawMetric>
+async fn collect<S>(tenants: S, cache: &Cache, ctx: &mut RequestContext) -> Vec<RawMetric>
 where
     S: futures::stream::Stream<Item = (TenantId, Arc<crate::tenant::Tenant>)>,
 {
@@ -342,7 +342,7 @@ impl TimelineSnapshot {
     /// [`Timeline::get_current_logical_size`]: crate::tenant::Timeline::get_current_logical_size
     fn collect(
         t: &Arc<crate::tenant::Timeline>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<Option<Self>> {
         if !t.is_active() {
             // no collection for broken or stopping needed, we will still keep the cached values

pageserver/src/context.rs

@@ -97,9 +97,10 @@ pub struct RequestContext {
     download_behavior: DownloadBehavior,
     access_stats_behavior: AccessStatsBehavior,
     page_content_kind: PageContentKind,
-    pub micros_spent_throttled: optional_counter::MicroSecondsCounterU32,
 }
 
+pub(crate) struct MicrosSpentThrottled(optional_counter::MicroSecondsCounterU32);
+
 /// The kind of access to the page cache.
 #[derive(Clone, Copy, PartialEq, Eq, Debug, enum_map::Enum, strum_macros::IntoStaticStr)]
 pub enum PageContentKind {
@@ -158,7 +159,7 @@ impl RequestContextBuilder {
         }
     }
 
-    pub fn extend(original: &RequestContext) -> Self {
+    pub fn extend(original: &mut RequestContext) -> Self {
         Self {
             // This is like a Copy, but avoid implementing Copy because ordinary users of
             // RequestContext should always move or ref it.

pageserver/src/http/routes.rs

@@ -352,7 +352,7 @@ async fn build_timeline_info(
     timeline: &Arc<Timeline>,
     include_non_incremental_logical_size: bool,
     force_await_initial_logical_size: bool,
-    ctx: &RequestContext,
+    ctx: &mut RequestContext,
 ) -> anyhow::Result<TimelineInfo> {
     crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
 
@@ -381,7 +381,7 @@ async fn build_timeline_info(
 
 async fn build_timeline_info_common(
     timeline: &Arc<Timeline>,
-    ctx: &RequestContext,
+    ctx: &mut RequestContext,
     logical_size_task_priority: tenant::timeline::GetLogicalSizePriority,
 ) -> anyhow::Result<TimelineInfo> {
     crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
@@ -502,7 +502,7 @@ async fn timeline_create_handler(
 
     let new_timeline_id = request_data.new_timeline_id;
 
-    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Error);
+    let mut ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Error);
 
     let state = get_state(&request);
 
@@ -527,7 +527,7 @@ async fn timeline_create_handler(
                 request_data.pg_version.unwrap_or(crate::DEFAULT_PG_VERSION),
                 request_data.existing_initdb_timeline_id,
                 state.broker_client.clone(),
-                &ctx,
+                &mut ctx,
             )
             .await
         {
@@ -535,7 +535,7 @@ async fn timeline_create_handler(
                 // Created. Construct a TimelineInfo for it.
                 let timeline_info = build_timeline_info_common(
                     &new_timeline,
-                    &ctx,
+                    &mut ctx,
                     tenant::timeline::GetLogicalSizePriority::User,
                 )
                 .await
@@ -593,7 +593,7 @@ async fn timeline_list_handler(
     check_permission(&request, Some(tenant_shard_id.tenant_id))?;
 
     let state = get_state(&request);
-    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
+    let mut ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
 
     let response_data = async {
         let tenant = state
@@ -610,7 +610,7 @@ async fn timeline_list_handler(
                 &timeline,
                 include_non_incremental_logical_size.unwrap_or(false),
                 force_await_initial_logical_size.unwrap_or(false),
-                &ctx,
+                &mut ctx,
             )
             .instrument(info_span!("build_timeline_info", timeline_id = %timeline.timeline_id))
             .await
@@ -680,7 +680,7 @@ async fn timeline_detail_handler(
     check_permission(&request, Some(tenant_shard_id.tenant_id))?;
 
     // Logical size calculation needs downloading.
-    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
+    let mut ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
     let state = get_state(&request);
 
     let timeline_info = async {
@@ -696,7 +696,7 @@ async fn timeline_detail_handler(
             &timeline,
             include_non_incremental_logical_size.unwrap_or(false),
             force_await_initial_logical_size.unwrap_or(false),
-            &ctx,
+            &mut ctx,
         )
         .await
         .context("get local timeline info")
@@ -735,13 +735,13 @@ async fn get_lsn_by_timestamp_handler(
         .map_err(ApiError::BadRequest)?;
     let timestamp_pg = postgres_ffi::to_pg_timestamp(timestamp);
 
-    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
+    let mut ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
 
     let timeline =
         active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
             .await?;
     let result = timeline
-        .find_lsn_for_timestamp(timestamp_pg, &cancel, &ctx)
+        .find_lsn_for_timestamp(timestamp_pg, &cancel, &mut ctx)
         .await?;
     #[derive(serde::Serialize, Debug)]
     struct Result {
@@ -786,11 +786,11 @@ async fn get_timestamp_of_lsn_handler(
         .with_context(|| format!("Invalid LSN: {lsn_str:?}"))
         .map_err(ApiError::BadRequest)?;
 
-    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
+    let  mut ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
     let timeline =
         active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
             .await?;
-    let result = timeline.get_timestamp_for_lsn(lsn, &ctx).await?;
+    let result = timeline.get_timestamp_for_lsn(lsn, &mut ctx).await?;
 
     match result {
         Some(time) => {
@@ -816,7 +816,7 @@ async fn tenant_attach_handler(
         None => TenantConfOpt::default(),
     };
 
-    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
+    let mut ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
 
     info!("Handling tenant attach {tenant_id}");
 
@@ -830,7 +830,7 @@ async fn tenant_attach_handler(
 
     let tenant = state
         .tenant_manager
-        .upsert_location(tenant_shard_id, location_conf, None, SpawnMode::Eager, &ctx)
+        .upsert_location(tenant_shard_id, location_conf, None, SpawnMode::Eager, &mut ctx)
         .await?;
 
     let Some(tenant) = tenant else {
@@ -921,11 +921,11 @@ async fn tenant_reset_handler(
 
     let drop_cache: Option<bool> = parse_query_param(&request, "drop_cache")?;
 
-    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
+    let mut ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
     let state = get_state(&request);
     state
         .tenant_manager
-        .reset_tenant(tenant_shard_id, drop_cache.unwrap_or(false), &ctx)
+        .reset_tenant(tenant_shard_id, drop_cache.unwrap_or(false), &mut ctx)
         .await
         .map_err(ApiError::InternalServerError)?;
 
@@ -939,7 +939,7 @@ async fn tenant_load_handler(
     let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
     check_permission(&request, Some(tenant_id))?;
 
-    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
+    let mut ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
 
     let maybe_body: Option<TenantLoadRequest> = json_request_or_empty_body(&mut request).await?;
 
@@ -956,7 +956,7 @@ async fn tenant_load_handler(
         state.broker_client.clone(),
         state.remote_storage.clone(),
         state.deletion_queue_client.clone(),
-        &ctx,
+        &mut ctx,
     )
     .instrument(info_span!("load", %tenant_id))
     .await?;
@@ -1120,7 +1120,7 @@ async fn tenant_size_handler(
         )));
     }
 
-    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
+    let mut ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
     let tenant = state
         .tenant_manager
         .get_attached_tenant_shard(tenant_shard_id)?;
@@ -1132,10 +1132,13 @@ async fn tenant_size_handler(
             retention_period,
             LogicalSizeCalculationCause::TenantSizeHandler,
             &cancel,
-            &ctx,
+            &mut ctx,
         )
         .await
-        .map_err(ApiError::InternalServerError)?;
+        .map_err(|e| match e {
+            crate::tenant::size::CalculateSyntheticSizeError::Cancelled => ApiError::ShuttingDown,
+            other => ApiError::InternalServerError(anyhow::anyhow!(other)),
+        })?;
 
     let mut sizes = None;
     let accepts_html = headers
@@ -1143,9 +1146,7 @@ async fn tenant_size_handler(
         .map(|v| v == "text/html")
         .unwrap_or_default();
     if !inputs_only.unwrap_or(false) {
-        let storage_model = inputs
-            .calculate_model()
-            .map_err(ApiError::InternalServerError)?;
+        let storage_model = inputs.calculate_model();
         let size = storage_model.calculate();
 
         // If request header expects html, return html
@@ -1192,7 +1193,7 @@ async fn tenant_shard_split_handler(
 
     let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
     let state = get_state(&request);
-    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
+    let mut ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
 
     let tenant = state
         .tenant_manager
@@ -1205,7 +1206,7 @@ async fn tenant_shard_split_handler(
             tenant,
             ShardCount::new(req.new_shard_count),
             req.new_stripe_size,
-            &ctx,
+            &mut ctx,
         )
         .await
         .map_err(ApiError::InternalServerError)?;
@@ -1385,7 +1386,7 @@ async fn tenant_create_handler(
 
     let generation = get_request_generation(state, request_data.generation)?;
 
-    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
+    let  mut ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
 
     let location_conf =
         LocationConf::attached_single(tenant_conf, generation, &request_data.shard_parameters);
@@ -1397,7 +1398,7 @@ async fn tenant_create_handler(
             location_conf,
             None,
             SpawnMode::Create,
-            &ctx,
+            &mut ctx,
         )
         .await?;
 
@@ -1497,7 +1498,7 @@ async fn put_tenant_location_config_handler(
     let lazy = parse_query_param(&request, "lazy")?.unwrap_or(false);
     check_permission(&request, Some(tenant_shard_id.tenant_id))?;
 
-    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
+    let mut ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
     let state = get_state(&request);
     let conf = state.conf;
 
@@ -1536,7 +1537,7 @@ async fn put_tenant_location_config_handler(
 
     let tenant = state
         .tenant_manager
-        .upsert_location(tenant_shard_id, location_conf, flush, spawn_mode, &ctx)
+        .upsert_location(tenant_shard_id, location_conf, flush, spawn_mode, &mut ctx)
         .await?;
     let stripe_size = tenant.as_ref().map(|t| t.get_shard_stripe_size());
     let attached = tenant.is_some();
@@ -1721,7 +1722,7 @@ async fn lsn_lease_handler(
     let lsn: Lsn = parse_query_param(&request, "lsn")?
         .ok_or_else(|| ApiError::BadRequest(anyhow!("missing 'lsn' query parameter")))?;
 
-    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
+    let mut ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
 
     let state = get_state(&request);
 
@@ -1729,7 +1730,7 @@ async fn lsn_lease_handler(
         active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
             .await?;
     let result = timeline
-        .make_lsn_lease(lsn, &ctx)
+        .make_lsn_lease(lsn, timeline.get_lsn_lease_length(), &mut ctx)
         .map_err(|e| ApiError::InternalServerError(e.context("lsn lease http handler")))?;
 
     json_response(StatusCode::OK, result)
@@ -1746,8 +1747,8 @@ async fn timeline_gc_handler(
 
     let gc_req: TimelineGcRequest = json_request(&mut request).await?;
 
-    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
-    let gc_result = mgr::immediate_gc(tenant_shard_id, timeline_id, gc_req, cancel, &ctx).await?;
+    let mut ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
+    let gc_result = mgr::immediate_gc(tenant_shard_id, timeline_id, gc_req, cancel, &mut ctx).await?;
 
     json_response(StatusCode::OK, gc_result)
 }
@@ -1774,10 +1775,10 @@ async fn timeline_compact_handler(
         parse_query_param::<_, bool>(&request, "wait_until_uploaded")?.unwrap_or(false);
 
     async {
-        let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
+        let mut ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
         let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
         timeline
-            .compact(&cancel, flags, &ctx)
+            .compact(&cancel, flags, &mut ctx)
             .await
             .map_err(|e| ApiError::InternalServerError(e.into()))?;
         if wait_until_uploaded {
@@ -1811,7 +1812,7 @@ async fn timeline_checkpoint_handler(
         parse_query_param::<_, bool>(&request, "wait_until_uploaded")?.unwrap_or(false);
 
     async {
-        let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
+        let mut ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
         let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
         timeline
             .freeze_and_flush()
@@ -1824,7 +1825,7 @@ async fn timeline_checkpoint_handler(
                 }
             })?;
         timeline
-            .compact(&cancel, flags, &ctx)
+            .compact(&cancel, flags, &mut ctx)
             .await
             .map_err(|e|
                 match e {
@@ -1917,8 +1918,8 @@ async fn timeline_detach_ancestor_handler(
 
         tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
 
-        let ctx = RequestContext::new(TaskKind::DetachAncestor, DownloadBehavior::Download);
-        let ctx = &ctx;
+        let mut ctx = RequestContext::new(TaskKind::DetachAncestor, DownloadBehavior::Download);
+        let ctx = &mut ctx;
 
         let timeline = tenant.get_timeline(timeline_id, true)?;
 
@@ -2002,10 +2003,10 @@ async fn getpage_at_lsn_handler(
         .ok_or_else(|| ApiError::BadRequest(anyhow!("missing 'lsn' query parameter")))?;
 
     async {
-        let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
+        let mut ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
         let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
 
-        let page = timeline.get(key.0, lsn, &ctx).await?;
+        let page = timeline.get(key.0, lsn, &mut ctx).await?;
 
         Result::<_, ApiError>::Ok(
             Response::builder()
@@ -2031,11 +2032,11 @@ async fn timeline_collect_keyspace(
     let at_lsn: Option<Lsn> = parse_query_param(&request, "at_lsn")?;
 
     async {
-        let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
+        let mut ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
         let timeline = active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id).await?;
         let at_lsn = at_lsn.unwrap_or_else(|| timeline.get_last_record_lsn());
         let (dense_ks, sparse_ks) = timeline
-            .collect_keyspace(at_lsn, &ctx)
+            .collect_keyspace(at_lsn, &mut ctx)
             .await
             .map_err(|e| ApiError::InternalServerError(e.into()))?;
 
@@ -2424,11 +2425,30 @@ async fn list_aux_files(
         active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
             .await?;
 
-    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
-    let files = timeline.list_aux_files(body.lsn, &ctx).await?;
+    let mut ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
+    let files = timeline.list_aux_files(body.lsn, &mut ctx).await?;
     json_response(StatusCode::OK, files)
 }
 
+async fn perf_info(
+    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")?;
+    check_permission(&request, Some(tenant_shard_id.tenant_id))?;
+
+    let state = get_state(&request);
+
+    let timeline =
+        active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
+            .await?;
+
+    let result = timeline.perf_info().await;
+
+    json_response(StatusCode::OK, result)
+}
+
 async fn ingest_aux_files(
     mut request: Request<Body>,
     _cancel: CancellationToken,
@@ -2447,15 +2467,15 @@ async fn ingest_aux_files(
     let mut modification = timeline.begin_modification(
         Lsn(timeline.get_last_record_lsn().0 + 8), /* advance LSN by 8 */
     );
-    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
+    let mut ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
     for (fname, content) in body.aux_files {
         modification
-            .put_file(&fname, content.as_bytes(), &ctx)
+            .put_file(&fname, content.as_bytes(), &mut ctx)
             .await
             .map_err(ApiError::InternalServerError)?;
     }
     modification
-        .commit(&ctx)
+        .commit(&mut ctx)
         .await
         .map_err(ApiError::InternalServerError)?;
 
@@ -2856,5 +2876,9 @@ pub fn make_router(
             |r| testing_api_handler("list_aux_files", r, list_aux_files),
         )
         .post("/v1/top_tenants", |r| api_handler(r, post_top_tenants))
+        .post(
+            "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/perf_info",
+            |r| testing_api_handler("perf_info", r, perf_info),
+        )
         .any(handler_404))
 }

pageserver/src/import_datadir.rs

@@ -53,7 +53,7 @@ pub async fn import_timeline_from_postgres_datadir(
     tline: &Timeline,
     pgdata_path: &Utf8Path,
     pgdata_lsn: Lsn,
-    ctx: &RequestContext,
+    ctx: &mut RequestContext,
 ) -> Result<()> {
     let mut pg_control: Option<ControlFileData> = None;
 
@@ -121,7 +121,7 @@ async fn import_rel(
     dboid: Oid,
     reader: &mut (impl AsyncRead + Unpin),
     len: usize,
-    ctx: &RequestContext,
+    ctx: &mut RequestContext,
 ) -> anyhow::Result<()> {
     // Does it look like a relation file?
     trace!("importing rel file {}", path.display());
@@ -210,7 +210,7 @@ async fn import_slru(
     path: &Path,
     reader: &mut (impl AsyncRead + Unpin),
     len: usize,
-    ctx: &RequestContext,
+    ctx: &mut RequestContext,
 ) -> anyhow::Result<()> {
     info!("importing slru file {path:?}");
 
@@ -268,7 +268,7 @@ async fn import_wal(
     tline: &Timeline,
     startpoint: Lsn,
     endpoint: Lsn,
-    ctx: &RequestContext,
+    ctx: &mut RequestContext,
 ) -> anyhow::Result<()> {
     let mut waldecoder = WalStreamDecoder::new(startpoint, tline.pg_version);
 
@@ -346,7 +346,7 @@ pub async fn import_basebackup_from_tar(
     tline: &Timeline,
     reader: &mut (impl AsyncRead + Send + Sync + Unpin),
     base_lsn: Lsn,
-    ctx: &RequestContext,
+    ctx: &mut RequestContext,
 ) -> Result<()> {
     info!("importing base at {base_lsn}");
     let mut modification = tline.begin_modification(base_lsn);
@@ -397,7 +397,7 @@ pub async fn import_wal_from_tar(
     reader: &mut (impl AsyncRead + Send + Sync + Unpin),
     start_lsn: Lsn,
     end_lsn: Lsn,
-    ctx: &RequestContext,
+    ctx: &mut RequestContext,
 ) -> Result<()> {
     // Set up walingest mutable state
     let mut waldecoder = WalStreamDecoder::new(start_lsn, tline.pg_version);
@@ -489,7 +489,7 @@ async fn import_file(
     file_path: &Path,
     reader: &mut (impl AsyncRead + Send + Sync + Unpin),
     len: usize,
-    ctx: &RequestContext,
+    ctx: &mut RequestContext,
 ) -> Result<Option<ControlFileData>> {
     let file_name = match file_path.file_name() {
         Some(name) => name.to_string_lossy(),

pageserver/src/metrics.rs

@@ -145,14 +145,6 @@ impl ReconstructTimeMetrics {
     }
 }
 
-pub(crate) static MATERIALIZED_PAGE_CACHE_HIT_DIRECT: Lazy<IntCounter> = Lazy::new(|| {
-    register_int_counter!(
-        "pageserver_materialized_cache_hits_direct_total",
-        "Number of cache hits from materialized page cache without redo",
-    )
-    .expect("failed to define a metric")
-});
-
 pub(crate) struct ReconstructDataTimeMetrics {
     singular: Histogram,
     vectored: Histogram,
@@ -182,14 +174,6 @@ pub(crate) static GET_RECONSTRUCT_DATA_TIME: Lazy<ReconstructDataTimeMetrics> =
     }
 });
 
-pub(crate) static MATERIALIZED_PAGE_CACHE_HIT: Lazy<IntCounter> = Lazy::new(|| {
-    register_int_counter!(
-        "pageserver_materialized_cache_hits_total",
-        "Number of cache hits from materialized page cache",
-    )
-    .expect("failed to define a metric")
-});
-
 pub(crate) struct GetVectoredLatency {
     map: EnumMap<TaskKind, Option<Histogram>>,
 }
@@ -298,12 +282,8 @@ pub(crate) static SCAN_LATENCY: Lazy<ScanLatency> = Lazy::new(|| {
 });
 
 pub(crate) struct PageCacheMetricsForTaskKind {
-    pub read_accesses_materialized_page: IntCounter,
     pub read_accesses_immutable: IntCounter,
-
     pub read_hits_immutable: IntCounter,
-    pub read_hits_materialized_page_exact: IntCounter,
-    pub read_hits_materialized_page_older_lsn: IntCounter,
 }
 
 pub(crate) struct PageCacheMetrics {
@@ -336,16 +316,6 @@ pub(crate) static PAGE_CACHE: Lazy<PageCacheMetrics> = Lazy::new(|| PageCacheMet
             let content_kind = <PageContentKind as enum_map::Enum>::from_usize(content_kind);
             let content_kind: &'static str = content_kind.into();
             PageCacheMetricsForTaskKind {
-                read_accesses_materialized_page: {
-                    PAGE_CACHE_READ_ACCESSES
-                        .get_metric_with_label_values(&[
-                            task_kind,
-                            "materialized_page",
-                            content_kind,
-                        ])
-                        .unwrap()
-                },
-
                 read_accesses_immutable: {
                     PAGE_CACHE_READ_ACCESSES
                         .get_metric_with_label_values(&[task_kind, "immutable", content_kind])
@@ -357,35 +327,13 @@ pub(crate) static PAGE_CACHE: Lazy<PageCacheMetrics> = Lazy::new(|| PageCacheMet
                         .get_metric_with_label_values(&[task_kind, "immutable", content_kind, "-"])
                         .unwrap()
                 },
-
-                read_hits_materialized_page_exact: {
-                    PAGE_CACHE_READ_HITS
-                        .get_metric_with_label_values(&[
-                            task_kind,
-                            "materialized_page",
-                            content_kind,
-                            "exact",
-                        ])
-                        .unwrap()
-                },
-
-                read_hits_materialized_page_older_lsn: {
-                    PAGE_CACHE_READ_HITS
-                        .get_metric_with_label_values(&[
-                            task_kind,
-                            "materialized_page",
-                            content_kind,
-                            "older_lsn",
-                        ])
-                        .unwrap()
-                },
             }
         }))
     })),
 });
 
 impl PageCacheMetrics {
-    pub(crate) fn for_ctx(&self, ctx: &RequestContext) -> &PageCacheMetricsForTaskKind {
+    pub(crate) fn for_ctx(&self, ctx: &mut RequestContext) -> &PageCacheMetricsForTaskKind {
         &self.map[ctx.task_kind()][ctx.page_content_kind()]
     }
 }
@@ -394,7 +342,6 @@ pub(crate) struct PageCacheSizeMetrics {
     pub max_bytes: UIntGauge,
 
     pub current_bytes_immutable: UIntGauge,
-    pub current_bytes_materialized_page: UIntGauge,
 }
 
 static PAGE_CACHE_SIZE_CURRENT_BYTES: Lazy<UIntGaugeVec> = Lazy::new(|| {
@@ -420,11 +367,6 @@ pub(crate) static PAGE_CACHE_SIZE: Lazy<PageCacheSizeMetrics> =
                 .get_metric_with_label_values(&["immutable"])
                 .unwrap()
         },
-        current_bytes_materialized_page: {
-            PAGE_CACHE_SIZE_CURRENT_BYTES
-                .get_metric_with_label_values(&["materialized_page"])
-                .unwrap()
-        },
     });
 
 pub(crate) mod page_cache_eviction_metrics {
@@ -1143,7 +1085,6 @@ impl GlobalAndPerTimelineHistogram {
 
 struct GlobalAndPerTimelineHistogramTimer<'a, 'c> {
     h: &'a GlobalAndPerTimelineHistogram,
-    ctx: &'c RequestContext,
     start: std::time::Instant,
     op: SmgrQueryType,
 }
@@ -1151,32 +1092,11 @@ struct GlobalAndPerTimelineHistogramTimer<'a, 'c> {
 impl<'a, 'c> Drop for GlobalAndPerTimelineHistogramTimer<'a, 'c> {
     fn drop(&mut self) {
         let elapsed = self.start.elapsed();
-        let ex_throttled = self
-            .ctx
-            .micros_spent_throttled
-            .close_and_checked_sub_from(elapsed);
-        let ex_throttled = match ex_throttled {
-            Ok(res) => res,
-            Err(error) => {
-                use utils::rate_limit::RateLimit;
-                static LOGGED: Lazy<Mutex<enum_map::EnumMap<SmgrQueryType, RateLimit>>> =
-                    Lazy::new(|| {
-                        Mutex::new(enum_map::EnumMap::from_array(std::array::from_fn(|_| {
-                            RateLimit::new(Duration::from_secs(10))
-                        })))
-                    });
-                let mut guard = LOGGED.lock().unwrap();
-                let rate_limit = &mut guard[self.op];
-                rate_limit.call(|| {
-                    warn!(op=?self.op, error, "error deducting time spent throttled; this message is logged at a global rate limit");
-                });
-                elapsed
-            }
-        };
-        self.h.observe(ex_throttled.as_secs_f64());
+        self.h.observe(elapsed.as_secs_f64());
     }
 }
 
+
 #[derive(
     Debug,
     Clone,
@@ -1291,33 +1211,11 @@ impl SmgrQueryTimePerTimeline {
         });
         Self { metrics }
     }
-    pub(crate) fn start_timer<'c: 'a, 'a>(
-        &'a self,
-        op: SmgrQueryType,
-        ctx: &'c RequestContext,
-    ) -> impl Drop + '_ {
+    pub(crate) fn start_timer<'a>(&'a self, op: SmgrQueryType) -> impl Drop + 'a {
         let metric = &self.metrics[op as usize];
         let start = Instant::now();
-        match ctx.micros_spent_throttled.open() {
-            Ok(()) => (),
-            Err(error) => {
-                use utils::rate_limit::RateLimit;
-                static LOGGED: Lazy<Mutex<enum_map::EnumMap<SmgrQueryType, RateLimit>>> =
-                    Lazy::new(|| {
-                        Mutex::new(enum_map::EnumMap::from_array(std::array::from_fn(|_| {
-                            RateLimit::new(Duration::from_secs(10))
-                        })))
-                    });
-                let mut guard = LOGGED.lock().unwrap();
-                let rate_limit = &mut guard[op];
-                rate_limit.call(|| {
-                    warn!(?op, error, "error opening micros_spent_throttled; this message is logged at a global rate limit");
-                });
-            }
-        }
         GlobalAndPerTimelineHistogramTimer {
             h: metric,
-            ctx,
             start,
             op,
         }
@@ -1384,7 +1282,7 @@ mod smgr_query_time_tests {
             assert_eq!(pre_per_tenant_timeline, 0);
 
             let ctx = RequestContext::new(TaskKind::UnitTest, DownloadBehavior::Download);
-            let timer = metrics.start_timer(*op, &ctx);
+            let timer = metrics.start_timer(*op);
             drop(timer);
 
             let (post_global, post_per_tenant_timeline) = get_counts();
@@ -1405,17 +1303,23 @@ static COMPUTE_STARTUP_BUCKETS: Lazy<[f64; 28]> = Lazy::new(|| {
     .map(|ms| (ms as f64) / 1000.0)
 });
 
-pub(crate) struct BasebackupQueryTime(HistogramVec);
+pub(crate) struct BasebackupQueryTime {
+    ok: Histogram,
+    error: Histogram,
+}
+
 pub(crate) static BASEBACKUP_QUERY_TIME: Lazy<BasebackupQueryTime> = Lazy::new(|| {
-    BasebackupQueryTime({
-        register_histogram_vec!(
-            "pageserver_basebackup_query_seconds",
-            "Histogram of basebackup queries durations, by result type",
-            &["result"],
-            COMPUTE_STARTUP_BUCKETS.to_vec(),
-        )
-        .expect("failed to define a metric")
-    })
+    let vec = register_histogram_vec!(
+        "pageserver_basebackup_query_seconds",
+        "Histogram of basebackup queries durations, by result type",
+        &["result"],
+        COMPUTE_STARTUP_BUCKETS.to_vec(),
+    )
+    .expect("failed to define a metric");
+    BasebackupQueryTime {
+        ok: vec.get_metric_with_label_values(&["ok"]).unwrap(),
+        error: vec.get_metric_with_label_values(&["error"]).unwrap(),
+    }
 });
 
 pub(crate) struct BasebackupQueryTimeOngoingRecording<'a, 'c> {
@@ -1470,12 +1374,11 @@ impl<'a, 'c> BasebackupQueryTimeOngoingRecording<'a, 'c> {
                 elapsed
             }
         };
-        let label_value = if res.is_ok() { "ok" } else { "error" };
-        let metric = self
-            .parent
-            .0
-            .get_metric_with_label_values(&[label_value])
-            .unwrap();
+        let metric = if res.is_ok() {
+            &self.parent.ok
+        } else {
+            &self.parent.error
+        };
         metric.observe(ex_throttled.as_secs_f64());
     }
 }
@@ -2918,13 +2821,11 @@ pub fn preinitialize_metrics() {
     // FIXME(4813): make it so that we have no top level metrics as this fn will easily fall out of
     // order:
     // - global metrics reside in a Lazy<PageserverMetrics>
-    //   - access via crate::metrics::PS_METRICS.materialized_page_cache_hit.inc()
+    //   - access via crate::metrics::PS_METRICS.some_metric.inc()
     // - could move the statics into TimelineMetrics::new()?
 
