clean up

Noticed this while debugging a test failure in #8673 which only occurs
with real S3 instead of mock S3: if you authenticate to S3 via
`AWS_PROFILE`, then it requires the `HOME` env var to be set so that it
can read inside the `~/.aws` directory.

The scrubber abstraction `StorageScrubber::scrubber_cli` in
`neon_fixtures.py` would otherwise not work. My earlier PR #6556 has
done similar things for the `neon_local` wrapper.

You can try:

```
aws sso login --profile dev
export ENABLE_REAL_S3_REMOTE_STORAGE=y REMOTE_STORAGE_S3_BUCKET=neon-github-ci-tests REMOTE_STORAGE_S3_REGION=eu-central-1 AWS_PROFILE=dev
RUST_BACKTRACE=1 BUILD_TYPE=debug DEFAULT_PG_VERSION=16 ./scripts/pytest -vv --tb=short -k test_scrubber_tenant_snapshot
```

before and after this patch: this patch fixes it.

.gitattributes

@@ -0,0 +1,2 @@
+# allows for nicer hunk headers with git show
+*.rs diff=rust

.github/actionlint.yml

@@ -8,6 +8,9 @@ self-hosted-runner:
     - small-arm64
     - us-east-2
 config-variables:
+  - BENCHMARK_PROJECT_ID_PUB
+  - BENCHMARK_PROJECT_ID_SUB
   - REMOTE_STORAGE_AZURE_CONTAINER
   - REMOTE_STORAGE_AZURE_REGION
   - SLACK_UPCOMING_RELEASE_CHANNEL_ID
+  - DEV_AWS_OIDC_ROLE_ARN

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

@@ -9,16 +9,13 @@ inputs:
     description: 'Region ID, if not set the project will be created in the default region'
     default: aws-us-east-2
   postgres_version:
-    description: 'Postgres version; default is 15'
-    default: '15'
+    description: 'Postgres version; default is 16'
+    default: '16'
   api_host:
     description: 'Neon API host'
     default: console-stage.neon.build
-  provisioner:
-    description: 'k8s-pod or k8s-neonvm'
-    default: 'k8s-pod'
   compute_units:
-    description: '[Min, Max] compute units; Min and Max are used for k8s-neonvm with autoscaling, for k8s-pod values Min and Max should be equal'
+    description: '[Min, Max] compute units'
     default: '[1, 1]'
 
 outputs:
@@ -37,10 +34,6 @@ runs:
       # A shell without `set -x` to not to expose password/dsn in logs
       shell: bash -euo pipefail {0}
       run: |
-        if [ "${PROVISIONER}" == "k8s-pod" ] && [ "${MIN_CU}" != "${MAX_CU}" ]; then
-          echo >&2 "For k8s-pod provisioner MIN_CU should be equal to MAX_CU"
-        fi
-
         project=$(curl \
           "https://${API_HOST}/api/v2/projects" \
           --fail \
@@ -52,7 +45,7 @@ runs:
               \"name\": \"Created by actions/neon-project-create; GITHUB_RUN_ID=${GITHUB_RUN_ID}\",
               \"pg_version\": ${POSTGRES_VERSION},
               \"region_id\": \"${REGION_ID}\",
-              \"provisioner\": \"${PROVISIONER}\",
+              \"provisioner\": \"k8s-neonvm\",
               \"autoscaling_limit_min_cu\": ${MIN_CU},
               \"autoscaling_limit_max_cu\": ${MAX_CU},
               \"settings\": { }
@@ -75,6 +68,5 @@ runs:
         API_KEY: ${{ inputs.api_key }}
         REGION_ID: ${{ inputs.region_id }}
         POSTGRES_VERSION: ${{ inputs.postgres_version }}
-        PROVISIONER: ${{ inputs.provisioner }}
         MIN_CU: ${{ fromJSON(inputs.compute_units)[0] }}
         MAX_CU: ${{ fromJSON(inputs.compute_units)[1] }}

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

@@ -114,6 +114,8 @@ runs:
         export PLATFORM=${PLATFORM:-github-actions-selfhosted}
         export POSTGRES_DISTRIB_DIR=${POSTGRES_DISTRIB_DIR:-/tmp/neon/pg_install}
         export DEFAULT_PG_VERSION=${PG_VERSION#v}
+        export LD_LIBRARY_PATH=${POSTGRES_DISTRIB_DIR}/v${DEFAULT_PG_VERSION}/lib
+        export BENCHMARK_CONNSTR=${BENCHMARK_CONNSTR:-}
 
         if [ "${BUILD_TYPE}" = "remote" ]; then
           export REMOTE_ENV=1
@@ -129,8 +131,8 @@ runs:
           exit 1
         fi
         if [[ "${{ inputs.run_in_parallel }}" == "true" ]]; then
-          # -n16 uses sixteen processes to run tests via pytest-xdist
-          EXTRA_PARAMS="-n16 $EXTRA_PARAMS"
+          # -n sets the number of parallel processes that pytest-xdist will run
+          EXTRA_PARAMS="-n12 $EXTRA_PARAMS"
 
           # --dist=loadgroup points tests marked with @pytest.mark.xdist_group
           # to the same worker to make @pytest.mark.order work with xdist
@@ -178,7 +180,15 @@ runs:
 
         # Wake up the cluster if we use remote neon instance
         if [ "${{ inputs.build_type }}" = "remote" ] && [ -n "${BENCHMARK_CONNSTR}" ]; then
-          ${POSTGRES_DISTRIB_DIR}/v${DEFAULT_PG_VERSION}/bin/psql ${BENCHMARK_CONNSTR} -c "SELECT version();"
+          QUERIES=("SELECT version()")
+          if [[ "${PLATFORM}" = "neon"* ]]; then
+            QUERIES+=("SHOW neon.tenant_id")
+            QUERIES+=("SHOW neon.timeline_id")
+          fi
+
+          for q in "${QUERIES[@]}"; do
+            ${POSTGRES_DISTRIB_DIR}/v${DEFAULT_PG_VERSION}/bin/psql ${BENCHMARK_CONNSTR} -c "${q}"
+          done
         fi
 
         # Run the tests.

.github/workflows/_build-and-test-locally.yml

@@ -0,0 +1,292 @@
+name: Build and Test Locally
+
+on:
+  workflow_call:
+    inputs:
+      arch:
+        description: 'x64 or arm64'
+        required: true
+        type: string
+      build-tag:
+        description: 'build tag'
+        required: true
+        type: string
+      build-tools-image:
+        description: 'build-tools image'
+        required: true
+        type: string
+      build-type:
+        description: 'debug or release'
+        required: true
+        type: string
+      pg-versions:
+        description: 'a json array of postgres versions to run regression tests on'
+        required: true
+        type: string
+
+defaults:
+  run:
+    shell: bash -euxo pipefail {0}
+
+env:
+  RUST_BACKTRACE: 1
+  COPT: '-Werror'
+  AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_DEV }}
+  AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_KEY_DEV }}
+
+jobs:
+  build-neon:
+    runs-on: ${{ fromJson(format('["self-hosted", "{0}"]', inputs.arch == 'arm64' && 'large-arm64' || 'large')) }}
+    container:
+      image: ${{ inputs.build-tools-image }}
+      credentials:
+        username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
+        password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
+      # 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
+    env:
+      BUILD_TYPE: ${{ inputs.build-type }}
+      GIT_VERSION: ${{ github.event.pull_request.head.sha || github.sha }}
+      BUILD_TAG: ${{ inputs.build-tag }}
+
+    steps:
+      - name: Fix git ownership
+        run: |
+          # Workaround for `fatal: detected dubious ownership in repository at ...`
+          #
+          # Use both ${{ github.workspace }} and ${GITHUB_WORKSPACE} because they're different on host and in containers
+          #   Ref https://github.com/actions/checkout/issues/785
+          #
+          git config --global --add safe.directory ${{ github.workspace }}
+          git config --global --add safe.directory ${GITHUB_WORKSPACE}
+          for r in 14 15 16; do
+            git config --global --add safe.directory "${{ github.workspace }}/vendor/postgres-v$r"
+            git config --global --add safe.directory "${GITHUB_WORKSPACE}/vendor/postgres-v$r"
+          done
+
+      - uses: actions/checkout@v4
+        with:
+          submodules: true
+          fetch-depth: 1
+
+      - name: Set pg 14 revision for caching
+        id: pg_v14_rev
+        run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-v14) >> $GITHUB_OUTPUT
+
+      - name: Set pg 15 revision for caching
+        id: pg_v15_rev
+        run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-v15) >> $GITHUB_OUTPUT
+
+      - name: Set pg 16 revision for caching
+        id: pg_v16_rev
+        run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-v16) >> $GITHUB_OUTPUT
+
+      # Set some environment variables used by all the steps.
+      #
+      # CARGO_FLAGS is extra options to pass to "cargo build", "cargo test" etc.
+      #   It also includes --features, if any
+      #
+      # CARGO_FEATURES is passed to "cargo metadata". It is separate from CARGO_FLAGS,
+      #   because "cargo metadata" doesn't accept --release or --debug options
+      #
+      # We run tests with addtional features, that are turned off by default (e.g. in release builds), see
+      # corresponding Cargo.toml files for their descriptions.
+      - name: Set env variables
+        run: |
+          CARGO_FEATURES="--features testing"
+          if [[ $BUILD_TYPE == "debug" ]]; then
+            cov_prefix="scripts/coverage --profraw-prefix=$GITHUB_JOB --dir=/tmp/coverage run"
+            CARGO_FLAGS="--locked"
+          elif [[ $BUILD_TYPE == "release" ]]; then
+            cov_prefix=""
+            CARGO_FLAGS="--locked --release"
+          fi
+          {
+            echo "cov_prefix=${cov_prefix}"
+            echo "CARGO_FEATURES=${CARGO_FEATURES}"
+            echo "CARGO_FLAGS=${CARGO_FLAGS}"
+            echo "CARGO_HOME=${GITHUB_WORKSPACE}/.cargo"
+          } >> $GITHUB_ENV
+
+      - name: Cache postgres v14 build
+        id: cache_pg_14
+        uses: actions/cache@v4
+        with:
+          path: pg_install/v14
+          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ inputs.build-type }}-pg-${{ steps.pg_v14_rev.outputs.pg_rev }}-${{ hashFiles('Makefile', 'Dockerfile.build-tools') }}
+
+      - name: Cache postgres v15 build
+        id: cache_pg_15
+        uses: actions/cache@v4
+        with:
+          path: pg_install/v15
+          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ inputs.build-type }}-pg-${{ steps.pg_v15_rev.outputs.pg_rev }}-${{ hashFiles('Makefile', 'Dockerfile.build-tools') }}
+
+      - name: Cache postgres v16 build
+        id: cache_pg_16
+        uses: actions/cache@v4
+        with:
+          path: pg_install/v16
+          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ inputs.build-type }}-pg-${{ steps.pg_v16_rev.outputs.pg_rev }}-${{ hashFiles('Makefile', 'Dockerfile.build-tools') }}
+
+      - name: Build postgres v14
+        if: steps.cache_pg_14.outputs.cache-hit != 'true'
+        run: mold -run make postgres-v14 -j$(nproc)
+
+      - name: Build postgres v15
+        if: steps.cache_pg_15.outputs.cache-hit != 'true'
+        run: mold -run make postgres-v15 -j$(nproc)
+
+      - name: Build postgres v16
+        if: steps.cache_pg_16.outputs.cache-hit != 'true'
+        run: mold -run make postgres-v16 -j$(nproc)
+
+      - name: Build neon extensions
+        run: mold -run make neon-pg-ext -j$(nproc)
+
+      - name: Build walproposer-lib
+        run: mold -run make walproposer-lib -j$(nproc)
+
+      - name: Run cargo build
+        run: |
+          PQ_LIB_DIR=$(pwd)/pg_install/v16/lib
+          export PQ_LIB_DIR
+          ${cov_prefix} mold -run cargo build $CARGO_FLAGS $CARGO_FEATURES --bins --tests
+
+      # Do install *before* running rust tests because they might recompile the
+      # binaries with different features/flags.
+      - name: Install rust binaries
+        run: |
+          # Install target binaries
+          mkdir -p /tmp/neon/bin/
+          binaries=$(
+            ${cov_prefix} cargo metadata $CARGO_FEATURES --format-version=1 --no-deps |
+            jq -r '.packages[].targets[] | select(.kind | index("bin")) | .name'
+          )
+          for bin in $binaries; do
+            SRC=target/$BUILD_TYPE/$bin
+            DST=/tmp/neon/bin/$bin
+            cp "$SRC" "$DST"
+          done
+
+          # Install test executables and write list of all binaries (for code coverage)
+          if [[ $BUILD_TYPE == "debug" ]]; then
+            # Keep bloated coverage data files away from the rest of the artifact
+            mkdir -p /tmp/coverage/
+
+            mkdir -p /tmp/neon/test_bin/
+
+            test_exe_paths=$(
+              ${cov_prefix} cargo test $CARGO_FLAGS $CARGO_FEATURES --message-format=json --no-run |
+              jq -r '.executable | select(. != null)'
+            )
+            for bin in $test_exe_paths; do
+              SRC=$bin
+              DST=/tmp/neon/test_bin/$(basename $bin)
+
+              # We don't need debug symbols for code coverage, so strip them out to make
+              # the artifact smaller.
+              strip "$SRC" -o "$DST"
+              echo "$DST" >> /tmp/coverage/binaries.list
+            done
+
+            for bin in $binaries; do
+              echo "/tmp/neon/bin/$bin" >> /tmp/coverage/binaries.list
+            done
+          fi
+
+      - name: Run rust tests
+        env:
+          NEXTEST_RETRIES: 3
+        run: |
+          PQ_LIB_DIR=$(pwd)/pg_install/v16/lib
+          export PQ_LIB_DIR
+          LD_LIBRARY_PATH=$(pwd)/pg_install/v16/lib
+          export LD_LIBRARY_PATH
+
+          #nextest does not yet support running doctests
+          cargo test --doc $CARGO_FLAGS $CARGO_FEATURES
+
+          for io_engine in std-fs tokio-epoll-uring ; do
+            NEON_PAGESERVER_UNIT_TEST_VIRTUAL_FILE_IOENGINE=$io_engine ${cov_prefix} cargo nextest run $CARGO_FLAGS $CARGO_FEATURES
+          done
+
+          # Run separate tests for real S3
+          export ENABLE_REAL_S3_REMOTE_STORAGE=nonempty
+          export REMOTE_STORAGE_S3_BUCKET=neon-github-ci-tests
+          export REMOTE_STORAGE_S3_REGION=eu-central-1
+          ${cov_prefix} cargo nextest run $CARGO_FLAGS $CARGO_FEATURES -E 'package(remote_storage)' -E 'test(test_real_s3)'
+
+          # Run separate tests for real Azure Blob Storage
+          # XXX: replace region with `eu-central-1`-like region
+          export ENABLE_REAL_AZURE_REMOTE_STORAGE=y
+          export AZURE_STORAGE_ACCOUNT="${{ secrets.AZURE_STORAGE_ACCOUNT_DEV }}"
+          export AZURE_STORAGE_ACCESS_KEY="${{ secrets.AZURE_STORAGE_ACCESS_KEY_DEV }}"
+          export REMOTE_STORAGE_AZURE_CONTAINER="${{ vars.REMOTE_STORAGE_AZURE_CONTAINER }}"
+          export REMOTE_STORAGE_AZURE_REGION="${{ vars.REMOTE_STORAGE_AZURE_REGION }}"
+          ${cov_prefix} cargo nextest run $CARGO_FLAGS $CARGO_FEATURES -E 'package(remote_storage)' -E 'test(test_real_azure)'
+
+      - name: Install postgres binaries
+        run: cp -a pg_install /tmp/neon/pg_install
+
+      - name: Upload Neon artifact
+        uses: ./.github/actions/upload
+        with:
+          name: neon-${{ runner.os }}-${{ runner.arch }}-${{ inputs.build-type }}-artifact
+          path: /tmp/neon
+
+      # XXX: keep this after the binaries.list is formed, so the coverage can properly work later
+      - name: Merge and upload coverage data
+        if: inputs.build-type == 'debug'
+        uses: ./.github/actions/save-coverage-data
+
+  regress-tests:
+    # Run test on x64 only
+    if: inputs.arch == 'x64'
+    needs: [ build-neon ]
+    runs-on: ${{ fromJson(format('["self-hosted", "{0}"]', inputs.arch == 'arm64' && 'large-arm64' || 'large')) }}
+    container:
+      image: ${{ inputs.build-tools-image }}
+      credentials:
+        username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
+        password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
+      # for changed limits, see comments on `options:` earlier in this file
+      options: --init --shm-size=512mb --ulimit memlock=67108864:67108864
+    strategy:
+      fail-fast: false
+      matrix:
+        pg_version: ${{ fromJson(inputs.pg-versions) }}
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          submodules: true
+          fetch-depth: 1
+
+      - name: Pytest regression tests
+        uses: ./.github/actions/run-python-test-set
+        timeout-minutes: 60
+        with:
+          build_type: ${{ inputs.build-type }}
+          test_selection: regress
+          needs_postgres_source: true
+          run_with_real_s3: true
+          real_s3_bucket: neon-github-ci-tests
+          real_s3_region: eu-central-1
+          rerun_flaky: true
+          pg_version: ${{ matrix.pg_version }}
+        env:
+          TEST_RESULT_CONNSTR: ${{ secrets.REGRESS_TEST_RESULT_CONNSTR_NEW }}
+          CHECK_ONDISK_DATA_COMPATIBILITY: nonempty
+          BUILD_TAG: ${{ inputs.build-tag }}
+          PAGESERVER_VIRTUAL_FILE_IO_ENGINE: tokio-epoll-uring
+
+      # Temporary disable this step until we figure out why it's so flaky
+      # Ref https://github.com/neondatabase/neon/issues/4540
+      - name: Merge and upload coverage data
+        if: |
+          false &&
+          inputs.build-type == 'debug' && matrix.pg_version == 'v16'
+        uses: ./.github/actions/save-coverage-data

.github/workflows/benchmarking.yml

@@ -56,24 +56,49 @@ concurrency:
 jobs:
   bench:
     if: ${{ github.event.inputs.run_only_pgvector_tests == 'false' || github.event.inputs.run_only_pgvector_tests == null }}
+    permissions:
+      contents: write
+      statuses: write
+      id-token: write # Required for OIDC authentication in azure runners
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - DEFAULT_PG_VERSION: 16
+            PLATFORM: "neon-staging"
+            region_id: ${{ github.event.inputs.region_id || 'aws-us-east-2' }}
+            RUNNER: [ self-hosted, us-east-2, x64 ]
+            IMAGE: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools:pinned
+          - DEFAULT_PG_VERSION: 16
+            PLATFORM: "azure-staging"
+            region_id: 'azure-eastus2'
+            RUNNER: [ self-hosted, eastus2, x64 ]
+            IMAGE: neondatabase/build-tools:pinned
     env:
       TEST_PG_BENCH_DURATIONS_MATRIX: "300"
       TEST_PG_BENCH_SCALES_MATRIX: "10,100"
       POSTGRES_DISTRIB_DIR: /tmp/neon/pg_install
-      DEFAULT_PG_VERSION: 14
+      DEFAULT_PG_VERSION: ${{ matrix.DEFAULT_PG_VERSION }}
       TEST_OUTPUT: /tmp/test_output
       BUILD_TYPE: remote
       SAVE_PERF_REPORT: ${{ github.event.inputs.save_perf_report || ( github.ref_name == 'main' ) }}
-      PLATFORM: "neon-staging"
+      PLATFORM: ${{ matrix.PLATFORM }}
 
-    runs-on: [ self-hosted, us-east-2, x64 ]
+    runs-on: ${{ matrix.RUNNER }}
     container:
-      image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools:pinned
+      image: ${{ matrix.IMAGE }}
       options: --init
 
     steps:
     - uses: actions/checkout@v4
 
+    - name: Configure AWS credentials # necessary on Azure runners
+      uses: aws-actions/configure-aws-credentials@v4
+      with:
+        aws-region: eu-central-1
+        role-to-assume: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}  
+        role-duration-seconds: 18000 # 5 hours
+
     - name: Download Neon artifact
       uses: ./.github/actions/download
       with:
@@ -85,7 +110,7 @@ jobs:
       id: create-neon-project
       uses: ./.github/actions/neon-project-create
       with:
-        region_id: ${{ github.event.inputs.region_id || 'aws-us-east-2' }}
+        region_id: ${{ matrix.region_id }}
         postgres_version: ${{ env.DEFAULT_PG_VERSION }}
         api_key: ${{ secrets.NEON_STAGING_API_KEY }}
 
@@ -96,10 +121,18 @@ jobs:
         test_selection: performance
         run_in_parallel: false
         save_perf_report: ${{ env.SAVE_PERF_REPORT }}
+        pg_version: ${{ env.DEFAULT_PG_VERSION }}
         # Set --sparse-ordering option of pytest-order plugin
         # to ensure tests are running in order of appears in the file.
         # It's important for test_perf_pgbench.py::test_pgbench_remote_* tests
-        extra_params: -m remote_cluster --sparse-ordering --timeout 5400 --ignore test_runner/performance/test_perf_olap.py --ignore test_runner/performance/test_perf_pgvector_queries.py
+        extra_params:
+          -m remote_cluster
+          --sparse-ordering
+          --timeout 14400
+          --ignore test_runner/performance/test_perf_olap.py
+          --ignore test_runner/performance/test_perf_pgvector_queries.py
+          --ignore test_runner/performance/test_logical_replication.py
+          --ignore test_runner/performance/test_physical_replication.py
       env:
         BENCHMARK_CONNSTR: ${{ steps.create-neon-project.outputs.dsn }}
         VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
@@ -125,18 +158,90 @@ jobs:
       env:
         SLACK_BOT_TOKEN: ${{ secrets.SLACK_BOT_TOKEN }}
 
+  replication-tests:
+    if: ${{ github.event.inputs.run_only_pgvector_tests == 'false' || github.event.inputs.run_only_pgvector_tests == null }}
+    env:
+      POSTGRES_DISTRIB_DIR: /tmp/neon/pg_install
+      DEFAULT_PG_VERSION: 16
+      TEST_OUTPUT: /tmp/test_output
+      BUILD_TYPE: remote
+      SAVE_PERF_REPORT: ${{ github.event.inputs.save_perf_report || ( github.ref_name == 'main' ) }}
+      PLATFORM: "neon-staging"
+
+    runs-on: [ self-hosted, us-east-2, x64 ]
+    container:
+      image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools:pinned
+      options: --init
+
+    steps:
+    - uses: actions/checkout@v4
+
+    
+    - name: Download Neon artifact
+      uses: ./.github/actions/download
+      with:
+        name: neon-${{ runner.os }}-${{ runner.arch }}-release-artifact
+        path: /tmp/neon/
+        prefix: latest
+
+    - name: Run Logical Replication benchmarks
+      uses: ./.github/actions/run-python-test-set
+      with:
+        build_type: ${{ env.BUILD_TYPE }}
+        test_selection: performance/test_logical_replication.py
+        run_in_parallel: false
+        save_perf_report: ${{ env.SAVE_PERF_REPORT }}
+        extra_params: -m remote_cluster --timeout 5400
+        pg_version: ${{ env.DEFAULT_PG_VERSION }}
+      env:
+        VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
+        PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
+        NEON_API_KEY: ${{ secrets.NEON_STAGING_API_KEY }}
+        BENCHMARK_PROJECT_ID_PUB: ${{ vars.BENCHMARK_PROJECT_ID_PUB }}
+        BENCHMARK_PROJECT_ID_SUB: ${{ vars.BENCHMARK_PROJECT_ID_SUB }}
+
+    - name: Run Physical Replication benchmarks
+      uses: ./.github/actions/run-python-test-set
+      with:
+        build_type: ${{ env.BUILD_TYPE }}
+        test_selection: performance/test_physical_replication.py
+        run_in_parallel: false
+        save_perf_report: ${{ env.SAVE_PERF_REPORT }}
+        extra_params: -m remote_cluster --timeout 5400
+        pg_version: ${{ env.DEFAULT_PG_VERSION }}
+      env:
+        VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
+        PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
+        NEON_API_KEY: ${{ secrets.NEON_STAGING_API_KEY }}
+
+    - name: Create Allure report
+      if: ${{ !cancelled() }}
+      uses: ./.github/actions/allure-report-generate
+
+    - name: Post to a Slack channel
+      if: ${{ github.event.schedule && failure() }}
+      uses: slackapi/slack-github-action@v1
+      with:
+        channel-id: "C033QLM5P7D" # dev-staging-stream
+        slack-message: "Periodic replication testing: ${{ job.status }}\n${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
+      env:
+        SLACK_BOT_TOKEN: ${{ secrets.SLACK_BOT_TOKEN }}
+
   generate-matrices:
     if: ${{ github.event.inputs.run_only_pgvector_tests == 'false' || github.event.inputs.run_only_pgvector_tests == null }}
     # Create matrices for the benchmarking jobs, so we run benchmarks on rds only once a week (on Saturday)
     #
     # Available platforms:
-    # - neon-captest-new: Freshly created project (1 CU)
-    # - neon-captest-freetier: Use freetier-sized compute (0.25 CU)
-    # - neon-captest-reuse: Reusing existing project
+    # - neonvm-captest-new: Freshly created project (1 CU)
+    # - neonvm-captest-freetier: Use freetier-sized compute (0.25 CU)
+    # - neonvm-captest-azure-new: Freshly created project (1 CU) in azure region
+    # - neonvm-captest-azure-freetier: Use freetier-sized compute (0.25 CU) in azure region
+    # - neonvm-captest-reuse: Reusing existing project
     # - rds-aurora: Aurora Postgres Serverless v2 with autoscaling from 0.5 to 2 ACUs
     # - rds-postgres: RDS Postgres db.m5.large instance (2 vCPU, 8 GiB) with gp3 EBS storage
     env:
       RUN_AWS_RDS_AND_AURORA: ${{ github.event.inputs.run_AWS_RDS_AND_AURORA || 'false' }}
+      DEFAULT_REGION_ID: ${{ github.event.inputs.region_id || 'aws-us-east-2' }}
     runs-on: ubuntu-22.04
     outputs:
       pgbench-compare-matrix: ${{ steps.pgbench-compare-matrix.outputs.matrix }}
@@ -147,23 +252,36 @@ jobs:
     - name: Generate matrix for pgbench benchmark
       id: pgbench-compare-matrix
       run: |
+        region_id_default=${{ env.DEFAULT_REGION_ID }}
+        runner_default='["self-hosted", "us-east-2", "x64"]'
+        runner_azure='["self-hosted", "eastus2", "x64"]'
+        image_default="369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools:pinned"
         matrix='{
+          "pg_version" : [
+            16
+          ],
+          "region_id" : [
+            "'"$region_id_default"'"
+            ],
           "platform": [
-            "neon-captest-new",
-            "neon-captest-reuse",
+            "neonvm-captest-new",
+            "neonvm-captest-reuse",
             "neonvm-captest-new"
           ],
           "db_size": [ "10gb" ],
-          "include": [{ "platform": "neon-captest-freetier",         "db_size": "3gb"  },
-                      { "platform": "neon-captest-new",              "db_size": "50gb" },
-                      { "platform": "neonvm-captest-freetier",       "db_size": "3gb"  },
-                      { "platform": "neonvm-captest-new",            "db_size": "50gb" },
-                      { "platform": "neonvm-captest-sharding-reuse", "db_size": "50gb" }]
+          "runner": ['"$runner_default"'],
+          "image": [ "'"$image_default"'" ],
+          "include": [{ "pg_version": 16, "region_id": "'"$region_id_default"'", "platform": "neonvm-captest-freetier",       "db_size": "3gb" ,"runner": '"$runner_default"', "image": "'"$image_default"'" },
+                      { "pg_version": 16, "region_id": "'"$region_id_default"'", "platform": "neonvm-captest-new",            "db_size": "10gb","runner": '"$runner_default"', "image": "'"$image_default"'" },
+                      { "pg_version": 16, "region_id": "'"$region_id_default"'", "platform": "neonvm-captest-new",            "db_size": "50gb","runner": '"$runner_default"', "image": "'"$image_default"'" },
+                      { "pg_version": 16, "region_id": "azure-eastus2",          "platform": "neonvm-azure-captest-freetier", "db_size": "3gb" ,"runner": '"$runner_azure"',   "image": "neondatabase/build-tools:pinned" },
+                      { "pg_version": 16, "region_id": "azure-eastus2",          "platform": "neonvm-azure-captest-new",      "db_size": "10gb","runner": '"$runner_azure"',   "image": "neondatabase/build-tools:pinned" },
+                      { "pg_version": 16, "region_id": "azure-eastus2",          "platform": "neonvm-azure-captest-new",      "db_size": "50gb","runner": '"$runner_azure"',   "image": "neondatabase/build-tools:pinned" },
+                      { "pg_version": 16, "region_id": "'"$region_id_default"'", "platform": "neonvm-captest-sharding-reuse", "db_size": "50gb","runner": '"$runner_default"', "image": "'"$image_default"'" }]
         }'
 
         if [ "$(date +%A)" = "Saturday" ]; then
-          matrix=$(echo "$matrix" | jq '.include += [{ "platform": "rds-postgres", "db_size": "10gb"},
-                                                     { "platform": "rds-aurora",   "db_size": "50gb"}]')
+          matrix=$(echo "$matrix" | jq '.include += [{ "pg_version": 14, "region_id": "'"$region_id_default"'", "platform": "rds-postgres", "db_size": "10gb","runner": '"$runner_default"', "image": "'"$image_default"'" }]')
         fi
 
         echo "matrix=$(echo "$matrix" | jq --compact-output '.')" >> $GITHUB_OUTPUT
@@ -173,7 +291,7 @@ jobs:
       run: |
         matrix='{
           "platform": [
-            "neon-captest-reuse"
+            "neonvm-captest-reuse"
           ]
         }'
 
@@ -189,7 +307,7 @@ jobs:
       run: |
         matrix='{
           "platform": [
-            "neon-captest-reuse"
+            "neonvm-captest-reuse"
           ],
           "scale": [
             "10"
@@ -206,6 +324,10 @@ jobs:
   pgbench-compare:
     if: ${{ github.event.inputs.run_only_pgvector_tests == 'false' || github.event.inputs.run_only_pgvector_tests == null }}
     needs: [ generate-matrices ]
+    permissions:
+      contents: write
+      statuses: write
+      id-token: write # Required for OIDC authentication in azure runners
 
     strategy:
       fail-fast: false
@@ -215,15 +337,15 @@ jobs:
       TEST_PG_BENCH_DURATIONS_MATRIX: "60m"
       TEST_PG_BENCH_SCALES_MATRIX: ${{ matrix.db_size }}
       POSTGRES_DISTRIB_DIR: /tmp/neon/pg_install
-      DEFAULT_PG_VERSION: 14
+      DEFAULT_PG_VERSION: ${{ matrix.pg_version }}
       TEST_OUTPUT: /tmp/test_output
       BUILD_TYPE: remote
       SAVE_PERF_REPORT: ${{ github.event.inputs.save_perf_report || ( github.ref_name == 'main' ) }}
       PLATFORM: ${{ matrix.platform }}
 
-    runs-on: [ self-hosted, us-east-2, x64 ]
+    runs-on: ${{ matrix.runner }}
     container:
-      image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools:pinned
+      image: ${{ matrix.image }}
       options: --init
 
     # Increase timeout to 8h, default timeout is 6h
@@ -232,6 +354,13 @@ jobs:
     steps:
     - uses: actions/checkout@v4
 
+    - name: Configure AWS credentials # necessary on Azure runners
+      uses: aws-actions/configure-aws-credentials@v4
+      with:
+        aws-region: eu-central-1
+        role-to-assume: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+        role-duration-seconds: 18000 # 5 hours
+        
     - name: Download Neon artifact
       uses: ./.github/actions/download
       with:
@@ -239,33 +368,27 @@ jobs:
         path: /tmp/neon/
         prefix: latest
 
-    - name: Add Postgres binaries to PATH
-      run: |
-        ${POSTGRES_DISTRIB_DIR}/v${DEFAULT_PG_VERSION}/bin/pgbench --version
-        echo "${POSTGRES_DISTRIB_DIR}/v${DEFAULT_PG_VERSION}/bin" >> $GITHUB_PATH
-
     - name: Create Neon Project
-      if: contains(fromJson('["neon-captest-new", "neon-captest-freetier", "neonvm-captest-new", "neonvm-captest-freetier"]'), matrix.platform)
+      if: contains(fromJson('["neonvm-captest-new", "neonvm-captest-freetier", "neonvm-azure-captest-freetier", "neonvm-azure-captest-new"]'), matrix.platform)
       id: create-neon-project
       uses: ./.github/actions/neon-project-create
       with:
-        region_id: ${{ github.event.inputs.region_id || 'aws-us-east-2' }}
+        region_id: ${{ matrix.region_id }}
         postgres_version: ${{ env.DEFAULT_PG_VERSION }}
         api_key: ${{ secrets.NEON_STAGING_API_KEY }}
-        compute_units: ${{ (matrix.platform == 'neon-captest-freetier' && '[0.25, 0.25]') || '[1, 1]' }}
-        provisioner: ${{ (contains(matrix.platform, 'neonvm-') && 'k8s-neonvm') || 'k8s-pod' }}
+        compute_units: ${{ (contains(matrix.platform, 'captest-freetier') && '[0.25, 0.25]') || '[1, 1]' }}
 
     - name: Set up Connection String
       id: set-up-connstr
       run: |
         case "${PLATFORM}" in
-          neon-captest-reuse)
+          neonvm-captest-reuse)
             CONNSTR=${{ secrets.BENCHMARK_CAPTEST_CONNSTR }}
             ;;
           neonvm-captest-sharding-reuse)
             CONNSTR=${{ secrets.BENCHMARK_CAPTEST_SHARDING_CONNSTR }}
             ;;
-          neon-captest-new | neon-captest-freetier | neonvm-captest-new | neonvm-captest-freetier)
+          neonvm-captest-new | neonvm-captest-freetier | neonvm-azure-captest-new | neonvm-azure-captest-freetier)
             CONNSTR=${{ steps.create-neon-project.outputs.dsn }}
             ;;
           rds-aurora)
@@ -282,16 +405,6 @@ jobs:
 
         echo "connstr=${CONNSTR}" >> $GITHUB_OUTPUT
 
-        QUERIES=("SELECT version()")
-        if [[ "${PLATFORM}" = "neon"* ]]; then
-          QUERIES+=("SHOW neon.tenant_id")
-          QUERIES+=("SHOW neon.timeline_id")
-        fi
-
-        for q in "${QUERIES[@]}"; do
-          psql ${CONNSTR} -c "${q}"
-        done
-
     - name: Benchmark init
       uses: ./.github/actions/run-python-test-set
       with:
@@ -300,6 +413,7 @@ jobs:
         run_in_parallel: false
         save_perf_report: ${{ env.SAVE_PERF_REPORT }}
         extra_params: -m remote_cluster --timeout 21600 -k test_pgbench_remote_init
+        pg_version: ${{ env.DEFAULT_PG_VERSION }}
       env:
         BENCHMARK_CONNSTR: ${{ steps.set-up-connstr.outputs.connstr }}
         VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
@@ -313,6 +427,7 @@ jobs:
         run_in_parallel: false
         save_perf_report: ${{ env.SAVE_PERF_REPORT }}
         extra_params: -m remote_cluster --timeout 21600 -k test_pgbench_remote_simple_update
+        pg_version: ${{ env.DEFAULT_PG_VERSION }}
       env:
         BENCHMARK_CONNSTR: ${{ steps.set-up-connstr.outputs.connstr }}
         VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
@@ -326,6 +441,7 @@ jobs:
         run_in_parallel: false
         save_perf_report: ${{ env.SAVE_PERF_REPORT }}
         extra_params: -m remote_cluster --timeout 21600 -k test_pgbench_remote_select_only
+        pg_version: ${{ env.DEFAULT_PG_VERSION }}
       env:
         BENCHMARK_CONNSTR: ${{ steps.set-up-connstr.outputs.connstr }}
         VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
@@ -352,6 +468,21 @@ jobs:
         SLACK_BOT_TOKEN: ${{ secrets.SLACK_BOT_TOKEN }}
 
   pgbench-pgvector:
+    permissions:
+      contents: write
+      statuses: write
+      id-token: write # Required for OIDC authentication in azure runners
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - PLATFORM: "neonvm-captest-pgvector"
+            RUNNER: [ self-hosted, us-east-2, x64 ]
+            IMAGE: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools:pinned
+          - PLATFORM: "azure-captest-pgvector"
+            RUNNER: [ self-hosted, eastus2, x64 ]
+            IMAGE: neondatabase/build-tools:pinned
+
     env:
       TEST_PG_BENCH_DURATIONS_MATRIX: "15m"
       TEST_PG_BENCH_SCALES_MATRIX: "1"
@@ -359,43 +490,60 @@ jobs:
       DEFAULT_PG_VERSION: 16
       TEST_OUTPUT: /tmp/test_output
       BUILD_TYPE: remote
+      LD_LIBRARY_PATH: /home/nonroot/pg/usr/lib/x86_64-linux-gnu
       SAVE_PERF_REPORT: ${{ github.event.inputs.save_perf_report || ( github.ref_name == 'main' ) }}
-      PLATFORM: "neon-captest-pgvector"
+      PLATFORM: ${{ matrix.PLATFORM }}
 
-    runs-on: [ self-hosted, us-east-2, x64 ]
+    runs-on: ${{ matrix.RUNNER }}
     container:
-      image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools:pinned
+      image: ${{ matrix.IMAGE }}
       options: --init
 
     steps:
     - uses: actions/checkout@v4
 
-    - name: Download Neon artifact
-      uses: ./.github/actions/download
-      with:
-        name: neon-${{ runner.os }}-${{ runner.arch }}-release-artifact
-        path: /tmp/neon/
-        prefix: latest
-
-    - name: Add Postgres binaries to PATH
+    # until https://github.com/neondatabase/neon/issues/8275 is fixed we temporarily install postgresql-16
+    # instead of using Neon artifacts containing pgbench
+    - name: Install postgresql-16 where pytest expects it
       run: |
-        ${POSTGRES_DISTRIB_DIR}/v${DEFAULT_PG_VERSION}/bin/pgbench --version
-        echo "${POSTGRES_DISTRIB_DIR}/v${DEFAULT_PG_VERSION}/bin" >> $GITHUB_PATH
+        cd /home/nonroot
+        wget -q https://apt.postgresql.org/pub/repos/apt/pool/main/p/postgresql-16/libpq5_16.4-1.pgdg110%2B1_amd64.deb
+        wget -q https://apt.postgresql.org/pub/repos/apt/pool/main/p/postgresql-16/postgresql-client-16_16.4-1.pgdg110%2B1_amd64.deb
+        wget -q https://apt.postgresql.org/pub/repos/apt/pool/main/p/postgresql-16/postgresql-16_16.4-1.pgdg110%2B1_amd64.deb 
+        dpkg -x libpq5_16.4-1.pgdg110+1_amd64.deb pg
+        dpkg -x postgresql-client-16_16.4-1.pgdg110+1_amd64.deb pg
+        dpkg -x postgresql-16_16.4-1.pgdg110+1_amd64.deb pg
+        mkdir -p /tmp/neon/pg_install/v16/bin
+        ln -s /home/nonroot/pg/usr/lib/postgresql/16/bin/pgbench /tmp/neon/pg_install/v16/bin/pgbench  
+        ln -s /home/nonroot/pg/usr/lib/postgresql/16/bin/psql /tmp/neon/pg_install/v16/bin/psql  
+        ln -s /home/nonroot/pg/usr/lib/x86_64-linux-gnu /tmp/neon/pg_install/v16/lib 
+        /tmp/neon/pg_install/v16/bin/pgbench --version
+        /tmp/neon/pg_install/v16/bin/psql --version
 
     - name: Set up Connection String
       id: set-up-connstr
       run: |
-        CONNSTR=${{ secrets.BENCHMARK_PGVECTOR_CONNSTR }}
-        
-        echo "connstr=${CONNSTR}" >> $GITHUB_OUTPUT
+        case "${PLATFORM}" in
+          neonvm-captest-pgvector)
+            CONNSTR=${{ secrets.BENCHMARK_PGVECTOR_CONNSTR }}
+            ;;
+          azure-captest-pgvector)
+            CONNSTR=${{ secrets.BENCHMARK_PGVECTOR_CONNSTR_AZURE }}
+            ;;
+          *)
+            echo >&2 "Unknown PLATFORM=${PLATFORM}"
+            exit 1
+            ;;
+        esac
 
-        QUERIES=("SELECT version()")
-        QUERIES+=("SHOW neon.tenant_id")
-        QUERIES+=("SHOW neon.timeline_id")
+        echo "connstr=${CONNSTR}" >> $GITHUB_OUTPUT
         
-        for q in "${QUERIES[@]}"; do
-          psql ${CONNSTR} -c "${q}"
-        done
+    - name: Configure AWS credentials # necessary on Azure runners to read/write from/to S3
+      uses: aws-actions/configure-aws-credentials@v4
+      with:
+        aws-region: eu-central-1
+        role-to-assume: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+        role-duration-seconds: 18000 # 5 hours
 
     - name: Benchmark pgvector hnsw indexing
       uses: ./.github/actions/run-python-test-set
@@ -405,6 +553,7 @@ jobs:
         run_in_parallel: false
         save_perf_report: ${{ env.SAVE_PERF_REPORT }}
         extra_params: -m remote_cluster --timeout 21600 -k test_pgvector_indexing
+        pg_version: ${{ env.DEFAULT_PG_VERSION }}
       env:
         VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
         PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
@@ -417,7 +566,8 @@ jobs:
         test_selection: performance/test_perf_pgvector_queries.py
         run_in_parallel: false
         save_perf_report: ${{ env.SAVE_PERF_REPORT }}
-        extra_params: -m remote_cluster --timeout 21600 
+        extra_params: -m remote_cluster --timeout 21600
+        pg_version: ${{ env.DEFAULT_PG_VERSION }}
       env:
         BENCHMARK_CONNSTR: ${{ steps.set-up-connstr.outputs.connstr }}
         VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
@@ -432,11 +582,10 @@ jobs:
       uses: slackapi/slack-github-action@v1
       with:
         channel-id: "C033QLM5P7D" # dev-staging-stream
-        slack-message: "Periodic perf testing neon-captest-pgvector: ${{ job.status }}\n${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
+        slack-message: "Periodic perf testing ${PLATFORM}: ${{ job.status }}\n${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
       env:
         SLACK_BOT_TOKEN: ${{ secrets.SLACK_BOT_TOKEN }}
 
-
   clickbench-compare:
     # ClichBench DB for rds-aurora and rds-Postgres deployed to the same clusters
     # we use for performance testing in pgbench-compare.
@@ -477,16 +626,11 @@ jobs:
         path: /tmp/neon/
         prefix: latest
 
-    - name: Add Postgres binaries to PATH
-      run: |
-        ${POSTGRES_DISTRIB_DIR}/v${DEFAULT_PG_VERSION}/bin/pgbench --version
-        echo "${POSTGRES_DISTRIB_DIR}/v${DEFAULT_PG_VERSION}/bin" >> $GITHUB_PATH
-
     - name: Set up Connection String
       id: set-up-connstr
       run: |
         case "${PLATFORM}" in
-          neon-captest-reuse)
+          neonvm-captest-reuse)
             CONNSTR=${{ secrets.BENCHMARK_CAPTEST_CLICKBENCH_10M_CONNSTR }}
             ;;
           rds-aurora)
@@ -496,23 +640,13 @@ jobs:
             CONNSTR=${{ secrets.BENCHMARK_RDS_POSTGRES_CLICKBENCH_10M_CONNSTR }}
             ;;
           *)
-            echo >&2 "Unknown PLATFORM=${PLATFORM}. Allowed only 'neon-captest-reuse', 'rds-aurora', or 'rds-postgres'"
+            echo >&2 "Unknown PLATFORM=${PLATFORM}. Allowed only 'neonvm-captest-reuse', 'rds-aurora', or 'rds-postgres'"
             exit 1
             ;;
         esac
 
         echo "connstr=${CONNSTR}" >> $GITHUB_OUTPUT
 
-        QUERIES=("SELECT version()")
-        if [[ "${PLATFORM}" = "neon"* ]]; then
-          QUERIES+=("SHOW neon.tenant_id")
-          QUERIES+=("SHOW neon.timeline_id")
-        fi
-
-        for q in "${QUERIES[@]}"; do
-          psql ${CONNSTR} -c "${q}"
-        done
-
     - name: ClickBench benchmark
       uses: ./.github/actions/run-python-test-set
       with:
@@ -580,15 +714,10 @@ jobs:
         path: /tmp/neon/
         prefix: latest
 
-    - name: Add Postgres binaries to PATH
-      run: |
-        ${POSTGRES_DISTRIB_DIR}/v${DEFAULT_PG_VERSION}/bin/pgbench --version
-        echo "${POSTGRES_DISTRIB_DIR}/v${DEFAULT_PG_VERSION}/bin" >> $GITHUB_PATH
-
     - name: Get Connstring Secret Name
       run: |
         case "${PLATFORM}" in
-          neon-captest-reuse)
+          neonvm-captest-reuse)
             ENV_PLATFORM=CAPTEST_TPCH
             ;;
           rds-aurora)
@@ -598,7 +727,7 @@ jobs:
             ENV_PLATFORM=RDS_AURORA_TPCH
             ;;
           *)
-            echo >&2 "Unknown PLATFORM=${PLATFORM}. Allowed only 'neon-captest-reuse', 'rds-aurora', or 'rds-postgres'"
+            echo >&2 "Unknown PLATFORM=${PLATFORM}. Allowed only 'neonvm-captest-reuse', 'rds-aurora', or 'rds-postgres'"
             exit 1
             ;;
         esac
@@ -613,16 +742,6 @@ jobs:
 
         echo "connstr=${CONNSTR}" >> $GITHUB_OUTPUT
 
-        QUERIES=("SELECT version()")
-        if [[ "${PLATFORM}" = "neon"* ]]; then
-          QUERIES+=("SHOW neon.tenant_id")
-          QUERIES+=("SHOW neon.timeline_id")
-        fi
-
-        for q in "${QUERIES[@]}"; do
-          psql ${CONNSTR} -c "${q}"
-        done
-
     - name: Run TPC-H benchmark
       uses: ./.github/actions/run-python-test-set
       with:
@@ -681,16 +800,11 @@ jobs:
         path: /tmp/neon/
         prefix: latest
 
-    - name: Add Postgres binaries to PATH
-      run: |
-        ${POSTGRES_DISTRIB_DIR}/v${DEFAULT_PG_VERSION}/bin/pgbench --version
-        echo "${POSTGRES_DISTRIB_DIR}/v${DEFAULT_PG_VERSION}/bin" >> $GITHUB_PATH
-
     - name: Set up Connection String
       id: set-up-connstr
       run: |
         case "${PLATFORM}" in
-          neon-captest-reuse)
+          neonvm-captest-reuse)
             CONNSTR=${{ secrets.BENCHMARK_USER_EXAMPLE_CAPTEST_CONNSTR }}
             ;;
           rds-aurora)
@@ -700,23 +814,13 @@ jobs:
             CONNSTR=${{ secrets.BENCHMARK_USER_EXAMPLE_RDS_POSTGRES_CONNSTR }}
             ;;
           *)
-            echo >&2 "Unknown PLATFORM=${PLATFORM}. Allowed only 'neon-captest-reuse', 'rds-aurora', or 'rds-postgres'"
+            echo >&2 "Unknown PLATFORM=${PLATFORM}. Allowed only 'neonvm-captest-reuse', 'rds-aurora', or 'rds-postgres'"
             exit 1
             ;;
         esac
 
         echo "connstr=${CONNSTR}" >> $GITHUB_OUTPUT
 
-        QUERIES=("SELECT version()")
-        if [[ "${PLATFORM}" = "neon"* ]]; then
-          QUERIES+=("SHOW neon.tenant_id")
-          QUERIES+=("SHOW neon.timeline_id")
-        fi
-
-        for q in "${QUERIES[@]}"; do
-          psql ${CONNSTR} -c "${q}"
-        done
-
     - name: Run user examples
       uses: ./.github/actions/run-python-test-set
       with:
@@ -725,6 +829,7 @@ jobs:
         run_in_parallel: false
         save_perf_report: ${{ env.SAVE_PERF_REPORT }}
         extra_params: -m remote_cluster --timeout 21600 -k test_user_examples
+        pg_version: ${{ env.DEFAULT_PG_VERSION }}
       env:
         VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
         PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"

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

@@ -63,25 +63,34 @@ jobs:
           mkdir -p /tmp/.docker-custom
           echo DOCKER_CONFIG=/tmp/.docker-custom >> $GITHUB_ENV
 
-      - uses: docker/setup-buildx-action@v2
+      - uses: docker/setup-buildx-action@v3
+        with:
+          cache-binary: false
 
-      - uses: docker/login-action@v2
+      - uses: docker/login-action@v3
         with:
           username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
           password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
 
