Merge pull request #8723 from neondatabase/rc/proxy/2024-08-14

Proxy release 2024-08-14

.github/actionlint.yml

@@ -8,7 +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_CONTAINER_NEW
   - REMOTE_STORAGE_AZURE_REGION
   - SLACK_UPCOMING_RELEASE_CHANNEL_ID
+  - DEV_AWS_OIDC_ROLE_ARN

.github/actions/set-docker-config-dir/action.yml

@@ -0,0 +1,36 @@
+name: "Set custom docker config directory"
+description: "Create a directory for docker config and set DOCKER_CONFIG"
+
+# Use custom DOCKER_CONFIG directory to avoid conflicts with default settings
+runs:
+  using: "composite"
+  steps:
+  - name: Show warning on GitHub-hosted runners
+    if: runner.environment == 'github-hosted'
+    shell: bash -euo pipefail {0}
+    run: |
+      # Using the following environment variables to find a path to the workflow file
+      # ${GITHUB_WORKFLOW_REF} - octocat/hello-world/.github/workflows/my-workflow.yml@refs/heads/my_branch
+      # ${GITHUB_REPOSITORY}   - octocat/hello-world
+      # ${GITHUB_REF}          - refs/heads/my_branch
+      # From https://docs.github.com/en/actions/writing-workflows/choosing-what-your-workflow-does/variables
+
+      filename_with_ref=${GITHUB_WORKFLOW_REF#"$GITHUB_REPOSITORY/"}
+      filename=${filename_with_ref%"@$GITHUB_REF"}
+
+      # https://docs.github.com/en/actions/writing-workflows/choosing-what-your-workflow-does/workflow-commands-for-github-actions#setting-a-warning-message
+      title='Unnecessary usage of `.github/actions/set-docker-config-dir`'
+      message='No need to use `.github/actions/set-docker-config-dir` action on GitHub-hosted runners'
+      echo "::warning file=${filename},title=${title}::${message}"
+
+  - uses: pyTooling/Actions/with-post-step@74afc5a42a17a046c90c68cb5cfa627e5c6c5b6b # v1.0.7
+    env:
+      DOCKER_CONFIG: .docker-custom-${{ github.run_id }}-${{ github.run_attempt }}
+    with:
+      main: |
+        mkdir -p "${DOCKER_CONFIG}"
+        echo DOCKER_CONFIG=${DOCKER_CONFIG} | tee -a $GITHUB_ENV
+      post: |
+        if [ -d "${DOCKER_CONFIG}" ]; then
+          rm -r "${DOCKER_CONFIG}"
+        fi

.github/workflows/_benchmarking_preparation.yml

@@ -0,0 +1,152 @@
+name: Prepare benchmarking databases by restoring dumps
+
+on:
+  workflow_call:
+    # no inputs needed
+    
+defaults:
+  run:
+    shell: bash -euxo pipefail {0}
+
+jobs:
+  setup-databases:
+    strategy:
+      fail-fast: false
+      matrix:
+        platform: [ aws-rds-postgres, aws-aurora-serverless-v2-postgres, neon ] 
+        database: [ clickbench, tpch, userexample ]
+  
+    env:
+      LD_LIBRARY_PATH: /tmp/neon/pg_install/v16/lib
+      PLATFORM: ${{ matrix.platform }}
+      PG_BINARIES: /tmp/neon/pg_install/v16/bin
+
+    runs-on: [ self-hosted, us-east-2, x64 ]
+    container:
+      image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools:pinned
+      options: --init
+
+    steps:
+    - name: Set up Connection String
+      id: set-up-prep-connstr
+      run: |
+        case "${PLATFORM}" in
+          neon)
+            CONNSTR=${{ secrets.BENCHMARK_CAPTEST_CONNSTR }} 
+            ;;
+          aws-rds-postgres)
+            CONNSTR=${{ secrets.BENCHMARK_RDS_POSTGRES_CONNSTR }} 
+            ;;
+          aws-aurora-serverless-v2-postgres)
+            CONNSTR=${{ secrets.BENCHMARK_RDS_AURORA_CONNSTR }} 
+            ;;
+          *)
+            echo >&2 "Unknown PLATFORM=${PLATFORM}"
+            exit 1
+            ;;
+        esac
+
+        echo "connstr=${CONNSTR}" >> $GITHUB_OUTPUT  
+
+    - name: Download Neon artifact
+      uses: ./.github/actions/download
+      with:
+        name: neon-${{ runner.os }}-${{ runner.arch }}-release-artifact
+        path: /tmp/neon/
+        prefix: latest
+
+    # we create a table that has one row for each database that we want to restore with the status whether the restore is done    
+    - name: Create benchmark_restore_status table if it does not exist
+      env:
+        BENCHMARK_CONNSTR: ${{ steps.set-up-prep-connstr.outputs.connstr }}
+        DATABASE_NAME: ${{ matrix.database }}
+      # to avoid a race condition of multiple jobs trying to create the table at the same time, 
+      # we use an advisory lock
+      run: |
+        ${PG_BINARIES}/psql "${{ env.BENCHMARK_CONNSTR }}" -c "
+        SELECT pg_advisory_lock(4711);  
+        CREATE TABLE IF NOT EXISTS benchmark_restore_status (
+        databasename text primary key,
+        restore_done boolean
+        );
+        SELECT pg_advisory_unlock(4711);
+        "
+    
+    - name: Check if restore is already done
+      id: check-restore-done
+      env:
+        BENCHMARK_CONNSTR: ${{ steps.set-up-prep-connstr.outputs.connstr }}
+        DATABASE_NAME: ${{ matrix.database }}
+      run: |
+        skip=false
+        if ${PG_BINARIES}/psql "${{ env.BENCHMARK_CONNSTR }}" -tAc "SELECT 1 FROM benchmark_restore_status WHERE databasename='${{ env.DATABASE_NAME }}' AND restore_done=true;" | grep -q 1; then
+          echo "Restore already done for database ${{ env.DATABASE_NAME }} on platform ${{ env.PLATFORM }}. Skipping this database."
+          skip=true
+        fi
+        echo "skip=${skip}" | tee -a $GITHUB_OUTPUT
+
+    - name: Check and create database if it does not exist
+      if: steps.check-restore-done.outputs.skip != 'true'
+      env:
+        BENCHMARK_CONNSTR: ${{ steps.set-up-prep-connstr.outputs.connstr }}
+        DATABASE_NAME: ${{ matrix.database }}
+      run: |
+        DB_EXISTS=$(${PG_BINARIES}/psql "${{ env.BENCHMARK_CONNSTR }}" -tAc "SELECT 1 FROM pg_database WHERE datname='${{ env.DATABASE_NAME }}'")
+        if [ "$DB_EXISTS" != "1" ]; then
+          echo "Database ${{ env.DATABASE_NAME }} does not exist. Creating it..."
+          ${PG_BINARIES}/psql "${{ env.BENCHMARK_CONNSTR }}" -c "CREATE DATABASE \"${{ env.DATABASE_NAME }}\";"
+        else
+          echo "Database ${{ env.DATABASE_NAME }} already exists."
+        fi
+
+    - name: Download dump from S3 to /tmp/dumps
+      if: steps.check-restore-done.outputs.skip != 'true'
+      env:
+        DATABASE_NAME: ${{ matrix.database }}
+      run: |
+        mkdir -p /tmp/dumps
+        aws s3 cp s3://neon-github-dev/performance/pgdumps/$DATABASE_NAME/$DATABASE_NAME.pg_dump /tmp/dumps/ 
+
+    - name: Replace database name in connection string
+      if: steps.check-restore-done.outputs.skip != 'true'
+      id: replace-dbname
+      env:
+        DATABASE_NAME: ${{ matrix.database }}
+        BENCHMARK_CONNSTR: ${{ steps.set-up-prep-connstr.outputs.connstr }}
+      run: |
+        # Extract the part before the database name
+        base_connstr="${BENCHMARK_CONNSTR%/*}"
+        # Extract the query parameters (if any) after the database name
+        query_params="${BENCHMARK_CONNSTR#*\?}"
+        # Reconstruct the new connection string
+        if [ "$query_params" != "$BENCHMARK_CONNSTR" ]; then
+          new_connstr="${base_connstr}/${DATABASE_NAME}?${query_params}"
+        else
+          new_connstr="${base_connstr}/${DATABASE_NAME}"
+        fi
+        echo "database_connstr=${new_connstr}" >> $GITHUB_OUTPUT  
+
+    - name: Restore dump
+      if: steps.check-restore-done.outputs.skip != 'true'
+      env:
+        DATABASE_NAME: ${{ matrix.database }}
+        DATABASE_CONNSTR: ${{ steps.replace-dbname.outputs.database_connstr }}
+        # the following works only with larger computes: 
+        # PGOPTIONS: "-c maintenance_work_mem=8388608 -c max_parallel_maintenance_workers=7"
+        # we add the || true because:
+        # the dumps were created with Neon and contain neon extensions that are not 
+        # available in RDS, so we will always report an error, but we can ignore it
+      run: |
+        ${PG_BINARIES}/pg_restore --clean --if-exists --no-owner --jobs=4 \
+        -d "${DATABASE_CONNSTR}" /tmp/dumps/${DATABASE_NAME}.pg_dump || true
+
+    - name: Update benchmark_restore_status table
+      if: steps.check-restore-done.outputs.skip != 'true'
+      env:
+        BENCHMARK_CONNSTR: ${{ steps.set-up-prep-connstr.outputs.connstr }}
+        DATABASE_NAME: ${{ matrix.database }}
+      run: |
+        ${PG_BINARIES}/psql "${{ env.BENCHMARK_CONNSTR }}" -c "
+        INSERT INTO benchmark_restore_status (databasename, restore_done) VALUES ('${{ env.DATABASE_NAME }}', true)
+        ON CONFLICT (databasename) DO UPDATE SET restore_done = true;
+        "

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

@@ -223,9 +223,9 @@ jobs:
           # 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="${{ secrets.AZURE_STORAGE_ACCOUNT_DEV_NEW }}"
-          export AZURE_STORAGE_ACCESS_KEY="${{ secrets.AZURE_STORAGE_ACCESS_KEY_DEV_NEW }}"
-          export REMOTE_STORAGE_AZURE_CONTAINER="${{ vars.REMOTE_STORAGE_AZURE_CONTAINER_NEW }}"
+          export AZURE_STORAGE_ACCOUNT="${{ 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)'
 

.github/workflows/benchmarking.yml

@@ -56,6 +56,10 @@ 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:
@@ -63,9 +67,13 @@ jobs:
           - 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"
@@ -76,14 +84,21 @@ jobs:
       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
 
     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:
@@ -147,7 +162,7 @@ jobs:
     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: 14
+      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' ) }}
@@ -161,6 +176,7 @@ jobs:
     steps:
     - uses: actions/checkout@v4
 
+    
     - name: Download Neon artifact
       uses: ./.github/actions/download
       with:
@@ -168,7 +184,7 @@ jobs:
         path: /tmp/neon/
         prefix: latest
 
-    - name: Run benchmark
+    - name: Run Logical Replication benchmarks
       uses: ./.github/actions/run-python-test-set
       with:
         build_type: ${{ env.BUILD_TYPE }}
@@ -176,12 +192,15 @@ jobs:
         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 benchmark
+    - name: Run Physical Replication benchmarks
       uses: ./.github/actions/run-python-test-set
       with:
         build_type: ${{ env.BUILD_TYPE }}
@@ -234,6 +253,9 @@ jobs:
       id: pgbench-compare-matrix
       run: |
         region_id_default=${{ 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
@@ -247,16 +269,20 @@ jobs:
             "neonvm-captest-new"
           ],
           "db_size": [ "10gb" ],
-          "include": [{ "pg_version": 16, "region_id": "'"$region_id_default"'", "platform": "neonvm-captest-freetier",       "db_size": "3gb"  },
-                      { "pg_version": 16, "region_id": "'"$region_id_default"'", "platform": "neonvm-captest-new",            "db_size": "50gb" },
-                      { "pg_version": 16, "region_id": "azure-eastus2",          "platform": "neonvm-azure-captest-freetier", "db_size": "3gb"  },
-                      { "pg_version": 16, "region_id": "azure-eastus2",          "platform": "neonvm-azure-captest-new",      "db_size": "10gb" },
-                      { "pg_version": 16, "region_id": "azure-eastus2",          "platform": "neonvm-azure-captest-new",      "db_size": "50gb" },
-                      { "pg_version": 16, "region_id": "'"$region_id_default"'", "platform": "neonvm-captest-sharding-reuse", "db_size": "50gb" }]
+          "runner": ['"$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 "$matrix" | jq '.include += [{ "pg_version": 14, "region_id": "'"$region_id_default"'", "platform": "rds-postgres", "db_size": "10gb"}]')
+        if [ "$(date +%A)" = "Saturday" ] || [ ${RUN_AWS_RDS_AND_AURORA} = "true" ]; then
+          matrix=$(echo "$matrix" | jq '.include += [{ "pg_version": 16, "region_id": "'"$region_id_default"'", "platform": "rds-postgres", "db_size": "10gb","runner": '"$runner_default"', "image": "'"$image_default"'" },
+                                                     { "pg_version": 16, "region_id": "'"$region_id_default"'", "platform": "rds-aurora", "db_size": "10gb","runner": '"$runner_default"', "image": "'"$image_default"'" }]')
         fi
 
         echo "matrix=$(echo "$matrix" | jq --compact-output '.')" >> $GITHUB_OUTPUT
@@ -296,9 +322,17 @@ jobs:
 
         echo "matrix=$(echo "$matrix" | jq --compact-output '.')" >> $GITHUB_OUTPUT
 
+  prepare_AWS_RDS_databases:
+    uses: ./.github/workflows/_benchmarking_preparation.yml
+    secrets: inherit
+  
   pgbench-compare:
     if: ${{ github.event.inputs.run_only_pgvector_tests == 'false' || github.event.inputs.run_only_pgvector_tests == null }}
-    needs: [ generate-matrices ]
+    needs: [ generate-matrices, prepare_AWS_RDS_databases ]
+    permissions:
+      contents: write
+      statuses: write
+      id-token: write # Required for OIDC authentication in azure runners
 
     strategy:
       fail-fast: false
@@ -314,9 +348,9 @@ jobs:
       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
@@ -325,6 +359,13 @@ 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:
@@ -432,12 +473,20 @@ 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"
@@ -450,9 +499,9 @@ jobs:
       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
 
     steps:
@@ -463,12 +512,12 @@ jobs:
     - name: Install postgresql-16 where pytest expects it
       run: |
         cd /home/nonroot
-        wget -q https://apt.postgresql.org/pub/repos/apt/pool/main/p/postgresql-16/libpq5_16.3-1.pgdg110%2B1_amd64.deb
-        wget -q https://apt.postgresql.org/pub/repos/apt/pool/main/p/postgresql-16/postgresql-client-16_16.3-1.pgdg110%2B1_amd64.deb
-        wget -q https://apt.postgresql.org/pub/repos/apt/pool/main/p/postgresql-16/postgresql-16_16.3-1.pgdg110%2B1_amd64.deb 
-        dpkg -x libpq5_16.3-1.pgdg110+1_amd64.deb pg
-        dpkg -x postgresql-client-16_16.3-1.pgdg110+1_amd64.deb pg
-        dpkg -x postgresql-16_16.3-1.pgdg110+1_amd64.deb pg
+        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  
@@ -493,6 +542,13 @@ jobs:
         esac
 
         echo "connstr=${CONNSTR}" >> $GITHUB_OUTPUT
+        
+    - 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
@@ -521,7 +577,7 @@ jobs:
         BENCHMARK_CONNSTR: ${{ steps.set-up-connstr.outputs.connstr }}
         VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
         PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
-
+    
     - name: Create Allure report
       if: ${{ !cancelled() }}
       uses: ./.github/actions/allure-report-generate
@@ -544,7 +600,7 @@ jobs:
     # *_CLICKBENCH_CONNSTR: Genuine ClickBench DB with ~100M rows
     # *_CLICKBENCH_10M_CONNSTR: DB with the first 10M rows of ClickBench DB
     if: ${{ !cancelled() && (github.event.inputs.run_only_pgvector_tests == 'false' || github.event.inputs.run_only_pgvector_tests == null) }}
-    needs: [ generate-matrices, pgbench-compare ]
+    needs: [ generate-matrices, pgbench-compare, prepare_AWS_RDS_databases ]
 
     strategy:
       fail-fast: false
@@ -552,7 +608,7 @@ jobs:
 
     env:
       POSTGRES_DISTRIB_DIR: /tmp/neon/pg_install
-      DEFAULT_PG_VERSION: 14
+      DEFAULT_PG_VERSION: 16
       TEST_OUTPUT: /tmp/test_output
       TEST_OLAP_COLLECT_EXPLAIN: ${{ github.event.inputs.collect_olap_explain }}
       TEST_OLAP_COLLECT_PG_STAT_STATEMENTS: ${{ github.event.inputs.collect_pg_stat_statements }}
@@ -604,6 +660,7 @@ jobs:
         run_in_parallel: false
         save_perf_report: ${{ env.SAVE_PERF_REPORT }}
         extra_params: -m remote_cluster --timeout 21600 -k test_clickbench
+        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 }}"
@@ -633,7 +690,7 @@ jobs:
     #
     # *_TPCH_S10_CONNSTR: DB generated with scale factor 10 (~10 GB)
     if: ${{ !cancelled() && (github.event.inputs.run_only_pgvector_tests == 'false' || github.event.inputs.run_only_pgvector_tests == null) }}
-    needs: [ generate-matrices, clickbench-compare ]
+    needs: [ generate-matrices, clickbench-compare, prepare_AWS_RDS_databases ]
 
     strategy:
       fail-fast: false
@@ -641,7 +698,7 @@ jobs:
 
     env:
       POSTGRES_DISTRIB_DIR: /tmp/neon/pg_install
-      DEFAULT_PG_VERSION: 14
+      DEFAULT_PG_VERSION: 16
       TEST_OUTPUT: /tmp/test_output
       BUILD_TYPE: remote
       SAVE_PERF_REPORT: ${{ github.event.inputs.save_perf_report || ( github.ref_name == 'main' ) }}
@@ -673,7 +730,7 @@ jobs:
             ENV_PLATFORM=RDS_AURORA_TPCH
             ;;
           rds-postgres)
-            ENV_PLATFORM=RDS_AURORA_TPCH
+            ENV_PLATFORM=RDS_POSTGRES_TPCH
             ;;
           *)
             echo >&2 "Unknown PLATFORM=${PLATFORM}. Allowed only 'neonvm-captest-reuse', 'rds-aurora', or 'rds-postgres'"
@@ -699,6 +756,7 @@ jobs:
         run_in_parallel: false
         save_perf_report: ${{ env.SAVE_PERF_REPORT }}
         extra_params: -m remote_cluster --timeout 21600 -k test_tpch
+        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 }}"
@@ -720,7 +778,7 @@ jobs:
 
   user-examples-compare:
     if: ${{ !cancelled() && (github.event.inputs.run_only_pgvector_tests == 'false' || github.event.inputs.run_only_pgvector_tests == null) }}
-    needs: [ generate-matrices, tpch-compare ]
+    needs: [ generate-matrices, tpch-compare, prepare_AWS_RDS_databases ]
 
     strategy:
       fail-fast: false
@@ -728,7 +786,7 @@ jobs:
 
     env:
       POSTGRES_DISTRIB_DIR: /tmp/neon/pg_install
-      DEFAULT_PG_VERSION: 14
+      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' ) }}

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

@@ -56,13 +56,7 @@ jobs:
 
       - uses: actions/checkout@v4
 
-      # Use custom DOCKER_CONFIG directory to avoid conflicts with default settings
-      # The default value is ~/.docker
-      - name: Set custom docker config directory
-        run: |
-          mkdir -p /tmp/.docker-custom
-          echo DOCKER_CONFIG=/tmp/.docker-custom >> $GITHUB_ENV
-
+      - uses: ./.github/actions/set-docker-config-dir
       - uses: docker/setup-buildx-action@v3
         with:
           cache-binary: false
@@ -89,11 +83,6 @@ jobs:
           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
-
   merge-images:
     needs: [ build-image ]
     runs-on: ubuntu-22.04

