state machines for better error management, small futures, and clearer flows

.dockerignore

@@ -1,27 +1,27 @@
 *
 
-# Files
-!Cargo.lock
-!Cargo.toml
-!Makefile
 !rust-toolchain.toml
-!scripts/combine_control_files.py
-!scripts/ninstall.sh
-!vm-cgconfig.conf
+!Cargo.toml
+!Cargo.lock
+!Makefile
 
-# Directories
 !.cargo/
 !.config/
-!compute_tools/
 !control_plane/
+!compute_tools/
 !libs/
-!neon_local/
 !pageserver/
 !pgxn/
 !proxy/
-!s3_scrubber/
 !safekeeper/
+!s3_scrubber/
 !storage_broker/
 !trace/
-!vendor/postgres-*/
+!vendor/postgres-v14/
+!vendor/postgres-v15/
+!vendor/postgres-v16/
 !workspace_hack/
+!neon_local/
+!scripts/ninstall.sh
+!scripts/combine_control_files.py
+!vm-cgconfig.conf

.github/actionlint.yml

@@ -4,8 +4,6 @@ self-hosted-runner:
     - dev
     - gen3
     - large
-    # Remove `macos-14` from the list after https://github.com/rhysd/actionlint/pull/392 is merged.
-    - macos-14
     - small
     - us-east-2
 config-variables:

.github/actions/allure-report-generate/action.yml

@@ -179,6 +179,23 @@ runs:
           aws s3 rm "s3://${BUCKET}/${LOCK_FILE}"
         fi
 
+    - name: Store Allure test stat in the DB
+      if: ${{ !cancelled() && inputs.store-test-results-into-db == 'true' }}
+      shell: bash -euxo pipefail {0}
+      env:
+        COMMIT_SHA: ${{ github.event.pull_request.head.sha || github.sha }}
+        REPORT_JSON_URL: ${{ steps.generate-report.outputs.report-json-url }}
+      run: |
+        export DATABASE_URL=${REGRESS_TEST_RESULT_CONNSTR}
+
+        ./scripts/pysync
+
+        poetry run python3 scripts/ingest_regress_test_result.py \
+          --revision ${COMMIT_SHA} \
+          --reference ${GITHUB_REF} \
+          --build-type unified \
+          --ingest ${WORKDIR}/report/data/suites.json
+
     - name: Store Allure test stat in the DB (new)
       if: ${{ !cancelled() && inputs.store-test-results-into-db == 'true' }}
       shell: bash -euxo pipefail {0}

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

@@ -44,10 +44,6 @@ inputs:
     description: 'Postgres version to use for tests'
     required: false
     default: 'v14'
-  benchmark_durations:
-    description: 'benchmark durations JSON'
-    required: false
-    default: '{}'
 
 runs:
   using: "composite"
@@ -164,7 +160,7 @@ runs:
         # We use pytest-split plugin to run benchmarks in parallel on different CI runners
         if [ "${TEST_SELECTION}" = "test_runner/performance" ] && [ "${{ inputs.build_type }}" != "remote" ]; then
           mkdir -p $TEST_OUTPUT
-          echo '${{ inputs.benchmark_durations || '{}' }}' > $TEST_OUTPUT/benchmark_durations.json
+          poetry run ./scripts/benchmark_durations.py "${TEST_RESULT_CONNSTR}" --days 10 --output "$TEST_OUTPUT/benchmark_durations.json"
 
           EXTRA_PARAMS="--durations-path $TEST_OUTPUT/benchmark_durations.json $EXTRA_PARAMS"
         fi

.github/workflows/build_and_push_docker_image.yml

@@ -69,15 +69,7 @@ jobs:
         run: echo "{\"credsStore\":\"ecr-login\"}" > /kaniko/.docker/config.json
 
       - name: Kaniko build
-        run: |
-          /kaniko/executor \
-            --reproducible \
-            --snapshotMode=redo \
-            --skip-unused-stages \
-            --dockerfile ${{ inputs.dockerfile-path }} \
-            --cache=true \
-            --cache-repo 369495373322.dkr.ecr.eu-central-1.amazonaws.com/cache \
-            --destination 369495373322.dkr.ecr.eu-central-1.amazonaws.com/${{ inputs.image-name }}:${{ needs.tag.outputs.build-tools-tag }}-amd64
+        run: /kaniko/executor --reproducible --snapshotMode=redo --skip-unused-stages --dockerfile ${{ inputs.dockerfile-path }} --cache=true --cache-repo 369495373322.dkr.ecr.eu-central-1.amazonaws.com/cache  --destination 369495373322.dkr.ecr.eu-central-1.amazonaws.com/${{ inputs.image-name }}:${{ needs.tag.outputs.build-tools-tag }}-amd64
 
   kaniko-arm:
     if: needs.check-if-build-tools-dockerfile-changed.outputs.docker_file_changed == 'true'
@@ -93,15 +85,7 @@ jobs:
         run: echo "{\"credsStore\":\"ecr-login\"}" > /kaniko/.docker/config.json
 
       - name: Kaniko build
-        run: |
-          /kaniko/executor \
-            --reproducible \
-            --snapshotMode=redo \
-            --skip-unused-stages \
-            --dockerfile ${{ inputs.dockerfile-path }} \
-            --cache=true \
-            --cache-repo 369495373322.dkr.ecr.eu-central-1.amazonaws.com/cache \
-            --destination 369495373322.dkr.ecr.eu-central-1.amazonaws.com/${{ inputs.image-name }}:${{ needs.tag.outputs.build-tools-tag }}-arm64
+        run: /kaniko/executor --reproducible --snapshotMode=redo --skip-unused-stages --dockerfile ${{ inputs.dockerfile-path }} --cache=true --cache-repo 369495373322.dkr.ecr.eu-central-1.amazonaws.com/cache --destination 369495373322.dkr.ecr.eu-central-1.amazonaws.com/${{ inputs.image-name }}:${{ needs.tag.outputs.build-tools-tag }}-arm64
 
   manifest:
     if: needs.check-if-build-tools-dockerfile-changed.outputs.docker_file_changed == 'true'
@@ -115,10 +99,7 @@ jobs:
 
     steps:
       - name: Create manifest
-        run: |
-          docker manifest create 369495373322.dkr.ecr.eu-central-1.amazonaws.com/${{ inputs.image-name }}:${{ needs.tag.outputs.build-tools-tag }} \
-                         --amend 369495373322.dkr.ecr.eu-central-1.amazonaws.com/${{ inputs.image-name }}:${{ needs.tag.outputs.build-tools-tag }}-amd64 \
-                         --amend 369495373322.dkr.ecr.eu-central-1.amazonaws.com/${{ inputs.image-name }}:${{ needs.tag.outputs.build-tools-tag }}-arm64
+        run: docker manifest create 369495373322.dkr.ecr.eu-central-1.amazonaws.com/${{ inputs.image-name }}:${{ needs.tag.outputs.build-tools-tag }} --amend 369495373322.dkr.ecr.eu-central-1.amazonaws.com/${{ inputs.image-name }}:${{ needs.tag.outputs.build-tools-tag }}-amd64 --amend 369495373322.dkr.ecr.eu-central-1.amazonaws.com/${{ inputs.image-name }}:${{ needs.tag.outputs.build-tools-tag }}-arm64
 
       - name: Push manifest
         run: docker manifest push 369495373322.dkr.ecr.eu-central-1.amazonaws.com/${{ inputs.image-name }}:${{ needs.tag.outputs.build-tools-tag }}

.github/workflows/build_and_test.yml

@@ -21,8 +21,6 @@ env:
   COPT: '-Werror'
   AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_DEV }}
   AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_KEY_DEV }}
-  # A concurrency group that we use for e2e-tests runs, matches `concurrency.group` above with `github.repository` as a prefix
-  E2E_CONCURRENCY_GROUP: ${{ github.repository }}-${{ github.workflow }}-${{ github.ref_name }}-${{ github.ref_name == 'main' && github.sha || 'anysha' }}
 
 jobs:
   check-permissions:
@@ -46,20 +44,6 @@ jobs:
 
         exit 1
 
-  cancel-previous-e2e-tests:
-    needs: [ check-permissions ]
-    if: github.event_name == 'pull_request'
-    runs-on: ubuntu-latest
-
-    steps:
-      - name: Cancel previous e2e-tests runs for this PR
-        env:
-          GH_TOKEN: ${{ secrets.CI_ACCESS_TOKEN }}
-        run: |
-          gh workflow --repo neondatabase/cloud \
-            run cancel-previous-in-concurrency-group.yml \
-              --field concurrency_group="${{ env.E2E_CONCURRENCY_GROUP }}"
-
   tag:
     needs: [ check-permissions ]
     runs-on: [ self-hosted, gen3, small ]
@@ -132,7 +116,7 @@ jobs:
 
   check-codestyle-rust:
     needs: [ check-permissions, build-buildtools-image ]
-    runs-on: [ self-hosted, gen3, small ]
+    runs-on: [ self-hosted, gen3, large ]
     container:
       image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools:${{ needs.build-buildtools-image.outputs.build-tools-tag }}
       options: --init
@@ -202,11 +186,7 @@ jobs:
     runs-on: [ self-hosted, gen3, large ]
     container:
       image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools:${{ needs.build-buildtools-image.outputs.build-tools-tag }}
-      # Raise locked memory limit for tokio-epoll-uring.
-      # On 5.10 LTS kernels < 5.10.162 (and generally mainline kernels < 5.12),
-      # io_uring will account the memory of the CQ and SQ as locked.
-      # More details: https://github.com/neondatabase/neon/issues/6373#issuecomment-1905814391
-      options: --init --shm-size=512mb --ulimit memlock=67108864:67108864
+      options: --init
     strategy:
       fail-fast: false
       matrix:
@@ -360,12 +340,8 @@ jobs:
           ${cov_prefix} mold -run cargo build $CARGO_FLAGS $CARGO_FEATURES --bins --tests
 
       - name: Run rust tests
-        env:
-          NEXTEST_RETRIES: 3
         run: |
-          for io_engine in std-fs tokio-epoll-uring ; do
-            NEON_PAGESERVER_UNIT_TEST_VIRTUAL_FILE_IOENGINE=$io_engine ${cov_prefix} cargo nextest run $CARGO_FLAGS $CARGO_FEATURES
-          done
+          ${cov_prefix} cargo nextest run $CARGO_FLAGS $CARGO_FEATURES
 
           # Run separate tests for real S3
           export ENABLE_REAL_S3_REMOTE_STORAGE=nonempty
@@ -443,8 +419,8 @@ jobs:
     runs-on: [ self-hosted, gen3, large ]
     container:
       image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools:${{ needs.build-buildtools-image.outputs.build-tools-tag }}
-      # for changed limits, see comments on `options:` earlier in this file
-      options: --init --shm-size=512mb --ulimit memlock=67108864:67108864
+      # Default shared memory is 64mb
+      options: --init --shm-size=512mb
     strategy:
       fail-fast: false
       matrix:
@@ -472,57 +448,23 @@ jobs:
           TEST_RESULT_CONNSTR: ${{ secrets.REGRESS_TEST_RESULT_CONNSTR_NEW }}
           CHECK_ONDISK_DATA_COMPATIBILITY: nonempty
           BUILD_TAG: ${{ needs.tag.outputs.build-tag }}
-          PAGESERVER_VIRTUAL_FILE_IO_ENGINE: std-fs
 
       - name: Merge and upload coverage data
         if: matrix.build_type == 'debug' && matrix.pg_version == 'v14'
         uses: ./.github/actions/save-coverage-data
 
-  get-benchmarks-durations:
-    outputs:
-      json: ${{ steps.get-benchmark-durations.outputs.json }}
-    needs: [ check-permissions, build-buildtools-image ]
-    runs-on: [ self-hosted, gen3, small ]
-    container:
-      image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools:${{ needs.build-buildtools-image.outputs.build-tools-tag }}
-      options: --init
-    if: github.ref_name == 'main' || contains(github.event.pull_request.labels.*.name, 'run-benchmarks')
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v3
-
-      - name: Cache poetry deps
-        uses: actions/cache@v3
-        with:
-          path: ~/.cache/pypoetry/virtualenvs
-          key: v1-${{ runner.os }}-python-deps-${{ hashFiles('poetry.lock') }}
-
-      - name: Install Python deps
-        run: ./scripts/pysync
-
-      - name: get benchmark durations
-        id: get-benchmark-durations
-        env:
-          TEST_RESULT_CONNSTR: ${{ secrets.REGRESS_TEST_RESULT_CONNSTR_NEW }}
-        run: |
-          poetry run ./scripts/benchmark_durations.py "${TEST_RESULT_CONNSTR}" \
-                                                      --days 10 \
-                                                      --output /tmp/benchmark_durations.json
-          echo "json=$(jq --compact-output '.' /tmp/benchmark_durations.json)" >> $GITHUB_OUTPUT
-
   benchmarks:
-    needs: [ check-permissions, build-neon, build-buildtools-image, get-benchmarks-durations ]
+    needs: [ check-permissions, build-neon, build-buildtools-image ]
     runs-on: [ self-hosted, gen3, small ]
     container:
       image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools:${{ needs.build-buildtools-image.outputs.build-tools-tag }}
-      # for changed limits, see comments on `options:` earlier in this file
-      options: --init --shm-size=512mb --ulimit memlock=67108864:67108864
+      # Default shared memory is 64mb
+      options: --init --shm-size=512mb
     if: github.ref_name == 'main' || contains(github.event.pull_request.labels.*.name, 'run-benchmarks')
     strategy:
       fail-fast: false
       matrix:
-        # the amount of groups (N) should be reflected in `extra_params: --splits N ...`
-        pytest_split_group: [ 1, 2, 3, 4, 5 ]
+        pytest_split_group: [ 1, 2, 3, 4 ]
         build_type: [ release ]
     steps:
       - name: Checkout
@@ -535,13 +477,11 @@ jobs:
           test_selection: performance
           run_in_parallel: false
           save_perf_report: ${{ github.ref_name == 'main' }}
-          extra_params: --splits 5 --group ${{ matrix.pytest_split_group }}
-          benchmark_durations: ${{ needs.get-benchmarks-durations.outputs.json }}
+          extra_params: --splits ${{ strategy.job-total }} --group ${{ matrix.pytest_split_group }}
         env:
           VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
           PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
           TEST_RESULT_CONNSTR: "${{ secrets.REGRESS_TEST_RESULT_CONNSTR_NEW }}"
-          PAGESERVER_VIRTUAL_FILE_IO_ENGINE: std-fs
       # XXX: no coverage data handling here, since benchmarks are run on release builds,
       # while coverage is currently collected for the debug ones
 
@@ -564,6 +504,7 @@ jobs:
         with:
           store-test-results-into-db: true
         env:
+          REGRESS_TEST_RESULT_CONNSTR: ${{ secrets.REGRESS_TEST_RESULT_CONNSTR }}
           REGRESS_TEST_RESULT_CONNSTR_NEW: ${{ secrets.REGRESS_TEST_RESULT_CONNSTR_NEW }}
 
       - uses: actions/github-script@v6
@@ -641,6 +582,17 @@ jobs:
             --input-objects=/tmp/coverage/binaries.list \
             --format=lcov
 
+      - name: Upload coverage report
+        id: upload-coverage-report
+        env:
+          BUCKET: neon-github-public-dev
+          COMMIT_SHA: ${{ github.event.pull_request.head.sha || github.sha }}
+        run: |
+          aws s3 cp --only-show-errors --recursive /tmp/coverage/report s3://${BUCKET}/code-coverage/${COMMIT_SHA}
+
+          REPORT_URL=https://${BUCKET}.s3.amazonaws.com/code-coverage/${COMMIT_SHA}/index.html
+          echo "report-url=${REPORT_URL}" >> $GITHUB_OUTPUT
+
       - name: Build coverage report NEW
         id: upload-coverage-report-new
         env:
@@ -677,11 +629,21 @@ jobs:
 
       - uses: actions/github-script@v6
         env:
+          REPORT_URL: ${{ steps.upload-coverage-report.outputs.report-url }}
           REPORT_URL_NEW: ${{ steps.upload-coverage-report-new.outputs.report-url }}
           COMMIT_SHA: ${{ github.event.pull_request.head.sha || github.sha }}
         with:
           script: |
-            const { REPORT_URL_NEW, COMMIT_SHA } = process.env
+            const { REPORT_URL, REPORT_URL_NEW, COMMIT_SHA } = process.env
+
+            await github.rest.repos.createCommitStatus({
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              sha: `${COMMIT_SHA}`,
+              state: 'success',
+              target_url: `${REPORT_URL}`,
+              context: 'Code coverage report',
+            })
 
             await github.rest.repos.createCommitStatus({
               owner: context.repo.owner,
@@ -733,8 +695,7 @@ jobs:
                 \"commit_hash\": \"$COMMIT_SHA\",
                 \"remote_repo\": \"${{ github.repository }}\",
                 \"storage_image_tag\": \"${{ needs.tag.outputs.build-tag }}\",
-                \"compute_image_tag\": \"${{ needs.tag.outputs.build-tag }}\",
-                \"concurrency_group\": \"${{ env.E2E_CONCURRENCY_GROUP }}\"
+                \"compute_image_tag\": \"${{ needs.tag.outputs.build-tag }}\"
               }
             }"
 
@@ -905,7 +866,7 @@ jobs:
       run:
         shell: sh -eu {0}
     env:
-      VM_BUILDER_VERSION: v0.23.2
+      VM_BUILDER_VERSION: v0.21.0
 
     steps:
       - name: Checkout

.github/workflows/neon_extra_builds.yml

@@ -26,7 +26,7 @@ jobs:
       contains(github.event.pull_request.labels.*.name, 'run-extra-build-*') ||
       github.ref_name == 'main'
     timeout-minutes: 90
-    runs-on: macos-14
+    runs-on: macos-latest
 
     env:
       # Use release build only, to have less debug info around
@@ -60,21 +60,21 @@ jobs:
         uses: actions/cache@v3
         with:
           path: pg_install/v14
-          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ env.BUILD_TYPE }}-pg-${{ steps.pg_v14_rev.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
+          key: v1-${{ runner.os }}-${{ env.BUILD_TYPE }}-pg-${{ steps.pg_v14_rev.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
 
       - name: Cache postgres v15 build
         id: cache_pg_15
         uses: actions/cache@v3
         with:
           path: pg_install/v15
-          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ env.BUILD_TYPE }}-pg-${{ steps.pg_v15_rev.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
+          key: v1-${{ runner.os }}-${{ env.BUILD_TYPE }}-pg-${{ steps.pg_v15_rev.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
 
       - name: Cache postgres v16 build
         id: cache_pg_16
         uses: actions/cache@v3
         with:
           path: pg_install/v16
-          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ env.BUILD_TYPE }}-pg-${{ steps.pg_v16_rev.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
+          key: v1-${{ runner.os }}-${{ env.BUILD_TYPE }}-pg-${{ steps.pg_v16_rev.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
 
       - name: Set extra env for macOS
         run: |
@@ -89,7 +89,7 @@ jobs:
             !~/.cargo/registry/src
             ~/.cargo/git
             target
-          key: v1-${{ runner.os }}-${{ runner.arch }}-cargo-${{ hashFiles('./Cargo.lock') }}-${{ hashFiles('./rust-toolchain.toml') }}-rust
+          key: v1-${{ runner.os }}-cargo-${{ hashFiles('./Cargo.lock') }}-${{ hashFiles('./rust-toolchain.toml') }}-rust
 
       - name: Build postgres v14
         if: steps.cache_pg_14.outputs.cache-hit != 'true'
@@ -110,7 +110,7 @@ jobs:
         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
+        run: cargo build --all --release
 
       - name: Check that no warnings are produced
         run: ./run_clippy.sh
@@ -124,12 +124,12 @@ jobs:
       # Hence keeping target/ (and general cache size) smaller
       BUILD_TYPE: release
       CARGO_FEATURES: --features testing
-      CARGO_FLAGS: --release
+      CARGO_FLAGS: --locked --release
       AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_DEV }}
       AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_KEY_DEV }}
 
     container:
-      image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools:pinned
+      image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:pinned
       options: --init
 
     steps:
@@ -210,20 +210,18 @@ jobs:
 
       - name: Run cargo build
         run: |
-          mold -run cargo build --locked $CARGO_FLAGS $CARGO_FEATURES --bins --tests
+          mold -run cargo build $CARGO_FLAGS $CARGO_FEATURES --bins --tests
 
       - name: Run cargo test
-        env:
-          NEXTEST_RETRIES: 3
         run: |
-          cargo nextest run $CARGO_FEATURES
+          cargo test $CARGO_FLAGS $CARGO_FEATURES
 
           # Run separate tests for real S3
           export ENABLE_REAL_S3_REMOTE_STORAGE=nonempty
           export REMOTE_STORAGE_S3_BUCKET=neon-github-ci-tests
           export REMOTE_STORAGE_S3_REGION=eu-central-1
           # Avoid `$CARGO_FEATURES` since there's no `testing` feature in the e2e tests now
-          cargo nextest run --package remote_storage --test test_real_s3
+          cargo test $CARGO_FLAGS --package remote_storage --test test_real_s3
 
           # Run separate tests for real Azure Blob Storage
           # XXX: replace region with `eu-central-1`-like region
@@ -233,7 +231,7 @@ jobs:
           export REMOTE_STORAGE_AZURE_CONTAINER="${{ vars.REMOTE_STORAGE_AZURE_CONTAINER }}"
           export REMOTE_STORAGE_AZURE_REGION="${{ vars.REMOTE_STORAGE_AZURE_REGION }}"
           # Avoid `$CARGO_FEATURES` since there's no `testing` feature in the e2e tests now
-          cargo nextest run --package remote_storage --test test_real_azure
+          cargo test $CARGO_FLAGS --package remote_storage --test test_real_azure
 
   check-codestyle-rust-arm:
     timeout-minutes: 90

.github/workflows/update_build_tools_image.yml

@@ -20,51 +20,111 @@ defaults:
   run:
     shell: bash -euo pipefail {0}
 
+env:
+  AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_DEV }}
+  AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_KEY_DEV }}
+
 permissions: {}
 
 jobs:
   tag-image:
     runs-on: [ self-hosted, gen3, small ]
+    container: golang:1.19-bullseye
 
     env:
-      ECR_IMAGE: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools
-      DOCKER_HUB_IMAGE: docker.io/neondatabase/build-tools
+      IMAGE: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools
+      FROM_TAG: ${{ inputs.from-tag }}
+      TO_TAG: ${{ inputs.to-tag }}
+    outputs:
+      next-digest-buildtools: ${{ steps.next-digest.outputs.next-digest-buildtools }}
+      prev-digest-buildtools: ${{ steps.prev-digest.outputs.prev-digest-buildtools }}
+
+    steps:
+      - name: Install Crane & ECR helper
+        run: |
+          go install github.com/google/go-containerregistry/cmd/crane@a54d64203cffcbf94146e04069aae4a97f228ee2 # v0.16.1
+          go install github.com/awslabs/amazon-ecr-credential-helper/ecr-login/cli/docker-credential-ecr-login@adf1bafd791ae7d4ff098108b1e91f36a4da5404 # v0.7.1
+
+      - name: Configure ECR login
+        run: |
+          mkdir /github/home/.docker/
+          echo "{\"credsStore\":\"ecr-login\"}" > /github/home/.docker/config.json
+
+      - name: Get source image digest
+        id: next-digest
+        run: |
+          NEXT_DIGEST=$(crane digest ${IMAGE}:${FROM_TAG} || true)
+          if [ -z "${NEXT_DIGEST}" ]; then
+            echo >&2 "Image ${IMAGE}:${FROM_TAG} does not exist"
+            exit 1
+          fi
+
+          echo "Current ${IMAGE}@${FROM_TAG} image is ${IMAGE}@${NEXT_DIGEST}"
+          echo "next-digest-buildtools=$NEXT_DIGEST" >> $GITHUB_OUTPUT
+
+      - name: Get destination image digest (if already exists)
+        id: prev-digest
+        run: |
+          PREV_DIGEST=$(crane digest ${IMAGE}:${TO_TAG} || true)
+          if [ -z "${PREV_DIGEST}" ]; then
+            echo >&2 "Image ${IMAGE}:${TO_TAG} does not exist (it's ok)"
+          else
+            echo >&2 "Current ${IMAGE}@${TO_TAG} image is ${IMAGE}@${PREV_DIGEST}"
+
+            echo "prev-digest-buildtools=$PREV_DIGEST" >> $GITHUB_OUTPUT
+          fi
+
+      - name: Tag image
+        run: |
+          crane tag "${IMAGE}:${FROM_TAG}" "${TO_TAG}"
+
+  rollback-tag-image:
+    needs:  tag-image
+    if: ${{ !success() }}
+
+    runs-on: [ self-hosted, gen3, small ]
+    container: golang:1.19-bullseye
+
+    env:
+      IMAGE: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools
       FROM_TAG: ${{ inputs.from-tag }}
       TO_TAG: ${{ inputs.to-tag }}
 
     steps:
-      # Use custom DOCKER_CONFIG directory to avoid conflicts with default settings
-      # The default value is ~/.docker
-      - name: Set custom docker config directory
+      - name: Install Crane & ECR helper
         run: |
-          mkdir -p .docker-custom
-          echo DOCKER_CONFIG=$(pwd)/.docker-custom >> $GITHUB_ENV
+          go install github.com/google/go-containerregistry/cmd/crane@a54d64203cffcbf94146e04069aae4a97f228ee2 # v0.16.1
+          go install github.com/awslabs/amazon-ecr-credential-helper/ecr-login/cli/docker-credential-ecr-login@adf1bafd791ae7d4ff098108b1e91f36a4da5404 # v0.7.1
 
-      - uses: docker/login-action@v2
-        with:
-          username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
-          password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
-
-      - uses: docker/login-action@v2
-        with:
-          registry: 369495373322.dkr.ecr.eu-central-1.amazonaws.com
-          username: ${{ secrets.AWS_ACCESS_KEY_DEV }}
-          password: ${{ secrets.AWS_SECRET_KEY_DEV }}
-
-      - uses: actions/setup-go@v5
-        with:
-          go-version: '1.21'
-
-      - name: Install crane
+      - name: Configure ECR login
         run: |
-          go install github.com/google/go-containerregistry/cmd/crane@a0658aa1d0cc7a7f1bcc4a3af9155335b6943f40 # v0.18.0
+          mkdir /github/home/.docker/
+          echo "{\"credsStore\":\"ecr-login\"}" > /github/home/.docker/config.json
 
-      - name: Copy images
+      - name: Restore previous tag if needed
         run: |
-          crane copy "${ECR_IMAGE}:${FROM_TAG}" "${ECR_IMAGE}:${TO_TAG}"
-          crane copy "${ECR_IMAGE}:${FROM_TAG}" "${DOCKER_HUB_IMAGE}:${TO_TAG}"
+          NEXT_DIGEST="${{ needs.tag-image.outputs.next-digest-buildtools }}"
+          PREV_DIGEST="${{ needs.tag-image.outputs.prev-digest-buildtools }}"
 
-      - name: Remove custom docker config directory
-        if: always()
-        run: |
-          rm -rf .docker-custom
+          if [ -z "${NEXT_DIGEST}" ]; then
+            echo >&2 "Image ${IMAGE}:${FROM_TAG} does not exist, nothing to rollback"
+            exit 0
+          fi
+
+          if [ -z "${PREV_DIGEST}" ]; then
+            # I guess we should delete the tag here/untag the image, but crane does not support it
+            # - https://github.com/google/go-containerregistry/issues/999
+
+            echo >&2 "Image ${IMAGE}:${TO_TAG} did not exist, but it was created by the job, no need to rollback"
+
+            exit 0
+          fi
+
+          CURRENT_DIGEST=$(crane digest "${IMAGE}:${TO_TAG}")
+          if [ "${CURRENT_DIGEST}" == "${NEXT_DIGEST}" ]; then
+            crane tag "${IMAGE}@${PREV_DIGEST}" "${TO_TAG}"
+
+            echo >&2 "Successfully restored ${TO_TAG} tag from ${IMAGE}@${CURRENT_DIGEST} to ${IMAGE}@${PREV_DIGEST}"
+          else
+            echo >&2 "Image ${IMAGE}:${TO_TAG}@${CURRENT_DIGEST} is not required to be restored"
+          fi

CONTRIBUTING.md

@@ -20,7 +20,7 @@ ln -s ../../pre-commit.py .git/hooks/pre-commit
 
 This will run following checks on staged files before each commit:
 - `rustfmt`
-- checks for Python files, see [obligatory checks](/docs/sourcetree.md#obligatory-checks).
+- checks for python files, see [obligatory checks](/docs/sourcetree.md#obligatory-checks).
 
 There is also a separate script `./run_clippy.sh` that runs `cargo clippy` on the whole project
 and `./scripts/reformat` that runs all formatting tools to ensure the project is up to date.

Cargo.toml

@@ -48,12 +48,11 @@ azure_storage_blobs = "0.18"
 flate2 = "1.0.26"
 async-stream = "0.3"
 async-trait = "0.1"
-aws-config = { version = "1.1.4", default-features = false, features=["rustls"] }
-aws-sdk-s3 = "1.14"
-aws-sdk-secretsmanager = { version = "1.14.0" }
-aws-smithy-async = { version = "1.1.4", default-features = false, features=["rt-tokio"] }
-aws-smithy-types = "1.1.4"
-aws-credential-types = "1.1.4"
+aws-config = { version = "1.0", default-features = false, features=["rustls"] }
+aws-sdk-s3 = "1.0"
+aws-smithy-async = { version = "1.0", default-features = false, features=["rt-tokio"] }
+aws-smithy-types = "1.0"
+aws-credential-types = "1.0"
 axum = { version = "0.6.20", features = ["ws"] }
 base64 = "0.13.0"
 bincode = "1.3"
@@ -65,6 +64,7 @@ camino = "1.1.6"
 cfg-if = "1.0.0"
 chrono = { version = "0.4", default-features = false, features = ["clock"] }
 clap = { version = "4.0", features = ["derive"] }
+close_fds = "0.3.2"
 comfy-table = "6.1"
 const_format = "0.2"
 crc32c = "0.6"
@@ -95,7 +95,6 @@ inotify = "0.10.2"
 ipnet = "2.9.0"
 itertools = "0.10"
 jsonwebtoken = "9"
-lasso = "0.7"
 libc = "0.2"
 md5 = "0.7.0"
 memoffset = "0.8"
@@ -150,11 +149,8 @@ tar = "0.4"
 task-local-extensions = "0.1.4"
 test-context = "0.1"
 thiserror = "1.0"
-tikv-jemallocator = "0.5"
-tikv-jemalloc-ctl = "0.5"
 tls-listener = { version = "0.7", features = ["rustls", "hyper-h1"] }
 tokio = { version = "1.17", features = ["macros"] }
-tokio-epoll-uring = { git = "https://github.com/neondatabase/tokio-epoll-uring.git" , branch = "main" }
 tokio-io-timeout = "1.2.0"
 tokio-postgres-rustls = "0.10.0"
 tokio-rustls = "0.24"
@@ -168,7 +164,6 @@ tracing = "0.1"
 tracing-error = "0.2.0"
 tracing-opentelemetry = "0.20.0"
 tracing-subscriber = { version = "0.3", default_features = false, features = ["smallvec", "fmt", "tracing-log", "std", "env-filter", "json"] }
-twox-hash = { version = "1.6.3", default-features = false }
 url = "2.2"
 uuid = { version = "1.6.1", features = ["v4", "v7", "serde"] }
 walkdir = "2.3.2"

Dockerfile

@@ -53,7 +53,6 @@ RUN set -e \
       --bin pagectl  \
       --bin safekeeper  \
       --bin storage_broker  \
-      --bin attachment_service  \
       --bin proxy  \
       --bin neon_local \
       --locked --release \
@@ -81,7 +80,6 @@ COPY --from=build --chown=neon:neon /home/nonroot/target/release/pageserver
 COPY --from=build --chown=neon:neon /home/nonroot/target/release/pagectl             /usr/local/bin
 COPY --from=build --chown=neon:neon /home/nonroot/target/release/safekeeper          /usr/local/bin
 COPY --from=build --chown=neon:neon /home/nonroot/target/release/storage_broker      /usr/local/bin
-COPY --from=build --chown=neon:neon /home/nonroot/target/release/attachment_service  /usr/local/bin
 COPY --from=build --chown=neon:neon /home/nonroot/target/release/proxy               /usr/local/bin
 COPY --from=build --chown=neon:neon /home/nonroot/target/release/neon_local          /usr/local/bin
 

Dockerfile.compute-node

@@ -52,7 +52,7 @@ RUN cd postgres && \
     # We need to grant EXECUTE on pg_stat_statements_reset() to neon_superuser.
     # In vanilla postgres this function is limited to Postgres role superuser.
     # In neon we have neon_superuser role that is not a superuser but replaces superuser in some cases.
-    # We could add the additional grant statements to the postgres repository but it would be hard to maintain,
+    # We could add the additional grant statements to the postgres repository but it would be hard to maintain, 
     # whenever we need to pick up a new postgres version and we want to limit the changes in our postgres fork,
     # so we do it here.
     old_list="pg_stat_statements--1.0--1.1.sql pg_stat_statements--1.1--1.2.sql pg_stat_statements--1.2--1.3.sql pg_stat_statements--1.3--1.4.sql pg_stat_statements--1.4--1.5.sql pg_stat_statements--1.4.sql pg_stat_statements--1.5--1.6.sql"; \
@@ -63,14 +63,14 @@ RUN cd postgres && \
             echo 'GRANT EXECUTE ON FUNCTION pg_stat_statements_reset() TO neon_superuser;' >> $file; \
         fi; \
     done; \
-    # the second loop is for pg_stat_statement extension versions >= 1.7,
+    # the second loop is for pg_stat_statement extension versions >= 1.7, 
     # where pg_stat_statement_reset() got 3 additional arguments
     for file in /usr/local/pgsql/share/extension/pg_stat_statements--*.sql; do \
         filename=$(basename "$file"); \
         if ! echo "$old_list" | grep -q -F "$filename"; then \
             echo 'GRANT EXECUTE ON FUNCTION pg_stat_statements_reset(Oid, Oid, bigint) TO neon_superuser;' >> $file; \
         fi; \
-    done
+    done      
 
 #########################################################################################
 #
@@ -144,23 +144,30 @@ RUN wget https://github.com/pgRouting/pgrouting/archive/v3.4.2.tar.gz -O pgrouti
 FROM build-deps AS plv8-build
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
+ARG PG_VERSION
 RUN apt update && \
     apt install -y ninja-build python3-dev libncurses5 binutils clang
 
-RUN wget https://github.com/plv8/plv8/archive/refs/tags/v3.1.10.tar.gz -O plv8.tar.gz && \
-    echo "7096c3290928561f0d4901b7a52794295dc47f6303102fae3f8e42dd575ad97d plv8.tar.gz" | sha256sum --check && \
+RUN case "${PG_VERSION}" in \
+      "v14" | "v15") \
+        export PLV8_VERSION=3.1.5 \
+        export PLV8_CHECKSUM=1e108d5df639e4c189e1c5bdfa2432a521c126ca89e7e5a969d46899ca7bf106 \
+        ;; \
+      "v16") \
+        export PLV8_VERSION=3.1.8 \
+        export PLV8_CHECKSUM=92b10c7db39afdae97ff748c9ec54713826af222c459084ad002571b79eb3f49 \
+        ;; \
+      *) \
+        echo "Export the valid PG_VERSION variable" && exit 1 \
+        ;; \
+    esac && \
+    wget https://github.com/plv8/plv8/archive/refs/tags/v${PLV8_VERSION}.tar.gz -O plv8.tar.gz && \
+    echo "${PLV8_CHECKSUM} plv8.tar.gz" | sha256sum --check && \
     mkdir plv8-src && cd plv8-src && tar xvzf ../plv8.tar.gz --strip-components=1 -C . && \
-    # generate and copy upgrade scripts
-    mkdir -p upgrade && ./generate_upgrade.sh 3.1.10 && \
-    cp upgrade/* /usr/local/pgsql/share/extension/ && \
     export PATH="/usr/local/pgsql/bin:$PATH" && \
     make DOCKER=1 -j $(getconf _NPROCESSORS_ONLN) install && \
     rm -rf /plv8-* && \
     find /usr/local/pgsql/ -name "plv8-*.so" | xargs strip && \
-    # don't break computes with installed old version of plv8
-    cd /usr/local/pgsql/lib/ && \
-    ln -s plv8-3.1.10.so plv8-3.1.5.so && \
-    ln -s plv8-3.1.10.so plv8-3.1.8.so && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/plv8.control && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/plcoffee.control && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/plls.control
@@ -520,7 +527,8 @@ RUN apt-get update && \
         libboost-regex1.74-dev \
         libboost-serialization1.74-dev \
         libboost-system1.74-dev \
-        libeigen3-dev
+        libeigen3-dev \
+        libfreetype6-dev
 
 ENV PATH "/usr/local/pgsql/bin/:/usr/local/pgsql/:$PATH"
 RUN wget https://github.com/rdkit/rdkit/archive/refs/tags/Release_2023_03_3.tar.gz -O rdkit.tar.gz && \
@@ -545,8 +553,6 @@ RUN wget https://github.com/rdkit/rdkit/archive/refs/tags/Release_2023_03_3.tar.
         -D PostgreSQL_TYPE_INCLUDE_DIR=`pg_config --includedir-server` \
         -D PostgreSQL_LIBRARY_DIR=`pg_config --libdir` \
         -D RDK_INSTALL_INTREE=OFF \
-        -D RDK_INSTALL_COMIC_FONTS=OFF \
-        -D RDK_BUILD_FREETYPE_SUPPORT=OFF \
         -D CMAKE_BUILD_TYPE=Release \
         . && \
     make -j $(getconf _NPROCESSORS_ONLN) && \
@@ -901,7 +907,7 @@ COPY --from=compute-tools --chown=postgres /home/nonroot/target/release-line-deb
 # libgeos, libgdal, libsfcgal1, libproj and libprotobuf-c1 for PostGIS
 # libxml2, libxslt1.1 for xml2
 # libzstd1 for zstd
-# libboost* for rdkit
+# libboost*, libfreetype6, and zlib1g for rdkit
 # ca-certificates for communicating with s3 by compute_ctl
 RUN apt update &&  \
     apt install --no-install-recommends -y \
@@ -914,6 +920,7 @@ RUN apt update &&  \
         libboost-serialization1.74.0 \
         libboost-system1.74.0 \
         libossp-uuid16 \
+        libfreetype6 \
         libgeos-c1v5 \
         libgdal28 \
         libproj19 \
@@ -925,6 +932,7 @@ RUN apt update &&  \
         libcurl4-openssl-dev \
         locales \
         procps \
+        zlib1g \
         ca-certificates && \
     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

Makefile

@@ -51,8 +51,6 @@ CARGO_BUILD_FLAGS += $(filter -j1,$(MAKEFLAGS))
 CARGO_CMD_PREFIX += $(if $(filter n,$(MAKEFLAGS)),,+)
 # Force cargo not to print progress bar
 CARGO_CMD_PREFIX += CARGO_TERM_PROGRESS_WHEN=never CI=1
-# Set PQ_LIB_DIR to make sure `attachment_service` get linked with bundled libpq (through diesel)
-CARGO_CMD_PREFIX += PQ_LIB_DIR=$(POSTGRES_INSTALL_DIR)/v16/lib
 
 #
 # Top level Makefile to build Neon and PostgreSQL
@@ -176,10 +174,10 @@ neon-pg-ext-clean-%:
 
 # Build walproposer as a static library. walproposer source code is located
 # in the pgxn/neon directory.
-#
+# 
 # We also need to include libpgport.a and libpgcommon.a, because walproposer
 # uses some functions from those libraries.
-#
+# 
 # Some object files are removed from libpgport.a and libpgcommon.a because
 # they depend on openssl and other libraries that are not included in our
 # Rust build.

README.md

@@ -14,8 +14,8 @@ Alternatively, compile and run the project [locally](#running-local-installation
 A Neon installation consists of compute nodes and the Neon storage engine. Compute nodes are stateless PostgreSQL nodes backed by the Neon storage engine.
 
 The Neon storage engine consists of two major components:
-- Pageserver: Scalable storage backend for the compute nodes.
-- Safekeepers: The safekeepers form a redundant WAL service that received WAL from the compute node, and stores it durably until it has been processed by the pageserver and uploaded to cloud storage.
+- Pageserver. Scalable storage backend for the compute nodes.
+- Safekeepers. The safekeepers form a redundant WAL service that received WAL from the compute node, and stores it durably until it has been processed by the pageserver and uploaded to cloud storage.
 
 See developer documentation in [SUMMARY.md](/docs/SUMMARY.md) for more information.
 
@@ -81,9 +81,9 @@ The project uses [rust toolchain file](./rust-toolchain.toml) to define the vers
 
 This file is automatically picked up by [`rustup`](https://rust-lang.github.io/rustup/overrides.html#the-toolchain-file) that installs (if absent) and uses the toolchain version pinned in the file.
 
-rustup users who want to build with another toolchain can use the [`rustup override`](https://rust-lang.github.io/rustup/overrides.html#directory-overrides) command to set a specific toolchain for the project's directory.
+rustup users who want to build with another toolchain can use [`rustup override`](https://rust-lang.github.io/rustup/overrides.html#directory-overrides) command to set a specific toolchain for the project's directory.
 
-non-rustup users most probably are not getting the same toolchain automatically from the file, so are responsible to manually verify that their toolchain matches the version in the file.
+non-rustup users most probably are not getting the same toolchain automatically from the file, so are responsible to manually verify their toolchain matches the version in the file.
 Newer rustc versions most probably will work fine, yet older ones might not be supported due to some new features used by the project or the crates.
 
 #### Building on Linux
@@ -124,7 +124,7 @@ make -j`sysctl -n hw.logicalcpu` -s
 To run the `psql` client, install the `postgresql-client` package or modify `PATH` and `LD_LIBRARY_PATH` to include `pg_install/bin` and `pg_install/lib`, respectively.
 
 To run the integration tests or Python scripts (not required to use the code), install
-Python (3.9 or higher), and install the python3 packages using `./scripts/pysync` (requires [poetry>=1.3](https://python-poetry.org/)) in the project directory.
+Python (3.9 or higher), and install python3 packages using `./scripts/pysync` (requires [poetry>=1.3](https://python-poetry.org/)) in the project directory.
 
