code reading notes from last weekend

There used to be some pg version dependencies in these extensions, but
now that there isn't, follow the simpler pattern used in other
extensions. No change in the produced images.

.cargo/config.toml

@@ -3,6 +3,16 @@
 # by the RUSTDOCFLAGS env var in CI.
 rustdocflags = ["-Arustdoc::private_intra_doc_links"]
 
+# Enable frame pointers. This may have a minor performance overhead, but makes it easier and more
+# efficient to obtain stack traces (and thus CPU/heap profiles). It may also avoid seg faults that
+# we've seen with libunwind-based profiling. See also:
+#
+# * <https://www.brendangregg.com/blog/2024-03-17/the-return-of-the-frame-pointers.html>
+# * <https://github.com/rust-lang/rust/pull/122646>
+#
+# NB: the RUSTFLAGS envvar will replace this. Make sure to update e.g. Dockerfile as well.
+rustflags = ["-Cforce-frame-pointers=yes"]
+
 [alias]
 build_testing = ["build", "--features", "testing"]
 neon = ["run", "--bin", "neon_local"]

.github/file-filters.yaml

@@ -0,0 +1,12 @@
+rust_code: ['**/*.rs', '**/Cargo.toml', '**/Cargo.lock']
+
+v14: ['vendor/postgres-v14/**', 'Makefile', 'pgxn/**']
+v15: ['vendor/postgres-v15/**', 'Makefile', 'pgxn/**']
+v16: ['vendor/postgres-v16/**', 'Makefile', 'pgxn/**']
+v17: ['vendor/postgres-v17/**', 'Makefile', 'pgxn/**']
+
+rebuild_neon_extra:
+    - .github/workflows/neon_extra_builds.yml
+
+rebuild_macos:
+    - .github/workflows/build-macos.yml

.github/workflows/actionlint.yml

@@ -33,7 +33,7 @@ jobs:
           # SC2086 - Double quote to prevent globbing and word splitting. - https://www.shellcheck.net/wiki/SC2086
           SHELLCHECK_OPTS: --exclude=SC2046,SC2086
         with:
-          fail_on_error: true
+          fail_level: error
           filter_mode: nofilter
           level: error
 

.github/workflows/benchmarking.yml

@@ -308,6 +308,7 @@ jobs:
           "image": [ "'"$image_default"'" ],
           "include": [{ "pg_version": 16, "region_id": "'"$region_id_default"'", "platform": "neonvm-captest-freetier",       "db_size": "3gb" ,"runner": '"$runner_default"', "image": "'"$image_default"'" },
                       { "pg_version": 16, "region_id": "'"$region_id_default"'", "platform": "neonvm-captest-new",            "db_size": "10gb","runner": '"$runner_default"', "image": "'"$image_default"'" },
+                      { "pg_version": 16, "region_id": "'"$region_id_default"'", "platform": "neonvm-captest-new-many-tables","db_size": "10gb","runner": '"$runner_default"', "image": "'"$image_default"'" },
                       { "pg_version": 16, "region_id": "'"$region_id_default"'", "platform": "neonvm-captest-new",            "db_size": "50gb","runner": '"$runner_default"', "image": "'"$image_default"'" },
                       { "pg_version": 16, "region_id": "azure-eastus2",          "platform": "neonvm-azure-captest-freetier", "db_size": "3gb" ,"runner": '"$runner_azure"',   "image": "neondatabase/build-tools:pinned-bookworm" },
                       { "pg_version": 16, "region_id": "azure-eastus2",          "platform": "neonvm-azure-captest-new",      "db_size": "10gb","runner": '"$runner_azure"',   "image": "neondatabase/build-tools:pinned-bookworm" },
@@ -410,7 +411,7 @@ jobs:
         aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
 
     - name: Create Neon Project
-      if: contains(fromJson('["neonvm-captest-new", "neonvm-captest-freetier", "neonvm-azure-captest-freetier", "neonvm-azure-captest-new"]'), matrix.platform)
+      if: contains(fromJson('["neonvm-captest-new", "neonvm-captest-new-many-tables", "neonvm-captest-freetier", "neonvm-azure-captest-freetier", "neonvm-azure-captest-new"]'), matrix.platform)
       id: create-neon-project
       uses: ./.github/actions/neon-project-create
       with:
@@ -429,7 +430,7 @@ jobs:
           neonvm-captest-sharding-reuse)
             CONNSTR=${{ secrets.BENCHMARK_CAPTEST_SHARDING_CONNSTR }}
             ;;
-          neonvm-captest-new | neonvm-captest-freetier | neonvm-azure-captest-new | neonvm-azure-captest-freetier)
+          neonvm-captest-new | neonvm-captest-new-many-tables | neonvm-captest-freetier | neonvm-azure-captest-new | neonvm-azure-captest-freetier)
             CONNSTR=${{ steps.create-neon-project.outputs.dsn }}
             ;;
           rds-aurora)
@@ -446,6 +447,26 @@ jobs:
 
         echo "connstr=${CONNSTR}" >> $GITHUB_OUTPUT
 
+    # we want to compare Neon project OLTP throughput and latency at scale factor 10 GB 
+    # without (neonvm-captest-new)
+    # and with (neonvm-captest-new-many-tables) many relations in the database
+    - name: Create many relations before the run
+      if: contains(fromJson('["neonvm-captest-new-many-tables"]'), matrix.platform)
+      uses: ./.github/actions/run-python-test-set
+      with:
+        build_type: ${{ env.BUILD_TYPE }}
+        test_selection: performance
+        run_in_parallel: false
+        save_perf_report: ${{ env.SAVE_PERF_REPORT }}
+        extra_params: -m remote_cluster --timeout 21600 -k test_perf_many_relations
+        pg_version: ${{ env.DEFAULT_PG_VERSION }}
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+      env:
+        BENCHMARK_CONNSTR: ${{ steps.set-up-connstr.outputs.connstr }}
+        VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
+        PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
+        TEST_NUM_RELATIONS: 10000
+
     - name: Benchmark init
       uses: ./.github/actions/run-python-test-set
       with:

.github/workflows/build-macos.yml

@@ -0,0 +1,241 @@
+name: Check neon with MacOS builds
+
+on:
+  workflow_call:
+    inputs:
+      pg_versions:
+        description: "Array of the pg versions to build for, for example: ['v14', 'v17']"
+        type: string
+        default: '[]'
+        required: false
+      rebuild_rust_code:
+        description: "Rebuild Rust code"
+        type: boolean
+        default: false
+        required: false
+      rebuild_everything:
+        description: "If true, rebuild for all versions"
+        type: boolean
+        default: false
+        required: false
+
+env:
+  RUST_BACKTRACE: 1
+  COPT: '-Werror'
+
+# TODO: move `check-*` and `files-changed` jobs to the "Caller" Workflow
+# We should care about that as Github has limitations:
+# - You can connect up to four levels of workflows
+# - You can call a maximum of 20 unique reusable workflows from a single workflow file.
+# https://docs.github.com/en/actions/sharing-automations/reusing-workflows#limitations
+jobs:
+  build-pgxn:
+    if: |
+      (inputs.pg_versions != '[]' || inputs.rebuild_everything) && (
+        contains(github.event.pull_request.labels.*.name, 'run-extra-build-macos')  ||
+        contains(github.event.pull_request.labels.*.name, 'run-extra-build-*') ||
+        github.ref_name == 'main'
+      )
+    timeout-minutes: 30
+    runs-on: macos-15
+    strategy:
+      matrix:
+        postgres-version: ${{ inputs.rebuild_everything && fromJson('["v14", "v15", "v16", "v17"]') || fromJSON(inputs.pg_versions) }}
+    env:
+      # Use release build only, to have less debug info around
+      # Hence keeping target/ (and general cache size) smaller
+      BUILD_TYPE: release
+    steps:
+      - name: Checkout main repo
+        uses: actions/checkout@v4
+
+      - name: Set pg ${{ matrix.postgres-version }} for caching
+        id: pg_rev
+        run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-${{ matrix.postgres-version }}) | tee -a "${GITHUB_OUTPUT}"
+
+      - name: Cache postgres ${{ matrix.postgres-version }} build
+        id: cache_pg
+        uses: actions/cache@v4
+        with:
+          path: pg_install/${{ matrix.postgres-version }}
+          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ env.BUILD_TYPE }}-pg-${{ matrix.postgres-version }}-${{ steps.pg_rev.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
+
+      - name: Checkout submodule vendor/postgres-${{ matrix.postgres-version }}
+        if: steps.cache_pg.outputs.cache-hit != 'true'
+        run: |
+          git submodule init vendor/postgres-${{ matrix.postgres-version }}
+          git submodule update --depth 1 --recursive
+
+      - name: Install build dependencies
+        if: steps.cache_pg.outputs.cache-hit != 'true'
+        run: |
+          brew install flex bison openssl protobuf icu4c
+
+      - name: Set extra env for macOS
+        if: steps.cache_pg.outputs.cache-hit != 'true'
+        run: |
+          echo 'LDFLAGS=-L/usr/local/opt/openssl@3/lib' >> $GITHUB_ENV
+          echo 'CPPFLAGS=-I/usr/local/opt/openssl@3/include' >> $GITHUB_ENV
+
+      - name: Build Postgres ${{ matrix.postgres-version }}
+        if: steps.cache_pg.outputs.cache-hit != 'true'
+        run: |
+          make postgres-${{ matrix.postgres-version }} -j$(sysctl -n hw.ncpu)
+
+      - name: Build Neon Pg Ext ${{ matrix.postgres-version }}
+        if: steps.cache_pg.outputs.cache-hit != 'true'
+        run: |
+          make "neon-pg-ext-${{ matrix.postgres-version }}" -j$(sysctl -n hw.ncpu)
+
+      - name: Get postgres headers ${{ matrix.postgres-version }}
+        if: steps.cache_pg.outputs.cache-hit != 'true'
+        run: |
+          make postgres-headers-${{ matrix.postgres-version }} -j$(sysctl -n hw.ncpu)
+
+  build-walproposer-lib:
+    if: |
+      (inputs.pg_versions != '[]' || inputs.rebuild_everything) && (
+        contains(github.event.pull_request.labels.*.name, 'run-extra-build-macos')  ||
+        contains(github.event.pull_request.labels.*.name, 'run-extra-build-*') ||
+        github.ref_name == 'main'
+      )
+    timeout-minutes: 30
+    runs-on: macos-15
+    needs: [build-pgxn]
+    env:
+      # Use release build only, to have less debug info around
+      # Hence keeping target/ (and general cache size) smaller
+      BUILD_TYPE: release
+    steps:
+      - name: Checkout main repo
+        uses: actions/checkout@v4
+
+      - name: Set pg v17 for caching
+        id: pg_rev
+        run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-v17) | tee -a "${GITHUB_OUTPUT}"
+
+      - name: Cache postgres v17 build
+        id: cache_pg
+        uses: actions/cache@v4
+        with:
+          path: pg_install/v17
+          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ env.BUILD_TYPE }}-pg-v17-${{ steps.pg_rev.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
+
+      - name: Cache walproposer-lib
+        id: cache_walproposer_lib
+        uses: actions/cache@v4
+        with:
+          path: pg_install/build/walproposer-lib
+          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ env.BUILD_TYPE }}-walproposer_lib-v17-${{ steps.pg_rev.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
+
+      - name: Checkout submodule vendor/postgres-v17
+        if: steps.cache_walproposer_lib.outputs.cache-hit != 'true'
+        run: |
+          git submodule init vendor/postgres-v17
+          git submodule update --depth 1 --recursive
+
+      - name: Install build dependencies
+        if: steps.cache_walproposer_lib.outputs.cache-hit != 'true'
+        run: |
+          brew install flex bison openssl protobuf icu4c
+
+      - name: Set extra env for macOS
+        if: steps.cache_walproposer_lib.outputs.cache-hit != 'true'
+        run: |
+          echo 'LDFLAGS=-L/usr/local/opt/openssl@3/lib' >> $GITHUB_ENV
+          echo 'CPPFLAGS=-I/usr/local/opt/openssl@3/include' >> $GITHUB_ENV
+
+      - name: Build walproposer-lib (only for v17)
+        if: steps.cache_walproposer_lib.outputs.cache-hit != 'true'
+        run:
+          make walproposer-lib -j$(sysctl -n hw.ncpu)
+
+  cargo-build:
+    if: |
+      (inputs.pg_versions != '[]' || inputs.rebuild_rust_code || inputs.rebuild_everything) && (
+        contains(github.event.pull_request.labels.*.name, 'run-extra-build-macos')  ||
+        contains(github.event.pull_request.labels.*.name, 'run-extra-build-*') ||
+        github.ref_name == 'main'
+      )
+    timeout-minutes: 30
+    runs-on: macos-15
+    needs: [build-pgxn, build-walproposer-lib]
+    env:
+      # Use release build only, to have less debug info around
+      # Hence keeping target/ (and general cache size) smaller
+      BUILD_TYPE: release
+    steps:
+      - name: Checkout main repo
+        uses: actions/checkout@v4
+        with:
+          submodules: true
+
+      - name: Set pg v14 for caching
+        id: pg_rev_v14
+        run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-v14) | tee -a "${GITHUB_OUTPUT}"
+      - name: Set pg v15 for caching
+        id: pg_rev_v15
+        run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-v15) | tee -a "${GITHUB_OUTPUT}"
+      - name: Set pg v16 for caching
+        id: pg_rev_v16
+        run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-v16) | tee -a "${GITHUB_OUTPUT}"
+      - name: Set pg v17 for caching
+        id: pg_rev_v17
+        run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-v17) | tee -a "${GITHUB_OUTPUT}"
+
+      - name: Cache postgres v14 build
+        id: cache_pg
+        uses: actions/cache@v4
+        with:
+          path: pg_install/v14
+          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ env.BUILD_TYPE }}-pg-v14-${{ steps.pg_rev_v14.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
+      - name: Cache postgres v15 build
+        id: cache_pg_v15
+        uses: actions/cache@v4
+        with:
+          path: pg_install/v15
+          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ env.BUILD_TYPE }}-pg-v15-${{ steps.pg_rev_v15.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
+      - name: Cache postgres v16 build
+        id: cache_pg_v16
+        uses: actions/cache@v4
+        with:
+          path: pg_install/v16
+          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ env.BUILD_TYPE }}-pg-v16-${{ steps.pg_rev_v16.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
+      - name: Cache postgres v17 build
+        id: cache_pg_v17
+        uses: actions/cache@v4
+        with:
+          path: pg_install/v17
+          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ env.BUILD_TYPE }}-pg-v17-${{ steps.pg_rev_v17.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
+
+      - name: Cache cargo deps (only for v17)
+        uses: actions/cache@v4
+        with:
+          path: |
+            ~/.cargo/registry
+            !~/.cargo/registry/src
+            ~/.cargo/git
+            target
+          key: v1-${{ runner.os }}-${{ runner.arch }}-cargo-${{ hashFiles('./Cargo.lock') }}-${{ hashFiles('./rust-toolchain.toml') }}-rust
+
+      - name: Cache walproposer-lib
+        id: cache_walproposer_lib
+        uses: actions/cache@v4
+        with:
+          path: pg_install/build/walproposer-lib
+          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ env.BUILD_TYPE }}-walproposer_lib-v17-${{ steps.pg_rev_v17.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
+
+      - name: Install build dependencies
+        run: |
+          brew install flex bison openssl protobuf icu4c
+
+      - name: Set extra env for macOS
+        run: |
+          echo 'LDFLAGS=-L/usr/local/opt/openssl@3/lib' >> $GITHUB_ENV
+          echo 'CPPFLAGS=-I/usr/local/opt/openssl@3/include' >> $GITHUB_ENV
+
+      - name: Run cargo build (only for v17)
+        run: PQ_LIB_DIR=$(pwd)/pg_install/v17/lib cargo build --all --release -j$(sysctl -n hw.ncpu)
+
+      - name: Check that no warnings are produced (only for v17)
+        run: ./run_clippy.sh

.github/workflows/build_and_test.yml

@@ -212,7 +212,7 @@ jobs:
           fi
           echo "CLIPPY_COMMON_ARGS=${CLIPPY_COMMON_ARGS}" >> $GITHUB_ENV
       - name: Run cargo clippy (debug)
-        run: cargo hack --feature-powerset clippy $CLIPPY_COMMON_ARGS
+        run: cargo hack --features default --ignore-unknown-features --feature-powerset clippy $CLIPPY_COMMON_ARGS
 
       - name: Check documentation generation
         run: cargo doc --workspace --no-deps --document-private-items
@@ -538,7 +538,7 @@ jobs:
 
   trigger-e2e-tests:
     if: ${{ !github.event.pull_request.draft || contains( github.event.pull_request.labels.*.name, 'run-e2e-tests-in-draft') || github.ref_name == 'main' || github.ref_name == 'release' || github.ref_name == 'release-proxy' || github.ref_name == 'release-compute' }}
-    needs: [ check-permissions, promote-images, tag ]
+    needs: [ check-permissions, promote-images-dev, tag ]
     uses: ./.github/workflows/trigger-e2e-tests.yml
     secrets: inherit
 
@@ -728,30 +728,6 @@ jobs:
           tags: |
             neondatabase/neon-test-extensions-${{ matrix.version.pg }}:${{needs.tag.outputs.build-tag}}-${{ matrix.version.debian }}-${{ matrix.arch }}
 
-      - name: Build compute-tools image
-        # compute-tools are Postgres independent, so build it only once
-        # We pick 16, because that builds on debian 11 with older glibc (and is
-        # thus compatible with newer glibc), rather than 17 on Debian 12, as
-        # that isn't guaranteed to be compatible with Debian 11
-        if: matrix.version.pg == 'v16'
-        uses: docker/build-push-action@v6
-        with:
-          target: compute-tools-image
-          context: .
-          build-args: |
-            GIT_VERSION=${{ github.event.pull_request.head.sha || github.sha }}
-            BUILD_TAG=${{ needs.tag.outputs.build-tag }}
-            TAG=${{ needs.build-build-tools-image.outputs.image-tag }}-${{ matrix.version.debian }}
-            DEBIAN_VERSION=${{ matrix.version.debian }}
-          provenance: false
-          push: true
-          pull: true
-          file: compute/compute-node.Dockerfile
-          cache-from: type=registry,ref=cache.neon.build/compute-node-${{ matrix.version.pg }}:cache-${{ matrix.version.debian }}-${{ matrix.arch }}
-          cache-to: ${{ github.ref_name == 'main' && format('type=registry,ref=cache.neon.build/compute-tools-{0}:cache-{1}-{2},mode=max', matrix.version.pg, matrix.version.debian, matrix.arch) || '' }}
-          tags: |
-            neondatabase/compute-tools:${{ needs.tag.outputs.build-tag }}-${{ matrix.version.debian }}-${{ matrix.arch }}
-
   compute-node-image:
     needs: [ compute-node-image-arch, tag ]
     permissions:
@@ -794,14 +770,6 @@ jobs:
                                              neondatabase/neon-test-extensions-${{ matrix.version.pg }}:${{ needs.tag.outputs.build-tag }}-${{ matrix.version.debian }}-x64 \
                                              neondatabase/neon-test-extensions-${{ matrix.version.pg }}:${{ needs.tag.outputs.build-tag }}-${{ matrix.version.debian }}-arm64
 
-      - name: Create multi-arch compute-tools image
-        if: matrix.version.pg == 'v16'
-        run: |
-          docker buildx imagetools create -t neondatabase/compute-tools:${{ needs.tag.outputs.build-tag }} \
-                                          -t neondatabase/compute-tools:${{ needs.tag.outputs.build-tag }}-${{ matrix.version.debian }} \
-                                             neondatabase/compute-tools:${{ needs.tag.outputs.build-tag }}-${{ matrix.version.debian }}-x64 \
-                                             neondatabase/compute-tools:${{ needs.tag.outputs.build-tag }}-${{ matrix.version.debian }}-arm64
-
       - name: Configure AWS credentials
         uses: aws-actions/configure-aws-credentials@v4
         with:
@@ -817,12 +785,6 @@ jobs:
           docker buildx imagetools create -t 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-${{ matrix.version.pg }}:${{ needs.tag.outputs.build-tag }} \
                                                                                 neondatabase/compute-node-${{ matrix.version.pg }}:${{ needs.tag.outputs.build-tag }}
 
-      - name: Push multi-arch compute-tools image to ECR
-        if: matrix.version.pg == 'v16'
-        run: |
-          docker buildx imagetools create -t 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-tools:${{ needs.tag.outputs.build-tag }} \
-                                                                                neondatabase/compute-tools:${{ needs.tag.outputs.build-tag }}
-
   vm-compute-node-image:
     needs: [ check-permissions, tag, compute-node-image ]
     runs-on: [ self-hosted, large ]
@@ -930,8 +892,8 @@ jobs:
           docker compose -f ./docker-compose/docker-compose.yml logs || 0
           docker compose -f ./docker-compose/docker-compose.yml down
 
-  promote-images:
-    needs: [ check-permissions, tag, test-images, vm-compute-node-image ]
+  promote-images-dev:
+    needs: [ check-permissions, tag, vm-compute-node-image ]
     runs-on: ubuntu-22.04
 
     permissions:
@@ -965,6 +927,35 @@ jobs:
                                                neondatabase/vm-compute-node-${version}:${{ needs.tag.outputs.build-tag }}
           done
 
+  promote-images-prod:
+    needs: [ check-permissions, tag, test-images, vm-compute-node-image ]
+    runs-on: ubuntu-22.04
+    if: github.ref_name == 'main' || github.ref_name == 'release' || github.ref_name == 'release-proxy' || github.ref_name == 'release-compute'
+
+    permissions:
+      id-token: write # aws-actions/configure-aws-credentials
+      statuses: write
+      contents: read
+
+    env:
+      VERSIONS: v14 v15 v16 v17
+
+    steps:
+      - name: Configure AWS credentials
+        uses: aws-actions/configure-aws-credentials@v4
+        with:
+          aws-region: eu-central-1
+          role-to-assume: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+          role-duration-seconds: 3600
+
+      - name: Login to Amazon Dev ECR
+        uses: aws-actions/amazon-ecr-login@v2
+
+      - uses: docker/login-action@v3
+        with:
+          username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
+          password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
+
       - name: Add latest tag to images
         if: github.ref_name == 'main'
         run: |
@@ -972,9 +963,6 @@ jobs:
             docker buildx imagetools create -t $repo/neon:latest \
                                                $repo/neon:${{ needs.tag.outputs.build-tag }}
 
-            docker buildx imagetools create -t $repo/compute-tools:latest \
-                                               $repo/compute-tools:${{ needs.tag.outputs.build-tag }}
-
             for version in ${VERSIONS}; do
               docker buildx imagetools create -t $repo/compute-node-${version}:latest \
                                                  $repo/compute-node-${version}:${{ needs.tag.outputs.build-tag }}
@@ -1003,31 +991,31 @@ jobs:
       - name: Copy all images to prod ECR
         if: github.ref_name == 'release' || github.ref_name == 'release-proxy' || github.ref_name == 'release-compute'
         run: |
-          for image in neon compute-tools {vm-,}compute-node-{v14,v15,v16,v17}; do
+          for image in neon {vm-,}compute-node-{v14,v15,v16,v17}; do
             docker buildx imagetools create -t 093970136003.dkr.ecr.eu-central-1.amazonaws.com/${image}:${{ needs.tag.outputs.build-tag }} \
                                                369495373322.dkr.ecr.eu-central-1.amazonaws.com/${image}:${{ needs.tag.outputs.build-tag }}
           done
 
   push-to-acr-dev:
     if: github.ref_name == 'main'
-    needs: [ tag, promote-images ]
+    needs: [ tag, promote-images-dev ]
     uses: ./.github/workflows/_push-to-acr.yml
     with:
       client_id: ${{ vars.AZURE_DEV_CLIENT_ID }}
       image_tag: ${{ needs.tag.outputs.build-tag }}
-      images: neon compute-tools vm-compute-node-v14 vm-compute-node-v15 vm-compute-node-v16 vm-compute-node-v17 compute-node-v14 compute-node-v15 compute-node-v16 compute-node-v17
+      images: neon vm-compute-node-v14 vm-compute-node-v15 vm-compute-node-v16 vm-compute-node-v17 compute-node-v14 compute-node-v15 compute-node-v16 compute-node-v17
       registry_name: ${{ vars.AZURE_DEV_REGISTRY_NAME }}
       subscription_id: ${{ vars.AZURE_DEV_SUBSCRIPTION_ID }}
       tenant_id: ${{ vars.AZURE_TENANT_ID }}
 
   push-to-acr-prod:
     if: github.ref_name == 'release' || github.ref_name == 'release-proxy' || github.ref_name == 'release-compute'
-    needs: [ tag, promote-images ]
+    needs: [ tag, promote-images-prod ]
     uses: ./.github/workflows/_push-to-acr.yml
     with:
       client_id: ${{ vars.AZURE_PROD_CLIENT_ID }}
       image_tag: ${{ needs.tag.outputs.build-tag }}
-      images: neon compute-tools vm-compute-node-v14 vm-compute-node-v15 vm-compute-node-v16 vm-compute-node-v17 compute-node-v14 compute-node-v15 compute-node-v16 compute-node-v17
+      images: neon vm-compute-node-v14 vm-compute-node-v15 vm-compute-node-v16 vm-compute-node-v17 compute-node-v14 compute-node-v15 compute-node-v16 compute-node-v17
       registry_name: ${{ vars.AZURE_PROD_REGISTRY_NAME }}
       subscription_id: ${{ vars.AZURE_PROD_SUBSCRIPTION_ID }}
       tenant_id: ${{ vars.AZURE_TENANT_ID }}
@@ -1112,7 +1100,7 @@ jobs:
           exit 1
 
   deploy:
-    needs: [ check-permissions, promote-images, tag, build-and-test-locally, trigger-custom-extensions-build-and-wait, push-to-acr-dev, push-to-acr-prod ]
+    needs: [ check-permissions, promote-images-prod, tag, build-and-test-locally, trigger-custom-extensions-build-and-wait, push-to-acr-dev, push-to-acr-prod ]
     # `!failure() && !cancelled()` is required because the workflow depends on the job that can be skipped: `push-to-acr-dev` and `push-to-acr-prod`
     if: (github.ref_name == 'main' || github.ref_name == 'release' || github.ref_name == 'release-proxy' || github.ref_name == 'release-compute') && !failure() && !cancelled()
     permissions:
@@ -1333,7 +1321,7 @@ jobs:
           done
 
   pin-build-tools-image:
-    needs: [ build-build-tools-image, promote-images, build-and-test-locally ]
+    needs: [ build-build-tools-image, promote-images-prod, build-and-test-locally ]
     if: github.ref_name == 'main'
     uses: ./.github/workflows/pin-build-tools-image.yml
     with:
@@ -1356,7 +1344,7 @@ jobs:
       - build-and-test-locally
       - check-codestyle-python
       - check-codestyle-rust
-      - promote-images
+      - promote-images-dev
       - test-images
       - trigger-custom-extensions-build-and-wait
     runs-on: ubuntu-22.04

.github/workflows/neon_extra_builds.yml

@@ -31,19 +31,15 @@ jobs:
     uses: ./.github/workflows/build-build-tools-image.yml
     secrets: inherit
 
-  check-macos-build:
-    needs: [ check-permissions ]
-    if: |
-      contains(github.event.pull_request.labels.*.name, 'run-extra-build-macos')  ||
-      contains(github.event.pull_request.labels.*.name, 'run-extra-build-*') ||
-      github.ref_name == 'main'
-    timeout-minutes: 90
-    runs-on: macos-15
-
-    env:
-      # Use release build only, to have less debug info around
-      # Hence keeping target/ (and general cache size) smaller
-      BUILD_TYPE: release
+  files-changed:
+    name: Detect what files changed
+    runs-on: ubuntu-22.04
+    timeout-minutes: 3
+    outputs:
+      v17: ${{ steps.files_changed.outputs.v17 }}
+      postgres_changes: ${{ steps.postgres_changes.outputs.changes }}
+      rebuild_rust_code: ${{ steps.files_changed.outputs.rust_code }}
+      rebuild_everything: ${{ steps.files_changed.outputs.rebuild_neon_extra || steps.files_changed.outputs.rebuild_macos }}
 
     steps:
       - name: Checkout
@@ -51,106 +47,45 @@ jobs:
         with:
           submodules: true
 
-      - name: Install macOS postgres dependencies
-        run: brew install flex bison openssl protobuf icu4c
-
-      - name: Set pg 14 revision for caching
-        id: pg_v14_rev
-        run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-v14) >> $GITHUB_OUTPUT
-
-      - name: Set pg 15 revision for caching
-        id: pg_v15_rev
-        run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-v15) >> $GITHUB_OUTPUT
-
-      - name: Set pg 16 revision for caching
-        id: pg_v16_rev
-        run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-v16) >> $GITHUB_OUTPUT
-
-      - name: Set pg 17 revision for caching
-        id: pg_v17_rev
-        run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-v17) >> $GITHUB_OUTPUT
-
-      - name: Cache postgres v14 build
-        id: cache_pg_14
-        uses: actions/cache@v4
+      - name: Check for Postgres changes
+        uses: dorny/paths-filter@1441771bbfdd59dcd748680ee64ebd8faab1a242  #v3
+        id: files_changed
         with:
-          path: pg_install/v14
-          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ env.BUILD_TYPE }}-pg-${{ steps.pg_v14_rev.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
+          token: ${{ github.token }}
+          filters: .github/file-filters.yaml
+          base: ${{ github.event_name != 'pull_request' && (github.event.merge_group.base_ref || github.ref_name) || '' }}
+          ref: ${{ github.event_name != 'pull_request' && (github.event.merge_group.head_ref || github.ref) || '' }}
 
-      - name: Cache postgres v15 build
-        id: cache_pg_15
-        uses: actions/cache@v4
-        with:
-          path: pg_install/v15
-          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ env.BUILD_TYPE }}-pg-${{ steps.pg_v15_rev.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
-
-      - name: Cache postgres v16 build
-        id: cache_pg_16
-        uses: actions/cache@v4
-        with:
-          path: pg_install/v16
-          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ env.BUILD_TYPE }}-pg-${{ steps.pg_v16_rev.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
-
-      - name: Cache postgres v17 build
-        id: cache_pg_17
-        uses: actions/cache@v4
-        with:
-          path: pg_install/v17
-          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ env.BUILD_TYPE }}-pg-${{ steps.pg_v17_rev.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
-
-      - name: Set extra env for macOS
+      - name: Filter out only v-string for build matrix
+        id: postgres_changes
         run: |
-          echo 'LDFLAGS=-L/usr/local/opt/openssl@3/lib' >> $GITHUB_ENV
-          echo 'CPPFLAGS=-I/usr/local/opt/openssl@3/include' >> $GITHUB_ENV
+          v_strings_only_as_json_array=$(echo ${{ steps.files_changed.outputs.chnages }} | jq '.[]|select(test("v\\d+"))' | jq --slurp -c)
+          echo "changes=${v_strings_only_as_json_array}" | tee -a "${GITHUB_OUTPUT}"
 
-      - name: Cache cargo deps
-        uses: actions/cache@v4
-        with:
-          path: |
-            ~/.cargo/registry
-            !~/.cargo/registry/src
-            ~/.cargo/git
-            target
-          key: v1-${{ runner.os }}-${{ runner.arch }}-cargo-${{ hashFiles('./Cargo.lock') }}-${{ hashFiles('./rust-toolchain.toml') }}-rust
-
-      - name: Build postgres v14
-        if: steps.cache_pg_14.outputs.cache-hit != 'true'
-        run: make postgres-v14 -j$(sysctl -n hw.ncpu)
-
-      - name: Build postgres v15
-        if: steps.cache_pg_15.outputs.cache-hit != 'true'
-        run: make postgres-v15 -j$(sysctl -n hw.ncpu)
-
-      - name: Build postgres v16
-        if: steps.cache_pg_16.outputs.cache-hit != 'true'
-        run: make postgres-v16 -j$(sysctl -n hw.ncpu)
-
-      - name: Build postgres v17
-        if: steps.cache_pg_17.outputs.cache-hit != 'true'
-        run: make postgres-v17 -j$(sysctl -n hw.ncpu)
-
-      - name: Build neon extensions
-        run: make neon-pg-ext -j$(sysctl -n hw.ncpu)
-
-      - name: Build walproposer-lib
-        run: make walproposer-lib -j$(sysctl -n hw.ncpu)
-
-      - name: Run cargo build
-        run: PQ_LIB_DIR=$(pwd)/pg_install/v16/lib cargo build --all --release
-
-      - name: Check that no warnings are produced
-        run: ./run_clippy.sh
+  check-macos-build:
+    needs: [ check-permissions, files-changed ]
+    if: |
+      contains(github.event.pull_request.labels.*.name, 'run-extra-build-macos')  ||
+      contains(github.event.pull_request.labels.*.name, 'run-extra-build-*') ||
+      github.ref_name == 'main'
+    uses: ./.github/workflows/build-macos.yml
+    with:
+      pg_versions: ${{ needs.files-changed.outputs.postgres_changes }}
+      rebuild_rust_code: ${{ needs.files-changed.outputs.rebuild_rust_code }}
+      rebuild_everything: ${{ fromJson(needs.files-changed.outputs.rebuild_everything) }}
 
   gather-rust-build-stats:
-    needs: [ check-permissions, build-build-tools-image ]
+    needs: [ check-permissions, build-build-tools-image, files-changed ]
     permissions:
       id-token: write # aws-actions/configure-aws-credentials
       statuses: write
       contents: write
     if: |
-      contains(github.event.pull_request.labels.*.name, 'run-extra-build-stats') ||
-      contains(github.event.pull_request.labels.*.name, 'run-extra-build-*') ||
-      github.ref_name == 'main'
+      (needs.files-changed.outputs.v17 == 'true' || needs.files-changed.outputs.rebuild_everything == 'true') && (
+        contains(github.event.pull_request.labels.*.name, 'run-extra-build-stats') ||
+        contains(github.event.pull_request.labels.*.name, 'run-extra-build-*') ||
+        github.ref_name == 'main'
+      )
     runs-on: [ self-hosted, large ]
     container:
       image: ${{ needs.build-build-tools-image.outputs.image }}-bookworm

.github/workflows/trigger-e2e-tests.yml

@@ -68,7 +68,7 @@ jobs:
       GH_TOKEN: ${{ secrets.CI_ACCESS_TOKEN }}
       TAG: ${{ needs.tag.outputs.build-tag }}
     steps:
-      - name: Wait for `promote-images` job to finish
+      - name: Wait for `promote-images-dev` job to finish
         # It's important to have a timeout here, the script in the step can run infinitely
         timeout-minutes: 60
         run: |
@@ -79,17 +79,17 @@ jobs:
           # For PRs we use the run id as the tag
           BUILD_AND_TEST_RUN_ID=${TAG}
           while true; do
-            conclusion=$(gh run --repo ${GITHUB_REPOSITORY} view ${BUILD_AND_TEST_RUN_ID} --json jobs --jq '.jobs[] | select(.name == "promote-images") | .conclusion')
+            conclusion=$(gh run --repo ${GITHUB_REPOSITORY} view ${BUILD_AND_TEST_RUN_ID} --json jobs --jq '.jobs[] | select(.name == "promote-images-dev") | .conclusion')
             case "$conclusion" in
               success)
                 break
                 ;;
               failure | cancelled | skipped)
-                echo "The 'promote-images' job didn't succeed: '${conclusion}'. Exiting..."
+                echo "The 'promote-images-dev' job didn't succeed: '${conclusion}'. Exiting..."
                 exit 1
                 ;;
               *)
-                echo "The 'promote-images' hasn't succeed yet. Waiting..."
+                echo "The 'promote-images-dev' hasn't succeed yet. Waiting..."
                 sleep 60
                 ;;
             esac

Cargo.lock

@@ -718,13 +718,13 @@ dependencies = [
 
 [[package]]
 name = "axum"
-version = "0.7.5"
+version = "0.7.9"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "3a6c9af12842a67734c9a2e355436e5d03b22383ed60cf13cd0c18fbfe3dcbcf"
+checksum = "edca88bc138befd0323b20752846e6587272d3b03b0343c8ea28a6f819e6e71f"
 dependencies = [
  "async-trait",
  "axum-core",
- "base64 0.21.1",
+ "base64 0.22.1",
  "bytes",
  "futures-util",
  "http 1.1.0",
@@ -746,8 +746,8 @@ dependencies = [
  "sha1",
  "sync_wrapper 1.0.1",
  "tokio",
- "tokio-tungstenite",
- "tower",
+ "tokio-tungstenite 0.24.0",
+ "tower 0.5.2",
  "tower-layer",
  "tower-service",
  "tracing",
@@ -1267,6 +1267,7 @@ dependencies = [
  "aws-config",
  "aws-sdk-kms",
  "aws-sdk-s3",
+ "axum",
  "base64 0.13.1",
  "bytes",
  "camino",
@@ -1274,9 +1275,10 @@ dependencies = [
  "chrono",
  "clap",
  "compute_api",
+ "fail",
  "flate2",
  "futures",
- "hyper 0.14.30",
+ "http 1.1.0",
  "metrics",
  "nix 0.27.1",
  "notify",
@@ -1302,6 +1304,8 @@ dependencies = [
  "tokio-postgres",
  "tokio-stream",
  "tokio-util",
+ "tower 0.5.2",
+ "tower-http",
  "tracing",
  "tracing-opentelemetry",
  "tracing-subscriber",
@@ -1649,6 +1653,20 @@ dependencies = [
  "parking_lot_core 0.9.8",
 ]
 
+[[package]]
+name = "dashmap"
+version = "6.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5041cc499144891f3790297212f32a74fb938e5136a14943f338ef9e0ae276cf"
+dependencies = [
+ "cfg-if",
+ "crossbeam-utils",
+ "hashbrown 0.14.5",
+ "lock_api",
+ "once_cell",
+ "parking_lot_core 0.9.8",
+]
+
 [[package]]
 name = "data-encoding"
 version = "2.4.0"
@@ -1732,9 +1750,9 @@ checksum = "ab03c107fafeb3ee9f5925686dbb7a73bc76e3932abb0d2b365cb64b169cf04c"
 
 [[package]]
 name = "diesel"
-version = "2.2.3"
+version = "2.2.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "65e13bab2796f412722112327f3e575601a3e9cdcbe426f0d30dbf43f3f5dc71"
+checksum = "ccf1bedf64cdb9643204a36dd15b19a6ce8e7aa7f7b105868e9f1fad5ffa7d12"
 dependencies = [
  "bitflags 2.4.1",
  "byteorder",
@@ -1948,6 +1966,15 @@ dependencies = [
  "syn 2.0.90",
 ]
 
+[[package]]
+name = "env_filter"
+version = "0.1.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "186e05a59d4c50738528153b83b0b0194d3a29507dfec16eccd4b342903397d0"
+dependencies = [
+ "log",
+]
+
 [[package]]
 name = "env_logger"
 version = "0.10.2"
@@ -1961,6 +1988,16 @@ dependencies = [
  "termcolor",
 ]
 
+[[package]]
+name = "env_logger"
+version = "0.11.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6c012a26a7f605efc424dd53697843a72be7dc86ad2d01f7814337794a12231d"
+dependencies = [
+ "env_filter",
+ "log",
+]
+
 [[package]]
 name = "equator"
 version = "0.2.2"
@@ -2719,7 +2756,7 @@ dependencies = [
  "pin-project-lite",
  "socket2",
  "tokio",
- "tower",
+ "tower 0.4.13",
  "tower-service",
  "tracing",
 ]
@@ -2944,6 +2981,28 @@ dependencies = [
  "str_stack",
 ]
 