     // counters
     [
-        &MATERIALIZED_PAGE_CACHE_HIT,
-        &MATERIALIZED_PAGE_CACHE_HIT_DIRECT,
         &UNEXPECTED_ONDEMAND_DOWNLOADS,
         &WALRECEIVER_STARTED_CONNECTIONS,
         &WALRECEIVER_BROKER_UPDATES,
@@ -2986,4 +2887,5 @@ pub fn preinitialize_metrics() {
     // Custom
     Lazy::force(&RECONSTRUCT_TIME);
     Lazy::force(&tenant_throttling::TIMELINE_GET);
+    Lazy::force(&BASEBACKUP_QUERY_TIME);
 }

pageserver/src/page_cache.rs

@@ -17,7 +17,6 @@
 //!
 //! Two types of pages are supported:
 //!
-//! * **Materialized pages**, filled & used by page reconstruction
 //! * **Immutable File pages**, filled & used by [`crate::tenant::block_io`] and [`crate::tenant::ephemeral_file`].
 //!
 //! Note that [`crate::tenant::ephemeral_file::EphemeralFile`] is generally mutable, but, it's append-only.
@@ -28,9 +27,6 @@
 //! Page cache maps from a cache key to a buffer slot.
 //! The cache key uniquely identifies the piece of data that is being cached.
 //!
-//! The cache key for **materialized pages** is  [`TenantShardId`], [`TimelineId`], [`Key`], and [`Lsn`].
-//! Use [`PageCache::memorize_materialized_page`] and [`PageCache::lookup_materialized_page`] for fill & access.
-//!
 //! The cache key for **immutable file** pages is [`FileId`] and a block number.
 //! Users of page cache that wish to page-cache an arbitrary (immutable!) on-disk file do the following:
 //! * Have a mechanism to deterministically associate the on-disk file with a [`FileId`].
@@ -82,13 +78,10 @@ use std::{
 
 use anyhow::Context;
 use once_cell::sync::OnceCell;
-use pageserver_api::shard::TenantShardId;
-use utils::{id::TimelineId, lsn::Lsn};
 
 use crate::{
     context::RequestContext,
     metrics::{page_cache_eviction_metrics, PageCacheSizeMetrics},
-    repository::Key,
 };
 
 static PAGE_CACHE: OnceCell<PageCache> = OnceCell::new();
@@ -139,33 +132,7 @@ pub fn next_file_id() -> FileId {
 #[derive(Debug, PartialEq, Eq, Clone)]
 #[allow(clippy::enum_variant_names)]
 enum CacheKey {
-    MaterializedPage {
-        hash_key: MaterializedPageHashKey,
-        lsn: Lsn,
-    },
-    ImmutableFilePage {
-        file_id: FileId,
-        blkno: u32,
-    },
-}
-
-#[derive(Debug, PartialEq, Eq, Hash, Clone)]
-struct MaterializedPageHashKey {
-    /// Why is this TenantShardId rather than TenantId?
-    ///
-    /// Usually, the materialized value of a page@lsn is identical on any shard in the same tenant.  However, this
-    /// this not the case for certain internally-generated pages (e.g. relation sizes).  In future, we may make this
-    /// key smaller by omitting the shard, if we ensure that reads to such pages always skip the cache, or are
-    /// special-cased in some other way.
-    tenant_shard_id: TenantShardId,
-    timeline_id: TimelineId,
-    key: Key,
-}
-
-#[derive(Clone)]
-struct Version {
-    lsn: Lsn,
-    slot_idx: usize,
+    ImmutableFilePage { file_id: FileId, blkno: u32 },
 }
 
 struct Slot {
@@ -236,17 +203,6 @@ impl SlotInner {
 }
 
 pub struct PageCache {
-    /// This contains the mapping from the cache key to buffer slot that currently
-    /// contains the page, if any.
-    ///
-    /// TODO: This is protected by a single lock. If that becomes a bottleneck,
-    /// this HashMap can be replaced with a more concurrent version, there are
-    /// plenty of such crates around.
-    ///
-    /// If you add support for caching different kinds of objects, each object kind
-    /// can have a separate mapping map, next to this field.
-    materialized_page_map: std::sync::RwLock<HashMap<MaterializedPageHashKey, Vec<Version>>>,
-
     immutable_page_map: std::sync::RwLock<HashMap<(FileId, u32), usize>>,
 
     /// The actual buffers with their metadata.
@@ -371,175 +327,14 @@ pub enum ReadBufResult<'a> {
 }
 
 impl PageCache {
-    //
-    // Section 1.1: Public interface functions for looking up and memorizing materialized page
-    // versions in the page cache
-    //
-
-    /// Look up a materialized page version.
-    ///
-    /// The 'lsn' is an upper bound, this will return the latest version of
-    /// the given block, but not newer than 'lsn'. Returns the actual LSN of the
-    /// returned page.
-    pub async fn lookup_materialized_page(
-        &self,
-        tenant_shard_id: TenantShardId,
-        timeline_id: TimelineId,
-        key: &Key,
-        lsn: Lsn,
-        ctx: &RequestContext,
-    ) -> Option<(Lsn, PageReadGuard)> {
-        let Ok(permit) = self.try_get_pinned_slot_permit().await else {
-            return None;
-        };
-
-        crate::metrics::PAGE_CACHE
-            .for_ctx(ctx)
-            .read_accesses_materialized_page
-            .inc();
-
-        let mut cache_key = CacheKey::MaterializedPage {
-            hash_key: MaterializedPageHashKey {
-                tenant_shard_id,
-                timeline_id,
-                key: *key,
-            },
-            lsn,
-        };
-
-        if let Some(guard) = self
-            .try_lock_for_read(&mut cache_key, &mut Some(permit))
-            .await
-        {
-            if let CacheKey::MaterializedPage {
-                hash_key: _,
-                lsn: available_lsn,
-            } = cache_key
-            {
-                if available_lsn == lsn {
-                    crate::metrics::PAGE_CACHE
-                        .for_ctx(ctx)
-                        .read_hits_materialized_page_exact
-                        .inc();
-                } else {
-                    crate::metrics::PAGE_CACHE
-                        .for_ctx(ctx)
-                        .read_hits_materialized_page_older_lsn
-                        .inc();
-                }
-                Some((available_lsn, guard))
-            } else {
-                panic!("unexpected key type in slot");
-            }
-        } else {
-            None
-        }
-    }
-
-    ///
-    /// Store an image of the given page in the cache.
-    ///
-    pub async fn memorize_materialized_page(
-        &self,
-        tenant_shard_id: TenantShardId,
-        timeline_id: TimelineId,
-        key: Key,
-        lsn: Lsn,
-        img: &[u8],
-    ) -> anyhow::Result<()> {
-        let cache_key = CacheKey::MaterializedPage {
-            hash_key: MaterializedPageHashKey {
-                tenant_shard_id,
-                timeline_id,
-                key,
-            },
-            lsn,
-        };
-
-        let mut permit = Some(self.try_get_pinned_slot_permit().await?);
-        loop {
-            // First check if the key already exists in the cache.
-            if let Some(slot_idx) = self.search_mapping_exact(&cache_key) {
-                // The page was found in the mapping. Lock the slot, and re-check
-                // that it's still what we expected (because we don't released the mapping
-                // lock already, another thread could have evicted the page)
-                let slot = &self.slots[slot_idx];
-                let inner = slot.inner.write().await;
-                if inner.key.as_ref() == Some(&cache_key) {
-                    slot.inc_usage_count();
-                    debug_assert!(
-                        {
-                            let guard = inner.permit.lock().unwrap();
-                            guard.upgrade().is_none()
-                        },
-                        "we hold a write lock, so, no one else should have a permit"
-                    );
-                    debug_assert_eq!(inner.buf.len(), img.len());
-                    // We already had it in cache. Another thread must've put it there
-                    // concurrently. Check that it had the same contents that we
-                    // replayed.
-                    assert!(inner.buf == img);
-                    return Ok(());
-                }
-            }
-            debug_assert!(permit.is_some());
-
-            // Not found. Find a victim buffer
-            let (slot_idx, mut inner) = self
-                .find_victim(permit.as_ref().unwrap())
-                .await
-                .context("Failed to find evict victim")?;
-
-            // Insert mapping for this. At this point, we may find that another
-            // thread did the same thing concurrently. In that case, we evicted
-            // our victim buffer unnecessarily. Put it into the free list and
-            // continue with the slot that the other thread chose.
-            if let Some(_existing_slot_idx) = self.try_insert_mapping(&cache_key, slot_idx) {
-                // TODO: put to free list
-
-                // We now just loop back to start from beginning. This is not
-                // optimal, we'll perform the lookup in the mapping again, which
-                // is not really necessary because we already got
-                // 'existing_slot_idx'.  But this shouldn't happen often enough
-                // to matter much.
-                continue;
-            }
-
-            // Make the slot ready
-            let slot = &self.slots[slot_idx];
-            inner.key = Some(cache_key.clone());
-            slot.set_usage_count(1);
-            // Create a write guard for the slot so we go through the expected motions.
-            debug_assert!(
-                {
-                    let guard = inner.permit.lock().unwrap();
-                    guard.upgrade().is_none()
-                },
-                "we hold a write lock, so, no one else should have a permit"
-            );
-            let mut write_guard = PageWriteGuard {
-                state: PageWriteGuardState::Invalid {
-                    _permit: permit.take().unwrap(),
-                    inner,
-                },
-            };
-            write_guard.copy_from_slice(img);
-            let _ = write_guard.mark_valid();
-            return Ok(());
-        }
-    }
-
-    // Section 1.2: Public interface functions for working with immutable file pages.
-
     pub async fn read_immutable_buf(
         &self,
         file_id: FileId,
         blkno: u32,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<ReadBufResult> {
-        let mut cache_key = CacheKey::ImmutableFilePage { file_id, blkno };
-
-        self.lock_for_read(&mut cache_key, ctx).await
+        self.lock_for_read(&(CacheKey::ImmutableFilePage { file_id, blkno }), ctx)
+            .await
     }
 
     //
@@ -573,19 +368,11 @@ impl PageCache {
 
     /// Look up a page in the cache.
     ///
-    /// If the search criteria is not exact, *cache_key is updated with the key
-    /// for exact key of the returned page. (For materialized pages, that means
-    /// that the LSN in 'cache_key' is updated with the LSN of the returned page
-    /// version.)
-    ///
-    /// If no page is found, returns None and *cache_key is left unmodified.
-    ///
     async fn try_lock_for_read(
         &self,
-        cache_key: &mut CacheKey,
+        cache_key: &CacheKey,
         permit: &mut Option<PinnedSlotsPermit>,
     ) -> Option<PageReadGuard> {
-        let cache_key_orig = cache_key.clone();
         if let Some(slot_idx) = self.search_mapping(cache_key) {
             // The page was found in the mapping. Lock the slot, and re-check
             // that it's still what we expected (because we released the mapping
@@ -598,9 +385,6 @@ impl PageCache {
                     _permit: inner.coalesce_readers_permit(permit.take().unwrap()),
                     slot_guard: inner,
                 });
-            } else {
-                // search_mapping might have modified the search key; restore it.
-                *cache_key = cache_key_orig;
             }
         }
         None
@@ -637,15 +421,12 @@ impl PageCache {
     ///
     async fn lock_for_read(
         &self,
-        cache_key: &mut CacheKey,
-        ctx: &RequestContext,
+        cache_key: &CacheKey,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<ReadBufResult> {
         let mut permit = Some(self.try_get_pinned_slot_permit().await?);
 
         let (read_access, hit) = match cache_key {
-            CacheKey::MaterializedPage { .. } => {
-                unreachable!("Materialized pages use lookup_materialized_page")
-            }
             CacheKey::ImmutableFilePage { .. } => (
                 &crate::metrics::PAGE_CACHE
                     .for_ctx(ctx)
@@ -717,52 +498,15 @@ impl PageCache {
 
     /// Search for a page in the cache using the given search key.
     ///
-    /// Returns the slot index, if any. If the search criteria is not exact,
-    /// *cache_key is updated with the actual key of the found page.
+    /// Returns the slot index, if any.
     ///
     /// NOTE: We don't hold any lock on the mapping on return, so the slot might
     /// get recycled for an unrelated page immediately after this function
     /// returns.  The caller is responsible for re-checking that the slot still
     /// contains the page with the same key before using it.
     ///
-    fn search_mapping(&self, cache_key: &mut CacheKey) -> Option<usize> {
+    fn search_mapping(&self, cache_key: &CacheKey) -> Option<usize> {
         match cache_key {
-            CacheKey::MaterializedPage { hash_key, lsn } => {
-                let map = self.materialized_page_map.read().unwrap();
-                let versions = map.get(hash_key)?;
-
-                let version_idx = match versions.binary_search_by_key(lsn, |v| v.lsn) {
-                    Ok(version_idx) => version_idx,
-                    Err(0) => return None,
-                    Err(version_idx) => version_idx - 1,
-                };
-                let version = &versions[version_idx];
-                *lsn = version.lsn;
-                Some(version.slot_idx)
-            }
-            CacheKey::ImmutableFilePage { file_id, blkno } => {
-                let map = self.immutable_page_map.read().unwrap();
-                Some(*map.get(&(*file_id, *blkno))?)
-            }
-        }
-    }
-
-    /// Search for a page in the cache using the given search key.
-    ///
-    /// Like 'search_mapping, but performs an "exact" search. Used for
-    /// allocating a new buffer.
-    fn search_mapping_exact(&self, key: &CacheKey) -> Option<usize> {
-        match key {
-            CacheKey::MaterializedPage { hash_key, lsn } => {
-                let map = self.materialized_page_map.read().unwrap();
-                let versions = map.get(hash_key)?;
-
-                if let Ok(version_idx) = versions.binary_search_by_key(lsn, |v| v.lsn) {
-                    Some(versions[version_idx].slot_idx)
-                } else {
-                    None
-                }
-            }
             CacheKey::ImmutableFilePage { file_id, blkno } => {
                 let map = self.immutable_page_map.read().unwrap();
                 Some(*map.get(&(*file_id, *blkno))?)
@@ -775,27 +519,6 @@ impl PageCache {
     ///
     fn remove_mapping(&self, old_key: &CacheKey) {
         match old_key {
-            CacheKey::MaterializedPage {
-                hash_key: old_hash_key,
-                lsn: old_lsn,
-            } => {
-                let mut map = self.materialized_page_map.write().unwrap();
-                if let Entry::Occupied(mut old_entry) = map.entry(old_hash_key.clone()) {
-                    let versions = old_entry.get_mut();
-
-                    if let Ok(version_idx) = versions.binary_search_by_key(old_lsn, |v| v.lsn) {
-                        versions.remove(version_idx);
-                        self.size_metrics
-                            .current_bytes_materialized_page
-                            .sub_page_sz(1);
-                        if versions.is_empty() {
-                            old_entry.remove_entry();
-                        }
-                    }
-                } else {
-                    panic!("could not find old key in mapping")
-                }
-            }
             CacheKey::ImmutableFilePage { file_id, blkno } => {
                 let mut map = self.immutable_page_map.write().unwrap();
                 map.remove(&(*file_id, *blkno))
@@ -812,30 +535,6 @@ impl PageCache {
     /// of the existing mapping and leaves it untouched.
     fn try_insert_mapping(&self, new_key: &CacheKey, slot_idx: usize) -> Option<usize> {
         match new_key {
-            CacheKey::MaterializedPage {
-                hash_key: new_key,
-                lsn: new_lsn,
-            } => {
-                let mut map = self.materialized_page_map.write().unwrap();
-                let versions = map.entry(new_key.clone()).or_default();
-                match versions.binary_search_by_key(new_lsn, |v| v.lsn) {
-                    Ok(version_idx) => Some(versions[version_idx].slot_idx),
-                    Err(version_idx) => {
-                        versions.insert(
-                            version_idx,
-                            Version {
-                                lsn: *new_lsn,
-                                slot_idx,
-                            },
-                        );
-                        self.size_metrics
-                            .current_bytes_materialized_page
-                            .add_page_sz(1);
-                        None
-                    }
-                }
-            }
-
             CacheKey::ImmutableFilePage { file_id, blkno } => {
                 let mut map = self.immutable_page_map.write().unwrap();
                 match map.entry((*file_id, *blkno)) {
@@ -949,7 +648,6 @@ impl PageCache {
         let size_metrics = &crate::metrics::PAGE_CACHE_SIZE;
         size_metrics.max_bytes.set_page_sz(num_pages);
         size_metrics.current_bytes_immutable.set_page_sz(0);
-        size_metrics.current_bytes_materialized_page.set_page_sz(0);
 
         let slots = page_buffer
             .chunks_exact_mut(PAGE_SZ)
@@ -968,7 +666,6 @@ impl PageCache {
             .collect();
 
         Self {
-            materialized_page_map: Default::default(),
             immutable_page_map: Default::default(),
             slots,
             next_evict_slot: AtomicUsize::new(0),

pageserver/src/page_service.rs

@@ -554,7 +554,7 @@ impl PageServerHandler {
         tenant_id: TenantId,
         timeline_id: TimelineId,
         protocol_version: PagestreamProtocolVersion,
-        ctx: RequestContext,
+        mut ctx: RequestContext,
     ) -> Result<(), QueryError>
     where
         IO: AsyncRead + AsyncWrite + Send + Sync + Unpin,
@@ -624,7 +624,7 @@ impl PageServerHandler {
                     fail::fail_point!("ps::handle-pagerequest-message::exists");
                     let span = tracing::info_span!("handle_get_rel_exists_request", rel = %req.rel, req_lsn = %req.request_lsn);
                     (
-                        self.handle_get_rel_exists_request(tenant_id, timeline_id, &req, &ctx)
+                        self.handle_get_rel_exists_request(tenant_id, timeline_id, &req, &mut ctx)
                             .instrument(span.clone())
                             .await,
                         span,
@@ -634,7 +634,7 @@ impl PageServerHandler {
                     fail::fail_point!("ps::handle-pagerequest-message::nblocks");
                     let span = tracing::info_span!("handle_get_nblocks_request", rel = %req.rel, req_lsn = %req.request_lsn);
                     (
-                        self.handle_get_nblocks_request(tenant_id, timeline_id, &req, &ctx)
+                        self.handle_get_nblocks_request(tenant_id, timeline_id, &req, &mut ctx)
                             .instrument(span.clone())
                             .await,
                         span,
@@ -645,7 +645,7 @@ impl PageServerHandler {
                     // shard_id is filled in by the handler
                     let span = tracing::info_span!("handle_get_page_at_lsn_request", rel = %req.rel, blkno = %req.blkno, req_lsn = %req.request_lsn);
                     (
-                        self.handle_get_page_at_lsn_request(tenant_id, timeline_id, &req, &ctx)
+                        self.handle_get_page_at_lsn_request(tenant_id, timeline_id, &req, &mut ctx)
                             .instrument(span.clone())
                             .await,
                         span,
@@ -655,7 +655,7 @@ impl PageServerHandler {
                     fail::fail_point!("ps::handle-pagerequest-message::dbsize");
                     let span = tracing::info_span!("handle_db_size_request", dbnode = %req.dbnode, req_lsn = %req.request_lsn);
                     (
-                        self.handle_db_size_request(tenant_id, timeline_id, &req, &ctx)
+                        self.handle_db_size_request(tenant_id, timeline_id, &req, &mut ctx)
                             .instrument(span.clone())
                             .await,
                         span,
@@ -665,7 +665,7 @@ impl PageServerHandler {
                     fail::fail_point!("ps::handle-pagerequest-message::slrusegment");
                     let span = tracing::info_span!("handle_get_slru_segment_request", kind = %req.kind, segno = %req.segno, req_lsn = %req.request_lsn);
                     (
-                        self.handle_get_slru_segment_request(tenant_id, timeline_id, &req, &ctx)
+                        self.handle_get_slru_segment_request(tenant_id, timeline_id, &req, &mut ctx)
                             .instrument(span.clone())
                             .await,
                         span,
@@ -728,7 +728,7 @@ impl PageServerHandler {
         base_lsn: Lsn,
         _end_lsn: Lsn,
         pg_version: u32,
-        ctx: RequestContext,
+        mut ctx: RequestContext,
     ) -> Result<(), QueryError>
     where
         IO: AsyncRead + AsyncWrite + Send + Sync + Unpin,
@@ -741,7 +741,7 @@ impl PageServerHandler {
             .get_active_tenant_with_timeout(tenant_id, ShardSelector::Zero, ACTIVE_TENANT_TIMEOUT)
             .await?;
         let timeline = tenant
-            .create_empty_timeline(timeline_id, base_lsn, pg_version, &ctx)
+            .create_empty_timeline(timeline_id, base_lsn, pg_version, &mut ctx)
             .await?;
 
         // TODO mark timeline as not ready until it reaches end_lsn.
@@ -766,7 +766,7 @@ impl PageServerHandler {
                 &mut copyin_reader,
                 base_lsn,
                 self.broker_client.clone(),
-                &ctx,
+                &mut ctx,
             )
             .await?;
 
@@ -791,7 +791,7 @@ impl PageServerHandler {
         timeline_id: TimelineId,
         start_lsn: Lsn,
         end_lsn: Lsn,
-        ctx: RequestContext,
+        mut ctx: RequestContext,
     ) -> Result<(), QueryError>
     where
         IO: AsyncRead + AsyncWrite + Send + Sync + Unpin,
@@ -814,7 +814,7 @@ impl PageServerHandler {
         pgb.write_message_noflush(&BeMessage::CopyInResponse)?;
         self.flush_cancellable(pgb, &timeline.cancel).await?;
         let mut copyin_reader = pin!(StreamReader::new(self.copyin_stream(pgb, &timeline.cancel)));
-        import_wal_from_tar(&timeline, &mut copyin_reader, start_lsn, end_lsn, &ctx).await?;
+        import_wal_from_tar(&timeline, &mut copyin_reader, start_lsn, end_lsn, &mut ctx).await?;
         info!("wal import complete");
 
         // Read the end of the tar archive.
@@ -867,7 +867,7 @@ impl PageServerHandler {
         request_lsn: Lsn,
         not_modified_since: Lsn,
         latest_gc_cutoff_lsn: &RcuReadGuard<Lsn>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<Lsn, PageStreamError> {
         let last_record_lsn = timeline.get_last_record_lsn();
 
@@ -926,7 +926,7 @@ impl PageServerHandler {
         tenant_shard_id: TenantShardId,
         timeline_id: TimelineId,
         lsn: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<(), QueryError>
     where
         IO: AsyncRead + AsyncWrite + Send + Sync + Unpin,
@@ -935,7 +935,7 @@ impl PageServerHandler {
         let timeline = self
             .get_active_tenant_timeline(tenant_shard_id.tenant_id, timeline_id, shard_selector)
             .await?;
-        let lease = timeline.make_lsn_lease(lsn, ctx)?;
+        let lease = timeline.make_lsn_lease(lsn, timeline.get_lsn_lease_length(), ctx)?;
         let valid_until = lease
             .valid_until
             .duration_since(SystemTime::UNIX_EPOCH)
@@ -958,12 +958,12 @@ impl PageServerHandler {
         tenant_id: TenantId,
         timeline_id: TimelineId,
         req: &PagestreamExistsRequest,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<PagestreamBeMessage, PageStreamError> {
         let timeline = self.get_timeline_shard_zero(tenant_id, timeline_id).await?;
         let _timer = timeline
             .query_metrics
-            .start_timer(metrics::SmgrQueryType::GetRelExists, ctx);
+            .start_timer(metrics::SmgrQueryType::GetRelExists);
 