-      - uses: docker/build-push-action@v4
+      - uses: docker/login-action@v3
+        with:
+          registry: cache.neon.build
+          username: ${{ secrets.NEON_CI_DOCKERCACHE_USERNAME }}
+          password: ${{ secrets.NEON_CI_DOCKERCACHE_PASSWORD }}
+
+      - uses: docker/build-push-action@v6
         with:
           context: .
           provenance: false
           push: true
           pull: true
           file: Dockerfile.build-tools
-          cache-from: type=registry,ref=neondatabase/build-tools:cache-${{ matrix.arch }}
-          cache-to: ${{ github.ref_name == 'main' && format('type=registry,ref=neondatabase/build-tools:cache-{0},mode=max', matrix.arch) || '' }}
+          cache-from: type=registry,ref=cache.neon.build/build-tools:cache-${{ matrix.arch }}
+          cache-to: ${{ github.ref_name == 'main' && format('type=registry,ref=cache.neon.build/build-tools:cache-{0},mode=max', matrix.arch) || '' }}
           tags: neondatabase/build-tools:${{ inputs.image-tag }}-${{ matrix.arch }}
 
       - name: Remove custom docker config directory
+        if: always()
         run: |
           rm -rf /tmp/.docker-custom
 

.github/workflows/build_and_test.yml

@@ -30,7 +30,7 @@ jobs:
     if: ${{ !contains(github.event.pull_request.labels.*.name, 'run-no-ci') }}
     uses: ./.github/workflows/check-permissions.yml
     with:
-      github-event-name: ${{ github.event_name}}
+      github-event-name: ${{ github.event_name }}
 
   cancel-previous-e2e-tests:
     needs: [ check-permissions ]
@@ -125,7 +125,11 @@ jobs:
 
   check-codestyle-rust:
     needs: [ check-permissions, build-build-tools-image ]
-    runs-on: [ self-hosted, gen3, small ]
+    strategy:
+      matrix:
+        arch: [ x64, arm64 ]
+    runs-on: ${{ fromJson(format('["self-hosted", "{0}"]', matrix.arch == 'arm64' && 'small-arm64' || 'small')) }}
+
     container:
       image: ${{ needs.build-build-tools-image.outputs.image }}
       credentials:
@@ -193,284 +197,30 @@ jobs:
         if: ${{ !cancelled() }}
         run: cargo deny check --hide-inclusion-graph
 
-  build-neon:
-    needs: [ check-permissions, tag, build-build-tools-image ]
-    runs-on: [ self-hosted, gen3, large ]
-    container:
-      image: ${{ needs.build-build-tools-image.outputs.image }}
-      credentials:
-        username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
-        password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
-      # 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
+  build-and-test-locally:
+    needs: [ tag, build-build-tools-image ]
     strategy:
       fail-fast: false
       matrix:
-        build_type: [ debug, release ]
-    env:
-      BUILD_TYPE: ${{ matrix.build_type }}
-      GIT_VERSION: ${{ github.event.pull_request.head.sha || github.sha }}
-      BUILD_TAG: ${{ needs.tag.outputs.build-tag }}
-
-    steps:
-      - name: Fix git ownership
-        run: |
-          # Workaround for `fatal: detected dubious ownership in repository at ...`
-          #
-          # Use both ${{ github.workspace }} and ${GITHUB_WORKSPACE} because they're different on host and in containers
-          #   Ref https://github.com/actions/checkout/issues/785
-          #
-          git config --global --add safe.directory ${{ github.workspace }}
-          git config --global --add safe.directory ${GITHUB_WORKSPACE}
-          for r in 14 15 16; do
-            git config --global --add safe.directory "${{ github.workspace }}/vendor/postgres-v$r"
-            git config --global --add safe.directory "${GITHUB_WORKSPACE}/vendor/postgres-v$r"
-          done
-
-      - name: Checkout
-        uses: actions/checkout@v4
-        with:
-          submodules: true
-          fetch-depth: 1
-
-      - name: Set pg 14 revision for caching
-        id: pg_v14_rev
-        run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-v14) >> $GITHUB_OUTPUT
-
-      - name: Set pg 15 revision for caching
-        id: pg_v15_rev
-        run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-v15) >> $GITHUB_OUTPUT
-
-      - name: Set pg 16 revision for caching
-        id: pg_v16_rev
-        run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-v16) >> $GITHUB_OUTPUT
-
-      # Set some environment variables used by all the steps.
-      #
-      # CARGO_FLAGS is extra options to pass to "cargo build", "cargo test" etc.
-      #   It also includes --features, if any
-      #
-      # CARGO_FEATURES is passed to "cargo metadata". It is separate from CARGO_FLAGS,
-      #   because "cargo metadata" doesn't accept --release or --debug options
-      #
-      # We run tests with addtional features, that are turned off by default (e.g. in release builds), see
-      # corresponding Cargo.toml files for their descriptions.
-      - name: Set env variables
-        run: |
-          CARGO_FEATURES="--features testing"
-          if [[ $BUILD_TYPE == "debug" ]]; then
-            cov_prefix="scripts/coverage --profraw-prefix=$GITHUB_JOB --dir=/tmp/coverage run"
-            CARGO_FLAGS="--locked"
-          elif [[ $BUILD_TYPE == "release" ]]; then
-            cov_prefix=""
-            CARGO_FLAGS="--locked --release"
-          fi
-          {
-            echo "cov_prefix=${cov_prefix}"
-            echo "CARGO_FEATURES=${CARGO_FEATURES}"
-            echo "CARGO_FLAGS=${CARGO_FLAGS}"
-            echo "CARGO_HOME=${GITHUB_WORKSPACE}/.cargo"
-          } >> $GITHUB_ENV
-
-      # Disabled for now
-      # Don't include the ~/.cargo/registry/src directory. It contains just
-      # uncompressed versions of the crates in ~/.cargo/registry/cache
-      # directory, and it's faster to let 'cargo' to rebuild it from the
-      # compressed crates.
-#      - name: Cache cargo deps
-#        id: cache_cargo
-#        uses: actions/cache@v4
-#        with:
-#          path: |
-#            ~/.cargo/registry/
-#            !~/.cargo/registry/src
-#            ~/.cargo/git/
-#            target/
-#          # Fall back to older versions of the key, if no cache for current Cargo.lock was found
-#          key: |
-#            v1-${{ runner.os }}-${{ runner.arch }}-${{ matrix.build_type }}-cargo-${{ hashFiles('rust-toolchain.toml') }}-${{ hashFiles('Cargo.lock') }}
-#            v1-${{ runner.os }}-${{ runner.arch }}-${{ matrix.build_type }}-cargo-${{ hashFiles('rust-toolchain.toml') }}-
-
-      - name: Cache postgres v14 build
-        id: cache_pg_14
-        uses: actions/cache@v4
-        with:
-          path: pg_install/v14
-          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ matrix.build_type }}-pg-${{ steps.pg_v14_rev.outputs.pg_rev }}-${{ hashFiles('Makefile', 'Dockerfile.build-tools') }}
-
-      - name: Cache postgres v15 build
-        id: cache_pg_15
-        uses: actions/cache@v4
-        with:
-          path: pg_install/v15
-          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ matrix.build_type }}-pg-${{ steps.pg_v15_rev.outputs.pg_rev }}-${{ hashFiles('Makefile', 'Dockerfile.build-tools') }}
-
-      - name: Cache postgres v16 build
-        id: cache_pg_16
-        uses: actions/cache@v4
-        with:
-          path: pg_install/v16
-          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ matrix.build_type }}-pg-${{ steps.pg_v16_rev.outputs.pg_rev }}-${{ hashFiles('Makefile', 'Dockerfile.build-tools') }}
-
-      - name: Build postgres v14
-        if: steps.cache_pg_14.outputs.cache-hit != 'true'
-        run: mold -run make postgres-v14 -j$(nproc)
-
-      - name: Build postgres v15
-        if: steps.cache_pg_15.outputs.cache-hit != 'true'
-        run: mold -run make postgres-v15 -j$(nproc)
-
-      - name: Build postgres v16
-        if: steps.cache_pg_16.outputs.cache-hit != 'true'
-        run: mold -run make postgres-v16 -j$(nproc)
-
-      - name: Build neon extensions
-        run: mold -run make neon-pg-ext -j$(nproc)
-
-      - name: Build walproposer-lib
-        run: mold -run make walproposer-lib -j$(nproc)
-
-      - name: Run cargo build
-        run: |
-          ${cov_prefix} mold -run cargo build $CARGO_FLAGS $CARGO_FEATURES --bins --tests
-
-      # Do install *before* running rust tests because they might recompile the
-      # binaries with different features/flags.
-      - name: Install rust binaries
-        run: |
-          # Install target binaries
-          mkdir -p /tmp/neon/bin/
-          binaries=$(
-            ${cov_prefix} cargo metadata $CARGO_FEATURES --format-version=1 --no-deps |
-            jq -r '.packages[].targets[] | select(.kind | index("bin")) | .name'
-          )
-          for bin in $binaries; do
-            SRC=target/$BUILD_TYPE/$bin
-            DST=/tmp/neon/bin/$bin
-            cp "$SRC" "$DST"
-          done
-
-          # Install test executables and write list of all binaries (for code coverage)
-          if [[ $BUILD_TYPE == "debug" ]]; then
-            # Keep bloated coverage data files away from the rest of the artifact
-            mkdir -p /tmp/coverage/
-
-            mkdir -p /tmp/neon/test_bin/
-
-            test_exe_paths=$(
-              ${cov_prefix} cargo test $CARGO_FLAGS $CARGO_FEATURES --message-format=json --no-run |
-              jq -r '.executable | select(. != null)'
-            )
-            for bin in $test_exe_paths; do
-              SRC=$bin
-              DST=/tmp/neon/test_bin/$(basename $bin)
-
-              # We don't need debug symbols for code coverage, so strip them out to make
-              # the artifact smaller.
-              strip "$SRC" -o "$DST"
-              echo "$DST" >> /tmp/coverage/binaries.list
-            done
-
-            for bin in $binaries; do
-              echo "/tmp/neon/bin/$bin" >> /tmp/coverage/binaries.list
-            done
-          fi
-
-      - name: Run rust tests
-        env:
-          NEXTEST_RETRIES: 3
-        run: |
-          #nextest does not yet support running doctests
-          cargo test --doc $CARGO_FLAGS $CARGO_FEATURES
-
-          for io_engine in std-fs tokio-epoll-uring ; do
-            NEON_PAGESERVER_UNIT_TEST_VIRTUAL_FILE_IOENGINE=$io_engine ${cov_prefix} cargo nextest run $CARGO_FLAGS $CARGO_FEATURES
-          done
-
-          # Run separate tests for real S3
-          export ENABLE_REAL_S3_REMOTE_STORAGE=nonempty
-          export REMOTE_STORAGE_S3_BUCKET=neon-github-ci-tests
-          export REMOTE_STORAGE_S3_REGION=eu-central-1
-          ${cov_prefix} cargo nextest run $CARGO_FLAGS $CARGO_FEATURES -E 'package(remote_storage)' -E 'test(test_real_s3)'
-
-          # Run separate tests for real Azure Blob Storage
-          # XXX: replace region with `eu-central-1`-like region
-          export ENABLE_REAL_AZURE_REMOTE_STORAGE=y
-          export AZURE_STORAGE_ACCOUNT="${{ secrets.AZURE_STORAGE_ACCOUNT_DEV }}"
-          export AZURE_STORAGE_ACCESS_KEY="${{ secrets.AZURE_STORAGE_ACCESS_KEY_DEV }}"
-          export REMOTE_STORAGE_AZURE_CONTAINER="${{ vars.REMOTE_STORAGE_AZURE_CONTAINER }}"
-          export REMOTE_STORAGE_AZURE_REGION="${{ vars.REMOTE_STORAGE_AZURE_REGION }}"
-          ${cov_prefix} cargo nextest run $CARGO_FLAGS $CARGO_FEATURES -E 'package(remote_storage)' -E 'test(test_real_azure)'
-
-      - name: Install postgres binaries
-        run: cp -a pg_install /tmp/neon/pg_install
-
-      - name: Upload Neon artifact
-        uses: ./.github/actions/upload
-        with:
-          name: neon-${{ runner.os }}-${{ runner.arch }}-${{ matrix.build_type }}-artifact
-          path: /tmp/neon
-
-      # XXX: keep this after the binaries.list is formed, so the coverage can properly work later
-      - name: Merge and upload coverage data
-        if: matrix.build_type == 'debug'
-        uses: ./.github/actions/save-coverage-data
-
-  regress-tests:
-    needs: [ check-permissions, build-neon, build-build-tools-image, tag ]
-    runs-on: [ self-hosted, gen3, large ]
-    container:
-      image: ${{ needs.build-build-tools-image.outputs.image }}
-      credentials:
-        username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
-        password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
-      # for changed limits, see comments on `options:` earlier in this file
-      options: --init --shm-size=512mb --ulimit memlock=67108864:67108864
-    strategy:
-      fail-fast: false
-      matrix:
-        build_type: [ debug, release ]
-        pg_version: [ v14, v15, v16 ]
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-        with:
-          submodules: true
-          fetch-depth: 1
-
-      - name: Pytest regression tests
-        uses: ./.github/actions/run-python-test-set
-        timeout-minutes: 60
-        with:
-          build_type: ${{ matrix.build_type }}
-          test_selection: regress
-          needs_postgres_source: true
-          run_with_real_s3: true
-          real_s3_bucket: neon-github-ci-tests
-          real_s3_region: eu-central-1
-          rerun_flaky: true
-          pg_version: ${{ matrix.pg_version }}
-        env:
-          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: tokio-epoll-uring
-          PAGESERVER_GET_VECTORED_IMPL: vectored
-          PAGESERVER_GET_IMPL: vectored
-          PAGESERVER_VALIDATE_VEC_GET: true
-
-      # Temporary disable this step until we figure out why it's so flaky
-      # Ref https://github.com/neondatabase/neon/issues/4540
-      - name: Merge and upload coverage data
-        if: |
-          false &&
-          matrix.build_type == 'debug' && matrix.pg_version == 'v14'
-        uses: ./.github/actions/save-coverage-data
+        arch: [ x64 ]
+        # Do not build or run tests in debug for release branches
+        build-type: ${{ fromJson((startsWith(github.ref_name, 'release' && github.event_name == 'push')) && '["release"]' || '["debug", "release"]') }}
+        include:
+          - build-type: release
+            arch: arm64
+    uses: ./.github/workflows/_build-and-test-locally.yml
+    with:
+      arch: ${{ matrix.arch }}
+      build-tools-image: ${{ needs.build-build-tools-image.outputs.image }}
+      build-tag: ${{ needs.tag.outputs.build-tag }}
+      build-type: ${{ matrix.build-type }}
+      # Run tests on all Postgres versions in release builds and only on the latest version in debug builds
+      pg-versions: ${{ matrix.build-type == 'release' && '["v14", "v15", "v16"]' || '["v16"]' }}
+    secrets: inherit
 
+  # Keep `benchmarks` job outside of `build-and-test-locally` workflow to make job failures non-blocking
   get-benchmarks-durations:
+    if: github.ref_name == 'main' || contains(github.event.pull_request.labels.*.name, 'run-benchmarks')
     outputs:
       json: ${{ steps.get-benchmark-durations.outputs.json }}
     needs: [ check-permissions, build-build-tools-image ]
@@ -481,7 +231,6 @@ jobs:
         username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
         password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
       options: --init
-    if: github.ref_name == 'main' || contains(github.event.pull_request.labels.*.name, 'run-benchmarks')
     steps:
       - name: Checkout
         uses: actions/checkout@v4
@@ -506,7 +255,8 @@ jobs:
           echo "json=$(jq --compact-output '.' /tmp/benchmark_durations.json)" >> $GITHUB_OUTPUT
 
   benchmarks:
-    needs: [ check-permissions, build-neon, build-build-tools-image, get-benchmarks-durations ]
+    if: github.ref_name == 'main' || contains(github.event.pull_request.labels.*.name, 'run-benchmarks')
+    needs: [ check-permissions, build-and-test-locally, build-build-tools-image, get-benchmarks-durations ]
     runs-on: [ self-hosted, gen3, small ]
     container:
       image: ${{ needs.build-build-tools-image.outputs.image }}
@@ -515,7 +265,6 @@ jobs:
         password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
       # for changed limits, see comments on `options:` earlier in this file
       options: --init --shm-size=512mb --ulimit memlock=67108864:67108864
-    if: github.ref_name == 'main' || contains(github.event.pull_request.labels.*.name, 'run-benchmarks')
     strategy:
       fail-fast: false
       matrix:
@@ -540,9 +289,6 @@ jobs:
           PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
           TEST_RESULT_CONNSTR: "${{ secrets.REGRESS_TEST_RESULT_CONNSTR_NEW }}"
           PAGESERVER_VIRTUAL_FILE_IO_ENGINE: tokio-epoll-uring
-          PAGESERVER_GET_VECTORED_IMPL: vectored
-          PAGESERVER_GET_IMPL: vectored
-          PAGESERVER_VALIDATE_VEC_GET: false
       # XXX: no coverage data handling here, since benchmarks are run on release builds,
       # while coverage is currently collected for the debug ones
 
@@ -563,7 +309,7 @@ jobs:
         SLACK_BOT_TOKEN: ${{ secrets.SLACK_BOT_TOKEN }}
 
   create-test-report:
-    needs: [ check-permissions, regress-tests, coverage-report, benchmarks, build-build-tools-image ]
+    needs: [ check-permissions, build-and-test-locally, coverage-report, build-build-tools-image, benchmarks ]
     if: ${{ !cancelled() && contains(fromJSON('["skipped", "success"]'), needs.check-permissions.result) }}
     outputs:
       report-url: ${{ steps.create-allure-report.outputs.report-url }}
@@ -614,7 +360,7 @@ jobs:
             })
 
   coverage-report:
-    needs: [ check-permissions, regress-tests, build-build-tools-image ]
+    needs: [ check-permissions, build-build-tools-image, build-and-test-locally ]
     runs-on: [ self-hosted, gen3, small ]
     container:
       image: ${{ needs.build-build-tools-image.outputs.image }}
@@ -744,14 +490,22 @@ jobs:
         run: |
           mkdir -p .docker-custom
           echo DOCKER_CONFIG=$(pwd)/.docker-custom >> $GITHUB_ENV
-      - uses: docker/setup-buildx-action@v2
+      - uses: docker/setup-buildx-action@v3
+        with:
+          cache-binary: false
 
       - uses: docker/login-action@v3
         with:
           username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
           password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
 
-      - uses: docker/build-push-action@v5
+      - uses: docker/login-action@v3
+        with:
+          registry: cache.neon.build
+          username: ${{ secrets.NEON_CI_DOCKERCACHE_USERNAME }}
+          password: ${{ secrets.NEON_CI_DOCKERCACHE_PASSWORD }}
+
+      - uses: docker/build-push-action@v6
         with:
           context: .
           build-args: |
@@ -762,8 +516,8 @@ jobs:
           push: true
           pull: true
           file: Dockerfile
-          cache-from: type=registry,ref=neondatabase/neon:cache-${{ matrix.arch }}
-          cache-to: ${{ github.ref_name == 'main' && format('type=registry,ref=neondatabase/neon:cache-{0},mode=max', matrix.arch) || '' }}
+          cache-from: type=registry,ref=cache.neon.build/neon:cache-${{ matrix.arch }}
+          cache-to: ${{ github.ref_name == 'main' && format('type=registry,ref=cache.neon.build/neon:cache-{0},mode=max', matrix.arch) || '' }}
           tags: |
             neondatabase/neon:${{ needs.tag.outputs.build-tag }}-${{ matrix.arch }}
 
@@ -822,11 +576,12 @@ jobs:
         run: |
           mkdir -p .docker-custom
           echo DOCKER_CONFIG=$(pwd)/.docker-custom >> $GITHUB_ENV
-      - uses: docker/setup-buildx-action@v2
+      - uses: docker/setup-buildx-action@v3
         with:
+          cache-binary: false
           # Disable parallelism for docker buildkit.
           # As we already build everything with `make -j$(nproc)`, running it in additional level of parallelisam blows up the Runner.
-          config-inline: |
+          buildkitd-config-inline: |
             [worker.oci]
               max-parallelism = 1
 
@@ -841,8 +596,14 @@ jobs:
           username: ${{ secrets.AWS_ACCESS_KEY_DEV }}
           password: ${{ secrets.AWS_SECRET_KEY_DEV }}
 
+      - uses: docker/login-action@v3
+        with:
+          registry: cache.neon.build
+          username: ${{ secrets.NEON_CI_DOCKERCACHE_USERNAME }}
+          password: ${{ secrets.NEON_CI_DOCKERCACHE_PASSWORD }}
+
       - name: Build compute-node image
-        uses: docker/build-push-action@v5
+        uses: docker/build-push-action@v6
         with:
           context: .
           build-args: |
@@ -854,14 +615,14 @@ jobs:
           push: true
           pull: true
           file: Dockerfile.compute-node
-          cache-from: type=registry,ref=neondatabase/compute-node-${{ matrix.version }}:cache-${{ matrix.arch }}
-          cache-to: ${{ github.ref_name == 'main' && format('type=registry,ref=neondatabase/compute-node-{0}:cache-{1},mode=max', matrix.version, matrix.arch) || '' }}
+          cache-from: type=registry,ref=cache.neon.build/compute-node-${{ matrix.version }}:cache-${{ matrix.arch }}
+          cache-to: ${{ github.ref_name == 'main' && format('type=registry,ref=cache.neon.build/compute-node-{0}:cache-{1},mode=max', matrix.version, matrix.arch) || '' }}
           tags: |
             neondatabase/compute-node-${{ matrix.version }}:${{ needs.tag.outputs.build-tag }}-${{ matrix.arch }}
 
       - name: Build neon extensions test image
         if: matrix.version == 'v16'
-        uses: docker/build-push-action@v5
+        uses: docker/build-push-action@v6
         with:
           context: .
           build-args: |
@@ -874,15 +635,15 @@ jobs:
           pull: true
           file: Dockerfile.compute-node
           target: neon-pg-ext-test
-          cache-from: type=registry,ref=neondatabase/neon-test-extensions-${{ matrix.version }}:cache-${{ matrix.arch }}
-          cache-to: ${{ github.ref_name == 'main' && format('type=registry,ref=neondatabase/neon-test-extensions-{0}:cache-{1},mode=max', matrix.version, matrix.arch) || '' }}
+          cache-from: type=registry,ref=cache.neon.build/neon-test-extensions-${{ matrix.version }}:cache-${{ matrix.arch }}
+          cache-to: ${{ github.ref_name == 'main' && format('type=registry,ref=cache.neon.build/neon-test-extensions-{0}:cache-{1},mode=max', matrix.version, matrix.arch) || '' }}
           tags: |
             neondatabase/neon-test-extensions-${{ matrix.version }}:${{needs.tag.outputs.build-tag}}-${{ matrix.arch }}
 
       - name: Build compute-tools image
         # compute-tools are Postgres independent, so build it only once
         if: matrix.version == 'v16'
-        uses: docker/build-push-action@v5
+        uses: docker/build-push-action@v6
         with:
           target: compute-tools-image
           context: .
@@ -1075,6 +836,9 @@ jobs:
           rm -rf .docker-custom
 
   promote-images:
+    permissions:
+      contents: read  # This is required for actions/checkout
+      id-token: write # This is required for Azure Login to work.
     needs: [ check-permissions, tag, test-images, vm-compute-node-image ]
     runs-on: ubuntu-22.04
 
@@ -1101,6 +865,28 @@ jobs:
                                                neondatabase/vm-compute-node-${version}:${{ needs.tag.outputs.build-tag }}
           done
 
+      - name: Azure login
+        if: github.ref_name == 'main'
+        uses: azure/login@6c251865b4e6290e7b78be643ea2d005bc51f69a  # @v2.1.1
+        with:
+          client-id: ${{ secrets.AZURE_DEV_CLIENT_ID }}
+          tenant-id: ${{ secrets.AZURE_TENANT_ID }}
+          subscription-id: ${{ secrets.AZURE_DEV_SUBSCRIPTION_ID }}
+
+      - name: Login to ACR
+        if: github.ref_name == 'main'
+        run: |
+          az acr login --name=neoneastus2
+
+      - name: Copy docker images to ACR-dev
+        if: github.ref_name == 'main'
+        run: |
+          for image in neon compute-tools {vm-,}compute-node-{v14,v15,v16}; do
+            docker buildx imagetools create \
+              -t neoneastus2.azurecr.io/neondatabase/${image}:${{ needs.tag.outputs.build-tag }} \
+                                        neondatabase/${image}:${{ needs.tag.outputs.build-tag }}
+          done
+
       - name: Add latest tag to images
         if: github.ref_name == 'main'
         run: |
@@ -1213,7 +999,7 @@ jobs:
           exit 1
 
   deploy:
-    needs: [ check-permissions, promote-images, tag, regress-tests, trigger-custom-extensions-build-and-wait ]
+    needs: [ check-permissions, promote-images, tag, build-and-test-locally, trigger-custom-extensions-build-and-wait ]
     if: github.ref_name == 'main' || github.ref_name == 'release'|| github.ref_name == 'release-proxy'
 
     runs-on: [ self-hosted, gen3, small ]
@@ -1314,7 +1100,7 @@ jobs:
             })
 
   promote-compatibility-data:
-    needs: [ check-permissions, promote-images, tag, regress-tests ]
+    needs: [ check-permissions, promote-images, tag, build-and-test-locally ]
     if: github.ref_name == 'release'
 
     runs-on: [ self-hosted, gen3, small ]
@@ -1326,6 +1112,7 @@ jobs:
         env:
           BUCKET: neon-github-public-dev
           PREFIX: artifacts/latest
+          COMMIT_SHA: ${{ github.event.pull_request.head.sha || github.sha }}
         run: |
           # Update compatibility snapshot for the release
           for pg_version in v14 v15 v16; do
@@ -1339,7 +1126,7 @@ jobs:
 
           # Update Neon artifact for the release (reuse already uploaded artifact)
           for build_type in debug release; do
-            OLD_PREFIX=artifacts/${GITHUB_RUN_ID}
+            OLD_PREFIX=artifacts/${COMMIT_SHA}/${GITHUB_RUN_ID}
             FILENAME=neon-${{ runner.os }}-${{ runner.arch }}-${build_type}-artifact.tar.zst
 
             S3_KEY=$(aws s3api list-objects-v2 --bucket ${BUCKET} --prefix ${OLD_PREFIX} | jq -r '.Contents[]?.Key' | grep ${FILENAME} | sort --version-sort | tail -1 || true)
@@ -1352,9 +1139,37 @@ jobs:
           done
 
   pin-build-tools-image:
-    needs: [ build-build-tools-image, promote-images, regress-tests ]
+    needs: [ build-build-tools-image, promote-images, build-and-test-locally ]
     if: github.ref_name == 'main'
     uses: ./.github/workflows/pin-build-tools-image.yml
     with:
       from-tag: ${{ needs.build-build-tools-image.outputs.image-tag }}
     secrets: inherit
+
+  # This job simplifies setting branch protection rules (in GitHub UI)
+  # by allowing to set only this job instead of listing many others.
+  # It also makes it easier to rename or parametrise jobs (using matrix)
+  # which requires changes in branch protection rules
+  #
+  # Note, that we can't add external check (like `neon-cloud-e2e`) we still need to use GitHub UI for that.
+  #
+  # https://github.com/neondatabase/neon/settings/branch_protection_rules
+  conclusion:
+    if: always()
+    # Format `needs` differently to make the list more readable.
+    # Usually we do `needs: [...]`
+    needs:
+      - check-codestyle-python
+      - check-codestyle-rust
+      - build-and-test-locally
+      - test-images
+    runs-on: ubuntu-22.04
+    steps:
+      # The list of possible results:
+      # https://docs.github.com/en/actions/learn-github-actions/contexts#needs-context
+      - name: Fail the job if any of the dependencies do not succeed
+        run: exit 1
+        if: |
+          contains(needs.*.result, 'failure')
+          || contains(needs.*.result, 'cancelled')
+          || contains(needs.*.result, 'skipped')

.github/workflows/neon_extra_builds.yml

@@ -133,214 +133,6 @@ jobs:
       - name: Check that no warnings are produced
         run: ./run_clippy.sh
 
-  check-linux-arm-build:
-    needs: [ check-permissions, build-build-tools-image ]
-    timeout-minutes: 90
-    runs-on: [ self-hosted, small-arm64 ]
-
-    env:
-      # Use release build only, to have less debug info around
-      # Hence keeping target/ (and general cache size) smaller
-      BUILD_TYPE: release
-      CARGO_FEATURES: --features testing
-      CARGO_FLAGS: --release
-      AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_DEV }}
-      AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_KEY_DEV }}
-
-    container:
-      image: ${{ needs.build-build-tools-image.outputs.image }}
-      credentials:
-        username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
-        password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
-      options: --init
-
-    steps:
-      - name: Fix git ownership
-        run: |
-          # Workaround for `fatal: detected dubious ownership in repository at ...`
-          #
-          # Use both ${{ github.workspace }} and ${GITHUB_WORKSPACE} because they're different on host and in containers
-          #   Ref https://github.com/actions/checkout/issues/785
-          #
-          git config --global --add safe.directory ${{ github.workspace }}
-          git config --global --add safe.directory ${GITHUB_WORKSPACE}
-          for r in 14 15 16; do
-            git config --global --add safe.directory "${{ github.workspace }}/vendor/postgres-v$r"
-            git config --global --add safe.directory "${GITHUB_WORKSPACE}/vendor/postgres-v$r"
-          done
-
-      - name: Checkout
-        uses: actions/checkout@v4
-        with:
-          submodules: true
-          fetch-depth: 1
-
-      - name: Set pg 14 revision for caching
-        id: pg_v14_rev
-        run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-v14) >> $GITHUB_OUTPUT
-
-      - name: Set pg 15 revision for caching
-        id: pg_v15_rev
-        run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-v15) >> $GITHUB_OUTPUT
-
-      - name: Set pg 16 revision for caching
-        id: pg_v16_rev
-        run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-v16) >> $GITHUB_OUTPUT
-
-      - name: Set env variables
-        run: |
-          echo "CARGO_HOME=${GITHUB_WORKSPACE}/.cargo" >> $GITHUB_ENV
-
-      - name: Cache postgres v14 build
-        id: cache_pg_14
-        uses: actions/cache@v4
-        with:
-          path: pg_install/v14
-          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ 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@v4
-        with:
-          path: pg_install/v15
-          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ env.BUILD_TYPE }}-pg-${{ steps.pg_v15_rev.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
-
-      - name: Cache postgres v16 build
-        id: cache_pg_16
-        uses: actions/cache@v4
-        with:
-          path: pg_install/v16
-          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ env.BUILD_TYPE }}-pg-${{ steps.pg_v16_rev.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
-
-      - name: Build postgres v14
-        if: steps.cache_pg_14.outputs.cache-hit != 'true'
-        run: mold -run make postgres-v14 -j$(nproc)
-
-      - name: Build postgres v15
-        if: steps.cache_pg_15.outputs.cache-hit != 'true'
-        run: mold -run make postgres-v15 -j$(nproc)
-
-      - name: Build postgres v16
-        if: steps.cache_pg_16.outputs.cache-hit != 'true'
-        run: mold -run make postgres-v16 -j$(nproc)
-
-      - name: Build neon extensions
-        run: mold -run make neon-pg-ext -j$(nproc)
-
-      - name: Build walproposer-lib
-        run: mold -run make walproposer-lib -j$(nproc)
-
-      - name: Run cargo build
-        run: |
-          mold -run cargo build --locked $CARGO_FLAGS $CARGO_FEATURES --bins --tests -j$(nproc)
-
-      - name: Run cargo test
-        env:
-          NEXTEST_RETRIES: 3
-        run: |
-          cargo nextest run $CARGO_FEATURES -j$(nproc)
-
-          # 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 `$CARGO_FEATURES` since there's no `testing` feature in the e2e tests now
-          cargo nextest run --package remote_storage --test test_real_s3 -j$(nproc)
-
-          # Run separate tests for real Azure Blob Storage
-          # XXX: replace region with `eu-central-1`-like region
-          export ENABLE_REAL_AZURE_REMOTE_STORAGE=y
-          export AZURE_STORAGE_ACCOUNT="${{ secrets.AZURE_STORAGE_ACCOUNT_DEV }}"
-          export AZURE_STORAGE_ACCESS_KEY="${{ secrets.AZURE_STORAGE_ACCESS_KEY_DEV }}"
-          export REMOTE_STORAGE_AZURE_CONTAINER="${{ vars.REMOTE_STORAGE_AZURE_CONTAINER }}"
-          export REMOTE_STORAGE_AZURE_REGION="${{ vars.REMOTE_STORAGE_AZURE_REGION }}"
-          # Avoid `$CARGO_FEATURES` since there's no `testing` feature in the e2e tests now
-          cargo nextest run --package remote_storage --test test_real_azure -j$(nproc)
-
-  check-codestyle-rust-arm:
-    needs: [ check-permissions, build-build-tools-image ]
-    timeout-minutes: 90
-    runs-on: [ self-hosted, small-arm64 ]
-
-    container:
-      image: ${{ needs.build-build-tools-image.outputs.image }}
-      credentials:
-        username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
-        password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
-      options: --init
-
-    strategy:
-      fail-fast: false
-      matrix:
-        build_type: [ debug, release ]
-
-    steps:
-      - name: Fix git ownership
-        run: |
-          # Workaround for `fatal: detected dubious ownership in repository at ...`
-          #
-          # Use both ${{ github.workspace }} and ${GITHUB_WORKSPACE} because they're different on host and in containers
-          #   Ref https://github.com/actions/checkout/issues/785
-          #
-          git config --global --add safe.directory ${{ github.workspace }}
-          git config --global --add safe.directory ${GITHUB_WORKSPACE}
-          for r in 14 15 16; do
-            git config --global --add safe.directory "${{ github.workspace }}/vendor/postgres-v$r"
-            git config --global --add safe.directory "${GITHUB_WORKSPACE}/vendor/postgres-v$r"
-          done
-
-      - name: Checkout
-        uses: actions/checkout@v4
-        with:
-          submodules: true
-          fetch-depth: 1
-
-      # Some of our rust modules use FFI and need those to be checked
-      - name: Get postgres headers
-        run: make postgres-headers -j$(nproc)
-
-      # cargo hack runs the given cargo subcommand (clippy in this case) for all feature combinations.
-      # This will catch compiler & clippy warnings in all feature combinations.
-      # TODO: use cargo hack for build and test as well, but, that's quite expensive.
-      # NB: keep clippy args in sync with ./run_clippy.sh
-      - run: |
-          CLIPPY_COMMON_ARGS="$( source .neon_clippy_args; echo "$CLIPPY_COMMON_ARGS")"
-          if [ "$CLIPPY_COMMON_ARGS" = "" ]; then
-            echo "No clippy args found in .neon_clippy_args"
-            exit 1
-          fi
-          echo "CLIPPY_COMMON_ARGS=${CLIPPY_COMMON_ARGS}" >> $GITHUB_ENV
-
-      - name: Run cargo clippy (debug)
-        if: matrix.build_type == 'debug'
-        run: cargo hack --feature-powerset clippy $CLIPPY_COMMON_ARGS
-      - name: Run cargo clippy (release)
-        if: matrix.build_type == 'release'
-        run: cargo hack --feature-powerset clippy --release $CLIPPY_COMMON_ARGS
-
-      - name: Check documentation generation
-        if: matrix.build_type == 'release'
-        run: cargo doc --workspace --no-deps --document-private-items -j$(nproc)
-        env:
-            RUSTDOCFLAGS: "-Dwarnings -Arustdoc::private_intra_doc_links"
-
-      # Use `${{ !cancelled() }}` to run quck tests after the longer clippy run
-      - name: Check formatting
-        if: ${{ !cancelled() && matrix.build_type == 'release' }}
-        run: cargo fmt --all -- --check
-
-      # https://github.com/facebookincubator/cargo-guppy/tree/bec4e0eb29dcd1faac70b1b5360267fc02bf830e/tools/cargo-hakari#2-keep-the-workspace-hack-up-to-date-in-ci
-      - name: Check rust dependencies
-        if: ${{ !cancelled() && matrix.build_type == 'release' }}
-        run: |
-          cargo hakari generate --diff  # workspace-hack Cargo.toml is up-to-date
-          cargo hakari manage-deps --dry-run  # all workspace crates depend on workspace-hack
-
-      # https://github.com/EmbarkStudios/cargo-deny
-      - name: Check rust licenses/bans/advisories/sources
-        if: ${{ !cancelled() && matrix.build_type == 'release' }}
-        run: cargo deny check
-
   gather-rust-build-stats:
     needs: [ check-permissions, build-build-tools-image ]
     if: |
@@ -357,8 +149,6 @@ jobs:
 
     env:
       BUILD_TYPE: release
-      # remove the cachepot wrapper and build without crate caches
-      RUSTC_WRAPPER: ""
       # build with incremental compilation produce partial results
       # so do not attempt to cache this build, also disable the incremental compilation
       CARGO_INCREMENTAL: 0
@@ -378,7 +168,7 @@ jobs:
         run: make walproposer-lib -j$(nproc)
 
       - name: Produce the build stats
-        run: cargo build --all --release --timings -j$(nproc)
+        run: PQ_LIB_DIR=$(pwd)/pg_install/v16/lib cargo build --all --release --timings -j$(nproc)
 
       - name: Upload the build stats
         id: upload-stats

.github/workflows/periodic_pagebench.yml

@@ -0,0 +1,155 @@
+name: Periodic pagebench performance test on dedicated EC2 machine in eu-central-1 region
+
+on:
+  schedule:
+    # * is a special character in YAML so you have to quote this string
+    #          ┌───────────── minute (0 - 59)
+    #          │ ┌───────────── hour (0 - 23)
+    #          │ │ ┌───────────── day of the month (1 - 31)
+    #          │ │ │ ┌───────────── month (1 - 12 or JAN-DEC)
+    #          │ │ │ │ ┌───────────── day of the week (0 - 6 or SUN-SAT)
+    - cron:  '0 18 * * *' # Runs at 6 PM UTC every day
+  workflow_dispatch: # Allows manual triggering of the workflow
+    inputs:
+      commit_hash:
+        type: string
+        description: 'The long neon repo commit hash for the system under test (pageserver) to be tested.'
+        required: false
+        default: ''
+
+defaults:
+  run:
+    shell: bash -euo pipefail {0}
+
+concurrency:
+  group: ${{ github.workflow }}
+  cancel-in-progress: false
+
+jobs:
+  trigger_bench_on_ec2_machine_in_eu_central_1:
+    runs-on: [ self-hosted, gen3, small ]
+    container:
+      image: neondatabase/build-tools:pinned
+      credentials:
+        username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
+        password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
+      options: --init
+    timeout-minutes: 360  # Set the timeout to 6 hours
+    env:
+      API_KEY: ${{ secrets.PERIODIC_PAGEBENCH_EC2_RUNNER_API_KEY }}
+      RUN_ID: ${{ github.run_id }}
+      AWS_ACCESS_KEY_ID: ${{ secrets.AWS_EC2_US_TEST_RUNNER_ACCESS_KEY_ID }}
+      AWS_SECRET_ACCESS_KEY : ${{ secrets.AWS_EC2_US_TEST_RUNNER_ACCESS_KEY_SECRET }}
+      AWS_DEFAULT_REGION : "eu-central-1"
+      AWS_INSTANCE_ID : "i-02a59a3bf86bc7e74"
+    steps:
+    # we don't need the neon source code because we run everything remotely
+    # however we still need the local github actions to run the allure step below
+    - uses: actions/checkout@v4
+
+    - name: Show my own (github runner) external IP address - usefull for IP allowlisting
+      run: curl https://ifconfig.me
+
+    - name: Start EC2 instance and wait for the instance to boot up
+      run: |
+        aws ec2 start-instances --instance-ids $AWS_INSTANCE_ID
+        aws ec2 wait instance-running --instance-ids $AWS_INSTANCE_ID
+        sleep 60 # sleep some time to allow cloudinit and our API server to start up
+
+    - name: Determine public IP of the EC2 instance and set env variable EC2_MACHINE_URL_US
+      run: |
+        public_ip=$(aws ec2 describe-instances --instance-ids $AWS_INSTANCE_ID --query 'Reservations[*].Instances[*].PublicIpAddress' --output text)
+        echo "Public IP of the EC2 instance: $public_ip"
+        echo "EC2_MACHINE_URL_US=https://${public_ip}:8443" >> $GITHUB_ENV
+
+    - name: Determine commit hash
+      env:
+        INPUT_COMMIT_HASH: ${{ github.event.inputs.commit_hash }}
+      run: |
+        if [ -z "$INPUT_COMMIT_HASH" ]; then
+          echo "COMMIT_HASH=$(curl -s https://api.github.com/repos/neondatabase/neon/commits/main | jq -r '.sha')" >> $GITHUB_ENV
+        else
+          echo "COMMIT_HASH=$INPUT_COMMIT_HASH" >> $GITHUB_ENV
+        fi
+
+    - name: Start Bench with run_id   
+      run: |
+        curl -k -X 'POST' \
+        "${EC2_MACHINE_URL_US}/start_test/${GITHUB_RUN_ID}" \
+        -H 'accept: application/json' \
+        -H 'Content-Type: application/json' \
+        -H "Authorization: Bearer $API_KEY" \
+        -d "{\"neonRepoCommitHash\": \"${COMMIT_HASH}\"}"
+
+    - name: Poll Test Status
+      id: poll_step
+      run: |
+        status=""
+        while [[ "$status" != "failure" && "$status" != "success" ]]; do
+          response=$(curl -k -X 'GET' \
+          "${EC2_MACHINE_URL_US}/test_status/${GITHUB_RUN_ID}" \
+          -H 'accept: application/json' \
+          -H "Authorization: Bearer $API_KEY")
+          echo "Response: $response"
+          set +x
+          status=$(echo $response | jq -r '.status')
+          echo "Test status: $status"
+          if [[ "$status" == "failure" ]]; then
+            echo "Test failed"
+            exit 1 # Fail the job step if status is failure
+          elif [[ "$status" == "success" || "$status" == "null" ]]; then
+            break
+          elif [[ "$status" == "too_many_runs" ]]; then
+            echo "Too many runs already running"
+            echo "too_many_runs=true" >> "$GITHUB_OUTPUT"
+            exit 1
+          fi
+
+          sleep 60 # Poll every 60 seconds
+        done
+
+    - name: Retrieve Test Logs
+      if: always() && steps.poll_step.outputs.too_many_runs != 'true'
+      run: |
+        curl -k -X 'GET' \
+        "${EC2_MACHINE_URL_US}/test_log/${GITHUB_RUN_ID}" \
+        -H 'accept: application/gzip' \
+        -H "Authorization: Bearer $API_KEY" \
+        --output "test_log_${GITHUB_RUN_ID}.gz"
+    
+    - name: Unzip Test Log and Print it into this job's log
+      if: always() && steps.poll_step.outputs.too_many_runs != 'true'
+      run: |
+        gzip -d "test_log_${GITHUB_RUN_ID}.gz"
+        cat "test_log_${GITHUB_RUN_ID}"
+
+    - name: Create Allure report
+      env:
+        AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_DEV }}
+        AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_KEY_DEV }}
+      if: ${{ !cancelled() }}
+      uses: ./.github/actions/allure-report-generate
+
+    - name: Post to a Slack channel
+      if: ${{ github.event.schedule && failure() }}
+      uses: slackapi/slack-github-action@v1
+      with:
+        channel-id: "C033QLM5P7D" # dev-staging-stream
+        slack-message: "Periodic pagebench testing on dedicated hardware: ${{ job.status }}\n${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
+      env:
+        SLACK_BOT_TOKEN: ${{ secrets.SLACK_BOT_TOKEN }}
+
+    - name: Cleanup Test Resources
+      if: always() 
+      run: |
+        curl -k -X 'POST' \
+        "${EC2_MACHINE_URL_US}/cleanup_test/${GITHUB_RUN_ID}" \
+        -H 'accept: application/json' \
+        -H "Authorization: Bearer $API_KEY" \
+        -d ''
+
+    - name: Stop EC2 instance and wait for the instance to be stopped
+      if: always() && steps.poll_step.outputs.too_many_runs != 'true'
+      run: |
+        aws ec2 stop-instances --instance-ids $AWS_INSTANCE_ID
+        aws ec2 wait instance-stopped --instance-ids $AWS_INSTANCE_ID

.github/workflows/pg-clients.yml

@@ -0,0 +1,211 @@
+name: Test Postgres client libraries
+
+on:
+  schedule:
+    # * is a special character in YAML so you have to quote this string
+    #          ┌───────────── minute (0 - 59)
+    #          │ ┌───────────── hour (0 - 23)
+    #          │ │ ┌───────────── day of the month (1 - 31)
+    #          │ │ │ ┌───────────── month (1 - 12 or JAN-DEC)
+    #          │ │ │ │ ┌───────────── day of the week (0 - 6 or SUN-SAT)
+    - cron:  '23 02 * * *' # run once a day, timezone is utc
+  pull_request:
+    paths:
+      - '.github/workflows/pg-clients.yml'
+      - 'test_runner/pg_clients/**'
+      - 'test_runner/logical_repl/**'
+      - 'poetry.lock'
+  workflow_dispatch:
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref_name }}
+  cancel-in-progress: ${{ github.event_name == 'pull_request' }}
+
+defaults:
+  run:
+    shell: bash -euxo pipefail {0}
+
+env:
+  DEFAULT_PG_VERSION: 16
+  PLATFORM: neon-captest-new
+  AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_DEV }}
+  AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_KEY_DEV }}
+  AWS_DEFAULT_REGION: eu-central-1
+
+jobs:
+  check-permissions:
+    if: ${{ !contains(github.event.pull_request.labels.*.name, 'run-no-ci') }}
+    uses: ./.github/workflows/check-permissions.yml
+    with:
+      github-event-name: ${{ github.event_name }}
+
+  check-build-tools-image:
+    needs: [ check-permissions ]
+    uses: ./.github/workflows/check-build-tools-image.yml
+
+  build-build-tools-image:
+    needs: [ check-build-tools-image ]
+    uses: ./.github/workflows/build-build-tools-image.yml
+    with:
+      image-tag: ${{ needs.check-build-tools-image.outputs.image-tag }}
+    secrets: inherit
+
+  test-logical-replication:
+    needs: [ build-build-tools-image ]
+    runs-on: ubuntu-22.04
+
+    container:
+      image: ${{ needs.build-build-tools-image.outputs.image }}
+      credentials:
+        username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
+        password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
+      options: --init --user root
+    services:
+      clickhouse:
+        image: clickhouse/clickhouse-server:24.6.3.64
+        ports:
+          - 9000:9000
+          - 8123:8123
+      zookeeper:
+        image: quay.io/debezium/zookeeper:2.7
+        ports:
+          - 2181:2181
+      kafka:
+        image: quay.io/debezium/kafka:2.7
+        env:
+          ZOOKEEPER_CONNECT: "zookeeper:2181"
+          KAFKA_ADVERTISED_LISTENERS: PLAINTEXT://kafka:9092
+          KAFKA_BROKER_ID: 1
+          KAFKA_OFFSETS_TOPIC_REPLICATION_FACTOR: 1
+          KAFKA_JMX_PORT: 9991
+        ports:
+          - 9092:9092
+      debezium:
+        image: quay.io/debezium/connect:2.7
+        env:
+          BOOTSTRAP_SERVERS: kafka:9092
+          GROUP_ID: 1
+          CONFIG_STORAGE_TOPIC: debezium-config
+          OFFSET_STORAGE_TOPIC: debezium-offset
+          STATUS_STORAGE_TOPIC: debezium-status
+          DEBEZIUM_CONFIG_CONNECTOR_CLASS: io.debezium.connector.postgresql.PostgresConnector
+        ports:
+          - 8083:8083
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Download Neon artifact
+        uses: ./.github/actions/download
+        with:
+          name: neon-${{ runner.os }}-${{ runner.arch }}-release-artifact
+          path: /tmp/neon/
+          prefix: latest
+
+      - name: Create Neon Project
+        id: create-neon-project
+        uses: ./.github/actions/neon-project-create
+        with:
+          api_key: ${{ secrets.NEON_STAGING_API_KEY }}
+          postgres_version: ${{ env.DEFAULT_PG_VERSION }}
+
+      - name: Run tests
+        uses: ./.github/actions/run-python-test-set
+        with:
+          build_type: remote
+          test_selection: logical_repl
+          run_in_parallel: false
+          extra_params: -m remote_cluster
+          pg_version: ${{ env.DEFAULT_PG_VERSION }}
+        env:
+          BENCHMARK_CONNSTR: ${{ steps.create-neon-project.outputs.dsn }}
+
+      - name: Delete Neon Project
+        if: always()
+        uses: ./.github/actions/neon-project-delete
+        with:
+          project_id: ${{ steps.create-neon-project.outputs.project_id }}
+          api_key: ${{ secrets.NEON_STAGING_API_KEY }}
+
+      - name: Create Allure report
+        if: ${{ !cancelled() }}
+        id: create-allure-report
+        uses: ./.github/actions/allure-report-generate
+        with:
+          store-test-results-into-db: true
+        env:
+          REGRESS_TEST_RESULT_CONNSTR_NEW: ${{ secrets.REGRESS_TEST_RESULT_CONNSTR_NEW }}
+
+      - name: Post to a Slack channel
+        if: github.event.schedule && failure()
+        uses: slackapi/slack-github-action@v1
+        with:
+          channel-id: "C06KHQVQ7U3" # on-call-qa-staging-stream
+          slack-message: |
+            Testing the logical replication: <${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}|${{ job.status }}> (<${{ steps.create-allure-report.outputs.report-url }}|test report>)
+        env:
+          SLACK_BOT_TOKEN: ${{ secrets.SLACK_BOT_TOKEN }}
+
+  test-postgres-client-libs:
+    needs: [ build-build-tools-image ]
+    runs-on: ubuntu-22.04
+
+    container:
+      image: ${{ needs.build-build-tools-image.outputs.image }}
+      credentials:
+        username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
+        password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
+      options: --init --user root
+
+    steps:
+    - uses: actions/checkout@v4
+
+    - name: Download Neon artifact
+      uses: ./.github/actions/download
+      with:
+        name: neon-${{ runner.os }}-${{ runner.arch }}-release-artifact
+        path: /tmp/neon/
+        prefix: latest
+
+    - name: Create Neon Project
+      id: create-neon-project
+      uses: ./.github/actions/neon-project-create
+      with:
+        api_key: ${{ secrets.NEON_STAGING_API_KEY }}
+        postgres_version: ${{ env.DEFAULT_PG_VERSION }}
+
+    - name: Run tests
+      uses: ./.github/actions/run-python-test-set
+      with:
+        build_type: remote
+        test_selection: pg_clients
+        run_in_parallel: false
+        extra_params: -m remote_cluster
+        pg_version: ${{ env.DEFAULT_PG_VERSION }}
+      env:
+        BENCHMARK_CONNSTR: ${{ steps.create-neon-project.outputs.dsn }}
+
+    - name: Delete Neon Project
+      if: always()
+      uses: ./.github/actions/neon-project-delete
+      with:
+        project_id: ${{ steps.create-neon-project.outputs.project_id }}
+        api_key: ${{ secrets.NEON_STAGING_API_KEY }}
+
+    - name: Create Allure report
+      if: ${{ !cancelled() }}
+      id: create-allure-report
+      uses: ./.github/actions/allure-report-generate
+      with:
+        store-test-results-into-db: true
+      env:
+        REGRESS_TEST_RESULT_CONNSTR_NEW: ${{ secrets.REGRESS_TEST_RESULT_CONNSTR_NEW }}
+
+    - name: Post to a Slack channel
+      if: github.event.schedule && failure()
+      uses: slackapi/slack-github-action@v1
+      with:
+        channel-id: "C06KHQVQ7U3" # on-call-qa-staging-stream
+        slack-message: |
+          Testing Postgres clients: <${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}|${{ job.status }}> (<${{ steps.create-allure-report.outputs.report-url }}|test report>)
+      env:
+        SLACK_BOT_TOKEN: ${{ secrets.SLACK_BOT_TOKEN }}