+[[package]]
+name = "inferno"
+version = "0.12.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "75a5d75fee4d36809e6b021e4b96b686e763d365ffdb03af2bd00786353f84fe"
+dependencies = [
+ "ahash",
+ "clap",
+ "crossbeam-channel",
+ "crossbeam-utils",
+ "dashmap 6.1.0",
+ "env_logger 0.11.2",
+ "indexmap 2.0.1",
+ "itoa",
+ "log",
+ "num-format",
+ "once_cell",
+ "quick-xml 0.37.1",
+ "rgb",
+ "str_stack",
+]
+
 [[package]]
 name = "inotify"
 version = "0.9.6"
@@ -3151,7 +3210,7 @@ version = "0.7.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "4644821e1c3d7a560fe13d842d13f587c07348a1a05d3a797152d41c90c56df2"
 dependencies = [
- "dashmap",
+ "dashmap 5.5.0",
  "hashbrown 0.13.2",
 ]
 
@@ -3259,9 +3318,9 @@ checksum = "b87248edafb776e59e6ee64a79086f65890d3510f2c656c000bf2a7e8a0aea40"
 
 [[package]]
 name = "matchit"
-version = "0.8.2"
+version = "0.8.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "540f1c43aed89909c0cc0cc604e3bb2f7e7a341a3728a9e6cfe760e733cd11ed"
+checksum = "47e1ffaa40ddd1f3ed91f717a33c8c0ee23fff369e3aa8772b9605cc1d22f4c3"
 
 [[package]]
 name = "md-5"
@@ -3689,23 +3748,23 @@ checksum = "ff011a302c396a5197692431fc1948019154afc178baf7d8e37367442a4601cf"
 
 [[package]]
 name = "opentelemetry"
-version = "0.26.0"
+version = "0.27.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "570074cc999d1a58184080966e5bd3bf3a9a4af650c3b05047c2621e7405cd17"
+checksum = "ab70038c28ed37b97d8ed414b6429d343a8bbf44c9f79ec854f3a643029ba6d7"
 dependencies = [
  "futures-core",
  "futures-sink",
  "js-sys",
- "once_cell",
  "pin-project-lite",
  "thiserror",
+ "tracing",
 ]
 
 [[package]]
 name = "opentelemetry-http"
-version = "0.26.0"
+version = "0.27.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "6351496aeaa49d7c267fb480678d85d1cd30c5edb20b497c48c56f62a8c14b99"
+checksum = "10a8a7f5f6ba7c1b286c2fbca0454eaba116f63bbe69ed250b642d36fbb04d80"
 dependencies = [
  "async-trait",
  "bytes",
@@ -3716,9 +3775,9 @@ dependencies = [
 
 [[package]]
 name = "opentelemetry-otlp"
-version = "0.26.0"
+version = "0.27.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "29e1f9c8b032d4f635c730c0efcf731d5e2530ea13fa8bef7939ddc8420696bd"
+checksum = "91cf61a1868dacc576bf2b2a1c3e9ab150af7272909e80085c3173384fe11f76"
 dependencies = [
  "async-trait",
  "futures-core",
@@ -3734,9 +3793,9 @@ dependencies = [
 
 [[package]]
 name = "opentelemetry-proto"
-version = "0.26.1"
+version = "0.27.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "c9d3968ce3aefdcca5c27e3c4ea4391b37547726a70893aab52d3de95d5f8b34"
+checksum = "a6e05acbfada5ec79023c85368af14abd0b307c015e9064d249b2a950ef459a6"
 dependencies = [
  "opentelemetry",
  "opentelemetry_sdk",
@@ -3746,22 +3805,21 @@ dependencies = [
 
 [[package]]
 name = "opentelemetry-semantic-conventions"
-version = "0.26.0"
+version = "0.27.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "db945c1eaea8ac6a9677185357480d215bb6999faa9f691d0c4d4d641eab7a09"
+checksum = "bc1b6902ff63b32ef6c489e8048c5e253e2e4a803ea3ea7e783914536eb15c52"
 
 [[package]]
 name = "opentelemetry_sdk"
-version = "0.26.0"
+version = "0.27.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "d2c627d9f4c9cdc1f21a29ee4bfbd6028fcb8bcf2a857b43f3abdf72c9c862f3"
+checksum = "231e9d6ceef9b0b2546ddf52335785ce41252bc7474ee8ba05bfad277be13ab8"
 dependencies = [
  "async-trait",
  "futures-channel",
  "futures-executor",
  "futures-util",
  "glob",
- "once_cell",
  "opentelemetry",
  "percent-encoding",
  "rand 0.8.5",
@@ -3769,6 +3827,7 @@ dependencies = [
  "thiserror",
  "tokio",
  "tokio-stream",
+ "tracing",
 ]
 
 [[package]]
@@ -4417,7 +4476,7 @@ dependencies = [
  "bytes",
  "crc32c",
  "criterion",
- "env_logger",
+ "env_logger 0.10.2",
  "log",
  "memoffset 0.9.0",
  "once_cell",
@@ -4458,7 +4517,7 @@ dependencies = [
  "cfg-if",
  "criterion",
  "findshlibs",
- "inferno",
+ "inferno 0.11.21",
  "libc",
  "log",
  "nix 0.26.4",
@@ -4493,9 +4552,9 @@ checksum = "5b40af805b3121feab8a3c29f04d8ad262fa8e0561883e7653e024ae4479e6de"
 
 [[package]]
 name = "pq-sys"
-version = "0.4.8"
+version = "0.6.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "31c0052426df997c0cbd30789eb44ca097e3541717a7b8fa36b1c464ee7edebd"
+checksum = "f6cc05d7ea95200187117196eee9edd0644424911821aeb28a18ce60ea0b8793"
 dependencies = [
  "vcpkg",
 ]
@@ -4684,9 +4743,9 @@ dependencies = [
  "clap",
  "compute_api",
  "consumption_metrics",
- "dashmap",
+ "dashmap 5.5.0",
  "ecdsa 0.16.9",
- "env_logger",
+ "env_logger 0.10.2",
  "fallible-iterator",
  "flate2",
  "framed-websockets",
@@ -4757,7 +4816,7 @@ dependencies = [
  "tokio-postgres",
  "tokio-postgres2",
  "tokio-rustls 0.26.0",
- "tokio-tungstenite",
+ "tokio-tungstenite 0.21.0",
  "tokio-util",
  "tracing",
  "tracing-subscriber",
@@ -4793,6 +4852,15 @@ dependencies = [
  "serde",
 ]
 
+[[package]]
+name = "quick-xml"
+version = "0.37.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f22f29bdff3987b4d8632ef95fd6424ec7e4e0a57e2f4fc63e489e75357f6a03"
+dependencies = [
+ "memchr",
+]
+
 [[package]]
 name = "quote"
 version = "1.0.37"
@@ -5177,15 +5245,15 @@ dependencies = [
 
 [[package]]
 name = "reqwest-tracing"
-version = "0.5.4"
+version = "0.5.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ff82cf5730a1311fb9413b0bc2b8e743e0157cd73f010ab4ec374a923873b6a2"
+checksum = "73e6153390585f6961341b50e5a1931d6be6dee4292283635903c26ef9d980d2"
 dependencies = [
  "anyhow",
  "async-trait",
  "getrandom 0.2.11",
  "http 1.1.0",
- "matchit 0.8.2",
+ "matchit 0.8.4",
  "opentelemetry",
  "reqwest",
  "reqwest-middleware",
@@ -6799,7 +6867,19 @@ dependencies = [
  "futures-util",
  "log",
  "tokio",
- "tungstenite",
+ "tungstenite 0.21.0",
+]
+
+[[package]]
+name = "tokio-tungstenite"
+version = "0.24.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "edc5f74e248dc973e0dbb7b74c7e0d6fcc301c694ff50049504004ef4d0cdcd9"
+dependencies = [
+ "futures-util",
+ "log",
+ "tokio",
+ "tungstenite 0.24.0",
 ]
 
 [[package]]
@@ -6880,7 +6960,7 @@ dependencies = [
  "tokio",
  "tokio-rustls 0.26.0",
  "tokio-stream",
- "tower",
+ "tower 0.4.13",
  "tower-layer",
  "tower-service",
  "tracing",
@@ -6921,16 +7001,49 @@ dependencies = [
 ]
 
 [[package]]
-name = "tower-layer"
-version = "0.3.2"
+name = "tower"
+version = "0.5.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "c20c8dbed6283a09604c3e69b4b7eeb54e298b8a600d4d5ecb5ad39de609f1d0"
+checksum = "d039ad9159c98b70ecfd540b2573b97f7f52c3e8d9f8ad57a24b916a536975f9"
+dependencies = [
+ "futures-core",
+ "futures-util",
+ "pin-project-lite",
+ "sync_wrapper 1.0.1",
+ "tokio",
+ "tower-layer",
+ "tower-service",
+ "tracing",
+]
+
+[[package]]
+name = "tower-http"
+version = "0.6.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "403fa3b783d4b626a8ad51d766ab03cb6d2dbfc46b1c5d4448395e6628dc9697"
+dependencies = [
+ "bitflags 2.4.1",
+ "bytes",
+ "http 1.1.0",
+ "http-body 1.0.0",
+ "pin-project-lite",
+ "tower-layer",
+ "tower-service",
+ "tracing",
+ "uuid",
+]
+
+[[package]]
+name = "tower-layer"
+version = "0.3.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "121c2a6cda46980bb0fcd1647ffaf6cd3fc79a013de288782836f6df9c48780e"
 
 [[package]]
 name = "tower-service"
-version = "0.3.2"
+version = "0.3.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "b6bc1c9ce2b5135ac7f93c72918fc37feb872bdc6a5533a8b85eb4b86bfdae52"
+checksum = "8df9b6e13f2d32c91b9bd719c00d1958837bc7dec474d94952798cc8e69eeec3"
 
 [[package]]
 name = "tracing"
@@ -6999,9 +7112,9 @@ dependencies = [
 
 [[package]]
 name = "tracing-opentelemetry"
-version = "0.27.0"
+version = "0.28.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "dc58af5d3f6c5811462cabb3289aec0093f7338e367e5a33d28c0433b3c7360b"
+checksum = "97a971f6058498b5c0f1affa23e7ea202057a7301dbff68e968b2d578bcbd053"
 dependencies = [
  "js-sys",
  "once_cell",
@@ -7085,6 +7198,24 @@ dependencies = [
  "utf-8",
 ]
 
+[[package]]
+name = "tungstenite"
+version = "0.24.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "18e5b8366ee7a95b16d32197d0b2604b43a0be89dc5fac9f8e96ccafbaedda8a"
+dependencies = [
+ "byteorder",
+ "bytes",
+ "data-encoding",
+ "http 1.1.0",
+ "httparse",
+ "log",
+ "rand 0.8.5",
+ "sha1",
+ "thiserror",
+ "utf-8",
+]
+
 [[package]]
 name = "twox-hash"
 version = "1.6.3"
@@ -7252,6 +7383,7 @@ dependencies = [
  "hex-literal",
  "humantime",
  "hyper 0.14.30",
+ "inferno 0.12.0",
  "itertools 0.10.5",
  "jemalloc_pprof",
  "jsonwebtoken",
@@ -7355,7 +7487,7 @@ dependencies = [
  "anyhow",
  "camino-tempfile",
  "clap",
- "env_logger",
+ "env_logger 0.10.2",
  "log",
  "postgres",
  "postgres_ffi",
@@ -7866,7 +7998,8 @@ dependencies = [
  "tokio-util",
  "toml_edit",
  "tonic",
- "tower",
+ "tower 0.4.13",
+ "tower 0.5.2",
  "tracing",
  "tracing-core",
  "url",

Cargo.toml

@@ -65,7 +65,7 @@ aws-smithy-types = "1.2"
 aws-credential-types = "1.2.0"
 aws-sigv4 = { version = "1.2", features = ["sign-http"] }
 aws-types = "1.3"
-axum = { version = "0.7.5", features = ["ws"] }
+axum = { version = "0.7.9", features = ["ws"] }
 base64 = "0.13.0"
 bincode = "1.3"
 bindgen = "0.70"
@@ -110,6 +110,7 @@ hyper-util = "0.1"
 tokio-tungstenite = "0.21.0"
 indexmap = "2"
 indoc = "2"
+inferno = "0.12.0"
 ipnet = "2.10.0"
 itertools = "0.10"
 itoa = "1.0.11"
@@ -126,16 +127,16 @@ notify = "6.0.0"
 num_cpus = "1.15"
 num-traits = "0.2.15"
 once_cell = "1.13"
-opentelemetry = "0.26"
-opentelemetry_sdk = "0.26"
-opentelemetry-otlp = { version = "0.26", default-features=false, features = ["http-proto", "trace", "http", "reqwest-client"] }
-opentelemetry-semantic-conventions = "0.26"
+opentelemetry = "0.27"
+opentelemetry_sdk = "0.27"
+opentelemetry-otlp = { version = "0.27", default-features = false, features = ["http-proto", "trace", "http", "reqwest-client"] }
+opentelemetry-semantic-conventions = "0.27"
 parking_lot = "0.12"
 parquet = { version = "53", default-features = false, features = ["zstd"] }
 parquet_derive = "53"
 pbkdf2 = { version = "0.12.1", features = ["simple", "std"] }
 pin-project-lite = "0.2"
-pprof = { version = "0.14", features = ["criterion", "flamegraph", "protobuf", "protobuf-codec"] }
+pprof = { version = "0.14", features = ["criterion", "flamegraph", "frame-pointer", "protobuf", "protobuf-codec"] }
 procfs = "0.16"
 prometheus = {version = "0.13", default-features=false, features = ["process"]} # removes protobuf dependency
 prost = "0.13"
@@ -143,7 +144,7 @@ rand = "0.8"
 redis = { version = "0.25.2", features = ["tokio-rustls-comp", "keep-alive"] }
 regex = "1.10.2"
 reqwest = { version = "0.12", default-features = false, features = ["rustls-tls"] }
-reqwest-tracing = { version = "0.5", features = ["opentelemetry_0_26"] }
+reqwest-tracing = { version = "0.5", features = ["opentelemetry_0_27"] }
 reqwest-middleware = "0.4"
 reqwest-retry = "0.7"
 routerify = "3"
@@ -187,10 +188,12 @@ tokio-util = { version = "0.7.10", features = ["io", "rt"] }
 toml = "0.8"
 toml_edit = "0.22"
 tonic = {version = "0.12.3", features = ["tls", "tls-roots"]}
-tower-service = "0.3.2"
+tower = { version = "0.5.2", default-features = false }
+tower-http = { version = "0.6.2", features = ["request-id", "trace"] }
+tower-service = "0.3.3"
 tracing = "0.1"
 tracing-error = "0.2"
-tracing-opentelemetry = "0.27"
+tracing-opentelemetry = "0.28"
 tracing-subscriber = { version = "0.3", default-features = false, features = ["smallvec", "fmt", "tracing-log", "std", "env-filter", "json"] }
 try-lock = "0.2.5"
 twox-hash = { version = "1.6.3", default-features = false }
@@ -266,6 +269,8 @@ tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", br
 [profile.release]
 # This is useful for profiling and, to some extent, debug.
 # Besides, debug info should not affect the performance.
+#
+# NB: we also enable frame pointers for improved profiling, see .cargo/config.toml.
 debug = true
 
 # disable debug symbols for all packages except this one to decrease binaries size

Dockerfile

@@ -45,7 +45,7 @@ COPY --chown=nonroot . .
 
 ARG ADDITIONAL_RUSTFLAGS
 RUN set -e \
-    && PQ_LIB_DIR=$(pwd)/pg_install/v${STABLE_PG_VERSION}/lib RUSTFLAGS="-Clinker=clang -Clink-arg=-fuse-ld=mold -Clink-arg=-Wl,--no-rosegment ${ADDITIONAL_RUSTFLAGS}" cargo build \
+    && PQ_LIB_DIR=$(pwd)/pg_install/v${STABLE_PG_VERSION}/lib RUSTFLAGS="-Clinker=clang -Clink-arg=-fuse-ld=mold -Clink-arg=-Wl,--no-rosegment -Cforce-frame-pointers=yes ${ADDITIONAL_RUSTFLAGS}" cargo build \
       --bin pg_sni_router  \
       --bin pageserver  \
       --bin pagectl  \
@@ -69,6 +69,8 @@ RUN set -e \
         libreadline-dev \
         libseccomp-dev \
         ca-certificates \
+	# System postgres for use with client libraries (e.g. in storage controller)
+        postgresql-15 \
     && rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/* \
     && useradd -d /data neon \
     && chown -R neon:neon /data
@@ -101,11 +103,6 @@ RUN mkdir -p /data/.neon/ && \
   > /data/.neon/pageserver.toml && \
   chown -R neon:neon /data/.neon
 
-# When running a binary that links with libpq, default to using our most recent postgres version.  Binaries
-# that want a particular postgres version will select it explicitly: this is just a default.
-ENV LD_LIBRARY_PATH=/usr/local/v${DEFAULT_PG_VERSION}/lib
-
-
 VOLUME ["/data"]
 USER neon
 EXPOSE 6400

build-tools.Dockerfile

@@ -258,7 +258,7 @@ 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=1.83.0
+ENV RUSTC_VERSION=1.84.0
 ENV RUSTUP_HOME="/home/nonroot/.rustup"
 ENV PATH="/home/nonroot/.cargo/bin:${PATH}"
 ARG RUSTFILT_VERSION=0.2.1

compute/compute-node.Dockerfile

@@ -976,22 +976,9 @@ RUN apt update && apt install --no-install-recommends --no-install-suggests -y p
 
 FROM rust-extensions-build-pgrx12 AS pg-jsonschema-pg-build
 ARG PG_VERSION
-# version 0.3.3 supports v17
 # last release v0.3.3 - Oct 16, 2024
-#
-# there were no breaking changes
-# so we can use the same version for all postgres versions
-RUN case "${PG_VERSION}" in \
-    "v14" | "v15" | "v16" | "v17") \
-        export PG_JSONSCHEMA_VERSION=0.3.3 \
-        export PG_JSONSCHEMA_CHECKSUM=40c2cffab4187e0233cb8c3bde013be92218c282f95f4469c5282f6b30d64eac \
-    ;; \
-    *) \
-        echo "unexpected PostgreSQL version" && exit 1 \
-    ;; \
-    esac && \
-    wget https://github.com/supabase/pg_jsonschema/archive/refs/tags/v${PG_JSONSCHEMA_VERSION}.tar.gz -O pg_jsonschema.tar.gz && \
-    echo "${PG_JSONSCHEMA_CHECKSUM} pg_jsonschema.tar.gz" | sha256sum --check && \
+RUN wget https://github.com/supabase/pg_jsonschema/archive/refs/tags/v0.3.3.tar.gz -O pg_jsonschema.tar.gz && \
+    echo "40c2cffab4187e0233cb8c3bde013be92218c282f95f4469c5282f6b30d64eac pg_jsonschema.tar.gz" | sha256sum --check && \
     mkdir pg_jsonschema-src && cd pg_jsonschema-src && tar xzf ../pg_jsonschema.tar.gz --strip-components=1 -C . && \
     # see commit 252b3685a27a0f4c31a0f91e983c6314838e89e8
     # `unsafe-postgres` feature allows to build pgx extensions
@@ -1012,22 +999,9 @@ RUN case "${PG_VERSION}" in \
 FROM rust-extensions-build-pgrx12 AS pg-graphql-pg-build
 ARG PG_VERSION
 
-# version 1.5.9 supports v17
 # last release v1.5.9 - Oct 16, 2024
-#
-# there were no breaking changes
-# so we can use the same version for all postgres versions
-RUN case "${PG_VERSION}" in \
-    "v14" | "v15" | "v16" | "v17") \
-        export PG_GRAPHQL_VERSION=1.5.9 \
-        export PG_GRAPHQL_CHECKSUM=cf768385a41278be1333472204fc0328118644ae443182cf52f7b9b23277e497 \
-    ;; \
-    *) \
-        echo "unexpected PostgreSQL version" && exit 1 \
-    ;; \
-    esac && \
-    wget https://github.com/supabase/pg_graphql/archive/refs/tags/v${PG_GRAPHQL_VERSION}.tar.gz -O pg_graphql.tar.gz && \
-    echo "${PG_GRAPHQL_CHECKSUM} pg_graphql.tar.gz" | sha256sum --check && \
+RUN wget https://github.com/supabase/pg_graphql/archive/refs/tags/v1.5.9.tar.gz -O pg_graphql.tar.gz && \
+    echo "cf768385a41278be1333472204fc0328118644ae443182cf52f7b9b23277e497 pg_graphql.tar.gz" | sha256sum --check && \
     mkdir pg_graphql-src && cd pg_graphql-src && tar xzf ../pg_graphql.tar.gz --strip-components=1 -C . && \
     sed -i 's/pgrx = "=0.12.6"/pgrx = { version = "0.12.6", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
     cargo pgrx install --release && \
@@ -1167,22 +1141,13 @@ FROM rust-extensions-build AS pg-mooncake-build
 ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
-# The topmost commit in the `neon` branch at the time of writing this
-# https://github.com/Mooncake-Labs/pg_mooncake/commits/neon/
-# https://github.com/Mooncake-Labs/pg_mooncake/commit/077c92c452bb6896a7b7776ee95f039984f076af
-ENV PG_MOONCAKE_VERSION=077c92c452bb6896a7b7776ee95f039984f076af
 ENV PATH="/usr/local/pgsql/bin/:$PATH"
 
-RUN case "${PG_VERSION}" in \
-        'v14') \
-            echo "pg_mooncake is not supported on Postgres ${PG_VERSION}" && exit 0;; \
-    esac && \
-    git clone --depth 1 --branch neon https://github.com/Mooncake-Labs/pg_mooncake.git pg_mooncake-src && \
-    cd pg_mooncake-src && \
-    git checkout "${PG_MOONCAKE_VERSION}" && \
-    git submodule update --init --depth 1 --recursive && \
-    make BUILD_TYPE=release -j $(getconf _NPROCESSORS_ONLN) && \
-    make BUILD_TYPE=release -j $(getconf _NPROCESSORS_ONLN) install && \
+RUN wget https://github.com/Mooncake-Labs/pg_mooncake/releases/download/v0.1.0/pg_mooncake-0.1.0.tar.gz -O pg_mooncake.tar.gz && \
+    echo "eafd059b77f541f11525eb8affcd66a176968cbd8fe7c0d436e733f2aa4da59f pg_mooncake.tar.gz" | sha256sum --check && \
+    mkdir pg_mooncake-src && cd pg_mooncake-src && tar xzf ../pg_mooncake.tar.gz --strip-components=1 -C . && \
+    make release -j $(getconf _NPROCESSORS_ONLN) && \
+    make install -j $(getconf _NPROCESSORS_ONLN) && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/pg_mooncake.control
 
 #########################################################################################
@@ -1267,25 +1232,11 @@ RUN make -j $(getconf _NPROCESSORS_ONLN) \
     make -j $(getconf _NPROCESSORS_ONLN) \
         PG_CONFIG=/usr/local/pgsql/bin/pg_config \
         -C pgxn/neon_rmgr \
-        -s install && \
-    case "${PG_VERSION}" in \
-        "v14" | "v15") \
-        ;; \
-        "v16" | "v17") \
-            echo "Skipping HNSW for PostgreSQL ${PG_VERSION}" && exit 0 \
-        ;; \
-        *) \
-            echo "unexpected PostgreSQL version" && exit 1 \
-        ;; \
-        esac && \
-    make -j $(getconf _NPROCESSORS_ONLN) \
-        PG_CONFIG=/usr/local/pgsql/bin/pg_config \
-        -C pgxn/hnsw \
         -s install
 
 #########################################################################################
 #
-# Compile and run the Neon-specific `compute_ctl` and `fast_import` binaries
+# Compile the Neon-specific `compute_ctl`, `fast_import`, and `local_proxy` binaries
 #
 #########################################################################################
 FROM $REPOSITORY/$IMAGE:$TAG AS compute-tools
@@ -1295,18 +1246,7 @@ ENV BUILD_TAG=$BUILD_TAG
 USER nonroot
 # Copy entire project to get Cargo.* files with proper dependencies for the whole project
 COPY --chown=nonroot . .
-RUN cd compute_tools && mold -run cargo build --locked --profile release-line-debug-size-lto
-
-#########################################################################################
-#
-# Final compute-tools image
-#
-#########################################################################################
-
-FROM debian:$DEBIAN_FLAVOR AS compute-tools-image
-
-COPY --from=compute-tools /home/nonroot/target/release-line-debug-size-lto/compute_ctl /usr/local/bin/compute_ctl
-COPY --from=compute-tools /home/nonroot/target/release-line-debug-size-lto/fast_import /usr/local/bin/fast_import
+RUN mold -run cargo build --locked --profile release-line-debug-size-lto --bin compute_ctl --bin fast_import --bin local_proxy
 
 #########################################################################################
 #
@@ -1338,20 +1278,6 @@ RUN set -e \
     && make -j $(nproc) dist_man_MANS= \
     && make install dist_man_MANS=
 
-#########################################################################################
-#
-# Compile the Neon-specific `local_proxy` binary
-#
-#########################################################################################
-FROM $REPOSITORY/$IMAGE:$TAG AS local_proxy
-ARG BUILD_TAG
-ENV BUILD_TAG=$BUILD_TAG
-
-USER nonroot
-# Copy entire project to get Cargo.* files with proper dependencies for the whole project
-COPY --chown=nonroot . .
-RUN mold -run cargo build --locked --profile release-line-debug-size-lto --bin local_proxy
-
 #########################################################################################
 #
 # Layers "postgres-exporter" and "sql-exporter"
@@ -1491,7 +1417,7 @@ COPY --from=pgbouncer         /usr/local/pgbouncer/bin/pgbouncer /usr/local/bin/
 COPY --chmod=0666 --chown=postgres compute/etc/pgbouncer.ini /etc/pgbouncer.ini
 
 # local_proxy and its config
-COPY --from=local_proxy --chown=postgres /home/nonroot/target/release-line-debug-size-lto/local_proxy /usr/local/bin/local_proxy
+COPY --from=compute-tools --chown=postgres /home/nonroot/target/release-line-debug-size-lto/local_proxy /usr/local/bin/local_proxy
 RUN mkdir -p /etc/local_proxy && chown postgres:postgres /etc/local_proxy
 
 # Metrics exporter binaries and  configuration files
@@ -1556,28 +1482,30 @@ RUN apt update && \
         locales \
         procps \
         ca-certificates \
+        curl \
+        unzip \
         $VERSION_INSTALLS && \
     apt clean && rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/* && \
     localedef -i en_US -c -f UTF-8 -A /usr/share/locale/locale.alias en_US.UTF-8
 
-# s5cmd 2.2.2 from https://github.com/peak/s5cmd/releases/tag/v2.2.2
-# used by fast_import
+# aws cli is used by fast_import (curl and unzip above are at this time only used for this installation step)
 ARG TARGETARCH
-ADD https://github.com/peak/s5cmd/releases/download/v2.2.2/s5cmd_2.2.2_linux_$TARGETARCH.deb /tmp/s5cmd.deb
 RUN set -ex; \
-    \
-    # Determine the expected checksum based on TARGETARCH
     if [ "${TARGETARCH}" = "amd64" ]; then \
-        CHECKSUM="392c385320cd5ffa435759a95af77c215553d967e4b1c0fffe52e4f14c29cf85"; \
+        TARGETARCH_ALT="x86_64"; \
+        CHECKSUM="c9a9df3770a3ff9259cb469b6179e02829687a464e0824d5c32d378820b53a00"; \
     elif [ "${TARGETARCH}" = "arm64" ]; then \
-        CHECKSUM="939bee3cf4b5604ddb00e67f8c157b91d7c7a5b553d1fbb6890fad32894b7b46"; \
+        TARGETARCH_ALT="aarch64"; \
+        CHECKSUM="8181730be7891582b38b028112e81b4899ca817e8c616aad807c9e9d1289223a"; \
     else \
         echo "Unsupported architecture: ${TARGETARCH}"; exit 1; \
     fi; \
-    \
-    # Compute and validate the checksum
-    echo "${CHECKSUM}  /tmp/s5cmd.deb" | sha256sum -c -
-RUN dpkg -i /tmp/s5cmd.deb && rm /tmp/s5cmd.deb
+    curl -L "https://awscli.amazonaws.com/awscli-exe-linux-${TARGETARCH_ALT}-2.17.5.zip" -o /tmp/awscliv2.zip; \
+    echo "${CHECKSUM}  /tmp/awscliv2.zip" | sha256sum -c -; \
+    unzip /tmp/awscliv2.zip -d /tmp/awscliv2; \
+    /tmp/awscliv2/aws/install; \
+    rm -rf /tmp/awscliv2.zip /tmp/awscliv2; \
+    true
 
 ENV LANG=en_US.utf8
 USER postgres

compute_tools/Cargo.toml

@@ -7,7 +7,7 @@ license.workspace = true
 [features]
 default = []
 # Enables test specific features.
-testing = []
+testing = ["fail/failpoints"]
 
 [dependencies]
 base64.workspace = true
@@ -15,13 +15,15 @@ aws-config.workspace = true
 aws-sdk-s3.workspace = true
 aws-sdk-kms.workspace = true
 anyhow.workspace = true
+axum = { workspace = true, features = [] }
 camino.workspace = true
 chrono.workspace = true
 cfg-if.workspace = true
 clap.workspace = true
+fail.workspace = true
 flate2.workspace = true
 futures.workspace = true
-hyper0 = { workspace = true, features = ["full"] }
+http.workspace = true
 metrics.workspace = true
 nix.workspace = true
 notify.workspace = true
@@ -36,6 +38,8 @@ serde_with.workspace = true
 serde_json.workspace = true
 signal-hook.workspace = true
 tar.workspace = true
+tower.workspace = true
+tower-http.workspace = true
 reqwest = { workspace = true, features = ["json"] }
 tokio = { workspace = true, features = ["rt", "rt-multi-thread"] }
 tokio-postgres.workspace = true

compute_tools/src/bin/compute_ctl.rs

@@ -60,19 +60,22 @@ use compute_tools::compute::{
 };
 use compute_tools::configurator::launch_configurator;
 use compute_tools::extension_server::get_pg_version_string;
-use compute_tools::http::api::launch_http_server;
+use compute_tools::http::launch_http_server;
 use compute_tools::logger::*;
 use compute_tools::monitor::launch_monitor;
 use compute_tools::params::*;
 use compute_tools::spec::*;
 use compute_tools::swap::resize_swap;
 use rlimit::{setrlimit, Resource};
+use utils::failpoint_support;
 
 // this is an arbitrary build tag. Fine as a default / for testing purposes
 // in-case of not-set environment var
 const BUILD_TAG_DEFAULT: &str = "latest";
 
 fn main() -> Result<()> {
+    let scenario = failpoint_support::init();
+
     let (build_tag, clap_args) = init()?;
 
     // enable core dumping for all child processes
@@ -100,17 +103,14 @@ fn main() -> Result<()> {
 
     maybe_delay_exit(delay_exit);
 
+    scenario.teardown();
+
     deinit_and_exit(wait_pg_result);
 }
 
 fn init() -> Result<(String, clap::ArgMatches)> {
     init_tracing_and_logging(DEFAULT_LOG_LEVEL)?;
 
-    opentelemetry::global::set_error_handler(|err| {
-        tracing::info!("OpenTelemetry error: {err}");
-    })
-    .expect("global error handler lock poisoned");
-
     let mut signals = Signals::new([SIGINT, SIGTERM, SIGQUIT])?;
     thread::spawn(move || {
         for sig in signals.forever() {
@@ -419,9 +419,14 @@ fn start_postgres(
         "running compute with features: {:?}",
         state.pspec.as_ref().unwrap().spec.features
     );
-    // before we release the mutex, fetch the swap size (if any) for later.
-    let swap_size_bytes = state.pspec.as_ref().unwrap().spec.swap_size_bytes;
-    let disk_quota_bytes = state.pspec.as_ref().unwrap().spec.disk_quota_bytes;
+    // before we release the mutex, fetch some parameters for later.
+    let &ComputeSpec {
+        swap_size_bytes,
+        disk_quota_bytes,
+        #[cfg(target_os = "linux")]
+        disable_lfc_resizing,
+        ..
+    } = &state.pspec.as_ref().unwrap().spec;
     drop(state);
 
     // Launch remaining service threads
@@ -483,7 +488,10 @@ fn start_postgres(
     let mut pg = None;
     if !prestartup_failed {
         pg = match compute.start_compute() {
-            Ok(pg) => Some(pg),
+            Ok(pg) => {
+                info!(postmaster_pid = %pg.0.id(), "Postgres was started");
+                Some(pg)
+            }
             Err(err) => {
                 error!("could not start the compute node: {:#}", err);
                 compute.set_failed_status(err);
@@ -526,11 +534,18 @@ fn start_postgres(
             // This token is used internally by the monitor to clean up all threads
             let token = CancellationToken::new();
 
+            // don't pass postgres connection string to vm-monitor if we don't want it to resize LFC
+            let pgconnstr = if disable_lfc_resizing.unwrap_or(false) {
+                None
+            } else {
+                file_cache_connstr.cloned()
+            };
+
             let vm_monitor = rt.as_ref().map(|rt| {
                 rt.spawn(vm_monitor::start(
                     Box::leak(Box::new(vm_monitor::Args {
                         cgroup: cgroup.cloned(),
-                        pgconnstr: file_cache_connstr.cloned(),
+                        pgconnstr,
                         addr: vm_monitor_addr.clone(),
                     })),
                     token.clone(),
@@ -574,6 +589,8 @@ fn wait_postgres(pg: Option<PostgresHandle>) -> Result<WaitPostgresResult> {
     // propagate to Postgres and it will be shut down as well.
     let mut exit_code = None;
     if let Some((mut pg, logs_handle)) = pg {
+        info!(postmaster_pid = %pg.id(), "Waiting for Postgres to exit");
+
         let ecode = pg
             .wait()
             .expect("failed to start waiting on Postgres process");

compute_tools/src/bin/fast_import.rs

@@ -17,7 +17,7 @@
 //!
 //! # Local Testing
 //!
-//! - Comment out most of the pgxns in The Dockerfile.compute-tools to speed up the build.
+//! - Comment out most of the pgxns in compute-node.Dockerfile to speed up the build.
 //! - Build the image with the following command:
 //!
 //! ```bash
@@ -34,12 +34,12 @@ use nix::unistd::Pid;
 use tracing::{info, info_span, warn, Instrument};
 use utils::fs_ext::is_directory_empty;
 
+#[path = "fast_import/aws_s3_sync.rs"]
+mod aws_s3_sync;
 #[path = "fast_import/child_stdio_to_log.rs"]
 mod child_stdio_to_log;
 #[path = "fast_import/s3_uri.rs"]
 mod s3_uri;
-#[path = "fast_import/s5cmd.rs"]
-mod s5cmd;
 
 #[derive(clap::Parser)]
 struct Args {
@@ -326,7 +326,7 @@ pub(crate) async fn main() -> anyhow::Result<()> {
     }
 
     info!("upload pgdata");
-    s5cmd::sync(Utf8Path::new(&pgdata_dir), &s3_prefix.append("/"))
+    aws_s3_sync::sync(Utf8Path::new(&pgdata_dir), &s3_prefix.append("/pgdata/"))
         .await
         .context("sync dump directory to destination")?;
 
@@ -334,10 +334,10 @@ pub(crate) async fn main() -> anyhow::Result<()> {
     {
         let status_dir = working_directory.join("status");
         std::fs::create_dir(&status_dir).context("create status directory")?;
-        let status_file = status_dir.join("status");
+        let status_file = status_dir.join("pgdata");
         std::fs::write(&status_file, serde_json::json!({"done": true}).to_string())
             .context("write status file")?;
-        s5cmd::sync(&status_file, &s3_prefix.append("/status/pgdata"))
+        aws_s3_sync::sync(&status_dir, &s3_prefix.append("/status/"))
             .await
             .context("sync status directory to destination")?;
     }

compute_tools/src/bin/fast_import/aws_s3_sync.rs

@@ -4,24 +4,21 @@ use camino::Utf8Path;
 use super::s3_uri::S3Uri;
 
 pub(crate) async fn sync(local: &Utf8Path, remote: &S3Uri) -> anyhow::Result<()> {
-    let mut builder = tokio::process::Command::new("s5cmd");
-    // s5cmd uses aws-sdk-go v1, hence doesn't support AWS_ENDPOINT_URL
-    if let Some(val) = std::env::var_os("AWS_ENDPOINT_URL") {
-        builder.arg("--endpoint-url").arg(val);
-    }
+    let mut builder = tokio::process::Command::new("aws");
     builder
+        .arg("s3")
         .arg("sync")
         .arg(local.as_str())
         .arg(remote.to_string());
     let st = builder
         .spawn()
-        .context("spawn s5cmd")?
+        .context("spawn aws s3 sync")?
         .wait()
         .await
-        .context("wait for s5cmd")?;
+        .context("wait for aws s3 sync")?;
     if st.success() {
         Ok(())
     } else {
-        Err(anyhow::anyhow!("s5cmd failed"))
+        Err(anyhow::anyhow!("aws s3 sync failed"))
     }
 }

compute_tools/src/catalog.rs

@@ -36,11 +36,11 @@ pub async fn get_dbs_and_roles(compute: &Arc<ComputeNode>) -> anyhow::Result<Cat
 
 #[derive(Debug, thiserror::Error)]
 pub enum SchemaDumpError {
-    #[error("Database does not exist.")]
+    #[error("database does not exist")]
     DatabaseDoesNotExist,
-    #[error("Failed to execute pg_dump.")]
+    #[error("failed to execute pg_dump")]
     IO(#[from] std::io::Error),
-    #[error("Unexpected error.")]
+    #[error("unexpected I/O error")]
     Unexpected,
 }
 

compute_tools/src/compute.rs

@@ -15,7 +15,7 @@ use std::time::Instant;
 
 use anyhow::{Context, Result};
 use chrono::{DateTime, Utc};
-use compute_api::spec::{PgIdent, Role};
+use compute_api::spec::{Database, PgIdent, Role};
 use futures::future::join_all;
 use futures::stream::FuturesUnordered;
 use futures::StreamExt;
@@ -45,8 +45,10 @@ use crate::spec_apply::ApplySpecPhase::{
     DropInvalidDatabases, DropRoles, HandleNeonExtension, HandleOtherExtensions,
     RenameAndDeleteDatabases, RenameRoles, RunInEachDatabase,
 };
+use crate::spec_apply::PerDatabasePhase;
 use crate::spec_apply::PerDatabasePhase::{
-    ChangeSchemaPerms, DeleteDBRoleReferences, HandleAnonExtension,
+    ChangeSchemaPerms, DeleteDBRoleReferences, DropSubscriptionsForDeletedDatabases,
+    HandleAnonExtension,
 };
 use crate::spec_apply::{apply_operations, MutableApplyContext, DB};
 use crate::sync_sk::{check_if_synced, ping_safekeeper};
@@ -834,7 +836,7 @@ impl ComputeNode {
         conf
     }
 
-    async fn get_maintenance_client(
+    pub async fn get_maintenance_client(
         conf: &tokio_postgres::Config,
     ) -> Result<tokio_postgres::Client> {
         let mut conf = conf.clone();
@@ -943,6 +945,78 @@ impl ComputeNode {
                 dbs: databases,
             }));
 