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
         let lsn = Self::wait_or_get_last_lsn(
@@ -990,13 +990,13 @@ impl PageServerHandler {
         tenant_id: TenantId,
         timeline_id: TimelineId,
         req: &PagestreamNblocksRequest,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<PagestreamBeMessage, PageStreamError> {
         let timeline = self.get_timeline_shard_zero(tenant_id, timeline_id).await?;
 
         let _timer = timeline
             .query_metrics
-            .start_timer(metrics::SmgrQueryType::GetRelSize, ctx);
+            .start_timer(metrics::SmgrQueryType::GetRelSize);
 
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
         let lsn = Self::wait_or_get_last_lsn(
@@ -1023,13 +1023,13 @@ impl PageServerHandler {
         tenant_id: TenantId,
         timeline_id: TimelineId,
         req: &PagestreamDbSizeRequest,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<PagestreamBeMessage, PageStreamError> {
         let timeline = self.get_timeline_shard_zero(tenant_id, timeline_id).await?;
 
         let _timer = timeline
             .query_metrics
-            .start_timer(metrics::SmgrQueryType::GetDbSize, ctx);
+            .start_timer(metrics::SmgrQueryType::GetDbSize);
 
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
         let lsn = Self::wait_or_get_last_lsn(
@@ -1173,7 +1173,7 @@ impl PageServerHandler {
         tenant_id: TenantId,
         timeline_id: TimelineId,
         req: &PagestreamGetPageRequest,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<PagestreamBeMessage, PageStreamError> {
         let timeline = match self.get_cached_timeline_for_page(req) {
             Ok(tl) => {
@@ -1206,7 +1206,7 @@ impl PageServerHandler {
 
         let _timer = timeline
             .query_metrics
-            .start_timer(metrics::SmgrQueryType::GetPageAtLsn, ctx);
+            .start_timer(metrics::SmgrQueryType::GetPageAtLsn);
 
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
         let lsn = Self::wait_or_get_last_lsn(
@@ -1233,13 +1233,13 @@ impl PageServerHandler {
         tenant_id: TenantId,
         timeline_id: TimelineId,
         req: &PagestreamGetSlruSegmentRequest,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<PagestreamBeMessage, PageStreamError> {
         let timeline = self.get_timeline_shard_zero(tenant_id, timeline_id).await?;
 
         let _timer = timeline
             .query_metrics
-            .start_timer(metrics::SmgrQueryType::GetSlruSegment, ctx);
+            .start_timer(metrics::SmgrQueryType::GetSlruSegment);
 
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
         let lsn = Self::wait_or_get_last_lsn(
@@ -1275,7 +1275,7 @@ impl PageServerHandler {
         prev_lsn: Option<Lsn>,
         full_backup: bool,
         gzip: bool,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<(), QueryError>
     where
         IO: AsyncRead + AsyncWrite + Send + Sync + Unpin,
@@ -1534,7 +1534,7 @@ where
 
         fail::fail_point!("ps::connection-start::process-query");
 
-        let ctx = self.connection_ctx.attached_child();
+        let mut ctx = self.connection_ctx.attached_child();
         debug!("process query {query_string:?}");
         let parts = query_string.split_whitespace().collect::<Vec<_>>();
         if let Some(params) = parts.strip_prefix(&["pagestream_v2"]) {
@@ -1624,7 +1624,7 @@ where
                 }
             };
 
-            let metric_recording = metrics::BASEBACKUP_QUERY_TIME.start_recording(&ctx);
+            let metric_recording = metrics::BASEBACKUP_QUERY_TIME.start_recording(&mut ctx);
             let res = async {
                 self.handle_basebackup_request(
                     pgb,
@@ -1634,7 +1634,7 @@ where
                     None,
                     false,
                     gzip,
-                    &ctx,
+                    &mut ctx,
                 )
                 .await?;
                 pgb.write_message_noflush(&BeMessage::CommandComplete(b"SELECT 1"))?;
@@ -1732,7 +1732,7 @@ where
                 prev_lsn,
                 true,
                 false,
-                &ctx,
+                &mut ctx,
             )
             .await?;
             pgb.write_message_noflush(&BeMessage::CommandComplete(b"SELECT 1"))?;
@@ -1860,7 +1860,7 @@ where
                 .with_context(|| format!("Failed to parse Lsn from {}", params[2]))?;
 
             match self
-                .handle_make_lsn_lease(pgb, tenant_shard_id, timeline_id, lsn, &ctx)
+                .handle_make_lsn_lease(pgb, tenant_shard_id, timeline_id, lsn, &mut ctx)
                 .await
             {
                 Ok(()) => pgb.write_message_noflush(&BeMessage::CommandComplete(b"SELECT 1"))?,

pageserver/src/pgdatadir_mapping.rs

@@ -188,7 +188,7 @@ impl Timeline {
         tag: RelTag,
         blknum: BlockNumber,
         version: Version<'_>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<Bytes, PageReconstructError> {
         if tag.relnode == 0 {
             return Err(PageReconstructError::Other(
@@ -218,7 +218,7 @@ impl Timeline {
         spcnode: Oid,
         dbnode: Oid,
         version: Version<'_>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<usize, PageReconstructError> {
         let mut total_blocks = 0;
 
@@ -236,7 +236,7 @@ impl Timeline {
         &self,
         tag: RelTag,
         version: Version<'_>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<BlockNumber, PageReconstructError> {
         if tag.relnode == 0 {
             return Err(PageReconstructError::Other(
@@ -272,7 +272,7 @@ impl Timeline {
         &self,
         tag: RelTag,
         version: Version<'_>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<bool, PageReconstructError> {
         if tag.relnode == 0 {
             return Err(PageReconstructError::Other(
@@ -307,7 +307,7 @@ impl Timeline {
         spcnode: Oid,
         dbnode: Oid,
         version: Version<'_>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<HashSet<RelTag>, PageReconstructError> {
         // fetch directory listing
         let key = rel_dir_to_key(spcnode, dbnode);
@@ -335,7 +335,7 @@ impl Timeline {
         kind: SlruKind,
         segno: u32,
         lsn: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<Bytes, PageReconstructError> {
         let n_blocks = self
             .get_slru_segment_size(kind, segno, Version::Lsn(lsn), ctx)
@@ -357,7 +357,7 @@ impl Timeline {
         segno: u32,
         blknum: BlockNumber,
         lsn: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<Bytes, PageReconstructError> {
         let key = slru_block_to_key(kind, segno, blknum);
         self.get(key, lsn, ctx).await
@@ -369,7 +369,7 @@ impl Timeline {
         kind: SlruKind,
         segno: u32,
         version: Version<'_>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<BlockNumber, PageReconstructError> {
         let key = slru_segment_size_to_key(kind, segno);
         let mut buf = version.get(self, key, ctx).await?;
@@ -382,7 +382,7 @@ impl Timeline {
         kind: SlruKind,
         segno: u32,
         version: Version<'_>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<bool, PageReconstructError> {
         // fetch directory listing
         let key = slru_dir_to_key(kind);
@@ -408,7 +408,7 @@ impl Timeline {
         &self,
         search_timestamp: TimestampTz,
         cancel: &CancellationToken,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<LsnForTimestamp, PageReconstructError> {
         pausable_failpoint!("find-lsn-for-timestamp-pausable");
 
@@ -499,7 +499,7 @@ impl Timeline {
         probe_lsn: Lsn,
         found_smaller: &mut bool,
         found_larger: &mut bool,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<bool, PageReconstructError> {
         self.map_all_timestamps(probe_lsn, ctx, |timestamp| {
             if timestamp >= search_timestamp {
@@ -519,7 +519,7 @@ impl Timeline {
     pub(crate) async fn get_timestamp_for_lsn(
         &self,
         probe_lsn: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<Option<TimestampTz>, PageReconstructError> {
         let mut max: Option<TimestampTz> = None;
         self.map_all_timestamps(probe_lsn, ctx, |timestamp| {
@@ -542,7 +542,7 @@ impl Timeline {
     async fn map_all_timestamps<T: Default>(
         &self,
         probe_lsn: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
         mut f: impl FnMut(TimestampTz) -> ControlFlow<T>,
     ) -> Result<T, PageReconstructError> {
         for segno in self
@@ -575,7 +575,7 @@ impl Timeline {
     pub(crate) async fn get_slru_keyspace(
         &self,
         version: Version<'_>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<KeySpace, PageReconstructError> {
         let mut accum = KeySpaceAccum::new();
 
@@ -604,7 +604,7 @@ impl Timeline {
         &self,
         kind: SlruKind,
         version: Version<'_>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<HashSet<u32>, PageReconstructError> {
         // fetch directory entry
         let key = slru_dir_to_key(kind);
@@ -621,7 +621,7 @@ impl Timeline {
         spcnode: Oid,
         dbnode: Oid,
         version: Version<'_>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<Bytes, PageReconstructError> {
         let key = relmap_file_key(spcnode, dbnode);
 
@@ -632,7 +632,7 @@ impl Timeline {
     pub(crate) async fn list_dbdirs(
         &self,
         lsn: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<HashMap<(Oid, Oid), bool>, PageReconstructError> {
         // fetch directory entry
         let buf = self.get(DBDIR_KEY, lsn, ctx).await?;
@@ -647,7 +647,7 @@ impl Timeline {
         &self,
         xid: TransactionId,
         lsn: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<Bytes, PageReconstructError> {
         let key = twophase_file_key(xid);
         let buf = self.get(key, lsn, ctx).await?;
@@ -657,7 +657,7 @@ impl Timeline {
     pub(crate) async fn list_twophase_files(
         &self,
         lsn: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<HashSet<TransactionId>, PageReconstructError> {
         // fetch directory entry
         let buf = self.get(TWOPHASEDIR_KEY, lsn, ctx).await?;
@@ -671,7 +671,7 @@ impl Timeline {
     pub(crate) async fn get_control_file(
         &self,
         lsn: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<Bytes, PageReconstructError> {
         self.get(CONTROLFILE_KEY, lsn, ctx).await
     }
@@ -679,7 +679,7 @@ impl Timeline {
     pub(crate) async fn get_checkpoint(
         &self,
         lsn: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<Bytes, PageReconstructError> {
         self.get(CHECKPOINT_KEY, lsn, ctx).await
     }
@@ -687,7 +687,7 @@ impl Timeline {
     async fn list_aux_files_v1(
         &self,
         lsn: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<HashMap<String, Bytes>, PageReconstructError> {
         match self.get(AUX_FILES_KEY, lsn, ctx).await {
             Ok(buf) => match AuxFilesDirectory::des(&buf).context("deserialization failure") {
@@ -705,7 +705,7 @@ impl Timeline {
     async fn list_aux_files_v2(
         &self,
         lsn: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<HashMap<String, Bytes>, PageReconstructError> {
         let kv = self
             .scan(KeySpace::single(Key::metadata_aux_key_range()), lsn, ctx)
@@ -729,7 +729,7 @@ impl Timeline {
     pub(crate) async fn trigger_aux_file_size_computation(
         &self,
         lsn: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<(), PageReconstructError> {
         let current_policy = self.last_aux_file_policy.load();
         if let Some(AuxFilePolicy::V2) | Some(AuxFilePolicy::CrossValidation) = current_policy {
@@ -741,7 +741,7 @@ impl Timeline {
     pub(crate) async fn list_aux_files(
         &self,
         lsn: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<HashMap<String, Bytes>, PageReconstructError> {
         let current_policy = self.last_aux_file_policy.load();
         match current_policy {
@@ -779,7 +779,7 @@ impl Timeline {
     pub(crate) async fn get_replorigins(
         &self,
         lsn: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<HashMap<RepOriginId, Lsn>, PageReconstructError> {
         let kv = self
             .scan(KeySpace::single(repl_origin_key_range()), lsn, ctx)
@@ -809,7 +809,7 @@ impl Timeline {
     pub(crate) async fn get_current_logical_size_non_incremental(
         &self,
         lsn: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<u64, CalculateLogicalSizeError> {
         debug_assert_current_span_has_tenant_and_timeline_id_no_shard_id();
 
@@ -845,7 +845,7 @@ impl Timeline {
     pub(crate) async fn collect_keyspace(
         &self,
         lsn: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<(KeySpace, SparseKeySpace), CollectKeySpaceError> {
         // Iterate through key ranges, greedily packing them into partitions
         let mut result = KeySpaceAccum::new();
@@ -919,6 +919,14 @@ impl Timeline {
             result.add_key(AUX_FILES_KEY);
         }
 
+        #[cfg(test)]
+        {
+            let guard = self.extra_test_dense_keyspace.load();
+            for kr in &guard.ranges {
+                result.add_range(kr.clone());
+            }
+        }
+
         Ok((
             result.to_keyspace(),
             /* AUX sparse key space */
@@ -1137,7 +1145,7 @@ impl<'a> DatadirModification<'a> {
         spcnode: Oid,
         dbnode: Oid,
         img: Bytes,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<()> {
         // Add it to the directory (if it doesn't exist already)
         let buf = self.get(DBDIR_KEY, ctx).await?;
@@ -1174,7 +1182,7 @@ impl<'a> DatadirModification<'a> {
         &mut self,
         xid: TransactionId,
         img: Bytes,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<()> {
         // Add it to the directory entry
         let buf = self.get(TWOPHASEDIR_KEY, ctx).await?;
@@ -1221,7 +1229,7 @@ impl<'a> DatadirModification<'a> {
         &mut self,
         spcnode: Oid,
         dbnode: Oid,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<()> {
         let total_blocks = self
             .tline
@@ -1258,7 +1266,7 @@ impl<'a> DatadirModification<'a> {
         &mut self,
         rel: RelTag,
         nblocks: BlockNumber,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<(), RelationError> {
         if rel.relnode == 0 {
             return Err(RelationError::InvalidRelnode);
@@ -1320,7 +1328,7 @@ impl<'a> DatadirModification<'a> {
         &mut self,
         rel: RelTag,
         nblocks: BlockNumber,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<()> {
         anyhow::ensure!(rel.relnode != 0, RelationError::InvalidRelnode);
         if self
@@ -1354,7 +1362,7 @@ impl<'a> DatadirModification<'a> {
         &mut self,
         rel: RelTag,
         nblocks: BlockNumber,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<()> {
         anyhow::ensure!(rel.relnode != 0, RelationError::InvalidRelnode);
 
@@ -1376,7 +1384,11 @@ impl<'a> DatadirModification<'a> {
     }
 
     /// Drop a relation.
-    pub async fn put_rel_drop(&mut self, rel: RelTag, ctx: &RequestContext) -> anyhow::Result<()> {
+    pub async fn put_rel_drop(
+        &mut self,
+        rel: RelTag,
+        ctx: &mut RequestContext,
+    ) -> anyhow::Result<()> {
         anyhow::ensure!(rel.relnode != 0, RelationError::InvalidRelnode);
 
         // Remove it from the directory entry
@@ -1412,7 +1424,7 @@ impl<'a> DatadirModification<'a> {
         kind: SlruKind,
         segno: u32,
         nblocks: BlockNumber,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<()> {
         // Add it to the directory entry
         let dir_key = slru_dir_to_key(kind);
@@ -1458,7 +1470,7 @@ impl<'a> DatadirModification<'a> {
         &mut self,
         kind: SlruKind,
         segno: u32,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<()> {
         // Remove it from the directory entry
         let dir_key = slru_dir_to_key(kind);
@@ -1491,7 +1503,7 @@ impl<'a> DatadirModification<'a> {
     pub async fn drop_twophase_file(
         &mut self,
         xid: TransactionId,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<()> {
         // Remove it from the directory entry
         let buf = self.get(TWOPHASEDIR_KEY, ctx).await?;
@@ -1530,7 +1542,7 @@ impl<'a> DatadirModification<'a> {
         &mut self,
         path: &str,
         content: &[u8],
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<()> {
         let switch_policy = self.tline.get_switch_aux_file_policy();
 
@@ -1723,7 +1735,7 @@ impl<'a> DatadirModification<'a> {
     /// retains all the metadata, but data pages are flushed. That's again OK
     /// for bulk import, where you are just loading data pages and won't try to
     /// modify the same pages twice.
-    pub async fn flush(&mut self, ctx: &RequestContext) -> anyhow::Result<()> {
+    pub async fn flush(&mut self, ctx: &mut RequestContext) -> anyhow::Result<()> {
         // Unless we have accumulated a decent amount of changes, it's not worth it
         // to scan through the pending_updates list.
         let pending_nblocks = self.pending_nblocks;
@@ -1769,7 +1781,7 @@ impl<'a> DatadirModification<'a> {
     /// underlying timeline.
     /// All the modifications in this atomic update are stamped by the specified LSN.
     ///
-    pub async fn commit(&mut self, ctx: &RequestContext) -> anyhow::Result<()> {
+    pub async fn commit(&mut self, ctx: &mut RequestContext) -> anyhow::Result<()> {
         let mut writer = self.tline.writer().await;
 
         let pending_nblocks = self.pending_nblocks;
@@ -1820,7 +1832,7 @@ impl<'a> DatadirModification<'a> {
 
     // Internal helper functions to batch the modifications
 
-    async fn get(&self, key: Key, ctx: &RequestContext) -> Result<Bytes, PageReconstructError> {
+    async fn get(&self, key: Key, ctx: &mut RequestContext) -> Result<Bytes, PageReconstructError> {
         // Have we already updated the same key? Read the latest pending updated
         // version in that case.
         //
@@ -1887,7 +1899,7 @@ impl<'a> Version<'a> {
         &self,
         timeline: &Timeline,
         key: Key,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<Bytes, PageReconstructError> {
         match self {
             Version::Lsn(lsn) => timeline.get(key, *lsn, ctx).await,

pageserver/src/repository.rs

@@ -240,6 +240,7 @@ pub struct GcResult {
     pub layers_needed_by_cutoff: u64,
     pub layers_needed_by_pitr: u64,
     pub layers_needed_by_branches: u64,
+    pub layers_needed_by_leases: u64,
     pub layers_not_updated: u64,
     pub layers_removed: u64, // # of layer files removed because they have been made obsolete by newer ondisk files.
 
@@ -269,6 +270,7 @@ impl AddAssign for GcResult {
         self.layers_needed_by_pitr += other.layers_needed_by_pitr;
         self.layers_needed_by_cutoff += other.layers_needed_by_cutoff;
         self.layers_needed_by_branches += other.layers_needed_by_branches;
+        self.layers_needed_by_leases += other.layers_needed_by_leases;
         self.layers_not_updated += other.layers_not_updated;
         self.layers_removed += other.layers_removed;
 

pageserver/src/tenant.rs

@@ -31,6 +31,7 @@ use remote_storage::DownloadError;
 use remote_storage::GenericRemoteStorage;
 use remote_storage::TimeoutOrCancel;
 use std::fmt;
+use std::time::SystemTime;
 use storage_broker::BrokerClientChannel;
 use tokio::io::BufReader;
 use tokio::sync::watch;
@@ -65,9 +66,9 @@ use self::timeline::uninit::TimelineCreateGuard;
 use self::timeline::uninit::TimelineExclusionError;
 use self::timeline::uninit::UninitializedTimeline;
 use self::timeline::EvictionTaskTenantState;
+use self::timeline::GcCutoffs;
 use self::timeline::TimelineResources;
 use self::timeline::WaitLsnError;
-use self::timeline::{GcCutoffs, GcInfo};
 use crate::config::PageServerConf;
 use crate::context::{DownloadBehavior, RequestContext};
 use crate::deletion_queue::DeletionQueueClient;
@@ -509,11 +510,24 @@ pub(crate) enum GcError {
     #[error(transparent)]
     Remote(anyhow::Error),
 
+    // An error reading while calculating GC cutoffs
+    #[error(transparent)]
+    GcCutoffs(PageReconstructError),
+
     // If GC was invoked for a particular timeline, this error means it didn't exist
     #[error("timeline not found")]
     TimelineNotFound,
 }
 
+impl From<PageReconstructError> for GcError {
+    fn from(value: PageReconstructError) -> Self {
+        match value {
+            PageReconstructError::Cancelled => Self::TimelineCancelled,
+            other => Self::GcCutoffs(other),
+        }
+    }
+}
+
 impl Tenant {
     /// Yet another helper for timeline initialization.
     ///
@@ -533,7 +547,7 @@ impl Tenant {
         metadata: TimelineMetadata,
         ancestor: Option<Arc<Timeline>>,
         last_aux_file_policy: Option<AuxFilePolicy>,
-        _ctx: &RequestContext,
+        _ctx: &mut RequestContext,
     ) -> anyhow::Result<()> {
         let tenant_id = self.tenant_shard_id;
 
@@ -642,7 +656,7 @@ impl Tenant {
         init_order: Option<InitializationOrder>,
         tenants: &'static std::sync::RwLock<TenantsMap>,
         mode: SpawnMode,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<Arc<Tenant>> {
         let wal_redo_manager = Arc::new(WalRedoManager::from(PostgresRedoManager::new(
             conf,
@@ -951,7 +965,7 @@ impl Tenant {
         self: &Arc<Tenant>,
         preload: Option<TenantPreload>,
         mode: SpawnMode,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<()> {
         span::debug_assert_current_span_has_tenant_id();
 
@@ -1033,7 +1047,6 @@ impl Tenant {
                 remote_metadata,
                 TimelineResources {
                     remote_client,
-                    deletion_queue_client: self.deletion_queue_client.clone(),
                     timeline_get_throttle: self.timeline_get_throttle.clone(),
                 },
                 ctx,
@@ -1059,7 +1072,6 @@ impl Tenant {
                 timeline_id,
                 &index_part.metadata,
                 remote_timeline_client,
-                self.deletion_queue_client.clone(),
             )
             .instrument(tracing::info_span!("timeline_delete", %timeline_id))
             .await
@@ -1163,7 +1175,7 @@ impl Tenant {
         index_part: IndexPart,
         remote_metadata: TimelineMetadata,
         resources: TimelineResources,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<()> {
         span::debug_assert_current_span_has_tenant_id();
 
@@ -1346,7 +1358,7 @@ impl Tenant {
         new_timeline_id: TimelineId,
         initdb_lsn: Lsn,
         pg_version: u32,
-        _ctx: &RequestContext,
+        _ctx: &mut RequestContext,
     ) -> anyhow::Result<UninitializedTimeline> {
         anyhow::ensure!(
             self.is_active(),
@@ -1389,7 +1401,7 @@ impl Tenant {
         new_timeline_id: TimelineId,
         initdb_lsn: Lsn,
         pg_version: u32,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<Arc<Timeline>> {
         let uninit_tl = self
             .create_empty_timeline(new_timeline_id, initdb_lsn, pg_version, ctx)
@@ -1428,7 +1440,7 @@ impl Tenant {
         new_timeline_id: TimelineId,
         initdb_lsn: Lsn,
         pg_version: u32,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
         delta_layer_desc: Vec<Vec<(pageserver_api::key::Key, Lsn, crate::repository::Value)>>,
         image_layer_desc: Vec<(Lsn, Vec<(pageserver_api::key::Key, bytes::Bytes)>)>,
         end_lsn: Lsn,
@@ -1465,7 +1477,7 @@ impl Tenant {
         pg_version: u32,
         load_existing_initdb: Option<TimelineId>,
         broker_client: storage_broker::BrokerClientChannel,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<Arc<Timeline>, CreateTimelineError> {
         if !self.is_active() {
             if matches!(self.current_state(), TenantState::Stopping { .. }) {
@@ -1638,7 +1650,7 @@ impl Tenant {
         horizon: u64,
         pitr: Duration,
         cancel: &CancellationToken,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<GcResult, GcError> {
         // Don't start doing work during shutdown
         if let TenantState::Stopping { .. } = self.current_state() {
@@ -1670,7 +1682,7 @@ impl Tenant {
     async fn compaction_iteration(
         &self,
         cancel: &CancellationToken,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<(), timeline::CompactionError> {
         // Don't start doing work during shutdown, or when broken, we do not need those in the logs
         if !self.is_active() {
@@ -1767,7 +1779,7 @@ impl Tenant {
         self: &Arc<Self>,
         broker_client: BrokerClientChannel,
         background_jobs_can_start: Option<&completion::Barrier>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) {
         span::debug_assert_current_span_has_tenant_id();
 
@@ -2417,6 +2429,13 @@ impl Tenant {
         }
     }
 
+    pub fn get_lsn_lease_length(&self) -> Duration {
+        let tenant_conf = self.tenant_conf.load().tenant_conf.clone();
+        tenant_conf
+            .lsn_lease_length
+            .unwrap_or(self.conf.default_tenant_conf.lsn_lease_length)
+    }
+
     pub fn set_new_tenant_config(&self, new_tenant_conf: TenantConfOpt) {
         // Use read-copy-update in order to avoid overwriting the location config
         // state if this races with [`Tenant::set_new_location_config`]. Note that
@@ -2815,7 +2834,7 @@ impl Tenant {
         horizon: u64,
         pitr: Duration,
         cancel: &CancellationToken,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<GcResult, GcError> {
         let mut totals: GcResult = Default::default();
         let now = Instant::now();
@@ -2875,7 +2894,7 @@ impl Tenant {
     pub(crate) async fn refresh_gc_info(
         &self,
         cancel: &CancellationToken,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<Vec<Arc<Timeline>>, GcError> {
         // since this method can now be called at different rates than the configured gc loop, it
         // might be that these configuration values get applied faster than what it was previously,
@@ -2896,7 +2915,7 @@ impl Tenant {
         horizon: u64,
         pitr: Duration,
         cancel: &CancellationToken,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<Vec<Arc<Timeline>>, GcError> {
         // before taking the gc_cs lock, do the heavier weight finding of gc_cutoff points for
         // currently visible timelines.
@@ -2921,17 +2940,9 @@ impl Tenant {
                 .checked_sub(horizon)
                 .unwrap_or(Lsn(0));
 
-            let res = timeline.find_gc_cutoffs(cutoff, pitr, cancel, ctx).await;
-
-            match res {
-                Ok(cutoffs) => {
-                    let old = gc_cutoffs.insert(timeline.timeline_id, cutoffs);
-                    assert!(old.is_none());
-                }
-                Err(e) => {
-                    tracing::warn!(timeline_id = %timeline.timeline_id, "ignoring failure to find gc cutoffs: {e:#}");
-                }
-            }
+            let cutoffs = timeline.find_gc_cutoffs(cutoff, pitr, cancel, ctx).await?;
+            let old = gc_cutoffs.insert(timeline.timeline_id, cutoffs);
+            assert!(old.is_none());
         }
 
         if !self.is_active() || self.cancel.is_cancelled() {
@@ -3007,12 +3018,13 @@ impl Tenant {
             {
                 let mut target = timeline.gc_info.write().unwrap();
 