 
 #### Running neon database
@@ -166,7 +166,7 @@ Starting postgres at 'postgresql://cloud_admin@127.0.0.1:55432/postgres'
 
 2. Now, it is possible to connect to postgres and run some queries:
 ```text
-> psql -p 55432 -h 127.0.0.1 -U cloud_admin postgres
+> psql -p55432 -h 127.0.0.1 -U cloud_admin postgres
 postgres=# CREATE TABLE t(key int primary key, value text);
 CREATE TABLE
 postgres=# insert into t values(1,1);
@@ -205,7 +205,7 @@ Starting postgres at 'postgresql://cloud_admin@127.0.0.1:55434/postgres'
 
 # this new postgres instance will have all the data from 'main' postgres,
 # but all modifications would not affect data in original postgres
-> psql -p 55434 -h 127.0.0.1 -U cloud_admin postgres
+> psql -p55434 -h 127.0.0.1 -U cloud_admin postgres
 postgres=# select * from t;
  key | value
 -----+-------
@@ -216,7 +216,7 @@ postgres=# insert into t values(2,2);
 INSERT 0 1
 
 # check that the new change doesn't affect the 'main' postgres
-> psql -p 55432 -h 127.0.0.1 -U cloud_admin postgres
+> psql -p55432 -h 127.0.0.1 -U cloud_admin postgres
 postgres=# select * from t;
  key | value
 -----+-------
@@ -224,7 +224,7 @@ postgres=# select * from t;
 (1 row)
 ```
 
-4. If you want to run tests afterwards (see below), you must stop all the running pageserver, safekeeper, and postgres instances
+4. If you want to run tests afterward (see below), you must stop all the running of the pageserver, safekeeper, and postgres instances
    you have just started. You can terminate them all with one command:
 ```sh
 > cargo neon stop
@@ -243,7 +243,7 @@ CARGO_BUILD_FLAGS="--features=testing" make
 ```
 