+            // Apply special pre drop database phase.
+            // NOTE: we use the code of RunInEachDatabase phase for parallelism
+            // and connection management, but we don't really run it in *each* database,
+            // only in databases, we're about to drop.
+            info!("Applying PerDatabase (pre-dropdb) phase");
+            let concurrency_token = Arc::new(tokio::sync::Semaphore::new(concurrency));
+
+            // Run the phase for each database that we're about to drop.
+            let db_processes = spec
+                .delta_operations
+                .iter()
+                .flatten()
+                .filter_map(move |op| {
+                    if op.action.as_str() == "delete_db" {
+                        Some(op.name.clone())
+                    } else {
+                        None
+                    }
+                })
+                .map(|dbname| {
+                    let spec = spec.clone();
+                    let ctx = ctx.clone();
+                    let jwks_roles = jwks_roles.clone();
+                    let mut conf = conf.as_ref().clone();
+                    let concurrency_token = concurrency_token.clone();
+                    // We only need dbname field for this phase, so set other fields to dummy values
+                    let db = DB::UserDB(Database {
+                        name: dbname.clone(),
+                        owner: "cloud_admin".to_string(),
+                        options: None,
+                        restrict_conn: false,
+                        invalid: false,
+                    });
+
+                    debug!("Applying per-database phases for Database {:?}", &db);
+
+                    match &db {
+                        DB::SystemDB => {}
+                        DB::UserDB(db) => {
+                            conf.dbname(db.name.as_str());
+                        }
+                    }
+
+                    let conf = Arc::new(conf);
+                    let fut = Self::apply_spec_sql_db(
+                        spec.clone(),
+                        conf,
+                        ctx.clone(),
+                        jwks_roles.clone(),
+                        concurrency_token.clone(),
+                        db,
+                        [DropSubscriptionsForDeletedDatabases].to_vec(),
+                    );
+
+                    Ok(spawn(fut))
+                })
+                .collect::<Vec<Result<_, anyhow::Error>>>();
+
+            for process in db_processes.into_iter() {
+                let handle = process?;
+                if let Err(e) = handle.await? {
+                    // Handle the error case where the database does not exist
+                    // We do not check whether the DB exists or not in the deletion phase,
+                    // so we shouldn't be strict about it in pre-deletion cleanup as well.
+                    if e.to_string().contains("does not exist") {
+                        warn!("Error dropping subscription: {}", e);
+                    } else {
+                        return Err(e);
+                    }
+                };
+            }
+
             for phase in [
                 CreateSuperUser,
                 DropInvalidDatabases,
@@ -962,7 +1036,7 @@ impl ComputeNode {
                 .await?;
             }
 
-            info!("Applying RunInEachDatabase phase");
+            info!("Applying RunInEachDatabase2 phase");
             let concurrency_token = Arc::new(tokio::sync::Semaphore::new(concurrency));
 
             let db_processes = spec
@@ -997,6 +1071,12 @@ impl ComputeNode {
                         jwks_roles.clone(),
                         concurrency_token.clone(),
                         db,
+                        [
+                            DeleteDBRoleReferences,
+                            ChangeSchemaPerms,
+                            HandleAnonExtension,
+                        ]
+                        .to_vec(),
                     );
 
                     Ok(spawn(fut))
@@ -1043,16 +1123,13 @@ impl ComputeNode {
         jwks_roles: Arc<HashSet<String>>,
         concurrency_token: Arc<tokio::sync::Semaphore>,
         db: DB,
+        subphases: Vec<PerDatabasePhase>,
     ) -> Result<()> {
         let _permit = concurrency_token.acquire().await?;
 
         let mut client_conn = None;
 
-        for subphase in [
-            DeleteDBRoleReferences,
-            ChangeSchemaPerms,
-            HandleAnonExtension,
-        ] {
+        for subphase in subphases {
             apply_operations(
                 spec.clone(),
                 ctx.clone(),
@@ -1181,8 +1258,19 @@ impl ComputeNode {
             let mut conf = postgres::config::Config::from(conf);
             conf.application_name("compute_ctl:migrations");
 
-            let mut client = conf.connect(NoTls)?;
-            handle_migrations(&mut client).context("apply_config handle_migrations")
+            match conf.connect(NoTls) {
+                Ok(mut client) => {
+                    if let Err(e) = handle_migrations(&mut client) {
+                        error!("Failed to run migrations: {}", e);
+                    }
+                }
+                Err(e) => {
+                    error!(
+                        "Failed to connect to the compute for running migrations: {}",
+                        e
+                    );
+                }
+            };
         });
 
         Ok::<(), anyhow::Error>(())

compute_tools/src/http/api.rs

@@ -1,591 +0,0 @@
-use std::convert::Infallible;
-use std::net::IpAddr;
-use std::net::Ipv6Addr;
-use std::net::SocketAddr;
-use std::sync::Arc;
-use std::thread;
-
-use crate::catalog::SchemaDumpError;
-use crate::catalog::{get_database_schema, get_dbs_and_roles};
-use crate::compute::forward_termination_signal;
-use crate::compute::{ComputeNode, ComputeState, ParsedSpec};
-use crate::installed_extensions;
-use compute_api::requests::{ConfigurationRequest, ExtensionInstallRequest, SetRoleGrantsRequest};
-use compute_api::responses::{
-    ComputeStatus, ComputeStatusResponse, ExtensionInstallResult, GenericAPIError,
-    SetRoleGrantsResponse,
-};
-
-use anyhow::Result;
-use hyper::header::CONTENT_TYPE;
-use hyper::service::{make_service_fn, service_fn};
-use hyper::{Body, Method, Request, Response, Server, StatusCode};
-use metrics::proto::MetricFamily;
-use metrics::Encoder;
-use metrics::TextEncoder;
-use tokio::task;
-use tracing::{debug, error, info, warn};
-use tracing_utils::http::OtelName;
-use utils::http::request::must_get_query_param;
-
-fn status_response_from_state(state: &ComputeState) -> ComputeStatusResponse {
-    ComputeStatusResponse {
-        start_time: state.start_time,
-        tenant: state
-            .pspec
-            .as_ref()
-            .map(|pspec| pspec.tenant_id.to_string()),
-        timeline: state
-            .pspec
-            .as_ref()
-            .map(|pspec| pspec.timeline_id.to_string()),
-        status: state.status,
-        last_active: state.last_active,
-        error: state.error.clone(),
-    }
-}
-
-// Service function to handle all available routes.
-async fn routes(req: Request<Body>, compute: &Arc<ComputeNode>) -> Response<Body> {
-    //
-    // NOTE: The URI path is currently included in traces. That's OK because
-    // it doesn't contain any variable parts or sensitive information. But
-    // please keep that in mind if you change the routing here.
-    //
-    match (req.method(), req.uri().path()) {
-        // Serialized compute state.
-        (&Method::GET, "/status") => {
-            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()))
-        }
-
-        // Startup metrics in JSON format. Keep /metrics reserved for a possible
-        // future use for Prometheus metrics format.
-        (&Method::GET, "/metrics.json") => {
-            info!("serving /metrics.json GET request");
-            let metrics = compute.state.lock().unwrap().metrics.clone();
-            Response::new(Body::from(serde_json::to_string(&metrics).unwrap()))
-        }
-
-        // Prometheus metrics
-        (&Method::GET, "/metrics") => {
-            debug!("serving /metrics GET request");
-
-            // When we call TextEncoder::encode() below, it will immediately
-            // return an error if a metric family has no metrics, so we need to
-            // preemptively filter out metric families with no metrics.
-            let metrics = installed_extensions::collect()
-                .into_iter()
-                .filter(|m| !m.get_metric().is_empty())
-                .collect::<Vec<MetricFamily>>();
-
-            let encoder = TextEncoder::new();
-            let mut buffer = vec![];
-
-            if let Err(err) = encoder.encode(&metrics, &mut buffer) {
-                let msg = format!("error handling /metrics request: {err}");
-                error!(msg);
-                return render_json_error(&msg, StatusCode::INTERNAL_SERVER_ERROR);
-            }
-
-            match Response::builder()
-                .status(StatusCode::OK)
-                .header(CONTENT_TYPE, encoder.format_type())
-                .body(Body::from(buffer))
-            {
-                Ok(response) => response,
-                Err(err) => {
-                    let msg = format!("error handling /metrics request: {err}");
-                    error!(msg);
-                    render_json_error(&msg, StatusCode::INTERNAL_SERVER_ERROR)
-                }
-            }
-        }
-        // Collect Postgres current usage insights
-        (&Method::GET, "/insights") => {
-            info!("serving /insights GET request");
-            let status = compute.get_status();
-            if status != ComputeStatus::Running {
-                let msg = format!("compute is not running, current status: {:?}", status);
-                error!(msg);
-                return Response::new(Body::from(msg));
-            }
-
-            let insights = compute.collect_insights().await;
-            Response::new(Body::from(insights))
-        }
-
-        (&Method::POST, "/check_writability") => {
-            info!("serving /check_writability POST request");
-            let status = compute.get_status();
-            if status != ComputeStatus::Running {
-                let msg = format!(
-                    "invalid compute status for check_writability request: {:?}",
-                    status
-                );
-                error!(msg);
-                return Response::new(Body::from(msg));
-            }
-
-            let res = crate::checker::check_writability(compute).await;
-            match res {
-                Ok(_) => Response::new(Body::from("true")),
-                Err(e) => {
-                    error!("check_writability failed: {}", e);
-                    Response::new(Body::from(e.to_string()))
-                }
-            }
-        }
-
-        (&Method::POST, "/extensions") => {
-            info!("serving /extensions POST request");
-            let status = compute.get_status();
-            if status != ComputeStatus::Running {
-                let msg = format!(
-                    "invalid compute status for extensions request: {:?}",
-                    status
-                );
-                error!(msg);
-                return render_json_error(&msg, StatusCode::PRECONDITION_FAILED);
-            }
-
-            let request = hyper::body::to_bytes(req.into_body()).await.unwrap();
-            let request = serde_json::from_slice::<ExtensionInstallRequest>(&request).unwrap();
-            let res = compute
-                .install_extension(&request.extension, &request.database, request.version)
-                .await;
-            match res {
-                Ok(version) => render_json(Body::from(
-                    serde_json::to_string(&ExtensionInstallResult {
-                        extension: request.extension,
-                        version,
-                    })
-                    .unwrap(),
-                )),
-                Err(e) => {
-                    error!("install_extension failed: {}", e);
-                    render_json_error(&e.to_string(), StatusCode::INTERNAL_SERVER_ERROR)
-                }
-            }
-        }
-
-        (&Method::GET, "/info") => {
-            let num_cpus = num_cpus::get_physical();
-            info!("serving /info GET request. num_cpus: {}", num_cpus);
-            Response::new(Body::from(
-                serde_json::json!({
-                    "num_cpus": num_cpus,
-                })
-                .to_string(),
-            ))
-        }
-
-        // Accept spec in JSON format and request compute configuration. If
-        // anything goes wrong after we set the compute status to `ConfigurationPending`
-        // and update compute state with new spec, we basically leave compute
-        // in the potentially wrong state. That said, it's control-plane's
-        // responsibility to watch compute state after reconfiguration request
-        // and to clean restart in case of errors.
-        (&Method::POST, "/configure") => {
-            info!("serving /configure POST request");
-            match handle_configure_request(req, compute).await {
-                Ok(msg) => Response::new(Body::from(msg)),
-                Err((msg, code)) => {
-                    error!("error handling /configure request: {msg}");
-                    render_json_error(&msg, code)
-                }
-            }
-        }
-
-        (&Method::POST, "/terminate") => {
-            info!("serving /terminate POST request");
-            match handle_terminate_request(compute).await {
-                Ok(()) => Response::new(Body::empty()),
-                Err((msg, code)) => {
-                    error!("error handling /terminate request: {msg}");
-                    render_json_error(&msg, code)
-                }
-            }
-        }
-
-        (&Method::GET, "/dbs_and_roles") => {
-            info!("serving /dbs_and_roles GET request",);
-            match get_dbs_and_roles(compute).await {
-                Ok(res) => render_json(Body::from(serde_json::to_string(&res).unwrap())),
-                Err(_) => {
-                    render_json_error("can't get dbs and roles", StatusCode::INTERNAL_SERVER_ERROR)
-                }
-            }
-        }
-
-        (&Method::GET, "/database_schema") => {
-            let database = match must_get_query_param(&req, "database") {
-                Err(e) => return e.into_response(),
-                Ok(database) => database,
-            };
-            info!("serving /database_schema GET request with database: {database}",);
-            match get_database_schema(compute, &database).await {
-                Ok(res) => render_plain(Body::wrap_stream(res)),
-                Err(SchemaDumpError::DatabaseDoesNotExist) => {
-                    render_json_error("database does not exist", StatusCode::NOT_FOUND)
-                }
-                Err(e) => {
-                    error!("can't get schema dump: {}", e);
-                    render_json_error("can't get schema dump", StatusCode::INTERNAL_SERVER_ERROR)
-                }
-            }
-        }
-
-        (&Method::POST, "/grants") => {
-            info!("serving /grants POST request");
-            let status = compute.get_status();
-            if status != ComputeStatus::Running {
-                let msg = format!(
-                    "invalid compute status for set_role_grants request: {:?}",
-                    status
-                );
-                error!(msg);
-                return render_json_error(&msg, StatusCode::PRECONDITION_FAILED);
-            }
-
-            let request = hyper::body::to_bytes(req.into_body()).await.unwrap();
-            let request = serde_json::from_slice::<SetRoleGrantsRequest>(&request).unwrap();
-
-            let res = compute
-                .set_role_grants(
-                    &request.database,
-                    &request.schema,
-                    &request.privileges,
-                    &request.role,
-                )
-                .await;
-            match res {
-                Ok(()) => render_json(Body::from(
-                    serde_json::to_string(&SetRoleGrantsResponse {
-                        database: request.database,
-                        schema: request.schema,
-                        role: request.role,
-                        privileges: request.privileges,
-                    })
-                    .unwrap(),
-                )),
-                Err(e) => render_json_error(
-                    &format!("could not grant role privileges to the schema: {e}"),
-                    // TODO: can we filter on role/schema not found errors
-                    // and return appropriate error code?
-                    StatusCode::INTERNAL_SERVER_ERROR,
-                ),
-            }
-        }
-
-        // get the list of installed extensions
-        // currently only used in python tests
-        // TODO: call it from cplane
-        (&Method::GET, "/installed_extensions") => {
-            info!("serving /installed_extensions GET request");
-            let status = compute.get_status();
-            if status != ComputeStatus::Running {
-                let msg = format!(
-                    "invalid compute status for extensions request: {:?}",
-                    status
-                );
-                error!(msg);
-                return Response::new(Body::from(msg));
-            }
-
-            let conf = compute.get_conn_conf(None);
-            let res =
-                task::spawn_blocking(move || installed_extensions::get_installed_extensions(conf))
-                    .await
-                    .unwrap();
-
-            match res {
-                Ok(res) => render_json(Body::from(serde_json::to_string(&res).unwrap())),
-                Err(e) => render_json_error(
-                    &format!("could not get list of installed extensions: {}", e),
-                    StatusCode::INTERNAL_SERVER_ERROR,
-                ),
-            }
-        }
-
-        // download extension files from remote extension storage on demand
-        (&Method::POST, route) if route.starts_with("/extension_server/") => {
-            info!("serving {:?} POST request", route);
-            info!("req.uri {:?}", req.uri());
-
-            // don't even try to download extensions
-            // if no remote storage is configured
-            if compute.ext_remote_storage.is_none() {
-                info!("no extensions remote storage configured");
-                let mut resp = Response::new(Body::from("no remote storage configured"));
-                *resp.status_mut() = StatusCode::INTERNAL_SERVER_ERROR;
-                return resp;
-            }
-
-            let mut is_library = false;
-            if let Some(params) = req.uri().query() {
-                info!("serving {:?} POST request with params: {}", route, params);
-                if params == "is_library=true" {
-                    is_library = true;
-                } else {
-                    let mut resp = Response::new(Body::from("Wrong request parameters"));
-                    *resp.status_mut() = StatusCode::BAD_REQUEST;
-                    return resp;
-                }
-            }
-            let filename = route.split('/').last().unwrap().to_string();
-            info!("serving /extension_server POST request, filename: {filename:?} is_library: {is_library}");
-
-            // get ext_name and path from spec
-            // don't lock compute_state for too long
-            let ext = {
-                let compute_state = compute.state.lock().unwrap();
-                let pspec = compute_state.pspec.as_ref().expect("spec must be set");
-                let spec = &pspec.spec;
-
-                // debug only
-                info!("spec: {:?}", spec);
-
-                let remote_extensions = match spec.remote_extensions.as_ref() {
-                    Some(r) => r,
-                    None => {
-                        info!("no remote extensions spec was provided");
-                        let mut resp = Response::new(Body::from("no remote storage configured"));
-                        *resp.status_mut() = StatusCode::INTERNAL_SERVER_ERROR;
-                        return resp;
-                    }
-                };
-
-                remote_extensions.get_ext(
-                    &filename,
-                    is_library,
-                    &compute.build_tag,
-                    &compute.pgversion,
-                )
-            };
-
-            match ext {
-                Ok((ext_name, ext_path)) => {
-                    match compute.download_extension(ext_name, ext_path).await {
-                        Ok(_) => Response::new(Body::from("OK")),
-                        Err(e) => {
-                            error!("extension download failed: {}", e);
-                            let mut resp = Response::new(Body::from(e.to_string()));
-                            *resp.status_mut() = StatusCode::INTERNAL_SERVER_ERROR;
-                            resp
-                        }
-                    }
-                }
-                Err(e) => {
-                    warn!("extension download failed to find extension: {}", e);
-                    let mut resp = Response::new(Body::from("failed to find file"));
-                    *resp.status_mut() = StatusCode::INTERNAL_SERVER_ERROR;
-                    resp
-                }
-            }
-        }
-
-        // Return the `404 Not Found` for any other routes.
-        _ => {
-            let mut not_found = Response::new(Body::from("404 Not Found"));
-            *not_found.status_mut() = StatusCode::NOT_FOUND;
-            not_found
-        }
-    }
-}
-
-async fn handle_configure_request(
-    req: Request<Body>,
-    compute: &Arc<ComputeNode>,
-) -> Result<String, (String, StatusCode)> {
-    if !compute.live_config_allowed {
-        return Err((
-            "live configuration is not allowed for this compute node".to_string(),
-            StatusCode::PRECONDITION_FAILED,
-        ));
-    }
-
-    let body_bytes = hyper::body::to_bytes(req.into_body()).await.unwrap();
-    let spec_raw = String::from_utf8(body_bytes.to_vec()).unwrap();
-    if let Ok(request) = serde_json::from_str::<ConfigurationRequest>(&spec_raw) {
-        let spec = request.spec;
-
-        let parsed_spec = match ParsedSpec::try_from(spec) {
-            Ok(ps) => ps,
-            Err(msg) => return Err((msg, StatusCode::BAD_REQUEST)),
-        };
-
-        // XXX: wrap state update under lock in code blocks. Otherwise,
-        // we will try to `Send` `mut state` into the spawned thread
-        // bellow, which will cause error:
-        // ```
-        // error: future cannot be sent between threads safely
-        // ```
-        {
-            let mut state = compute.state.lock().unwrap();
-            if state.status != ComputeStatus::Empty && state.status != ComputeStatus::Running {
-                let msg = format!(
-                    "invalid compute status for configuration request: {:?}",
-                    state.status.clone()
-                );
-                return Err((msg, StatusCode::PRECONDITION_FAILED));
-            }
-            state.pspec = Some(parsed_spec);
-            state.set_status(ComputeStatus::ConfigurationPending, &compute.state_changed);
-            drop(state);
-            info!("set new spec and notified waiters");
-        }
-
-        // Spawn a blocking thread to wait for compute to become Running.
-        // This is needed to do not block the main pool of workers and
-        // be able to serve other requests while some particular request
-        // is waiting for compute to finish configuration.
-        let c = compute.clone();
-        task::spawn_blocking(move || {
-            let mut state = c.state.lock().unwrap();
-            while state.status != ComputeStatus::Running {
-                state = c.state_changed.wait(state).unwrap();
-                info!(
-                    "waiting for compute to become Running, current status: {:?}",
-                    state.status
-                );
-
-                if state.status == ComputeStatus::Failed {
-                    let err = state.error.as_ref().map_or("unknown error", |x| x);
-                    let msg = format!("compute configuration failed: {:?}", err);
-                    return Err((msg, StatusCode::INTERNAL_SERVER_ERROR));
-                }
-            }
-
-            Ok(())
-        })
-        .await
-        .unwrap()?;
-
-        // Return current compute state if everything went well.
-        let state = compute.state.lock().unwrap().clone();
-        let status_response = status_response_from_state(&state);
-        Ok(serde_json::to_string(&status_response).unwrap())
-    } else {
-        Err(("invalid spec".to_string(), StatusCode::BAD_REQUEST))
-    }
-}
-
-fn render_json_error(e: &str, status: StatusCode) -> Response<Body> {
-    let error = GenericAPIError {
-        error: e.to_string(),
-    };
-    Response::builder()
-        .status(status)
-        .header(CONTENT_TYPE, "application/json")
-        .body(Body::from(serde_json::to_string(&error).unwrap()))
-        .unwrap()
-}
-
-fn render_json(body: Body) -> Response<Body> {
-    Response::builder()
-        .header(CONTENT_TYPE, "application/json")
-        .body(body)
-        .unwrap()
-}
-
-fn render_plain(body: Body) -> Response<Body> {
-    Response::builder()
-        .header(CONTENT_TYPE, "text/plain")
-        .body(body)
-        .unwrap()
-}
-
-async fn handle_terminate_request(compute: &Arc<ComputeNode>) -> Result<(), (String, StatusCode)> {
-    {
-        let mut state = compute.state.lock().unwrap();
-        if state.status == ComputeStatus::Terminated {
-            return Ok(());
-        }
-        if state.status != ComputeStatus::Empty && state.status != ComputeStatus::Running {
-            let msg = format!(
-                "invalid compute status for termination request: {}",
-                state.status
-            );
-            return Err((msg, StatusCode::PRECONDITION_FAILED));
-        }
-        state.set_status(ComputeStatus::TerminationPending, &compute.state_changed);
-        drop(state);
-    }
-
-    forward_termination_signal();
-    info!("sent signal and notified waiters");
-
-    // Spawn a blocking thread to wait for compute to become Terminated.
-    // This is needed to do not block the main pool of workers and
-    // be able to serve other requests while some particular request
-    // is waiting for compute to finish configuration.
-    let c = compute.clone();
-    task::spawn_blocking(move || {
-        let mut state = c.state.lock().unwrap();
-        while state.status != ComputeStatus::Terminated {
-            state = c.state_changed.wait(state).unwrap();
-            info!(
-                "waiting for compute to become {}, current status: {:?}",
-                ComputeStatus::Terminated,
-                state.status
-            );
-        }
-
-        Ok(())
-    })
-    .await
-    .unwrap()?;
-    info!("terminated Postgres");
-    Ok(())
-}
-
-// Main Hyper HTTP server function that runs it and blocks waiting on it forever.
-#[tokio::main]
-async fn serve(port: u16, state: Arc<ComputeNode>) {
-    // this usually binds to both IPv4 and IPv6 on linux
-    // see e.g. https://github.com/rust-lang/rust/pull/34440
-    let addr = SocketAddr::new(IpAddr::from(Ipv6Addr::UNSPECIFIED), port);
-
-    let make_service = make_service_fn(move |_conn| {
-        let state = state.clone();
-        async move {
-            Ok::<_, Infallible>(service_fn(move |req: Request<Body>| {
-                let state = state.clone();
-                async move {
-                    Ok::<_, Infallible>(
-                        // NOTE: We include the URI path in the string. It
-                        // doesn't contain any variable parts or sensitive
-                        // information in this API.
-                        tracing_utils::http::tracing_handler(
-                            req,
-                            |req| routes(req, &state),
-                            OtelName::UriPath,
-                        )
-                        .await,
-                    )
-                }
-            }))
-        }
-    });
-
-    info!("starting HTTP server on {}", addr);
-
-    let server = Server::bind(&addr).serve(make_service);
-
-    // Run this server forever
-    if let Err(e) = server.await {
-        error!("server error: {}", e);
-    }
-}
-
-/// Launch a separate Hyper HTTP API server thread and return its `JoinHandle`.
-pub fn launch_http_server(port: u16, state: &Arc<ComputeNode>) -> Result<thread::JoinHandle<()>> {
-    let state = Arc::clone(state);
-
-    Ok(thread::Builder::new()
-        .name("http-endpoint".into())
-        .spawn(move || serve(port, state))?)
-}

compute_tools/src/http/extract/json.rs

@@ -0,0 +1,48 @@
+use std::ops::{Deref, DerefMut};
+
+use axum::{
+    async_trait,
+    extract::{rejection::JsonRejection, FromRequest, Request},
+};
+use compute_api::responses::GenericAPIError;
+use http::StatusCode;
+
+/// Custom `Json` extractor, so that we can format errors into
+/// `JsonResponse<GenericAPIError>`.
+#[derive(Debug, Clone, Copy, Default)]
+pub(crate) struct Json<T>(pub T);
+
+#[async_trait]
+impl<S, T> FromRequest<S> for Json<T>
+where
+    axum::Json<T>: FromRequest<S, Rejection = JsonRejection>,
+    S: Send + Sync,
+{
+    type Rejection = (StatusCode, axum::Json<GenericAPIError>);
+
+    async fn from_request(req: Request, state: &S) -> Result<Self, Self::Rejection> {
+        match axum::Json::<T>::from_request(req, state).await {
+            Ok(value) => Ok(Self(value.0)),
+            Err(rejection) => Err((
+                rejection.status(),
+                axum::Json(GenericAPIError {
+                    error: rejection.body_text().to_lowercase(),
+                }),
+            )),
+        }
+    }
+}
+
+impl<T> Deref for Json<T> {
+    type Target = T;
+
+    fn deref(&self) -> &Self::Target {
+        &self.0
+    }
+}
+
+impl<T> DerefMut for Json<T> {
+    fn deref_mut(&mut self) -> &mut Self::Target {
+        &mut self.0
+    }
+}

compute_tools/src/http/extract/mod.rs

@@ -0,0 +1,7 @@
+pub(crate) mod json;
+pub(crate) mod path;
+pub(crate) mod query;
+
+pub(crate) use json::Json;
+pub(crate) use path::Path;
+pub(crate) use query::Query;

compute_tools/src/http/extract/path.rs

@@ -0,0 +1,48 @@
+use std::ops::{Deref, DerefMut};
+
+use axum::{
+    async_trait,
+    extract::{rejection::PathRejection, FromRequestParts},
+};
+use compute_api::responses::GenericAPIError;
+use http::{request::Parts, StatusCode};
+
+/// Custom `Path` extractor, so that we can format errors into
+/// `JsonResponse<GenericAPIError>`.
+#[derive(Debug, Clone, Copy, Default)]
+pub(crate) struct Path<T>(pub T);
+
+#[async_trait]
+impl<S, T> FromRequestParts<S> for Path<T>
+where
+    axum::extract::Path<T>: FromRequestParts<S, Rejection = PathRejection>,
+    S: Send + Sync,
+{
+    type Rejection = (StatusCode, axum::Json<GenericAPIError>);
+
+    async fn from_request_parts(parts: &mut Parts, state: &S) -> Result<Self, Self::Rejection> {
+        match axum::extract::Path::<T>::from_request_parts(parts, state).await {
+            Ok(value) => Ok(Self(value.0)),
+            Err(rejection) => Err((
+                rejection.status(),
+                axum::Json(GenericAPIError {
+                    error: rejection.body_text().to_ascii_lowercase(),
+                }),
+            )),
+        }
+    }
+}
+
+impl<T> Deref for Path<T> {
+    type Target = T;
+
+    fn deref(&self) -> &Self::Target {
+        &self.0
+    }
+}
+
+impl<T> DerefMut for Path<T> {
+    fn deref_mut(&mut self) -> &mut Self::Target {
+        &mut self.0
+    }
+}

compute_tools/src/http/extract/query.rs

@@ -0,0 +1,48 @@
+use std::ops::{Deref, DerefMut};
+
+use axum::{
+    async_trait,
+    extract::{rejection::QueryRejection, FromRequestParts},
+};
+use compute_api::responses::GenericAPIError;
+use http::{request::Parts, StatusCode};
+
+/// Custom `Query` extractor, so that we can format errors into
+/// `JsonResponse<GenericAPIError>`.
+#[derive(Debug, Clone, Copy, Default)]
+pub(crate) struct Query<T>(pub T);
+
+#[async_trait]
+impl<S, T> FromRequestParts<S> for Query<T>
+where
+    axum::extract::Query<T>: FromRequestParts<S, Rejection = QueryRejection>,
+    S: Send + Sync,
+{
+    type Rejection = (StatusCode, axum::Json<GenericAPIError>);
+
+    async fn from_request_parts(parts: &mut Parts, state: &S) -> Result<Self, Self::Rejection> {
+        match axum::extract::Query::<T>::from_request_parts(parts, state).await {
+            Ok(value) => Ok(Self(value.0)),
+            Err(rejection) => Err((
+                rejection.status(),
+                axum::Json(GenericAPIError {
+                    error: rejection.body_text().to_ascii_lowercase(),
+                }),
+            )),
+        }
+    }
+}
+
+impl<T> Deref for Query<T> {
+    type Target = T;
+
+    fn deref(&self) -> &Self::Target {
+        &self.0
+    }
+}
+
+impl<T> DerefMut for Query<T> {
+    fn deref_mut(&mut self) -> &mut Self::Target {
+        &mut self.0
+    }
+}

compute_tools/src/http/mod.rs

@@ -1 +1,56 @@
-pub mod api;
+use axum::{body::Body, response::Response};
+use compute_api::responses::{ComputeStatus, GenericAPIError};
+use http::{header::CONTENT_TYPE, StatusCode};
+use serde::Serialize;
+use tracing::error;
+
+pub use server::launch_http_server;
+
+mod extract;
+mod routes;
+mod server;
+
+/// Convenience response builder for JSON responses
+struct JsonResponse;
+
+impl JsonResponse {
+    /// Helper for actually creating a response
+    fn create_response(code: StatusCode, body: impl Serialize) -> Response {
+        Response::builder()
+            .status(code)
+            .header(CONTENT_TYPE.as_str(), "application/json")
+            .body(Body::from(serde_json::to_string(&body).unwrap()))
+            .unwrap()
+    }
+
+    /// Create a successful error response
+    pub(self) fn success(code: StatusCode, body: impl Serialize) -> Response {
+        assert!({
+            let code = code.as_u16();
+
+            (200..300).contains(&code)
+        });
+
+        Self::create_response(code, body)
+    }
+
+    /// Create an error response
+    pub(self) fn error(code: StatusCode, error: impl ToString) -> Response {
+        assert!(code.as_u16() >= 400);
+
+        let message = error.to_string();
+        error!(message);
+
+        Self::create_response(code, &GenericAPIError { error: message })
+    }
+
+    /// Create an error response related to the compute being in an invalid state
+    pub(self) fn invalid_status(status: ComputeStatus) -> Response {
+        Self::create_response(
+            StatusCode::PRECONDITION_FAILED,
+            &GenericAPIError {
+                error: format!("invalid compute status: {status}"),
+            },
+        )
+    }
+}

compute_tools/src/http/openapi_spec.yaml

@@ -37,7 +37,7 @@ paths:
               schema:
                 $ref: "#/components/schemas/ComputeMetrics"
 
-  /metrics
+  /metrics:
     get:
       tags:
       - Info

compute_tools/src/http/routes/check_writability.rs

@@ -0,0 +1,20 @@
+use std::sync::Arc;
+
+use axum::{extract::State, response::Response};
+use compute_api::responses::ComputeStatus;
+use http::StatusCode;
+
+use crate::{checker::check_writability, compute::ComputeNode, http::JsonResponse};
+
+/// Check that the compute is currently running.
+pub(in crate::http) async fn is_writable(State(compute): State<Arc<ComputeNode>>) -> Response {
+    let status = compute.get_status();
+    if status != ComputeStatus::Running {
+        return JsonResponse::invalid_status(status);
+    }
+
+    match check_writability(&compute).await {
+        Ok(_) => JsonResponse::success(StatusCode::OK, true),
+        Err(e) => JsonResponse::error(StatusCode::INTERNAL_SERVER_ERROR, e),
+    }
+}

compute_tools/src/http/routes/configure.rs

@@ -0,0 +1,91 @@
+use std::sync::Arc;
+
+use axum::{extract::State, response::Response};
+use compute_api::{
+    requests::ConfigurationRequest,
+    responses::{ComputeStatus, ComputeStatusResponse},
+};
+use http::StatusCode;
+use tokio::task;
+use tracing::info;
+
+use crate::{
+    compute::{ComputeNode, ParsedSpec},
+    http::{extract::Json, JsonResponse},
+};
+
+// Accept spec in JSON format and request compute configuration. If anything
+// goes wrong after we set the compute status to `ConfigurationPending` and
+// update compute state with new spec, we basically leave compute in the
+// potentially wrong state. That said, it's control-plane's responsibility to
+// watch compute state after reconfiguration request and to clean restart in
+// case of errors.
+pub(in crate::http) async fn configure(
+    State(compute): State<Arc<ComputeNode>>,
+    request: Json<ConfigurationRequest>,
+) -> Response {
+    if !compute.live_config_allowed {
+        return JsonResponse::error(
+            StatusCode::PRECONDITION_FAILED,
+            "live configuration is not allowed for this compute node".to_string(),
+        );
+    }
+
+    let pspec = match ParsedSpec::try_from(request.spec.clone()) {
+        Ok(p) => p,
+        Err(e) => return JsonResponse::error(StatusCode::BAD_REQUEST, e),
+    };
+
+    // XXX: wrap state update under lock in a code block. Otherwise, we will try
+    // to `Send` `mut state` into the spawned thread bellow, which will cause
+    // the following rustc error:
+    //
+    // error: future cannot be sent between threads safely
+    {
+        let mut state = compute.state.lock().unwrap();
+        if !matches!(state.status, ComputeStatus::Empty | ComputeStatus::Running) {
+            return JsonResponse::invalid_status(state.status);
+        }
+
+        state.pspec = Some(pspec);
+        state.set_status(ComputeStatus::ConfigurationPending, &compute.state_changed);
+        drop(state);
+    }
+
+    // Spawn a blocking thread to wait for compute to become Running. This is
+    // needed to do not block the main pool of workers and be able to serve
+    // other requests while some particular request is waiting for compute to
+    // finish configuration.
+    let c = compute.clone();
+    let completed = task::spawn_blocking(move || {
+        let mut state = c.state.lock().unwrap();
+        while state.status != ComputeStatus::Running {
+            state = c.state_changed.wait(state).unwrap();
+            info!(
+                "waiting for compute to become {}, current status: {}",
+                ComputeStatus::Running,
+                state.status
+            );
+
+            if state.status == ComputeStatus::Failed {
+                let err = state.error.as_ref().map_or("unknown error", |x| x);
+                let msg = format!("compute configuration failed: {:?}", err);
+                return Err(msg);
+            }
+        }
+
+        Ok(())
+    })
+    .await
+    .unwrap();
+
+    if let Err(e) = completed {
+        return JsonResponse::error(StatusCode::INTERNAL_SERVER_ERROR, e);
+    }
+
+    // Return current compute state if everything went well.
+    let state = compute.state.lock().unwrap().clone();
+    let body = ComputeStatusResponse::from(&state);
+
+    JsonResponse::success(StatusCode::OK, body)
+}

compute_tools/src/http/routes/database_schema.rs

@@ -0,0 +1,34 @@
+use std::sync::Arc;
+
+use axum::{body::Body, extract::State, response::Response};
+use http::{header::CONTENT_TYPE, StatusCode};
+use serde::Deserialize;
+
+use crate::{
+    catalog::{get_database_schema, SchemaDumpError},
+    compute::ComputeNode,
+    http::{extract::Query, JsonResponse},
+};
+
+#[derive(Debug, Clone, Deserialize)]
+pub(in crate::http) struct DatabaseSchemaParams {
+    database: String,
+}
+
+/// Get a schema dump of the requested database.
+pub(in crate::http) async fn get_schema_dump(
+    params: Query<DatabaseSchemaParams>,
+    State(compute): State<Arc<ComputeNode>>,
+) -> Response {
+    match get_database_schema(&compute, &params.database).await {
+        Ok(schema) => Response::builder()
+            .status(StatusCode::OK)
+            .header(CONTENT_TYPE.as_str(), "application/json")
+            .body(Body::from_stream(schema))
+            .unwrap(),
+        Err(SchemaDumpError::DatabaseDoesNotExist) => {
+            JsonResponse::error(StatusCode::NOT_FOUND, SchemaDumpError::DatabaseDoesNotExist)
+        }
+        Err(e) => JsonResponse::error(StatusCode::INTERNAL_SERVER_ERROR, e),
+    }
+}

compute_tools/src/http/routes/dbs_and_roles.rs

@@ -0,0 +1,16 @@
+use std::sync::Arc;
+
+use axum::{extract::State, response::Response};
+use http::StatusCode;
+
+use crate::{catalog::get_dbs_and_roles, compute::ComputeNode, http::JsonResponse};
+
+/// Get the databases and roles from the compute.
+pub(in crate::http) async fn get_catalog_objects(
+    State(compute): State<Arc<ComputeNode>>,
+) -> Response {
+    match get_dbs_and_roles(&compute).await {
+        Ok(catalog_objects) => JsonResponse::success(StatusCode::OK, catalog_objects),
+        Err(e) => JsonResponse::error(StatusCode::INTERNAL_SERVER_ERROR, e),
+    }
+}

compute_tools/src/http/routes/extension_server.rs

@@ -0,0 +1,67 @@
+use std::sync::Arc;
+
+use axum::{
+    extract::State,
+    response::{IntoResponse, Response},
+};
+use http::StatusCode;
+use serde::Deserialize;
+
+use crate::{
+    compute::ComputeNode,
+    http::{
+        extract::{Path, Query},
+        JsonResponse,
+    },
+};
+
+#[derive(Debug, Clone, Deserialize)]
+pub(in crate::http) struct ExtensionServerParams {
+    is_library: Option<bool>,
+}
+
+/// Download a remote extension.
+pub(in crate::http) async fn download_extension(
+    Path(filename): Path<String>,
+    params: Query<ExtensionServerParams>,
+    State(compute): State<Arc<ComputeNode>>,
+) -> Response {
+    // Don't even try to download extensions if no remote storage is configured
+    if compute.ext_remote_storage.is_none() {
+        return JsonResponse::error(
+            StatusCode::PRECONDITION_FAILED,
+            "remote storage is not configured",
+        );
+    }
+
+    let ext = {
+        let state = compute.state.lock().unwrap();
+        let pspec = state.pspec.as_ref().unwrap();
+        let spec = &pspec.spec;
+
+        let remote_extensions = match spec.remote_extensions.as_ref() {
+            Some(r) => r,
+            None => {
+                return JsonResponse::error(
+                    StatusCode::CONFLICT,
+                    "information about remote extensions is unavailable",
+                );
+            }
+        };
+
+        remote_extensions.get_ext(
+            &filename,
+            params.is_library.unwrap_or(false),
+            &compute.build_tag,
+            &compute.pgversion,
+        )
+    };
+
+    match ext {
+        Ok((ext_name, ext_path)) => match compute.download_extension(ext_name, ext_path).await {
+            Ok(_) => StatusCode::OK.into_response(),
+            Err(e) => JsonResponse::error(StatusCode::INTERNAL_SERVER_ERROR, e),
+        },
+        Err(e) => JsonResponse::error(StatusCode::NOT_FOUND, e),
+    }
+}