.github/workflows/build_and_test.yml

@@ -309,7 +309,7 @@ jobs:
         SLACK_BOT_TOKEN: ${{ secrets.SLACK_BOT_TOKEN }}
 
   create-test-report:
-    needs: [ check-permissions, build-and-test-locally, coverage-report, 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 }}
@@ -484,12 +484,7 @@ jobs:
           submodules: true
           fetch-depth: 0
 
-      # Use custom DOCKER_CONFIG directory to avoid conflicts with default settings
-      # The default value is ~/.docker
-      - name: Set custom docker config directory
-        run: |
-          mkdir -p .docker-custom
-          echo DOCKER_CONFIG=$(pwd)/.docker-custom >> $GITHUB_ENV
+      - uses: ./.github/actions/set-docker-config-dir
       - uses: docker/setup-buildx-action@v3
         with:
           cache-binary: false
@@ -521,11 +516,6 @@ jobs:
           tags: |
             neondatabase/neon:${{ needs.tag.outputs.build-tag }}-${{ matrix.arch }}
 
-      - name: Remove custom docker config directory
-        if: always()
-        run: |
-          rm -rf .docker-custom
-
   neon-image:
     needs: [ neon-image-arch, tag ]
     runs-on: ubuntu-22.04
@@ -570,12 +560,7 @@ jobs:
           submodules: true
           fetch-depth: 0
 
-      # Use custom DOCKER_CONFIG directory to avoid conflicts with default settings
-      # The default value is ~/.docker
-      - name: Set custom docker config directory
-        run: |
-          mkdir -p .docker-custom
-          echo DOCKER_CONFIG=$(pwd)/.docker-custom >> $GITHUB_ENV
+      - uses: ./.github/actions/set-docker-config-dir
       - uses: docker/setup-buildx-action@v3
         with:
           cache-binary: false
@@ -658,11 +643,6 @@ jobs:
           tags: |
             neondatabase/compute-tools:${{ needs.tag.outputs.build-tag }}-${{ matrix.arch }}
 
-      - name: Remove custom docker config directory
-        if: always()
-        run: |
-          rm -rf .docker-custom
-
   compute-node-image:
     needs: [ compute-node-image-arch, tag ]
     runs-on: ubuntu-22.04
@@ -735,13 +715,7 @@ jobs:
           curl -fL https://github.com/neondatabase/autoscaling/releases/download/$VM_BUILDER_VERSION/vm-builder -o vm-builder
           chmod +x vm-builder
 
-      # Use custom DOCKER_CONFIG directory to avoid conflicts with default settings
-      # The default value is ~/.docker
-      - name: Set custom docker config directory
-        run: |
-          mkdir -p .docker-custom
-          echo DOCKER_CONFIG=$(pwd)/.docker-custom >> $GITHUB_ENV
-
+      - uses: ./.github/actions/set-docker-config-dir
       - uses: docker/login-action@v3
         with:
           username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
@@ -764,11 +738,6 @@ jobs:
         run: |
           docker push neondatabase/vm-compute-node-${{ matrix.version }}:${{ needs.tag.outputs.build-tag }}
 
-      - name: Remove custom docker config directory
-        if: always()
-        run: |
-          rm -rf .docker-custom
-
   test-images:
     needs: [ check-permissions, tag, neon-image, compute-node-image ]
     strategy:
@@ -784,13 +753,7 @@ jobs:
         with:
           fetch-depth: 0
 
-      # Use custom DOCKER_CONFIG directory to avoid conflicts with default settings
-      # The default value is ~/.docker
-      - name: Set custom docker config directory
-        run: |
-          mkdir -p .docker-custom
-          echo DOCKER_CONFIG=$(pwd)/.docker-custom >> $GITHUB_ENV
-
+      - uses: ./.github/actions/set-docker-config-dir
       - uses: docker/login-action@v3
         with:
           username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
@@ -830,11 +793,6 @@ jobs:
           docker compose -f ./docker-compose/docker-compose.yml logs || 0
           docker compose -f ./docker-compose/docker-compose.yml down
 
-      - name: Remove custom docker config directory
-        if: always()
-        run: |
-          rm -rf .docker-custom
-
   promote-images:
     permissions:
       contents: read  # This is required for actions/checkout
@@ -871,7 +829,7 @@ jobs:
         with:
           client-id: ${{ secrets.AZURE_DEV_CLIENT_ID }}
           tenant-id: ${{ secrets.AZURE_TENANT_ID }}
-          subscription-id: ${{ secrets.AZURE_SUBSCRIPTION_ID }}
+          subscription-id: ${{ secrets.AZURE_DEV_SUBSCRIPTION_ID }}
 
       - name: Login to ACR
         if: github.ref_name == 'main'

.github/workflows/label-for-external-users.yml

@@ -0,0 +1,35 @@
+name: Add `external` label to issues and PRs created by external users
+
+on:
+  issues:
+    types:
+      - opened
+  pull_request:
+    types:
+      - opened
+
+# No permission for GITHUB_TOKEN by default; the **minimal required** set of permissions should be granted in each job.
+permissions: {}
+
+env:
+  LABEL: external
+
+jobs:
+  add-label:
+    # This workflow uses `author_association` for PRs and issues to determine if the user is an external user.
+    # Possible values for `author_association`: https://docs.github.com/en/graphql/reference/enums#commentauthorassociation
+    if: ${{ !contains(fromJSON('["OWNER", "MEMBER", "COLLABORATOR"]'), github.event[github.event_name == 'pull_request' && 'pull_request' || 'issue'].author_association) }}
+
+    runs-on: ubuntu-22.04
+    permissions:
+      pull-requests: write
+      issues: write
+
+    steps:
+    - name: Label new ${{ github.event_name }}
+      env:
+        GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        ITEM_NUMBER: ${{ github.event[github.event_name == 'pull_request' && 'pull_request' || 'issue'].number }}
+        GH_CLI_COMMAND: ${{ github.event_name == 'pull_request' && 'pr' || 'issue' }}
+      run: |
+        gh ${GH_CLI_COMMAND} --repo ${GITHUB_REPOSITORY} edit --add-label=${LABEL} ${ITEM_NUMBER}

.github/workflows/neon_extra_builds.yml

@@ -149,8 +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

.github/workflows/pg-clients.yml

@@ -66,7 +66,31 @@ jobs:
         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
 

.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

Cargo.lock

@@ -1418,7 +1418,7 @@ dependencies = [
  "clap",
  "criterion-plot",
  "is-terminal",
- "itertools",
+ "itertools 0.10.5",
  "num-traits",
  "once_cell",
  "oorandom",
@@ -1439,7 +1439,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "6b50826342786a51a89e2da3a28f1c32b06e387201bc2d19791f622c673706b1"
 dependencies = [
  "cast",
- "itertools",
+ "itertools 0.10.5",
 ]
 
 [[package]]
@@ -2134,6 +2134,12 @@ dependencies = [
  "slab",
 ]
 
+[[package]]
+name = "gen_ops"
+version = "0.4.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "304de19db7028420975a296ab0fcbbc8e69438c4ed254a1e41e2a7f37d5f0e0a"
+
 [[package]]
 name = "generic-array"
 version = "0.14.7"
@@ -2710,17 +2716,6 @@ version = "3.0.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "8bb03732005da905c88227371639bf1ad885cc712789c011c31c5fb3ab3ccf02"
 
-[[package]]
-name = "io-lifetimes"
-version = "1.0.11"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "eae7b9aee968036d54dce06cebaefd919e4472e753296daccd6d344e3e2df0c2"
-dependencies = [
- "hermit-abi",
- "libc",
- "windows-sys 0.48.0",
-]
-
 [[package]]
 name = "io-uring"
 version = "0.6.2"
@@ -2739,14 +2734,13 @@ checksum = "8f518f335dce6725a761382244631d86cf0ccb2863413590b31338feb467f9c3"
 
 [[package]]
 name = "is-terminal"
-version = "0.4.7"
+version = "0.4.12"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "adcf93614601c8129ddf72e2d5633df827ba6551541c6d8c59520a371475be1f"
+checksum = "f23ff5ef2b80d608d61efee834934d862cd92461afc0560dedf493e4c033738b"
 dependencies = [
  "hermit-abi",
- "io-lifetimes",
- "rustix 0.37.25",
- "windows-sys 0.48.0",
+ "libc",
+ "windows-sys 0.52.0",
 ]
 
 [[package]]
@@ -2758,6 +2752,15 @@ dependencies = [
  "either",
 ]
 
+[[package]]
+name = "itertools"
+version = "0.12.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ba291022dbbd398a455acf126c1e341954079855bc60dfdda641363bd6922569"
+dependencies = [
+ "either",
+]
+
 [[package]]
 name = "itoa"
 version = "1.0.6"
@@ -2872,18 +2875,6 @@ version = "0.2.8"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "4ec2a862134d2a7d32d7983ddcdd1c4923530833c9f2ea1a44fc5fa473989058"
 
-[[package]]
-name = "linux-raw-sys"
-version = "0.1.4"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "f051f77a7c8e6957c0696eac88f26b0117e54f52d3fc682ab19397a8812846a4"
-
-[[package]]
-name = "linux-raw-sys"
-version = "0.3.8"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ef53942eb7bf7ff43a617b3e2c1c4a5ecf5944a7c1bc12d7ee39bbb15e5c1519"
-
 [[package]]
 name = "linux-raw-sys"
 version = "0.4.13"
@@ -3001,7 +2992,7 @@ checksum = "7c4b80445aeb08e832d87bf1830049a924cdc1d6b7ef40b6b9b365bff17bf8ec"
 dependencies = [
  "libc",
  "measured",
- "procfs 0.16.0",
+ "procfs",
 ]
 
 [[package]]
@@ -3046,7 +3037,7 @@ dependencies = [
  "measured",
  "measured-process",
  "once_cell",
- "procfs 0.14.2",
+ "procfs",
  "prometheus",
  "rand 0.8.5",
  "rand_distr",
@@ -3575,7 +3566,7 @@ dependencies = [
  "humantime",
  "humantime-serde",
  "hyper 0.14.26",
- "itertools",
+ "itertools 0.10.5",
  "leaky-bucket",
  "md5",
  "metrics",
@@ -3593,8 +3584,9 @@ dependencies = [
  "postgres_connection",
  "postgres_ffi",
  "pq_proto",
- "procfs 0.14.2",
+ "procfs",
  "rand 0.8.5",
+ "range-set-blaze",
  "regex",
  "remote_storage",
  "reqwest 0.12.4",
@@ -3645,7 +3637,7 @@ dependencies = [
  "hex",
  "humantime",
  "humantime-serde",
- "itertools",
+ "itertools 0.10.5",
  "postgres_ffi",
  "rand 0.8.5",
  "serde",
@@ -3703,7 +3695,7 @@ dependencies = [
  "hex-literal",
  "humantime",
  "humantime-serde",
- "itertools",
+ "itertools 0.10.5",
  "metrics",
  "once_cell",
  "pageserver_api",
@@ -3968,7 +3960,7 @@ dependencies = [
 [[package]]
 name = "postgres"
 version = "0.19.4"
-source = "git+https://github.com/neondatabase/rust-postgres.git?branch=neon#cff6927e4f58b1af6ecc2ee7279df1f2ff537295"
+source = "git+https://github.com/neondatabase/rust-postgres.git?branch=neon#20031d7a9ee1addeae6e0968e3899ae6bf01cee2"
 dependencies = [
  "bytes",
  "fallible-iterator",
@@ -3981,7 +3973,7 @@ dependencies = [
 [[package]]
 name = "postgres-protocol"
 version = "0.6.4"
-source = "git+https://github.com/neondatabase/rust-postgres.git?branch=neon#cff6927e4f58b1af6ecc2ee7279df1f2ff537295"
+source = "git+https://github.com/neondatabase/rust-postgres.git?branch=neon#20031d7a9ee1addeae6e0968e3899ae6bf01cee2"
 dependencies = [
  "base64 0.20.0",
  "byteorder",
@@ -4000,7 +3992,7 @@ dependencies = [
 [[package]]
 name = "postgres-types"
 version = "0.2.4"
-source = "git+https://github.com/neondatabase/rust-postgres.git?branch=neon#cff6927e4f58b1af6ecc2ee7279df1f2ff537295"
+source = "git+https://github.com/neondatabase/rust-postgres.git?branch=neon#20031d7a9ee1addeae6e0968e3899ae6bf01cee2"
 dependencies = [
  "bytes",
  "fallible-iterator",
@@ -4035,7 +4027,7 @@ name = "postgres_connection"
 version = "0.1.0"
 dependencies = [
  "anyhow",
- "itertools",
+ "itertools 0.10.5",
  "once_cell",
  "postgres",
  "tokio-postgres",
@@ -4093,7 +4085,7 @@ version = "0.1.0"
 dependencies = [
  "byteorder",
  "bytes",
- "itertools",
+ "itertools 0.10.5",
  "pin-project-lite",
  "postgres-protocol",
  "rand 0.8.5",
@@ -4139,21 +4131,6 @@ dependencies = [
  "unicode-ident",
 ]
 
-[[package]]
-name = "procfs"
-version = "0.14.2"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "b1de8dacb0873f77e6aefc6d71e044761fcc68060290f5b1089fcdf84626bb69"
-dependencies = [
- "bitflags 1.3.2",
- "byteorder",
- "chrono",
- "flate2",
- "hex",
- "lazy_static",
- "rustix 0.36.16",
-]
-
 [[package]]
 name = "procfs"
 version = "0.16.0"
@@ -4161,10 +4138,12 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "731e0d9356b0c25f16f33b5be79b1c57b562f141ebfcdb0ad8ac2c13a24293b4"
 dependencies = [
  "bitflags 2.4.1",
+ "chrono",
+ "flate2",
  "hex",
  "lazy_static",
  "procfs-core",
- "rustix 0.38.28",
+ "rustix",
 ]
 
 [[package]]
@@ -4174,14 +4153,15 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "2d3554923a69f4ce04c4a754260c338f505ce22642d3830e049a399fc2059a29"
 dependencies = [
  "bitflags 2.4.1",
+ "chrono",
  "hex",
 ]
 
 [[package]]
 name = "prometheus"
-version = "0.13.3"
+version = "0.13.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "449811d15fbdf5ceb5c1144416066429cf82316e2ec8ce0c1f6f8a02e7bbcf8c"
+checksum = "3d33c28a30771f7f96db69893f78b857f7450d7e0237e9c8fc6427a81bae7ed1"
 dependencies = [
  "cfg-if",
  "fnv",
@@ -4189,7 +4169,7 @@ dependencies = [
  "libc",
  "memchr",
  "parking_lot 0.12.1",
- "procfs 0.14.2",
+ "procfs",
  "thiserror",
 ]
 
@@ -4211,7 +4191,7 @@ checksum = "119533552c9a7ffacc21e099c24a0ac8bb19c2a2a3f363de84cd9b844feab270"
 dependencies = [
  "bytes",
  "heck 0.4.1",
- "itertools",
+ "itertools 0.10.5",
  "lazy_static",
  "log",
  "multimap",
@@ -4232,7 +4212,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "e5d2d8d10f3c6ded6da8b05b5fb3b8a5082514344d56c9f871412d29b4e075b4"
 dependencies = [
  "anyhow",
- "itertools",
+ "itertools 0.10.5",
  "proc-macro2",
  "quote",
  "syn 1.0.109",
@@ -4289,7 +4269,7 @@ dependencies = [
  "hyper-util",
  "indexmap 2.0.1",
  "ipnet",
- "itertools",
+ "itertools 0.10.5",
  "lasso",
  "md5",
  "measured",
@@ -4344,6 +4324,7 @@ dependencies = [
  "tracing-opentelemetry",
  "tracing-subscriber",
  "tracing-utils",
+ "try-lock",
  "typed-json",
  "url",
  "urlencoding",
@@ -4465,6 +4446,18 @@ dependencies = [
  "rand_core 0.5.1",
 ]
 
+[[package]]
+name = "range-set-blaze"
+version = "0.1.16"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8421b5d459262eabbe49048d362897ff3e3830b44eac6cfe341d6acb2f0f13d2"
+dependencies = [
+ "gen_ops",
+ "itertools 0.12.1",
+ "num-integer",
+ "num-traits",
+]
+
 [[package]]
 name = "rayon"
 version = "1.7.0"
@@ -4633,7 +4626,7 @@ dependencies = [
  "humantime",
  "humantime-serde",
  "hyper 0.14.26",
- "itertools",
+ "itertools 0.10.5",
  "metrics",
  "once_cell",
  "pin-project-lite",
@@ -4943,34 +4936,6 @@ dependencies = [
  "nom",
 ]
 
-[[package]]
-name = "rustix"
-version = "0.36.16"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "6da3636faa25820d8648e0e31c5d519bbb01f72fdf57131f0f5f7da5fed36eab"
-dependencies = [
- "bitflags 1.3.2",
- "errno",
- "io-lifetimes",
- "libc",
- "linux-raw-sys 0.1.4",
- "windows-sys 0.45.0",
-]
-
-[[package]]
-name = "rustix"
-version = "0.37.25"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "d4eb579851244c2c03e7c24f501c3432bed80b8f720af1d6e5b0e0f01555a035"
-dependencies = [
- "bitflags 1.3.2",
- "errno",
- "io-lifetimes",
- "libc",
- "linux-raw-sys 0.3.8",
- "windows-sys 0.48.0",
-]
-
 [[package]]
 name = "rustix"
 version = "0.38.28"
@@ -5730,7 +5695,7 @@ dependencies = [
  "hex",
  "humantime",
  "hyper 0.14.26",
- "itertools",
+ "itertools 0.10.5",
  "lasso",
  "measured",
  "metrics",
@@ -5739,6 +5704,7 @@ dependencies = [
  "pageserver_client",
  "postgres_connection",
  "r2d2",
+ "rand 0.8.5",
  "reqwest 0.12.4",
  "routerify",
  "scopeguard",
@@ -5794,9 +5760,10 @@ dependencies = [
  "either",
  "futures",
  "futures-util",
+ "git-version",
  "hex",
  "humantime",
- "itertools",
+ "itertools 0.10.5",
  "once_cell",
  "pageserver",
  "pageserver_api",
@@ -5973,15 +5940,15 @@ dependencies = [
 
 [[package]]
 name = "tempfile"
-version = "3.5.0"
+version = "3.9.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "b9fbec84f381d5795b08656e4912bec604d162bff9291d6189a78f4c8ab87998"
+checksum = "01ce4141aa927a6d1bd34a041795abd0db1cccba5d5f24b009f694bdf3a1f3fa"
 dependencies = [
  "cfg-if",
- "fastrand 1.9.0",
- "redox_syscall 0.3.5",
- "rustix 0.37.25",
- "windows-sys 0.45.0",
+ "fastrand 2.0.0",
+ "redox_syscall 0.4.1",
+ "rustix",
+ "windows-sys 0.52.0",
 ]
 
 [[package]]
@@ -6220,7 +6187,7 @@ dependencies = [
 [[package]]
 name = "tokio-postgres"
 version = "0.7.7"
-source = "git+https://github.com/neondatabase/rust-postgres.git?branch=neon#cff6927e4f58b1af6ecc2ee7279df1f2ff537295"
+source = "git+https://github.com/neondatabase/rust-postgres.git?branch=neon#20031d7a9ee1addeae6e0968e3899ae6bf01cee2"
 dependencies = [
  "async-trait",
  "byteorder",
@@ -6597,9 +6564,9 @@ dependencies = [
 
 [[package]]
 name = "try-lock"
-version = "0.2.4"
+version = "0.2.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "3528ecfd12c466c6f163363caf2d02a71161dd5e1cc6ae7b34207ea2d42d81ed"
+checksum = "e421abadd41a4225275504ea4d6566923418b7f05506fbc9c0fe86ba7396114b"
 
 [[package]]
 name = "tungstenite"
@@ -7178,15 +7145,6 @@ dependencies = [
  "windows_x86_64_msvc 0.42.2",
 ]
 
-[[package]]
-name = "windows-sys"
-version = "0.45.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "75283be5efb2831d37ea142365f009c02ec203cd29a3ebecbc093d52315b66d0"
-dependencies = [
- "windows-targets 0.42.2",
-]
-
 [[package]]
 name = "windows-sys"
 version = "0.48.0"
@@ -7205,21 +7163,6 @@ dependencies = [
  "windows-targets 0.52.4",
 ]
 