 By default, this runs both debug and release modes, and all supported postgres versions. When
-testing locally, it is convenient to run just one set of permutations, like this:
+testing locally, it is convenient to run just run one set of permutations, like this:
 
 ```sh
 DEFAULT_PG_VERSION=15 BUILD_TYPE=release ./scripts/pytest

compute_tools/src/compute.rs

@@ -207,7 +207,6 @@ fn maybe_cgexec(cmd: &str) -> Command {
 
 /// Create special neon_superuser role, that's a slightly nerfed version of a real superuser
 /// that we give to customers
-#[instrument(skip_all)]
 fn create_neon_superuser(spec: &ComputeSpec, client: &mut Client) -> Result<()> {
     let roles = spec
         .cluster
@@ -320,7 +319,7 @@ impl ComputeNode {
     // Get basebackup from the libpq connection to pageserver using `connstr` and
     // unarchive it to `pgdata` directory overriding all its previous content.
     #[instrument(skip_all, fields(%lsn))]
-    fn try_get_basebackup(&self, compute_state: &ComputeState, lsn: Lsn) -> Result<()> {
+    fn get_basebackup(&self, compute_state: &ComputeState, lsn: Lsn) -> Result<()> {
         let spec = compute_state.pspec.as_ref().expect("spec must be set");
         let start_time = Instant::now();
 
@@ -391,34 +390,6 @@ impl ComputeNode {
         Ok(())
     }
 
-    // Gets the basebackup in a retry loop
-    #[instrument(skip_all, fields(%lsn))]
-    pub fn get_basebackup(&self, compute_state: &ComputeState, lsn: Lsn) -> Result<()> {
-        let mut retry_period_ms = 500;
-        let mut attempts = 0;
-        let max_attempts = 5;
-        loop {
-            let result = self.try_get_basebackup(compute_state, lsn);
-            match result {
-                Ok(_) => {
-                    return result;
-                }
-                Err(ref e) if attempts < max_attempts => {
-                    warn!(
-                        "Failed to get basebackup: {} (attempt {}/{})",
-                        e, attempts, max_attempts
-                    );
-                    std::thread::sleep(std::time::Duration::from_millis(retry_period_ms));
-                    retry_period_ms *= 2;
-                }
-                Err(_) => {
-                    return result;
-                }
-            }
-            attempts += 1;
-        }
-    }
-
     pub async fn check_safekeepers_synced_async(
         &self,
         compute_state: &ComputeState,

compute_tools/src/monitor.rs

@@ -138,34 +138,6 @@ fn watch_compute_activity(compute: &ComputeNode) {
                     }
                 }
                 //
-                // Don't suspend compute if there is an active logical replication subscription
-                //
-                // `where pid is not null` – to filter out read only computes and subscription on branches
-                //
-                let logical_subscriptions_query =
-                    "select count(*) from pg_stat_subscription where pid is not null;";
-                match cli.query_one(logical_subscriptions_query, &[]) {
-                    Ok(row) => match row.try_get::<&str, i64>("count") {
-                        Ok(num_subscribers) => {
-                            if num_subscribers > 0 {
-                                compute.update_last_active(Some(Utc::now()));
-                                continue;
-                            }
-                        }
-                        Err(e) => {
-                            warn!("failed to parse `pg_stat_subscription` count: {:?}", e);
-                            continue;
-                        }
-                    },
-                    Err(e) => {
-                        warn!(
-                            "failed to get list of active logical replication subscriptions: {:?}",
-                            e
-                        );
-                        continue;
-                    }
-                }
-                //
                 // Do not suspend compute if autovacuum is running
                 //
                 let autovacuum_count_query = "select count(*) from pg_stat_activity where backend_type = 'autovacuum worker'";

compute_tools/src/spec.rs

@@ -758,14 +758,6 @@ BEGIN
     END LOOP;
 END $$;
 "#,
-        r#"
-DO $$
-BEGIN
-    IF (SELECT setting::numeric >= 160000 FROM pg_settings WHERE name = 'server_version_num') THEN
-        EXECUTE 'GRANT pg_create_subscription TO neon_superuser';
-    END IF;
-END
-$$;"#,
     ];
 
     let mut query = "CREATE SCHEMA IF NOT EXISTS neon_migration";

control_plane/Cargo.toml

@@ -10,8 +10,6 @@ async-trait.workspace = true
 camino.workspace = true
 clap.workspace = true
 comfy-table.workspace = true
-diesel = { version = "2.1.4", features = ["postgres"]}
-diesel_migrations = { version = "2.1.0", features = ["postgres"]}
 futures.workspace = true
 git-version.workspace = true
 nix.workspace = true
@@ -21,7 +19,6 @@ hex.workspace = true
 hyper.workspace = true
 regex.workspace = true
 reqwest = { workspace = true, features = ["blocking", "json"] }
-scopeguard.workspace = true
 serde.workspace = true
 serde_json.workspace = true
 serde_with.workspace = true

control_plane/attachment_service/Cargo.toml

@@ -6,8 +6,6 @@ license.workspace = true
 
 [dependencies]
 anyhow.workspace = true
-aws-config.workspace = true
-aws-sdk-secretsmanager.workspace = true
 camino.workspace = true
 clap.workspace = true
 futures.workspace = true
@@ -16,7 +14,6 @@ hyper.workspace = true
 pageserver_api.workspace = true
 pageserver_client.workspace = true
 postgres_connection.workspace = true
-reqwest.workspace = true
 serde.workspace = true
 serde_json.workspace = true
 thiserror.workspace = true
@@ -24,8 +21,9 @@ tokio.workspace = true
 tokio-util.workspace = true
 tracing.workspace = true
 
-diesel = { version = "2.1.4", features = ["serde_json", "postgres"] }
-diesel_migrations = { version = "2.1.0" }
+# TODO: remove this after DB persistence is added, it is only used for
+# a parsing function when loading pageservers from neon_local LocalEnv
+postgres_backend.workspace = true
 
 utils = { path = "../../libs/utils/" }
 metrics = { path = "../../libs/metrics/" }

control_plane/attachment_service/migrations/00000000000000_diesel_initial_setup/down.sql

@@ -1,6 +0,0 @@
--- This file was automatically created by Diesel to setup helper functions
--- and other internal bookkeeping. This file is safe to edit, any future
--- changes will be added to existing projects as new migrations.
-
-DROP FUNCTION IF EXISTS diesel_manage_updated_at(_tbl regclass);
-DROP FUNCTION IF EXISTS diesel_set_updated_at();

control_plane/attachment_service/migrations/00000000000000_diesel_initial_setup/up.sql

@@ -1,36 +0,0 @@
--- This file was automatically created by Diesel to setup helper functions
--- and other internal bookkeeping. This file is safe to edit, any future
--- changes will be added to existing projects as new migrations.
-
-
-
-
--- Sets up a trigger for the given table to automatically set a column called
--- `updated_at` whenever the row is modified (unless `updated_at` was included
--- in the modified columns)
---
--- # Example
---
--- ```sql
--- CREATE TABLE users (id SERIAL PRIMARY KEY, updated_at TIMESTAMP NOT NULL DEFAULT NOW());
---
--- SELECT diesel_manage_updated_at('users');
--- ```
-CREATE OR REPLACE FUNCTION diesel_manage_updated_at(_tbl regclass) RETURNS VOID AS $$
-BEGIN
-    EXECUTE format('CREATE TRIGGER set_updated_at BEFORE UPDATE ON %s
-                    FOR EACH ROW EXECUTE PROCEDURE diesel_set_updated_at()', _tbl);
-END;
-$$ LANGUAGE plpgsql;
-
-CREATE OR REPLACE FUNCTION diesel_set_updated_at() RETURNS trigger AS $$
-BEGIN
-    IF (
-        NEW IS DISTINCT FROM OLD AND
-        NEW.updated_at IS NOT DISTINCT FROM OLD.updated_at
-    ) THEN
-        NEW.updated_at := current_timestamp;
-    END IF;
-    RETURN NEW;
-END;
-$$ LANGUAGE plpgsql;

control_plane/attachment_service/migrations/2024-01-07-211257_create_tenant_shards/down.sql

@@ -1 +0,0 @@
-DROP TABLE tenant_shards;

control_plane/attachment_service/migrations/2024-01-07-211257_create_tenant_shards/up.sql

@@ -1,12 +0,0 @@
-CREATE TABLE tenant_shards (
-  tenant_id VARCHAR NOT NULL,
-  shard_number INTEGER NOT NULL,
-  shard_count INTEGER NOT NULL,
-  PRIMARY KEY(tenant_id, shard_number, shard_count),
-  shard_stripe_size INTEGER NOT NULL,
-  generation INTEGER NOT NULL,
-  generation_pageserver BIGINT NOT NULL,
-  placement_policy VARCHAR NOT NULL,
-  -- config is JSON encoded, opaque to the database.
-  config TEXT NOT NULL
-);

control_plane/attachment_service/migrations/2024-01-07-212945_create_nodes/down.sql

@@ -1 +0,0 @@
-DROP TABLE nodes;

control_plane/attachment_service/migrations/2024-01-07-212945_create_nodes/up.sql

@@ -1,10 +0,0 @@
-CREATE TABLE nodes (
-  node_id BIGINT PRIMARY KEY NOT NULL,
-
-  scheduling_policy VARCHAR NOT NULL,
-
-  listen_http_addr VARCHAR NOT NULL,
-  listen_http_port INTEGER NOT NULL,
-  listen_pg_addr VARCHAR NOT NULL,
-  listen_pg_port INTEGER NOT NULL
-);

control_plane/attachment_service/src/compute_hook.rs

@@ -1,76 +1,24 @@
-use std::{collections::HashMap, time::Duration};
+use std::collections::HashMap;
 
-use control_plane::endpoint::{ComputeControlPlane, EndpointStatus};
+use control_plane::endpoint::ComputeControlPlane;
 use control_plane::local_env::LocalEnv;
-use hyper::{Method, StatusCode};
-use pageserver_api::shard::{ShardCount, ShardIndex, ShardNumber, TenantShardId};
+use pageserver_api::shard::{ShardCount, ShardIndex, TenantShardId};
 use postgres_connection::parse_host_port;
-use serde::{Deserialize, Serialize};
-use tokio_util::sync::CancellationToken;
-use utils::{
-    backoff::{self},
-    id::{NodeId, TenantId},
-};
-
-use crate::service::Config;
-
-const BUSY_DELAY: Duration = Duration::from_secs(1);
-const SLOWDOWN_DELAY: Duration = Duration::from_secs(5);
-
-pub(crate) const API_CONCURRENCY: usize = 32;
+use utils::id::{NodeId, TenantId};
 
 pub(super) struct ComputeHookTenant {
     shards: Vec<(ShardIndex, NodeId)>,
 }
 
-#[derive(Serialize, Deserialize, Debug)]
-struct ComputeHookNotifyRequestShard {
-    node_id: NodeId,
-    shard_number: ShardNumber,
-}
-
-/// Request body that we send to the control plane to notify it of where a tenant is attached
-#[derive(Serialize, Deserialize, Debug)]
-struct ComputeHookNotifyRequest {
-    tenant_id: TenantId,
-    shards: Vec<ComputeHookNotifyRequestShard>,
-}
-
-/// Error type for attempts to call into the control plane compute notification hook
-#[derive(thiserror::Error, Debug)]
-pub(crate) enum NotifyError {
-    // Request was not send successfully, e.g. transport error
-    #[error("Sending request: {0}")]
-    Request(#[from] reqwest::Error),
-    // Request could not be serviced right now due to ongoing Operation in control plane, but should be possible soon.
-    #[error("Control plane tenant busy")]
-    Busy,
-    // Explicit 429 response asking us to retry less frequently
-    #[error("Control plane overloaded")]
-    SlowDown,
-    // A 503 response indicates the control plane can't handle the request right now
-    #[error("Control plane unavailable (status {0})")]
-    Unavailable(StatusCode),
-    // API returned unexpected non-success status.  We will retry, but log a warning.
-    #[error("Control plane returned unexpected status {0}")]
-    Unexpected(StatusCode),
-    // We shutdown while sending
-    #[error("Shutting down")]
-    ShuttingDown,
-    // A response indicates we will never succeed, such as 400 or 404
-    #[error("Non-retryable error {0}")]
-    Fatal(StatusCode),
-}
-
 impl ComputeHookTenant {
-    async fn maybe_reconfigure(&mut self, tenant_id: TenantId) -> Option<ComputeHookNotifyRequest> {
+    pub(super) async fn maybe_reconfigure(&mut self, tenant_id: TenantId) -> anyhow::Result<()> {
         // Find the highest shard count and drop any shards that aren't
         // for that shard count.
         let shard_count = self.shards.iter().map(|(k, _v)| k.shard_count).max();
         let Some(shard_count) = shard_count else {
             // No shards, nothing to do.
             tracing::info!("ComputeHookTenant::maybe_reconfigure: no shards");
-            return None;
+            return Ok(());
         };
 
         self.shards.retain(|(k, _v)| k.shard_count == shard_count);
@@ -78,18 +26,38 @@ impl ComputeHookTenant {
             .sort_by_key(|(shard, _node_id)| shard.shard_number);
 
         if self.shards.len() == shard_count.0 as usize || shard_count == ShardCount(0) {
-            // We have pageservers for all the shards: emit a configuration update
-            return Some(ComputeHookNotifyRequest {
-                tenant_id,
-                shards: self
-                    .shards
-                    .iter()
-                    .map(|(shard, node_id)| ComputeHookNotifyRequestShard {
-                        shard_number: shard.shard_number,
-                        node_id: *node_id,
-                    })
-                    .collect(),
-            });
+            // We have pageservers for all the shards: proceed to reconfigure compute
+            let env = match LocalEnv::load_config() {
+                Ok(e) => e,
+                Err(e) => {
+                    tracing::warn!(
+                        "Couldn't load neon_local config, skipping compute update ({e})"
+                    );
+                    return Ok(());
+                }
+            };
+            let cplane = ComputeControlPlane::load(env.clone())
+                .expect("Error loading compute control plane");
+
+            let compute_pageservers = self
+                .shards
+                .iter()
+                .map(|(_shard, node_id)| {
+                    let ps_conf = env
+                        .get_pageserver_conf(*node_id)
+                        .expect("Unknown pageserver");
+                    let (pg_host, pg_port) = parse_host_port(&ps_conf.listen_pg_addr)
+                        .expect("Unable to parse listen_pg_addr");
+                    (pg_host, pg_port.unwrap_or(5432))
+                })
+                .collect::<Vec<_>>();
+
+            for (endpoint_name, endpoint) in &cplane.endpoints {
+                if endpoint.tenant_id == tenant_id && endpoint.status() == "running" {
+                    tracing::info!("🔁 Reconfiguring endpoint {}", endpoint_name,);
+                    endpoint.reconfigure(compute_pageservers.clone()).await?;
+                }
+            }
         } else {
             tracing::info!(
                 "ComputeHookTenant::maybe_reconfigure: not enough shards ({}/{})",
@@ -98,7 +66,7 @@ impl ComputeHookTenant {
             );
         }
 
-        None
+        Ok(())
     }
 }
 
@@ -106,173 +74,22 @@ impl ComputeHookTenant {
 /// mapping.  It aggregates updates for the shards in a tenant, and when appropriate reconfigures
 /// the compute connection string.
 pub(super) struct ComputeHook {
-    config: Config,
     state: tokio::sync::Mutex<HashMap<TenantId, ComputeHookTenant>>,
-    authorization_header: Option<String>,
 }
 
 impl ComputeHook {
-    pub(super) fn new(config: Config) -> Self {
-        let authorization_header = config
-            .control_plane_jwt_token
-            .clone()
-            .map(|jwt| format!("Bearer {}", jwt));
-
+    pub(super) fn new() -> Self {
         Self {
             state: Default::default(),
-            config,
-            authorization_header,
         }
     }
 
-    /// For test environments: use neon_local's LocalEnv to update compute
-    async fn do_notify_local(
-        &self,
-        reconfigure_request: ComputeHookNotifyRequest,
-    ) -> anyhow::Result<()> {
-        let env = match LocalEnv::load_config() {
-            Ok(e) => e,
-            Err(e) => {
-                tracing::warn!("Couldn't load neon_local config, skipping compute update ({e})");
-                return Ok(());
-            }
-        };
-        let cplane =
-            ComputeControlPlane::load(env.clone()).expect("Error loading compute control plane");
-        let ComputeHookNotifyRequest { tenant_id, shards } = reconfigure_request;
-
-        let compute_pageservers = shards
-            .into_iter()
-            .map(|shard| {
-                let ps_conf = env
-                    .get_pageserver_conf(shard.node_id)
-                    .expect("Unknown pageserver");
-                let (pg_host, pg_port) = parse_host_port(&ps_conf.listen_pg_addr)
-                    .expect("Unable to parse listen_pg_addr");
-                (pg_host, pg_port.unwrap_or(5432))
-            })
-            .collect::<Vec<_>>();
-
-        for (endpoint_name, endpoint) in &cplane.endpoints {
-            if endpoint.tenant_id == tenant_id && endpoint.status() == EndpointStatus::Running {
-                tracing::info!("🔁 Reconfiguring endpoint {}", endpoint_name,);
-                endpoint.reconfigure(compute_pageservers.clone()).await?;
-            }
-        }
-
-        Ok(())
-    }
-
-    async fn do_notify_iteration(
-        &self,
-        client: &reqwest::Client,
-        url: &String,
-        reconfigure_request: &ComputeHookNotifyRequest,
-        cancel: &CancellationToken,
-    ) -> Result<(), NotifyError> {
-        let req = client.request(Method::POST, url);
-        let req = if let Some(value) = &self.authorization_header {
-            req.header(reqwest::header::AUTHORIZATION, value)
-        } else {
-            req
-        };
-
-        tracing::debug!(
-            "Sending notify request to {} ({:?})",
-            url,
-            reconfigure_request
-        );
-        let send_result = req.json(&reconfigure_request).send().await;
-        let response = match send_result {
-            Ok(r) => r,
-            Err(e) => return Err(e.into()),
-        };
-
-        // Treat all 2xx responses as success
-        if response.status() >= StatusCode::OK && response.status() < StatusCode::MULTIPLE_CHOICES {
-            if response.status() != StatusCode::OK {
-                // Non-200 2xx response: it doesn't make sense to retry, but this is unexpected, so
-                // log a warning.
-                tracing::warn!(
-                    "Unexpected 2xx response code {} from control plane",
-                    response.status()
-                );
-            }
-
-            return Ok(());
-        }
-
-        // Error response codes
-        match response.status() {
-            StatusCode::TOO_MANY_REQUESTS => {
-                // TODO: 429 handling should be global: set some state visible to other requests
-                // so that they will delay before starting, rather than all notifications trying
-                // once before backing off.
-                tokio::time::timeout(SLOWDOWN_DELAY, cancel.cancelled())
-                    .await
-                    .ok();
-                Err(NotifyError::SlowDown)
-            }
-            StatusCode::LOCKED => {
-                // Delay our retry if busy: the usual fast exponential backoff in backoff::retry
-                // is not appropriate
-                tokio::time::timeout(BUSY_DELAY, cancel.cancelled())
-                    .await
-                    .ok();
-                Err(NotifyError::Busy)
-            }
-            StatusCode::SERVICE_UNAVAILABLE
-            | StatusCode::GATEWAY_TIMEOUT
-            | StatusCode::BAD_GATEWAY => Err(NotifyError::Unavailable(response.status())),
-            StatusCode::BAD_REQUEST | StatusCode::UNAUTHORIZED | StatusCode::FORBIDDEN => {
-                Err(NotifyError::Fatal(response.status()))
-            }
-            _ => Err(NotifyError::Unexpected(response.status())),
-        }
-    }
-
-    async fn do_notify(
-        &self,
-        url: &String,
-        reconfigure_request: ComputeHookNotifyRequest,
-        cancel: &CancellationToken,
-    ) -> Result<(), NotifyError> {
-        let client = reqwest::Client::new();
-        backoff::retry(
-            || self.do_notify_iteration(&client, url, &reconfigure_request, cancel),
-            |e| matches!(e, NotifyError::Fatal(_)),
-            3,
-            10,
-            "Send compute notification",
-            cancel,
-        )
-        .await
-        .ok_or_else(|| NotifyError::ShuttingDown)
-        .and_then(|x| x)
-    }
-
-    /// Call this to notify the compute (postgres) tier of new pageservers to use
-    /// for a tenant.  notify() is called by each shard individually, and this function
-    /// will decide whether an update to the tenant is sent.  An update is sent on the
-    /// condition that:
-    /// - We know a pageserver for every shard.
-    /// - All the shards have the same shard_count (i.e. we are not mid-split)
-    ///
-    /// Cancellation token enables callers to drop out, e.g. if calling from a Reconciler
-    /// that is cancelled.
-    ///
-    /// This function is fallible, including in the case that the control plane is transiently
-    /// unavailable.  A limited number of retries are done internally to efficiently hide short unavailability
-    /// periods, but we don't retry forever.  The **caller** is responsible for handling failures and
-    /// ensuring that they eventually call again to ensure that the compute is eventually notified of
-    /// the proper pageserver nodes for a tenant.
-    #[tracing::instrument(skip_all, fields(tenant_shard_id, node_id))]
     pub(super) async fn notify(
         &self,
         tenant_shard_id: TenantShardId,
         node_id: NodeId,
-        cancel: &CancellationToken,
-    ) -> Result<(), NotifyError> {
+    ) -> anyhow::Result<()> {
+        tracing::info!("ComputeHook::notify: {}->{}", tenant_shard_id, node_id);
         let mut locked = self.state.lock().await;
         let entry = locked
             .entry(tenant_shard_id.tenant_id)
@@ -294,25 +111,6 @@ impl ComputeHook {
             entry.shards.push((shard_index, node_id));
         }
 
-        let reconfigure_request = entry.maybe_reconfigure(tenant_shard_id.tenant_id).await;
-        let Some(reconfigure_request) = reconfigure_request else {
-            // The tenant doesn't yet have pageservers for all its shards: we won't notify anything
-            // until it does.
-            tracing::debug!("Tenant isn't yet ready to emit a notification",);
-            return Ok(());
-        };
-
-        if let Some(notify_url) = &self.config.compute_hook_url {
-            self.do_notify(notify_url, reconfigure_request, cancel)
-                .await
-        } else {
-            self.do_notify_local(reconfigure_request)
-                .await
-                .map_err(|e| {
-                    // This path is for testing only, so munge the error into our prod-style error type.
-                    tracing::error!("Local notification hook failed: {e}");
-                    NotifyError::Fatal(StatusCode::INTERNAL_SERVER_ERROR)
-                })
-        }
+        entry.maybe_reconfigure(tenant_shard_id.tenant_id).await
     }
 }

control_plane/attachment_service/src/http.rs

@@ -1,18 +1,14 @@
 use crate::reconciler::ReconcileError;
-use crate::service::{Service, STARTUP_RECONCILE_TIMEOUT};
+use crate::service::Service;
 use hyper::{Body, Request, Response};
 use hyper::{StatusCode, Uri};
-use pageserver_api::models::{
-    TenantCreateRequest, TenantLocationConfigRequest, TimelineCreateRequest,
-};
+use pageserver_api::models::{TenantCreateRequest, TimelineCreateRequest};
 use pageserver_api::shard::TenantShardId;
-use pageserver_client::mgmt_api;
 use std::sync::Arc;
-use std::time::{Duration, Instant};
 use utils::auth::SwappableJwtAuth;
 use utils::http::endpoint::{auth_middleware, request_span};
 use utils::http::request::parse_request_param;
-use utils::id::{TenantId, TimelineId};
+use utils::id::TenantId;
 
 use utils::{
     http::{
@@ -108,163 +104,34 @@ async fn handle_inspect(mut req: Request<Body>) -> Result<Response<Body>, ApiErr
     json_response(StatusCode::OK, state.service.inspect(inspect_req))
 }
 
-async fn handle_tenant_create(
-    service: Arc<Service>,
-    mut req: Request<Body>,
-) -> Result<Response<Body>, ApiError> {
+async fn handle_tenant_create(mut req: Request<Body>) -> Result<Response<Body>, ApiError> {
     let create_req = json_request::<TenantCreateRequest>(&mut req).await?;
-    json_response(StatusCode::OK, service.tenant_create(create_req).await?)
-}
-
-// For tenant and timeline deletions, which both implement an "initially return 202, then 404 once
-// we're done" semantic, we wrap with a retry loop to expose a simpler API upstream.  This avoids
-// needing to track a "deleting" state for tenants.
-async fn deletion_wrapper<R, F>(service: Arc<Service>, f: F) -> Result<Response<Body>, ApiError>
-where
-    R: std::future::Future<Output = Result<StatusCode, ApiError>> + Send + 'static,
-    F: Fn(Arc<Service>) -> R + Send + Sync + 'static,
-{
-    let started_at = Instant::now();
-    // To keep deletion reasonably snappy for small tenants, initially check after 1 second if deletion
-    // completed.
-    let mut retry_period = Duration::from_secs(1);
-    // On subsequent retries, wait longer.
-    let max_retry_period = Duration::from_secs(5);
-    // Enable callers with a 30 second request timeout to reliably get a response
-    let max_wait = Duration::from_secs(25);
-
-    loop {
-        let status = f(service.clone()).await?;
-        match status {
-            StatusCode::ACCEPTED => {
-                tracing::info!("Deletion accepted, waiting to try again...");
-                tokio::time::sleep(retry_period).await;
-                retry_period = max_retry_period;
-            }
-            StatusCode::NOT_FOUND => {
-                tracing::info!("Deletion complete");
-                return json_response(StatusCode::OK, ());
-            }
-            _ => {
-                tracing::warn!("Unexpected status {status}");
-                return json_response(status, ());
-            }
-        }
-
-        let now = Instant::now();
-        if now + retry_period > started_at + max_wait {
-            tracing::info!("Deletion timed out waiting for 404");
-            // REQUEST_TIMEOUT would be more appropriate, but CONFLICT is already part of
-            // the pageserver's swagger definition for this endpoint, and has the same desired
-            // effect of causing the control plane to retry later.
-            return json_response(StatusCode::CONFLICT, ());
-        }
-    }
-}
-
-async fn handle_tenant_location_config(
-    service: Arc<Service>,
-    mut req: Request<Body>,
-) -> Result<Response<Body>, ApiError> {
-    let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
-    let config_req = json_request::<TenantLocationConfigRequest>(&mut req).await?;
+    let state = get_state(&req);
     json_response(
         StatusCode::OK,
-        service
-            .tenant_location_config(tenant_id, config_req)
-            .await?,
+        state.service.tenant_create(create_req).await?,
     )
 }
 
-async fn handle_tenant_delete(
-    service: Arc<Service>,
-    req: Request<Body>,
-) -> Result<Response<Body>, ApiError> {
-    let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
-
-    deletion_wrapper(service, move |service| async move {
-        service.tenant_delete(tenant_id).await
-    })
-    .await
-}
-
-async fn handle_tenant_timeline_create(
-    service: Arc<Service>,
-    mut req: Request<Body>,
-) -> Result<Response<Body>, ApiError> {
+async fn handle_tenant_timeline_create(mut req: Request<Body>) -> Result<Response<Body>, ApiError> {
     let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
     let create_req = json_request::<TimelineCreateRequest>(&mut req).await?;
+
+    let state = get_state(&req);
     json_response(
         StatusCode::OK,
-        service
+        state
+            .service
             .tenant_timeline_create(tenant_id, create_req)
             .await?,
     )
 }
 
-async fn handle_tenant_timeline_delete(
-    service: Arc<Service>,
-    req: Request<Body>,
-) -> Result<Response<Body>, ApiError> {
+async fn handle_tenant_locate(req: Request<Body>) -> Result<Response<Body>, ApiError> {
     let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
-    let timeline_id: TimelineId = parse_request_param(&req, "timeline_id")?;
+    let state = get_state(&req);
 
-    deletion_wrapper(service, move |service| async move {
-        service.tenant_timeline_delete(tenant_id, timeline_id).await
-    })
-    .await
-}
-
-async fn handle_tenant_timeline_passthrough(
-    service: Arc<Service>,
-    req: Request<Body>,
-) -> Result<Response<Body>, ApiError> {
-    let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
-
-    let Some(path) = req.uri().path_and_query() else {
-        // This should never happen, our request router only calls us if there is a path
-        return Err(ApiError::BadRequest(anyhow::anyhow!("Missing path")));
-    };
-
-    tracing::info!("Proxying request for tenant {} ({})", tenant_id, path);
-
-    // Find the node that holds shard zero
-    let (base_url, tenant_shard_id) = service.tenant_shard0_baseurl(tenant_id)?;
-
-    // Callers will always pass an unsharded tenant ID.  Before proxying, we must
-    // rewrite this to a shard-aware shard zero ID.
-    let path = format!("{}", path);
-    let tenant_str = tenant_id.to_string();
-    let tenant_shard_str = format!("{}", tenant_shard_id);
-    let path = path.replace(&tenant_str, &tenant_shard_str);
-
-    let client = mgmt_api::Client::new(base_url, service.get_config().jwt_token.as_deref());
-    let resp = client.get_raw(path).await.map_err(|_e|
-        // FIXME: give APiError a proper Unavailable variant.  We return 503 here because
-        // if we can't successfully send a request to the pageserver, we aren't available.
-        ApiError::ShuttingDown)?;
-
-    // We have a reqest::Response, would like a http::Response
-    let mut builder = hyper::Response::builder()
-        .status(resp.status())
-        .version(resp.version());
-    for (k, v) in resp.headers() {
-        builder = builder.header(k, v);
-    }
-
-    let response = builder
-        .body(Body::wrap_stream(resp.bytes_stream()))
-        .map_err(|e| ApiError::InternalServerError(e.into()))?;
-
-    Ok(response)
-}
-
-async fn handle_tenant_locate(
-    service: Arc<Service>,
-    req: Request<Body>,
-) -> Result<Response<Body>, ApiError> {
-    let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
-    json_response(StatusCode::OK, service.tenant_locate(tenant_id)?)
+    json_response(StatusCode::OK, state.service.tenant_locate(tenant_id)?)
 }
 
 async fn handle_node_register(mut req: Request<Body>) -> Result<Response<Body>, ApiError> {
@@ -274,11 +141,6 @@ async fn handle_node_register(mut req: Request<Body>) -> Result<Response<Body>,
     json_response(StatusCode::OK, ())
 }
 
-async fn handle_node_list(req: Request<Body>) -> Result<Response<Body>, ApiError> {
-    let state = get_state(&req);
-    json_response(StatusCode::OK, state.service.node_list().await?)
-}
-
 async fn handle_node_configure(mut req: Request<Body>) -> Result<Response<Body>, ApiError> {
     let node_id: NodeId = parse_request_param(&req, "node_id")?;
     let config_req = json_request::<NodeConfigureRequest>(&mut req).await?;
@@ -292,15 +154,14 @@ async fn handle_node_configure(mut req: Request<Body>) -> Result<Response<Body>,
     json_response(StatusCode::OK, state.service.node_configure(config_req)?)
 }
 
-async fn handle_tenant_shard_migrate(
-    service: Arc<Service>,
-    mut req: Request<Body>,
-) -> Result<Response<Body>, ApiError> {
+async fn handle_tenant_shard_migrate(mut req: Request<Body>) -> Result<Response<Body>, ApiError> {
     let tenant_shard_id: TenantShardId = parse_request_param(&req, "tenant_shard_id")?;
     let migrate_req = json_request::<TenantShardMigrateRequest>(&mut req).await?;
+    let state = get_state(&req);
     json_response(
         StatusCode::OK,
-        service
+        state
+            .service
             .tenant_shard_migrate(tenant_shard_id, migrate_req)
             .await?,
     )
@@ -317,35 +178,6 @@ impl From<ReconcileError> for ApiError {
     }
 }
 
-/// Common wrapper for request handlers that call into Service and will operate on tenants: they must only
-/// be allowed to run if Service has finished its initial reconciliation.
-async fn tenant_service_handler<R, H>(request: Request<Body>, handler: H) -> R::Output
-where
-    R: std::future::Future<Output = Result<Response<Body>, ApiError>> + Send + 'static,
-    H: FnOnce(Arc<Service>, Request<Body>) -> R + Send + Sync + 'static,
-{
-    let state = get_state(&request);
-    let service = state.service.clone();
-
-    let startup_complete = service.startup_complete.clone();
-    if tokio::time::timeout(STARTUP_RECONCILE_TIMEOUT, startup_complete.wait())
-        .await
-        .is_err()
-    {
-        // This shouldn't happen: it is the responsibilty of [`Service::startup_reconcile`] to use appropriate
-        // timeouts around its remote calls, to bound its runtime.
-        return Err(ApiError::Timeout(
-            "Timed out waiting for service readiness".into(),
-        ));
-    }
-
-    request_span(
-        request,
-        |request| async move { handler(service, request).await },
-    )
-    .await
-}
-
 pub fn make_router(
     service: Arc<Service>,
     auth: Option<Arc<SwappableJwtAuth>>,
@@ -364,59 +196,23 @@ pub fn make_router(
 
     router
         .data(Arc::new(HttpState::new(service, auth)))
-        // Non-prefixed generic endpoints (status, metrics)
         .get("/status", |r| request_span(r, handle_status))
-        // Upcalls for the pageserver: point the pageserver's `control_plane_api` config to this prefix
-        .post("/upcall/v1/re-attach", |r| {
-            request_span(r, handle_re_attach)
-        })
-        .post("/upcall/v1/validate", |r| request_span(r, handle_validate))
-        // Test/dev/debug endpoints
-        .post("/debug/v1/attach-hook", |r| {
-            request_span(r, handle_attach_hook)
-        })
-        .post("/debug/v1/inspect", |r| request_span(r, handle_inspect))
-        .get("/control/v1/tenant/:tenant_id/locate", |r| {
-            tenant_service_handler(r, handle_tenant_locate)
-        })
-        // Node operations
-        .post("/control/v1/node", |r| {
-            request_span(r, handle_node_register)
-        })
-        .get("/control/v1/node", |r| request_span(r, handle_node_list))
-        .put("/control/v1/node/:node_id/config", |r| {
+        .post("/re-attach", |r| request_span(r, handle_re_attach))
+        .post("/validate", |r| request_span(r, handle_validate))
+        .post("/attach-hook", |r| request_span(r, handle_attach_hook))
+        .post("/inspect", |r| request_span(r, handle_inspect))
+        .post("/node", |r| request_span(r, handle_node_register))
+        .put("/node/:node_id/config", |r| {
             request_span(r, handle_node_configure)
         })
-        // Tenant Shard operations
-        .put("/control/v1/tenant/:tenant_shard_id/migrate", |r| {
-            tenant_service_handler(r, handle_tenant_shard_migrate)
+        .post("/tenant", |r| request_span(r, handle_tenant_create))
+        .post("/tenant/:tenant_id/timeline", |r| {
+            request_span(r, handle_tenant_timeline_create)
         })
-        // Tenant operations
-        // The ^/v1/ endpoints act as a "Virtual Pageserver", enabling shard-naive clients to call into
-        // this service to manage tenants that actually consist of many tenant shards, as if they are a single entity.
-        .post("/v1/tenant", |r| {
-            tenant_service_handler(r, handle_tenant_create)
+        .get("/tenant/:tenant_id/locate", |r| {
+            request_span(r, handle_tenant_locate)
         })
-        .delete("/v1/tenant/:tenant_id", |r| {
-            tenant_service_handler(r, handle_tenant_delete)
-        })
-        .put("/v1/tenant/:tenant_id/location_config", |r| {
-            tenant_service_handler(r, handle_tenant_location_config)
-        })
-        // Timeline operations
-        .delete("/v1/tenant/:tenant_id/timeline/:timeline_id", |r| {
-            tenant_service_handler(r, handle_tenant_timeline_delete)
-        })
-        .post("/v1/tenant/:tenant_id/timeline", |r| {
-            tenant_service_handler(r, handle_tenant_timeline_create)
-        })
-        // Tenant detail GET passthrough to shard zero
-        .get("/v1/tenant/:tenant_id", |r| {
-            tenant_service_handler(r, handle_tenant_timeline_passthrough)
-        })
-        // Timeline GET passthrough to shard zero.  Note that the `*` in the URL is a wildcard: any future
-        // timeline GET APIs will be implicitly included.
-        .get("/v1/tenant/:tenant_id/timeline*", |r| {
-            tenant_service_handler(r, handle_tenant_timeline_passthrough)
+        .put("/tenant/:tenant_shard_id/migrate", |r| {
+            request_span(r, handle_tenant_shard_migrate)
         })
 }

control_plane/attachment_service/src/lib.rs

@@ -7,7 +7,6 @@ mod node;
 pub mod persistence;
 mod reconciler;
 mod scheduler;
-mod schema;
 pub mod service;
 mod tenant_state;
 
@@ -18,8 +17,6 @@ enum PlacementPolicy {
     /// Production-ready way to attach a tenant: one attached pageserver and
     /// some number of secondaries.
     Double(usize),
-    /// Do not attach to any pageservers
-    Detached,
 }
 
 #[derive(Ord, PartialOrd, Eq, PartialEq, Copy, Clone)]

control_plane/attachment_service/src/main.rs

@@ -4,28 +4,23 @@
 /// This enables running & testing pageservers without a full-blown
 /// deployment of the Neon cloud platform.
 ///
-use anyhow::{anyhow, Context};
+use anyhow::anyhow;
 use attachment_service::http::make_router;
 use attachment_service::persistence::Persistence;
 use attachment_service::service::{Config, Service};
-use aws_config::{self, BehaviorVersion, Region};
 use camino::Utf8PathBuf;
 use clap::Parser;
-use diesel::Connection;
 use metrics::launch_timestamp::LaunchTimestamp;
 use std::sync::Arc;
-use tokio::signal::unix::SignalKind;
 use utils::auth::{JwtAuth, SwappableJwtAuth};
 use utils::logging::{self, LogFormat};
+use utils::signals::{ShutdownSignals, Signal};
 
 use utils::{project_build_tag, project_git_version, tcp_listener};
 
 project_git_version!(GIT_VERSION);
 project_build_tag!(BUILD_TAG);
 
-use diesel_migrations::{embed_migrations, EmbeddedMigrations};
-pub const MIGRATIONS: EmbeddedMigrations = embed_migrations!("./migrations");
-
 #[derive(Parser)]
 #[command(author, version, about, long_about = None)]
 #[command(arg_required_else_help(true))]
@@ -34,153 +29,17 @@ struct Cli {
     #[arg(short, long)]
     listen: std::net::SocketAddr,
 
-    /// Public key for JWT authentication of clients
+    /// Path to public key for JWT authentication of clients
     #[arg(long)]
-    public_key: Option<String>,
+    public_key: Option<camino::Utf8PathBuf>,
 
     /// Token for authenticating this service with the pageservers it controls
-    #[arg(long)]
+    #[arg(short, long)]
     jwt_token: Option<String>,
 
-    /// Token for authenticating this service with the control plane, when calling
-    /// the compute notification endpoint
-    #[arg(long)]
-    control_plane_jwt_token: Option<String>,
-
-    /// URL to control plane compute notification endpoint
-    #[arg(long)]
-    compute_hook_url: Option<String>,
-
     /// Path to the .json file to store state (will be created if it doesn't exist)
     #[arg(short, long)]
-    path: Option<Utf8PathBuf>,
-
-    /// URL to connect to postgres, like postgresql://localhost:1234/attachment_service
-    #[arg(long)]
-    database_url: Option<String>,
-}
-
-/// Secrets may either be provided on the command line (for testing), or loaded from AWS SecretManager: this
-/// type encapsulates the logic to decide which and do the loading.
-struct Secrets {
-    database_url: String,
-    public_key: Option<JwtAuth>,
-    jwt_token: Option<String>,
-    control_plane_jwt_token: Option<String>,
-}
-
-impl Secrets {
-    const DATABASE_URL_SECRET: &'static str = "rds-neon-storage-controller-url";
-    const PAGESERVER_JWT_TOKEN_SECRET: &'static str =
-        "neon-storage-controller-pageserver-jwt-token";
-    const CONTROL_PLANE_JWT_TOKEN_SECRET: &'static str =
-        "neon-storage-controller-control-plane-jwt-token";
-    const PUBLIC_KEY_SECRET: &'static str = "neon-storage-controller-public-key";
-
-    async fn load(args: &Cli) -> anyhow::Result<Self> {
-        match &args.database_url {
-            Some(url) => Self::load_cli(url, args),
-            None => Self::load_aws_sm().await,
-        }
-    }
-
-    async fn load_aws_sm() -> anyhow::Result<Self> {
-        let Ok(region) = std::env::var("AWS_REGION") else {
-            anyhow::bail!("AWS_REGION is not set, cannot load secrets automatically: either set this, or use CLI args to supply secrets");
-        };
-        let config = aws_config::defaults(BehaviorVersion::v2023_11_09())
-            .region(Region::new(region.clone()))
-            .load()
-            .await;
-
-        let asm = aws_sdk_secretsmanager::Client::new(&config);
-
-        let Some(database_url) = asm
-            .get_secret_value()
-            .secret_id(Self::DATABASE_URL_SECRET)
-            .send()
-            .await?
-            .secret_string()
-            .map(str::to_string)
-        else {
-            anyhow::bail!(
-                "Database URL secret not found at {region}/{}",
-                Self::DATABASE_URL_SECRET
-            )
-        };
-
-        let jwt_token = asm
-            .get_secret_value()
-            .secret_id(Self::PAGESERVER_JWT_TOKEN_SECRET)
-            .send()
-            .await?
-            .secret_string()
-            .map(str::to_string);
-        if jwt_token.is_none() {
-            tracing::warn!("No pageserver JWT token set: this will only work if authentication is disabled on the pageserver");
-        }
-
-        let control_plane_jwt_token = asm
-            .get_secret_value()
-            .secret_id(Self::CONTROL_PLANE_JWT_TOKEN_SECRET)
-            .send()
-            .await?
-            .secret_string()
-            .map(str::to_string);
-        if jwt_token.is_none() {
-            tracing::warn!("No control plane JWT token set: this will only work if authentication is disabled on the pageserver");
-        }
-
-        let public_key = asm
-            .get_secret_value()
-            .secret_id(Self::PUBLIC_KEY_SECRET)
-            .send()
-            .await?
-            .secret_string()
-            .map(str::to_string);
-        let public_key = match public_key {
-            Some(key) => Some(JwtAuth::from_key(key)?),
-            None => {
-                tracing::warn!(
-                    "No public key set: inccoming HTTP requests will not be authenticated"
-                );
-                None
-            }
-        };
-
-        Ok(Self {
-            database_url,
-            public_key,
-            jwt_token,
-            control_plane_jwt_token,
-        })
-    }
-
-    fn load_cli(database_url: &str, args: &Cli) -> anyhow::Result<Self> {
-        let public_key = match &args.public_key {
-            None => None,
-            Some(key) => Some(JwtAuth::from_key(key.clone()).context("Loading public key")?),
-        };
-        Ok(Self {
-            database_url: database_url.to_owned(),
-            public_key,
-            jwt_token: args.jwt_token.clone(),
-            control_plane_jwt_token: args.control_plane_jwt_token.clone(),
-        })
-    }
-}
-
-async fn migration_run(database_url: &str) -> anyhow::Result<()> {
-    use diesel::PgConnection;
-    use diesel_migrations::{HarnessWithOutput, MigrationHarness};
-    let mut conn = PgConnection::establish(database_url)?;
-
-    HarnessWithOutput::write_to_stdout(&mut conn)
-        .run_pending_migrations(MIGRATIONS)
-        .map(|_| ())
-        .map_err(|e| anyhow::anyhow!(e))?;
-
-    Ok(())
+    path: Utf8PathBuf,
 }
 
 #[tokio::main]
@@ -199,61 +58,43 @@ async fn main() -> anyhow::Result<()> {
         GIT_VERSION,
         launch_ts.to_string(),
         BUILD_TAG,
-        args.path.as_ref().unwrap_or(&Utf8PathBuf::from("<none>")),
+        args.path,
         args.listen
     );
 
-    let secrets = Secrets::load(&args).await?;
-
     let config = Config {
-        jwt_token: secrets.jwt_token,
-        control_plane_jwt_token: secrets.control_plane_jwt_token,
-        compute_hook_url: args.compute_hook_url,
+        jwt_token: args.jwt_token,
     };
 
-    // After loading secrets & config, but before starting anything else, apply database migrations
-    migration_run(&secrets.database_url)
-        .await
-        .context("Running database migrations")?;
+    let persistence = Arc::new(Persistence::new(&args.path).await);
 
-    let json_path = args.path;
-    let persistence = Arc::new(Persistence::new(secrets.database_url, json_path.clone()));
-
-    let service = Service::spawn(config, persistence.clone()).await?;
+    let service = Service::spawn(config, persistence).await?;
 
     let http_listener = tcp_listener::bind(args.listen)?;
 
-    let auth = secrets
-        .public_key
-        .map(|jwt_auth| Arc::new(SwappableJwtAuth::new(jwt_auth)));
+    let auth = if let Some(public_key_path) = &args.public_key {
+        let jwt_auth = JwtAuth::from_key_path(public_key_path)?;
+        Some(Arc::new(SwappableJwtAuth::new(jwt_auth)))
+    } else {
+        None
+    };
     let router = make_router(service, auth)
         .build()
         .map_err(|err| anyhow!(err))?;
-    let router_service = utils::http::RouterService::new(router).unwrap();
-    let server = hyper::Server::from_tcp(http_listener)?.serve(router_service);
+    let service = utils::http::RouterService::new(router).unwrap();
+    let server = hyper::Server::from_tcp(http_listener)?.serve(service);
 
     tracing::info!("Serving on {0}", args.listen);
 
     tokio::task::spawn(server);
 
-    // Wait until we receive a signal
-    let mut sigint = tokio::signal::unix::signal(SignalKind::interrupt())?;
-    let mut sigquit = tokio::signal::unix::signal(SignalKind::quit())?;
-    let mut sigterm = tokio::signal::unix::signal(SignalKind::terminate())?;
-    tokio::select! {
-        _ = sigint.recv() => {},
-        _ = sigterm.recv() => {},
-        _ = sigquit.recv() => {},
-    }
-    tracing::info!("Terminating on signal");
-
-    if json_path.is_some() {
-        // Write out a JSON dump on shutdown: this is used in compat tests to avoid passing
-        // full postgres dumps around.
-        if let Err(e) = persistence.write_tenants_json().await {
-            tracing::error!("Failed to write JSON on shutdown: {e}")
+    ShutdownSignals::handle(|signal| match signal {
+        Signal::Interrupt | Signal::Terminate | Signal::Quit => {
+            tracing::info!("Got {}. Terminating", signal.name());
+            // We're just a test helper: no graceful shutdown.
+            std::process::exit(0);
         }
-    }
+    })?;
 
-    std::process::exit(0);
+    Ok(())
 }

control_plane/attachment_service/src/node.rs

@@ -1,8 +1,6 @@
 use control_plane::attachment_service::{NodeAvailability, NodeSchedulingPolicy};
 use utils::id::NodeId;
 
-use crate::persistence::NodePersistence;
-
 #[derive(Clone)]
 pub(crate) struct Node {
     pub(crate) id: NodeId,
@@ -36,15 +34,4 @@ impl Node {
             NodeSchedulingPolicy::Pause => false,
         }
     }
-
-    pub(crate) fn to_persistent(&self) -> NodePersistence {
-        NodePersistence {
-            node_id: self.id.0 as i64,
-            scheduling_policy: self.scheduling.into(),
-            listen_http_addr: self.listen_http_addr.clone(),
-            listen_http_port: self.listen_http_port as i32,
-            listen_pg_addr: self.listen_pg_addr.clone(),
-            listen_pg_port: self.listen_pg_port as i32,
-        }
-    }
 }

control_plane/attachment_service/src/persistence.rs

@@ -1,295 +1,181 @@
-use std::collections::HashMap;
-use std::str::FromStr;
+use std::{collections::HashMap, str::FromStr};
 
-use camino::Utf8Path;
-use camino::Utf8PathBuf;
-use control_plane::attachment_service::{NodeAvailability, NodeSchedulingPolicy};
-use diesel::pg::PgConnection;
-use diesel::prelude::*;
-use diesel::Connection;
-use pageserver_api::models::TenantConfig;
-use pageserver_api::shard::{ShardCount, ShardNumber, TenantShardId};
+use camino::{Utf8Path, Utf8PathBuf};
+use control_plane::{
+    attachment_service::{NodeAvailability, NodeSchedulingPolicy},
+    local_env::LocalEnv,
+};
+use pageserver_api::{
+    models::TenantConfig,
+    shard::{ShardCount, ShardNumber, TenantShardId},
+};
+use postgres_connection::parse_host_port;
 use serde::{Deserialize, Serialize};
-use utils::generation::Generation;
-use utils::id::{NodeId, TenantId};
+use utils::{
+    generation::Generation,
+    id::{NodeId, TenantId},
+};
 
-use crate::node::Node;
-use crate::PlacementPolicy;
+use crate::{node::Node, PlacementPolicy};
 
-/// ## What do we store?
-///
-/// The attachment service does not store most of its state durably.
-///
-/// The essential things to store durably are:
-/// - generation numbers, as these must always advance monotonically to ensure data safety.
-/// - Tenant's PlacementPolicy and TenantConfig, as the source of truth for these is something external.
-/// - Node's scheduling policies, as the source of truth for these is something external.
-///
-/// Other things we store durably as an implementation detail:
-/// - Node's host/port: this could be avoided it we made nodes emit a self-registering heartbeat,
-///   but it is operationally simpler to make this service the authority for which nodes
-///   it talks to.
-///
-/// ## Performance/efficiency
-///
-/// The attachment service does not go via the database for most things: there are
-/// a couple of places where we must, and where efficiency matters:
-/// - Incrementing generation numbers: the Reconciler has to wait for this to complete
-///   before it can attach a tenant, so this acts as a bound on how fast things like
-///   failover can happen.
-/// - Pageserver re-attach: we will increment many shards' generations when this happens,
-///   so it is important to avoid e.g. issuing O(N) queries.
-///
-/// Database calls relating to nodes have low performance requirements, as they are very rarely
-/// updated, and reads of nodes are always from memory, not the database.  We only require that
-/// we can UPDATE a node's scheduling mode reasonably quickly to mark a bad node offline.
+/// Placeholder for storage.  This will be replaced with a database client.
 pub struct Persistence {
-    database_url: String,
-
-    // In test environments, we support loading+saving a JSON file.  This is temporary, for the benefit of
-    // test_compatibility.py, so that we don't have to commit to making the database contents fully backward/forward
-    // compatible just yet.
-    json_path: Option<Utf8PathBuf>,
+    state: std::sync::Mutex<PersistentState>,
 }
 
-/// Legacy format, for use in JSON compat objects in test environment
+// Top level state available to all HTTP handlers
 #[derive(Serialize, Deserialize)]
-struct JsonPersistence {
+struct PersistentState {
     tenants: HashMap<TenantShardId, TenantShardPersistence>,
+
+    #[serde(skip)]
+    path: Utf8PathBuf,
 }
 
-#[derive(thiserror::Error, Debug)]
-pub(crate) enum DatabaseError {
-    #[error(transparent)]
-    Query(#[from] diesel::result::Error),
-    #[error(transparent)]
-    Connection(#[from] diesel::result::ConnectionError),
-    #[error("Logical error: {0}")]
-    Logical(String),
+/// A convenience for serializing the state inside a sync lock, and then
+/// writing it to disk outside of the lock.  This will go away when switching
+/// to a database backend.
+struct PendingWrite {
+    bytes: Vec<u8>,
+    path: Utf8PathBuf,
 }
 
-pub(crate) type DatabaseResult<T> = Result<T, DatabaseError>;
+impl PendingWrite {
+    async fn commit(&self) -> anyhow::Result<()> {
+        tokio::fs::write(&self.path, &self.bytes).await?;
 
-impl Persistence {
-    pub fn new(database_url: String, json_path: Option<Utf8PathBuf>) -> Self {
-        Self {
-            database_url,
-            json_path,
+        Ok(())
+    }
+}
+
+impl PersistentState {
+    fn save(&self) -> PendingWrite {
+        PendingWrite {
+            bytes: serde_json::to_vec(self).expect("Serialization error"),
+            path: self.path.clone(),
         }
     }
 
-    /// Call the provided function in a tokio blocking thread, with a Diesel database connection.
-    async fn with_conn<F, R>(&self, func: F) -> DatabaseResult<R>
-    where
-        F: Fn(&mut PgConnection) -> DatabaseResult<R> + Send + 'static,
-        R: Send + 'static,
-    {
-        let database_url = self.database_url.clone();
-        tokio::task::spawn_blocking(move || -> DatabaseResult<R> {
-            // TODO: connection pooling, such as via diesel::r2d2
-            let mut conn = PgConnection::establish(&database_url)?;
-            func(&mut conn)
-        })
-        .await
-        .expect("Task panic")
-    }
+    async fn load(path: &Utf8Path) -> anyhow::Result<Self> {
+        let bytes = tokio::fs::read(path).await?;
+        let mut decoded = serde_json::from_slice::<Self>(&bytes)?;
+        decoded.path = path.to_owned();
 
-    /// When a node is first registered, persist it before using it for anything
-    pub(crate) async fn insert_node(&self, node: &Node) -> DatabaseResult<()> {
-        let np = node.to_persistent();
-        self.with_conn(move |conn| -> DatabaseResult<()> {
-            diesel::insert_into(crate::schema::nodes::table)
-                .values(&np)
-                .execute(conn)?;
-            Ok(())
-        })
-        .await
-    }
-
-    /// At startup, populate the list of nodes which our shards may be placed on
-    pub(crate) async fn list_nodes(&self) -> DatabaseResult<Vec<Node>> {
-        let nodes: Vec<Node> = self
-            .with_conn(move |conn| -> DatabaseResult<_> {
-                Ok(crate::schema::nodes::table
-                    .load::<NodePersistence>(conn)?
-                    .into_iter()
-                    .map(|n| Node {
-                        id: NodeId(n.node_id as u64),
-                        // At startup we consider a node offline until proven otherwise.
-                        availability: NodeAvailability::Offline,
-                        scheduling: NodeSchedulingPolicy::from_str(&n.scheduling_policy)
-                            .expect("Bad scheduling policy in DB"),
-                        listen_http_addr: n.listen_http_addr,
-                        listen_http_port: n.listen_http_port as u16,
-                        listen_pg_addr: n.listen_pg_addr,
-                        listen_pg_port: n.listen_pg_port as u16,
-                    })
-                    .collect::<Vec<Node>>())
-            })
-            .await?;
-
-        tracing::info!("list_nodes: loaded {} nodes", nodes.len());
-
-        Ok(nodes)
-    }
-
-    /// At startup, load the high level state for shards, such as their config + policy.  This will
-    /// be enriched at runtime with state discovered on pageservers.
-    pub(crate) async fn list_tenant_shards(&self) -> DatabaseResult<Vec<TenantShardPersistence>> {
-        let loaded = self
-            .with_conn(move |conn| -> DatabaseResult<_> {
-                Ok(crate::schema::tenant_shards::table.load::<TenantShardPersistence>(conn)?)
-            })
-            .await?;
-
-        if loaded.is_empty() {
-            if let Some(path) = &self.json_path {
-                if tokio::fs::try_exists(path)
-                    .await
-                    .map_err(|e| DatabaseError::Logical(format!("Error stat'ing JSON file: {e}")))?
-                {
-                    tracing::info!("Importing from legacy JSON format at {path}");
-                    return self.list_tenant_shards_json(path).await;
-                }
-            }
-        }
-        Ok(loaded)
-    }
-
-    /// Shim for automated compatibility tests: load tenants from a JSON file instead of database
-    pub(crate) async fn list_tenant_shards_json(
-        &self,
-        path: &Utf8Path,
-    ) -> DatabaseResult<Vec<TenantShardPersistence>> {
-        let bytes = tokio::fs::read(path)
-            .await
-            .map_err(|e| DatabaseError::Logical(format!("Failed to load JSON: {e}")))?;
-
-        let mut decoded = serde_json::from_slice::<JsonPersistence>(&bytes)
-            .map_err(|e| DatabaseError::Logical(format!("Deserialization error: {e}")))?;
         for (tenant_id, tenant) in &mut decoded.tenants {
             // Backward compat: an old attachments.json from before PR #6251, replace
             // empty strings with proper defaults.
             if tenant.tenant_id.is_empty() {
-                tenant.tenant_id = tenant_id.to_string();
-                tenant.config = serde_json::to_string(&TenantConfig::default())
-                    .map_err(|e| DatabaseError::Logical(format!("Serialization error: {e}")))?;
-                tenant.placement_policy = serde_json::to_string(&PlacementPolicy::default())
-                    .map_err(|e| DatabaseError::Logical(format!("Serialization error: {e}")))?;
+                tenant.tenant_id = format!("{}", tenant_id);
+                tenant.config = serde_json::to_string(&TenantConfig::default())?;
+                tenant.placement_policy = serde_json::to_string(&PlacementPolicy::default())?;
             }
         }
 
-        let tenants: Vec<TenantShardPersistence> = decoded.tenants.into_values().collect();
-
-        // Synchronize database with what is in the JSON file
-        self.insert_tenant_shards(tenants.clone()).await?;
-
-        Ok(tenants)
+        Ok(decoded)
     }
 
-    /// For use in testing environments, where we dump out JSON on shutdown.
-    pub async fn write_tenants_json(&self) -> anyhow::Result<()> {
-        let Some(path) = &self.json_path else {
-            anyhow::bail!("Cannot write JSON if path isn't set (test environment bug)");
-        };
-        tracing::info!("Writing state to {path}...");
-        let tenants = self.list_tenant_shards().await?;
-        let mut tenants_map = HashMap::new();
-        for tsp in tenants {
-            let tenant_shard_id = TenantShardId {
-                tenant_id: TenantId::from_str(tsp.tenant_id.as_str())?,
-                shard_number: ShardNumber(tsp.shard_number as u8),
-                shard_count: ShardCount(tsp.shard_count as u8),
-            };
-
-            tenants_map.insert(tenant_shard_id, tsp);
+    async fn load_or_new(path: &Utf8Path) -> Self {
+        match Self::load(path).await {
+            Ok(s) => {
+                tracing::info!("Loaded state file at {}", path);
+                s
+            }
+            Err(e)
+                if e.downcast_ref::<std::io::Error>()
+                    .map(|e| e.kind() == std::io::ErrorKind::NotFound)
+                    .unwrap_or(false) =>
+            {
+                tracing::info!("Will create state file at {}", path);
+                Self {
+                    tenants: HashMap::new(),
+                    path: path.to_owned(),
+                }
+            }
+            Err(e) => {
+                panic!("Failed to load state from '{}': {e:#} (maybe your .neon/ dir was written by an older version?)", path)
+            }
         }
-        let json = serde_json::to_string(&JsonPersistence {
-            tenants: tenants_map,
-        })?;
+    }
+}
 
-        tokio::fs::write(path, &json).await?;
-        tracing::info!("Wrote {} bytes to {path}...", json.len());
+impl Persistence {
+    pub async fn new(path: &Utf8Path) -> Self {
+        let state = PersistentState::load_or_new(path).await;
+        Self {
+            state: std::sync::Mutex::new(state),
+        }
+    }
 
+    /// When registering a node, persist it so that on next start we will be able to
+    /// iterate over known nodes to synchronize their tenant shard states with our observed state.
+    pub(crate) async fn insert_node(&self, _node: &Node) -> anyhow::Result<()> {
+        // TODO: node persitence will come with database backend
         Ok(())
     }
 
+    /// At startup, we populate the service's list of nodes, and use this list to call into
+    /// each node to do an initial reconciliation of the state of the world with our in-memory
+    /// observed state.
+    pub(crate) async fn list_nodes(&self) -> anyhow::Result<Vec<Node>> {
+        let env = LocalEnv::load_config()?;
+        // TODO: node persitence will come with database backend
+
+        // XXX hack: enable test_backward_compatibility to work by populating our list of
+        // nodes from LocalEnv when it is not present in persistent storage.  Otherwise at
+        // first startup in the compat test, we may have shards but no nodes.
+        let mut result = Vec::new();
+        tracing::info!(
+            "Loaded {} pageserver nodes from LocalEnv",
+            env.pageservers.len()
+        );
+        for ps_conf in env.pageservers {
+            let (pg_host, pg_port) =
+                parse_host_port(&ps_conf.listen_pg_addr).expect("Unable to parse listen_pg_addr");
+            let (http_host, http_port) = parse_host_port(&ps_conf.listen_http_addr)
+                .expect("Unable to parse listen_http_addr");
+            result.push(Node {
+                id: ps_conf.id,
+                listen_pg_addr: pg_host.to_string(),
+                listen_pg_port: pg_port.unwrap_or(5432),
+                listen_http_addr: http_host.to_string(),
+                listen_http_port: http_port.unwrap_or(80),
+                availability: NodeAvailability::Active,
+                scheduling: NodeSchedulingPolicy::Active,
+            });
+        }
+
+        Ok(result)
+    }
+
+    /// At startup, we populate our map of tenant shards from persistent storage.
+    pub(crate) async fn list_tenant_shards(&self) -> anyhow::Result<Vec<TenantShardPersistence>> {
+        let locked = self.state.lock().unwrap();
+        Ok(locked.tenants.values().cloned().collect())
+    }
+
     /// Tenants must be persisted before we schedule them for the first time.  This enables us
     /// to correctly retain generation monotonicity, and the externally provided placement policy & config.
     pub(crate) async fn insert_tenant_shards(
         &self,
         shards: Vec<TenantShardPersistence>,
-    ) -> DatabaseResult<()> {
-        use crate::schema::tenant_shards::dsl::*;
-        self.with_conn(move |conn| -> DatabaseResult<()> {
-            conn.transaction(|conn| -> QueryResult<()> {
-                for tenant in &shards {
-                    diesel::insert_into(tenant_shards)
-                        .values(tenant)
-                        .execute(conn)?;
-                }
-                Ok(())
-            })?;
-            Ok(())
-        })
-        .await
-    }
+    ) -> anyhow::Result<()> {
+        let write = {
+            let mut locked = self.state.lock().unwrap();
+            for shard in shards {
+                let tenant_shard_id = TenantShardId {
+                    tenant_id: TenantId::from_str(shard.tenant_id.as_str())?,
+                    shard_number: ShardNumber(shard.shard_number as u8),
+                    shard_count: ShardCount(shard.shard_count as u8),
+                };
 
-    /// Ordering: call this _after_ deleting the tenant on pageservers, but _before_ dropping state for
-    /// the tenant from memory on this server.
-    #[allow(unused)]
-    pub(crate) async fn delete_tenant(&self, del_tenant_id: TenantId) -> DatabaseResult<()> {
-        use crate::schema::tenant_shards::dsl::*;
-        self.with_conn(move |conn| -> DatabaseResult<()> {
-            diesel::delete(tenant_shards)
-                .filter(tenant_id.eq(del_tenant_id.to_string()))
-                .execute(conn)?;
+                locked.tenants.insert(tenant_shard_id, shard);
+            }
+            locked.save()
+        };
 
-            Ok(())
-        })
-        .await
-    }
+        write.commit().await?;
 
-    /// When a tenant invokes the /re-attach API, this function is responsible for doing an efficient
-    /// batched increment of the generations of all tenants whose generation_pageserver is equal to
-    /// the node that called /re-attach.
-    #[tracing::instrument(skip_all, fields(node_id))]
-    pub(crate) async fn re_attach(
-        &self,
-        node_id: NodeId,
-    ) -> DatabaseResult<HashMap<TenantShardId, Generation>> {
-        use crate::schema::tenant_shards::dsl::*;
-        let updated = self
-            .with_conn(move |conn| {
-                let rows_updated = diesel::update(tenant_shards)
-                    .filter(generation_pageserver.eq(node_id.0 as i64))
-                    .set(generation.eq(generation + 1))
-                    .execute(conn)?;
-
-                tracing::info!("Incremented {} tenants' generations", rows_updated);
-
-                // TODO: UPDATE+SELECT in one query
-
-                let updated = tenant_shards
-                    .filter(generation_pageserver.eq(node_id.0 as i64))
-                    .select(TenantShardPersistence::as_select())
-                    .load(conn)?;
-                Ok(updated)
-            })
-            .await?;
-
-        let mut result = HashMap::new();
-        for tsp in updated {
-            let tenant_shard_id = TenantShardId {
-                tenant_id: TenantId::from_str(tsp.tenant_id.as_str())
-                    .map_err(|e| DatabaseError::Logical(format!("Malformed tenant id: {e}")))?,
-                shard_number: ShardNumber(tsp.shard_number as u8),
-                shard_count: ShardCount(tsp.shard_count as u8),
-            };
-            result.insert(tenant_shard_id, Generation::new(tsp.generation as u32));
-        }
-
-        Ok(result)
+        Ok(())
     }
 
     /// Reconciler calls this immediately before attaching to a new pageserver, to acquire a unique, monotonically
@@ -300,48 +186,57 @@ impl Persistence {
         tenant_shard_id: TenantShardId,
         node_id: NodeId,
     ) -> anyhow::Result<Generation> {
-        use crate::schema::tenant_shards::dsl::*;
-        let updated = self
-            .with_conn(move |conn| {
-                let updated = diesel::update(tenant_shards)
-                    .filter(tenant_id.eq(tenant_shard_id.tenant_id.to_string()))
-                    .filter(shard_number.eq(tenant_shard_id.shard_number.0 as i32))
-                    .filter(shard_count.eq(tenant_shard_id.shard_count.0 as i32))
-                    .set((
-                        generation.eq(generation + 1),
-                        generation_pageserver.eq(node_id.0 as i64),
-                    ))
-                    // TODO: only returning() the generation column
-                    .returning(TenantShardPersistence::as_returning())
-                    .get_result(conn)?;
+        let (write, gen) = {
+            let mut locked = self.state.lock().unwrap();
+            let Some(shard) = locked.tenants.get_mut(&tenant_shard_id) else {
+                anyhow::bail!("Tried to increment generation of unknown shard");
+            };
 
-                Ok(updated)
-            })
-            .await?;
+            shard.generation += 1;
+            shard.generation_pageserver = Some(node_id);
 
-        Ok(Generation::new(updated.generation as u32))
+            let gen = Generation::new(shard.generation);
+            (locked.save(), gen)
+        };
+
+        write.commit().await?;
+        Ok(gen)
     }
 
     pub(crate) async fn detach(&self, tenant_shard_id: TenantShardId) -> anyhow::Result<()> {
-        use crate::schema::tenant_shards::dsl::*;
-        self.with_conn(move |conn| {
-            let updated = diesel::update(tenant_shards)
-                .filter(tenant_id.eq(tenant_shard_id.tenant_id.to_string()))
-                .filter(shard_number.eq(tenant_shard_id.shard_number.0 as i32))
-                .filter(shard_count.eq(tenant_shard_id.shard_count.0 as i32))
-                .set((
-                    generation_pageserver.eq(i64::MAX),
-                    placement_policy.eq(serde_json::to_string(&PlacementPolicy::Detached).unwrap()),
-                ))
-                .execute(conn)?;
-
-            Ok(updated)
-        })
-        .await?;
-
+        let write = {
+            let mut locked = self.state.lock().unwrap();
+            let Some(shard) = locked.tenants.get_mut(&tenant_shard_id) else {
+                anyhow::bail!("Tried to increment generation of unknown shard");
+            };
+            shard.generation_pageserver = None;
+            locked.save()
+        };
+        write.commit().await?;
         Ok(())
     }
 
+    pub(crate) async fn re_attach(
+        &self,
+        node_id: NodeId,
+    ) -> anyhow::Result<HashMap<TenantShardId, Generation>> {
+        let (write, result) = {
+            let mut result = HashMap::new();
+            let mut locked = self.state.lock().unwrap();
+            for (tenant_shard_id, shard) in locked.tenants.iter_mut() {
+                if shard.generation_pageserver == Some(node_id) {
+                    shard.generation += 1;
+                    result.insert(*tenant_shard_id, Generation::new(shard.generation));
+                }
+            }
+
+            (locked.save(), result)
+        };
+
+        write.commit().await?;
+        Ok(result)
+    }
+
     // TODO: when we start shard splitting, we must durably mark the tenant so that
     // on restart, we know that we must go through recovery (list shards that exist
     // and pick up where we left off and/or revert to parent shards).
@@ -359,8 +254,7 @@ impl Persistence {
 }
 
 /// Parts of [`crate::tenant_state::TenantState`] that are stored durably
-#[derive(Queryable, Selectable, Insertable, Serialize, Deserialize, Clone)]
-#[diesel(table_name = crate::schema::tenant_shards)]
+#[derive(Serialize, Deserialize, Clone)]
 pub(crate) struct TenantShardPersistence {
     #[serde(default)]
     pub(crate) tenant_id: String,
@@ -371,28 +265,16 @@ pub(crate) struct TenantShardPersistence {
     #[serde(default)]
     pub(crate) shard_stripe_size: i32,
 
-    // Latest generation number: next time we attach, increment this
-    // and use the incremented number when attaching
-    pub(crate) generation: i32,
-
     // Currently attached pageserver
     #[serde(rename = "pageserver")]
-    pub(crate) generation_pageserver: i64,
+    pub(crate) generation_pageserver: Option<NodeId>,
+
+    // Latest generation number: next time we attach, increment this
+    // and use the incremented number when attaching
+    pub(crate) generation: u32,
 
     #[serde(default)]
     pub(crate) placement_policy: String,
     #[serde(default)]
     pub(crate) config: String,
 }
-
-/// Parts of [`crate::node::Node`] that are stored durably
-#[derive(Serialize, Deserialize, Queryable, Selectable, Insertable)]
-#[diesel(table_name = crate::schema::nodes)]
-pub(crate) struct NodePersistence {
-    pub(crate) node_id: i64,
-    pub(crate) scheduling_policy: String,
-    pub(crate) listen_http_addr: String,
-    pub(crate) listen_http_port: i32,
-    pub(crate) listen_pg_addr: String,
-    pub(crate) listen_pg_port: i32,
-}

control_plane/attachment_service/src/reconciler.rs

@@ -14,7 +14,7 @@ use utils::generation::Generation;
 use utils::id::{NodeId, TimelineId};
 use utils::lsn::Lsn;
 
-use crate::compute_hook::{ComputeHook, NotifyError};
+use crate::compute_hook::ComputeHook;
 use crate::node::Node;
 use crate::tenant_state::{IntentState, ObservedState, ObservedStateLocation};
 
@@ -37,15 +37,9 @@ pub(super) struct Reconciler {
     pub(crate) pageservers: Arc<HashMap<NodeId, Node>>,
 
     /// A hook to notify the running postgres instances when we change the location
-    /// of a tenant.  Use this via [`Self::compute_notify`] to update our failure flag
-    /// and guarantee eventual retries.
+    /// of a tenant
     pub(crate) compute_hook: Arc<ComputeHook>,
 
-    /// To avoid stalling if the cloud control plane is unavailable, we may proceed
-    /// past failures in [`ComputeHook::notify`], but we _must_ remember that we failed
-    /// so that we can set [`crate::tenant_state::TenantState::pending_compute_notification`] to ensure a later retry.
-    pub(crate) compute_notify_failure: bool,
-
     /// A means to abort background reconciliation: it is essential to
     /// call this when something changes in the original TenantState that
     /// will make this reconciliation impossible or unnecessary, for
@@ -58,9 +52,7 @@ pub(super) struct Reconciler {
 }
 
 #[derive(thiserror::Error, Debug)]
-pub(crate) enum ReconcileError {
-    #[error(transparent)]
-    Notify(#[from] NotifyError),
+pub enum ReconcileError {
     #[error(transparent)]
     Other(#[from] anyhow::Error),
 }
@@ -325,19 +317,9 @@ impl Reconciler {
         }
 
         tracing::info!("🔁 Notifying compute to use pageserver {}", dest_ps_id);
-
-        // During a live migration it is unhelpful to proceed if we couldn't notify compute: if we detach
-        // the origin without notifying compute, we will render the tenant unavailable.
-        while let Err(e) = self.compute_notify().await {
-            match e {
-                NotifyError::Fatal(_) => return Err(anyhow::anyhow!(e)),
-                _ => {
-                    tracing::warn!(
-                        "Live migration blocked by compute notification error, retrying: {e}"
-                    );
-                }
-            }
-        }
+        self.compute_hook
+            .notify(self.tenant_shard_id, dest_ps_id)
+            .await?;
 
         // Downgrade the origin to secondary.  If the tenant's policy is PlacementPolicy::Single, then
         // this location will be deleted in the general case reconciliation that runs after this.
@@ -418,7 +400,15 @@ impl Reconciler {
                     wanted_conf.generation = self.generation.into();
                     tracing::info!("Observed configuration requires update.");
                     self.location_config(node_id, wanted_conf, None).await?;
-                    self.compute_notify().await?;
+                    if let Err(e) = self
+                        .compute_hook
+                        .notify(self.tenant_shard_id, node_id)
+                        .await
+                    {
+                        tracing::warn!(
+                            "Failed to notify compute of newly attached pageserver {node_id}: {e}"
+                        );
+                    }
                 }
             }
         }
@@ -471,29 +461,6 @@ impl Reconciler {
 
         Ok(())
     }
-
-    pub(crate) async fn compute_notify(&mut self) -> Result<(), NotifyError> {
-        // Whenever a particular Reconciler emits a notification, it is always notifying for the intended
-        // destination.
-        if let Some(node_id) = self.intent.attached {
-            let result = self
-                .compute_hook
-                .notify(self.tenant_shard_id, node_id, &self.cancel)
-                .await;
-            if let Err(e) = &result {
-                // It is up to the caller whether they want to drop out on this error, but they don't have to:
-                // in general we should avoid letting unavailability of the cloud control plane stop us from
-                // making progress.
-                tracing::warn!("Failed to notify compute of attached pageserver {node_id}: {e}");
-                // Set this flag so that in our ReconcileResult we will set the flag on the shard that it
-                // needs to retry at some point.
-                self.compute_notify_failure = true;
-            }
-            result
-        } else {
-            Ok(())
-        }
-    }
 }
 
 pub(crate) fn attached_location_conf(

control_plane/attachment_service/src/schema.rs

@@ -1,27 +0,0 @@
-// @generated automatically by Diesel CLI.
-
-diesel::table! {
-    nodes (node_id) {
-        node_id -> Int8,
-        scheduling_policy -> Varchar,
-        listen_http_addr -> Varchar,
-        listen_http_port -> Int4,
-        listen_pg_addr -> Varchar,
-        listen_pg_port -> Int4,
-    }
-}
-
-diesel::table! {
-    tenant_shards (tenant_id, shard_number, shard_count) {
-        tenant_id -> Varchar,
-        shard_number -> Int4,
-        shard_count -> Int4,
-        shard_stripe_size -> Int4,
-        generation -> Int4,
-        generation_pageserver -> Int8,
-        placement_policy -> Varchar,
-        config -> Text,
-    }
-}
-
-diesel::allow_tables_to_appear_in_same_query!(nodes, tenant_shards,);

control_plane/attachment_service/src/tenant_state.rs

@@ -71,12 +71,6 @@ pub(crate) struct TenantState {
     /// TODO: generalize to an array of recent events
     /// TOOD: use a ArcSwap instead of mutex for faster reads?
     pub(crate) last_error: std::sync::Arc<std::sync::Mutex<String>>,
-
-    /// If we have a pending compute notification that for some reason we weren't able to send,
-    /// set this to true. If this is set, calls to [`Self::maybe_reconcile`] will run a task to retry
-    /// sending it.  This is the mechanism by which compute notifications are included in the scope
-    /// of state that we publish externally in an eventually consistent way.
-    pub(crate) pending_compute_notification: bool,
 }
 
 #[derive(Default, Clone, Debug)]
@@ -170,9 +164,6 @@ pub(crate) struct ReconcileResult {
     pub(crate) tenant_shard_id: TenantShardId,
     pub(crate) generation: Generation,
     pub(crate) observed: ObservedState,
-
-    /// Set [`TenantState::pending_compute_notification`] from this flag
-    pub(crate) pending_compute_notification: bool,
 }
 
 impl IntentState {
@@ -235,7 +226,6 @@ impl TenantState {
             waiter: Arc::new(SeqWait::new(Sequence(0))),
             error_waiter: Arc::new(SeqWait::new(Sequence(0))),
             last_error: Arc::default(),
-            pending_compute_notification: false,
         }
     }
 
@@ -322,18 +312,6 @@ impl TenantState {
                     modified = true;
                 }
             }
-            Detached => {
-                // Should have no attached or secondary pageservers
-                if self.intent.attached.is_some() {
-                    self.intent.attached = None;
-                    modified = true;
-                }
-
-                if !self.intent.secondary.is_empty() {
-                    self.intent.secondary.clear();
-                    modified = true;
-                }
-            }
         }
 
         if modified {
@@ -343,38 +321,6 @@ impl TenantState {
         Ok(())
     }
 
-    /// Query whether the tenant's observed state for attached node matches its intent state, and if so,
-    /// yield the node ID.  This is appropriate for emitting compute hook notifications: we are checking that
-    /// the node in question is not only where we intend to attach, but that the tenant is indeed already attached there.
-    ///
-    /// Reconciliation may still be needed for other aspects of state such as secondaries (see [`Self::dirty`]): this
-    /// funciton should not be used to decide whether to reconcile.
-    pub(crate) fn stably_attached(&self) -> Option<NodeId> {
-        if let Some(attach_intent) = self.intent.attached {
-            match self.observed.locations.get(&attach_intent) {
-                Some(loc) => match &loc.conf {
-                    Some(conf) => match conf.mode {
-                        LocationConfigMode::AttachedMulti
-                        | LocationConfigMode::AttachedSingle
-                        | LocationConfigMode::AttachedStale => {
-                            // Our intent and observed state agree that this node is in an attached state.
-                            Some(attach_intent)
-                        }
-                        // Our observed config is not an attached state
-                        _ => None,
-                    },
-                    // Our observed state is None, i.e. in flux
-                    None => None,
-                },
-                // We have no observed state for this node
-                None => None,
-            }
-        } else {
-            // Our intent is not to attach
-            None
-        }
-    }
-
     fn dirty(&self) -> bool {
         if let Some(node_id) = self.intent.attached {
             let wanted_conf = attached_location_conf(self.generation, &self.shard, &self.config);
@@ -396,12 +342,6 @@ impl TenantState {
             }
         }
 
-        // Even if there is no pageserver work to be done, if we have a pending notification to computes,
-        // wake up a reconciler to send it.
-        if self.pending_compute_notification {
-            return true;
-        }
-
         false
     }
 
@@ -463,13 +403,11 @@ impl TenantState {
             service_config: service_config.clone(),
             cancel: cancel.clone(),
             persistence: persistence.clone(),
-            compute_notify_failure: false,
         };
 
         let reconcile_seq = self.sequence;
 
         tracing::info!("Spawning Reconciler for sequence {}", self.sequence);
-        let must_notify = self.pending_compute_notification;
         let join_handle = tokio::task::spawn(async move {
             // Wait for any previous reconcile task to complete before we start
             if let Some(old_handle) = old_handle {
@@ -488,16 +426,7 @@ impl TenantState {
                 return;
             }
 
-            // Attempt to make observed state match intent state
             let result = reconciler.reconcile().await;
-
-            // If we know we had a pending compute notification from some previous action, send a notification irrespective
-            // of whether the above reconcile() did any work
-            if result.is_ok() && must_notify {
-                // If this fails we will send the need to retry in [`ReconcileResult::pending_compute_notification`]
-                reconciler.compute_notify().await.ok();
-            }
-
             result_tx
                 .send(ReconcileResult {
                     sequence: reconcile_seq,
@@ -505,7 +434,6 @@ impl TenantState {
                     tenant_shard_id: reconciler.tenant_shard_id,
                     generation: reconciler.generation,
                     observed: reconciler.observed,
-                    pending_compute_notification: reconciler.compute_notify_failure,
                 })
                 .ok();
         });

control_plane/src/attachment_service.rs

@@ -1,11 +1,5 @@
 use crate::{background_process, local_env::LocalEnv};
-use camino::{Utf8Path, Utf8PathBuf};
-use diesel::{
-    backend::Backend,
-    query_builder::{AstPass, QueryFragment, QueryId},
-    Connection, PgConnection, QueryResult, RunQueryDsl,
-};
-use diesel_migrations::{HarnessWithOutput, MigrationHarness};
+use camino::Utf8PathBuf;
 use hyper::Method;
 use pageserver_api::{
     models::{ShardParameters, TenantCreateRequest, TimelineCreateRequest, TimelineInfo},
@@ -13,11 +7,10 @@ use pageserver_api::{
 };
 use pageserver_client::mgmt_api::ResponseErrorMessageExt;
 use postgres_backend::AuthType;
+use postgres_connection::parse_host_port;
 use serde::{de::DeserializeOwned, Deserialize, Serialize};
-use std::{env, str::FromStr};
-use tokio::process::Command;
+use std::{path::PathBuf, process::Child, str::FromStr};
 use tracing::instrument;
-use url::Url;
 use utils::{
     auth::{Claims, Scope},
     id::{NodeId, TenantId},
@@ -26,17 +19,14 @@ use utils::{
 pub struct AttachmentService {
     env: LocalEnv,
     listen: String,
-    path: Utf8PathBuf,
+    path: PathBuf,
     jwt_token: Option<String>,
-    public_key: Option<String>,
-    postgres_port: u16,
+    public_key_path: Option<Utf8PathBuf>,
     client: reqwest::Client,
 }
 
 const COMMAND: &str = "attachment_service";
 
-const ATTACHMENT_SERVICE_POSTGRES_VERSION: u32 = 16;
-
 #[derive(Serialize, Deserialize)]
 pub struct AttachHookRequest {
     pub tenant_shard_id: TenantShardId,
@@ -60,7 +50,6 @@ pub struct InspectResponse {
 
 #[derive(Serialize, Deserialize)]
 pub struct TenantCreateResponseShard {
-    pub shard_id: TenantShardId,
     pub node_id: NodeId,
     pub generation: u32,
 }
@@ -180,9 +169,7 @@ pub struct TenantShardMigrateResponse {}
 
 impl AttachmentService {
     pub fn from_env(env: &LocalEnv) -> Self {
-        let path = Utf8PathBuf::from_path_buf(env.base_data_dir.clone())
-            .unwrap()
-            .join("attachments.json");
+        let path = env.base_data_dir.join("attachments.json");
 
         // Makes no sense to construct this if pageservers aren't going to use it: assume
         // pageservers have control plane API set
@@ -194,20 +181,13 @@ impl AttachmentService {
             listen_url.port().unwrap()
         );
 
-        // Convention: NeonEnv in python tests reserves the next port after the control_plane_api
-        // port, for use by our captive postgres.
-        let postgres_port = listen_url
-            .port()
-            .expect("Control plane API setting should always have a port")
-            + 1;
-
         // Assume all pageservers have symmetric auth configuration: this service
         // expects to use one JWT token to talk to all of them.
         let ps_conf = env
             .pageservers
             .first()
             .expect("Config is validated to contain at least one pageserver");
-        let (jwt_token, public_key) = match ps_conf.http_auth_type {
+        let (jwt_token, public_key_path) = match ps_conf.http_auth_type {
             AuthType::Trust => (None, None),
             AuthType::NeonJWT => {
                 let jwt_token = env
@@ -219,26 +199,7 @@ impl AttachmentService {
                 let public_key_path =
                     camino::Utf8PathBuf::try_from(env.base_data_dir.join("auth_public_key.pem"))
                         .unwrap();
-
-                // This service takes keys as a string rather than as a path to a file/dir: read the key into memory.
-                let public_key = if std::fs::metadata(&public_key_path)
-                    .expect("Can't stat public key")
-                    .is_dir()
-                {
-                    // Our config may specify a directory: this is for the pageserver's ability to handle multiple
-                    // keys.  We only use one key at a time, so, arbitrarily load the first one in the directory.
-                    let mut dir =
-                        std::fs::read_dir(&public_key_path).expect("Can't readdir public key path");
-                    let dent = dir
-                        .next()
-                        .expect("Empty key dir")
-                        .expect("Error reading key dir");
-
-                    std::fs::read_to_string(dent.path()).expect("Can't read public key")
-                } else {
-                    std::fs::read_to_string(&public_key_path).expect("Can't read public key")
-                };
-                (Some(jwt_token), Some(public_key))
+                (Some(jwt_token), Some(public_key_path))
             }
         };
 
@@ -247,8 +208,7 @@ impl AttachmentService {
             path,
             listen,
             jwt_token,
-            public_key,
-            postgres_port,
+            public_key_path,
             client: reqwest::ClientBuilder::new()
                 .build()
                 .expect("Failed to construct http client"),
@@ -260,229 +220,22 @@ impl AttachmentService {
             .expect("non-Unicode path")
     }
 
-    /// PIDFile for the postgres instance used to store attachment service state
-    fn postgres_pid_file(&self) -> Utf8PathBuf {
-        Utf8PathBuf::from_path_buf(
-            self.env
-                .base_data_dir
-                .join("attachment_service_postgres.pid"),
-        )
-        .expect("non-Unicode path")
-    }
+    pub async fn start(&self) -> anyhow::Result<Child> {
+        let path_str = self.path.to_string_lossy();
 
-    /// In order to access database migrations, we need to find the Neon source tree
-    async fn find_source_root(&self) -> anyhow::Result<Utf8PathBuf> {
-        // We assume that either prd or our binary is in the source tree. The former is usually
-        // true for automated test runners, the latter is usually true for developer workstations. Often
-        // both are true, which is fine.
-        let candidate_start_points = [
-            // Current working directory
-            Utf8PathBuf::from_path_buf(std::env::current_dir()?).unwrap(),
-            // Directory containing the binary we're running inside
-            Utf8PathBuf::from_path_buf(env::current_exe()?.parent().unwrap().to_owned()).unwrap(),
-        ];
-
-        // For each candidate start point, search through ancestors looking for a neon.git source tree root
-        for start_point in &candidate_start_points {
-            // Start from the build dir: assumes we are running out of a built neon source tree
-            for path in start_point.ancestors() {
-                // A crude approximation: the root of the source tree is whatever contains a "control_plane"
-                // subdirectory.
-                let control_plane = path.join("control_plane");
-                if tokio::fs::try_exists(&control_plane).await? {
-                    return Ok(path.to_owned());
-                }
-            }
-        }
-
-        // Fall-through
-        Err(anyhow::anyhow!(
-            "Could not find control_plane src dir, after searching ancestors of {candidate_start_points:?}"
-        ))
-    }
-
-    /// Find the directory containing postgres binaries, such as `initdb` and `pg_ctl`
-    ///
-    /// This usually uses ATTACHMENT_SERVICE_POSTGRES_VERSION of postgres, but will fall back
-    /// to other versions if that one isn't found.  Some automated tests create circumstances
-    /// where only one version is available in pg_distrib_dir, such as `test_remote_extensions`.
-    pub async fn get_pg_bin_dir(&self) -> anyhow::Result<Utf8PathBuf> {
-        let prefer_versions = [ATTACHMENT_SERVICE_POSTGRES_VERSION, 15, 14];
-
-        for v in prefer_versions {
-            let path = Utf8PathBuf::from_path_buf(self.env.pg_bin_dir(v)?).unwrap();
-            if tokio::fs::try_exists(&path).await? {
-                return Ok(path);
-            }
-        }
-
-        // Fall through
-        anyhow::bail!(
-            "Postgres binaries not found in {}",
-            self.env.pg_distrib_dir.display()
-        );
-    }
-
-    /// Readiness check for our postgres process
-    async fn pg_isready(&self, pg_bin_dir: &Utf8Path) -> anyhow::Result<bool> {
-        let bin_path = pg_bin_dir.join("pg_isready");
-        let args = ["-h", "localhost", "-p", &format!("{}", self.postgres_port)];
-        let exitcode = Command::new(bin_path).args(args).spawn()?.wait().await?;
-
-        Ok(exitcode.success())
-    }
-
-    /// Create our database if it doesn't exist, and run migrations.
-    ///
-    /// This function is equivalent to the `diesel setup` command in the diesel CLI.  We implement
-    /// the same steps by hand to avoid imposing a dependency on installing diesel-cli for developers
-    /// who just want to run `cargo neon_local` without knowing about diesel.
-    ///
-    /// Returns the database url
-    pub async fn setup_database(&self) -> anyhow::Result<String> {
-        let database_url = format!(
-            "postgresql://localhost:{}/attachment_service",
-            self.postgres_port
-        );
-        println!("Running attachment service database setup...");
-        fn change_database_of_url(database_url: &str, default_database: &str) -> (String, String) {
-            let base = ::url::Url::parse(database_url).unwrap();
-            let database = base.path_segments().unwrap().last().unwrap().to_owned();
-            let mut new_url = base.join(default_database).unwrap();
-            new_url.set_query(base.query());
-            (database, new_url.into())
-        }
-
-        #[derive(Debug, Clone)]
-        pub struct CreateDatabaseStatement {
-            db_name: String,
-        }
-
-        impl CreateDatabaseStatement {
-            pub fn new(db_name: &str) -> Self {
-                CreateDatabaseStatement {
-                    db_name: db_name.to_owned(),
-                }
-            }
-        }
-
-        impl<DB: Backend> QueryFragment<DB> for CreateDatabaseStatement {
-            fn walk_ast<'b>(&'b self, mut out: AstPass<'_, 'b, DB>) -> QueryResult<()> {
-                out.push_sql("CREATE DATABASE ");
-                out.push_identifier(&self.db_name)?;
-                Ok(())
-            }
-        }
-
-        impl<Conn> RunQueryDsl<Conn> for CreateDatabaseStatement {}
-
-        impl QueryId for CreateDatabaseStatement {
-            type QueryId = ();
-
-            const HAS_STATIC_QUERY_ID: bool = false;
-        }
-        if PgConnection::establish(&database_url).is_err() {
-            let (database, postgres_url) = change_database_of_url(&database_url, "postgres");
-            println!("Creating database: {database}");
-            let mut conn = PgConnection::establish(&postgres_url)?;
-            CreateDatabaseStatement::new(&database).execute(&mut conn)?;
-        }
-        let mut conn = PgConnection::establish(&database_url)?;
-
-        let migrations_dir = self
-            .find_source_root()
-            .await?
-            .join("control_plane/attachment_service/migrations");
-
-        let migrations = diesel_migrations::FileBasedMigrations::from_path(migrations_dir)?;
-        println!("Running migrations in {}", migrations.path().display());
-        HarnessWithOutput::write_to_stdout(&mut conn)
-            .run_pending_migrations(migrations)
-            .map(|_| ())
-            .map_err(|e| anyhow::anyhow!(e))?;
-
-        println!("Migrations complete");
-
-        Ok(database_url)
-    }
-
-    pub async fn start(&self) -> anyhow::Result<()> {
-        // Start a vanilla Postgres process used by the attachment service for persistence.
-        let pg_data_path = Utf8PathBuf::from_path_buf(self.env.base_data_dir.clone())
-            .unwrap()
-            .join("attachment_service_db");
-        let pg_bin_dir = self.get_pg_bin_dir().await?;
-        let pg_log_path = pg_data_path.join("postgres.log");
-
-        if !tokio::fs::try_exists(&pg_data_path).await? {
-            // Initialize empty database
-            let initdb_path = pg_bin_dir.join("initdb");
-            let mut child = Command::new(&initdb_path)
-                .args(["-D", pg_data_path.as_ref()])
-                .spawn()
-                .expect("Failed to spawn initdb");
-            let status = child.wait().await?;
-            if !status.success() {
-                anyhow::bail!("initdb failed with status {status}");
-            }
-
-            tokio::fs::write(
-                &pg_data_path.join("postgresql.conf"),
-                format!("port = {}", self.postgres_port),
-            )
-            .await?;
-        };
-
-        println!("Starting attachment service database...");
-        let db_start_args = [
-            "-w",
-            "-D",
-            pg_data_path.as_ref(),
-            "-l",
-            pg_log_path.as_ref(),
-            "start",
-        ];
-
-        background_process::start_process(
-            "attachment_service_db",
-            &self.env.base_data_dir,
-            pg_bin_dir.join("pg_ctl").as_std_path(),
-            db_start_args,
-            [],
-            background_process::InitialPidFile::Create(self.postgres_pid_file()),
-            || self.pg_isready(&pg_bin_dir),
-        )
-        .await?;
-
-        // Run migrations on every startup, in case something changed.
-        let database_url = self.setup_database().await?;
-
-        let mut args = vec![
-            "-l",
-            &self.listen,
-            "-p",
-            self.path.as_ref(),
-            "--database-url",
-            &database_url,
-        ]
-        .into_iter()
-        .map(|s| s.to_string())
-        .collect::<Vec<_>>();
+        let mut args = vec!["-l", &self.listen, "-p", &path_str]
+            .into_iter()
+            .map(|s| s.to_string())
+            .collect::<Vec<_>>();
         if let Some(jwt_token) = &self.jwt_token {
             args.push(format!("--jwt-token={jwt_token}"));
         }
 
-        if let Some(public_key) = &self.public_key {
-            args.push(format!("--public-key=\"{public_key}\""));
+        if let Some(public_key_path) = &self.public_key_path {
+            args.push(format!("--public-key={public_key_path}"));
         }
 
-        if let Some(control_plane_compute_hook_api) = &self.env.control_plane_compute_hook_api {
-            args.push(format!(
-                "--compute-hook-url={control_plane_compute_hook_api}"
-            ));
-        }
-
-        background_process::start_process(
+        let result = background_process::start_process(
             COMMAND,
             &self.env.base_data_dir,
             &self.env.attachment_service_bin(),
@@ -499,46 +252,29 @@ impl AttachmentService {
                 }
             },
         )
-        .await?;
+        .await;
 
-        Ok(())
-    }
-
-    pub async fn stop(&self, immediate: bool) -> anyhow::Result<()> {
-        background_process::stop_process(immediate, COMMAND, &self.pid_file())?;
-
-        let pg_data_path = self.env.base_data_dir.join("attachment_service_db");
-        let pg_bin_dir = self.get_pg_bin_dir().await?;
-
-        println!("Stopping attachment service database...");
-        let pg_stop_args = ["-D", &pg_data_path.to_string_lossy(), "stop"];
-        let stop_status = Command::new(pg_bin_dir.join("pg_ctl"))
-            .args(pg_stop_args)
-            .spawn()?
-            .wait()
+        for ps_conf in &self.env.pageservers {
+            let (pg_host, pg_port) =
+                parse_host_port(&ps_conf.listen_pg_addr).expect("Unable to parse listen_pg_addr");
+            let (http_host, http_port) = parse_host_port(&ps_conf.listen_http_addr)
+                .expect("Unable to parse listen_http_addr");
+            self.node_register(NodeRegisterRequest {
+                node_id: ps_conf.id,
+                listen_pg_addr: pg_host.to_string(),
+                listen_pg_port: pg_port.unwrap_or(5432),
+                listen_http_addr: http_host.to_string(),
+                listen_http_port: http_port.unwrap_or(80),
+            })
             .await?;
-        if !stop_status.success() {
-            let pg_status_args = ["-D", &pg_data_path.to_string_lossy(), "status"];
-            let status_exitcode = Command::new(pg_bin_dir.join("pg_ctl"))
-                .args(pg_status_args)
-                .spawn()?
-                .wait()
-                .await?;
-
-            // pg_ctl status returns this exit code if postgres is not running: in this case it is
-            // fine that stop failed.  Otherwise it is an error that stop failed.
-            const PG_STATUS_NOT_RUNNING: i32 = 3;
-            if Some(PG_STATUS_NOT_RUNNING) == status_exitcode.code() {
-                println!("Attachment service data base is already stopped");
-                return Ok(());
-            } else {
-                anyhow::bail!("Failed to stop attachment service database: {stop_status}")
-            }
         }
 
-        Ok(())
+        result
     }
 
+    pub fn stop(&self, immediate: bool) -> anyhow::Result<()> {
+        background_process::stop_process(immediate, COMMAND, &self.pid_file())
+    }
     /// Simple HTTP request wrapper for calling into attachment service
     async fn dispatch<RQ, RS>(
         &self,
@@ -550,15 +286,13 @@ impl AttachmentService {
         RQ: Serialize + Sized,
         RS: DeserializeOwned + Sized,
     {
-        // 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 url = Url::from_str(&format!(
-            "http://{}:{}/{path}",
-            listen_url.host_str().unwrap(),
-            listen_url.port().unwrap()
-        ))
-        .unwrap();
+        let url = self
+            .env
+            .control_plane_api
+            .clone()
+            .unwrap()
+            .join(&path)
+            .unwrap();
 
         let mut builder = self.client.request(method, url);
         if let Some(body) = body {
@@ -595,7 +329,7 @@ impl AttachmentService {
         let response = self
             .dispatch::<_, AttachHookResponse>(
                 Method::POST,
-                "debug/v1/attach-hook".to_string(),
+                "attach-hook".to_string(),
                 Some(request),
             )
             .await?;
@@ -611,11 +345,7 @@ impl AttachmentService {
         let request = InspectRequest { tenant_shard_id };
 
         let response = self
-            .dispatch::<_, InspectResponse>(
-                Method::POST,
-                "debug/v1/inspect".to_string(),
-                Some(request),
-            )
+            .dispatch::<_, InspectResponse>(Method::POST, "inspect".to_string(), Some(request))
             .await?;
 
         Ok(response.attachment)
@@ -626,18 +356,14 @@ impl AttachmentService {
         &self,
         req: TenantCreateRequest,
     ) -> anyhow::Result<TenantCreateResponse> {
-        self.dispatch(Method::POST, "v1/tenant".to_string(), Some(req))
+        self.dispatch(Method::POST, "tenant".to_string(), Some(req))
             .await
     }
 
     #[instrument(skip(self))]
     pub async fn tenant_locate(&self, tenant_id: TenantId) -> anyhow::Result<TenantLocateResponse> {
-        self.dispatch::<(), _>(
-            Method::GET,
-            format!("control/v1/tenant/{tenant_id}/locate"),
-            None,
-        )
-        .await
+        self.dispatch::<(), _>(Method::GET, format!("tenant/{tenant_id}/locate"), None)
+            .await
     }
 
     #[instrument(skip(self))]
@@ -659,7 +385,7 @@ impl AttachmentService {
 
     #[instrument(skip_all, fields(node_id=%req.node_id))]
     pub async fn node_register(&self, req: NodeRegisterRequest) -> anyhow::Result<()> {
-        self.dispatch::<_, ()>(Method::POST, "control/v1/node".to_string(), Some(req))
+        self.dispatch::<_, ()>(Method::POST, "node".to_string(), Some(req))
             .await
     }
 
@@ -667,7 +393,7 @@ impl AttachmentService {
     pub async fn node_configure(&self, req: NodeConfigureRequest) -> anyhow::Result<()> {
         self.dispatch::<_, ()>(
             Method::PUT,
-            format!("control/v1/node/{}/config", req.node_id),
+            format!("node/{}/config", req.node_id),
             Some(req),
         )
         .await
@@ -687,7 +413,7 @@ impl AttachmentService {
     ) -> anyhow::Result<TimelineInfo> {
         self.dispatch(
             Method::POST,
-            format!("v1/tenant/{tenant_id}/timeline"),
+            format!("tenant/{tenant_id}/timeline"),
             Some(req),
         )
         .await

control_plane/src/background_process.rs

@@ -17,7 +17,7 @@ use std::io::Write;
 use std::os::unix::prelude::AsRawFd;
 use std::os::unix::process::CommandExt;
 use std::path::Path;
-use std::process::Command;
+use std::process::{Child, Command};
 use std::time::Duration;
 use std::{fs, io, thread};
 
@@ -60,7 +60,7 @@ pub async fn start_process<F, Fut, AI, A, EI>(
     envs: EI,
     initial_pid_file: InitialPidFile,
     process_status_check: F,
-) -> anyhow::Result<()>
+) -> anyhow::Result<Child>
 where
     F: Fn() -> Fut,
     Fut: std::future::Future<Output = anyhow::Result<bool>>,
@@ -98,7 +98,7 @@ where
         InitialPidFile::Expect(path) => path,
     };
 
-    let spawned_process = filled_cmd.spawn().with_context(|| {
+    let mut spawned_process = filled_cmd.spawn().with_context(|| {
         format!("Could not spawn {process_name}, see console output and log files for details.")
     })?;
     let pid = spawned_process.id();
@@ -106,26 +106,12 @@ where
         i32::try_from(pid)
             .with_context(|| format!("Subprocess {process_name} has invalid pid {pid}"))?,
     );
-    // set up a scopeguard to kill & wait for the child in case we panic or bail below
-    let spawned_process = scopeguard::guard(spawned_process, |mut spawned_process| {
-        println!("SIGKILL & wait the started process");
-        (|| {
-            // TODO: use another signal that can be caught by the child so it can clean up any children it spawned (e..g, walredo).
-            spawned_process.kill().context("SIGKILL child")?;
-            spawned_process.wait().context("wait() for child process")?;
-            anyhow::Ok(())
-        })()
-        .with_context(|| format!("scopeguard kill&wait child {process_name:?}"))
-        .unwrap();
-    });
 
     for retries in 0..RETRIES {
         match process_started(pid, pid_file_to_check, &process_status_check).await {
             Ok(true) => {
-                println!("\n{process_name} started and passed status check, pid: {pid}");
-                // leak the child process, it'll outlive this neon_local invocation
-                drop(scopeguard::ScopeGuard::into_inner(spawned_process));
-                return Ok(());
+                println!("\n{process_name} started, pid: {pid}");
+                return Ok(spawned_process);
             }
             Ok(false) => {
                 if retries == NOTICE_AFTER_RETRIES {
@@ -140,15 +126,16 @@ where
                 thread::sleep(Duration::from_millis(RETRY_INTERVAL_MILLIS));
             }
             Err(e) => {
-                println!("error starting process {process_name:?}: {e:#}");
+                println!("{process_name} failed to start: {e:#}");
+                if let Err(e) = spawned_process.kill() {
+                    println!("Could not stop {process_name} subprocess: {e:#}")
+                };
                 return Err(e);
             }
         }
     }
     println!();
-    anyhow::bail!(
-        "{process_name} did not start+pass status checks within {RETRY_UNTIL_SECS} seconds"
-    );
+    anyhow::bail!("{process_name} did not start in {RETRY_UNTIL_SECS} seconds");
 }
 
 /// Stops the process, using the pid file given. Returns Ok also if the process is already not running.
@@ -256,9 +243,7 @@ fn fill_remote_storage_secrets_vars(mut cmd: &mut Command) -> &mut Command {
     for env_key in [
         "AWS_ACCESS_KEY_ID",
         "AWS_SECRET_ACCESS_KEY",
-        "AWS_PROFILE",
-        // HOME is needed in combination with `AWS_PROFILE` to pick up the SSO sessions.
-        "HOME",
+        "AWS_SESSION_TOKEN",
         "AZURE_STORAGE_ACCOUNT",
         "AZURE_STORAGE_ACCESS_KEY",
     ] {

control_plane/src/bin/neon_local.rs

@@ -51,7 +51,7 @@ project_git_version!(GIT_VERSION);
 
 const DEFAULT_PG_VERSION: &str = "15";
 
-const DEFAULT_PAGESERVER_CONTROL_PLANE_API: &str = "http://127.0.0.1:1234/upcall/v1/";
+const DEFAULT_PAGESERVER_CONTROL_PLANE_API: &str = "http://127.0.0.1:1234/";
 
 fn default_conf(num_pageservers: u16) -> String {
     let mut template = format!(
@@ -135,7 +135,7 @@ fn main() -> Result<()> {
             "tenant" => rt.block_on(handle_tenant(sub_args, &mut env)),
             "timeline" => rt.block_on(handle_timeline(sub_args, &mut env)),
             "start" => rt.block_on(handle_start_all(sub_args, &env)),
-            "stop" => rt.block_on(handle_stop_all(sub_args, &env)),
+            "stop" => handle_stop_all(sub_args, &env),
             "pageserver" => rt.block_on(handle_pageserver(sub_args, &env)),
             "attachment_service" => rt.block_on(handle_attachment_service(sub_args, &env)),
             "safekeeper" => rt.block_on(handle_safekeeper(sub_args, &env)),
@@ -795,7 +795,7 @@ async fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Re
                     &endpoint.timeline_id.to_string(),
                     branch_name,
                     lsn_str.as_str(),
-                    &format!("{}", endpoint.status()),
+                    endpoint.status(),
                 ]);
             }
 
@@ -1056,9 +1056,8 @@ fn get_pageserver(env: &local_env::LocalEnv, args: &ArgMatches) -> Result<PageSe
 async fn handle_pageserver(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
     match sub_match.subcommand() {
         Some(("start", subcommand_args)) => {
-            let register = subcommand_args.get_one::<bool>("register").unwrap_or(&true);
             if let Err(e) = get_pageserver(env, subcommand_args)?
-                .start(&pageserver_config_overrides(subcommand_args), *register)
+                .start(&pageserver_config_overrides(subcommand_args))
                 .await
             {
                 eprintln!("pageserver start failed: {e}");
@@ -1087,7 +1086,24 @@ async fn handle_pageserver(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
             }
 
             if let Err(e) = pageserver
-                .start(&pageserver_config_overrides(subcommand_args), false)
+                .start(&pageserver_config_overrides(subcommand_args))
+                .await
+            {
+                eprintln!("pageserver start failed: {e}");
+                exit(1);
+            }
+        }
+
+        Some(("migrate", subcommand_args)) => {
+            let pageserver = get_pageserver(env, subcommand_args)?;
+            //TODO what shutdown strategy should we use here?
+            if let Err(e) = pageserver.stop(false) {
+                eprintln!("pageserver stop failed: {}", e);
+                exit(1);
+            }
+
+            if let Err(e) = pageserver
+                .start(&pageserver_config_overrides(subcommand_args))
                 .await
             {
                 eprintln!("pageserver start failed: {e}");
@@ -1145,7 +1161,7 @@ async fn handle_attachment_service(
                 .map(|s| s.as_str())
                 == Some("immediate");
 
-            if let Err(e) = svc.stop(immediate).await {
+            if let Err(e) = svc.stop(immediate) {
                 eprintln!("stop failed: {}", e);
                 exit(1);
             }
@@ -1241,7 +1257,7 @@ async fn handle_start_all(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
         let attachment_service = AttachmentService::from_env(env);
         if let Err(e) = attachment_service.start().await {
             eprintln!("attachment_service start failed: {:#}", e);
-            try_stop_all(env, true).await;
+            try_stop_all(env, true);
             exit(1);
         }
     }
@@ -1249,11 +1265,11 @@ async fn handle_start_all(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
     for ps_conf in &env.pageservers {
         let pageserver = PageServerNode::from_env(env, ps_conf);
         if let Err(e) = pageserver
-            .start(&pageserver_config_overrides(sub_match), true)
+            .start(&pageserver_config_overrides(sub_match))
             .await
         {
             eprintln!("pageserver {} start failed: {:#}", ps_conf.id, e);
-            try_stop_all(env, true).await;
+            try_stop_all(env, true);
             exit(1);
         }
     }
@@ -1262,23 +1278,23 @@ async fn handle_start_all(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
         let safekeeper = SafekeeperNode::from_env(env, node);
         if let Err(e) = safekeeper.start(vec![]).await {
             eprintln!("safekeeper {} start failed: {:#}", safekeeper.id, e);
-            try_stop_all(env, false).await;
+            try_stop_all(env, false);
             exit(1);
         }
     }
     Ok(())
 }
 
-async fn handle_stop_all(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
+fn handle_stop_all(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
     let immediate =
         sub_match.get_one::<String>("stop-mode").map(|s| s.as_str()) == Some("immediate");
 
-    try_stop_all(env, immediate).await;
+    try_stop_all(env, immediate);
 
     Ok(())
 }
 
-async fn try_stop_all(env: &local_env::LocalEnv, immediate: bool) {
+fn try_stop_all(env: &local_env::LocalEnv, immediate: bool) {
     // Stop all endpoints
     match ComputeControlPlane::load(env.clone()) {
         Ok(cplane) => {
@@ -1313,7 +1329,7 @@ async fn try_stop_all(env: &local_env::LocalEnv, immediate: bool) {
 
     if env.control_plane_api.is_some() {
         let attachment_service = AttachmentService::from_env(env);
-        if let Err(e) = attachment_service.stop(immediate).await {
+        if let Err(e) = attachment_service.stop(immediate) {
             eprintln!("attachment service stop failed: {e:#}");
         }
     }
@@ -1533,11 +1549,7 @@ fn cli() -> Command {
                 .subcommand(Command::new("status"))
                 .subcommand(Command::new("start")
                     .about("Start local pageserver")
-                    .arg(pageserver_config_args.clone()).arg(Arg::new("register")
-                    .long("register")
-                    .default_value("true").required(false)
-                    .value_parser(value_parser!(bool))
-                    .value_name("register"))
+                    .arg(pageserver_config_args.clone())
                 )
                 .subcommand(Command::new("stop")
                     .about("Stop local pageserver")

control_plane/src/endpoint.rs

@@ -184,7 +184,7 @@ impl ComputeControlPlane {
                 v.tenant_id == tenant_id
                     && v.timeline_id == timeline_id
                     && v.mode == mode
-                    && v.status() != EndpointStatus::Stopped
+                    && v.status() != "stopped"
             });
 
             if let Some((key, _)) = duplicates.next() {
@@ -223,26 +223,6 @@ pub struct Endpoint {
     features: Vec<ComputeFeature>,
 }
 
-#[derive(PartialEq, Eq)]
-pub enum EndpointStatus {
-    Running,
-    Stopped,
-    Crashed,
-    RunningNoPidfile,
-}
-
-impl std::fmt::Display for EndpointStatus {
-    fn fmt(&self, writer: &mut std::fmt::Formatter) -> std::fmt::Result {
-        let s = match self {
-            Self::Running => "running",
-            Self::Stopped => "stopped",
-            Self::Crashed => "crashed",
-            Self::RunningNoPidfile => "running, no pidfile",
-        };
-        write!(writer, "{}", s)
-    }
-}
-
 impl Endpoint {
     fn from_dir_entry(entry: std::fs::DirEntry, env: &LocalEnv) -> Result<Endpoint> {
         if !entry.file_type()?.is_dir() {
@@ -400,16 +380,16 @@ impl Endpoint {
         self.endpoint_path().join("pgdata")
     }
 
-    pub fn status(&self) -> EndpointStatus {
+    pub fn status(&self) -> &str {
         let timeout = Duration::from_millis(300);
         let has_pidfile = self.pgdata().join("postmaster.pid").exists();
         let can_connect = TcpStream::connect_timeout(&self.pg_address, timeout).is_ok();
 
         match (has_pidfile, can_connect) {
-            (true, true) => EndpointStatus::Running,
-            (false, false) => EndpointStatus::Stopped,
-            (true, false) => EndpointStatus::Crashed,
-            (false, true) => EndpointStatus::RunningNoPidfile,
+            (true, true) => "running",
+            (false, false) => "stopped",
+            (true, false) => "crashed",
+            (false, true) => "running, no pidfile",
         }
     }
 
@@ -458,7 +438,7 @@ impl Endpoint {
     }
 
     fn wait_for_compute_ctl_to_exit(&self, send_sigterm: bool) -> Result<()> {
-        // TODO use background_process::stop_process instead: https://github.com/neondatabase/neon/pull/6482
+        // TODO use background_process::stop_process instead
         let pidfile_path = self.endpoint_path().join("compute_ctl.pid");
         let pid: u32 = std::fs::read_to_string(pidfile_path)?.parse()?;
         let pid = nix::unistd::Pid::from_raw(pid as i32);
@@ -501,7 +481,7 @@ impl Endpoint {
         remote_ext_config: Option<&String>,
         shard_stripe_size: usize,
     ) -> Result<()> {
-        if self.status() == EndpointStatus::Running {
+        if self.status() == "running" {
             anyhow::bail!("The endpoint is already running");
         }
 
@@ -603,21 +583,9 @@ impl Endpoint {
         }
 
         let child = cmd.spawn()?;
-        // set up a scopeguard to kill & wait for the child in case we panic or bail below
-        let child = scopeguard::guard(child, |mut child| {
-            println!("SIGKILL & wait the started process");
-            (|| {
-                // TODO: use another signal that can be caught by the child so it can clean up any children it spawned
-                child.kill().context("SIGKILL child")?;
-                child.wait().context("wait() for child process")?;
-                anyhow::Ok(())
-            })()
-            .with_context(|| format!("scopeguard kill&wait child {child:?}"))
-            .unwrap();
-        });
 
         // Write down the pid so we can wait for it when we want to stop
-        // TODO use background_process::start_process instead: https://github.com/neondatabase/neon/pull/6482
+        // TODO use background_process::start_process instead
         let pid = child.id();
         let pidfile_path = self.endpoint_path().join("compute_ctl.pid");
         std::fs::write(pidfile_path, pid.to_string())?;
@@ -666,9 +634,6 @@ impl Endpoint {
             std::thread::sleep(ATTEMPT_INTERVAL);
         }
 
-        // disarm the scopeguard, let the child outlive this function (and neon_local invoction)
-        drop(scopeguard::ScopeGuard::into_inner(child));
-
         Ok(())
     }
 

control_plane/src/local_env.rs

@@ -72,16 +72,11 @@ pub struct LocalEnv {
     #[serde(default)]
     pub safekeepers: Vec<SafekeeperConf>,
 
-    // Control plane upcall API for pageserver: if None, we will not run attachment_service.  If set, this will
+    // Control plane location: if None, we will not run attachment_service.  If set, this will
     // be propagated into each pageserver's configuration.
     #[serde(default)]
     pub control_plane_api: Option<Url>,
 
-    // Control plane upcall API for attachment service.  If set, this will be propagated into the
-    // attachment service's configuration.
-    #[serde(default)]
-    pub control_plane_compute_hook_api: Option<Url>,
-
     /// Keep human-readable aliases in memory (and persist them to config), to hide ZId hex strings from the user.
     #[serde(default)]
     // A `HashMap<String, HashMap<TenantId, TimelineId>>` would be more appropriate here,
@@ -228,11 +223,7 @@ impl LocalEnv {
     }
 
     pub fn attachment_service_bin(&self) -> PathBuf {
-        // Irrespective of configuration, attachment service binary is always
-        // run from the same location as neon_local.  This means that for compatibility
-        // tests that run old pageserver/safekeeper, they still run latest attachment service.
-        let neon_local_bin_dir = env::current_exe().unwrap().parent().unwrap().to_owned();
-        neon_local_bin_dir.join("attachment_service")
+        self.neon_distrib_dir.join("attachment_service")
     }
 
     pub fn safekeeper_bin(&self) -> PathBuf {

control_plane/src/pageserver.rs

@@ -11,7 +11,7 @@ use std::io;
 use std::io::Write;
 use std::num::NonZeroU64;
 use std::path::PathBuf;
-use std::process::Command;
+use std::process::{Child, Command};
 use std::time::Duration;
 
 use anyhow::{bail, Context};
@@ -30,7 +30,6 @@ use utils::{
     lsn::Lsn,
 };
 
-use crate::attachment_service::{AttachmentService, NodeRegisterRequest};
 use crate::local_env::PageServerConf;
 use crate::{background_process, local_env::LocalEnv};
 
@@ -162,8 +161,8 @@ impl PageServerNode {
             .expect("non-Unicode path")
     }
 
-    pub async fn start(&self, config_overrides: &[&str], register: bool) -> anyhow::Result<()> {
-        self.start_node(config_overrides, false, register).await
+    pub async fn start(&self, config_overrides: &[&str]) -> anyhow::Result<Child> {
+        self.start_node(config_overrides, false).await
     }
 
     fn pageserver_init(&self, config_overrides: &[&str]) -> anyhow::Result<()> {
@@ -208,8 +207,7 @@ impl PageServerNode {
         &self,
         config_overrides: &[&str],
         update_config: bool,
-        register: bool,
-    ) -> anyhow::Result<()> {
+    ) -> anyhow::Result<Child> {
         // TODO: using a thread here because start_process() is not async but we need to call check_status()
         let datadir = self.repo_path();
         print!(
@@ -246,26 +244,7 @@ impl PageServerNode {
                 }
             },
         )
-        .await?;
-
-        if register {
-            let attachment_service = AttachmentService::from_env(&self.env);
-            let (pg_host, pg_port) =
-                parse_host_port(&self.conf.listen_pg_addr).expect("Unable to parse listen_pg_addr");
-            let (http_host, http_port) = parse_host_port(&self.conf.listen_http_addr)
-                .expect("Unable to parse listen_http_addr");
-            attachment_service
-                .node_register(NodeRegisterRequest {
-                    node_id: self.conf.id,
-                    listen_pg_addr: pg_host.to_string(),
-                    listen_pg_port: pg_port.unwrap_or(5432),
-                    listen_http_addr: http_host.to_string(),
-                    listen_http_port: http_port.unwrap_or(80),
-                })
-                .await?;
-        }
-
-        Ok(())
+        .await
     }
 
     fn pageserver_basic_args<'a>(
@@ -395,11 +374,6 @@ impl PageServerNode {
                 .transpose()
                 .context("Failed to parse 'gc_feedback' as bool")?,
             heatmap_period: settings.remove("heatmap_period").map(|x| x.to_string()),
-            lazy_slru_download: settings
-                .remove("lazy_slru_download")
-                .map(|x| x.parse::<bool>())
-                .transpose()
-                .context("Failed to parse 'lazy_slru_download' as bool")?,
         };
         if !settings.is_empty() {
             bail!("Unrecognized tenant settings: {settings:?}")
@@ -500,11 +474,6 @@ impl PageServerNode {
                     .transpose()
                     .context("Failed to parse 'gc_feedback' as bool")?,
                 heatmap_period: settings.remove("heatmap_period").map(|x| x.to_string()),
-                lazy_slru_download: settings
-                    .remove("lazy_slru_download")
-                    .map(|x| x.parse::<bool>())
-                    .transpose()
-                    .context("Failed to parse 'lazy_slru_download' as bool")?,
             }
         };
 

control_plane/src/safekeeper.rs

@@ -7,6 +7,7 @@
 //! ```
 use std::io::Write;
 use std::path::PathBuf;
+use std::process::Child;
 use std::{io, result};
 
 use anyhow::Context;
@@ -103,7 +104,7 @@ impl SafekeeperNode {
             .expect("non-Unicode path")
     }
 
-    pub async fn start(&self, extra_opts: Vec<String>) -> anyhow::Result<()> {
+    pub async fn start(&self, extra_opts: Vec<String>) -> anyhow::Result<Child> {
         print!(
             "Starting safekeeper at '{}' in '{}'",
             self.pg_connection_config.raw_address(),

diesel.toml

@@ -1,9 +0,0 @@
-# For documentation on how to configure this file,
-# see https://diesel.rs/guides/configuring-diesel-cli
-
-[print_schema]
-file = "control_plane/attachment_service/src/schema.rs"
-custom_type_derives = ["diesel::query_builder::QueryId"]
-
-[migrations_directory]
-dir = "control_plane/attachment_service/migrations"

docs/docker.md

@@ -21,7 +21,7 @@ We build all images after a successful `release` tests run and push automaticall
 
 ## Docker Compose example
 
-You can see a [docker compose](https://docs.docker.com/compose/) example to create a neon cluster in [/docker-compose/docker-compose.yml](/docker-compose/docker-compose.yml). It creates the following containers.
+You can see a [docker compose](https://docs.docker.com/compose/) example to create a neon cluster in [/docker-compose/docker-compose.yml](/docker-compose/docker-compose.yml). It creates the following conatainers.
 
 - pageserver x 1
 - safekeeper x 3
@@ -38,7 +38,7 @@ You can specify version of neon cluster using following environment values.
 - TAG: the tag version of [docker image](https://registry.hub.docker.com/r/neondatabase/neon/tags) (default is latest), which is tagged in [CI test](/.github/workflows/build_and_test.yml)
 ```
 $ cd docker-compose/
-$ docker-compose down   # remove the containers if exists
+$ docker-compose down   # remove the conainers if exists
 $ PG_VERSION=15 TAG=2937 docker-compose up --build -d  # You can specify the postgres and image version
 Creating network "dockercompose_default" with the default driver
 Creating docker-compose_storage_broker_1       ... done