compute_tools/src/http/routes/extensions.rs

@@ -0,0 +1,45 @@
+use std::sync::Arc;
+
+use axum::{extract::State, response::Response};
+use compute_api::{
+    requests::ExtensionInstallRequest,
+    responses::{ComputeStatus, ExtensionInstallResponse},
+};
+use http::StatusCode;
+
+use crate::{
+    compute::ComputeNode,
+    http::{extract::Json, JsonResponse},
+};
+
+/// Install a extension.
+pub(in crate::http) async fn install_extension(
+    State(compute): State<Arc<ComputeNode>>,
+    request: Json<ExtensionInstallRequest>,
+) -> Response {
+    let status = compute.get_status();
+    if status != ComputeStatus::Running {
+        return JsonResponse::invalid_status(status);
+    }
+
+    match compute
+        .install_extension(
+            &request.extension,
+            &request.database,
+            request.version.to_string(),
+        )
+        .await
+    {
+        Ok(version) => JsonResponse::success(
+            StatusCode::CREATED,
+            Some(ExtensionInstallResponse {
+                extension: request.extension.clone(),
+                version,
+            }),
+        ),
+        Err(e) => JsonResponse::error(
+            StatusCode::INTERNAL_SERVER_ERROR,
+            format!("failed to install extension: {e}"),
+        ),
+    }
+}

compute_tools/src/http/routes/failpoints.rs

@@ -0,0 +1,35 @@
+use axum::response::{IntoResponse, Response};
+use http::StatusCode;
+use tracing::info;
+use utils::failpoint_support::{apply_failpoint, ConfigureFailpointsRequest};
+
+use crate::http::{extract::Json, JsonResponse};
+
+/// Configure failpoints for testing purposes.
+pub(in crate::http) async fn configure_failpoints(
+    failpoints: Json<ConfigureFailpointsRequest>,
+) -> Response {
+    if !fail::has_failpoints() {
+        return JsonResponse::error(
+            StatusCode::PRECONDITION_FAILED,
+            "Cannot manage failpoints because neon was compiled without failpoints support",
+        );
+    }
+
+    for fp in &*failpoints {
+        info!("cfg failpoint: {} {}", fp.name, fp.actions);
+
+        // We recognize one extra "action" that's not natively recognized
+        // by the failpoints crate: exit, to immediately kill the process
+        let cfg_result = apply_failpoint(&fp.name, &fp.actions);
+
+        if let Err(e) = cfg_result {
+            return JsonResponse::error(
+                StatusCode::BAD_REQUEST,
+                format!("failed to configure failpoints: {e}"),
+            );
+        }
+    }
+
+    StatusCode::OK.into_response()
+}

compute_tools/src/http/routes/grants.rs

@@ -0,0 +1,48 @@
+use std::sync::Arc;
+
+use axum::{extract::State, response::Response};
+use compute_api::{
+    requests::SetRoleGrantsRequest,
+    responses::{ComputeStatus, SetRoleGrantsResponse},
+};
+use http::StatusCode;
+
+use crate::{
+    compute::ComputeNode,
+    http::{extract::Json, JsonResponse},
+};
+
+/// Add grants for a role.
+pub(in crate::http) async fn add_grant(
+    State(compute): State<Arc<ComputeNode>>,
+    request: Json<SetRoleGrantsRequest>,
+) -> Response {
+    let status = compute.get_status();
+    if status != ComputeStatus::Running {
+        return JsonResponse::invalid_status(status);
+    }
+
+    match compute
+        .set_role_grants(
+            &request.database,
+            &request.schema,
+            &request.privileges,
+            &request.role,
+        )
+        .await
+    {
+        Ok(()) => JsonResponse::success(
+            StatusCode::CREATED,
+            Some(SetRoleGrantsResponse {
+                database: request.database.clone(),
+                schema: request.schema.clone(),
+                role: request.role.clone(),
+                privileges: request.privileges.clone(),
+            }),
+        ),
+        Err(e) => JsonResponse::error(
+            StatusCode::INTERNAL_SERVER_ERROR,
+            format!("failed to grant role privileges to the schema: {e}"),
+        ),
+    }
+}

compute_tools/src/http/routes/info.rs

@@ -0,0 +1,11 @@
+use axum::response::Response;
+use compute_api::responses::InfoResponse;
+use http::StatusCode;
+
+use crate::http::JsonResponse;
+
+/// Get information about the physical characteristics about the compute.
+pub(in crate::http) async fn get_info() -> Response {
+    let num_cpus = num_cpus::get_physical();
+    JsonResponse::success(StatusCode::OK, &InfoResponse { num_cpus })
+}

compute_tools/src/http/routes/insights.rs

@@ -0,0 +1,18 @@
+use std::sync::Arc;
+
+use axum::{extract::State, response::Response};
+use compute_api::responses::ComputeStatus;
+use http::StatusCode;
+
+use crate::{compute::ComputeNode, http::JsonResponse};
+
+/// Collect current Postgres usage insights.
+pub(in crate::http) async fn get_insights(State(compute): State<Arc<ComputeNode>>) -> Response {
+    let status = compute.get_status();
+    if status != ComputeStatus::Running {
+        return JsonResponse::invalid_status(status);
+    }
+
+    let insights = compute.collect_insights().await;
+    JsonResponse::success(StatusCode::OK, insights)
+}

compute_tools/src/http/routes/installed_extensions.rs

@@ -0,0 +1,33 @@
+use std::sync::Arc;
+
+use axum::{extract::State, response::Response};
+use compute_api::responses::ComputeStatus;
+use http::StatusCode;
+use tokio::task;
+
+use crate::{compute::ComputeNode, http::JsonResponse, installed_extensions};
+
+/// Get a list of installed extensions.
+pub(in crate::http) async fn get_installed_extensions(
+    State(compute): State<Arc<ComputeNode>>,
+) -> Response {
+    let status = compute.get_status();
+    if status != ComputeStatus::Running {
+        return JsonResponse::invalid_status(status);
+    }
+
+    let conf = compute.get_conn_conf(None);
+    let res = task::spawn_blocking(move || installed_extensions::get_installed_extensions(conf))
+        .await
+        .unwrap();
+
+    match res {
+        Ok(installed_extensions) => {
+            JsonResponse::success(StatusCode::OK, Some(installed_extensions))
+        }
+        Err(e) => JsonResponse::error(
+            StatusCode::INTERNAL_SERVER_ERROR,
+            format!("failed to get list of installed extensions: {e}"),
+        ),
+    }
+}

compute_tools/src/http/routes/metrics.rs

@@ -0,0 +1,32 @@
+use axum::{body::Body, response::Response};
+use http::header::CONTENT_TYPE;
+use http::StatusCode;
+use metrics::proto::MetricFamily;
+use metrics::Encoder;
+use metrics::TextEncoder;
+
+use crate::{http::JsonResponse, installed_extensions};
+
+/// Expose Prometheus metrics.
+pub(in crate::http) async fn get_metrics() -> Response {
+    // When we call TextEncoder::encode() below, it will immediately return an
+    // error if a metric family has no metrics, so we need to preemptively
+    // filter out metric families with no metrics.
+    let metrics = installed_extensions::collect()
+        .into_iter()
+        .filter(|m| !m.get_metric().is_empty())
+        .collect::<Vec<MetricFamily>>();
+
+    let encoder = TextEncoder::new();
+    let mut buffer = vec![];
+
+    if let Err(e) = encoder.encode(&metrics, &mut buffer) {
+        return JsonResponse::error(StatusCode::INTERNAL_SERVER_ERROR, e);
+    }
+
+    Response::builder()
+        .status(StatusCode::OK)
+        .header(CONTENT_TYPE, encoder.format_type())
+        .body(Body::from(buffer))
+        .unwrap()
+}

compute_tools/src/http/routes/metrics_json.rs

@@ -0,0 +1,12 @@
+use std::sync::Arc;
+
+use axum::{extract::State, response::Response};
+use http::StatusCode;
+
+use crate::{compute::ComputeNode, http::JsonResponse};
+
+/// Get startup metrics.
+pub(in crate::http) async fn get_metrics(State(compute): State<Arc<ComputeNode>>) -> Response {
+    let metrics = compute.state.lock().unwrap().metrics.clone();
+    JsonResponse::success(StatusCode::OK, metrics)
+}

compute_tools/src/http/routes/mod.rs

@@ -0,0 +1,38 @@
+use compute_api::responses::ComputeStatusResponse;
+
+use crate::compute::ComputeState;
+
+pub(in crate::http) mod check_writability;
+pub(in crate::http) mod configure;
+pub(in crate::http) mod database_schema;
+pub(in crate::http) mod dbs_and_roles;
+pub(in crate::http) mod extension_server;
+pub(in crate::http) mod extensions;
+pub(in crate::http) mod failpoints;
+pub(in crate::http) mod grants;
+pub(in crate::http) mod info;
+pub(in crate::http) mod insights;
+pub(in crate::http) mod installed_extensions;
+pub(in crate::http) mod metrics;
+pub(in crate::http) mod metrics_json;
+pub(in crate::http) mod status;
+pub(in crate::http) mod terminate;
+
+impl From<&ComputeState> for ComputeStatusResponse {
+    fn from(state: &ComputeState) -> Self {
+        ComputeStatusResponse {
+            start_time: state.start_time,
+            tenant: state
+                .pspec
+                .as_ref()
+                .map(|pspec| pspec.tenant_id.to_string()),
+            timeline: state
+                .pspec
+                .as_ref()
+                .map(|pspec| pspec.timeline_id.to_string()),
+            status: state.status,
+            last_active: state.last_active,
+            error: state.error.clone(),
+        }
+    }
+}

compute_tools/src/http/routes/status.rs

@@ -0,0 +1,14 @@
+use std::{ops::Deref, sync::Arc};
+
+use axum::{extract::State, http::StatusCode, response::Response};
+use compute_api::responses::ComputeStatusResponse;
+
+use crate::{compute::ComputeNode, http::JsonResponse};
+
+/// Retrieve the state of the comute.
+pub(in crate::http) async fn get_status(State(compute): State<Arc<ComputeNode>>) -> Response {
+    let state = compute.state.lock().unwrap();
+    let body = ComputeStatusResponse::from(state.deref());
+
+    JsonResponse::success(StatusCode::OK, body)
+}

compute_tools/src/http/routes/terminate.rs

@@ -0,0 +1,58 @@
+use std::sync::Arc;
+
+use axum::{
+    extract::State,
+    response::{IntoResponse, Response},
+};
+use compute_api::responses::ComputeStatus;
+use http::StatusCode;
+use tokio::task;
+use tracing::info;
+
+use crate::{
+    compute::{forward_termination_signal, ComputeNode},
+    http::JsonResponse,
+};
+
+/// Terminate the compute.
+pub(in crate::http) async fn terminate(State(compute): State<Arc<ComputeNode>>) -> Response {
+    {
+        let mut state = compute.state.lock().unwrap();
+        if state.status == ComputeStatus::Terminated {
+            return StatusCode::CREATED.into_response();
+        }
+
+        if !matches!(state.status, ComputeStatus::Empty | ComputeStatus::Running) {
+            return JsonResponse::invalid_status(state.status);
+        }
+
+        state.set_status(ComputeStatus::TerminationPending, &compute.state_changed);
+        drop(state);
+    }
+
+    forward_termination_signal();
+    info!("sent signal and notified waiters");
+
+    // Spawn a blocking thread to wait for compute to become Terminated.
+    // This is needed to do not block the main pool of workers and
+    // be able to serve other requests while some particular request
+    // is waiting for compute to finish configuration.
+    let c = compute.clone();
+    task::spawn_blocking(move || {
+        let mut state = c.state.lock().unwrap();
+        while state.status != ComputeStatus::Terminated {
+            state = c.state_changed.wait(state).unwrap();
+            info!(
+                "waiting for compute to become {}, current status: {:?}",
+                ComputeStatus::Terminated,
+                state.status
+            );
+        }
+    })
+    .await
+    .unwrap();
+
+    info!("terminated Postgres");
+
+    StatusCode::OK.into_response()
+}

compute_tools/src/http/server.rs

@@ -0,0 +1,165 @@
+use std::{
+    net::{IpAddr, Ipv6Addr, SocketAddr},
+    sync::{
+        atomic::{AtomicU64, Ordering},
+        Arc,
+    },
+    thread,
+    time::Duration,
+};
+
+use anyhow::Result;
+use axum::{
+    response::{IntoResponse, Response},
+    routing::{get, post},
+    Router,
+};
+use http::StatusCode;
+use tokio::net::TcpListener;
+use tower::ServiceBuilder;
+use tower_http::{
+    request_id::{MakeRequestId, PropagateRequestIdLayer, RequestId, SetRequestIdLayer},
+    trace::TraceLayer,
+};
+use tracing::{debug, error, info, Span};
+
+use super::routes::{
+    check_writability, configure, database_schema, dbs_and_roles, extension_server, extensions,
+    grants, info as info_route, insights, installed_extensions, metrics, metrics_json, status,
+    terminate,
+};
+use crate::compute::ComputeNode;
+
+async fn handle_404() -> Response {
+    StatusCode::NOT_FOUND.into_response()
+}
+
+#[derive(Clone, Default)]
+struct ComputeMakeRequestId(Arc<AtomicU64>);
+
+impl MakeRequestId for ComputeMakeRequestId {
+    fn make_request_id<B>(
+        &mut self,
+        _request: &http::Request<B>,
+    ) -> Option<tower_http::request_id::RequestId> {
+        let request_id = self
+            .0
+            .fetch_add(1, Ordering::SeqCst)
+            .to_string()
+            .parse()
+            .unwrap();
+
+        Some(RequestId::new(request_id))
+    }
+}
+
+/// Run the HTTP server and wait on it forever.
+#[tokio::main]
+async fn serve(port: u16, compute: Arc<ComputeNode>) {
+    const X_REQUEST_ID: &str = "x-request-id";
+
+    let mut app = Router::new()
+        .route("/check_writability", post(check_writability::is_writable))
+        .route("/configure", post(configure::configure))
+        .route("/database_schema", get(database_schema::get_schema_dump))
+        .route("/dbs_and_roles", get(dbs_and_roles::get_catalog_objects))
+        .route(
+            "/extension_server/*filename",
+            post(extension_server::download_extension),
+        )
+        .route("/extensions", post(extensions::install_extension))
+        .route("/grants", post(grants::add_grant))
+        .route("/info", get(info_route::get_info))
+        .route("/insights", get(insights::get_insights))
+        .route(
+            "/installed_extensions",
+            get(installed_extensions::get_installed_extensions),
+        )
+        .route("/metrics", get(metrics::get_metrics))
+        .route("/metrics.json", get(metrics_json::get_metrics))
+        .route("/status", get(status::get_status))
+        .route("/terminate", post(terminate::terminate))
+        .fallback(handle_404)
+        .layer(
+            ServiceBuilder::new()
+                .layer(SetRequestIdLayer::x_request_id(
+                    ComputeMakeRequestId::default(),
+                ))
+                .layer(
+                    TraceLayer::new_for_http()
+                        .on_request(|request: &http::Request<_>, _span: &Span| {
+                            let request_id = request
+                                .headers()
+                                .get(X_REQUEST_ID)
+                                .unwrap()
+                                .to_str()
+                                .unwrap();
+
+                            match request.uri().path() {
+                                "/metrics" => {
+                                    debug!(%request_id, "{} {}", request.method(), request.uri())
+                                }
+                                _ => info!(%request_id, "{} {}", request.method(), request.uri()),
+                            };
+                        })
+                        .on_response(
+                            |response: &http::Response<_>, latency: Duration, _span: &Span| {
+                                let request_id = response
+                                    .headers()
+                                    .get(X_REQUEST_ID)
+                                    .unwrap()
+                                    .to_str()
+                                    .unwrap();
+
+                                info!(
+                                    %request_id,
+                                    code = response.status().as_u16(),
+                                    latency = latency.as_millis()
+                                )
+                            },
+                        ),
+                )
+                .layer(PropagateRequestIdLayer::x_request_id()),
+        )
+        .with_state(compute);
+
+    // Add in any testing support
+    if cfg!(feature = "testing") {
+        use super::routes::failpoints;
+
+        app = app.route("/failpoints", post(failpoints::configure_failpoints))
+    }
+
+    // This usually binds to both IPv4 and IPv6 on Linux, see
+    // https://github.com/rust-lang/rust/pull/34440 for more information
+    let addr = SocketAddr::new(IpAddr::from(Ipv6Addr::UNSPECIFIED), port);
+    let listener = match TcpListener::bind(&addr).await {
+        Ok(listener) => listener,
+        Err(e) => {
+            error!(
+                "failed to bind the compute_ctl HTTP server to port {}: {}",
+                port, e
+            );
+            return;
+        }
+    };
+
+    if let Ok(local_addr) = listener.local_addr() {
+        info!("compute_ctl HTTP server listening on {}", local_addr);
+    } else {
+        info!("compute_ctl HTTP server listening on port {}", port);
+    }
+
+    if let Err(e) = axum::serve(listener, app).await {
+        error!("compute_ctl HTTP server error: {}", e);
+    }
+}
+
+/// Launch a separate HTTP server thread and return its `JoinHandle`.
+pub fn launch_http_server(port: u16, state: &Arc<ComputeNode>) -> Result<thread::JoinHandle<()>> {
+    let state = Arc::clone(state);
+
+    Ok(thread::Builder::new()
+        .name("http-server".into())
+        .spawn(move || serve(port, state))?)
+}

compute_tools/src/lib.rs

@@ -3,8 +3,6 @@
 #![deny(unsafe_code)]
 #![deny(clippy::undocumented_unsafe_blocks)]
 
-extern crate hyper0 as hyper;
-
 pub mod checker;
 pub mod config;
 pub mod configurator;

compute_tools/src/migration.rs

@@ -1,13 +1,16 @@
 use anyhow::{Context, Result};
-use postgres::Client;
+use fail::fail_point;
+use postgres::{Client, Transaction};
 use tracing::info;
 
+/// Runs a series of migrations on a target database
 pub(crate) struct MigrationRunner<'m> {
     client: &'m mut Client,
     migrations: &'m [&'m str],
 }
 
 impl<'m> MigrationRunner<'m> {
+    /// Create a new migration runner
     pub fn new(client: &'m mut Client, migrations: &'m [&'m str]) -> Self {
         // The neon_migration.migration_id::id column is a bigint, which is equivalent to an i64
         assert!(migrations.len() + 1 < i64::MAX as usize);
@@ -15,87 +18,110 @@ impl<'m> MigrationRunner<'m> {
         Self { client, migrations }
     }
 
+    /// Get the current value neon_migration.migration_id
     fn get_migration_id(&mut self) -> Result<i64> {
-        let query = "SELECT id FROM neon_migration.migration_id";
         let row = self
             .client
-            .query_one(query, &[])
-            .context("run_migrations get migration_id")?;
+            .query_one("SELECT id FROM neon_migration.migration_id", &[])?;
 
         Ok(row.get::<&str, i64>("id"))
     }
 
-    fn update_migration_id(&mut self, migration_id: i64) -> Result<()> {
-        let setval = format!("UPDATE neon_migration.migration_id SET id={}", migration_id);
+    /// Update the neon_migration.migration_id value
+    ///
+    /// This function has a fail point called compute-migration, which can be
+    /// used if you would like to fail the application of a series of migrations
+    /// at some point.
+    fn update_migration_id(txn: &mut Transaction, migration_id: i64) -> Result<()> {
+        // We use this fail point in order to check that failing in the
+        // middle of applying a series of migrations fails in an expected
+        // manner
+        if cfg!(feature = "testing") {
+            let fail = (|| {
+                fail_point!("compute-migration", |fail_migration_id| {
+                    migration_id == fail_migration_id.unwrap().parse::<i64>().unwrap()
+                });
 
+                false
+            })();
+
+            if fail {
+                return Err(anyhow::anyhow!(format!(
+                    "migration {} was configured to fail because of a failpoint",
+                    migration_id
+                )));
+            }
+        }
+
+        txn.query(
+            "UPDATE neon_migration.migration_id SET id = $1",
+            &[&migration_id],
+        )
+        .with_context(|| format!("update neon_migration.migration_id to {migration_id}"))?;
+
+        Ok(())
+    }
+
+    /// Prepare the migrations the target database for handling migrations
+    fn prepare_database(&mut self) -> Result<()> {
         self.client
-            .simple_query(&setval)
-            .context("run_migrations update id")?;
+            .simple_query("CREATE SCHEMA IF NOT EXISTS neon_migration")?;
+        self.client.simple_query("CREATE TABLE IF NOT EXISTS neon_migration.migration_id (key INT NOT NULL PRIMARY KEY, id bigint NOT NULL DEFAULT 0)")?;
+        self.client.simple_query(
+            "INSERT INTO neon_migration.migration_id VALUES (0, 0) ON CONFLICT DO NOTHING",
+        )?;
+        self.client
+            .simple_query("ALTER SCHEMA neon_migration OWNER TO cloud_admin")?;
+        self.client
+            .simple_query("REVOKE ALL ON SCHEMA neon_migration FROM PUBLIC")?;
 
         Ok(())
     }
 
-    fn prepare_migrations(&mut self) -> Result<()> {
-        let query = "CREATE SCHEMA IF NOT EXISTS neon_migration";
-        self.client.simple_query(query)?;
+    /// Run an individual migration
+    fn run_migration(txn: &mut Transaction, migration_id: i64, migration: &str) -> Result<()> {
+        if migration.starts_with("-- SKIP") {
+            info!("Skipping migration id={}", migration_id);
 
-        let query = "CREATE TABLE IF NOT EXISTS neon_migration.migration_id (key INT NOT NULL PRIMARY KEY, id bigint NOT NULL DEFAULT 0)";
-        self.client.simple_query(query)?;
+            // Even though we are skipping the migration, updating the
+            // migration ID should help keep logic easy to understand when
+            // trying to understand the state of a cluster.
+            Self::update_migration_id(txn, migration_id)?;
+        } else {
+            info!("Running migration id={}:\n{}\n", migration_id, migration);
 
-        let query = "INSERT INTO neon_migration.migration_id VALUES (0, 0) ON CONFLICT DO NOTHING";
-        self.client.simple_query(query)?;
+            txn.simple_query(migration)
+                .with_context(|| format!("apply migration {migration_id}"))?;
 
-        let query = "ALTER SCHEMA neon_migration OWNER TO cloud_admin";
-        self.client.simple_query(query)?;
-
-        let query = "REVOKE ALL ON SCHEMA neon_migration FROM PUBLIC";
-        self.client.simple_query(query)?;
+            Self::update_migration_id(txn, migration_id)?;
+        }
 
         Ok(())
     }
 
+    /// Run the configured set of migrations
     pub fn run_migrations(mut self) -> Result<()> {
-        self.prepare_migrations()?;
+        self.prepare_database()
+            .context("prepare database to handle migrations")?;
 
         let mut current_migration = self.get_migration_id()? as usize;
         while current_migration < self.migrations.len() {
-            macro_rules! migration_id {
-                ($cm:expr) => {
-                    ($cm + 1) as i64
-                };
-            }
+            // The index lags the migration ID by 1, so the current migration
+            // ID is also the next index
+            let migration_id = (current_migration + 1) as i64;
 
-            let migration = self.migrations[current_migration];
+            let mut txn = self
+                .client
+                .transaction()
+                .with_context(|| format!("begin transaction for migration {migration_id}"))?;
 
-            if migration.starts_with("-- SKIP") {
-                info!("Skipping migration id={}", migration_id!(current_migration));
-            } else {
-                info!(
-                    "Running migration id={}:\n{}\n",
-                    migration_id!(current_migration),
-                    migration
-                );
+            Self::run_migration(&mut txn, migration_id, self.migrations[current_migration])
+                .with_context(|| format!("running migration {migration_id}"))?;
 
-                self.client
-                    .simple_query("BEGIN")
-                    .context("begin migration")?;
+            txn.commit()
+                .with_context(|| format!("commit transaction for migration {migration_id}"))?;
 
-                self.client.simple_query(migration).with_context(|| {
-                    format!(
-                        "run_migrations migration id={}",
-                        migration_id!(current_migration)
-                    )
-                })?;
-
-                // Migration IDs start at 1
-                self.update_migration_id(migration_id!(current_migration))?;
-
-                self.client
-                    .simple_query("COMMIT")
-                    .context("commit migration")?;
-
-                info!("Finished migration id={}", migration_id!(current_migration));
-            }
+            info!("Finished migration id={}", migration_id);
 
             current_migration += 1;
         }

compute_tools/src/migrations/tests/0001-neon_superuser_bypass_rls.sql

@@ -0,0 +1,9 @@
+DO $$
+DECLARE
+    bypassrls boolean;
+BEGIN
+    SELECT rolbypassrls INTO bypassrls FROM pg_roles WHERE rolname = 'neon_superuser';
+    IF NOT bypassrls THEN
+        RAISE EXCEPTION 'neon_superuser cannot bypass RLS';
+    END IF;
+END $$;

compute_tools/src/migrations/tests/0002-alter_roles.sql

@@ -0,0 +1,25 @@
+DO $$
+DECLARE
+    role record;
+BEGIN
+    FOR role IN
+        SELECT rolname AS name, rolinherit AS inherit
+        FROM pg_roles
+        WHERE pg_has_role(rolname, 'neon_superuser', 'member')
+    LOOP
+        IF NOT role.inherit THEN
+            RAISE EXCEPTION '% cannot inherit', quote_ident(role.name);
+        END IF;
+    END LOOP;
+
+    FOR role IN
+        SELECT rolname AS name, rolbypassrls AS bypassrls
+        FROM pg_roles
+        WHERE NOT pg_has_role(rolname, 'neon_superuser', 'member')
+            AND NOT starts_with(rolname, 'pg_')
+    LOOP
+        IF role.bypassrls THEN
+            RAISE EXCEPTION  '% can bypass RLS', quote_ident(role.name);
+        END IF;
+    END LOOP;
+END $$;

compute_tools/src/migrations/tests/0003-grant_pg_create_subscription_to_neon_superuser.sql

@@ -0,0 +1,10 @@
+DO $$
+BEGIN
+    IF (SELECT current_setting('server_version_num')::numeric < 160000) THEN
+        RETURN;
+    END IF;
+
+    IF NOT (SELECT pg_has_role('neon_superuser', 'pg_create_subscription', 'member')) THEN
+        RAISE EXCEPTION 'neon_superuser cannot execute pg_create_subscription';
+    END IF;
+END $$;

compute_tools/src/migrations/tests/0004-grant_pg_monitor_to_neon_superuser.sql

@@ -0,0 +1,19 @@
+DO $$
+DECLARE
+    monitor record;
+BEGIN
+    SELECT pg_has_role('neon_superuser', 'pg_monitor', 'member') AS member,
+            admin_option AS admin
+        INTO monitor
+        FROM pg_auth_members
+        WHERE roleid = 'pg_monitor'::regrole
+            AND member = 'pg_monitor'::regrole;
+
+    IF NOT monitor.member THEN
+        RAISE EXCEPTION 'neon_superuser is not a member of pg_monitor';
+    END IF;
+
+    IF NOT monitor.admin THEN
+        RAISE EXCEPTION 'neon_superuser cannot grant pg_monitor';
+    END IF;
+END $$;

compute_tools/src/migrations/tests/0005-grant_all_on_tables_to_neon_superuser.sql

@@ -0,0 +1,2 @@
+-- This test was never written becuase at the time migration tests were added
+-- the accompanying migration was already skipped.

compute_tools/src/migrations/tests/0006-grant_all_on_sequences_to_neon_superuser.sql

@@ -0,0 +1,2 @@
+-- This test was never written becuase at the time migration tests were added
+-- the accompanying migration was already skipped.

compute_tools/src/migrations/tests/0007-grant_all_on_tables_to_neon_superuser_with_grant_option.sql

@@ -0,0 +1,2 @@
+-- This test was never written becuase at the time migration tests were added
+-- the accompanying migration was already skipped.

compute_tools/src/migrations/tests/0008-grant_all_on_sequences_to_neon_superuser_with_grant_option.sql

@@ -0,0 +1,2 @@
+-- This test was never written becuase at the time migration tests were added
+-- the accompanying migration was already skipped.

compute_tools/src/migrations/tests/0009-revoke_replication_for_previously_allowed_roles.sql

@@ -0,0 +1,2 @@
+-- This test was never written becuase at the time migration tests were added
+-- the accompanying migration was already skipped.

compute_tools/src/migrations/tests/0010-grant_snapshot_synchronization_funcs_to_neon_superuser.sql

@@ -0,0 +1,13 @@
+DO $$
+DECLARE
+    can_execute boolean;
+BEGIN
+    SELECT bool_and(has_function_privilege('neon_superuser', oid, 'execute'))
+       INTO can_execute
+       FROM pg_proc
+       WHERE proname IN ('pg_export_snapshot', 'pg_log_standby_snapshot')
+           AND pronamespace = 'pg_catalog'::regnamespace;
+    IF NOT can_execute THEN
+        RAISE EXCEPTION 'neon_superuser cannot execute both pg_export_snapshot and pg_log_standby_snapshot';
+    END IF;
+END $$;

compute_tools/src/migrations/tests/0011-grant_pg_show_replication_origin_status_to_neon_superuser.sql

@@ -0,0 +1,13 @@
+DO $$
+DECLARE
+    can_execute boolean;
+BEGIN
+    SELECT has_function_privilege('neon_superuser', oid, 'execute')
+       INTO can_execute
+       FROM pg_proc
+       WHERE proname = 'pg_show_replication_origin_status'
+           AND pronamespace = 'pg_catalog'::regnamespace;
+    IF NOT can_execute THEN
+        RAISE EXCEPTION 'neon_superuser cannot execute pg_show_replication_origin_status';
+    END IF;
+END $$;

compute_tools/src/spec_apply.rs

@@ -47,6 +47,7 @@ pub enum PerDatabasePhase {
     DeleteDBRoleReferences,
     ChangeSchemaPerms,
     HandleAnonExtension,
+    DropSubscriptionsForDeletedDatabases,
 }
 
 #[derive(Clone, Debug)]
@@ -74,7 +75,7 @@ pub struct MutableApplyContext {
     pub dbs: HashMap<String, Database>,
 }
 
-/// Appply the operations that belong to the given spec apply phase.
+/// Apply the operations that belong to the given spec apply phase.
 ///
 /// Commands within a single phase are executed in order of Iterator yield.
 /// Commands of ApplySpecPhase::RunInEachDatabase will execute in the database
@@ -326,13 +327,12 @@ async fn get_operations<'a>(
 
                             // Use FORCE to drop database even if there are active connections.
                             // We run this from `cloud_admin`, so it should have enough privileges.
+                            //
                             // NB: there could be other db states, which prevent us from dropping
                             // the database. For example, if db is used by any active subscription
                             // or replication slot.
-                            // TODO: deal with it once we allow logical replication. Proper fix should
-                            // involve returning an error code to the control plane, so it could
-                            // figure out that this is a non-retryable error, return it to the user
-                            // and fail operation permanently.
+                            // Such cases are handled in the DropSubscriptionsForDeletedDatabases
+                            // phase. We do all the cleanup before actually dropping the database.
                             let drop_db_query: String = format!(
                                 "DROP DATABASE IF EXISTS {} WITH (FORCE)",
                                 &op.name.pg_quote()
@@ -444,6 +444,30 @@ async fn get_operations<'a>(
         }
         ApplySpecPhase::RunInEachDatabase { db, subphase } => {
             match subphase {
+                PerDatabasePhase::DropSubscriptionsForDeletedDatabases => {
+                    match &db {
+                        DB::UserDB(db) => {
+                            let drop_subscription_query: String = format!(
+                                include_str!("sql/drop_subscription_for_drop_dbs.sql"),
+                                datname_str = escape_literal(&db.name),
+                            );
+
+                            let operations = vec![Operation {
+                                query: drop_subscription_query,
+                                comment: Some(format!(
+                                    "optionally dropping subscriptions for DB {}",
+                                    db.name,
+                                )),
+                            }]
+                            .into_iter();
+
+                            Ok(Box::new(operations))
+                        }
+                        // skip this cleanup for the system databases
+                        // because users can't drop them
+                        DB::SystemDB => Ok(Box::new(empty())),
+                    }
+                }
                 PerDatabasePhase::DeleteDBRoleReferences => {
                     let ctx = ctx.read().await;
 
@@ -474,7 +498,19 @@ async fn get_operations<'a>(
                                         ),
                                         comment: None,
                                     },
+                                    // Revoke some potentially blocking privileges (Neon-specific currently)
+                                    Operation {
+                                        query: format!(
+                                            include_str!("sql/pre_drop_role_revoke_privileges.sql"),
+                                            role_name = quoted,
+                                        ),
+                                        comment: None,
+                                    },
                                     // This now will only drop privileges of the role
+                                    // TODO: this is obviously not 100% true because of the above case,
+                                    // there could be still some privileges that are not revoked. Maybe this
+                                    // only drops privileges that were granted *by this* role, not *to this* role,
+                                    // but this has to be checked.
                                     Operation {
                                         query: format!("DROP OWNED BY {}", quoted),
                                         comment: None,

compute_tools/src/sql/drop_subscription_for_drop_dbs.sql

@@ -0,0 +1,11 @@
+DO $$
+DECLARE
+    subname TEXT;
+BEGIN
+    FOR subname IN SELECT pg_subscription.subname FROM pg_subscription WHERE subdbid = (SELECT oid FROM pg_database WHERE datname = {datname_str}) LOOP
+        EXECUTE format('ALTER SUBSCRIPTION %I DISABLE;', subname);
+        EXECUTE format('ALTER SUBSCRIPTION %I SET (slot_name = NONE);', subname);
+        EXECUTE format('DROP SUBSCRIPTION %I;', subname);
+    END LOOP;
+END;
+$$;

compute_tools/src/sql/pre_drop_role_revoke_privileges.sql

@@ -0,0 +1,28 @@
+SET SESSION ROLE neon_superuser;
+
+DO $$
+DECLARE
+    schema TEXT;
+    revoke_query TEXT;
+BEGIN
+    FOR schema IN
+        SELECT schema_name
+        FROM information_schema.schemata
+        -- So far, we only had issues with 'public' schema. Probably, because we do some additional grants,
+        -- e.g., make DB owner the owner of 'public' schema automatically (when created via API).
+        -- See https://github.com/neondatabase/cloud/issues/13582 for the context.
+        -- Still, keep the loop because i) it efficiently handles the case when there is no 'public' schema,
+        -- ii) it's easy to add more schemas to the list if needed.
+        WHERE schema_name IN ('public')
+    LOOP
+        revoke_query := format(
+            'REVOKE ALL PRIVILEGES ON ALL TABLES IN SCHEMA %I FROM {role_name} GRANTED BY neon_superuser;',
+            schema
+        );
+
+        EXECUTE revoke_query;
+    END LOOP;
+END;
+$$;
+
+RESET ROLE;

control_plane/src/bin/neon_local.rs

@@ -19,6 +19,7 @@ use control_plane::storage_controller::{
     NeonStorageControllerStartArgs, NeonStorageControllerStopArgs, StorageController,
 };
 use control_plane::{broker, local_env};
+use nix::fcntl::{flock, FlockArg};
 use pageserver_api::config::{
     DEFAULT_HTTP_LISTEN_PORT as DEFAULT_PAGESERVER_HTTP_PORT,
     DEFAULT_PG_LISTEN_PORT as DEFAULT_PAGESERVER_PG_PORT,
@@ -36,6 +37,8 @@ use safekeeper_api::{
 };
 use std::borrow::Cow;
 use std::collections::{BTreeSet, HashMap};
+use std::fs::File;
+use std::os::fd::AsRawFd;
 use std::path::PathBuf;
 use std::process::exit;
 use std::str::FromStr;
@@ -689,6 +692,21 @@ struct TimelineTreeEl {
     pub children: BTreeSet<TimelineId>,
 }
 
+/// A flock-based guard over the neon_local repository directory
+struct RepoLock {
+    _file: File,
+}
+
+impl RepoLock {
+    fn new() -> Result<Self> {
+        let repo_dir = File::open(local_env::base_path())?;
+        let repo_dir_fd = repo_dir.as_raw_fd();
+        flock(repo_dir_fd, FlockArg::LockExclusive)?;
+
+        Ok(Self { _file: repo_dir })
+    }
+}
+
 // Main entry point for the 'neon_local' CLI utility
 //
 // This utility helps to manage neon installation. That includes following:
@@ -700,9 +718,14 @@ fn main() -> Result<()> {
     let cli = Cli::parse();
 
     // Check for 'neon init' command first.
-    let subcommand_result = if let NeonLocalCmd::Init(args) = cli.command {
-        handle_init(&args).map(|env| Some(Cow::Owned(env)))
+    let (subcommand_result, _lock) = if let NeonLocalCmd::Init(args) = cli.command {
+        (handle_init(&args).map(|env| Some(Cow::Owned(env))), None)
     } else {
+        // This tool uses a collection of simple files to store its state, and consequently
+        // it is not generally safe to run multiple commands concurrently.  Rather than expect
+        // all callers to know this, use a lock file to protect against concurrent execution.
+        let _repo_lock = RepoLock::new().unwrap();
+
         // all other commands need an existing config
         let env = LocalEnv::load_config(&local_env::base_path()).context("Error loading config")?;
         let original_env = env.clone();
@@ -728,11 +751,12 @@ fn main() -> Result<()> {
             NeonLocalCmd::Mappings(subcmd) => handle_mappings(&subcmd, env),
         };
 
-        if &original_env != env {
+        let subcommand_result = if &original_env != env {
             subcommand_result.map(|()| Some(Cow::Borrowed(env)))
         } else {
             subcommand_result.map(|()| None)
-        }
+        };
+        (subcommand_result, Some(_repo_lock))
     };
 
     match subcommand_result {
@@ -922,7 +946,7 @@ fn handle_init(args: &InitCmdArgs) -> anyhow::Result<LocalEnv> {
     } else {
         // User (likely interactive) did not provide a description of the environment, give them the default
         NeonLocalInitConf {
-            control_plane_api: Some(Some(DEFAULT_PAGESERVER_CONTROL_PLANE_API.parse().unwrap())),
+            control_plane_api: Some(DEFAULT_PAGESERVER_CONTROL_PLANE_API.parse().unwrap()),
             broker: NeonBroker {
                 listen_addr: DEFAULT_BROKER_ADDR.parse().unwrap(),
             },
@@ -1718,18 +1742,15 @@ async fn handle_start_all_impl(
             broker::start_broker_process(env, &retry_timeout).await
         });
 
-        // Only start the storage controller if the pageserver is configured to need it
-        if env.control_plane_api.is_some() {
-            js.spawn(async move {
-                let storage_controller = StorageController::from_env(env);
-                storage_controller
-                    .start(NeonStorageControllerStartArgs::with_default_instance_id(
-                        retry_timeout,
-                    ))
-                    .await
-                    .map_err(|e| e.context("start storage_controller"))
-            });
-        }
+        js.spawn(async move {
+            let storage_controller = StorageController::from_env(env);
+            storage_controller
+                .start(NeonStorageControllerStartArgs::with_default_instance_id(
+                    retry_timeout,
+                ))
+                .await
+                .map_err(|e| e.context("start storage_controller"))
+        });
 
         for ps_conf in &env.pageservers {
             js.spawn(async move {
@@ -1774,10 +1795,6 @@ async fn neon_start_status_check(
     const RETRY_INTERVAL: Duration = Duration::from_millis(100);
     const NOTICE_AFTER_RETRIES: Duration = Duration::from_secs(5);
 
-    if env.control_plane_api.is_none() {
-        return Ok(());
-    }
-
     let storcon = StorageController::from_env(env);
 
     let retries = retry_timeout.as_millis() / RETRY_INTERVAL.as_millis();

control_plane/src/endpoint.rs

@@ -62,7 +62,7 @@ use crate::local_env::LocalEnv;
 use crate::postgresql_conf::PostgresConf;
 use crate::storage_controller::StorageController;
 
-use compute_api::responses::{ComputeState, ComputeStatus};
+use compute_api::responses::{ComputeStatus, ComputeStatusResponse};
 use compute_api::spec::{Cluster, ComputeFeature, ComputeMode, ComputeSpec};
 