+                let now = SystemTime::now();
+                target.leases.retain(|_, lease| !lease.is_expired(&now));
+
                 match gc_cutoffs.remove(&timeline.timeline_id) {
                     Some(cutoffs) => {
-                        *target = GcInfo {
-                            retain_lsns: branchpoints,
-                            cutoffs,
-                        };
+                        target.retain_lsns = branchpoints;
+                        target.cutoffs = cutoffs;
                     }
                     None => {
                         // reasons for this being unavailable:
@@ -3041,7 +3053,7 @@ impl Tenant {
         src_timeline: &Arc<Timeline>,
         dst_id: TimelineId,
         ancestor_lsn: Option<Lsn>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<Arc<Timeline>, CreateTimelineError> {
         let create_guard = self.create_timeline_create_guard(dst_id).unwrap();
         let tl = self
@@ -3059,7 +3071,7 @@ impl Tenant {
         src_timeline: &Arc<Timeline>,
         dst_id: TimelineId,
         ancestor_lsn: Option<Lsn>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
         delta_layer_desc: Vec<Vec<(pageserver_api::key::Key, Lsn, crate::repository::Value)>>,
         image_layer_desc: Vec<(Lsn, Vec<(pageserver_api::key::Key, bytes::Bytes)>)>,
         end_lsn: Lsn,
@@ -3096,7 +3108,7 @@ impl Tenant {
         dst_id: TimelineId,
         start_lsn: Option<Lsn>,
         timeline_create_guard: TimelineCreateGuard<'_>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<Arc<Timeline>, CreateTimelineError> {
         self.branch_timeline_impl(src_timeline, dst_id, start_lsn, timeline_create_guard, ctx)
             .await
@@ -3108,7 +3120,7 @@ impl Tenant {
         dst_id: TimelineId,
         start_lsn: Option<Lsn>,
         timeline_create_guard: TimelineCreateGuard<'_>,
-        _ctx: &RequestContext,
+        _ctx: &mut RequestContext,
     ) -> Result<Arc<Timeline>, CreateTimelineError> {
         let src_id = src_timeline.timeline_id;
 
@@ -3221,7 +3233,7 @@ impl Tenant {
         timeline_id: TimelineId,
         pg_version: u32,
         load_existing_initdb: Option<TimelineId>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<Arc<Timeline>> {
         let create_guard = self.create_timeline_create_guard(timeline_id).unwrap();
         self.bootstrap_timeline(
@@ -3293,7 +3305,7 @@ impl Tenant {
         pg_version: u32,
         load_existing_initdb: Option<TimelineId>,
         timeline_create_guard: TimelineCreateGuard<'_>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<Arc<Timeline>> {
         // create a `tenant/{tenant_id}/timelines/basebackup-{timeline_id}.{TEMP_FILE_SUFFIX}/`
         // temporary directory for basebackup files for the given timeline.
@@ -3395,6 +3407,12 @@ impl Tenant {
         let tenant_shard_id = raw_timeline.owning_tenant.tenant_shard_id;
         let unfinished_timeline = raw_timeline.raw_timeline()?;
 
+        // Flush the new layer files to disk, before we make the timeline as available to
+        // the outside world.
+        //
+        // Flush loop needs to be spawned in order to be able to flush.
+        unfinished_timeline.maybe_spawn_flush_loop();
+
         import_datadir::import_timeline_from_postgres_datadir(
             unfinished_timeline,
             &pgdata_path,
@@ -3406,12 +3424,6 @@ impl Tenant {
             format!("Failed to import pgdatadir for timeline {tenant_shard_id}/{timeline_id}")
         })?;
 
-        // Flush the new layer files to disk, before we make the timeline as available to
-        // the outside world.
-        //
-        // Flush loop needs to be spawned in order to be able to flush.
-        unfinished_timeline.maybe_spawn_flush_loop();
-
         fail::fail_point!("before-checkpoint-new-timeline", |_| {
             anyhow::bail!("failpoint before-checkpoint-new-timeline");
         });
@@ -3443,7 +3455,6 @@ impl Tenant {
         );
         TimelineResources {
             remote_client,
-            deletion_queue_client: self.deletion_queue_client.clone(),
             timeline_get_throttle: self.timeline_get_throttle.clone(),
         }
     }
@@ -3552,8 +3563,8 @@ impl Tenant {
         max_retention_period: Option<u64>,
         cause: LogicalSizeCalculationCause,
         cancel: &CancellationToken,
-        ctx: &RequestContext,
-    ) -> anyhow::Result<size::ModelInputs> {
+        ctx: &mut RequestContext,
+    ) -> Result<size::ModelInputs, size::CalculateSyntheticSizeError> {
         let logical_sizes_at_once = self
             .conf
             .concurrent_tenant_size_logical_size_queries
@@ -3568,8 +3579,8 @@ impl Tenant {
         // See more for on the issue #2748 condenced out of the initial PR review.
         let mut shared_cache = tokio::select! {
             locked = self.cached_logical_sizes.lock() => locked,
-            _ = cancel.cancelled() => anyhow::bail!("cancelled"),
-            _ = self.cancel.cancelled() => anyhow::bail!("tenant is shutting down"),
+            _ = cancel.cancelled() => return Err(size::CalculateSyntheticSizeError::Cancelled),
+            _ = self.cancel.cancelled() => return Err(size::CalculateSyntheticSizeError::Cancelled),
         };
 
         size::gather_inputs(
@@ -3592,11 +3603,11 @@ impl Tenant {
         &self,
         cause: LogicalSizeCalculationCause,
         cancel: &CancellationToken,
-        ctx: &RequestContext,
-    ) -> anyhow::Result<u64> {
+        ctx: &mut RequestContext,
+    ) -> Result<u64, size::CalculateSyntheticSizeError> {
         let inputs = self.gather_size_inputs(None, cause, cancel, ctx).await?;
 
-        let size = inputs.calculate()?;
+        let size = inputs.calculate();
 
         self.set_cached_synthetic_size(size);
 
@@ -3745,7 +3756,7 @@ async fn run_initdb(
 pub async fn dump_layerfile_from_path(
     path: &Utf8Path,
     verbose: bool,
-    ctx: &RequestContext,
+    ctx: &mut RequestContext,
 ) -> anyhow::Result<()> {
     use std::os::unix::fs::FileExt;
 
@@ -3831,6 +3842,8 @@ pub(crate) mod harness {
                     tenant_conf.image_layer_creation_check_threshold,
                 ),
                 switch_aux_file_policy: Some(tenant_conf.switch_aux_file_policy),
+                lsn_lease_length: Some(tenant_conf.lsn_lease_length),
+                lsn_lease_length_for_ts: Some(tenant_conf.lsn_lease_length_for_ts),
             }
         }
     }
@@ -3947,7 +3960,7 @@ pub(crate) mod harness {
         #[instrument(skip_all, fields(tenant_id=%self.tenant_shard_id.tenant_id, shard_id=%self.tenant_shard_id.shard_slug()))]
         pub(crate) async fn do_try_load(
             &self,
-            ctx: &RequestContext,
+            ctx: &mut RequestContext,
         ) -> anyhow::Result<Arc<Tenant>> {
             let walredo_mgr = Arc::new(WalRedoManager::from(TestRedoManager));
 
@@ -4041,13 +4054,16 @@ mod tests {
     use crate::repository::{Key, Value};
     use crate::tenant::harness::*;
     use crate::tenant::timeline::CompactFlags;
+    use crate::walrecord::NeonWalRecord;
     use crate::DEFAULT_PG_VERSION;
     use bytes::{Bytes, BytesMut};
     use hex_literal::hex;
+    use itertools::Itertools;
     use pageserver_api::key::{AUX_FILES_KEY, AUX_KEY_PREFIX, NON_INHERITED_RANGE};
     use pageserver_api::keyspace::KeySpace;
     use pageserver_api::models::{CompactionAlgorithm, CompactionAlgorithmSettings};
     use rand::{thread_rng, Rng};
+    use storage_layer::PersistentLayerKey;
     use tests::storage_layer::ValuesReconstructState;
     use tests::timeline::{GetVectoredError, ShutdownMode};
     use utils::bin_ser::BeSer;
@@ -4203,7 +4219,7 @@ mod tests {
     async fn make_some_layers(
         tline: &Timeline,
         start_lsn: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<()> {
         let mut lsn = start_lsn;
         {
@@ -4691,7 +4707,7 @@ mod tests {
     async fn bulk_insert_compact_gc(
         tenant: &Tenant,
         timeline: &Arc<Timeline>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
         lsn: Lsn,
         repeat: usize,
         key_count: usize,
@@ -4703,7 +4719,7 @@ mod tests {
     async fn bulk_insert_maybe_compact_gc(
         tenant: &Tenant,
         timeline: &Arc<Timeline>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
         mut lsn: Lsn,
         repeat: usize,
         key_count: usize,
@@ -5262,6 +5278,9 @@ mod tests {
         let cancel = CancellationToken::new();
 
         let mut test_key = Key::from_hex("010000000033333333444444445500000000").unwrap();
+        let mut test_key_end = test_key;
+        test_key_end.field6 = NUM_KEYS as u32;
+        tline.add_extra_test_dense_keyspace(KeySpace::single(test_key..test_key_end));
 
         let mut keyspace = KeySpaceAccum::new();
 
@@ -6221,8 +6240,8 @@ mod tests {
 
         let cancel = CancellationToken::new();
 
-        let mut base_key = Key::from_hex("000000000033333333444444445500000000").unwrap();
-        base_key.field1 = AUX_KEY_PREFIX;
+        let base_key = Key::from_hex("620000000033333333444444445500000000").unwrap();
+        assert_eq!(base_key.field1, AUX_KEY_PREFIX); // in case someone accidentally changed the prefix...
         let mut test_key = base_key;
         let mut lsn = Lsn(0x10);
 
@@ -6230,7 +6249,7 @@ mod tests {
             tline: &Timeline,
             keyspace: &KeySpace,
             lsn: Lsn,
-            ctx: &RequestContext,
+            ctx: &mut RequestContext,
         ) -> anyhow::Result<(BTreeMap<Key, Result<Bytes, PageReconstructError>>, usize)> {
             let mut reconstruct_state = ValuesReconstructState::default();
             let res = tline
@@ -6327,6 +6346,7 @@ mod tests {
                 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?;
+        tline.add_extra_test_dense_keyspace(KeySpace::single(base_key..(base_key_nonexist.next())));
 
         let child = tenant
             .branch_timeline_test_with_layers(
@@ -6345,7 +6365,7 @@ mod tests {
             tline: &Arc<Timeline>,
             key: Key,
             lsn: Lsn,
-            ctx: &RequestContext,
+            ctx: &mut RequestContext,
         ) -> Result<Option<Bytes>, GetVectoredError> {
             let mut reconstruct_state = ValuesReconstructState::new();
             let mut res = tline
@@ -6441,7 +6461,7 @@ mod tests {
             tline: &Arc<Timeline>,
             key: Key,
             lsn: Lsn,
-            ctx: &RequestContext,
+            ctx: &mut RequestContext,
         ) -> Result<Option<Bytes>, GetVectoredError> {
             let mut reconstruct_state = ValuesReconstructState::new();
             let mut res = tline
@@ -6495,7 +6515,7 @@ mod tests {
         tline: &Arc<Timeline>,
         key: Key,
         lsn: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<Option<Bytes>, GetVectoredError> {
         let mut reconstruct_state = ValuesReconstructState::new();
         let mut res = tline
@@ -6584,8 +6604,8 @@ mod tests {
     }
 
     #[tokio::test]
-    async fn test_metadata_tombstone_image_creation() -> anyhow::Result<()> {
-        let harness = TenantHarness::create("test_metadata_tombstone_image_creation")?;
+    async fn test_metadata_tombstone_image_creation() {
+        let harness = TenantHarness::create("test_metadata_tombstone_image_creation").unwrap();
         let (tenant, ctx) = harness.load().await;
 
         let key0 = Key::from_hex("620000000033333333444444445500000000").unwrap();
@@ -6613,7 +6633,8 @@ mod tests {
                 vec![(Lsn(0x10), vec![(key1, test_img("metadata key 1"))])],
                 Lsn(0x30),
             )
-            .await?;
+            .await
+            .unwrap();
 
         let cancel = CancellationToken::new();
 
@@ -6628,23 +6649,24 @@ mod tests {
                 },
                 &ctx,
             )
-            .await?;
+            .await
+            .unwrap();
 
         // Image layers are created at last_record_lsn
         let images = tline
             .inspect_image_layers(Lsn(0x30), &ctx)
-            .await?
+            .await
+            .unwrap()
             .into_iter()
             .filter(|(k, _)| k.is_metadata_key())
             .collect::<Vec<_>>();
         assert_eq!(images.len(), 2); // the image layer should only contain two existing keys, tombstones should be removed.
-
-        Ok(())
     }
 
     #[tokio::test]
-    async fn test_metadata_tombstone_empty_image_creation() -> anyhow::Result<()> {
-        let harness = TenantHarness::create("test_metadata_tombstone_image_creation")?;
+    async fn test_metadata_tombstone_empty_image_creation() {
+        let harness =
+            TenantHarness::create("test_metadata_tombstone_empty_image_creation").unwrap();
         let (tenant, ctx) = harness.load().await;
 
         let key1 = Key::from_hex("620000000033333333444444445500000001").unwrap();
@@ -6666,7 +6688,8 @@ mod tests {
                 vec![(Lsn(0x10), vec![(key1, test_img("metadata key 1"))])],
                 Lsn(0x30),
             )
-            .await?;
+            .await
+            .unwrap();
 
         let cancel = CancellationToken::new();
 
@@ -6681,16 +6704,338 @@ mod tests {
                 },
                 &ctx,
             )
-            .await?;
+            .await
+            .unwrap();
 
         // Image layers are created at last_record_lsn
         let images = tline
             .inspect_image_layers(Lsn(0x30), &ctx)
-            .await?
+            .await
+            .unwrap()
             .into_iter()
             .filter(|(k, _)| k.is_metadata_key())
             .collect::<Vec<_>>();
         assert_eq!(images.len(), 0); // the image layer should not contain tombstones, or it is not created
+    }
+
+    #[tokio::test]
+    async fn test_simple_bottom_most_compaction_images() -> anyhow::Result<()> {
+        let harness = TenantHarness::create("test_simple_bottom_most_compaction_images")?;
+        let (tenant, ctx) = harness.load().await;
+
+        fn get_key(id: u32) -> Key {
+            // using aux key here b/c they are guaranteed to be inside `collect_keyspace`.
+            let mut key = Key::from_hex("620000000033333333444444445500000000").unwrap();
+            key.field6 = id;
+            key
+        }
+
+        // We create one bottom-most image layer, a delta layer D1 crossing the GC horizon, D2 below the horizon, and D3 above the horizon.
+        //
+        //  | D1 |                       | D3 |
+        // -|    |-- gc horizon -----------------
+        //  |    |                | D2 |
+        // --------- img layer ------------------
+        //
+        // What we should expact from this compaction is:
+        //  | Part of D1 |               | D3 |
+        // --------- img layer with D1+D2 at GC horizon------------------
+
+        // img layer at 0x10
+        let img_layer = (0..10)
+            .map(|id| (get_key(id), test_img(&format!("value {id}@0x10"))))
+            .collect_vec();
+
+        let delta1 = vec![
+            // TODO: we should test a real delta record here, which requires us to add a variant of NeonWalRecord for testing purpose.
+            (
+                get_key(1),
+                Lsn(0x20),
+                Value::Image(test_img("value 1@0x20")),
+            ),
+            (
+                get_key(2),
+                Lsn(0x30),
+                Value::Image(test_img("value 2@0x30")),
+            ),
+            (
+                get_key(3),
+                Lsn(0x40),
+                Value::Image(test_img("value 3@0x40")),
+            ),
+        ];
+        let delta2 = vec![
+            (
+                get_key(5),
+                Lsn(0x20),
+                Value::Image(test_img("value 5@0x20")),
+            ),
+            (
+                get_key(6),
+                Lsn(0x20),
+                Value::Image(test_img("value 6@0x20")),
+            ),
+        ];
+        let delta3 = vec![
+            (
+                get_key(8),
+                Lsn(0x40),
+                Value::Image(test_img("value 8@0x40")),
+            ),
+            (
+                get_key(9),
+                Lsn(0x40),
+                Value::Image(test_img("value 9@0x40")),
+            ),
+        ];
+
+        let tline = tenant
+            .create_test_timeline_with_layers(
+                TIMELINE_ID,
+                Lsn(0x10),
+                DEFAULT_PG_VERSION,
+                &ctx,
+                vec![delta1, delta2, delta3], // delta layers
+                vec![(Lsn(0x10), img_layer)], // image layers
+                Lsn(0x50),
+            )
+            .await?;
+        {
+            // Update GC info
+            let mut guard = tline.gc_info.write().unwrap();
+            guard.cutoffs.pitr = Lsn(0x30);
+            guard.cutoffs.horizon = Lsn(0x30);
+        }
+
+        let cancel = CancellationToken::new();
+        tline.compact_with_gc(&cancel, &ctx).await.unwrap();
+
+        // 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)
+            .await
+            .unwrap()
+            .into_iter()
+            .filter(|(k, _)| k.is_metadata_key())
+            .collect::<Vec<_>>();
+
+        assert_eq!(image_at_gc_horizon.len(), 10);
+        let expected_lsn = [0x10, 0x20, 0x30, 0x10, 0x10, 0x20, 0x20, 0x10, 0x10, 0x10];
+        for idx in 0..10 {
+            assert_eq!(
+                image_at_gc_horizon[idx],
+                (
+                    get_key(idx as u32),
+                    test_img(&format!("value {idx}@{:#x}", expected_lsn[idx]))
+                )
+            );
+        }
+
+        // Check if old layers are removed / new layers have the expected LSN
+        let mut all_layers = tline.inspect_historic_layers().await.unwrap();
+        all_layers.sort_by(|k1, k2| {
+            (
+                k1.is_delta,
+                k1.key_range.start,
+                k1.key_range.end,
+                k1.lsn_range.start,
+                k1.lsn_range.end,
+            )
+                .cmp(&(
+                    k2.is_delta,
+                    k2.key_range.start,
+                    k2.key_range.end,
+                    k2.lsn_range.start,
+                    k2.lsn_range.end,
+                ))
+        });
+        assert_eq!(
+            all_layers,
+            vec![
+                // Image layer at GC horizon
+                PersistentLayerKey {
+                    key_range: Key::MIN..get_key(10),
+                    lsn_range: Lsn(0x30)..Lsn(0x31),
+                    is_delta: false
+                },
+                // The delta layer that is cut in the middle
+                PersistentLayerKey {
+                    key_range: Key::MIN..get_key(9),
+                    lsn_range: Lsn(0x30)..Lsn(0x41),
+                    is_delta: true
+                },
+                // The delta layer we created and should not be picked for the compaction
+                PersistentLayerKey {
+                    key_range: get_key(8)..get_key(10),
+                    lsn_range: Lsn(0x40)..Lsn(0x41),
+                    is_delta: true
+                }
+            ]
+        );
+
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn test_neon_test_record() -> anyhow::Result<()> {
+        let harness = TenantHarness::create("test_neon_test_record")?;
+        let (tenant, ctx) = harness.load().await;
+
+        fn get_key(id: u32) -> Key {
+            // using aux key here b/c they are guaranteed to be inside `collect_keyspace`.
+            let mut key = Key::from_hex("620000000033333333444444445500000000").unwrap();
+            key.field6 = id;
+            key
+        }
+
+        let delta1 = vec![
+            (
+                get_key(1),
+                Lsn(0x20),
+                Value::WalRecord(NeonWalRecord::wal_append(",0x20")),
+            ),
+            (
+                get_key(1),
+                Lsn(0x30),
+                Value::WalRecord(NeonWalRecord::wal_append(",0x30")),
+            ),
+            (get_key(2), Lsn(0x10), Value::Image("0x10".into())),
+            (
+                get_key(2),
+                Lsn(0x20),
+                Value::WalRecord(NeonWalRecord::wal_append(",0x20")),
+            ),
+            (
+                get_key(2),
+                Lsn(0x30),
+                Value::WalRecord(NeonWalRecord::wal_append(",0x30")),
+            ),
+            (get_key(3), Lsn(0x10), Value::Image("0x10".into())),
+            (
+                get_key(3),
+                Lsn(0x20),
+                Value::WalRecord(NeonWalRecord::wal_clear()),
+            ),
+            (get_key(4), Lsn(0x10), Value::Image("0x10".into())),
+            (
+                get_key(4),
+                Lsn(0x20),
+                Value::WalRecord(NeonWalRecord::wal_init()),
+            ),
+        ];
+        let image1 = vec![(get_key(1), "0x10".into())];
+
+        let tline = tenant
+            .create_test_timeline_with_layers(
+                TIMELINE_ID,
+                Lsn(0x10),
+                DEFAULT_PG_VERSION,
+                &ctx,
+                vec![delta1],              // delta layers
+                vec![(Lsn(0x10), image1)], // image layers
+                Lsn(0x50),
+            )
+            .await?;
+
+        assert_eq!(
+            tline.get(get_key(1), Lsn(0x50), &ctx).await?,
+            Bytes::from_static(b"0x10,0x20,0x30")
+        );
+        assert_eq!(
+            tline.get(get_key(2), Lsn(0x50), &ctx).await?,
+            Bytes::from_static(b"0x10,0x20,0x30")
+        );
+        // assert_eq!(tline.get(get_key(3), Lsn(0x50), &ctx).await?, Bytes::new());
+        // assert_eq!(tline.get(get_key(4), Lsn(0x50), &ctx).await?, Bytes::new());
+
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn test_lsn_lease() -> anyhow::Result<()> {
+        let (tenant, ctx) = TenantHarness::create("test_lsn_lease")?.load().await;
+        let key = Key::from_hex("010000000033333333444444445500000000").unwrap();
+
+        let end_lsn = Lsn(0x100);
+        let image_layers = (0x20..=0x90)
+            .step_by(0x10)
+            .map(|n| {
+                (
+                    Lsn(n),
+                    vec![(key, test_img(&format!("data key at {:x}", n)))],
+                )
+            })
+            .collect();
+
+        let timeline = tenant
+            .create_test_timeline_with_layers(
+                TIMELINE_ID,
+                Lsn(0x10),
+                DEFAULT_PG_VERSION,
+                &ctx,
+                Vec::new(),
+                image_layers,
+                end_lsn,
+            )
+            .await?;
+
+        let leased_lsns = [0x30, 0x50, 0x70];
+        let mut leases = Vec::new();
+        let _: anyhow::Result<_> = leased_lsns.iter().try_for_each(|n| {
+            leases.push(timeline.make_lsn_lease(Lsn(*n), timeline.get_lsn_lease_length(), &ctx)?);
+            Ok(())
+        });
+
+        // Renewing with shorter lease should not change the lease.
+        let updated_lease_0 =
+            timeline.make_lsn_lease(Lsn(leased_lsns[0]), Duration::from_secs(0), &ctx)?;
+        assert_eq!(updated_lease_0.valid_until, leases[0].valid_until);
+
+        // Renewing with a long lease should renew lease with later expiration time.
+        let updated_lease_1 = timeline.make_lsn_lease(
+            Lsn(leased_lsns[1]),
+            timeline.get_lsn_lease_length() * 2,
+            &ctx,
+        )?;
+
+        assert!(updated_lease_1.valid_until > leases[1].valid_until);
+
+        // Force set disk consistent lsn so we can get the cutoff at `end_lsn`.
+        info!(
+            "latest_gc_cutoff_lsn: {}",
+            *timeline.get_latest_gc_cutoff_lsn()
+        );
+        timeline.force_set_disk_consistent_lsn(end_lsn);
+
+        let res = tenant
+            .gc_iteration(
+                Some(TIMELINE_ID),
+                0,
+                Duration::ZERO,
+                &CancellationToken::new(),
+                &ctx,
+            )
+            .await?;
+
+        // Keeping everything <= Lsn(0x80) b/c leases:
+        // 0/10: initdb layer
+        // (0/20..=0/70).step_by(0x10): image layers added when creating the timeline.
+        assert_eq!(res.layers_needed_by_leases, 7);
+        // Keeping 0/90 b/c it is the latest layer.
+        assert_eq!(res.layers_not_updated, 1);
+        // Removed 0/80.
+        assert_eq!(res.layers_removed, 1);
+
+        // Make lease on a already GC-ed LSN.
+        // 0/80 does not have a valid lease + is below latest_gc_cutoff
+        assert!(Lsn(0x80) < *timeline.get_latest_gc_cutoff_lsn());
+        let res = timeline.make_lsn_lease(Lsn(0x80), timeline.get_lsn_lease_length(), &ctx);
+        assert!(res.is_err());
+
+        // Should still be able to renew a currently valid lease
+        // Assumption: original lease to is still valid for 0/50.
+        let _ =
+            timeline.make_lsn_lease(Lsn(leased_lsns[1]), timeline.get_lsn_lease_length(), &ctx)?;
 
         Ok(())
     }

pageserver/src/tenant/blob_io.rs

@@ -26,7 +26,7 @@ impl<'a> BlockCursor<'a> {
     pub async fn read_blob(
         &self,
         offset: u64,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<Vec<u8>, std::io::Error> {
         let mut buf = Vec::new();
         self.read_blob_into_buf(offset, &mut buf, ctx).await?;
@@ -38,7 +38,7 @@ impl<'a> BlockCursor<'a> {
         &self,
         offset: u64,
         dstbuf: &mut Vec<u8>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<(), std::io::Error> {
         let mut blknum = (offset / PAGE_SZ as u64) as u32;
         let mut off = (offset % PAGE_SZ as u64) as usize;
@@ -130,7 +130,7 @@ impl<const BUFFERED: bool> BlobWriter<BUFFERED> {
     async fn write_all_unbuffered<B: BoundedBuf<Buf = Buf>, Buf: IoBuf + Send>(
         &mut self,
         src_buf: B,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> (B::Buf, Result<(), Error>) {
         let (src_buf, res) = self.inner.write_all(src_buf, ctx).await;
         let nbytes = match res {
@@ -143,7 +143,7 @@ impl<const BUFFERED: bool> BlobWriter<BUFFERED> {
 