docs/pageserver-storage.md

@@ -64,7 +64,7 @@ Storage.
 
 The LayerMap tracks what layers exist in a timeline.
 
-Currently, the layer map is just a resizable array (Vec). On a GetPage@LSN or
+Currently, the layer map is just a resizeable array (Vec). On a GetPage@LSN or
 other read request, the layer map scans through the array to find the right layer
 that contains the data for the requested page. The read-code in LayeredTimeline
 is aware of the ancestor, and returns data from the ancestor timeline if it's

docs/pageserver-thread-mgmt.md

@@ -22,7 +22,7 @@ timeline to shutdown. It will also wait for them to finish.
 
 A task registered in the task registry can check if it has been
 requested to shut down, by calling `is_shutdown_requested()`. There's
-also a `shutdown_watcher()` Future that can be used with `tokio::select!`
+also a `shudown_watcher()` Future that can be used with `tokio::select!`
 or similar, to wake up on shutdown.
 
 

docs/pageserver-walredo.md

@@ -74,4 +74,4 @@ somewhat wasteful, but because most WAL records only affect one page,
 the overhead is acceptable.
 
 The WAL redo always happens for one particular page. If the WAL record
-contains changes to other pages, they are ignored.
+coantains changes to other pages, they are ignored.

docs/rfcs/017-console-split.md

@@ -1,420 +0,0 @@
-# Splitting cloud console
-
-Created on 17.06.2022
-
-## Summary
-
-Currently we have `cloud` repository that contains code implementing public API for our clients as well as code for managing storage and internal infrastructure services. We can split everything user-related from everything storage-related to make it easier to test and maintain.
-
-This RFC proposes to introduce a new control-plane service with HTTP API. The overall architecture will look like this:
-
-```markup
-.                    x
-       external area x internal area
-       (our clients) x (our services)
-                     x
-                     x                                                      ┌───────────────────────┐
-                     x ┌───────────────┐   >    ┌─────────────────────┐     │      Storage (EC2)    │
-                     x │  console db   │   >    │  control-plane db   │     │                       │
-                     x └───────────────┘   >    └─────────────────────┘     │ - safekeepers         │
-                     x         ▲           >               ▲                │ - pageservers         │
-                     x         │           >               │                │                       │
-┌──────────────────┐ x ┌───────┴───────┐   >               │                │     Dependencies      │
-│    browser UI    ├──►│               │   >    ┌──────────┴──────────┐     │                       │
-└──────────────────┘ x │               │   >    │                     │     │ - etcd                │
-                     x │    console    ├───────►│    control-plane    ├────►│ - S3                  │
-┌──────────────────┐ x │               │   >    │  (deployed in k8s)  │     │ - more?               │
-│public API clients├──►│               │   >    │                     │     │                       │
-└──────────────────┘ x └───────┬───────┘   >    └──────────┬──────────┘     └───────────────────────┘
-                     x         │           >          ▲    │                            ▲
-                     x         │           >          │    │                            │
-                     x ┌───────┴───────┐   >          │    │                ┌───────────┴───────────┐
-                     x │ dependencies  │   >          │    │                │                       │
-                     x │- analytics    │   >          │    └───────────────►│       computes        │
-                     x │- auth         │   >          │                     │   (deployed in k8s)   │
-                     x │- billing      │   >          │                     │                       │
-                     x └───────────────┘   >          │                     └───────────────────────┘
-                     x                     >          │                                 ▲
-                     x                     >    ┌─────┴───────────────┐                 │
-┌──────────────────┐ x                     >    │                     │                 │
-│                  │ x                     >    │        proxy        ├─────────────────┘
-│     postgres     ├───────────────────────────►│  (deployed in k8s)  │
-│      users       │ x                     >    │                     │
-│                  │ x                     >    └─────────────────────┘
-└──────────────────┘ x                     >
-                                           >
-                                           >
-                             closed-source > open-source
-                                           >
-                                           >
-```
-
-Notes:
-
-- diagram is simplified in the less-important places
-- directed arrows are strict and mean that connections in the reverse direction are forbidden
-
-This split is quite complex and this RFC proposes several smaller steps to achieve the larger goal: 
-
-1. Start by refactoring the console code, the goal is to have console and control-plane code in the different directories without dependencies on each other.
-2. Do similar refactoring for tables in the console database, remove queries selecting data from both console and control-plane; move control-plane tables to a separate database.
-3. Implement control-plane HTTP API serving on a separate TCP port; make all console→control-plane calls to go through that HTTP API.
-4. Move control-plane source code to the neon repo; start control-plane as a separate service.
-
-## Motivation
-
-These are the two most important problems we want to solve:
-
-- Publish open-source implementation of all our cloud/storage features
-- Make a unified control-plane that is used in all cloud (serverless) and local (tests) setups
-
-Right now we have some closed-source code in the cloud repo. That code contains implementation for running Neon computes in k8s and without that code it’s impossible to automatically scale PostgreSQL computes. That means that we don’t have an open-source serverless PostgreSQL at the moment.
-
-After splitting and open-sourcing control-plane service we will have source code and Docker images for all storage services. That control-plane service should have HTTP API for creating and managing tenants (including all our storage features), while proxy will listen for incoming connections and create computes on-demand.
-
-Improving our test suite is an important task, but requires a lot of prerequisites and may require a separate RFC. Possible implementation of that is described in the section [Next steps](#next-steps).
-
-Another piece of motivation can be a better involvement of storage development team into a control-plane. By splitting control-plane from the console, it can be more convenient to test and develop control-plane with paying less attention to “business” features, such as user management, billing and analytics.
-
-For example, console currently requires authentication providers such as GitHub OAuth to work at all, as well as nodejs to be able to build it locally. It will be more convenient to build and run it locally without these requirements.
-
-## Proposed implementation
-
-### Current state of things
-
-Let’s start with defining the current state of things at the moment of this proposal. We have three repositories containing source code:
-
-- open-source `postgres` — our fork of postgres
-- open-source `neon` — our main repository for storage source code
-- closed-source `cloud` — mostly console backend and UI frontend
-
-This proposal aims not to change anything at the existing code in `neon` and `postgres` repositories, but to create control-plane service and move it’s source code from `cloud` to the `neon` repository. That means that we need to split code in `cloud` repo only, and will consider only this repository for exploring its source code.
-
-Let’s look at the miscellaneous things in the `cloud` repo which are NOT part of the console application, i.e. NOT the Go source code that is compiled to the `./console` binary. There we have:
-
-- command-line tools, such as cloudbench, neonadmin
-- markdown documentation
-- cloud operations scripts (helm, terraform, ansible)
-- configs and other things
-- e2e python tests
-- incidents playbooks
-- UI frontend
-- Make build scripts, code generation scripts
-- database migrations
-- swagger definitions
-
-And also let’s take a look at what we have in the console source code, which is the service we’d like to split:
-
-- API Servers
-    - Public API v2
-    - Management API v2
-    - Public API v1
-    - Admin API v1 (same port as Public API v1)
-    - Management API v1
-- Workers
-    - Monitor Compute Activity
-    - Watch Failed Operations
-    - Availability Checker
-    - Business Metrics Collector
-- Internal Services
-    - Auth Middleware, UserIsAdmin, Cookies
-    - Cable Websocket Server
-    - Admin Services
-        - Global Settings, Operations, Pageservers, Platforms, Projects, Safekeepers, Users
-    - Authenticate Proxy
-    - API Keys
-    - App Controller, serving UI HTML
-    - Auth Controller
-    - Branches
-    - Projects
-    - Psql Connect + Passwordless login
-    - Users
-    - Cloud Metrics
-    - User Metrics
-    - Invites
-    - Pageserver/Safekeeper management
-    - Operations, k8s/docker/common logic
-    - Platforms, Regions
-    - Project State
-    - Projects Roles, SCRAM
-    - Global Settings
-- Other things
-    - segment analytics integration
-    - sentry integration
-    - other common utilities packages
-
-### Drawing the splitting line
-
-The most challenging and the most important thing is to define the line that will split new control-plane service from the existing cloud service. If we don’t get it right, then we can end up with having a lot more issues without many benefits.
-
-We propose to define that line as follows:
-
-- everything user-related stays in the console service
-- everything storage-related should be in the control-plane service
-- something that falls in between should be decided where to go, but most likely should stay in the console service
-- some similar parts should be in both services, such as admin/management/db_migrations
-
-We call user-related all requests that can be connected to some user. The general idea is don’t have any user_id in the control-plane service and operate exclusively on tenant_id+timeline_id, the same way as existing storage services work now (compute, safekeeper, pageserver).
-
-Storage-related things can be defined as doing any of the following:
-
-- using k8s API
-- doing requests to any of the storage services (proxy, compute, safekeeper, pageserver, etc..)
-- tracking current status of tenants/timelines, managing lifetime of computes
-
-Based on that idea, we can say that new control-plane service should have the following components:
-
-- single HTTP API for everything
-    - Create and manage tenants and timelines
-    - Manage global settings and storage configuration (regions, platforms, safekeepers, pageservers)
-    - Admin API for storage health inspection and debugging
-- Workers
-    - Monitor Compute Activity
-    - Watch Failed Operations
-    - Availability Checker
-- Internal Services
-    - Admin Services
-        - Global Settings, Operations, Pageservers, Platforms, Tenants, Safekeepers
-    - Authenticate Proxy
-    - Branches
-    - Psql Connect
-    - Cloud Metrics
-    - Pageserver/Safekeeper management
-    - Operations, k8s/docker/common logic
-    - Platforms, Regions
-    - Tenant State
-    - Compute Roles, SCRAM
-    - Global Settings
-
----
-
-And other components should probably stay in the console service:
-
-- API Servers (no changes here)
-    - Public API v2
-    - Management API v2
-    - Public API v1
-    - Admin API v1 (same port as Public API v1)
-    - Management API v1
-- Workers
-    - Business Metrics Collector
-- Internal Services
-    - Auth Middleware, UserIsAdmin, Cookies
-    - Cable Websocket Server
-    - Admin Services
-        - Users admin stays the same
-        - Other admin services can redirect requests to the control-plane
-    - API Keys
-    - App Controller, serving UI HTML
-    - Auth Controller
-    - Projects
-    - User Metrics
-    - Invites
-    - Users
-    - Passwordless login
-- Other things
-    - segment analytics integration
-    - sentry integration
-    - other common utilities packages
-
-There are also miscellaneous things that are useful for all kinds of services. So we can say that these things can be in both services:
-
-- markdown documentation
-- e2e python tests
-- make build scripts, code generation scripts
-- database migrations
-- swagger definitions
-
-The single entrypoint to the storage should be control-plane API. After we define that API, we can have code-generated implementation for the client and for the server. The general idea is to move code implementing storage components from the console to the API implementation inside the new control-plane service.
-
-After the code is moved to the new service, we can fill the created void by making API calls to the new service:
-
-- authorization of the client
-- mapping user_id + project_id to the tenant_id
-- calling the control-plane API
-
-### control-plane API
-
-Currently we have the following projects API in the console:
-
-```
-GET /projects/{project_id}
-PATCH /projects/{project_id}
-POST /projects/{project_id}/branches
-GET /projects/{project_id}/databases
-POST /projects/{project_id}/databases
-GET /projects/{project_id}/databases/{database_id}
-PUT /projects/{project_id}/databases/{database_id}
-DELETE /projects/{project_id}/databases/{database_id}
-POST /projects/{project_id}/delete
-GET /projects/{project_id}/issue_token
-GET /projects/{project_id}/operations
-GET /projects/{project_id}/operations/{operation_id}
-POST /projects/{project_id}/query
-GET /projects/{project_id}/roles
-POST /projects/{project_id}/roles
-GET /projects/{project_id}/roles/{role_name}
-DELETE /projects/{project_id}/roles/{role_name}
-POST /projects/{project_id}/roles/{role_name}/reset_password
-POST /projects/{project_id}/start
-POST /projects/{project_id}/stop
-POST /psql_session/{psql_session_id}
-```
-
-It looks fine and we probably already have clients relying on it. So we should not change it, at least for now. But most of these endpoints (if not all) are related to storage, and it can suggest us what control-plane API should look like:
-
-```
-GET /tenants/{tenant_id}
-PATCH /tenants/{tenant_id}
-POST /tenants/{tenant_id}/branches
-GET /tenants/{tenant_id}/databases
-POST /tenants/{tenant_id}/databases
-GET /tenants/{tenant_id}/databases/{database_id}
-PUT /tenants/{tenant_id}/databases/{database_id}
-DELETE /tenants/{tenant_id}/databases/{database_id}
-POST /tenants/{tenant_id}/delete
-GET /tenants/{tenant_id}/issue_token
-GET /tenants/{tenant_id}/operations
-GET /tenants/{tenant_id}/operations/{operation_id}
-POST /tenants/{tenant_id}/query
-GET /tenants/{tenant_id}/roles
-POST /tenants/{tenant_id}/roles
-GET /tenants/{tenant_id}/roles/{role_name}
-DELETE /tenants/{tenant_id}/roles/{role_name}
-POST /tenants/{tenant_id}/roles/{role_name}/reset_password
-POST /tenants/{tenant_id}/start
-POST /tenants/{tenant_id}/stop
-POST /psql_session/{psql_session_id}
-```
-
-One of the options here is to use gRPC instead of the HTTP, which has some useful features, but there are some strong points towards using plain HTTP:
-
-- HTTP API is easier to use for the clients
-- we already have HTTP API in pageserver/safekeeper/console
-- we probably want control-plane API to be similar to the console API, available in the cloud
-
-### Getting updates from the storage
-
-There can be some valid cases, when we would like to know what is changed in the storage. For example, console might want to know when user has queried and started compute and when compute was scaled to zero after that, to know how much user should pay for the service. Another example is to get info about reaching the disk space limits. Yet another example is to do analytics, such as how many users had at least one active project in a month.
-
-All of the above cases can happen without using the console, just by accessing compute through the proxy.
-
-To solve this, we can have a log of events occurring in the storage (event logs). That is very similar to operations table we have right now, the only difference is that events are immutable and we cannot change them after saving to the database. For example, we might want to have events for the following activities:
-
-- We finished processing some HTTP API query, such as resetting the password
-- We changed some state, such as started or stopped a compute
-- Operation is created
-- Operation is started for the first time
-- Operation is failed for the first time
-- Operation is finished
-
-Once we save these events to the database, we can create HTTP API to subscribe to these events. That API can look like this:
-
-```
-GET /events/<cursor>
-
-{
-  "events": [...],
-  "next_cursor": 123
-}
-```
-
-It should be possible to replay event logs from some point of time, to get a state of almost anything from the storage services. That means that if we maintain some state in the control-plane database and we have a reason to have the same state in the console database, it is possible by polling events from the control-plane API and changing the state in the console database according to the events.
-
-### Next steps
-
-After implementing control-plane HTTP API and starting control-plane as a separate service, we might want to think of exploiting benefits of the new architecture, such as reorganizing test infrastructure. Possible options are listed in the  [Next steps](#next-steps-1).
-
-## Non Goals
-
-RFC doesn’t cover the actual cloud deployment scripts and schemas, such as terraform, ansible, k8s yaml’s and so on.
-
-## Impacted components
-
-Mostly console, but can also affect some storage service.
-
-## Scalability
-
-We should support starting several instances of the new control-plane service at the same time.
-
-At the same time, it should be possible to use only single instance of control-plane, which can be useful for local tests.
-
-## Security implications
-
-New control-plane service is an internal service, so no external requests can reach it. But at the same time, it contains API to do absolutely anything with any of the tenants. That means that bad internal actor can potentially read and write all of the tenants. To make this safer, we can have one of these:
-
-- Simple option is to protect all requests with a single private key, so that no one can make requests without having that one key.
-- Another option is to have a separate token for every tenant and store these tokens in another secure place. This way it’s harder to access all tenants at once, because they have the different tokens.
-
-## Alternative implementation
-
-There was an idea to create a k8s operator for managing storage services and computes, but author of this RFC is not really familiar with it.
-
-Regarding less alternative ideas, there are another options for the name of the new control-plane service:
-
-- storage-ctl
-- cloud
-- cloud-ctl
-
-## Pros/cons of proposed approaches (TODO)
-
-Pros:
-
-- All storage features are completely open-source
-- Better tests coverage, less difference between cloud and local setups
-- Easier to develop storage and cloud features, because there is no need to setup console for that
-- Easier to deploy storage-only services to the any cloud
-
-Cons:
-
-- All storage features are completely open-source
-- Distributed services mean more code to connect different services and potential network issues
-- Console needs to have a dependency on storage API, there can be complications with developing new feature in a branch
-- More code to JOIN data from different services (console and control-plane)
-
-## Definition of Done
-
-We have a new control-plane service running in the k8s. Source code for that control-plane service is located in the open-source neon repo.
-
-## Next steps
-
-After we’ve reached DoD, we can make further improvements.
-
-First thing that can benefit from the split is local testing. The same control-plane service can implement starting computes as a local processes instead of k8s deployments. If it will also support starting pageservers/safekeepers/proxy for the local setup, then it can completely replace `./neon_local` binary, which is currently used for testing. The local testing environment can look like this:
-
-```
-┌─────────────────────┐     ┌───────────────────────┐
-│                     │     │      Storage (local)  │
-│  control-plane db   │     │                       │
-│   (local process)   │     │ - safekeepers         │
-│                     │     │ - pageservers         │
-└──────────▲──────────┘     │                       │
-           │                │     Dependencies      │
-┌──────────┴──────────┐     │                       │
-│                     │     │ - etcd                │
-│    control-plane    ├────►│ - S3                  │
-│   (local process)   │     │ - more?               │
-│                     │     │                       │
-└──────────┬──────────┘     └───────────────────────┘
-       ▲   │                            ▲
-       │   │                            │
-       │   │                ┌───────────┴───────────┐
-       │   │                │                       │
-       │   └───────────────►│       computes        │
-       │                    │   (local processes)   │
-       │                    │                       │
-┌──────┴──────────────┐     └───────────────────────┘
-│                     │                 ▲
-│        proxy        │                 │
-│   (local process)   ├─────────────────┘
-│                     │
-└─────────────────────┘
-```
-
-The key thing here is that control-plane local service have the same API and almost the same implementation as the one deployed in the k8s. This allows to run the same e2e tests against both cloud and local setups.
-
-For the python test_runner tests everything can stay mostly the same. To do that, we just need to replace `./neon_local` cli commands with API calls to the control-plane.
-
-The benefit here will be in having fast local tests that are really close to our cloud setup. Bugs in k8s queries are still cannot be found when running computes as a local processes, but it should be really easy to start k8s locally (for example in k3s) and run the same tests with control-plane connected to the local k8s.
-
-Talking about console and UI tests, after the split there should be a way to test these without spinning up all the storage locally. New control-plane service has a well-defined API, allowing us to mock it. This way we can create UI tests to verify the right calls are issued after specific UI interactions and verify that we render correct messages when API returns errors.

docs/rfcs/018-storage-messaging-2.md

@@ -78,7 +78,7 @@ with grpc streams and tokio mpsc channels. The implementation description is at
 
 It is just 500 lines of code and core functionality is complete. 1-1 pub sub
 gives about 120k received messages per second; having multiple subscribers in
-different connections quickly scales to 1 million received messages per second.
+different connecitons quickly scales to 1 million received messages per second.
 I had concerns about many concurrent streams in singe connection, but 2^20
 subscribers still work (though eat memory, with 10 publishers 20GB are consumed;
 in this implementation each publisher holds full copy of all subscribers). There
@@ -95,12 +95,12 @@ other members, with best-effort this is simple.
 ### Security implications
 
 Communication happens in a private network that is not exposed to users;
-additionally we can add auth to the broker.
+additionaly we can add auth to the broker.
 