 // contents of a endpoint.json file
@@ -316,6 +316,10 @@ impl Endpoint {
         // and can cause errors like 'no unpinned buffers available', see
         // <https://github.com/neondatabase/neon/issues/9956>
         conf.append("shared_buffers", "1MB");
+        // Postgres defaults to effective_io_concurrency=1, which does not exercise the pageserver's
+        // batching logic.  Set this to 2 so that we exercise the code a bit without letting
+        // individual tests do a lot of concurrent work on underpowered test machines
+        conf.append("effective_io_concurrency", "2");
         conf.append("fsync", "off");
         conf.append("max_connections", "100");
         conf.append("wal_level", "logical");
@@ -581,6 +585,7 @@ impl Endpoint {
             features: self.features.clone(),
             swap_size_bytes: None,
             disk_quota_bytes: None,
+            disable_lfc_resizing: None,
             cluster: Cluster {
                 cluster_id: None, // project ID: not used
                 name: None,       // project name: not used
@@ -734,7 +739,7 @@ impl Endpoint {
     }
 
     // Call the /status HTTP API
-    pub async fn get_status(&self) -> Result<ComputeState> {
+    pub async fn get_status(&self) -> Result<ComputeStatusResponse> {
         let client = reqwest::Client::new();
 
         let response = client

control_plane/src/local_env.rs

@@ -76,7 +76,7 @@ pub struct LocalEnv {
 
     // Control plane upcall API for pageserver: if None, we will not run storage_controller  If set, this will
     // be propagated into each pageserver's configuration.
-    pub control_plane_api: Option<Url>,
+    pub control_plane_api: Url,
 
     // Control plane upcall API for storage controller.  If set, this will be propagated into the
     // storage controller's configuration.
@@ -133,7 +133,7 @@ pub struct NeonLocalInitConf {
     pub storage_controller: Option<NeonStorageControllerConf>,
     pub pageservers: Vec<NeonLocalInitPageserverConf>,
     pub safekeepers: Vec<SafekeeperConf>,
-    pub control_plane_api: Option<Option<Url>>,
+    pub control_plane_api: Option<Url>,
     pub control_plane_compute_hook_api: Option<Option<Url>>,
 }
 
@@ -180,7 +180,7 @@ impl NeonStorageControllerConf {
     const DEFAULT_MAX_WARMING_UP_INTERVAL: std::time::Duration = std::time::Duration::from_secs(30);
 
     // Very tight heartbeat interval to speed up tests
-    const DEFAULT_HEARTBEAT_INTERVAL: std::time::Duration = std::time::Duration::from_millis(100);
+    const DEFAULT_HEARTBEAT_INTERVAL: std::time::Duration = std::time::Duration::from_millis(1000);
 }
 
 impl Default for NeonStorageControllerConf {
@@ -535,7 +535,7 @@ impl LocalEnv {
                 storage_controller,
                 pageservers,
                 safekeepers,
-                control_plane_api,
+                control_plane_api: control_plane_api.unwrap(),
                 control_plane_compute_hook_api,
                 branch_name_mappings,
             }
@@ -638,7 +638,7 @@ impl LocalEnv {
                 storage_controller: self.storage_controller.clone(),
                 pageservers: vec![], // it's skip_serializing anyway
                 safekeepers: self.safekeepers.clone(),
-                control_plane_api: self.control_plane_api.clone(),
+                control_plane_api: Some(self.control_plane_api.clone()),
                 control_plane_compute_hook_api: self.control_plane_compute_hook_api.clone(),
                 branch_name_mappings: self.branch_name_mappings.clone(),
             },
@@ -768,7 +768,7 @@ impl LocalEnv {
             storage_controller: storage_controller.unwrap_or_default(),
             pageservers: pageservers.iter().map(Into::into).collect(),
             safekeepers,
-            control_plane_api: control_plane_api.unwrap_or_default(),
+            control_plane_api: control_plane_api.unwrap(),
             control_plane_compute_hook_api: control_plane_compute_hook_api.unwrap_or_default(),
             branch_name_mappings: Default::default(),
         };

control_plane/src/pageserver.rs

@@ -95,21 +95,19 @@ impl PageServerNode {
 
         let mut overrides = vec![pg_distrib_dir_param, broker_endpoint_param];
 
-        if let Some(control_plane_api) = &self.env.control_plane_api {
-            overrides.push(format!(
-                "control_plane_api='{}'",
-                control_plane_api.as_str()
-            ));
+        overrides.push(format!(
+            "control_plane_api='{}'",
+            self.env.control_plane_api.as_str()
+        ));
 
-            // Storage controller uses the same auth as pageserver: if JWT is enabled
-            // for us, we will also need it to talk to them.
-            if matches!(conf.http_auth_type, AuthType::NeonJWT) {
-                let jwt_token = self
-                    .env
-                    .generate_auth_token(&Claims::new(None, Scope::GenerationsApi))
-                    .unwrap();
-                overrides.push(format!("control_plane_api_token='{}'", jwt_token));
-            }
+        // Storage controller uses the same auth as pageserver: if JWT is enabled
+        // for us, we will also need it to talk to them.
+        if matches!(conf.http_auth_type, AuthType::NeonJWT) {
+            let jwt_token = self
+                .env
+                .generate_auth_token(&Claims::new(None, Scope::GenerationsApi))
+                .unwrap();
+            overrides.push(format!("control_plane_api_token='{}'", jwt_token));
         }
 
         if !conf.other.contains_key("remote_storage") {

control_plane/src/storage_controller.rs

@@ -338,7 +338,7 @@ impl StorageController {
                         .port(),
                 )
             } else {
-                let listen_url = self.env.control_plane_api.clone().unwrap();
+                let listen_url = self.env.control_plane_api.clone();
 
                 let listen = format!(
                     "{}:{}",
@@ -708,7 +708,7 @@ impl StorageController {
         } else {
             // The configured URL has the /upcall path prefix for pageservers to use: we will strip that out
             // for general purpose API access.
-            let listen_url = self.env.control_plane_api.clone().unwrap();
+            let listen_url = self.env.control_plane_api.clone();
             Url::from_str(&format!(
                 "http://{}:{}/{path}",
                 listen_url.host_str().unwrap(),

control_plane/storcon_cli/src/main.rs

@@ -5,7 +5,8 @@ use clap::{Parser, Subcommand};
 use pageserver_api::{
     controller_api::{
         AvailabilityZone, NodeAvailabilityWrapper, NodeDescribeResponse, NodeShardResponse,
-        ShardSchedulingPolicy, TenantCreateRequest, TenantDescribeResponse, TenantPolicyRequest,
+        SafekeeperDescribeResponse, ShardSchedulingPolicy, TenantCreateRequest,
+        TenantDescribeResponse, TenantPolicyRequest,
     },
     models::{
         EvictionPolicy, EvictionPolicyLayerAccessThreshold, LocationConfigSecondary,
@@ -211,6 +212,8 @@ enum Command {
         #[arg(long)]
         timeout: humantime::Duration,
     },
+    /// List safekeepers known to the storage controller
+    Safekeepers {},
 }
 
 #[derive(Parser)]
@@ -1020,6 +1023,40 @@ async fn main() -> anyhow::Result<()> {
                 "Fill was cancelled for node {node_id}. Schedulling policy is now {final_policy:?}"
             );
         }
+        Command::Safekeepers {} => {
+            let mut resp = storcon_client
+                .dispatch::<(), Vec<SafekeeperDescribeResponse>>(
+                    Method::GET,
+                    "control/v1/safekeeper".to_string(),
+                    None,
+                )
+                .await?;
+
+            resp.sort_by(|a, b| a.id.cmp(&b.id));
+
+            let mut table = comfy_table::Table::new();
+            table.set_header([
+                "Id",
+                "Version",
+                "Host",
+                "Port",
+                "Http Port",
+                "AZ Id",
+                "Scheduling",
+            ]);
+            for sk in resp {
+                table.add_row([
+                    format!("{}", sk.id),
+                    format!("{}", sk.version),
+                    sk.host,
+                    format!("{}", sk.port),
+                    format!("{}", sk.http_port),
+                    sk.availability_zone_id.clone(),
+                    String::from(sk.scheduling_policy),
+                ]);
+            }
+            println!("{table}");
+        }
     }
 
     Ok(())

docs/docker.md

@@ -7,15 +7,11 @@ Currently we build two main images:
 - [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. Built from [/compute-node/Dockerfile](/compute/compute-node.Dockerfile).
 
-And additional intermediate image:
-
-- [neondatabase/compute-tools](https://hub.docker.com/repository/docker/neondatabase/compute-tools) — compute node configuration management tools.
-
 ## 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-v16` (and -v15 and -v14)
+1. `neondatabase/compute-node-v17` (and -16, -v15, -v14)
 
 2. `neondatabase/neon`
 

docs/storage_broker.md

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

libs/compute_api/src/responses.rs

@@ -15,6 +15,17 @@ pub struct GenericAPIError {
     pub error: String,
 }
 
+#[derive(Debug, Clone, Serialize)]
+pub struct InfoResponse {
+    pub num_cpus: usize,
+}
+
+#[derive(Debug, Clone, Serialize)]
+pub struct ExtensionInstallResponse {
+    pub extension: PgIdent,
+    pub version: ExtVersion,
+}
+
 /// Response of the /status API
 #[derive(Serialize, Debug, Deserialize)]
 #[serde(rename_all = "snake_case")]
@@ -28,16 +39,6 @@ pub struct ComputeStatusResponse {
     pub error: Option<String>,
 }
 
-#[derive(Deserialize, Serialize)]
-#[serde(rename_all = "snake_case")]
-pub struct ComputeState {
-    pub status: ComputeStatus,
-    /// Timestamp of the last Postgres activity
-    #[serde(serialize_with = "rfc3339_serialize")]
-    pub last_active: Option<DateTime<Utc>>,
-    pub error: Option<String>,
-}
-
 #[derive(Serialize, Clone, Copy, Debug, Deserialize, PartialEq, Eq)]
 #[serde(rename_all = "snake_case")]
 pub enum ComputeStatus {
@@ -78,7 +79,7 @@ impl Display for ComputeStatus {
     }
 }
 
-fn rfc3339_serialize<S>(x: &Option<DateTime<Utc>>, s: S) -> Result<S::Ok, S::Error>
+pub fn rfc3339_serialize<S>(x: &Option<DateTime<Utc>>, s: S) -> Result<S::Ok, S::Error>
 where
     S: Serializer,
 {

libs/compute_api/src/spec.rs

@@ -67,6 +67,15 @@ pub struct ComputeSpec {
     #[serde(default)]
     pub disk_quota_bytes: Option<u64>,
 
+    /// Disables the vm-monitor behavior that resizes LFC on upscale/downscale, instead relying on
+    /// the initial size of LFC.
+    ///
+    /// This is intended for use when the LFC size is being overridden from the default but
+    /// autoscaling is still enabled, and we don't want the vm-monitor to interfere with the custom
+    /// LFC sizing.
+    #[serde(default)]
+    pub disable_lfc_resizing: Option<bool>,
+
     /// Expected cluster state at the end of transition process.
     pub cluster: Cluster,
     pub delta_operations: Option<Vec<DeltaOp>>,

libs/pageserver_api/src/controller_api.rs

@@ -320,6 +320,38 @@ impl From<NodeSchedulingPolicy> for String {
     }
 }
 
+#[derive(Serialize, Deserialize, Clone, Copy, Eq, PartialEq, Debug)]
+pub enum SkSchedulingPolicy {
+    Active,
+    Disabled,
+    Decomissioned,
+}
+
+impl FromStr for SkSchedulingPolicy {
+    type Err = anyhow::Error;
+
+    fn from_str(s: &str) -> Result<Self, Self::Err> {
+        Ok(match s {
+            "active" => Self::Active,
+            "disabled" => Self::Disabled,
+            "decomissioned" => Self::Decomissioned,
+            _ => return Err(anyhow::anyhow!("Unknown scheduling state '{s}'")),
+        })
+    }
+}
+
+impl From<SkSchedulingPolicy> for String {
+    fn from(value: SkSchedulingPolicy) -> String {
+        use SkSchedulingPolicy::*;
+        match value {
+            Active => "active",
+            Disabled => "disabled",
+            Decomissioned => "decomissioned",
+        }
+        .to_string()
+    }
+}
+
 /// Controls how tenant shards are mapped to locations on pageservers, e.g. whether
 /// to create secondary locations.
 #[derive(Clone, Serialize, Deserialize, Debug, PartialEq, Eq)]
@@ -372,6 +404,24 @@ pub struct MetadataHealthListOutdatedResponse {
     pub health_records: Vec<MetadataHealthRecord>,
 }
 
+/// Publicly exposed safekeeper description
+///
+/// The `active` flag which we have in the DB is not included on purpose: it is deprecated.
+#[derive(Serialize, Deserialize, Clone)]
+pub struct SafekeeperDescribeResponse {
+    pub id: NodeId,
+    pub region_id: String,
+    /// 1 is special, it means just created (not currently posted to storcon).
+    /// Zero or negative is not really expected.
+    /// Otherwise the number from `release-$(number_of_commits_on_branch)` tag.
+    pub version: i64,
+    pub host: String,
+    pub port: i32,
+    pub http_port: i32,
+    pub availability_zone_id: String,
+    pub scheduling_policy: SkSchedulingPolicy,
+}
+
 #[cfg(test)]
 mod test {
     use super::*;

libs/pageserver_api/src/models.rs

@@ -6,6 +6,7 @@ pub mod utilization;
 use camino::Utf8PathBuf;
 pub use utilization::PageserverUtilization;
 
+use core::ops::Range;
 use std::{
     collections::HashMap,
     fmt::Display,
@@ -28,6 +29,7 @@ use utils::{
 };
 
 use crate::{
+    key::Key,
     reltag::RelTag,
     shard::{ShardCount, ShardStripeSize, TenantShardId},
 };
@@ -210,6 +212,70 @@ pub enum TimelineState {
     Broken { reason: String, backtrace: String },
 }
 
+#[serde_with::serde_as]
+#[derive(Debug, Clone, serde::Deserialize, serde::Serialize)]
+pub struct CompactLsnRange {
+    pub start: Lsn,
+    pub end: Lsn,
+}
+
+#[serde_with::serde_as]
+#[derive(Debug, Clone, serde::Deserialize, serde::Serialize)]
+pub struct CompactKeyRange {
+    #[serde_as(as = "serde_with::DisplayFromStr")]
+    pub start: Key,
+    #[serde_as(as = "serde_with::DisplayFromStr")]
+    pub end: Key,
+}
+
+impl From<Range<Lsn>> for CompactLsnRange {
+    fn from(range: Range<Lsn>) -> Self {
+        Self {
+            start: range.start,
+            end: range.end,
+        }
+    }
+}
+
+impl From<Range<Key>> for CompactKeyRange {
+    fn from(range: Range<Key>) -> Self {
+        Self {
+            start: range.start,
+            end: range.end,
+        }
+    }
+}
+
+impl From<CompactLsnRange> for Range<Lsn> {
+    fn from(range: CompactLsnRange) -> Self {
+        range.start..range.end
+    }
+}
+
+impl From<CompactKeyRange> for Range<Key> {
+    fn from(range: CompactKeyRange) -> Self {
+        range.start..range.end
+    }
+}
+
+impl CompactLsnRange {
+    pub fn above(lsn: Lsn) -> Self {
+        Self {
+            start: lsn,
+            end: Lsn::MAX,
+        }
+    }
+}
+
+#[derive(Debug, Clone, Serialize)]
+pub struct CompactInfoResponse {
+    pub compact_key_range: Option<CompactKeyRange>,
+    pub compact_lsn_range: Option<CompactLsnRange>,
+    pub sub_compaction: bool,
+    pub running: bool,
+    pub job_id: usize,
+}
+
 #[derive(Serialize, Deserialize, Clone)]
 pub struct TimelineCreateRequest {
     pub new_timeline_id: TimelineId,
@@ -1396,75 +1462,91 @@ impl TryFrom<u8> for PagestreamBeMessageTag {
 // interface allows sending both LSNs, and let the pageserver do the right thing. There was no
 // difference in the responses between V1 and V2.
 //
-#[derive(Clone, Copy)]
+// V3 version of protocol adds request ID to all requests. This request ID is also included in response
+// as well as other fields from requests, which allows to verify that we receive response for our request.
+// We copy fields from request to response to make checking more reliable: request ID is formed from process ID
+// and local counter, so in principle there can be duplicated requests IDs if process PID is reused.
+//
+#[derive(Debug, PartialEq, Eq, Clone, Copy)]
 pub enum PagestreamProtocolVersion {
     V2,
+    V3,
 }
 
-#[derive(Debug, PartialEq, Eq)]
+pub type RequestId = u64;
+
+#[derive(Debug, PartialEq, Eq, Clone, Copy)]
+pub struct PagestreamRequest {
+    pub reqid: RequestId,
+    pub request_lsn: Lsn,
+    pub not_modified_since: Lsn,
+}
+
+#[derive(Debug, PartialEq, Eq, Clone, Copy)]
 pub struct PagestreamExistsRequest {
-    pub request_lsn: Lsn,
-    pub not_modified_since: Lsn,
+    pub hdr: PagestreamRequest,
     pub rel: RelTag,
 }
 
-#[derive(Debug, PartialEq, Eq)]
+#[derive(Debug, PartialEq, Eq, Clone, Copy)]
 pub struct PagestreamNblocksRequest {
-    pub request_lsn: Lsn,
-    pub not_modified_since: Lsn,
+    pub hdr: PagestreamRequest,
     pub rel: RelTag,
 }
 
-#[derive(Debug, PartialEq, Eq)]
+#[derive(Debug, PartialEq, Eq, Clone, Copy)]
 pub struct PagestreamGetPageRequest {
-    pub request_lsn: Lsn,
-    pub not_modified_since: Lsn,
+    pub hdr: PagestreamRequest,
     pub rel: RelTag,
     pub blkno: u32,
 }
 
-#[derive(Debug, PartialEq, Eq)]
+#[derive(Debug, PartialEq, Eq, Clone, Copy)]
 pub struct PagestreamDbSizeRequest {
-    pub request_lsn: Lsn,
-    pub not_modified_since: Lsn,
+    pub hdr: PagestreamRequest,
     pub dbnode: u32,
 }
 
-#[derive(Debug, PartialEq, Eq)]
+#[derive(Debug, PartialEq, Eq, Clone, Copy)]
 pub struct PagestreamGetSlruSegmentRequest {
-    pub request_lsn: Lsn,
-    pub not_modified_since: Lsn,
+    pub hdr: PagestreamRequest,
     pub kind: u8,
     pub segno: u32,
 }
 
 #[derive(Debug)]
 pub struct PagestreamExistsResponse {
+    pub req: PagestreamExistsRequest,
     pub exists: bool,
 }
 
 #[derive(Debug)]
 pub struct PagestreamNblocksResponse {
+    pub req: PagestreamNblocksRequest,
     pub n_blocks: u32,
 }
 
 #[derive(Debug)]
 pub struct PagestreamGetPageResponse {
+    pub req: PagestreamGetPageRequest,
     pub page: Bytes,
 }
 
 #[derive(Debug)]
 pub struct PagestreamGetSlruSegmentResponse {
+    pub req: PagestreamGetSlruSegmentRequest,
     pub segment: Bytes,
 }
 
 #[derive(Debug)]
 pub struct PagestreamErrorResponse {
+    pub req: PagestreamRequest,
     pub message: String,
 }
 
 #[derive(Debug)]
 pub struct PagestreamDbSizeResponse {
+    pub req: PagestreamDbSizeRequest,
     pub db_size: i64,
 }
 
@@ -1481,15 +1563,16 @@ pub struct TenantHistorySize {
 
 impl PagestreamFeMessage {
     /// Serialize a compute -> pageserver message. This is currently only used in testing
-    /// tools. Always uses protocol version 2.
+    /// tools. Always uses protocol version 3.
     pub fn serialize(&self) -> Bytes {
         let mut bytes = BytesMut::new();
 
         match self {
             Self::Exists(req) => {
                 bytes.put_u8(0);
-                bytes.put_u64(req.request_lsn.0);
-                bytes.put_u64(req.not_modified_since.0);
+                bytes.put_u64(req.hdr.reqid);
+                bytes.put_u64(req.hdr.request_lsn.0);
+                bytes.put_u64(req.hdr.not_modified_since.0);
                 bytes.put_u32(req.rel.spcnode);
                 bytes.put_u32(req.rel.dbnode);
                 bytes.put_u32(req.rel.relnode);
@@ -1498,8 +1581,9 @@ impl PagestreamFeMessage {
 
             Self::Nblocks(req) => {
                 bytes.put_u8(1);
-                bytes.put_u64(req.request_lsn.0);
-                bytes.put_u64(req.not_modified_since.0);
+                bytes.put_u64(req.hdr.reqid);
+                bytes.put_u64(req.hdr.request_lsn.0);
+                bytes.put_u64(req.hdr.not_modified_since.0);
                 bytes.put_u32(req.rel.spcnode);
                 bytes.put_u32(req.rel.dbnode);
                 bytes.put_u32(req.rel.relnode);
@@ -1508,8 +1592,9 @@ impl PagestreamFeMessage {
 
             Self::GetPage(req) => {
                 bytes.put_u8(2);
-                bytes.put_u64(req.request_lsn.0);
-                bytes.put_u64(req.not_modified_since.0);
+                bytes.put_u64(req.hdr.reqid);
+                bytes.put_u64(req.hdr.request_lsn.0);
+                bytes.put_u64(req.hdr.not_modified_since.0);
                 bytes.put_u32(req.rel.spcnode);
                 bytes.put_u32(req.rel.dbnode);
                 bytes.put_u32(req.rel.relnode);
@@ -1519,15 +1604,17 @@ impl PagestreamFeMessage {
 
             Self::DbSize(req) => {
                 bytes.put_u8(3);
-                bytes.put_u64(req.request_lsn.0);
-                bytes.put_u64(req.not_modified_since.0);
+                bytes.put_u64(req.hdr.reqid);
+                bytes.put_u64(req.hdr.request_lsn.0);
+                bytes.put_u64(req.hdr.not_modified_since.0);
                 bytes.put_u32(req.dbnode);
             }
 
             Self::GetSlruSegment(req) => {
                 bytes.put_u8(4);
-                bytes.put_u64(req.request_lsn.0);
-                bytes.put_u64(req.not_modified_since.0);
+                bytes.put_u64(req.hdr.reqid);
+                bytes.put_u64(req.hdr.request_lsn.0);
+                bytes.put_u64(req.hdr.not_modified_since.0);
                 bytes.put_u8(req.kind);
                 bytes.put_u32(req.segno);
             }
@@ -1536,21 +1623,35 @@ impl PagestreamFeMessage {
         bytes.into()
     }
 
-    pub fn parse<R: std::io::Read>(body: &mut R) -> anyhow::Result<PagestreamFeMessage> {
+    pub fn parse<R: std::io::Read>(
+        body: &mut R,
+        protocol_version: PagestreamProtocolVersion,
+    ) -> anyhow::Result<PagestreamFeMessage> {
         // these correspond to the NeonMessageTag enum in pagestore_client.h
         //
         // TODO: consider using protobuf or serde bincode for less error prone
         // serialization.
         let msg_tag = body.read_u8()?;
-
-        // these two fields are the same for every request type
-        let request_lsn = Lsn::from(body.read_u64::<BigEndian>()?);
-        let not_modified_since = Lsn::from(body.read_u64::<BigEndian>()?);
+        let (reqid, request_lsn, not_modified_since) = match protocol_version {
+            PagestreamProtocolVersion::V2 => (
+                0,
+                Lsn::from(body.read_u64::<BigEndian>()?),
+                Lsn::from(body.read_u64::<BigEndian>()?),
+            ),
+            PagestreamProtocolVersion::V3 => (
+                body.read_u64::<BigEndian>()?,
+                Lsn::from(body.read_u64::<BigEndian>()?),
+                Lsn::from(body.read_u64::<BigEndian>()?),
+            ),
+        };
 
         match msg_tag {
             0 => Ok(PagestreamFeMessage::Exists(PagestreamExistsRequest {
-                request_lsn,
-                not_modified_since,
+                hdr: PagestreamRequest {
+                    reqid,
+                    request_lsn,
+                    not_modified_since,
+                },
                 rel: RelTag {
                     spcnode: body.read_u32::<BigEndian>()?,
                     dbnode: body.read_u32::<BigEndian>()?,
@@ -1559,8 +1660,11 @@ impl PagestreamFeMessage {
                 },
             })),
             1 => Ok(PagestreamFeMessage::Nblocks(PagestreamNblocksRequest {
-                request_lsn,
-                not_modified_since,
+                hdr: PagestreamRequest {
+                    reqid,
+                    request_lsn,
+                    not_modified_since,
+                },
                 rel: RelTag {
                     spcnode: body.read_u32::<BigEndian>()?,
                     dbnode: body.read_u32::<BigEndian>()?,
@@ -1569,8 +1673,11 @@ impl PagestreamFeMessage {
                 },
             })),
             2 => Ok(PagestreamFeMessage::GetPage(PagestreamGetPageRequest {
-                request_lsn,
-                not_modified_since,
+                hdr: PagestreamRequest {
+                    reqid,
+                    request_lsn,
+                    not_modified_since,
+                },
                 rel: RelTag {
                     spcnode: body.read_u32::<BigEndian>()?,
                     dbnode: body.read_u32::<BigEndian>()?,
@@ -1580,14 +1687,20 @@ impl PagestreamFeMessage {
                 blkno: body.read_u32::<BigEndian>()?,
             })),
             3 => Ok(PagestreamFeMessage::DbSize(PagestreamDbSizeRequest {
-                request_lsn,
-                not_modified_since,
+                hdr: PagestreamRequest {
+                    reqid,
+                    request_lsn,
+                    not_modified_since,
+                },
                 dbnode: body.read_u32::<BigEndian>()?,
             })),
             4 => Ok(PagestreamFeMessage::GetSlruSegment(
                 PagestreamGetSlruSegmentRequest {
-                    request_lsn,
-                    not_modified_since,
+                    hdr: PagestreamRequest {
+                        reqid,
+                        request_lsn,
+                        not_modified_since,
+                    },
                     kind: body.read_u8()?,
                     segno: body.read_u32::<BigEndian>()?,
                 },
@@ -1598,43 +1711,114 @@ impl PagestreamFeMessage {
 }
 
 impl PagestreamBeMessage {
-    pub fn serialize(&self) -> Bytes {
+    pub fn serialize(&self, protocol_version: PagestreamProtocolVersion) -> Bytes {
         let mut bytes = BytesMut::new();
 
         use PagestreamBeMessageTag as Tag;
-        match self {
-            Self::Exists(resp) => {
-                bytes.put_u8(Tag::Exists as u8);
-                bytes.put_u8(resp.exists as u8);
-            }
+        match protocol_version {
+            PagestreamProtocolVersion::V2 => {
+                match self {
+                    Self::Exists(resp) => {
+                        bytes.put_u8(Tag::Exists as u8);
+                        bytes.put_u8(resp.exists as u8);
+                    }
 
-            Self::Nblocks(resp) => {
-                bytes.put_u8(Tag::Nblocks as u8);
-                bytes.put_u32(resp.n_blocks);
-            }
+                    Self::Nblocks(resp) => {
+                        bytes.put_u8(Tag::Nblocks as u8);
+                        bytes.put_u32(resp.n_blocks);
+                    }
 
-            Self::GetPage(resp) => {
-                bytes.put_u8(Tag::GetPage as u8);
-                bytes.put(&resp.page[..]);
-            }
+                    Self::GetPage(resp) => {
+                        bytes.put_u8(Tag::GetPage as u8);
+                        bytes.put(&resp.page[..])
+                    }
 
-            Self::Error(resp) => {
-                bytes.put_u8(Tag::Error as u8);
-                bytes.put(resp.message.as_bytes());
-                bytes.put_u8(0); // null terminator
-            }
-            Self::DbSize(resp) => {
-                bytes.put_u8(Tag::DbSize as u8);
-                bytes.put_i64(resp.db_size);
-            }
+                    Self::Error(resp) => {
+                        bytes.put_u8(Tag::Error as u8);
+                        bytes.put(resp.message.as_bytes());
+                        bytes.put_u8(0); // null terminator
+                    }
+                    Self::DbSize(resp) => {
+                        bytes.put_u8(Tag::DbSize as u8);
+                        bytes.put_i64(resp.db_size);
+                    }
 
-            Self::GetSlruSegment(resp) => {
-                bytes.put_u8(Tag::GetSlruSegment as u8);
-                bytes.put_u32((resp.segment.len() / BLCKSZ as usize) as u32);
-                bytes.put(&resp.segment[..]);
+                    Self::GetSlruSegment(resp) => {
+                        bytes.put_u8(Tag::GetSlruSegment as u8);
+                        bytes.put_u32((resp.segment.len() / BLCKSZ as usize) as u32);
+                        bytes.put(&resp.segment[..]);
+                    }
+                }
+            }
+            PagestreamProtocolVersion::V3 => {
+                match self {
+                    Self::Exists(resp) => {
+                        bytes.put_u8(Tag::Exists as u8);
+                        bytes.put_u64(resp.req.hdr.reqid);
+                        bytes.put_u64(resp.req.hdr.request_lsn.0);
+                        bytes.put_u64(resp.req.hdr.not_modified_since.0);
+                        bytes.put_u32(resp.req.rel.spcnode);
+                        bytes.put_u32(resp.req.rel.dbnode);
+                        bytes.put_u32(resp.req.rel.relnode);
+                        bytes.put_u8(resp.req.rel.forknum);
+                        bytes.put_u8(resp.exists as u8);
+                    }
+
+                    Self::Nblocks(resp) => {
+                        bytes.put_u8(Tag::Nblocks as u8);
+                        bytes.put_u64(resp.req.hdr.reqid);
+                        bytes.put_u64(resp.req.hdr.request_lsn.0);
+                        bytes.put_u64(resp.req.hdr.not_modified_since.0);
+                        bytes.put_u32(resp.req.rel.spcnode);
+                        bytes.put_u32(resp.req.rel.dbnode);
+                        bytes.put_u32(resp.req.rel.relnode);
+                        bytes.put_u8(resp.req.rel.forknum);
+                        bytes.put_u32(resp.n_blocks);
+                    }
+
+                    Self::GetPage(resp) => {
+                        bytes.put_u8(Tag::GetPage as u8);
+                        bytes.put_u64(resp.req.hdr.reqid);
+                        bytes.put_u64(resp.req.hdr.request_lsn.0);
+                        bytes.put_u64(resp.req.hdr.not_modified_since.0);
+                        bytes.put_u32(resp.req.rel.spcnode);
+                        bytes.put_u32(resp.req.rel.dbnode);
+                        bytes.put_u32(resp.req.rel.relnode);
+                        bytes.put_u8(resp.req.rel.forknum);
+                        bytes.put_u32(resp.req.blkno);
+                        bytes.put(&resp.page[..])
+                    }
+
+                    Self::Error(resp) => {
+                        bytes.put_u8(Tag::Error as u8);
+                        bytes.put_u64(resp.req.reqid);
+                        bytes.put_u64(resp.req.request_lsn.0);
+                        bytes.put_u64(resp.req.not_modified_since.0);
+                        bytes.put(resp.message.as_bytes());
+                        bytes.put_u8(0); // null terminator
+                    }
+                    Self::DbSize(resp) => {
+                        bytes.put_u8(Tag::DbSize as u8);
+                        bytes.put_u64(resp.req.hdr.reqid);
+                        bytes.put_u64(resp.req.hdr.request_lsn.0);
+                        bytes.put_u64(resp.req.hdr.not_modified_since.0);
+                        bytes.put_u32(resp.req.dbnode);
+                        bytes.put_i64(resp.db_size);
+                    }
+
+                    Self::GetSlruSegment(resp) => {
+                        bytes.put_u8(Tag::GetSlruSegment as u8);
+                        bytes.put_u64(resp.req.hdr.reqid);
+                        bytes.put_u64(resp.req.hdr.request_lsn.0);
+                        bytes.put_u64(resp.req.hdr.not_modified_since.0);
+                        bytes.put_u8(resp.req.kind);
+                        bytes.put_u32(resp.req.segno);
+                        bytes.put_u32((resp.segment.len() / BLCKSZ as usize) as u32);
+                        bytes.put(&resp.segment[..]);
+                    }
+                }
             }
         }
-
         bytes.into()
     }
 
@@ -1646,38 +1830,131 @@ impl PagestreamBeMessage {
         let ok =
             match Tag::try_from(msg_tag).map_err(|tag: u8| anyhow::anyhow!("invalid tag {tag}"))? {
                 Tag::Exists => {
-                    let exists = buf.read_u8()?;
+                    let reqid = buf.read_u64::<BigEndian>()?;
+                    let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
+                    let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
+                    let rel = RelTag {
+                        spcnode: buf.read_u32::<BigEndian>()?,
+                        dbnode: buf.read_u32::<BigEndian>()?,
+                        relnode: buf.read_u32::<BigEndian>()?,
+                        forknum: buf.read_u8()?,
+                    };
+                    let exists = buf.read_u8()? != 0;
                     Self::Exists(PagestreamExistsResponse {
-                        exists: exists != 0,
+                        req: PagestreamExistsRequest {
+                            hdr: PagestreamRequest {
+                                reqid,
+                                request_lsn,
+                                not_modified_since,
+                            },
+                            rel,
+                        },
+                        exists,
                     })
                 }
                 Tag::Nblocks => {
+                    let reqid = buf.read_u64::<BigEndian>()?;
+                    let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
+                    let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
+                    let rel = RelTag {
+                        spcnode: buf.read_u32::<BigEndian>()?,
+                        dbnode: buf.read_u32::<BigEndian>()?,
+                        relnode: buf.read_u32::<BigEndian>()?,
+                        forknum: buf.read_u8()?,
+                    };
                     let n_blocks = buf.read_u32::<BigEndian>()?;
-                    Self::Nblocks(PagestreamNblocksResponse { n_blocks })
+                    Self::Nblocks(PagestreamNblocksResponse {
+                        req: PagestreamNblocksRequest {
+                            hdr: PagestreamRequest {
+                                reqid,
+                                request_lsn,
+                                not_modified_since,
+                            },
+                            rel,
+                        },
+                        n_blocks,
+                    })
                 }
                 Tag::GetPage => {
+                    let reqid = buf.read_u64::<BigEndian>()?;
+                    let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
+                    let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
+                    let rel = RelTag {
+                        spcnode: buf.read_u32::<BigEndian>()?,
+                        dbnode: buf.read_u32::<BigEndian>()?,
+                        relnode: buf.read_u32::<BigEndian>()?,
+                        forknum: buf.read_u8()?,
+                    };
+                    let blkno = buf.read_u32::<BigEndian>()?;
                     let mut page = vec![0; 8192]; // TODO: use MaybeUninit
                     buf.read_exact(&mut page)?;
-                    PagestreamBeMessage::GetPage(PagestreamGetPageResponse { page: page.into() })
+                    Self::GetPage(PagestreamGetPageResponse {
+                        req: PagestreamGetPageRequest {
+                            hdr: PagestreamRequest {
+                                reqid,
+                                request_lsn,
+                                not_modified_since,
+                            },
+                            rel,
+                            blkno,
+                        },
+                        page: page.into(),
+                    })
                 }
                 Tag::Error => {
+                    let reqid = buf.read_u64::<BigEndian>()?;
+                    let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
+                    let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
                     let mut msg = Vec::new();
                     buf.read_until(0, &mut msg)?;
                     let cstring = std::ffi::CString::from_vec_with_nul(msg)?;
                     let rust_str = cstring.to_str()?;
-                    PagestreamBeMessage::Error(PagestreamErrorResponse {
+                    Self::Error(PagestreamErrorResponse {
+                        req: PagestreamRequest {
+                            reqid,
+                            request_lsn,
+                            not_modified_since,
+                        },
                         message: rust_str.to_owned(),
                     })
                 }
                 Tag::DbSize => {
+                    let reqid = buf.read_u64::<BigEndian>()?;
+                    let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
+                    let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
+                    let dbnode = buf.read_u32::<BigEndian>()?;
                     let db_size = buf.read_i64::<BigEndian>()?;
-                    Self::DbSize(PagestreamDbSizeResponse { db_size })
+                    Self::DbSize(PagestreamDbSizeResponse {
+                        req: PagestreamDbSizeRequest {
+                            hdr: PagestreamRequest {
+                                reqid,
+                                request_lsn,
+                                not_modified_since,
+                            },
+                            dbnode,
+                        },
+                        db_size,
+                    })
                 }
                 Tag::GetSlruSegment => {
+                    let reqid = buf.read_u64::<BigEndian>()?;
+                    let request_lsn = Lsn(buf.read_u64::<BigEndian>()?);
+                    let not_modified_since = Lsn(buf.read_u64::<BigEndian>()?);
+                    let kind = buf.read_u8()?;
+                    let segno = buf.read_u32::<BigEndian>()?;
                     let n_blocks = buf.read_u32::<BigEndian>()?;
                     let mut segment = vec![0; n_blocks as usize * BLCKSZ as usize];
                     buf.read_exact(&mut segment)?;
                     Self::GetSlruSegment(PagestreamGetSlruSegmentResponse {
+                        req: PagestreamGetSlruSegmentRequest {
+                            hdr: PagestreamRequest {
+                                reqid,
+                                request_lsn,
+                                not_modified_since,
+                            },
+                            kind,
+                            segno,
+                        },
                         segment: segment.into(),
                     })
                 }
@@ -1716,8 +1993,11 @@ mod tests {
         // Test serialization/deserialization of PagestreamFeMessage
         let messages = vec![
             PagestreamFeMessage::Exists(PagestreamExistsRequest {
-                request_lsn: Lsn(4),
-                not_modified_since: Lsn(3),
+                hdr: PagestreamRequest {
+                    reqid: 0,
+                    request_lsn: Lsn(4),
+                    not_modified_since: Lsn(3),
+                },
                 rel: RelTag {
                     forknum: 1,
                     spcnode: 2,
@@ -1726,8 +2006,11 @@ mod tests {
                 },
             }),
             PagestreamFeMessage::Nblocks(PagestreamNblocksRequest {
-                request_lsn: Lsn(4),
-                not_modified_since: Lsn(4),
+                hdr: PagestreamRequest {
+                    reqid: 0,
+                    request_lsn: Lsn(4),
+                    not_modified_since: Lsn(4),
+                },
                 rel: RelTag {
                     forknum: 1,
                     spcnode: 2,
@@ -1736,8 +2019,11 @@ mod tests {
                 },
             }),
             PagestreamFeMessage::GetPage(PagestreamGetPageRequest {
-                request_lsn: Lsn(4),
-                not_modified_since: Lsn(3),
+                hdr: PagestreamRequest {
+                    reqid: 0,
+                    request_lsn: Lsn(4),
+                    not_modified_since: Lsn(3),
+                },
                 rel: RelTag {
                     forknum: 1,
                     spcnode: 2,
@@ -1747,14 +2033,19 @@ mod tests {
                 blkno: 7,
             }),
             PagestreamFeMessage::DbSize(PagestreamDbSizeRequest {
-                request_lsn: Lsn(4),
-                not_modified_since: Lsn(3),
+                hdr: PagestreamRequest {
+                    reqid: 0,
+                    request_lsn: Lsn(4),
+                    not_modified_since: Lsn(3),
+                },
                 dbnode: 7,
             }),
         ];
         for msg in messages {
             let bytes = msg.serialize();
-            let reconstructed = PagestreamFeMessage::parse(&mut bytes.reader()).unwrap();
+            let reconstructed =
+                PagestreamFeMessage::parse(&mut bytes.reader(), PagestreamProtocolVersion::V3)
+                    .unwrap();
             assert!(msg == reconstructed);
         }
     }

libs/postgres_ffi/src/wal_generator.rs

@@ -106,11 +106,11 @@ impl<R: RecordGenerator> WalGenerator<R> {
     const TIMELINE_ID: u32 = 1;
 