     #[inline(always)]
     /// Flushes the internal buffer to the underlying `VirtualFile`.
-    pub async fn flush_buffer(&mut self, ctx: &RequestContext) -> Result<(), Error> {
+    pub async fn flush_buffer(&mut self, ctx: &mut RequestContext) -> Result<(), Error> {
         let buf = std::mem::take(&mut self.buf);
         let (mut buf, res) = self.inner.write_all(buf, ctx).await;
         res?;
@@ -166,7 +166,7 @@ impl<const BUFFERED: bool> BlobWriter<BUFFERED> {
     async fn write_all<B: BoundedBuf<Buf = Buf>, Buf: IoBuf + Send>(
         &mut self,
         src_buf: B,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> (B::Buf, Result<(), Error>) {
         if !BUFFERED {
             assert!(self.buf.is_empty());
@@ -218,7 +218,7 @@ impl<const BUFFERED: bool> BlobWriter<BUFFERED> {
     pub async fn write_blob<B: BoundedBuf<Buf = Buf>, Buf: IoBuf + Send>(
         &mut self,
         srcbuf: B,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> (B::Buf, Result<u64, Error>) {
         let offset = self.offset;
 
@@ -267,7 +267,7 @@ impl BlobWriter<true> {
     ///
     /// This function flushes the internal buffer before giving access
     /// to the underlying `VirtualFile`.
-    pub async fn into_inner(mut self, ctx: &RequestContext) -> Result<VirtualFile, Error> {
+    pub async fn into_inner(mut self, ctx: &mut RequestContext) -> Result<VirtualFile, Error> {
         self.flush_buffer(ctx).await?;
         Ok(self.inner)
     }

pageserver/src/tenant/block_io.rs

@@ -92,7 +92,7 @@ impl<'a> BlockReaderRef<'a> {
     async fn read_blk(
         &self,
         blknum: u32,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<BlockLease, std::io::Error> {
         use BlockReaderRef::*;
         match self {
@@ -150,7 +150,7 @@ impl<'a> BlockCursor<'a> {
     pub async fn read_blk(
         &self,
         blknum: u32,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<BlockLease, std::io::Error> {
         self.reader.read_blk(blknum, ctx).await
     }
@@ -177,7 +177,7 @@ impl<'a> FileBlockReader<'a> {
         &self,
         buf: PageWriteGuard<'static>,
         blkno: u32,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<PageWriteGuard<'static>, std::io::Error> {
         assert!(buf.len() == PAGE_SZ);
         self.file
@@ -192,7 +192,7 @@ impl<'a> FileBlockReader<'a> {
     pub async fn read_blk<'b>(
         &self,
         blknum: u32,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<BlockLease<'b>, std::io::Error> {
         let cache = page_cache::get();
         match cache

pageserver/src/tenant/config.rs

@@ -13,6 +13,7 @@ use pageserver_api::models::AuxFilePolicy;
 use pageserver_api::models::CompactionAlgorithm;
 use pageserver_api::models::CompactionAlgorithmSettings;
 use pageserver_api::models::EvictionPolicy;
+use pageserver_api::models::LsnLease;
 use pageserver_api::models::{self, ThrottleConfig};
 use pageserver_api::shard::{ShardCount, ShardIdentity, ShardNumber, ShardStripeSize};
 use serde::de::IntoDeserializer;
@@ -377,6 +378,16 @@ pub struct TenantConf {
     /// There is a `last_aux_file_policy` flag which gets persisted in `index_part.json` once the first aux
     /// file is written.
     pub switch_aux_file_policy: AuxFilePolicy,
+
+    /// The length for an explicit LSN lease request.
+    /// Layers needed to reconstruct pages at LSN will not be GC-ed during this interval.
+    #[serde(with = "humantime_serde")]
+    pub lsn_lease_length: Duration,
+
+    /// The length for an implicit LSN lease granted as part of `get_lsn_by_timestamp` request.
+    /// Layers needed to reconstruct pages at LSN will not be GC-ed during this interval.
+    #[serde(with = "humantime_serde")]
+    pub lsn_lease_length_for_ts: Duration,
 }
 
 /// Same as TenantConf, but this struct preserves the information about
@@ -476,6 +487,16 @@ pub struct TenantConfOpt {
     #[serde(skip_serializing_if = "Option::is_none")]
     #[serde(default)]
     pub switch_aux_file_policy: Option<AuxFilePolicy>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
+    #[serde(with = "humantime_serde")]
+    #[serde(default)]
+    pub lsn_lease_length: Option<Duration>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
+    #[serde(with = "humantime_serde")]
+    #[serde(default)]
+    pub lsn_lease_length_for_ts: Option<Duration>,
 }
 
 impl TenantConfOpt {
@@ -538,6 +559,12 @@ impl TenantConfOpt {
             switch_aux_file_policy: self
                 .switch_aux_file_policy
                 .unwrap_or(global_conf.switch_aux_file_policy),
+            lsn_lease_length: self
+                .lsn_lease_length
+                .unwrap_or(global_conf.lsn_lease_length),
+            lsn_lease_length_for_ts: self
+                .lsn_lease_length_for_ts
+                .unwrap_or(global_conf.lsn_lease_length_for_ts),
         }
     }
 }
@@ -582,6 +609,8 @@ impl Default for TenantConf {
             timeline_get_throttle: crate::tenant::throttle::Config::disabled(),
             image_layer_creation_check_threshold: DEFAULT_IMAGE_LAYER_CREATION_CHECK_THRESHOLD,
             switch_aux_file_policy: AuxFilePolicy::default_tenant_config(),
+            lsn_lease_length: LsnLease::DEFAULT_LENGTH,
+            lsn_lease_length_for_ts: LsnLease::DEFAULT_LENGTH_FOR_TS,
         }
     }
 }
@@ -657,6 +686,8 @@ impl From<TenantConfOpt> for models::TenantConfig {
             timeline_get_throttle: value.timeline_get_throttle.map(ThrottleConfig::from),
             image_layer_creation_check_threshold: value.image_layer_creation_check_threshold,
             switch_aux_file_policy: value.switch_aux_file_policy,
+            lsn_lease_length: value.lsn_lease_length.map(humantime),
+            lsn_lease_length_for_ts: value.lsn_lease_length_for_ts.map(humantime),
         }
     }
 }

pageserver/src/tenant/delete.rs

@@ -404,7 +404,7 @@ impl DeleteTenantFlow {
         tenant: &Arc<Tenant>,
         preload: Option<TenantPreload>,
         tenants: &'static std::sync::RwLock<TenantsMap>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<(), DeleteTenantError> {
         let (_, progress) = completion::channel();
 

pageserver/src/tenant/disk_btree.rs

@@ -22,7 +22,7 @@ use async_stream::try_stream;
 use byteorder::{ReadBytesExt, BE};
 use bytes::{BufMut, Bytes, BytesMut};
 use either::Either;
-use futures::Stream;
+use futures::{Stream, StreamExt};
 use hex;
 use std::{
     cmp::Ordering,
@@ -242,7 +242,7 @@ where
     ///
     /// Read the value for given key. Returns the value, or None if it doesn't exist.
     ///
-    pub async fn get(&self, search_key: &[u8; L], ctx: &RequestContext) -> Result<Option<u64>> {
+    pub async fn get(&self, search_key: &[u8; L], ctx: &mut RequestContext) -> Result<Option<u64>> {
         let mut result: Option<u64> = None;
         self.visit(
             search_key,
@@ -259,6 +259,16 @@ where
         Ok(result)
     }
 
+    pub fn iter<'a>(
+        &'a self,
+        start_key: &'a [u8; L],
+        ctx: &'a RequestContext,
+    ) -> DiskBtreeIterator<'a> {
+        DiskBtreeIterator {
+            stream: Box::pin(self.get_stream_from(start_key, ctx)),
+        }
+    }
+
     /// Return a stream which yields all key, value pairs from the index
     /// starting from the first key greater or equal to `start_key`.
     ///
@@ -268,7 +278,7 @@ where
     pub fn get_stream_from<'a>(
         &'a self,
         start_key: &'a [u8; L],
-        ctx: &'a RequestContext,
+        ctx: &'a mut RequestContext,
     ) -> impl Stream<Item = std::result::Result<(Vec<u8>, u64), DiskBtreeError>> + 'a {
         try_stream! {
             let mut stack = Vec::new();
@@ -353,7 +363,7 @@ where
         search_key: &[u8; L],
         dir: VisitDirection,
         mut visitor: V,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<bool>
     where
         V: FnMut(&[u8], u64) -> bool,
@@ -496,6 +506,19 @@ where
     }
 }
 
+pub struct DiskBtreeIterator<'a> {
+    #[allow(clippy::type_complexity)]
+    stream: std::pin::Pin<
+        Box<dyn Stream<Item = std::result::Result<(Vec<u8>, u64), DiskBtreeError>> + 'a>,
+    >,
+}
+
+impl<'a> DiskBtreeIterator<'a> {
+    pub async fn next(&mut self) -> Option<std::result::Result<(Vec<u8>, u64), DiskBtreeError>> {
+        self.stream.next().await
+    }
+}
+
 ///
 /// Public builder object, for creating a new tree.
 ///
@@ -1088,6 +1111,17 @@ pub(crate) mod tests {
                 == all_data.get(&u128::MAX).cloned()
         );
 
+        // Test iterator and get_stream API
+        let mut iter = reader.iter(&[0; 16], &ctx);
+        let mut cnt = 0;
+        while let Some(res) = iter.next().await {
+            let (key, val) = res?;
+            let key = u128::from_be_bytes(key.as_slice().try_into().unwrap());
+            assert_eq!(val, *all_data.get(&key).unwrap());
+            cnt += 1;
+        }
+        assert_eq!(cnt, all_data.len());
+
         Ok(())
     }
 

pageserver/src/tenant/ephemeral_file.rs

@@ -28,7 +28,7 @@ impl EphemeralFile {
         conf: &PageServerConf,
         tenant_shard_id: TenantShardId,
         timeline_id: TimelineId,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<EphemeralFile, io::Error> {
         static NEXT_FILENAME: AtomicU64 = AtomicU64::new(1);
         let filename_disambiguator =
@@ -68,7 +68,7 @@ impl EphemeralFile {
     pub(crate) async fn read_blk(
         &self,
         blknum: u32,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<BlockLease, io::Error> {
         self.rw.read_blk(blknum, ctx).await
     }
@@ -76,7 +76,7 @@ impl EphemeralFile {
     pub(crate) async fn write_blob(
         &mut self,
         srcbuf: &[u8],
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<u64, io::Error> {
         let pos = self.rw.bytes_written();
 

pageserver/src/tenant/ephemeral_file/page_caching.rs

@@ -38,7 +38,7 @@ impl RW {
     pub(crate) async fn write_all_borrowed(
         &mut self,
         srcbuf: &[u8],
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<usize, io::Error> {
         // It doesn't make sense to proactively fill the page cache on the Pageserver write path
         // because Compute is unlikely to access recently written data.
@@ -52,7 +52,7 @@ impl RW {
     pub(crate) async fn read_blk(
         &self,
         blknum: u32,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<BlockLease, io::Error> {
         match self.rw.read_blk(blknum).await? {
             zero_padded_read_write::ReadResult::NeedsReadFromWriter { writer } => {
@@ -138,7 +138,7 @@ impl crate::virtual_file::owned_buffers_io::write::OwnedAsyncWriter for PreWarmi
     >(
         &mut self,
         buf: B,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> std::io::Result<(usize, B::Buf)> {
         let buf = buf.slice(..);
         let saved_bounds = buf.bounds(); // save for reconstructing the Slice from iobuf after the IO is done

pageserver/src/tenant/ephemeral_file/zero_padded_read_write.rs

@@ -64,7 +64,7 @@ where
     pub async fn write_all_borrowed(
         &mut self,
         buf: &[u8],
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> std::io::Result<usize> {
         self.buffered_writer.write_buffered_borrowed(buf, ctx).await
     }

pageserver/src/tenant/layer_map.rs

@@ -850,7 +850,7 @@ impl LayerMap {
 
     /// debugging function to print out the contents of the layer map
     #[allow(unused)]
-    pub async fn dump(&self, verbose: bool, ctx: &RequestContext) -> Result<()> {
+    pub async fn dump(&self, verbose: bool, ctx: &mut RequestContext) -> Result<()> {
         println!("Begin dump LayerMap");
 
         println!("open_layer:");

pageserver/src/tenant/metadata.rs

@@ -1,15 +1,23 @@
-//! Every image of a certain timeline from [`crate::tenant::Tenant`]
-//! has a metadata that needs to be stored persistently.
+//! Describes the legacy now hopefully no longer modified per-timeline metadata stored in
+//! `index_part.json` managed by [`remote_timeline_client`]. For many tenants and their timelines,
+//! this struct and it's original serialization format is still needed because they were written a
+//! long time ago.
 //!
-//! Later, the file gets used in [`remote_timeline_client`] as a part of
-//! external storage import and export operations.
+//! Instead of changing and adding versioning to this, just change [`IndexPart`] with soft json
+//! versioning.
 //!
-//! The module contains all structs and related helper methods related to timeline metadata.
+//! To clean up this module we need to migrate all index_part.json files to a later version.
+//! While doing this, we need to be mindful about s3 based recovery as well, so it might take
+//! however long we keep the old versions to be able to delete the old code. After that, we can
+//! remove everything else than [`TimelineMetadataBodyV2`], rename it as `TimelineMetadata` and
+//! move it to `index.rs`. Before doing all of this, we need to keep the structures for backwards
+//! compatibility.
 //!
 //! [`remote_timeline_client`]: super::remote_timeline_client
+//! [`IndexPart`]: super::remote_timeline_client::index::IndexPart
 
 use anyhow::ensure;
-use serde::{de::Error, Deserialize, Serialize, Serializer};
+use serde::{Deserialize, Serialize};
 use utils::bin_ser::SerializeError;
 use utils::{bin_ser::BeSer, id::TimelineId, lsn::Lsn};
 
@@ -17,17 +25,37 @@ use utils::{bin_ser::BeSer, id::TimelineId, lsn::Lsn};
 const METADATA_FORMAT_VERSION: u16 = 4;
 
 /// Previous supported format versions.
+///
+/// In practice, none of these should remain, all are [`METADATA_FORMAT_VERSION`], but confirming
+/// that requires a scrubber run which is yet to be done.
 const METADATA_OLD_FORMAT_VERSION: u16 = 3;
 
-/// We assume that a write of up to METADATA_MAX_SIZE bytes is atomic.
+/// When the file existed on disk we assumed that a write of up to METADATA_MAX_SIZE bytes is atomic.
 ///
 /// This is the same assumption that PostgreSQL makes with the control file,
+///
 /// see PG_CONTROL_MAX_SAFE_SIZE
 const METADATA_MAX_SIZE: usize = 512;
 
-/// Metadata stored on disk for each timeline
+/// Legacy metadata stored as a component of `index_part.json` per timeline.
 ///
-/// The fields correspond to the values we hold in memory, in Timeline.
+/// Do not make new changes to this type or the module. In production, we have two different kinds
+/// of serializations of this type: bincode and json. Bincode version reflects what used to be
+/// stored on disk in earlier versions and does internal crc32 checksumming.
+///
+/// This type should not implement `serde::Serialize` or `serde::Deserialize` because there would
+/// be a confusion whether you want the old version ([`TimelineMetadata::from_bytes`]) or the modern
+/// as-exists in `index_part.json` ([`self::modern_serde`]).
+///
+/// ```compile_fail
+/// #[derive(serde::Serialize)]
+/// struct DoNotDoThis(pageserver::tenant::metadata::TimelineMetadata);
+/// ```
+///
+/// ```compile_fail
+/// #[derive(serde::Deserialize)]
+/// struct NeitherDoThis(pageserver::tenant::metadata::TimelineMetadata);
+/// ```
 #[derive(Debug, Clone, PartialEq, Eq)]
 pub struct TimelineMetadata {
     hdr: TimelineMetadataHeader,
@@ -40,6 +68,49 @@ struct TimelineMetadataHeader {
     size: u16,           // size of serialized metadata
     format_version: u16, // metadata format version (used for compatibility checks)
 }
+
+impl TryFrom<&TimelineMetadataBodyV2> for TimelineMetadataHeader {
+    type Error = Crc32CalculationFailed;
+
+    fn try_from(value: &TimelineMetadataBodyV2) -> Result<Self, Self::Error> {
+        #[derive(Default)]
+        struct Crc32Sink {
+            crc: u32,
+            count: usize,
+        }
+
+        impl std::io::Write for Crc32Sink {
+            fn write(&mut self, buf: &[u8]) -> std::io::Result<usize> {
+                self.crc = crc32c::crc32c_append(self.crc, buf);
+                self.count += buf.len();
+                Ok(buf.len())
+            }
+
+            fn flush(&mut self) -> std::io::Result<()> {
+                Ok(())
+            }
+        }
+
+        // jump through hoops to calculate the crc32 so that TimelineMetadata::ne works
+        // across serialization versions
+        let mut sink = Crc32Sink::default();
+        <TimelineMetadataBodyV2 as utils::bin_ser::BeSer>::ser_into(value, &mut sink)
+            .map_err(Crc32CalculationFailed)?;
+
+        let size = METADATA_HDR_SIZE + sink.count;
+
+        Ok(TimelineMetadataHeader {
+            checksum: sink.crc,
+            size: size as u16,
+            format_version: METADATA_FORMAT_VERSION,
+        })
+    }
+}
+
+#[derive(thiserror::Error, Debug)]
+#[error("re-serializing for crc32 failed")]
+struct Crc32CalculationFailed(#[source] utils::bin_ser::SerializeError);
+
 const METADATA_HDR_SIZE: usize = std::mem::size_of::<TimelineMetadataHeader>();
 
 #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
@@ -111,6 +182,12 @@ impl TimelineMetadata {
         }
     }
 
+    #[cfg(test)]
+    pub(crate) fn with_recalculated_checksum(mut self) -> anyhow::Result<Self> {
+        self.hdr = TimelineMetadataHeader::try_from(&self.body)?;
+        Ok(self)
+    }
+
     fn upgrade_timeline_metadata(metadata_bytes: &[u8]) -> anyhow::Result<Self> {
         let mut hdr = TimelineMetadataHeader::des(&metadata_bytes[0..METADATA_HDR_SIZE])?;
 
@@ -261,32 +338,8 @@ impl TimelineMetadata {
     }
 }
 
-impl<'de> Deserialize<'de> for TimelineMetadata {
-    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
-    where
-        D: serde::Deserializer<'de>,
-    {
-        let bytes = Vec::<u8>::deserialize(deserializer)?;
-        Self::from_bytes(bytes.as_slice()).map_err(D::Error::custom)
-    }
-}
-
-impl Serialize for TimelineMetadata {
-    fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
-    where
-        S: Serializer,
-    {
-        let bytes = self.to_bytes().map_err(serde::ser::Error::custom)?;
-        bytes.serialize(serializer)
-    }
-}
-
 pub(crate) mod modern_serde {
-    use crate::tenant::metadata::METADATA_FORMAT_VERSION;
-
-    use super::{
-        TimelineMetadata, TimelineMetadataBodyV2, TimelineMetadataHeader, METADATA_HDR_SIZE,
-    };
+    use super::{TimelineMetadata, TimelineMetadataBodyV2, TimelineMetadataHeader};
     use serde::{Deserialize, Serialize};
 
     pub(crate) fn deserialize<'de, D>(deserializer: D) -> Result<TimelineMetadata, D::Error>
@@ -322,71 +375,15 @@ pub(crate) mod modern_serde {
 
                 let de = serde::de::value::MapAccessDeserializer::new(map);
                 let body = TimelineMetadataBodyV2::deserialize(de)?;
+                let hdr = TimelineMetadataHeader::try_from(&body).map_err(A::Error::custom)?;
 
-                // jump through hoops to calculate the crc32 so that TimelineMetadata::ne works
-                // across serialization versions
-                let mut sink = Crc32Sink::default();
-                <TimelineMetadataBodyV2 as utils::bin_ser::BeSer>::ser_into(&body, &mut sink)
-                    .map_err(|e| A::Error::custom(Crc32CalculationFailed(e)))?;
-
-                let size = METADATA_HDR_SIZE + sink.count;
-
-                Ok(TimelineMetadata {
-                    hdr: TimelineMetadataHeader {
-                        checksum: sink.crc,
-                        size: size as u16,
-                        format_version: METADATA_FORMAT_VERSION,
-                    },
-                    body,
-                })
+                Ok(TimelineMetadata { hdr, body })
             }
         }
 
         deserializer.deserialize_any(Visitor)
     }
 
-    #[derive(Default)]
-    struct Crc32Sink {
-        crc: u32,
-        count: usize,
-    }
-
-    impl std::io::Write for Crc32Sink {
-        fn write(&mut self, buf: &[u8]) -> std::io::Result<usize> {
-            self.crc = crc32c::crc32c_append(self.crc, buf);
-            self.count += buf.len();
-            Ok(buf.len())
-        }
-
-        fn flush(&mut self) -> std::io::Result<()> {
-            Ok(())
-        }
-    }
-
-    #[derive(thiserror::Error)]
-    #[error("re-serializing for crc32 failed")]
-    struct Crc32CalculationFailed<E>(#[source] E);
-
-    // this should be true for one release, after that we can change it to false
-    // remember to check the IndexPart::metadata field TODO comment as well
-    const LEGACY_BINCODED_BYTES: bool = true;
-
-    #[derive(serde::Serialize)]
-    #[serde(transparent)]
-    struct LegacyPaddedBytes<'a>(&'a TimelineMetadata);
-
-    struct JustTheBodyV2<'a>(&'a TimelineMetadata);
-
-    impl serde::Serialize for JustTheBodyV2<'_> {
-        fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
-        where
-            S: serde::Serializer,
-        {
-            // header is not needed, upon reading we've upgraded all v1 to v2
-            self.0.body.serialize(serializer)
-        }
-    }
-
     pub(crate) fn serialize<S>(
         metadata: &TimelineMetadata,
         serializer: S,
@@ -394,25 +391,23 @@ pub(crate) mod modern_serde {
     where
         S: serde::Serializer,
     {
-        // we cannot use TimelineMetadata::serialize for now because it'll do
-        // TimelineMetadata::to_bytes
-        if LEGACY_BINCODED_BYTES {
-            LegacyPaddedBytes(metadata).serialize(serializer)
-        } else {
-            JustTheBodyV2(metadata).serialize(serializer)
-        }
+        // header is not needed, upon reading we've upgraded all v1 to v2
+        metadata.body.serialize(serializer)
     }
 
     #[test]
     fn deserializes_bytes_as_well_as_equivalent_body_v2() {
         #[derive(serde::Deserialize, serde::Serialize)]
-        struct Wrapper(#[serde(deserialize_with = "deserialize")] TimelineMetadata);
+        struct Wrapper(
+            #[serde(deserialize_with = "deserialize", serialize_with = "serialize")]
+            TimelineMetadata,
+        );
 
         let too_many_bytes = "[216,111,252,208,0,54,0,4,0,0,0,0,1,73,253,144,1,0,0,0,0,1,73,253,24,0,0,0,0,0,0,0,0,0,0,0,0,0,1,73,253,24,0,0,0,0,1,73,253,24,0,0,0,15,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]";
 
         let wrapper_from_bytes = serde_json::from_str::<Wrapper>(too_many_bytes).unwrap();
 
-        let serialized = serde_json::to_value(JustTheBodyV2(&wrapper_from_bytes.0)).unwrap();
+        let serialized = serde_json::to_value(&wrapper_from_bytes).unwrap();
 
         assert_eq!(
             serialized,
@@ -553,59 +548,6 @@ mod tests {
         );
     }
 