.github/workflows/pg_clients.yml

@@ -1,98 +0,0 @@
-name: Test Postgres client libraries
-
-on:
-  schedule:
-    # * is a special character in YAML so you have to quote this string
-    #          ┌───────────── minute (0 - 59)
-    #          │ ┌───────────── hour (0 - 23)
-    #          │ │ ┌───────────── day of the month (1 - 31)
-    #          │ │ │ ┌───────────── month (1 - 12 or JAN-DEC)
-    #          │ │ │ │ ┌───────────── day of the week (0 - 6 or SUN-SAT)
-    - cron:  '23 02 * * *' # run once a day, timezone is utc
-
-  workflow_dispatch:
-
-concurrency:
-  # Allow only one workflow per any non-`main` branch.
-  group: ${{ github.workflow }}-${{ github.ref_name }}-${{ github.ref_name == 'main' && github.sha || 'anysha' }}
-  cancel-in-progress: true
-
-jobs:
-  test-postgres-client-libs:
-    # TODO: switch to gen2 runner, requires docker
-    runs-on: ubuntu-22.04
-
-    env:
-      DEFAULT_PG_VERSION: 14
-      TEST_OUTPUT: /tmp/test_output
-
-    steps:
-    - name: Checkout
-      uses: actions/checkout@v4
-
-    - uses: actions/setup-python@v4
-      with:
-        python-version: 3.9
-
-    - name: Install Poetry
-      uses: snok/install-poetry@v1
-
-    - name: Cache poetry deps
-      uses: actions/cache@v4
-      with:
-        path: ~/.cache/pypoetry/virtualenvs
-        key: v2-${{ runner.os }}-${{ runner.arch }}-python-deps-ubunutu-latest-${{ hashFiles('poetry.lock') }}
-
-    - name: Install Python deps
-      shell: bash -euxo pipefail {0}
-      run: ./scripts/pysync
-
-    - name: Create Neon Project
-      id: create-neon-project
-      uses: ./.github/actions/neon-project-create
-      with:
-        api_key: ${{ secrets.NEON_STAGING_API_KEY }}
-        postgres_version: ${{ env.DEFAULT_PG_VERSION }}
-
-    - name: Run pytest
-      env:
-        REMOTE_ENV: 1
-        BENCHMARK_CONNSTR: ${{ steps.create-neon-project.outputs.dsn }}
-        POSTGRES_DISTRIB_DIR: /tmp/neon/pg_install
-      shell: bash -euxo pipefail {0}
-      run: |
-        # Test framework expects we have psql binary;
-        # but since we don't really need it in this test, let's mock it
-        mkdir -p "$POSTGRES_DISTRIB_DIR/v${DEFAULT_PG_VERSION}/bin" && touch "$POSTGRES_DISTRIB_DIR/v${DEFAULT_PG_VERSION}/bin/psql";
-        ./scripts/pytest \
-          --junitxml=$TEST_OUTPUT/junit.xml \
-          --tb=short \
-          --verbose \
-          -m "remote_cluster" \
-          -rA "test_runner/pg_clients"
-
-    - name: Delete Neon Project
-      if: ${{ always() }}
-      uses: ./.github/actions/neon-project-delete
-      with:
-        project_id: ${{ steps.create-neon-project.outputs.project_id }}
-        api_key: ${{ secrets.NEON_STAGING_API_KEY }}
-
-    # We use GitHub's action upload-artifact because `ubuntu-latest` doesn't have configured AWS CLI.
-    # It will be fixed after switching to gen2 runner
-    - name: Upload python test logs
-      if: always()
-      uses: actions/upload-artifact@v4
-      with:
-        retention-days: 7
-        name: python-test-pg_clients-${{ runner.os }}-${{ runner.arch }}-stage-logs
-        path: ${{ env.TEST_OUTPUT }}
-
-    - name: Post to a Slack channel
-      if: ${{ github.event.schedule && failure() }}
-      uses: slackapi/slack-github-action@v1
-      with:
-        channel-id: "C033QLM5P7D" # dev-staging-stream
-        slack-message: "Testing Postgres clients: ${{ job.status }}\n${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
-      env:
-        SLACK_BOT_TOKEN: ${{ secrets.SLACK_BOT_TOKEN }}

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

@@ -7,12 +7,20 @@ on:
         description: 'Source tag'
         required: true
         type: string
+      force:
+        description: 'Force the image to be pinned'
+        default: false
+        type: boolean
   workflow_call:
     inputs:
       from-tag:
         description: 'Source tag'
         required: true
         type: string
+      force:
+        description: 'Force the image to be pinned'
+        default: false
+        type: boolean
 
 defaults:
   run:
@@ -22,15 +30,18 @@ concurrency:
   group: pin-build-tools-image-${{ inputs.from-tag }}
   cancel-in-progress: false
 
+# No permission for GITHUB_TOKEN by default; the **minimal required** set of permissions should be granted in each job.
 permissions: {}
 
-jobs:
-  tag-image:
-    runs-on: ubuntu-22.04
+env:
+  FROM_TAG: ${{ inputs.from-tag }}
+  TO_TAG: pinned
 
-    env:
-      FROM_TAG: ${{ inputs.from-tag }}
-      TO_TAG: pinned
+jobs:
+  check-manifests:
+    runs-on: ubuntu-22.04
+    outputs:
+      skip: ${{ steps.check-manifests.outputs.skip }}
 
     steps:
       - name: Check if we really need to pin the image
@@ -47,27 +58,44 @@ jobs:
 
           echo "skip=${skip}" | tee -a $GITHUB_OUTPUT
 
+  tag-image:
+    needs: check-manifests
+
+    # use format(..) to catch both inputs.force = true AND inputs.force = 'true'
+    if: needs.check-manifests.outputs.skip == 'false' || format('{0}', inputs.force) == 'true'
+
+    runs-on: ubuntu-22.04
+
+    permissions:
+      id-token: write # for `azure/login`
+
+    steps:
       - uses: docker/login-action@v3
-        if: steps.check-manifests.outputs.skip == 'false'
+
         with:
           username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
           password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
 
-      - name: Tag build-tools with `${{ env.TO_TAG }}` in Docker Hub
-        if: steps.check-manifests.outputs.skip == 'false'
-        run: |
-          docker buildx imagetools create -t neondatabase/build-tools:${TO_TAG} \
-                                             neondatabase/build-tools:${FROM_TAG}
-
       - uses: docker/login-action@v3
-        if: steps.check-manifests.outputs.skip == 'false'
         with:
           registry: 369495373322.dkr.ecr.eu-central-1.amazonaws.com
           username: ${{ secrets.AWS_ACCESS_KEY_DEV }}
           password: ${{ secrets.AWS_SECRET_KEY_DEV }}
 
-      - name: Tag build-tools with `${{ env.TO_TAG }}` in ECR
-        if: steps.check-manifests.outputs.skip == 'false'
+      - name: Azure login
+        uses: azure/login@6c251865b4e6290e7b78be643ea2d005bc51f69a  # @v2.1.1
+        with:
+          client-id: ${{ secrets.AZURE_DEV_CLIENT_ID }}
+          tenant-id: ${{ secrets.AZURE_TENANT_ID }}
+          subscription-id: ${{ secrets.AZURE_DEV_SUBSCRIPTION_ID }}
+
+      - name: Login to ACR
+        run: |
+          az acr login --name=neoneastus2
+
+      - name: Tag build-tools with `${{ env.TO_TAG }}` in Docker Hub, ECR, and ACR
         run: |
           docker buildx imagetools create -t 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools:${TO_TAG} \
+                                          -t neoneastus2.azurecr.io/neondatabase/build-tools:${TO_TAG} \
+                                          -t neondatabase/build-tools:${TO_TAG} \
                                              neondatabase/build-tools:${FROM_TAG}

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

@@ -13,8 +13,6 @@ defaults:
 env:
   # A concurrency group that we use for e2e-tests runs, matches `concurrency.group` above with `github.repository` as a prefix
   E2E_CONCURRENCY_GROUP: ${{ github.repository }}-e2e-tests-${{ github.ref_name }}-${{ github.ref_name == 'main' && github.sha || 'anysha' }}
-  AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_DEV }}
-  AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_KEY_DEV }}
 
 jobs:
   cancel-previous-e2e-tests:
@@ -64,19 +62,35 @@ jobs:
     needs: [ tag ]
     runs-on: ubuntu-22.04
     env:
+      EVENT_ACTION: ${{ github.event.action }}
+      GH_TOKEN: ${{ secrets.CI_ACCESS_TOKEN }}
       TAG: ${{ needs.tag.outputs.build-tag }}
     steps:
-      - name: check if ecr image are present
-        env:
-          AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_DEV }}
-          AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_KEY_DEV }}
+      - name: Wait for `promote-images` job to finish
+        # It's important to have a timeout here, the script in the step can run infinitely
+        timeout-minutes: 60
         run: |
-          for REPO in neon compute-tools compute-node-v14 vm-compute-node-v14 compute-node-v15 vm-compute-node-v15 compute-node-v16 vm-compute-node-v16; do
-            OUTPUT=$(aws ecr describe-images --repository-name ${REPO} --region eu-central-1 --query "imageDetails[?imageTags[?contains(@, '${TAG}')]]" --output text)
-            if [ "$OUTPUT" == "" ]; then
-              echo "$REPO with image tag $TAG not found" >> $GITHUB_OUTPUT
-              exit 1
-            fi
+          if [ "${GITHUB_EVENT_NAME}" != "pull_request" ] || [ "${EVENT_ACTION}" != "ready_for_review" ]; then
+            exit 0
+          fi
+
+          # For PRs we use the run id as the tag
+          BUILD_AND_TEST_RUN_ID=${TAG}
+          while true; do
+            conclusion=$(gh run --repo ${GITHUB_REPOSITORY} view ${BUILD_AND_TEST_RUN_ID} --json jobs --jq '.jobs[] | select(.name == "promote-images") | .conclusion')
+            case "$conclusion" in
+              success)
+                break
+                ;;
+              failure | cancelled | skipped)
+                echo "The 'promote-images' job didn't succeed: '${conclusion}'. Exiting..."
+                exit 1
+                ;;
+              *)
+                echo "The 'promote-images' hasn't succeed yet. Waiting..."
+                sleep 60
+                ;;
+            esac
           done
 
       - name: Set e2e-platforms

CODEOWNERS

@@ -1,13 +1,13 @@
 /compute_tools/ @neondatabase/control-plane @neondatabase/compute
 /storage_controller @neondatabase/storage
 /libs/pageserver_api/ @neondatabase/storage
-/libs/postgres_ffi/ @neondatabase/compute @neondatabase/safekeepers
+/libs/postgres_ffi/ @neondatabase/compute @neondatabase/storage
 /libs/remote_storage/ @neondatabase/storage
-/libs/safekeeper_api/ @neondatabase/safekeepers
+/libs/safekeeper_api/ @neondatabase/storage
 /libs/vm_monitor/ @neondatabase/autoscaling
 /pageserver/ @neondatabase/storage
 /pgxn/ @neondatabase/compute
-/pgxn/neon/ @neondatabase/compute @neondatabase/safekeepers
+/pgxn/neon/ @neondatabase/compute @neondatabase/storage
 /proxy/ @neondatabase/proxy
-/safekeeper/ @neondatabase/safekeepers
+/safekeeper/ @neondatabase/storage
 /vendor/ @neondatabase/compute

Cargo.toml

@@ -9,13 +9,16 @@ members = [
     "pageserver/ctl",
     "pageserver/client",
     "pageserver/pagebench",
-    "proxy",
+    "proxy/core",
+    "proxy/sasl",
+    "proxy/proxy",
+    "proxy/pg_sni_router",
     "safekeeper",
     "storage_broker",
     "storage_controller",
+    "storage_controller/client",
     "storage_scrubber",
     "workspace_hack",
-    "trace",
     "libs/compute_api",
     "libs/pageserver_api",
     "libs/postgres_ffi",
@@ -84,7 +87,6 @@ enumset = "1.0.12"
 fail = "0.5.0"
 fallible-iterator = "0.2"
 framed-websockets = { version = "0.1.0", git = "https://github.com/neondatabase/framed-websockets" }
-fs2 = "0.4.3"
 futures = "0.3"
 futures-core = "0.3"
 futures-util = "0.3"
@@ -111,8 +113,8 @@ lasso = "0.7"
 leaky-bucket = "1.0.1"
 libc = "0.2"
 md5 = "0.7.0"
-measured = { version = "0.0.21", features=["lasso"] }
-measured-process = { version = "0.0.21" }
+measured = { version = "0.0.22", features=["lasso"] }
+measured-process = { version = "0.0.22" }
 memoffset = "0.8"
 nix = { version = "0.27", features = ["fs", "process", "socket", "signal", "poll"] }
 notify = "6.0.0"
@@ -127,7 +129,7 @@ parquet = { version = "51.0.0", default-features = false, features = ["zstd"] }
 parquet_derive = "51.0.0"
 pbkdf2 = { version = "0.12.1", features = ["simple", "std"] }
 pin-project-lite = "0.2"
-procfs = "0.14"
+procfs = "0.16"
 prometheus = {version = "0.13", default-features=false, features = ["process"]} # removes protobuf dependency
 prost = "0.11"
 rand = "0.8"
@@ -184,14 +186,17 @@ tower-service = "0.3.2"
 tracing = "0.1"
 tracing-error = "0.2.0"
 tracing-opentelemetry = "0.21.0"
-tracing-subscriber = { version = "0.3", default-features = false, features = ["smallvec", "fmt", "tracing-log", "std", "env-filter", "json", "ansi"] }
+tracing-subscriber = { version = "0.3", default-features = false, features = ["smallvec", "fmt", "tracing-log", "std", "env-filter", "json"] }
+try-lock = "0.2.5"
 twox-hash = { version = "1.6.3", default-features = false }
+typed-json = "0.1"
 url = "2.2"
 urlencoding = "2.1"
 uuid = { version = "1.6.1", features = ["v4", "v7", "serde"] }
 walkdir = "2.3.2"
 rustls-native-certs = "0.7"
 x509-parser = "0.15"
+whoami = "1.5.1"
 
 ## TODO replace this with tracing
 env_logger = "0.10"
@@ -203,9 +208,6 @@ postgres-protocol = { git = "https://github.com/neondatabase/rust-postgres.git",
 postgres-types = { git = "https://github.com/neondatabase/rust-postgres.git", branch="neon" }
 tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", branch="neon" }
 
-## Other git libraries
-heapless = { default-features=false, features=[], git = "https://github.com/japaric/heapless.git", rev = "644653bf3b831c6bb4963be2de24804acf5e5001" } # upstream release pending
-
 ## Local libraries
 compute_api = { version = "0.1", path = "./libs/compute_api/" }
 consumption_metrics = { version = "0.1", path = "./libs/consumption_metrics/" }
@@ -221,6 +223,7 @@ remote_storage = { version = "0.1", path = "./libs/remote_storage/" }
 safekeeper_api = { version = "0.1", path = "./libs/safekeeper_api" }
 desim = { version = "0.1", path = "./libs/desim" }
 storage_broker = { version = "0.1", path = "./storage_broker/" } # Note: main broker code is inside the binary crate, so linking with the library shouldn't be heavy.
+storage_controller_client = { path = "./storage_controller/client" }
 tenant_size_model = { version = "0.1", path = "./libs/tenant_size_model/" }
 tracing-utils = { version = "0.1", path = "./libs/tracing-utils/" }
 utils = { version = "0.1", path = "./libs/utils/" }

Dockerfile

@@ -17,7 +17,7 @@ COPY --chown=nonroot pgxn pgxn
 COPY --chown=nonroot Makefile Makefile
 COPY --chown=nonroot scripts/ninstall.sh scripts/ninstall.sh
 
-ENV BUILD_TYPE release
+ENV BUILD_TYPE=release
 RUN set -e \
     && mold -run make -j $(nproc) -s neon-pg-ext \
     && rm -rf pg_install/build \
@@ -29,25 +29,14 @@ WORKDIR /home/nonroot
 ARG GIT_VERSION=local
 ARG BUILD_TAG
 
-# Enable https://github.com/paritytech/cachepot to cache Rust crates' compilation results in Docker builds.
-# Set up cachepot to use an AWS S3 bucket for cache results, to reuse it between `docker build` invocations.
-# cachepot falls back to local filesystem if S3 is misconfigured, not failing the build
-ARG RUSTC_WRAPPER=cachepot
-ENV AWS_REGION=eu-central-1
-ENV CACHEPOT_S3_KEY_PREFIX=cachepot
-ARG CACHEPOT_BUCKET=neon-github-dev
-#ARG AWS_ACCESS_KEY_ID
-#ARG AWS_SECRET_ACCESS_KEY
-
 COPY --from=pg-build /home/nonroot/pg_install/v14/include/postgresql/server pg_install/v14/include/postgresql/server
 COPY --from=pg-build /home/nonroot/pg_install/v15/include/postgresql/server pg_install/v15/include/postgresql/server
 COPY --from=pg-build /home/nonroot/pg_install/v16/include/postgresql/server pg_install/v16/include/postgresql/server
+COPY --from=pg-build /home/nonroot/pg_install/v16/lib                       pg_install/v16/lib
 COPY --chown=nonroot . .
 
-# Show build caching stats to check if it was used in the end.
-# Has to be the part of the same RUN since cachepot daemon is killed in the end of this RUN, losing the compilation stats.
 RUN set -e \
-    && RUSTFLAGS="-Clinker=clang -Clink-arg=-fuse-ld=mold -Clink-arg=-Wl,--no-rosegment" cargo build  \
+    && PQ_LIB_DIR=$(pwd)/pg_install/v16/lib RUSTFLAGS="-Clinker=clang -Clink-arg=-fuse-ld=mold -Clink-arg=-Wl,--no-rosegment" cargo build \
       --bin pg_sni_router  \
       --bin pageserver  \
       --bin pagectl  \
@@ -56,8 +45,8 @@ RUN set -e \
       --bin storage_controller  \
       --bin proxy  \
       --bin neon_local \
-      --locked --release \
-    && cachepot -s
+      --bin storage_scrubber \
+      --locked --release
 
 # Build final image
 #
@@ -82,6 +71,7 @@ COPY --from=build --chown=neon:neon /home/nonroot/target/release/storage_broker
 COPY --from=build --chown=neon:neon /home/nonroot/target/release/storage_controller  /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
+COPY --from=build --chown=neon:neon /home/nonroot/target/release/storage_scrubber    /usr/local/bin
 
 COPY --from=pg-build /home/nonroot/pg_install/v14 /usr/local/v14/
 COPY --from=pg-build /home/nonroot/pg_install/v15 /usr/local/v15/
@@ -90,20 +80,24 @@ COPY --from=pg-build /home/nonroot/postgres_install.tar.gz /data/
 
 # By default, pageserver uses `.neon/` working directory in WORKDIR, so create one and fill it with the dummy config.
 # Now, when `docker run ... pageserver` is run, it can start without errors, yet will have some default dummy values.
-RUN mkdir -p /data/.neon/ && chown -R neon:neon /data/.neon/ \
-    && /usr/local/bin/pageserver -D /data/.neon/ --init \
-       -c "id=1234" \
-       -c "broker_endpoint='http://storage_broker:50051'" \
-       -c "pg_distrib_dir='/usr/local/'" \
-       -c "listen_pg_addr='0.0.0.0:6400'" \
-       -c "listen_http_addr='0.0.0.0:9898'"
+RUN mkdir -p /data/.neon/ && \
+  echo "id=1234" > "/data/.neon/identity.toml" && \
+  echo "broker_endpoint='http://storage_broker:50051'\n" \
+       "pg_distrib_dir='/usr/local/'\n" \
+       "listen_pg_addr='0.0.0.0:6400'\n" \
+       "listen_http_addr='0.0.0.0:9898'\n" \
+  > /data/.neon/pageserver.toml && \
+  chown -R neon:neon /data/.neon
 
 # When running a binary that links with libpq, default to using our most recent postgres version.  Binaries
 # that want a particular postgres version will select it explicitly: this is just a default.
-ENV LD_LIBRARY_PATH /usr/local/v16/lib
+ENV LD_LIBRARY_PATH=/usr/local/v16/lib
 
 
 VOLUME ["/data"]
 USER neon
 EXPOSE 6400
 EXPOSE 9898
+
+CMD ["/usr/local/bin/pageserver", "-D", "/data/.neon"]
+

Dockerfile.build-tools

@@ -1,5 +1,13 @@
 FROM debian:bullseye-slim
 
+# Use ARG as a build-time environment variable here to allow.
+# It's not supposed to be set outside.
+# Alternatively it can be obtained using the following command
+# ```
+# . /etc/os-release && echo "${VERSION_CODENAME}"
+# ```
+ARG DEBIAN_VERSION_CODENAME=bullseye
+
 # Add nonroot user
 RUN useradd -ms /bin/bash nonroot -b /home
 SHELL ["/bin/bash", "-c"]
@@ -26,7 +34,6 @@ RUN set -e \
         liblzma-dev \
         libncurses5-dev \
         libncursesw5-dev \
-        libpq-dev \
         libreadline-dev \
         libseccomp-dev \
         libsqlite3-dev \
@@ -51,7 +58,7 @@ RUN set -e \
     && rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/*
 
 # protobuf-compiler (protoc)
-ENV PROTOC_VERSION 25.1
+ENV PROTOC_VERSION=25.1
 RUN curl -fsSL "https://github.com/protocolbuffers/protobuf/releases/download/v${PROTOC_VERSION}/protoc-${PROTOC_VERSION}-linux-$(uname -m | sed 's/aarch64/aarch_64/g').zip" -o "protoc.zip" \
     && unzip -q protoc.zip -d protoc \
     && mv protoc/bin/protoc /usr/local/bin/protoc \
@@ -67,12 +74,24 @@ RUN curl -sL "https://github.com/peak/s5cmd/releases/download/v${S5CMD_VERSION}/
 # LLVM
 ENV LLVM_VERSION=18
 RUN curl -fsSL 'https://apt.llvm.org/llvm-snapshot.gpg.key' | apt-key add - \
-    && echo "deb http://apt.llvm.org/bullseye/ llvm-toolchain-bullseye-${LLVM_VERSION} main" > /etc/apt/sources.list.d/llvm.stable.list \
+    && echo "deb http://apt.llvm.org/${DEBIAN_VERSION_CODENAME}/ llvm-toolchain-${DEBIAN_VERSION_CODENAME}-${LLVM_VERSION} main" > /etc/apt/sources.list.d/llvm.stable.list \
     && apt update \
     && apt install -y clang-${LLVM_VERSION} llvm-${LLVM_VERSION} \
     && bash -c 'for f in /usr/bin/clang*-${LLVM_VERSION} /usr/bin/llvm*-${LLVM_VERSION}; do ln -s "${f}" "${f%-${LLVM_VERSION}}"; done' \
     && rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/*
 
+# Install docker
+RUN curl -fsSL https://download.docker.com/linux/ubuntu/gpg | gpg --dearmor -o /usr/share/keyrings/docker-archive-keyring.gpg \
+    && echo "deb [arch=$(dpkg --print-architecture) signed-by=/usr/share/keyrings/docker-archive-keyring.gpg] https://download.docker.com/linux/debian ${DEBIAN_VERSION_CODENAME} stable" > /etc/apt/sources.list.d/docker.list \
+    && apt update \
+    && apt install -y docker-ce docker-ce-cli \
+    && rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/*
+
+# Configure sudo & docker
+RUN usermod -aG sudo nonroot && \
+    echo '%sudo ALL=(ALL) NOPASSWD:ALL' >> /etc/sudoers && \
+    usermod -aG docker nonroot
+
 # AWS CLI
 RUN curl "https://awscli.amazonaws.com/awscli-exe-linux-$(uname -m).zip" -o "awscliv2.zip" \
     && unzip -q awscliv2.zip \
@@ -80,7 +99,7 @@ RUN curl "https://awscli.amazonaws.com/awscli-exe-linux-$(uname -m).zip" -o "aws
     && rm awscliv2.zip
 
 # Mold: A Modern Linker
-ENV MOLD_VERSION v2.31.0
+ENV MOLD_VERSION=v2.33.0
 RUN set -e \
     && git clone https://github.com/rui314/mold.git \
     && mkdir mold/build \
@@ -149,7 +168,7 @@ USER nonroot:nonroot
 WORKDIR /home/nonroot
 
 # Python
-ENV PYTHON_VERSION=3.9.18 \
+ENV PYTHON_VERSION=3.9.19 \
     PYENV_ROOT=/home/nonroot/.pyenv \
     PATH=/home/nonroot/.pyenv/shims:/home/nonroot/.pyenv/bin:/home/nonroot/.poetry/bin:$PATH
 RUN set -e \
@@ -173,9 +192,14 @@ WORKDIR /home/nonroot
 
 # Rust
 # Please keep the version of llvm (installed above) in sync with rust llvm (`rustc --version --verbose | grep LLVM`)
-ENV RUSTC_VERSION=1.79.0
+ENV RUSTC_VERSION=1.80.1
 ENV RUSTUP_HOME="/home/nonroot/.rustup"
 ENV PATH="/home/nonroot/.cargo/bin:${PATH}"
+ARG RUSTFILT_VERSION=0.2.1
+ARG CARGO_HAKARI_VERSION=0.9.30
+ARG CARGO_DENY_VERSION=0.16.1
+ARG CARGO_HACK_VERSION=0.6.31
+ARG CARGO_NEXTEST_VERSION=0.9.72
 RUN curl -sSO https://static.rust-lang.org/rustup/dist/$(uname -m)-unknown-linux-gnu/rustup-init && whoami && \
 	chmod +x rustup-init && \
 	./rustup-init -y --default-toolchain ${RUSTC_VERSION} && \
@@ -184,15 +208,13 @@ RUN curl -sSO https://static.rust-lang.org/rustup/dist/$(uname -m)-unknown-linux
     . "$HOME/.cargo/env" && \
     cargo --version && rustup --version && \
     rustup component add llvm-tools-preview rustfmt clippy && \
-    cargo install --git https://github.com/paritytech/cachepot && \
-    cargo install rustfilt && \
-    cargo install cargo-hakari && \
-    cargo install cargo-deny --locked && \
-    cargo install cargo-hack && \
-    cargo install cargo-nextest && \
+    cargo install rustfilt            --version ${RUSTFILT_VERSION} && \
+    cargo install cargo-hakari        --version ${CARGO_HAKARI_VERSION} && \
+    cargo install cargo-deny --locked --version ${CARGO_DENY_VERSION} && \
+    cargo install cargo-hack          --version ${CARGO_HACK_VERSION} && \
+    cargo install cargo-nextest       --version ${CARGO_NEXTEST_VERSION} && \
     rm -rf /home/nonroot/.cargo/registry && \
     rm -rf /home/nonroot/.cargo/git
-ENV RUSTC_WRAPPER=cachepot
 
 # Show versions
 RUN whoami \

Dockerfile.compute-node

@@ -94,7 +94,7 @@ RUN wget https://gitlab.com/Oslandia/SFCGAL/-/archive/v1.3.10/SFCGAL-v1.3.10.tar
     DESTDIR=/sfcgal make install -j $(getconf _NPROCESSORS_ONLN) && \
     make clean && cp -R /sfcgal/* /
 
-ENV PATH "/usr/local/pgsql/bin:$PATH"
+ENV PATH="/usr/local/pgsql/bin:$PATH"
 
 RUN wget https://download.osgeo.org/postgis/source/postgis-3.3.3.tar.gz -O postgis.tar.gz && \
     echo "74eb356e3f85f14233791013360881b6748f78081cc688ff9d6f0f673a762d13 postgis.tar.gz" | sha256sum --check && \
@@ -311,9 +311,12 @@ RUN wget https://github.com/iCyberon/pg_hashids/archive/refs/tags/v1.2.1.tar.gz
 FROM build-deps AS rum-pg-build
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
+COPY patches/rum.patch /rum.patch
+
 RUN wget https://github.com/postgrespro/rum/archive/refs/tags/1.3.13.tar.gz -O rum.tar.gz && \
     echo "6ab370532c965568df6210bd844ac6ba649f53055e48243525b0b7e5c4d69a7d rum.tar.gz" | sha256sum --check && \
     mkdir rum-src && cd rum-src && tar xzf ../rum.tar.gz --strip-components=1 -C . && \
+    patch -p1 < /rum.patch && \
     make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config USE_PGXS=1 && \
     make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config USE_PGXS=1 && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/rum.control
@@ -408,7 +411,7 @@ FROM build-deps AS timescaledb-pg-build
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 ARG PG_VERSION
-ENV PATH "/usr/local/pgsql/bin:$PATH"
+ENV PATH="/usr/local/pgsql/bin:$PATH"
 
 RUN case "${PG_VERSION}" in \
       "v14" | "v15") \
@@ -441,7 +444,7 @@ FROM build-deps AS pg-hint-plan-pg-build
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 ARG PG_VERSION
-ENV PATH "/usr/local/pgsql/bin:$PATH"
+ENV PATH="/usr/local/pgsql/bin:$PATH"
 
 RUN case "${PG_VERSION}" in \
       "v14") \
@@ -477,7 +480,7 @@ RUN case "${PG_VERSION}" in \
 FROM build-deps AS pg-cron-pg-build
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
-ENV PATH "/usr/local/pgsql/bin/:$PATH"
+ENV PATH="/usr/local/pgsql/bin/:$PATH"
 RUN wget https://github.com/citusdata/pg_cron/archive/refs/tags/v1.6.0.tar.gz -O pg_cron.tar.gz && \
     echo "383a627867d730222c272bfd25cd5e151c578d73f696d32910c7db8c665cc7db pg_cron.tar.gz" | sha256sum --check && \
     mkdir pg_cron-src && cd pg_cron-src && tar xzf ../pg_cron.tar.gz --strip-components=1 -C . && \
@@ -503,7 +506,7 @@ RUN apt-get update && \
         libboost-system1.74-dev \
         libeigen3-dev
 
-ENV PATH "/usr/local/pgsql/bin/:/usr/local/pgsql/:$PATH"
+ENV PATH="/usr/local/pgsql/bin/:/usr/local/pgsql/:$PATH"
 RUN wget https://github.com/rdkit/rdkit/archive/refs/tags/Release_2023_03_3.tar.gz -O rdkit.tar.gz && \
     echo "bdbf9a2e6988526bfeb8c56ce3cdfe2998d60ac289078e2215374288185e8c8d rdkit.tar.gz" | sha256sum --check && \
     mkdir rdkit-src && cd rdkit-src && tar xzf ../rdkit.tar.gz --strip-components=1 -C . && \
@@ -543,7 +546,7 @@ RUN wget https://github.com/rdkit/rdkit/archive/refs/tags/Release_2023_03_3.tar.
 FROM build-deps AS pg-uuidv7-pg-build
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
-ENV PATH "/usr/local/pgsql/bin/:$PATH"
+ENV PATH="/usr/local/pgsql/bin/:$PATH"
 RUN wget https://github.com/fboulnois/pg_uuidv7/archive/refs/tags/v1.0.1.tar.gz -O pg_uuidv7.tar.gz && \
     echo "0d0759ab01b7fb23851ecffb0bce27822e1868a4a5819bfd276101c716637a7a pg_uuidv7.tar.gz" | sha256sum --check && \
     mkdir pg_uuidv7-src && cd pg_uuidv7-src && tar xzf ../pg_uuidv7.tar.gz --strip-components=1 -C . && \
@@ -560,7 +563,7 @@ RUN wget https://github.com/fboulnois/pg_uuidv7/archive/refs/tags/v1.0.1.tar.gz
 FROM build-deps AS pg-roaringbitmap-pg-build
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
-ENV PATH "/usr/local/pgsql/bin/:$PATH"
+ENV PATH="/usr/local/pgsql/bin/:$PATH"
 RUN wget https://github.com/ChenHuajun/pg_roaringbitmap/archive/refs/tags/v0.5.4.tar.gz -O pg_roaringbitmap.tar.gz && \
     echo "b75201efcb1c2d1b014ec4ae6a22769cc7a224e6e406a587f5784a37b6b5a2aa pg_roaringbitmap.tar.gz" | sha256sum --check && \
     mkdir pg_roaringbitmap-src && cd pg_roaringbitmap-src && tar xzf ../pg_roaringbitmap.tar.gz --strip-components=1 -C . && \
@@ -577,7 +580,7 @@ RUN wget https://github.com/ChenHuajun/pg_roaringbitmap/archive/refs/tags/v0.5.4
 FROM build-deps AS pg-semver-pg-build
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
-ENV PATH "/usr/local/pgsql/bin/:$PATH"
+ENV PATH="/usr/local/pgsql/bin/:$PATH"
 RUN wget https://github.com/theory/pg-semver/archive/refs/tags/v0.32.1.tar.gz -O pg_semver.tar.gz && \
     echo "fbdaf7512026d62eec03fad8687c15ed509b6ba395bff140acd63d2e4fbe25d7 pg_semver.tar.gz" | sha256sum --check && \
     mkdir pg_semver-src && cd pg_semver-src && tar xzf ../pg_semver.tar.gz --strip-components=1 -C . && \
@@ -595,7 +598,7 @@ FROM build-deps AS pg-embedding-pg-build
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
 ARG PG_VERSION
-ENV PATH "/usr/local/pgsql/bin/:$PATH"
+ENV PATH="/usr/local/pgsql/bin/:$PATH"
 RUN case "${PG_VERSION}" in \
       "v14" | "v15") \
         export PG_EMBEDDING_VERSION=0.3.5 \
@@ -619,7 +622,7 @@ RUN case "${PG_VERSION}" in \
 FROM build-deps AS pg-anon-pg-build
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
-ENV PATH "/usr/local/pgsql/bin/:$PATH"
+ENV PATH="/usr/local/pgsql/bin/:$PATH"
 RUN wget  https://github.com/neondatabase/postgresql_anonymizer/archive/refs/tags/neon_1.1.1.tar.gz -O pg_anon.tar.gz && \
     echo "321ea8d5c1648880aafde850a2c576e4a9e7b9933a34ce272efc839328999fa9  pg_anon.tar.gz" | sha256sum --check && \
     mkdir pg_anon-src && cd pg_anon-src && tar xzf ../pg_anon.tar.gz --strip-components=1 -C . && \
@@ -654,7 +657,7 @@ RUN curl -sSO https://static.rust-lang.org/rustup/dist/$(uname -m)-unknown-linux
     chmod +x rustup-init && \
     ./rustup-init -y --no-modify-path --profile minimal --default-toolchain stable && \
     rm rustup-init && \
-    cargo install --locked --version 0.10.2 cargo-pgrx && \
+    cargo install --locked --version 0.11.3 cargo-pgrx && \
     /bin/bash -c 'cargo pgrx init --pg${PG_VERSION:1}=/usr/local/pgsql/bin/pg_config'
 
 USER root
@@ -669,10 +672,15 @@ USER root
 FROM rust-extensions-build AS pg-jsonschema-pg-build
 ARG PG_VERSION
 
-RUN wget https://github.com/supabase/pg_jsonschema/archive/refs/tags/v0.2.0.tar.gz -O pg_jsonschema.tar.gz && \
-    echo "9118fc508a6e231e7a39acaa6f066fcd79af17a5db757b47d2eefbe14f7794f0 pg_jsonschema.tar.gz" | sha256sum --check && \
+RUN wget https://github.com/supabase/pg_jsonschema/archive/refs/tags/v0.3.1.tar.gz -O pg_jsonschema.tar.gz && \
+    echo "61df3db1ed83cf24f6aa39c826f8818bfa4f0bd33b587fd6b2b1747985642297 pg_jsonschema.tar.gz" | sha256sum --check && \
     mkdir pg_jsonschema-src && cd pg_jsonschema-src && tar xzf ../pg_jsonschema.tar.gz --strip-components=1 -C . && \
-    sed -i 's/pgrx = "0.10.2"/pgrx = { version = "0.10.2", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    # see commit 252b3685a27a0f4c31a0f91e983c6314838e89e8
+    # `unsafe-postgres` feature allows to build pgx extensions
+    # against postgres forks that decided to change their ABI name (like us).
+    # With that we can build extensions without forking them and using stock
+    # pgx. As this feature is new few manual version bumps were required.
+    sed -i 's/pgrx = "0.11.3"/pgrx = { version = "0.11.3", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
     cargo pgrx install --release && \
     echo "trusted = true" >> /usr/local/pgsql/share/extension/pg_jsonschema.control
 
@@ -686,10 +694,10 @@ RUN wget https://github.com/supabase/pg_jsonschema/archive/refs/tags/v0.2.0.tar.
 FROM rust-extensions-build AS pg-graphql-pg-build
 ARG PG_VERSION
 
-RUN wget https://github.com/supabase/pg_graphql/archive/refs/tags/v1.4.0.tar.gz -O pg_graphql.tar.gz && \
-    echo "bd8dc7230282b3efa9ae5baf053a54151ed0e66881c7c53750e2d0c765776edc pg_graphql.tar.gz" | sha256sum --check && \
+RUN wget https://github.com/supabase/pg_graphql/archive/refs/tags/v1.5.7.tar.gz -O pg_graphql.tar.gz && \
+    echo "2b3e567a5b31019cb97ae0e33263c1bcc28580be5a444ac4c8ece5c4be2aea41 pg_graphql.tar.gz" | sha256sum --check && \
     mkdir pg_graphql-src && cd pg_graphql-src && tar xzf ../pg_graphql.tar.gz --strip-components=1 -C . && \
-    sed -i 's/pgrx = "=0.10.2"/pgrx = { version = "0.10.2", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    sed -i 's/pgrx = "=0.11.3"/pgrx = { version = "0.11.3", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
     cargo pgrx install --release && \
     # it's needed to enable extension because it uses untrusted C language
     sed -i 's/superuser = false/superuser = true/g' /usr/local/pgsql/share/extension/pg_graphql.control && \
@@ -709,6 +717,9 @@ ARG PG_VERSION
 RUN wget https://github.com/kelvich/pg_tiktoken/archive/26806147b17b60763039c6a6878884c41a262318.tar.gz -O pg_tiktoken.tar.gz && \
     echo "e64e55aaa38c259512d3e27c572da22c4637418cf124caba904cd50944e5004e pg_tiktoken.tar.gz" | sha256sum --check && \
     mkdir pg_tiktoken-src && cd pg_tiktoken-src && tar xzf ../pg_tiktoken.tar.gz --strip-components=1 -C . && \
+    # TODO update pgrx version in the pg_tiktoken repo and remove this line
+    sed -i 's/pgrx = { version = "=0.10.2",/pgrx = { version = "0.11.3",/g' Cargo.toml && \
+    sed -i 's/pgrx-tests = "=0.10.2"/pgrx-tests = "0.11.3"/g' Cargo.toml && \
     cargo pgrx install --release && \
     echo "trusted = true" >> /usr/local/pgsql/share/extension/pg_tiktoken.control
 
@@ -722,14 +733,10 @@ RUN wget https://github.com/kelvich/pg_tiktoken/archive/26806147b17b60763039c6a6
 FROM rust-extensions-build AS pg-pgx-ulid-build
 ARG PG_VERSION
 
-RUN wget https://github.com/pksunkara/pgx_ulid/archive/refs/tags/v0.1.3.tar.gz -O pgx_ulid.tar.gz && \
-    echo "ee5db82945d2d9f2d15597a80cf32de9dca67b897f605beb830561705f12683c pgx_ulid.tar.gz" | sha256sum --check && \
+RUN wget https://github.com/pksunkara/pgx_ulid/archive/refs/tags/v0.1.5.tar.gz -O pgx_ulid.tar.gz && \
+    echo "9d1659a2da65af0133d5451c454de31b37364e3502087dadf579f790bc8bef17 pgx_ulid.tar.gz" | sha256sum --check && \
     mkdir pgx_ulid-src && cd pgx_ulid-src && tar xzf ../pgx_ulid.tar.gz --strip-components=1 -C . && \
-    echo "******************* Apply a patch for Postgres 16 support; delete in the next release ******************" && \
-    wget https://github.com/pksunkara/pgx_ulid/commit/f84954cf63fc8c80d964ac970d9eceed3c791196.patch && \
-    patch -p1 < f84954cf63fc8c80d964ac970d9eceed3c791196.patch && \
-    echo "********************************************************************************************************" && \
-    sed -i 's/pgrx       = "=0.10.2"/pgrx = { version = "=0.10.2", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    sed -i 's/pgrx       = "^0.11.2"/pgrx = { version = "=0.11.3", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
     cargo pgrx install --release && \
     echo "trusted = true" >> /usr/local/pgsql/share/extension/ulid.control
 
@@ -743,7 +750,7 @@ RUN wget https://github.com/pksunkara/pgx_ulid/archive/refs/tags/v0.1.3.tar.gz -
 FROM build-deps AS wal2json-pg-build
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
-ENV PATH "/usr/local/pgsql/bin/:$PATH"
+ENV PATH="/usr/local/pgsql/bin/:$PATH"
 RUN wget https://github.com/eulerto/wal2json/archive/refs/tags/wal2json_2_5.tar.gz && \
     echo "b516653575541cf221b99cf3f8be9b6821f6dbcfc125675c85f35090f824f00e wal2json_2_5.tar.gz" | sha256sum --check && \
     mkdir wal2json-src && cd wal2json-src && tar xzf ../wal2json_2_5.tar.gz --strip-components=1 -C . && \
@@ -759,7 +766,7 @@ RUN wget https://github.com/eulerto/wal2json/archive/refs/tags/wal2json_2_5.tar.
 FROM build-deps AS pg-ivm-build
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
-ENV PATH "/usr/local/pgsql/bin/:$PATH"
+ENV PATH="/usr/local/pgsql/bin/:$PATH"
 RUN wget https://github.com/sraoss/pg_ivm/archive/refs/tags/v1.7.tar.gz -O pg_ivm.tar.gz && \
     echo "ebfde04f99203c7be4b0e873f91104090e2e83e5429c32ac242d00f334224d5e pg_ivm.tar.gz" | sha256sum --check && \
     mkdir pg_ivm-src && cd pg_ivm-src && tar xzf ../pg_ivm.tar.gz --strip-components=1 -C . && \
@@ -776,7 +783,7 @@ RUN wget https://github.com/sraoss/pg_ivm/archive/refs/tags/v1.7.tar.gz -O pg_iv
 FROM build-deps AS pg-partman-build
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
-ENV PATH "/usr/local/pgsql/bin/:$PATH"
+ENV PATH="/usr/local/pgsql/bin/:$PATH"
 RUN wget https://github.com/pgpartman/pg_partman/archive/refs/tags/v5.0.1.tar.gz -O pg_partman.tar.gz && \
     echo "75b541733a9659a6c90dbd40fccb904a630a32880a6e3044d0c4c5f4c8a65525 pg_partman.tar.gz" | sha256sum --check && \
     mkdir pg_partman-src && cd pg_partman-src && tar xzf ../pg_partman.tar.gz --strip-components=1 -C . && \
@@ -926,7 +933,8 @@ COPY --from=pgjwt-pg-build /pgjwt.tar.gz /ext-src
 #COPY --from=pg-tiktoken-pg-build /home/nonroot/pg_tiktoken.tar.gz /ext-src
 COPY --from=hypopg-pg-build /hypopg.tar.gz /ext-src
 COPY --from=pg-hashids-pg-build /pg_hashids.tar.gz /ext-src
-#COPY --from=rum-pg-build /rum.tar.gz /ext-src
+COPY --from=rum-pg-build /rum.tar.gz /ext-src
+COPY patches/rum.patch /ext-src
 #COPY --from=pgtap-pg-build /pgtap.tar.gz /ext-src
 COPY --from=ip4r-pg-build /ip4r.tar.gz /ext-src
 COPY --from=prefix-pg-build /prefix.tar.gz /ext-src
@@ -938,7 +946,7 @@ COPY patches/pg_hintplan.patch /ext-src
 COPY --from=pg-cron-pg-build /pg_cron.tar.gz /ext-src
 COPY patches/pg_cron.patch /ext-src
 #COPY --from=pg-pgx-ulid-build /home/nonroot/pgx_ulid.tar.gz /ext-src
-COPY --from=rdkit-pg-build /rdkit.tar.gz /ext-src
+#COPY --from=rdkit-pg-build /rdkit.tar.gz /ext-src
 COPY --from=pg-uuidv7-pg-build /pg_uuidv7.tar.gz /ext-src
 COPY --from=pg-roaringbitmap-pg-build /pg_roaringbitmap.tar.gz /ext-src
 COPY --from=pg-semver-pg-build /pg_semver.tar.gz /ext-src
@@ -953,6 +961,7 @@ RUN cd /ext-src/ && for f in *.tar.gz; \
     rm -rf $dname; mkdir $dname; tar xzf $f --strip-components=1 -C $dname \
     || exit 1; rm -f $f; done
 RUN cd /ext-src/pgvector-src && patch -p1 <../pgvector.patch
+RUN cd /ext-src/rum-src && patch -p1 <../rum.patch
 # cmake is required for the h3 test
 RUN apt-get update && apt-get install -y cmake
 RUN patch -p1 < /ext-src/pg_hintplan.patch
@@ -1025,6 +1034,6 @@ RUN apt update &&  \
     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
 
-ENV LANG en_US.utf8
+ENV LANG=en_US.utf8
 USER postgres
 ENTRYPOINT ["/usr/local/bin/compute_ctl"]

Makefile

@@ -69,6 +69,8 @@ CARGO_CMD_PREFIX += CARGO_TERM_PROGRESS_WHEN=never CI=1
 # Set PQ_LIB_DIR to make sure `storage_controller` get linked with bundled libpq (through diesel)
 CARGO_CMD_PREFIX += PQ_LIB_DIR=$(POSTGRES_INSTALL_DIR)/v16/lib
 
+CACHEDIR_TAG_CONTENTS := "Signature: 8a477f597d28d172789f06886806bc55"
+
 #
 # Top level Makefile to build Neon and PostgreSQL
 #
@@ -79,15 +81,24 @@ all: neon postgres neon-pg-ext
 #
 # The 'postgres_ffi' depends on the Postgres headers.
 .PHONY: neon
-neon: postgres-headers walproposer-lib
+neon: postgres-headers walproposer-lib cargo-target-dir
 	+@echo "Compiling Neon"
 	$(CARGO_CMD_PREFIX) cargo build $(CARGO_BUILD_FLAGS)
+.PHONY: cargo-target-dir
+cargo-target-dir:
+	# https://github.com/rust-lang/cargo/issues/14281
+	mkdir -p target
+	test -e target/CACHEDIR.TAG || echo "$(CACHEDIR_TAG_CONTENTS)" > target/CACHEDIR.TAG
 