     /// Creates a new WAL generator with the given record generator.
-    pub fn new(record_generator: R) -> WalGenerator<R> {
+    pub fn new(record_generator: R, start_lsn: Lsn) -> WalGenerator<R> {
         Self {
             record_generator,
-            lsn: Lsn(0),
-            prev_lsn: Lsn(0),
+            lsn: start_lsn,
+            prev_lsn: start_lsn,
         }
     }
 

libs/proxy/tokio-postgres2/src/connection.rs

@@ -33,10 +33,14 @@ pub struct Response {
 #[derive(PartialEq, Debug)]
 enum State {
     Active,
-    Terminating,
     Closing,
 }
 
+enum WriteReady {
+    Terminating,
+    WaitingOnRead,
+}
+
 /// A connection to a PostgreSQL database.
 ///
 /// This is one half of what is returned when a new connection is established. It performs the actual IO with the
@@ -51,7 +55,6 @@ pub struct Connection<S, T> {
     /// HACK: we need this in the Neon Proxy to forward params.
     pub parameters: HashMap<String, String>,
     receiver: mpsc::UnboundedReceiver<Request>,
-    pending_request: Option<RequestMessages>,
     pending_responses: VecDeque<BackendMessage>,
     responses: VecDeque<Response>,
     state: State,
@@ -72,7 +75,6 @@ where
             stream,
             parameters,
             receiver,
-            pending_request: None,
             pending_responses,
             responses: VecDeque::new(),
             state: State::Active,
@@ -93,26 +95,23 @@ where
             .map(|o| o.map(|r| r.map_err(Error::io)))
     }
 
-    fn poll_read(&mut self, cx: &mut Context<'_>) -> Result<Option<AsyncMessage>, Error> {
-        if self.state != State::Active {
-            trace!("poll_read: done");
-            return Ok(None);
-        }
-
+    /// Read and process messages from the connection to postgres.
+    /// client <- postgres
+    fn poll_read(&mut self, cx: &mut Context<'_>) -> Poll<Result<AsyncMessage, Error>> {
         loop {
             let message = match self.poll_response(cx)? {
                 Poll::Ready(Some(message)) => message,
-                Poll::Ready(None) => return Err(Error::closed()),
+                Poll::Ready(None) => return Poll::Ready(Err(Error::closed())),
                 Poll::Pending => {
                     trace!("poll_read: waiting on response");
-                    return Ok(None);
+                    return Poll::Pending;
                 }
             };
 
             let (mut messages, request_complete) = match message {
                 BackendMessage::Async(Message::NoticeResponse(body)) => {
                     let error = DbError::parse(&mut body.fields()).map_err(Error::parse)?;
-                    return Ok(Some(AsyncMessage::Notice(error)));
+                    return Poll::Ready(Ok(AsyncMessage::Notice(error)));
                 }
                 BackendMessage::Async(Message::NotificationResponse(body)) => {
                     let notification = Notification {
@@ -120,7 +119,7 @@ where
                         channel: body.channel().map_err(Error::parse)?.to_string(),
                         payload: body.message().map_err(Error::parse)?.to_string(),
                     };
-                    return Ok(Some(AsyncMessage::Notification(notification)));
+                    return Poll::Ready(Ok(AsyncMessage::Notification(notification)));
                 }
                 BackendMessage::Async(Message::ParameterStatus(body)) => {
                     self.parameters.insert(
@@ -139,8 +138,10 @@ where
             let mut response = match self.responses.pop_front() {
                 Some(response) => response,
                 None => match messages.next().map_err(Error::parse)? {
-                    Some(Message::ErrorResponse(error)) => return Err(Error::db(error)),
-                    _ => return Err(Error::unexpected_message()),
+                    Some(Message::ErrorResponse(error)) => {
+                        return Poll::Ready(Err(Error::db(error)))
+                    }
+                    _ => return Poll::Ready(Err(Error::unexpected_message())),
                 },
             };
 
@@ -164,18 +165,14 @@ where
                         request_complete,
                     });
                     trace!("poll_read: waiting on sender");
-                    return Ok(None);
+                    return Poll::Pending;
                 }
             }
         }
     }
 
+    /// Fetch the next client request and enqueue the response sender.
     fn poll_request(&mut self, cx: &mut Context<'_>) -> Poll<Option<RequestMessages>> {
-        if let Some(messages) = self.pending_request.take() {
-            trace!("retrying pending request");
-            return Poll::Ready(Some(messages));
-        }
-
         if self.receiver.is_closed() {
             return Poll::Ready(None);
         }
@@ -193,74 +190,80 @@ where
         }
     }
 
-    fn poll_write(&mut self, cx: &mut Context<'_>) -> Result<bool, Error> {
+    /// Process client requests and write them to the postgres connection, flushing if necessary.
+    /// client -> postgres
+    fn poll_write(&mut self, cx: &mut Context<'_>) -> Poll<Result<WriteReady, Error>> {
         loop {
-            if self.state == State::Closing {
-                trace!("poll_write: done");
-                return Ok(false);
-            }
-
             if Pin::new(&mut self.stream)
                 .poll_ready(cx)
                 .map_err(Error::io)?
                 .is_pending()
             {
                 trace!("poll_write: waiting on socket");
-                return Ok(false);
+
+                // poll_ready is self-flushing.
+                return Poll::Pending;
             }
 
-            let request = match self.poll_request(cx) {
-                Poll::Ready(Some(request)) => request,
-                Poll::Ready(None) if self.responses.is_empty() && self.state == State::Active => {
+            match self.poll_request(cx) {
+                // send the message to postgres
+                Poll::Ready(Some(RequestMessages::Single(request))) => {
+                    Pin::new(&mut self.stream)
+                        .start_send(request)
+                        .map_err(Error::io)?;
+                }
+                // No more messages from the client, and no more responses to wait for.
+                // Send a terminate message to postgres
+                Poll::Ready(None) if self.responses.is_empty() => {
                     trace!("poll_write: at eof, terminating");
-                    self.state = State::Terminating;
                     let mut request = BytesMut::new();
                     frontend::terminate(&mut request);
-                    RequestMessages::Single(FrontendMessage::Raw(request.freeze()))
+                    let request = FrontendMessage::Raw(request.freeze());
+
+                    Pin::new(&mut self.stream)
+                        .start_send(request)
+                        .map_err(Error::io)?;
+
+                    trace!("poll_write: sent eof, closing");
+                    trace!("poll_write: done");
+                    return Poll::Ready(Ok(WriteReady::Terminating));
                 }
+                // No more messages from the client, but there are still some responses to wait for.
                 Poll::Ready(None) => {
                     trace!(
                         "poll_write: at eof, pending responses {}",
                         self.responses.len()
                     );
-                    return Ok(true);
+                    ready!(self.poll_flush(cx))?;
+                    return Poll::Ready(Ok(WriteReady::WaitingOnRead));
                 }
+                // Still waiting for a message from the client.
                 Poll::Pending => {
                     trace!("poll_write: waiting on request");
-                    return Ok(true);
-                }
-            };
-
-            match request {
-                RequestMessages::Single(request) => {
-                    Pin::new(&mut self.stream)
-                        .start_send(request)
-                        .map_err(Error::io)?;
-                    if self.state == State::Terminating {
-                        trace!("poll_write: sent eof, closing");
-                        self.state = State::Closing;
-                    }
+                    ready!(self.poll_flush(cx))?;
+                    return Poll::Pending;
                 }
             }
         }
     }
 
-    fn poll_flush(&mut self, cx: &mut Context<'_>) -> Result<(), Error> {
+    fn poll_flush(&mut self, cx: &mut Context<'_>) -> Poll<Result<(), Error>> {
         match Pin::new(&mut self.stream)
             .poll_flush(cx)
             .map_err(Error::io)?
         {
-            Poll::Ready(()) => trace!("poll_flush: flushed"),
-            Poll::Pending => trace!("poll_flush: waiting on socket"),
+            Poll::Ready(()) => {
+                trace!("poll_flush: flushed");
+                Poll::Ready(Ok(()))
+            }
+            Poll::Pending => {
+                trace!("poll_flush: waiting on socket");
+                Poll::Pending
+            }
         }
-        Ok(())
     }
 
     fn poll_shutdown(&mut self, cx: &mut Context<'_>) -> Poll<Result<(), Error>> {
-        if self.state != State::Closing {
-            return Poll::Pending;
-        }
-
         match Pin::new(&mut self.stream)
             .poll_close(cx)
             .map_err(Error::io)?
@@ -289,18 +292,30 @@ where
         &mut self,
         cx: &mut Context<'_>,
     ) -> Poll<Option<Result<AsyncMessage, Error>>> {
-        let message = self.poll_read(cx)?;
-        let want_flush = self.poll_write(cx)?;
-        if want_flush {
-            self.poll_flush(cx)?;
+        if self.state != State::Closing {
+            // if the state is still active, try read from and write to postgres.
+            let message = self.poll_read(cx)?;
+            let closing = self.poll_write(cx)?;
+            if let Poll::Ready(WriteReady::Terminating) = closing {
+                self.state = State::Closing;
+            }
+
+            if let Poll::Ready(message) = message {
+                return Poll::Ready(Some(Ok(message)));
+            }
+
+            // poll_read returned Pending.
+            // poll_write returned Pending or Ready(WriteReady::WaitingOnRead).
+            // if poll_write returned Ready(WriteReady::WaitingOnRead), then we are waiting to read more data from postgres.
+            if self.state != State::Closing {
+                return Poll::Pending;
+            }
         }
-        match message {
-            Some(message) => Poll::Ready(Some(Ok(message))),
-            None => match self.poll_shutdown(cx) {
-                Poll::Ready(Ok(())) => Poll::Ready(None),
-                Poll::Ready(Err(e)) => Poll::Ready(Some(Err(e))),
-                Poll::Pending => Poll::Pending,
-            },
+
+        match self.poll_shutdown(cx) {
+            Poll::Ready(Ok(())) => Poll::Ready(None),
+            Poll::Ready(Err(e)) => Poll::Ready(Some(Err(e))),
+            Poll::Pending => Poll::Pending,
         }
     }
 }

libs/remote_storage/src/config.rs

@@ -115,13 +115,15 @@ fn default_max_keys_per_list_response() -> Option<i32> {
 }
 
 fn default_azure_conn_pool_size() -> usize {
-    // Conservative default: no connection pooling.  At time of writing this is the Azure
-    // SDK's default as well, due to historic reports of hard-to-reproduce issues
+    // By default, the Azure SDK does no connection pooling, due to historic reports of hard-to-reproduce issues
     // (https://github.com/hyperium/hyper/issues/2312)
     //
     // However, using connection pooling is important to avoid exhausting client ports when
     // doing huge numbers of requests (https://github.com/neondatabase/cloud/issues/20971)
-    0
+    //
+    // We therefore enable a modest pool size by default: this may be configured to zero if
+    // issues like the alleged upstream hyper issue appear.
+    8
 }
 
 impl Debug for S3Config {

libs/tracing-utils/src/lib.rs

@@ -38,7 +38,6 @@ pub mod http;
 
 use opentelemetry::trace::TracerProvider;
 use opentelemetry::KeyValue;
-use opentelemetry_sdk::Resource;
 use tracing::Subscriber;
 use tracing_subscriber::registry::LookupSpan;
 use tracing_subscriber::Layer;
@@ -121,7 +120,10 @@ where
     S: Subscriber + for<'span> LookupSpan<'span>,
 {
     // Sets up exporter from the OTEL_EXPORTER_* environment variables.
-    let exporter = opentelemetry_otlp::new_exporter().http();
+    let exporter = opentelemetry_otlp::SpanExporter::builder()
+        .with_http()
+        .build()
+        .expect("could not initialize opentelemetry exporter");
 
     // TODO: opentelemetry::global::set_error_handler() with custom handler that
     //       bypasses default tracing layers, but logs regular looking log
@@ -132,17 +134,13 @@ where
         opentelemetry_sdk::propagation::TraceContextPropagator::new(),
     );
 
-    let tracer = opentelemetry_otlp::new_pipeline()
-        .tracing()
-        .with_exporter(exporter)
-        .with_trace_config(opentelemetry_sdk::trace::Config::default().with_resource(
-            Resource::new(vec![KeyValue::new(
-                opentelemetry_semantic_conventions::resource::SERVICE_NAME,
-                service_name,
-            )]),
-        ))
-        .install_batch(opentelemetry_sdk::runtime::Tokio)
-        .expect("could not initialize opentelemetry exporter")
+    let tracer = opentelemetry_sdk::trace::TracerProvider::builder()
+        .with_batch_exporter(exporter, opentelemetry_sdk::runtime::Tokio)
+        .with_resource(opentelemetry_sdk::Resource::new(vec![KeyValue::new(
+            opentelemetry_semantic_conventions::resource::SERVICE_NAME,
+            service_name,
+        )]))
+        .build()
         .tracer("global");
 
     tracing_opentelemetry::layer().with_tracer(tracer)

libs/utils/Cargo.toml

@@ -26,6 +26,7 @@ git-version.workspace = true
 hex = { workspace = true, features = ["serde"] }
 humantime.workspace = true
 hyper0 = { workspace = true, features = ["full"] }
+inferno.workspace = true
 itertools.workspace = true
 fail.workspace = true
 futures = { workspace = true }

libs/utils/src/failpoint_support.rs

@@ -9,7 +9,7 @@ use serde::{Deserialize, Serialize};
 use tokio_util::sync::CancellationToken;
 use tracing::*;
 
-/// Declare a failpoint that can use the `pause` failpoint action.
+/// Declare a failpoint that can use to `pause` failpoint action.
 /// We don't want to block the executor thread, hence, spawn_blocking + await.
 #[macro_export]
 macro_rules! pausable_failpoint {
@@ -181,7 +181,7 @@ pub async fn failpoints_handler(
 ) -> Result<Response<Body>, ApiError> {
     if !fail::has_failpoints() {
         return Err(ApiError::BadRequest(anyhow::anyhow!(
-            "Cannot manage failpoints because storage was compiled without failpoints support"
+            "Cannot manage failpoints because neon was compiled without failpoints support"
         )));
     }
 

libs/utils/src/http/endpoint.rs

@@ -417,6 +417,7 @@ pub async fn profile_heap_handler(req: Request<Body>) -> Result<Response<Body>,
     enum Format {
         Jemalloc,
         Pprof,
+        Svg,
     }
 
     // Parameters.
@@ -424,9 +425,24 @@ pub async fn profile_heap_handler(req: Request<Body>) -> Result<Response<Body>,
         None => Format::Pprof,
         Some("jemalloc") => Format::Jemalloc,
         Some("pprof") => Format::Pprof,
+        Some("svg") => Format::Svg,
         Some(format) => return Err(ApiError::BadRequest(anyhow!("invalid format {format}"))),
     };
 
+    // Functions and mappings to strip when symbolizing pprof profiles. If true,
+    // also remove child frames.
+    static STRIP_FUNCTIONS: Lazy<Vec<(Regex, bool)>> = Lazy::new(|| {
+        vec![
+            (Regex::new("^__rust").unwrap(), false),
+            (Regex::new("^_start$").unwrap(), false),
+            (Regex::new("^irallocx_prof").unwrap(), true),
+            (Regex::new("^prof_alloc_prep").unwrap(), true),
+            (Regex::new("^std::rt::lang_start").unwrap(), false),
+            (Regex::new("^std::sys::backtrace::__rust").unwrap(), false),
+        ]
+    });
+    const STRIP_MAPPINGS: &[&str] = &["libc", "libgcc", "pthread", "vdso"];
+
     // Obtain profiler handle.
     let mut prof_ctl = jemalloc_pprof::PROF_CTL
         .as_ref()
@@ -464,24 +480,9 @@ pub async fn profile_heap_handler(req: Request<Body>) -> Result<Response<Body>,
                 // Symbolize the profile.
                 // TODO: consider moving this upstream to jemalloc_pprof and avoiding the
                 // serialization roundtrip.
-                static STRIP_FUNCTIONS: Lazy<Vec<(Regex, bool)>> = Lazy::new(|| {
-                    // Functions to strip from profiles. If true, also remove child frames.
-                    vec![
-                        (Regex::new("^__rust").unwrap(), false),
-                        (Regex::new("^_start$").unwrap(), false),
-                        (Regex::new("^irallocx_prof").unwrap(), true),
-                        (Regex::new("^prof_alloc_prep").unwrap(), true),
-                        (Regex::new("^std::rt::lang_start").unwrap(), false),
-                        (Regex::new("^std::sys::backtrace::__rust").unwrap(), false),
-                    ]
-                });
                 let profile = pprof::decode(&bytes)?;
                 let profile = pprof::symbolize(profile)?;
-                let profile = pprof::strip_locations(
-                    profile,
-                    &["libc", "libgcc", "pthread", "vdso"],
-                    &STRIP_FUNCTIONS,
-                );
+                let profile = pprof::strip_locations(profile, STRIP_MAPPINGS, &STRIP_FUNCTIONS);
                 pprof::encode(&profile)
             })
             .await
@@ -494,6 +495,27 @@ pub async fn profile_heap_handler(req: Request<Body>) -> Result<Response<Body>,
                 .body(Body::from(data))
                 .map_err(|err| ApiError::InternalServerError(err.into()))
         }
+
+        Format::Svg => {
+            let body = tokio::task::spawn_blocking(move || {
+                let bytes = prof_ctl.dump_pprof()?;
+                let profile = pprof::decode(&bytes)?;
+                let profile = pprof::symbolize(profile)?;
+                let profile = pprof::strip_locations(profile, STRIP_MAPPINGS, &STRIP_FUNCTIONS);
+                let mut opts = inferno::flamegraph::Options::default();
+                opts.title = "Heap inuse".to_string();
+                opts.count_name = "bytes".to_string();
+                pprof::flamegraph(profile, &mut opts)
+            })
+            .await
+            .map_err(|join_err| ApiError::InternalServerError(join_err.into()))?
+            .map_err(ApiError::InternalServerError)?;
+            Response::builder()
+                .status(200)
+                .header(CONTENT_TYPE, "image/svg+xml")
+                .body(Body::from(body))
+                .map_err(|err| ApiError::InternalServerError(err.into()))
+        }
     }
 }
 

libs/utils/src/pprof.rs

@@ -1,8 +1,9 @@
+use anyhow::bail;
 use flate2::write::{GzDecoder, GzEncoder};
 use flate2::Compression;
 use itertools::Itertools as _;
 use once_cell::sync::Lazy;
-use pprof::protos::{Function, Line, Message as _, Profile};
+use pprof::protos::{Function, Line, Location, Message as _, Profile};
 use regex::Regex;
 
 use std::borrow::Cow;
@@ -188,3 +189,59 @@ pub fn strip_locations(
 
     profile
 }
+
+/// Generates an SVG flamegraph from a symbolized pprof profile.
+pub fn flamegraph(
+    profile: Profile,
+    opts: &mut inferno::flamegraph::Options,
+) -> anyhow::Result<Vec<u8>> {
+    if profile.mapping.iter().any(|m| !m.has_functions) {
+        bail!("profile not symbolized");
+    }
+
+    // Index locations, functions, and strings.
+    let locations: HashMap<u64, Location> =
+        profile.location.into_iter().map(|l| (l.id, l)).collect();
+    let functions: HashMap<u64, Function> =
+        profile.function.into_iter().map(|f| (f.id, f)).collect();
+    let strings = profile.string_table;
+
+    // Resolve stacks as function names, and sum sample values per stack. Also reverse the stack,
+    // since inferno expects it bottom-up.
+    let mut stacks: HashMap<Vec<&str>, i64> = HashMap::new();
+    for sample in profile.sample {
+        let mut stack = Vec::with_capacity(sample.location_id.len());
+        for location in sample.location_id.into_iter().rev() {
+            let Some(location) = locations.get(&location) else {
+                bail!("missing location {location}");
+            };
+            for line in location.line.iter().rev() {
+                let Some(function) = functions.get(&line.function_id) else {
+                    bail!("missing function {}", line.function_id);
+                };
+                let Some(name) = strings.get(function.name as usize) else {
+                    bail!("missing string {}", function.name);
+                };
+                stack.push(name.as_str());
+            }
+        }
+        let Some(&value) = sample.value.first() else {
+            bail!("missing value");
+        };
+        *stacks.entry(stack).or_default() += value;
+    }
+
+    // Construct stack lines for inferno.
+    let lines = stacks
+        .into_iter()
+        .map(|(stack, value)| (stack.into_iter().join(";"), value))
+        .map(|(stack, value)| format!("{stack} {value}"))
+        .sorted()
+        .collect_vec();
+
+    // Construct the flamegraph.
+    let mut bytes = Vec::new();
+    let lines = lines.iter().map(|line| line.as_str());
+    inferno::flamegraph::from_lines(opts, lines, &mut bytes)?;
+    Ok(bytes)
+}

libs/utils/src/sync/spsc_fold.rs

@@ -96,7 +96,11 @@ impl<T: Send> Sender<T> {
                     }
                 }
                 State::SenderWaitsForReceiverToConsume(_data) => {
-                    // Really, we shouldn't be polled until receiver has consumed and wakes us.
+                    // SAFETY: send is single threaded due to `&mut self` requirement,
+                    // therefore register is not concurrent.
+                    unsafe {
+                        self.state.wake_sender.register(cx.waker());
+                    }
                     Poll::Pending
                 }
                 State::ReceiverGone => Poll::Ready(Err(SendError::ReceiverGone)),
@@ -449,4 +453,38 @@ mod tests {
         let err = recv_task.await.unwrap().expect_err("should error");
         assert!(matches!(err, RecvError::SenderGone));
     }
+
+    #[tokio::test(start_paused = true)]
+    async fn test_receiver_drop_while_waiting_for_receiver_to_consume_unblocks_sender() {
+        let (mut sender, receiver) = channel();
+
+        let state = receiver.state.clone();
+
+        sender.send((), |_, _| unreachable!()).await.unwrap();
+
+        assert!(matches!(&*state.value.lock().unwrap(), &State::HasData(_)));
+
+        let unmergeable = sender.send((), |_, _| Err(()));
+        let mut unmergeable = std::pin::pin!(unmergeable);
+        tokio::select! {
+            _ = tokio::time::sleep(FOREVER) => {},
+            _ = &mut unmergeable => {
+                panic!("unmergeable should not complete");
+            },
+        }
+
+        assert!(matches!(
+            &*state.value.lock().unwrap(),
+            &State::SenderWaitsForReceiverToConsume(_)
+        ));
+
+        drop(receiver);
+
+        assert!(matches!(
+            &*state.value.lock().unwrap(),
+            &State::ReceiverGone
+        ));
+
+        unmergeable.await.unwrap_err();
+    }
 }

libs/wal_decoder/src/models.rs

@@ -95,6 +95,14 @@ impl InterpretedWalRecord {
             && self.metadata_record.is_none()
             && matches!(self.flush_uncommitted, FlushUncommittedRecords::No)
     }
+
+    /// Checks if the WAL record is observed (i.e. contains only metadata
+    /// for observed values)
+    pub fn is_observed(&self) -> bool {
+        self.batch.is_observed()
+            && self.metadata_record.is_none()
+            && matches!(self.flush_uncommitted, FlushUncommittedRecords::No)
+    }
 }
 
 /// The interpreted part of the Postgres WAL record which requires metadata

libs/wal_decoder/src/serialized_batch.rs

@@ -501,6 +501,11 @@ impl SerializedValueBatch {
         !self.has_data() && self.metadata.is_empty()
     }
 
+    /// Checks if the batch contains only observed values
+    pub fn is_observed(&self) -> bool {
+        !self.has_data() && !self.metadata.is_empty()
+    }
+
     /// Checks if the batch contains data
     ///
     /// Note that if this returns false, it may still contain observed values or

pageserver/client/src/page_service.rs

@@ -60,7 +60,7 @@ impl Client {
     ) -> anyhow::Result<PagestreamClient> {
         let copy_both: tokio_postgres::CopyBothDuplex<bytes::Bytes> = self
             .client
-            .copy_both_simple(&format!("pagestream_v2 {tenant_id} {timeline_id}"))
+            .copy_both_simple(&format!("pagestream_v3 {tenant_id} {timeline_id}"))
             .await?;
         let Client {
             cancel_on_client_drop,

pageserver/pagebench/src/cmd/getpage_latest_lsn.rs

@@ -2,7 +2,7 @@ use anyhow::Context;
 use camino::Utf8PathBuf;
 use pageserver_api::key::Key;
 use pageserver_api::keyspace::KeySpaceAccum;
-use pageserver_api::models::PagestreamGetPageRequest;
+use pageserver_api::models::{PagestreamGetPageRequest, PagestreamRequest};
 
 use pageserver_api::shard::TenantShardId;
 use tokio_util::sync::CancellationToken;
@@ -322,12 +322,15 @@ async fn main_impl(
                         .to_rel_block()
                         .expect("we filter non-rel-block keys out above");
                     PagestreamGetPageRequest {
-                        request_lsn: if rng.gen_bool(args.req_latest_probability) {
-                            Lsn::MAX
-                        } else {
-                            r.timeline_lsn
+                        hdr: PagestreamRequest {
+                            reqid: 0,
+                            request_lsn: if rng.gen_bool(args.req_latest_probability) {
+                                Lsn::MAX
+                            } else {
+                                r.timeline_lsn
+                            },
+                            not_modified_since: r.timeline_lsn,
                         },
-                        not_modified_since: r.timeline_lsn,
                         rel: rel_tag,
                         blkno: block_no,
                     }

pageserver/src/http/routes.rs

@@ -2039,6 +2039,26 @@ async fn timeline_cancel_compact_handler(
     .await
 }
 
+// Get compact info of a timeline
+async fn timeline_compact_info_handler(
+    request: Request<Body>,
+    _cancel: CancellationToken,
+) -> Result<Response<Body>, ApiError> {
+    let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
+    let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
+    check_permission(&request, Some(tenant_shard_id.tenant_id))?;
+    let state = get_state(&request);
+    async {
+        let tenant = state
+            .tenant_manager
+            .get_attached_tenant_shard(tenant_shard_id)?;
+        let resp = tenant.get_scheduled_compaction_tasks(timeline_id);
+        json_response(StatusCode::OK, resp)
+    }
+    .instrument(info_span!("timeline_compact_info", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
+    .await
+}
+
 // Run compaction immediately on given timeline.
 async fn timeline_compact_handler(
     mut request: Request<Body>,
@@ -3400,6 +3420,10 @@ pub fn make_router(
             "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/do_gc",
             |r| api_handler(r, timeline_gc_handler),
         )
+        .get(
+            "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/compact",
+            |r| api_handler(r, timeline_compact_info_handler),
+        )
         .put(
             "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/compact",
             |r| api_handler(r, timeline_compact_handler),

pageserver/src/metrics.rs

@@ -3,7 +3,7 @@ use metrics::{
     register_counter_vec, register_gauge_vec, register_histogram, register_histogram_vec,
     register_int_counter, register_int_counter_pair_vec, register_int_counter_vec,
     register_int_gauge, register_int_gauge_vec, register_uint_gauge, register_uint_gauge_vec,
-    Counter, CounterVec, GaugeVec, Histogram, HistogramVec, IntCounter, IntCounterPair,
+    Counter, CounterVec, Gauge, GaugeVec, Histogram, HistogramVec, IntCounter, IntCounterPair,
     IntCounterPairVec, IntCounterVec, IntGauge, IntGaugeVec, UIntGauge, UIntGaugeVec,
 };
 use once_cell::sync::Lazy;
@@ -91,15 +91,6 @@ pub(crate) static STORAGE_TIME_GLOBAL: Lazy<HistogramVec> = Lazy::new(|| {
     .expect("failed to define a metric")
 });
 
-pub(crate) static READ_NUM_LAYERS_VISITED: Lazy<Histogram> = Lazy::new(|| {
-    register_histogram!(
-        "pageserver_layers_visited_per_read_global",
-        "Number of layers visited to reconstruct one key",
-        vec![1.0, 4.0, 8.0, 16.0, 32.0, 64.0, 128.0, 256.0, 512.0, 1024.0],
-    )
-    .expect("failed to define a metric")
-});
-
 pub(crate) static VEC_READ_NUM_LAYERS_VISITED: Lazy<Histogram> = Lazy::new(|| {
     register_histogram!(
         "pageserver_layers_visited_per_vectored_read_global",
@@ -445,6 +436,15 @@ pub(crate) static WAIT_LSN_TIME: Lazy<Histogram> = Lazy::new(|| {
     .expect("failed to define a metric")
 });
 
+static FLUSH_WAIT_UPLOAD_TIME: Lazy<GaugeVec> = Lazy::new(|| {
+    register_gauge_vec!(
+        "pageserver_flush_wait_upload_seconds",
+        "Time spent waiting for preceding uploads during layer flush",
+        &["tenant_id", "shard_id", "timeline_id"]
+    )
+    .expect("failed to define a metric")
+});
+
 static LAST_RECORD_LSN: Lazy<IntGaugeVec> = Lazy::new(|| {
     register_int_gauge_vec!(
         "pageserver_last_record_lsn",
@@ -1845,6 +1845,7 @@ pub(crate) static LIVE_CONNECTIONS: Lazy<IntCounterPairVec> = Lazy::new(|| {
 
 #[derive(Clone, Copy, enum_map::Enum, IntoStaticStr)]
 pub(crate) enum ComputeCommandKind {
+    PageStreamV3,
     PageStreamV2,
     Basebackup,
     Fullbackup,
@@ -2328,13 +2329,15 @@ macro_rules! redo_bytes_histogram_count_buckets {
 pub(crate) struct WalIngestMetrics {
     pub(crate) bytes_received: IntCounter,
     pub(crate) records_received: IntCounter,
+    pub(crate) records_observed: IntCounter,
     pub(crate) records_committed: IntCounter,
     pub(crate) records_filtered: IntCounter,
     pub(crate) gap_blocks_zeroed_on_rel_extend: IntCounter,
     pub(crate) clear_vm_bits_unknown: IntCounterVec,
 }
 
-pub(crate) static WAL_INGEST: Lazy<WalIngestMetrics> = Lazy::new(|| WalIngestMetrics {
+pub(crate) static WAL_INGEST: Lazy<WalIngestMetrics> = Lazy::new(|| {
+    WalIngestMetrics {
     bytes_received: register_int_counter!(
         "pageserver_wal_ingest_bytes_received",
         "Bytes of WAL ingested from safekeepers",
@@ -2345,6 +2348,11 @@ pub(crate) static WAL_INGEST: Lazy<WalIngestMetrics> = Lazy::new(|| WalIngestMet
         "Number of WAL records received from safekeepers"
     )
     .expect("failed to define a metric"),
+    records_observed: register_int_counter!(
+        "pageserver_wal_ingest_records_observed",
+        "Number of WAL records observed from safekeepers. These are metadata only records for shard 0."
+    )
+    .expect("failed to define a metric"),
     records_committed: register_int_counter!(
         "pageserver_wal_ingest_records_committed",
         "Number of WAL records which resulted in writes to pageserver storage"
@@ -2366,6 +2374,7 @@ pub(crate) static WAL_INGEST: Lazy<WalIngestMetrics> = Lazy::new(|| WalIngestMet
         &["entity"],
     )
     .expect("failed to define a metric"),
+}
 });
 
 pub(crate) static PAGESERVER_TIMELINE_WAL_RECORDS_RECEIVED: Lazy<IntCounterVec> = Lazy::new(|| {
@@ -2577,6 +2586,7 @@ pub(crate) struct TimelineMetrics {
     shard_id: String,
     timeline_id: String,
     pub flush_time_histo: StorageTimeMetrics,
+    pub flush_wait_upload_time_gauge: Gauge,
     pub compact_time_histo: StorageTimeMetrics,
     pub create_images_time_histo: StorageTimeMetrics,
     pub logical_size_histo: StorageTimeMetrics,
@@ -2622,6 +2632,9 @@ impl TimelineMetrics {
             &shard_id,
             &timeline_id,
         );
+        let flush_wait_upload_time_gauge = FLUSH_WAIT_UPLOAD_TIME
+            .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
+            .unwrap();
         let compact_time_histo = StorageTimeMetrics::new(
             StorageTimeOperation::Compact,
             &tenant_id,
@@ -2767,6 +2780,7 @@ impl TimelineMetrics {
             shard_id,
             timeline_id,
             flush_time_histo,
+            flush_wait_upload_time_gauge,
             compact_time_histo,
             create_images_time_histo,
             logical_size_histo,
@@ -2816,6 +2830,14 @@ impl TimelineMetrics {
         self.resident_physical_size_gauge.get()
     }
 
+    pub(crate) fn flush_wait_upload_time_gauge_add(&self, duration: f64) {
+        self.flush_wait_upload_time_gauge.add(duration);
+        crate::metrics::FLUSH_WAIT_UPLOAD_TIME
+            .get_metric_with_label_values(&[&self.tenant_id, &self.shard_id, &self.timeline_id])
+            .unwrap()
+            .add(duration);
+    }
+
     pub(crate) fn shutdown(&self) {
         let was_shutdown = self
             .shutdown
@@ -2833,6 +2855,7 @@ impl TimelineMetrics {
         let shard_id = &self.shard_id;
         let _ = LAST_RECORD_LSN.remove_label_values(&[tenant_id, shard_id, timeline_id]);
         let _ = DISK_CONSISTENT_LSN.remove_label_values(&[tenant_id, shard_id, timeline_id]);
+        let _ = FLUSH_WAIT_UPLOAD_TIME.remove_label_values(&[tenant_id, shard_id, timeline_id]);
         let _ = STANDBY_HORIZON.remove_label_values(&[tenant_id, shard_id, timeline_id]);
         {
             RESIDENT_PHYSICAL_SIZE_GLOBAL.sub(self.resident_physical_size_get());
@@ -3862,7 +3885,6 @@ pub fn preinitialize_metrics(conf: &'static PageServerConf) {
 
     // histograms
     [
-        &READ_NUM_LAYERS_VISITED,
         &VEC_READ_NUM_LAYERS_VISITED,
         &WAIT_LSN_TIME,
         &WAL_REDO_TIME,

pageserver/src/page_service.rs

@@ -17,7 +17,7 @@ use pageserver_api::models::{
     PagestreamErrorResponse, PagestreamExistsRequest, PagestreamExistsResponse,
     PagestreamFeMessage, PagestreamGetPageRequest, PagestreamGetSlruSegmentRequest,
     PagestreamGetSlruSegmentResponse, PagestreamNblocksRequest, PagestreamNblocksResponse,
-    PagestreamProtocolVersion,
+    PagestreamProtocolVersion, PagestreamRequest,
 };
 use pageserver_api::shard::TenantShardId;
 use postgres_backend::{
@@ -67,7 +67,7 @@ use crate::tenant::PageReconstructError;
 use crate::tenant::Timeline;
 use crate::{basebackup, timed_after_cancellation};
 use pageserver_api::key::rel_block_to_key;
-use pageserver_api::reltag::{BlockNumber, RelTag, SlruKind};
+use pageserver_api::reltag::SlruKind;
 use postgres_ffi::pg_constants::DEFAULTTABLESPACE_OID;
 use postgres_ffi::BLCKSZ;
 
@@ -537,6 +537,23 @@ impl From<WaitLsnError> for QueryError {
     }
 }
 
+#[derive(thiserror::Error, Debug)]
+struct BatchedPageStreamError {
+    req: PagestreamRequest,
+    err: PageStreamError,
+}
+
+impl std::fmt::Display for BatchedPageStreamError {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        self.err.fmt(f)
+    }
+}
+
+struct BatchedGetPageRequest {
+    req: PagestreamGetPageRequest,
+    timer: SmgrOpTimer,
+}
+
 enum BatchedFeMessage {
     Exists {
         span: Span,
@@ -554,7 +571,7 @@ enum BatchedFeMessage {
         span: Span,
         shard: timeline::handle::Handle<TenantManagerTypes>,
         effective_request_lsn: Lsn,
-        pages: smallvec::SmallVec<[(RelTag, BlockNumber, SmgrOpTimer); 1]>,
+        pages: smallvec::SmallVec<[BatchedGetPageRequest; 1]>,
     },
     DbSize {
         span: Span,
@@ -570,7 +587,7 @@ enum BatchedFeMessage {
     },
     RespondError {
         span: Span,
-        error: PageStreamError,
+        error: BatchedPageStreamError,
     },
 }
 
@@ -595,12 +612,15 @@ impl BatchedFeMessage {
             BatchedFeMessage::GetPage { shard, pages, .. } => (
                 shard,
                 pages.len(),
-                itertools::Either::Right(pages.iter_mut().map(|(_, _, timer)| timer)),
+                itertools::Either::Right(pages.iter_mut().map(|p| &mut p.timer)),
             ),
             BatchedFeMessage::RespondError { .. } => return Ok(()),
         };
         let throttled = tokio::select! {
             throttled = shard.pagestream_throttle.throttle(tokens) => { throttled }
+            _ = shard.cancel.cancelled() => {
+                return Err(QueryError::Shutdown);
+            }
             _ = cancel.cancelled() => {
                 return Err(QueryError::Shutdown);
             }
@@ -654,6 +674,7 @@ impl PageServerHandler {
         )
     }
 
+    #[allow(clippy::too_many_arguments)]
     async fn pagestream_read_message<IO>(
         pgb: &mut PostgresBackendReader<IO>,
         tenant_id: TenantId,
@@ -661,6 +682,7 @@ impl PageServerHandler {
         timeline_handles: &mut TimelineHandles,
         cancel: &CancellationToken,
         ctx: &RequestContext,
+        protocol_version: PagestreamProtocolVersion,
         parent_span: Span,
     ) -> Result<Option<BatchedFeMessage>, QueryError>
     where
@@ -695,11 +717,12 @@ impl PageServerHandler {
         fail::fail_point!("ps::handle-pagerequest-message");
 
         // parse request
-        let neon_fe_msg = PagestreamFeMessage::parse(&mut copy_data_bytes.reader())?;
+        let neon_fe_msg =
+            PagestreamFeMessage::parse(&mut copy_data_bytes.reader(), protocol_version)?;
 
         let batched_msg = match neon_fe_msg {
             PagestreamFeMessage::Exists(req) => {
-                let span = tracing::info_span!(parent: parent_span, "handle_get_rel_exists_request", rel = %req.rel, req_lsn = %req.request_lsn);
+                let span = tracing::info_span!(parent: parent_span, "handle_get_rel_exists_request", rel = %req.rel, req_lsn = %req.hdr.request_lsn);
                 let shard = timeline_handles
                     .get(tenant_id, timeline_id, ShardSelector::Zero)
                     .instrument(span.clone()) // sets `shard_id` field
@@ -715,7 +738,7 @@ impl PageServerHandler {
                 }
             }
             PagestreamFeMessage::Nblocks(req) => {
-                let span = tracing::info_span!(parent: parent_span, "handle_get_nblocks_request", rel = %req.rel, req_lsn = %req.request_lsn);
+                let span = tracing::info_span!(parent: parent_span, "handle_get_nblocks_request", rel = %req.rel, req_lsn = %req.hdr.request_lsn);
                 let shard = timeline_handles
                     .get(tenant_id, timeline_id, ShardSelector::Zero)
                     .instrument(span.clone()) // sets `shard_id` field
@@ -731,7 +754,7 @@ impl PageServerHandler {
                 }
             }
             PagestreamFeMessage::DbSize(req) => {
-                let span = tracing::info_span!(parent: parent_span, "handle_db_size_request", dbnode = %req.dbnode, req_lsn = %req.request_lsn);
+                let span = tracing::info_span!(parent: parent_span, "handle_db_size_request", dbnode = %req.dbnode, req_lsn = %req.hdr.request_lsn);
                 let shard = timeline_handles
                     .get(tenant_id, timeline_id, ShardSelector::Zero)
                     .instrument(span.clone()) // sets `shard_id` field
@@ -747,7 +770,7 @@ impl PageServerHandler {
                 }
             }
             PagestreamFeMessage::GetSlruSegment(req) => {
-                let span = tracing::info_span!(parent: parent_span, "handle_get_slru_segment_request", kind = %req.kind, segno = %req.segno, req_lsn = %req.request_lsn);
+                let span = tracing::info_span!(parent: parent_span, "handle_get_slru_segment_request", kind = %req.kind, segno = %req.segno, req_lsn = %req.hdr.request_lsn);
                 let shard = timeline_handles
                     .get(tenant_id, timeline_id, ShardSelector::Zero)
                     .instrument(span.clone()) // sets `shard_id` field
@@ -762,25 +785,23 @@ impl PageServerHandler {
                     req,
                 }
             }
-            PagestreamFeMessage::GetPage(PagestreamGetPageRequest {
-                request_lsn,
-                not_modified_since,
-                rel,
-                blkno,
-            }) => {
-                let span = tracing::info_span!(parent: parent_span, "handle_get_page_at_lsn_request_batched", req_lsn = %request_lsn);
+            PagestreamFeMessage::GetPage(req) => {
+                let span = tracing::info_span!(parent: parent_span, "handle_get_page_at_lsn_request_batched", req_lsn = %req.hdr.request_lsn);
 
                 macro_rules! respond_error {
                     ($error:expr) => {{
                         let error = BatchedFeMessage::RespondError {
                             span,
-                            error: $error,
+                            error: BatchedPageStreamError {
+                                req: req.hdr,
+                                err: $error,
+                            },
                         };
                         Ok(Some(error))
                     }};
                 }
 