-    #[test]
-    fn test_metadata_bincode_serde() {
-        let original_metadata = TimelineMetadata::new(
-            Lsn(0x200),
-            Some(Lsn(0x100)),
-            Some(TIMELINE_ID),
-            Lsn(0),
-            Lsn(0),
-            Lsn(0),
-            // Any version will do here, so use the default
-            crate::DEFAULT_PG_VERSION,
-        );
-        let metadata_bytes = original_metadata
-            .to_bytes()
-            .expect("Cannot create bytes array from metadata");
-
-        let metadata_bincode_be_bytes = original_metadata
-            .ser()
-            .expect("Cannot serialize the metadata");
-
-        // 8 bytes for the length of the vector
-        assert_eq!(metadata_bincode_be_bytes.len(), 8 + metadata_bytes.len());
-
-        let expected_bincode_bytes = {
-            let mut temp = vec![];
-            let len_bytes = metadata_bytes.len().to_be_bytes();
-            temp.extend_from_slice(&len_bytes);
-            temp.extend_from_slice(&metadata_bytes);
-            temp
-        };
-        assert_eq!(metadata_bincode_be_bytes, expected_bincode_bytes);
-
-        let deserialized_metadata = TimelineMetadata::des(&metadata_bincode_be_bytes).unwrap();
-        // Deserialized metadata has the metadata header, which is different from the serialized one.
-        //   Reference: TimelineMetaData::to_bytes()
-        let expected_metadata = {
-            let mut temp_metadata = original_metadata;
-            let body_bytes = temp_metadata
-                .body
-                .ser()
-                .expect("Cannot serialize the metadata body");
-            let metadata_size = METADATA_HDR_SIZE + body_bytes.len();
-            let hdr = TimelineMetadataHeader {
-                size: metadata_size as u16,
-                format_version: METADATA_FORMAT_VERSION,
-                checksum: crc32c::crc32c(&body_bytes),
-            };
-            temp_metadata.hdr = hdr;
-            temp_metadata
-        };
-        assert_eq!(deserialized_metadata, expected_metadata);
-    }
-
     #[test]
     fn test_metadata_bincode_serde_ensure_roundtrip() {
         let original_metadata = TimelineMetadata::new(
@@ -619,8 +561,6 @@ mod tests {
             crate::DEFAULT_PG_VERSION,
         );
         let expected_bytes = vec![
-            /* bincode length encoding bytes */
-            0, 0, 0, 0, 0, 0, 2, 0, // 8 bytes for the length of the serialized vector
             /* TimelineMetadataHeader */
             4, 37, 101, 34, 0, 70, 0, 4, // checksum, size, format_version (4 + 2 + 2)
             /* TimelineMetadataBodyV2 */
@@ -650,7 +590,7 @@ mod tests {
             0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
             0, 0, 0, 0, 0, 0, 0,
         ];
-        let metadata_ser_bytes = original_metadata.ser().unwrap();
+        let metadata_ser_bytes = original_metadata.to_bytes().unwrap();
         assert_eq!(metadata_ser_bytes, expected_bytes);
 
         let expected_metadata = {
@@ -668,7 +608,7 @@ mod tests {
             temp_metadata.hdr = hdr;
             temp_metadata
         };
-        let des_metadata = TimelineMetadata::des(&metadata_ser_bytes).unwrap();
+        let des_metadata = TimelineMetadata::from_bytes(&metadata_ser_bytes).unwrap();
         assert_eq!(des_metadata, expected_metadata);
     }
 }

pageserver/src/tenant/mgr.rs

@@ -696,7 +696,7 @@ fn tenant_spawn(
     init_order: Option<InitializationOrder>,
     tenants: &'static std::sync::RwLock<TenantsMap>,
     mode: SpawnMode,
-    ctx: &RequestContext,
+    ctx: &mut RequestContext,
 ) -> anyhow::Result<Arc<Tenant>> {
     anyhow::ensure!(
         tenant_path.is_dir(),
@@ -956,7 +956,7 @@ impl TenantManager {
         new_location_config: LocationConf,
         flush: Option<Duration>,
         mut spawn_mode: SpawnMode,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<Option<Arc<Tenant>>, UpsertLocationError> {
         debug_assert_current_span_has_tenant_id();
         info!("configuring tenant location to state {new_location_config:?}");
@@ -1247,7 +1247,7 @@ impl TenantManager {
         &self,
         tenant_shard_id: TenantShardId,
         drop_cache: bool,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<()> {
         let mut slot_guard = tenant_map_acquire_slot(&tenant_shard_id, TenantSlotAcquireMode::Any)?;
         let Some(old_slot) = slot_guard.get_old_value() else {
@@ -1509,7 +1509,7 @@ impl TenantManager {
         tenant: Arc<Tenant>,
         new_shard_count: ShardCount,
         new_stripe_size: Option<ShardStripeSize>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<Vec<TenantShardId>> {
         let tenant_shard_id = *tenant.get_tenant_shard_id();
         let r = self
@@ -1539,7 +1539,7 @@ impl TenantManager {
         tenant: Arc<Tenant>,
         new_shard_count: ShardCount,
         new_stripe_size: Option<ShardStripeSize>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<Vec<TenantShardId>> {
         let tenant_shard_id = *tenant.get_tenant_shard_id();
 
@@ -1994,7 +1994,7 @@ impl TenantManager {
         tenant_shard_id: TenantShardId,
         timeline_id: TimelineId,
         prepared: PreparedTimelineDetach,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<Vec<TimelineId>, anyhow::Error> {
         struct RevertOnDropSlot(Option<SlotGuard>);
 
@@ -2229,7 +2229,7 @@ pub(crate) async fn load_tenant(
     broker_client: storage_broker::BrokerClientChannel,
     remote_storage: GenericRemoteStorage,
     deletion_queue_client: DeletionQueueClient,
-    ctx: &RequestContext,
+    ctx: &mut RequestContext,
 ) -> Result<(), TenantMapInsertError> {
     // This is a legacy API (replaced by `/location_conf`).  It does not support sharding
     let tenant_shard_id = TenantShardId::unsharded(tenant_id);
@@ -2837,7 +2837,7 @@ pub(crate) async fn immediate_gc(
     timeline_id: TimelineId,
     gc_req: TimelineGcRequest,
     cancel: CancellationToken,
-    ctx: &RequestContext,
+    ctx: &mut RequestContext,
 ) -> Result<GcResult, ApiError> {
     let tenant = {
         let guard = TENANTS.read().unwrap();

pageserver/src/tenant/remote_timeline_client.rs

@@ -518,7 +518,7 @@ impl RemoteTimelineClient {
         layer_metadata: &LayerFileMetadata,
         local_path: &Utf8Path,
         cancel: &CancellationToken,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<u64> {
         let downloaded_size = {
             let _unfinished_gauge_guard = self.metrics.call_begin(

pageserver/src/tenant/remote_timeline_client/download.rs

@@ -52,7 +52,7 @@ pub async fn download_layer_file<'a>(
     layer_metadata: &'a LayerFileMetadata,
     local_path: &Utf8Path,
     cancel: &CancellationToken,
-    ctx: &RequestContext,
+    ctx: &mut RequestContext,
 ) -> Result<u64, DownloadError> {
     debug_assert_current_span_has_tenant_and_timeline_id();
 
@@ -78,8 +78,33 @@ pub async fn download_layer_file<'a>(
     // If pageserver crashes the temp file will be deleted on startup and re-downloaded.
     let temp_file_path = path_with_suffix_extension(local_path, TEMP_DOWNLOAD_EXTENSION);
 
+    struct DownloadObjectClosure<'a> {
+        storage: &'a GenericRemoteStorage,
+        remote_path: &'a RemotePath,
+        temp_file_path: &'a Utf8PathBuf,
+        cancel: &'a CancellationToken,
+        ctx: &'a mut RequestContext,
+    }
+    impl backoff::Op<u64, DownloadError> for DownloadObjectClosure<'_> {
+        async fn call(&mut self) -> Result<u64, DownloadError> {
+            let DownloadObjectClosure {
+                storage,
+                remote_path,
+                temp_file_path,
+                cancel,
+                ctx,
+            } = self;
+            download_object(storage, remote_path, temp_file_path, cancel, ctx).await
+        }
+    }
     let bytes_amount = download_retry(
-        || async { download_object(storage, &remote_path, &temp_file_path, cancel, ctx).await },
+        DownloadObjectClosure {
+            storage,
+            remote_path: &remote_path,
+            temp_file_path: &temp_file_path,
+            cancel,
+            ctx,
+        },
         &format!("download {remote_path:?}"),
         cancel,
     )
@@ -107,9 +132,9 @@ pub async fn download_layer_file<'a>(
     // the in-memory state of the filesystem already has the layer file in its final place,
     // and subsequent pageserver code could think it's durable while it really isn't.
     let work = {
-        let ctx = ctx.detached_child(ctx.task_kind(), ctx.download_behavior());
+        let mut ctx = ctx.detached_child(ctx.task_kind(), ctx.download_behavior());
         async move {
-            let timeline_dir = VirtualFile::open(&timeline_path, &ctx)
+            let timeline_dir = VirtualFile::open(&timeline_path, &mut ctx)
                 .await
                 .fatal_err("VirtualFile::open for timeline dir fsync");
             timeline_dir
@@ -140,7 +165,7 @@ async fn download_object<'a>(
     src_path: &RemotePath,
     dst_path: &Utf8PathBuf,
     cancel: &CancellationToken,
-    #[cfg_attr(target_os = "macos", allow(unused_variables))] ctx: &RequestContext,
+    #[cfg_attr(target_os = "macos", allow(unused_variables))] ctx: &mut RequestContext,
 ) -> Result<u64, DownloadError> {
     let res = match crate::virtual_file::io_engine::get() {
         crate::virtual_file::io_engine::IoEngine::NotSet => panic!("unset"),
@@ -568,15 +593,11 @@ pub(crate) async fn download_initdb_tar_zst(
 /// with backoff.
 ///
 /// (See similar logic for uploads in `perform_upload_task`)
-pub(super) async fn download_retry<T, O, F>(
-    op: O,
+pub(super) async fn download_retry<T>(
+    op: impl backoff::Op<T, DownloadError>,
     description: &str,
     cancel: &CancellationToken,
-) -> Result<T, DownloadError>
-where
-    O: FnMut() -> F,
-    F: Future<Output = Result<T, DownloadError>>,
-{
+) -> Result<T, DownloadError> {
     backoff::retry(
         op,
         DownloadError::is_permanent,

pageserver/src/tenant/remote_timeline_client/index.rs

@@ -38,14 +38,17 @@ pub struct IndexPart {
     /// that latest version stores.
     pub layer_metadata: HashMap<LayerName, LayerFileMetadata>,
 
-    // 'disk_consistent_lsn' is a copy of the 'disk_consistent_lsn' in the metadata.
-    // It's duplicated for convenience when reading the serialized structure, but is
-    // private because internally we would read from metadata instead.
+    /// Because of the trouble of eyeballing the legacy "metadata" field, we copied the
+    /// "disk_consistent_lsn" out. After version 7 this is no longer needed, but the name cannot be
+    /// reused.
     pub(super) disk_consistent_lsn: Lsn,
 
-    // TODO: later make this "rename" to "alias", rename field as "legacy_metadata"
+    // TODO: rename as "metadata" next week, keep the alias = "metadata_bytes", bump version Adding
+    // the "alias = metadata" was forgotten in #7693, so we have to use "rewrite = metadata_bytes"
+    // for backwards compatibility.
     #[serde(
         rename = "metadata_bytes",
+        alias = "metadata",
         with = "crate::tenant::metadata::modern_serde"
     )]
     pub metadata: TimelineMetadata,
@@ -76,10 +79,11 @@ impl IndexPart {
     /// - 4: timeline_layers is fully removed.
     /// - 5: lineage was added
     /// - 6: last_aux_file_policy is added.
-    const LATEST_VERSION: usize = 6;
+    /// - 7: metadata_bytes is no longer written, but still read
+    const LATEST_VERSION: usize = 7;
 
     // Versions we may see when reading from a bucket.
-    pub const KNOWN_VERSIONS: &'static [usize] = &[1, 2, 3, 4, 5, 6];
+    pub const KNOWN_VERSIONS: &'static [usize] = &[1, 2, 3, 4, 5, 6, 7];
 
     pub const FILE_NAME: &'static str = "index_part.json";
 
@@ -95,7 +99,7 @@ impl IndexPart {
         }
     }
 
-    pub fn get_version(&self) -> usize {
+    pub fn version(&self) -> usize {
         self.version
     }
 
@@ -217,9 +221,9 @@ impl Lineage {
 
 #[cfg(test)]
 mod tests {
-    use std::str::FromStr;
-
     use super::*;
+    use std::str::FromStr;
+    use utils::id::TimelineId;
 
     #[test]
     fn v1_indexpart_is_parsed() {
@@ -338,8 +342,7 @@ mod tests {
             ]),
             disk_consistent_lsn: "0/16960E8".parse::<Lsn>().unwrap(),
             metadata: TimelineMetadata::from_bytes(&[113,11,159,210,0,54,0,4,0,0,0,0,1,105,96,232,1,0,0,0,0,1,105,96,112,0,0,0,0,0,0,0,0,0,0,0,0,0,1,105,96,112,0,0,0,0,1,105,96,112,0,0,0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]).unwrap(),
-            deleted_at: Some(chrono::NaiveDateTime::parse_from_str(
-                "2023-07-31T09:00:00.123000000", "%Y-%m-%dT%H:%M:%S.%f").unwrap()),
+            deleted_at: Some(parse_naive_datetime("2023-07-31T09:00:00.123000000")),
             lineage: Lineage::default(),
             last_aux_file_policy: None,
         };
@@ -515,8 +518,7 @@ mod tests {
             ]),
             disk_consistent_lsn: "0/16960E8".parse::<Lsn>().unwrap(),
             metadata: TimelineMetadata::from_bytes(&[113,11,159,210,0,54,0,4,0,0,0,0,1,105,96,232,1,0,0,0,0,1,105,96,112,0,0,0,0,0,0,0,0,0,0,0,0,0,1,105,96,112,0,0,0,0,1,105,96,112,0,0,0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]).unwrap(),
-            deleted_at: Some(chrono::NaiveDateTime::parse_from_str(
-                "2023-07-31T09:00:00.123000000", "%Y-%m-%dT%H:%M:%S.%f").unwrap()),
+            deleted_at: Some(parse_naive_datetime("2023-07-31T09:00:00.123000000")),
             lineage: Lineage {
                 reparenting_history_truncated: false,
                 reparenting_history: vec![TimelineId::from_str("e1bfd8c633d713d279e6fcd2bcc15b6d").unwrap()],
@@ -529,6 +531,60 @@ mod tests {
         assert_eq!(part, expected);
     }
 
+    #[test]
+    fn v7_indexpart_is_parsed() {
+        let example = r#"{
+            "version": 7,
+            "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
+            },
+            "deleted_at": "2023-07-31T09:00:00.123"
+        }"#;
+
+        let expected = IndexPart {
+            version: 7,
+            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: Some(parse_naive_datetime("2023-07-31T09:00:00.123000000")),
+            lineage: Default::default(),
+            last_aux_file_policy: Default::default(),
+        };
+
+        let part = IndexPart::from_s3_bytes(example.as_bytes()).unwrap();
+        assert_eq!(part, expected);
+    }
+
     fn parse_naive_datetime(s: &str) -> NaiveDateTime {
         chrono::NaiveDateTime::parse_from_str(s, "%Y-%m-%dT%H:%M:%S.%f").unwrap()
     }

pageserver/src/tenant/secondary/downloader.rs

@@ -506,14 +506,14 @@ impl<'a> TenantDownloader<'a> {
         }
     }
 
-    async fn download(&self, ctx: &RequestContext) -> Result<(), UpdateError> {
+    async fn download(&self, ctx: &mut RequestContext) -> Result<(), UpdateError> {
         debug_assert_current_span_has_tenant_id();
 
         // For the duration of a download, we must hold the SecondaryTenant::gate, to ensure
         // cover our access to local storage.
         let Ok(_guard) = self.secondary_state.gate.enter() else {
             // Shutting down
-            return Ok(());
+            return Err(UpdateError::Cancelled);
         };
 
         let tenant_shard_id = self.secondary_state.get_tenant_shard_id();
@@ -831,7 +831,7 @@ impl<'a> TenantDownloader<'a> {
         &self,
         timeline: HeatMapTimeline,
         timeline_state: SecondaryDetailTimeline,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<(), UpdateError> {
         debug_assert_current_span_has_tenant_and_timeline_id();
         let tenant_shard_id = self.secondary_state.get_tenant_shard_id();
@@ -846,7 +846,7 @@ impl<'a> TenantDownloader<'a> {
         for layer in timeline.layers {
             if self.secondary_state.cancel.is_cancelled() {
                 tracing::debug!("Cancelled -- dropping out of layer loop");
-                return Ok(());
+                return Err(UpdateError::Cancelled);
             }
 
             // Existing on-disk layers: just update their access time.
@@ -978,7 +978,7 @@ impl<'a> TenantDownloader<'a> {
         tenant_shard_id: &TenantShardId,
         timeline_id: &TimelineId,
         layer: HeatMapLayer,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<Option<HeatMapLayer>, UpdateError> {
         // Failpoint for simulating slow remote storage
         failpoint_support::sleep_millis_async!(

pageserver/src/tenant/size.rs

@@ -3,7 +3,6 @@ use std::collections::hash_map::Entry;
 use std::collections::{HashMap, HashSet};
 use std::sync::Arc;
 
-use anyhow::{bail, Context};
 use tokio::sync::oneshot::error::RecvError;
 use tokio::sync::Semaphore;
 use tokio_util::sync::CancellationToken;
@@ -11,7 +10,7 @@ use tokio_util::sync::CancellationToken;
 use crate::context::RequestContext;
 use crate::pgdatadir_mapping::CalculateLogicalSizeError;
 
-use super::{LogicalSizeCalculationCause, Tenant};
+use super::{GcError, LogicalSizeCalculationCause, Tenant};
 use crate::tenant::Timeline;
 use utils::id::TimelineId;
 use utils::lsn::Lsn;
@@ -43,6 +42,40 @@ pub struct SegmentMeta {
     pub kind: LsnKind,
 }
 
+#[derive(thiserror::Error, Debug)]
+pub(crate) enum CalculateSyntheticSizeError {
+    /// Something went wrong internally to the calculation of logical size at a particular branch point
+    #[error("Failed to calculated logical size on timeline {timeline_id} at {lsn}: {error}")]
+    LogicalSize {
+        timeline_id: TimelineId,
+        lsn: Lsn,
+        error: CalculateLogicalSizeError,
+    },
+
+    /// Something went wrong internally when calculating GC parameters at start of size calculation
+    #[error(transparent)]
+    GcInfo(GcError),
+
+    /// Totally unexpected errors, like panics joining a task
+    #[error(transparent)]
+    Fatal(anyhow::Error),
+
+    /// Tenant shut down while calculating size
+    #[error("Cancelled")]
+    Cancelled,
+}
+
+impl From<GcError> for CalculateSyntheticSizeError {
+    fn from(value: GcError) -> Self {
+        match value {
+            GcError::TenantCancelled | GcError::TimelineCancelled => {
+                CalculateSyntheticSizeError::Cancelled
+            }
+            other => CalculateSyntheticSizeError::GcInfo(other),
+        }
+    }
+}
+
 impl SegmentMeta {
     fn size_needed(&self) -> bool {
         match self.kind {
@@ -115,13 +148,10 @@ pub(super) async fn gather_inputs(
     logical_size_cache: &mut HashMap<(TimelineId, Lsn), u64>,
     cause: LogicalSizeCalculationCause,
     cancel: &CancellationToken,
-    ctx: &RequestContext,
-) -> anyhow::Result<ModelInputs> {
+    ctx: &mut RequestContext,
+) -> Result<ModelInputs, CalculateSyntheticSizeError> {
     // refresh is needed to update gc related pitr_cutoff and horizon_cutoff
-    tenant
-        .refresh_gc_info(cancel, ctx)
-        .await
-        .context("Failed to refresh gc_info before gathering inputs")?;
+    tenant.refresh_gc_info(cancel, ctx).await?;
 
     // Collect information about all the timelines
     let mut timelines = tenant.list_timelines();
@@ -327,6 +357,12 @@ pub(super) async fn gather_inputs(
     )
     .await?;
 
+    if tenant.cancel.is_cancelled() {
+        // If we're shutting down, return an error rather than a sparse result that might include some
+        // timelines from before we started shutting down
+        return Err(CalculateSyntheticSizeError::Cancelled);
+    }
+
     Ok(ModelInputs {
         segments,
         timeline_inputs,
@@ -335,17 +371,16 @@ pub(super) async fn gather_inputs(
 
 /// Augment 'segments' with logical sizes
 ///
-/// this will probably conflict with on-demand downloaded layers, or at least force them all
-/// to be downloaded
-///
+/// This will leave segments' sizes as None if the Timeline associated with the segment is deleted concurrently
+/// (i.e. we cannot read its logical size at a particular LSN).
 async fn fill_logical_sizes(
     timelines: &[Arc<Timeline>],
     segments: &mut [SegmentMeta],
     limit: &Arc<Semaphore>,
     logical_size_cache: &mut HashMap<(TimelineId, Lsn), u64>,
     cause: LogicalSizeCalculationCause,
-    ctx: &RequestContext,
-) -> anyhow::Result<()> {
+    ctx: &mut RequestContext,
+) -> Result<(), CalculateSyntheticSizeError> {
     let timeline_hash: HashMap<TimelineId, Arc<Timeline>> = HashMap::from_iter(
         timelines
             .iter()
@@ -387,7 +422,7 @@ async fn fill_logical_sizes(
     }
 
     // Perform the size lookups
-    let mut have_any_error = false;
+    let mut have_any_error = None;
     while let Some(res) = joinset.join_next().await {
         // each of these come with Result<anyhow::Result<_>, JoinError>
         // because of spawn + spawn_blocking
@@ -398,21 +433,36 @@ async fn fill_logical_sizes(
             Err(join_error) => {
                 // cannot really do anything, as this panic is likely a bug
                 error!("task that calls spawn_ondemand_logical_size_calculation panicked: {join_error:#}");
-                have_any_error = true;
+
+                have_any_error = Some(CalculateSyntheticSizeError::Fatal(
+                    anyhow::anyhow!(join_error)
+                        .context("task that calls spawn_ondemand_logical_size_calculation"),
+                ));
             }
             Ok(Err(recv_result_error)) => {
                 // cannot really do anything, as this panic is likely a bug
                 error!("failed to receive logical size query result: {recv_result_error:#}");
-                have_any_error = true;
+                have_any_error = Some(CalculateSyntheticSizeError::Fatal(
+                    anyhow::anyhow!(recv_result_error)
+                        .context("Receiving logical size query result"),
+                ));
             }
             Ok(Ok(TimelineAtLsnSizeResult(timeline, lsn, Err(error)))) => {
-                if !matches!(error, CalculateLogicalSizeError::Cancelled) {
+                if matches!(error, CalculateLogicalSizeError::Cancelled) {
+                    // Skip this: it's okay if one timeline among many is shutting down while we
+                    // calculate inputs for the overall tenant.
+                    continue;
+                } else {
                     warn!(
                         timeline_id=%timeline.timeline_id,
                         "failed to calculate logical size at {lsn}: {error:#}"
                     );
+                    have_any_error = Some(CalculateSyntheticSizeError::LogicalSize {
+                        timeline_id: timeline.timeline_id,
+                        lsn,
+                        error,
+                    });
                 }
-                have_any_error = true;
             }
             Ok(Ok(TimelineAtLsnSizeResult(timeline, lsn, Ok(size)))) => {
                 debug!(timeline_id=%timeline.timeline_id, %lsn, size, "size calculated");
@@ -426,10 +476,10 @@ async fn fill_logical_sizes(
     // prune any keys not needed anymore; we record every used key and added key.
     logical_size_cache.retain(|key, _| sizes_needed.contains_key(key));
 
-    if have_any_error {
+    if let Some(error) = have_any_error {
         // we cannot complete this round, because we are missing data.
         // we have however cached all we were able to request calculation on.
-        anyhow::bail!("failed to calculate some logical_sizes");
+        return Err(error);
     }
 
     // Insert the looked up sizes to the Segments
@@ -443,33 +493,28 @@ async fn fill_logical_sizes(
 
         if let Some(Some(size)) = sizes_needed.get(&(timeline_id, lsn)) {
             seg.segment.size = Some(*size);
-        } else {
-            bail!("could not find size at {} in timeline {}", lsn, timeline_id);
         }
     }
     Ok(())
 }
 
 impl ModelInputs {
-    pub fn calculate_model(&self) -> anyhow::Result<tenant_size_model::StorageModel> {
+    pub fn calculate_model(&self) -> tenant_size_model::StorageModel {
         // Convert SegmentMetas into plain Segments
-        let storage = StorageModel {
+        StorageModel {
             segments: self
                 .segments
                 .iter()
                 .map(|seg| seg.segment.clone())
                 .collect(),
-        };
-
-        Ok(storage)
+        }
     }
 
     // calculate total project size
-    pub fn calculate(&self) -> anyhow::Result<u64> {
-        let storage = self.calculate_model()?;
+    pub fn calculate(&self) -> u64 {
+        let storage = self.calculate_model();
         let sizes = storage.calculate();
-
-        Ok(sizes.total_size)
+        sizes.total_size
     }
 }
 
@@ -656,7 +701,7 @@ fn verify_size_for_multiple_branches() {
 "#;
     let inputs: ModelInputs = serde_json::from_str(doc).unwrap();
 
-    assert_eq!(inputs.calculate().unwrap(), 37_851_408);
+    assert_eq!(inputs.calculate(), 37_851_408);
 }
 
 #[test]
@@ -711,7 +756,7 @@ fn verify_size_for_one_branch() {
 
     let model: ModelInputs = serde_json::from_str(doc).unwrap();
 
-    let res = model.calculate_model().unwrap().calculate();
+    let res = model.calculate_model().calculate();
 
     println!("calculated synthetic size: {}", res.total_size);
     println!("result: {:?}", serde_json::to_string(&res.segments));

pageserver/src/tenant/storage_layer.rs

@@ -318,7 +318,7 @@ pub(crate) struct LayerFringe {
 #[derive(Debug)]
 struct LayerKeyspace {
     layer: ReadableLayer,
-    target_keyspace: Vec<KeySpace>,
+    target_keyspace: KeySpaceRandomAccum,
 }
 
 impl LayerFringe {
@@ -342,17 +342,13 @@ impl LayerFringe {
                 _,
                 LayerKeyspace {
                     layer,
-                    target_keyspace,
+                    mut target_keyspace,
                 },
-            )) => {
-                let mut keyspace = KeySpaceRandomAccum::new();
-                for ks in target_keyspace {
-                    for part in ks.ranges {
-                        keyspace.add_range(part);
-                    }
-                }
-                Some((layer, keyspace.consume_keyspace(), read_desc.lsn_range))
-            }
+            )) => Some((
+                layer,
+                target_keyspace.consume_keyspace(),
+                read_desc.lsn_range,
+            )),
             None => unreachable!("fringe internals are always consistent"),
         }
     }
@@ -367,16 +363,18 @@ impl LayerFringe {
         let entry = self.layers.entry(layer_id.clone());
         match entry {
             Entry::Occupied(mut entry) => {
-                entry.get_mut().target_keyspace.push(keyspace);
+                entry.get_mut().target_keyspace.add_keyspace(keyspace);
             }
             Entry::Vacant(entry) => {
                 self.planned_reads_by_lsn.push(ReadDesc {
                     lsn_range,
                     layer_id: layer_id.clone(),
                 });
+                let mut accum = KeySpaceRandomAccum::new();
+                accum.add_keyspace(keyspace);
                 entry.insert(LayerKeyspace {
                     layer,
-                    target_keyspace: vec![keyspace],
+                    target_keyspace: accum,
                 });
             }
         }
@@ -427,7 +425,7 @@ impl ReadableLayer {
         keyspace: KeySpace,
         lsn_range: Range<Lsn>,
         reconstruct_state: &mut ValuesReconstructState,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<(), GetVectoredError> {
         match self {
             ReadableLayer::PersistentLayer(layer) => {
@@ -576,7 +574,7 @@ impl LayerAccessStats {
         });
     }
 