 ### PostgreSQL parts
 # Some rules are duplicated for Postgres v14 and 15. We may want to refactor
 # to avoid the duplication in the future, but it's tolerable for now.
 #
 $(POSTGRES_INSTALL_DIR)/build/%/config.status:
+
+	mkdir -p $(POSTGRES_INSTALL_DIR)
+	test -e $(POSTGRES_INSTALL_DIR)/CACHEDIR.TAG || echo "$(CACHEDIR_TAG_CONTENTS)" > $(POSTGRES_INSTALL_DIR)/CACHEDIR.TAG
+
 	+@echo "Configuring Postgres $* build"
 	@test -s $(ROOT_PROJECT_DIR)/vendor/postgres-$*/configure || { \
 		echo "\nPostgres submodule not found in $(ROOT_PROJECT_DIR)/vendor/postgres-$*/, execute "; \

compute_tools/Cargo.toml

@@ -4,6 +4,11 @@ version = "0.1.0"
 edition.workspace = true
 license.workspace = true
 
+[features]
+default = []
+# Enables test specific features.
+testing = []
+
 [dependencies]
 anyhow.workspace = true
 async-compression.workspace = true
@@ -44,3 +49,4 @@ vm_monitor = { version = "0.1", path = "../libs/vm_monitor/" }
 zstd = "0.13"
 bytes = "1.0"
 rust-ini = "0.20.0"
+rlimit = "0.10.1"

compute_tools/src/bin/compute_ctl.rs

@@ -6,7 +6,7 @@
 //! - Every start is a fresh start, so the data directory is removed and
 //!   initialized again on each run.
 //! - If remote_extension_config is provided, it will be used to fetch extensions list
-//!  and download `shared_preload_libraries` from the remote storage.
+//!   and download `shared_preload_libraries` from the remote storage.
 //! - Next it will put configuration files into the `PGDATA` directory.
 //! - Sync safekeepers and get commit LSN.
 //! - Get `basebackup` from pageserver using the returned on the previous step LSN.
@@ -33,7 +33,6 @@
 //!             -b /usr/local/bin/postgres \
 //!             -r http://pg-ext-s3-gateway \
 //! ```
-//!
 use std::collections::HashMap;
 use std::fs::File;
 use std::path::Path;
@@ -64,6 +63,7 @@ use compute_tools::monitor::launch_monitor;
 use compute_tools::params::*;
 use compute_tools::spec::*;
 use compute_tools::swap::resize_swap;
+use rlimit::{setrlimit, Resource};
 
 // this is an arbitrary build tag. Fine as a default / for testing purposes
 // in-case of not-set environment var
@@ -72,6 +72,9 @@ const BUILD_TAG_DEFAULT: &str = "latest";
 fn main() -> Result<()> {
     let (build_tag, clap_args) = init()?;
 
+    // enable core dumping for all child processes
+    setrlimit(Resource::CORE, rlimit::INFINITY, rlimit::INFINITY)?;
+
     let (pg_handle, start_pg_result) = {
         // Enter startup tracing context
         let _startup_context_guard = startup_context_from_env();

compute_tools/src/compute.rs

@@ -56,6 +56,7 @@ pub struct ComputeNode {
     /// - we push new spec and it does reconfiguration
     /// - but then something happens and compute pod / VM is destroyed,
     ///   so k8s controller starts it again with the **old** spec
+    ///
     /// and the same for empty computes:
     /// - we started compute without any spec
     /// - we push spec and it does configuration
@@ -399,7 +400,15 @@ impl ComputeNode {
     pub fn get_basebackup(&self, compute_state: &ComputeState, lsn: Lsn) -> Result<()> {
         let mut retry_period_ms = 500.0;
         let mut attempts = 0;
-        let max_attempts = 10;
+        const DEFAULT_ATTEMPTS: u16 = 10;
+        #[cfg(feature = "testing")]
+        let max_attempts = if let Ok(v) = env::var("NEON_COMPUTE_TESTING_BASEBACKUP_RETRIES") {
+            u16::from_str(&v).unwrap()
+        } else {
+            DEFAULT_ATTEMPTS
+        };
+        #[cfg(not(feature = "testing"))]
+        let max_attempts = DEFAULT_ATTEMPTS;
         loop {
             let result = self.try_get_basebackup(compute_state, lsn);
             match result {
@@ -798,7 +807,11 @@ impl ComputeNode {
         // In this case we need to connect with old `zenith_admin` name
         // and create new user. We cannot simply rename connected user,
         // but we can create a new one and grant it all privileges.
-        let connstr = self.connstr.clone();
+        let mut connstr = self.connstr.clone();
+        connstr
+            .query_pairs_mut()
+            .append_pair("application_name", "apply_config");
+
         let mut client = match Client::connect(connstr.as_str(), NoTls) {
             Err(e) => match e.code() {
                 Some(&SqlState::INVALID_PASSWORD)
@@ -867,15 +880,19 @@ impl ComputeNode {
 
         // Run migrations separately to not hold up cold starts
         thread::spawn(move || {
+            let mut connstr = connstr.clone();
+            connstr
+                .query_pairs_mut()
+                .append_pair("application_name", "migrations");
+
             let mut client = Client::connect(connstr.as_str(), NoTls)?;
             handle_migrations(&mut client).context("apply_config handle_migrations")
         });
         Ok(())
     }
 
-    // We could've wrapped this around `pg_ctl reload`, but right now we don't use
-    // `pg_ctl` for start / stop, so this just seems much easier to do as we already
-    // have opened connection to Postgres and superuser access.
+    // Wrapped this around `pg_ctl reload`, but right now we don't use
+    // `pg_ctl` for start / stop.
     #[instrument(skip_all)]
     fn pg_reload_conf(&self) -> Result<()> {
         let pgctl_bin = Path::new(&self.pgbin).parent().unwrap().join("pg_ctl");
@@ -1108,7 +1125,7 @@ impl ComputeNode {
     // EKS worker nodes have following core dump settings:
     //   /proc/sys/kernel/core_pattern -> core
     //   /proc/sys/kernel/core_uses_pid -> 1
-    //   ulimint -c -> unlimited
+    //   ulimit -c -> unlimited
     // which results in core dumps being written to postgres data directory as core.<pid>.
     //
     // Use that as a default location and pattern, except macos where core dumps are written
@@ -1387,7 +1404,9 @@ pub fn forward_termination_signal() {
     let pg_pid = PG_PID.load(Ordering::SeqCst);
     if pg_pid != 0 {
         let pg_pid = nix::unistd::Pid::from_raw(pg_pid as i32);
-        // use 'immediate' shutdown (SIGQUIT): https://www.postgresql.org/docs/current/server-shutdown.html
-        kill(pg_pid, Signal::SIGQUIT).ok();
+        // Use 'fast' shutdown (SIGINT) because it also creates a shutdown checkpoint, which is important for
+        // ROs to get a list of running xacts faster instead of going through the CLOG.
+        // See https://www.postgresql.org/docs/current/server-shutdown.html for the list of modes and signals.
+        kill(pg_pid, Signal::SIGINT).ok();
     }
 }

compute_tools/src/lib.rs

@@ -11,6 +11,7 @@ pub mod logger;
 pub mod catalog;
 pub mod compute;
 pub mod extension_server;
+mod migration;
 pub mod monitor;
 pub mod params;
 pub mod pg_helpers;

compute_tools/src/migration.rs

@@ -0,0 +1,105 @@
+use anyhow::{Context, Result};
+use postgres::Client;
+use tracing::info;
+
+pub(crate) struct MigrationRunner<'m> {
+    client: &'m mut Client,
+    migrations: &'m [&'m str],
+}
+
+impl<'m> MigrationRunner<'m> {
+    pub fn new(client: &'m mut Client, migrations: &'m [&'m str]) -> Self {
+        // The neon_migration.migration_id::id column is a bigint, which is equivalent to an i64
+        assert!(migrations.len() + 1 < i64::MAX as usize);
+
+        Self { client, migrations }
+    }
+
+    fn get_migration_id(&mut self) -> Result<i64> {
+        let query = "SELECT id FROM neon_migration.migration_id";
+        let row = self
+            .client
+            .query_one(query, &[])
+            .context("run_migrations get migration_id")?;
+
+        Ok(row.get::<&str, i64>("id"))
+    }
+
+    fn update_migration_id(&mut self, migration_id: i64) -> Result<()> {
+        let setval = format!("UPDATE neon_migration.migration_id SET id={}", migration_id);
+
+        self.client
+            .simple_query(&setval)
+            .context("run_migrations update id")?;
+
+        Ok(())
+    }
+
+    fn prepare_migrations(&mut self) -> Result<()> {
+        let query = "CREATE SCHEMA IF NOT EXISTS neon_migration";
+        self.client.simple_query(query)?;
+
+        let query = "CREATE TABLE IF NOT EXISTS neon_migration.migration_id (key INT NOT NULL PRIMARY KEY, id bigint NOT NULL DEFAULT 0)";
+        self.client.simple_query(query)?;
+
+        let query = "INSERT INTO neon_migration.migration_id VALUES (0, 0) ON CONFLICT DO NOTHING";
+        self.client.simple_query(query)?;
+
+        let query = "ALTER SCHEMA neon_migration OWNER TO cloud_admin";
+        self.client.simple_query(query)?;
+
+        let query = "REVOKE ALL ON SCHEMA neon_migration FROM PUBLIC";
+        self.client.simple_query(query)?;
+
+        Ok(())
+    }
+
+    pub fn run_migrations(mut self) -> Result<()> {
+        self.prepare_migrations()?;
+
+        let mut current_migration = self.get_migration_id()? as usize;
+        while current_migration < self.migrations.len() {
+            macro_rules! migration_id {
+                ($cm:expr) => {
+                    ($cm + 1) as i64
+                };
+            }
+
+            let migration = self.migrations[current_migration];
+
+            if migration.starts_with("-- SKIP") {
+                info!("Skipping migration id={}", migration_id!(current_migration));
+            } else {
+                info!(
+                    "Running migration id={}:\n{}\n",
+                    migration_id!(current_migration),
+                    migration
+                );
+
+                self.client
+                    .simple_query("BEGIN")
+                    .context("begin migration")?;
+
+                self.client.simple_query(migration).with_context(|| {
+                    format!(
+                        "run_migrations migration id={}",
+                        migration_id!(current_migration)
+                    )
+                })?;
+
+                // Migration IDs start at 1
+                self.update_migration_id(migration_id!(current_migration))?;
+
+                self.client
+                    .simple_query("COMMIT")
+                    .context("commit migration")?;
+
+                info!("Finished migration id={}", migration_id!(current_migration));
+            }
+
+            current_migration += 1;
+        }
+
+        Ok(())
+    }
+}

compute_tools/src/migrations/0010-grant_snapshot_synchronization_funcs_to_neon_superuser.sql

@@ -0,0 +1,7 @@
+DO $$
+BEGIN
+    IF (SELECT setting::numeric >= 160000 FROM pg_settings WHERE name = 'server_version_num') THEN
+       EXECUTE 'GRANT EXECUTE ON FUNCTION pg_export_snapshot TO neon_superuser';
+       EXECUTE 'GRANT EXECUTE ON FUNCTION pg_log_standby_snapshot TO neon_superuser';
+    END IF;
+END $$;

compute_tools/src/pg_helpers.rs

@@ -489,7 +489,7 @@ pub fn handle_postgres_logs(stderr: std::process::ChildStderr) -> JoinHandle<()>
 /// Read Postgres logs from `stderr` until EOF. Buffer is flushed on one of the following conditions:
 /// - next line starts with timestamp
 /// - EOF
-/// - no new lines were written for the last second
+/// - no new lines were written for the last 100 milliseconds
 async fn handle_postgres_logs_async(stderr: tokio::process::ChildStderr) -> Result<()> {
     let mut lines = tokio::io::BufReader::new(stderr).lines();
     let timeout_duration = Duration::from_millis(100);

compute_tools/src/spec.rs

@@ -10,6 +10,7 @@ use tracing::{error, info, info_span, instrument, span_enabled, warn, Level};
 
 use crate::config;
 use crate::logger::inlinify;
+use crate::migration::MigrationRunner;
 use crate::params::PG_HBA_ALL_MD5;
 use crate::pg_helpers::*;
 
@@ -776,84 +777,25 @@ pub fn handle_migrations(client: &mut Client) -> Result<()> {
 
     // Add new migrations in numerical order.
     let migrations = [
-        include_str!("./migrations/0000-neon_superuser_bypass_rls.sql"),
-        include_str!("./migrations/0001-alter_roles.sql"),
-        include_str!("./migrations/0002-grant_pg_create_subscription_to_neon_superuser.sql"),
-        include_str!("./migrations/0003-grant_pg_monitor_to_neon_superuser.sql"),
-        include_str!("./migrations/0004-grant_all_on_tables_to_neon_superuser.sql"),
-        include_str!("./migrations/0005-grant_all_on_sequences_to_neon_superuser.sql"),
+        include_str!("./migrations/0001-neon_superuser_bypass_rls.sql"),
+        include_str!("./migrations/0002-alter_roles.sql"),
+        include_str!("./migrations/0003-grant_pg_create_subscription_to_neon_superuser.sql"),
+        include_str!("./migrations/0004-grant_pg_monitor_to_neon_superuser.sql"),
+        include_str!("./migrations/0005-grant_all_on_tables_to_neon_superuser.sql"),
+        include_str!("./migrations/0006-grant_all_on_sequences_to_neon_superuser.sql"),
         include_str!(
-            "./migrations/0006-grant_all_on_tables_to_neon_superuser_with_grant_option.sql"
+            "./migrations/0007-grant_all_on_tables_to_neon_superuser_with_grant_option.sql"
         ),
         include_str!(
-            "./migrations/0007-grant_all_on_sequences_to_neon_superuser_with_grant_option.sql"
+            "./migrations/0008-grant_all_on_sequences_to_neon_superuser_with_grant_option.sql"
+        ),
+        include_str!("./migrations/0009-revoke_replication_for_previously_allowed_roles.sql"),
+        include_str!(
+            "./migrations/0010-grant_snapshot_synchronization_funcs_to_neon_superuser.sql"
         ),
-        include_str!("./migrations/0008-revoke_replication_for_previously_allowed_roles.sql"),
     ];
 
-    let mut func = || {
-        let query = "CREATE SCHEMA IF NOT EXISTS neon_migration";
-        client.simple_query(query)?;
-
-        let query = "CREATE TABLE IF NOT EXISTS neon_migration.migration_id (key INT NOT NULL PRIMARY KEY, id bigint NOT NULL DEFAULT 0)";
-        client.simple_query(query)?;
-
-        let query = "INSERT INTO neon_migration.migration_id VALUES (0, 0) ON CONFLICT DO NOTHING";
-        client.simple_query(query)?;
-
-        let query = "ALTER SCHEMA neon_migration OWNER TO cloud_admin";
-        client.simple_query(query)?;
-
-        let query = "REVOKE ALL ON SCHEMA neon_migration FROM PUBLIC";
-        client.simple_query(query)?;
-        Ok::<_, anyhow::Error>(())
-    };
-    func().context("handle_migrations prepare")?;
-
-    let query = "SELECT id FROM neon_migration.migration_id";
-    let row = client
-        .query_one(query, &[])
-        .context("handle_migrations get migration_id")?;
-    let mut current_migration: usize = row.get::<&str, i64>("id") as usize;
-    let starting_migration_id = current_migration;
-
-    let query = "BEGIN";
-    client
-        .simple_query(query)
-        .context("handle_migrations begin")?;
-
-    while current_migration < migrations.len() {
-        let migration = &migrations[current_migration];
-        if migration.starts_with("-- SKIP") {
-            info!("Skipping migration id={}", current_migration);
-        } else {
-            info!(
-                "Running migration id={}:\n{}\n",
-                current_migration, migration
-            );
-            client.simple_query(migration).with_context(|| {
-                format!("handle_migrations current_migration={}", current_migration)
-            })?;
-        }
-        current_migration += 1;
-    }
-    let setval = format!(
-        "UPDATE neon_migration.migration_id SET id={}",
-        migrations.len()
-    );
-    client
-        .simple_query(&setval)
-        .context("handle_migrations update id")?;
-
-    let query = "COMMIT";
-    client
-        .simple_query(query)
-        .context("handle_migrations commit")?;
-
-    info!(
-        "Ran {} migrations",
-        (migrations.len() - starting_migration_id)
-    );
+    MigrationRunner::new(client, &migrations).run_migrations()?;
 
     Ok(())
 }

control_plane/Cargo.toml

@@ -40,6 +40,7 @@ safekeeper_api.workspace = true
 postgres_connection.workspace = true
 storage_broker.workspace = true
 utils.workspace = true
+whoami.workspace = true
 
 compute_api.workspace = true
 workspace_hack.workspace = true

control_plane/src/background_process.rs

@@ -289,7 +289,7 @@ fn fill_remote_storage_secrets_vars(mut cmd: &mut Command) -> &mut Command {
 
 fn fill_env_vars_prefixed_neon(mut cmd: &mut Command) -> &mut Command {
     for (var, val) in std::env::vars() {
-        if var.starts_with("NEON_PAGESERVER_") {
+        if var.starts_with("NEON_") {
             cmd = cmd.env(var, val);
         }
     }

control_plane/src/bin/neon_local.rs

@@ -21,7 +21,9 @@ use pageserver_api::config::{
     DEFAULT_HTTP_LISTEN_PORT as DEFAULT_PAGESERVER_HTTP_PORT,
     DEFAULT_PG_LISTEN_PORT as DEFAULT_PAGESERVER_PG_PORT,
 };
-use pageserver_api::controller_api::{PlacementPolicy, TenantCreateRequest};
+use pageserver_api::controller_api::{
+    NodeAvailabilityWrapper, PlacementPolicy, TenantCreateRequest,
+};
 use pageserver_api::models::{ShardParameters, TimelineCreateRequest, TimelineInfo};
 use pageserver_api::shard::{ShardCount, ShardStripeSize, TenantShardId};
 use postgres_backend::AuthType;
@@ -1250,9 +1252,70 @@ async fn handle_start_all(
             exit(1);
         }
     }
+
+    neon_start_status_check(env, retry_timeout).await?;
+
     Ok(())
 }
 
+async fn neon_start_status_check(
+    env: &local_env::LocalEnv,
+    retry_timeout: &Duration,
+) -> anyhow::Result<()> {
+    const RETRY_INTERVAL: Duration = Duration::from_millis(100);
+    const NOTICE_AFTER_RETRIES: Duration = Duration::from_secs(5);
+
+    if env.control_plane_api.is_none() {
+        return Ok(());
+    }
+
+    let storcon = StorageController::from_env(env);
+
+    let retries = retry_timeout.as_millis() / RETRY_INTERVAL.as_millis();
+    let notice_after_retries = retry_timeout.as_millis() / NOTICE_AFTER_RETRIES.as_millis();
+
+    println!("\nRunning neon status check");
+
+    for retry in 0..retries {
+        if retry == notice_after_retries {
+            println!("\nNeon status check has not passed yet, continuing to wait")
+        }
+
+        let mut passed = true;
+        let mut nodes = storcon.node_list().await?;
+        let mut pageservers = env.pageservers.clone();
+
+        if nodes.len() != pageservers.len() {
+            continue;
+        }
+
+        nodes.sort_by_key(|ps| ps.id);
+        pageservers.sort_by_key(|ps| ps.id);
+
+        for (idx, pageserver) in pageservers.iter().enumerate() {
+            let node = &nodes[idx];
+            if node.id != pageserver.id {
+                passed = false;
+                break;
+            }
+
+            if !matches!(node.availability, NodeAvailabilityWrapper::Active) {
+                passed = false;
+                break;
+            }
+        }
+
+        if passed {
+            println!("\nNeon started and passed status check");
+            return Ok(());
+        }
+
+        tokio::time::sleep(RETRY_INTERVAL).await;
+    }
+
+    anyhow::bail!("\nNeon passed status check")
+}
+
 async 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");

control_plane/src/broker.rs

@@ -1,9 +1,9 @@
 //! Code to manage the storage broker
 //!
-//! In the local test environment, the data for each safekeeper is stored in
+//! In the local test environment, the storage broker stores its data directly in
 //!
 //! ```text
-//!   .neon/safekeepers/<safekeeper id>
+//!   .neon
 //! ```
 use std::time::Duration;
 

control_plane/src/local_env.rs

@@ -151,23 +151,31 @@ pub struct NeonBroker {
 pub struct NeonStorageControllerConf {
     /// Heartbeat timeout before marking a node offline
     #[serde(with = "humantime_serde")]
-    pub max_unavailable: Duration,
+    pub max_offline: Duration,
+
+    #[serde(with = "humantime_serde")]
+    pub max_warming_up: Duration,
 
     /// Threshold for auto-splitting a tenant into shards
     pub split_threshold: Option<u64>,
+
+    pub max_secondary_lag_bytes: Option<u64>,
 }
 
 impl NeonStorageControllerConf {
     // Use a shorter pageserver unavailability interval than the default to speed up tests.
-    const DEFAULT_MAX_UNAVAILABLE_INTERVAL: std::time::Duration =
-        std::time::Duration::from_secs(10);
+    const DEFAULT_MAX_OFFLINE_INTERVAL: std::time::Duration = std::time::Duration::from_secs(10);
+
+    const DEFAULT_MAX_WARMING_UP_INTERVAL: std::time::Duration = std::time::Duration::from_secs(30);
 }
 
 impl Default for NeonStorageControllerConf {
     fn default() -> Self {
         Self {
-            max_unavailable: Self::DEFAULT_MAX_UNAVAILABLE_INTERVAL,
+            max_offline: Self::DEFAULT_MAX_OFFLINE_INTERVAL,
+            max_warming_up: Self::DEFAULT_MAX_WARMING_UP_INTERVAL,
             split_threshold: None,
+            max_secondary_lag_bytes: None,
         }
     }
 }
@@ -325,11 +333,16 @@ impl LocalEnv {
         }
     }
 
-    pub fn pg_bin_dir(&self, pg_version: u32) -> anyhow::Result<PathBuf> {
-        Ok(self.pg_distrib_dir(pg_version)?.join("bin"))
+    pub fn pg_dir(&self, pg_version: u32, dir_name: &str) -> anyhow::Result<PathBuf> {
+        Ok(self.pg_distrib_dir(pg_version)?.join(dir_name))
     }
+
+    pub fn pg_bin_dir(&self, pg_version: u32) -> anyhow::Result<PathBuf> {
+        self.pg_dir(pg_version, "bin")
+    }
+
     pub fn pg_lib_dir(&self, pg_version: u32) -> anyhow::Result<PathBuf> {
-        Ok(self.pg_distrib_dir(pg_version)?.join("lib"))
+        self.pg_dir(pg_version, "lib")
     }
 
     pub fn pageserver_bin(&self) -> PathBuf {
@@ -504,7 +517,6 @@ impl LocalEnv {
                 #[derive(serde::Serialize, serde::Deserialize)]
                 // (allow unknown fields, unlike PageServerConf)
                 struct PageserverConfigTomlSubset {
-                    id: NodeId,
                     listen_pg_addr: String,
                     listen_http_addr: String,
                     pg_auth_type: AuthType,
@@ -516,18 +528,30 @@ impl LocalEnv {
                         .with_context(|| format!("read {:?}", config_toml_path))?,
                 )
                 .context("parse pageserver.toml")?;
+                let identity_toml_path = dentry.path().join("identity.toml");
+                #[derive(serde::Serialize, serde::Deserialize)]
+                struct IdentityTomlSubset {
+                    id: NodeId,
+                }
+                let identity_toml: IdentityTomlSubset = toml_edit::de::from_str(
+                    &std::fs::read_to_string(&identity_toml_path)
+                        .with_context(|| format!("read {:?}", identity_toml_path))?,
+                )
+                .context("parse identity.toml")?;
                 let PageserverConfigTomlSubset {
-                    id: config_toml_id,
                     listen_pg_addr,
                     listen_http_addr,
                     pg_auth_type,
                     http_auth_type,
                 } = config_toml;
+                let IdentityTomlSubset {
+                    id: identity_toml_id,
+                } = identity_toml;
                 let conf = PageServerConf {
                     id: {
                         anyhow::ensure!(
-                            config_toml_id == id,
-                            "id mismatch: config_toml.id={config_toml_id} id={id}",
+                            identity_toml_id == id,
+                            "id mismatch: identity.toml:id={identity_toml_id} pageserver_(.*) id={id}",
                         );
                         id
                     },

control_plane/src/pageserver.rs

@@ -1,8 +1,10 @@
 //! Code to manage pageservers
 //!
-//! In the local test environment, the pageserver stores its data directly in
+//! In the local test environment, the data for each pageserver is stored in
 //!
-//!   .neon/
+//! ```text
+//!   .neon/pageserver_<pageserver_id>
+//! ```
 //!
 use std::collections::HashMap;
 
@@ -15,7 +17,6 @@ use std::time::Duration;
 
 use anyhow::{bail, Context};
 use camino::Utf8PathBuf;
-use futures::SinkExt;
 use pageserver_api::models::{
     self, AuxFilePolicy, LocationConfig, TenantHistorySize, TenantInfo, TimelineInfo,
 };
@@ -24,6 +25,7 @@ use pageserver_client::mgmt_api;
 use postgres_backend::AuthType;
 use postgres_connection::{parse_host_port, PgConnectionConfig};
 use utils::auth::{Claims, Scope};
+use utils::id::NodeId;
 use utils::{
     id::{TenantId, TimelineId},
     lsn::Lsn,
@@ -73,6 +75,10 @@ impl PageServerNode {
         }
     }
 
+    fn pageserver_make_identity_toml(&self, node_id: NodeId) -> toml_edit::Document {
+        toml_edit::Document::from_str(&format!("id={node_id}")).unwrap()
+    }
+
     fn pageserver_init_make_toml(
         &self,
         conf: NeonLocalInitPageserverConf,
@@ -121,10 +127,13 @@ impl PageServerNode {
         }
 
         // Apply the user-provided overrides
-        overrides.push(
-            toml_edit::ser::to_string_pretty(&conf)
-                .expect("we deserialized this from toml earlier"),
-        );
+        overrides.push({
+            let mut doc =
+                toml_edit::ser::to_document(&conf).expect("we deserialized this from toml earlier");
+            // `id` is written out to `identity.toml` instead of `pageserver.toml`
+            doc.remove("id").expect("it's part of the struct");
+            doc.to_string()
+        });
 
         // Turn `overrides` into a toml document.
         // TODO: above code is legacy code, it should be refactored to use toml_edit directly.
@@ -185,6 +194,19 @@ impl PageServerNode {
             .write_all(config.to_string().as_bytes())
             .context("write pageserver toml")?;
         drop(config_file);
+
+        let identity_file_path = datadir.join("identity.toml");
+        let mut identity_file = std::fs::OpenOptions::new()
+            .create_new(true)
+            .write(true)
+            .open(identity_file_path)
+            .with_context(|| format!("open identity toml for write: {config_file_path:?}"))?;
+        let identity_toml = self.pageserver_make_identity_toml(node_id);
+        identity_file
+            .write_all(identity_toml.to_string().as_bytes())
+            .context("write identity toml")?;
+        drop(identity_toml);
+
         // TODO: invoke a TBD config-check command to validate that pageserver will start with the written config
 
         // Write metadata file, used by pageserver on startup to register itself with
@@ -350,11 +372,6 @@ impl PageServerNode {
                 .map(|x| x.parse::<NonZeroU64>())
                 .transpose()
                 .context("Failed to parse 'max_lsn_wal_lag' as non zero integer")?,
-            trace_read_requests: settings
-                .remove("trace_read_requests")
-                .map(|x| x.parse::<bool>())
-                .transpose()
-                .context("Failed to parse 'trace_read_requests' as bool")?,
             eviction_policy: settings
                 .remove("eviction_policy")
                 .map(serde_json::from_str)
@@ -455,11 +472,6 @@ impl PageServerNode {
                     .map(|x| x.parse::<NonZeroU64>())
                     .transpose()
                     .context("Failed to parse 'max_lsn_wal_lag' as non zero integer")?,
-                trace_read_requests: settings
-                    .remove("trace_read_requests")
-                    .map(|x| x.parse::<bool>())
-                    .transpose()
-                    .context("Failed to parse 'trace_read_requests' as bool")?,
                 eviction_policy: settings
                     .remove("eviction_policy")
                     .map(serde_json::from_str)
@@ -566,60 +578,39 @@ impl PageServerNode {
         pg_wal: Option<(Lsn, PathBuf)>,
         pg_version: u32,
     ) -> anyhow::Result<()> {
-        let (client, conn) = self.page_server_psql_client().await?;
-        // The connection object performs the actual communication with the database,
-        // so spawn it off to run on its own.
-        tokio::spawn(async move {
-            if let Err(e) = conn.await {
-                eprintln!("connection error: {}", e);
-            }
-        });
-        let client = std::pin::pin!(client);
-
         // Init base reader
         let (start_lsn, base_tarfile_path) = base;
         let base_tarfile = tokio::fs::File::open(base_tarfile_path).await?;
-        let base_tarfile = tokio_util::io::ReaderStream::new(base_tarfile);
+        let base_tarfile =
+            mgmt_api::ReqwestBody::wrap_stream(tokio_util::io::ReaderStream::new(base_tarfile));
 
         // Init wal reader if necessary
         let (end_lsn, wal_reader) = if let Some((end_lsn, wal_tarfile_path)) = pg_wal {
             let wal_tarfile = tokio::fs::File::open(wal_tarfile_path).await?;
-            let wal_reader = tokio_util::io::ReaderStream::new(wal_tarfile);
+            let wal_reader =
+                mgmt_api::ReqwestBody::wrap_stream(tokio_util::io::ReaderStream::new(wal_tarfile));
             (end_lsn, Some(wal_reader))
         } else {
             (start_lsn, None)
         };
 
-        let copy_in = |reader, cmd| {
-            let client = &client;
-            async move {
-                let writer = client.copy_in(&cmd).await?;
-                let writer = std::pin::pin!(writer);
-                let mut writer = writer.sink_map_err(|e| {
-                    std::io::Error::new(std::io::ErrorKind::Other, format!("{e}"))
-                });
-                let mut reader = std::pin::pin!(reader);
-                writer.send_all(&mut reader).await?;
-                writer.into_inner().finish().await?;
-                anyhow::Ok(())
-            }
-        };
-
         // Import base
-        copy_in(
-            base_tarfile,
-            format!(
-                "import basebackup {tenant_id} {timeline_id} {start_lsn} {end_lsn} {pg_version}"
-            ),
-        )
-        .await?;
-        // Import wal if necessary
-        if let Some(wal_reader) = wal_reader {
-            copy_in(
-                wal_reader,
-                format!("import wal {tenant_id} {timeline_id} {start_lsn} {end_lsn}"),
+        self.http_client
+            .import_basebackup(
+                tenant_id,
+                timeline_id,
+                start_lsn,
+                end_lsn,
+                pg_version,
+                base_tarfile,
             )
             .await?;
+
+        // Import wal if necessary
+        if let Some(wal_reader) = wal_reader {
+            self.http_client
+                .import_wal(tenant_id, timeline_id, start_lsn, end_lsn, wal_reader)
+                .await?;
         }
 
         Ok(())

control_plane/src/storage_controller.rs

@@ -5,8 +5,9 @@ use crate::{
 use camino::{Utf8Path, Utf8PathBuf};
 use pageserver_api::{
     controller_api::{
-        NodeConfigureRequest, NodeRegisterRequest, TenantCreateRequest, TenantCreateResponse,
-        TenantLocateResponse, TenantShardMigrateRequest, TenantShardMigrateResponse,
+        NodeConfigureRequest, NodeDescribeResponse, NodeRegisterRequest, TenantCreateRequest,
+        TenantCreateResponse, TenantLocateResponse, TenantShardMigrateRequest,
+        TenantShardMigrateResponse,
     },
     models::{
         TenantShardSplitRequest, TenantShardSplitResponse, TimelineCreateRequest, TimelineInfo,
@@ -29,7 +30,6 @@ use utils::{
 pub struct StorageController {
     env: LocalEnv,
     listen: String,
-    path: Utf8PathBuf,
     private_key: Option<Vec<u8>>,
     public_key: Option<String>,
     postgres_port: u16,
@@ -41,6 +41,8 @@ const COMMAND: &str = "storage_controller";
 
 const STORAGE_CONTROLLER_POSTGRES_VERSION: u32 = 16;
 
+const DB_NAME: &str = "storage_controller";
+
 #[derive(Serialize, Deserialize)]
 pub struct AttachHookRequest {
     pub tenant_shard_id: TenantShardId,
@@ -65,10 +67,6 @@ pub struct InspectResponse {
 
 impl StorageController {
     pub fn from_env(env: &LocalEnv) -> Self {
-        let path = Utf8PathBuf::from_path_buf(env.base_data_dir.clone())
-            .unwrap()
-            .join("attachments.json");
-
         // Makes no sense to construct this if pageservers aren't going to use it: assume
         // pageservers have control plane API set
         let listen_url = env.control_plane_api.clone().unwrap();
@@ -128,7 +126,6 @@ impl StorageController {
 
         Self {
             env: env.clone(),
-            path,
             listen,
             private_key,
             public_key,
@@ -155,16 +152,16 @@ impl StorageController {
         .expect("non-Unicode path")
     }
 
-    /// Find the directory containing postgres binaries, such as `initdb` and `pg_ctl`
+    /// Find the directory containing postgres subdirectories, such `bin` and `lib`
     ///
     /// This usually uses STORAGE_CONTROLLER_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> {
+    async fn get_pg_dir(&self, dir_name: &str) -> anyhow::Result<Utf8PathBuf> {
         let prefer_versions = [STORAGE_CONTROLLER_POSTGRES_VERSION, 15, 14];
 
         for v in prefer_versions {
-            let path = Utf8PathBuf::from_path_buf(self.env.pg_bin_dir(v)?).unwrap();
+            let path = Utf8PathBuf::from_path_buf(self.env.pg_dir(v, dir_name)?).unwrap();
             if tokio::fs::try_exists(&path).await? {
                 return Ok(path);
             }
@@ -172,11 +169,20 @@ impl StorageController {
 
         // Fall through
         anyhow::bail!(
-            "Postgres binaries not found in {}",
-            self.env.pg_distrib_dir.display()
+            "Postgres directory '{}' not found in {}",
+            dir_name,
+            self.env.pg_distrib_dir.display(),
         );
     }
 
+    pub async fn get_pg_bin_dir(&self) -> anyhow::Result<Utf8PathBuf> {
+        self.get_pg_dir("bin").await
+    }
+
+    pub async fn get_pg_lib_dir(&self) -> anyhow::Result<Utf8PathBuf> {
+        self.get_pg_dir("lib").await
+    }
+
     /// 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");
@@ -194,7 +200,6 @@ impl StorageController {
     ///
     /// Returns the database url
     pub async fn setup_database(&self) -> anyhow::Result<String> {
-        const DB_NAME: &str = "storage_controller";
         let database_url = format!("postgresql://localhost:{}/{DB_NAME}", self.postgres_port);
 
         let pg_bin_dir = self.get_pg_bin_dir().await?;
@@ -223,18 +228,47 @@ impl StorageController {
         Ok(database_url)
     }
 
+    pub async fn connect_to_database(
+        &self,
+    ) -> anyhow::Result<(
+        tokio_postgres::Client,
+        tokio_postgres::Connection<tokio_postgres::Socket, tokio_postgres::tls::NoTlsStream>,
+    )> {
+        tokio_postgres::Config::new()
+            .host("localhost")
+            .port(self.postgres_port)
+            // The user is the ambient operating system user name.
+            // That is an impurity which we want to fix in => TODO https://github.com/neondatabase/neon/issues/8400
+            //
+            // Until we get there, use the ambient operating system user name.
+            // Recent tokio-postgres versions default to this if the user isn't specified.
+            // But tokio-postgres fork doesn't have this upstream commit:
+            // https://github.com/sfackler/rust-postgres/commit/cb609be758f3fb5af537f04b584a2ee0cebd5e79
+            // => we should rebase our fork => TODO https://github.com/neondatabase/neon/issues/8399
+            .user(&whoami::username())
+            .dbname(DB_NAME)
+            .connect(tokio_postgres::NoTls)
+            .await
+            .map_err(anyhow::Error::new)
+    }
+
     pub async fn start(&self, retry_timeout: &Duration) -> anyhow::Result<()> {
         // Start a vanilla Postgres process used by the storage controller for persistence.
         let pg_data_path = Utf8PathBuf::from_path_buf(self.env.base_data_dir.clone())
             .unwrap()
             .join("storage_controller_db");
         let pg_bin_dir = self.get_pg_bin_dir().await?;
+        let pg_lib_dir = self.get_pg_lib_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)
+                .envs(vec![
+                    ("LD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
+                    ("DYLD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
+                ])
                 .args(["-D", pg_data_path.as_ref()])
                 .spawn()
                 .expect("Failed to spawn initdb");
@@ -242,18 +276,21 @@ impl StorageController {
             if !status.success() {
                 anyhow::bail!("initdb failed with status {status}");
             }
-
-            // Write a minimal config file:
-            // - Specify the port, since this is chosen dynamically
-            // - Switch off fsync, since we're running on lightweight test environments and when e.g. scale testing
-            //   the storage controller we don't want a slow local disk to interfere with that.
-            tokio::fs::write(
-                &pg_data_path.join("postgresql.conf"),
-                format!("port = {}\nfsync=off\n", self.postgres_port),
-            )
-            .await?;
         };
 
+        // Write a minimal config file:
+        // - Specify the port, since this is chosen dynamically
+        // - Switch off fsync, since we're running on lightweight test environments and when e.g. scale testing
+        //   the storage controller we don't want a slow local disk to interfere with that.
+        //
+        // NB: it's important that we rewrite this file on each start command so we propagate changes
+        // from `LocalEnv`'s config file (`.neon/config`).
+        tokio::fs::write(
+            &pg_data_path.join("postgresql.conf"),
+            format!("port = {}\nfsync=off\n", self.postgres_port),
+        )
+        .await?;
+
         println!("Starting storage controller database...");
         let db_start_args = [
             "-w",
@@ -269,7 +306,10 @@ impl StorageController {
             &self.env.base_data_dir,
             pg_bin_dir.join("pg_ctl").as_std_path(),
             db_start_args,
-            [],
+            vec![
+                ("LD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
+                ("DYLD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
+            ],
             background_process::InitialPidFile::Create(self.postgres_pid_file()),
             retry_timeout,
             || self.pg_isready(&pg_bin_dir),
@@ -279,16 +319,45 @@ impl StorageController {
         // Run migrations on every startup, in case something changed.
         let database_url = self.setup_database().await?;
 
+        // We support running a startup SQL script to fiddle with the database before we launch storcon.
+        // This is used by the test suite.
+        let startup_script_path = self
+            .env
+            .base_data_dir
+            .join("storage_controller_db.startup.sql");
+        let startup_script = match tokio::fs::read_to_string(&startup_script_path).await {
+            Ok(script) => {
+                tokio::fs::remove_file(startup_script_path).await?;
+                script
+            }
+            Err(e) => {
+                if e.kind() == std::io::ErrorKind::NotFound {
+                    // always run some startup script so that this code path doesn't bit rot
+                    "BEGIN; COMMIT;".to_string()
+                } else {
+                    anyhow::bail!("Failed to read startup script: {e}")
+                }
+            }
+        };
+        let (mut client, conn) = self.connect_to_database().await?;
+        let conn = tokio::spawn(conn);
+        let tx = client.build_transaction();
+        let tx = tx.start().await?;
+        tx.batch_execute(&startup_script).await?;
+        tx.commit().await?;
+        drop(client);
+        conn.await??;
+
         let mut args = vec![
             "-l",
             &self.listen,
-            "-p",
-            self.path.as_ref(),
             "--dev",
             "--database-url",
             &database_url,
-            "--max-unavailable-interval",
-            &humantime::Duration::from(self.config.max_unavailable).to_string(),
+            "--max-offline-interval",
+            &humantime::Duration::from(self.config.max_offline).to_string(),
+            "--max-warming-up-interval",
+            &humantime::Duration::from(self.config.max_warming_up).to_string(),
         ]
         .into_iter()
         .map(|s| s.to_string())
@@ -314,6 +383,10 @@ impl StorageController {
             args.push(format!("--split-threshold={split_threshold}"))
         }
 
+        if let Some(lag) = self.config.max_secondary_lag_bytes.as_ref() {
+            args.push(format!("--max-secondary-lag-bytes={lag}"))
+        }
+
         args.push(format!(
             "--neon-local-repo-dir={}",
             self.env.base_data_dir.display()
@@ -324,7 +397,10 @@ impl StorageController {
             &self.env.base_data_dir,
             &self.env.storage_controller_bin(),
             args,
-            [],
+            vec![
+                ("LD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
+                ("DYLD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
+            ],
             background_process::InitialPidFile::Create(self.pid_file()),
             retry_timeout,
             || async {
@@ -556,6 +632,15 @@ impl StorageController {
         .await
     }
 
+    pub async fn node_list(&self) -> anyhow::Result<Vec<NodeDescribeResponse>> {
+        self.dispatch::<(), Vec<NodeDescribeResponse>>(
+            Method::GET,
+            "control/v1/node".to_string(),
+            None,
+        )
+        .await
+    }
+
     #[instrument(skip(self))]
     pub async fn ready(&self) -> anyhow::Result<()> {
         self.dispatch::<(), ()>(Method::GET, "ready".to_string(), None)

control_plane/storcon_cli/Cargo.toml

@@ -17,6 +17,7 @@ pageserver_client.workspace = true
 reqwest.workspace = true
 serde.workspace = true
 serde_json = { workspace = true, features = ["raw_value"] }
+storage_controller_client.workspace = true
 thiserror.workspace = true
 tokio.workspace = true
 tracing.workspace = true

control_plane/storcon_cli/src/main.rs

@@ -14,15 +14,15 @@ use pageserver_api::{
     },
     shard::{ShardStripeSize, TenantShardId},
 };
-use pageserver_client::mgmt_api::{self, ResponseErrorMessageExt};
+use pageserver_client::mgmt_api::{self};
 use reqwest::{Method, StatusCode, Url};
-use serde::{de::DeserializeOwned, Serialize};
 use utils::id::{NodeId, TenantId};
 
 use pageserver_api::controller_api::{
     NodeConfigureRequest, NodeRegisterRequest, NodeSchedulingPolicy, PlacementPolicy,
     TenantShardMigrateRequest, TenantShardMigrateResponse,
 };
+use storage_controller_client::control_api::Client;
 
 #[derive(Subcommand, Debug)]
 enum Command {
@@ -56,6 +56,10 @@ enum Command {
         #[arg(long)]
         scheduling: Option<NodeSchedulingPolicy>,
     },
+    NodeDelete {
+        #[arg(long)]
+        node_id: NodeId,
+    },
     /// Modify a tenant's policies in the storage controller
     TenantPolicy {
         #[arg(long)]
@@ -245,64 +249,6 @@ impl FromStr for NodeAvailabilityArg {
     }
 }
 
-struct Client {
-    base_url: Url,
-    jwt_token: Option<String>,
-    client: reqwest::Client,
-}
-
-impl Client {
-    fn new(base_url: Url, jwt_token: Option<String>) -> Self {
-        Self {
-            base_url,
-            jwt_token,
-            client: reqwest::ClientBuilder::new()
-                .build()
-                .expect("Failed to construct http client"),
-        }
-    }
-
-    /// Simple HTTP request wrapper for calling into storage controller
-    async fn dispatch<RQ, RS>(
-        &self,
-        method: Method,
-        path: String,
-        body: Option<RQ>,
-    ) -> mgmt_api::Result<RS>
-    where
-        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 url = Url::from_str(&format!(
-            "http://{}:{}/{path}",
-            self.base_url.host_str().unwrap(),
-            self.base_url.port().unwrap()
-        ))
-        .unwrap();
-
-        let mut builder = self.client.request(method, url);
-        if let Some(body) = body {
-            builder = builder.json(&body)
-        }
-        if let Some(jwt_token) = &self.jwt_token {
-            builder = builder.header(
-                reqwest::header::AUTHORIZATION,
-                format!("Bearer {jwt_token}"),
-            );
-        }
-
-        let response = builder.send().await.map_err(mgmt_api::Error::ReceiveBody)?;
-        let response = response.error_from_body().await?;
-
-        response
-            .json()
-            .await
-            .map_err(pageserver_client::mgmt_api::Error::ReceiveBody)
-    }
-}
-
 #[tokio::main]
 async fn main() -> anyhow::Result<()> {
     let cli = Cli::parse();
@@ -337,7 +283,7 @@ async fn main() -> anyhow::Result<()> {
         }
         Command::TenantCreate { tenant_id } => {
             storcon_client
-                .dispatch(
+                .dispatch::<_, ()>(
                     Method::POST,
                     "v1/tenant".to_string(),
                     Some(TenantCreateRequest {
@@ -357,13 +303,16 @@ async fn main() -> anyhow::Result<()> {
             tracing::info!("Delete status: {}", status);
         }
         Command::Nodes {} => {
-            let resp = storcon_client
+            let mut resp = storcon_client
                 .dispatch::<(), Vec<NodeDescribeResponse>>(
                     Method::GET,
                     "control/v1/node".to_string(),
                     None,
                 )
                 .await?;
+
+            resp.sort_by(|a, b| a.listen_http_addr.cmp(&b.listen_http_addr));
+
             let mut table = comfy_table::Table::new();
             table.set_header(["Id", "Hostname", "Scheduling", "Availability"]);
             for node in resp {
@@ -395,13 +344,16 @@ async fn main() -> anyhow::Result<()> {
                 .await?;
         }
         Command::Tenants {} => {
-            let resp = storcon_client
+            let mut resp = storcon_client
                 .dispatch::<(), Vec<TenantDescribeResponse>>(
                     Method::GET,
                     "control/v1/tenant".to_string(),
                     None,
                 )
                 .await?;
+
+            resp.sort_by(|a, b| a.tenant_id.cmp(&b.tenant_id));
+
             let mut table = comfy_table::Table::new();
             table.set_header([
                 "TenantId",
@@ -650,6 +602,11 @@ async fn main() -> anyhow::Result<()> {
                 .dispatch::<(), ()>(Method::POST, format!("debug/v1/node/{node_id}/drop"), None)
                 .await?;
         }
+        Command::NodeDelete { node_id } => {
+            storcon_client
+                .dispatch::<(), ()>(Method::DELETE, format!("control/v1/node/{node_id}"), None)
+                .await?;
+        }
         Command::TenantSetTimeBasedEviction {
             tenant_id,
             period,

deny.toml

@@ -4,6 +4,7 @@
 # to your expectations and requirements.
 