-[[package]]
-name = "windows-targets"
-version = "0.42.2"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "8e5180c00cd44c9b1c88adb3693291f1cd93605ded80c250a75d472756b4d071"
-dependencies = [
- "windows_aarch64_gnullvm 0.42.2",
- "windows_aarch64_msvc 0.42.2",
- "windows_i686_gnu 0.42.2",
- "windows_i686_msvc 0.42.2",
- "windows_x86_64_gnu 0.42.2",
- "windows_x86_64_gnullvm 0.42.2",
- "windows_x86_64_msvc 0.42.2",
-]
-
 [[package]]
 name = "windows-targets"
 version = "0.48.0"
@@ -7449,7 +7392,7 @@ dependencies = [
  "hmac",
  "hyper 0.14.26",
  "indexmap 1.9.3",
- "itertools",
+ "itertools 0.10.5",
  "libc",
  "log",
  "memchr",

Cargo.toml

@@ -126,7 +126,7 @@ 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"
@@ -184,6 +184,7 @@ 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"] }
+try-lock = "0.2.5"
 twox-hash = { version = "1.6.3", default-features = false }
 typed-json = "0.1"
 url = "2.2"

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 \
@@ -29,24 +29,12 @@ 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 \
     && 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  \
@@ -58,8 +46,7 @@ RUN set -e \
       --bin proxy  \
       --bin neon_local \
       --bin storage_scrubber \
-      --locked --release \
-    && cachepot -s
+      --locked --release
 
 # Build final image
 #
@@ -104,7 +91,7 @@ RUN mkdir -p /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"]
@@ -112,5 +99,5 @@ USER neon
 EXPOSE 6400
 EXPOSE 9898
 
-CMD /usr/local/bin/pageserver -D /data/.neon
+CMD ["/usr/local/bin/pageserver", "-D", "/data/.neon"]
 

Dockerfile.build-tools

@@ -58,7 +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-$(uname -m | sed 's/aarch64/aarch_64/g').zip" -o "protoc.zip" \
     && unzip -q protoc.zip -d protoc \
     && mv protoc/bin/protoc /usr/local/bin/protoc \
@@ -99,7 +99,7 @@ RUN curl "https://awscli.amazonaws.com/awscli-exe-linux-$(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 \
@@ -168,7 +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 \
@@ -192,9 +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=1.80.0
+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 ${RUSTC_VERSION} && \
@@ -203,15 +208,13 @@ RUN curl -sSO https://static.rust-lang.org/rustup/dist/$(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 && \
@@ -411,7 +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") \
@@ -444,7 +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 "${PG_VERSION}" in \
       "v14") \
@@ -480,7 +480,7 @@ RUN case "${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 . && \
@@ -506,7 +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 . && \
@@ -546,7 +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 . && \
@@ -563,7 +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 . && \
@@ -580,7 +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 . && \
@@ -598,7 +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 "${PG_VERSION}" in \
       "v14" | "v15") \
         export PG_EMBEDDING_VERSION=0.3.5 \
@@ -622,7 +622,7 @@ RUN case "${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 . && \
@@ -750,7 +750,7 @@ RUN wget https://github.com/pksunkara/pgx_ulid/archive/refs/tags/v0.1.5.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 . && \
@@ -766,7 +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 . && \
@@ -783,7 +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 . && \
@@ -933,7 +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
@@ -945,7 +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
@@ -960,6 +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
@@ -1032,6 +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"]

README.md

@@ -313,5 +313,3 @@ To get more familiar with this aspect, refer to:
 - Read [CONTRIBUTING.md](/CONTRIBUTING.md) to learn about project code style and practices.
 - To get familiar with a source tree layout, use [sourcetree.md](/docs/sourcetree.md).
 - To learn more about PostgreSQL internals, check http://www.interdb.jp/pg/index.html
-
-.

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

compute_tools/src/compute.rs

@@ -400,7 +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 {

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/endpoint.rs

@@ -824,11 +824,12 @@ impl Endpoint {
         // cleanup work to do after postgres stops, like syncing safekeepers,
         // etc.
         //
-        // If destroying, send it SIGTERM before waiting. Sometimes we do *not*
-        // want this cleanup: tests intentionally do stop when majority of
-        // safekeepers is down, so sync-safekeepers would hang otherwise. This
-        // could be a separate flag though.
-        self.wait_for_compute_ctl_to_exit(destroy)?;
+        // If destroying or stop mode is immediate, send it SIGTERM before
+        // waiting. Sometimes we do *not* want this cleanup: tests intentionally
+        // do stop when majority of safekeepers is down, so sync-safekeepers
+        // would hang otherwise. This could be a separate flag though.
+        let send_sigterm = destroy || mode == "immediate";
+        self.wait_for_compute_ctl_to_exit(send_sigterm)?;
         if destroy {
             println!(
                 "Destroying postgres data directory '{}'",

control_plane/src/local_env.rs

@@ -158,6 +158,8 @@ pub struct NeonStorageControllerConf {
 
     /// Threshold for auto-splitting a tenant into shards
     pub split_threshold: Option<u64>,
+
+    pub max_secondary_lag_bytes: Option<u64>,
 }
 
 impl NeonStorageControllerConf {
@@ -173,6 +175,7 @@ impl Default for NeonStorageControllerConf {
             max_offline: Self::DEFAULT_MAX_OFFLINE_INTERVAL,
             max_warming_up: Self::DEFAULT_MAX_WARMING_UP_INTERVAL,
             split_threshold: None,
+            max_secondary_lag_bytes: None,
         }
     }
 }

control_plane/src/storage_controller.rs

@@ -383,6 +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()

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 +21,13 @@ 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 = []
 
 # 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 +40,6 @@ 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/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/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/035-safekeeper-dynamic-membership-change.md

@@ -0,0 +1,495 @@
+# Safekeeper dynamic membership change
+
+To quickly recover from safekeeper node failures and do rebalancing we need to
+be able to change set of safekeepers the timeline resides on. The procedure must
+be safe (not lose committed log) regardless of safekeepers and compute state. It
+should be able to progress if any majority of old safekeeper set, any majority
+of new safekeeper set and compute are up and connected. This is known as a
+consensus membership change. It always involves two phases: 1) switch old
+majority to old + new configuration, preventing commits without acknowledge from
+the new set 2) bootstrap the new set by ensuring majority of the new set has all
+data which ever could have been committed before the first phase completed;
+after that switch is safe to finish. Without two phases switch to the new set
+which quorum might not intersect with quorum of the old set (and typical case of
+ABC -> ABD switch is an example of that, because quorums AC and BD don't
+intersect). Furthermore, procedure is typically carried out by the consensus
+leader, and so enumeration of configurations which establishes order between
+them is done through consensus log.
+
+In our case consensus leader is compute (walproposer), and we don't want to wake
+up all computes for the change. Neither we want to fully reimplement the leader
+logic second time outside compute. Because of that the proposed algorithm relies
+for issuing configurations on the external fault tolerant (distributed) strongly
+consisent storage with simple API: CAS (compare-and-swap) on the single key.
+Properly configured postgres suits this.
+
+In the system consensus is implemented at the timeline level, so algorithm below
+applies to the single timeline.
+
+## Algorithm
+
+### Definitions
+
+A configuration is
+
+```
+struct Configuration {
+    generation: Generation, // a number uniquely identifying configuration
+    sk_set: Vec<NodeId>, // current safekeeper set
+    new_sk_set: Optional<Vec<NodeId>>,
+}
+```
+
+Configuration with `new_set` present is used for the intermediate step during
+the change and called joint configuration. Generations establish order of
+generations: we say `c1` is higher than `c2` if `c1.generation` >
+`c2.generation`.
+
+### Persistently stored data changes
+
+Safekeeper starts storing its current configuration in the control file. Update
+of is atomic, so in-memory value always matches the persistent one.
+
+External CAS providing storage (let's call it configuration storage here) also
+stores configuration for each timeline. It is initialized with generation 1 and
+initial set of safekeepers during timeline creation. Executed CAS on it must
+never be lost.
+
+### Compute <-> safekeeper protocol changes
+
+`ProposerGreeting` message carries walproposer's configuration if it is already
+established (see below), else null.  `AcceptorGreeting` message carries
+safekeeper's current `Configuration`. All further messages (`VoteRequest`,
+`VoteResponse`, `ProposerElected`, `AppendRequest`, `AppendResponse`) carry
+generation number, of walproposer in case of wp->sk message or of safekeeper in
+case of sk->wp message.
+
+### Safekeeper changes
+
+Basic rule: once safekeeper observes configuration higher than his own it
+immediately switches to it. It must refuse all messages with lower generation
+that his. It also refuses messages if it is not member of the current generation
+(that is, of either `sk_set` of `sk_new_set`), though it is likely not unsafe to
+process them (walproposer should ignore result anyway).
+
+If there is non null configuration in `ProposerGreeting` and it is higher than
+current safekeeper one, safekeeper switches to it.
+
+Safekeeper sends its current configuration in its first message to walproposer
+`AcceptorGreeting`. It refuses all other walproposer messages if the
+configuration generation in them is less than its current one. Namely, it
+refuses to vote, to truncate WAL in `handle_elected` and to accept WAL. In
+response it sends its current configuration generation to let walproposer know.
+
+Safekeeper gets `PUT /v1/tenants/{tenant_id}/timelines/{timeline_id}/configuration` 
+accepting `Configuration`. Safekeeper switches to the given conf it is higher than its
+current one and ignores it otherwise. In any case it replies with
+```
+struct ConfigurationSwitchResponse {
+    conf: Configuration,
+    term: Term,
+    last_log_term: Term,
+    flush_lsn: Lsn,
+}
+```
+
+### Compute (walproposer) changes
+
+Basic rule is that joint configuration requires votes from majorities in the
+both `set` and `new_sk_set`.
+
+Compute receives list of safekeepers to connect to from the control plane as
+currently and tries to communicate with all of them. However, the list does not
+define consensus members. Instead, on start walproposer tracks highest
+configuration it receives from `AcceptorGreeting`s. Once it assembles greetings
+from majority of `sk_set` and majority of `new_sk_set` (if it is present), it
+establishes this configuration as its own and moves to voting. 
+
+It should stop talking to safekeepers not listed in the configuration at this
+point, though it is not unsafe to continue doing so.
+
+To be elected it must receive votes from both majorites if `new_sk_set` is present.
+Similarly, to commit WAL it must receive flush acknowledge from both majorities.
+
+If walproposer hears from safekeeper configuration higher than his own (i.e.
+refusal to accept due to configuration change) it simply restarts.
+
+### Change algorithm
+
+The following algorithm can be executed anywhere having access to configuration
+storage and safekeepers. It is safe to interrupt / restart it and run multiple
+instances of it concurrently, though likely one of them won't make
+progress then. It accepts `desired_set: Vec<NodeId>` as input. 
+
+Algorithm will refuse to make the change if it encounters previous interrupted
+change attempt, but in this case it will try to finish it.
+
+It will eventually converge if old majority, new majority and configuration
+storage are reachable.
+
+1) Fetch current timeline configuration from the configuration storage.
+2) If it is already joint one and `new_set` is different from `desired_set`
+   refuse to change. However, assign join conf to (in memory) var
+   `join_conf` and proceed to step 4 to finish the ongoing change.
+3) Else, create joint `joint_conf: Configuration`: increment current conf number
+   `n` and put `desired_set` to `new_sk_set`. Persist it in the configuration
+   storage by doing CAS on the current generation: change happens only if
+   current configuration number is still `n`. Apart from guaranteeing uniqueness
+   of configurations, CAS linearizes them, ensuring that new configuration is
+   created only following the previous one when we know that the transition is
+   safe. Failed CAS aborts the procedure.
+4) Call `PUT` `configuration` on safekeepers from the current set,
+   delivering them `joint_conf`. Collecting responses from majority is required
+   to proceed. If any response returned generation higher than 
+   `joint_conf.generation`, abort (another switch raced us). Otherwise, choose
+   max `<last_log_term, flush_lsn>` among responses and establish it as
+   (in memory) `sync_position`. Also choose max `term` and establish it as (in
+   memory) `sync_term`. We can't finish the switch until majority of the new set
+   catches up to this `sync_position` because data before it could be committed
+   without ack from the new set. Similarly, we'll bump term on new majority
+   to `sync_term` so that two computes with the same term are never elected.
+4) Initialize timeline on safekeeper(s) from `new_sk_set` where it
+   doesn't exist yet by doing `pull_timeline` from the majority of the 
+   current set. Doing that on majority of `new_sk_set` is enough to
+   proceed, but it is reasonable to ensure that all `new_sk_set` members
+   are initialized -- if some of them are down why are we migrating there?
+5) Call `POST` `bump_term(sync_term)` on safekeepers from the new set. 
+   Success on majority is enough.
+6) Repeatedly call `PUT` `configuration` on safekeepers from the new set,
+   delivering them `joint_conf` and collecting their positions. This will
+   switch them to the `joint_conf` which generally won't be needed 
+   because `pull_timeline` already includes it and plus additionally would be
+   broadcast by compute. More importantly, we may proceed to the next step
+   only when `<last_log_term, flush_lsn>` on the majority of the new set reached 
+   `sync_position`. Similarly, on the happy path no waiting is not needed because 
+   `pull_timeline` already includes it. However, we should double
+    check to be safe. For example, timeline could have been created earlier e.g.
+    manually or after try-to-migrate, abort, try-to-migrate-again sequence. 
+7) Create `new_conf: Configuration` incrementing `join_conf` generation and having new 
+   safekeeper set as `sk_set` and None `new_sk_set`. Write it to configuration 
+   storage under one more CAS.
+8) Call `PUT` `configuration` on safekeepers from the new set,
+   delivering them `new_conf`. It is enough to deliver it to the majority 
+   of the new set; the rest can be updated by compute.
+
+I haven't put huge effort to make the description above very precise, because it
+is natural language prone to interpretations anyway. Instead I'd like to make TLA+
+spec of it.
+
+Description above focuses on safety. To make the flow practical and live, here a few more 
+considerations.
+1) It makes sense to ping new set to ensure it we are migrating to live node(s) before 
+  step 3.
+2) If e.g. accidentally wrong new sk set has been specified, before CAS in step `6` is completed 
+   it is safe to rollback to the old conf with one more CAS.
+3) On step 4 timeline might be already created on members of the new set for various reasons; 
+   the simplest is the procedure restart. There are more complicated scenarious like mentioned
+   in step 5. Deleting and re-doing `pull_timeline` is generally unsafe without involving 
+   generations, so seems simpler to treat existing timeline as success. However, this also 
+   has a disadvantage: you might imagine an surpassingly unlikely schedule where condition in
+   the step 5 is never reached until compute is (re)awaken up to synchronize new member(s).
+   I don't think we'll observe this in practice, but can add waking up compute if needed.
+4) In the end timeline should be locally deleted on the safekeeper(s) which are
+   in the old set but not in the new one, unless they are unreachable. To be
+   safe this also should be done under generation number (deletion proceeds only if 
+   current configuration is <= than one in request and safekeeper is not memeber of it).
+5) If current conf fetched on step 1 is already not joint and members equal to `desired_set`,
+   jump to step 7, using it as `new_conf`.
+
+## Implementation
+
+The procedure ought to be driven from somewhere. Obvious candidates are control
+plane and storage_controller; and as each of them already has db we don't want
+yet another storage. I propose to manage safekeepers in storage_controller
+because 1) since it is in rust it simplifies simulation testing (more on this
+below) 2) it already manages pageservers. 
+
+This assumes that migration will be fully usable only after we migrate all
+tenants/timelines to storage_controller. It is discussible whether we want also
+to manage pageserver attachments for all of these, but likely we do.
+
+This requires us to define storcon <-> cplane interface.
+
+### storage_controller <-> control plane interface
+
+First of all, control plane should
+[change](https://neondb.slack.com/archives/C03438W3FLZ/p1719226543199829)
+storing safekeepers per timeline instead of per tenant because we can't migrate
+tenants atomically. 
+
+The important question is how updated configuration is delivered from
+storage_controller to control plane to provide it to computes. As always, there
+are two options, pull and push. Let's do it the same push as with pageserver
+`/notify-attach` because 1) it keeps storage_controller out of critical compute
+start path 2) provides easier upgrade: there won't be such a thing as 'timeline
+managed by control plane / storcon', cplane just takes the value out of its db
+when needed 3) uniformity. It makes storage_controller responsible for retrying notifying
+control plane until it succeeds.
+
+So, cplane `/notify-safekeepers` for the timeline accepts `Configuration` and
+updates it in the db if the provided conf generation is higher (the cplane db
+should also store generations for this). Similarly to [`/notify-attach`](https://www.notion.so/neondatabase/Storage-Controller-Control-Plane-interface-6de56dd310a043bfa5c2f5564fa98365), it
+should update db which makes the call successful, and then try to schedule
+`apply_config` if possible, it is ok if not. storage_controller 
+should rate limit calling the endpoint, but likely this won't be needed, as migration
+throughput is limited by `pull_timeline`.
+
+Timeline (branch) creation in cplane should call storage_controller POST
+`tenant/:tenant_id/timeline` like it currently does for sharded tenants.
+Response should be augmented with `safekeeper_conf: Configuration`. The call
+should be retried until succeeds.
+
+Timeline deletion and tenant deletion in cplane should call appropriate
+storage_controller endpoints like it currently does for sharded tenants. The
+calls should be retried until they succeed.
+
+### storage_controller implementation
+
+Current 'load everything on startup and keep in memory' easy design is fine.
+Single timeline shouldn't take more than 100 bytes (it's 16 byte tenant_id, 16
+byte timeline_id, int generation, vec of ~3 safekeeper ids plus some flags), so
+10^6 of timelines shouldn't take more than 100MB.
+
+Similar to pageserver attachment Intents storage_controller would have in-memory
+`MigrationRequest` (or its absense) for each timeline and pool of tasks trying
+to make these request reality; this ensures one instance of storage_controller
+won't do several migrations on the same timeline concurrently. In the first
+version it is simpler to have more manual control and no retries, i.e. migration
+failure removes the request. Later we can build retries and automatic
+scheduling/migration. `MigrationRequest` is
+```
+enum MigrationRequest {
+    To(Vec<NodeId>),
+    FinishPending,
+}
+```
+
+`FinishPending` requests to run the procedure to ensure state is clean: current
+configuration is not joint and majority of safekeepers are aware of it, but do
+not attempt to migrate anywhere. If current configuration fetched on step 1 is
+not joint it jumps to step 7. It should be run at startup for all timelines (but
+similarly, in the first version it is ok to trigger it manually).
+
+#### Schema
+
+`safekeepers` table mirroring current `nodes` should be added, except that for
+`scheduling_policy` field (seems like `status` is a better name for it): it is enough
+to have at least in the beginning only 3 fields: 1) `active` 2) `offline` 3)
+`decomissioned`.
+
+`timelines` table:
+```
+table! {
+    // timeline_id is primary key
+    timelines (tenant_id, timeline_id) {
+        timeline_id -> Varchar,
+        tenant_id -> Varchar,
+        generation -> Int4,
+        sk_set -> Array<Int4>, // list of safekeeper ids
+        new_sk_set -> Nullable<Array<Int4>>, // list of safekeeper ids, null if not joint conf
+        cplane_notified_generation -> Int4,
+    }
+}
+```
+
+#### API
+
+Node management is similar to pageserver:
+1) POST `/control/v1/safekeepers` upserts safekeeper.
+2) GET `/control/v1/safekeepers` lists safekeepers.
+3) GET `/control/v1/safekeepers/:node_id` gets safekeeper.
+4) PUT `/control/v1/safekepers/:node_id/status` changes status to e.g.
+   `offline` or `decomissioned`. Initially it is simpler not to schedule any
+    migrations here.
+
+Safekeeper deploy scripts should register safekeeper at storage_contorller as
+they currently do with cplane, under the same id.
+
+Timeline creation/deletion: already existing POST `tenant/:tenant_id/timeline`
+would 1) choose initial set of safekeepers; 2) write to the db initial
+`Configuration` with `INSERT ON CONFLICT DO NOTHING` returning existing row in
+case of conflict; 3) create timeline on the majority of safekeepers (already
+created is ok).
+
+We don't want to block timeline creation when one safekeeper is down. Currently
+this is solved by compute implicitly creating timeline on any safekeeper it is
+connected to. This creates ugly timeline state on safekeeper when timeline is
+created, but start LSN is not defined yet. It would be nice to remove this; to
+do that, controller can in the background retry to create timeline on
+safekeeper(s) which missed that during initial creation call. It can do that
+through `pull_timeline` from majority so it doesn't need to remember
+`parent_lsn` in its db.
+
+Timeline deletion removes the row from the db and forwards deletion to the
+current configuration members. Without additional actions deletions might leak,
+see below on this; initially let's ignore these, reporting to cplane success if
+at least one safekeeper deleted the timeline (this will remove s3 data).
+
+Tenant deletion repeats timeline deletion for all timelines.
+
+Migration API: the first version is the simplest and the most imperative:
+1) PUT `/control/v1/safekeepers/migrate` schedules `MigrationRequest`s to move
+all timelines from one safekeeper to another. It accepts json
+```
+{
+    "src_sk": u32,
+    "dst_sk": u32,
+    "limit": Optional<u32>,
+}
+```
+
+Returns list of scheduled requests.
+
+2) PUT `/control/v1/tenant/:tenant_id/timeline/:timeline_id/safekeeper_migrate` schedules `MigrationRequest`
+   to move single timeline to given set of safekeepers:
+```
+{
+    "desired_set": Vec<u32>,
+}
+```
+
+Returns scheduled request.
+
+Similar call should be added for the tenant.
+
+It would be great to have some way of subscribing to the results (apart from
+looking at logs/metrics).
+
+Migration is executed as described above. One subtlety is that (local) deletion on
+source safekeeper might fail, which is not a problem if we are going to
+decomission the node but leaves garbage otherwise. I'd propose in the first version
+1) Don't attempt deletion at all if node status is `offline`.
+2) If it failed, just issue warning.
+And add PUT `/control/v1/safekeepers/:node_id/scrub` endpoint which would find and 
+remove garbage timelines for manual use. It will 1) list all timelines on the 
+safekeeper 2) compare each one against configuration storage: if timeline 
+doesn't exist at all (had been deleted), it can be deleted. Otherwise, it can 
+be deleted under generation number if node is not member of current generation.
+
+Automating this is untrivial; we'd need to register all potential missing
+deletions <tenant_id, timeline_id, generation, node_id> in the same transaction
+which switches configurations. Similarly when timeline is fully deleted to
+prevent cplane operation from blocking when some safekeeper is not available
+deletion should be also registered.
+
+One more task pool should infinitely retry notifying control plane about changed
+safekeeper sets.
+
+3) GET `/control/v1/tenant/:tenant_id/timeline/:timeline_id/` should return
+   current in memory state of the timeline and pending `MigrationRequest`,
+   if any.
+
+4) PUT `/control/v1/tenant/:tenant_id/timeline/:timeline_id/safekeeper_migrate_abort` tries to abort the
+   migration by switching configuration from the joint to the one with (previous) `sk_set` under CAS
+   (incrementing generation as always).
+
+#### Dealing with multiple instances of storage_controller
+
+Operations described above executed concurrently might create some errors but do
+not prevent progress, so while we normally don't want to run multiple instances
+of storage_controller it is fine to have it temporarily, e.g. during redeploy.
+
+Any interactions with db update in-memory controller state, e.g. if migration
+request failed because different one is in progress, controller remembers that
+and tries to finish it.
+
+## Testing
+
+`neon_local` should be switched to use storage_controller, playing role of
+control plane.
+
+There should be following layers of tests:
+1) Model checked TLA+ spec specifies the algorithm and verifies its basic safety.
+
+2) To cover real code and at the same time test many schedules we should have
+   simulation tests. For that, configuration storage, storage_controller <->
+   safekeeper communication and pull_timeline need to be mocked and main switch
+   procedure wrapped to as a node (thread) in simulation tests, using these
+   mocks. Test would inject migrations like it currently injects
+   safekeeper/walproposer restars. Main assert is the same -- committed WAL must
+   not be lost.
+
+3) Since simulation testing injects at relatively high level points (not
+   syscalls), it omits some code, in particular `pull_timeline`. Thus it is
+   better to have basic tests covering whole system as well. Extended version of
+   `test_restarts_under_load` would do: start background load and do migration 
+   under it, then restart endpoint and check that no reported commits 
+   had been lost. I'd also add one more creating classic network split scenario, with
+   one compute talking to AC and another to BD while migration from nodes ABC to ABD
+   happens.
+
+4) Simple e2e test should ensure that full flow including cplane notification works.
+
+## Order of implementation and rollout
+
+Note that 
+- Control plane parts and integration with it is fully independent from everything else
+  (tests would use simulation and neon_local).
+- There is a lot of infra work making storage_controller aware of timelines and safekeepers
+  and its impl/rollout should be separate from migration itself.
+- Initially walproposer can just stop working while it observers joint configuration.
+  Such window would be typically very short anyway.
+
+To rollout smoothly, both walproposer and safekeeper should have flag
+`configurations_enabled`; when set to false, they would work as currently, i.e.
+walproposer is able to commit on whatever safekeeper set it is provided. Until
+all timelines are managed by storcon we'd need to use current script to migrate
+and update/drop entries in the storage_controller database if it has any.
+
+Safekeepers would need to be able to talk both current and new protocol version
+with compute to reduce number of computes restarted in prod once v2 protocol is
+deployed (though before completely switching we'd need to force this).
+
+Let's have the following rollout order:
+- storage_controller becomes aware of safekeepers;
+- storage_controller gets timeline creation for new timelines and deletion requests, but
+  doesn't manage all timelines yet. Migration can be tested on these new timelines.
+  To keep control plane and storage_controller databases in sync while control 
+  plane still chooses the safekeepers initially (until all timelines are imported
+  it can choose better), `TimelineCreateRequest` can get optional safekeepers
+  field with safekeepers chosen by cplane.
+- Then we can import all existing timelines from control plane to
+  storage_controller and gradually enable configurations region by region.
+
+
+Very rough implementation order:
+- Add concept of configurations to safekeepers (including control file),
+  implement v3 protocol.
+- Implement walproposer changes, including protocol.
+- Implement storconn part. Use it in neon_local (and pytest).
+- Make cplane store safekeepers per timeline instead of per tenant.
+- Implement cplane/storcon integration. Route branch creation/deletion 
+  through storcon. Then we can test migration of new branches.
+- Finally import existing branches. Then we can drop cplane 
+  safekeeper selection code. Gradually enable configurations at 
+  computes and safekeepers. Before that, all computes must talk only
+  v3 protocol version.
+
+## Integration with evicted timelines
+
+Currently, `pull_timeline` doesn't work correctly with evicted timelines because
+copy would point to original partial file. To fix let's just do s3 copy of the
+file. It is a bit stupid as generally unnecessary work, but it makes sense to
+implement proper migration before doing smarter timeline archival. [Issue](https://github.com/neondatabase/neon/issues/8542)
+
+## Possible optimizations
+
+Steps above suggest walproposer restart (with re-election) and thus reconnection
+to safekeepers. Since by bumping term on new majority we ensure that leader
+terms are unique even across generation switches it is possible to preserve
+connections. However, it is more complicated, reconnection is very fast and it
+is much more important to avoid compute restart than millisecond order of write
+stall.
+
+Multiple joint consensus: algorithm above rejects attempt to change membership
+while another attempt is in progress. It is possible to overlay them and AFAIK
+Aurora does this but similarly I don't think this is needed.
+
+## Misc
+
+We should use Compute <-> safekeeper protocol change to include other (long
+yearned) modifications:
+- send data in network order to make arm work.
+- remove term_start_lsn from AppendRequest
+- add horizon to TermHistory
+- add to ProposerGreeting number of connection from this wp to sk