-    fn record_access(&self, access_kind: LayerAccessKind, ctx: &RequestContext) {
+    fn record_access(&self, access_kind: LayerAccessKind, ctx: &mut RequestContext) {
         if ctx.access_stats_behavior() == AccessStatsBehavior::Skip {
             return;
         }

pageserver/src/tenant/storage_layer/delta_layer.rs

@@ -219,7 +219,6 @@ pub struct DeltaLayerInner {
     // values copied from summary
     index_start_blk: u32,
     index_root_blk: u32,
-    lsn_range: Range<Lsn>,
 
     file: VirtualFile,
     file_id: FileId,
@@ -250,7 +249,7 @@ impl AsLayerDesc for DeltaLayer {
 }
 
 impl DeltaLayer {
-    pub(crate) async fn dump(&self, verbose: bool, ctx: &RequestContext) -> Result<()> {
+    pub(crate) async fn dump(&self, verbose: bool, ctx: &mut RequestContext) -> Result<()> {
         self.desc.dump();
 
         if !verbose {
@@ -293,7 +292,7 @@ impl DeltaLayer {
     async fn load(
         &self,
         access_kind: LayerAccessKind,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<&Arc<DeltaLayerInner>> {
         self.access_stats.record_access(access_kind, ctx);
         // Quick exit if already loaded
@@ -303,7 +302,7 @@ impl DeltaLayer {
             .with_context(|| format!("Failed to load delta layer {}", self.path()))
     }
 
-    async fn load_inner(&self, ctx: &RequestContext) -> Result<Arc<DeltaLayerInner>> {
+    async fn load_inner(&self, ctx: &mut RequestContext) -> Result<Arc<DeltaLayerInner>> {
         let path = self.path();
 
         let loaded = DeltaLayerInner::load(&path, None, None, ctx)
@@ -394,7 +393,7 @@ impl DeltaLayerWriterInner {
         tenant_shard_id: TenantShardId,
         key_start: Key,
         lsn_range: Range<Lsn>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<Self> {
         // Create the file initially with a temporary filename. We don't know
         // the end key yet, so we cannot form the final filename yet. We will
@@ -436,7 +435,7 @@ impl DeltaLayerWriterInner {
         key: Key,
         lsn: Lsn,
         val: Value,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<()> {
         let (_, res) = self
             .put_value_bytes(key, lsn, Value::ser(&val)?, val.will_init(), ctx)
@@ -450,7 +449,7 @@ impl DeltaLayerWriterInner {
         lsn: Lsn,
         val: Vec<u8>,
         will_init: bool,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> (Vec<u8>, anyhow::Result<()>) {
         assert!(self.lsn_range.start <= lsn);
         let (val, res) = self.blob_writer.write_blob(val, ctx).await;
@@ -477,7 +476,7 @@ impl DeltaLayerWriterInner {
         self,
         key_end: Key,
         timeline: &Arc<Timeline>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<ResidentLayer> {
         let temp_path = self.path.clone();
         let result = self.finish0(key_end, timeline, ctx).await;
@@ -494,7 +493,7 @@ impl DeltaLayerWriterInner {
         self,
         key_end: Key,
         timeline: &Arc<Timeline>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<ResidentLayer> {
         let index_start_blk =
             ((self.blob_writer.size() + PAGE_SZ as u64 - 1) / PAGE_SZ as u64) as u32;
@@ -604,7 +603,7 @@ impl DeltaLayerWriter {
         tenant_shard_id: TenantShardId,
         key_start: Key,
         lsn_range: Range<Lsn>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<Self> {
         Ok(Self {
             inner: Some(
@@ -631,7 +630,7 @@ impl DeltaLayerWriter {
         key: Key,
         lsn: Lsn,
         val: Value,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<()> {
         self.inner
             .as_mut()
@@ -646,7 +645,7 @@ impl DeltaLayerWriter {
         lsn: Lsn,
         val: Vec<u8>,
         will_init: bool,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> (Vec<u8>, anyhow::Result<()>) {
         self.inner
             .as_mut()
@@ -666,7 +665,7 @@ impl DeltaLayerWriter {
         mut self,
         key_end: Key,
         timeline: &Arc<Timeline>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<ResidentLayer> {
         self.inner
             .take()
@@ -705,7 +704,7 @@ impl DeltaLayer {
     pub async fn rewrite_summary<F>(
         path: &Utf8Path,
         rewrite: F,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<(), RewriteSummaryError>
     where
         F: Fn(Summary) -> Summary,
@@ -745,7 +744,7 @@ impl DeltaLayerInner {
         path: &Utf8Path,
         summary: Option<Summary>,
         max_vectored_read_bytes: Option<MaxVectoredReadBytes>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<Result<Self, anyhow::Error>, anyhow::Error> {
         let file = match VirtualFile::open(path, ctx).await {
             Ok(file) => file,
@@ -785,7 +784,6 @@ impl DeltaLayerInner {
             file_id,
             index_start_blk: actual_summary.index_start_blk,
             index_root_blk: actual_summary.index_root_blk,
-            lsn_range: actual_summary.lsn_range,
             max_vectored_read_bytes,
         }))
     }
@@ -795,7 +793,7 @@ impl DeltaLayerInner {
         key: Key,
         lsn_range: Range<Lsn>,
         reconstruct_state: &mut ValueReconstructState,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<ValueReconstructResult> {
         let mut need_image = true;
         // Scan the page versions backwards, starting from `lsn`.
@@ -826,13 +824,13 @@ impl DeltaLayerInner {
 
                     !blob_ref.will_init()
                 },
-                &RequestContextBuilder::extend(ctx)
+                &mut RequestContextBuilder::extend(ctx)
                     .page_content_kind(PageContentKind::DeltaLayerBtreeNode)
                     .build(),
             )
             .await?;
 
-        let ctx = &RequestContextBuilder::extend(ctx)
+        let ctx = &mut RequestContextBuilder::extend(ctx)
             .page_content_kind(PageContentKind::DeltaLayerValue)
             .build();
 
@@ -891,7 +889,7 @@ impl DeltaLayerInner {
         keyspace: KeySpace,
         lsn_range: Range<Lsn>,
         reconstruct_state: &mut ValuesReconstructState,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<(), GetVectoredError> {
         let block_reader = FileBlockReader::new(&self.file, self.file_id);
         let index_reader = DiskBtreeReader::<_, DELTA_KEY_SIZE>::new(
@@ -911,7 +909,7 @@ impl DeltaLayerInner {
 
         let reads = Self::plan_reads(
             &keyspace,
-            lsn_range,
+            lsn_range.clone(),
             data_end_offset,
             index_reader,
             planner,
@@ -924,11 +922,50 @@ impl DeltaLayerInner {
         self.do_reads_and_update_state(reads, reconstruct_state, ctx)
             .await;
 
-        reconstruct_state.on_lsn_advanced(&keyspace, self.lsn_range.start);
+        reconstruct_state.on_lsn_advanced(&keyspace, lsn_range.start);
 
         Ok(())
     }
 
+    /// Load all key-values in the delta layer, should be replaced by an iterator-based interface in the future.
+    #[cfg(test)]
+    pub(super) async fn load_key_values(
+        &self,
+        ctx: &mut RequestContext,
+    ) -> anyhow::Result<Vec<(Key, Lsn, Value)>> {
+        let block_reader = FileBlockReader::new(&self.file, self.file_id);
+        let index_reader = DiskBtreeReader::<_, DELTA_KEY_SIZE>::new(
+            self.index_start_blk,
+            self.index_root_blk,
+            block_reader,
+        );
+        let mut result = Vec::new();
+        let mut stream =
+            Box::pin(self.stream_index_forwards(&index_reader, &[0; DELTA_KEY_SIZE], ctx));
+        let block_reader = FileBlockReader::new(&self.file, self.file_id);
+        let cursor = block_reader.block_cursor();
+        let mut buf = Vec::new();
+        while let Some(item) = stream.next().await {
+            let (key, lsn, pos) = item?;
+            // TODO: dedup code with get_reconstruct_value
+            // TODO: ctx handling and sharding
+            cursor
+                .read_blob_into_buf(pos.pos(), &mut buf, ctx)
+                .await
+                .with_context(|| {
+                    format!("Failed to read blob from virtual file {}", self.file.path)
+                })?;
+            let val = Value::des(&buf).with_context(|| {
+                format!(
+                    "Failed to deserialize file blob from virtual file {}",
+                    self.file.path
+                )
+            })?;
+            result.push((key, lsn, val));
+        }
+        Ok(result)
+    }
+
     async fn plan_reads<Reader>(
         keyspace: &KeySpace,
         lsn_range: Range<Lsn>,
@@ -936,12 +973,12 @@ impl DeltaLayerInner {
         index_reader: DiskBtreeReader<Reader, DELTA_KEY_SIZE>,
         mut planner: VectoredReadPlanner,
         reconstruct_state: &mut ValuesReconstructState,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<Vec<VectoredRead>>
     where
         Reader: BlockReader,
     {
-        let ctx = RequestContextBuilder::extend(ctx)
+        let mut ctx = RequestContextBuilder::extend(ctx)
             .page_content_kind(PageContentKind::DeltaLayerBtreeNode)
             .build();
 
@@ -949,7 +986,7 @@ impl DeltaLayerInner {
             let mut range_end_handled = false;
 
             let start_key = DeltaKey::from_key_lsn(&range.start, lsn_range.start);
-            let index_stream = index_reader.get_stream_from(&start_key.0, &ctx);
+            let index_stream = index_reader.get_stream_from(&start_key.0, &mut ctx);
             let mut index_stream = std::pin::pin!(index_stream);
 
             while let Some(index_entry) = index_stream.next().await {
@@ -1025,7 +1062,7 @@ impl DeltaLayerInner {
         &self,
         reads: Vec<VectoredRead>,
         reconstruct_state: &mut ValuesReconstructState,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) {
         let vectored_blob_reader = VectoredBlobReader::new(&self.file);
         let mut ignore_key_with_err = None;
@@ -1103,7 +1140,7 @@ impl DeltaLayerInner {
 
     pub(super) async fn load_keys<'a>(
         &'a self,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<Vec<DeltaEntry<'a>>> {
         let block_reader = FileBlockReader::new(&self.file, self.file_id);
         let tree_reader = DiskBtreeReader::<_, DELTA_KEY_SIZE>::new(
@@ -1142,7 +1179,7 @@ impl DeltaLayerInner {
                     all_keys.push(entry);
                     true
                 },
-                &RequestContextBuilder::extend(ctx)
+                &mut RequestContextBuilder::extend(ctx)
                     .page_content_kind(PageContentKind::DeltaLayerBtreeNode)
                     .build(),
             )
@@ -1162,7 +1199,7 @@ impl DeltaLayerInner {
         &self,
         writer: &mut DeltaLayerWriter,
         until: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<usize> {
         use crate::tenant::vectored_blob_io::{
             BlobMeta, VectoredReadBuilder, VectoredReadExtended,
@@ -1350,7 +1387,7 @@ impl DeltaLayerInner {
         Ok(records)
     }
 
-    pub(super) async fn dump(&self, ctx: &RequestContext) -> anyhow::Result<()> {
+    pub(super) async fn dump(&self, ctx: &mut RequestContext) -> anyhow::Result<()> {
         println!(
             "index_start_blk: {}, root {}",
             self.index_start_blk, self.index_root_blk
@@ -1367,7 +1404,7 @@ impl DeltaLayerInner {
 
         let keys = self.load_keys(ctx).await?;
 
-        async fn dump_blob(val: &ValueRef<'_>, ctx: &RequestContext) -> anyhow::Result<String> {
+        async fn dump_blob(val: &ValueRef<'_>, ctx: &mut RequestContext) -> anyhow::Result<String> {
             let buf = val.reader.read_blob(val.blob_ref.pos(), ctx).await?;
             let val = Value::des(&buf)?;
             let desc = match val {
@@ -1476,7 +1513,7 @@ pub struct ValueRef<'a> {
 
 impl<'a> ValueRef<'a> {
     /// Loads the value from disk
-    pub async fn load(&self, ctx: &RequestContext) -> Result<Value> {
+    pub async fn load(&self, ctx: &mut RequestContext) -> Result<Value> {
         // theoretically we *could* record an access time for each, but it does not really matter
         let buf = self.reader.read_blob(self.blob_ref.pos(), ctx).await?;
         let val = Value::des(&buf)?;
@@ -1490,7 +1527,7 @@ impl<T: AsRef<DeltaLayerInner>> Adapter<T> {
     pub(crate) async fn read_blk(
         &self,
         blknum: u32,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<BlockLease, std::io::Error> {
         let block_reader = FileBlockReader::new(&self.0.as_ref().file, self.0.as_ref().file_id);
         block_reader.read_blk(blknum, ctx).await
@@ -2023,7 +2060,7 @@ mod test {
         source: &DeltaLayerInner,
         truncated: &DeltaLayerInner,
         truncated_at: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) {
         use futures::future::ready;
         use futures::stream::TryStreamExt;

pageserver/src/tenant/storage_layer/image_layer.rs

@@ -177,7 +177,7 @@ impl std::fmt::Debug for ImageLayerInner {
 }
 
 impl ImageLayerInner {
-    pub(super) async fn dump(&self, ctx: &RequestContext) -> anyhow::Result<()> {
+    pub(super) async fn dump(&self, ctx: &mut RequestContext) -> anyhow::Result<()> {
         let block_reader = FileBlockReader::new(&self.file, self.file_id);
         let tree_reader = DiskBtreeReader::<_, KEY_SIZE>::new(
             self.index_start_blk,
@@ -217,7 +217,7 @@ impl AsLayerDesc for ImageLayer {
 }
 
 impl ImageLayer {
-    pub(crate) async fn dump(&self, verbose: bool, ctx: &RequestContext) -> Result<()> {
+    pub(crate) async fn dump(&self, verbose: bool, ctx: &mut RequestContext) -> Result<()> {
         self.desc.dump();
 
         if !verbose {
@@ -254,7 +254,7 @@ impl ImageLayer {
     async fn load(
         &self,
         access_kind: LayerAccessKind,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<&ImageLayerInner> {
         self.access_stats.record_access(access_kind, ctx);
         self.inner
@@ -263,7 +263,7 @@ impl ImageLayer {
             .with_context(|| format!("Failed to load image layer {}", self.path()))
     }
 
-    async fn load_inner(&self, ctx: &RequestContext) -> Result<ImageLayerInner> {
+    async fn load_inner(&self, ctx: &mut RequestContext) -> Result<ImageLayerInner> {
         let path = self.path();
 
         let loaded = ImageLayerInner::load(&path, self.desc.image_layer_lsn(), None, None, ctx)
@@ -336,7 +336,7 @@ impl ImageLayer {
     pub async fn rewrite_summary<F>(
         path: &Utf8Path,
         rewrite: F,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<(), RewriteSummaryError>
     where
         F: Fn(Summary) -> Summary,
@@ -377,7 +377,7 @@ impl ImageLayerInner {
         lsn: Lsn,
         summary: Option<Summary>,
         max_vectored_read_bytes: Option<MaxVectoredReadBytes>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<Result<Self, anyhow::Error>, anyhow::Error> {
         let file = match VirtualFile::open(path, ctx).await {
             Ok(file) => file,
@@ -428,7 +428,7 @@ impl ImageLayerInner {
         &self,
         key: Key,
         reconstruct_state: &mut ValueReconstructState,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<ValueReconstructResult> {
         let block_reader = FileBlockReader::new(&self.file, self.file_id);
         let tree_reader =
@@ -439,7 +439,7 @@ impl ImageLayerInner {
         if let Some(offset) = tree_reader
             .get(
                 &keybuf,
-                &RequestContextBuilder::extend(ctx)
+                &mut RequestContextBuilder::extend(ctx)
                     .page_content_kind(PageContentKind::ImageLayerBtreeNode)
                     .build(),
             )
@@ -449,7 +449,7 @@ impl ImageLayerInner {
                 .block_cursor()
                 .read_blob(
                     offset,
-                    &RequestContextBuilder::extend(ctx)
+                    &mut RequestContextBuilder::extend(ctx)
                         .page_content_kind(PageContentKind::ImageLayerValue)
                         .build(),
                 )
@@ -470,7 +470,7 @@ impl ImageLayerInner {
         &self,
         keyspace: KeySpace,
         reconstruct_state: &mut ValuesReconstructState,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<(), GetVectoredError> {
         let reads = self
             .plan_reads(keyspace, None, ctx)
@@ -485,6 +485,34 @@ impl ImageLayerInner {
         Ok(())
     }
 
+    /// Load all key-values in the delta layer, should be replaced by an iterator-based interface in the future.
+    #[cfg(test)]
+    pub(super) async fn load_key_values(
+        &self,
+        ctx: &mut RequestContext,
+    ) -> anyhow::Result<Vec<(Key, Lsn, Value)>> {
+        let block_reader = FileBlockReader::new(&self.file, self.file_id);
+        let tree_reader =
+            DiskBtreeReader::new(self.index_start_blk, self.index_root_blk, &block_reader);
+        let mut result = Vec::new();
+        let mut stream = Box::pin(tree_reader.get_stream_from(&[0; KEY_SIZE], ctx));
+        let block_reader = FileBlockReader::new(&self.file, self.file_id);
+        let cursor = block_reader.block_cursor();
+        while let Some(item) = stream.next().await {
+            // TODO: dedup code with get_reconstruct_value
+            let (raw_key, offset) = item?;
+            let key = Key::from_slice(&raw_key[..KEY_SIZE]);
+            // TODO: ctx handling and sharding
+            let blob = cursor
+                .read_blob(offset, ctx)
+                .await
+                .with_context(|| format!("failed to read value from offset {}", offset))?;
+            let value = Bytes::from(blob);
+            result.push((key, self.lsn, Value::Image(value)));
+        }
+        Ok(result)
+    }
+
     /// Traverse the layer's index to build read operations on the overlap of the input keyspace
     /// and the keys in this layer.
     ///
@@ -494,7 +522,7 @@ impl ImageLayerInner {
         &self,
         keyspace: KeySpace,
         shard_identity: Option<&ShardIdentity>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<Vec<VectoredRead>> {
         let mut planner = VectoredReadPlanner::new(
             self.max_vectored_read_bytes
@@ -507,7 +535,7 @@ impl ImageLayerInner {
         let tree_reader =
             DiskBtreeReader::new(self.index_start_blk, self.index_root_blk, block_reader);
 
-        let ctx = RequestContextBuilder::extend(ctx)
+        let mut ctx = RequestContextBuilder::extend(ctx)
             .page_content_kind(PageContentKind::ImageLayerBtreeNode)
             .build();
 
@@ -516,7 +544,7 @@ impl ImageLayerInner {
             let mut search_key: [u8; KEY_SIZE] = [0u8; KEY_SIZE];
             range.start.write_to_byte_slice(&mut search_key);
 
-            let index_stream = tree_reader.get_stream_from(&search_key, &ctx);
+            let index_stream = tree_reader.get_stream_from(&search_key, &mut ctx);
             let mut index_stream = std::pin::pin!(index_stream);
 
             while let Some(index_entry) = index_stream.next().await {
@@ -559,7 +587,7 @@ impl ImageLayerInner {
         &self,
         shard_identity: &ShardIdentity,
         writer: &mut ImageLayerWriter,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<usize> {
         // Fragment the range into the regions owned by this ShardIdentity
         let plan = self
@@ -601,7 +629,7 @@ impl ImageLayerInner {
         &self,
         reads: Vec<VectoredRead>,
         reconstruct_state: &mut ValuesReconstructState,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) {
         let max_vectored_read_bytes = self
             .max_vectored_read_bytes
@@ -696,7 +724,7 @@ impl ImageLayerWriterInner {
         tenant_shard_id: TenantShardId,
         key_range: &Range<Key>,
         lsn: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<Self> {
         // Create the file initially with a temporary filename.
         // We'll atomically rename it to the final name when we're done.
@@ -751,7 +779,7 @@ impl ImageLayerWriterInner {
         &mut self,
         key: Key,
         img: Bytes,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<()> {
         ensure!(self.key_range.contains(&key));
         let (_img, res) = self.blob_writer.write_blob(img, ctx).await;
@@ -771,7 +799,7 @@ impl ImageLayerWriterInner {
     async fn finish(
         self,
         timeline: &Arc<Timeline>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<ResidentLayer> {
         let index_start_blk =
             ((self.blob_writer.size() + PAGE_SZ as u64 - 1) / PAGE_SZ as u64) as u32;
@@ -871,7 +899,7 @@ impl ImageLayerWriter {
         tenant_shard_id: TenantShardId,
         key_range: &Range<Key>,
         lsn: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<ImageLayerWriter> {
         Ok(Self {
             inner: Some(
@@ -890,7 +918,7 @@ impl ImageLayerWriter {
         &mut self,
         key: Key,
         img: Bytes,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<()> {
         self.inner.as_mut().unwrap().put_image(key, img, ctx).await
     }
@@ -901,7 +929,7 @@ impl ImageLayerWriter {
     pub(crate) async fn finish(
         mut self,
         timeline: &Arc<Timeline>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<super::ResidentLayer> {
         self.inner.take().unwrap().finish(timeline, ctx).await
     }

pageserver/src/tenant/storage_layer/inmemory_layer.rs

@@ -52,7 +52,7 @@ pub struct InMemoryLayer {
 
     /// Frozen layers have an exclusive end LSN.
     /// Writes are only allowed when this is `None`.
-    end_lsn: OnceLock<Lsn>,
+    pub(crate) end_lsn: OnceLock<Lsn>,
 
     /// Used for traversal path. Cached representation of the in-memory layer before frozen.
     local_path_str: Arc<str>,
@@ -256,7 +256,7 @@ impl InMemoryLayer {
     /// debugging function to print out the contents of the layer
     ///
     /// this is likely completly unused
-    pub async fn dump(&self, verbose: bool, ctx: &RequestContext) -> Result<()> {
+    pub async fn dump(&self, verbose: bool, ctx: &mut RequestContext) -> Result<()> {
         let inner = self.inner.read().await;
 
         let end_str = self.end_lsn_or_max();
@@ -308,12 +308,12 @@ impl InMemoryLayer {
         key: Key,
         lsn_range: Range<Lsn>,
         reconstruct_state: &mut ValueReconstructState,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<ValueReconstructResult> {
         ensure!(lsn_range.start >= self.start_lsn);
         let mut need_image = true;
 
-        let ctx = RequestContextBuilder::extend(ctx)
+        let mut ctx = RequestContextBuilder::extend(ctx)
             .page_content_kind(PageContentKind::InMemoryLayer)
             .build();
 
@@ -325,7 +325,7 @@ impl InMemoryLayer {
         if let Some(vec_map) = inner.index.get(&key) {
             let slice = vec_map.slice_range(lsn_range);
             for (entry_lsn, pos) in slice.iter().rev() {
-                let buf = reader.read_blob(*pos, &ctx).await?;
+                let buf = reader.read_blob(*pos, &mut ctx).await?;
                 let value = Value::des(&buf)?;
                 match value {
                     Value::Image(img) => {
@@ -365,9 +365,9 @@ impl InMemoryLayer {
         keyspace: KeySpace,
         end_lsn: Lsn,
         reconstruct_state: &mut ValuesReconstructState,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<(), GetVectoredError> {
-        let ctx = RequestContextBuilder::extend(ctx)
+        let mut ctx = RequestContextBuilder::extend(ctx)
             .page_content_kind(PageContentKind::InMemoryLayer)
             .build();
 
@@ -410,7 +410,7 @@ impl InMemoryLayer {
                 continue;
             }
 
-            let buf = reader.read_blob(block_read.block_offset, &ctx).await;
+            let buf = reader.read_blob(block_read.block_offset, &mut ctx).await;
             if let Err(e) = buf {
                 reconstruct_state
                     .on_key_error(block_read.key, PageReconstructError::from(anyhow!(e)));
@@ -473,7 +473,7 @@ impl InMemoryLayer {
         timeline_id: TimelineId,
         tenant_shard_id: TenantShardId,
         start_lsn: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<InMemoryLayer> {
         trace!("initializing new empty InMemoryLayer for writing on timeline {timeline_id} at {start_lsn}");
 
@@ -512,7 +512,7 @@ impl InMemoryLayer {
         key: Key,
         lsn: Lsn,
         buf: &[u8],
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<()> {
         let mut inner = self.inner.write().await;
         self.assert_writable();
@@ -525,7 +525,7 @@ impl InMemoryLayer {
         key: Key,
         lsn: Lsn,
         buf: &[u8],
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<()> {
         trace!("put_value key {} at {}/{}", key, self.timeline_id, lsn);
 
@@ -534,7 +534,7 @@ impl InMemoryLayer {
                 .file
                 .write_blob(
                     buf,
-                    &RequestContextBuilder::extend(ctx)
+                    &mut RequestContextBuilder::extend(ctx)
                         .page_content_kind(PageContentKind::InMemoryLayer)
                         .build(),
                 )
@@ -606,7 +606,7 @@ impl InMemoryLayer {
     pub(crate) async fn write_to_disk(
         &self,
         timeline: &Arc<Timeline>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
         key_range: Option<Range<Key>>,
     ) -> Result<Option<ResidentLayer>> {
         // Grab the lock in read-mode. We hold it over the I/O, but because this

pageserver/src/tenant/storage_layer/layer.rs

@@ -331,7 +331,7 @@ impl Layer {
         key: Key,
         lsn_range: Range<Lsn>,
         reconstruct_data: &mut ValueReconstructState,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<ValueReconstructResult> {
         use anyhow::ensure;
 
@@ -361,7 +361,7 @@ impl Layer {
         keyspace: KeySpace,
         lsn_range: Range<Lsn>,
         reconstruct_data: &mut ValuesReconstructState,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<(), GetVectoredError> {
         let layer = self
             .0
@@ -388,6 +388,23 @@ impl Layer {
             })
     }
 