 # Root options
+[graph]
 targets = [
     { triple = "x86_64-unknown-linux-gnu" },
     { triple = "aarch64-unknown-linux-gnu" },
@@ -12,6 +13,7 @@ targets = [
 ]
 all-features = false
 no-default-features = false
+[output]
 feature-depth = 1
 
 # This section is considered when running `cargo deny check advisories`
@@ -19,17 +21,16 @@ feature-depth = 1
 # https://embarkstudios.github.io/cargo-deny/checks/advisories/cfg.html
 [advisories]
 db-urls = ["https://github.com/rustsec/advisory-db"]
-vulnerability = "deny"
-unmaintained = "warn"
 yanked = "warn"
-notice = "warn"
-ignore = []
+
+[[advisories.ignore]]
+id = "RUSTSEC-2023-0071"
+reason = "the marvin attack only affects private key decryption, not public key signature verification"
 
 # This section is considered when running `cargo deny check licenses`
 # More documentation for the licenses section can be found here:
 # https://embarkstudios.github.io/cargo-deny/checks/licenses/cfg.html
 [licenses]
-unlicensed = "deny"
 allow = [
     "Apache-2.0",
     "Artistic-2.0",
@@ -42,10 +43,6 @@ allow = [
     "OpenSSL",
     "Unicode-DFS-2016",
 ]
-deny = []
-copyleft = "warn"
-allow-osi-fsf-free = "neither"
-default = "deny"
 confidence-threshold = 0.8
 exceptions = [
     # Zlib license has some restrictions if we decide to change sth

docker-compose/compute_wrapper/shell/compute.sh

@@ -33,7 +33,7 @@ echo $result | jq .
 
 generate_id timeline_id
 PARAMS=(
-     -sb 
+     -sbf
      -X POST
      -H "Content-Type: application/json"
      -d "{\"new_timeline_id\": \"${timeline_id}\", \"pg_version\": ${PG_VERSION}}"

docker-compose/docker-compose.yml

@@ -31,25 +31,14 @@ services:
     restart: always
     image: ${REPOSITORY:-neondatabase}/neon:${TAG:-latest}
     environment:
-      - BROKER_ENDPOINT='http://storage_broker:50051'
       - AWS_ACCESS_KEY_ID=minio
       - AWS_SECRET_ACCESS_KEY=password
       #- RUST_BACKTRACE=1
     ports:
        #- 6400:6400  # pg protocol handler
        - 9898:9898 # http endpoints
-    entrypoint:
-      - "/bin/sh"
-      - "-c"
-    command:
-      - "/usr/local/bin/pageserver -D /data/.neon/
-                                   -c \"broker_endpoint=$$BROKER_ENDPOINT\"
-                                   -c \"listen_pg_addr='0.0.0.0:6400'\"
-                                   -c \"listen_http_addr='0.0.0.0:9898'\"
-                                   -c \"remote_storage={endpoint='http://minio:9000',
-                                                        bucket_name='neon',
-                                                        bucket_region='eu-north-1',
-                                                        prefix_in_bucket='/pageserver/'}\""
+    volumes:
+      - ./pageserver_config:/data/.neon/
     depends_on:
       - storage_broker
       - minio_create_buckets

docker-compose/docker_compose_test.sh

@@ -78,7 +78,7 @@ for pg_version in 14 15 16; do
         docker cp $TMPDIR/data $COMPUTE_CONTAINER_NAME:/ext-src/pg_hint_plan-src/
         rm -rf $TMPDIR
         # We are running tests now
-        if docker exec -e SKIP=rum-src,timescaledb-src,rdkit-src,postgis-src,pgx_ulid-src,pgtap-src,pg_tiktoken-src,pg_jsonschema-src,pg_graphql-src,kq_imcx-src,wal2json_2_5-src \
+        if docker exec -e SKIP=timescaledb-src,rdkit-src,postgis-src,pgx_ulid-src,pgtap-src,pg_tiktoken-src,pg_jsonschema-src,pg_graphql-src,kq_imcx-src,wal2json_2_5-src \
             $TEST_CONTAINER_NAME /run-tests.sh | tee testout.txt
         then
             cleanup

docker-compose/pageserver_config/identity.toml

@@ -0,0 +1 @@
+id=1234

docker-compose/pageserver_config/pageserver.toml

@@ -0,0 +1,5 @@
+broker_endpoint='http://storage_broker:50051'
+pg_distrib_dir='/usr/local/'
+listen_pg_addr='0.0.0.0:6400'
+listen_http_addr='0.0.0.0:9898'
+remote_storage={ endpoint='http://minio:9000', bucket_name='neon', bucket_region='eu-north-1', prefix_in_bucket='/pageserver' }

docker-compose/run-tests.sh

@@ -1,15 +1,15 @@
 #!/bin/bash
 set -x
 
-cd /ext-src
+cd /ext-src || exit 2
 FAILED=
-LIST=$((echo ${SKIP} | sed 's/,/\n/g'; ls -d *-src) | sort | uniq -u)
+LIST=$( (echo "${SKIP//","/"\n"}"; ls -d -- *-src) | sort | uniq -u)
 for d in ${LIST}
 do
-       [ -d ${d} ] || continue
+       [ -d "${d}" ] || continue
     psql -c "select 1" >/dev/null || break
-       make -C ${d} installcheck || FAILED="${d} ${FAILED}"
+       USE_PGXS=1 make -C "${d}" installcheck || FAILED="${d} ${FAILED}"
 done
 [ -z "${FAILED}" ] && exit 0
-echo ${FAILED}
+echo "${FAILED}"
 exit 1

docs/SUMMARY.md

@@ -1,13 +1,18 @@
 # Summary
 
+# Looking for `neon.tech` docs?
+
+This page linkes to a selection of technical content about the open source code in this repository.
+
+Please visit https://neon.tech/docs for documentation about using the Neon service, which is based on the code
+in this repository.
+
+# Architecture
+
 [Introduction]()
 - [Separation of Compute and Storage](./separation-compute-storage.md)
 
-# Architecture
-
 - [Compute]()
-  - [WAL proposer]()
-  - [WAL Backpressure]()
   - [Postgres changes](./core_changes.md)
 
 - [Pageserver](./pageserver.md)
@@ -16,33 +21,15 @@
     - [WAL Redo](./pageserver-walredo.md)
     - [Page cache](./pageserver-pagecache.md)
     - [Storage](./pageserver-storage.md)
-        - [Datadir mapping]()
-        - [Layer files]()
-        - [Branching]()
-        - [Garbage collection]()
-    - [Cloud Storage]()
     - [Processing a GetPage request](./pageserver-processing-getpage.md)
     - [Processing WAL](./pageserver-processing-wal.md)
-	- [Management API]()
-	- [Tenant Rebalancing]()
 
 - [WAL Service](walservice.md)
   - [Consensus protocol](safekeeper-protocol.md)
-  - [Management API]()
-  - [Rebalancing]()
-
-- [Control Plane]()
-
-- [Proxy]()
 
 - [Source view](./sourcetree.md)
   - [docker.md](./docker.md) — Docker images and building pipeline.
   - [Error handling and logging](./error-handling.md)
-  - [Testing]()
-    - [Unit testing]()
-    - [Integration testing]()
-    - [Benchmarks]()
-
 
 - [Glossary](./glossary.md)
 
@@ -58,28 +45,6 @@
 
 # RFCs
 
-- [RFCs](./rfcs/README.md)
-
-- [002-storage](rfcs/002-storage.md)
-- [003-laptop-cli](rfcs/003-laptop-cli.md)
-- [004-durability](rfcs/004-durability.md)
-- [005-zenith_local](rfcs/005-zenith_local.md)
-- [006-laptop-cli-v2-CLI](rfcs/006-laptop-cli-v2-CLI.md)
-- [006-laptop-cli-v2-repository-structure](rfcs/006-laptop-cli-v2-repository-structure.md)
-- [007-serverless-on-laptop](rfcs/007-serverless-on-laptop.md)
-- [008-push-pull](rfcs/008-push-pull.md)
-- [009-snapshot-first-storage-cli](rfcs/009-snapshot-first-storage-cli.md)
-- [009-snapshot-first-storage](rfcs/009-snapshot-first-storage.md)
-- [009-snapshot-first-storage-pitr](rfcs/009-snapshot-first-storage-pitr.md)
-- [010-storage_details](rfcs/010-storage_details.md)
-- [011-retention-policy](rfcs/011-retention-policy.md)
-- [012-background-tasks](rfcs/012-background-tasks.md)
-- [013-term-history](rfcs/013-term-history.md)
-- [014-safekeepers-gossip](rfcs/014-safekeepers-gossip.md)
-- [014-storage-lsm](rfcs/014-storage-lsm.md)
-- [015-storage-messaging](rfcs/015-storage-messaging.md)
-- [016-connection-routing](rfcs/016-connection-routing.md)
-- [017-timeline-data-management](rfcs/017-timeline-data-management.md)
-- [018-storage-messaging-2](rfcs/018-storage-messaging-2.md)
-- [019-tenant-timeline-lifecycles](rfcs/019-tenant-timeline-lifecycles.md)
-- [cluster-size-limits](rfcs/cluster-size-limits.md)
+Major changes are documented in RFCS:
+- See [RFCs](./rfcs/README.md) for more information
+- view the RFCs at https://github.com/neondatabase/neon/tree/main/docs/rfcs

docs/rfcs/033-storage-controller-drain-and-fill.md

@@ -0,0 +1,345 @@
+# Graceful Restarts of Storage Controller Managed Clusters
+
+## Summary
+This RFC describes new storage controller APIs for draining and filling tenant shards from/on pageserver nodes.
+It also covers how these new APIs should be used by an orchestrator (e.g. Ansible) in order to implement
+graceful cluster restarts.
+
+## Motivation
+
+Pageserver restarts cause read availablity downtime for tenants.
+
+For example pageserver-3 @ us-east-1 was unavailable for a randomly
+picked tenant (which requested on-demand activation) for around 30 seconds
+during the restart at 2024-04-03 16:37 UTC.
+
+Note that lots of shutdowns on loaded pageservers do not finish within the
+[10 second systemd enforced timeout](https://github.com/neondatabase/aws/blob/0a5280b383e43c063d43cbf87fa026543f6d6ad4/.github/ansible/systemd/pageserver.service#L16). This means we are shutting down without flushing ephemeral layers
+and have to reingest data in order to serve requests after restarting, potentially making first request latencies worse.
+
+This problem is not yet very acutely felt in storage controller managed pageservers since
+tenant density is much lower there. However, we are planning on eventually migrating all
+pageservers to storage controller management, so it makes sense to solve the issue proactively.
+
+## Requirements
+
+- Pageserver re-deployments cause minimal downtime for tenants
+- The storage controller exposes HTTP API hooks for draining and filling tenant shards
+from a given pageserver. Said hooks can be used by an orchestrator proces or a human operator.
+- The storage controller exposes some HTTP API to cancel draining and filling background operations.
+- Failures to drain or fill the node should not be fatal. In such cases, cluster restarts should proceed
+as usual (with downtime).
+- Progress of draining/filling is visible through metrics
+
+## Non Goals
+
+- Integration with the control plane
+- Graceful restarts for large non-HA tenants.
+
+## Impacted Components
+
+- storage controller
+- deployment orchestrator (i.e. Ansible)
+- pageserver (indirectly)
+
+## Terminology
+
+** Draining ** is the process through which all tenant shards that can be migrated from a given pageserver
+are distributed across the rest of the cluster.
+
+** Filling ** is the symmetric opposite of draining. In this process tenant shards are migrated onto a given
+pageserver until the cluster reaches a resonable, quiescent distribution of tenant shards across pageservers.
+
+** Node scheduling policies ** act as constraints to the scheduler. For instance, when a
+node is set in the `Paused` policy, no further shards will be scheduled on it.
+
+** Node ** is a pageserver. Term is used interchangeably in this RFC.
+
+** Deployment orchestrator ** is a generic term for whatever drives our deployments.
+Currently, it's an Ansible playbook.
+
+## Background
+
+### Storage Controller Basics (skip if already familiar)
+
+Fundamentally, the storage controller is a reconciler which aims to move from the observed mapping between pageservers and tenant shards to an intended mapping. Pageserver nodes and tenant shards metadata is durably persisted in a database, but note that the mapping between the two entities is not durably persisted. Instead, this mapping (*observed state*) is constructed at startup by sending `GET location_config` requests to registered pageservers.
+
+An internal scheduler maps tenant shards to pageservers while respecting certain constraints. The result of scheduling is the *intent state*. When the intent state changes, a *reconciliation* will inform pageservers about the new assigment via `PUT location_config` requests and will notify the compute via the configured hook.
+
+### Background Optimizations
+
+The storage controller performs scheduling optimizations in the background. It will
+migrate attachments to warm secondaries and replace secondaries in order to balance
+the cluster out.
+
+### Reconciliations Concurrency Limiting
+
+There's a hard limit on the number of reconciles that the storage controller
+can have in flight at any given time. To get an idea of scales, the limit is
+128 at the time of writing.
+
+## Implementation
+
+Note: this section focuses on the core functionality of the graceful restart process.
+It doesn't neccesarily describe the most efficient approach. Optimizations are described
+separately in a later section.
+
+### Overall Flow
+
+This section describes how to implement graceful restarts from the perspective
+of Ansible, the deployment orchestrator. Pageservers are already restarted sequentially.
+The orchestrator shall implement the following epilogue and prologue steps for each
+pageserver restart:
+
+#### Prologue
+
+The orchestrator shall first fetch the pageserver node id from the control plane or
+the pageserver it aims to restart directly. Next, it issues an HTTP request
+to the storage controller in order to start the drain of said pageserver node.
+All error responses are retried with a short back-off. When a 202 (Accepted)
+HTTP code is returned, the drain has started. Now the orchestrator polls the
+node status endpoint exposed by the storage controller in order to await the
+end of the drain process. When the `policy` field of the node status response
+becomes `PauseForRestart`, the drain has completed and the orchestrator can
+proceed with restarting the pageserver.
+
+The prologue is subject to an overall timeout. It will have a value in the ballpark
+of minutes. As storage controller managed pageservers become more loaded this timeout
+will likely have to increase.
+
+#### Epilogue
+
+After restarting the pageserver, the orchestrator issues an HTTP request
+to the storage controller to kick off the filling process. This API call
+may be retried for all error codes with a short backoff. This also serves
+as a synchronization primitive as the fill will be refused if the pageserver
+has not yet re-attached to the storage controller. When a 202(Accepted) HTTP
+code is returned, the fill has started. Now the orchestrator polls the node
+status endpoint exposed by the storage controller in order to await the end of
+the filling process. When the `policy` field of the node status response becomes
+`Active`, the fill has completed and the orchestrator may proceed to the next pageserver.
+
+Again, the epilogue is subject to an overall timeout. We can start off with
+using the same timeout as for the prologue, but can also consider relying on
+the storage controller's background optimizations with a shorter timeout.
+
+In the case that the deployment orchestrator times out, it attempts to cancel
+the fill. This operation shall be retried with a short back-off. If it ultimately
+fails it will require manual intervention to set the nodes scheduling policy to
+`NodeSchedulingPolicy::Active`. Not doing that is not immediately problematic,
+but it constrains the scheduler as mentioned previously.
+
+### Node Scheduling Policy State Machine
+
+The state machine below encodes the behaviours discussed above and
+the various failover situations described in a later section.
+
+Assuming no failures and/or timeouts the flow should be:
+`Active -> Draining -> PauseForRestart -> Active -> Filling -> Active`
+
+```
+                          Operator requested drain
+               +-----------------------------------------+
+               |                                         |
+       +-------+-------+                         +-------v-------+
+       |               |                         |               |
+       |     Pause     |             +----------->    Draining   +----------+
+       |               |             |           |               |          |
+       +---------------+             |           +-------+-------+          |
+                                     |                   |                  |
+                                     |                   |                  |
+                      Drain requested|                   |                  |
+                                     |                   |Drain complete    | Drain failed
+                                     |                   |                  | Cancelled/PS reattach/Storcon restart
+                                     |                   |                  |
+                             +-------+-------+           |                  |
+                             |               |           |                  |
+               +-------------+    Active     <-----------+------------------+
+               |             |               |           |
+Fill requested |             +---^---^-------+           |
+               |                 |   |                   |
+               |                 |   |                   |
+               |                 |   |                   |
+               |   Fill completed|   |                   |
+               |                 |   |PS reattach        |
+               |                 |   |after restart      |
+       +-------v-------+         |   |           +-------v-------+
+       |               |         |   |           |               |
+       |    Filling    +---------+   +-----------+PauseForRestart|
+       |               |                         |               |
+       +---------------+                         +---------------+
+```
+
+### Draining/Filling APIs
+
+The storage controller API to trigger the draining of a given node is:
+`PUT /v1/control/node/:node_id/{drain,fill}`.
+
+The following HTTP non-success return codes are used.
+All of them are safely retriable from the perspective of the storage controller.
+- 404: Requested node was not found
+- 503: Requested node is known to the storage controller, but unavailable
+- 412: Drain precondition failed: there is no other node to drain to or the node's schedulling policy forbids draining
+- 409: A {drain, fill} is already in progress. Only one such background operation
+is allowed per node.
+
+When the drain is accepted and commenced a 202 HTTP code is returned.
+
+Drains and fills shall be cancellable by the deployment orchestrator or a
+human operator via: `DELETE /v1/control/node/:node_id/{drain,fill}`. A 200
+response is returned when the cancelation is successful. Errors are retriable.
+
+### Drain Process
+
+Before accpeting a drain request the following validations is applied:
+* Ensure that the node is known the storage controller
+* Ensure that the schedulling policy is `NodeSchedulingPolicy::Active` or `NodeSchedulingPolicy::Pause`
+* Ensure that another drain or fill is not already running on the node
+* Ensure that a drain is possible (i.e. check that there is at least one
+schedulable node to drain to)
+
+After accepting the drain, the scheduling policy of the node is set to
+`NodeSchedulingPolicy::Draining` and persisted in both memory and the database.
+This disallows the optimizer from adding or removing shards from the node which
+is desirable to avoid them racing.
+
+Next, a separate Tokio task is spawned to manage the draining. For each tenant
+shard attached to the node being drained, demote the node to a secondary and
+attempt to schedule the node away. Scheduling might fail due to unsatisfiable
+constraints, but that is fine. Draining is a best effort process since it might
+not always be possible to cut over all shards.
+
+Importantly, this task manages the concurrency of issued reconciles in order to
+avoid drowning out the target pageservers and to allow other important reconciles
+to proceed.
+
+Once the triggered reconciles have finished or timed out, set the node's scheduling
+policy to `NodeSchedulingPolicy::PauseForRestart` to signal the end of the drain.
+
+A note on non HA tenants: These tenants do not have secondaries, so by the description
+above, they would not be migrated. It makes sense to skip them (especially the large ones)
+since, depending on tenant size, this might be more disruptive than the restart since the
+pageserver we've moved to do will need to on-demand download the entire working set for the tenant.
+We can consider expanding to small non-HA tenants in the future.
+
+### Fill Process
+
+Before accpeting a fill request the following validations is applied:
+* Ensure that the node is known the storage controller
+* Ensure that the schedulling policy is `NodeSchedulingPolicy::Active`.
+This is the only acceptable policy for the fill starting state. When a node re-attaches,
+it set the scheduling policy to `NodeSchedulingPolicy::Active` if it was equal to
+`NodeSchedulingPolicy::PauseForRestart` or `NodeSchedulingPolicy::Draining` (possible end states for a node drain).
+* Ensure that another drain or fill is not already running on the node
+
+After accepting the drain, the scheduling policy of the node is set to
+`NodeSchedulingPolicy::Filling` and persisted in both memory and the database.
+This disallows the optimizer from adding or removing shards from the node which
+is desirable to avoid them racing.
+
+Next, a separate Tokio task is spawned to manage the draining. For each tenant
+shard where the filled node is a secondary, promote the secondary. This is done
+until we run out of shards or the counts of attached shards become balanced across
+the cluster.
+
+Like for draining, the concurrency of spawned reconciles is limited.
+
+### Failure Modes & Handling
+
+Failures are generally handled by transition back into the `Active`
+(neutral) state. This simplifies the implementation greatly at the
+cost of adding transitions to the state machine. For example, we
+could detect the `Draining` state upon restart and proceed with a drain,
+but how should the storage controller know that's what the orchestrator
+needs still?
+
+#### Storage Controller Crash
+
+When the storage controller starts up reset the node scheduling policy
+of all nodes in states `Draining`, `Filling` or `PauseForRestart` to
+`Active`. The rationale is that when the storage controller restarts,
+we have lost context of what the deployment orchestrator wants. It also
+has the benefit of making things easier to reason about.
+
+#### Pageserver Crash During Drain
+
+The pageserver will attempt to re-attach during restart at which
+point the node scheduling policy will be set back to `Active`, thus
+reenabling the scheduler to use the node.
+
+#### Non-drained Pageserver Crash During Drain
+
+What should happen when a pageserver we are draining to crashes during the
+process. Two reasonable options are: cancel the drain and focus on the failover
+*or* do both, but prioritise failover. Since the number of concurrent reconciles
+produced by drains/fills are limited, we get the later behaviour for free.
+My suggestion is we take this approach, but the cancellation option is trivial
+to implement as well.
+
+#### Pageserver Crash During Fill
+
+The pageserver will attempt to re-attach during restart at which
+point the node scheduling policy will be set back to `Active`, thus
+reenabling the scheduler to use the node.
+
+#### Pageserver Goes unavailable During Drain/Fill
+
+The drain and fill jobs handle this by stopping early. When the pageserver
+is detected as online by storage controller heartbeats, reset its scheduling
+policy to `Active`. If a restart happens instead, see the pageserver crash
+failure mode.
+
+#### Orchestrator Drain Times Out
+
+Orchestrator will still proceed with the restart.
+When the pageserver re-attaches, the scheduling policy is set back to
+`Active`.
+
+#### Orchestrator Fill Times Out
+
+Orchestrator will attempt to cancel the fill operation. If that fails,
+the fill will continue until it quiesces and the node will be left
+in the `Filling` scheduling policy. This hinders the scheduler, but is
+otherwise harmless. A human operator can handle this by setting the scheduling
+policy to `Active`, or we can bake in a fill timeout into the storage controller.
+
+## Optimizations
+
+### Location Warmth
+
+When cutting over to a secondary, the storage controller will wait for it to
+become "warm" (i.e. download enough of the tenants data). This means that some
+reconciliations can take significantly longer than others and hold up precious
+reconciliations units. As an optimization, the drain stage can only cut over
+tenants that are already "warm". Similarly, the fill stage can prioritise the
+"warmest" tenants in the fill.
+
+Given that the number of tenants by the storage controller will be fairly low
+for the foreseable future, the first implementation could simply query the tenants
+for secondary status. This doesn't scale well with increasing tenant counts, so
+eventually we will need new pageserver API endpoints to report the sets of
+"warm" and "cold" nodes.
+
+## Alternatives Considered
+
+### Draining and Filling Purely as Scheduling Constraints
+
+At its core, the storage controller is a big background loop that detects changes
+in the environment and reacts on them. One could express draining and filling
+of nodes purely in terms of constraining the scheduler (as opposed to having
+such background tasks).
+
+While theoretically nice, I think that's harder to implement and more importantly operate and reason about.
+Consider cancellation of a drain/fill operation. We would have to update the scheduler state, create
+an entirely new schedule (intent state) and start work on applying that. It gets trickier if we wish
+to cancel the reconciliation tasks spawned by drain/fill nodes. How would we know which ones belong
+to the conceptual drain/fill? One could add labels to reconciliations, but it gets messy in my opinion.
+
+It would also mean that reconciliations themselves have side effects that persist in the database
+(persist something to the databse when the drain is done), which I'm not conceptually fond of.
+
+## Proof of Concept
+
+This RFC is accompanied by a POC which implements nearly everything mentioned here
+apart from the optimizations and some of the failure handling:
+https://github.com/neondatabase/neon/pull/7682

docs/rfcs/034-ancestor-deletion.md

@@ -0,0 +1,252 @@
+# Ancestor Timeline Deletion
+
+Created on: 2024-02-23
+
+Author: John Spray
+
+# Summary
+
+When a tenant creates a new timeline that they will treat as their 'main' history,
+it is awkward to permanently retain an 'old main' timeline as its ancestor. Currently
+this is necessary because it is forbidden to delete a timeline which has descendents.
+
+A new pageserver API is proposed to 'adopt' data from a parent timeline into
+one of its children, such that the link between ancestor and child can be severed,
+leaving the parent in a state where it may then be deleted.
+
+# Motivation
+
+Retaining parent timelines currently has two costs:
+
+- Cognitive load on users, who have to remember which is the "real" main timeline.
+- Storage capacity cost, as the parent timeline will retain layers up to the
+  child's timeline point, even if the child fully covers its keyspace with image
+  layers and will never actually read from the parent.
+
+# Solution
+
+A new pageserver API `PUT /v1/tenant/:tenant_id/timeline/:timeline_id/detach_ancestor`
+will be added. The `timeline_id` in this URL is that of the _child_ timeline that we
+wish to detach from its parent.
+
+On success, this API will leave the following state:
+
+- The detached child timeline will no longer have an ancestor, and will contain all
+  the data needed to service reads without recursing into an ancestor.
+- Any other children of the parent whose timeline points were at a lower LSN than
+  the detached child timeline will be modified to have the child timeline as their
+  new parent.
+- The parent timeline will still exist, but the child will no longer have it as an
+  ancestor. If this was the last timeline that depended on the parent, then the
+  parent will become deletable.
+
+This API's implementation will consist of a series of retryable steps, such that
+on failures/timeout it can safely be called again to reach the target state.
+
+## Example
+
+### Before
+
+The user has "rolled back" their project to LSN X, resulting in a "new main"
+timeline. The parent "old main" timeline still exists, and they would like
+to clean it up.
+
+They have two other timelines A and B. A is from before the rollback point,
+and B is from after the rollback point.
+
+```
+----"old main" timeline-------X-------------------------------------------->
+                |             |                         |
+                |-> child A   |                         |
+                              |-> "new main" timeline   |
+                                                        -> child B
+
+```
+
+### After calling detach ancestor API
+
+The "new main" timeline is no longer dependent on old main, and neither
+is child A, because it had a branch point before X.
+
+The user may now choose to delete child B and "old main" to get to
+a pristine state. Child B is likely to be unwanted since the user
+chose to roll back to X, and it branches from after X. However, we
+don't assume this in the API; it is up to the user to delete it.
+
+```
+|----"old main" timeline---------------------------------------------------->
+                                                         |
+                                                         |
+                                                         |
+                                                         -> child B
+
+|----"new main" timeline--------->
+                 |
+                 |-> child A
+
+
+```
+
+### After removing timelines
+
+We end up with a totally clean state that leaves no trace that a rollback
+ever happened: there is only one root timeline.
+
+```
+| ----"new main" timeline----------->
+                |
+                |-> child A
+
+
+```
+
+## Caveats
+
+Important things for API users to bear in mind:
+
+- this API does not delete the parent timeline: you must still do that explicitly.
+- if there are other child timelines ahead of the branch point of the detached
+  child, the parent won't be deletable: you must either delete or detach those
+  children.
+- do _not_ simply loop over all children and detach them all: this can have an
+  extremely high storage cost. The detach ancestor API is intended for use on a single
+  timeline to make it the new "main".
+- The detach ancestor API should also not be
+  exposed directly to the user as button/API, because they might decide
+  to click it for all the children and thereby generate many copies of the
+  parent's data -- the detach ancestor API should be used as part
+  of a high level "clean up after rollback" feature.
+
+## `detach_ancestor` API implementation
+
+Terms used in the following sections:
+
+- "the child": the timeline whose ID is specified in the detach ancestor API URL, also
+  called "new main" in the example.
+- "the parent": the parent of "the child". Also called "old main" in the example.
+- "the branch point" the ancestor_lsn of "the child"
+
+### Phase 1: write out adopted layers to S3
+
+The child will "adopt" layers from the parent, such that its end state contains
+all the parent's history as well as its own.
+
+For all layers in the parent's layer map whose high LSN is below the branch
+point, issue S3 CopyObject requests to duplicate them into the child timeline's
+prefix. Do not add them to the child's layer map yet.
+
+For delta layers in the parent's layer map which straddle the branch point, read them
+and write out only content up to the branch point into new layer objects.
+
+This is a long running operation if the parent has many layers: it should be
+implemented in a way that resumes rather than restarting from scratch, if the API
+times out and is called again.
+
+As an optimization, if there are no other timelines that will be adopted into
+the child, _and_ the child's image layers already full cover the branch LSN,
+then we may skip adopting layers.
+
+### Phase 2: update the child's index
+
+Having written out all needed layers in phase 1, atomically link them all
+into the child's IndexPart and upload to S3. This may be done while the
+child Timeline is still running.
+
+### Phase 3: modify timelines ancestry
+
+Modify the child's ancestor to None, and upload its IndexPart to persist the change.
+
+For all timelines which have the same parent as the child, and have a branch
+point lower than our branch point, switch their ancestor_timeline to the child,
+and upload their IndexPart to persist the change.
+
+## Alternatives considered
+
+### Generate full image layer on child, rather than adopting parent deltas
+
+This would work for the case of a single child, but would prevent re-targeting
+other timelines that depended on the parent. If we detached many children this
+way, the storage cost would become prohibitive (consider a 1TB database with
+100 child timelines: it would cost 100TiB if they all generated their own image layers).
+
+### Don't rewrite anything: just fake it in the API
+
+We could add a layer of indirection that let a child "pretend" that it had no
+ancestor, when in reality it still had the parent. The pageserver API could
+accept deletion of ancestor timelines, and just update child metadata to make
+them look like they have no ancestor.
+
+This would not achieve the desired reduction in storage cost, and may well be more
+complex to maintain than simply implementing the API described in this RFC.
+
+### Avoid copying objects: enable child index to use parent layers directly
+
+We could teach IndexPart to store a TimelineId for each layer, such that a child
+timeline could reference a parent's layers directly, rather than copying them
+into the child's prefix.
+
+This would impose a cost for the normal case of indices that only target the
+timeline's own layers, add complexity, and break the useful simplifying
+invariant that timelines "own" their own path. If child timelines were
+referencing layers from the parent, we would have to ensure that the parent
+never runs GC/compaction again, which would make the API less flexible (the
+proposal in this RFC enables deletion of the parent but doesn't require it.)
+
+## Performance
+
+### Adopting layers
+
+- CopyObject is a relatively cheap operation, but we may need to issue tens of thousands
+  of such requests: this can take up to tens of seconds and will compete for RemoteStorage
+  semaphore units with other activity on the pageserver.
+- If we are running on storage backend that doesn't implement CopyObject, then
+  this part will be much more expensive as we would stream all layer content
+  through the pageserver. This is no different to issuing a lot
+  of reads to a timeline that does not have a warm local cache: it will move
+  a lot of gigabytes, but that shouldn't break anything.
+- Generating truncated layers for delta that straddle the branch point will
+  require streaming read/write of all the layers in question.
+
+### Updating timeline ancestry
+
+The simplest way to update timeline ancestry will probably be to stop and start
+all the Timeline objects: this is preferable to the complexity of making their
+ancestry mutable at runtime.
+
+There will be a corresponding "stutter" in the availability of the timelines,
+of the order 10-100ms, which is the time taken to upload their IndexPart, and
+restart the Timeline.
+
+# Interaction with other features
+
+## Concurrent timeline creation
+
+If new historic timelines are created using the parent as an ancestor while the
+detach ancestor API is running, they will not be re-parented to the child. This
+doesn't break anything, but it leaves the parent in a state where it might not
+be possible to delete it.
+
+Since timeline creations are an explicit user action, this is not something we need to
+worry about as the storage layer: a user who wants to delete their parent timeline will not create
+new children, and if they do, they can choose to delete those children to
+enable deleting the parent.
+
+For the least surprise to the user, before starting the detach ancestor branch
+operation, the control plane should wait until all branches are created and not
+allow any branches to be created before the branch point on the ancestor branch
+while the operation is ongoing.
+
+## WAL based disaster recovery
+
+WAL based disaster recovery currently supports only restoring of the main
+branch. Enabling WAL based disaster recovery in the future requires that we
+keep a record which timeline generated the WAL and at which LSN was a parent
+detached. Keep a list of timeline ids and the LSN in which they were detached in
+the `index_part.json`. Limit the size of the list to 100 first entries, after
+which the WAL disaster recovery will not be possible.
+
+## Sharded tenants
+
+For sharded tenants, calls to the detach ancestor API will pass through the storage
+controller, which will handle them the same as timeline creations: invoke first
+on shard zero, and then on all the other shards.

docs/rfcs/035-timeline-archive.md

@@ -0,0 +1,507 @@
+# Timeline Archival
+
+## Summary
+
+This RFC describes a mechanism for pageservers to eliminate local storage + compute work
+for timelines which are not in use, in response to external API calls to "archive" a timeline.
+
+The archived state roughly corresponds to fully offloading a timeline to object storage, such
+that its cost is purely the cost of that object storage.
+
+## Motivation
+
+Archived timelines serve multiple purposes:
+- Act as a 'snapshot' for workloads that would like to retain restorable copies of their
+  database from longer ago than their PITR window.
+- Enable users to create huge numbers of branches (e.g. one per github PR) without having
+  to diligently clean them up later to avoid overloading the pageserver (currently we support
+  up to ~500 branches per tenant).
+
+### Prior art
+
+Most storage and database systems have some form of snapshot, which can be implemented several ways:
+1. full copies of data (e.g. an EBS snapshot to S3)
+2. shallow snapshots which are CoW relative to the original version of the data, e.g. on a typical NFS appliance, or a filesystem like CephFS.
+3. a series of snapshots which are CoW or de-duplicated relative to one another.
+
+Today's Neon branches are approximately like `2.`, although due to implementation details branches
+often end up storing much more data than they really need, as parent branches assume that all data
+at the branch point is needed.  The layers pinned in the parent branch may have a much larger size
+than the physical size of a compressed image layer representing the data at the branch point.
+
+## Requirements
+
+- Enter & exit the archived state in response to external admin API calls
+- API calls to modify the archived state are atomic and durable
+- An archived timeline should eventually (once out of PITR window) use an efficient compressed
+  representation, and avoid retaining arbitrarily large data in its parent branch.
+- Remote object GETs during tenant start may be O(N) with the number of _active_ branches,
+  but must not scale with the number of _archived_ branches.
+- Background I/O for archived branches should only be done a limited number of times to evolve them
+  to a long-term-efficient state (e.g. rewriting to image layers).  There should be no ongoing "housekeeping"
+  overhead for archived branches, including operations related to calculating sizes for billing.
+- The pageserver should put no load on the safekeeper for archived branches.
+- Performance of un-archiving a branch must make good use of S3/disk bandwidth to restore the branch
+  to a performant state in a short time (linear with the branch's logical size)
+
+## Non Goals
+
+- Archived branches are not a literal `fullbackup` postgres snapshot: they are still stored
+  in Neon's internal format.
+- Compute cold starts after activating an archived branch will not have comparable performance to
+  cold starts on an active branch.
+- Archived branches will not use any new/additional compression or de-duplication beyond what
+  is already implemented for image layers (zstd per page).
+- The pageserver will not "auto start" archived branches in response to page_service API requests: they
+  are only activated explicitly via the HTTP API.
+- We will not implement a total offload of archived timelines from safekeepers: their control file (small) will
+  remain on local disk, although existing eviction mechanisms will remove any segments from local disk.
+- We will not expose any prometheus metrics for archived timelines, or make them visible in any
+  detailed HTTP APIs other than the specific API for listing archived timelines.
+- A parent branch may not be archived unless all its children are.
+
+## Impacted Components
+
+pageserver, storage controller
+
+## Terminology
+
+**Archived**: a branch is _archived_ when an HTTP API request to archive it has succeeded: the caller
+may assume that this branch is now very cheap to store, although this may not be physically so until the
+branch proceeds to the offloaded state.
+
+**Active** branches are branches which are available for use by page_service clients, and have a relatively
+high cost due to consuming local storage.
+
+**Offloaded** branches are a subset of _archived_ branches, which have had their local state removed such
+that they now consume minimal runtime resources and have a cost similar to the cost of object storage.
+
+**Activate** (verb): transition from Archived to Active
+
+**Archive** (verb): transition from Active to Archived
+
+**Offload** (verb): transition from Archived to Offloaded
+
+**Offload manifest**: an object stored in S3 that describes timelines which pageservers do not load.
+
+**Warm up** (verb): operation done on an active branch, by downloading its active layers.  Once a branch is
+warmed up, good performance will be available to page_service clients.
+
+## Implementation
+
+### High level flow
+
+We may think of a timeline which is archived and then activated as proceeding through a series of states:
+
+```mermaid
+stateDiagram
+  [*] --> Active(warm)
+  Active(warm) --> Archived
+  Archived --> Offloaded
+  Archived --> Active(warm)
+  Offloaded --> Active(cold)
+  Active(cold) --> Active(warm)
+```
+
+Note that the transition from Archived to Active(warm) is expected to be fairly rare: the most common lifecycles
+of branches will be:
+- Very frequent: Short lived branches: Active -> Deleted
+- Frequent: Long-lived branches: Active -> Archived -> Offloaded -> Deleted
+- Rare: Branches used to restore old state: Active ->Archived -> Offloaded -> Active
+
+These states are _not_ all stored as a single physical state on the timeline, but rather represent the combination
+of:
+- the timeline's lifecycle state: active or archived, stored in the timeline's index
+- its offload state: whether pageserver has chosen to drop local storage of the timeline and write it into the
+  manifest of offloaded timelines.
+- cache state (whether it's warm or cold).
+
+### Storage format changes
+
+There are two storage format changes:
+1. `index_part.json` gets a new attribute `state` that describes whether the timeline is to
+   be considered active or archived.
+2. A new tenant-level _manifest_ object `tenant_manifest-v1.json` describes which timelines a tenant does not need to load
+   at startup (and is available for storing other small, rarely changing tenant-wide attributes in future)
+
+The manifest object will have a format like this:
+```
+{
+  "offload_timelines": [
+    {
+      "timeline_id": ...
+      "last_record_lsn": ...
+      "last_record_lsn_time": ...
+      "pitr_interval": ...
+      "last_gc_lsn": ...  # equal to last_record_lsn if this branch has no history (i.e. a snapshot)
+      "logical_size": ...  # The size at last_record_lsn
+      "physical_size" ...
+      "parent": Option<{
+        "timeline_id"...
+        "lsn"... # Branch point LSN on the parent
+        "requires_data": bool # True if this branch depends on layers in its parent, identify it here
+
+      }>
+    }
+  ]
+}
+```
+
+The information about a timeline in its offload state is intentionally minimal: just enough to decide:
+- Whether it requires [archive optimization](#archive-branch-optimization) by rewriting as a set of image layers: we may infer this
+  by checking if now > last_record_lsn_time - pitr_interval, and pitr_lsn < last_record_lsn.
+- Whether a parent branch should include this offloaded branch in its GC inputs to avoid removing
+  layers that the archived branch depends on
+- Whether requests to delete this `timeline_id` should be executed (i.e. if a deletion request
+  is received for a timeline_id that isn't in the site of live `Timelines` or in the manifest, then
+  we don't need to go to S3 for the deletion.
+- How much archived space to report in consumption metrics
+
+The contents of the manifest's offload list will also be stored as an attribute of `Tenant`, such that the total
+set of timelines may be found by the union of `Tenant::timelines` (non-offloaded timelines) and `Tenant::offloaded`
+(offloaded timelines).
+
+For split-brain protection, the manifest object will be written with a generation suffix, in the same way as
+index_part objects are (see [generation numbers RFC](025-generation-numbers.md)).  This will add some complexity, but
+give us total safety against two pageservers with the same tenant attached fighting over the object.  Existing code
+for finding the latest generation and for cleaning up old generations (in the scrubber) will be generalized to cover
+the manifest file.
+
+### API & Timeline state
+
+Timelines will store a lifecycle state (enum of Active or Archived) in their IndexPart.  This will
+be controlled by a new per-timeline `configure` endpoint.  This is intentionally generic naming, which
+may be used in future to control other per-timeline attributes (e.g. in future we may make PITR interval
+a per-timeline configuration).
+
+`PUT /v1/tenants/{tenant_id}/timelines/{timeline_id}/configure`
+```
+{
+  'state': 'active|archive'
+}
+```
+
+When archiving a timeline, this API will complete as soon as the timeline's state has been set in index_part, and that index has been uploaded.
+
+When activating a timeline, this API will complete as soon as the timeline's state has been set in index_part,
+**and** the `Timeline` object has been instantiated and activated.  This will require reading the timeline's
+index, but not any data: it should be about as fast as a couple of small S3 requests.
+
+The API will be available with identical path via the storage controller: calling this on a sharded tenant
+will simply map the API call to all the shards.
+
+Archived timelines may never have descendent timelines which are active.  This will be enforced at the API level,
+such that activating a timeline requires that all its ancestors are active, and archiving a timeline requires
+that all its descendents are archived.  It is the callers responsibility to walk the hierarchy of timelines
+in the proper order if they would like to archive whole trees of branches.
+
+Because archive timelines will be excluded from the usual timeline listing APIs, a new API specifically
+for archived timelines will be added: this is for use in support/debug:
+
+```
+GET /v1/tenants/{tenant_id}/archived_timelines
+
+{
+  ...same per-timeline content as the tenant manifest...
+}
+
+```
+
+### Tenant attach changes
+
+Currently, during Tenant::spawn we list all the timelines in the S3 bucket, and then for each timeline
+we load their index_part.json.  To avoid the number of GETs scaling linearly with the number of archived
+timelines, we must have a single object that tells us which timelines do not need to be loaded.  The
+number of ListObjects requests while listing timelines will still scale O(N), but this is less problematic
+because each request covers 1000 timelines.
+
+This is **not** literally the same as the set of timelines who have state=archived.  Rather, it is
+the set of timelines which have been offloaded in the background after their state was set to archived.
+
+We may simply skip loading these timelines: there will be no special state of `Timeline`, they just won't
+exist from the perspective of an active `Tenant` apart from in deletion: timeline deletion will need
+to check for offloaded timelines as well as active timelines, to avoid wrongly returning 404 on trying
+to delete an offloaded timeline.
+
+### Warm-up API
+
+`PUT /v1/tenants/{tenant_id}/timelines/{timeline_id}/download?wait_ms=1234`
+
+This API will be similar to the existing `download_remote_layers` API, but smarter:
+- It will not download _all_ remote layers, just the visible set (i.e. layers needed for a read)
+- It will download layers in the visible set until reaching `wait_ms`, then return a struct describing progress
+  of downloads, so that the caller can poll.
+
+The _visible set_ mentioned above will be calculated by the pageserver in the background, by taking the set
+of readable LSNs (i.e. branch points and heads of branches), and walking the layer map to work out which layers
+can possibly be read from these LSNs.  This concept of layer visibility is more generally useful for cache
+eviction and heatmaps, as well as in this specific case of warming up a timeline.
+
+The caller does not have to wait for the warm up API, or call it at all.  But it is strongly advised
+to call it, because otherwise populating local contents for a timeline can take a long time when waiting
+for SQL queries to coincidentally hit all the layers, and during that time query latency remains quite
+volatile.
+
+### Background work
+
+Archived branches are not subject to normal compaction.  Instead, when the compaction loop encounters
+an archived branch, it will consider rewriting the branch to just image layers if the branch has no history
+([archive branch optimization](#archive-branch-optimization)), or offloading the timeline from local disk
+if its state permits that.
+
+Additionally, the tenant compaction task will walk the state of already offloaded timelines to consider
+optimizing their storage, e.g. if a timeline had some history when offloaded, but since then its PITR
+has elapsed and it can now be rewritten to image layers.
+
+#### Archive branch offload
+
+Recall that when we archive a timeline via the HTTP API, this only sets a state: it doesn't do
+any actual work.
+
+This work is done in the background compaction loop.  It makes sense to tag this work on to the compaction
+loop, because it is spiritually aligned: offloading data for archived branches improves storage efficiency.
+
+The condition for offload is simple:
+ - a `Timeline` object exists with state `Archived`
+ - the timeline does not have any non-offloaded children.
+ 
+ Regarding the condition that children must be offloaded, this will always be eventually true, because
+ we enforce at the API level that children of archived timelines must themselves be archived, and all
+ archived timelines will eventually be offloaded.
+
+Offloading a timeline is simple:
+- Read the timeline's attributes that we will store in its offloaded state (especially its logical size)
+- Call `shutdown()` on the timeline and remove it from the `Tenant` (as if we were about to delete it)
+- Erase all the timeline's content from local storage (`remove_dir_all` on its path)
+- Write the tenant manifest to S3 to prevent this timeline being loaded on next start.
+
+#### Archive branch optimization (flattening)
+
+When we offloaded a branch, it might have had some history that prevented rewriting it to a single
+point in time set of image layers.  For example, a branch might have several days of writes and a 7
+day PITR: when we archive it, it still has those days of history.
+
+Once the PITR has expired, we have an opportunity to reduce the physical footprint of the branch by:
+- Writing compressed image layers within the archived branch, as these are more efficient as a way of storing
+  a point in time compared with delta layers
+- Updating the branch's offload metadata to indicate that this branch no longer depends on its ancestor
+  for data, i.e. the ancestor is free to GC layers files at+below the branch point
+
+Fully compacting an archived branch into image layers at a single LSN may be thought of as *flattening* the
+branch, such that it is now a one-dimensional keyspace rather than a two-dimensional key/lsn space. It becomes
+a true snapshot at that LSN.
+
+It is not always more efficient to flatten a branch than to keep some extra history on the parent: this
+is described in more detail in [optimizations](#delaying-storage-optimization-if-retaining-parent-layers-is-cheaper)
+
+Archive branch optimization should be done _before_ background offloads during compaction, because there may
+be timelines which are ready to be offloaded but also would benefit from the optimization step before
+being offloaded.  For example, a branch which has already fallen out of PITR window and has no history
+of its own may be immediately re-written as a series of image layers before being offloaded.
+
+### Consumption metrics
+
+Archived timelines and offloaded timelines will be excluded from the synthetic size calculation, in anticipating
+that billing structures based on consumption metrics are highly likely to apply different $/GB rates to archived
+vs. ordinary content.
+
+Archived and offloaded timelines' logical size will be reported under the existing `timeline_logical_size`
+variant of `MetricsKey`: receivers are then free to bill on this metric as they please.
+
+### Secondary locations
+
+Archived timelines (including offloaded timelines) will be excluded from heatmaps, and thereby
+when a timeline is archived, after the next cycle of heatmap upload & secondary download, its contents
+will be dropped from secondary locations.
+
+### Sharding
+
+Archiving or activating a timeline will be done symmetrically across all shards in a tenant, in
+the same way that timeline creation and deletion is done.  There are no special rules about ordering:
+the storage controller may dispatch concurrent calls to all shards when archiving or activating a timeline.
+
+Since consumption metrics are only transmitted from shard zero, the state of archival on this shard
+will be authoritative for consumption metrics.
+
+## Error cases
+
+### Errors in sharded tenants
+
+If one shard in a tenant fails an operation but others succeed, the tenant may end up in a mixed
+state, where a timeline is archived on some shards but not on others.  
+
+We will not bother implementing a rollback mechanism for this: errors in archiving/activating a timeline
+are either transient (e.g. S3 unavailable, shutting down), or the fault of the caller (NotFound, BadRequest).
+In the transient case callers are expected to retry until success, or to make appropriate API calls to clear
+up their mistake.  We rely on this good behavior of callers to eventually get timelines into a consistent
+state across all shards.  If callers do leave a timeline in an inconsistent state across shards, this doesn't
+break anything, it's just "weird".
+
+This is similar to the status quo for timeline creation and deletion: callers are expected to retry
+these operations until they succeed.
+
+### Archiving/activating
+
+Archiving/activating a timeline can fail in a limited number of ways:
+1. I/O error storing/reading the timeline's updated index
+    - These errors are always retryable: a fundamental design assumption of the pageserver is that remote
+      storage errors are always transient. 
+2. NotFound if the timeline doesn't exist
+    - Callers of the API are expected to avoid calling deletion and archival APIs concurrently.
+    - The storage controller has runtime locking to prevent races such as deleting a timeline while
+      archiving it.
+3. BadRequest if the rules around ancestors/descendents of archived timelines would be violated
+    - Callers are expected to do their own checks to avoid hitting this case.  If they make
+      a mistake and encounter this error, they should give up.
+
+### Offloading
+
+Offloading can only fail if remote storage is unavailable, which would prevent us from writing the
+tenant manifest.  In such error cases, we give up in the expectation that offloading will be tried 
+again at the next iteration of the compaction loop.
+
+### Archive branch optimization
+
+Optimization is a special form of compaction, so can encounter all the same errors as regular compaction
+can: it should return Result<(), CompactionError>, and as with compaction it will be retried on
+the next iteration of the compaction loop.
+
+## Optimizations
+
+### Delaying storage optimization if retaining parent layers is cheaper
+
+Optimizing archived branches to image layers and thereby enabling parent branch GC to progress
+is a safe default: archived branches cannot over-fill a pageserver's local disk, and once they
+are offloaded to S3 they're totally safe, inert things.
+
+However, in some cases it can be advantageous to retain extra history on their parent branch rather
+than flattening the archived branch.  For example, if a 1TB parent branch is rather slow-changing (1GB
+of data per day), and archive branches are being created nightly, then writing out full 1TB image layers
+for each nightly branch is inefficient compared with just keeping more history on the main branch.
+
+Getting this right requires consideration of:
+- Compaction: if keeping more history on the main branch is going to prompt the main branch's compaction to
+  write out extra image layers, then it might make more sense to just write out the image layers on
+  the archived branch.
+- Metadata bloat: keeping extra history on a parent branch doesn't just cost GB of storage, it makes
+  the layer map (and index_part) bigger.  There are practical limits beyond which writing an indefinitely
+  large layer map can cause problems elsewhere.
+
+This optimization can probably be implemented quite cheaply with some basic heuristics like:
+- don't bother doing optimization on an archive branch if the LSN distance between
+  its branch point and the end of the PITR window is <5% of the logical size of the archive branch.
+- ...but, Don't keep more history on the main branch than double the PITR
+
+### Creating a timeline in archived state (a snapshot)
+
+Sometimes, one might want to create a branch with no history, which will not be written to
+before it is archived.  This is a snapshot, although we do not require a special snapshot API,
+since a snapshot can be represented as a timeline with no history.
+
+This can be accomplished by simply creating a timeline and then immediately archiving it, but
+that is somewhat wasteful: this timeline it will spin up various tasks and open a connection to the storage
+broker to try and ingest WAL, before being shutdown in the subsequent archival call.  To explicitly
+support this common special case, we may add a parameter to the timeline creation API which
+creates a timeline directly into the archived state.
+
+Such a timeline creation will do exactly two I/Os at creation time:
+- write the index_part object to record the timeline's existence
+- when the timeline is offloaded in the next iteration of the compaction loop (~20s later),
+  write the tenant manifest.
+
+Later, when the timeline falls off the end of the PITR interval, the usual offload logic will wake
+up the 'snapshot' branch and write out image layers.
+
+## Future Work
+
+### Enabling `fullbackup` dumps from archive branches
+
+It would be useful to be able to export an archive branch to another system, or for use in a local
+postgres database.
+
+This could be implemented as a general capability for all branches, in which case it would "just work"
+for archive branches by activating them.  However, downloading all the layers in a branch just to generate
+a fullbackup is a bit inefficient: we could implement a special case for flattened archived branches
+which streams image layers from S3 and outputs the fullbackup stream without writing the layers out to disk.
+
+Implementing `fullbackup` is a bit more complicated than this because of sharding, but solving that problem
+is unrelated to the topic of archived branches (it probably involves having each shard write out a fullbackup 
+stream to S3 in an intermediate format and, then having one node stitch them together).
+
+### Tagging layers from archived branches
+
+When we know a layer is an image layer written for an archived branch that has fallen off the PITR window,
+we may add tags to the S3 objects to enable writing lifecycle policies that transition such layers to even
+cheaper storage.
+
+This could be done for all archived layers, or it could be driven by the archival API, to give the pageserver
+external hints on which branches are likely to be reactivated, and which branches are good candidates for
+tagging for low performance storage.
+
+Tagging+lifecycles is just one mechanism: one might also directly use S3 storage classes.  Other clouds' object
+stores have similar mechanisms.
+
+### Storing sequences of archive branches as deltas
+
+When archived branches are used as scheduled snapshots, we could store them even more efficiently
+by encoding them as deltas relative to each other (i.e. for nightly snapshots, when we do the
+storage optimization for Tuesday's snapshot, we would read Monday's snapshot and store only the modified
+pages). This is the kind of encoding that many backup storage systems use.
+
+The utility of this depends a lot on the churn rate of the data, and the cost of doing the delta encoding
+vs. just writing out a simple stream of the entire database.  For smaller databases, writing out a full
+copy is pretty trivial (e.g. writing a compressed copy of a 10GiB database to S3 can take under 10 seconds,
+so the complexity tradeoff of diff-encoding it is dubious).
+
+One does not necessarily have to read-back the previous snapshot in order to encoded the next one: if the
+pageserver knows about the schedule, it can intentionally retain extra history on the main branch so that
+we can say: "A branch exists from Monday night.  I have Monday night's data still active in the main branch,
+so now I can read at the Monday LSN and the Tuesday LSN, calculate the delta, and store it as Tuesday's
+delta snapshot".
+
+Clearly this all requires careful housekeeping to retain the relationship between branches that depend on
+each other: perhaps this would be done by making the archive branches have child/parent relationships with
+each other, or perhaps we would permit them to remain children of their original parent, but additionally
+have a relationship with the snapshot they're encoded relative to.
+
+Activating a branch that is diff-encoded may require activating several earlier branches too, so figuring
+out how frequently to write a full copy is important.  This is essentially a zoomed-out version of what
+we do with delta layers and image layers within a timeline, except each "layer" is a whole timeline.
+
+
+## FAQ/Alternatives
+
+### Store all timelines in the tenant manifest
+
+Rather than special-casing offloaded timelines in the offload manifest, we could store a total
+manifest of all timelines, eliminating the need for the pageserver to list timelines in S3 on
+startup.
+
+That would be a more invasive change (require hooking in to timeline creation), and would
+generate much more I/O to this manifest for tenants that had many branches _and_ frequent
+create/delete cycles for short lived branches.  Restricting the manifest to offloaded timelines
+means that we only have to cope with the rate at which long-lived timelines are archived, rather
+than the rate at which sort lived timelines are created & destroyed.
+
+### Automatically archiving/activating timelines without external API calls
+
+We could implement TTL driven offload of timelines, waking them up when a page request
+arrives.
+
+This has downsides:
+- Opacity: if we do TTL-driven offload inside the pageserver, then the end user doesn't
+  know which of their branches are in this state, and might get a surprise when they try
+  to use such a branch.
+- Price fluctuation: if the archival of a branch is used in end user pricing, then users
+  prefer clarity & consistency.  Ideally a branch's storage should cost the same from the moment it
+  is created, rather than having a usage-dependency storage price.
+- Complexity: enabling the page service to call up into the Tenant to activate a timeline
+  would be awkward, compared with an external entry point.
+
+### Make offloaded a state of Timeline
+
+To reduce the operator-facing complexity of having some timelines APIs that only return
+non-offloaded timelines, we could build the offloaded state into the Timeline type.
+
+`timeline.rs` is already one of the most egregiously long source files in the tree, so
+this is rejected on the basis that we need to avoid making that complexity worse.

docs/storage_controller.md

@@ -44,7 +44,7 @@ If you need to modify the database schema, here’s how to create a migration:
 - Use `diesel migration generate <name>` to create a new migration
 - Populate the SQL files in the `migrations/` subdirectory
 - Use `DATABASE_URL=... diesel migration run` to apply the migration you just wrote: this will update the `[schema.rs](http://schema.rs)` file automatically.
-  - This requires a running database: the easiest way to do that is to just run `cargo neon init ; cargo neon start`, which will leave a database available at `postgresql://localhost:1235/attachment_service`
+  - This requires a running database: the easiest way to do that is to just run `cargo neon init ; cargo neon start`, which will leave a database available at `postgresql://localhost:1235/storage_controller`
 - Commit the migration files and the changes to schema.rs
 - If you need to iterate, you can rewind migrations with `diesel migration revert -a` and then `diesel migration run` again.
 - The migrations are build into the storage controller binary, and automatically run at startup after it is deployed, so once you’ve committed a migration no further steps are needed.

docs/synthetic-size.md

@@ -21,9 +21,9 @@ 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
-the size of an individual branch?" for discussion on those
-difficulties.
+straightforward to attribute size to individual branches. See [What is
+the size of an individual branch?](#what-is-the-size-of-an-individual-branch)
+for a discussion of those difficulties.
 