libs/pageserver_api/src/key.rs

@@ -22,6 +22,11 @@ pub struct Key {
     pub field6: u32,
 }
 
+/// When working with large numbers of Keys in-memory, it is more efficient to handle them as i128 than as
+/// a struct of fields.
+#[derive(Clone, Copy, Hash, PartialEq, Eq, Ord, PartialOrd)]
+pub struct CompactKey(i128);
+
 /// The storage key size.
 pub const KEY_SIZE: usize = 18;
 
@@ -107,7 +112,10 @@ 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)
@@ -127,6 +135,14 @@ impl Key {
         }
     }
 
+    pub fn to_compact(&self) -> CompactKey {
+        CompactKey(self.to_i128())
+    }
+
+    pub fn from_compact(k: CompactKey) -> Self {
+        Self::from_i128(k.0)
+    }
+
     pub const fn next(&self) -> Key {
         self.add(1)
     }
@@ -196,6 +212,13 @@ impl fmt::Display for Key {
     }
 }
 
+impl fmt::Display for CompactKey {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        let k = Key::from_compact(*self);
+        k.fmt(f)
+    }
+}
+
 impl Key {
     pub const MIN: Key = Key {
         field1: u8::MIN,

libs/pageserver_api/src/models.rs

@@ -637,6 +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)]
@@ -940,6 +947,8 @@ pub struct TopTenantShardsResponse {
 }
 
 pub mod virtual_file {
+    use std::path::PathBuf;
+
     #[derive(
         Copy,
         Clone,
@@ -958,6 +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
@@ -1427,6 +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(),
@@ -1449,6 +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(),

libs/pageserver_api/src/models/detach_ancestor.rs

@@ -1,6 +1,8 @@
+use std::collections::HashSet;
+
 use utils::id::TimelineId;
 
 #[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/models/utilization.rs

@@ -1,4 +1,5 @@
-use utils::serde_system_time::SystemTime;
+use std::time::SystemTime;
+use utils::{serde_percent::Percent, serde_system_time};
 
 /// Pageserver current utilization and scoring for how good candidate the pageserver would be for
 /// the next tenant.
@@ -9,19 +10,88 @@ use utils::serde_system_time::SystemTime;
 /// not handle full u64 values properly.
 #[derive(serde::Serialize, serde::Deserialize, Debug, Clone)]
 pub struct PageserverUtilization {
-    /// Used disk space
+    /// Used disk space (physical, ground truth from statfs())
     #[serde(serialize_with = "ser_saturating_u63")]
     pub disk_usage_bytes: u64,
     /// Free disk space
     #[serde(serialize_with = "ser_saturating_u63")]
     pub free_space_bytes: u64,
-    /// Lower is better score for how good candidate for a next tenant would this pageserver be.
-    #[serde(serialize_with = "ser_saturating_u63")]
+
+    /// Wanted disk space, based on the tenant shards currently present on this pageserver: this
+    /// is like disk_usage_bytes, but it is stable and does not change with the cache state of
+    /// tenants, whereas disk_usage_bytes may reach the disk eviction `max_usage_pct` and stay
+    /// there, or may be unrealistically low if the pageserver has attached tenants which haven't
+    /// downloaded layers yet.
+    #[serde(serialize_with = "ser_saturating_u63", default)]
+    pub disk_wanted_bytes: u64,
+
+    // What proportion of total disk space will this pageserver use before it starts evicting data?
+    #[serde(default = "unity_percent")]
+    pub disk_usable_pct: Percent,
+
+    // How many shards are currently on this node?
+    #[serde(default)]
+    pub shard_count: u32,
+
+    // How many shards should this node be able to handle at most?
+    #[serde(default)]
+    pub max_shard_count: u32,
+
+    /// Cached result of [`Self::score`]
     pub utilization_score: u64,
+
     /// When was this snapshot captured, pageserver local time.
     ///
     /// Use millis to give confidence that the value is regenerated often enough.
-    pub captured_at: SystemTime,
+    pub captured_at: serde_system_time::SystemTime,
+}
+
+fn unity_percent() -> Percent {
+    Percent::new(0).unwrap()
+}
+
+impl PageserverUtilization {
+    const UTILIZATION_FULL: u64 = 1000000;
+
+    /// Calculate a utilization score.  The result is to be inrepreted as a fraction of
+    /// Self::UTILIZATION_FULL.
+    ///
+    /// Lower values are more affine to scheduling more work on this node.
+    /// - UTILIZATION_FULL represents an ideal node which is fully utilized but should not receive any more work.
+    /// - 0.0 represents an empty node.
+    /// - Negative values are forbidden
+    /// - Values over UTILIZATION_FULL indicate an overloaded node, which may show degraded performance due to
+    ///   layer eviction.
+    pub fn score(&self) -> u64 {
+        let disk_usable_capacity = ((self.disk_usage_bytes + self.free_space_bytes)
+            * self.disk_usable_pct.get() as u64)
+            / 100;
+        let disk_utilization_score =
+            self.disk_wanted_bytes * Self::UTILIZATION_FULL / disk_usable_capacity;
+
+        let shard_utilization_score =
+            self.shard_count as u64 * Self::UTILIZATION_FULL / self.max_shard_count as u64;
+        std::cmp::max(disk_utilization_score, shard_utilization_score)
+    }
+
+    pub fn refresh_score(&mut self) {
+        self.utilization_score = self.score();
+    }
+
+    /// A utilization structure that has a full utilization score: use this as a placeholder when
+    /// you need a utilization but don't have real values yet.
+    pub fn full() -> Self {
+        Self {
+            disk_usage_bytes: 1,
+            free_space_bytes: 0,
+            disk_wanted_bytes: 1,
+            disk_usable_pct: Percent::new(100).unwrap(),
+            shard_count: 1,
+            max_shard_count: 1,
+            utilization_score: Self::UTILIZATION_FULL,
+            captured_at: serde_system_time::SystemTime(SystemTime::now()),
+        }
+    }
 }
 
 /// openapi knows only `format: int64`, so avoid outputting a non-parseable value by generated clients.
@@ -49,15 +119,19 @@ mod tests {
         let doc = PageserverUtilization {
             disk_usage_bytes: u64::MAX,
             free_space_bytes: 0,
-            utilization_score: u64::MAX,
-            captured_at: SystemTime(
+            disk_wanted_bytes: u64::MAX,
+            utilization_score: 13,
+            disk_usable_pct: Percent::new(90).unwrap(),
+            shard_count: 100,
+            max_shard_count: 200,
+            captured_at: serde_system_time::SystemTime(
                 std::time::SystemTime::UNIX_EPOCH + Duration::from_secs(1708509779),
             ),
         };
 
         let s = serde_json::to_string(&doc).unwrap();
 
-        let expected = r#"{"disk_usage_bytes":9223372036854775807,"free_space_bytes":0,"utilization_score":9223372036854775807,"captured_at":"2024-02-21T10:02:59.000Z"}"#;
+        let expected = "{\"disk_usage_bytes\":9223372036854775807,\"free_space_bytes\":0,\"disk_wanted_bytes\":9223372036854775807,\"disk_usable_pct\":90,\"shard_count\":100,\"max_shard_count\":200,\"utilization_score\":13,\"captured_at\":\"2024-02-21T10:02:59.000Z\"}";
 
         assert_eq!(s, expected);
     }

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/remote_storage/src/lib.rs

@@ -144,6 +144,7 @@ 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,

libs/utils/src/completion.rs

@@ -5,13 +5,40 @@ use tokio_util::task::{task_tracker::TaskTrackerToken, TaskTracker};
 /// Can be cloned, moved and kept around in futures as "guard objects".
 #[derive(Clone)]
 pub struct Completion {
-    _token: TaskTrackerToken,
+    token: TaskTrackerToken,
+}
+
+impl std::fmt::Debug for Completion {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.debug_struct("Completion")
+            .field("siblings", &self.token.task_tracker().len())
+            .finish()
+    }
+}
+
+impl Completion {
+    /// Returns true if this completion is associated with the given barrier.
+    pub fn blocks(&self, barrier: &Barrier) -> bool {
+        TaskTracker::ptr_eq(self.token.task_tracker(), &barrier.0)
+    }
+
+    pub fn barrier(&self) -> Barrier {
+        Barrier(self.token.task_tracker().clone())
+    }
 }
 
 /// Barrier will wait until all clones of [`Completion`] have been dropped.
 #[derive(Clone)]
 pub struct Barrier(TaskTracker);
 
+impl std::fmt::Debug for Barrier {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.debug_struct("Barrier")
+            .field("remaining", &self.0.len())
+            .finish()
+    }
+}
+
 impl Default for Barrier {
     fn default() -> Self {
         let (_, rx) = channel();
@@ -51,5 +78,5 @@ pub fn channel() -> (Completion, Barrier) {
     tracker.close();
 
     let token = tracker.token();
-    (Completion { _token: token }, Barrier(tracker))
+    (Completion { token }, Barrier(tracker))
 }

libs/utils/src/lib.rs

@@ -128,7 +128,7 @@ pub mod circuit_breaker;
 ///
 /// #############################################################################################
 /// 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/sync/gate.rs

@@ -78,8 +78,9 @@ impl Drop for GateGuard {
     }
 }
 
-#[derive(Debug)]
+#[derive(Debug, thiserror::Error)]
 pub enum GateError {
+    #[error("gate is closed")]
     GateClosed,
 }
 

pageserver/Cargo.toml

@@ -49,6 +49,7 @@ postgres_backend.workspace = true
 postgres-protocol.workspace = true
 postgres-types.workspace = true
 rand.workspace = true
+range-set-blaze = { version = "0.1.16", features = ["alloc"] }
 regex.workspace = true
 scopeguard.workspace = true
 serde.workspace = true
@@ -107,3 +108,7 @@ harness = false
 [[bench]]
 name = "bench_walredo"
 harness = false
+
+[[bench]]
+name = "bench_ingest"
+harness = false