 ## Alternative: get existing pub-sub
 
 We could take some existing pub sub solution, e.g. RabbitMQ, Redis. But in this
-case IMV simplicity of our own outweighs external dependency costs (RabbitMQ is
+case IMV simplicity of our own outweights external dependency costs (RabbitMQ is
 much more complicated and needs VM; Redis Rust client maintenance is not
 ideal...). Also note that projects like CockroachDB and TiDB are based on gRPC
 as well.

docs/rfcs/019-tenant-timeline-lifecycles.md

@@ -74,7 +74,7 @@ TenantMaintenanceGuard: Like ActiveTenantGuard, but can be held even when the
 tenant is not in Active state. Used for operations like attach/detach. Perhaps
 allow only one such guard on a Tenant at a time.
 
-Similarly for Timelines. We don't currently have a "state" on Timeline, but I think
+Similarly for Timelines. We don't currentl have a "state" on Timeline, but I think
 we need at least two states: Active and Stopping. The Stopping state is used at
 deletion, to prevent new TimelineActiveGuards from appearing, while you wait for
 existing TimelineActiveGuards to die out.
@@ -85,7 +85,7 @@ have a TenantActiveGuard, and the tenant's state changes from Active to
 Stopping, the is_shutdown_requested() function should return true, and
 shutdown_watcher() future should return.
 
-This signaling doesn't necessarily need to cover all cases. For example, if you
+This signaling doesn't neessarily need to cover all cases. For example, if you
 have a block of code in spawn_blocking(), it might be acceptable if
 is_shutdown_requested() doesn't return true even though the tenant is in
 Stopping state, as long as the code finishes reasonably fast.

docs/rfcs/020-pageserver-s3-coordination.md

@@ -37,7 +37,7 @@ sequenceDiagram
 ```
 
 At this point it is not possible to restore from index, it contains L2 which
-is no longer available in s3 and doesn't contain L3 added by compaction by the
+is no longer available in s3 and doesnt contain L3 added by compaction by the
 first pageserver. So if any of the pageservers restart initial sync will fail
 (or in on-demand world it will fail a bit later during page request from
 missing layer)
@@ -74,7 +74,7 @@ One possible solution for relocation case is to orchestrate background jobs
 from outside. The oracle who runs migration can turn off background jobs on
 PS1 before migration and then run migration -> enable them on PS2. The problem
 comes if migration fails. In this case in order to resume background jobs
-oracle needs to guarantee that PS2 doesn't run background jobs and if it doesn't
+oracle needs to guarantee that PS2 doesnt run background jobs and if it doesnt
 respond then PS1 is stuck unable to run compaction/gc. This cannot be solved
 without human ensuring that no upload from PS2 can happen. In order to be able
 to resolve this automatically CAS is required on S3 side so pageserver can
@@ -128,7 +128,7 @@ During discussion it seems that we converged on the approach consisting of:
   whether we need to apply change to the index state or not.
 - Responsibility for running background jobs is assigned externally. Pageserver
   keeps locally persistent flag for each tenant that indicates whether this
-  pageserver is considered as primary one or not. TODO what happens if we
+  pageserver is considered as primary one or not. TODO what happends if we
   crash and cannot start for some extended period of time? Control plane can
   assign ownership to some other pageserver. Pageserver needs some way to check
   if its still the blessed one. Maybe by explicit request to control plane on
@@ -138,7 +138,7 @@ Requirement for deterministic layer generation was considered overly strict
 because of two reasons:
 
 - It can limit possible optimizations e g when pageserver wants to reshuffle
-  some data locally and doesn't want to coordinate this
+  some data locally and doesnt want to coordinate this
 - The deterministic algorithm itself can change so during deployments for some
   time there will be two different version running at the same time which can
   cause non determinism
@@ -164,7 +164,7 @@ sequenceDiagram
     CP->>PS1: Yes
     deactivate CP
     PS1->>S3: Fetch PS1 index.
-    note over PS1: Continue operations, start background jobs
+    note over PS1: Continue operations, start backround jobs
     note over PS1,PS2: PS1 starts up and still and is not a leader anymore
     PS1->>CP: Am I still the leader for Tenant X?
     CP->>PS1: No
@@ -203,7 +203,7 @@ sequenceDiagram
 ### Eviction
 
 When two pageservers operate on a tenant for extended period of time follower
-doesn't perform write operations in s3. When layer is evicted follower relies
+doesnt perform write operations in s3. When layer is evicted follower relies
 on updates from primary to get info about layers it needs to cover range for
 evicted layer.
 

docs/rfcs/022-pageserver-delete-from-s3.md

@@ -4,7 +4,7 @@ Created on 08.03.23
 
 ## Motivation
 
-Currently we don't delete pageserver part of the data from s3 when project is deleted. (The same is true for safekeepers, but this outside of the scope of this RFC).
+Currently we dont delete pageserver part of the data from s3 when project is deleted. (The same is true for safekeepers, but this outside of the scope of this RFC).
 
 This RFC aims to spin a discussion to come to a robust deletion solution that wont put us in into a corner for features like postponed deletion (when we keep data for user to be able to restore a project if it was deleted by accident)
 
@@ -75,9 +75,9 @@ Remote one is needed for cases when pageserver is lost during deletion so other
 
 Why local mark file is needed?
 
-If we don't have one, we have two choices, delete local data before deleting the remote part or do that after.
+If we dont have one, we have two choices, delete local data before deleting the remote part or do that after.
 
-If we delete local data before remote then during restart pageserver wont pick up remote tenant at all because nothing is available locally (pageserver looks for remote counterparts of locally available tenants).
+If we delete local data before remote then during restart pageserver wont pick up remote tenant at all because nothing is available locally (pageserver looks for remote conuterparts of locally available tenants).
 
 If we delete local data after remote then at the end of the sequence when remote mark file is deleted if pageserver restart happens then the state is the same to situation when pageserver just missing data on remote without knowing the fact that this data is intended to be deleted. In this case the current behavior is upload everything local-only to remote.
 
@@ -145,7 +145,7 @@ sequenceDiagram
         CP->>PS: Retry delete tenant
         PS->>CP: Not modified
     else Mark is missing
-        note over PS: Continue to operate the tenant as if deletion didn't happen
+        note over PS: Continue to operate the tenant as if deletion didnt happen
 
         note over CP: Eventually console should <br> retry delete request
 
@@ -168,7 +168,7 @@ sequenceDiagram
     PS->>CP: True
 ```
 
-Similar sequence applies when both local and remote marks were persisted but Control Plane still didn't receive a response.
+Similar sequence applies when both local and remote marks were persisted but Control Plane still didnt receive a response.
 
 If pageserver crashes after both mark files were deleted then it will reply to control plane status poll request with 404 which should be treated by control plane as success.
 
@@ -187,7 +187,7 @@ If pageseserver is lost then the deleted tenant should be attached to different
 
 ##### Restrictions for tenant that is in progress of being deleted
 
-I propose to add another state to tenant/timeline - PendingDelete. This state shouldn't allow executing any operations aside from polling the deletion status.
+I propose to add another state to tenant/timeline - PendingDelete. This state shouldnt allow executing any operations aside from polling the deletion status.
 
 #### Summary
 
@@ -237,7 +237,7 @@ New branch gets created
 PS1 starts up (is it possible or we just recycle it?)
 PS1 is unaware of the new branch. It can either fall back to s3 ls, or ask control plane.
 
-So here comes the dependency of storage on control plane. During restart storage needs to know which timelines are valid for operation. If there is nothing on s3 that can answer that question storage needs to ask control plane.
+So here comes the dependency of storage on control plane. During restart storage needs to know which timelines are valid for operation. If there is nothing on s3 that can answer that question storage neeeds to ask control plane.
 
 ### Summary
 
@@ -250,7 +250,7 @@ Cons:
 
 Pros:
 
-- Easier to reason about if you don't have to account for pageserver restarts
+- Easier to reason about if you dont have to account for pageserver restarts
 
 ### Extra notes
 
@@ -262,7 +262,7 @@ Delayed deletion can be done with both approaches. As discussed with Anna (@step
 
 After discussion in comments I see that we settled on two options (though a bit different from ones described in rfc). First one is the same - pageserver owns as much as possible. The second option is that pageserver owns markers thing, but actual deletion happens in control plane by repeatedly calling ls + delete.
 
-To my mind the only benefit of the latter approach is possible code reuse between safekeepers and pageservers. Otherwise poking around integrating s3 library into control plane, configuring shared knowledge about paths in s3 - are the downsides. Another downside of relying on control plane is the testing process. Control plane resides in different repository so it is quite hard to test pageserver related changes there. e2e test suite there doesn't support shutting down pageservers, which are separate docker containers there instead of just processes.
+To my mind the only benefit of the latter approach is possible code reuse between safekeepers and pageservers. Otherwise poking around integrating s3 library into control plane, configuring shared knowledge abouth paths in s3 - are the downsides. Another downside of relying on control plane is the testing process. Control plane resides in different repository so it is quite hard to test pageserver related changes there. e2e test suite there doesnt support shutting down pageservers, which are separate docker containers there instead of just processes.
 
 With pageserver owning everything we still give the retry logic to control plane but its easier to duplicate if needed compared to sharing inner s3 workings. We will have needed tests for retry logic in neon repo.
 

docs/rfcs/023-the-state-of-pageserver-tenant-relocation.md

@@ -75,7 +75,7 @@ sequenceDiagram
 ```
 
 At this point it is not possible to restore the state from index, it contains L2 which
-is no longer available in s3 and doesn't contain L3 added by compaction by the
+is no longer available in s3 and doesnt contain L3 added by compaction by the
 first pageserver. So if any of the pageservers restart, initial sync will fail
 (or in on-demand world it will fail a bit later during page request from
 missing layer)
@@ -171,7 +171,7 @@ sequenceDiagram
 
 Another problem is a possibility of concurrent branch creation calls.
 
-I e during migration create_branch can be called on old pageserver and newly created branch wont be seen on new pageserver. Prior art includes prototyping an approach of trying to mirror such branches, but currently it lost its importance, because now attach is fast because we don't need to download all data, and additionally to the best of my knowledge of control plane internals (cc @ololobus to confirm) operations on one project are executed sequentially, so it is not possible to have such case. So branch create operation will be executed only when relocation is completed. As a safety measure we can forbid branch creation for tenants that are in readonly remote state.
+I e during migration create_branch can be called on old pageserver and newly created branch wont be seen on new pageserver. Prior art includes prototyping an approach of trying to mirror such branches, but currently it lost its importance, because now attach is fast because we dont need to download all data, and additionally to the best of my knowledge of control plane internals (cc @ololobus to confirm) operations on one project are executed sequentially, so it is not possible to have such case. So branch create operation will be executed only when relocation is completed. As a safety measure we can forbid branch creation for tenants that are in readonly remote state.
 
 ## Simplistic approach
 

docs/rfcs/024-extension-loading.md

@@ -55,7 +55,7 @@ When PostgreSQL requests a file, `compute_ctl` downloads it.
 PostgreSQL requests files in the following cases:
 - When loading a preload library set in `local_preload_libraries`
 - When explicitly loading a library with `LOAD`
-- When creating extension with `CREATE EXTENSION` (download sql scripts, (optional) extension data files and (optional) library files)))
+- Wnen creating extension with `CREATE EXTENSION` (download sql scripts, (optional) extension data files and (optional) library files)))
 
 
 #### Summary

docs/rfcs/025-generation-numbers.md

@@ -26,7 +26,7 @@ plane guarantee prevents robust response to failures, as if a pageserver is unre
 we may not detach from it. The mechanism in this RFC fixes this, by making it safe to
 attach to a new, different pageserver even if an unresponsive pageserver may be running.
 
-Further lack of safety during split-brain conditions blocks two important features where occasional
+Futher, lack of safety during split-brain conditions blocks two important features where occasional
 split-brain conditions are part of the design assumptions:
 
 - seamless tenant migration ([RFC PR](https://github.com/neondatabase/neon/pull/5029))
@@ -490,11 +490,11 @@ The above makes it safe for control plane to change the assignment of
 tenant to pageserver in control plane while a timeline creation is ongoing.
 The reason is that the creation request against the new assigned pageserver
 uses a new generation number. However, care must be taken by control plane
-to ensure that a "timeline creation successful" response from some pageserver
+to ensure that a "timeline creation successul" response from some pageserver
 is checked for the pageserver's generation for that timeline's tenant still being the latest.
 If it is not the latest, the response does not constitute a successful timeline creation.
 It is acceptable to discard such responses, the scrubber will clean up the S3 state.
-It is better to issue a timeline deletion request to the stale attachment.
+It is better to issue a timelien deletion request to the stale attachment.
 
 #### Timeline Deletion
 
@@ -633,7 +633,7 @@ As outlined in the Part 1 on correctness, it is critical that deletions are only
 executed once the key is not referenced anywhere in S3.
 This property is obviously upheld by the scheme above.
 
-#### We Accept Object Leakage In Acceptable Circumstances
+#### We Accept Object Leakage In Acceptable Circumcstances
 
 If we crash in the flow above between (2) and (3), we lose track of unreferenced object.
 Further, enqueuing a single to the persistent queue may not be durable immediately to amortize cost of flush to disk.

docs/rfcs/026-pageserver-s3-mvcc.md

@@ -162,7 +162,7 @@ struct Tenant {
   ...
 
   txns: HashMap<TxnId, Transaction>,
-  // the most recently started txn's id; only most recently started can win
+  // the most recently started txn's id; only most recently sarted can win
   next_winner_txn: Option<TxnId>,
 }
 struct Transaction {
@@ -186,7 +186,7 @@ A transaction T in state Committed has subsequent transactions that may or may n
 
 So, for garbage collection, we need to assess transactions in state Committed and RejectAcknowledged:
 
-- Committed: delete objects on the deadlist.
+- Commited: delete objects on the deadlist.
     - We don’t need a LIST request here, the deadlist is sufficient. So, it’s really cheap.
     - This is **not true MVCC garbage collection**; by deleting the objects on Committed transaction T ’s deadlist, we might delete data referenced by other transactions that were concurrent with T, i.e., they started while T was still open. However, the fact that T is committed means that the other transactions are RejectPending or RejectAcknowledged, so, they don’t matter. Pageservers executing these doomed RejectPending transactions must handle 404 for GETs gracefully, e.g., by trying to commit txn so they observe the rejection they’re destined to get anyways. 404’s for RejectAcknowledged is handled below.
 - RejectAcknowledged: delete all objects created in that txn, and discard deadlists.
@@ -242,15 +242,15 @@ If a pageserver is unresponsive from Control Plane’s / Compute’s perspective
 
 At this point, availability is restored and user pain relieved.
 
-What’s left is to somehow close the doomed transaction of the unresponsive pageserver, so that it becomes RejectAcknowledged, and GC can make progress. Since S3 is cheap, we can afford to wait a really long time here, especially if we put a soft bound on the amount of data a transaction may produce before it must commit. Procedure:
+What’s left is to somehow close the doomed transaction of the unresponsive pageserver, so that it beomes RejectAcknowledged, and GC can make progress. Since S3 is cheap, we can afford to wait a really long time here, especially if we put a soft bound on the amount of data a transaction may produce before it must commit. Procedure:
 
 1. Ensure the unresponsive pageserver is taken out of rotation for new attachments. That probably should happen as part of the routine above.
 2. Make a human operator investigate decide what to do (next morning, NO ONCALL ALERT):
     1. Inspect the instance, investigate logs, understand root cause.
     2. Try to re-establish connectivity between pageserver and Control Plane so that pageserver can retry commits, get rejected, ack rejection ⇒ enable GC.
-    3. Use below procedure to decommission pageserver.
+    3. Use below procedure to decomission pageserver.
 
-### Decommissioning A Pageserver (Dead or Alive-but-Unresponsive)
+### Decomissioning A Pageserver (Dead or Alive-but-Unrespsonive)
 
 The solution, enabled by this proposal:
 
@@ -310,7 +310,7 @@ Issues that we discussed:
     1. In abstract terms, this proposal provides a linearized history for a given S3 prefix.
     2. In concrete terms, this proposal provides a linearized history per tenant.
     3. There can be multiple writers at a given time, but only one of them will win to become part of the linearized history.
-4. ************************************************************************************Alternative ideas mentioned during meetings that should be turned into a written proposal like this one:************************************************************************************
+4. ************************************************************************************Alternative ideas mentioned during meetings that should be turned into a written prospoal like this one:************************************************************************************
     1. @Dmitry Rodionov : having linearized storage of index_part.json in some database that allows serializable transactions / atomic compare-and-swap PUT
     2. @Dmitry Rodionov :
     3. @Stas : something like this scheme, but somehow find a way to equate attachment duration with transaction duration, without losing work if pageserver dies months after attachment.

docs/rfcs/027-crash-consistent-layer-map-through-index-part.md

@@ -54,7 +54,7 @@ If the compaction algorithm doesn't change between the two compaction runs, is d
 *However*:
 1. the file size of the overwritten L1s may not be identical, and
 2. the bit pattern of the overwritten L1s may not be identical, and,
-3. in the future, we may want to make the compaction code non-deterministic, influenced by past access patterns, or otherwise change it, resulting in L1 overwrites with a different set of delta records than before the overwrite
+3. in the future, we may want to make the compaction code non-determinstic, influenced by past access patterns, or otherwise change it, resulting in L1 overwrites with a different set of delta records than before the overwrite
 
 The items above are a problem for the [split-brain protection RFC](https://github.com/neondatabase/neon/pull/4919) because it assumes that layer files in S3 are only ever deleted, but never replaced (overPUTted).
 
@@ -63,7 +63,7 @@ But node B based its world view on the version of node A's `index_part.json` fro
 That earlier `index_part.json`` contained the file size of the pre-overwrite L1.
 If the overwritten L1 has a different file size, node B will refuse to read data from the overwritten L1.
 Effectively, the data in the L1 has become inaccessible to node B.
-If node B already uploaded an index part itself, all subsequent attachments will use node B's index part, and run into the same problem.
+If node B already uploaded an index part itself, all subsequent attachments will use node B's index part, and run into the same probem.
 
 If we ever introduce checksums instead of checking just the file size, then a mismatching bit pattern (2) will cause similar problems.
 
@@ -121,7 +121,7 @@ Multi-object changes that previously created and removed files in timeline dir a
 * atomic `index_part.json` update in S3, as per guarantee that S3 PUT is atomic
 * local timeline dir state:
   * irrelevant for layer map content => irrelevant for atomic updates / crash consistency
-  * if we crash after index part PUT, local layer files will be used, so, no on-demand downloads needed for them
+  * if we crash after index part PUT, local layer files will be used, so, no on-demand downloads neede for them
   * if we crash before index part PUT, local layer files will be deleted
 
 ## Trade-Offs
@@ -140,7 +140,7 @@ Assuming upload queue allows for unlimited queue depth (that's what it does toda
 * wal ingest: currently unbounded
 * L0 => L1 compaction: CPU time proportional to `O(sum(L0 size))` and upload work proportional to `O()`
   * Compaction threshold is 10 L0s and each L0 can be up to 256M in size. Target size for L1 is 128M.
-  * In practice, most L0s are tiny due to 10minute `DEFAULT_CHECKPOINT_TIMEOUT`.
+  * In practive, most L0s are tiny due to 10minute `DEFAULT_CHECKPOINT_TIMEOUT`.
 * image layer generation: CPU time `O(sum(input data))` + upload work `O(sum(new image layer size))`
   * I have no intuition how expensive / long-running it is in reality.
 * gc: `update_gc_info`` work (not substantial, AFAIK)
@@ -158,7 +158,7 @@ Pageserver crashes are very rare ; it would likely be acceptable to re-do the lo
 However, regular pageserver restart happen frequently, e.g., during weekly deploys.
 
 In general, pageserver restart faces the problem of tenants that "take too long" to shut down.
-They are a problem because other tenants that shut down quickly are unavailable while we wait for the slow tenants to shut down.
+They are a problem because other tenants that shut down quickly are unavailble while we wait for the slow tenants to shut down.
 We currently allot 10 seconds for graceful shutdown until we SIGKILL the pageserver process (as per `pageserver.service` unit file).
 A longer budget would expose tenants that are done early to a longer downtime.
 A short budget would risk throwing away more work that'd have to be re-done after restart.
@@ -236,7 +236,7 @@ tenants/$tenant/timelines/$timeline/$key_and_lsn_range
 tenants/$tenant/timelines/$timeline/$layer_file_id-$key_and_lsn_range
 ```
 
-To guarantee uniqueness, the unique number is a sequence number, stored in `index_part.json`.
+To guarantee uniqueness, the unqiue number is a sequence number, stored in `index_part.json`.
 
 This alternative does not solve atomic layer map updates.
 In our crash-during-compaction scenario above, the compaction run after the crash will not overwrite the L1s, but write/PUT new files with new sequence numbers.
@@ -246,11 +246,11 @@ We'd need to write a deduplication pass that checks if perfectly overlapping lay
 However, this alternative is appealing because it systematically prevents overwrites at a lower level than this RFC.
 
 So, this alternative is sufficient for the needs of the split-brain safety RFC (immutable layer files locally and in S3).
-But it doesn't solve the problems with crash-during-compaction outlined earlier in this RFC, and in fact, makes it much more acute.
+But it doesn't solve the problems with crash-during-compaction outlined earlier in this RFC, and in fact, makes it much more accute.
 The proposed design in this RFC addresses both.
 
 So, if this alternative sounds appealing, we should implement the proposal in this RFC first, then implement this alternative on top.
-That way, we avoid a phase where the crash-during-compaction problem is acute.
+That way, we avoid a phase where the crash-during-compaction problem is accute.
 
 ## Related issues
 

docs/rfcs/028-pageserver-migration.md

@@ -596,4 +596,4 @@ pageservers are updated to be aware of it.
 
 As well as simplifying implementation, putting heatmaps in S3 will be useful
 for future analytics purposes -- gathering aggregated statistics on activity
-patterns across many tenants may be done directly from data in S3.
+pattersn across many tenants may be done directly from data in S3.

docs/rfcs/029-pageserver-wal-disaster-recovery.md

@@ -147,7 +147,7 @@ Separating corrupt writes from non-corrupt ones is a hard problem in general,
 and if the application was involved in making the corrupt write, a recovery
 would also involve the application. Therefore, corruption that has made it into
 the WAL is outside of the scope of this feature. However, the WAL replay can be
-issued to right before the point in time where the corruption occurred. Then the
+issued to right before the point in time where the corruption occured. Then the
 data loss is isolated to post-corruption writes only.
 
 ## Impacted components (e.g. pageserver, safekeeper, console, etc)
@@ -161,7 +161,7 @@ limits and billing we apply to existing timelines.
 
 ## Proposed implementation
 
-The first problem to keep in mind is the reproducibility of `initdb`.
+The first problem to keep in mind is the reproducability of `initdb`.
 So an initial step would be to upload `initdb` snapshots to S3.
 
 After that, we'd have the endpoint spawn a background process which

docs/rfcs/030-vectored-timeline-get.md

@@ -69,7 +69,7 @@ However, unlike above, an ideal solution will
   * This means, read each `DiskBtree` page at most once.
 * Facilitate merging of the reads we issue to the OS and eventually NVMe.
 
-Each of these items above represents a significant amount of work.
+Each of these items above represents a signficant amount of work.
 
 ## Performance
 

docs/synthetic-size.md

@@ -21,7 +21,7 @@ implementation where we keep more data than we would need to, do not
 change the synthetic size or incur any costs to the user.
 
 The synthetic size is calculated for the whole project. It is not
-straightforward to attribute size to individual branches. See "What is
+straighforward to attribute size to individual branches. See "What is
 the size of an individual branch?" for discussion on those
 difficulties.
 
@@ -248,7 +248,7 @@ and truncate the WAL.
 
 Synthetic size is calculated for the whole project, and includes all
 branches. There is no such thing as the size of a branch, because it
-is not straightforward to attribute the parts of size to individual
+is not straighforward to attribute the parts of size to individual
 branches.
 
 ## Example: attributing size to branches

libs/metrics/Cargo.toml

@@ -9,10 +9,5 @@ prometheus.workspace = true
 libc.workspace = true
 once_cell.workspace = true
 chrono.workspace = true
-twox-hash.workspace = true
 
 workspace_hack.workspace = true
-
-[dev-dependencies]
-rand = "0.8"
-rand_distr = "0.4.3"

libs/metrics/src/hll.rs

@@ -1,523 +0,0 @@
-//! HyperLogLog is an algorithm for the count-distinct problem,
-//! approximating the number of distinct elements in a multiset.
-//! Calculating the exact cardinality of the distinct elements
-//! of a multiset requires an amount of memory proportional to
-//! the cardinality, which is impractical for very large data sets.
-//! Probabilistic cardinality estimators, such as the HyperLogLog algorithm,
-//! use significantly less memory than this, but can only approximate the cardinality.
-
-use std::{
-    collections::HashMap,
-    hash::{BuildHasher, BuildHasherDefault, Hash, Hasher},
-    sync::{atomic::AtomicU8, Arc, RwLock},
-};
-
-use prometheus::{
-    core::{self, Describer},
-    proto, Opts,
-};
-use twox_hash::xxh3;
-
-/// Create an [`HyperLogLogVec`] and registers to default registry.
-#[macro_export(local_inner_macros)]
-macro_rules! register_hll_vec {
-    ($N:literal, $OPTS:expr, $LABELS_NAMES:expr $(,)?) => {{
-        let hll_vec = $crate::HyperLogLogVec::<$N>::new($OPTS, $LABELS_NAMES).unwrap();
-        $crate::register(Box::new(hll_vec.clone())).map(|_| hll_vec)
-    }};
-
-    ($N:literal, $NAME:expr, $HELP:expr, $LABELS_NAMES:expr $(,)?) => {{
-        $crate::register_hll_vec!($N, $crate::opts!($NAME, $HELP), $LABELS_NAMES)
-    }};
-}
-
-/// Create an [`HyperLogLog`] and registers to default registry.
-#[macro_export(local_inner_macros)]
-macro_rules! register_hll {
-    ($N:literal, $OPTS:expr $(,)?) => {{
-        let hll = $crate::HyperLogLog::<$N>::with_opts($OPTS).unwrap();
-        $crate::register(Box::new(hll.clone())).map(|_| hll)
-    }};
-
-    ($N:literal, $NAME:expr, $HELP:expr $(,)?) => {{
-        $crate::register_hll!($N, $crate::opts!($NAME, $HELP), $LABELS_NAMES)
-    }};
-}
-
-/// HLL is a probabilistic cardinality measure.
-///
-/// How to use this time-series for a metric name `my_metrics_total_hll`:
-///
-/// ```promql
-/// # harmonic mean
-/// 1 / (
-///     sum (
-///         2 ^ -(
-///             # HLL merge operation
-///             max (my_metrics_total_hll{}) by (hll_shard, other_labels...)
-///         )
-///     ) without (hll_shard)
-/// )
-/// * alpha
-/// * shards_count
-/// * shards_count
-/// ```
-///
-/// If you want an estimate over time, you can use the following query:
-///
-/// ```promql
-/// # harmonic mean
-/// 1 / (
-///     sum (
-///         2 ^ -(
-///             # HLL merge operation
-///             max (
-///                 max_over_time(my_metrics_total_hll{}[$__rate_interval])
-///             ) by (hll_shard, other_labels...)
-///         )
-///     ) without (hll_shard)
-/// )
-/// * alpha
-/// * shards_count
-/// * shards_count
-/// ```
-///
-/// In the case of low cardinality, you might want to use the linear counting approximation:
-///
-/// ```promql
-/// # LinearCounting(m, V) = m log (m / V)
-/// shards_count * ln(shards_count /
-///     # calculate V = how many shards contain a 0
-///     count(max (proxy_connecting_endpoints{}) by (hll_shard, protocol) == 0) without (hll_shard)
-/// )
-/// ```
-///
-/// See <https://en.wikipedia.org/wiki/HyperLogLog#Practical_considerations> for estimates on alpha
-#[derive(Clone)]
-pub struct HyperLogLogVec<const N: usize> {
-    core: Arc<HyperLogLogVecCore<N>>,
-}
-
-struct HyperLogLogVecCore<const N: usize> {
-    pub children: RwLock<HashMap<u64, HyperLogLog<N>, BuildHasherDefault<xxh3::Hash64>>>,
-    pub desc: core::Desc,
-    pub opts: Opts,
-}
-
-impl<const N: usize> core::Collector for HyperLogLogVec<N> {
-    fn desc(&self) -> Vec<&core::Desc> {
-        vec![&self.core.desc]
-    }
-
-    fn collect(&self) -> Vec<proto::MetricFamily> {
-        let mut m = proto::MetricFamily::default();
-        m.set_name(self.core.desc.fq_name.clone());
-        m.set_help(self.core.desc.help.clone());
-        m.set_field_type(proto::MetricType::GAUGE);
-
-        let mut metrics = Vec::new();
-        for child in self.core.children.read().unwrap().values() {
-            child.core.collect_into(&mut metrics);
-        }
-        m.set_metric(metrics);
-
-        vec![m]
-    }
-}
-
-impl<const N: usize> HyperLogLogVec<N> {
-    /// Create a new [`HyperLogLogVec`] based on the provided
-    /// [`Opts`] and partitioned by the given label names. At least one label name must be
-    /// provided.
-    pub fn new(opts: Opts, label_names: &[&str]) -> prometheus::Result<Self> {
-        assert!(N.is_power_of_two());
-        let variable_names = label_names.iter().map(|s| (*s).to_owned()).collect();
-        let opts = opts.variable_labels(variable_names);
-
-        let desc = opts.describe()?;
-        let v = HyperLogLogVecCore {
-            children: RwLock::new(HashMap::default()),
-            desc,
-            opts,
-        };
-
-        Ok(Self { core: Arc::new(v) })
-    }
-
-    /// `get_metric_with_label_values` returns the [`HyperLogLog<P>`] for the given slice
-    /// of label values (same order as the VariableLabels in Desc). If that combination of
-    /// label values is accessed for the first time, a new [`HyperLogLog<P>`] is created.
-    ///
-    /// An error is returned if the number of label values is not the same as the
-    /// number of VariableLabels in Desc.
-    pub fn get_metric_with_label_values(
-        &self,
-        vals: &[&str],
-    ) -> prometheus::Result<HyperLogLog<N>> {
-        self.core.get_metric_with_label_values(vals)
-    }
-
-    /// `with_label_values` works as `get_metric_with_label_values`, but panics if an error
-    /// occurs.
-    pub fn with_label_values(&self, vals: &[&str]) -> HyperLogLog<N> {
-        self.get_metric_with_label_values(vals).unwrap()
-    }
-}
-
-impl<const N: usize> HyperLogLogVecCore<N> {
-    pub fn get_metric_with_label_values(
-        &self,
-        vals: &[&str],
-    ) -> prometheus::Result<HyperLogLog<N>> {
-        let h = self.hash_label_values(vals)?;
-
-        if let Some(metric) = self.children.read().unwrap().get(&h).cloned() {
-            return Ok(metric);
-        }
-
-        self.get_or_create_metric(h, vals)
-    }
-
-    pub(crate) fn hash_label_values(&self, vals: &[&str]) -> prometheus::Result<u64> {
-        if vals.len() != self.desc.variable_labels.len() {
-            return Err(prometheus::Error::InconsistentCardinality {
-                expect: self.desc.variable_labels.len(),
-                got: vals.len(),
-            });
-        }
-
-        let mut h = xxh3::Hash64::default();
-        for val in vals {
-            h.write(val.as_bytes());
-        }
-
-        Ok(h.finish())
-    }
-
-    fn get_or_create_metric(
-        &self,
-        hash: u64,
-        label_values: &[&str],
-    ) -> prometheus::Result<HyperLogLog<N>> {
-        let mut children = self.children.write().unwrap();
-        // Check exist first.
-        if let Some(metric) = children.get(&hash).cloned() {
-            return Ok(metric);
-        }
-
-        let metric = HyperLogLog::with_opts_and_label_values(&self.opts, label_values)?;
-        children.insert(hash, metric.clone());
-        Ok(metric)
-    }
-}
-
-/// HLL is a probabilistic cardinality measure.
-///
-/// How to use this time-series for a metric name `my_metrics_total_hll`:
-///
-/// ```promql
-/// # harmonic mean
-/// 1 / (
-///     sum (
-///         2 ^ -(
-///             # HLL merge operation
-///             max (my_metrics_total_hll{}) by (hll_shard, other_labels...)
-///         )
-///     ) without (hll_shard)
-/// )
-/// * alpha
-/// * shards_count
-/// * shards_count
-/// ```
-///
-/// If you want an estimate over time, you can use the following query:
-///
-/// ```promql
-/// # harmonic mean
-/// 1 / (
-///     sum (
-///         2 ^ -(
-///             # HLL merge operation
-///             max (
-///                 max_over_time(my_metrics_total_hll{}[$__rate_interval])
-///             ) by (hll_shard, other_labels...)
-///         )
-///     ) without (hll_shard)
-/// )
-/// * alpha
-/// * shards_count
-/// * shards_count
-/// ```
-///
-/// In the case of low cardinality, you might want to use the linear counting approximation:
-///
-/// ```promql
-/// # LinearCounting(m, V) = m log (m / V)
-/// shards_count * ln(shards_count /
-///     # calculate V = how many shards contain a 0
-///     count(max (proxy_connecting_endpoints{}) by (hll_shard, protocol) == 0) without (hll_shard)
-/// )
-/// ```
-///
-/// See <https://en.wikipedia.org/wiki/HyperLogLog#Practical_considerations> for estimates on alpha
-#[derive(Clone)]
-pub struct HyperLogLog<const N: usize> {
-    core: Arc<HyperLogLogCore<N>>,
-}
-
-impl<const N: usize> HyperLogLog<N> {
-    /// Create a [`HyperLogLog`] with the `name` and `help` arguments.
-    pub fn new<S1: Into<String>, S2: Into<String>>(name: S1, help: S2) -> prometheus::Result<Self> {
-        assert!(N.is_power_of_two());
-        let opts = Opts::new(name, help);
-        Self::with_opts(opts)
-    }
-
-    /// Create a [`HyperLogLog`] with the `opts` options.
-    pub fn with_opts(opts: Opts) -> prometheus::Result<Self> {
-        Self::with_opts_and_label_values(&opts, &[])
-    }
-
-    fn with_opts_and_label_values(opts: &Opts, label_values: &[&str]) -> prometheus::Result<Self> {
-        let desc = opts.describe()?;
-        let labels = make_label_pairs(&desc, label_values)?;
-
-        let v = HyperLogLogCore {
-            shards: [0; N].map(AtomicU8::new),
-            desc,
-            labels,
-        };
-        Ok(Self { core: Arc::new(v) })
-    }
-
-    pub fn measure(&self, item: &impl Hash) {
-        // changing the hasher will break compatibility with previous measurements.
-        self.record(BuildHasherDefault::<xxh3::Hash64>::default().hash_one(item));
-    }
-
-    fn record(&self, hash: u64) {
-        let p = N.ilog2() as u8;
-        let j = hash & (N as u64 - 1);
-        let rho = (hash >> p).leading_zeros() as u8 + 1 - p;
-        self.core.shards[j as usize].fetch_max(rho, std::sync::atomic::Ordering::Relaxed);
-    }
-}
-
-struct HyperLogLogCore<const N: usize> {
-    shards: [AtomicU8; N],
-    desc: core::Desc,
-    labels: Vec<proto::LabelPair>,
-}
-
-impl<const N: usize> core::Collector for HyperLogLog<N> {
-    fn desc(&self) -> Vec<&core::Desc> {
-        vec![&self.core.desc]
-    }
-
-    fn collect(&self) -> Vec<proto::MetricFamily> {
-        let mut m = proto::MetricFamily::default();
-        m.set_name(self.core.desc.fq_name.clone());
-        m.set_help(self.core.desc.help.clone());
-        m.set_field_type(proto::MetricType::GAUGE);
-
-        let mut metrics = Vec::new();
-        self.core.collect_into(&mut metrics);
-        m.set_metric(metrics);
-
-        vec![m]
-    }
-}
-
-impl<const N: usize> HyperLogLogCore<N> {
-    fn collect_into(&self, metrics: &mut Vec<proto::Metric>) {
-        self.shards.iter().enumerate().for_each(|(i, x)| {
-            let mut shard_label = proto::LabelPair::default();
-            shard_label.set_name("hll_shard".to_owned());
-            shard_label.set_value(format!("{i}"));
-
-            // We reset the counter to 0 so we can perform a cardinality measure over any time slice in prometheus.
-
-            // This seems like it would be a race condition,
-            // but HLL is not impacted by a write in one shard happening in between.
-            // This is because in PromQL we will be implementing a harmonic mean of all buckets.
-            // we will also merge samples in a time series using `max by (hll_shard)`.
-
-            // TODO: maybe we shouldn't reset this on every collect, instead, only after a time window.
-            // this would mean that a dev port-forwarding the metrics url won't break the sampling.
-            let v = x.swap(0, std::sync::atomic::Ordering::Relaxed);
-
-            let mut m = proto::Metric::default();
-            let mut c = proto::Gauge::default();
-            c.set_value(v as f64);
-            m.set_gauge(c);
-
-            let mut labels = Vec::with_capacity(self.labels.len() + 1);
-            labels.extend_from_slice(&self.labels);
-            labels.push(shard_label);
-
-            m.set_label(labels);
-            metrics.push(m);
-        })
-    }
-}
-
-fn make_label_pairs(
-    desc: &core::Desc,
-    label_values: &[&str],
-) -> prometheus::Result<Vec<proto::LabelPair>> {
-    if desc.variable_labels.len() != label_values.len() {
-        return Err(prometheus::Error::InconsistentCardinality {
-            expect: desc.variable_labels.len(),
-            got: label_values.len(),
-        });
-    }
-
-    let total_len = desc.variable_labels.len() + desc.const_label_pairs.len();
-    if total_len == 0 {
-        return Ok(vec![]);
-    }
-
-    if desc.variable_labels.is_empty() {
-        return Ok(desc.const_label_pairs.clone());
-    }
-
-    let mut label_pairs = Vec::with_capacity(total_len);
-    for (i, n) in desc.variable_labels.iter().enumerate() {
-        let mut label_pair = proto::LabelPair::default();
-        label_pair.set_name(n.clone());
-        label_pair.set_value(label_values[i].to_owned());
-        label_pairs.push(label_pair);
-    }
-
-    for label_pair in &desc.const_label_pairs {
-        label_pairs.push(label_pair.clone());
-    }
-    label_pairs.sort();
-    Ok(label_pairs)
-}
-
-#[cfg(test)]
-mod tests {
-    use std::collections::HashSet;
-
-    use prometheus::{proto, Opts};
-    use rand::{rngs::StdRng, Rng, SeedableRng};
-    use rand_distr::{Distribution, Zipf};
-
-    use crate::HyperLogLogVec;
-
-    fn collect(hll: &HyperLogLogVec<32>) -> Vec<proto::Metric> {
-        let mut metrics = vec![];
-        hll.core
-            .children
-            .read()
-            .unwrap()
-            .values()
-            .for_each(|c| c.core.collect_into(&mut metrics));
-        metrics
-    }
-    fn get_cardinality(metrics: &[proto::Metric], filter: impl Fn(&proto::Metric) -> bool) -> f64 {
-        let mut buckets = [0.0; 32];
-        for metric in metrics.chunks_exact(32) {
-            if filter(&metric[0]) {
-                for (i, m) in metric.iter().enumerate() {
-                    buckets[i] = f64::max(buckets[i], m.get_gauge().get_value());
-                }
-            }
-        }
-
-        buckets
-            .into_iter()
-            .map(|f| 2.0f64.powf(-f))
-            .sum::<f64>()
-            .recip()
-            * 0.697
-            * 32.0
-            * 32.0
-    }
-
-    fn test_cardinality(n: usize, dist: impl Distribution<f64>) -> ([usize; 3], [f64; 3]) {
-        let hll = HyperLogLogVec::<32>::new(Opts::new("foo", "bar"), &["x"]).unwrap();
-
-        let mut iter = StdRng::seed_from_u64(0x2024_0112).sample_iter(dist);
-        let mut set_a = HashSet::new();
-        let mut set_b = HashSet::new();
-
-        for x in iter.by_ref().take(n) {
-            set_a.insert(x.to_bits());
-            hll.with_label_values(&["a"]).measure(&x.to_bits());
-        }
-        for x in iter.by_ref().take(n) {
-            set_b.insert(x.to_bits());
-            hll.with_label_values(&["b"]).measure(&x.to_bits());
-        }
-        let merge = &set_a | &set_b;
-
-        let metrics = collect(&hll);
-        let len = get_cardinality(&metrics, |_| true);
-        let len_a = get_cardinality(&metrics, |l| l.get_label()[0].get_value() == "a");
-        let len_b = get_cardinality(&metrics, |l| l.get_label()[0].get_value() == "b");
-
-        ([merge.len(), set_a.len(), set_b.len()], [len, len_a, len_b])
-    }
-
-    #[test]
-    fn test_cardinality_small() {
-        let (actual, estimate) = test_cardinality(100, Zipf::new(100, 1.2f64).unwrap());
-
-        assert_eq!(actual, [46, 30, 32]);
-        assert!(51.3 < estimate[0] && estimate[0] < 51.4);
-        assert!(44.0 < estimate[1] && estimate[1] < 44.1);
-        assert!(39.0 < estimate[2] && estimate[2] < 39.1);
-    }
-
-    #[test]
-    fn test_cardinality_medium() {
-        let (actual, estimate) = test_cardinality(10000, Zipf::new(10000, 1.2f64).unwrap());
-
-        assert_eq!(actual, [2529, 1618, 1629]);
-        assert!(2309.1 < estimate[0] && estimate[0] < 2309.2);
-        assert!(1566.6 < estimate[1] && estimate[1] < 1566.7);
-        assert!(1629.5 < estimate[2] && estimate[2] < 1629.6);
-    }
-
-    #[test]
-    fn test_cardinality_large() {
-        let (actual, estimate) = test_cardinality(1_000_000, Zipf::new(1_000_000, 1.2f64).unwrap());
-
-        assert_eq!(actual, [129077, 79579, 79630]);
-        assert!(126067.2 < estimate[0] && estimate[0] < 126067.3);
-        assert!(83076.8 < estimate[1] && estimate[1] < 83076.9);
-        assert!(64251.2 < estimate[2] && estimate[2] < 64251.3);
-    }
-
-    #[test]
-    fn test_cardinality_small2() {
-        let (actual, estimate) = test_cardinality(100, Zipf::new(200, 0.8f64).unwrap());
-
-        assert_eq!(actual, [92, 58, 60]);
-        assert!(116.1 < estimate[0] && estimate[0] < 116.2);
-        assert!(81.7 < estimate[1] && estimate[1] < 81.8);
-        assert!(69.3 < estimate[2] && estimate[2] < 69.4);
-    }
-
-    #[test]
-    fn test_cardinality_medium2() {
-        let (actual, estimate) = test_cardinality(10000, Zipf::new(20000, 0.8f64).unwrap());
-
-        assert_eq!(actual, [8201, 5131, 5051]);
-        assert!(6846.4 < estimate[0] && estimate[0] < 6846.5);
-        assert!(5239.1 < estimate[1] && estimate[1] < 5239.2);
-        assert!(4292.8 < estimate[2] && estimate[2] < 4292.9);
-    }
-
-    #[test]
-    fn test_cardinality_large2() {
-        let (actual, estimate) = test_cardinality(1_000_000, Zipf::new(2_000_000, 0.8f64).unwrap());
-
-        assert_eq!(actual, [777847, 482069, 482246]);
-        assert!(699437.4 < estimate[0] && estimate[0] < 699437.5);
-        assert!(374948.9 < estimate[1] && estimate[1] < 374949.0);
-        assert!(434609.7 < estimate[2] && estimate[2] < 434609.8);
-    }
-}