 The synthetic size is designed to:
 
@@ -40,8 +40,9 @@ The synthetic size is designed to:
 - logical size is the size of a branch *at a given point in
   time*. It's the total size of all tables in all databases, as you
   see with "\l+" in psql for example, plus the Postgres SLRUs and some
-  small amount of metadata. NOTE that currently, Neon does not include
-  the SLRUs and metadata in the logical size. See comment to `get_current_logical_size_non_incremental()`.
+  small amount of metadata. Note that currently, Neon does not include
+  the SLRUs and metadata in the logical size. Refer to the comment in
+  [`get_current_logical_size_non_incremental()`](/pageserver/src/pgdatadir_mapping.rs#L813-L814).
 
 - a "point in time" is defined as an LSN value. You can convert a
   timestamp to an LSN, but the storage internally works with LSNs.

libs/metrics/src/hll.rs

@@ -13,11 +13,7 @@ use std::{
 
 use measured::{
     label::{LabelGroupVisitor, LabelName, LabelValue, LabelVisitor},
-    metric::{
-        group::{Encoding, MetricValue},
-        name::MetricNameEncoder,
-        Metric, MetricType, MetricVec,
-    },
+    metric::{counter::CounterState, name::MetricNameEncoder, Metric, MetricType, MetricVec},
     text::TextEncoder,
     LabelGroup,
 };
@@ -144,6 +140,7 @@ impl<const N: usize> HyperLogLogState<N> {
         })
     }
 }
+
 impl<W: std::io::Write, const N: usize> measured::metric::MetricEncoding<TextEncoder<W>>
     for HyperLogLogState<N>
 {
@@ -182,12 +179,13 @@ impl<W: std::io::Write, const N: usize> measured::metric::MetricEncoding<TextEnc
             .into_iter()
             .enumerate()
             .try_for_each(|(hll_shard, val)| {
-                enc.write_metric_value(
-                    name.by_ref(),
+                CounterState::new(val as u64).collect_into(
+                    &(),
                     labels.by_ref().compose_with(HllShardLabel {
                         hll_shard: hll_shard as i64,
                     }),
-                    MetricValue::Int(val as i64),
+                    name.by_ref(),
+                    enc,
                 )
             })
     }

libs/metrics/src/lib.rs

@@ -9,7 +9,7 @@ use measured::{
     metric::{
         counter::CounterState,
         gauge::GaugeState,
-        group::{Encoding, MetricValue},
+        group::Encoding,
         name::{MetricName, MetricNameEncoder},
         MetricEncoding, MetricFamilyEncoding,
     },
@@ -171,8 +171,11 @@ fn write_gauge<Enc: Encoding>(
     labels: impl LabelGroup,
     name: impl MetricNameEncoder,
     enc: &mut Enc,
-) -> Result<(), Enc::Err> {
-    enc.write_metric_value(name, labels, MetricValue::Int(x))
+) -> Result<(), Enc::Err>
+where
+    GaugeState: MetricEncoding<Enc>,
+{
+    GaugeState::new(x).collect_into(&(), labels, name, enc)
 }
 
 #[derive(Default)]
@@ -544,15 +547,6 @@ impl<T: Encoding> Encoding for Inc<T> {
     fn write_help(&mut self, name: impl MetricNameEncoder, help: &str) -> Result<(), Self::Err> {
         self.0.write_help(name, help)
     }
-
-    fn write_metric_value(
-        &mut self,
-        name: impl MetricNameEncoder,
-        labels: impl LabelGroup,
-        value: MetricValue,
-    ) -> Result<(), Self::Err> {
-        self.0.write_metric_value(name, labels, value)
-    }
 }
 
 impl<T: Encoding> MetricEncoding<Inc<T>> for MeasuredCounterPairState
@@ -579,15 +573,6 @@ impl<T: Encoding> Encoding for Dec<T> {
     fn write_help(&mut self, name: impl MetricNameEncoder, help: &str) -> Result<(), Self::Err> {
         self.0.write_help(name, help)
     }
-
-    fn write_metric_value(
-        &mut self,
-        name: impl MetricNameEncoder,
-        labels: impl LabelGroup,
-        value: MetricValue,
-    ) -> Result<(), Self::Err> {
-        self.0.write_metric_value(name, labels, value)
-    }
 }
 
 /// Write the dec counter to the encoder

libs/pageserver_api/src/controller_api.rs

@@ -1,4 +1,6 @@
+use std::collections::HashSet;
 use std::str::FromStr;
+use std::time::{Duration, Instant};
 
 /// Request/response types for the storage controller
 /// API (`/control/v1` prefix).  Implemented by the server
@@ -87,7 +89,7 @@ pub struct TenantLocateResponse {
     pub shard_params: ShardParameters,
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Serialize, Deserialize, Debug)]
 pub struct TenantDescribeResponse {
     pub tenant_id: TenantId,
     pub shards: Vec<TenantDescribeResponseShard>,
@@ -110,7 +112,7 @@ pub struct NodeDescribeResponse {
     pub listen_pg_port: u16,
 }
 
-#[derive(Serialize, Deserialize)]
+#[derive(Serialize, Deserialize, Debug)]
 pub struct TenantDescribeResponseShard {
     pub tenant_shard_id: TenantShardId,
 
@@ -150,11 +152,16 @@ impl UtilizationScore {
     }
 }
 
-#[derive(Serialize, Deserialize, Clone, Copy, Debug)]
+#[derive(Serialize, Clone, Copy, Debug)]
 #[serde(into = "NodeAvailabilityWrapper")]
 pub enum NodeAvailability {
     // Normal, happy state
     Active(UtilizationScore),
+    // Node is warming up, but we expect it to become available soon. Covers
+    // the time span between the re-attach response being composed on the storage controller
+    // and the first successful heartbeat after the processing of the re-attach response
+    // finishes on the pageserver.
+    WarmingUp(Instant),
     // Offline: Tenants shouldn't try to attach here, but they may assume that their
     // secondary locations on this node still exist.  Newly added nodes are in this
     // state until we successfully contact them.
@@ -164,7 +171,10 @@ pub enum NodeAvailability {
 impl PartialEq for NodeAvailability {
     fn eq(&self, other: &Self) -> bool {
         use NodeAvailability::*;
-        matches!((self, other), (Active(_), Active(_)) | (Offline, Offline))
+        matches!(
+            (self, other),
+            (Active(_), Active(_)) | (Offline, Offline) | (WarmingUp(_), WarmingUp(_))
+        )
     }
 }
 
@@ -176,6 +186,7 @@ impl Eq for NodeAvailability {}
 #[derive(Serialize, Deserialize, Clone, Copy, Debug)]
 pub enum NodeAvailabilityWrapper {
     Active,
+    WarmingUp,
     Offline,
 }
 
@@ -185,6 +196,7 @@ impl From<NodeAvailabilityWrapper> for NodeAvailability {
             // Assume the worst utilisation score to begin with. It will later be updated by
             // the heartbeats.
             NodeAvailabilityWrapper::Active => NodeAvailability::Active(UtilizationScore::worst()),
+            NodeAvailabilityWrapper::WarmingUp => NodeAvailability::WarmingUp(Instant::now()),
             NodeAvailabilityWrapper::Offline => NodeAvailability::Offline,
         }
     }
@@ -194,6 +206,7 @@ impl From<NodeAvailability> for NodeAvailabilityWrapper {
     fn from(val: NodeAvailability) -> Self {
         match val {
             NodeAvailability::Active(_) => NodeAvailabilityWrapper::Active,
+            NodeAvailability::WarmingUp(_) => NodeAvailabilityWrapper::WarmingUp,
             NodeAvailability::Offline => NodeAvailabilityWrapper::Offline,
         }
     }
@@ -282,6 +295,42 @@ pub enum PlacementPolicy {
 #[derive(Serialize, Deserialize, Debug)]
 pub struct TenantShardMigrateResponse {}
 
+/// Metadata health record posted from scrubber.
+#[derive(Serialize, Deserialize, Debug)]
+pub struct MetadataHealthRecord {
+    pub tenant_shard_id: TenantShardId,
+    pub healthy: bool,
+    pub last_scrubbed_at: chrono::DateTime<chrono::Utc>,
+}
+
+#[derive(Serialize, Deserialize, Debug)]
+pub struct MetadataHealthUpdateRequest {
+    pub healthy_tenant_shards: HashSet<TenantShardId>,
+    pub unhealthy_tenant_shards: HashSet<TenantShardId>,
+}
+
+#[derive(Serialize, Deserialize, Debug)]
+pub struct MetadataHealthUpdateResponse {}
+
+#[derive(Serialize, Deserialize, Debug)]
+
+pub struct MetadataHealthListUnhealthyResponse {
+    pub unhealthy_tenant_shards: Vec<TenantShardId>,
+}
+
+#[derive(Serialize, Deserialize, Debug)]
+
+pub struct MetadataHealthListOutdatedRequest {
+    #[serde(with = "humantime_serde")]
+    pub not_scrubbed_for: Duration,
+}
+
+#[derive(Serialize, Deserialize, Debug)]
+
+pub struct MetadataHealthListOutdatedResponse {
+    pub health_records: Vec<MetadataHealthRecord>,
+}
+
 #[cfg(test)]
 mod test {
     use super::*;

libs/pageserver_api/src/key.rs

@@ -29,7 +29,7 @@ pub const KEY_SIZE: usize = 18;
 /// See [`Key::to_i128`] for more information on the encoding.
 pub const METADATA_KEY_SIZE: usize = 16;
 
-/// The key prefix start range for the metadata keys. All keys with the first byte >= 0x40 is a metadata key.
+/// The key prefix start range for the metadata keys. All keys with the first byte >= 0x60 is a metadata key.
 pub const METADATA_KEY_BEGIN_PREFIX: u8 = 0x60;
 pub const METADATA_KEY_END_PREFIX: u8 = 0x7F;
 
@@ -107,7 +107,10 @@ impl Key {
     /// As long as Neon does not support tablespace (because of lack of access to local file system),
     /// we can assume that only some predefined namespace OIDs are used which can fit in u16
     pub fn to_i128(&self) -> i128 {
-        assert!(self.field2 <= 0xFFFF || self.field2 == 0xFFFFFFFF || self.field2 == 0x22222222);
+        assert!(
+            self.field2 <= 0xFFFF || self.field2 == 0xFFFFFFFF || self.field2 == 0x22222222,
+            "invalid key: {self}",
+        );
         (((self.field1 & 0x7F) as i128) << 120)
             | (((self.field2 & 0xFFFF) as i128) << 104)
             | ((self.field3 as i128) << 72)

libs/pageserver_api/src/keyspace.rs

@@ -17,6 +17,16 @@ pub struct KeySpace {
     pub ranges: Vec<Range<Key>>,
 }
 
+impl std::fmt::Display for KeySpace {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(f, "[")?;
+        for range in &self.ranges {
+            write!(f, "{}..{},", range.start, range.end)?;
+        }
+        write!(f, "]")
+    }
+}
+
 /// A wrapper type for sparse keyspaces.
 #[derive(Clone, Debug, Default, PartialEq, Eq)]
 pub struct SparseKeySpace(pub KeySpace);

libs/pageserver_api/src/models.rs

@@ -5,10 +5,10 @@ pub mod utilization;
 pub use utilization::PageserverUtilization;
 
 use std::{
-    borrow::Cow,
     collections::HashMap,
     io::{BufRead, Read},
     num::{NonZeroU64, NonZeroUsize},
+    str::FromStr,
     sync::atomic::AtomicUsize,
     time::{Duration, SystemTime},
 };
@@ -19,7 +19,6 @@ use serde::{Deserialize, Serialize};
 use serde_with::serde_as;
 use utils::{
     completion,
-    history_buffer::HistoryBufferWithDropCounter,
     id::{NodeId, TenantId, TimelineId},
     lsn::Lsn,
     serde_system_time,
@@ -228,6 +227,11 @@ pub struct TimelineCreateRequest {
     pub pg_version: Option<u32>,
 }
 
+#[derive(Serialize, Deserialize, Clone)]
+pub struct LsnLeaseRequest {
+    pub lsn: Lsn,
+}
+
 #[derive(Serialize, Deserialize)]
 pub struct TenantShardSplitRequest {
     pub new_shard_count: u8,
@@ -288,7 +292,6 @@ pub struct TenantConfig {
     pub walreceiver_connect_timeout: Option<String>,
     pub lagging_wal_timeout: Option<String>,
     pub max_lsn_wal_lag: Option<NonZeroU64>,
-    pub trace_read_requests: Option<bool>,
     pub eviction_policy: Option<EvictionPolicy>,
     pub min_resident_size_override: Option<u64>,
     pub evictions_low_residence_duration_metric_threshold: Option<String>,
@@ -432,6 +435,41 @@ pub enum CompactionAlgorithm {
     Tiered,
 }
 
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
+pub enum ImageCompressionAlgorithm {
+    // Disabled for writes, support decompressing during read path
+    Disabled,
+    /// Zstandard compression. Level 0 means and None mean the same (default level). Levels can be negative as well.
+    /// For details, see the [manual](http://facebook.github.io/zstd/zstd_manual.html).
+    Zstd {
+        level: Option<i8>,
+    },
+}
+
+impl FromStr for ImageCompressionAlgorithm {
+    type Err = anyhow::Error;
+    fn from_str(s: &str) -> Result<Self, Self::Err> {
+        let mut components = s.split(['(', ')']);
+        let first = components
+            .next()
+            .ok_or_else(|| anyhow::anyhow!("empty string"))?;
+        match first {
+            "disabled" => Ok(ImageCompressionAlgorithm::Disabled),
+            "zstd" => {
+                let level = if let Some(v) = components.next() {
+                    let v: i8 = v.parse()?;
+                    Some(v)
+                } else {
+                    None
+                };
+
+                Ok(ImageCompressionAlgorithm::Zstd { level })
+            }
+            _ => anyhow::bail!("invalid specifier '{first}'"),
+        }
+    }
+}
+
 #[derive(Eq, PartialEq, Debug, Clone, Serialize, Deserialize)]
 pub struct CompactionAlgorithmSettings {
     pub kind: CompactionAlgorithm,
@@ -599,6 +637,13 @@ pub struct TenantInfo {
     pub current_physical_size: Option<u64>, // physical size is only included in `tenant_status` endpoint
     pub attachment_status: TenantAttachmentStatus,
     pub generation: u32,
+
+    /// Opaque explanation if gc is being blocked.
+    ///
+    /// Only looked up for the individual tenant detail, not the listing. This is purely for
+    /// debugging, not included in openapi.
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub gc_blocking: Option<String>,
 }
 
 #[derive(Serialize, Deserialize, Clone)]
@@ -611,6 +656,17 @@ pub struct TenantDetails {
     pub timelines: Vec<TimelineId>,
 }
 
+#[derive(Serialize, Deserialize, PartialEq, Eq, Clone, Copy, Debug)]
+pub enum TimelineArchivalState {
+    Archived,
+    Unarchived,
+}
+
+#[derive(Serialize, Deserialize, PartialEq, Eq, Clone)]
+pub struct TimelineArchivalConfigRequest {
+    pub state: TimelineArchivalState,
+}
+
 /// This represents the output of the "timeline_detail" and "timeline_list" API calls.
 #[derive(Debug, Serialize, Deserialize, Clone)]
 pub struct TimelineInfo {
@@ -643,6 +699,16 @@ pub struct TimelineInfo {
     pub current_physical_size: Option<u64>, // is None when timeline is Unloaded
     pub current_logical_size_non_incremental: Option<u64>,
 
+    /// How many bytes of WAL are within this branch's pitr_interval.  If the pitr_interval goes
+    /// beyond the branch's branch point, we only count up to the branch point.
+    pub pitr_history_size: u64,
+
+    /// Whether this branch's branch point is within its ancestor's PITR interval (i.e. any
+    /// ancestor data used by this branch would have been retained anyway).  If this is false, then
+    /// this branch may be imposing a cost on the ancestor by causing it to retain layers that it would
+    /// otherwise be able to GC.
+    pub within_ancestor_pitr: bool,
+
     pub timeline_dir_layer_file_size_sum: Option<u64>,
 
     pub wal_source_connstr: Option<String>,
@@ -665,58 +731,7 @@ pub struct LayerMapInfo {
     pub historic_layers: Vec<HistoricLayerInfo>,
 }
 
-#[derive(Debug, Hash, PartialEq, Eq, Clone, Copy, Serialize, Deserialize, enum_map::Enum)]
-#[repr(usize)]
-pub enum LayerAccessKind {
-    GetValueReconstructData,
-    Iter,
-    KeyIter,
-    Dump,
-}
-
-#[derive(Debug, Clone, Serialize, Deserialize)]
-pub struct LayerAccessStatFullDetails {
-    pub when_millis_since_epoch: u64,
-    pub task_kind: Cow<'static, str>,
-    pub access_kind: LayerAccessKind,
-}
-
-/// An event that impacts the layer's residence status.
-#[serde_as]
-#[derive(Debug, Clone, Serialize, Deserialize)]
-pub struct LayerResidenceEvent {
-    /// The time when the event occurred.
-    /// NB: this timestamp is captured while the residence status changes.
-    /// So, it might be behind/ahead of the actual residence change by a short amount of time.
-    ///
-    #[serde(rename = "timestamp_millis_since_epoch")]
-    #[serde_as(as = "serde_with::TimestampMilliSeconds")]
-    pub timestamp: SystemTime,
-    /// The new residence status of the layer.
-    pub status: LayerResidenceStatus,
-    /// The reason why we had to record this event.
-    pub reason: LayerResidenceEventReason,
-}
-
-/// The reason for recording a given [`LayerResidenceEvent`].
-#[derive(Debug, Clone, Copy, Serialize, Deserialize)]
-pub enum LayerResidenceEventReason {
-    /// The layer map is being populated, e.g. during timeline load or attach.
-    /// This includes [`RemoteLayer`] objects created in [`reconcile_with_remote`].
-    /// We need to record such events because there is no persistent storage for the events.
-    ///
-    // https://github.com/rust-lang/rust/issues/74481
-    /// [`RemoteLayer`]: ../../tenant/storage_layer/struct.RemoteLayer.html
-    /// [`reconcile_with_remote`]: ../../tenant/struct.Timeline.html#method.reconcile_with_remote
-    LayerLoad,
-    /// We just created the layer (e.g., freeze_and_flush or compaction).
-    /// Such layers are always [`LayerResidenceStatus::Resident`].
-    LayerCreate,
-    /// We on-demand downloaded or evicted the given layer.
-    ResidenceChange,
-}
-
-/// The residence status of the layer, after the given [`LayerResidenceEvent`].
+/// The residence status of a layer
 #[derive(Debug, Clone, Copy, Serialize, Deserialize)]
 pub enum LayerResidenceStatus {
     /// Residence status for a layer file that exists locally.
@@ -726,23 +741,16 @@ pub enum LayerResidenceStatus {
     Evicted,
 }
 
-impl LayerResidenceEvent {
-    pub fn new(status: LayerResidenceStatus, reason: LayerResidenceEventReason) -> Self {
-        Self {
-            status,
-            reason,
-            timestamp: SystemTime::now(),
-        }
-    }
-}
-
+#[serde_as]
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct LayerAccessStats {
-    pub access_count_by_access_kind: HashMap<LayerAccessKind, u64>,
-    pub task_kind_access_flag: Vec<Cow<'static, str>>,
-    pub first: Option<LayerAccessStatFullDetails>,
-    pub accesses_history: HistoryBufferWithDropCounter<LayerAccessStatFullDetails, 16>,
-    pub residence_events_history: HistoryBufferWithDropCounter<LayerResidenceEvent, 16>,
+    #[serde_as(as = "serde_with::TimestampMilliSeconds")]
+    pub access_time: SystemTime,
+
+    #[serde_as(as = "serde_with::TimestampMilliSeconds")]
+    pub residence_time: SystemTime,
+
+    pub visible: bool,
 }
 
 #[derive(Debug, Clone, Serialize, Deserialize)]
@@ -939,6 +947,8 @@ pub struct TopTenantShardsResponse {
 }
 
 pub mod virtual_file {
+    use std::path::PathBuf;
+
     #[derive(
         Copy,
         Clone,
@@ -957,6 +967,53 @@ pub mod virtual_file {
         #[cfg(target_os = "linux")]
         TokioEpollUring,
     }
+
+    /// Direct IO modes for a pageserver.
+    #[derive(Debug, PartialEq, Eq, Clone, serde::Deserialize, serde::Serialize, Default)]
+    #[serde(tag = "mode", rename_all = "kebab-case", deny_unknown_fields)]
+    pub enum DirectIoMode {
+        /// Direct IO disabled (uses usual buffered IO).
+        #[default]
+        Disabled,
+        /// Direct IO disabled (performs checks and perf simulations).
+        Evaluate {
+            /// Alignment check level
+            alignment_check: DirectIoAlignmentCheckLevel,
+            /// Latency padded for performance simulation.
+            latency_padding: DirectIoLatencyPadding,
+        },
+        /// Direct IO enabled.
+        Enabled {
+            /// Actions to perform on alignment error.
+            on_alignment_error: DirectIoOnAlignmentErrorAction,
+        },
+    }
+
+    #[derive(Debug, PartialEq, Eq, Clone, serde::Deserialize, serde::Serialize, Default)]
+    #[serde(rename_all = "kebab-case")]
+    pub enum DirectIoAlignmentCheckLevel {
+        #[default]
+        Error,
+        Log,
+        None,
+    }
+
+    #[derive(Debug, PartialEq, Eq, Clone, serde::Deserialize, serde::Serialize, Default)]
+    #[serde(rename_all = "kebab-case")]
+    pub enum DirectIoOnAlignmentErrorAction {
+        Error,
+        #[default]
+        FallbackToBuffered,
+    }
+
+    #[derive(Debug, PartialEq, Eq, Clone, serde::Deserialize, serde::Serialize, Default)]
+    #[serde(tag = "type", rename_all = "kebab-case")]
+    pub enum DirectIoLatencyPadding {
+        /// Pad virtual file operations with IO to a fake file.
+        FakeFileRW { path: PathBuf },
+        #[default]
+        None,
+    }
 }
 
 // Wrapped in libpq CopyData
@@ -1426,6 +1483,7 @@ mod tests {
             current_physical_size: Some(42),
             attachment_status: TenantAttachmentStatus::Attached,
             generation: 1,
+            gc_blocking: None,
         };
         let expected_active = json!({
             "id": original_active.id.to_string(),
@@ -1448,6 +1506,7 @@ mod tests {
             current_physical_size: Some(42),
             attachment_status: TenantAttachmentStatus::Attached,
             generation: 1,
+            gc_blocking: None,
         };
         let expected_broken = json!({
             "id": original_broken.id.to_string(),
@@ -1614,4 +1673,25 @@ mod tests {
             AuxFilePolicy::CrossValidation
         );
     }
+
+    #[test]
+    fn test_image_compression_algorithm_parsing() {
+        use ImageCompressionAlgorithm::*;
+        assert_eq!(
+            ImageCompressionAlgorithm::from_str("disabled").unwrap(),
+            Disabled
+        );
+        assert_eq!(
+            ImageCompressionAlgorithm::from_str("zstd").unwrap(),
+            Zstd { level: None }
+        );
+        assert_eq!(
+            ImageCompressionAlgorithm::from_str("zstd(18)").unwrap(),
+            Zstd { level: Some(18) }
+        );
+        assert_eq!(
+            ImageCompressionAlgorithm::from_str("zstd(-3)").unwrap(),
+            Zstd { level: Some(-3) }
+        );
+    }
 }

libs/pageserver_api/src/models/detach_ancestor.rs

@@ -1,6 +1,8 @@
+use std::collections::HashSet;
+
 use utils::id::TimelineId;
 
-#[derive(Default, serde::Serialize)]
+#[derive(Debug, Default, PartialEq, serde::Serialize, serde::Deserialize)]
 pub struct AncestorDetached {
-    pub reparented_timelines: Vec<TimelineId>,
+    pub reparented_timelines: HashSet<TimelineId>,
 }

libs/pageserver_api/src/shard.rs

@@ -1,59 +1,42 @@
-use std::{ops::RangeInclusive, str::FromStr};
+//! See docs/rfcs/031-sharding-static.md for an overview of sharding.
+//!
+//! This module contains a variety of types used to represent the concept of sharding
+//! a Neon tenant across multiple physical shards.  Since there are quite a few of these,
+//! we provide an summary here.
+//!
+//! Types used to describe shards:
+//! - [`ShardCount`] describes how many shards make up a tenant, plus the magic `unsharded` value
+//!   which identifies a tenant which is not shard-aware.  This means its storage paths do not include
+//!   a shard suffix.
+//! - [`ShardNumber`] is simply the zero-based index of a shard within a tenant.
+//! - [`ShardIndex`] is the 2-tuple of `ShardCount` and `ShardNumber`, it's just like a `TenantShardId`
+//!   without the tenant ID.  This is useful for things that are implicitly scoped to a particular
+//!   tenant, such as layer files.
+//! - [`ShardIdentity`]` is the full description of a particular shard's parameters, in sufficient
+//!   detail to convert a [`Key`] to a [`ShardNumber`] when deciding where to write/read.
+//! - The [`ShardSlug`] is a terse formatter for ShardCount and ShardNumber, written as
+//!   four hex digits.  An unsharded tenant is `0000`.
+//! - [`TenantShardId`] is the unique ID of a particular shard within a particular tenant
+//!
+//! Types used to describe the parameters for data distribution in a sharded tenant:
+//! - [`ShardStripeSize`] controls how long contiguous runs of [`Key`]s (stripes) are when distributed across
+//!   multiple shards.  Its value is given in 8kiB pages.
+//! - [`ShardLayout`] describes the data distribution scheme, and at time of writing is
+//!   always zero: this is provided for future upgrades that might introduce different
+//!   data distribution schemes.
+//!
+//! Examples:
+//! - A legacy unsharded tenant has one shard with ShardCount(0), ShardNumber(0), and its slug is 0000
+//! - A single sharded tenant has one shard with ShardCount(1), ShardNumber(0), and its slug is 0001
+//! - In a tenant with 4 shards, each shard has ShardCount(N), ShardNumber(i) where i in 0..N-1 (inclusive),
+//!   and their slugs are 0004, 0104, 0204, and 0304.
 
 use crate::{key::Key, models::ShardParameters};
-use hex::FromHex;
 use postgres_ffi::relfile_utils::INIT_FORKNUM;
 use serde::{Deserialize, Serialize};
-use utils::id::TenantId;
 
-/// See docs/rfcs/031-sharding-static.md for an overview of sharding.
-///
-/// This module contains a variety of types used to represent the concept of sharding
-/// a Neon tenant across multiple physical shards.  Since there are quite a few of these,
-/// we provide an summary here.
-///
-/// Types used to describe shards:
-/// - [`ShardCount`] describes how many shards make up a tenant, plus the magic `unsharded` value
-///   which identifies a tenant which is not shard-aware.  This means its storage paths do not include
-///   a shard suffix.
-/// - [`ShardNumber`] is simply the zero-based index of a shard within a tenant.
-/// - [`ShardIndex`] is the 2-tuple of `ShardCount` and `ShardNumber`, it's just like a `TenantShardId`
-///   without the tenant ID.  This is useful for things that are implicitly scoped to a particular
-///   tenant, such as layer files.
-/// - [`ShardIdentity`]` is the full description of a particular shard's parameters, in sufficient
-///   detail to convert a [`Key`] to a [`ShardNumber`] when deciding where to write/read.
-/// - The [`ShardSlug`] is a terse formatter for ShardCount and ShardNumber, written as
-///   four hex digits.  An unsharded tenant is `0000`.
-/// - [`TenantShardId`] is the unique ID of a particular shard within a particular tenant
-///
-/// Types used to describe the parameters for data distribution in a sharded tenant:
-/// - [`ShardStripeSize`] controls how long contiguous runs of [`Key`]s (stripes) are when distributed across
-///   multiple shards.  Its value is given in 8kiB pages.
-/// - [`ShardLayout`] describes the data distribution scheme, and at time of writing is
-///   always zero: this is provided for future upgrades that might introduce different
-///   data distribution schemes.
-///
-/// Examples:
-/// - A legacy unsharded tenant has one shard with ShardCount(0), ShardNumber(0), and its slug is 0000
-/// - A single sharded tenant has one shard with ShardCount(1), ShardNumber(0), and its slug is 0001
-/// - In a tenant with 4 shards, each shard has ShardCount(N), ShardNumber(i) where i in 0..N-1 (inclusive),
-///   and their slugs are 0004, 0104, 0204, and 0304.
-
-#[derive(Ord, PartialOrd, Eq, PartialEq, Clone, Copy, Serialize, Deserialize, Debug, Hash)]
-pub struct ShardNumber(pub u8);
-
-#[derive(Ord, PartialOrd, Eq, PartialEq, Clone, Copy, Serialize, Deserialize, Debug, Hash)]
-pub struct ShardCount(u8);
-
-/// Combination of ShardNumber and ShardCount.  For use within the context of a particular tenant,
-/// when we need to know which shard we're dealing with, but do not need to know the full
-/// ShardIdentity (because we won't be doing any page->shard mapping), and do not need to know
-/// the fully qualified TenantShardId.
-#[derive(Eq, PartialEq, PartialOrd, Ord, Clone, Copy, Hash)]
-pub struct ShardIndex {
-    pub shard_number: ShardNumber,
-    pub shard_count: ShardCount,
-}
+#[doc(inline)]
+pub use ::utils::shard::*;
 
 /// The ShardIdentity contains enough information to map a [`Key`] to a [`ShardNumber`],
 /// and to check whether that [`ShardNumber`] is the same as the current shard.
@@ -65,362 +48,6 @@ pub struct ShardIdentity {
     layout: ShardLayout,
 }
 
-/// Formatting helper, for generating the `shard_id` label in traces.
-struct ShardSlug<'a>(&'a TenantShardId);
-
-/// TenantShardId globally identifies a particular shard in a particular tenant.
-///
-/// These are written as `<TenantId>-<ShardSlug>`, for example:
-///   # The second shard in a two-shard tenant
-///   072f1291a5310026820b2fe4b2968934-0102
-///
-/// If the `ShardCount` is _unsharded_, the `TenantShardId` is written without
-/// a shard suffix and is equivalent to the encoding of a `TenantId`: this enables
-/// an unsharded [`TenantShardId`] to be used interchangably with a [`TenantId`].
-///
-/// The human-readable encoding of an unsharded TenantShardId, such as used in API URLs,
-/// is both forward and backward compatible with TenantId: a legacy TenantId can be
-/// decoded as a TenantShardId, and when re-encoded it will be parseable
-/// as a TenantId.
-#[derive(Eq, PartialEq, PartialOrd, Ord, Clone, Copy, Hash)]
-pub struct TenantShardId {
-    pub tenant_id: TenantId,
-    pub shard_number: ShardNumber,
-    pub shard_count: ShardCount,
-}
-
-impl ShardCount {
-    pub const MAX: Self = Self(u8::MAX);
-
-    /// The internal value of a ShardCount may be zero, which means "1 shard, but use
-    /// legacy format for TenantShardId that excludes the shard suffix", also known
-    /// as [`TenantShardId::unsharded`].
-    ///
-    /// This method returns the actual number of shards, i.e. if our internal value is
-    /// zero, we return 1 (unsharded tenants have 1 shard).
-    pub fn count(&self) -> u8 {
-        if self.0 > 0 {
-            self.0
-        } else {
-            1
-        }
-    }
-
-    /// The literal internal value: this is **not** the number of shards in the
-    /// tenant, as we have a special zero value for legacy unsharded tenants.  Use
-    /// [`Self::count`] if you want to know the cardinality of shards.
-    pub fn literal(&self) -> u8 {
-        self.0
-    }
-
-    /// Whether the `ShardCount` is for an unsharded tenant, so uses one shard but
-    /// uses the legacy format for `TenantShardId`. See also the documentation for
-    /// [`Self::count`].
-    pub fn is_unsharded(&self) -> bool {
-        self.0 == 0
-    }
-
-    /// `v` may be zero, or the number of shards in the tenant.  `v` is what
-    /// [`Self::literal`] would return.
-    pub const fn new(val: u8) -> Self {
-        Self(val)
-    }
-}
-
-impl ShardNumber {
-    pub const MAX: Self = Self(u8::MAX);
-}
-
-impl TenantShardId {
-    pub fn unsharded(tenant_id: TenantId) -> Self {
-        Self {
-            tenant_id,
-            shard_number: ShardNumber(0),
-            shard_count: ShardCount(0),
-        }
-    }
-
-    /// The range of all TenantShardId that belong to a particular TenantId.  This is useful when
-    /// you have a BTreeMap of TenantShardId, and are querying by TenantId.
-    pub fn tenant_range(tenant_id: TenantId) -> RangeInclusive<Self> {
-        RangeInclusive::new(
-            Self {
-                tenant_id,
-                shard_number: ShardNumber(0),
-                shard_count: ShardCount(0),
-            },
-            Self {
-                tenant_id,
-                shard_number: ShardNumber::MAX,
-                shard_count: ShardCount::MAX,
-            },
-        )
-    }
-
-    pub fn shard_slug(&self) -> impl std::fmt::Display + '_ {
-        ShardSlug(self)
-    }
-
-    /// Convenience for code that has special behavior on the 0th shard.
-    pub fn is_shard_zero(&self) -> bool {
-        self.shard_number == ShardNumber(0)
-    }
-
-    /// The "unsharded" value is distinct from simply having a single shard: it represents
-    /// a tenant which is not shard-aware at all, and whose storage paths will not include
-    /// a shard suffix.
-    pub fn is_unsharded(&self) -> bool {
-        self.shard_number == ShardNumber(0) && self.shard_count.is_unsharded()
-    }
-
-    /// Convenience for dropping the tenant_id and just getting the ShardIndex: this
-    /// is useful when logging from code that is already in a span that includes tenant ID, to
-    /// keep messages reasonably terse.
-    pub fn to_index(&self) -> ShardIndex {
-        ShardIndex {
-            shard_number: self.shard_number,
-            shard_count: self.shard_count,
-        }
-    }
-
-    /// Calculate the children of this TenantShardId when splitting the overall tenant into
-    /// the given number of shards.
-    pub fn split(&self, new_shard_count: ShardCount) -> Vec<TenantShardId> {
-        let effective_old_shard_count = std::cmp::max(self.shard_count.0, 1);
-        let mut child_shards = Vec::new();
-        for shard_number in 0..ShardNumber(new_shard_count.0).0 {
-            // Key mapping is based on a round robin mapping of key hash modulo shard count,
-            // so our child shards are the ones which the same keys would map to.
-            if shard_number % effective_old_shard_count == self.shard_number.0 {
-                child_shards.push(TenantShardId {
-                    tenant_id: self.tenant_id,
-                    shard_number: ShardNumber(shard_number),
-                    shard_count: new_shard_count,
-                })
-            }
-        }
-
-        child_shards
-    }
-}
-
-impl<'a> std::fmt::Display for ShardSlug<'a> {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        write!(
-            f,
-            "{:02x}{:02x}",
-            self.0.shard_number.0, self.0.shard_count.0
-        )
-    }
-}
-
-impl std::fmt::Display for TenantShardId {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        if self.shard_count != ShardCount(0) {
-            write!(f, "{}-{}", self.tenant_id, self.shard_slug())
-        } else {
-            // Legacy case (shard_count == 0) -- format as just the tenant id.  Note that this
-            // is distinct from the normal single shard case (shard count == 1).
-            self.tenant_id.fmt(f)
-        }
-    }
-}
-
-impl std::fmt::Debug for TenantShardId {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        // Debug is the same as Display: the compact hex representation
-        write!(f, "{}", self)
-    }
-}
-
-impl std::str::FromStr for TenantShardId {
-    type Err = hex::FromHexError;
-
-    fn from_str(s: &str) -> Result<Self, Self::Err> {
-        // Expect format: 16 byte TenantId, '-', 1 byte shard number, 1 byte shard count
-        if s.len() == 32 {
-            // Legacy case: no shard specified
-            Ok(Self {
-                tenant_id: TenantId::from_str(s)?,
-                shard_number: ShardNumber(0),
-                shard_count: ShardCount(0),
-            })
-        } else if s.len() == 37 {
-            let bytes = s.as_bytes();
-            let tenant_id = TenantId::from_hex(&bytes[0..32])?;
-            let mut shard_parts: [u8; 2] = [0u8; 2];
-            hex::decode_to_slice(&bytes[33..37], &mut shard_parts)?;
-            Ok(Self {
-                tenant_id,
-                shard_number: ShardNumber(shard_parts[0]),
-                shard_count: ShardCount(shard_parts[1]),
-            })
-        } else {
-            Err(hex::FromHexError::InvalidStringLength)
-        }
-    }
-}
-
-impl From<[u8; 18]> for TenantShardId {
-    fn from(b: [u8; 18]) -> Self {
-        let tenant_id_bytes: [u8; 16] = b[0..16].try_into().unwrap();
-
-        Self {
-            tenant_id: TenantId::from(tenant_id_bytes),
-            shard_number: ShardNumber(b[16]),
-            shard_count: ShardCount(b[17]),
-        }
-    }
-}
-
-impl ShardIndex {
-    pub fn new(number: ShardNumber, count: ShardCount) -> Self {
-        Self {
-            shard_number: number,
-            shard_count: count,
-        }
-    }
-    pub fn unsharded() -> Self {
-        Self {
-            shard_number: ShardNumber(0),
-            shard_count: ShardCount(0),
-        }
-    }
-
-    /// The "unsharded" value is distinct from simply having a single shard: it represents
-    /// a tenant which is not shard-aware at all, and whose storage paths will not include
-    /// a shard suffix.
-    pub fn is_unsharded(&self) -> bool {
-        self.shard_number == ShardNumber(0) && self.shard_count == ShardCount(0)
-    }
-
-    /// For use in constructing remote storage paths: concatenate this with a TenantId
-    /// to get a fully qualified TenantShardId.
-    ///
-    /// Backward compat: this function returns an empty string if Self::is_unsharded, such
-    /// that the legacy pre-sharding remote key format is preserved.
-    pub fn get_suffix(&self) -> String {
-        if self.is_unsharded() {
-            "".to_string()
-        } else {
-            format!("-{:02x}{:02x}", self.shard_number.0, self.shard_count.0)
-        }
-    }
-}
-
-impl std::fmt::Display for ShardIndex {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        write!(f, "{:02x}{:02x}", self.shard_number.0, self.shard_count.0)
-    }
-}
-
-impl std::fmt::Debug for ShardIndex {
-    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        // Debug is the same as Display: the compact hex representation
-        write!(f, "{}", self)
-    }
-}
-
-impl std::str::FromStr for ShardIndex {
-    type Err = hex::FromHexError;
-
-    fn from_str(s: &str) -> Result<Self, Self::Err> {
-        // Expect format: 1 byte shard number, 1 byte shard count
-        if s.len() == 4 {
-            let bytes = s.as_bytes();
-            let mut shard_parts: [u8; 2] = [0u8; 2];
-            hex::decode_to_slice(bytes, &mut shard_parts)?;
-            Ok(Self {
-                shard_number: ShardNumber(shard_parts[0]),
-                shard_count: ShardCount(shard_parts[1]),
-            })
-        } else {
-            Err(hex::FromHexError::InvalidStringLength)
-        }
-    }
-}
-
-impl From<[u8; 2]> for ShardIndex {
-    fn from(b: [u8; 2]) -> Self {
-        Self {
-            shard_number: ShardNumber(b[0]),
-            shard_count: ShardCount(b[1]),
-        }
-    }
-}
-
-impl Serialize for TenantShardId {
-    fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
-    where
-        S: serde::Serializer,
-    {
-        if serializer.is_human_readable() {
-            serializer.collect_str(self)
-        } else {
-            // Note: while human encoding of [`TenantShardId`] is backward and forward
-            // compatible, this binary encoding is not.
-            let mut packed: [u8; 18] = [0; 18];
-            packed[0..16].clone_from_slice(&self.tenant_id.as_arr());
-            packed[16] = self.shard_number.0;
-            packed[17] = self.shard_count.0;
-
-            packed.serialize(serializer)
-        }
-    }
-}
-
-impl<'de> Deserialize<'de> for TenantShardId {
-    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
-    where
-        D: serde::Deserializer<'de>,
-    {
-        struct IdVisitor {
-            is_human_readable_deserializer: bool,
-        }
-
-        impl<'de> serde::de::Visitor<'de> for IdVisitor {
-            type Value = TenantShardId;
-
-            fn expecting(&self, formatter: &mut std::fmt::Formatter) -> std::fmt::Result {
-                if self.is_human_readable_deserializer {
-                    formatter.write_str("value in form of hex string")
-                } else {
-                    formatter.write_str("value in form of integer array([u8; 18])")
-                }
-            }
-
-            fn visit_seq<A>(self, seq: A) -> Result<Self::Value, A::Error>
-            where
-                A: serde::de::SeqAccess<'de>,
-            {
-                let s = serde::de::value::SeqAccessDeserializer::new(seq);
-                let id: [u8; 18] = Deserialize::deserialize(s)?;
-                Ok(TenantShardId::from(id))
-            }
-
-            fn visit_str<E>(self, v: &str) -> Result<Self::Value, E>
-            where
-                E: serde::de::Error,
-            {
-                TenantShardId::from_str(v).map_err(E::custom)
-            }
-        }
-
-        if deserializer.is_human_readable() {
-            deserializer.deserialize_str(IdVisitor {
-                is_human_readable_deserializer: true,
-            })
-        } else {
-            deserializer.deserialize_tuple(
-                18,
-                IdVisitor {
-                    is_human_readable_deserializer: false,
-                },
-            )
-        }
-    }
-}
-
 /// Stripe size in number of pages
 #[derive(Clone, Copy, Serialize, Deserialize, Eq, PartialEq, Debug)]
 pub struct ShardStripeSize(pub u32);
@@ -585,77 +212,6 @@ impl ShardIdentity {
     }
 }
 
-impl Serialize for ShardIndex {
-    fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
-    where
-        S: serde::Serializer,
-    {
-        if serializer.is_human_readable() {
-            serializer.collect_str(self)
-        } else {
-            // Binary encoding is not used in index_part.json, but is included in anticipation of
-            // switching various structures (e.g. inter-process communication, remote metadata) to more
-            // compact binary encodings in future.
-            let mut packed: [u8; 2] = [0; 2];
-            packed[0] = self.shard_number.0;
-            packed[1] = self.shard_count.0;
-            packed.serialize(serializer)
-        }
-    }
-}
-
-impl<'de> Deserialize<'de> for ShardIndex {
-    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
-    where
-        D: serde::Deserializer<'de>,
-    {
-        struct IdVisitor {
-            is_human_readable_deserializer: bool,
-        }
-
-        impl<'de> serde::de::Visitor<'de> for IdVisitor {
-            type Value = ShardIndex;
-
-            fn expecting(&self, formatter: &mut std::fmt::Formatter) -> std::fmt::Result {
-                if self.is_human_readable_deserializer {
-                    formatter.write_str("value in form of hex string")
-                } else {
-                    formatter.write_str("value in form of integer array([u8; 2])")
-                }
-            }
-
-            fn visit_seq<A>(self, seq: A) -> Result<Self::Value, A::Error>
-            where
-                A: serde::de::SeqAccess<'de>,
-            {
-                let s = serde::de::value::SeqAccessDeserializer::new(seq);
-                let id: [u8; 2] = Deserialize::deserialize(s)?;
-                Ok(ShardIndex::from(id))
-            }
-
-            fn visit_str<E>(self, v: &str) -> Result<Self::Value, E>
-            where
-                E: serde::de::Error,
-            {
-                ShardIndex::from_str(v).map_err(E::custom)
-            }
-        }
-
-        if deserializer.is_human_readable() {
-            deserializer.deserialize_str(IdVisitor {
-                is_human_readable_deserializer: true,
-            })
-        } else {
-            deserializer.deserialize_tuple(
-                2,
-                IdVisitor {
-                    is_human_readable_deserializer: false,
-                },
-            )
-        }
-    }
-}
-
 /// Whether this key is always held on shard 0 (e.g. shard 0 holds all SLRU keys
 /// in order to be able to serve basebackup requests without peer communication).
 fn key_is_shard0(key: &Key) -> bool {
@@ -737,7 +293,9 @@ pub fn describe(
 
 #[cfg(test)]
 mod tests {
-    use utils::Hex;
+    use std::str::FromStr;
+
+    use utils::{id::TenantId, Hex};
 
     use super::*;
 

libs/postgres_backend/Cargo.toml

@@ -13,6 +13,7 @@ rustls.workspace = true
 serde.workspace = true
 thiserror.workspace = true
 tokio.workspace = true
+tokio-util.workspace = true
 tokio-rustls.workspace = true
 tracing.workspace = true
 
@@ -23,4 +24,4 @@ workspace_hack.workspace = true
 once_cell.workspace = true
 rustls-pemfile.workspace = true
 tokio-postgres.workspace = true
-tokio-postgres-rustls.workspace = true
+tokio-postgres-rustls.workspace = true

libs/postgres_backend/src/lib.rs

@@ -16,6 +16,7 @@ use std::{fmt, io};
 use std::{future::Future, str::FromStr};
 use tokio::io::{AsyncRead, AsyncWrite};
 use tokio_rustls::TlsAcceptor;
+use tokio_util::sync::CancellationToken;
 use tracing::{debug, error, info, trace, warn};
 
 use pq_proto::framed::{ConnectionError, Framed, FramedReader, FramedWriter};
@@ -400,21 +401,15 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> PostgresBackend<IO> {
     }
 