-                let key = rel_block_to_key(rel, blkno);
+                let key = rel_block_to_key(req.rel, req.blkno);
                 let shard = match timeline_handles
                     .get(tenant_id, timeline_id, ShardSelector::Page(key))
                     .instrument(span.clone()) // sets `shard_id` field
@@ -814,8 +835,8 @@ impl PageServerHandler {
 
                 let effective_request_lsn = match Self::wait_or_get_last_lsn(
                     &shard,
-                    request_lsn,
-                    not_modified_since,
+                    req.hdr.request_lsn,
+                    req.hdr.not_modified_since,
                     &shard.get_latest_gc_cutoff_lsn(),
                     ctx,
                 )
@@ -831,7 +852,7 @@ impl PageServerHandler {
                     span,
                     shard,
                     effective_request_lsn,
-                    pages: smallvec::smallvec![(rel, blkno, timer)],
+                    pages: smallvec::smallvec![BatchedGetPageRequest { req, timer }],
                 }
             }
         };
@@ -910,6 +931,7 @@ impl PageServerHandler {
         pgb_writer: &mut PostgresBackend<IO>,
         batch: BatchedFeMessage,
         cancel: &CancellationToken,
+        protocol_version: PagestreamProtocolVersion,
         ctx: &RequestContext,
     ) -> Result<(), QueryError>
     where
@@ -917,7 +939,7 @@ impl PageServerHandler {
     {
         // invoke handler function
         let (handler_results, span): (
-            Vec<Result<(PagestreamBeMessage, SmgrOpTimer), PageStreamError>>,
+            Vec<Result<(PagestreamBeMessage, SmgrOpTimer), BatchedPageStreamError>>,
             _,
         ) = match batch {
             BatchedFeMessage::Exists {
@@ -932,7 +954,8 @@ impl PageServerHandler {
                         .handle_get_rel_exists_request(&shard, &req, ctx)
                         .instrument(span.clone())
                         .await
-                        .map(|msg| (msg, timer))],
+                        .map(|msg| (msg, timer))
+                        .map_err(|err| BatchedPageStreamError { err, req: req.hdr })],
                     span,
                 )
             }
@@ -948,7 +971,8 @@ impl PageServerHandler {
                         .handle_get_nblocks_request(&shard, &req, ctx)
                         .instrument(span.clone())
                         .await
-                        .map(|msg| (msg, timer))],
+                        .map(|msg| (msg, timer))
+                        .map_err(|err| BatchedPageStreamError { err, req: req.hdr })],
                     span,
                 )
             }
@@ -990,7 +1014,8 @@ impl PageServerHandler {
                         .handle_db_size_request(&shard, &req, ctx)
                         .instrument(span.clone())
                         .await
-                        .map(|msg| (msg, timer))],
+                        .map(|msg| (msg, timer))
+                        .map_err(|err| BatchedPageStreamError { err, req: req.hdr })],
                     span,
                 )
             }
@@ -1006,7 +1031,8 @@ impl PageServerHandler {
                         .handle_get_slru_segment_request(&shard, &req, ctx)
                         .instrument(span.clone())
                         .await
-                        .map(|msg| (msg, timer))],
+                        .map(|msg| (msg, timer))
+                        .map_err(|err| BatchedPageStreamError { err, req: req.hdr })],
                     span,
                 )
             }
@@ -1022,7 +1048,7 @@ impl PageServerHandler {
         // Other handler errors are sent back as an error message and we stay in pagestream protocol.
         for handler_result in handler_results {
             let (response_msg, timer) = match handler_result {
-                Err(e) => match &e {
+                Err(e) => match &e.err {
                     PageStreamError::Shutdown => {
                         // If we fail to fulfil a request during shutdown, which may be _because_ of
                         // shutdown, then do not send the error to the client.  Instead just drop the
@@ -1041,13 +1067,14 @@ impl PageServerHandler {
                         // print the all details to the log with {:#}, but for the client the
                         // error message is enough.  Do not log if shutting down, as the anyhow::Error
                         // here includes cancellation which is not an error.
-                        let full = utils::error::report_compact_sources(&e);
+                        let full = utils::error::report_compact_sources(&e.err);
                         span.in_scope(|| {
                             error!("error reading relation or page version: {full:#}")
                         });
                         (
                             PagestreamBeMessage::Error(PagestreamErrorResponse {
-                                message: e.to_string(),
+                                req: e.req,
+                                message: e.err.to_string(),
                             }),
                             None, // TODO: measure errors
                         )
@@ -1060,7 +1087,9 @@ impl PageServerHandler {
             // marshal & transmit response message
             //
 
-            pgb_writer.write_message_noflush(&BeMessage::CopyData(&response_msg.serialize()))?;
+            pgb_writer.write_message_noflush(&BeMessage::CopyData(
+                &response_msg.serialize(protocol_version),
+            ))?;
 
             // We purposefully don't count flush time into the timer.
             //
@@ -1123,7 +1152,7 @@ impl PageServerHandler {
         pgb: &mut PostgresBackend<IO>,
         tenant_id: TenantId,
         timeline_id: TimelineId,
-        _protocol_version: PagestreamProtocolVersion,
+        protocol_version: PagestreamProtocolVersion,
         ctx: RequestContext,
     ) -> Result<(), QueryError>
     where
@@ -1163,6 +1192,7 @@ impl PageServerHandler {
                     timeline_handles,
                     request_span,
                     pipelining_config,
+                    protocol_version,
                     &ctx,
                 )
                 .await
@@ -1175,6 +1205,7 @@ impl PageServerHandler {
                     timeline_id,
                     timeline_handles,
                     request_span,
+                    protocol_version,
                     &ctx,
                 )
                 .await
@@ -1201,6 +1232,7 @@ impl PageServerHandler {
         timeline_id: TimelineId,
         mut timeline_handles: TimelineHandles,
         request_span: Span,
+        protocol_version: PagestreamProtocolVersion,
         ctx: &RequestContext,
     ) -> (
         (PostgresBackendReader<IO>, TimelineHandles),
@@ -1218,6 +1250,7 @@ impl PageServerHandler {
                 &mut timeline_handles,
                 &cancel,
                 ctx,
+                protocol_version,
                 request_span.clone(),
             )
             .await;
@@ -1238,7 +1271,7 @@ impl PageServerHandler {
             }
 
             let err = self
-                .pagesteam_handle_batched_message(pgb_writer, msg, &cancel, ctx)
+                .pagesteam_handle_batched_message(pgb_writer, msg, &cancel, protocol_version, ctx)
                 .await;
             match err {
                 Ok(()) => {}
@@ -1261,6 +1294,7 @@ impl PageServerHandler {
         mut timeline_handles: TimelineHandles,
         request_span: Span,
         pipelining_config: PageServicePipeliningConfigPipelined,
+        protocol_version: PagestreamProtocolVersion,
         ctx: &RequestContext,
     ) -> (
         (PostgresBackendReader<IO>, TimelineHandles),
@@ -1358,6 +1392,7 @@ impl PageServerHandler {
                         &mut timeline_handles,
                         &cancel_batcher,
                         &ctx,
+                        protocol_version,
                         request_span.clone(),
                     )
                     .await;
@@ -1403,8 +1438,14 @@ impl PageServerHandler {
                     batch
                         .throttle_and_record_start_processing(&self.cancel)
                         .await?;
-                    self.pagesteam_handle_batched_message(pgb_writer, batch, &cancel, &ctx)
-                        .await?;
+                    self.pagesteam_handle_batched_message(
+                        pgb_writer,
+                        batch,
+                        &cancel,
+                        protocol_version,
+                        &ctx,
+                    )
+                    .await?;
                 }
             }
         });
@@ -1578,8 +1619,8 @@ impl PageServerHandler {
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
         let lsn = Self::wait_or_get_last_lsn(
             timeline,
-            req.request_lsn,
-            req.not_modified_since,
+            req.hdr.request_lsn,
+            req.hdr.not_modified_since,
             &latest_gc_cutoff_lsn,
             ctx,
         )
@@ -1590,6 +1631,7 @@ impl PageServerHandler {
             .await?;
 
         Ok(PagestreamBeMessage::Exists(PagestreamExistsResponse {
+            req: *req,
             exists,
         }))
     }
@@ -1604,8 +1646,8 @@ impl PageServerHandler {
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
         let lsn = Self::wait_or_get_last_lsn(
             timeline,
-            req.request_lsn,
-            req.not_modified_since,
+            req.hdr.request_lsn,
+            req.hdr.not_modified_since,
             &latest_gc_cutoff_lsn,
             ctx,
         )
@@ -1616,6 +1658,7 @@ impl PageServerHandler {
             .await?;
 
         Ok(PagestreamBeMessage::Nblocks(PagestreamNblocksResponse {
+            req: *req,
             n_blocks,
         }))
     }
@@ -1630,8 +1673,8 @@ impl PageServerHandler {
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
         let lsn = Self::wait_or_get_last_lsn(
             timeline,
-            req.request_lsn,
-            req.not_modified_since,
+            req.hdr.request_lsn,
+            req.hdr.not_modified_since,
             &latest_gc_cutoff_lsn,
             ctx,
         )
@@ -1643,6 +1686,7 @@ impl PageServerHandler {
         let db_size = total_blocks as i64 * BLCKSZ as i64;
 
         Ok(PagestreamBeMessage::DbSize(PagestreamDbSizeResponse {
+            req: *req,
             db_size,
         }))
     }
@@ -1652,9 +1696,9 @@ impl PageServerHandler {
         &mut self,
         timeline: &Timeline,
         effective_lsn: Lsn,
-        requests: smallvec::SmallVec<[(RelTag, BlockNumber, SmgrOpTimer); 1]>,
+        requests: smallvec::SmallVec<[BatchedGetPageRequest; 1]>,
         ctx: &RequestContext,
-    ) -> Vec<Result<(PagestreamBeMessage, SmgrOpTimer), PageStreamError>> {
+    ) -> Vec<Result<(PagestreamBeMessage, SmgrOpTimer), BatchedPageStreamError>> {
         debug_assert_current_span_has_tenant_and_timeline_id();
 
         timeline
@@ -1663,7 +1707,7 @@ impl PageServerHandler {
 
         let results = timeline
             .get_rel_page_at_lsn_batched(
-                requests.iter().map(|(reltag, blkno, _)| (reltag, blkno)),
+                requests.iter().map(|p| (&p.req.rel, &p.req.blkno)),
                 effective_lsn,
                 ctx,
             )
@@ -1675,16 +1719,20 @@ impl PageServerHandler {
             requests
                 .into_iter()
                 .zip(results.into_iter())
-                .map(|((_, _, timer), res)| {
+                .map(|(req, res)| {
                     res.map(|page| {
                         (
                             PagestreamBeMessage::GetPage(models::PagestreamGetPageResponse {
+                                req: req.req,
                                 page,
                             }),
-                            timer,
+                            req.timer,
                         )
                     })
-                    .map_err(PageStreamError::from)
+                    .map_err(|e| BatchedPageStreamError {
+                        err: PageStreamError::from(e),
+                        req: req.req.hdr,
+                    })
                 }),
         )
     }
@@ -1699,8 +1747,8 @@ impl PageServerHandler {
         let latest_gc_cutoff_lsn = timeline.get_latest_gc_cutoff_lsn();
         let lsn = Self::wait_or_get_last_lsn(
             timeline,
-            req.request_lsn,
-            req.not_modified_since,
+            req.hdr.request_lsn,
+            req.hdr.not_modified_since,
             &latest_gc_cutoff_lsn,
             ctx,
         )
@@ -1711,7 +1759,7 @@ impl PageServerHandler {
         let segment = timeline.get_slru_segment(kind, req.segno, lsn, ctx).await?;
 
         Ok(PagestreamBeMessage::GetSlruSegment(
-            PagestreamGetSlruSegmentResponse { segment },
+            PagestreamGetSlruSegmentResponse { req: *req, segment },
         ))
     }
 
@@ -1906,6 +1954,7 @@ struct FullBackupCmd {
 struct PageStreamCmd {
     tenant_id: TenantId,
     timeline_id: TimelineId,
+    protocol_version: PagestreamProtocolVersion,
 }
 
 /// `lease lsn tenant timeline lsn`
@@ -1926,7 +1975,7 @@ enum PageServiceCmd {
 }
 
 impl PageStreamCmd {
-    fn parse(query: &str) -> anyhow::Result<Self> {
+    fn parse(query: &str, protocol_version: PagestreamProtocolVersion) -> anyhow::Result<Self> {
         let parameters = query.split_whitespace().collect_vec();
         if parameters.len() != 2 {
             bail!(
@@ -1941,6 +1990,7 @@ impl PageStreamCmd {
         Ok(Self {
             tenant_id,
             timeline_id,
+            protocol_version,
         })
     }
 }
@@ -2078,7 +2128,14 @@ impl PageServiceCmd {
             bail!("cannot parse query: {query}")
         };
         match cmd.to_ascii_lowercase().as_str() {
-            "pagestream_v2" => Ok(Self::PageStream(PageStreamCmd::parse(other)?)),
+            "pagestream_v2" => Ok(Self::PageStream(PageStreamCmd::parse(
+                other,
+                PagestreamProtocolVersion::V2,
+            )?)),
+            "pagestream_v3" => Ok(Self::PageStream(PageStreamCmd::parse(
+                other,
+                PagestreamProtocolVersion::V3,
+            )?)),
             "basebackup" => Ok(Self::BaseBackup(BaseBackupCmd::parse(other)?)),
             "fullbackup" => Ok(Self::FullBackup(FullBackupCmd::parse(other)?)),
             "lease" => {
@@ -2160,25 +2217,21 @@ where
             PageServiceCmd::PageStream(PageStreamCmd {
                 tenant_id,
                 timeline_id,
+                protocol_version,
             }) => {
                 tracing::Span::current()
                     .record("tenant_id", field::display(tenant_id))
                     .record("timeline_id", field::display(timeline_id));
 
                 self.check_permission(Some(tenant_id))?;
+                let command_kind = match protocol_version {
+                    PagestreamProtocolVersion::V2 => ComputeCommandKind::PageStreamV2,
+                    PagestreamProtocolVersion::V3 => ComputeCommandKind::PageStreamV3,
+                };
+                COMPUTE_COMMANDS_COUNTERS.for_command(command_kind).inc();
 
-                COMPUTE_COMMANDS_COUNTERS
-                    .for_command(ComputeCommandKind::PageStreamV2)
-                    .inc();
-
-                self.handle_pagerequests(
-                    pgb,
-                    tenant_id,
-                    timeline_id,
-                    PagestreamProtocolVersion::V2,
-                    ctx,
-                )
-                .await?;
+                self.handle_pagerequests(pgb, tenant_id, timeline_id, protocol_version, ctx)
+                    .await?;
             }
             PageServiceCmd::BaseBackup(BaseBackupCmd {
                 tenant_id,
@@ -2357,7 +2410,8 @@ mod tests {
             cmd,
             PageServiceCmd::PageStream(PageStreamCmd {
                 tenant_id,
-                timeline_id
+                timeline_id,
+                protocol_version: PagestreamProtocolVersion::V2,
             })
         );
         let cmd = PageServiceCmd::parse(&format!("basebackup {tenant_id} {timeline_id}")).unwrap();

pageserver/src/pgdatadir_mapping.rs

@@ -627,7 +627,7 @@ impl Timeline {
             // cannot overflow, high and low are both smaller than u64::MAX / 2
             let mid = (high + low) / 2;
 
-            let cmp = self
+            let cmp = match self
                 .is_latest_commit_timestamp_ge_than(
                     search_timestamp,
                     Lsn(mid * 8),
@@ -635,7 +635,16 @@ impl Timeline {
                     &mut found_larger,
                     ctx,
                 )
-                .await?;
+                .await
+            {
+                Ok(res) => res,
+                Err(PageReconstructError::MissingKey(e)) => {
+                    warn!("Missing key while find_lsn_for_timestamp. Either we might have already garbage-collected that data or the key is really missing. Last error: {:#}", e);
+                    // Return that we didn't find any requests smaller than the LSN, and logging the error.
+                    return Ok(LsnForTimestamp::Past(min_lsn));
+                }
+                Err(e) => return Err(e),
+            };
 
             if cmp {
                 high = mid;
@@ -643,6 +652,7 @@ impl Timeline {
                 low = mid + 1;
             }
         }
+
         // If `found_smaller == true`, `low = t + 1` where `t` is the target LSN,
         // so the LSN of the last commit record before or at `search_timestamp`.
         // Remove one from `low` to get `t`.

pageserver/src/tenant.rs

@@ -21,6 +21,7 @@ use enumset::EnumSet;
 use futures::stream::FuturesUnordered;
 use futures::StreamExt;
 use pageserver_api::models;
+use pageserver_api::models::CompactInfoResponse;
 use pageserver_api::models::LsnLease;
 use pageserver_api::models::TimelineArchivalState;
 use pageserver_api::models::TimelineState;
@@ -37,20 +38,17 @@ use remote_timeline_client::manifest::{
 };
 use remote_timeline_client::UploadQueueNotReadyError;
 use std::collections::BTreeMap;
-use std::collections::VecDeque;
 use std::fmt;
 use std::future::Future;
 use std::sync::atomic::AtomicBool;
 use std::sync::Weak;
 use std::time::SystemTime;
 use storage_broker::BrokerClientChannel;
-use timeline::compaction::GcCompactJob;
-use timeline::compaction::ScheduledCompactionTask;
+use timeline::compaction::GcCompactionQueue;
 use timeline::import_pgdata;
 use timeline::offload::offload_timeline;
-use timeline::CompactFlags;
+use timeline::offload::OffloadError;
 use timeline::CompactOptions;
-use timeline::CompactionError;
 use timeline::ShutdownMode;
 use tokio::io::BufReader;
 use tokio::sync::watch;
@@ -346,10 +344,8 @@ pub struct Tenant {
     /// Overhead of mutex is acceptable because compaction is done with a multi-second period.
     compaction_circuit_breaker: std::sync::Mutex<CircuitBreaker>,
 
-    /// Scheduled compaction tasks. Currently, this can only be populated by triggering
-    /// a manual gc-compaction from the manual compaction API.
-    scheduled_compaction_tasks:
-        std::sync::Mutex<HashMap<TimelineId, VecDeque<ScheduledCompactionTask>>>,
+    /// Scheduled gc-compaction tasks.
+    scheduled_compaction_tasks: std::sync::Mutex<HashMap<TimelineId, Arc<GcCompactionQueue>>>,
 
     /// If the tenant is in Activating state, notify this to encourage it
     /// to proceed to Active as soon as possible, rather than waiting for lazy
@@ -2039,7 +2035,7 @@ impl Tenant {
     ) -> Result<Arc<Timeline>, TimelineArchivalError> {
         info!("unoffloading timeline");
 
-        // We activate the timeline below manually, so this must be called on an active timeline.
+        // We activate the timeline below manually, so this must be called on an active tenant.
         // We expect callers of this function to ensure this.
         match self.current_state() {
             TenantState::Activating { .. }
@@ -2604,9 +2600,15 @@ impl Tenant {
                 WaitCompletionError::NotInitialized(
                     e, // If the queue is already stopped, it's a shutdown error.
                 ) if e.is_stopping() => CreateTimelineError::ShuttingDown,
-                e => CreateTimelineError::Other(e.into()),
-            })
-            .context("wait for timeline initial uploads to complete")?;
+                WaitCompletionError::NotInitialized(_) => {
+                    // This is a bug: we should never try to wait for uploads before initializing the timeline
+                    debug_assert!(false);
+                    CreateTimelineError::Other(anyhow::anyhow!("timeline not initialized"))
+                }
+                WaitCompletionError::UploadQueueShutDownOrStopped => {
+                    CreateTimelineError::ShuttingDown
+                }
+            })?;
 
         // The creating task is responsible for activating the timeline.
         // We do this after `wait_completion()` so that we don't spin up tasks that start
@@ -2990,113 +2992,35 @@ impl Tenant {
                 if has_pending_l0_compaction_task {
                     Some(true)
                 } else {
-                    let mut has_pending_scheduled_compaction_task;
-                    let next_scheduled_compaction_task = {
-                        let mut guard = self.scheduled_compaction_tasks.lock().unwrap();
-                        if let Some(tline_pending_tasks) = guard.get_mut(timeline_id) {
-                            if !tline_pending_tasks.is_empty() {
-                                info!(
-                                    "{} tasks left in the compaction schedule queue",
-                                    tline_pending_tasks.len()
-                                );
-                            }
-                            let next_task = tline_pending_tasks.pop_front();
-                            has_pending_scheduled_compaction_task = !tline_pending_tasks.is_empty();
-                            next_task
-                        } else {
-                            has_pending_scheduled_compaction_task = false;
-                            None
-                        }
+                    let queue = {
+                        let guard = self.scheduled_compaction_tasks.lock().unwrap();
+                        guard.get(timeline_id).cloned()
                     };
-                    if let Some(mut next_scheduled_compaction_task) = next_scheduled_compaction_task
-                    {
-                        if !next_scheduled_compaction_task
-                            .options
-                            .flags
-                            .contains(CompactFlags::EnhancedGcBottomMostCompaction)
-                        {
-                            warn!("ignoring scheduled compaction task: scheduled task must be gc compaction: {:?}", next_scheduled_compaction_task.options);
-                        } else if next_scheduled_compaction_task.options.sub_compaction {
-                            info!("running scheduled enhanced gc bottom-most compaction with sub-compaction, splitting compaction jobs");
-                            let jobs: Vec<GcCompactJob> = timeline
-                                .gc_compaction_split_jobs(
-                                    GcCompactJob::from_compact_options(
-                                        next_scheduled_compaction_task.options.clone(),
-                                    ),
-                                    next_scheduled_compaction_task
-                                        .options
-                                        .sub_compaction_max_job_size_mb,
-                                )
-                                .await
-                                .map_err(CompactionError::Other)?;
-                            if jobs.is_empty() {
-                                info!("no jobs to run, skipping scheduled compaction task");
-                            } else {
-                                has_pending_scheduled_compaction_task = true;
-                                let jobs_len = jobs.len();
-                                let mut guard = self.scheduled_compaction_tasks.lock().unwrap();
-                                let tline_pending_tasks = guard.entry(*timeline_id).or_default();
-                                for (idx, job) in jobs.into_iter().enumerate() {
-                                    // Unfortunately we need to convert the `GcCompactJob` back to `CompactionOptions`
-                                    // until we do further refactors to allow directly call `compact_with_gc`.
-                                    let mut flags: EnumSet<CompactFlags> = EnumSet::default();
-                                    flags |= CompactFlags::EnhancedGcBottomMostCompaction;
-                                    if job.dry_run {
-                                        flags |= CompactFlags::DryRun;
-                                    }
-                                    let options = CompactOptions {
-                                        flags,
-                                        sub_compaction: false,
-                                        compact_key_range: Some(job.compact_key_range.into()),
-                                        compact_lsn_range: Some(job.compact_lsn_range.into()),
-                                        sub_compaction_max_job_size_mb: None,
-                                    };
-                                    tline_pending_tasks.push_back(if idx == jobs_len - 1 {
-                                        ScheduledCompactionTask {
-                                            options,
-                                            // The last job in the queue sends the signal and releases the gc guard
-                                            result_tx: next_scheduled_compaction_task
-                                                .result_tx
-                                                .take(),
-                                            gc_block: next_scheduled_compaction_task
-                                                .gc_block
-                                                .take(),
-                                        }
-                                    } else {
-                                        ScheduledCompactionTask {
-                                            options,
-                                            result_tx: None,
-                                            gc_block: None,
-                                        }
-                                    });
-                                }
-                                info!("scheduled enhanced gc bottom-most compaction with sub-compaction, split into {} jobs", jobs_len);
-                            }
-                        } else {
-                            let _ = timeline
-                                .compact_with_options(
-                                    cancel,
-                                    next_scheduled_compaction_task.options,
-                                    ctx,
-                                )
-                                .instrument(info_span!("scheduled_compact_timeline", %timeline_id))
-                                .await?;
-                            if let Some(tx) = next_scheduled_compaction_task.result_tx.take() {
-                                // TODO: we can send compaction statistics in the future
-                                tx.send(()).ok();
-                            }
-                        }
+                    if let Some(queue) = queue {
+                        let has_pending_tasks = queue
+                            .iteration(cancel, ctx, &self.gc_block, timeline)
+                            .await?;
+                        Some(has_pending_tasks)
+                    } else {
+                        Some(false)
                     }
-                    Some(has_pending_scheduled_compaction_task)
                 }
             } else {
                 None
             };
             has_pending_task |= pending_task_left.unwrap_or(false);
             if pending_task_left == Some(false) && *can_offload {
-                offload_timeline(self, timeline)
+                pausable_failpoint!("before-timeline-auto-offload");
+                match offload_timeline(self, timeline)
                     .instrument(info_span!("offload_timeline", %timeline_id))
-                    .await?;
+                    .await
+                {
+                    Err(OffloadError::NotArchived) => {
+                        // Ignore this, we likely raced with unarchival
+                        Ok(())
+                    }
+                    other => other,
+                }?;
             }
         }
 
@@ -3109,17 +3033,32 @@ impl Tenant {
     }
 
     /// Cancel scheduled compaction tasks
-    pub(crate) fn cancel_scheduled_compaction(
+    pub(crate) fn cancel_scheduled_compaction(&self, timeline_id: TimelineId) {
+        let mut guard = self.scheduled_compaction_tasks.lock().unwrap();
+        if let Some(q) = guard.get_mut(&timeline_id) {
+            q.cancel_scheduled();
+        }
+    }
+
+    pub(crate) fn get_scheduled_compaction_tasks(
         &self,
         timeline_id: TimelineId,
-    ) -> Vec<ScheduledCompactionTask> {
-        let mut guard = self.scheduled_compaction_tasks.lock().unwrap();
-        if let Some(tline_pending_tasks) = guard.get_mut(&timeline_id) {
-            let current_tline_pending_tasks = std::mem::take(tline_pending_tasks);
-            current_tline_pending_tasks.into_iter().collect()
-        } else {
-            Vec::new()
+    ) -> Vec<CompactInfoResponse> {
+        let res = {
+            let guard = self.scheduled_compaction_tasks.lock().unwrap();
+            guard.get(&timeline_id).map(|q| q.remaining_jobs())
+        };
+        let Some((running, remaining)) = res else {
+            return Vec::new();
+        };
+        let mut result = Vec::new();
+        if let Some((id, running)) = running {
+            result.extend(running.into_compact_info_resp(id, true));
         }
+        for (id, job) in remaining {
+            result.extend(job.into_compact_info_resp(id, false));
+        }
+        result
     }
 
     /// Schedule a compaction task for a timeline.
@@ -3128,20 +3067,12 @@ impl Tenant {
         timeline_id: TimelineId,
         options: CompactOptions,
     ) -> anyhow::Result<tokio::sync::oneshot::Receiver<()>> {
-        let gc_guard = match self.gc_block.start().await {
-            Ok(guard) => guard,
-            Err(e) => {
-                bail!("cannot run gc-compaction because gc is blocked: {}", e);
-            }
-        };
         let (tx, rx) = tokio::sync::oneshot::channel();
         let mut guard = self.scheduled_compaction_tasks.lock().unwrap();
-        let tline_pending_tasks = guard.entry(timeline_id).or_default();
-        tline_pending_tasks.push_back(ScheduledCompactionTask {
-            options,
-            result_tx: Some(tx),
-            gc_block: Some(gc_guard),
-        });
+        let q = guard
+            .entry(timeline_id)
+            .or_insert_with(|| Arc::new(GcCompactionQueue::new()));
+        q.schedule_manual_compaction(options, Some(tx));
         Ok(rx)
     }
 
@@ -4465,13 +4396,17 @@ impl Tenant {
         let mut gc_cutoffs: HashMap<TimelineId, GcCutoffs> =
             HashMap::with_capacity(timelines.len());
 
+        // Ensures all timelines use the same start time when computing the time cutoff.
+        let now_ts_for_pitr_calc = SystemTime::now();
         for timeline in timelines.iter() {
             let cutoff = timeline
                 .get_last_record_lsn()
                 .checked_sub(horizon)
                 .unwrap_or(Lsn(0));
 
-            let cutoffs = timeline.find_gc_cutoffs(cutoff, pitr, cancel, ctx).await?;
+            let cutoffs = timeline
+                .find_gc_cutoffs(now_ts_for_pitr_calc, cutoff, pitr, cancel, ctx)
+                .await?;
             let old = gc_cutoffs.insert(timeline.timeline_id, cutoffs);
             assert!(old.is_none());
         }
@@ -5759,13 +5694,13 @@ mod tests {
     use timeline::{CompactOptions, DeltaLayerTestDesc};
     use utils::id::TenantId;
 
+    #[cfg(feature = "testing")]
+    use models::CompactLsnRange;
     #[cfg(feature = "testing")]
     use pageserver_api::record::NeonWalRecord;
     #[cfg(feature = "testing")]
     use timeline::compaction::{KeyHistoryRetention, KeyLogAtLsn};
     #[cfg(feature = "testing")]
-    use timeline::CompactLsnRange;
-    #[cfg(feature = "testing")]
     use timeline::GcInfo;
 
     static TEST_KEY: Lazy<Key> =
@@ -9634,7 +9569,7 @@ mod tests {
     #[cfg(feature = "testing")]
     #[tokio::test]
     async fn test_simple_bottom_most_compaction_on_branch() -> anyhow::Result<()> {
-        use timeline::CompactLsnRange;
+        use models::CompactLsnRange;
 
         let harness = TenantHarness::create("test_simple_bottom_most_compaction_on_branch").await?;
         let (tenant, ctx) = harness.load().await;

pageserver/src/tenant/checks.rs

@@ -1,12 +1,15 @@
 use std::collections::BTreeSet;
 
 use itertools::Itertools;
+use pageserver_compaction::helpers::overlaps_with;
 
 use super::storage_layer::LayerName;
 
 /// Checks whether a layer map is valid (i.e., is a valid result of the current compaction algorithm if nothing goes wrong).
 ///
-/// The function checks if we can split the LSN range of a delta layer only at the LSNs of the delta layers. For example,
+/// The function implements a fast path check and a slow path check.
+///
+/// The fast path checks if we can split the LSN range of a delta layer only at the LSNs of the delta layers. For example,
 ///
 /// ```plain
 /// |       |                 |       |
@@ -25,31 +28,47 @@ use super::storage_layer::LayerName;
 /// |       |    |   4   |    |       |
 ///
 /// If layer 2 and 4 contain the same single key, this is also a valid layer map.
+///
+/// However, if a partial compaction is still going on, it is possible that we get a layer map not satisfying the above condition.
+/// Therefore, we fallback to simply check if any of the two delta layers overlap. (See "A slow path...")
 pub fn check_valid_layermap(metadata: &[LayerName]) -> Option<String> {
     let mut lsn_split_point = BTreeSet::new(); // TODO: use a better data structure (range tree / range set?)
     let mut all_delta_layers = Vec::new();
     for name in metadata {
         if let LayerName::Delta(layer) = name {
-            if layer.key_range.start.next() != layer.key_range.end {
-                all_delta_layers.push(layer.clone());
-            }
+            all_delta_layers.push(layer.clone());
         }
     }
     for layer in &all_delta_layers {
-        let lsn_range = &layer.lsn_range;
-        lsn_split_point.insert(lsn_range.start);
-        lsn_split_point.insert(lsn_range.end);
+        if layer.key_range.start.next() != layer.key_range.end {
+            let lsn_range = &layer.lsn_range;
+            lsn_split_point.insert(lsn_range.start);
+            lsn_split_point.insert(lsn_range.end);
+        }
     }
-    for layer in &all_delta_layers {
+    for (idx, layer) in all_delta_layers.iter().enumerate() {
+        if layer.key_range.start.next() == layer.key_range.end {
+            continue;
+        }
         let lsn_range = layer.lsn_range.clone();
         let intersects = lsn_split_point.range(lsn_range).collect_vec();
         if intersects.len() > 1 {
-            let err = format!(
-                "layer violates the layer map LSN split assumption: layer {} intersects with LSN [{}]",
-                layer,
-                intersects.into_iter().map(|lsn| lsn.to_string()).join(", ")
-            );
-            return Some(err);
+            // A slow path to check if the layer intersects with any other delta layer.
+            for (other_idx, other_layer) in all_delta_layers.iter().enumerate() {
+                if other_idx == idx {
+                    // do not check self intersects with self
+                    continue;
+                }
+                if overlaps_with(&layer.lsn_range, &other_layer.lsn_range)
+                    && overlaps_with(&layer.key_range, &other_layer.key_range)
+                {
+                    let err = format!(
+                            "layer violates the layer map LSN split assumption: layer {} intersects with layer {}",
+                            layer, other_layer
+                        );
+                    return Some(err);
+                }
+            }
         }
     }
     None

pageserver/src/tenant/remote_timeline_client.rs

@@ -304,6 +304,15 @@ pub enum WaitCompletionError {
 #[derive(Debug, thiserror::Error)]
 #[error("Upload queue either in unexpected state or hasn't downloaded manifest yet")]
 pub struct UploadQueueNotReadyError;
+
+#[derive(Debug, thiserror::Error)]
+pub enum ShutdownIfArchivedError {
+    #[error(transparent)]
+    NotInitialized(NotInitialized),
+    #[error("timeline is not archived")]
+    NotArchived,
+}
+
 /// Behavioral modes that enable seamless live migration.
 ///
 /// See docs/rfcs/028-pageserver-migration.md to understand how these fit in.
@@ -816,6 +825,55 @@ impl RemoteTimelineClient {
         Ok(need_wait)
     }
 
+    /// Shuts the timeline client down, but only if the timeline is archived.
+    ///
+    /// This function and [`Self::schedule_index_upload_for_timeline_archival_state`] use the
+    /// same lock to prevent races between unarchival and offloading: unarchival requires the
+    /// upload queue to be initialized, and leaves behind an upload queue where either dirty
+    /// or clean has archived_at of `None`. offloading leaves behind an uninitialized upload
+    /// queue.
+    pub(crate) async fn shutdown_if_archived(
+        self: &Arc<Self>,
+    ) -> Result<(), ShutdownIfArchivedError> {
+        {
+            let mut guard = self.upload_queue.lock().unwrap();
+            let upload_queue = guard
+                .initialized_mut()
+                .map_err(ShutdownIfArchivedError::NotInitialized)?;
+
+            match (
+                upload_queue.dirty.archived_at.is_none(),
+                upload_queue.clean.0.archived_at.is_none(),
+            ) {
+                // The expected case: the timeline is archived and we don't want to unarchive
+                (false, false) => {}
+                (true, false) => {
+                    tracing::info!("can't shut down timeline: timeline slated for unarchival");
+                    return Err(ShutdownIfArchivedError::NotArchived);
+                }
+                (dirty_archived, true) => {
+                    tracing::info!(%dirty_archived, "can't shut down timeline: timeline not archived in remote storage");
+                    return Err(ShutdownIfArchivedError::NotArchived);
+                }
+            }
+
+            // Set the shutting_down flag while the guard from the archival check is held.
+            // This prevents a race with unarchival, as initialized_mut will not return
+            // an upload queue from this point.
+            // Also launch the queued tasks like shutdown() does.
+            if !upload_queue.shutting_down {
+                upload_queue.shutting_down = true;
+                upload_queue.queued_operations.push_back(UploadOp::Shutdown);
+                // this operation is not counted similar to Barrier
+                self.launch_queued_tasks(upload_queue);
+            }
+        }
+
+        self.shutdown().await;
+
+        Ok(())
+    }
+
     /// Launch an index-file upload operation in the background, setting `import_pgdata` field.
     pub(crate) fn schedule_index_upload_for_import_pgdata_state_update(
         self: &Arc<Self>,
@@ -1943,6 +2001,30 @@ impl RemoteTimelineClient {
                 return;
             }
 
+            // Assert that we don't modify a layer that's referenced by the current index.
+            if cfg!(debug_assertions) {
+                let modified = match &task.op {
+                    UploadOp::UploadLayer(layer, layer_metadata, _) => {
+                        vec![(layer.layer_desc().layer_name(), layer_metadata)]
+                    }
+                    UploadOp::Delete(delete) => {
+                        delete.layers.iter().map(|(n, m)| (n.clone(), m)).collect()
+                    }
+                    // These don't modify layers.
+                    UploadOp::UploadMetadata { .. } => Vec::new(),
+                    UploadOp::Barrier(_) => Vec::new(),
+                    UploadOp::Shutdown => Vec::new(),
+                };
+                if let Ok(queue) = self.upload_queue.lock().unwrap().initialized_mut() {
+                    for (ref name, metadata) in modified {
+                        debug_assert!(
+                            !queue.clean.0.references(name, metadata),
+                            "layer {name} modified while referenced by index",
+                        );
+                    }
+                }
+            }
+
             let upload_result: anyhow::Result<()> = match &task.op {
                 UploadOp::UploadLayer(ref layer, ref layer_metadata, mode) => {
                     if let Some(OpType::FlushDeletion) = mode {
@@ -2509,6 +2591,21 @@ pub fn remote_layer_path(
     RemotePath::from_string(&path).expect("Failed to construct path")
 }
 
+/// Returns true if a and b have the same layer path within a tenant/timeline. This is essentially
+/// remote_layer_path(a) == remote_layer_path(b) without the string allocations.
+///
+/// TODO: there should be a variant of LayerName for the physical path that contains information
+/// about the shard and generation, such that this could be replaced by a simple comparison.
+pub fn is_same_remote_layer_path(
+    aname: &LayerName,
+    ameta: &LayerFileMetadata,
+    bname: &LayerName,
+    bmeta: &LayerFileMetadata,
+) -> bool {
+    // NB: don't assert remote_layer_path(a) == remote_layer_path(b); too expensive even for debug.
+    aname == bname && ameta.shard == bmeta.shard && ameta.generation == bmeta.generation
+}
+
 pub fn remote_initdb_archive_path(tenant_id: &TenantId, timeline_id: &TimelineId) -> RemotePath {
     RemotePath::from_string(&format!(
         "tenants/{tenant_id}/{TIMELINES_SEGMENT_NAME}/{timeline_id}/{INITDB_PATH}"

pageserver/src/tenant/remote_timeline_client/index.rs

@@ -8,14 +8,14 @@ use std::collections::HashMap;
 use chrono::NaiveDateTime;
 use pageserver_api::models::AuxFilePolicy;
 use serde::{Deserialize, Serialize};
-use utils::id::TimelineId;
 