libs/metrics/src/lib.rs

@@ -28,9 +28,7 @@ use prometheus::{Registry, Result};
 pub mod launch_timestamp;
 mod wrappers;
 pub use wrappers::{CountedReader, CountedWriter};
-mod hll;
 pub mod metric_vec_duration;
-pub use hll::{HyperLogLog, HyperLogLogVec};
 
 pub type UIntGauge = GenericGauge<AtomicU64>;
 pub type UIntGaugeVec = GenericGaugeVec<AtomicU64>;

libs/pageserver_api/Cargo.toml

@@ -20,7 +20,6 @@ strum_macros.workspace = true
 hex.workspace = true
 thiserror.workspace = true
 humantime-serde.workspace = true
-chrono.workspace = true
 
 workspace_hack.workspace = true
 

libs/pageserver_api/src/keyspace.rs

@@ -63,84 +63,16 @@ impl KeySpace {
         KeyPartitioning { parts }
     }
 
-    /// Update the keyspace such that it doesn't contain any range
-    /// that is overlapping with `other`. This can involve splitting or
-    /// removing of existing ranges.
-    pub fn remove_overlapping_with(&mut self, other: &KeySpace) {
-        let (self_start, self_end) = match (self.start(), self.end()) {
-            (Some(start), Some(end)) => (start, end),
-            _ => {
-                // self is empty
-                return;
-            }
-        };
-
-        // Key spaces are sorted by definition, so skip ahead to the first
-        // potentially intersecting range. Similarly, ignore ranges that start
-        // after the current keyspace ends.
-        let other_ranges = other
-            .ranges
-            .iter()
-            .skip_while(|range| self_start >= range.end)
-            .take_while(|range| self_end > range.start);
-
-        for range in other_ranges {
-            while let Some(overlap_at) = self.overlaps_at(range) {
-                let overlapped = self.ranges[overlap_at].clone();
-
-                if overlapped.start < range.start && overlapped.end <= range.end {
-                    // Higher part of the range is completely overlapped.
-                    self.ranges[overlap_at].end = range.start;
-                }
-                if overlapped.start >= range.start && overlapped.end > range.end {
-                    // Lower part of the range is completely overlapped.
-                    self.ranges[overlap_at].start = range.end;
-                }
-                if overlapped.start < range.start && overlapped.end > range.end {
-                    // Middle part of the range is overlapped.
-                    self.ranges[overlap_at].end = range.start;
-                    self.ranges
-                        .insert(overlap_at + 1, range.end..overlapped.end);
-                }
-                if overlapped.start >= range.start && overlapped.end <= range.end {
-                    // Whole range is overlapped
-                    self.ranges.remove(overlap_at);
-                }
-            }
-        }
-    }
-
-    pub fn start(&self) -> Option<Key> {
-        self.ranges.first().map(|range| range.start)
-    }
-
-    pub fn end(&self) -> Option<Key> {
-        self.ranges.last().map(|range| range.end)
-    }
-
-    #[allow(unused)]
-    pub fn total_size(&self) -> usize {
-        self.ranges
-            .iter()
-            .map(|range| key_range_size(range) as usize)
-            .sum()
-    }
-
-    fn overlaps_at(&self, range: &Range<Key>) -> Option<usize> {
-        match self.ranges.binary_search_by_key(&range.end, |r| r.start) {
-            Ok(0) => None,
-            Err(0) => None,
-            Ok(index) if self.ranges[index - 1].end > range.start => Some(index - 1),
-            Err(index) if self.ranges[index - 1].end > range.start => Some(index - 1),
-            _ => None,
-        }
-    }
-
     ///
     /// Check if key space contains overlapping range
     ///
     pub fn overlaps(&self, range: &Range<Key>) -> bool {
-        self.overlaps_at(range).is_some()
+        match self.ranges.binary_search_by_key(&range.end, |r| r.start) {
+            Ok(0) => false,
+            Err(0) => false,
+            Ok(index) => self.ranges[index - 1].end > range.start,
+            Err(index) => self.ranges[index - 1].end > range.start,
+        }
     }
 }
 
@@ -172,7 +104,6 @@ pub struct KeySpaceAccum {
     accum: Option<Range<Key>>,
 
     ranges: Vec<Range<Key>>,
-    size: u64,
 }
 
 impl KeySpaceAccum {
@@ -180,7 +111,6 @@ impl KeySpaceAccum {
         Self {
             accum: None,
             ranges: Vec::new(),
-            size: 0,
         }
     }
 
@@ -191,8 +121,6 @@ impl KeySpaceAccum {
 
     #[inline(always)]
     pub fn add_range(&mut self, range: Range<Key>) {
-        self.size += key_range_size(&range) as u64;
-
         match self.accum.as_mut() {
             Some(accum) => {
                 if range.start == accum.end {
@@ -218,23 +146,6 @@ impl KeySpaceAccum {
             ranges: self.ranges,
         }
     }
-
-    pub fn consume_keyspace(&mut self) -> KeySpace {
-        if let Some(accum) = self.accum.take() {
-            self.ranges.push(accum);
-        }
-
-        let mut prev_accum = KeySpaceAccum::new();
-        std::mem::swap(self, &mut prev_accum);
-
-        KeySpace {
-            ranges: prev_accum.ranges,
-        }
-    }
-
-    pub fn size(&self) -> u64 {
-        self.size
-    }
 }
 
 ///
@@ -343,30 +254,6 @@ mod tests {
         }
     }
 
-    #[test]
-    fn keyspace_consume() {
-        let ranges = vec![kr(0..10), kr(20..35), kr(40..45)];
-
-        let mut accum = KeySpaceAccum::new();
-        for range in &ranges {
-            accum.add_range(range.clone());
-        }
-
-        let expected_size: u64 = ranges.iter().map(|r| key_range_size(r) as u64).sum();
-        assert_eq!(accum.size(), expected_size);
-
-        assert_ks_eq(&accum.consume_keyspace(), ranges.clone());
-        assert_eq!(accum.size(), 0);
-
-        assert_ks_eq(&accum.consume_keyspace(), vec![]);
-        assert_eq!(accum.size(), 0);
-
-        for range in &ranges {
-            accum.add_range(range.clone());
-        }
-        assert_ks_eq(&accum.to_keyspace(), ranges);
-    }
-
     #[test]
     fn keyspace_add_range() {
         // two separate ranges
@@ -509,118 +396,4 @@ mod tests {
         //        xxxxxxxxxxx
         assert!(ks.overlaps(&kr(0..30))); // XXXXX This fails currently!
     }
-
-    #[test]
-    fn test_remove_full_overlapps() {
-        let mut key_space1 = KeySpace {
-            ranges: vec![
-                Key::from_i128(1)..Key::from_i128(4),
-                Key::from_i128(5)..Key::from_i128(8),
-                Key::from_i128(10)..Key::from_i128(12),
-            ],
-        };
-        let key_space2 = KeySpace {
-            ranges: vec![
-                Key::from_i128(2)..Key::from_i128(3),
-                Key::from_i128(6)..Key::from_i128(7),
-                Key::from_i128(11)..Key::from_i128(13),
-            ],
-        };
-        key_space1.remove_overlapping_with(&key_space2);
-        assert_eq!(
-            key_space1.ranges,
-            vec![
-                Key::from_i128(1)..Key::from_i128(2),
-                Key::from_i128(3)..Key::from_i128(4),
-                Key::from_i128(5)..Key::from_i128(6),
-                Key::from_i128(7)..Key::from_i128(8),
-                Key::from_i128(10)..Key::from_i128(11)
-            ]
-        );
-    }
-
-    #[test]
-    fn test_remove_partial_overlaps() {
-        // Test partial ovelaps
-        let mut key_space1 = KeySpace {
-            ranges: vec![
-                Key::from_i128(1)..Key::from_i128(5),
-                Key::from_i128(7)..Key::from_i128(10),
-                Key::from_i128(12)..Key::from_i128(15),
-            ],
-        };
-        let key_space2 = KeySpace {
-            ranges: vec![
-                Key::from_i128(3)..Key::from_i128(6),
-                Key::from_i128(8)..Key::from_i128(11),
-                Key::from_i128(14)..Key::from_i128(17),
-            ],
-        };
-        key_space1.remove_overlapping_with(&key_space2);
-        assert_eq!(
-            key_space1.ranges,
-            vec![
-                Key::from_i128(1)..Key::from_i128(3),
-                Key::from_i128(7)..Key::from_i128(8),
-                Key::from_i128(12)..Key::from_i128(14),
-            ]
-        );
-    }
-
-    #[test]
-    fn test_remove_no_overlaps() {
-        let mut key_space1 = KeySpace {
-            ranges: vec![
-                Key::from_i128(1)..Key::from_i128(5),
-                Key::from_i128(7)..Key::from_i128(10),
-                Key::from_i128(12)..Key::from_i128(15),
-            ],
-        };
-        let key_space2 = KeySpace {
-            ranges: vec![
-                Key::from_i128(6)..Key::from_i128(7),
-                Key::from_i128(11)..Key::from_i128(12),
-                Key::from_i128(15)..Key::from_i128(17),
-            ],
-        };
-        key_space1.remove_overlapping_with(&key_space2);
-        assert_eq!(
-            key_space1.ranges,
-            vec![
-                Key::from_i128(1)..Key::from_i128(5),
-                Key::from_i128(7)..Key::from_i128(10),
-                Key::from_i128(12)..Key::from_i128(15),
-            ]
-        );
-    }
-
-    #[test]
-    fn test_remove_one_range_overlaps_multiple() {
-        let mut key_space1 = KeySpace {
-            ranges: vec![
-                Key::from_i128(1)..Key::from_i128(3),
-                Key::from_i128(3)..Key::from_i128(6),
-                Key::from_i128(6)..Key::from_i128(10),
-                Key::from_i128(12)..Key::from_i128(15),
-                Key::from_i128(17)..Key::from_i128(20),
-                Key::from_i128(20)..Key::from_i128(30),
-                Key::from_i128(30)..Key::from_i128(40),
-            ],
-        };
-        let key_space2 = KeySpace {
-            ranges: vec![Key::from_i128(9)..Key::from_i128(19)],
-        };
-        key_space1.remove_overlapping_with(&key_space2);
-        assert_eq!(
-            key_space1.ranges,
-            vec![
-                Key::from_i128(1)..Key::from_i128(3),
-                Key::from_i128(3)..Key::from_i128(6),
-                Key::from_i128(6)..Key::from_i128(9),
-                Key::from_i128(19)..Key::from_i128(20),
-                Key::from_i128(20)..Key::from_i128(30),
-                Key::from_i128(30)..Key::from_i128(40),
-            ]
-        );
-    }
 }

libs/pageserver_api/src/models.rs

@@ -8,7 +8,6 @@ use std::{
 };
 
 use byteorder::{BigEndian, ReadBytesExt};
-use postgres_ffi::BLCKSZ;
 use serde::{Deserialize, Serialize};
 use serde_with::serde_as;
 use strum_macros;
@@ -272,7 +271,6 @@ pub struct TenantConfig {
     pub evictions_low_residence_duration_metric_threshold: Option<String>,
     pub gc_feedback: Option<bool>,
     pub heatmap_period: Option<String>,
-    pub lazy_slru_download: Option<bool>,
 }
 
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
@@ -366,19 +364,6 @@ pub struct TenantLocationConfigRequest {
     pub config: LocationConfig, // as we have a flattened field, we should reject all unknown fields in it
 }
 
-#[derive(Serialize, Deserialize, Debug)]
-#[serde(deny_unknown_fields)]
-pub struct TenantShardLocation {
-    pub shard_id: TenantShardId,
-    pub node_id: NodeId,
-}
-
-#[derive(Serialize, Deserialize, Debug)]
-#[serde(deny_unknown_fields)]
-pub struct TenantLocationConfigResponse {
-    pub shards: Vec<TenantShardLocation>,
-}
-
 #[derive(Serialize, Deserialize, Debug)]
 #[serde(deny_unknown_fields)]
 pub struct TenantConfigRequest {
@@ -454,8 +439,6 @@ pub struct TenantDetails {
     #[serde(flatten)]
     pub tenant_info: TenantInfo,
 
-    pub walredo: Option<WalRedoManagerStatus>,
-
     pub timelines: Vec<TimelineId>,
 }
 
@@ -643,12 +626,6 @@ pub struct TimelineGcRequest {
     pub gc_horizon: Option<u64>,
 }
 
-#[derive(Debug, Clone, Serialize, Deserialize)]
-pub struct WalRedoManagerStatus {
-    pub last_redo_at: Option<chrono::DateTime<chrono::Utc>>,
-    pub pid: Option<u32>,
-}
-
 // Wrapped in libpq CopyData
 #[derive(PartialEq, Eq, Debug)]
 pub enum PagestreamFeMessage {
@@ -656,7 +633,6 @@ pub enum PagestreamFeMessage {
     Nblocks(PagestreamNblocksRequest),
     GetPage(PagestreamGetPageRequest),
     DbSize(PagestreamDbSizeRequest),
-    GetSlruSegment(PagestreamGetSlruSegmentRequest),
 }
 
 // Wrapped in libpq CopyData
@@ -667,7 +643,6 @@ pub enum PagestreamBeMessage {
     GetPage(PagestreamGetPageResponse),
     Error(PagestreamErrorResponse),
     DbSize(PagestreamDbSizeResponse),
-    GetSlruSegment(PagestreamGetSlruSegmentResponse),
 }
 
 // Keep in sync with `pagestore_client.h`
@@ -678,7 +653,6 @@ enum PagestreamBeMessageTag {
     GetPage = 102,
     Error = 103,
     DbSize = 104,
-    GetSlruSegment = 105,
 }
 impl TryFrom<u8> for PagestreamBeMessageTag {
     type Error = u8;
@@ -689,7 +663,6 @@ impl TryFrom<u8> for PagestreamBeMessageTag {
             102 => Ok(PagestreamBeMessageTag::GetPage),
             103 => Ok(PagestreamBeMessageTag::Error),
             104 => Ok(PagestreamBeMessageTag::DbSize),
-            105 => Ok(PagestreamBeMessageTag::GetSlruSegment),
             _ => Err(value),
         }
     }
@@ -724,14 +697,6 @@ pub struct PagestreamDbSizeRequest {
     pub dbnode: u32,
 }
 
-#[derive(Debug, PartialEq, Eq)]
-pub struct PagestreamGetSlruSegmentRequest {
-    pub latest: bool,
-    pub lsn: Lsn,
-    pub kind: u8,
-    pub segno: u32,
-}
-
 #[derive(Debug)]
 pub struct PagestreamExistsResponse {
     pub exists: bool,
@@ -747,11 +712,6 @@ pub struct PagestreamGetPageResponse {
     pub page: Bytes,
 }
 
-#[derive(Debug)]
-pub struct PagestreamGetSlruSegmentResponse {
-    pub segment: Bytes,
-}
-
 #[derive(Debug)]
 pub struct PagestreamErrorResponse {
     pub message: String,
@@ -815,14 +775,6 @@ impl PagestreamFeMessage {
                 bytes.put_u64(req.lsn.0);
                 bytes.put_u32(req.dbnode);
             }
-
-            Self::GetSlruSegment(req) => {
-                bytes.put_u8(4);
-                bytes.put_u8(u8::from(req.latest));
-                bytes.put_u64(req.lsn.0);
-                bytes.put_u8(req.kind);
-                bytes.put_u32(req.segno);
-            }
         }
 
         bytes.into()
@@ -873,14 +825,6 @@ impl PagestreamFeMessage {
                 lsn: Lsn::from(body.read_u64::<BigEndian>()?),
                 dbnode: body.read_u32::<BigEndian>()?,
             })),
-            4 => Ok(PagestreamFeMessage::GetSlruSegment(
-                PagestreamGetSlruSegmentRequest {
-                    latest: body.read_u8()? != 0,
-                    lsn: Lsn::from(body.read_u64::<BigEndian>()?),
-                    kind: body.read_u8()?,
-                    segno: body.read_u32::<BigEndian>()?,
-                },
-            )),
             _ => bail!("unknown smgr message tag: {:?}", msg_tag),
         }
     }
@@ -916,12 +860,6 @@ impl PagestreamBeMessage {
                 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[..]);
-            }
         }
 
         bytes.into()
@@ -962,14 +900,6 @@ impl PagestreamBeMessage {
                     let db_size = buf.read_i64::<BigEndian>()?;
                     Self::DbSize(PagestreamDbSizeResponse { db_size })
                 }
-                Tag::GetSlruSegment => {
-                    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 {
-                        segment: segment.into(),
-                    })
-                }
             };
         let remaining = buf.into_inner();
         if !remaining.is_empty() {
@@ -988,7 +918,6 @@ impl PagestreamBeMessage {
             Self::GetPage(_) => "GetPage",
             Self::Error(_) => "Error",
             Self::DbSize(_) => "DbSize",
-            Self::GetSlruSegment(_) => "GetSlruSegment",
         }
     }
 }

libs/pageserver_api/src/reltag.rs

@@ -111,23 +111,9 @@ impl RelTag {
 /// These files are divided into segments, which are divided into
 /// pages of the same BLCKSZ as used for relation files.
 ///
-#[derive(
-    Debug,
-    Clone,
-    Copy,
-    Hash,
-    Serialize,
-    Deserialize,
-    PartialEq,
-    Eq,
-    PartialOrd,
-    Ord,
-    strum_macros::EnumIter,
-    strum_macros::FromRepr,
-)]
-#[repr(u8)]
+#[derive(Debug, Clone, Copy, Hash, Serialize, Deserialize, PartialEq, Eq, PartialOrd, Ord)]
 pub enum SlruKind {
-    Clog = 0,
+    Clog,
     MultiXactMembers,
     MultiXactOffsets,
 }

libs/postgres_ffi/src/xlog_utils.rs

@@ -207,16 +207,10 @@ pub fn find_end_of_wal(
                 let seg_offs = curr_lsn.segment_offset(wal_seg_size);
                 segment.seek(SeekFrom::Start(seg_offs as u64))?;
                 // loop inside segment
-                while curr_lsn.segment_number(wal_seg_size) == segno {
+                loop {
                     let bytes_read = segment.read(&mut buf)?;
                     if bytes_read == 0 {
-                        debug!(
-                            "find_end_of_wal reached end at {:?}, EOF in segment {:?} at offset {}",
-                            result,
-                            seg_file_path,
-                            curr_lsn.segment_offset(wal_seg_size)
-                        );
-                        return Ok(result);
+                        break; // EOF
                     }
                     curr_lsn += bytes_read as u64;
                     decoder.feed_bytes(&buf[0..bytes_read]);

libs/pq_proto/src/framed.rs

@@ -82,6 +82,19 @@ impl<S> Framed<S> {
             write_buf: self.write_buf,
         })
     }
+
+    /// Return new Framed with stream type transformed by f. For dynamic dispatch.
+    pub fn map_stream_sync<S2, F>(self, f: F) -> Framed<S2>
+    where
+        F: FnOnce(S) -> S2,
+    {
+        let stream = f(self.stream);
+        Framed {
+            stream,
+            read_buf: self.read_buf,
+            write_buf: self.write_buf,
+        }
+    }
 }
 
 impl<S: AsyncRead + Unpin> Framed<S> {

libs/remote_storage/src/azure_blob.rs

@@ -8,7 +8,6 @@ use std::pin::Pin;
 use std::str::FromStr;
 use std::sync::Arc;
 use std::time::Duration;
-use std::time::SystemTime;
 
 use super::REMOTE_STORAGE_PREFIX_SEPARATOR;
 use anyhow::Result;
@@ -24,11 +23,9 @@ use futures::stream::Stream;
 use futures_util::StreamExt;
 use http_types::{StatusCode, Url};
 use tokio::time::Instant;
-use tokio_util::sync::CancellationToken;
 use tracing::debug;
 
 use crate::s3_bucket::RequestKind;
-use crate::TimeTravelError;
 use crate::{
     AzureConfig, ConcurrencyLimiter, Download, DownloadError, Listing, ListingMode, RemotePath,
     RemoteStorage, StorageMetadata,
@@ -186,6 +183,7 @@ fn to_download_error(error: azure_core::Error) -> DownloadError {
     }
 }
 
+#[async_trait::async_trait]
 impl RemoteStorage for AzureBlobStorage {
     async fn list(
         &self,
@@ -373,18 +371,6 @@ impl RemoteStorage for AzureBlobStorage {
             copy_status = status;
         }
     }
-
-    async fn time_travel_recover(
-        &self,
-        _prefix: Option<&RemotePath>,
-        _timestamp: SystemTime,
-        _done_if_after: SystemTime,
-        _cancel: &CancellationToken,
-    ) -> Result<(), TimeTravelError> {
-        // TODO use Azure point in time recovery feature for this
-        // https://learn.microsoft.com/en-us/azure/storage/blobs/point-in-time-restore-overview
-        Err(TimeTravelError::Unimplemented)
-    }
 }
 
 pin_project_lite::pin_project! {

libs/remote_storage/src/lib.rs

@@ -25,7 +25,6 @@ use bytes::Bytes;
 use futures::stream::Stream;
 use serde::{Deserialize, Serialize};
 use tokio::sync::Semaphore;
-use tokio_util::sync::CancellationToken;
 use toml_edit::Item;
 use tracing::info;
 
@@ -143,7 +142,7 @@ pub struct Listing {
 /// Storage (potentially remote) API to manage its state.
 /// This storage tries to be unaware of any layered repository context,
 /// providing basic CRUD operations for storage files.
-#[allow(async_fn_in_trait)]
+#[async_trait::async_trait]
 pub trait RemoteStorage: Send + Sync + 'static {
     /// Lists all top level subdirectories for a given prefix
     /// Note: here we assume that if the prefix is passed it was obtained via remote_object_id
@@ -211,15 +210,6 @@ pub trait RemoteStorage: Send + Sync + 'static {
 
     /// Copy a remote object inside a bucket from one path to another.
     async fn copy(&self, from: &RemotePath, to: &RemotePath) -> anyhow::Result<()>;
-
-    /// Resets the content of everything with the given prefix to the given state
-    async fn time_travel_recover(
-        &self,
-        prefix: Option<&RemotePath>,
-        timestamp: SystemTime,
-        done_if_after: SystemTime,
-        cancel: &CancellationToken,
-    ) -> Result<(), TimeTravelError>;
 }
 
 pub type DownloadStream = Pin<Box<dyn Stream<Item = std::io::Result<Bytes>> + Unpin + Send + Sync>>;
@@ -269,57 +259,17 @@ impl std::fmt::Display for DownloadError {
 
 impl std::error::Error for DownloadError {}
 
-#[derive(Debug)]
-pub enum TimeTravelError {
-    /// Validation or other error happened due to user input.
-    BadInput(anyhow::Error),
-    /// The used remote storage does not have time travel recovery implemented
-    Unimplemented,
-    /// The number of versions/deletion markers is above our limit.
-    TooManyVersions,
-    /// A cancellation token aborted the process, typically during
-    /// request closure or process shutdown.
-    Cancelled,
-    /// Other errors
-    Other(anyhow::Error),
-}
-
-impl std::fmt::Display for TimeTravelError {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        match self {
-            TimeTravelError::BadInput(e) => {
-                write!(
-                    f,
-                    "Failed to time travel recover a prefix due to user input: {e}"
-                )
-            }
-            TimeTravelError::Unimplemented => write!(
-                f,
-                "time travel recovery is not implemented for the current storage backend"
-            ),
-            TimeTravelError::Cancelled => write!(f, "Cancelled, shutting down"),
-            TimeTravelError::TooManyVersions => {
-                write!(f, "Number of versions/delete markers above limit")
-            }
-            TimeTravelError::Other(e) => write!(f, "Failed to time travel recover a prefix: {e:?}"),
-        }
-    }
-}
-
-impl std::error::Error for TimeTravelError {}
-
 /// Every storage, currently supported.
 /// Serves as a simple way to pass around the [`RemoteStorage`] without dealing with generics.
 #[derive(Clone)]
-// Require Clone for `Other` due to https://github.com/rust-lang/rust/issues/26925
-pub enum GenericRemoteStorage<Other: Clone = Arc<UnreliableWrapper>> {
+pub enum GenericRemoteStorage {
     LocalFs(LocalFs),
     AwsS3(Arc<S3Bucket>),
     AzureBlob(Arc<AzureBlobStorage>),
-    Unreliable(Other),
+    Unreliable(Arc<UnreliableWrapper>),
 }
 
-impl<Other: RemoteStorage> GenericRemoteStorage<Arc<Other>> {
+impl GenericRemoteStorage {
     pub async fn list(
         &self,
         prefix: Option<&RemotePath>,
@@ -436,33 +386,6 @@ impl<Other: RemoteStorage> GenericRemoteStorage<Arc<Other>> {
             Self::Unreliable(s) => s.copy(from, to).await,
         }
     }
-
-    pub async fn time_travel_recover(
-        &self,
-        prefix: Option<&RemotePath>,
-        timestamp: SystemTime,
-        done_if_after: SystemTime,
-        cancel: &CancellationToken,
-    ) -> Result<(), TimeTravelError> {
-        match self {
-            Self::LocalFs(s) => {
-                s.time_travel_recover(prefix, timestamp, done_if_after, cancel)
-                    .await
-            }
-            Self::AwsS3(s) => {
-                s.time_travel_recover(prefix, timestamp, done_if_after, cancel)
-                    .await
-            }
-            Self::AzureBlob(s) => {
-                s.time_travel_recover(prefix, timestamp, done_if_after, cancel)
-                    .await
-            }
-            Self::Unreliable(s) => {
-                s.time_travel_recover(prefix, timestamp, done_if_after, cancel)
-                    .await
-            }
-        }
-    }
 }
 
 impl GenericRemoteStorage {
@@ -473,12 +396,7 @@ impl GenericRemoteStorage {
                 Self::LocalFs(LocalFs::new(root.clone())?)
             }
             RemoteStorageKind::AwsS3(s3_config) => {
-                // The profile and access key id are only printed here for debugging purposes,
-                // their values don't indicate the eventually taken choice for auth.
-                let profile = std::env::var("AWS_PROFILE").unwrap_or_else(|_| "<none>".into());
-                let access_key_id =
-                    std::env::var("AWS_ACCESS_KEY_ID").unwrap_or_else(|_| "<none>".into());
-                info!("Using s3 bucket '{}' in region '{}' as a remote storage, prefix in bucket: '{:?}', bucket endpoint: '{:?}', profile: {profile}, access_key_id: {access_key_id}",
+                info!("Using s3 bucket '{}' in region '{}' as a remote storage, prefix in bucket: '{:?}', bucket endpoint: '{:?}'",
                       s3_config.bucket_name, s3_config.bucket_region, s3_config.prefix_in_bucket, s3_config.endpoint);
                 Self::AwsS3(Arc::new(S3Bucket::new(s3_config)?))
             }
@@ -755,7 +673,6 @@ impl ConcurrencyLimiter {
             RequestKind::List => &self.read,
             RequestKind::Delete => &self.write,
             RequestKind::Copy => &self.write,
-            RequestKind::TimeTravel => &self.write,
         }
     }
 

libs/remote_storage/src/local_fs.rs

@@ -4,7 +4,7 @@
 //! This storage used in tests, but can also be used in cases when a certain persistent
 //! volume is mounted to the local FS.
 
-use std::{borrow::Cow, future::Future, io::ErrorKind, pin::Pin, time::SystemTime};
+use std::{borrow::Cow, future::Future, io::ErrorKind, pin::Pin};
 
 use anyhow::{bail, ensure, Context};
 use bytes::Bytes;
@@ -14,13 +14,11 @@ use tokio::{
     fs,
     io::{self, AsyncReadExt, AsyncSeekExt, AsyncWriteExt},
 };
-use tokio_util::{io::ReaderStream, sync::CancellationToken};
+use tokio_util::io::ReaderStream;
 use tracing::*;
 use utils::{crashsafe::path_with_suffix_extension, fs_ext::is_directory_empty};
 
-use crate::{
-    Download, DownloadError, DownloadStream, Listing, ListingMode, RemotePath, TimeTravelError,
-};
+use crate::{Download, DownloadError, DownloadStream, Listing, ListingMode, RemotePath};
 
 use super::{RemoteStorage, StorageMetadata};
 
@@ -159,6 +157,7 @@ impl LocalFs {
     }
 }
 
+#[async_trait::async_trait]
 impl RemoteStorage for LocalFs {
     async fn list(
         &self,
@@ -424,17 +423,6 @@ impl RemoteStorage for LocalFs {
         })?;
         Ok(())
     }
-
-    #[allow(clippy::diverging_sub_expression)]
-    async fn time_travel_recover(
-        &self,
-        _prefix: Option<&RemotePath>,
-        _timestamp: SystemTime,
-        _done_if_after: SystemTime,
-        _cancel: &CancellationToken,
-    ) -> Result<(), TimeTravelError> {
-        Err(TimeTravelError::Unimplemented)
-    }
 }
 
 fn storage_metadata_path(original_path: &Utf8Path) -> Utf8PathBuf {

libs/remote_storage/src/s3_bucket.rs

@@ -6,14 +6,12 @@
 
 use std::{
     borrow::Cow,
-    collections::HashMap,
     pin::Pin,
     sync::Arc,
     task::{Context, Poll},
-    time::SystemTime,
 };
 
-use anyhow::{anyhow, Context as _};
+use anyhow::Context as _;
 use aws_config::{
     environment::credentials::EnvironmentVariableCredentialsProvider,
     imds::credentials::ImdsCredentialsProvider,
@@ -29,24 +27,22 @@ use aws_sdk_s3::{
     config::{AsyncSleep, Builder, IdentityCache, Region, SharedAsyncSleep},
     error::SdkError,
     operation::get_object::GetObjectError,
-    types::{Delete, DeleteMarkerEntry, ObjectIdentifier, ObjectVersion},
+    types::{Delete, ObjectIdentifier},
     Client,
 };
 use aws_smithy_async::rt::sleep::TokioSleep;
 
+use aws_smithy_types::body::SdkBody;
 use aws_smithy_types::byte_stream::ByteStream;
-use aws_smithy_types::{body::SdkBody, DateTime};
 use bytes::Bytes;
 use futures::stream::Stream;
 use hyper::Body;
 use scopeguard::ScopeGuard;
-use tokio_util::sync::CancellationToken;
-use utils::backoff;
 
 use super::StorageMetadata;
 use crate::{
     ConcurrencyLimiter, Download, DownloadError, Listing, ListingMode, RemotePath, RemoteStorage,
-    S3Config, TimeTravelError, MAX_KEYS_PER_DELETE, REMOTE_STORAGE_PREFIX_SEPARATOR,
+    S3Config, MAX_KEYS_PER_DELETE, REMOTE_STORAGE_PREFIX_SEPARATOR,
 };
 
 pub(super) mod metrics;
@@ -274,59 +270,6 @@ impl S3Bucket {
             }
         }
     }
-
-    async fn delete_oids(
-        &self,
-        kind: RequestKind,
-        delete_objects: &[ObjectIdentifier],
-    ) -> anyhow::Result<()> {
-        for chunk in delete_objects.chunks(MAX_KEYS_PER_DELETE) {
-            let started_at = start_measuring_requests(kind);
-
-            let resp = self
-                .client
-                .delete_objects()
-                .bucket(self.bucket_name.clone())
-                .delete(
-                    Delete::builder()
-                        .set_objects(Some(chunk.to_vec()))
-                        .build()?,
-                )
-                .send()
-                .await;
-
-            let started_at = ScopeGuard::into_inner(started_at);
-            metrics::BUCKET_METRICS
-                .req_seconds
-                .observe_elapsed(kind, &resp, started_at);
-
-            let resp = resp?;
-            metrics::BUCKET_METRICS
-                .deleted_objects_total
-                .inc_by(chunk.len() as u64);
-            if let Some(errors) = resp.errors {
-                // Log a bounded number of the errors within the response:
-                // these requests can carry 1000 keys so logging each one
-                // would be too verbose, especially as errors may lead us
-                // to retry repeatedly.
-                const LOG_UP_TO_N_ERRORS: usize = 10;
-                for e in errors.iter().take(LOG_UP_TO_N_ERRORS) {
-                    tracing::warn!(
-                        "DeleteObjects key {} failed: {}: {}",
-                        e.key.as_ref().map(Cow::from).unwrap_or("".into()),
-                        e.code.as_ref().map(Cow::from).unwrap_or("".into()),
-                        e.message.as_ref().map(Cow::from).unwrap_or("".into())
-                    );
-                }
-
-                return Err(anyhow::format_err!(
-                    "Failed to delete {} objects",
-                    errors.len()
-                ));
-            }
-        }
-        Ok(())
-    }
 }
 
 pin_project_lite::pin_project! {
@@ -430,6 +373,7 @@ impl<S: Stream<Item = std::io::Result<Bytes>>> Stream for TimedDownload<S> {
     }
 }
 
+#[async_trait::async_trait]
 impl RemoteStorage for S3Bucket {
     async fn list(
         &self,
@@ -625,222 +569,64 @@ impl RemoteStorage for S3Bucket {
             delete_objects.push(obj_id);
         }
 
-        self.delete_oids(kind, &delete_objects).await
+        for chunk in delete_objects.chunks(MAX_KEYS_PER_DELETE) {
+            let started_at = start_measuring_requests(kind);
+
+            let resp = self
+                .client
+                .delete_objects()
+                .bucket(self.bucket_name.clone())
+                .delete(
+                    Delete::builder()
+                        .set_objects(Some(chunk.to_vec()))
+                        .build()?,
+                )
+                .send()
+                .await;
+
+            let started_at = ScopeGuard::into_inner(started_at);
+            metrics::BUCKET_METRICS
+                .req_seconds
+                .observe_elapsed(kind, &resp, started_at);
+
+            match resp {
+                Ok(resp) => {
+                    metrics::BUCKET_METRICS
+                        .deleted_objects_total
+                        .inc_by(chunk.len() as u64);
+                    if let Some(errors) = resp.errors {
+                        // Log a bounded number of the errors within the response:
+                        // these requests can carry 1000 keys so logging each one
+                        // would be too verbose, especially as errors may lead us
+                        // to retry repeatedly.
+                        const LOG_UP_TO_N_ERRORS: usize = 10;
+                        for e in errors.iter().take(LOG_UP_TO_N_ERRORS) {
+                            tracing::warn!(
+                                "DeleteObjects key {} failed: {}: {}",
+                                e.key.as_ref().map(Cow::from).unwrap_or("".into()),
+                                e.code.as_ref().map(Cow::from).unwrap_or("".into()),
+                                e.message.as_ref().map(Cow::from).unwrap_or("".into())
+                            );
+                        }
+
+                        return Err(anyhow::format_err!(
+                            "Failed to delete {} objects",
+                            errors.len()
+                        ));
+                    }
+                }
+                Err(e) => {
+                    return Err(e.into());
+                }
+            }
+        }
+        Ok(())
     }
 
     async fn delete(&self, path: &RemotePath) -> anyhow::Result<()> {
         let paths = std::array::from_ref(path);
         self.delete_objects(paths).await
     }
-
-    async fn time_travel_recover(
-        &self,
-        prefix: Option<&RemotePath>,
-        timestamp: SystemTime,
-        done_if_after: SystemTime,
-        cancel: &CancellationToken,
-    ) -> Result<(), TimeTravelError> {
-        let kind = RequestKind::TimeTravel;
-        let _guard = self.permit(kind).await;
-
-        let timestamp = DateTime::from(timestamp);
-        let done_if_after = DateTime::from(done_if_after);
-
-        tracing::trace!("Target time: {timestamp:?}, done_if_after {done_if_after:?}");
-
-        // get the passed prefix or if it is not set use prefix_in_bucket value
-        let prefix = prefix
-            .map(|p| self.relative_path_to_s3_object(p))
-            .or_else(|| self.prefix_in_bucket.clone());
-
-        let warn_threshold = 3;
-        let max_retries = 10;
-        let is_permanent = |_e: &_| false;
-
-        let mut key_marker = None;
-        let mut version_id_marker = None;
-        let mut versions_and_deletes = Vec::new();
-
-        loop {
-            let response = backoff::retry(
-                || async {
-                    self.client
-                        .list_object_versions()
-                        .bucket(self.bucket_name.clone())
-                        .set_prefix(prefix.clone())
-                        .set_key_marker(key_marker.clone())
-                        .set_version_id_marker(version_id_marker.clone())
-                        .send()
-                        .await
-                        .map_err(|e| TimeTravelError::Other(e.into()))
-                },
-                is_permanent,
-                warn_threshold,
-                max_retries,
-                "listing object versions for time_travel_recover",
-                cancel,
-            )
-            .await
-            .ok_or_else(|| TimeTravelError::Cancelled)
-            .and_then(|x| x)?;
-
-            tracing::trace!(
-                "  Got List response version_id_marker={:?}, key_marker={:?}",
-                response.version_id_marker,
-                response.key_marker
-            );
-            let versions = response
-                .versions
-                .unwrap_or_default()
-                .into_iter()
-                .map(VerOrDelete::from_version);
-            let deletes = response
-                .delete_markers
-                .unwrap_or_default()
-                .into_iter()
-                .map(VerOrDelete::from_delete_marker);
-            itertools::process_results(versions.chain(deletes), |n_vds| {
-                versions_and_deletes.extend(n_vds)
-            })
-            .map_err(TimeTravelError::Other)?;
-            fn none_if_empty(v: Option<String>) -> Option<String> {
-                v.filter(|v| !v.is_empty())
-            }
-            version_id_marker = none_if_empty(response.next_version_id_marker);
-            key_marker = none_if_empty(response.next_key_marker);
-            if version_id_marker.is_none() {
-                // The final response is not supposed to be truncated
-                if response.is_truncated.unwrap_or_default() {
-                    return Err(TimeTravelError::Other(anyhow::anyhow!(
-                        "Received truncated ListObjectVersions response for prefix={prefix:?}"
-                    )));
-                }
-                break;
-            }
-            // Limit the number of versions deletions, mostly so that we don't
-            // keep requesting forever if the list is too long, as we'd put the
-            // list in RAM.
-            // Building a list of 100k entries that reaches the limit roughly takes
-            // 40 seconds, and roughly corresponds to tenants of 2 TiB physical size.
-            const COMPLEXITY_LIMIT: usize = 100_000;
-            if versions_and_deletes.len() >= COMPLEXITY_LIMIT {
-                return Err(TimeTravelError::TooManyVersions);
-            }
-        }
-
-        tracing::info!(
-            "Built list for time travel with {} versions and deletions",
-            versions_and_deletes.len()
-        );
-
-        // Work on the list of references instead of the objects directly,
-        // otherwise we get lifetime errors in the sort_by_key call below.
-        let mut versions_and_deletes = versions_and_deletes.iter().collect::<Vec<_>>();
-
-        versions_and_deletes.sort_by_key(|vd| (&vd.key, &vd.last_modified));
-
-        let mut vds_for_key = HashMap::<_, Vec<_>>::new();
-
-        for vd in &versions_and_deletes {
-            let VerOrDelete {
-                version_id, key, ..
-            } = &vd;
-            if version_id == "null" {
-                return Err(TimeTravelError::Other(anyhow!("Received ListVersions response for key={key} with version_id='null', \
-                    indicating either disabled versioning, or legacy objects with null version id values")));
-            }
-            tracing::trace!(
-                "Parsing version key={key} version_id={version_id} kind={:?}",
-                vd.kind
-            );
-
-            vds_for_key.entry(key).or_default().push(vd);
-        }
-        for (key, versions) in vds_for_key {
-            let last_vd = versions.last().unwrap();
-            if last_vd.last_modified > done_if_after {
-                tracing::trace!("Key {key} has version later than done_if_after, skipping");
-                continue;
-            }
-            // the version we want to restore to.
-            let version_to_restore_to =
-                match versions.binary_search_by_key(&timestamp, |tpl| tpl.last_modified) {
-                    Ok(v) => v,
-                    Err(e) => e,
-                };
-            if version_to_restore_to == versions.len() {
-                tracing::trace!("Key {key} has no changes since timestamp, skipping");
-                continue;
-            }
-            let mut do_delete = false;
-            if version_to_restore_to == 0 {
-                // All versions more recent, so the key didn't exist at the specified time point.
-                tracing::trace!(
-                    "All {} versions more recent for {key}, deleting",
-                    versions.len()
-                );
-                do_delete = true;
-            } else {
-                match &versions[version_to_restore_to - 1] {
-                    VerOrDelete {
-                        kind: VerOrDeleteKind::Version,
-                        version_id,
-                        ..
-                    } => {
-                        tracing::trace!("Copying old version {version_id} for {key}...");
-                        // Restore the state to the last version by copying
-                        let source_id =
-                            format!("{}/{key}?versionId={version_id}", self.bucket_name);
-
-                        backoff::retry(
-                            || async {
-                                self.client
-                                    .copy_object()
-                                    .bucket(self.bucket_name.clone())
-                                    .key(key)
-                                    .copy_source(&source_id)
-                                    .send()
-                                    .await
-                                    .map_err(|e| TimeTravelError::Other(e.into()))
-                            },
-                            is_permanent,
-                            warn_threshold,
-                            max_retries,
-                            "copying object version for time_travel_recover",
-                            cancel,
-                        )
-                        .await
-                        .ok_or_else(|| TimeTravelError::Cancelled)
-                        .and_then(|x| x)?;
-                        tracing::info!(%version_id, %key, "Copied old version in S3");
-                    }
-                    VerOrDelete {
-                        kind: VerOrDeleteKind::DeleteMarker,
-                        ..
-                    } => {
-                        do_delete = true;
-                    }
-                }
-            };
-            if do_delete {
-                if matches!(last_vd.kind, VerOrDeleteKind::DeleteMarker) {
-                    // Key has since been deleted (but there was some history), no need to do anything
-                    tracing::trace!("Key {key} already deleted, skipping.");
-                } else {
-                    tracing::trace!("Deleting {key}...");
-
-                    let oid = ObjectIdentifier::builder()
-                        .key(key.to_owned())
-                        .build()
-                        .map_err(|e| TimeTravelError::Other(anyhow::Error::new(e)))?;
-                    self.delete_oids(kind, &[oid])
-                        .await
-                        .map_err(TimeTravelError::Other)?;
-                }
-            }
-        }
-        Ok(())
-    }
 }
 