pageserver/benches/bench_ingest.rs

@@ -0,0 +1,239 @@
+use std::{env, num::NonZeroUsize};
+
+use bytes::Bytes;
+use camino::Utf8PathBuf;
+use criterion::{criterion_group, criterion_main, Criterion};
+use pageserver::{
+    config::PageServerConf,
+    context::{DownloadBehavior, RequestContext},
+    l0_flush::{L0FlushConfig, L0FlushGlobalState},
+    page_cache,
+    repository::Value,
+    task_mgr::TaskKind,
+    tenant::storage_layer::InMemoryLayer,
+    virtual_file,
+};
+use pageserver_api::{key::Key, shard::TenantShardId};
+use utils::{
+    bin_ser::BeSer,
+    id::{TenantId, TimelineId},
+};
+
+// A very cheap hash for generating non-sequential keys.
+fn murmurhash32(mut h: u32) -> u32 {
+    h ^= h >> 16;
+    h = h.wrapping_mul(0x85ebca6b);
+    h ^= h >> 13;
+    h = h.wrapping_mul(0xc2b2ae35);
+    h ^= h >> 16;
+    h
+}
+
+enum KeyLayout {
+    /// Sequential unique keys
+    Sequential,
+    /// Random unique keys
+    Random,
+    /// Random keys, but only use the bits from the mask of them
+    RandomReuse(u32),
+}
+
+enum WriteDelta {
+    Yes,
+    No,
+}
+
+async fn ingest(
+    conf: &'static PageServerConf,
+    put_size: usize,
+    put_count: usize,
+    key_layout: KeyLayout,
+    write_delta: WriteDelta,
+) -> anyhow::Result<()> {
+    let mut lsn = utils::lsn::Lsn(1000);
+    let mut key = Key::from_i128(0x0);
+
+    let timeline_id = TimelineId::generate();
+    let tenant_id = TenantId::generate();
+    let tenant_shard_id = TenantShardId::unsharded(tenant_id);
+
+    tokio::fs::create_dir_all(conf.timeline_path(&tenant_shard_id, &timeline_id)).await?;
+
+    let ctx = RequestContext::new(TaskKind::DebugTool, DownloadBehavior::Error);
+
+    let gate = utils::sync::gate::Gate::default();
+    let entered = gate.enter().unwrap();
+
+    let layer =
+        InMemoryLayer::create(conf, timeline_id, tenant_shard_id, lsn, entered, &ctx).await?;
+
+    let data = Value::Image(Bytes::from(vec![0u8; put_size])).ser()?;
+    let ctx = RequestContext::new(
+        pageserver::task_mgr::TaskKind::WalReceiverConnectionHandler,
+        pageserver::context::DownloadBehavior::Download,
+    );
+
+    for i in 0..put_count {
+        lsn += put_size as u64;
+
+        // Generate lots of keys within a single relation, which simulates the typical bulk ingest case: people
+        // usually care the most about write performance when they're blasting a huge batch of data into a huge table.
+        match key_layout {
+            KeyLayout::Sequential => {
+                // Use sequential order to illustrate the experience a user is likely to have
+                // when ingesting bulk data.
+                key.field6 = i as u32;
+            }
+            KeyLayout::Random => {
+                // Use random-order keys to avoid giving a false advantage to data structures that are
+                // faster when inserting on the end.
+                key.field6 = murmurhash32(i as u32);
+            }
+            KeyLayout::RandomReuse(mask) => {
+                // Use low bits only, to limit cardinality
+                key.field6 = murmurhash32(i as u32) & mask;
+            }
+        }
+
+        layer.put_value(key.to_compact(), lsn, &data, &ctx).await?;
+    }
+    layer.freeze(lsn + 1).await;
+
+    if matches!(write_delta, WriteDelta::Yes) {
+        let l0_flush_state = L0FlushGlobalState::new(L0FlushConfig::Direct {
+            max_concurrency: NonZeroUsize::new(1).unwrap(),
+        });
+        let (_desc, path) = layer
+            .write_to_disk(&ctx, None, l0_flush_state.inner())
+            .await?
+            .unwrap();
+        tokio::fs::remove_file(path).await?;
+    }
+
+    Ok(())
+}
+
+/// Wrapper to instantiate a tokio runtime
+fn ingest_main(
+    conf: &'static PageServerConf,
+    put_size: usize,
+    put_count: usize,
+    key_layout: KeyLayout,
+    write_delta: WriteDelta,
+) {
+    let runtime = tokio::runtime::Builder::new_current_thread()
+        .enable_all()
+        .build()
+        .unwrap();
+
+    runtime.block_on(async move {
+        let r = ingest(conf, put_size, put_count, key_layout, write_delta).await;
+        if let Err(e) = r {
+            panic!("{e:?}");
+        }
+    });
+}
+
+/// Declare a series of benchmarks for the Pageserver's ingest write path.
+///
+/// This benchmark does not include WAL decode: it starts at InMemoryLayer::put_value, and ends either
+/// at freezing the ephemeral layer, or writing the ephemeral layer out to an L0 (depending on whether WriteDelta is set).
+///
+/// Genuine disk I/O is used, so expect results to differ depending on storage.  However, when running on
+/// a fast disk, CPU is the bottleneck at time of writing.
+fn criterion_benchmark(c: &mut Criterion) {
+    let temp_dir_parent: Utf8PathBuf = env::current_dir().unwrap().try_into().unwrap();
+    let temp_dir = camino_tempfile::tempdir_in(temp_dir_parent).unwrap();
+    eprintln!("Data directory: {}", temp_dir.path());
+
+    let conf: &'static PageServerConf = Box::leak(Box::new(
+        pageserver::config::PageServerConf::dummy_conf(temp_dir.path().to_path_buf()),
+    ));
+    virtual_file::init(16384, virtual_file::io_engine_for_bench());
+    page_cache::init(conf.page_cache_size);
+
+    {
+        let mut group = c.benchmark_group("ingest-small-values");
+        let put_size = 100usize;
+        let put_count = 128 * 1024 * 1024 / put_size;
+        group.throughput(criterion::Throughput::Bytes((put_size * put_count) as u64));
+        group.sample_size(10);
+        group.bench_function("ingest 128MB/100b seq", |b| {
+            b.iter(|| {
+                ingest_main(
+                    conf,
+                    put_size,
+                    put_count,
+                    KeyLayout::Sequential,
+                    WriteDelta::Yes,
+                )
+            })
+        });
+        group.bench_function("ingest 128MB/100b rand", |b| {
+            b.iter(|| {
+                ingest_main(
+                    conf,
+                    put_size,
+                    put_count,
+                    KeyLayout::Random,
+                    WriteDelta::Yes,
+                )
+            })
+        });
+        group.bench_function("ingest 128MB/100b rand-1024keys", |b| {
+            b.iter(|| {
+                ingest_main(
+                    conf,
+                    put_size,
+                    put_count,
+                    KeyLayout::RandomReuse(0x3ff),
+                    WriteDelta::Yes,
+                )
+            })
+        });
+        group.bench_function("ingest 128MB/100b seq, no delta", |b| {
+            b.iter(|| {
+                ingest_main(
+                    conf,
+                    put_size,
+                    put_count,
+                    KeyLayout::Sequential,
+                    WriteDelta::No,
+                )
+            })
+        });
+    }
+
+    {
+        let mut group = c.benchmark_group("ingest-big-values");
+        let put_size = 8192usize;
+        let put_count = 128 * 1024 * 1024 / put_size;
+        group.throughput(criterion::Throughput::Bytes((put_size * put_count) as u64));
+        group.sample_size(10);
+        group.bench_function("ingest 128MB/8k seq", |b| {
+            b.iter(|| {
+                ingest_main(
+                    conf,
+                    put_size,
+                    put_count,
+                    KeyLayout::Sequential,
+                    WriteDelta::Yes,
+                )
+            })
+        });
+        group.bench_function("ingest 128MB/8k seq, no delta", |b| {
+            b.iter(|| {
+                ingest_main(
+                    conf,
+                    put_size,
+                    put_count,
+                    KeyLayout::Sequential,
+                    WriteDelta::No,
+                )
+            })
+        });
+    }
+}
+
+criterion_group!(benches, criterion_benchmark);
+criterion_main!(benches);

pageserver/benches/bench_layer_map.rs

@@ -1,3 +1,4 @@
+use criterion::measurement::WallTime;
 use pageserver::keyspace::{KeyPartitioning, KeySpace};
 use pageserver::repository::Key;
 use pageserver::tenant::layer_map::LayerMap;
@@ -15,7 +16,11 @@ use utils::id::{TenantId, TimelineId};
 
 use utils::lsn::Lsn;
 
-use criterion::{black_box, criterion_group, criterion_main, Criterion};
+use criterion::{black_box, criterion_group, criterion_main, BenchmarkGroup, Criterion};
+
+fn fixture_path(relative: &str) -> PathBuf {
+    PathBuf::from(env!("CARGO_MANIFEST_DIR")).join(relative)
+}
 
 fn build_layer_map(filename_dump: PathBuf) -> LayerMap {
     let mut layer_map = LayerMap::default();
@@ -109,7 +114,7 @@ fn uniform_key_partitioning(layer_map: &LayerMap, _lsn: Lsn) -> KeyPartitioning
 // between each test run.
 fn bench_from_captest_env(c: &mut Criterion) {
     // TODO consider compressing this file
-    let layer_map = build_layer_map(PathBuf::from("benches/odd-brook-layernames.txt"));
+    let layer_map = build_layer_map(fixture_path("benches/odd-brook-layernames.txt"));
     let queries: Vec<(Key, Lsn)> = uniform_query_pattern(&layer_map);
 
     // Test with uniform query pattern
@@ -139,7 +144,7 @@ fn bench_from_captest_env(c: &mut Criterion) {
 fn bench_from_real_project(c: &mut Criterion) {
     // Init layer map
     let now = Instant::now();
-    let layer_map = build_layer_map(PathBuf::from("benches/odd-brook-layernames.txt"));
+    let layer_map = build_layer_map(fixture_path("benches/odd-brook-layernames.txt"));
     println!("Finished layer map init in {:?}", now.elapsed());
 
     // Choose uniformly distributed queries
@@ -242,7 +247,72 @@ fn bench_sequential(c: &mut Criterion) {
     group.finish();
 }
 
+fn bench_visibility_with_map(
+    group: &mut BenchmarkGroup<WallTime>,
+    layer_map: LayerMap,
+    read_points: Vec<Lsn>,
+    bench_name: &str,
+) {
+    group.bench_function(bench_name, |b| {
+        b.iter(|| black_box(layer_map.get_visibility(read_points.clone())));
+    });
+}
+
+// Benchmark using synthetic data. Arrange image layers on stacked diagonal lines.
+fn bench_visibility(c: &mut Criterion) {
+    let mut group = c.benchmark_group("visibility");
+    {
+        // Init layer map. Create 100_000 layers arranged in 1000 diagonal lines.
+        let now = Instant::now();
+        let mut layer_map = LayerMap::default();
+        let mut updates = layer_map.batch_update();
+        for i in 0..100_000 {
+            let i32 = (i as u32) % 100;
+            let zero = Key::from_hex("000000000000000000000000000000000000").unwrap();
+            let layer = PersistentLayerDesc::new_img(
+                TenantShardId::unsharded(TenantId::generate()),
+                TimelineId::generate(),
+                zero.add(10 * i32)..zero.add(10 * i32 + 1),
+                Lsn(i),
+                0,
+            );
+            updates.insert_historic(layer);
+        }
+        updates.flush();
+        println!("Finished layer map init in {:?}", now.elapsed());
+
+        let mut read_points = Vec::new();
+        for i in (0..100_000).step_by(1000) {
+            read_points.push(Lsn(i));
+        }
+
+        bench_visibility_with_map(&mut group, layer_map, read_points, "sequential");
+    }
+
+    {
+        let layer_map = build_layer_map(fixture_path("benches/odd-brook-layernames.txt"));
+        let read_points = vec![Lsn(0x1C760FA190)];
+        bench_visibility_with_map(&mut group, layer_map, read_points, "real_map");
+
+        let layer_map = build_layer_map(fixture_path("benches/odd-brook-layernames.txt"));
+        let read_points = vec![
+            Lsn(0x1C760FA190),
+            Lsn(0x000000931BEAD539),
+            Lsn(0x000000931BF63011),
+            Lsn(0x000000931B33AE68),
+            Lsn(0x00000038E67ABFA0),
+            Lsn(0x000000931B33AE68),
+            Lsn(0x000000914E3F38F0),
+            Lsn(0x000000931B33AE68),
+        ];
+        bench_visibility_with_map(&mut group, layer_map, read_points, "real_map_many_branches");
+    }
+
+    group.finish();
+}
+
 criterion_group!(group_1, bench_from_captest_env);
 criterion_group!(group_2, bench_from_real_project);
 criterion_group!(group_3, bench_sequential);
-criterion_main!(group_1, group_2, group_3);
+criterion_group!(group_4, bench_visibility);
+criterion_main!(group_1, group_2, group_3, group_4);

pageserver/src/bin/pageserver.rs

@@ -17,11 +17,9 @@ use pageserver::config::PageserverIdentity;
 use pageserver::control_plane_client::ControlPlaneClient;
 use pageserver::disk_usage_eviction_task::{self, launch_disk_usage_global_eviction_task};
 use pageserver::metrics::{STARTUP_DURATION, STARTUP_IS_LOADING};
-use pageserver::task_mgr::WALRECEIVER_RUNTIME;
+use pageserver::task_mgr::{COMPUTE_REQUEST_RUNTIME, WALRECEIVER_RUNTIME};
 use pageserver::tenant::{secondary, TenantSharedResources};
-use pageserver::{
-    CancellableTask, ConsumptionMetricsTasks, HttpEndpointListener, LibpqEndpointListener,
-};
+use pageserver::{CancellableTask, ConsumptionMetricsTasks, HttpEndpointListener};
 use remote_storage::GenericRemoteStorage;
 use tokio::signal::unix::SignalKind;
 use tokio::time::Instant;
@@ -31,11 +29,9 @@ use tracing::*;
 use metrics::set_build_info_metric;
 use pageserver::{
     config::PageServerConf,
-    context::{DownloadBehavior, RequestContext},
     deletion_queue::DeletionQueue,
     http, page_cache, page_service, task_mgr,
-    task_mgr::TaskKind,
-    task_mgr::{BACKGROUND_RUNTIME, COMPUTE_REQUEST_RUNTIME, MGMT_REQUEST_RUNTIME},
+    task_mgr::{BACKGROUND_RUNTIME, MGMT_REQUEST_RUNTIME},
     tenant::mgr,
     virtual_file,
 };
@@ -127,8 +123,7 @@ fn main() -> anyhow::Result<()> {
 
     // after setting up logging, log the effective IO engine choice and read path implementations
     info!(?conf.virtual_file_io_engine, "starting with virtual_file IO engine");
-    info!(?conf.get_impl, "starting with get page implementation");
-    info!(?conf.get_vectored_impl, "starting with vectored get page implementation");
+    info!(?conf.virtual_file_direct_io, "starting with virtual_file Direct IO settings");
     info!(?conf.compact_level0_phase1_value_access, "starting with setting for compact_level0_phase1_value_access");
 
     let tenants_path = conf.tenants_path();
@@ -594,30 +589,13 @@ fn start_pageserver(
 
     // Spawn a task to listen for libpq connections. It will spawn further tasks
     // for each connection. We created the listener earlier already.
-    let libpq_listener = {
-        let cancel = CancellationToken::new();
-        let libpq_ctx = RequestContext::todo_child(
-            TaskKind::LibpqEndpointListener,
-            // listener task shouldn't need to download anything. (We will
-            // create a separate sub-contexts for each connection, with their
-            // own download behavior. This context is used only to listen and
-            // accept connections.)
-            DownloadBehavior::Error,
-        );
-
-        let task = COMPUTE_REQUEST_RUNTIME.spawn(task_mgr::exit_on_panic_or_error(
-            "libpq listener",
-            page_service::libpq_listener_main(
-                tenant_manager.clone(),
-                pg_auth,
-                pageserver_listener,
-                conf.pg_auth_type,
-                libpq_ctx,
-                cancel.clone(),
-            ),
-        ));
-        LibpqEndpointListener(CancellableTask { task, cancel })
-    };
+    let page_service = page_service::spawn(conf, tenant_manager.clone(), pg_auth, {
+        let _entered = COMPUTE_REQUEST_RUNTIME.enter(); // TcpListener::from_std requires it
+        pageserver_listener
+            .set_nonblocking(true)
+            .context("set listener to nonblocking")?;
+        tokio::net::TcpListener::from_std(pageserver_listener).context("create tokio listener")?
+    });
 
     let mut shutdown_pageserver = Some(shutdown_pageserver.drop_guard());
 
@@ -645,7 +623,7 @@ fn start_pageserver(
             shutdown_pageserver.take();
             pageserver::shutdown_pageserver(
                 http_endpoint_listener,
-                libpq_listener,
+                page_service,
                 consumption_metrics_tasks,
                 disk_usage_eviction_task,
                 &tenant_manager,

pageserver/src/config.rs

@@ -29,12 +29,12 @@ use utils::{
     logging::LogFormat,
 };
 
+use crate::l0_flush::L0FlushConfig;
+use crate::tenant::config::TenantConfOpt;
 use crate::tenant::timeline::compaction::CompactL0Phase1ValueAccess;
 use crate::tenant::vectored_blob_io::MaxVectoredReadBytes;
-use crate::tenant::{config::TenantConfOpt, timeline::GetImpl};
 use crate::tenant::{TENANTS_SEGMENT_NAME, TIMELINES_SEGMENT_NAME};
 use crate::{disk_usage_eviction_task::DiskUsageEvictionTaskConfig, virtual_file::io_engine};
-use crate::{l0_flush::L0FlushConfig, tenant::timeline::GetVectoredImpl};
 use crate::{tenant::config::TenantConf, virtual_file};
 use crate::{TENANT_HEATMAP_BASENAME, TENANT_LOCATION_CONFIG_NAME, TIMELINE_DELETE_MARK_SUFFIX};
 
@@ -133,14 +133,8 @@ pub mod defaults {
 
 #virtual_file_io_engine = '{DEFAULT_VIRTUAL_FILE_IO_ENGINE}'
 
-#get_vectored_impl = '{DEFAULT_GET_VECTORED_IMPL}'
-
-#get_impl = '{DEFAULT_GET_IMPL}'
-
 #max_vectored_read_bytes = '{DEFAULT_MAX_VECTORED_READ_BYTES}'
 
-#validate_vectored_get = '{DEFAULT_VALIDATE_VECTORED_GET}'
-
 [tenant_config]
 #checkpoint_distance = {DEFAULT_CHECKPOINT_DISTANCE} # in bytes
 #checkpoint_timeout = {DEFAULT_CHECKPOINT_TIMEOUT}
@@ -278,14 +272,8 @@ pub struct PageServerConf {
 
     pub virtual_file_io_engine: virtual_file::IoEngineKind,
 
-    pub get_vectored_impl: GetVectoredImpl,
-
-    pub get_impl: GetImpl,
-
     pub max_vectored_read_bytes: MaxVectoredReadBytes,
 
-    pub validate_vectored_get: bool,
-
     pub image_compression: ImageCompressionAlgorithm,
 
     /// How many bytes of ephemeral layer content will we allow per kilobyte of RAM.  When this
@@ -300,6 +288,9 @@ pub struct PageServerConf {
     /// This flag is temporary and will be removed after gradual rollout.
     /// See <https://github.com/neondatabase/neon/issues/8184>.
     pub compact_level0_phase1_value_access: CompactL0Phase1ValueAccess,
+
+    /// Direct IO settings
+    pub virtual_file_direct_io: virtual_file::DirectIoMode,
 }
 
 /// We do not want to store this in a PageServerConf because the latter may be logged
@@ -393,14 +384,8 @@ struct PageServerConfigBuilder {
 
     virtual_file_io_engine: BuilderValue<virtual_file::IoEngineKind>,
 
-    get_vectored_impl: BuilderValue<GetVectoredImpl>,
-
-    get_impl: BuilderValue<GetImpl>,
-
     max_vectored_read_bytes: BuilderValue<MaxVectoredReadBytes>,
 
-    validate_vectored_get: BuilderValue<bool>,
-
     image_compression: BuilderValue<ImageCompressionAlgorithm>,
 
     ephemeral_bytes_per_memory_kb: BuilderValue<usize>,
@@ -408,6 +393,8 @@ struct PageServerConfigBuilder {
     l0_flush: BuilderValue<L0FlushConfig>,
 
     compact_level0_phase1_value_access: BuilderValue<CompactL0Phase1ValueAccess>,
+
+    virtual_file_direct_io: BuilderValue<virtual_file::DirectIoMode>,
 }
 
 impl PageServerConfigBuilder {
@@ -488,16 +475,14 @@ impl PageServerConfigBuilder {
 
             virtual_file_io_engine: Set(DEFAULT_VIRTUAL_FILE_IO_ENGINE.parse().unwrap()),
 
-            get_vectored_impl: Set(DEFAULT_GET_VECTORED_IMPL.parse().unwrap()),
-            get_impl: Set(DEFAULT_GET_IMPL.parse().unwrap()),
             max_vectored_read_bytes: Set(MaxVectoredReadBytes(
                 NonZeroUsize::new(DEFAULT_MAX_VECTORED_READ_BYTES).unwrap(),
             )),
             image_compression: Set(DEFAULT_IMAGE_COMPRESSION),
-            validate_vectored_get: Set(DEFAULT_VALIDATE_VECTORED_GET),
             ephemeral_bytes_per_memory_kb: Set(DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB),
             l0_flush: Set(L0FlushConfig::default()),
             compact_level0_phase1_value_access: Set(CompactL0Phase1ValueAccess::default()),
+            virtual_file_direct_io: Set(virtual_file::DirectIoMode::default()),
         }
     }
 }
@@ -653,22 +638,10 @@ impl PageServerConfigBuilder {
         self.virtual_file_io_engine = BuilderValue::Set(value);
     }
 
-    pub fn get_vectored_impl(&mut self, value: GetVectoredImpl) {
-        self.get_vectored_impl = BuilderValue::Set(value);
-    }
-
-    pub fn get_impl(&mut self, value: GetImpl) {
-        self.get_impl = BuilderValue::Set(value);
-    }
-
     pub fn get_max_vectored_read_bytes(&mut self, value: MaxVectoredReadBytes) {
         self.max_vectored_read_bytes = BuilderValue::Set(value);
     }
 