+    /// Get all key/values in the layer. Should be replaced with an iterator-based API in the future.
+    #[cfg(test)]
+    pub(crate) async fn load_key_values(
+        &self,
+        ctx: &mut RequestContext,
+    ) -> anyhow::Result<Vec<(Key, Lsn, crate::repository::Value)>> {
+        let layer = self
+            .0
+            .get_or_maybe_download(true, Some(ctx))
+            .await
+            .map_err(|err| match err {
+                DownloadError::DownloadCancelled => GetVectoredError::Cancelled,
+                other => GetVectoredError::Other(anyhow::anyhow!(other)),
+            })?;
+        layer.load_key_values(&self.0, ctx).await
+    }
+
     /// Download the layer if evicted.
     ///
     /// Will not error when the layer is already downloaded.
@@ -462,7 +479,7 @@ impl Layer {
 
     /// Traditional debug dumping facility
     #[allow(unused)]
-    pub(crate) async fn dump(&self, verbose: bool, ctx: &RequestContext) -> anyhow::Result<()> {
+    pub(crate) async fn dump(&self, verbose: bool, ctx: &mut RequestContext) -> anyhow::Result<()> {
         self.0.desc.dump();
 
         if verbose {
@@ -881,7 +898,7 @@ impl LayerInner {
     async fn get_or_maybe_download(
         self: &Arc<Self>,
         allow_download: bool,
-        ctx: Option<&RequestContext>,
+        ctx: Option<&mut RequestContext>,
     ) -> Result<Arc<DownloadedLayer>, DownloadError> {
         let (weak, permit) = {
             // get_or_init_detached can:
@@ -971,7 +988,7 @@ impl LayerInner {
             return Err(DownloadError::NotFile(ft));
         }
 
-        if let Some(ctx) = ctx {
+        if let Some(ref ctx) = ctx {
             self.check_expected_download(ctx)?;
         }
 
@@ -1032,7 +1049,7 @@ impl LayerInner {
         self: &Arc<Self>,
         timeline: Arc<Timeline>,
         permit: heavier_once_cell::InitPermit,
-        ctx: RequestContext,
+        mut ctx: RequestContext,
     ) -> Result<Arc<DownloadedLayer>, DownloadError> {
         debug_assert_current_span_has_tenant_and_timeline_id();
 
@@ -1062,7 +1079,7 @@ impl LayerInner {
                     .await
                     .unwrap();
 
-                let res = this.download_and_init(timeline, permit, &ctx).await;
+                let res = this.download_and_init(timeline, permit, &mut ctx).await;
 
                 if let Err(res) = tx.send(res) {
                     match res {
@@ -1105,7 +1122,7 @@ impl LayerInner {
         self: &Arc<LayerInner>,
         timeline: Arc<Timeline>,
         permit: heavier_once_cell::InitPermit,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<Arc<DownloadedLayer>> {
         let result = timeline
             .remote_client
@@ -1645,9 +1662,9 @@ impl DownloadedLayer {
     async fn get<'a>(
         &'a self,
         owner: &Arc<LayerInner>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<&'a LayerKind> {
-        let init = || async {
+        let init = async {
             assert_eq!(
                 Weak::as_ptr(&self.owner),
                 Arc::as_ptr(owner),
@@ -1702,7 +1719,7 @@ impl DownloadedLayer {
             }
         };
         self.kind
-            .get_or_try_init(init)
+            .get_or_try_init(move || init)
             // return transient errors using `?`
             .await?
             .as_ref()
@@ -1719,7 +1736,7 @@ impl DownloadedLayer {
         lsn_range: Range<Lsn>,
         reconstruct_data: &mut ValueReconstructState,
         owner: &Arc<LayerInner>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<ValueReconstructResult> {
         use LayerKind::*;
 
@@ -1741,7 +1758,7 @@ impl DownloadedLayer {
         lsn_range: Range<Lsn>,
         reconstruct_data: &mut ValuesReconstructState,
         owner: &Arc<LayerInner>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<(), GetVectoredError> {
         use LayerKind::*;
 
@@ -1757,7 +1774,21 @@ impl DownloadedLayer {
         }
     }
 
-    async fn dump(&self, owner: &Arc<LayerInner>, ctx: &RequestContext) -> anyhow::Result<()> {
+    #[cfg(test)]
+    async fn load_key_values(
+        &self,
+        owner: &Arc<LayerInner>,
+        ctx: &mut RequestContext,
+    ) -> anyhow::Result<Vec<(Key, Lsn, crate::repository::Value)>> {
+        use LayerKind::*;
+
+        match self.get(owner, ctx).await? {
+            Delta(d) => d.load_key_values(ctx).await,
+            Image(i) => i.load_key_values(ctx).await,
+        }
+    }
+
+    async fn dump(&self, owner: &Arc<LayerInner>, ctx: &mut RequestContext) -> anyhow::Result<()> {
         use LayerKind::*;
         match self.get(owner, ctx).await? {
             Delta(d) => d.dump(ctx).await?,
@@ -1806,7 +1837,7 @@ impl ResidentLayer {
     #[tracing::instrument(level = tracing::Level::DEBUG, skip_all, fields(layer=%self))]
     pub(crate) async fn load_keys<'a>(
         &'a self,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<Vec<DeltaEntry<'a>>> {
         use LayerKind::*;
 
@@ -1835,7 +1866,7 @@ impl ResidentLayer {
         &'a self,
         shard_identity: &ShardIdentity,
         writer: &mut ImageLayerWriter,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<usize> {
         use LayerKind::*;
 
@@ -1850,7 +1881,7 @@ impl ResidentLayer {
         &self,
         writer: &mut super::delta_layer::DeltaLayerWriter,
         until: Lsn,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<usize> {
         use LayerKind::*;
 
@@ -1876,7 +1907,7 @@ impl ResidentLayer {
     #[cfg(test)]
     pub(crate) async fn as_delta(
         &self,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<&delta_layer::DeltaLayerInner> {
         use LayerKind::*;
         match self.downloaded.get(&self.owner.0, ctx).await? {

pageserver/src/tenant/tasks.rs

@@ -73,7 +73,7 @@ static PERMIT_GAUGES: once_cell::sync::Lazy<
 /// Cancellation safe.
 pub(crate) async fn concurrent_background_tasks_rate_limit_permit(
     loop_kind: BackgroundLoopKind,
-    _ctx: &RequestContext,
+    _ctx: &mut RequestContext,
 ) -> tokio::sync::SemaphorePermit<'static> {
     let _guard = PERMIT_GAUGES[loop_kind].guard();
 
@@ -346,6 +346,7 @@ async fn gc_loop(tenant: Arc<Tenant>, cancel: CancellationToken) {
         // cutoff specified as time.
         let ctx =
             RequestContext::todo_child(TaskKind::GarbageCollector, DownloadBehavior::Download);
+
         let mut first = true;
         loop {
             tokio::select! {
@@ -362,6 +363,14 @@ async fn gc_loop(tenant: Arc<Tenant>, cancel: CancellationToken) {
 
             if first {
                 first = false;
+
+                if delay_by_lease_length(tenant.get_lsn_lease_length(), &cancel)
+                    .await
+                    .is_err()
+                {
+                    break;
+                }
+
                 if random_init_delay(period, &cancel).await.is_err() {
                     break;
                 }
@@ -531,6 +540,21 @@ pub(crate) async fn random_init_delay(
     }
 }
 
+/// Delays GC by defaul lease length at restart.
+///
+/// We do this as the leases mapping are not persisted to disk. By delaying GC by default
+/// length, we gurantees that all the leases we granted before the restart will expire
+/// when we run GC for the first time after the restart.
+pub(crate) async fn delay_by_lease_length(
+    length: Duration,
+    cancel: &CancellationToken,
+) -> Result<(), Cancelled> {
+    match tokio::time::timeout(length, cancel.cancelled()).await {
+        Ok(_) => Err(Cancelled),
+        Err(_) => Ok(()),
+    }
+}
+
 /// Attention: the `task` and `period` beocme labels of a pageserver-wide prometheus metric.
 pub(crate) fn warn_when_period_overrun(
     elapsed: Duration,

pageserver/src/tenant/throttle.rs

@@ -157,19 +157,6 @@ where
                 .fetch_add(wait_time.as_micros() as u64, Ordering::Relaxed);
             let observation = Observation { wait_time };
             self.metric.observe_throttling(&observation);
-            match ctx.micros_spent_throttled.add(wait_time) {
-                Ok(res) => res,
-                Err(error) => {
-                    use once_cell::sync::Lazy;
-                    use utils::rate_limit::RateLimit;
-                    static WARN_RATE_LIMIT: Lazy<Mutex<RateLimit>> =
-                        Lazy::new(|| Mutex::new(RateLimit::new(Duration::from_secs(10))));
-                    let mut guard = WARN_RATE_LIMIT.lock().unwrap();
-                    guard.call(move || {
-                        warn!(error, "error adding time spent throttled; this message is logged at a global rate limit");
-                    });
-                }
-            }
             Some(wait_time)
         } else {
             None

pageserver/src/tenant/timeline/analysis.rs

@@ -0,0 +1,90 @@
+use std::{collections::BTreeSet, ops::Range};
+
+use utils::lsn::Lsn;
+
+use super::Timeline;
+
+#[derive(serde::Serialize)]
+pub(crate) struct RangeAnalysis {
+    start: String,
+    end: String,
+    has_image: bool,
+    num_of_deltas_above_image: usize,
+    total_num_of_deltas: usize,
+}
+
+impl Timeline {
+    pub(crate) async fn perf_info(&self) -> Vec<RangeAnalysis> {
+        // First, collect all split points of the layers.
+        let mut split_points = BTreeSet::new();
+        let mut delta_ranges = Vec::new();
+        let mut image_ranges = Vec::new();
+
+        let all_layer_files = {
+            let guard = self.layers.read().await;
+            guard.all_persistent_layers()
+        };
+        let lsn = self.get_last_record_lsn();
+
+        for key in all_layer_files {
+            split_points.insert(key.key_range.start);
+            split_points.insert(key.key_range.end);
+            if key.is_delta {
+                delta_ranges.push((key.key_range.clone(), key.lsn_range.clone()));
+            } else {
+                image_ranges.push((key.key_range.clone(), key.lsn_range.start));
+            }
+        }
+
+        // For each split range, compute the estimated read amplification.
+        let split_points = split_points.into_iter().collect::<Vec<_>>();
+
+        let mut result = Vec::new();
+
+        for i in 0..(split_points.len() - 1) {
+            let start = split_points[i];
+            let end = split_points[i + 1];
+            // Find the latest image layer that contains the information.
+            let mut maybe_image_layers = image_ranges
+                .iter()
+                // We insert split points for all image layers, and therefore a `contains` check for the start point should be enough.
+                .filter(|(key_range, img_lsn)| key_range.contains(&start) && img_lsn <= &lsn)
+                .cloned()
+                .collect::<Vec<_>>();
+            maybe_image_layers.sort_by(|a, b| a.1.cmp(&b.1));
+            let image_layer = maybe_image_layers.last().cloned();
+            let lsn_filter_start = image_layer
+                .as_ref()
+                .map(|(_, lsn)| *lsn)
+                .unwrap_or(Lsn::INVALID);
+
+            fn overlaps_with(lsn_range_a: &Range<Lsn>, lsn_range_b: &Range<Lsn>) -> bool {
+                !(lsn_range_a.end <= lsn_range_b.start || lsn_range_a.start >= lsn_range_b.end)
+            }
+
+            let maybe_delta_layers = delta_ranges
+                .iter()
+                .filter(|(key_range, lsn_range)| {
+                    key_range.contains(&start) && overlaps_with(&(lsn_filter_start..lsn), lsn_range)
+                })
+                .cloned()
+                .collect::<Vec<_>>();
+
+            let pitr_delta_layers = delta_ranges
+                .iter()
+                .filter(|(key_range, _)| key_range.contains(&start))
+                .cloned()
+                .collect::<Vec<_>>();
+
+            result.push(RangeAnalysis {
+                start: start.to_string(),
+                end: end.to_string(),
+                has_image: image_layer.is_some(),
+                num_of_deltas_above_image: maybe_delta_layers.len(),
+                total_num_of_deltas: pitr_delta_layers.len(),
+            });
+        }
+
+        result
+    }
+}

pageserver/src/tenant/timeline/compaction.rs

@@ -49,7 +49,7 @@ impl Timeline {
         self: &Arc<Self>,
         _cancel: &CancellationToken,
         flags: EnumSet<CompactFlags>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<(), CompactionError> {
         // High level strategy for compaction / image creation:
         //
@@ -175,7 +175,7 @@ impl Timeline {
     async fn compact_shard_ancestors(
         self: &Arc<Self>,
         rewrite_max: usize,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<()> {
         let mut drop_layers = Vec::new();
         let mut layers_to_rewrite: Vec<Layer> = Vec::new();
@@ -359,7 +359,7 @@ impl Timeline {
     async fn compact_level0(
         self: &Arc<Self>,
         target_file_size: u64,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<(), CompactionError> {
         let CompactLevel0Phase1Result {
             new_layers,
@@ -400,7 +400,7 @@ impl Timeline {
         guard: tokio::sync::OwnedRwLockReadGuard<LayerManager>,
         mut stats: CompactLevel0Phase1StatsBuilder,
         target_file_size: u64,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<CompactLevel0Phase1Result, CompactionError> {
         stats.read_lock_held_spawn_blocking_startup_micros =
             stats.read_lock_acquisition_micros.till_now(); // set by caller
@@ -907,7 +907,7 @@ impl Timeline {
     pub(crate) async fn compact_tiered(
         self: &Arc<Self>,
         _cancel: &CancellationToken,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<(), CompactionError> {
         let fanout = self.get_compaction_threshold() as u64;
         let target_file_size = self.get_checkpoint_distance();
@@ -952,6 +952,178 @@ impl Timeline {
         adaptor.flush_updates().await?;
         Ok(())
     }
+
+    /// An experimental compaction building block that combines compaction with garbage collection.
+    ///
+    /// The current implementation picks all delta + image layers that are below or intersecting with
+    /// the GC horizon without considering retain_lsns. Then, it does a full compaction over all these delta
+    /// layers and image layers, which generates image layers on the gc horizon, drop deltas below gc horizon,
+    /// and create delta layers with all deltas >= gc horizon.
+    #[cfg(test)]
+    pub(crate) async fn compact_with_gc(
+        self: &Arc<Self>,
+        _cancel: &CancellationToken,
+        ctx: &mut RequestContext,
+    ) -> Result<(), CompactionError> {
+        use crate::tenant::storage_layer::ValueReconstructState;
+        // Step 0: pick all delta layers + image layers below/intersect with the GC horizon.
+        // The layer selection has the following properties:
+        // 1. If a layer is in the selection, all layers below it are in the selection.
+        // 2. Inferred from (1), for each key in the layer selection, the value can be reconstructed only with the layers in the layer selection.
+        let (layer_selection, gc_cutoff) = {
+            let guard = self.layers.read().await;
+            let layers = guard.layer_map();
+            let gc_info = self.gc_info.read().unwrap();
+            let gc_cutoff = Lsn::min(gc_info.cutoffs.horizon, gc_info.cutoffs.pitr);
+            let mut selected_layers = Vec::new();
+            // TODO: consider retain_lsns
+            drop(gc_info);
+            for desc in layers.iter_historic_layers() {
+                if desc.get_lsn_range().start <= gc_cutoff {
+                    selected_layers.push(guard.get_from_desc(&desc));
+                }
+            }
+            (selected_layers, gc_cutoff)
+        };
+        // Step 1: (In the future) construct a k-merge iterator over all layers. For now, simply collect all keys + LSNs.
+        let mut all_key_values = Vec::new();
+        for layer in &layer_selection {
+            all_key_values.extend(layer.load_key_values(ctx).await?);
+        }
+        // Key small to large, LSN low to high, if the same LSN has both image and delta due to the merge of delta layers and
+        // image layers, make image appear later than delta.
+        struct ValueWrapper<'a>(&'a crate::repository::Value);
+        impl Ord for ValueWrapper<'_> {
+            fn cmp(&self, other: &Self) -> std::cmp::Ordering {
+                use crate::repository::Value;
+                use std::cmp::Ordering;
+                match (self.0, other.0) {
+                    (Value::Image(_), Value::WalRecord(_)) => Ordering::Greater,
+                    (Value::WalRecord(_), Value::Image(_)) => Ordering::Less,
+                    _ => Ordering::Equal,
+                }
+            }
+        }
+        impl PartialOrd for ValueWrapper<'_> {
+            fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
+                Some(self.cmp(other))
+            }
+        }
+        impl PartialEq for ValueWrapper<'_> {
+            fn eq(&self, other: &Self) -> bool {
+                self.cmp(other) == std::cmp::Ordering::Equal
+            }
+        }
+        impl Eq for ValueWrapper<'_> {}
+        all_key_values.sort_by(|(k1, l1, v1), (k2, l2, v2)| {
+            (k1, l1, ValueWrapper(v1)).cmp(&(k2, l2, ValueWrapper(v2)))
+        });
+        let max_lsn = all_key_values
+            .iter()
+            .map(|(_, lsn, _)| lsn)
+            .max()
+            .copied()
+            .unwrap()
+            + 1;
+        // Step 2: Produce images+deltas. TODO: ensure newly-produced delta does not overlap with other deltas.
+        // Data of the same key.
+        let mut accumulated_values = Vec::new();
+        let mut last_key = all_key_values.first().unwrap().0; // TODO: assert all_key_values not empty
+
+        /// Take a list of images and deltas, produce an image at the GC horizon, and a list of deltas above the GC horizon.
+        async fn flush_accumulated_states(
+            tline: &Arc<Timeline>,
+            key: Key,
+            accumulated_values: &[&(Key, Lsn, crate::repository::Value)],
+            horizon: Lsn,
+        ) -> anyhow::Result<(Vec<(Key, Lsn, crate::repository::Value)>, bytes::Bytes)> {
+            let mut base_image = None;
+            let mut keys_above_horizon = Vec::new();
+            let mut delta_above_base_image = Vec::new();
+            // We have a list of deltas/images. We want to create image layers while collect garbages.
+            for (key, lsn, val) in accumulated_values.iter().rev() {
+                if *lsn > horizon {
+                    keys_above_horizon.push((*key, *lsn, val.clone())); // TODO: ensure one LSN corresponds to either delta or image instead of both
+                } else if *lsn <= horizon {
+                    match val {
+                        crate::repository::Value::Image(image) => {
+                            if lsn <= &horizon {
+                                base_image = Some((*lsn, image.clone()));
+                                break;
+                            }
+                        }
+                        crate::repository::Value::WalRecord(wal) => {
+                            delta_above_base_image.push((*lsn, wal.clone()));
+                        }
+                    }
+                }
+            }
+            delta_above_base_image.reverse();
+            keys_above_horizon.reverse();
+            let state = ValueReconstructState {
+                img: base_image,
+                records: delta_above_base_image,
+            };
+            let img = tline.reconstruct_value(key, horizon, state).await?;
+            Ok((keys_above_horizon, img))
+        }
+
+        let mut delta_layer_writer = DeltaLayerWriter::new(
+            self.conf,
+            self.timeline_id,
+            self.tenant_shard_id,
+            all_key_values.first().unwrap().0,
+            gc_cutoff..max_lsn, // TODO: off by one?
+            ctx,
+        )
+        .await?;
+        let mut image_layer_writer = ImageLayerWriter::new(
+            self.conf,
+            self.timeline_id,
+            self.tenant_shard_id,
+            &(all_key_values.first().unwrap().0..all_key_values.last().unwrap().0.next()),
+            gc_cutoff,
+            ctx,
+        )
+        .await?;
+
+        for item @ (key, _, _) in &all_key_values {
+            if &last_key == key {
+                accumulated_values.push(item);
+            } else {
+                let (deltas, image) =
+                    flush_accumulated_states(self, last_key, &accumulated_values, gc_cutoff)
+                        .await?;
+                image_layer_writer.put_image(last_key, image, ctx).await?;
+                for (key, lsn, val) in deltas {
+                    delta_layer_writer.put_value(key, lsn, val, ctx).await?;
+                }
+                accumulated_values.clear();
+                accumulated_values.push(item);
+                last_key = *key;
+            }
+        }
+        let (deltas, image) =
+            flush_accumulated_states(self, last_key, &accumulated_values, gc_cutoff).await?;
+        image_layer_writer.put_image(last_key, image, ctx).await?;
+        for (key, lsn, val) in deltas {
+            delta_layer_writer.put_value(key, lsn, val, ctx).await?;
+        }
+        accumulated_values.clear();
+        // TODO: split layers
+        let delta_layer = delta_layer_writer.finish(last_key, self, ctx).await?;
+        let image_layer = image_layer_writer.finish(self, ctx).await?;
+        // Step 3: Place back to the layer map.
+        {
+            let mut guard = self.layers.write().await;
+            guard.finish_gc_compaction(
+                &layer_selection,
+                &[delta_layer.clone(), image_layer.clone()],
+                &self.metrics,
+            )
+        };
+        Ok(())
+    }
 }
 
 struct TimelineAdaptor {
@@ -1018,7 +1190,7 @@ impl CompactionJobExecutor for TimelineAdaptor {
         &mut self,
         key_range: &Range<Key>,
         lsn_range: &Range<Lsn>,
-        _ctx: &RequestContext,
+        _ctx: &mut RequestContext,
     ) -> anyhow::Result<Vec<OwnArc<PersistentLayerDesc>>> {
         self.flush_updates().await?;
 
@@ -1039,7 +1211,7 @@ impl CompactionJobExecutor for TimelineAdaptor {
         &mut self,
         key_range: &Range<Key>,
         lsn: Lsn,
-        _ctx: &RequestContext,
+        _ctx: &mut RequestContext,
     ) -> anyhow::Result<Vec<Range<Key>>> {
         if lsn == self.keyspace.0 {
             Ok(pageserver_compaction::helpers::intersect_keyspace(
@@ -1075,7 +1247,7 @@ impl CompactionJobExecutor for TimelineAdaptor {
         &mut self,
         lsn: Lsn,
         key_range: &Range<Key>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<()> {
         Ok(self.create_image_impl(lsn, key_range, ctx).await?)
     }
@@ -1085,7 +1257,7 @@ impl CompactionJobExecutor for TimelineAdaptor {
         lsn_range: &Range<Lsn>,
         key_range: &Range<Key>,
         input_layers: &[ResidentDeltaLayer],
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> anyhow::Result<()> {
         debug!("Create new layer {}..{}", lsn_range.start, lsn_range.end);
 
@@ -1157,7 +1329,7 @@ impl CompactionJobExecutor for TimelineAdaptor {
     async fn delete_layer(
         &mut self,
         layer: &OwnArc<PersistentLayerDesc>,
-        _ctx: &RequestContext,
+        _ctx: &mut RequestContext,
     ) -> anyhow::Result<()> {
         self.layers_to_delete.push(layer.clone().0);
         Ok(())
@@ -1169,7 +1341,7 @@ impl TimelineAdaptor {
         &mut self,
         lsn: Lsn,
         key_range: &Range<Key>,
-        ctx: &RequestContext,
+        ctx: &mut RequestContext,
     ) -> Result<(), CreateImageLayersError> {
         let timer = self.timeline.metrics.create_images_time_histo.start_timer();
 
@@ -1296,7 +1468,7 @@ impl CompactionLayer<Key> for ResidentDeltaLayer {
 impl CompactionDeltaLayer<TimelineAdaptor> for ResidentDeltaLayer {
     type DeltaEntry<'a> = DeltaEntry<'a>;
 
-    async fn load_keys<'a>(&self, ctx: &RequestContext) -> anyhow::Result<Vec<DeltaEntry<'_>>> {
+    async fn load_keys<'a>(&self, ctx: &mut RequestContext) -> anyhow::Result<Vec<DeltaEntry<'_>>> {
         self.0.load_keys(ctx).await
     }
 }

pageserver/src/tenant/timeline/delete.rs

@@ -11,7 +11,6 @@ use utils::{crashsafe, fs_ext, id::TimelineId, pausable_failpoint};
 
 use crate::{
     config::PageServerConf,
-    deletion_queue::DeletionQueueClient,
     task_mgr::{self, TaskKind},
     tenant::{
         metadata::TimelineMetadata,
@@ -263,7 +262,6 @@ impl DeleteTimelineFlow {
         timeline_id: TimelineId,
         local_metadata: &TimelineMetadata,
         remote_client: RemoteTimelineClient,
-        deletion_queue_client: DeletionQueueClient,
     ) -> anyhow::Result<()> {
         // Note: here we even skip populating layer map. Timeline is essentially uninitialized.
         // RemoteTimelineClient is the only functioning part.
@@ -274,7 +272,6 @@ impl DeleteTimelineFlow {
                 None, // Ancestor is not needed for deletion.
                 TimelineResources {
                     remote_client,
-                    deletion_queue_client,
                     timeline_get_throttle: tenant.timeline_get_throttle.clone(),
                 },
                 // Important. We dont pass ancestor above because it can be missing.

pageserver/src/tenant/timeline/detach_ancestor.rs

@@ -87,7 +87,7 @@ pub(super) async fn prepare(
     detached: &Arc<Timeline>,
     tenant: &Tenant,
     options: Options,
-    ctx: &RequestContext,
+    ctx: &mut RequestContext,
 ) -> Result<(completion::Completion, PreparedTimelineDetach), Error> {
     use Error::*;
 
@@ -325,7 +325,7 @@ async fn upload_rewritten_layer(
     layer: &Layer,
     target: &Arc<Timeline>,
     cancel: &CancellationToken,
-    ctx: &RequestContext,
+    ctx: &mut RequestContext,
 ) -> Result<Option<Layer>, Error> {
     use Error::UploadRewritten;
     let copied = copy_lsn_prefix(end_lsn, layer, target, ctx).await?;
@@ -348,7 +348,7 @@ async fn copy_lsn_prefix(
     end_lsn: Lsn,
     layer: &Layer,
     target_timeline: &Arc<Timeline>,
-    ctx: &RequestContext,
+    ctx: &mut RequestContext,
 ) -> Result<Option<ResidentLayer>, Error> {
     use Error::{CopyDeltaPrefix, RewrittenDeltaDownloadFailed};
 
@@ -437,7 +437,7 @@ pub(super) async fn complete(
     detached: &Arc<Timeline>,
     tenant: &Tenant,
     prepared: PreparedTimelineDetach,
-    _ctx: &RequestContext,
+    _ctx: &mut RequestContext,
 ) -> Result<Vec<TimelineId>, anyhow::Error> {
     let PreparedTimelineDetach { layers } = prepared;