+use super::is_same_remote_layer_path;
 use crate::tenant::metadata::TimelineMetadata;
 use crate::tenant::storage_layer::LayerName;
 use crate::tenant::timeline::import_pgdata;
 use crate::tenant::Generation;
 use pageserver_api::shard::ShardIndex;
-
+use utils::id::TimelineId;
 use utils::lsn::Lsn;
 
 /// In-memory representation of an `index_part.json` file
@@ -45,10 +45,8 @@ pub struct IndexPart {
     #[serde(skip_serializing_if = "Option::is_none")]
     pub import_pgdata: Option<import_pgdata::index_part_format::Root>,
 
-    /// Per layer file name metadata, which can be present for a present or missing layer file.
-    ///
-    /// Older versions of `IndexPart` will not have this property or have only a part of metadata
-    /// that latest version stores.
+    /// Layer filenames and metadata. For an index persisted in remote storage, all layers must
+    /// exist in remote storage.
     pub layer_metadata: HashMap<LayerName, LayerFileMetadata>,
 
     /// Because of the trouble of eyeballing the legacy "metadata" field, we copied the
@@ -143,6 +141,17 @@ impl IndexPart {
     pub(crate) fn example() -> Self {
         Self::empty(TimelineMetadata::example())
     }
+
+    /// Returns true if the index contains a reference to the given layer (i.e. file path).
+    ///
+    /// TODO: there should be a variant of LayerName for the physical remote path that contains
+    /// information about the shard and generation, to avoid passing in metadata.
+    pub fn references(&self, name: &LayerName, metadata: &LayerFileMetadata) -> bool {
+        let Some(index_metadata) = self.layer_metadata.get(name) else {
+            return false;
+        };
+        is_same_remote_layer_path(name, metadata, name, index_metadata)
+    }
 }
 
 /// Metadata gathered for each of the layer files.

pageserver/src/tenant/timeline.rs

@@ -31,9 +31,9 @@ use pageserver_api::{
     },
     keyspace::{KeySpaceAccum, KeySpaceRandomAccum, SparseKeyPartitioning},
     models::{
-        CompactionAlgorithm, CompactionAlgorithmSettings, DownloadRemoteLayersTaskInfo,
-        DownloadRemoteLayersTaskSpawnRequest, EvictionPolicy, InMemoryLayerInfo, LayerMapInfo,
-        LsnLease, TimelineState,
+        CompactKeyRange, CompactLsnRange, CompactionAlgorithm, CompactionAlgorithmSettings,
+        DownloadRemoteLayersTaskInfo, DownloadRemoteLayersTaskSpawnRequest, EvictionPolicy,
+        InMemoryLayerInfo, LayerMapInfo, LsnLease, TimelineState,
     },
     reltag::BlockNumber,
     shard::{ShardIdentity, ShardNumber, TenantShardId},
@@ -144,15 +144,19 @@ use self::layer_manager::LayerManager;
 use self::logical_size::LogicalSize;
 use self::walreceiver::{WalReceiver, WalReceiverConf};
 
-use super::config::TenantConf;
-use super::remote_timeline_client::index::IndexPart;
-use super::remote_timeline_client::RemoteTimelineClient;
-use super::secondary::heatmap::{HeatMapLayer, HeatMapTimeline};
-use super::storage_layer::{LayerFringe, LayerVisibilityHint, ReadableLayer};
-use super::upload_queue::NotInitialized;
-use super::GcError;
 use super::{
-    debug_assert_current_span_has_tenant_and_timeline_id, AttachedTenantConf, MaybeOffloaded,
+    config::TenantConf, storage_layer::LayerVisibilityHint, upload_queue::NotInitialized,
+    MaybeOffloaded,
+};
+use super::{debug_assert_current_span_has_tenant_and_timeline_id, AttachedTenantConf};
+use super::{remote_timeline_client::index::IndexPart, storage_layer::LayerFringe};
+use super::{
+    remote_timeline_client::RemoteTimelineClient, remote_timeline_client::WaitCompletionError,
+    storage_layer::ReadableLayer,
+};
+use super::{
+    secondary::heatmap::{HeatMapLayer, HeatMapTimeline},
+    GcError,
 };
 
 #[cfg(test)]
@@ -788,63 +792,6 @@ pub(crate) struct CompactRequest {
     pub sub_compaction_max_job_size_mb: Option<u64>,
 }
 
-#[serde_with::serde_as]
-#[derive(Debug, Clone, serde::Deserialize)]
-pub(crate) struct CompactLsnRange {
-    pub start: Lsn,
-    pub end: Lsn,
-}
-
-#[serde_with::serde_as]
-#[derive(Debug, Clone, serde::Deserialize)]
-pub(crate) struct CompactKeyRange {
-    #[serde_as(as = "serde_with::DisplayFromStr")]
-    pub start: Key,
-    #[serde_as(as = "serde_with::DisplayFromStr")]
-    pub end: Key,
-}
-
-impl From<Range<Lsn>> for CompactLsnRange {
-    fn from(range: Range<Lsn>) -> Self {
-        Self {
-            start: range.start,
-            end: range.end,
-        }
-    }
-}
-
-impl From<Range<Key>> for CompactKeyRange {
-    fn from(range: Range<Key>) -> Self {
-        Self {
-            start: range.start,
-            end: range.end,
-        }
-    }
-}
-
-impl From<CompactLsnRange> for Range<Lsn> {
-    fn from(range: CompactLsnRange) -> Self {
-        range.start..range.end
-    }
-}
-
-impl From<CompactKeyRange> for Range<Key> {
-    fn from(range: CompactKeyRange) -> Self {
-        range.start..range.end
-    }
-}
-
-impl CompactLsnRange {
-    #[cfg(test)]
-    #[cfg(feature = "testing")]
-    pub fn above(lsn: Lsn) -> Self {
-        Self {
-            start: lsn,
-            end: Lsn::MAX,
-        }
-    }
-}
-
 #[derive(Debug, Clone, Default)]
 pub(crate) struct CompactOptions {
     pub flags: EnumSet<CompactFlags>,
@@ -3893,6 +3840,24 @@ impl Timeline {
             // release lock on 'layers'
         };
 
+        // Backpressure mechanism: wait with continuation of the flush loop until we have uploaded all layer files.
+        // This makes us refuse ingest until the new layers have been persisted to the remote
+        let start = Instant::now();
+        self.remote_client
+            .wait_completion()
+            .await
+            .map_err(|e| match e {
+                WaitCompletionError::UploadQueueShutDownOrStopped
+                | WaitCompletionError::NotInitialized(
+                    NotInitialized::ShuttingDown | NotInitialized::Stopped,
+                ) => FlushLayerError::Cancelled,
+                WaitCompletionError::NotInitialized(NotInitialized::Uninitialized) => {
+                    FlushLayerError::Other(anyhow!(e).into())
+                }
+            })?;
+        let duration = start.elapsed().as_secs_f64();
+        self.metrics.flush_wait_upload_time_gauge_add(duration);
+
         // FIXME: between create_delta_layer and the scheduling of the upload in `update_metadata_file`,
         // a compaction can delete the file and then it won't be available for uploads any more.
         // We still schedule the upload, resulting in an error, but ideally we'd somehow avoid this
@@ -4894,6 +4859,7 @@ impl Timeline {
 
     async fn find_gc_time_cutoff(
         &self,
+        now: SystemTime,
         pitr: Duration,
         cancel: &CancellationToken,
         ctx: &RequestContext,
@@ -4901,7 +4867,6 @@ impl Timeline {
         debug_assert_current_span_has_tenant_and_timeline_id();
         if self.shard_identity.is_shard_zero() {
             // Shard Zero has SLRU data and can calculate the PITR time -> LSN mapping itself
-            let now = SystemTime::now();
             let time_range = if pitr == Duration::ZERO {
                 humantime::parse_duration(DEFAULT_PITR_INTERVAL).expect("constant is invalid")
             } else {
@@ -4987,6 +4952,7 @@ impl Timeline {
     #[instrument(skip_all, fields(timeline_id=%self.timeline_id))]
     pub(super) async fn find_gc_cutoffs(
         &self,
+        now: SystemTime,
         space_cutoff: Lsn,
         pitr: Duration,
         cancel: &CancellationToken,
@@ -5014,7 +4980,7 @@ impl Timeline {
         // - if PITR interval is set, then this is our cutoff.
         // - if PITR interval is not set, then we do a lookup
         //   based on DEFAULT_PITR_INTERVAL, so that size-based retention does not result in keeping history around permanently on idle databases.
-        let time_cutoff = self.find_gc_time_cutoff(pitr, cancel, ctx).await?;
+        let time_cutoff = self.find_gc_time_cutoff(now, pitr, cancel, ctx).await?;
 
         Ok(match (pitr, time_cutoff) {
             (Duration::ZERO, Some(time_cutoff)) => {

pageserver/src/tenant/timeline/compaction.rs

@@ -4,7 +4,7 @@
 //!
 //! The old legacy algorithm is implemented directly in `timeline.rs`.
 
-use std::collections::{BinaryHeap, HashMap, HashSet};
+use std::collections::{BinaryHeap, HashMap, HashSet, VecDeque};
 use std::ops::{Deref, Range};
 use std::sync::Arc;
 
@@ -16,10 +16,12 @@ use super::{
 
 use anyhow::{anyhow, bail, Context};
 use bytes::Bytes;
+use enumset::EnumSet;
 use fail::fail_point;
 use itertools::Itertools;
 use pageserver_api::key::KEY_SIZE;
 use pageserver_api::keyspace::ShardedRange;
+use pageserver_api::models::CompactInfoResponse;
 use pageserver_api::shard::{ShardCount, ShardIdentity, TenantShardId};
 use serde::Serialize;
 use tokio_util::sync::CancellationToken;
@@ -29,6 +31,8 @@ use utils::id::TimelineId;
 use crate::context::{AccessStatsBehavior, RequestContext, RequestContextBuilder};
 use crate::page_cache;
 use crate::statvfs::Statvfs;
+use crate::tenant::checks::check_valid_layermap;
+use crate::tenant::gc_block::GcBlock;
 use crate::tenant::remote_timeline_client::WaitCompletionError;
 use crate::tenant::storage_layer::batch_split_writer::{
     BatchWriterResult, SplitDeltaLayerWriter, SplitImageLayerWriter,
@@ -62,16 +66,284 @@ use super::CompactionError;
 /// Maximum number of deltas before generating an image layer in bottom-most compaction.
 const COMPACTION_DELTA_THRESHOLD: usize = 5;
 
-/// A scheduled compaction task.
-pub(crate) struct ScheduledCompactionTask {
-    /// It's unfortunate that we need to store a compact options struct here because the only outer
-    /// API we can call here is `compact_with_options` which does a few setup calls before starting the
-    /// actual compaction job... We should refactor this to store `GcCompactionJob` in the future.
-    pub options: CompactOptions,
-    /// The channel to send the compaction result. If this is a subcompaction, the last compaction job holds the sender.
-    pub result_tx: Option<tokio::sync::oneshot::Sender<()>>,
-    /// Hold the GC block. If this is a subcompaction, the last compaction job holds the gc block guard.
-    pub gc_block: Option<gc_block::Guard>,
+#[derive(Debug, Clone, Copy, Hash, PartialEq, Eq)]
+pub struct GcCompactionJobId(pub usize);
+
+impl std::fmt::Display for GcCompactionJobId {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(f, "{}", self.0)
+    }
+}
+
+#[derive(Debug, Clone)]
+pub enum GcCompactionQueueItem {
+    Manual(CompactOptions),
+    SubCompactionJob(CompactOptions),
+    #[allow(dead_code)]
+    UpdateL2Lsn(Lsn),
+    Notify(GcCompactionJobId),
+}
+
+impl GcCompactionQueueItem {
+    pub fn into_compact_info_resp(
+        self,
+        id: GcCompactionJobId,
+        running: bool,
+    ) -> Option<CompactInfoResponse> {
+        match self {
+            GcCompactionQueueItem::Manual(options) => Some(CompactInfoResponse {
+                compact_key_range: options.compact_key_range,
+                compact_lsn_range: options.compact_lsn_range,
+                sub_compaction: options.sub_compaction,
+                running,
+                job_id: id.0,
+            }),
+            GcCompactionQueueItem::SubCompactionJob(options) => Some(CompactInfoResponse {
+                compact_key_range: options.compact_key_range,
+                compact_lsn_range: options.compact_lsn_range,
+                sub_compaction: options.sub_compaction,
+                running,
+                job_id: id.0,
+            }),
+            GcCompactionQueueItem::UpdateL2Lsn(_) => None,
+            GcCompactionQueueItem::Notify(_) => None,
+        }
+    }
+}
+
+struct GcCompactionQueueInner {
+    running: Option<(GcCompactionJobId, GcCompactionQueueItem)>,
+    queued: VecDeque<(GcCompactionJobId, GcCompactionQueueItem)>,
+    notify: HashMap<GcCompactionJobId, tokio::sync::oneshot::Sender<()>>,
+    gc_guards: HashMap<GcCompactionJobId, gc_block::Guard>,
+    last_id: GcCompactionJobId,
+}
+
+impl GcCompactionQueueInner {
+    fn next_id(&mut self) -> GcCompactionJobId {
+        let id = self.last_id;
+        self.last_id = GcCompactionJobId(id.0 + 1);
+        id
+    }
+}
+
+/// A structure to store gc_compaction jobs.
+pub struct GcCompactionQueue {
+    /// All items in the queue, and the currently-running job.
+    inner: std::sync::Mutex<GcCompactionQueueInner>,
+    /// Ensure only one thread is consuming the queue.
+    consumer_lock: tokio::sync::Mutex<()>,
+}
+
+impl GcCompactionQueue {
+    pub fn new() -> Self {
+        GcCompactionQueue {
+            inner: std::sync::Mutex::new(GcCompactionQueueInner {
+                running: None,
+                queued: VecDeque::new(),
+                notify: HashMap::new(),
+                gc_guards: HashMap::new(),
+                last_id: GcCompactionJobId(0),
+            }),
+            consumer_lock: tokio::sync::Mutex::new(()),
+        }
+    }
+
+    pub fn cancel_scheduled(&self) {
+        let mut guard = self.inner.lock().unwrap();
+        guard.queued.clear();
+        guard.notify.clear();
+        guard.gc_guards.clear();
+    }
+
+    /// Schedule a manual compaction job.
+    pub fn schedule_manual_compaction(
+        &self,
+        options: CompactOptions,
+        notify: Option<tokio::sync::oneshot::Sender<()>>,
+    ) -> GcCompactionJobId {
+        let mut guard = self.inner.lock().unwrap();
+        let id = guard.next_id();
+        guard
+            .queued
+            .push_back((id, GcCompactionQueueItem::Manual(options)));
+        if let Some(notify) = notify {
+            guard.notify.insert(id, notify);
+        }
+        info!("scheduled compaction job id={}", id);
+        id
+    }
+
+    /// Trigger an auto compaction.
+    #[allow(dead_code)]
+    pub fn trigger_auto_compaction(&self, _: &Arc<Timeline>) {}
+
+    /// Notify the caller the job has finished and unblock GC.
+    fn notify_and_unblock(&self, id: GcCompactionJobId) {
+        info!("compaction job id={} finished", id);
+        let mut guard = self.inner.lock().unwrap();
+        if let Some(blocking) = guard.gc_guards.remove(&id) {
+            drop(blocking)
+        }
+        if let Some(tx) = guard.notify.remove(&id) {
+            let _ = tx.send(());
+        }
+    }
+
+    async fn handle_sub_compaction(
+        &self,
+        id: GcCompactionJobId,
+        options: CompactOptions,
+        timeline: &Arc<Timeline>,
+        gc_block: &GcBlock,
+    ) -> Result<(), CompactionError> {
+        info!("running scheduled enhanced gc bottom-most compaction with sub-compaction, splitting compaction jobs");
+        let jobs: Vec<GcCompactJob> = timeline
+            .gc_compaction_split_jobs(
+                GcCompactJob::from_compact_options(options.clone()),
+                options.sub_compaction_max_job_size_mb,
+            )
+            .await
+            .map_err(CompactionError::Other)?;
+        if jobs.is_empty() {
+            info!("no jobs to run, skipping scheduled compaction task");
+            self.notify_and_unblock(id);
+        } else {
+            let gc_guard = match gc_block.start().await {
+                Ok(guard) => guard,
+                Err(e) => {
+                    return Err(CompactionError::Other(anyhow!(
+                        "cannot run gc-compaction because gc is blocked: {}",
+                        e
+                    )));
+                }
+            };
+
+            let jobs_len = jobs.len();
+            let mut pending_tasks = Vec::new();
+            for job in jobs {
+                // Unfortunately we need to convert the `GcCompactJob` back to `CompactionOptions`
+                // until we do further refactors to allow directly call `compact_with_gc`.
+                let mut flags: EnumSet<CompactFlags> = EnumSet::default();
+                flags |= CompactFlags::EnhancedGcBottomMostCompaction;
+                if job.dry_run {
+                    flags |= CompactFlags::DryRun;
+                }
+                let options = CompactOptions {
+                    flags,
+                    sub_compaction: false,
+                    compact_key_range: Some(job.compact_key_range.into()),
+                    compact_lsn_range: Some(job.compact_lsn_range.into()),
+                    sub_compaction_max_job_size_mb: None,
+                };
+                pending_tasks.push(GcCompactionQueueItem::SubCompactionJob(options));
+            }
+            pending_tasks.push(GcCompactionQueueItem::Notify(id));
+            {
+                let mut guard = self.inner.lock().unwrap();
+                guard.gc_guards.insert(id, gc_guard);
+                let mut tasks = Vec::new();
+                for task in pending_tasks {
+                    let id = guard.next_id();
+                    tasks.push((id, task));
+                }
+                tasks.reverse();
+                for item in tasks {
+                    guard.queued.push_front(item);
+                }
+            }
+            info!("scheduled enhanced gc bottom-most compaction with sub-compaction, split into {} jobs", jobs_len);
+        }
+        Ok(())
+    }
+
+    /// Take a job from the queue and process it. Returns if there are still pending tasks.
+    pub async fn iteration(
+        &self,
+        cancel: &CancellationToken,
+        ctx: &RequestContext,
+        gc_block: &GcBlock,
+        timeline: &Arc<Timeline>,
+    ) -> Result<bool, CompactionError> {
+        let _one_op_at_a_time_guard = self.consumer_lock.lock().await;
+        let has_pending_tasks;
+        let (id, item) = {
+            let mut guard = self.inner.lock().unwrap();
+            let Some((id, item)) = guard.queued.pop_front() else {
+                return Ok(false);
+            };
+            guard.running = Some((id, item.clone()));
+            has_pending_tasks = !guard.queued.is_empty();
+            (id, item)
+        };
+
+        match item {
+            GcCompactionQueueItem::Manual(options) => {
+                if !options
+                    .flags
+                    .contains(CompactFlags::EnhancedGcBottomMostCompaction)
+                {
+                    warn!("ignoring scheduled compaction task: scheduled task must be gc compaction: {:?}", options);
+                } else if options.sub_compaction {
+                    self.handle_sub_compaction(id, options, timeline, gc_block)
+                        .await?;
+                } else {
+                    let gc_guard = match gc_block.start().await {
+                        Ok(guard) => guard,
+                        Err(e) => {
+                            return Err(CompactionError::Other(anyhow!(
+                                "cannot run gc-compaction because gc is blocked: {}",
+                                e
+                            )));
+                        }
+                    };
+                    {
+                        let mut guard = self.inner.lock().unwrap();
+                        guard.gc_guards.insert(id, gc_guard);
+                    }
+                    let _ = timeline
+                        .compact_with_options(cancel, options, ctx)
+                        .instrument(info_span!("scheduled_compact_timeline", %timeline.timeline_id))
+                        .await?;
+                    self.notify_and_unblock(id);
+                }
+            }
+            GcCompactionQueueItem::SubCompactionJob(options) => {
+                let _ = timeline
+                    .compact_with_options(cancel, options, ctx)
+                    .instrument(info_span!("scheduled_compact_timeline", %timeline.timeline_id))
+                    .await?;
+            }
+            GcCompactionQueueItem::Notify(id) => {
+                self.notify_and_unblock(id);
+            }
+            GcCompactionQueueItem::UpdateL2Lsn(_) => {
+                unreachable!()
+            }
+        }
+        {
+            let mut guard = self.inner.lock().unwrap();
+            guard.running = None;
+        }
+        Ok(has_pending_tasks)
+    }
+
+    #[allow(clippy::type_complexity)]
+    pub fn remaining_jobs(
+        &self,
+    ) -> (
+        Option<(GcCompactionJobId, GcCompactionQueueItem)>,
+        VecDeque<(GcCompactionJobId, GcCompactionQueueItem)>,
+    ) {
+        let guard = self.inner.lock().unwrap();
+        (guard.running.clone(), guard.queued.clone())
+    }
+
+    #[allow(dead_code)]
+    pub fn remaining_jobs_num(&self) -> usize {
+        let guard = self.inner.lock().unwrap();
+        guard.queued.len() + if guard.running.is_some() { 1 } else { 0 }
+    }
 }
 
 /// A job description for the gc-compaction job. This structure describes the rectangle range that the job will
@@ -1798,6 +2070,24 @@ impl Timeline {
         Ok(())
     }
 
+    /// Get a watermark for gc-compaction, that is the lowest LSN that we can use as the `gc_horizon` for
+    /// the compaction algorithm. It is min(space_cutoff, time_cutoff, latest_gc_cutoff, standby_horizon).
+    /// Leases and retain_lsns are considered in the gc-compaction job itself so we don't need to account for them
+    /// here.
+    pub(crate) fn get_gc_compaction_watermark(self: &Arc<Self>) -> Lsn {
+        let gc_cutoff_lsn = {
+            let gc_info = self.gc_info.read().unwrap();
+            gc_info.min_cutoff()
+        };
+
+        // TODO: standby horizon should use leases so we don't really need to consider it here.
+        // let watermark = watermark.min(self.standby_horizon.load());
+
+        // TODO: ensure the child branches will not use anything below the watermark, or consider
+        // them when computing the watermark.
+        gc_cutoff_lsn.min(*self.get_latest_gc_cutoff_lsn())
+    }
+
     /// Split a gc-compaction job into multiple compaction jobs. The split is based on the key range and the estimated size of the compaction job.
     /// The function returns a list of compaction jobs that can be executed separately. If the upper bound of the compact LSN
     /// range is not specified, we will use the latest gc_cutoff as the upper bound, so that all jobs in the jobset acts
@@ -1810,7 +2100,7 @@ impl Timeline {
         let compact_below_lsn = if job.compact_lsn_range.end != Lsn::MAX {
             job.compact_lsn_range.end
         } else {
-            *self.get_latest_gc_cutoff_lsn() // use the real gc cutoff
+            self.get_gc_compaction_watermark()
         };
 
         // Split compaction job to about 4GB each
@@ -1823,7 +2113,7 @@ impl Timeline {
         // by estimating the amount of files read for a compaction job. We should also partition on LSN.
         let ((dense_ks, sparse_ks), _) = {
             let Ok(partition) = self.partitioning.try_lock() else {
-                bail!("failed to acquire partition lock");
+                bail!("failed to acquire partition lock during gc-compaction");
             };
             partition.clone()
         };
@@ -2005,7 +2295,7 @@ impl Timeline {
                 // Therefore, it can only clean up data that cannot be cleaned up with legacy gc, instead of
                 // cleaning everything that theoritically it could. In the future, it should use `self.gc_info`
                 // to get the truth data.
-                let real_gc_cutoff = *self.get_latest_gc_cutoff_lsn();
+                let real_gc_cutoff = self.get_gc_compaction_watermark();
                 // The compaction algorithm will keep all keys above the gc_cutoff while keeping only necessary keys below the gc_cutoff for
                 // each of the retain_lsn. Therefore, if the user-provided `compact_lsn_range.end` is larger than the real gc cutoff, we will use
                 // the real cutoff.
@@ -2156,15 +2446,14 @@ impl Timeline {
 
         // Step 1: construct a k-merge iterator over all layers.
         // Also, verify if the layer map can be split by drawing a horizontal line at every LSN start/end split point.
-        // disable the check for now because we need to adjust the check for partial compactions, will enable later.
-        // let layer_names = job_desc
-        //     .selected_layers
-        //     .iter()
-        //     .map(|layer| layer.layer_desc().layer_name())
-        //     .collect_vec();
-        // if let Some(err) = check_valid_layermap(&layer_names) {
-        //     warn!("gc-compaction layer map check failed because {}, this is normal if partial compaction is not finished yet", err);
-        // }
+        let layer_names = job_desc
+            .selected_layers
+            .iter()
+            .map(|layer| layer.layer_desc().layer_name())
+            .collect_vec();
+        if let Some(err) = check_valid_layermap(&layer_names) {
+            bail!("gc-compaction layer map check failed because {}, cannot proceed with compaction due to potential data loss", err);
+        }
         // The maximum LSN we are processing in this compaction loop
         let end_lsn = job_desc
             .selected_layers
@@ -2546,13 +2835,48 @@ impl Timeline {
         );
 
         // Step 3: Place back to the layer map.
+
+        // First, do a sanity check to ensure the newly-created layer map does not contain overlaps.
+        let all_layers = {
+            let guard = self.layers.read().await;
+            let layer_map = guard.layer_map()?;
+            layer_map.iter_historic_layers().collect_vec()
+        };
+
+        let mut final_layers = all_layers
+            .iter()
+            .map(|layer| layer.layer_name())
+            .collect::<HashSet<_>>();
+        for layer in &layer_selection {
+            final_layers.remove(&layer.layer_desc().layer_name());
+        }
+        for layer in &compact_to {
+            final_layers.insert(layer.layer_desc().layer_name());
+        }
+        let final_layers = final_layers.into_iter().collect_vec();
+
+        // TODO: move this check before we call `finish` on image layer writers. However, this will require us to get the layer name before we finish
+        // the writer, so potentially, we will need a function like `ImageLayerBatchWriter::get_all_pending_layer_keys` to get all the keys that are
+        // in the writer before finalizing the persistent layers. Now we would leave some dangling layers on the disk if the check fails.
+        if let Some(err) = check_valid_layermap(&final_layers) {
+            bail!("gc-compaction layer map check failed after compaction because {}, compaction result not applied to the layer map due to potential data loss", err);
+        }
+
+        // Between the sanity check and this compaction update, there could be new layers being flushed, but it should be fine because we only
+        // operate on L1 layers.
         {
-            // TODO: sanity check if the layer map is valid (i.e., should not have overlaps)
             let mut guard = self.layers.write().await;
             guard
                 .open_mut()?
                 .finish_gc_compaction(&layer_selection, &compact_to, &self.metrics)
         };
+
+        // Schedule an index-only upload to update the `latest_gc_cutoff` in the index_part.json.
+        // Otherwise, after restart, the index_part only contains the old `latest_gc_cutoff` and
+        // find_gc_cutoffs will try accessing things below the cutoff. TODO: ideally, this should
+        // be batched into `schedule_compaction_update`.
+        let disk_consistent_lsn = self.disk_consistent_lsn.load();
+        self.schedule_uploads(disk_consistent_lsn, None)?;
         self.remote_client
             .schedule_compaction_update(&layer_selection, &compact_to)?;
 

pageserver/src/tenant/timeline/delete.rs

@@ -194,7 +194,9 @@ impl DeleteTimelineFlow {
         super::debug_assert_current_span_has_tenant_and_timeline_id();
 
         let allow_offloaded_children = false;
-        let (timeline, mut guard) = Self::prepare(tenant, timeline_id, allow_offloaded_children)?;
+        let set_stopping = true;
+        let (timeline, mut guard) =
+            Self::prepare(tenant, timeline_id, allow_offloaded_children, set_stopping)?;
 
         guard.mark_in_progress()?;
 
@@ -334,6 +336,7 @@ impl DeleteTimelineFlow {
         tenant: &Tenant,
         timeline_id: TimelineId,
         allow_offloaded_children: bool,
+        set_stopping: bool,
     ) -> Result<(TimelineOrOffloaded, DeletionGuard), DeleteTimelineError> {
         // Note the interaction between this guard and deletion guard.
         // Here we attempt to lock deletion guard when we're holding a lock on timelines.
@@ -389,8 +392,10 @@ impl DeleteTimelineFlow {
             }
         };
 
-        if let TimelineOrOffloaded::Timeline(timeline) = &timeline {
-            timeline.set_state(TimelineState::Stopping);
+        if set_stopping {
+            if let TimelineOrOffloaded::Timeline(timeline) = &timeline {
+                timeline.set_state(TimelineState::Stopping);
+            }
         }
 
         Ok((timeline, delete_lock_guard))

pageserver/src/tenant/timeline/offload.rs

@@ -1,8 +1,11 @@
 use std::sync::Arc;
 
+use pageserver_api::models::{TenantState, TimelineState};
+
 use super::delete::{delete_local_timeline_directory, DeleteTimelineFlow, DeletionGuard};
 use super::Timeline;
 use crate::span::debug_assert_current_span_has_tenant_and_timeline_id;
+use crate::tenant::remote_timeline_client::ShutdownIfArchivedError;
 use crate::tenant::{OffloadedTimeline, Tenant, TenantManifestError, TimelineOrOffloaded};
 
 #[derive(thiserror::Error, Debug)]
@@ -34,28 +37,29 @@ pub(crate) async fn offload_timeline(
     tracing::info!("offloading archived timeline");
 
     let allow_offloaded_children = true;
-    let (timeline, guard) =
-        DeleteTimelineFlow::prepare(tenant, timeline.timeline_id, allow_offloaded_children)
-            .map_err(|e| OffloadError::Other(anyhow::anyhow!(e)))?;
+    let set_stopping = false;
+    let (timeline, guard) = DeleteTimelineFlow::prepare(
+        tenant,
+        timeline.timeline_id,
+        allow_offloaded_children,
+        set_stopping,
+    )
+    .map_err(|e| OffloadError::Other(anyhow::anyhow!(e)))?;
 
     let TimelineOrOffloaded::Timeline(timeline) = timeline else {
         tracing::error!("timeline already offloaded, but given timeline object");
         return Ok(());
     };
 
-    let is_archived = timeline.is_archived();
-    match is_archived {
-        Some(true) => (),
-        Some(false) => {
-            tracing::warn!("tried offloading a non-archived timeline");
-            return Err(OffloadError::NotArchived);
-        }
-        None => {
-            // This is legal: calls to this function can race with the timeline shutting down
-            tracing::info!("tried offloading a timeline whose remote storage is not initialized");
-            return Err(OffloadError::Cancelled);
+    match timeline.remote_client.shutdown_if_archived().await {
+        Ok(()) => {}
+        Err(ShutdownIfArchivedError::NotInitialized(_)) => {
+            // Either the timeline is being deleted, the operation is being retried, or we are shutting down.
+            // Don't return cancelled here to keep it idempotent.
         }
+        Err(ShutdownIfArchivedError::NotArchived) => return Err(OffloadError::NotArchived),
     }
+    timeline.set_state(TimelineState::Stopping);
 
     // Now that the Timeline is in Stopping state, request all the related tasks to shut down.
     timeline.shutdown(super::ShutdownMode::Reload).await;
@@ -70,6 +74,15 @@ pub(crate) async fn offload_timeline(
 
     {
         let mut offloaded_timelines = tenant.timelines_offloaded.lock().unwrap();
+        if matches!(
+            tenant.current_state(),
+            TenantState::Stopping { .. } | TenantState::Broken { .. }
+        ) {
+            // Cancel the operation if the tenant is shutting down. Do this while the
+            // timelines_offloaded lock is held to prevent a race with Tenant::shutdown
+            // for defusing the lock
+            return Err(OffloadError::Cancelled);
+        }
         offloaded_timelines.insert(
             timeline.timeline_id,
             Arc::new(

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

@@ -319,27 +319,11 @@ pub(super) async fn handle_walreceiver_connection(
             return Ok(());
         }
 
-        async fn commit(
-            modification: &mut DatadirModification<'_>,
-            uncommitted: &mut u64,
-            filtered: &mut u64,
-            ctx: &RequestContext,
-        ) -> anyhow::Result<()> {
-            WAL_INGEST
-                .records_committed
-                .inc_by(*uncommitted - *filtered);
-            modification.commit(ctx).await?;
-            *uncommitted = 0;
-            *filtered = 0;
-            Ok(())
-        }
-
         let status_update = match replication_message {
             ReplicationMessage::RawInterpretedWalRecords(raw) => {
                 WAL_INGEST.bytes_received.inc_by(raw.data().len() as u64);
 
                 let mut uncommitted_records = 0;
-                let mut filtered_records = 0;
 
                 // This is the end LSN of the raw WAL from which the records
                 // were interpreted.
@@ -380,31 +364,23 @@ pub(super) async fn handle_walreceiver_connection(
                     if matches!(interpreted.flush_uncommitted, FlushUncommittedRecords::Yes)
                         && uncommitted_records > 0
                     {
-                        commit(
-                            &mut modification,
-                            &mut uncommitted_records,
-                            &mut filtered_records,
-                            &ctx,
-                        )
-                        .await?;
+                        modification.commit(&ctx).await?;
+                        uncommitted_records = 0;
                     }
 
                     let local_next_record_lsn = interpreted.next_record_lsn;
-                    let ingested = walingest
+
+                    if interpreted.is_observed() {
+                        WAL_INGEST.records_observed.inc();
+                    }
+
+                    walingest
                         .ingest_record(interpreted, &mut modification, &ctx)
                         .await
                         .with_context(|| {
                             format!("could not ingest record at {local_next_record_lsn}")
                         })?;
 
-                    if !ingested {
-                        tracing::debug!(
-                            "ingest: filtered out record @ LSN {local_next_record_lsn}"
-                        );
-                        WAL_INGEST.records_filtered.inc();
-                        filtered_records += 1;
-                    }
-
                     uncommitted_records += 1;
 
                     // FIXME: this cannot be made pausable_failpoint without fixing the
@@ -418,13 +394,8 @@ pub(super) async fn handle_walreceiver_connection(
                         || modification.approx_pending_bytes()
                             > DatadirModification::MAX_PENDING_BYTES
                     {
-                        commit(
-                            &mut modification,
-                            &mut uncommitted_records,
-                            &mut filtered_records,
-                            &ctx,
-                        )
-                        .await?;
+                        modification.commit(&ctx).await?;
+                        uncommitted_records = 0;
                     }
                 }
 
@@ -442,13 +413,7 @@ pub(super) async fn handle_walreceiver_connection(
 
                 if uncommitted_records > 0 || needs_last_record_lsn_advance {
                     // Commit any uncommitted records
-                    commit(
-                        &mut modification,
-                        &mut uncommitted_records,
-                        &mut filtered_records,
-                        &ctx,
-                    )
-                    .await?;
+                    modification.commit(&ctx).await?;
                 }
 
                 if !caught_up && streaming_lsn >= end_of_wal {
@@ -469,6 +434,21 @@ pub(super) async fn handle_walreceiver_connection(
             }
 
             ReplicationMessage::XLogData(xlog_data) => {
+                async fn commit(
+                    modification: &mut DatadirModification<'_>,
+                    uncommitted: &mut u64,
+                    filtered: &mut u64,
+                    ctx: &RequestContext,
+                ) -> anyhow::Result<()> {
+                    WAL_INGEST
+                        .records_committed
+                        .inc_by(*uncommitted - *filtered);
+                    modification.commit(ctx).await?;
+                    *uncommitted = 0;
+                    *filtered = 0;
+                    Ok(())
+                }
+
                 // Pass the WAL data to the decoder, and see if we can decode
                 // more records as a result.
                 let data = xlog_data.data();

pgxn/hnsw/Makefile

@@ -1,26 +0,0 @@
-EXTENSION = hnsw
-EXTVERSION = 0.1.0
-
-MODULE_big = hnsw
-DATA = $(wildcard *--*.sql)
-OBJS = hnsw.o hnswalg.o
-
-TESTS = $(wildcard test/sql/*.sql)
-REGRESS = $(patsubst test/sql/%.sql,%,$(TESTS))
-REGRESS_OPTS = --inputdir=test --load-extension=hnsw
-
-# For auto-vectorization:
-# - GCC (needs -ftree-vectorize OR -O3) - https://gcc.gnu.org/projects/tree-ssa/vectorization.html
-PG_CFLAGS += -O3
-PG_CXXFLAGS +=  -O3 -std=c++11
-PG_LDFLAGS += -lstdc++
-
-all: $(EXTENSION)--$(EXTVERSION).sql
-
-PG_CONFIG ?= pg_config
-PGXS := $(shell $(PG_CONFIG) --pgxs)
-include $(PGXS)
-
-dist:
-	mkdir -p dist
-	git archive --format zip --prefix=$(EXTENSION)-$(EXTVERSION)/ --output dist/$(EXTENSION)-$(EXTVERSION).zip master

pgxn/hnsw/README.md

@@ -1,25 +0,0 @@
-# Revisiting the Inverted Indices for Billion-Scale Approximate Nearest Neighbors
-
-This ANN extension of Postgres is based
-on [ivf-hnsw](https://github.com/dbaranchuk/ivf-hnsw.git) implementation of [HNSW](https://www.pinecone.io/learn/hnsw),
-the code for the current state-of-the-art billion-scale nearest neighbor search system presented in the paper:
-
-[Revisiting the Inverted Indices for Billion-Scale Approximate Nearest Neighbors](http://openaccess.thecvf.com/content_ECCV_2018/html/Dmitry_Baranchuk_Revisiting_the_Inverted_ECCV_2018_paper.html),
-<br>
-Dmitry Baranchuk, Artem Babenko, Yury Malkov
-
-# Postgres extension
-
-HNSW index is hold in memory (built on demand) and it's maxial size is limited
-by `maxelements` index parameter. Another required parameter is nubmer of dimensions (if it is not specified in column type).
-Optional parameter `ef` specifies number of neighbors which are considered during index construction and search (corresponds `efConstruction` and `efSearch` parameters
-described in the article).
-
-# Example of usage:
-
-```
-create extension hnsw;
-create table embeddings(id integer primary key, payload real[]);
-create index on embeddings using hnsw(payload) with (maxelements=1000000, dims=100, m=32);
-select id from embeddings order by payload <-> array[1.0, 2.0,...] limit 100;
-```

pgxn/hnsw/hnsw--0.1.0.sql

@@ -1,29 +0,0 @@
--- complain if script is sourced in psql, rather than via CREATE EXTENSION
-\echo Use "CREATE EXTENSION hnsw" to load this file. \quit
-
--- functions
-
-CREATE FUNCTION l2_distance(real[], real[]) RETURNS real
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
--- operators
-
-CREATE OPERATOR <-> (
-	LEFTARG = real[], RIGHTARG = real[], PROCEDURE = l2_distance,
-	COMMUTATOR = '<->'
-);
-
--- access method
-
-CREATE FUNCTION hnsw_handler(internal) RETURNS index_am_handler
-	AS 'MODULE_PATHNAME' LANGUAGE C;
-
-CREATE ACCESS METHOD hnsw TYPE INDEX HANDLER hnsw_handler;
-
-COMMENT ON ACCESS METHOD hnsw IS 'hnsw index access method';
-
--- opclasses
-
-CREATE OPERATOR CLASS knn_ops
-	DEFAULT FOR TYPE real[] USING hnsw AS
-	OPERATOR 1 <-> (real[], real[]) FOR ORDER BY float_ops;