-    pub fn get_validate_vectored_get(&mut self, value: bool) {
-        self.validate_vectored_get = BuilderValue::Set(value);
-    }
-
     pub fn get_image_compression(&mut self, value: ImageCompressionAlgorithm) {
         self.image_compression = BuilderValue::Set(value);
     }
@@ -685,6 +658,10 @@ impl PageServerConfigBuilder {
         self.compact_level0_phase1_value_access = BuilderValue::Set(value);
     }
 
+    pub fn virtual_file_direct_io(&mut self, value: virtual_file::DirectIoMode) {
+        self.virtual_file_direct_io = BuilderValue::Set(value);
+    }
+
     pub fn build(self, id: NodeId) -> anyhow::Result<PageServerConf> {
         let default = Self::default_values();
 
@@ -735,14 +712,12 @@ impl PageServerConfigBuilder {
                 heatmap_upload_concurrency,
                 secondary_download_concurrency,
                 ingest_batch_size,
-                get_vectored_impl,
-                get_impl,
                 max_vectored_read_bytes,
-                validate_vectored_get,
                 image_compression,
                 ephemeral_bytes_per_memory_kb,
                 l0_flush,
                 compact_level0_phase1_value_access,
+                virtual_file_direct_io,
             }
             CUSTOM LOGIC
             {
@@ -991,21 +966,12 @@ impl PageServerConf {
                 "virtual_file_io_engine" => {
                     builder.virtual_file_io_engine(parse_toml_from_str("virtual_file_io_engine", item)?)
                 }
-                "get_vectored_impl" => {
-                    builder.get_vectored_impl(parse_toml_from_str("get_vectored_impl", item)?)
-                }
-                "get_impl" => {
-                    builder.get_impl(parse_toml_from_str("get_impl", item)?)
-                }
                 "max_vectored_read_bytes" => {
                     let bytes = parse_toml_u64("max_vectored_read_bytes", item)? as usize;
                     builder.get_max_vectored_read_bytes(
                         MaxVectoredReadBytes(
                             NonZeroUsize::new(bytes).expect("Max byte size of vectored read must be greater than 0")))
                 }
-                "validate_vectored_get" => {
-                    builder.get_validate_vectored_get(parse_toml_bool("validate_vectored_get", item)?)
-                }
                 "image_compression" => {
                     builder.get_image_compression(parse_toml_from_str("image_compression", item)?)
                 }
@@ -1018,6 +984,9 @@ impl PageServerConf {
                 "compact_level0_phase1_value_access" => {
                     builder.compact_level0_phase1_value_access(utils::toml_edit_ext::deserialize_item(item).context("compact_level0_phase1_value_access")?)
                 }
+                "virtual_file_direct_io" => {
+                    builder.virtual_file_direct_io(utils::toml_edit_ext::deserialize_item(item).context("virtual_file_direct_io")?)
+                }
                 _ => bail!("unrecognized pageserver option '{key}'"),
             }
         }
@@ -1092,17 +1061,15 @@ impl PageServerConf {
             secondary_download_concurrency: defaults::DEFAULT_SECONDARY_DOWNLOAD_CONCURRENCY,
             ingest_batch_size: defaults::DEFAULT_INGEST_BATCH_SIZE,
             virtual_file_io_engine: DEFAULT_VIRTUAL_FILE_IO_ENGINE.parse().unwrap(),
-            get_vectored_impl: defaults::DEFAULT_GET_VECTORED_IMPL.parse().unwrap(),
-            get_impl: defaults::DEFAULT_GET_IMPL.parse().unwrap(),
             max_vectored_read_bytes: MaxVectoredReadBytes(
                 NonZeroUsize::new(defaults::DEFAULT_MAX_VECTORED_READ_BYTES)
                     .expect("Invalid default constant"),
             ),
             image_compression: defaults::DEFAULT_IMAGE_COMPRESSION,
-            validate_vectored_get: defaults::DEFAULT_VALIDATE_VECTORED_GET,
             ephemeral_bytes_per_memory_kb: defaults::DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB,
             l0_flush: L0FlushConfig::default(),
             compact_level0_phase1_value_access: CompactL0Phase1ValueAccess::default(),
+            virtual_file_direct_io: virtual_file::DirectIoMode::default(),
         }
     }
 }
@@ -1334,17 +1301,15 @@ background_task_maximum_delay = '334 s'
                 secondary_download_concurrency: defaults::DEFAULT_SECONDARY_DOWNLOAD_CONCURRENCY,
                 ingest_batch_size: defaults::DEFAULT_INGEST_BATCH_SIZE,
                 virtual_file_io_engine: DEFAULT_VIRTUAL_FILE_IO_ENGINE.parse().unwrap(),
-                get_vectored_impl: defaults::DEFAULT_GET_VECTORED_IMPL.parse().unwrap(),
-                get_impl: defaults::DEFAULT_GET_IMPL.parse().unwrap(),
                 max_vectored_read_bytes: MaxVectoredReadBytes(
                     NonZeroUsize::new(defaults::DEFAULT_MAX_VECTORED_READ_BYTES)
                         .expect("Invalid default constant")
                 ),
-                validate_vectored_get: defaults::DEFAULT_VALIDATE_VECTORED_GET,
                 image_compression: defaults::DEFAULT_IMAGE_COMPRESSION,
                 ephemeral_bytes_per_memory_kb: defaults::DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB,
                 l0_flush: L0FlushConfig::default(),
                 compact_level0_phase1_value_access: CompactL0Phase1ValueAccess::default(),
+                virtual_file_direct_io: virtual_file::DirectIoMode::default(),
             },
             "Correct defaults should be used when no config values are provided"
         );
@@ -1409,17 +1374,15 @@ background_task_maximum_delay = '334 s'
                 secondary_download_concurrency: defaults::DEFAULT_SECONDARY_DOWNLOAD_CONCURRENCY,
                 ingest_batch_size: 100,
                 virtual_file_io_engine: DEFAULT_VIRTUAL_FILE_IO_ENGINE.parse().unwrap(),
-                get_vectored_impl: defaults::DEFAULT_GET_VECTORED_IMPL.parse().unwrap(),
-                get_impl: defaults::DEFAULT_GET_IMPL.parse().unwrap(),
                 max_vectored_read_bytes: MaxVectoredReadBytes(
                     NonZeroUsize::new(defaults::DEFAULT_MAX_VECTORED_READ_BYTES)
                         .expect("Invalid default constant")
                 ),
-                validate_vectored_get: defaults::DEFAULT_VALIDATE_VECTORED_GET,
                 image_compression: defaults::DEFAULT_IMAGE_COMPRESSION,
                 ephemeral_bytes_per_memory_kb: defaults::DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB,
                 l0_flush: L0FlushConfig::default(),
                 compact_level0_phase1_value_access: CompactL0Phase1ValueAccess::default(),
+                virtual_file_direct_io: virtual_file::DirectIoMode::default(),
             },
             "Should be able to parse all basic config values correctly"
         );

pageserver/src/http/openapi_spec.yml

@@ -308,6 +308,45 @@ paths:
             application/json:
               schema:
                 type: string
+
+  /v1/tenant/{tenant_shard_id}/timeline/{timeline_id}/block_gc:
+    parameters:
+      - name: tenant_shard_id
+        in: path
+        required: true
+        schema:
+          type: string
+      - name: timeline_id
+        in: path
+        required: true
+        schema:
+          type: string
+          format: hex
+    post:
+      description: Persistently add a gc blocking at the tenant level because of this timeline
+      responses:
+        "200":
+          description: OK
+
+  /v1/tenant/{tenant_shard_id}/timeline/{timeline_id}/unblock_gc:
+    parameters:
+      - name: tenant_shard_id
+        in: path
+        required: true
+        schema:
+          type: string
+      - name: timeline_id
+        in: path
+        required: true
+        schema:
+          type: string
+          format: hex
+    post:
+      description: Persistently remove a tenant level gc blocking for this timeline
+      responses:
+        "200":
+          description: OK
+
   /v1/tenant/{tenant_shard_id}/location_config:
     parameters:
       - name: tenant_shard_id
@@ -893,7 +932,7 @@ components:
           description: Whether to poll remote storage for layers to download.  If false, secondary locations don't download anything.
     ArchivalConfigRequest:
       type: object
-      required
+      required:
         - state
       properties:
         state:

pageserver/src/http/routes.rs

@@ -296,6 +296,11 @@ impl From<GetActiveTenantError> for ApiError {
             GetActiveTenantError::WaitForActiveTimeout { .. } => {
                 ApiError::ResourceUnavailable(format!("{}", e).into())
             }
+            GetActiveTenantError::SwitchedTenant => {
+                // in our HTTP handlers, this error doesn't happen
+                // TODO: separate error types
+                ApiError::ResourceUnavailable("switched tenant".into())
+            }
         }
     }
 }
@@ -930,6 +935,7 @@ async fn tenant_list_handler(
             generation: (*gen)
                 .into()
                 .expect("Tenants are always attached with a generation"),
+            gc_blocking: None,
         })
         .collect::<Vec<TenantInfo>>();
 
@@ -981,6 +987,7 @@ async fn tenant_status(
                     .generation()
                     .into()
                     .expect("Tenants are always attached with a generation"),
+                gc_blocking: tenant.gc_block.summary().map(|x| format!("{x:?}")),
             },
             walredo: tenant.wal_redo_manager_status(),
             timelines: tenant.list_timeline_ids(),
@@ -1155,7 +1162,10 @@ async fn layer_map_info_handler(
     let timeline =
         active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
             .await?;
-    let layer_map_info = timeline.layer_map_info(reset).await;
+    let layer_map_info = timeline
+        .layer_map_info(reset)
+        .await
+        .map_err(|_shutdown| ApiError::ShuttingDown)?;
 
     json_response(StatusCode::OK, layer_map_info)
 }
@@ -1221,6 +1231,72 @@ async fn evict_timeline_layer_handler(
     }
 }
 
+async fn timeline_gc_blocking_handler(
+    request: Request<Body>,
+    _cancel: CancellationToken,
+) -> Result<Response<Body>, ApiError> {
+    block_or_unblock_gc(request, true).await
+}
+
+async fn timeline_gc_unblocking_handler(
+    request: Request<Body>,
+    _cancel: CancellationToken,
+) -> Result<Response<Body>, ApiError> {
+    block_or_unblock_gc(request, false).await
+}
+
+/// Adding a block is `POST ../block_gc`, removing a block is `POST ../unblock_gc`.
+///
+/// Both are technically unsafe because they might fire off index uploads, thus they are POST.
+async fn block_or_unblock_gc(
+    request: Request<Body>,
+    block: bool,
+) -> Result<Response<Body>, ApiError> {
+    use crate::tenant::{
+        remote_timeline_client::WaitCompletionError, upload_queue::NotInitialized,
+    };
+    let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
+    check_permission(&request, Some(tenant_shard_id.tenant_id))?;
+    let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
+    let state = get_state(&request);
+
+    let tenant = state
+        .tenant_manager
+        .get_attached_tenant_shard(tenant_shard_id)?;
+
+    tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
+
+    let timeline = tenant.get_timeline(timeline_id, true)?;
+
+    let fut = async {
+        if block {
+            timeline.block_gc(&tenant).await.map(|_| ())
+        } else {
+            timeline.unblock_gc(&tenant).await
+        }
+    };
+
+    let span = tracing::info_span!(
+        "block_or_unblock_gc",
+        tenant_id = %tenant_shard_id.tenant_id,
+        shard_id = %tenant_shard_id.shard_slug(),
+        timeline_id = %timeline_id,
+        block = block,
+    );
+
+    let res = fut.instrument(span).await;
+
+    res.map_err(|e| {
+        if e.is::<NotInitialized>() || e.is::<WaitCompletionError>() {
+            ApiError::ShuttingDown
+        } else {
+            ApiError::InternalServerError(e)
+        }
+    })?;
+
+    json_response(StatusCode::OK, ())
+}
+
 /// Get tenant_size SVG graph along with the JSON data.
 fn synthetic_size_html_response(
     inputs: ModelInputs,
@@ -1711,9 +1787,11 @@ async fn timeline_checkpoint_handler(
         }
 
         if wait_until_uploaded {
+            tracing::info!("Waiting for uploads to complete...");
             timeline.remote_client.wait_completion().await
             // XXX map to correct ApiError for the cases where it's due to shutdown
             .context("wait completion").map_err(ApiError::InternalServerError)?;
+            tracing::info!("Uploads completed up to {}", timeline.get_remote_consistent_lsn_projected().unwrap_or(Lsn(0)));
         }
 
         json_response(StatusCode::OK, ())
@@ -1811,7 +1889,7 @@ async fn timeline_detach_ancestor_handler(
         // drop(tenant);
 
         let resp = match progress {
-            detach_ancestor::Progress::Prepared(_guard, prepared) => {
+            detach_ancestor::Progress::Prepared(attempt, prepared) => {
                 // it would be great to tag the guard on to the tenant activation future
                 let reparented_timelines = state
                     .tenant_manager
@@ -1819,10 +1897,10 @@ async fn timeline_detach_ancestor_handler(
                         tenant_shard_id,
                         timeline_id,
                         prepared,
+                        attempt,
                         ctx,
                     )
                     .await
-                    .context("timeline detach ancestor completion")
                     .map_err(ApiError::InternalServerError)?;
 
                 AncestorDetached {
@@ -2281,8 +2359,9 @@ async fn get_utilization(
     // regenerate at most 1Hz to allow polling at any rate.
     if !still_valid {
         let path = state.conf.tenants_path();
-        let doc = crate::utilization::regenerate(path.as_std_path())
-            .map_err(ApiError::InternalServerError)?;
+        let doc =
+            crate::utilization::regenerate(state.conf, path.as_std_path(), &state.tenant_manager)
+                .map_err(ApiError::InternalServerError)?;
 
         let mut buf = Vec::new();
         serde_json::to_writer(&mut buf, &doc)
@@ -2899,6 +2978,14 @@ pub fn make_router(
             "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/layer/:layer_file_name",
             |r| api_handler(r, evict_timeline_layer_handler),
         )
+        .post(
+            "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/block_gc",
+            |r| api_handler(r, timeline_gc_blocking_handler),
+        )
+        .post(
+            "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/unblock_gc",
+            |r| api_handler(r, timeline_gc_unblocking_handler),
+        )
         .post("/v1/tenant/:tenant_shard_id/heatmap_upload", |r| {
             api_handler(r, secondary_upload_handler)
         })

pageserver/src/l0_flush.rs

@@ -24,7 +24,7 @@ impl Default for L0FlushConfig {
 #[derive(Clone)]
 pub struct L0FlushGlobalState(Arc<Inner>);
 
-pub(crate) enum Inner {
+pub enum Inner {
     PageCached,
     Direct { semaphore: tokio::sync::Semaphore },
 }
@@ -40,7 +40,7 @@ impl L0FlushGlobalState {
         }
     }
 
-    pub(crate) fn inner(&self) -> &Arc<Inner> {
+    pub fn inner(&self) -> &Arc<Inner> {
         &self.0
     }
 }

pageserver/src/lib.rs

@@ -12,6 +12,8 @@ pub mod disk_usage_eviction_task;
 pub mod http;
 pub mod import_datadir;
 pub mod l0_flush;
+
+use futures::{stream::FuturesUnordered, StreamExt};
 pub use pageserver_api::keyspace;
 use tokio_util::sync::CancellationToken;
 pub mod aux_file;
@@ -30,14 +32,13 @@ pub mod walingest;
 pub mod walrecord;
 pub mod walredo;
 
-use crate::task_mgr::TaskKind;
 use camino::Utf8Path;
 use deletion_queue::DeletionQueue;
 use tenant::{
     mgr::{BackgroundPurges, TenantManager},
     secondary,
 };
-use tracing::info;
+use tracing::{info, info_span};
 
 /// Current storage format version
 ///
@@ -63,7 +64,6 @@ pub struct CancellableTask {
     pub cancel: CancellationToken,
 }
 pub struct HttpEndpointListener(pub CancellableTask);
-pub struct LibpqEndpointListener(pub CancellableTask);
 pub struct ConsumptionMetricsTasks(pub CancellableTask);
 pub struct DiskUsageEvictionTask(pub CancellableTask);
 impl CancellableTask {
@@ -77,7 +77,7 @@ impl CancellableTask {
 #[allow(clippy::too_many_arguments)]
 pub async fn shutdown_pageserver(
     http_listener: HttpEndpointListener,
-    libpq_listener: LibpqEndpointListener,
+    page_service: page_service::Listener,
     consumption_metrics_worker: ConsumptionMetricsTasks,
     disk_usage_eviction_task: Option<DiskUsageEvictionTask>,
     tenant_manager: &TenantManager,
@@ -87,10 +87,83 @@ pub async fn shutdown_pageserver(
     exit_code: i32,
 ) {
     use std::time::Duration;
+
+    // If the orderly shutdown below takes too long, we still want to make
+    // sure that all walredo processes are killed and wait()ed on by us, not systemd.
+    //
+    // (Leftover walredo processes are the hypothesized trigger for the systemd freezes
+    //  that we keep seeing in prod => https://github.com/neondatabase/cloud/issues/11387.
+    //
+    // We use a thread instead of a tokio task because the background runtime is likely busy
+    // with the final flushing / uploads. This activity here has priority, and due to lack
+    // of scheduling priority feature sin the tokio scheduler, using a separate thread is
+    // an effective priority booster.
+    let walredo_extraordinary_shutdown_thread_span = {
+        let span = info_span!(parent: None, "walredo_extraordinary_shutdown_thread");
+        span.follows_from(tracing::Span::current());
+        span
+    };
+    let walredo_extraordinary_shutdown_thread_cancel = CancellationToken::new();
+    let walredo_extraordinary_shutdown_thread = std::thread::spawn({
+        let walredo_extraordinary_shutdown_thread_cancel =
+            walredo_extraordinary_shutdown_thread_cancel.clone();
+        move || {
+            let rt = tokio::runtime::Builder::new_current_thread()
+                .enable_all()
+                .build()
+                .unwrap();
+            let _entered = rt.enter();
+            let _entered = walredo_extraordinary_shutdown_thread_span.enter();
+            if let Ok(()) = rt.block_on(tokio::time::timeout(
+                Duration::from_secs(8),
+                walredo_extraordinary_shutdown_thread_cancel.cancelled(),
+            )) {
+                info!("cancellation requested");
+                return;
+            }
+            let managers = tenant::WALREDO_MANAGERS
+                .lock()
+                .unwrap()
+                // prevents new walredo managers from being inserted
+                .take()
+                .expect("only we take()");
+            // Use FuturesUnordered to get in queue early for each manager's
+            // heavier_once_cell semaphore wait list.
+            // Also, for idle tenants that for some reason haven't
+            // shut down yet, it's quite likely that we're not going
+            // to get Poll::Pending once.
+            let mut futs: FuturesUnordered<_> = managers
+                .into_iter()
+                .filter_map(|(_, mgr)| mgr.upgrade())
+                .map(|mgr| async move { tokio::task::unconstrained(mgr.shutdown()).await })
+                .collect();
+            info!(count=%futs.len(), "built FuturesUnordered");
+            let mut last_log_at = std::time::Instant::now();
+            #[derive(Debug, Default)]
+            struct Results {
+                initiated: u64,
+                already: u64,
+            }
+            let mut results = Results::default();
+            while let Some(we_initiated) = rt.block_on(futs.next()) {
+                if we_initiated {
+                    results.initiated += 1;
+                } else {
+                    results.already += 1;
+                }
+                if last_log_at.elapsed() > Duration::from_millis(100) {
+                    info!(remaining=%futs.len(), ?results, "progress");
+                    last_log_at = std::time::Instant::now();
+                }
+            }
+            info!(?results, "done");
+        }
+    });
+
     // Shut down the libpq endpoint task. This prevents new connections from
     // being accepted.
-    timed(
-        libpq_listener.0.shutdown(),
+    let remaining_connections = timed(
+        page_service.stop_accepting(),
         "shutdown LibpqEndpointListener",
         Duration::from_secs(1),
     )
@@ -108,7 +181,7 @@ pub async fn shutdown_pageserver(
     // Shut down any page service tasks: any in-progress work for particular timelines or tenants
     // should already have been canclled via mgr::shutdown_all_tenants
     timed(
-        task_mgr::shutdown_tasks(Some(TaskKind::PageRequestHandler), None, None),
+        remaining_connections.shutdown(),
         "shutdown PageRequestHandlers",
         Duration::from_secs(1),
     )
@@ -162,6 +235,12 @@ pub async fn shutdown_pageserver(
         Duration::from_secs(1),
     )
     .await;
+
+    info!("cancel & join walredo_extraordinary_shutdown_thread");
+    walredo_extraordinary_shutdown_thread_cancel.cancel();
+    walredo_extraordinary_shutdown_thread.join().unwrap();
+    info!("walredo_extraordinary_shutdown_thread done");
+
     info!("Shut down successfully completed");
     std::process::exit(exit_code);
 }

pageserver/src/metrics.rs

@@ -525,6 +525,15 @@ static RESIDENT_PHYSICAL_SIZE: Lazy<UIntGaugeVec> = Lazy::new(|| {
     .expect("failed to define a metric")
 });
 