     /// Wrapper for run_message_loop() that shuts down socket when we are done
-    pub async fn run<F, S>(
+    pub async fn run(
         mut self,
         handler: &mut impl Handler<IO>,
-        shutdown_watcher: F,
-    ) -> Result<(), QueryError>
-    where
-        F: Fn() -> S + Clone,
-        S: Future,
-    {
-        let ret = self
-            .run_message_loop(handler, shutdown_watcher.clone())
-            .await;
+        cancel: &CancellationToken,
+    ) -> Result<(), QueryError> {
+        let ret = self.run_message_loop(handler, cancel).await;
 
         tokio::select! {
-            _ = shutdown_watcher() => {
+            _ = cancel.cancelled() => {
                 // do nothing; we most likely got already stopped by shutdown and will log it next.
             }
             _ = self.framed.shutdown() => {
@@ -444,21 +439,17 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> PostgresBackend<IO> {
         }
     }
 
-    async fn run_message_loop<F, S>(
+    async fn run_message_loop(
         &mut self,
         handler: &mut impl Handler<IO>,
-        shutdown_watcher: F,
-    ) -> Result<(), QueryError>
-    where
-        F: Fn() -> S,
-        S: Future,
-    {
+        cancel: &CancellationToken,
+    ) -> Result<(), QueryError> {
         trace!("postgres backend to {:?} started", self.peer_addr);
 
         tokio::select!(
             biased;
 
-            _ = shutdown_watcher() => {
+            _ = cancel.cancelled() => {
                 // We were requested to shut down.
                 tracing::info!("shutdown request received during handshake");
                 return Err(QueryError::Shutdown)
@@ -473,7 +464,7 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> PostgresBackend<IO> {
         let mut query_string = Bytes::new();
         while let Some(msg) = tokio::select!(
             biased;
-            _ = shutdown_watcher() => {
+            _ = cancel.cancelled() => {
                 // We were requested to shut down.
                 tracing::info!("shutdown request received in run_message_loop");
                 return Err(QueryError::Shutdown)
@@ -485,7 +476,7 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> PostgresBackend<IO> {
             let result = self.process_message(handler, msg, &mut query_string).await;
             tokio::select!(
                 biased;
-                _ = shutdown_watcher() => {
+                _ = cancel.cancelled() => {
                     // We were requested to shut down.
                     tracing::info!("shutdown request received during response flush");
 
@@ -672,11 +663,17 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> PostgresBackend<IO> {
         assert!(self.state < ProtoState::Authentication);
         let have_tls = self.tls_config.is_some();
         match msg {
-            FeStartupPacket::SslRequest => {
+            FeStartupPacket::SslRequest { direct } => {
                 debug!("SSL requested");
 
-                self.write_message(&BeMessage::EncryptionResponse(have_tls))
-                    .await?;
+                if !direct {
+                    self.write_message(&BeMessage::EncryptionResponse(have_tls))
+                        .await?;
+                } else if !have_tls {
+                    return Err(QueryError::Other(anyhow::anyhow!(
+                        "direct SSL negotiation but no TLS support"
+                    )));
+                }
 
                 if have_tls {
                     self.start_tls().await?;

libs/postgres_backend/tests/simple_select.rs

@@ -3,13 +3,14 @@ use once_cell::sync::Lazy;
 use postgres_backend::{AuthType, Handler, PostgresBackend, QueryError};
 use pq_proto::{BeMessage, RowDescriptor};
 use std::io::Cursor;
-use std::{future, sync::Arc};
+use std::sync::Arc;
 use tokio::io::{AsyncRead, AsyncWrite};
 use tokio::net::{TcpListener, TcpStream};
 use tokio_postgres::config::SslMode;
 use tokio_postgres::tls::MakeTlsConnect;
 use tokio_postgres::{Config, NoTls, SimpleQueryMessage};
 use tokio_postgres_rustls::MakeRustlsConnect;
+use tokio_util::sync::CancellationToken;
 
 // generate client, server test streams
 async fn make_tcp_pair() -> (TcpStream, TcpStream) {
@@ -50,7 +51,7 @@ async fn simple_select() {
 
     tokio::spawn(async move {
         let mut handler = TestHandler {};
-        pgbackend.run(&mut handler, future::pending::<()>).await
+        pgbackend.run(&mut handler, &CancellationToken::new()).await
     });
 
     let conf = Config::new();
@@ -102,7 +103,7 @@ async fn simple_select_ssl() {
 
     tokio::spawn(async move {
         let mut handler = TestHandler {};
-        pgbackend.run(&mut handler, future::pending::<()>).await
+        pgbackend.run(&mut handler, &CancellationToken::new()).await
     });
 
     let client_cfg = rustls::ClientConfig::builder()

libs/postgres_connection/src/lib.rs

@@ -144,7 +144,20 @@ impl PgConnectionConfig {
             // implement and this function is hardly a bottleneck. The function is only called around
             // establishing a new connection.
             #[allow(unstable_name_collisions)]
-            config.options(&encode_options(&self.options));
+            config.options(
+                &self
+                    .options
+                    .iter()
+                    .map(|s| {
+                        if s.contains(['\\', ' ']) {
+                            Cow::Owned(s.replace('\\', "\\\\").replace(' ', "\\ "))
+                        } else {
+                            Cow::Borrowed(s.as_str())
+                        }
+                    })
+                    .intersperse(Cow::Borrowed(" ")) // TODO: use impl from std once it's stabilized
+                    .collect::<String>(),
+            );
         }
         config
     }
@@ -165,21 +178,6 @@ impl PgConnectionConfig {
     }
 }
 
-#[allow(unstable_name_collisions)]
-fn encode_options(options: &[String]) -> String {
-    options
-        .iter()
-        .map(|s| {
-            if s.contains(['\\', ' ']) {
-                Cow::Owned(s.replace('\\', "\\\\").replace(' ', "\\ "))
-            } else {
-                Cow::Borrowed(s.as_str())
-            }
-        })
-        .intersperse(Cow::Borrowed(" ")) // TODO: use impl from std once it's stabilized
-        .collect::<String>()
-}
-
 impl fmt::Display for PgConnectionConfig {
     fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
         // The password is intentionally hidden and not part of this display string.
@@ -208,7 +206,7 @@ impl fmt::Debug for PgConnectionConfig {
 
 #[cfg(test)]
 mod tests_pg_connection_config {
-    use crate::{encode_options, PgConnectionConfig};
+    use crate::PgConnectionConfig;
     use once_cell::sync::Lazy;
     use url::Host;
 
@@ -257,12 +255,18 @@ mod tests_pg_connection_config {
 
     #[test]
     fn test_with_options() {
-        let options = encode_options(&[
-            "hello".to_owned(),
-            "world".to_owned(),
-            "with space".to_owned(),
-            "and \\ backslashes".to_owned(),
+        let cfg = PgConnectionConfig::new_host_port(STUB_HOST.clone(), 123).extend_options([
+            "hello",
+            "world",
+            "with space",
+            "and \\ backslashes",
         ]);
-        assert_eq!(options, "hello world with\\ space and\\ \\\\\\ backslashes");
+        assert_eq!(cfg.host(), &*STUB_HOST);
+        assert_eq!(cfg.port(), 123);
+        assert_eq!(cfg.raw_address(), "stub.host.example:123");
+        assert_eq!(
+            cfg.to_tokio_postgres_config().get_options(),
+            Some("hello world with\\ space and\\ \\\\\\ backslashes")
+        );
     }
 }

libs/postgres_ffi/src/controlfile_utils.rs

@@ -29,7 +29,7 @@ use anyhow::{bail, Result};
 use bytes::{Bytes, BytesMut};
 
 /// Equivalent to sizeof(ControlFileData) in C
-const SIZEOF_CONTROLDATA: usize = std::mem::size_of::<ControlFileData>();
+const SIZEOF_CONTROLDATA: usize = size_of::<ControlFileData>();
 
 impl ControlFileData {
     /// Compute the offset of the `crc` field within the `ControlFileData` struct.

libs/postgres_ffi/src/pg_constants.rs

@@ -31,7 +31,7 @@ pub const SMGR_TRUNCATE_FSM: u32 = 0x0004;
 //
 
 // Assumes 8 byte alignment
-const SIZEOF_PAGE_HEADER_DATA: usize = std::mem::size_of::<PageHeaderData>();
+const SIZEOF_PAGE_HEADER_DATA: usize = size_of::<PageHeaderData>();
 pub const MAXALIGN_SIZE_OF_PAGE_HEADER_DATA: usize = (SIZEOF_PAGE_HEADER_DATA + 7) & !7;
 
 //
@@ -191,7 +191,7 @@ pub const XLR_RMGR_INFO_MASK: u8 = 0xF0;
 pub const XLOG_TBLSPC_CREATE: u8 = 0x00;
 pub const XLOG_TBLSPC_DROP: u8 = 0x10;
 
-pub const SIZEOF_XLOGRECORD: u32 = std::mem::size_of::<XLogRecord>() as u32;
+pub const SIZEOF_XLOGRECORD: u32 = size_of::<XLogRecord>() as u32;
 
 //
 // from xlogrecord.h

libs/postgres_ffi/src/xlog_utils.rs

@@ -42,9 +42,9 @@ pub const XLP_FIRST_IS_CONTRECORD: u16 = 0x0001;
 pub const XLP_REM_LEN_OFFS: usize = 2 + 2 + 4 + 8;
 pub const XLOG_RECORD_CRC_OFFS: usize = 4 + 4 + 8 + 1 + 1 + 2;
 
-pub const XLOG_SIZE_OF_XLOG_SHORT_PHD: usize = std::mem::size_of::<XLogPageHeaderData>();
-pub const XLOG_SIZE_OF_XLOG_LONG_PHD: usize = std::mem::size_of::<XLogLongPageHeaderData>();
-pub const XLOG_SIZE_OF_XLOG_RECORD: usize = std::mem::size_of::<XLogRecord>();
+pub const XLOG_SIZE_OF_XLOG_SHORT_PHD: usize = size_of::<XLogPageHeaderData>();
+pub const XLOG_SIZE_OF_XLOG_LONG_PHD: usize = size_of::<XLogLongPageHeaderData>();
+pub const XLOG_SIZE_OF_XLOG_RECORD: usize = size_of::<XLogRecord>();
 #[allow(clippy::identity_op)]
 pub const SIZE_OF_XLOG_RECORD_DATA_HEADER_SHORT: usize = 1 * 2;
 
@@ -311,7 +311,7 @@ impl XLogLongPageHeaderData {
     }
 }
 
-pub const SIZEOF_CHECKPOINT: usize = std::mem::size_of::<CheckPoint>();
+pub const SIZEOF_CHECKPOINT: usize = size_of::<CheckPoint>();
 
 impl CheckPoint {
     pub fn encode(&self) -> Result<Bytes, SerializeError> {

libs/postgres_ffi/wal_craft/src/xlog_utils_test.rs

@@ -178,7 +178,7 @@ pub fn test_find_end_of_wal_last_crossing_segment() {
 /// currently 1024.
 #[test]
 pub fn test_update_next_xid() {
-    let checkpoint_buf = [0u8; std::mem::size_of::<CheckPoint>()];
+    let checkpoint_buf = [0u8; size_of::<CheckPoint>()];
     let mut checkpoint = CheckPoint::decode(&checkpoint_buf).unwrap();
 
     checkpoint.nextXid = FullTransactionId { value: 10 };
@@ -204,7 +204,7 @@ pub fn test_update_next_xid() {
 
 #[test]
 pub fn test_update_next_multixid() {
-    let checkpoint_buf = [0u8; std::mem::size_of::<CheckPoint>()];
+    let checkpoint_buf = [0u8; size_of::<CheckPoint>()];
     let mut checkpoint = CheckPoint::decode(&checkpoint_buf).unwrap();
 
     // simple case

libs/pq_proto/src/framed.rs

@@ -44,9 +44,9 @@ impl ConnectionError {
 /// Wraps async io `stream`, providing messages to write/flush + read Postgres
 /// messages.
 pub struct Framed<S> {
-    stream: S,
-    read_buf: BytesMut,
-    write_buf: BytesMut,
+    pub stream: S,
+    pub read_buf: BytesMut,
+    pub write_buf: BytesMut,
 }
 
 impl<S> Framed<S> {

libs/pq_proto/src/lib.rs

@@ -39,14 +39,39 @@ pub enum FeMessage {
     PasswordMessage(Bytes),
 }
 
+#[derive(Clone, Copy, PartialEq, PartialOrd)]
+pub struct ProtocolVersion(u32);
+
+impl ProtocolVersion {
+    pub const fn new(major: u16, minor: u16) -> Self {
+        Self((major as u32) << 16 | minor as u32)
+    }
+    pub const fn minor(self) -> u16 {
+        self.0 as u16
+    }
+    pub const fn major(self) -> u16 {
+        (self.0 >> 16) as u16
+    }
+}
+
+impl fmt::Debug for ProtocolVersion {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        f.debug_list()
+            .entry(&self.major())
+            .entry(&self.minor())
+            .finish()
+    }
+}
+
 #[derive(Debug)]
 pub enum FeStartupPacket {
     CancelRequest(CancelKeyData),
-    SslRequest,
+    SslRequest {
+        direct: bool,
+    },
     GssEncRequest,
     StartupMessage {
-        major_version: u32,
-        minor_version: u32,
+        version: ProtocolVersion,
         params: StartupMessageParams,
     },
 }
@@ -301,11 +326,23 @@ impl FeStartupPacket {
     /// different from [`FeMessage::parse`] because startup messages don't have
     /// message type byte; otherwise, its comments apply.
     pub fn parse(buf: &mut BytesMut) -> Result<Option<FeStartupPacket>, ProtocolError> {
+        /// <https://github.com/postgres/postgres/blob/ca481d3c9ab7bf69ff0c8d71ad3951d407f6a33c/src/include/libpq/pqcomm.h#L118>
         const MAX_STARTUP_PACKET_LENGTH: usize = 10000;
-        const RESERVED_INVALID_MAJOR_VERSION: u32 = 1234;
-        const CANCEL_REQUEST_CODE: u32 = 5678;
-        const NEGOTIATE_SSL_CODE: u32 = 5679;
-        const NEGOTIATE_GSS_CODE: u32 = 5680;
+        const RESERVED_INVALID_MAJOR_VERSION: u16 = 1234;
+        /// <https://github.com/postgres/postgres/blob/ca481d3c9ab7bf69ff0c8d71ad3951d407f6a33c/src/include/libpq/pqcomm.h#L132>
+        const CANCEL_REQUEST_CODE: ProtocolVersion = ProtocolVersion::new(1234, 5678);
+        /// <https://github.com/postgres/postgres/blob/ca481d3c9ab7bf69ff0c8d71ad3951d407f6a33c/src/include/libpq/pqcomm.h#L166>
+        const NEGOTIATE_SSL_CODE: ProtocolVersion = ProtocolVersion::new(1234, 5679);
+        /// <https://github.com/postgres/postgres/blob/ca481d3c9ab7bf69ff0c8d71ad3951d407f6a33c/src/include/libpq/pqcomm.h#L167>
+        const NEGOTIATE_GSS_CODE: ProtocolVersion = ProtocolVersion::new(1234, 5680);
+
+        // <https://github.com/postgres/postgres/blob/04bcf9e19a4261fe9c7df37c777592c2e10c32a7/src/backend/tcop/backend_startup.c#L378-L382>
+        // First byte indicates standard SSL handshake message
+        // (It can't be a Postgres startup length because in network byte order
+        // that would be a startup packet hundreds of megabytes long)
+        if buf.first() == Some(&0x16) {
+            return Ok(Some(FeStartupPacket::SslRequest { direct: true }));
+        }
 
         // need at least 4 bytes with packet len
         if buf.len() < 4 {
@@ -338,12 +375,10 @@ impl FeStartupPacket {
         let mut msg = buf.split_to(len).freeze();
         msg.advance(4); // consume len
 
-        let request_code = msg.get_u32();
-        let req_hi = request_code >> 16;
-        let req_lo = request_code & ((1 << 16) - 1);
+        let request_code = ProtocolVersion(msg.get_u32());
         // StartupMessage, CancelRequest, SSLRequest etc are differentiated by request code.
-        let message = match (req_hi, req_lo) {
-            (RESERVED_INVALID_MAJOR_VERSION, CANCEL_REQUEST_CODE) => {
+        let message = match request_code {
+            CANCEL_REQUEST_CODE => {
                 if msg.remaining() != 8 {
                     return Err(ProtocolError::BadMessage(
                         "CancelRequest message is malformed, backend PID / secret key missing"
@@ -355,21 +390,22 @@ impl FeStartupPacket {
                     cancel_key: msg.get_i32(),
                 })
             }
-            (RESERVED_INVALID_MAJOR_VERSION, NEGOTIATE_SSL_CODE) => {
+            NEGOTIATE_SSL_CODE => {
                 // Requested upgrade to SSL (aka TLS)
-                FeStartupPacket::SslRequest
+                FeStartupPacket::SslRequest { direct: false }
             }
-            (RESERVED_INVALID_MAJOR_VERSION, NEGOTIATE_GSS_CODE) => {
+            NEGOTIATE_GSS_CODE => {
                 // Requested upgrade to GSSAPI
                 FeStartupPacket::GssEncRequest
             }
-            (RESERVED_INVALID_MAJOR_VERSION, unrecognized_code) => {
+            version if version.major() == RESERVED_INVALID_MAJOR_VERSION => {
                 return Err(ProtocolError::Protocol(format!(
-                    "Unrecognized request code {unrecognized_code}"
+                    "Unrecognized request code {}",
+                    version.minor()
                 )));
             }
             // TODO bail if protocol major_version is not 3?
-            (major_version, minor_version) => {
+            version => {
                 // StartupMessage
 
                 let s = str::from_utf8(&msg).map_err(|_e| {
@@ -382,8 +418,7 @@ impl FeStartupPacket {
                 })?;
 
                 FeStartupPacket::StartupMessage {
-                    major_version,
-                    minor_version,
+                    version,
                     params: StartupMessageParams {
                         params: msg.slice_ref(s.as_bytes()),
                     },
@@ -522,6 +557,10 @@ pub enum BeMessage<'a> {
     RowDescription(&'a [RowDescriptor<'a>]),
     XLogData(XLogDataBody<'a>),
     NoticeResponse(&'a str),
+    NegotiateProtocolVersion {
+        version: ProtocolVersion,
+        options: &'a [&'a str],
+    },
     KeepAlive(WalSndKeepAlive),
 }
 
@@ -945,6 +984,18 @@ impl<'a> BeMessage<'a> {
                     buf.put_u8(u8::from(req.request_reply));
                 });
             }
+
+            BeMessage::NegotiateProtocolVersion { version, options } => {
+                buf.put_u8(b'v');
+                write_body(buf, |buf| {
+                    buf.put_u32(version.0);
+                    buf.put_u32(options.len() as u32);
+                    for option in options.iter() {
+                        write_cstr(option, buf)?;
+                    }
+                    Ok(())
+                })?
+            }
         }
         Ok(())
     }

libs/remote_storage/Cargo.toml

@@ -7,6 +7,7 @@ license.workspace = true
 [dependencies]
 anyhow.workspace = true
 async-trait.workspace = true
+async-stream.workspace = true
 once_cell.workspace = true
 aws-smithy-async.workspace = true
 aws-smithy-types.workspace = true

libs/remote_storage/src/azure_blob.rs

@@ -15,7 +15,7 @@ use std::time::SystemTime;
 use super::REMOTE_STORAGE_PREFIX_SEPARATOR;
 use anyhow::Result;
 use azure_core::request_options::{MaxResults, Metadata, Range};
-use azure_core::RetryOptions;
+use azure_core::{Continuable, RetryOptions};
 use azure_identity::DefaultAzureCredential;
 use azure_storage::StorageCredentials;
 use azure_storage_blobs::blob::CopyStatus;
@@ -33,6 +33,7 @@ use tracing::debug;
 use utils::backoff;
 
 use crate::metrics::{start_measuring_requests, AttemptOutcome, RequestKind};
+use crate::ListingObject;
 use crate::{
     config::AzureConfig, error::Cancelled, ConcurrencyLimiter, Download, DownloadError, Listing,
     ListingMode, RemotePath, RemoteStorage, StorageMetadata, TimeTravelError, TimeoutOrCancel,
@@ -40,6 +41,7 @@ use crate::{
 
 pub struct AzureBlobStorage {
     client: ContainerClient,
+    container_name: String,
     prefix_in_container: Option<String>,
     max_keys_per_list_response: Option<NonZeroU32>,
     concurrency_limiter: ConcurrencyLimiter,
@@ -85,6 +87,7 @@ impl AzureBlobStorage {
 
         Ok(AzureBlobStorage {
             client,
+            container_name: azure_config.container_name.to_owned(),
             prefix_in_container: azure_config.prefix_in_container.to_owned(),
             max_keys_per_list_response,
             concurrency_limiter: ConcurrencyLimiter::new(azure_config.concurrency_limit.get()),
@@ -238,6 +241,10 @@ impl AzureBlobStorage {
             _ = cancel.cancelled() => Err(Cancelled),
         }
     }
+
+    pub fn container_name(&self) -> &str {
+        &self.container_name
+    }
 }
 
 fn to_azure_metadata(metadata: StorageMetadata) -> Metadata {
@@ -261,30 +268,30 @@ fn to_download_error(error: azure_core::Error) -> DownloadError {
 }
 
 impl RemoteStorage for AzureBlobStorage {
-    async fn list(
+    fn list_streaming(
         &self,
         prefix: Option<&RemotePath>,
         mode: ListingMode,
         max_keys: Option<NonZeroU32>,
         cancel: &CancellationToken,
-    ) -> anyhow::Result<Listing, DownloadError> {
-        let _permit = self.permit(RequestKind::List, cancel).await?;
+    ) -> impl Stream<Item = Result<Listing, DownloadError>> {
+        // get the passed prefix or if it is not set use prefix_in_bucket value
+        let list_prefix = prefix
+            .map(|p| self.relative_path_to_name(p))
+            .or_else(|| self.prefix_in_container.clone())
+            .map(|mut p| {
+                // required to end with a separator
+                // otherwise request will return only the entry of a prefix
+                if matches!(mode, ListingMode::WithDelimiter)
+                    && !p.ends_with(REMOTE_STORAGE_PREFIX_SEPARATOR)
+                {
+                    p.push(REMOTE_STORAGE_PREFIX_SEPARATOR);
+                }
+                p
+            });
 
-        let op = async {
-            // get the passed prefix or if it is not set use prefix_in_bucket value
-            let list_prefix = prefix
-                .map(|p| self.relative_path_to_name(p))
-                .or_else(|| self.prefix_in_container.clone())
-                .map(|mut p| {
-                    // required to end with a separator
-                    // otherwise request will return only the entry of a prefix
-                    if matches!(mode, ListingMode::WithDelimiter)
-                        && !p.ends_with(REMOTE_STORAGE_PREFIX_SEPARATOR)
-                    {
-                        p.push(REMOTE_STORAGE_PREFIX_SEPARATOR);
-                    }
-                    p
-                });
+        async_stream::stream! {
+            let _permit = self.permit(RequestKind::List, cancel).await?;
 
             let mut builder = self.client.list_blobs();
 
@@ -300,21 +307,43 @@ impl RemoteStorage for AzureBlobStorage {
                 builder = builder.max_results(MaxResults::new(limit));
             }
 
-            let response = builder.into_stream();
-            let response = response.into_stream().map_err(to_download_error);
-            let response = tokio_stream::StreamExt::timeout(response, self.timeout);
-            let response = response.map(|res| match res {
-                Ok(res) => res,
-                Err(_elapsed) => Err(DownloadError::Timeout),
-            });
+            let mut next_marker = None;
 
-            let mut response = std::pin::pin!(response);
+            'outer: loop {
+                let mut builder = builder.clone();
+                if let Some(marker) = next_marker.clone() {
+                    builder = builder.marker(marker);
+                }
+                let response = builder.into_stream();
+                let response = response.into_stream().map_err(to_download_error);
+                let response = tokio_stream::StreamExt::timeout(response, self.timeout);
+                let response = response.map(|res| match res {
+                    Ok(res) => res,
+                    Err(_elapsed) => Err(DownloadError::Timeout),
+                });
 
-            let mut res = Listing::default();
+                let mut response = std::pin::pin!(response);
 
-            let mut max_keys = max_keys.map(|mk| mk.get());
-            while let Some(entry) = response.next().await {
-                let entry = entry?;
+                let mut max_keys = max_keys.map(|mk| mk.get());
+                let next_item = tokio::select! {
+                    op = response.next() => Ok(op),
+                    _ = cancel.cancelled() => Err(DownloadError::Cancelled),
+                }?;
+                let Some(entry) = next_item else {
+                    // The list is complete, so yield it.
+                    break;
+                };
+
+                let mut res = Listing::default();
+                let entry = match entry {
+                    Ok(entry) => entry,
+                    Err(e) => {
+                        // The error is potentially retryable, so we must rewind the loop after yielding.
+                        yield Err(e);
+                        continue;
+                    }
+                };
+                next_marker = entry.continuation();
                 let prefix_iter = entry
                     .blobs
                     .prefixes()
@@ -324,7 +353,12 @@ impl RemoteStorage for AzureBlobStorage {
                 let blob_iter = entry
                     .blobs
                     .blobs()
-                    .map(|k| self.name_to_relative_path(&k.name));
+                    .map(|k| ListingObject{
+                        key: self.name_to_relative_path(&k.name),
+                        last_modified: k.properties.last_modified.into(),
+                        size: k.properties.content_length,
+                    }
+                    );
 
                 for key in blob_iter {
                     res.keys.push(key);
@@ -333,19 +367,19 @@ impl RemoteStorage for AzureBlobStorage {
                         assert!(mk > 0);
                         mk -= 1;
                         if mk == 0 {
-                            return Ok(res); // limit reached
+                            yield Ok(res); // limit reached
+                            break 'outer;
                         }
                         max_keys = Some(mk);
                     }
                 }
+                yield Ok(res);
+
+                // We are done here
+                if next_marker.is_none() {
+                    break;
+                }
             }
-
-            Ok(res)
-        };
-
-        tokio::select! {
-            res = op => res,
-            _ = cancel.cancelled() => Err(DownloadError::Cancelled),
         }
     }
 

libs/remote_storage/src/config.rs

@@ -1,6 +1,5 @@
 use std::{fmt::Debug, num::NonZeroUsize, str::FromStr, time::Duration};
 
-use anyhow::bail;
 use aws_sdk_s3::types::StorageClass;
 use camino::Utf8PathBuf;
 
@@ -176,20 +175,8 @@ fn serialize_storage_class<S: serde::Serializer>(
 impl RemoteStorageConfig {
     pub const DEFAULT_TIMEOUT: Duration = std::time::Duration::from_secs(120);
 
-    pub fn from_toml(toml: &toml_edit::Item) -> anyhow::Result<Option<RemoteStorageConfig>> {
-        let document: toml_edit::Document = match toml {
-            toml_edit::Item::Table(toml) => toml.clone().into(),
-            toml_edit::Item::Value(toml_edit::Value::InlineTable(toml)) => {
-                toml.clone().into_table().into()
-            }
-            _ => bail!("toml not a table or inline table"),
-        };
-
-        if document.is_empty() {
-            return Ok(None);
-        }
-
-        Ok(Some(toml_edit::de::from_document(document)?))
+    pub fn from_toml(toml: &toml_edit::Item) -> anyhow::Result<RemoteStorageConfig> {
+        Ok(utils::toml_edit_ext::deserialize_item(toml)?)
     }
 }
 
@@ -197,7 +184,7 @@ impl RemoteStorageConfig {
 mod tests {
     use super::*;
 
-    fn parse(input: &str) -> anyhow::Result<Option<RemoteStorageConfig>> {
+    fn parse(input: &str) -> anyhow::Result<RemoteStorageConfig> {
         let toml = input.parse::<toml_edit::Document>().unwrap();
         RemoteStorageConfig::from_toml(toml.as_item())
     }
@@ -207,7 +194,7 @@ mod tests {
         let input = "local_path = '.'
 timeout = '5s'";
 
-        let config = parse(input).unwrap().expect("it exists");
+        let config = parse(input).unwrap();
 
         assert_eq!(
             config,
@@ -229,7 +216,7 @@ timeout = '5s'";
     timeout = '7s'
     ";
 
-        let config = parse(toml).unwrap().expect("it exists");
+        let config = parse(toml).unwrap();
 
         assert_eq!(
             config,
@@ -257,7 +244,7 @@ timeout = '5s'";
     timeout = '7s'
     ";
 
-        let config = parse(toml).unwrap().expect("it exists");
+        let config = parse(toml).unwrap();
 
         assert_eq!(
             config,

libs/remote_storage/src/lib.rs

@@ -26,7 +26,7 @@ use anyhow::Context;
 use camino::{Utf8Path, Utf8PathBuf};
 
 use bytes::Bytes;
-use futures::stream::Stream;
+use futures::{stream::Stream, StreamExt};
 use serde::{Deserialize, Serialize};
 use tokio::sync::Semaphore;
 use tokio_util::sync::CancellationToken;
@@ -144,15 +144,23 @@ impl RemotePath {
 ///
 /// The WithDelimiter mode will populate `prefixes` and `keys` in the result.  The
 /// NoDelimiter mode will only populate `keys`.
+#[derive(Copy, Clone)]
 pub enum ListingMode {
     WithDelimiter,
     NoDelimiter,
 }
 
+#[derive(PartialEq, Eq, Debug)]
+pub struct ListingObject {
+    pub key: RemotePath,
+    pub last_modified: SystemTime,
+    pub size: u64,
+}
+
 #[derive(Default)]
 pub struct Listing {
     pub prefixes: Vec<RemotePath>,
-    pub keys: Vec<RemotePath>,
+    pub keys: Vec<ListingObject>,
 }
 
 /// Storage (potentially remote) API to manage its state.
@@ -160,13 +168,18 @@ pub struct Listing {
 /// providing basic CRUD operations for storage files.
 #[allow(async_fn_in_trait)]
 pub trait RemoteStorage: Send + Sync + 'static {
-    /// List objects in remote storage, with semantics matching AWS S3's ListObjectsV2.
-    /// (see `<https://docs.aws.amazon.com/AmazonS3/latest/API/API_ListObjectsV2.html>`)
+    /// List objects in remote storage, with semantics matching AWS S3's [`ListObjectsV2`].
+    ///
+    /// The stream is guaranteed to return at least one element, even in the case of errors
+    /// (in that case it's an `Err()`), or an empty `Listing`.
+    ///
+    /// The stream is not ending if it returns an error, as long as [`is_permanent`] returns false on the error.
+    /// The `next` function can be retried, and maybe in a future retry, there will be success.
     ///
     /// Note that the prefix is relative to any `prefix_in_bucket` configured for the client, not
     /// from the absolute root of the bucket.
     ///
-    /// `mode` configures whether to use a delimiter.  Without a delimiter all keys
+    /// `mode` configures whether to use a delimiter.  Without a delimiter, all keys
     /// within the prefix are listed in the `keys` of the result.  With a delimiter, any "directories" at the top level of
     /// the prefix are returned in the `prefixes` of the result, and keys in the top level of the prefix are
     /// returned in `keys` ().
@@ -175,13 +188,32 @@ pub trait RemoteStorage: Send + Sync + 'static {
     /// will iteratively call listobjects until it runs out of keys.  Note that this is not safe to use on
     /// unlimted size buckets, as the full list of objects is allocated into a monolithic data structure.
     ///
+    /// [`ListObjectsV2`]: <https://docs.aws.amazon.com/AmazonS3/latest/API/API_ListObjectsV2.html>
+    /// [`is_permanent`]: DownloadError::is_permanent
+    fn list_streaming(
+        &self,
+        prefix: Option<&RemotePath>,
+        mode: ListingMode,
+        max_keys: Option<NonZeroU32>,
+        cancel: &CancellationToken,
+    ) -> impl Stream<Item = Result<Listing, DownloadError>> + Send;
+
     async fn list(
         &self,
         prefix: Option<&RemotePath>,
-        _mode: ListingMode,
+        mode: ListingMode,
         max_keys: Option<NonZeroU32>,
         cancel: &CancellationToken,
-    ) -> Result<Listing, DownloadError>;
+    ) -> Result<Listing, DownloadError> {
+        let mut stream = std::pin::pin!(self.list_streaming(prefix, mode, max_keys, cancel));
+        let mut combined = stream.next().await.expect("At least one item required")?;
+        while let Some(list) = stream.next().await {
+            let list = list?;
+            combined.keys.extend(list.keys.into_iter());
+            combined.prefixes.extend_from_slice(&list.prefixes);
+        }
+        Ok(combined)
+    }
 
     /// Streams the local file contents into remote into the remote storage entry.
     ///
@@ -288,8 +320,8 @@ impl Debug for Download {
 
 /// 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
+#[derive(Clone)]
 pub enum GenericRemoteStorage<Other: Clone = Arc<UnreliableWrapper>> {
     LocalFs(LocalFs),
     AwsS3(Arc<S3Bucket>),
@@ -298,13 +330,14 @@ pub enum GenericRemoteStorage<Other: Clone = Arc<UnreliableWrapper>> {
 }
 
 impl<Other: RemoteStorage> GenericRemoteStorage<Arc<Other>> {
+    // See [`RemoteStorage::list`].
     pub async fn list(
         &self,
         prefix: Option<&RemotePath>,
         mode: ListingMode,
         max_keys: Option<NonZeroU32>,
         cancel: &CancellationToken,
-    ) -> anyhow::Result<Listing, DownloadError> {
+    ) -> Result<Listing, DownloadError> {
         match self {
             Self::LocalFs(s) => s.list(prefix, mode, max_keys, cancel).await,
             Self::AwsS3(s) => s.list(prefix, mode, max_keys, cancel).await,
@@ -313,6 +346,23 @@ impl<Other: RemoteStorage> GenericRemoteStorage<Arc<Other>> {
         }
     }
 
+    // See [`RemoteStorage::list_streaming`].
+    pub fn list_streaming<'a>(
+        &'a self,
+        prefix: Option<&'a RemotePath>,
+        mode: ListingMode,
+        max_keys: Option<NonZeroU32>,
+        cancel: &'a CancellationToken,
+    ) -> impl Stream<Item = Result<Listing, DownloadError>> + 'a + Send {
+        match self {
+            Self::LocalFs(s) => Box::pin(s.list_streaming(prefix, mode, max_keys, cancel))
+                as Pin<Box<dyn Stream<Item = Result<Listing, DownloadError>> + Send>>,
+            Self::AwsS3(s) => Box::pin(s.list_streaming(prefix, mode, max_keys, cancel)),
+            Self::AzureBlob(s) => Box::pin(s.list_streaming(prefix, mode, max_keys, cancel)),
+            Self::Unreliable(s) => Box::pin(s.list_streaming(prefix, mode, max_keys, cancel)),
+        }
+    }
+
     /// See [`RemoteStorage::upload`]
     pub async fn upload(
         &self,
@@ -443,7 +493,7 @@ impl<Other: RemoteStorage> GenericRemoteStorage<Arc<Other>> {
 }
 
 impl GenericRemoteStorage {
-    pub fn from_config(storage_config: &RemoteStorageConfig) -> anyhow::Result<Self> {
+    pub async fn from_config(storage_config: &RemoteStorageConfig) -> anyhow::Result<Self> {
         let timeout = storage_config.timeout;
         Ok(match &storage_config.storage {
             RemoteStorageKind::LocalFs { local_path: path } => {
@@ -458,7 +508,7 @@ impl GenericRemoteStorage {
                     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}",
                       s3_config.bucket_name, s3_config.bucket_region, s3_config.prefix_in_bucket, s3_config.endpoint);
-                Self::AwsS3(Arc::new(S3Bucket::new(s3_config, timeout)?))
+                Self::AwsS3(Arc::new(S3Bucket::new(s3_config, timeout).await?))
             }
             RemoteStorageKind::AzureContainer(azure_config) => {
                 let storage_account = azure_config
@@ -504,6 +554,16 @@ impl GenericRemoteStorage {
             None => self.download(from, cancel).await,
         }
     }
+
+    /// The name of the bucket/container/etc.
+    pub fn bucket_name(&self) -> Option<&str> {
+        match self {
+            Self::LocalFs(_s) => None,
+            Self::AwsS3(s) => Some(s.bucket_name()),
+            Self::AzureBlob(s) => Some(s.container_name()),
+            Self::Unreliable(_s) => None,
+        }
+    }
 }
 
 /// Extra set of key-value pairs that contain arbitrary metadata about the storage entry.

libs/remote_storage/src/local_fs.rs

@@ -23,8 +23,8 @@ use tokio_util::{io::ReaderStream, sync::CancellationToken};
 use utils::crashsafe::path_with_suffix_extension;
 
 use crate::{
-    Download, DownloadError, Listing, ListingMode, RemotePath, TimeTravelError, TimeoutOrCancel,
-    REMOTE_STORAGE_PREFIX_SEPARATOR,
+    Download, DownloadError, Listing, ListingMode, ListingObject, RemotePath, TimeTravelError,
+    TimeoutOrCancel, REMOTE_STORAGE_PREFIX_SEPARATOR,
 };
 
 use super::{RemoteStorage, StorageMetadata};
@@ -331,6 +331,17 @@ impl LocalFs {
 }
 
 impl RemoteStorage for LocalFs {
+    fn list_streaming(
+        &self,
+        prefix: Option<&RemotePath>,
+        mode: ListingMode,
+        max_keys: Option<NonZeroU32>,
+        cancel: &CancellationToken,
+    ) -> impl Stream<Item = Result<Listing, DownloadError>> {
+        let listing = self.list(prefix, mode, max_keys, cancel);
+        futures::stream::once(listing)
+    }
+
     async fn list(
         &self,
         prefix: Option<&RemotePath>,
@@ -346,19 +357,29 @@ impl RemoteStorage for LocalFs {
                 .list_recursive(prefix)
                 .await
                 .map_err(DownloadError::Other)?;
-            let keys = keys
+            let objects = keys
                 .into_iter()
-                .filter(|k| {
+                .filter_map(|k| {
                     let path = k.with_base(&self.storage_root);
-                    !path.is_dir()
+                    if path.is_dir() {
+                        None
+                    } else {
+                        Some(ListingObject {
+                            key: k.clone(),
+                            // LocalFs is just for testing, so just specify a dummy time
+                            last_modified: SystemTime::now(),
+                            size: 0,
+                        })
+                    }
                 })
                 .collect();
 
             if let ListingMode::NoDelimiter = mode {
-                result.keys = keys;
+                result.keys = objects;
             } else {
                 let mut prefixes = HashSet::new();
-                for key in keys {
+                for object in objects {
+                    let key = object.key;
                     // If the part after the prefix includes a "/", take only the first part and put it in `prefixes`.
                     let relative_key = if let Some(prefix) = prefix {
                         let mut prefix = prefix.clone();
@@ -387,9 +408,12 @@ impl RemoteStorage for LocalFs {
                             .to_owned();
                         prefixes.insert(first_part);
                     } else {
-                        result
-                            .keys
-                            .push(RemotePath::from_string(&relative_key).unwrap());
+                        result.keys.push(ListingObject {
+                            key: RemotePath::from_string(&relative_key).unwrap(),
+                            // LocalFs is just for testing
+                            last_modified: SystemTime::now(),
+                            size: 0,
+                        });
                     }
                 }
                 result.prefixes = prefixes
@@ -939,7 +963,11 @@ mod fs_tests {
             .await?;
         assert!(listing.prefixes.is_empty());
         assert_eq!(
-            listing.keys.into_iter().collect::<HashSet<_>>(),
+            listing
+                .keys
+                .into_iter()
+                .map(|o| o.key)
+                .collect::<HashSet<_>>(),
             HashSet::from([uncle.clone(), child.clone(), child_sibling.clone()])
         );
 
@@ -964,7 +992,7 @@ mod fs_tests {
             )
             .await?;
         assert_eq!(
-            listing.keys,
+            listing.keys.into_iter().map(|o| o.key).collect::<Vec<_>>(),
             [RemotePath::from_string("uncle").unwrap()].to_vec()
         );
         assert_eq!(
@@ -981,7 +1009,7 @@ mod fs_tests {
                 &cancel,
             )
             .await?;
-        assert_eq!(listing.keys, [].to_vec());
+        assert_eq!(listing.keys, vec![]);
         assert_eq!(
             listing.prefixes,
             [RemotePath::from_string("grandparent").unwrap()].to_vec()
@@ -996,7 +1024,7 @@ mod fs_tests {
                 &cancel,
             )
             .await?;
-        assert_eq!(listing.keys, [].to_vec());
+        assert_eq!(listing.keys, vec![]);
         assert_eq!(
             listing.prefixes,
             [RemotePath::from_string("grandparent").unwrap()].to_vec()
@@ -1029,7 +1057,7 @@ mod fs_tests {
                 &cancel,
             )
             .await?;
-        assert_eq!(listing.keys, [].to_vec());
+        assert_eq!(listing.keys, vec![]);
 
         let mut found_prefixes = listing.prefixes.clone();
         found_prefixes.sort();

libs/remote_storage/src/s3_bucket.rs

@@ -16,16 +16,10 @@ use std::{
 
 use anyhow::{anyhow, Context as _};
 use aws_config::{
-    environment::credentials::EnvironmentVariableCredentialsProvider,
-    imds::credentials::ImdsCredentialsProvider,
-    meta::credentials::CredentialsProviderChain,
-    profile::ProfileFileCredentialsProvider,
-    provider_config::ProviderConfig,
+    default_provider::credentials::DefaultCredentialsChain,
     retry::{RetryConfigBuilder, RetryMode},
-    web_identity_token::WebIdentityTokenCredentialsProvider,
     BehaviorVersion,
 };
-use aws_credential_types::provider::SharedCredentialsProvider;
 use aws_sdk_s3::{
     config::{AsyncSleep, IdentityCache, Region, SharedAsyncSleep},
     error::SdkError,
@@ -50,8 +44,9 @@ use crate::{
     error::Cancelled,
     metrics::{start_counting_cancelled_wait, start_measuring_requests},
     support::PermitCarrying,
-    ConcurrencyLimiter, Download, DownloadError, Listing, ListingMode, RemotePath, RemoteStorage,
-    TimeTravelError, TimeoutOrCancel, MAX_KEYS_PER_DELETE, REMOTE_STORAGE_PREFIX_SEPARATOR,
+    ConcurrencyLimiter, Download, DownloadError, Listing, ListingMode, ListingObject, RemotePath,
+    RemoteStorage, TimeTravelError, TimeoutOrCancel, MAX_KEYS_PER_DELETE,
+    REMOTE_STORAGE_PREFIX_SEPARATOR,
 };
 
 use crate::metrics::AttemptOutcome;
@@ -76,40 +71,27 @@ struct GetObjectRequest {
 }
 impl S3Bucket {
     /// Creates the S3 storage, errors if incorrect AWS S3 configuration provided.
-    pub fn new(remote_storage_config: &S3Config, timeout: Duration) -> anyhow::Result<Self> {
+    pub async fn new(remote_storage_config: &S3Config, timeout: Duration) -> anyhow::Result<Self> {
         tracing::debug!(
             "Creating s3 remote storage for S3 bucket {}",
             remote_storage_config.bucket_name
         );
 
-        let region = Some(Region::new(remote_storage_config.bucket_region.clone()));
+        let region = Region::new(remote_storage_config.bucket_region.clone());
+        let region_opt = Some(region.clone());
 
-        let provider_conf = ProviderConfig::without_region().with_region(region.clone());
-
-        let credentials_provider = {
-            // uses "AWS_ACCESS_KEY_ID", "AWS_SECRET_ACCESS_KEY"
-            CredentialsProviderChain::first_try(
-                "env",
-                EnvironmentVariableCredentialsProvider::new(),
-            )
-            // uses "AWS_PROFILE" / `aws sso login --profile <profile>`
-            .or_else(
-                "profile-sso",
-                ProfileFileCredentialsProvider::builder()
-                    .configure(&provider_conf)
-                    .build(),
-            )
-            // uses "AWS_WEB_IDENTITY_TOKEN_FILE", "AWS_ROLE_ARN", "AWS_ROLE_SESSION_NAME"
-            // needed to access remote extensions bucket
-            .or_else(
-                "token",
-                WebIdentityTokenCredentialsProvider::builder()
-                    .configure(&provider_conf)
-                    .build(),
-            )
-            // uses imds v2
-            .or_else("imds", ImdsCredentialsProvider::builder().build())
-        };
+        // https://docs.aws.amazon.com/sdkref/latest/guide/standardized-credentials.html
+        // https://docs.rs/aws-config/latest/aws_config/default_provider/credentials/struct.DefaultCredentialsChain.html
+        // Incomplete list of auth methods used by this:
+        // * "AWS_ACCESS_KEY_ID", "AWS_SECRET_ACCESS_KEY"
+        // * "AWS_PROFILE" / `aws sso login --profile <profile>`
+        // * "AWS_WEB_IDENTITY_TOKEN_FILE", "AWS_ROLE_ARN", "AWS_ROLE_SESSION_NAME"
+        // * http (ECS/EKS) container credentials
+        // * imds v2
+        let credentials_provider = DefaultCredentialsChain::builder()
+            .region(region)
+            .build()
+            .await;
 
         // AWS SDK requires us to specify how the RetryConfig should sleep when it wants to back off
         let sleep_impl: Arc<dyn AsyncSleep> = Arc::new(TokioSleep::new());
@@ -118,9 +100,9 @@ impl S3Bucket {
             #[allow(deprecated)] /* TODO: https://github.com/neondatabase/neon/issues/7665 */
             BehaviorVersion::v2023_11_09(),
         )
-        .region(region)
+        .region(region_opt)
         .identity_cache(IdentityCache::lazy().build())
-        .credentials_provider(SharedCredentialsProvider::new(credentials_provider))
+        .credentials_provider(credentials_provider)
         .sleep_impl(SharedAsyncSleep::from(sleep_impl));
 
         let sdk_config: aws_config::SdkConfig = std::thread::scope(|s| {
@@ -405,6 +387,10 @@ impl S3Bucket {
         }
         Ok(())
     }
+
+    pub fn bucket_name(&self) -> &str {
+        &self.bucket_name
+    }
 }
 
 pin_project_lite::pin_project! {
@@ -482,17 +468,16 @@ impl<S: Stream<Item = std::io::Result<Bytes>>> Stream for TimedDownload<S> {
 }
 
 impl RemoteStorage for S3Bucket {
-    async fn list(
+    fn list_streaming(
         &self,
         prefix: Option<&RemotePath>,
         mode: ListingMode,
         max_keys: Option<NonZeroU32>,
         cancel: &CancellationToken,
-    ) -> Result<Listing, DownloadError> {
+    ) -> impl Stream<Item = Result<Listing, DownloadError>> {
         let kind = RequestKind::List;
         // s3 sdk wants i32
         let mut max_keys = max_keys.map(|mk| mk.get() as i32);
-        let mut result = Listing::default();
 
         // get the passed prefix or if it is not set use prefix_in_bucket value
         let list_prefix = prefix
@@ -504,89 +489,119 @@ impl RemoteStorage for S3Bucket {
                 })
             });
 
-        let _permit = self.permit(kind, cancel).await?;
+        async_stream::stream! {
+            let _permit = self.permit(kind, cancel).await?;
 
-        let mut continuation_token = None;
+            let mut continuation_token = None;
+            'outer: loop {
+                let started_at = start_measuring_requests(kind);
 
-        loop {
-            let started_at = start_measuring_requests(kind);
+                // min of two Options, returning Some if one is value and another is
+                // None (None is smaller than anything, so plain min doesn't work).
+                let request_max_keys = self
+                    .max_keys_per_list_response
+                    .into_iter()
+                    .chain(max_keys.into_iter())
+                    .min();
+                let mut request = self
+                    .client
+                    .list_objects_v2()
+                    .bucket(self.bucket_name.clone())
+                    .set_prefix(list_prefix.clone())
+                    .set_continuation_token(continuation_token.clone())
+                    .set_max_keys(request_max_keys);
 