 /// On drop (cancellation) count towards [`metrics::BucketMetrics::cancelled_waits`].
@@ -865,62 +651,6 @@ fn start_measuring_requests(
     })
 }
 
-// Save RAM and only store the needed data instead of the entire ObjectVersion/DeleteMarkerEntry
-struct VerOrDelete {
-    kind: VerOrDeleteKind,
-    last_modified: DateTime,
-    version_id: String,
-    key: String,
-}
-
-#[derive(Debug)]
-enum VerOrDeleteKind {
-    Version,
-    DeleteMarker,
-}
-
-impl VerOrDelete {
-    fn with_kind(
-        kind: VerOrDeleteKind,
-        last_modified: Option<DateTime>,
-        version_id: Option<String>,
-        key: Option<String>,
-    ) -> anyhow::Result<Self> {
-        let lvk = (last_modified, version_id, key);
-        let (Some(last_modified), Some(version_id), Some(key)) = lvk else {
-            anyhow::bail!(
-                "One (or more) of last_modified, key, and id is None. \
-            Is versioning enabled in the bucket? last_modified={:?}, version_id={:?}, key={:?}",
-                lvk.0,
-                lvk.1,
-                lvk.2,
-            );
-        };
-        Ok(Self {
-            kind,
-            last_modified,
-            version_id,
-            key,
-        })
-    }
-    fn from_version(v: ObjectVersion) -> anyhow::Result<Self> {
-        Self::with_kind(
-            VerOrDeleteKind::Version,
-            v.last_modified,
-            v.version_id,
-            v.key,
-        )
-    }
-    fn from_delete_marker(v: DeleteMarkerEntry) -> anyhow::Result<Self> {
-        Self::with_kind(
-            VerOrDeleteKind::DeleteMarker,
-            v.last_modified,
-            v.version_id,
-            v.key,
-        )
-    }
-}
-
 #[cfg(test)]
 mod tests {
     use camino::Utf8Path;

libs/remote_storage/src/s3_bucket/metrics.rs

@@ -12,7 +12,6 @@ pub(crate) enum RequestKind {
     Delete = 2,
     List = 3,
     Copy = 4,
-    TimeTravel = 5,
 }
 
 use RequestKind::*;
@@ -25,7 +24,6 @@ impl RequestKind {
             Delete => "delete_object",
             List => "list_objects",
             Copy => "copy_object",
-            TimeTravel => "time_travel_recover",
         }
     }
     const fn as_index(&self) -> usize {
@@ -33,7 +31,7 @@ impl RequestKind {
     }
 }
 
-pub(super) struct RequestTyped<C>([C; 6]);
+pub(super) struct RequestTyped<C>([C; 5]);
 
 impl<C> RequestTyped<C> {
     pub(super) fn get(&self, kind: RequestKind) -> &C {
@@ -42,8 +40,8 @@ impl<C> RequestTyped<C> {
 
     fn build_with(mut f: impl FnMut(RequestKind) -> C) -> Self {
         use RequestKind::*;
-        let mut it = [Get, Put, Delete, List, Copy, TimeTravel].into_iter();
-        let arr = std::array::from_fn::<C, 6, _>(|index| {
+        let mut it = [Get, Put, Delete, List, Copy].into_iter();
+        let arr = std::array::from_fn::<C, 5, _>(|index| {
             let next = it.next().unwrap();
             assert_eq!(index, next.as_index());
             f(next)

libs/remote_storage/src/simulate_failures.rs

@@ -3,19 +3,16 @@
 //! testing purposes.
 use bytes::Bytes;
 use futures::stream::Stream;
+use std::collections::hash_map::Entry;
 use std::collections::HashMap;
 use std::sync::Mutex;
-use std::time::SystemTime;
-use std::{collections::hash_map::Entry, sync::Arc};
-use tokio_util::sync::CancellationToken;
 
 use crate::{
-    Download, DownloadError, GenericRemoteStorage, Listing, ListingMode, RemotePath, RemoteStorage,
-    StorageMetadata, TimeTravelError,
+    Download, DownloadError, Listing, ListingMode, RemotePath, RemoteStorage, StorageMetadata,
 };
 
 pub struct UnreliableWrapper {
-    inner: GenericRemoteStorage<Arc<VoidStorage>>,
+    inner: crate::GenericRemoteStorage,
 
     // This many attempts of each operation will fail, then we let it succeed.
     attempts_to_fail: u64,
@@ -32,21 +29,11 @@ enum RemoteOp {
     Download(RemotePath),
     Delete(RemotePath),
     DeleteObjects(Vec<RemotePath>),
-    TimeTravelRecover(Option<RemotePath>),
 }
 
 impl UnreliableWrapper {
     pub fn new(inner: crate::GenericRemoteStorage, attempts_to_fail: u64) -> Self {
         assert!(attempts_to_fail > 0);
-        let inner = match inner {
-            GenericRemoteStorage::AwsS3(s) => GenericRemoteStorage::AwsS3(s),
-            GenericRemoteStorage::AzureBlob(s) => GenericRemoteStorage::AzureBlob(s),
-            GenericRemoteStorage::LocalFs(s) => GenericRemoteStorage::LocalFs(s),
-            // We could also make this a no-op, as in, extract the inner of the passed generic remote storage
-            GenericRemoteStorage::Unreliable(_s) => {
-                panic!("Can't wrap unreliable wrapper unreliably")
-            }
-        };
         UnreliableWrapper {
             inner,
             attempts_to_fail,
@@ -97,9 +84,7 @@ impl UnreliableWrapper {
     }
 }
 
-// We never construct this, so the type is not important, just has to not be UnreliableWrapper and impl RemoteStorage.
-type VoidStorage = crate::LocalFs;
-
+#[async_trait::async_trait]
 impl RemoteStorage for UnreliableWrapper {
     async fn list_prefixes(
         &self,
@@ -184,18 +169,4 @@ impl RemoteStorage for UnreliableWrapper {
         self.attempt(RemoteOp::Upload(to.clone()))?;
         self.inner.copy_object(from, to).await
     }
-
-    async fn time_travel_recover(
-        &self,
-        prefix: Option<&RemotePath>,
-        timestamp: SystemTime,
-        done_if_after: SystemTime,
-        cancel: &CancellationToken,
-    ) -> Result<(), TimeTravelError> {
-        self.attempt(RemoteOp::TimeTravelRecover(prefix.map(|p| p.to_owned())))
-            .map_err(|e| TimeTravelError::Other(anyhow::Error::new(e)))?;
-        self.inner
-            .time_travel_recover(prefix, timestamp, done_if_after, cancel)
-            .await
-    }
 }

libs/remote_storage/tests/test_real_s3.rs

@@ -1,21 +1,15 @@
+use std::collections::HashSet;
 use std::env;
-use std::fmt::{Debug, Display};
 use std::num::NonZeroUsize;
 use std::ops::ControlFlow;
 use std::sync::Arc;
-use std::time::{Duration, UNIX_EPOCH};
-use std::{collections::HashSet, time::SystemTime};
+use std::time::UNIX_EPOCH;
 
-use crate::common::{download_to_vec, upload_stream};
 use anyhow::Context;
-use camino::Utf8Path;
-use futures_util::Future;
 use remote_storage::{
     GenericRemoteStorage, RemotePath, RemoteStorageConfig, RemoteStorageKind, S3Config,
 };
-use test_context::test_context;
 use test_context::AsyncTestContext;
-use tokio_util::sync::CancellationToken;
 use tracing::info;
 
 mod common;
@@ -24,163 +18,11 @@ mod common;
 mod tests_s3;
 
 use common::{cleanup, ensure_logging_ready, upload_remote_data, upload_simple_remote_data};
-use utils::backoff;
 
 const ENABLE_REAL_S3_REMOTE_STORAGE_ENV_VAR_NAME: &str = "ENABLE_REAL_S3_REMOTE_STORAGE";
 
 const BASE_PREFIX: &str = "test";
 
-#[test_context(MaybeEnabledStorage)]
-#[tokio::test]
-async fn s3_time_travel_recovery_works(ctx: &mut MaybeEnabledStorage) -> anyhow::Result<()> {
-    let ctx = match ctx {
-        MaybeEnabledStorage::Enabled(ctx) => ctx,
-        MaybeEnabledStorage::Disabled => return Ok(()),
-    };
-    // Our test depends on discrepancies in the clock between S3 and the environment the tests
-    // run in. Therefore, wait a little bit before and after. The alternative would be
-    // to take the time from S3 response headers.
-    const WAIT_TIME: Duration = Duration::from_millis(3_000);
-
-    async fn retry<T, O, F, E>(op: O) -> Result<T, E>
-    where
-        E: Display + Debug + 'static,
-        O: FnMut() -> F,
-        F: Future<Output = Result<T, E>>,
-    {
-        let warn_threshold = 3;
-        let max_retries = 10;
-        backoff::retry(
-            op,
-            |_e| false,
-            warn_threshold,
-            max_retries,
-            "test retry",
-            &CancellationToken::new(),
-        )
-        .await
-        .expect("never cancelled")
-    }
-
-    async fn time_point() -> SystemTime {
-        tokio::time::sleep(WAIT_TIME).await;
-        let ret = SystemTime::now();
-        tokio::time::sleep(WAIT_TIME).await;
-        ret
-    }
-
-    async fn list_files(client: &Arc<GenericRemoteStorage>) -> anyhow::Result<HashSet<RemotePath>> {
-        Ok(retry(|| client.list_files(None))
-            .await
-            .context("list root files failure")?
-            .into_iter()
-            .collect::<HashSet<_>>())
-    }
-
-    let cancel = CancellationToken::new();
-
-    let path1 = RemotePath::new(Utf8Path::new(format!("{}/path1", ctx.base_prefix).as_str()))
-        .with_context(|| "RemotePath conversion")?;
-
-    let path2 = RemotePath::new(Utf8Path::new(format!("{}/path2", ctx.base_prefix).as_str()))
-        .with_context(|| "RemotePath conversion")?;
-
-    let path3 = RemotePath::new(Utf8Path::new(format!("{}/path3", ctx.base_prefix).as_str()))
-        .with_context(|| "RemotePath conversion")?;
-
-    retry(|| {
-        let (data, len) = upload_stream("remote blob data1".as_bytes().into());
-        ctx.client.upload(data, len, &path1, None)
-    })
-    .await?;
-
-    let t0_files = list_files(&ctx.client).await?;
-    let t0 = time_point().await;
-    println!("at t0: {t0_files:?}");
-
-    let old_data = "remote blob data2";
-
-    retry(|| {
-        let (data, len) = upload_stream(old_data.as_bytes().into());
-        ctx.client.upload(data, len, &path2, None)
-    })
-    .await?;
-
-    let t1_files = list_files(&ctx.client).await?;
-    let t1 = time_point().await;
-    println!("at t1: {t1_files:?}");
-
-    // A little check to ensure that our clock is not too far off from the S3 clock
-    {
-        let dl = retry(|| ctx.client.download(&path2)).await?;
-        let last_modified = dl.last_modified.unwrap();
-        let half_wt = WAIT_TIME.mul_f32(0.5);
-        let t0_hwt = t0 + half_wt;
-        let t1_hwt = t1 - half_wt;
-        if !(t0_hwt..=t1_hwt).contains(&last_modified) {
-            panic!("last_modified={last_modified:?} is not between t0_hwt={t0_hwt:?} and t1_hwt={t1_hwt:?}. \
-                This likely means a large lock discrepancy between S3 and the local clock.");
-        }
-    }
-
-    retry(|| {
-        let (data, len) = upload_stream("remote blob data3".as_bytes().into());
-        ctx.client.upload(data, len, &path3, None)
-    })
-    .await?;
-
-    let new_data = "new remote blob data2";
-
-    retry(|| {
-        let (data, len) = upload_stream(new_data.as_bytes().into());
-        ctx.client.upload(data, len, &path2, None)
-    })
-    .await?;
-
-    retry(|| ctx.client.delete(&path1)).await?;
-    let t2_files = list_files(&ctx.client).await?;
-    let t2 = time_point().await;
-    println!("at t2: {t2_files:?}");
-
-    // No changes after recovery to t2 (no-op)
-    let t_final = time_point().await;
-    ctx.client
-        .time_travel_recover(None, t2, t_final, &cancel)
-        .await?;
-    let t2_files_recovered = list_files(&ctx.client).await?;
-    println!("after recovery to t2: {t2_files_recovered:?}");
-    assert_eq!(t2_files, t2_files_recovered);
-    let path2_recovered_t2 = download_to_vec(ctx.client.download(&path2).await?).await?;
-    assert_eq!(path2_recovered_t2, new_data.as_bytes());
-
-    // after recovery to t1: path1 is back, path2 has the old content
-    let t_final = time_point().await;
-    ctx.client
-        .time_travel_recover(None, t1, t_final, &cancel)
-        .await?;
-    let t1_files_recovered = list_files(&ctx.client).await?;
-    println!("after recovery to t1: {t1_files_recovered:?}");
-    assert_eq!(t1_files, t1_files_recovered);
-    let path2_recovered_t1 = download_to_vec(ctx.client.download(&path2).await?).await?;
-    assert_eq!(path2_recovered_t1, old_data.as_bytes());
-
-    // after recovery to t0: everything is gone except for path1
-    let t_final = time_point().await;
-    ctx.client
-        .time_travel_recover(None, t0, t_final, &cancel)
-        .await?;
-    let t0_files_recovered = list_files(&ctx.client).await?;
-    println!("after recovery to t0: {t0_files_recovered:?}");
-    assert_eq!(t0_files, t0_files_recovered);
-
-    // cleanup
-
-    let paths = &[path1, path2, path3];
-    retry(|| ctx.client.delete_objects(paths)).await?;
-
-    Ok(())
-}
-
 struct EnabledS3 {
     client: Arc<GenericRemoteStorage>,
     base_prefix: &'static str,

libs/utils/src/auth.rs

@@ -127,10 +127,6 @@ impl JwtAuth {
         Ok(Self::new(decoding_keys))
     }
 
-    pub fn from_key(key: String) -> Result<Self> {
-        Ok(Self::new(vec![DecodingKey::from_ed_pem(key.as_bytes())?]))
-    }
-
     /// Attempt to decode the token with the internal decoding keys.
     ///
     /// The function tries the stored decoding keys in succession,

libs/utils/src/backoff.rs

@@ -37,53 +37,69 @@ pub fn exponential_backoff_duration_seconds(n: u32, base_increment: f64, max_sec
     }
 }
 
-/// Retries passed operation until one of the following conditions are met:
-/// - encountered error is considered as permanent (non-retryable)
-/// - retries have been exhausted
-/// - cancellation token has been cancelled
-///
-/// `is_permanent` closure should be used to provide distinction between permanent/non-permanent
-/// errors. When attempts cross `warn_threshold` function starts to emit log warnings.
+/// Configure cancellation for a retried operation: when to cancel (the token), and
+/// what kind of error to return on cancellation
+pub struct Cancel<E, CF>
+where
+    E: Display + Debug + 'static,
+    CF: Fn() -> E,
+{
+    token: CancellationToken,
+    on_cancel: CF,
+}
+
+impl<E, CF> Cancel<E, CF>
+where
+    E: Display + Debug + 'static,
+    CF: Fn() -> E,
+{
+    pub fn new(token: CancellationToken, on_cancel: CF) -> Self {
+        Self { token, on_cancel }
+    }
+}
+
+/// retries passed operation until one of the following conditions are met:
+/// Encountered error is considered as permanent (non-retryable)
+/// Retries have been exhausted.
+/// `is_permanent` closure should be used to provide distinction between permanent/non-permanent errors
+/// When attempts cross `warn_threshold` function starts to emit log warnings.
 /// `description` argument is added to log messages. Its value should identify the `op` is doing
-/// `cancel` cancels new attempts and the backoff sleep.
-///
-/// If attempts fail, they are being logged with `{:#}` which works for anyhow, but does not work
-/// for any other error type. Final failed attempt is logged with `{:?}`.
-///
-/// Returns `None` if cancellation was noticed during backoff or the terminal result.
-pub async fn retry<T, O, F, E>(
+/// `cancel` argument is required: any time we are looping on retry, we should be using a CancellationToken
+/// to drop out promptly on shutdown.
+pub async fn retry<T, O, F, E, CF>(
     mut op: O,
     is_permanent: impl Fn(&E) -> bool,
     warn_threshold: u32,
     max_retries: u32,
     description: &str,
-    cancel: &CancellationToken,
-) -> Option<Result<T, E>>
+    cancel: Cancel<E, CF>,
+) -> Result<T, E>
 where
     // Not std::error::Error because anyhow::Error doesnt implement it.
     // For context see https://github.com/dtolnay/anyhow/issues/63
     E: Display + Debug + 'static,
     O: FnMut() -> F,
     F: Future<Output = Result<T, E>>,
+    CF: Fn() -> E,
 {
     let mut attempts = 0;
     loop {
-        if cancel.is_cancelled() {
-            return None;
+        if cancel.token.is_cancelled() {
+            return Err((cancel.on_cancel)());
         }
 
         let result = op().await;
-        match &result {
+        match result {
             Ok(_) => {
                 if attempts > 0 {
                     tracing::info!("{description} succeeded after {attempts} retries");
                 }
-                return Some(result);
+                return result;
             }
 
             // These are "permanent" errors that should not be retried.
-            Err(e) if is_permanent(e) => {
-                return Some(result);
+            Err(ref e) if is_permanent(e) => {
+                return result;
             }
             // Assume that any other failure might be transient, and the operation might
             // succeed if we just keep trying.
@@ -93,12 +109,12 @@ where
             Err(err) if attempts < max_retries => {
                 tracing::warn!("{description} failed, will retry (attempt {attempts}): {err:#}");
             }
-            Err(err) => {
+            Err(ref err) => {
                 // Operation failed `max_attempts` times. Time to give up.
                 tracing::warn!(
                     "{description} still failed after {attempts} retries, giving up: {err:?}"
                 );
-                return Some(result);
+                return result;
             }
         }
         // sleep and retry
@@ -106,7 +122,7 @@ where
             attempts,
             DEFAULT_BASE_BACKOFF_SECONDS,
             DEFAULT_MAX_BACKOFF_SECONDS,
-            cancel,
+            &cancel.token,
         )
         .await;
         attempts += 1;
@@ -115,10 +131,12 @@ where
 
 #[cfg(test)]
 mod tests {
-    use super::*;
     use std::io;
+
     use tokio::sync::Mutex;
 
+    use super::*;
+
     #[test]
     fn backoff_defaults_produce_growing_backoff_sequence() {
         let mut current_backoff_value = None;
@@ -148,7 +166,7 @@ mod tests {
     #[tokio::test(start_paused = true)]
     async fn retry_always_error() {
         let count = Mutex::new(0);
-        retry(
+        let err_result = retry(
             || async {
                 *count.lock().await += 1;
                 Result::<(), io::Error>::Err(io::Error::from(io::ErrorKind::Other))
@@ -157,11 +175,11 @@ mod tests {
             1,
             1,
             "work",
-            &CancellationToken::new(),
+            Cancel::new(CancellationToken::new(), || -> io::Error { unreachable!() }),
         )
-        .await
-        .expect("not cancelled")
-        .expect_err("it can only fail");
+        .await;
+
+        assert!(err_result.is_err());
 
         assert_eq!(*count.lock().await, 2);
     }
@@ -183,11 +201,10 @@ mod tests {
             2,
             2,
             "work",
-            &CancellationToken::new(),
+            Cancel::new(CancellationToken::new(), || -> io::Error { unreachable!() }),
         )
         .await
-        .expect("not cancelled")
-        .expect("success on second try");
+        .unwrap();
     }
 
     #[tokio::test(start_paused = true)]
@@ -207,11 +224,10 @@ mod tests {
             2,
             2,
             "work",
-            &CancellationToken::new(),
+            Cancel::new(CancellationToken::new(), || -> io::Error { unreachable!() }),
         )
         .await
-        .expect("was not cancellation")
-        .expect_err("it was permanent error");
+        .unwrap_err();
 
         assert_eq!(*count.lock().await, 1);
     }

libs/utils/src/crashsafe.rs

@@ -112,55 +112,6 @@ pub async fn fsync_async(path: impl AsRef<Utf8Path>) -> Result<(), std::io::Erro
     tokio::fs::File::open(path.as_ref()).await?.sync_all().await
 }
 
-pub async fn fsync_async_opt(
-    path: impl AsRef<Utf8Path>,
-    do_fsync: bool,
-) -> Result<(), std::io::Error> {
-    if do_fsync {
-        fsync_async(path.as_ref()).await?;
-    }
-    Ok(())
-}
-
-/// Like postgres' durable_rename, renames file issuing fsyncs do make it
-/// durable. After return, file and rename are guaranteed to be persisted.
-///
-/// Unlike postgres, it only does fsyncs to 1) file to be renamed to make
-/// contents durable; 2) its directory entry to make rename durable 3) again to
-/// already renamed file, which is not required by standards but postgres does
-/// it, let's stick to that. Postgres additionally fsyncs newpath *before*
-/// rename if it exists to ensure that at least one of the files survives, but
-/// current callers don't need that.
-///
-/// virtual_file.rs has similar code, but it doesn't use vfs.
-///
-/// Useful links: <https://lwn.net/Articles/457667/>
-/// <https://www.postgresql.org/message-id/flat/56583BDD.9060302%402ndquadrant.com>
-/// <https://thunk.org/tytso/blog/2009/03/15/dont-fear-the-fsync/>
-pub async fn durable_rename(
-    old_path: impl AsRef<Utf8Path>,
-    new_path: impl AsRef<Utf8Path>,
-    do_fsync: bool,
-) -> io::Result<()> {
-    // first fsync the file
-    fsync_async_opt(old_path.as_ref(), do_fsync).await?;
-
-    // Time to do the real deal.
-    tokio::fs::rename(old_path.as_ref(), new_path.as_ref()).await?;
-
-    // Postgres'ish fsync of renamed file.
-    fsync_async_opt(new_path.as_ref(), do_fsync).await?;
-
-    // Now fsync the parent
-    let parent = match new_path.as_ref().parent() {
-        Some(p) => p,
-        None => Utf8Path::new("./"), // assume current dir if there is no parent
-    };
-    fsync_async_opt(parent, do_fsync).await?;
-
-    Ok(())
-}
-
 #[cfg(test)]
 mod tests {
 

libs/utils/src/http/error.rs

@@ -131,9 +131,7 @@ pub fn api_error_handler(api_error: ApiError) -> Response<Body> {
         ApiError::ResourceUnavailable(_) => info!("Error processing HTTP request: {api_error:#}"),
         ApiError::NotFound(_) => info!("Error processing HTTP request: {api_error:#}"),
         ApiError::InternalServerError(_) => error!("Error processing HTTP request: {api_error:?}"),
-        ApiError::ShuttingDown => info!("Shut down while processing HTTP request"),
-        ApiError::Timeout(_) => info!("Timeout while processing HTTP request: {api_error:#}"),
-        _ => info!("Error processing HTTP request: {api_error:#}"),
+        _ => error!("Error processing HTTP request: {api_error:#}"),
     }
 
     api_error.into_response()

libs/utils/src/sync/gate.rs

@@ -1,10 +1,4 @@
-use std::{
-    sync::{
-        atomic::{AtomicBool, Ordering},
-        Arc,
-    },
-    time::Duration,
-};
+use std::{sync::Arc, time::Duration};
 
 /// Gates are a concurrency helper, primarily used for implementing safe shutdown.
 ///
@@ -12,70 +6,62 @@ use std::{
 /// the resource calls `close()` when they want to ensure that all holders of guards
 /// have released them, and that no future guards will be issued.
 pub struct Gate {
-    inner: Arc<GateInner>,
+    /// Each caller of enter() takes one unit from the semaphore. In close(), we
+    /// take all the units to ensure all GateGuards are destroyed.
+    sem: Arc<tokio::sync::Semaphore>,
+
+    /// For observability only: a name that will be used to log warnings if a particular
+    /// gate is holding up shutdown
+    name: String,
 }
 
 impl std::fmt::Debug for Gate {
     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        f.debug_struct("Gate")
-            // use this for identification
-            .field("ptr", &Arc::as_ptr(&self.inner))
-            .field("inner", &self.inner)
-            .finish()
-    }
-}
-
-struct GateInner {
-    sem: tokio::sync::Semaphore,
-    closing: std::sync::atomic::AtomicBool,
-}
-
-impl std::fmt::Debug for GateInner {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        let avail = self.sem.available_permits();
-
-        let guards = u32::try_from(avail)
-            .ok()
-            // the sem only supports 32-bit ish amount, but lets play it safe
-            .and_then(|x| Gate::MAX_UNITS.checked_sub(x));
-
-        let closing = self.closing.load(Ordering::Relaxed);
-
-        if let Some(guards) = guards {
-            f.debug_struct("Gate")
-                .field("remaining_guards", &guards)
-                .field("closing", &closing)
-                .finish()
-        } else {
-            f.debug_struct("Gate")
-                .field("avail_permits", &avail)
-                .field("closing", &closing)
-                .finish()
-        }
+        write!(f, "Gate<{}>", self.name)
     }
 }
 
 /// RAII guard for a [`Gate`]: as long as this exists, calls to [`Gate::close`] will
 /// not complete.
 #[derive(Debug)]
-pub struct GateGuard {
-    // Record the span where the gate was entered, so that we can identify who was blocking Gate::close
-    span_at_enter: tracing::Span,
-    gate: Arc<GateInner>,
-}
+pub struct GateGuard(tokio::sync::OwnedSemaphorePermit);
 
-impl Drop for GateGuard {
-    fn drop(&mut self) {
-        if self.gate.closing.load(Ordering::Relaxed) {
-            self.span_at_enter.in_scope(
-                || tracing::info!(gate = ?Arc::as_ptr(&self.gate), "kept the gate from closing"),
-            );
+/// Observability helper: every `warn_period`, emit a log warning that we're still waiting on this gate
+async fn warn_if_stuck<Fut: std::future::Future>(
+    fut: Fut,
+    name: &str,
+    warn_period: std::time::Duration,
+) -> <Fut as std::future::Future>::Output {
+    let started = std::time::Instant::now();
+
+    let mut fut = std::pin::pin!(fut);
+
+    let mut warned = false;
+    let ret = loop {
+        match tokio::time::timeout(warn_period, &mut fut).await {
+            Ok(ret) => break ret,
+            Err(_) => {
+                tracing::warn!(
+                    gate = name,
+                    elapsed_ms = started.elapsed().as_millis(),
+                    "still waiting, taking longer than expected..."
+                );
+                warned = true;
+            }
         }
+    };
 
-        // when the permit was acquired, it was forgotten to allow us to manage it's lifecycle
-        // manually, so "return" the permit now.
-        self.gate.sem.add_permits(1);
+    // If we emitted a warning for slowness, also emit a message when we complete, so that
+    // someone debugging a shutdown can know for sure whether we have moved past this operation.
+    if warned {
+        tracing::info!(
+            gate = name,
+            elapsed_ms = started.elapsed().as_millis(),
+            "completed, after taking longer than expected"
+        )
     }
+
+    ret
 }
 
 #[derive(Debug)]
@@ -83,20 +69,16 @@ pub enum GateError {
     GateClosed,
 }
 
-impl Default for Gate {
-    fn default() -> Self {
-        Self {
-            inner: Arc::new(GateInner {
-                sem: tokio::sync::Semaphore::new(Self::MAX_UNITS as usize),
-                closing: AtomicBool::new(false),
-            }),
-        }
-    }
-}
-
 impl Gate {
     const MAX_UNITS: u32 = u32::MAX;
 
+    pub fn new(name: String) -> Self {
+        Self {
+            sem: Arc::new(tokio::sync::Semaphore::new(Self::MAX_UNITS as usize)),
+            name,
+        }
+    }
+
     /// Acquire a guard that will prevent close() calls from completing. If close()
     /// was already called, this will return an error which should be interpreted
     /// as "shutting down".
@@ -106,23 +88,11 @@ impl Gate {
     /// to avoid blocking close() indefinitely: typically types that contain a Gate will
     /// also contain a CancellationToken.
     pub fn enter(&self) -> Result<GateGuard, GateError> {
-        let permit = self
-            .inner
-            .sem
-            .try_acquire()
-            .map_err(|_| GateError::GateClosed)?;
-
-        // we now have the permit, let's disable the normal raii functionality and leave
-        // "returning" the permit to our GateGuard::drop.
-        //
-        // this is done to avoid the need for multiple Arcs (one for semaphore, next for other
-        // fields).
-        permit.forget();
-
-        Ok(GateGuard {
-            span_at_enter: tracing::Span::current(),
-            gate: self.inner.clone(),
-        })
+        self.sem
+            .clone()
+            .try_acquire_owned()
+            .map(GateGuard)
+            .map_err(|_| GateError::GateClosed)
     }
 
     /// Types with a shutdown() method and a gate should call this method at the
@@ -132,88 +102,48 @@ impl Gate {
     /// important that the holders of such guards are respecting a CancellationToken which has
     /// been cancelled before entering this function.
     pub async fn close(&self) {
-        let started_at = std::time::Instant::now();
-        let mut do_close = std::pin::pin!(self.do_close());
-
-        let nag_after = Duration::from_secs(1);
-
-        let Err(_timeout) = tokio::time::timeout(nag_after, &mut do_close).await else {
-            return;
-        };
-
-        tracing::info!(
-            gate = ?self.as_ptr(),
-            elapsed_ms = started_at.elapsed().as_millis(),
-            "closing is taking longer than expected"
-        );
-
-        // close operation is not trying to be cancellation safe as pageserver does not need it.
-        //
-        // note: "closing" is not checked in Gate::enter -- it exists just for observability,
-        // dropping of GateGuard after this will log who they were.
-        self.inner.closing.store(true, Ordering::Relaxed);
-
-        do_close.await;
-
-        tracing::info!(
-            gate = ?self.as_ptr(),
-            elapsed_ms = started_at.elapsed().as_millis(),
-            "close completed"
-        );
-    }
-
-    /// Used as an identity of a gate. This identity will be resolved to something useful when
-    /// it's actually closed in a hopefully sensible `tracing::Span` which will describe it even
-    /// more.
-    ///
-    /// `GateGuard::drop` also logs this pointer when it has realized it has been keeping the gate
-    /// open for too long.
-    fn as_ptr(&self) -> *const GateInner {
-        Arc::as_ptr(&self.inner)
+        warn_if_stuck(self.do_close(), &self.name, Duration::from_millis(1000)).await
     }
 
     /// Check if [`Self::close()`] has finished waiting for all [`Self::enter()`] users to finish.  This
     /// is usually analoguous for "Did shutdown finish?" for types that include a Gate, whereas checking
     /// the CancellationToken on such types is analogous to "Did shutdown start?"
     pub fn close_complete(&self) -> bool {
-        self.inner.sem.is_closed()
+        self.sem.is_closed()
     }
 
-    #[tracing::instrument(level = tracing::Level::DEBUG, skip_all, fields(gate = ?self.as_ptr()))]
     async fn do_close(&self) {
-        tracing::debug!("Closing Gate...");
-
-        match self.inner.sem.acquire_many(Self::MAX_UNITS).await {
-            Ok(_permit) => {
+        tracing::debug!(gate = self.name, "Closing Gate...");
+        match self.sem.acquire_many(Self::MAX_UNITS).await {
+            Ok(_units) => {
                 // While holding all units, close the semaphore.  All subsequent calls to enter() will fail.
-                self.inner.sem.close();
+                self.sem.close();
             }
-            Err(_closed) => {
+            Err(_) => {
                 // Semaphore closed: we are the only function that can do this, so it indicates a double-call.
                 // This is legal.  Timeline::shutdown for example is not protected from being called more than
                 // once.
-                tracing::debug!("Double close")
+                tracing::debug!(gate = self.name, "Double close")
             }
         }
-        tracing::debug!("Closed Gate.")
+        tracing::debug!(gate = self.name, "Closed Gate.")
     }
 }
 
 #[cfg(test)]
 mod tests {
+    use futures::FutureExt;
+
     use super::*;
 
     #[tokio::test]
-    async fn close_unused() {
-        // Having taken no guards, we should not be blocked in close
-        let gate = Gate::default();
+    async fn test_idle_gate() {
+        // Having taken no gates, we should not be blocked in close
+        let gate = Gate::new("test".to_string());
         gate.close().await;
-    }
 
-    #[tokio::test]
-    async fn close_idle() {
         // If a guard is dropped before entering, close should not be blocked
-        let gate = Gate::default();
+        let gate = Gate::new("test".to_string());
         let guard = gate.enter().unwrap();
         drop(guard);
         gate.close().await;
@@ -222,30 +152,25 @@ mod tests {
         gate.enter().expect_err("enter should fail after close");
     }
 
-    #[tokio::test(start_paused = true)]
-    async fn close_busy_gate() {
-        let gate = Gate::default();
-        let forever = Duration::from_secs(24 * 7 * 365);
+    #[tokio::test]
+    async fn test_busy_gate() {
+        let gate = Gate::new("test".to_string());
 
-        let guard =
-            tracing::info_span!("i am holding back the gate").in_scope(|| gate.enter().unwrap());
+        let guard = gate.enter().unwrap();
 
         let mut close_fut = std::pin::pin!(gate.close());
 
-        // Close should be waiting for guards to drop
-        tokio::time::timeout(forever, &mut close_fut)
-            .await
-            .unwrap_err();
+        // Close should be blocked
+        assert!(close_fut.as_mut().now_or_never().is_none());
 
         // Attempting to enter() should fail, even though close isn't done yet.
         gate.enter()
             .expect_err("enter should fail after entering close");
 
-        // this will now log, which we cannot verify except manually
         drop(guard);
 
         // Guard is gone, close should finish
-        close_fut.await;
+        assert!(close_fut.as_mut().now_or_never().is_some());
 
         // Attempting to enter() is still forbidden
         gate.enter().expect_err("enter should fail finishing close");

libs/utils/src/sync/heavier_once_cell.rs

@@ -1,6 +1,6 @@
 use std::sync::{
     atomic::{AtomicUsize, Ordering},
-    Arc,
+    Arc, Mutex, MutexGuard,
 };
 use tokio::sync::Semaphore;
 
@@ -12,7 +12,7 @@ use tokio::sync::Semaphore;
 ///
 /// [`OwnedSemaphorePermit`]: tokio::sync::OwnedSemaphorePermit
 pub struct OnceCell<T> {
-    inner: tokio::sync::RwLock<Inner<T>>,
+    inner: Mutex<Inner<T>>,
     initializers: AtomicUsize,
 }
 
@@ -50,7 +50,7 @@ impl<T> OnceCell<T> {
         let sem = Semaphore::new(1);
         sem.close();
         Self {
-            inner: tokio::sync::RwLock::new(Inner {
+            inner: Mutex::new(Inner {
                 init_semaphore: Arc::new(sem),
                 value: Some(value),
             }),
@@ -61,18 +61,18 @@ impl<T> OnceCell<T> {
     /// Returns a guard to an existing initialized value, or uniquely initializes the value before
     /// returning the guard.
     ///
-    /// Initializing might wait on any existing [`GuardMut::take_and_deinit`] deinitialization.
+    /// Initializing might wait on any existing [`Guard::take_and_deinit`] deinitialization.
     ///
     /// Initialization is panic-safe and cancellation-safe.
-    pub async fn get_mut_or_init<F, Fut, E>(&self, factory: F) -> Result<GuardMut<'_, T>, E>
+    pub async fn get_or_init<F, Fut, E>(&self, factory: F) -> Result<Guard<'_, T>, E>
     where
         F: FnOnce(InitPermit) -> Fut,
         Fut: std::future::Future<Output = Result<(T, InitPermit), E>>,
     {
         let sem = {
-            let guard = self.inner.write().await;
+            let guard = self.inner.lock().unwrap();
             if guard.value.is_some() {
-                return Ok(GuardMut(guard));
+                return Ok(Guard(guard));
             }
             guard.init_semaphore.clone()
         };
@@ -88,72 +88,29 @@ impl<T> OnceCell<T> {
                 let permit = InitPermit(permit);
                 let (value, _permit) = factory(permit).await?;
 
-                let guard = self.inner.write().await;
+                let guard = self.inner.lock().unwrap();
 
                 Ok(Self::set0(value, guard))
             }
             Err(_closed) => {
-                let guard = self.inner.write().await;
+                let guard = self.inner.lock().unwrap();
                 assert!(
                     guard.value.is_some(),
                     "semaphore got closed, must be initialized"
                 );
-                return Ok(GuardMut(guard));
+                return Ok(Guard(guard));
             }
         }
     }
 
-    /// Returns a guard to an existing initialized value, or uniquely initializes the value before
-    /// returning the guard.
-    ///
-    /// Initialization is panic-safe and cancellation-safe.
-    pub async fn get_or_init<F, Fut, E>(&self, factory: F) -> Result<GuardRef<'_, T>, E>
-    where
-        F: FnOnce(InitPermit) -> Fut,
-        Fut: std::future::Future<Output = Result<(T, InitPermit), E>>,
-    {
-        let sem = {
-            let guard = self.inner.read().await;
-            if guard.value.is_some() {
-                return Ok(GuardRef(guard));
-            }
-            guard.init_semaphore.clone()
-        };
-
-        let permit = {
-            // increment the count for the duration of queued
-            let _guard = CountWaitingInitializers::start(self);
-            sem.acquire_owned().await
-        };
-
-        match permit {
-            Ok(permit) => {
-                let permit = InitPermit(permit);
-                let (value, _permit) = factory(permit).await?;
-
-                let guard = self.inner.write().await;
-
-                Ok(Self::set0(value, guard).downgrade())
-            }
-            Err(_closed) => {
-                let guard = self.inner.read().await;
-                assert!(
-                    guard.value.is_some(),
-                    "semaphore got closed, must be initialized"
-                );
-                return Ok(GuardRef(guard));
-            }
-        }
-    }
-
-    /// Assuming a permit is held after previous call to [`GuardMut::take_and_deinit`], it can be used
+    /// Assuming a permit is held after previous call to [`Guard::take_and_deinit`], it can be used
     /// to complete initializing the inner value.
     ///
     /// # Panics
     ///
     /// If the inner has already been initialized.
-    pub async fn set(&self, value: T, _permit: InitPermit) -> GuardMut<'_, T> {
-        let guard = self.inner.write().await;
+    pub fn set(&self, value: T, _permit: InitPermit) -> Guard<'_, T> {
+        let guard = self.inner.lock().unwrap();
 
         // cannot assert that this permit is for self.inner.semaphore, but we can assert it cannot
         // give more permits right now.
@@ -165,31 +122,21 @@ impl<T> OnceCell<T> {
         Self::set0(value, guard)
     }
 
-    fn set0(value: T, mut guard: tokio::sync::RwLockWriteGuard<'_, Inner<T>>) -> GuardMut<'_, T> {
+    fn set0(value: T, mut guard: std::sync::MutexGuard<'_, Inner<T>>) -> Guard<'_, T> {
         if guard.value.is_some() {
             drop(guard);
             unreachable!("we won permit, must not be initialized");
         }
         guard.value = Some(value);
         guard.init_semaphore.close();
-        GuardMut(guard)
+        Guard(guard)
     }
 
     /// Returns a guard to an existing initialized value, if any.
-    pub async fn get_mut(&self) -> Option<GuardMut<'_, T>> {
-        let guard = self.inner.write().await;
+    pub fn get(&self) -> Option<Guard<'_, T>> {
+        let guard = self.inner.lock().unwrap();
         if guard.value.is_some() {
-            Some(GuardMut(guard))
-        } else {
-            None
-        }
-    }
-
-    /// Returns a guard to an existing initialized value, if any.
-    pub async fn get(&self) -> Option<GuardRef<'_, T>> {
-        let guard = self.inner.read().await;
-        if guard.value.is_some() {
-            Some(GuardRef(guard))
+            Some(Guard(guard))
         } else {
             None
         }
@@ -221,9 +168,9 @@ impl<'a, T> Drop for CountWaitingInitializers<'a, T> {
 /// Uninteresting guard object to allow short-lived access to inspect or clone the held,
 /// initialized value.
 #[derive(Debug)]
-pub struct GuardMut<'a, T>(tokio::sync::RwLockWriteGuard<'a, Inner<T>>);
+pub struct Guard<'a, T>(MutexGuard<'a, Inner<T>>);
 
-impl<T> std::ops::Deref for GuardMut<'_, T> {
+impl<T> std::ops::Deref for Guard<'_, T> {
     type Target = T;
 
     fn deref(&self) -> &Self::Target {
@@ -234,7 +181,7 @@ impl<T> std::ops::Deref for GuardMut<'_, T> {
     }
 }
 
-impl<T> std::ops::DerefMut for GuardMut<'_, T> {
+impl<T> std::ops::DerefMut for Guard<'_, T> {
     fn deref_mut(&mut self) -> &mut Self::Target {
         self.0
             .value
@@ -243,7 +190,7 @@ impl<T> std::ops::DerefMut for GuardMut<'_, T> {
     }
 }
 
-impl<'a, T> GuardMut<'a, T> {
+impl<'a, T> Guard<'a, T> {
     /// Take the current value, and a new permit for it's deinitialization.
     ///
     /// The permit will be on a semaphore part of the new internal value, and any following
@@ -261,24 +208,6 @@ impl<'a, T> GuardMut<'a, T> {
             .map(|v| (v, InitPermit(permit)))
             .expect("guard is not created unless value has been initialized")
     }
-
-    pub fn downgrade(self) -> GuardRef<'a, T> {
-        GuardRef(self.0.downgrade())
-    }
-}
-
-#[derive(Debug)]
-pub struct GuardRef<'a, T>(tokio::sync::RwLockReadGuard<'a, Inner<T>>);
-
-impl<T> std::ops::Deref for GuardRef<'_, T> {
-    type Target = T;
-
-    fn deref(&self) -> &Self::Target {
-        self.0
-            .value
-            .as_ref()
-            .expect("guard is not created unless value has been initialized")
-    }
 }
 
 /// Type held by OnceCell (de)initializing task.
@@ -319,7 +248,7 @@ mod tests {
                     barrier.wait().await;
                     let won = {
                         let g = cell
-                            .get_mut_or_init(|permit| {
+                            .get_or_init(|permit| {
                                 counters.factory_got_to_run.fetch_add(1, Ordering::Relaxed);
                                 async {
                                     counters.future_polled.fetch_add(1, Ordering::Relaxed);
@@ -366,11 +295,7 @@ mod tests {
             let cell = cell.clone();
             let deinitialization_started = deinitialization_started.clone();
             async move {
-                let (answer, _permit) = cell
-                    .get_mut()
-                    .await
-                    .expect("initialized to value")
-                    .take_and_deinit();
+                let (answer, _permit) = cell.get().expect("initialized to value").take_and_deinit();
                 assert_eq!(answer, initial);
 
                 deinitialization_started.wait().await;
@@ -381,7 +306,7 @@ mod tests {
         deinitialization_started.wait().await;
 
         let started_at = tokio::time::Instant::now();
-        cell.get_mut_or_init(|permit| async { Ok::<_, Infallible>((reinit, permit)) })
+        cell.get_or_init(|permit| async { Ok::<_, Infallible>((reinit, permit)) })
             .await
             .unwrap();
 