+static VISIBLE_PHYSICAL_SIZE: Lazy<UIntGaugeVec> = Lazy::new(|| {
+    register_uint_gauge_vec!(
+        "pageserver_visible_physical_size",
+        "The size of the layer files present in the pageserver's filesystem.",
+        &["tenant_id", "shard_id", "timeline_id"]
+    )
+    .expect("failed to define a metric")
+});
+
 pub(crate) static RESIDENT_PHYSICAL_SIZE_GLOBAL: Lazy<UIntGauge> = Lazy::new(|| {
     register_uint_gauge!(
         "pageserver_resident_physical_size_global",
@@ -2204,6 +2213,7 @@ pub(crate) struct TimelineMetrics {
     pub(crate) layer_count_delta: UIntGauge,
     pub standby_horizon_gauge: IntGauge,
     pub resident_physical_size_gauge: UIntGauge,
+    pub visible_physical_size_gauge: UIntGauge,
     /// copy of LayeredTimeline.current_logical_size
     pub current_logical_size_gauge: UIntGauge,
     pub aux_file_size_gauge: IntGauge,
@@ -2326,6 +2336,9 @@ impl TimelineMetrics {
         let resident_physical_size_gauge = RESIDENT_PHYSICAL_SIZE
             .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
             .unwrap();
+        let visible_physical_size_gauge = VISIBLE_PHYSICAL_SIZE
+            .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
+            .unwrap();
         // TODO: we shouldn't expose this metric
         let current_logical_size_gauge = CURRENT_LOGICAL_SIZE
             .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
@@ -2380,6 +2393,7 @@ impl TimelineMetrics {
             layer_count_delta,
             standby_horizon_gauge,
             resident_physical_size_gauge,
+            visible_physical_size_gauge,
             current_logical_size_gauge,
             aux_file_size_gauge,
             directory_entries_count_gauge,
@@ -2431,6 +2445,7 @@ impl TimelineMetrics {
             RESIDENT_PHYSICAL_SIZE_GLOBAL.sub(self.resident_physical_size_get());
             let _ = RESIDENT_PHYSICAL_SIZE.remove_label_values(&[tenant_id, shard_id, timeline_id]);
         }
+        let _ = VISIBLE_PHYSICAL_SIZE.remove_label_values(&[tenant_id, shard_id, timeline_id]);
         let _ = CURRENT_LOGICAL_SIZE.remove_label_values(&[tenant_id, shard_id, timeline_id]);
         if let Some(metric) = Lazy::get(&DIRECTORY_ENTRIES_COUNT) {
             let _ = metric.remove_label_values(&[tenant_id, shard_id, timeline_id]);

pageserver/src/statvfs.rs

@@ -56,7 +56,6 @@ impl Statvfs {
 }
 
 pub mod mock {
-    use anyhow::Context;
     use camino::Utf8Path;
     use regex::Regex;
     use tracing::log::info;
@@ -135,14 +134,30 @@ pub mod mock {
             {
                 continue;
             }
-            total += entry
-                .metadata()
-                .with_context(|| format!("get metadata of {:?}", entry.path()))?
-                .len();
+            let m = match entry.metadata() {
+                Ok(m) => m,
+                Err(e) if is_not_found(&e) => {
+                    // some temp file which got removed right as we are walking
+                    continue;
+                }
+                Err(e) => {
+                    return Err(anyhow::Error::new(e)
+                        .context(format!("get metadata of {:?}", entry.path())))
+                }
+            };
+            total += m.len();
         }
         Ok(total)
     }
 
+    fn is_not_found(e: &walkdir::Error) -> bool {
+        let Some(io_error) = e.io_error() else {
+            return false;
+        };
+        let kind = io_error.kind();
+        matches!(kind, std::io::ErrorKind::NotFound)
+    }
+
     pub struct Statvfs {
         pub blocks: u64,
         pub blocks_available: u64,

pageserver/src/tenant.rs

@@ -33,6 +33,7 @@ use remote_storage::GenericRemoteStorage;
 use remote_storage::TimeoutOrCancel;
 use std::collections::BTreeMap;
 use std::fmt;
+use std::sync::Weak;
 use std::time::SystemTime;
 use storage_broker::BrokerClientChannel;
 use tokio::io::BufReader;
@@ -40,6 +41,7 @@ use tokio::sync::watch;
 use tokio::task::JoinSet;
 use tokio_util::sync::CancellationToken;
 use tracing::*;
+use upload_queue::NotInitialized;
 use utils::backoff;
 use utils::circuit_breaker::CircuitBreaker;
 use utils::completion;
@@ -147,6 +149,7 @@ pub(crate) mod timeline;
 
 pub mod size;
 
+mod gc_block;
 pub(crate) mod throttle;
 
 pub(crate) use crate::span::debug_assert_current_span_has_tenant_and_timeline_id;
@@ -299,9 +302,19 @@ pub struct Tenant {
     pub(crate) timeline_get_throttle:
         Arc<throttle::Throttle<&'static crate::metrics::tenant_throttling::TimelineGet>>,
 
-    /// An ongoing timeline detach must be checked during attempts to GC or compact a timeline.
+    /// An ongoing timeline detach concurrency limiter.
+    ///
+    /// As a tenant will likely be restarted as part of timeline detach ancestor it makes no sense
+    /// to have two running at the same time. A different one can be started if an earlier one
+    /// has failed for whatever reason.
     ongoing_timeline_detach: std::sync::Mutex<Option<(TimelineId, utils::completion::Barrier)>>,
 
+    /// `index_part.json` based gc blocking reason tracking.
+    ///
+    /// New gc iterations must start a new iteration by acquiring `GcBlock::start` before
+    /// proceeding.
+    pub(crate) gc_block: gc_block::GcBlock,
+
     l0_flush_global_state: L0FlushGlobalState,
 }
 
@@ -312,14 +325,66 @@ impl std::fmt::Debug for Tenant {
 }
 
 pub(crate) enum WalRedoManager {
-    Prod(PostgresRedoManager),
+    Prod(WalredoManagerId, PostgresRedoManager),
     #[cfg(test)]
     Test(harness::TestRedoManager),
 }
 
-impl From<PostgresRedoManager> for WalRedoManager {
-    fn from(mgr: PostgresRedoManager) -> Self {
-        Self::Prod(mgr)
+#[derive(thiserror::Error, Debug)]
+#[error("pageserver is shutting down")]
+pub(crate) struct GlobalShutDown;
+
+impl WalRedoManager {
+    pub(crate) fn new(mgr: PostgresRedoManager) -> Result<Arc<Self>, GlobalShutDown> {
+        let id = WalredoManagerId::next();
+        let arc = Arc::new(Self::Prod(id, mgr));
+        let mut guard = WALREDO_MANAGERS.lock().unwrap();
+        match &mut *guard {
+            Some(map) => {
+                map.insert(id, Arc::downgrade(&arc));
+                Ok(arc)
+            }
+            None => Err(GlobalShutDown),
+        }
+    }
+}
+
+impl Drop for WalRedoManager {
+    fn drop(&mut self) {
+        match self {
+            Self::Prod(id, _) => {
+                let mut guard = WALREDO_MANAGERS.lock().unwrap();
+                if let Some(map) = &mut *guard {
+                    map.remove(id).expect("new() registers, drop() unregisters");
+                }
+            }
+            #[cfg(test)]
+            Self::Test(_) => {
+                // Not applicable to test redo manager
+            }
+        }
+    }
+}
+
+/// Global registry of all walredo managers so that [`crate::shutdown_pageserver`] can shut down
+/// the walredo processes outside of the regular order.
+///
+/// This is necessary to work around a systemd bug where it freezes if there are
+/// walredo processes left => <https://github.com/neondatabase/cloud/issues/11387>
+#[allow(clippy::type_complexity)]
+pub(crate) static WALREDO_MANAGERS: once_cell::sync::Lazy<
+    Mutex<Option<HashMap<WalredoManagerId, Weak<WalRedoManager>>>>,
+> = once_cell::sync::Lazy::new(|| Mutex::new(Some(HashMap::new())));
+#[derive(PartialEq, Eq, Hash, Clone, Copy, Debug)]
+pub(crate) struct WalredoManagerId(u64);
+impl WalredoManagerId {
+    pub fn next() -> Self {
+        static NEXT: std::sync::atomic::AtomicU64 = std::sync::atomic::AtomicU64::new(1);
+        let id = NEXT.fetch_add(1, std::sync::atomic::Ordering::Relaxed);
+        if id == 0 {
+            panic!("WalredoManagerId::new() returned 0, indicating wraparound, risking it's no longer unique");
+        }
+        Self(id)
     }
 }
 
@@ -331,19 +396,20 @@ impl From<harness::TestRedoManager> for WalRedoManager {
 }
 
 impl WalRedoManager {
-    pub(crate) async fn shutdown(&self) {
+    pub(crate) async fn shutdown(&self) -> bool {
         match self {
-            Self::Prod(mgr) => mgr.shutdown().await,
+            Self::Prod(_, mgr) => mgr.shutdown().await,
             #[cfg(test)]
             Self::Test(_) => {
                 // Not applicable to test redo manager
+                true
             }
         }
     }
 
     pub(crate) fn maybe_quiesce(&self, idle_timeout: Duration) {
         match self {
-            Self::Prod(mgr) => mgr.maybe_quiesce(idle_timeout),
+            Self::Prod(_, mgr) => mgr.maybe_quiesce(idle_timeout),
             #[cfg(test)]
             Self::Test(_) => {
                 // Not applicable to test redo manager
@@ -363,7 +429,7 @@ impl WalRedoManager {
         pg_version: u32,
     ) -> Result<bytes::Bytes, walredo::Error> {
         match self {
-            Self::Prod(mgr) => {
+            Self::Prod(_, mgr) => {
                 mgr.request_redo(key, lsn, base_img, records, pg_version)
                     .await
             }
@@ -377,7 +443,7 @@ impl WalRedoManager {
 
     pub(crate) fn status(&self) -> Option<WalRedoManagerStatus> {
         match self {
-            WalRedoManager::Prod(m) => Some(m.status()),
+            WalRedoManager::Prod(_, m) => Some(m.status()),
             #[cfg(test)]
             WalRedoManager::Test(_) => None,
         }
@@ -386,6 +452,8 @@ impl WalRedoManager {
 
 #[derive(Debug, thiserror::Error, PartialEq, Eq)]
 pub enum GetTimelineError {
+    #[error("Timeline is shutting down")]
+    ShuttingDown,
     #[error("Timeline {tenant_id}/{timeline_id} is not active, state: {state:?}")]
     NotActive {
         tenant_id: TenantShardId,
@@ -538,6 +606,21 @@ impl From<PageReconstructError> for GcError {
     }
 }
 
+impl From<NotInitialized> for GcError {
+    fn from(value: NotInitialized) -> Self {
+        match value {
+            NotInitialized::Uninitialized => GcError::Remote(value.into()),
+            NotInitialized::Stopped | NotInitialized::ShuttingDown => GcError::TimelineCancelled,
+        }
+    }
+}
+
+impl From<timeline::layer_manager::Shutdown> for GcError {
+    fn from(_: timeline::layer_manager::Shutdown) -> Self {
+        GcError::TimelineCancelled
+    }
+}
+
 #[derive(thiserror::Error, Debug)]
 pub(crate) enum LoadConfigError {
     #[error("TOML deserialization error: '{0}'")]
@@ -647,6 +730,7 @@ impl Tenant {
                     .read()
                     .await
                     .layer_map()
+                    .expect("currently loading, layer manager cannot be shutdown already")
                     .iter_historic_layers()
                     .next()
                     .is_some(),
@@ -675,11 +759,9 @@ impl Tenant {
         init_order: Option<InitializationOrder>,
         mode: SpawnMode,
         ctx: &RequestContext,
-    ) -> Arc<Tenant> {
-        let wal_redo_manager = Arc::new(WalRedoManager::from(PostgresRedoManager::new(
-            conf,
-            tenant_shard_id,
-        )));
+    ) -> Result<Arc<Tenant>, GlobalShutDown> {
+        let wal_redo_manager =
+            WalRedoManager::new(PostgresRedoManager::new(conf, tenant_shard_id))?;
 
         let TenantSharedResources {
             broker_client,
@@ -755,9 +837,9 @@ impl Tenant {
                             // The Stopping case is for when we have passed control on to DeleteTenantFlow:
                             // if it errors, we will call make_broken when tenant is already in Stopping.
                             assert!(
-                            matches!(*state, TenantState::Attaching | TenantState::Stopping { .. }),
-                            "the attach task owns the tenant state until activation is complete"
-                        );
+                                matches!(*state, TenantState::Attaching | TenantState::Stopping { .. }),
+                                "the attach task owns the tenant state until activation is complete"
+                            );
 
                             *state = TenantState::broken_from_reason(err.to_string());
                         });
@@ -878,7 +960,7 @@ impl Tenant {
             }
             .instrument(tracing::info_span!(parent: None, "attach", tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(), gen=?generation)),
         );
-        tenant
+        Ok(tenant)
     }
 
     #[instrument(skip_all)]
@@ -982,6 +1064,8 @@ impl Tenant {
             }
         }
 
+        let mut gc_blocks = HashMap::new();
+
         // For every timeline, download the metadata file, scan the local directory,
         // and build a layer map that contains an entry for each remote and local
         // layer file.
@@ -991,6 +1075,16 @@ impl Tenant {
                 .remove(&timeline_id)
                 .expect("just put it in above");
 
+            if let Some(blocking) = index_part.gc_blocking.as_ref() {
+                // could just filter these away, but it helps while testing
+                anyhow::ensure!(
+                    !blocking.reasons.is_empty(),
+                    "index_part for {timeline_id} is malformed: it should not have gc blocking with zero reasons"
+                );
+                let prev = gc_blocks.insert(timeline_id, blocking.reasons);
+                assert!(prev.is_none());
+            }
+
             // TODO again handle early failure
             self.load_remote_timeline(
                 timeline_id,
@@ -1035,6 +1129,8 @@ impl Tenant {
         // IndexPart is the source of truth.
         self.clean_up_timelines(&existent_timelines)?;
 
+        self.gc_block.set_scanned(gc_blocks);
+
         fail::fail_point!("attach-before-activate", |_| {
             anyhow::bail!("attach-before-activate");
         });
@@ -1580,7 +1676,7 @@ impl Tenant {
         self: Arc<Self>,
         timeline_id: TimelineId,
     ) -> Result<(), DeleteTimelineError> {
-        DeleteTimelineFlow::run(&self, timeline_id, false).await?;
+        DeleteTimelineFlow::run(&self, timeline_id).await?;
 
         Ok(())
     }
@@ -1625,6 +1721,14 @@ impl Tenant {
             }
         }
 
+        let _guard = match self.gc_block.start().await {
+            Ok(guard) => guard,
+            Err(reasons) => {
+                info!("Skipping GC: {reasons}");
+                return Ok(GcResult::default());
+            }
+        };
+
         self.gc_iteration_internal(target_timeline_id, horizon, pitr, cancel, ctx)
             .await
     }
@@ -2637,6 +2741,7 @@ impl Tenant {
             )),
             tenant_conf: Arc::new(ArcSwap::from_pointee(attached_conf)),
             ongoing_timeline_detach: std::sync::Mutex::default(),
+            gc_block: Default::default(),
             l0_flush_global_state,
         }
     }
@@ -2921,54 +3026,6 @@ impl Tenant {
         // because that will stall branch creation.
         let gc_cs = self.gc_cs.lock().await;
 
-        // Paranoia check: it is critical that GcInfo's list of child timelines is correct, to avoid incorrectly GC'ing data they
-        // depend on.  So although GcInfo is updated continuously by Timeline::new and Timeline::drop, we also calculate it here
-        // and fail out if it's inaccurate.
-        // (this can be removed later, it's a risk mitigation for https://github.com/neondatabase/neon/pull/8427)
-        {
-            let mut all_branchpoints: BTreeMap<TimelineId, Vec<(Lsn, TimelineId)>> =
-                BTreeMap::new();
-            timelines.iter().for_each(|timeline| {
-                if let Some(ancestor_timeline_id) = &timeline.get_ancestor_timeline_id() {
-                    let ancestor_children =
-                        all_branchpoints.entry(*ancestor_timeline_id).or_default();
-                    ancestor_children.push((timeline.get_ancestor_lsn(), timeline.timeline_id));
-                }
-            });
-
-            for timeline in &timelines {
-                let mut branchpoints: Vec<(Lsn, TimelineId)> = all_branchpoints
-                    .remove(&timeline.timeline_id)
-                    .unwrap_or_default();
-
-                branchpoints.sort_by_key(|b| b.0);
-
-                let target = timeline.gc_info.read().unwrap();
-
-                // We require that retain_lsns contains everything in `branchpoints`, but not that
-                // they are exactly equal: timeline deletions can race with us, so retain_lsns
-                // may contain some extra stuff.  It is safe to have extra timelines in there, because it
-                // just means that we retain slightly more data than we otherwise might.
-                let have_branchpoints = target.retain_lsns.iter().copied().collect::<HashSet<_>>();
-                for b in &branchpoints {
-                    if !have_branchpoints.contains(b) {
-                        tracing::error!(
-                            "Bug: `retain_lsns` is set incorrectly.  Expected be {:?}, but found {:?}",
-                            branchpoints,
-                            target.retain_lsns
-                        );
-                        debug_assert!(false);
-                        // Do not GC based on bad information!
-                        // (ab-use an existing GcError type rather than adding a new one, since this is a
-                        // "should never happen" check that will be removed soon).
-                        return Err(GcError::Remote(anyhow::anyhow!(
-                            "retain_lsns failed validation!"
-                        )));
-                    }
-                }
-            }
-        }
-
         // Ok, we now know all the branch points.
         // Update the GC information for each timeline.
         let mut gc_timelines = Vec::with_capacity(timelines.len());
@@ -3679,6 +3736,19 @@ impl Tenant {
     pub(crate) fn get_tenant_conf(&self) -> TenantConfOpt {
         self.tenant_conf.load().tenant_conf.clone()
     }
+
+    /// How much local storage would this tenant like to have?  It can cope with
+    /// less than this (via eviction and on-demand downloads), but this function enables
+    /// the Tenant to advertise how much storage it would prefer to have to provide fast I/O
+    /// by keeping important things on local disk.
+    pub(crate) fn local_storage_wanted(&self) -> u64 {
+        let mut wanted = 0;
+        let timelines = self.timelines.lock().unwrap();
+        for timeline in timelines.values() {
+            wanted += timeline.metrics.visible_physical_size_gauge.get();
+        }
+        wanted
+    }
 }
 
 /// Create the cluster temporarily in 'initdbpath' directory inside the repository
@@ -4038,7 +4108,7 @@ pub(crate) mod harness {
 