-            // min of two Options, returning Some if one is value and another is
-            // None (None is smaller than anything, so plain min doesn't work).
-            let request_max_keys = self
-                .max_keys_per_list_response
-                .into_iter()
-                .chain(max_keys.into_iter())
-                .min();
-            let mut request = self
-                .client
-                .list_objects_v2()
-                .bucket(self.bucket_name.clone())
-                .set_prefix(list_prefix.clone())
-                .set_continuation_token(continuation_token)
-                .set_max_keys(request_max_keys);
-
-            if let ListingMode::WithDelimiter = mode {
-                request = request.delimiter(REMOTE_STORAGE_PREFIX_SEPARATOR.to_string());
-            }
-
-            let request = request.send();
-
-            let response = tokio::select! {
-                res = request => res,
-                _ = tokio::time::sleep(self.timeout) => return Err(DownloadError::Timeout),
-                _ = cancel.cancelled() => return Err(DownloadError::Cancelled),
-            };
-
-            let response = response
-                .context("Failed to list S3 prefixes")
-                .map_err(DownloadError::Other);
-
-            let started_at = ScopeGuard::into_inner(started_at);
-
-            crate::metrics::BUCKET_METRICS
-                .req_seconds
-                .observe_elapsed(kind, &response, started_at);
-
-            let response = response?;
-
-            let keys = response.contents();
-            let empty = Vec::new();
-            let prefixes = response.common_prefixes.as_ref().unwrap_or(&empty);
-
-            tracing::debug!("list: {} prefixes, {} keys", prefixes.len(), keys.len());
-
-            for object in keys {
-                let object_path = object.key().expect("response does not contain a key");
-                let remote_path = self.s3_object_to_relative_path(object_path);
-                result.keys.push(remote_path);
-                if let Some(mut mk) = max_keys {
-                    assert!(mk > 0);
-                    mk -= 1;
-                    if mk == 0 {
-                        return Ok(result); // limit reached
-                    }
-                    max_keys = Some(mk);
+                if let ListingMode::WithDelimiter = mode {
+                    request = request.delimiter(REMOTE_STORAGE_PREFIX_SEPARATOR.to_string());
                 }
+
+                let request = request.send();
+
+                let response = tokio::select! {
+                    res = request => Ok(res),
+                    _ = tokio::time::sleep(self.timeout) => Err(DownloadError::Timeout),
+                    _ = cancel.cancelled() => Err(DownloadError::Cancelled),
+                }?;
+
+                let response = response
+                    .context("Failed to list S3 prefixes")
+                    .map_err(DownloadError::Other);
+
+                let started_at = ScopeGuard::into_inner(started_at);
+
+                crate::metrics::BUCKET_METRICS
+                    .req_seconds
+                    .observe_elapsed(kind, &response, started_at);
+
+                let response = match response {
+                    Ok(response) => response,
+                    Err(e) => {
+                        // The error is potentially retryable, so we must rewind the loop after yielding.
+                        yield Err(e);
+                        continue 'outer;
+                    },
+                };
+
+                let keys = response.contents();
+                let prefixes = response.common_prefixes.as_deref().unwrap_or_default();
+
+                tracing::debug!("list: {} prefixes, {} keys", prefixes.len(), keys.len());
+                let mut result = Listing::default();
+
+                for object in keys {
+                    let key = object.key().expect("response does not contain a key");
+                    let key = self.s3_object_to_relative_path(key);
+
+                    let last_modified = match object.last_modified.map(SystemTime::try_from) {
+                        Some(Ok(t)) => t,
+                        Some(Err(_)) => {
+                            tracing::warn!("Remote storage last_modified {:?} for {} is out of bounds",
+                                object.last_modified, key
+                        );
+                            SystemTime::now()
+                        },
+                        None => {
+                            SystemTime::now()
+                        }
+                    };
+
+                    let size = object.size.unwrap_or(0) as u64;
+
+                    result.keys.push(ListingObject{
+                        key,
+                        last_modified,
+                        size,
+                    });
+                    if let Some(mut mk) = max_keys {
+                        assert!(mk > 0);
+                        mk -= 1;
+                        if mk == 0 {
+                            // limit reached
+                            yield Ok(result);
+                            break 'outer;
+                        }
+                        max_keys = Some(mk);
+                    }
+                }
+
+                // S3 gives us prefixes like "foo/", we return them like "foo"
+                result.prefixes.extend(prefixes.iter().filter_map(|o| {
+                    Some(
+                        self.s3_object_to_relative_path(
+                            o.prefix()?
+                                .trim_end_matches(REMOTE_STORAGE_PREFIX_SEPARATOR),
+                        ),
+                    )
+                }));
+
+                yield Ok(result);
+
+                continuation_token = match response.next_continuation_token {
+                    Some(new_token) => Some(new_token),
+                    None => break,
+                };
             }
-
-            // S3 gives us prefixes like "foo/", we return them like "foo"
-            result.prefixes.extend(prefixes.iter().filter_map(|o| {
-                Some(
-                    self.s3_object_to_relative_path(
-                        o.prefix()?
-                            .trim_end_matches(REMOTE_STORAGE_PREFIX_SEPARATOR),
-                    ),
-                )
-            }));
-
-            continuation_token = match response.next_continuation_token {
-                Some(new_token) => Some(new_token),
-                None => break,
-            };
         }
-
-        Ok(result)
     }
 
     async fn upload(
@@ -1041,8 +1056,8 @@ mod tests {
 
     use crate::{RemotePath, S3Bucket, S3Config};
 
-    #[test]
-    fn relative_path() {
+    #[tokio::test]
+    async fn relative_path() {
         let all_paths = ["", "some/path", "some/path/"];
         let all_paths: Vec<RemotePath> = all_paths
             .iter()
@@ -1085,8 +1100,9 @@ mod tests {
                 max_keys_per_list_response: Some(5),
                 upload_storage_class: None,
             };
-            let storage =
-                S3Bucket::new(&config, std::time::Duration::ZERO).expect("remote storage init");
+            let storage = S3Bucket::new(&config, std::time::Duration::ZERO)
+                .await
+                .expect("remote storage init");
             for (test_path_idx, test_path) in all_paths.iter().enumerate() {
                 let result = storage.relative_path_to_s3_object(test_path);
                 let expected = expected_outputs[prefix_idx][test_path_idx];

libs/remote_storage/src/simulate_failures.rs

@@ -3,6 +3,7 @@
 //! testing purposes.
 use bytes::Bytes;
 use futures::stream::Stream;
+use futures::StreamExt;
 use std::collections::HashMap;
 use std::num::NonZeroU32;
 use std::sync::Mutex;
@@ -107,6 +108,23 @@ impl UnreliableWrapper {
 type VoidStorage = crate::LocalFs;
 
 impl RemoteStorage for UnreliableWrapper {
+    fn list_streaming(
+        &self,
+        prefix: Option<&RemotePath>,
+        mode: ListingMode,
+        max_keys: Option<NonZeroU32>,
+        cancel: &CancellationToken,
+    ) -> impl Stream<Item = Result<Listing, DownloadError>> + Send {
+        async_stream::stream! {
+            self.attempt(RemoteOp::ListPrefixes(prefix.cloned()))
+                .map_err(DownloadError::Other)?;
+            let mut stream = self.inner
+                .list_streaming(prefix, mode, max_keys, cancel);
+            while let Some(item) = stream.next().await {
+                yield item;
+            }
+        }
+    }
     async fn list(
         &self,
         prefix: Option<&RemotePath>,

libs/remote_storage/tests/common/mod.rs

@@ -152,7 +152,7 @@ pub(crate) async fn upload_remote_data(
     let mut upload_tasks = JoinSet::new();
     let cancel = CancellationToken::new();
 
-    for i in 1..upload_tasks_count + 1 {
+    for i in 1..=upload_tasks_count {
         let task_client = Arc::clone(client);
         let cancel = cancel.clone();
 

libs/remote_storage/tests/common/tests.rs

@@ -1,5 +1,6 @@
 use anyhow::Context;
 use camino::Utf8Path;
+use futures::StreamExt;
 use remote_storage::ListingMode;
 use remote_storage::RemotePath;
 use std::sync::Arc;
@@ -29,10 +30,10 @@ use super::{
 /// * with no prefix, it lists everything after its `${random_prefix_part}/` — that should be `${base_prefix_str}` value only
 /// * with `${base_prefix_str}/` prefix, it lists every `sub_prefix_${i}`
 ///
-/// With the real S3 enabled and `#[cfg(test)]` Rust configuration used, the S3 client test adds a `max-keys` param to limit the response keys.
-/// This way, we are able to test the pagination implicitly, by ensuring all results are returned from the remote storage and avoid uploading too many blobs to S3,
-/// since current default AWS S3 pagination limit is 1000.
-/// (see https://docs.aws.amazon.com/AmazonS3/latest/API/API_ListObjectsV2.html#API_ListObjectsV2_RequestSyntax)
+/// In the `MaybeEnabledStorageWithTestBlobs::setup`, we set the `max_keys_in_list_response` param to limit the keys in a single response.
+/// This way, we are able to test the pagination, by ensuring all results are returned from the remote storage and avoid uploading too many blobs to S3,
+/// as the current default AWS S3 pagination limit is 1000.
+/// (see <https://docs.aws.amazon.com/AmazonS3/latest/API/API_ListObjectsV2.html#API_ListObjectsV2_RequestSyntax>).
 ///
 /// Lastly, the test attempts to clean up and remove all uploaded S3 files.
 /// If any errors appear during the clean up, they get logged, but the test is not failed or stopped until clean up is finished.
@@ -87,6 +88,41 @@ async fn pagination_should_work(ctx: &mut MaybeEnabledStorageWithTestBlobs) -> a
         "remote storage nested prefixes list mismatches with the uploads. Remote only prefixes: {remote_only_prefixes:?}, missing uploaded prefixes: {missing_uploaded_prefixes:?}",
     );
 
+    // list_streaming
+
+    let prefix_with_slash = base_prefix.add_trailing_slash();
+    let mut nested_remote_prefixes_st = test_client.list_streaming(
+        Some(&prefix_with_slash),
+        ListingMode::WithDelimiter,
+        None,
+        &cancel,
+    );
+    let mut nested_remote_prefixes_combined = HashSet::new();
+    let mut segments = 0;
+    let mut segment_max_size = 0;
+    while let Some(st) = nested_remote_prefixes_st.next().await {
+        let st = st?;
+        segment_max_size = segment_max_size.max(st.prefixes.len());
+        nested_remote_prefixes_combined.extend(st.prefixes.into_iter());
+        segments += 1;
+    }
+    assert!(segments > 1, "less than 2 segments: {segments}");
+    assert!(
+        segment_max_size * 2 <= nested_remote_prefixes_combined.len(),
+        "double of segment_max_size={segment_max_size} larger number of remote prefixes of {}",
+        nested_remote_prefixes_combined.len()
+    );
+    let remote_only_prefixes = nested_remote_prefixes_combined
+        .difference(&expected_remote_prefixes)
+        .collect::<HashSet<_>>();
+    let missing_uploaded_prefixes = expected_remote_prefixes
+        .difference(&nested_remote_prefixes_combined)
+        .collect::<HashSet<_>>();
+    assert_eq!(
+        remote_only_prefixes.len() + missing_uploaded_prefixes.len(), 0,
+        "remote storage nested prefixes list mismatches with the uploads. Remote only prefixes: {remote_only_prefixes:?}, missing uploaded prefixes: {missing_uploaded_prefixes:?}",
+    );
+
     Ok(())
 }
 
@@ -120,6 +156,7 @@ async fn list_no_delimiter_works(
         .context("client list root files failure")?
         .keys
         .into_iter()
+        .map(|o| o.key)
         .collect::<HashSet<_>>();
     assert_eq!(
         root_files,
@@ -146,6 +183,7 @@ async fn list_no_delimiter_works(
         .context("client list nested files failure")?
         .keys
         .into_iter()
+        .map(|o| o.key)
         .collect::<HashSet<_>>();
     let trim_remote_blobs: HashSet<_> = ctx
         .remote_blobs

libs/remote_storage/tests/test_real_azure.rs

@@ -31,6 +31,7 @@ struct EnabledAzure {
 impl EnabledAzure {
     async fn setup(max_keys_in_list_response: Option<i32>) -> Self {
         let client = create_azure_client(max_keys_in_list_response)
+            .await
             .context("Azure client creation")
             .expect("Azure client creation failed");
 
@@ -187,7 +188,7 @@ impl AsyncTestContext for MaybeEnabledStorageWithSimpleTestBlobs {
     }
 }
 
-fn create_azure_client(
+async fn create_azure_client(
     max_keys_per_list_response: Option<i32>,
 ) -> anyhow::Result<Arc<GenericRemoteStorage>> {
     use rand::Rng;
@@ -221,6 +222,8 @@ fn create_azure_client(
         timeout: Duration::from_secs(120),
     };
     Ok(Arc::new(
-        GenericRemoteStorage::from_config(&remote_storage_config).context("remote storage init")?,
+        GenericRemoteStorage::from_config(&remote_storage_config)
+            .await
+            .context("remote storage init")?,
     ))
 }

libs/remote_storage/tests/test_real_s3.rs

@@ -81,6 +81,7 @@ async fn s3_time_travel_recovery_works(ctx: &mut MaybeEnabledStorage) -> anyhow:
                 .context("list root files failure")?
                 .keys
                 .into_iter()
+                .map(|o| o.key)
                 .collect::<HashSet<_>>(),
         )
     }
@@ -197,6 +198,7 @@ struct EnabledS3 {
 impl EnabledS3 {
     async fn setup(max_keys_in_list_response: Option<i32>) -> Self {
         let client = create_s3_client(max_keys_in_list_response)
+            .await
             .context("S3 client creation")
             .expect("S3 client creation failed");
 
@@ -352,7 +354,7 @@ impl AsyncTestContext for MaybeEnabledStorageWithSimpleTestBlobs {
     }
 }
 
-fn create_s3_client(
+async fn create_s3_client(
     max_keys_per_list_response: Option<i32>,
 ) -> anyhow::Result<Arc<GenericRemoteStorage>> {
     use rand::Rng;
@@ -385,7 +387,9 @@ fn create_s3_client(
         timeout: RemoteStorageConfig::DEFAULT_TIMEOUT,
     };
     Ok(Arc::new(
-        GenericRemoteStorage::from_config(&remote_storage_config).context("remote storage init")?,
+        GenericRemoteStorage::from_config(&remote_storage_config)
+            .await
+            .context("remote storage init")?,
     ))
 }
 

libs/tenant_size_model/src/calculation.rs

@@ -34,10 +34,10 @@ struct SegmentSize {
 }
 
 struct SizeAlternatives {
-    // cheapest alternative if parent is available.
+    /// cheapest alternative if parent is available.
     incremental: SegmentSize,
 
-    // cheapest alternative if parent node is not available
+    /// cheapest alternative if parent node is not available
     non_incremental: Option<SegmentSize>,
 }
 

libs/tenant_size_model/src/svg.rs

@@ -3,10 +3,17 @@ use std::fmt::Write;
 
 const SVG_WIDTH: f32 = 500.0;
 
+/// Different branch kind for SVG drawing.
+#[derive(PartialEq)]
+pub enum SvgBranchKind {
+    Timeline,
+    Lease,
+}
+
 struct SvgDraw<'a> {
     storage: &'a StorageModel,
     branches: &'a [String],
-    seg_to_branch: &'a [usize],
+    seg_to_branch: &'a [(usize, SvgBranchKind)],
     sizes: &'a [SegmentSizeResult],
 
     // layout
@@ -42,13 +49,18 @@ fn draw_legend(result: &mut String) -> anyhow::Result<()> {
         "<line x1=\"5\" y1=\"70\" x2=\"15\" y2=\"70\" stroke-width=\"1\" stroke=\"gray\" />"
     )?;
     writeln!(result, "<text x=\"20\" y=\"75\">WAL not retained</text>")?;
+    writeln!(
+        result,
+        "<line x1=\"10\" y1=\"85\" x2=\"10\" y2=\"95\" stroke-width=\"3\" stroke=\"blue\" />"
+    )?;
+    writeln!(result, "<text x=\"20\" y=\"95\">LSN lease</text>")?;
     Ok(())
 }
 
 pub fn draw_svg(
     storage: &StorageModel,
     branches: &[String],
-    seg_to_branch: &[usize],
+    seg_to_branch: &[(usize, SvgBranchKind)],
     sizes: &SizeResult,
 ) -> anyhow::Result<String> {
     let mut draw = SvgDraw {
@@ -100,7 +112,7 @@ impl<'a> SvgDraw<'a> {
 
         // Layout the timelines on Y dimension.
         // TODO
-        let mut y = 100.0;
+        let mut y = 120.0;
         let mut branch_y_coordinates = Vec::new();
         for _branch in self.branches {
             branch_y_coordinates.push(y);
@@ -109,7 +121,7 @@ impl<'a> SvgDraw<'a> {
 
         // Calculate coordinates for each point
         let seg_coordinates = std::iter::zip(segments, self.seg_to_branch)
-            .map(|(seg, branch_id)| {
+            .map(|(seg, (branch_id, _))| {
                 let x = (seg.lsn - min_lsn) as f32 / xscale;
                 let y = branch_y_coordinates[*branch_id];
                 (x, y)
@@ -175,6 +187,22 @@ impl<'a> SvgDraw<'a> {
 
         // draw a snapshot point if it's needed
         let (coord_x, coord_y) = self.seg_coordinates[seg_id];
+
+        let (_, kind) = &self.seg_to_branch[seg_id];
+        if kind == &SvgBranchKind::Lease {
+            let (x1, y1) = (coord_x, coord_y - 10.0);
+            let (x2, y2) = (coord_x, coord_y + 10.0);
+
+            let style = "stroke-width=\"3\" stroke=\"blue\"";
+
+            writeln!(
+                result,
+                "<line x1=\"{x1}\" y1=\"{y1}\" x2=\"{x2}\" y2=\"{y2}\" {style}>",
+            )?;
+            writeln!(result, "  <title>leased lsn at {}</title>", seg.lsn)?;
+            writeln!(result, "</line>")?;
+        }
+
         if self.sizes[seg_id].method == SegmentMethod::SnapshotHere {
             writeln!(
                 result,

libs/utils/Cargo.toml

@@ -20,7 +20,6 @@ bincode.workspace = true
 bytes.workspace = true
 camino.workspace = true
 chrono.workspace = true
-heapless.workspace = true
 hex = { workspace = true, features = ["serde"] }
 humantime.workspace = true
 hyper = { workspace = true, features = ["full"] }
@@ -40,6 +39,7 @@ thiserror.workspace = true
 tokio.workspace = true
 tokio-tar.workspace = true
 tokio-util.workspace = true
+toml_edit.workspace = true
 tracing.workspace = true
 tracing-error.workspace = true
 tracing-subscriber = { workspace = true, features = ["json", "registry"] }

libs/utils/src/auth.rs

@@ -18,21 +18,25 @@ const STORAGE_TOKEN_ALGORITHM: Algorithm = Algorithm::EdDSA;
 #[derive(Debug, Serialize, Deserialize, Clone, Copy, PartialEq)]
 #[serde(rename_all = "lowercase")]
 pub enum Scope {
-    // Provides access to all data for a specific tenant (specified in `struct Claims` below)
+    /// Provides access to all data for a specific tenant (specified in `struct Claims` below)
     // TODO: join these two?
     Tenant,
-    // Provides blanket access to all tenants on the pageserver plus pageserver-wide APIs.
-    // Should only be used e.g. for status check/tenant creation/list.
+    /// Provides blanket access to all tenants on the pageserver plus pageserver-wide APIs.
+    /// Should only be used e.g. for status check/tenant creation/list.
     PageServerApi,
-    // Provides blanket access to all data on the safekeeper plus safekeeper-wide APIs.
-    // Should only be used e.g. for status check.
-    // Currently also used for connection from any pageserver to any safekeeper.
+    /// Provides blanket access to all data on the safekeeper plus safekeeper-wide APIs.
+    /// Should only be used e.g. for status check.
+    /// Currently also used for connection from any pageserver to any safekeeper.
     SafekeeperData,
-    // The scope used by pageservers in upcalls to storage controller and cloud control plane
+    /// The scope used by pageservers in upcalls to storage controller and cloud control plane
     #[serde(rename = "generations_api")]
     GenerationsApi,
-    // Allows access to control plane managment API and some storage controller endpoints.
+    /// Allows access to control plane managment API and some storage controller endpoints.
     Admin,
+
+    /// Allows access to storage controller APIs used by the scrubber, to interrogate the state
+    /// of a tenant & post scrub results.
+    Scrubber,
 }
 
 /// JWT payload. See docs/authentication.md for the format

libs/utils/src/circuit_breaker.rs

@@ -0,0 +1,114 @@
+use std::{
+    fmt::Display,
+    time::{Duration, Instant},
+};
+
+use metrics::IntCounter;
+
+/// Circuit breakers are for operations that are expensive and fallible: if they fail repeatedly,
+/// we will stop attempting them for some period of time, to avoid denial-of-service from retries, and
+/// to mitigate the log spam from repeated failures.
+pub struct CircuitBreaker {
+    /// An identifier that enables us to log useful errors when a circuit is broken
+    name: String,
+
+    /// Consecutive failures since last success
+    fail_count: usize,
+
+    /// How many consecutive failures before we break the circuit
+    fail_threshold: usize,
+
+    /// If circuit is broken, when was it broken?
+    broken_at: Option<Instant>,
+
+    /// If set, we will auto-reset the circuit this long after it was broken.  If None, broken
+    /// circuits stay broken forever, or until success() is called.
+    reset_period: Option<Duration>,
+
+    /// If this is true, no actual circuit-breaking happens.  This is for overriding a circuit breaker
+    /// to permit something to keep running even if it would otherwise have tripped it.
+    short_circuit: bool,
+}
+
+impl CircuitBreaker {
+    pub fn new(name: String, fail_threshold: usize, reset_period: Option<Duration>) -> Self {
+        Self {
+            name,
+            fail_count: 0,
+            fail_threshold,
+            broken_at: None,
+            reset_period,
+            short_circuit: false,
+        }
+    }
+
+    /// Construct an unbreakable circuit breaker, for use in unit tests etc.
+    pub fn short_circuit() -> Self {
+        Self {
+            name: String::new(),
+            fail_threshold: 0,
+            fail_count: 0,
+            broken_at: None,
+            reset_period: None,
+            short_circuit: true,
+        }
+    }
+
+    pub fn fail<E>(&mut self, metric: &IntCounter, error: E)
+    where
+        E: Display,
+    {
+        if self.short_circuit {
+            return;
+        }
+
+        self.fail_count += 1;
+        if self.broken_at.is_none() && self.fail_count >= self.fail_threshold {
+            self.break_circuit(metric, error);
+        }
+    }
+
+    /// Call this after successfully executing an operation
+    pub fn success(&mut self, metric: &IntCounter) {
+        self.fail_count = 0;
+        if let Some(broken_at) = &self.broken_at {
+            tracing::info!(breaker=%self.name, "Circuit breaker failure ended (was broken for {})",
+                humantime::format_duration(broken_at.elapsed()));
+            self.broken_at = None;
+            metric.inc();
+        }
+    }
+
+    /// Call this before attempting an operation, and skip the operation if we are currently broken.
+    pub fn is_broken(&mut self) -> bool {
+        if self.short_circuit {
+            return false;
+        }
+
+        if let Some(broken_at) = self.broken_at {
+            match self.reset_period {
+                Some(reset_period) if broken_at.elapsed() > reset_period => {
+                    self.reset_circuit();
+                    false
+                }
+                _ => true,
+            }
+        } else {
+            false
+        }
+    }
+
+    fn break_circuit<E>(&mut self, metric: &IntCounter, error: E)
+    where
+        E: Display,
+    {
+        self.broken_at = Some(Instant::now());
+        tracing::error!(breaker=%self.name, "Circuit breaker broken!  Last error: {error}");
+        metric.inc();
+    }
+
+    fn reset_circuit(&mut self) {
+        self.broken_at = None;
+        self.fail_count = 0;
+    }
+}

libs/utils/src/history_buffer.rs

@@ -1,196 +0,0 @@
-//! A heapless buffer for events of sorts.
-
-use std::ops;
-
-use heapless::HistoryBuffer;
-
-#[derive(Debug, Clone)]
-pub struct HistoryBufferWithDropCounter<T, const L: usize> {
-    buffer: HistoryBuffer<T, L>,
-    drop_count: u64,
-}
-
-impl<T, const L: usize> HistoryBufferWithDropCounter<T, L> {
-    pub fn write(&mut self, data: T) {
-        let len_before = self.buffer.len();
-        self.buffer.write(data);
-        let len_after = self.buffer.len();
-        self.drop_count += u64::from(len_before == len_after);
-    }
-    pub fn drop_count(&self) -> u64 {
-        self.drop_count
-    }
-    pub fn map<U, F: Fn(&T) -> U>(&self, f: F) -> HistoryBufferWithDropCounter<U, L> {
-        let mut buffer = HistoryBuffer::new();
-        buffer.extend(self.buffer.oldest_ordered().map(f));
-        HistoryBufferWithDropCounter::<U, L> {
-            buffer,
-            drop_count: self.drop_count,
-        }
-    }
-}
-
-impl<T, const L: usize> Default for HistoryBufferWithDropCounter<T, L> {
-    fn default() -> Self {
-        Self {
-            buffer: HistoryBuffer::default(),
-            drop_count: 0,
-        }
-    }
-}
-
-impl<T, const L: usize> ops::Deref for HistoryBufferWithDropCounter<T, L> {
-    type Target = HistoryBuffer<T, L>;
-
-    fn deref(&self) -> &Self::Target {
-        &self.buffer
-    }
-}
-
-#[derive(serde::Serialize, serde::Deserialize)]
-struct SerdeRepr<T> {
-    buffer: Vec<T>,
-    buffer_size: usize,
-    drop_count: u64,
-}
-
-impl<'a, T, const L: usize> From<&'a HistoryBufferWithDropCounter<T, L>> for SerdeRepr<T>
-where
-    T: Clone + serde::Serialize,
-{
-    fn from(value: &'a HistoryBufferWithDropCounter<T, L>) -> Self {
-        let HistoryBufferWithDropCounter { buffer, drop_count } = value;
-        SerdeRepr {
-            buffer: buffer.iter().cloned().collect(),
-            buffer_size: L,
-            drop_count: *drop_count,
-        }
-    }
-}
-
-impl<T, const L: usize> serde::Serialize for HistoryBufferWithDropCounter<T, L>
-where
-    T: Clone + serde::Serialize,
-{
-    fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
-    where
-        S: serde::Serializer,
-    {
-        SerdeRepr::from(self).serialize(serializer)
-    }
-}
-
-impl<'de, T, const L: usize> serde::de::Deserialize<'de> for HistoryBufferWithDropCounter<T, L>
-where
-    T: Clone + serde::Deserialize<'de>,
-{
-    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
-    where
-        D: serde::Deserializer<'de>,
-    {
-        let SerdeRepr {
-            buffer: des_buffer,
-            drop_count,
-            buffer_size,
-        } = SerdeRepr::<T>::deserialize(deserializer)?;
-        if buffer_size != L {
-            use serde::de::Error;
-            return Err(D::Error::custom(format!(
-                "invalid buffer_size, expecting {L} got {buffer_size}"
-            )));
-        }
-        let mut buffer = HistoryBuffer::new();
-        buffer.extend(des_buffer);
-        Ok(HistoryBufferWithDropCounter { buffer, drop_count })
-    }
-}
-
-#[cfg(test)]
-mod test {
-    use super::HistoryBufferWithDropCounter;
-
-    #[test]
-    fn test_basics() {
-        let mut b = HistoryBufferWithDropCounter::<usize, 2>::default();
-        b.write(1);
-        b.write(2);
-        b.write(3);
-        assert!(b.iter().any(|e| *e == 2));
-        assert!(b.iter().any(|e| *e == 3));
-        assert!(!b.iter().any(|e| *e == 1));
-
-        // round-trip serde
-        let round_tripped: HistoryBufferWithDropCounter<usize, 2> =
-            serde_json::from_str(&serde_json::to_string(&b).unwrap()).unwrap();
-        assert_eq!(
-            round_tripped.iter().cloned().collect::<Vec<_>>(),
-            b.iter().cloned().collect::<Vec<_>>()
-        );
-    }
-
-    #[test]
-    fn test_drop_count_works() {
-        let mut b = HistoryBufferWithDropCounter::<_, 2>::default();
-        b.write(1);
-        assert_eq!(b.drop_count(), 0);
-        b.write(2);
-        assert_eq!(b.drop_count(), 0);
-        b.write(3);
-        assert_eq!(b.drop_count(), 1);
-        b.write(4);
-        assert_eq!(b.drop_count(), 2);
-    }
-
-    #[test]
-    fn test_clone_works() {
-        let mut b = HistoryBufferWithDropCounter::<_, 2>::default();
-        b.write(1);
-        b.write(2);
-        b.write(3);
-        assert_eq!(b.drop_count(), 1);
-        let mut c = b.clone();
-        assert_eq!(c.drop_count(), 1);
-        assert!(c.iter().any(|e| *e == 2));
-        assert!(c.iter().any(|e| *e == 3));
-        assert!(!c.iter().any(|e| *e == 1));
-
-        c.write(4);
-        assert!(c.iter().any(|e| *e == 4));
-        assert!(!b.iter().any(|e| *e == 4));
-    }
-
-    #[test]
-    fn test_map() {
-        let mut b = HistoryBufferWithDropCounter::<_, 2>::default();
-
-        b.write(1);
-        assert_eq!(b.drop_count(), 0);
-        {
-            let c = b.map(|i| i + 10);
-            assert_eq!(c.oldest_ordered().cloned().collect::<Vec<_>>(), vec![11]);
-            assert_eq!(c.drop_count(), 0);
-        }
-
-        b.write(2);
-        assert_eq!(b.drop_count(), 0);
-        {
-            let c = b.map(|i| i + 10);
-            assert_eq!(
-                c.oldest_ordered().cloned().collect::<Vec<_>>(),
-                vec![11, 12]
-            );
-            assert_eq!(c.drop_count(), 0);
-        }
-
-        b.write(3);
-        assert_eq!(b.drop_count(), 1);
-        {
-            let c = b.map(|i| i + 10);
-            assert_eq!(
-                c.oldest_ordered().cloned().collect::<Vec<_>>(),
-                vec![12, 13]
-            );
-            assert_eq!(c.drop_count(), 1);
-        }
-    }
-}

libs/utils/src/http/endpoint.rs

@@ -52,17 +52,17 @@ struct RequestId(String);
 /// There could be other ways to implement similar functionality:
 ///
 /// * procmacros placed on top of all handler methods
-/// With all the drawbacks of procmacros, brings no difference implementation-wise,
-/// and little code reduction compared to the existing approach.
+///   With all the drawbacks of procmacros, brings no difference implementation-wise,
+///   and little code reduction compared to the existing approach.
 ///
 /// * Another `TraitExt` with e.g. the `get_with_span`, `post_with_span` methods to do similar logic,
-/// implemented for [`RouterBuilder`].
-/// Could be simpler, but we don't want to depend on [`routerify`] more, targeting to use other library later.
+///   implemented for [`RouterBuilder`].
+///   Could be simpler, but we don't want to depend on [`routerify`] more, targeting to use other library later.
 ///
 /// * In theory, a span guard could've been created in a pre-request middleware and placed into a global collection, to be dropped
-/// later, in a post-response middleware.
-/// Due to suspendable nature of the futures, would give contradictive results which is exactly the opposite of what `tracing-futures`
-/// tries to achive with its `.instrument` used in the current approach.
+///   later, in a post-response middleware.
+///   Due to suspendable nature of the futures, would give contradictive results which is exactly the opposite of what `tracing-futures`
+///   tries to achive with its `.instrument` used in the current approach.
 ///
 /// If needed, a declarative macro to substitute the |r| ... closure boilerplate could be introduced.
 pub async fn request_span<R, H>(request: Request<Body>, handler: H) -> R::Output

libs/utils/src/http/request.rs

@@ -74,6 +74,15 @@ pub fn parse_query_param<E: fmt::Display, T: FromStr<Err = E>>(
         .transpose()
 }
 
+pub fn must_parse_query_param<E: fmt::Display, T: FromStr<Err = E>>(
+    request: &Request<Body>,
+    param_name: &str,
+) -> Result<T, ApiError> {
+    parse_query_param(request, param_name)?.ok_or_else(|| {
+        ApiError::BadRequest(anyhow!("no {param_name} specified in query parameters"))
+    })
+}
+
 pub async fn ensure_no_body(request: &mut Request<Body>) -> Result<(), ApiError> {
     match request.body_mut().data().await {
         Some(_) => Err(ApiError::BadRequest(anyhow!("Unexpected request body"))),

libs/utils/src/id.rs

@@ -302,17 +302,6 @@ pub struct TenantId(Id);
 
 id_newtype!(TenantId);
 
-/// Neon Connection Id identifies long-lived connections (for example a pagestream
-/// connection with the page_service). Is used for better logging and tracing
-///
-/// NOTE: It (de)serializes as an array of hex bytes, so the string representation would look
-/// like `[173,80,132,115,129,226,72,254,170,201,135,108,199,26,228,24]`.
-/// See [`Id`] for alternative ways to serialize it.
-#[derive(Clone, Copy, PartialEq, Eq, Hash, Serialize, Deserialize, PartialOrd, Ord)]
-pub struct ConnectionId(Id);
-
-id_newtype!(ConnectionId);
-
 // A pair uniquely identifying Neon instance.
 #[derive(Debug, Clone, Copy, PartialOrd, Ord, PartialEq, Eq, Hash, Serialize, Deserialize)]
 pub struct TenantTimelineId {

libs/utils/src/lib.rs

@@ -26,6 +26,8 @@ pub mod auth;
 // utility functions and helper traits for unified unique id generation/serialization etc.
 pub mod id;
 
+pub mod shard;
+
 mod hex;
 pub use hex::Hex;
 
@@ -57,8 +59,6 @@ pub mod signals;
 
 pub mod fs_ext;
 
-pub mod history_buffer;
-
 pub mod measured_stream;
 
 pub mod serde_percent;
@@ -94,6 +94,10 @@ pub mod env;
 
 pub mod poison;
 
+pub mod toml_edit_ext;
+
+pub mod circuit_breaker;
+
 /// This is a shortcut to embed git sha into binaries and avoid copying the same build script to all packages
 ///
 /// we have several cases:
@@ -124,7 +128,7 @@ pub mod poison;
 ///
 /// #############################################################################################
 /// TODO this macro is not the way the library is intended to be used, see <https://github.com/neondatabase/neon/issues/1565> for details.
-/// We use `cachepot` to reduce our current CI build times: <https://github.com/neondatabase/cloud/pull/1033#issuecomment-1100935036>
+/// We used `cachepot` to reduce our current CI build times: <https://github.com/neondatabase/cloud/pull/1033#issuecomment-1100935036>
 /// Yet, it seems to ignore the GIT_VERSION env variable, passed to Docker build, even with build.rs that contains
 /// `println!("cargo:rerun-if-env-changed=GIT_VERSION");` code for cachepot cache invalidation.
 /// The problem needs further investigation and regular `const` declaration instead of a macro.

libs/utils/src/shard.rs

@@ -0,0 +1,452 @@
+//! See `pageserver_api::shard` for description on sharding.
+
+use std::{ops::RangeInclusive, str::FromStr};
+
+use hex::FromHex;
+use serde::{Deserialize, Serialize};
+
+use crate::id::TenantId;
+
+#[derive(Ord, PartialOrd, Eq, PartialEq, Clone, Copy, Serialize, Deserialize, Debug, Hash)]
+pub struct ShardNumber(pub u8);
+
+#[derive(Ord, PartialOrd, Eq, PartialEq, Clone, Copy, Serialize, Deserialize, Debug, Hash)]
+pub struct ShardCount(pub u8);
+
+/// Combination of ShardNumber and ShardCount.  For use within the context of a particular tenant,
+/// when we need to know which shard we're dealing with, but do not need to know the full
+/// ShardIdentity (because we won't be doing any page->shard mapping), and do not need to know
+/// the fully qualified TenantShardId.
+#[derive(Eq, PartialEq, PartialOrd, Ord, Clone, Copy, Hash)]
+pub struct ShardIndex {
+    pub shard_number: ShardNumber,
+    pub shard_count: ShardCount,
+}
+
+/// Formatting helper, for generating the `shard_id` label in traces.
+pub struct ShardSlug<'a>(&'a TenantShardId);
+
+/// TenantShardId globally identifies a particular shard in a particular tenant.
+///
+/// These are written as `<TenantId>-<ShardSlug>`, for example:
+///   # The second shard in a two-shard tenant
+///   072f1291a5310026820b2fe4b2968934-0102
+///
+/// If the `ShardCount` is _unsharded_, the `TenantShardId` is written without
+/// a shard suffix and is equivalent to the encoding of a `TenantId`: this enables
+/// an unsharded [`TenantShardId`] to be used interchangably with a [`TenantId`].
+///
+/// The human-readable encoding of an unsharded TenantShardId, such as used in API URLs,
+/// is both forward and backward compatible with TenantId: a legacy TenantId can be
+/// decoded as a TenantShardId, and when re-encoded it will be parseable
+/// as a TenantId.
+#[derive(Eq, PartialEq, PartialOrd, Ord, Clone, Copy, Hash)]
+pub struct TenantShardId {
+    pub tenant_id: TenantId,
+    pub shard_number: ShardNumber,
+    pub shard_count: ShardCount,
+}
+
+impl ShardCount {
+    pub const MAX: Self = Self(u8::MAX);
+    pub const MIN: Self = Self(0);
+
+    /// The internal value of a ShardCount may be zero, which means "1 shard, but use
+    /// legacy format for TenantShardId that excludes the shard suffix", also known
+    /// as [`TenantShardId::unsharded`].
+    ///
+    /// This method returns the actual number of shards, i.e. if our internal value is
+    /// zero, we return 1 (unsharded tenants have 1 shard).
+    pub fn count(&self) -> u8 {
+        if self.0 > 0 {
+            self.0
+        } else {
+            1
+        }
+    }
+
+    /// The literal internal value: this is **not** the number of shards in the
+    /// tenant, as we have a special zero value for legacy unsharded tenants.  Use
+    /// [`Self::count`] if you want to know the cardinality of shards.
+    pub fn literal(&self) -> u8 {
+        self.0
+    }
+
+    /// Whether the `ShardCount` is for an unsharded tenant, so uses one shard but
+    /// uses the legacy format for `TenantShardId`. See also the documentation for
+    /// [`Self::count`].
+    pub fn is_unsharded(&self) -> bool {
+        self.0 == 0
+    }
+
+    /// `v` may be zero, or the number of shards in the tenant.  `v` is what
+    /// [`Self::literal`] would return.
+    pub const fn new(val: u8) -> Self {
+        Self(val)
+    }
+}
+
+impl ShardNumber {
+    pub const MAX: Self = Self(u8::MAX);
+}
+
+impl TenantShardId {
+    pub fn unsharded(tenant_id: TenantId) -> Self {
+        Self {
+            tenant_id,
+            shard_number: ShardNumber(0),
+            shard_count: ShardCount(0),
+        }
+    }
+
+    /// The range of all TenantShardId that belong to a particular TenantId.  This is useful when
+    /// you have a BTreeMap of TenantShardId, and are querying by TenantId.
+    pub fn tenant_range(tenant_id: TenantId) -> RangeInclusive<Self> {
+        RangeInclusive::new(
+            Self {
+                tenant_id,
+                shard_number: ShardNumber(0),
+                shard_count: ShardCount(0),
+            },
+            Self {
+                tenant_id,
+                shard_number: ShardNumber::MAX,
+                shard_count: ShardCount::MAX,
+            },
+        )
+    }
+
+    pub fn shard_slug(&self) -> impl std::fmt::Display + '_ {
+        ShardSlug(self)
+    }
+
+    /// Convenience for code that has special behavior on the 0th shard.
+    pub fn is_shard_zero(&self) -> bool {
+        self.shard_number == ShardNumber(0)
+    }
+
+    /// The "unsharded" value is distinct from simply having a single shard: it represents
+    /// a tenant which is not shard-aware at all, and whose storage paths will not include
+    /// a shard suffix.
+    pub fn is_unsharded(&self) -> bool {
+        self.shard_number == ShardNumber(0) && self.shard_count.is_unsharded()
+    }
+
+    /// Convenience for dropping the tenant_id and just getting the ShardIndex: this
+    /// is useful when logging from code that is already in a span that includes tenant ID, to
+    /// keep messages reasonably terse.
+    pub fn to_index(&self) -> ShardIndex {
+        ShardIndex {
+            shard_number: self.shard_number,
+            shard_count: self.shard_count,
+        }
+    }
+
+    /// Calculate the children of this TenantShardId when splitting the overall tenant into
+    /// the given number of shards.
+    pub fn split(&self, new_shard_count: ShardCount) -> Vec<TenantShardId> {
+        let effective_old_shard_count = std::cmp::max(self.shard_count.0, 1);
+        let mut child_shards = Vec::new();
+        for shard_number in 0..ShardNumber(new_shard_count.0).0 {
+            // Key mapping is based on a round robin mapping of key hash modulo shard count,
+            // so our child shards are the ones which the same keys would map to.
+            if shard_number % effective_old_shard_count == self.shard_number.0 {
+                child_shards.push(TenantShardId {
+                    tenant_id: self.tenant_id,
+                    shard_number: ShardNumber(shard_number),
+                    shard_count: new_shard_count,
+                })
+            }
+        }
+
+        child_shards
+    }
+}
+
+impl<'a> std::fmt::Display for ShardSlug<'a> {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(
+            f,
+            "{:02x}{:02x}",
+            self.0.shard_number.0, self.0.shard_count.0
+        )
+    }
+}
+
+impl std::fmt::Display for TenantShardId {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        if self.shard_count != ShardCount(0) {
+            write!(f, "{}-{}", self.tenant_id, self.shard_slug())
+        } else {
+            // Legacy case (shard_count == 0) -- format as just the tenant id.  Note that this
+            // is distinct from the normal single shard case (shard count == 1).
+            self.tenant_id.fmt(f)
+        }
+    }
+}
+
+impl std::fmt::Debug for TenantShardId {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        // Debug is the same as Display: the compact hex representation
+        write!(f, "{}", self)
+    }
+}
+
+impl std::str::FromStr for TenantShardId {
+    type Err = hex::FromHexError;
+
+    fn from_str(s: &str) -> Result<Self, Self::Err> {
+        // Expect format: 16 byte TenantId, '-', 1 byte shard number, 1 byte shard count
+        if s.len() == 32 {
+            // Legacy case: no shard specified
+            Ok(Self {
+                tenant_id: TenantId::from_str(s)?,
+                shard_number: ShardNumber(0),
+                shard_count: ShardCount(0),
+            })
+        } else if s.len() == 37 {
+            let bytes = s.as_bytes();
+            let tenant_id = TenantId::from_hex(&bytes[0..32])?;
+            let mut shard_parts: [u8; 2] = [0u8; 2];
+            hex::decode_to_slice(&bytes[33..37], &mut shard_parts)?;
+            Ok(Self {
+                tenant_id,
+                shard_number: ShardNumber(shard_parts[0]),
+                shard_count: ShardCount(shard_parts[1]),
+            })
+        } else {
+            Err(hex::FromHexError::InvalidStringLength)
+        }
+    }
+}
+
+impl From<[u8; 18]> for TenantShardId {
+    fn from(b: [u8; 18]) -> Self {
+        let tenant_id_bytes: [u8; 16] = b[0..16].try_into().unwrap();
+
+        Self {
+            tenant_id: TenantId::from(tenant_id_bytes),
+            shard_number: ShardNumber(b[16]),
+            shard_count: ShardCount(b[17]),
+        }
+    }
+}
+
+impl ShardIndex {
+    pub fn new(number: ShardNumber, count: ShardCount) -> Self {
+        Self {
+            shard_number: number,
+            shard_count: count,
+        }
+    }
+    pub fn unsharded() -> Self {
+        Self {
+            shard_number: ShardNumber(0),
+            shard_count: ShardCount(0),
+        }
+    }
+
+    /// The "unsharded" value is distinct from simply having a single shard: it represents
+    /// a tenant which is not shard-aware at all, and whose storage paths will not include
+    /// a shard suffix.
+    pub fn is_unsharded(&self) -> bool {
+        self.shard_number == ShardNumber(0) && self.shard_count == ShardCount(0)
+    }
+
+    /// For use in constructing remote storage paths: concatenate this with a TenantId
+    /// to get a fully qualified TenantShardId.
+    ///
+    /// Backward compat: this function returns an empty string if Self::is_unsharded, such
+    /// that the legacy pre-sharding remote key format is preserved.
+    pub fn get_suffix(&self) -> String {
+        if self.is_unsharded() {
+            "".to_string()
+        } else {
+            format!("-{:02x}{:02x}", self.shard_number.0, self.shard_count.0)
+        }
+    }
+}
+
+impl std::fmt::Display for ShardIndex {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(f, "{:02x}{:02x}", self.shard_number.0, self.shard_count.0)
+    }
+}
+
+impl std::fmt::Debug for ShardIndex {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        // Debug is the same as Display: the compact hex representation
+        write!(f, "{}", self)
+    }
+}
+
+impl std::str::FromStr for ShardIndex {
+    type Err = hex::FromHexError;
+
+    fn from_str(s: &str) -> Result<Self, Self::Err> {
+        // Expect format: 1 byte shard number, 1 byte shard count
+        if s.len() == 4 {
+            let bytes = s.as_bytes();
+            let mut shard_parts: [u8; 2] = [0u8; 2];
+            hex::decode_to_slice(bytes, &mut shard_parts)?;
+            Ok(Self {
+                shard_number: ShardNumber(shard_parts[0]),
+                shard_count: ShardCount(shard_parts[1]),
+            })
+        } else {
+            Err(hex::FromHexError::InvalidStringLength)
+        }
+    }
+}
+
+impl From<[u8; 2]> for ShardIndex {
+    fn from(b: [u8; 2]) -> Self {
+        Self {
+            shard_number: ShardNumber(b[0]),
+            shard_count: ShardCount(b[1]),
+        }
+    }
+}
+
+impl Serialize for TenantShardId {
+    fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
+    where
+        S: serde::Serializer,
+    {
+        if serializer.is_human_readable() {
+            serializer.collect_str(self)
+        } else {
+            // Note: while human encoding of [`TenantShardId`] is backward and forward
+            // compatible, this binary encoding is not.
+            let mut packed: [u8; 18] = [0; 18];
+            packed[0..16].clone_from_slice(&self.tenant_id.as_arr());
+            packed[16] = self.shard_number.0;
+            packed[17] = self.shard_count.0;
+
+            packed.serialize(serializer)
+        }
+    }
+}
+
+impl<'de> Deserialize<'de> for TenantShardId {
+    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
+    where
+        D: serde::Deserializer<'de>,
+    {
+        struct IdVisitor {
+            is_human_readable_deserializer: bool,
+        }
+
+        impl<'de> serde::de::Visitor<'de> for IdVisitor {
+            type Value = TenantShardId;
+
+            fn expecting(&self, formatter: &mut std::fmt::Formatter) -> std::fmt::Result {
+                if self.is_human_readable_deserializer {
+                    formatter.write_str("value in form of hex string")
+                } else {
+                    formatter.write_str("value in form of integer array([u8; 18])")
+                }
+            }
+
+            fn visit_seq<A>(self, seq: A) -> Result<Self::Value, A::Error>
+            where
+                A: serde::de::SeqAccess<'de>,
+            {
+                let s = serde::de::value::SeqAccessDeserializer::new(seq);
+                let id: [u8; 18] = Deserialize::deserialize(s)?;
+                Ok(TenantShardId::from(id))
+            }
+
+            fn visit_str<E>(self, v: &str) -> Result<Self::Value, E>
+            where
+                E: serde::de::Error,
+            {
+                TenantShardId::from_str(v).map_err(E::custom)
+            }
+        }
+
+        if deserializer.is_human_readable() {
+            deserializer.deserialize_str(IdVisitor {
+                is_human_readable_deserializer: true,
+            })
+        } else {
+            deserializer.deserialize_tuple(
+                18,
+                IdVisitor {
+                    is_human_readable_deserializer: false,
+                },
+            )
+        }
+    }
+}
+
+impl Serialize for ShardIndex {
+    fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
+    where
+        S: serde::Serializer,
+    {
+        if serializer.is_human_readable() {
+            serializer.collect_str(self)
+        } else {
+            // Binary encoding is not used in index_part.json, but is included in anticipation of
+            // switching various structures (e.g. inter-process communication, remote metadata) to more
+            // compact binary encodings in future.
+            let mut packed: [u8; 2] = [0; 2];
+            packed[0] = self.shard_number.0;
+            packed[1] = self.shard_count.0;
+            packed.serialize(serializer)
+        }
+    }
+}
+
+impl<'de> Deserialize<'de> for ShardIndex {
+    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
+    where
+        D: serde::Deserializer<'de>,
+    {
+        struct IdVisitor {
+            is_human_readable_deserializer: bool,
+        }
+
+        impl<'de> serde::de::Visitor<'de> for IdVisitor {
+            type Value = ShardIndex;
+
+            fn expecting(&self, formatter: &mut std::fmt::Formatter) -> std::fmt::Result {
+                if self.is_human_readable_deserializer {
+                    formatter.write_str("value in form of hex string")
+                } else {
+                    formatter.write_str("value in form of integer array([u8; 2])")
+                }
+            }
+
+            fn visit_seq<A>(self, seq: A) -> Result<Self::Value, A::Error>
+            where
+                A: serde::de::SeqAccess<'de>,
+            {
+                let s = serde::de::value::SeqAccessDeserializer::new(seq);
+                let id: [u8; 2] = Deserialize::deserialize(s)?;
+                Ok(ShardIndex::from(id))
+            }
+
+            fn visit_str<E>(self, v: &str) -> Result<Self::Value, E>
+            where
+                E: serde::de::Error,
+            {
+                ShardIndex::from_str(v).map_err(E::custom)
+            }
+        }
+
+        if deserializer.is_human_readable() {
+            deserializer.deserialize_str(IdVisitor {
+                is_human_readable_deserializer: true,
+            })
+        } else {
+            deserializer.deserialize_tuple(
+                2,
+                IdVisitor {
+                    is_human_readable_deserializer: false,
+                },
+            )
+        }
+    }
+}