@@ -393,21 +318,21 @@ mod tests {
 
         jh.await.unwrap();
 
-        assert_eq!(*cell.get_mut().await.unwrap(), reinit);
+        assert_eq!(*cell.get().unwrap(), reinit);
     }
 
-    #[tokio::test]
-    async fn reinit_with_deinit_permit() {
+    #[test]
+    fn reinit_with_deinit_permit() {
         let cell = Arc::new(OnceCell::new(42));
 
-        let (mol, permit) = cell.get_mut().await.unwrap().take_and_deinit();
-        cell.set(5, permit).await;
-        assert_eq!(*cell.get_mut().await.unwrap(), 5);
+        let (mol, permit) = cell.get().unwrap().take_and_deinit();
+        cell.set(5, permit);
+        assert_eq!(*cell.get().unwrap(), 5);
 
-        let (five, permit) = cell.get_mut().await.unwrap().take_and_deinit();
+        let (five, permit) = cell.get().unwrap().take_and_deinit();
         assert_eq!(5, five);
-        cell.set(mol, permit).await;
-        assert_eq!(*cell.get_mut().await.unwrap(), 42);
+        cell.set(mol, permit);
+        assert_eq!(*cell.get().unwrap(), 42);
     }
 
     #[tokio::test]
@@ -415,13 +340,13 @@ mod tests {
         let cell = OnceCell::default();
 
         for _ in 0..10 {
-            cell.get_mut_or_init(|_permit| async { Err("whatever error") })
+            cell.get_or_init(|_permit| async { Err("whatever error") })
                 .await
                 .unwrap_err();
         }
 
         let g = cell
-            .get_mut_or_init(|permit| async { Ok::<_, Infallible>(("finally success", permit)) })
+            .get_or_init(|permit| async { Ok::<_, Infallible>(("finally success", permit)) })
             .await
             .unwrap();
         assert_eq!(*g, "finally success");
@@ -433,7 +358,7 @@ mod tests {
 
         let barrier = tokio::sync::Barrier::new(2);
 
-        let initializer = cell.get_mut_or_init(|permit| async {
+        let initializer = cell.get_or_init(|permit| async {
             barrier.wait().await;
             futures::future::pending::<()>().await;
 
@@ -447,10 +372,10 @@ mod tests {
 
         // now initializer is dropped
 
-        assert!(cell.get_mut().await.is_none());
+        assert!(cell.get().is_none());
 
         let g = cell
-            .get_mut_or_init(|permit| async { Ok::<_, Infallible>(("now initialized", permit)) })
+            .get_or_init(|permit| async { Ok::<_, Infallible>(("now initialized", permit)) })
             .await
             .unwrap();
         assert_eq!(*g, "now initialized");

libs/utils/src/tracing_span_assert.rs

@@ -20,13 +20,13 @@
 //!
 //! // Then, in the main code:
 //!
-//! let span = tracing::info_span!("TestSpan", tenant_id = 1);
+//! let span = tracing::info_span!("TestSpan", test_id = 1);
 //! let _guard = span.enter();
 //!
 //! // ... down the call stack
 //!
-//! use utils::tracing_span_assert::{check_fields_present, ConstExtractor};
-//! let extractor = ConstExtractor::new("tenant_id");
+//! use utils::tracing_span_assert::{check_fields_present, MultiNameExtractor};
+//! let extractor = MultiNameExtractor::new("TestExtractor", ["test", "test_id"]);
 //! if let Err(missing) = check_fields_present!([&extractor]) {
 //!    // if you copypaste this to a custom assert method, remember to add #[track_caller]
 //!    // to get the "user" code location for the panic.
@@ -45,26 +45,27 @@ pub enum ExtractionResult {
 }
 
 pub trait Extractor: Send + Sync + std::fmt::Debug {
-    fn id(&self) -> &str;
+    fn name(&self) -> &str;
     fn extract(&self, fields: &tracing::field::FieldSet) -> ExtractionResult;
 }
 
 #[derive(Debug)]
-pub struct ConstExtractor {
-    field_name: &'static str,
+pub struct MultiNameExtractor<const L: usize> {
+    name: &'static str,
+    field_names: [&'static str; L],
 }
 
-impl ConstExtractor {
-    pub const fn new(field_name: &'static str) -> ConstExtractor {
-        ConstExtractor { field_name }
+impl<const L: usize> MultiNameExtractor<L> {
+    pub fn new(name: &'static str, field_names: [&'static str; L]) -> MultiNameExtractor<L> {
+        MultiNameExtractor { name, field_names }
     }
 }
-impl Extractor for ConstExtractor {
-    fn id(&self) -> &str {
-        self.field_name
+impl<const L: usize> Extractor for MultiNameExtractor<L> {
+    fn name(&self) -> &str {
+        self.name
     }
     fn extract(&self, fields: &tracing::field::FieldSet) -> ExtractionResult {
-        if fields.iter().any(|f| f.name() == self.field_name) {
+        if fields.iter().any(|f| self.field_names.contains(&f.name())) {
             ExtractionResult::Present
         } else {
             ExtractionResult::Absent
@@ -202,19 +203,19 @@ mod tests {
     }
     impl<'a> fmt::Debug for MemoryIdentity<'a> {
         fn fmt(&self, f: &mut fmt::Formatter<'_>) -> std::fmt::Result {
-            write!(f, "{:p}: {}", self.as_ptr(), self.0.id())
+            write!(f, "{:p}: {}", self.as_ptr(), self.0.name())
         }
     }
 
     struct Setup {
         _current_thread_subscriber_guard: tracing::subscriber::DefaultGuard,
-        tenant_extractor: ConstExtractor,
-        timeline_extractor: ConstExtractor,
+        tenant_extractor: MultiNameExtractor<2>,
+        timeline_extractor: MultiNameExtractor<2>,
     }
 
     fn setup_current_thread() -> Setup {
-        let tenant_extractor = ConstExtractor::new("tenant_id");
-        let timeline_extractor = ConstExtractor::new("timeline_id");
+        let tenant_extractor = MultiNameExtractor::new("TenantId", ["tenant_id", "tenant"]);
+        let timeline_extractor = MultiNameExtractor::new("TimelineId", ["timeline_id", "timeline"]);
 
         let registry = tracing_subscriber::registry()
             .with(tracing_subscriber::fmt::layer())
@@ -342,12 +343,12 @@ mod tests {
         let span = tracing::info_span!("foo", e = "some value");
         let _guard = span.enter();
 
-        let extractor = ConstExtractor::new("e");
+        let extractor = MultiNameExtractor::new("E", ["e"]);
         let res = check_fields_present0([&extractor]);
         assert!(matches!(res, Ok(Summary::Unconfigured)), "{res:?}");
 
         // similarly for a not found key
-        let extractor = ConstExtractor::new("foobar");
+        let extractor = MultiNameExtractor::new("F", ["foobar"]);
         let res = check_fields_present0([&extractor]);
         assert!(matches!(res, Ok(Summary::Unconfigured)), "{res:?}");
     }
@@ -367,14 +368,16 @@ mod tests {
         // normally this would work, but without any tracing-subscriber configured, both
         // check_field_present find nothing
         let _guard = subspan.enter();
-        let extractors: [&dyn Extractor; 2] =
-            [&ConstExtractor::new("e"), &ConstExtractor::new("f")];
+        let extractors: [&dyn Extractor; 2] = [
+            &MultiNameExtractor::new("E", ["e"]),
+            &MultiNameExtractor::new("F", ["f"]),
+        ];
 
         let res = check_fields_present0(extractors);
         assert!(matches!(res, Ok(Summary::Unconfigured)), "{res:?}");
 
         // similarly for a not found key
-        let extractor = ConstExtractor::new("g");
+        let extractor = MultiNameExtractor::new("G", ["g"]);
         let res = check_fields_present0([&extractor]);
         assert!(matches!(res, Ok(Summary::Unconfigured)), "{res:?}");
     }
@@ -407,7 +410,7 @@ mod tests {
         let span = tracing::info_span!("foo", e = "some value");
         let _guard = span.enter();
 
-        let extractors: [&dyn Extractor; 1] = [&ConstExtractor::new("e")];
+        let extractors: [&dyn Extractor; 1] = [&MultiNameExtractor::new("E", ["e"])];
 
         if span.is_disabled() {
             // the tests are running single threaded, or we got lucky and no other tests subscriber

pageserver/Cargo.toml

@@ -21,6 +21,7 @@ camino.workspace = true
 camino-tempfile.workspace = true
 chrono = { workspace = true, features = ["serde"] }
 clap = { workspace = true, features = ["string"] }
+close_fds.workspace = true
 const_format.workspace = true
 consumption_metrics.workspace = true
 crc32c.workspace = true
@@ -60,7 +61,6 @@ sync_wrapper.workspace = true
 tokio-tar.workspace = true
 thiserror.workspace = true
 tokio = { workspace = true, features = ["process", "sync", "fs", "rt", "io-util", "time"] }
-tokio-epoll-uring.workspace = true
 tokio-io-timeout.workspace = true
 tokio-postgres.workspace = true
 tokio-stream.workspace = true

pageserver/client/src/mgmt_api.rs

@@ -69,25 +69,6 @@ impl Client {
         resp.json().await.map_err(Error::ReceiveBody)
     }
 
-    /// Get an arbitrary path and returning a streaming Response.  This function is suitable
-    /// for pass-through/proxy use cases where we don't care what the response content looks
-    /// like.
-    ///
-    /// Use/add one of the properly typed methods below if you know aren't proxying, and
-    /// know what kind of response you expect.
-    pub async fn get_raw(&self, path: String) -> Result<reqwest::Response> {
-        debug_assert!(path.starts_with('/'));
-        let uri = format!("{}{}", self.mgmt_api_endpoint, path);
-
-        let req = self.client.request(Method::GET, uri);
-        let req = if let Some(value) = &self.authorization_header {
-            req.header(reqwest::header::AUTHORIZATION, value)
-        } else {
-            req
-        };
-        req.send().await.map_err(Error::ReceiveBody)
-    }
-
     pub async fn tenant_details(
         &self,
         tenant_shard_id: TenantShardId,
@@ -190,25 +171,6 @@ impl Client {
             .map_err(Error::ReceiveBody)
     }
 
-    /// The tenant deletion API can return 202 if deletion is incomplete, or
-    /// 404 if it is complete.  Callers are responsible for checking the status
-    /// code and retrying.  Error codes other than 404 will return Err().
-    pub async fn tenant_delete(&self, tenant_shard_id: TenantShardId) -> Result<StatusCode> {
-        let uri = format!("{}/v1/tenant/{tenant_shard_id}", self.mgmt_api_endpoint);
-
-        match self.request(Method::DELETE, &uri, ()).await {
-            Err(Error::ApiError(status_code, msg)) => {
-                if status_code == StatusCode::NOT_FOUND {
-                    Ok(StatusCode::NOT_FOUND)
-                } else {
-                    Err(Error::ApiError(status_code, msg))
-                }
-            }
-            Err(e) => Err(e),
-            Ok(response) => Ok(response.status()),
-        }
-    }
-
     pub async fn tenant_config(&self, req: &TenantConfigRequest) -> Result<()> {
         let uri = format!("{}/v1/tenant/config", self.mgmt_api_endpoint);
         self.request(Method::PUT, &uri, req).await?;
@@ -272,32 +234,6 @@ impl Client {
             .map_err(Error::ReceiveBody)
     }
 
-    /// The timeline deletion API can return 201 if deletion is incomplete, or
-    /// 403 if it is complete.  Callers are responsible for checking the status
-    /// code and retrying.  Error codes other than 403 will return Err().
-    pub async fn timeline_delete(
-        &self,
-        tenant_shard_id: TenantShardId,
-        timeline_id: TimelineId,
-    ) -> Result<StatusCode> {
-        let uri = format!(
-            "{}/v1/tenant/{tenant_shard_id}/timeline/{timeline_id}",
-            self.mgmt_api_endpoint
-        );
-
-        match self.request(Method::DELETE, &uri, ()).await {
-            Err(Error::ApiError(status_code, msg)) => {
-                if status_code == StatusCode::NOT_FOUND {
-                    Ok(StatusCode::NOT_FOUND)
-                } else {
-                    Err(Error::ApiError(status_code, msg))
-                }
-            }
-            Err(e) => Err(e),
-            Ok(response) => Ok(response.status()),
-        }
-    }
-
     pub async fn tenant_reset(&self, tenant_shard_id: TenantShardId) -> Result<()> {
         let uri = format!(
             "{}/v1/tenant/{}/reset",

pageserver/client/src/page_service.rs

@@ -156,8 +156,7 @@ impl PagestreamClient {
             PagestreamBeMessage::Error(e) => anyhow::bail!("Error: {:?}", e),
             PagestreamBeMessage::Exists(_)
             | PagestreamBeMessage::Nblocks(_)
-            | PagestreamBeMessage::DbSize(_)
-            | PagestreamBeMessage::GetSlruSegment(_) => {
+            | PagestreamBeMessage::DbSize(_) => {
                 anyhow::bail!(
                     "unexpected be message kind in response to getpage request: {}",
                     msg.kind()

pageserver/ctl/src/layer_map_analyzer.rs

@@ -18,7 +18,7 @@ use pageserver::tenant::block_io::FileBlockReader;
 use pageserver::tenant::disk_btree::{DiskBtreeReader, VisitDirection};
 use pageserver::tenant::storage_layer::delta_layer::{Summary, DELTA_KEY_SIZE};
 use pageserver::tenant::storage_layer::range_overlaps;
-use pageserver::virtual_file::{self, VirtualFile};
+use pageserver::virtual_file::VirtualFile;
 
 use utils::{bin_ser::BeSer, lsn::Lsn};
 
@@ -142,7 +142,7 @@ pub(crate) async fn main(cmd: &AnalyzeLayerMapCmd) -> Result<()> {
     let ctx = RequestContext::new(TaskKind::DebugTool, DownloadBehavior::Error);
 
     // Initialize virtual_file (file desriptor cache) and page cache which are needed to access layer persistent B-Tree.
-    pageserver::virtual_file::init(10, virtual_file::IoEngineKind::StdFs);
+    pageserver::virtual_file::init(10);
     pageserver::page_cache::init(100);
 
     let mut total_delta_layers = 0usize;

pageserver/ctl/src/layers.rs

@@ -59,7 +59,7 @@ pub(crate) enum LayerCmd {
 
 async fn read_delta_file(path: impl AsRef<Path>, ctx: &RequestContext) -> Result<()> {
     let path = Utf8Path::from_path(path.as_ref()).expect("non-Unicode path");
-    virtual_file::init(10, virtual_file::IoEngineKind::StdFs);
+    virtual_file::init(10);
     page_cache::init(100);
     let file = FileBlockReader::new(VirtualFile::open(path).await?);
     let summary_blk = file.read_blk(0, ctx).await?;
@@ -187,7 +187,7 @@ pub(crate) async fn main(cmd: &LayerCmd) -> Result<()> {
             new_tenant_id,
             new_timeline_id,
         } => {
-            pageserver::virtual_file::init(10, virtual_file::IoEngineKind::StdFs);
+            pageserver::virtual_file::init(10);
             pageserver::page_cache::init(100);
 
             let ctx = RequestContext::new(TaskKind::DebugTool, DownloadBehavior::Error);

pageserver/ctl/src/main.rs

@@ -123,7 +123,7 @@ fn read_pg_control_file(control_file_path: &Utf8Path) -> anyhow::Result<()> {
 
 async fn print_layerfile(path: &Utf8Path) -> anyhow::Result<()> {
     // Basic initialization of things that don't change after startup
-    virtual_file::init(10, virtual_file::IoEngineKind::StdFs);
+    virtual_file::init(10);
     page_cache::init(100);
     let ctx = RequestContext::new(TaskKind::DebugTool, DownloadBehavior::Error);
     dump_layerfile_from_path(path, true, &ctx).await

pageserver/pagebench/src/cmd/getpage_latest_lsn.rs

@@ -79,12 +79,6 @@ impl KeyRange {
     }
 }
 
-#[derive(PartialEq, Eq, Hash, Copy, Clone)]
-struct WorkerId {
-    timeline: TenantTimelineId,
-    num_client: usize, // from 0..args.num_clients
-}
-
 #[derive(serde::Serialize)]
 struct Output {
     total: request_stats::Output,
@@ -212,7 +206,7 @@ async fn main_impl(
 
     let live_stats = Arc::new(LiveStats::default());
 
-    let num_client_tasks = args.num_clients.get() * timelines.len();
+    let num_client_tasks = timelines.len();
     let num_live_stats_dump = 1;
     let num_work_sender_tasks = 1;
     let num_main_impl = 1;
@@ -241,25 +235,19 @@ async fn main_impl(
 
     let cancel = CancellationToken::new();
 
-    let mut work_senders: HashMap<WorkerId, _> = HashMap::new();
+    let mut work_senders: HashMap<TenantTimelineId, _> = HashMap::new();
     let mut tasks = Vec::new();
-    for timeline in timelines.iter().cloned() {
-        for num_client in 0..args.num_clients.get() {
-            let (sender, receiver) = tokio::sync::mpsc::channel(10); // TODO: not sure what the implications of this are
-            let worker_id = WorkerId {
-                timeline,
-                num_client,
-            };
-            work_senders.insert(worker_id, sender);
-            tasks.push(tokio::spawn(client(
-                args,
-                worker_id,
-                Arc::clone(&start_work_barrier),
-                receiver,
-                Arc::clone(&live_stats),
-                cancel.clone(),
-            )));
-        }
+    for tl in &timelines {
+        let (sender, receiver) = tokio::sync::mpsc::channel(10); // TODO: not sure what the implications of this are
+        work_senders.insert(*tl, sender);
+        tasks.push(tokio::spawn(client(
+            args,
+            *tl,
+            Arc::clone(&start_work_barrier),
+            receiver,
+            Arc::clone(&live_stats),
+            cancel.clone(),
+        )));
     }
 
     let work_sender: Pin<Box<dyn Send + Future<Output = ()>>> = {
@@ -283,10 +271,7 @@ async fn main_impl(
                         let (rel_tag, block_no) =
                             key_to_rel_block(key).expect("we filter non-rel-block keys out above");
                         (
-                            WorkerId {
-                                timeline: r.timeline,
-                                num_client: rng.gen_range(0..args.num_clients.get()),
-                            },
+                            r.timeline,
                             PagestreamGetPageRequest {
                                 latest: rng.gen_bool(args.req_latest_probability),
                                 lsn: r.timeline_lsn,
@@ -304,54 +289,56 @@ async fn main_impl(
             }),
             Some(rps_limit) => Box::pin(async move {
                 let period = Duration::from_secs_f64(1.0 / (rps_limit as f64));
-                let make_task: &dyn Fn(WorkerId) -> Pin<Box<dyn Send + Future<Output = ()>>> =
-                    &|worker_id| {
-                        let sender = work_senders.get(&worker_id).unwrap();
-                        let ranges: Vec<KeyRange> = all_ranges
-                            .iter()
-                            .filter(|r| r.timeline == worker_id.timeline)
-                            .cloned()
-                            .collect();
-                        let weights = rand::distributions::weighted::WeightedIndex::new(
-                            ranges.iter().map(|v| v.len()),
-                        )
-                        .unwrap();
+                let make_timeline_task: &dyn Fn(
+                    TenantTimelineId,
+                )
+                    -> Pin<Box<dyn Send + Future<Output = ()>>> = &|timeline| {
+                    let sender = work_senders.get(&timeline).unwrap();
+                    let ranges: Vec<KeyRange> = all_ranges
+                        .iter()
+                        .filter(|r| r.timeline == timeline)
+                        .cloned()
+                        .collect();
+                    let weights = rand::distributions::weighted::WeightedIndex::new(
+                        ranges.iter().map(|v| v.len()),
+                    )
+                    .unwrap();
 
-                        let cancel = cancel.clone();
-                        Box::pin(async move {
-                            let mut ticker = tokio::time::interval(period);
-                            ticker.set_missed_tick_behavior(
-                                /* TODO review this choice */
-                                tokio::time::MissedTickBehavior::Burst,
-                            );
-                            while !cancel.is_cancelled() {
-                                ticker.tick().await;
-                                let req = {
-                                    let mut rng = rand::thread_rng();
-                                    let r = &ranges[weights.sample(&mut rng)];
-                                    let key: i128 = rng.gen_range(r.start..r.end);
-                                    let key = Key::from_i128(key);
-                                    assert!(is_rel_block_key(&key));
-                                    let (rel_tag, block_no) = key_to_rel_block(key)
-                                        .expect("we filter non-rel-block keys out above");
-                                    PagestreamGetPageRequest {
-                                        latest: rng.gen_bool(args.req_latest_probability),
-                                        lsn: r.timeline_lsn,
-                                        rel: rel_tag,
-                                        blkno: block_no,
-                                    }
-                                };
-                                if sender.send(req).await.is_err() {
-                                    assert!(
-                                        cancel.is_cancelled(),
-                                        "client has gone away unexpectedly"
-                                    );
+                    let cancel = cancel.clone();
+                    Box::pin(async move {
+                        let mut ticker = tokio::time::interval(period);
+                        ticker.set_missed_tick_behavior(
+                            /* TODO review this choice */
+                            tokio::time::MissedTickBehavior::Burst,
+                        );
+                        while !cancel.is_cancelled() {
+                            ticker.tick().await;
+                            let req = {
+                                let mut rng = rand::thread_rng();
+                                let r = &ranges[weights.sample(&mut rng)];
+                                let key: i128 = rng.gen_range(r.start..r.end);
+                                let key = Key::from_i128(key);
+                                assert!(is_rel_block_key(&key));
+                                let (rel_tag, block_no) = key_to_rel_block(key)
+                                    .expect("we filter non-rel-block keys out above");
+                                PagestreamGetPageRequest {
+                                    latest: rng.gen_bool(args.req_latest_probability),
+                                    lsn: r.timeline_lsn,
+                                    rel: rel_tag,
+                                    blkno: block_no,
                                 }
+                            };
+                            if sender.send(req).await.is_err() {
+                                assert!(cancel.is_cancelled(), "client has gone away unexpectedly");
                             }
-                        })
-                    };
+                        }
+                    })
+                };
 
-                let tasks: Vec<_> = work_senders.keys().map(|tl| make_task(*tl)).collect();
+                let tasks: Vec<_> = work_senders
+                    .keys()
+                    .map(|tl| make_timeline_task(*tl))
+                    .collect();
 
                 start_work_barrier.wait().await;
 
@@ -403,16 +390,12 @@ async fn main_impl(
 #[instrument(skip_all)]
 async fn client(
     args: &'static Args,
-    id: WorkerId,
+    timeline: TenantTimelineId,
     start_work_barrier: Arc<Barrier>,
     mut work: tokio::sync::mpsc::Receiver<PagestreamGetPageRequest>,
     live_stats: Arc<LiveStats>,
     cancel: CancellationToken,
 ) {
-    let WorkerId {
-        timeline,
-        num_client: _,
-    } = id;
     let client = pageserver_client::page_service::Client::new(args.page_service_connstring.clone())
         .await
         .unwrap();
@@ -440,8 +423,8 @@ async fn client(
     tokio::select! {
         res = do_requests => { res },
         _ = cancel.cancelled() => {
-            // fallthrough to shutdown
+            client.shutdown().await;
+            return;
         }
     }
-    client.shutdown().await;
 }

pageserver/pagebench/src/util/request_stats.rs

@@ -66,10 +66,13 @@ impl serde::Serialize for LatencyPercentiles {
     {
         use serde::ser::SerializeMap;
         let mut ser = serializer.serialize_map(Some(LATENCY_PERCENTILES.len()))?;
-        for (p, v) in LATENCY_PERCENTILES.iter().zip(&self.latency_percentiles) {
+        for p in LATENCY_PERCENTILES {
             ser.serialize_entry(
                 &format!("p{p}"),
-                &format!("{}", humantime::format_duration(*v)),
+                &format!(
+                    "{}",
+                    &humantime::format_duration(self.latency_percentiles[0])
+                ),
             )?;
         }
         ser.end()

pageserver/src/basebackup.rs

@@ -11,9 +11,8 @@
 //! from data stored in object storage.
 //!
 use anyhow::{anyhow, bail, ensure, Context};
-use bytes::{BufMut, Bytes, BytesMut};
+use bytes::{BufMut, BytesMut};
 use fail::fail_point;
-use pageserver_api::key::{key_to_slru_block, Key};
 use postgres_ffi::pg_constants;
 use std::fmt::Write as FmtWrite;
 use std::time::SystemTime;
@@ -134,87 +133,6 @@ where
     ctx: &'a RequestContext,
 }
 
-/// A sink that accepts SLRU blocks ordered by key and forwards
-/// full segments to the archive.
-struct SlruSegmentsBuilder<'a, 'b, W>
-where
-    W: AsyncWrite + Send + Sync + Unpin,
-{
-    ar: &'a mut Builder<&'b mut W>,
-    buf: Vec<u8>,
-    current_segment: Option<(SlruKind, u32)>,
-}
-
-impl<'a, 'b, W> SlruSegmentsBuilder<'a, 'b, W>
-where
-    W: AsyncWrite + Send + Sync + Unpin,
-{
-    fn new(ar: &'a mut Builder<&'b mut W>) -> Self {
-        Self {
-            ar,
-            buf: Vec::new(),
-            current_segment: None,
-        }
-    }
-
-    async fn add_block(&mut self, key: &Key, block: Bytes) -> anyhow::Result<()> {
-        let (kind, segno, _) = key_to_slru_block(*key)?;
-
-        match kind {
-            SlruKind::Clog => {
-                ensure!(block.len() == BLCKSZ as usize || block.len() == BLCKSZ as usize + 8);
-            }
-            SlruKind::MultiXactMembers | SlruKind::MultiXactOffsets => {
-                ensure!(block.len() == BLCKSZ as usize);
-            }
-        }
-
-        let segment = (kind, segno);
-        match self.current_segment {
-            None => {
-                self.current_segment = Some(segment);
-                self.buf
-                    .extend_from_slice(block.slice(..BLCKSZ as usize).as_ref());
-            }
-            Some(current_seg) if current_seg == segment => {
-                self.buf
-                    .extend_from_slice(block.slice(..BLCKSZ as usize).as_ref());
-            }
-            Some(_) => {
-                self.flush().await?;
-
-                self.current_segment = Some(segment);
-                self.buf
-                    .extend_from_slice(block.slice(..BLCKSZ as usize).as_ref());
-            }
-        }
-
-        Ok(())
-    }
-
-    async fn flush(&mut self) -> anyhow::Result<()> {
-        let nblocks = self.buf.len() / BLCKSZ as usize;
-        let (kind, segno) = self.current_segment.take().unwrap();
-        let segname = format!("{}/{:>04X}", kind.to_str(), segno);
-        let header = new_tar_header(&segname, self.buf.len() as u64)?;
-        self.ar.append(&header, self.buf.as_slice()).await?;
-
-        trace!("Added to basebackup slru {} relsize {}", segname, nblocks);
-
-        self.buf.clear();
-
-        Ok(())
-    }
-
-    async fn finish(mut self) -> anyhow::Result<()> {
-        if self.current_segment.is_none() || self.buf.is_empty() {
-            return Ok(());
-        }
-
-        self.flush().await
-    }
-}
-
 impl<'a, W> Basebackup<'a, W>
 where
     W: AsyncWrite + Send + Sync + Unpin,
@@ -222,8 +140,6 @@ where
     async fn send_tarball(mut self) -> anyhow::Result<()> {
         // TODO include checksum
 
-        let lazy_slru_download = self.timeline.get_lazy_slru_download() && !self.full_backup;
-
         // Create pgdata subdirs structure
         for dir in PGDATA_SUBDIRS.iter() {
             let header = new_tar_header_dir(dir)?;
@@ -250,27 +166,20 @@ where
                     .context("could not add config file to basebackup tarball")?;
             }
         }
-        if !lazy_slru_download {
-            // Gather non-relational files from object storage pages.
-            let slru_partitions = self
+
+        // Gather non-relational files from object storage pages.
+        for kind in [
+            SlruKind::Clog,
+            SlruKind::MultiXactOffsets,
+            SlruKind::MultiXactMembers,
+        ] {
+            for segno in self
                 .timeline
-                .get_slru_keyspace(Version::Lsn(self.lsn), self.ctx)
+                .list_slru_segments(kind, Version::Lsn(self.lsn), self.ctx)
                 .await?
-                .partition(Timeline::MAX_GET_VECTORED_KEYS * BLCKSZ as u64);
-
-            let mut slru_builder = SlruSegmentsBuilder::new(&mut self.ar);
-
-            for part in slru_partitions.parts {
-                let blocks = self
-                    .timeline
-                    .get_vectored(&part.ranges, self.lsn, self.ctx)
-                    .await?;
-
-                for (key, block) in blocks {
-                    slru_builder.add_block(&key, block?).await?;
-                }
+            {
+                self.add_slru_segment(kind, segno).await?;
             }
-            slru_builder.finish().await?;
         }
 
         let mut min_restart_lsn: Lsn = Lsn::MAX;
@@ -396,6 +305,39 @@ where
         Ok(())
     }
 
+    //
+    // Generate SLRU segment files from repository.
+    //
+    async fn add_slru_segment(&mut self, slru: SlruKind, segno: u32) -> anyhow::Result<()> {
+        let nblocks = self
+            .timeline
+            .get_slru_segment_size(slru, segno, Version::Lsn(self.lsn), self.ctx)
+            .await?;
+
+        let mut slru_buf: Vec<u8> = Vec::with_capacity(nblocks as usize * BLCKSZ as usize);
+        for blknum in 0..nblocks {
+            let img = self
+                .timeline
+                .get_slru_page_at_lsn(slru, segno, blknum, self.lsn, self.ctx)
+                .await?;
+
+            if slru == SlruKind::Clog {
+                ensure!(img.len() == BLCKSZ as usize || img.len() == BLCKSZ as usize + 8);
+            } else {
+                ensure!(img.len() == BLCKSZ as usize);
+            }
+
+            slru_buf.extend_from_slice(&img[..BLCKSZ as usize]);
+        }
+
+        let segname = format!("{}/{:>04X}", slru.to_str(), segno);
+        let header = new_tar_header(&segname, slru_buf.len() as u64)?;
+        self.ar.append(&header, slru_buf.as_slice()).await?;
+
+        trace!("Added to basebackup slru {} relsize {}", segname, nblocks);
+        Ok(())
+    }
+
     //
     // Include database/tablespace directories.
     //

pageserver/src/bin/pageserver.rs

@@ -33,10 +33,12 @@ use pageserver::{
 use postgres_backend::AuthType;
 use utils::failpoint_support;
 use utils::logging::TracingErrorLayerEnablement;
+use utils::signals::ShutdownSignals;
 use utils::{
     auth::{JwtAuth, SwappableJwtAuth},
     logging, project_build_tag, project_git_version,
     sentry_init::init_sentry,
+    signals::Signal,
     tcp_listener,
 };
 
@@ -128,7 +130,7 @@ fn main() -> anyhow::Result<()> {
     let scenario = failpoint_support::init();
 
     // Basic initialization of things that don't change after startup
-    virtual_file::init(conf.max_file_descriptors, conf.virtual_file_io_engine);
+    virtual_file::init(conf.max_file_descriptors);
     page_cache::init(conf.page_cache_size);
 
     start_pageserver(launch_ts, conf).context("Failed to start pageserver")?;
@@ -654,42 +656,34 @@ fn start_pageserver(
     let mut shutdown_pageserver = Some(shutdown_pageserver.drop_guard());
 
     // All started up! Now just sit and wait for shutdown signal.
-    {
-        use signal_hook::consts::*;
-        let signal_handler = BACKGROUND_RUNTIME.spawn_blocking(move || {
-            let mut signals =
-                signal_hook::iterator::Signals::new([SIGINT, SIGTERM, SIGQUIT]).unwrap();
-            return signals
-                .forever()
-                .next()
-                .expect("forever() never returns None unless explicitly closed");
-        });
-        let signal = BACKGROUND_RUNTIME
-            .block_on(signal_handler)
-            .expect("join error");
-        match signal {
-            SIGQUIT => {
-                info!("Got signal {signal}. Terminating in immediate shutdown mode",);
-                std::process::exit(111);
-            }
-            SIGINT | SIGTERM => {
-                info!("Got signal {signal}. Terminating gracefully in fast shutdown mode",);
-
-                // This cancels the `shutdown_pageserver` cancellation tree.
-                // Right now that tree doesn't reach very far, and `task_mgr` is used instead.
-                // The plan is to change that over time.
-                shutdown_pageserver.take();
-                let bg_remote_storage = remote_storage.clone();
-                let bg_deletion_queue = deletion_queue.clone();
-                BACKGROUND_RUNTIME.block_on(pageserver::shutdown_pageserver(
-                    bg_remote_storage.map(|_| bg_deletion_queue),
-                    0,
-                ));
-                unreachable!()
-            }
-            _ => unreachable!(),
+    ShutdownSignals::handle(|signal| match signal {
+        Signal::Quit => {
+            info!(
+                "Got {}. Terminating in immediate shutdown mode",
+                signal.name()
+            );
+            std::process::exit(111);
         }
-    }
+
+        Signal::Interrupt | Signal::Terminate => {
+            info!(
+                "Got {}. Terminating gracefully in fast shutdown mode",
+                signal.name()
+            );
+
+            // This cancels the `shutdown_pageserver` cancellation tree.
+            // Right now that tree doesn't reach very far, and `task_mgr` is used instead.
+            // The plan is to change that over time.
+            shutdown_pageserver.take();
+            let bg_remote_storage = remote_storage.clone();
+            let bg_deletion_queue = deletion_queue.clone();
+            BACKGROUND_RUNTIME.block_on(pageserver::shutdown_pageserver(
+                bg_remote_storage.map(|_| bg_deletion_queue),
+                0,
+            ));
+            unreachable!()
+        }
+    })
 }
 
 fn create_remote_storage_client(

pageserver/src/config.rs

@@ -36,7 +36,6 @@ use crate::tenant::config::TenantConfOpt;
 use crate::tenant::{
     TENANTS_SEGMENT_NAME, TENANT_DELETED_MARKER_FILE_NAME, TIMELINES_SEGMENT_NAME,
 };
-use crate::virtual_file;
 use crate::{
     IGNORED_TENANT_FILE_NAME, METADATA_FILE_NAME, TENANT_CONFIG_NAME, TENANT_HEATMAP_BASENAME,
     TENANT_LOCATION_CONFIG_NAME, TIMELINE_DELETE_MARK_SUFFIX, TIMELINE_UNINIT_MARK_SUFFIX,
@@ -44,8 +43,6 @@ use crate::{
 
 use self::defaults::DEFAULT_CONCURRENT_TENANT_WARMUP;
 
-use self::defaults::DEFAULT_VIRTUAL_FILE_IO_ENGINE;
-
 pub mod defaults {
     use crate::tenant::config::defaults::*;
     use const_format::formatcp;
@@ -82,8 +79,6 @@ pub mod defaults {
 
     pub const DEFAULT_INGEST_BATCH_SIZE: u64 = 100;
 
-    pub const DEFAULT_VIRTUAL_FILE_IO_ENGINE: &str = "std-fs";
-
     ///
     /// Default built-in configuration file.
     ///
@@ -119,8 +114,6 @@ pub mod defaults {
 
 #ingest_batch_size = {DEFAULT_INGEST_BATCH_SIZE}
 
-#virtual_file_io_engine = '{DEFAULT_VIRTUAL_FILE_IO_ENGINE}'
-
 [tenant_config]
 #checkpoint_distance = {DEFAULT_CHECKPOINT_DISTANCE} # in bytes
 #checkpoint_timeout = {DEFAULT_CHECKPOINT_TIMEOUT}
@@ -254,8 +247,6 @@ pub struct PageServerConf {
 
     /// Maximum number of WAL records to be ingested and committed at the same time
     pub ingest_batch_size: u64,
-
-    pub virtual_file_io_engine: virtual_file::IoEngineKind,
 }
 
 /// We do not want to store this in a PageServerConf because the latter may be logged
@@ -340,8 +331,6 @@ struct PageServerConfigBuilder {
     secondary_download_concurrency: BuilderValue<usize>,
 
     ingest_batch_size: BuilderValue<u64>,
-
-    virtual_file_io_engine: BuilderValue<virtual_file::IoEngineKind>,
 }
 
 impl Default for PageServerConfigBuilder {
@@ -417,8 +406,6 @@ impl Default for PageServerConfigBuilder {
             secondary_download_concurrency: Set(DEFAULT_SECONDARY_DOWNLOAD_CONCURRENCY),
 
             ingest_batch_size: Set(DEFAULT_INGEST_BATCH_SIZE),
-
-            virtual_file_io_engine: Set(DEFAULT_VIRTUAL_FILE_IO_ENGINE.parse().unwrap()),
         }
     }
 }
@@ -575,10 +562,6 @@ impl PageServerConfigBuilder {
         self.ingest_batch_size = BuilderValue::Set(ingest_batch_size)
     }
 
-    pub fn virtual_file_io_engine(&mut self, value: virtual_file::IoEngineKind) {
-        self.virtual_file_io_engine = BuilderValue::Set(value);
-    }
-
     pub fn build(self) -> anyhow::Result<PageServerConf> {
         let concurrent_tenant_warmup = self
             .concurrent_tenant_warmup
@@ -686,9 +669,6 @@ impl PageServerConfigBuilder {
             ingest_batch_size: self
                 .ingest_batch_size
                 .ok_or(anyhow!("missing ingest_batch_size"))?,
-            virtual_file_io_engine: self
-                .virtual_file_io_engine
-                .ok_or(anyhow!("missing virtual_file_io_engine"))?,
         })
     }
 }
@@ -940,9 +920,6 @@ impl PageServerConf {
                     builder.secondary_download_concurrency(parse_toml_u64(key, item)? as usize)
                 },
                 "ingest_batch_size" => builder.ingest_batch_size(parse_toml_u64(key, item)?),
-                "virtual_file_io_engine" => {
-                    builder.virtual_file_io_engine(parse_toml_from_str("virtual_file_io_engine", item)?)
-                }
                 _ => bail!("unrecognized pageserver option '{key}'"),
             }
         }
@@ -1016,7 +993,6 @@ impl PageServerConf {
             heatmap_upload_concurrency: defaults::DEFAULT_HEATMAP_UPLOAD_CONCURRENCY,
             secondary_download_concurrency: defaults::DEFAULT_SECONDARY_DOWNLOAD_CONCURRENCY,
             ingest_batch_size: defaults::DEFAULT_INGEST_BATCH_SIZE,
-            virtual_file_io_engine: DEFAULT_VIRTUAL_FILE_IO_ENGINE.parse().unwrap(),
         }
     }
 }
@@ -1249,7 +1225,6 @@ background_task_maximum_delay = '334 s'
                 heatmap_upload_concurrency: defaults::DEFAULT_HEATMAP_UPLOAD_CONCURRENCY,
                 secondary_download_concurrency: defaults::DEFAULT_SECONDARY_DOWNLOAD_CONCURRENCY,
                 ingest_batch_size: defaults::DEFAULT_INGEST_BATCH_SIZE,
-                virtual_file_io_engine: DEFAULT_VIRTUAL_FILE_IO_ENGINE.parse().unwrap(),
             },
             "Correct defaults should be used when no config values are provided"
         );
@@ -1313,7 +1288,6 @@ background_task_maximum_delay = '334 s'
                 heatmap_upload_concurrency: defaults::DEFAULT_HEATMAP_UPLOAD_CONCURRENCY,
                 secondary_download_concurrency: defaults::DEFAULT_SECONDARY_DOWNLOAD_CONCURRENCY,
                 ingest_batch_size: 100,
-                virtual_file_io_engine: DEFAULT_VIRTUAL_FILE_IO_ENGINE.parse().unwrap(),
             },
             "Should be able to parse all basic config values correctly"
         );

pageserver/src/consumption_metrics/upload.rs

@@ -262,33 +262,35 @@ async fn upload(
 ) -> Result<(), UploadError> {
     let warn_after = 3;
     let max_attempts = 10;
-
-    // this is used only with tests so far
-    let last_value = if is_last { "true" } else { "false" };
-
     let res = utils::backoff::retry(
-        || async {
-            let res = client
-                .post(metric_collection_endpoint.clone())
-                .header(reqwest::header::CONTENT_TYPE, "application/json")
-                .header(LAST_IN_BATCH.clone(), last_value)
-                .body(body.clone())
-                .send()
-                .await;
+        move || {
+            let body = body.clone();
+            async move {
+                let res = client
+                    .post(metric_collection_endpoint.clone())
+                    .header(reqwest::header::CONTENT_TYPE, "application/json")
+                    .header(
+                        LAST_IN_BATCH.clone(),
+                        if is_last { "true" } else { "false" },
+                    )
+                    .body(body)
+                    .send()
+                    .await;
 
-            let res = res.and_then(|res| res.error_for_status());
+                let res = res.and_then(|res| res.error_for_status());
 
-            // 10 redirects are normally allowed, so we don't need worry about 3xx
-            match res {
-                Ok(_response) => Ok(()),
-                Err(e) => {
-                    let status = e.status().filter(|s| s.is_client_error());
-                    if let Some(status) = status {
-                        // rejection used to be a thing when the server could reject a
-                        // whole batch of metrics if one metric was bad.
-                        Err(UploadError::Rejected(status))
-                    } else {
-                        Err(UploadError::Reqwest(e))
+                // 10 redirects are normally allowed, so we don't need worry about 3xx
+                match res {
+                    Ok(_response) => Ok(()),
+                    Err(e) => {
+                        let status = e.status().filter(|s| s.is_client_error());
+                        if let Some(status) = status {
+                            // rejection used to be a thing when the server could reject a
+                            // whole batch of metrics if one metric was bad.
+                            Err(UploadError::Rejected(status))
+                        } else {
+                            Err(UploadError::Reqwest(e))
+                        }
                     }
                 }
             }
@@ -297,11 +299,9 @@ async fn upload(
         warn_after,
         max_attempts,
         "upload consumption_metrics",
-        cancel,
+        utils::backoff::Cancel::new(cancel.clone(), || UploadError::Cancelled),
     )
-    .await
-    .ok_or_else(|| UploadError::Cancelled)
-    .and_then(|x| x);
+    .await;
 
     match &res {
         Ok(_) => {}

pageserver/src/control_plane_client.rs

@@ -82,29 +82,46 @@ impl ControlPlaneClient {
         R: Serialize,
         T: DeserializeOwned,
     {
-        let res = backoff::retry(
+        #[derive(thiserror::Error, Debug)]
+        enum RemoteAttemptError {
+            #[error("shutdown")]
+            Shutdown,
+            #[error("remote: {0}")]
+            Remote(reqwest::Error),
+        }
+
+        match backoff::retry(
             || async {
                 let response = self
                     .http_client
                     .post(url.clone())
                     .json(&request)
                     .send()
-                    .await?;
+                    .await
+                    .map_err(RemoteAttemptError::Remote)?;
 
-                response.error_for_status_ref()?;
-                response.json::<T>().await
+                response
+                    .error_for_status_ref()
+                    .map_err(RemoteAttemptError::Remote)?;
+                response
+                    .json::<T>()
+                    .await
+                    .map_err(RemoteAttemptError::Remote)
             },
             |_| false,
             3,
             u32::MAX,
             "calling control plane generation validation API",
-            &self.cancel,
+            backoff::Cancel::new(self.cancel.clone(), || RemoteAttemptError::Shutdown),
         )
         .await
-        .ok_or(RetryForeverError::ShuttingDown)?
-        .expect("We retry forever, this should never be reached");
-
-        Ok(res)
+        {
+            Err(RemoteAttemptError::Shutdown) => Err(RetryForeverError::ShuttingDown),
+            Err(RemoteAttemptError::Remote(_)) => {
+                panic!("We retry forever, this should never be reached");
+            }
+            Ok(r) => Ok(r),
+        }
     }
 }
 

pageserver/src/deletion_queue.rs

@@ -700,6 +700,8 @@ impl DeletionQueue {
     }
 
     pub async fn shutdown(&mut self, timeout: Duration) {
+        self.cancel.cancel();
+
         match tokio::time::timeout(timeout, self.client.flush()).await {
             Ok(Ok(())) => {
                 tracing::info!("Deletion queue flushed successfully on shutdown")
@@ -713,10 +715,6 @@ impl DeletionQueue {
                 tracing::warn!("Timed out flushing deletion queue on shutdown")
             }
         }
-
-        // We only cancel _after_ flushing: otherwise we would be shutting down the
-        // components that do the flush.
-        self.cancel.cancel();
     }
 }