 #[cfg(test)]
 mod tests {
-    use std::collections::BTreeMap;
+    use std::collections::{BTreeMap, BTreeSet};
 
     use super::*;
     use crate::keyspace::KeySpaceAccum;
@@ -4590,10 +4660,10 @@ mod tests {
 
         let layer_map = tline.layers.read().await;
         let level0_deltas = layer_map
-            .layer_map()
-            .get_level0_deltas()
-            .into_iter()
-            .map(|desc| layer_map.get_from_desc(&desc))
+            .layer_map()?
+            .level0_deltas()
+            .iter()
+            .map(|desc| layer_map.get_from_desc(desc))
             .collect::<Vec<_>>();
 
         assert!(!level0_deltas.is_empty());
@@ -4713,7 +4783,7 @@ mod tests {
         lsn: Lsn,
         repeat: usize,
         key_count: usize,
-    ) -> anyhow::Result<()> {
+    ) -> anyhow::Result<HashMap<Key, BTreeSet<Lsn>>> {
         let compact = true;
         bulk_insert_maybe_compact_gc(tenant, timeline, ctx, lsn, repeat, key_count, compact).await
     }
@@ -4726,7 +4796,9 @@ mod tests {
         repeat: usize,
         key_count: usize,
         compact: bool,
-    ) -> anyhow::Result<()> {
+    ) -> anyhow::Result<HashMap<Key, BTreeSet<Lsn>>> {
+        let mut inserted: HashMap<Key, BTreeSet<Lsn>> = Default::default();
+
         let mut test_key = Key::from_hex("010000000033333333444444445500000000").unwrap();
         let mut blknum = 0;
 
@@ -4747,6 +4819,7 @@ mod tests {
                         ctx,
                     )
                     .await?;
+                inserted.entry(test_key).or_default().insert(lsn);
                 writer.finish_write(lsn);
                 drop(writer);
 
@@ -4771,7 +4844,7 @@ mod tests {
             assert_eq!(res.layers_removed, 0, "this never removes anything");
         }
 
-        Ok(())
+        Ok(inserted)
     }
 
     //
@@ -4818,14 +4891,16 @@ mod tests {
             .await?;
 
         let lsn = Lsn(0x10);
-        bulk_insert_compact_gc(&tenant, &tline, &ctx, lsn, 50, 10000).await?;
+        let inserted = bulk_insert_compact_gc(&tenant, &tline, &ctx, lsn, 50, 10000).await?;
 
         let guard = tline.layers.read().await;
-        guard.layer_map().dump(true, &ctx).await?;
+        let lm = guard.layer_map()?;
+
+        lm.dump(true, &ctx).await?;
 
         let mut reads = Vec::new();
         let mut prev = None;
-        guard.layer_map().iter_historic_layers().for_each(|desc| {
+        lm.iter_historic_layers().for_each(|desc| {
             if !desc.is_delta() {
                 prev = Some(desc.clone());
                 return;
@@ -4879,9 +4954,39 @@ mod tests {
                     &ctx,
                 )
                 .await;
-            tline
-                .validate_get_vectored_impl(&vectored_res, read, reads_lsn, &ctx)
-                .await;
+
+            let mut expected_lsns: HashMap<Key, Lsn> = Default::default();
+            let mut expect_missing = false;
+            let mut key = read.start().unwrap();
+            while key != read.end().unwrap() {
+                if let Some(lsns) = inserted.get(&key) {
+                    let expected_lsn = lsns.iter().rfind(|lsn| **lsn <= reads_lsn);
+                    match expected_lsn {
+                        Some(lsn) => {
+                            expected_lsns.insert(key, *lsn);
+                        }
+                        None => {
+                            expect_missing = true;
+                            break;
+                        }
+                    }
+                } else {
+                    expect_missing = true;
+                    break;
+                }
+
+                key = key.next();
+            }
+
+            if expect_missing {
+                assert!(matches!(vectored_res, Err(GetVectoredError::MissingKey(_))));
+            } else {
+                for (key, image) in vectored_res? {
+                    let expected_lsn = expected_lsns.get(&key).expect("determined above");
+                    let expected_image = test_img(&format!("{} at {}", key.field6, expected_lsn));
+                    assert_eq!(image?, expected_image);
+                }
+            }
         }
 
         Ok(())
@@ -4931,10 +5036,6 @@ mod tests {
             )
             .await;
 
-        child_timeline
-            .validate_get_vectored_impl(&vectored_res, aux_keyspace, read_lsn, &ctx)
-            .await;
-
         let images = vectored_res?;
         assert!(images.is_empty());
         Ok(())
@@ -5805,23 +5906,12 @@ mod tests {
             tline.freeze_and_flush().await?; // force create a delta layer
         }
 
-        let before_num_l0_delta_files = tline
-            .layers
-            .read()
-            .await
-            .layer_map()
-            .get_level0_deltas()
-            .len();
+        let before_num_l0_delta_files =
+            tline.layers.read().await.layer_map()?.level0_deltas().len();
 
         tline.compact(&cancel, EnumSet::empty(), &ctx).await?;
 
-        let after_num_l0_delta_files = tline
-            .layers
-            .read()
-            .await
-            .layer_map()
-            .get_level0_deltas()
-            .len();
+        let after_num_l0_delta_files = tline.layers.read().await.layer_map()?.level0_deltas().len();
 
         assert!(after_num_l0_delta_files < before_num_l0_delta_files, "after_num_l0_delta_files={after_num_l0_delta_files}, before_num_l0_delta_files={before_num_l0_delta_files}");
 
@@ -6845,7 +6935,10 @@ mod tests {
         }
 
         let cancel = CancellationToken::new();
-        tline.compact_with_gc(&cancel, &ctx).await.unwrap();
+        tline
+            .compact_with_gc(&cancel, EnumSet::new(), &ctx)
+            .await
+            .unwrap();
 
         for (idx, expected) in expected_result.iter().enumerate() {
             assert_eq!(
@@ -6909,7 +7002,11 @@ mod tests {
             vec![
                 // Image layer at GC horizon
                 PersistentLayerKey {
-                    key_range: Key::MIN..Key::MAX,
+                    key_range: {
+                        let mut key = Key::MAX;
+                        key.field6 -= 1;
+                        Key::MIN..key
+                    },
                     lsn_range: Lsn(0x30)..Lsn(0x31),
                     is_delta: false
                 },
@@ -6928,6 +7025,18 @@ mod tests {
             ]
         );
 
+        // increase GC horizon and compact again
+        {
+            // Update GC info
+            let mut guard = tline.gc_info.write().unwrap();
+            guard.cutoffs.time = Lsn(0x40);
+            guard.cutoffs.space = Lsn(0x40);
+        }
+        tline
+            .compact_with_gc(&cancel, EnumSet::new(), &ctx)
+            .await
+            .unwrap();
+
         Ok(())
     }
 
@@ -7260,7 +7369,10 @@ mod tests {
         }
 
         let cancel = CancellationToken::new();
-        tline.compact_with_gc(&cancel, &ctx).await.unwrap();
+        tline
+            .compact_with_gc(&cancel, EnumSet::new(), &ctx)
+            .await
+            .unwrap();
 
         for idx in 0..10 {
             assert_eq!(
@@ -7279,6 +7391,18 @@ mod tests {
             );
         }
 
+        // increase GC horizon and compact again
+        {
+            // Update GC info
+            let mut guard = tline.gc_info.write().unwrap();
+            guard.cutoffs.time = Lsn(0x40);
+            guard.cutoffs.space = Lsn(0x40);
+        }
+        tline
+            .compact_with_gc(&cancel, EnumSet::new(), &ctx)
+            .await
+            .unwrap();
+
         Ok(())
     }
 
@@ -7347,6 +7471,7 @@ mod tests {
                 Lsn(0x60),
                 &[Lsn(0x20), Lsn(0x40), Lsn(0x50)],
                 3,
+                None,
             )
             .await
             .unwrap();
@@ -7471,7 +7596,7 @@ mod tests {
             ),
         ];
         let res = tline
-            .generate_key_retention(key, &history, Lsn(0x60), &[Lsn(0x40), Lsn(0x50)], 3)
+            .generate_key_retention(key, &history, Lsn(0x60), &[Lsn(0x40), Lsn(0x50)], 3, None)
             .await
             .unwrap();
         let expected_res = KeyHistoryRetention {
@@ -7517,6 +7642,114 @@ mod tests {
         };
         assert_eq!(res, expected_res);
 
+        // In case of branch compaction, the branch itself does not have the full history, and we need to provide
+        // the ancestor image in the test case.
+
+        let history = vec![
+            (
+                key,
+                Lsn(0x20),
+                Value::WalRecord(NeonWalRecord::wal_append(";0x20")),
+            ),
+            (
+                key,
+                Lsn(0x30),
+                Value::WalRecord(NeonWalRecord::wal_append(";0x30")),
+            ),
+            (
+                key,
+                Lsn(0x40),
+                Value::WalRecord(NeonWalRecord::wal_append(";0x40")),
+            ),
+            (
+                key,
+                Lsn(0x70),
+                Value::WalRecord(NeonWalRecord::wal_append(";0x70")),
+            ),
+        ];
+        let res = tline
+            .generate_key_retention(
+                key,
+                &history,
+                Lsn(0x60),
+                &[],
+                3,
+                Some((key, Lsn(0x10), Bytes::copy_from_slice(b"0x10"))),
+            )
+            .await
+            .unwrap();
+        let expected_res = KeyHistoryRetention {
+            below_horizon: vec![(
+                Lsn(0x60),
+                KeyLogAtLsn(vec![(
+                    Lsn(0x60),
+                    Value::Image(Bytes::copy_from_slice(b"0x10;0x20;0x30;0x40")), // use the ancestor image to reconstruct the page
+                )]),
+            )],
+            above_horizon: KeyLogAtLsn(vec![(
+                Lsn(0x70),
+                Value::WalRecord(NeonWalRecord::wal_append(";0x70")),
+            )]),
+        };
+        assert_eq!(res, expected_res);
+
+        let history = vec![
+            (
+                key,
+                Lsn(0x20),
+                Value::WalRecord(NeonWalRecord::wal_append(";0x20")),
+            ),
+            (
+                key,
+                Lsn(0x40),
+                Value::WalRecord(NeonWalRecord::wal_append(";0x40")),
+            ),
+            (
+                key,
+                Lsn(0x60),
+                Value::WalRecord(NeonWalRecord::wal_append(";0x60")),
+            ),
+            (
+                key,
+                Lsn(0x70),
+                Value::WalRecord(NeonWalRecord::wal_append(";0x70")),
+            ),
+        ];
+        let res = tline
+            .generate_key_retention(
+                key,
+                &history,
+                Lsn(0x60),
+                &[Lsn(0x30)],
+                3,
+                Some((key, Lsn(0x10), Bytes::copy_from_slice(b"0x10"))),
+            )
+            .await
+            .unwrap();
+        let expected_res = KeyHistoryRetention {
+            below_horizon: vec![
+                (
+                    Lsn(0x30),
+                    KeyLogAtLsn(vec![(
+                        Lsn(0x20),
+                        Value::WalRecord(NeonWalRecord::wal_append(";0x20")),
+                    )]),
+                ),
+                (
+                    Lsn(0x60),
+                    KeyLogAtLsn(vec![(
+                        Lsn(0x60),
+                        Value::Image(Bytes::copy_from_slice(b"0x10;0x20;0x40;0x60")),
+                    )]),
+                ),
+            ],
+            above_horizon: KeyLogAtLsn(vec![(
+                Lsn(0x70),
+                Value::WalRecord(NeonWalRecord::wal_append(";0x70")),
+            )]),
+        };
+        assert_eq!(res, expected_res);
+
         Ok(())
     }
 
@@ -7674,6 +7907,10 @@ mod tests {
         ];
 
         let verify_result = || async {
+            let gc_horizon = {
+                let gc_info = tline.gc_info.read().unwrap();
+                gc_info.cutoffs.time
+            };
             for idx in 0..10 {
                 assert_eq!(
                     tline
@@ -7684,7 +7921,7 @@ mod tests {
                 );
                 assert_eq!(
                     tline
-                        .get(get_key(idx as u32), Lsn(0x30), &ctx)
+                        .get(get_key(idx as u32), gc_horizon, &ctx)
                         .await
                         .unwrap(),
                     &expected_result_at_gc_horizon[idx]
@@ -7709,7 +7946,232 @@ mod tests {
         verify_result().await;
 
         let cancel = CancellationToken::new();
-        tline.compact_with_gc(&cancel, &ctx).await.unwrap();
+        let mut dryrun_flags = EnumSet::new();
+        dryrun_flags.insert(CompactFlags::DryRun);
+
+        tline
+            .compact_with_gc(&cancel, dryrun_flags, &ctx)
+            .await
+            .unwrap();
+        // We expect layer map to be the same b/c the dry run flag, but we don't know whether there will be other background jobs
+        // cleaning things up, and therefore, we don't do sanity checks on the layer map during unit tests.
+        verify_result().await;
+
+        tline
+            .compact_with_gc(&cancel, EnumSet::new(), &ctx)
+            .await
+            .unwrap();
+        verify_result().await;
+
+        // compact again
+        tline
+            .compact_with_gc(&cancel, EnumSet::new(), &ctx)
+            .await
+            .unwrap();
+        verify_result().await;
+
+        // increase GC horizon and compact again
+        {
+            // Update GC info
+            let mut guard = tline.gc_info.write().unwrap();
+            guard.cutoffs.time = Lsn(0x38);
+            guard.cutoffs.space = Lsn(0x38);
+        }
+        tline
+            .compact_with_gc(&cancel, EnumSet::new(), &ctx)
+            .await
+            .unwrap();
+        verify_result().await; // no wals between 0x30 and 0x38, so we should obtain the same result
+
+        // not increasing the GC horizon and compact again
+        tline
+            .compact_with_gc(&cancel, EnumSet::new(), &ctx)
+            .await
+            .unwrap();
+        verify_result().await;
+
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn test_simple_bottom_most_compaction_on_branch() -> anyhow::Result<()> {
+        let harness = TenantHarness::create("test_simple_bottom_most_compaction_on_branch").await?;
+        let (tenant, ctx) = harness.load().await;
+
+        fn get_key(id: u32) -> Key {
+            let mut key = Key::from_hex("000000000033333333444444445500000000").unwrap();
+            key.field6 = id;
+            key
+        }
+
+        let img_layer = (0..10)
+            .map(|id| (get_key(id), Bytes::from(format!("value {id}@0x10"))))
+            .collect_vec();
+
+        let delta1 = vec![
+            (
+                get_key(1),
+                Lsn(0x20),
+                Value::WalRecord(NeonWalRecord::wal_append("@0x20")),
+            ),
+            (
+                get_key(2),
+                Lsn(0x30),
+                Value::WalRecord(NeonWalRecord::wal_append("@0x30")),
+            ),
+            (
+                get_key(3),
+                Lsn(0x28),
+                Value::WalRecord(NeonWalRecord::wal_append("@0x28")),
+            ),
+            (
+                get_key(3),
+                Lsn(0x30),
+                Value::WalRecord(NeonWalRecord::wal_append("@0x30")),
+            ),
+            (
+                get_key(3),
+                Lsn(0x40),
+                Value::WalRecord(NeonWalRecord::wal_append("@0x40")),
+            ),
+        ];
+        let delta2 = vec![
+            (
+                get_key(5),
+                Lsn(0x20),
+                Value::WalRecord(NeonWalRecord::wal_append("@0x20")),
+            ),
+            (
+                get_key(6),
+                Lsn(0x20),
+                Value::WalRecord(NeonWalRecord::wal_append("@0x20")),
+            ),
+        ];
+        let delta3 = vec![
+            (
+                get_key(8),
+                Lsn(0x48),
+                Value::WalRecord(NeonWalRecord::wal_append("@0x48")),
+            ),
+            (
+                get_key(9),
+                Lsn(0x48),
+                Value::WalRecord(NeonWalRecord::wal_append("@0x48")),
+            ),
+        ];
+
+        let parent_tline = tenant
+            .create_test_timeline_with_layers(
+                TIMELINE_ID,
+                Lsn(0x10),
+                DEFAULT_PG_VERSION,
+                &ctx,
+                vec![],                       // delta layers
+                vec![(Lsn(0x18), img_layer)], // image layers
+                Lsn(0x18),
+            )
+            .await?;
+
+        parent_tline.add_extra_test_dense_keyspace(KeySpace::single(get_key(0)..get_key(10)));
+
+        let branch_tline = tenant
+            .branch_timeline_test_with_layers(
+                &parent_tline,
+                NEW_TIMELINE_ID,
+                Some(Lsn(0x18)),
+                &ctx,
+                vec![
+                    DeltaLayerTestDesc::new_with_inferred_key_range(Lsn(0x20)..Lsn(0x48), delta1),
+                    DeltaLayerTestDesc::new_with_inferred_key_range(Lsn(0x20)..Lsn(0x48), delta2),
+                    DeltaLayerTestDesc::new_with_inferred_key_range(Lsn(0x48)..Lsn(0x50), delta3),
+                ], // delta layers
+                vec![], // image layers
+                Lsn(0x50),
+            )
+            .await?;
+
+        branch_tline.add_extra_test_dense_keyspace(KeySpace::single(get_key(0)..get_key(10)));
+
+        {
+            // Update GC info
+            let mut guard = parent_tline.gc_info.write().unwrap();
+            *guard = GcInfo {
+                retain_lsns: vec![(Lsn(0x18), branch_tline.timeline_id)],
+                cutoffs: GcCutoffs {
+                    time: Lsn(0x10),
+                    space: Lsn(0x10),
+                },
+                leases: Default::default(),
+                within_ancestor_pitr: false,
+            };
+        }
+
+        {
+            // Update GC info
+            let mut guard = branch_tline.gc_info.write().unwrap();
+            *guard = GcInfo {
+                retain_lsns: vec![(Lsn(0x40), branch_tline.timeline_id)],
+                cutoffs: GcCutoffs {
+                    time: Lsn(0x50),
+                    space: Lsn(0x50),
+                },
+                leases: Default::default(),
+                within_ancestor_pitr: false,
+            };
+        }
+
+        let expected_result_at_gc_horizon = [
+            Bytes::from_static(b"value 0@0x10"),
+            Bytes::from_static(b"value 1@0x10@0x20"),
+            Bytes::from_static(b"value 2@0x10@0x30"),
+            Bytes::from_static(b"value 3@0x10@0x28@0x30@0x40"),
+            Bytes::from_static(b"value 4@0x10"),
+            Bytes::from_static(b"value 5@0x10@0x20"),
+            Bytes::from_static(b"value 6@0x10@0x20"),
+            Bytes::from_static(b"value 7@0x10"),
+            Bytes::from_static(b"value 8@0x10@0x48"),
+            Bytes::from_static(b"value 9@0x10@0x48"),
+        ];
+
+        let expected_result_at_lsn_40 = [
+            Bytes::from_static(b"value 0@0x10"),
+            Bytes::from_static(b"value 1@0x10@0x20"),
+            Bytes::from_static(b"value 2@0x10@0x30"),
+            Bytes::from_static(b"value 3@0x10@0x28@0x30@0x40"),
+            Bytes::from_static(b"value 4@0x10"),
+            Bytes::from_static(b"value 5@0x10@0x20"),
+            Bytes::from_static(b"value 6@0x10@0x20"),
+            Bytes::from_static(b"value 7@0x10"),
+            Bytes::from_static(b"value 8@0x10"),
+            Bytes::from_static(b"value 9@0x10"),
+        ];
+
+        let verify_result = || async {
+            for idx in 0..10 {
+                assert_eq!(
+                    branch_tline
+                        .get(get_key(idx as u32), Lsn(0x50), &ctx)
+                        .await
+                        .unwrap(),
+                    &expected_result_at_gc_horizon[idx]
+                );
+                assert_eq!(
+                    branch_tline
+                        .get(get_key(idx as u32), Lsn(0x40), &ctx)
+                        .await
+                        .unwrap(),
+                    &expected_result_at_lsn_40[idx]
+                );
+            }
+        };
+
+        verify_result().await;
+
+        let cancel = CancellationToken::new();
+        branch_tline
+            .compact_with_gc(&cancel, EnumSet::new(), &ctx)
+            .await
+            .unwrap();
 
         verify_result().await;
 

pageserver/src/tenant/ephemeral_file.rs

@@ -29,6 +29,7 @@ impl EphemeralFile {
         conf: &PageServerConf,
         tenant_shard_id: TenantShardId,
         timeline_id: TimelineId,
+        gate_guard: utils::sync::gate::GateGuard,
         ctx: &RequestContext,
     ) -> Result<EphemeralFile, io::Error> {
         static NEXT_FILENAME: AtomicU64 = AtomicU64::new(1);
@@ -51,10 +52,12 @@ impl EphemeralFile {
         )
         .await?;
 
+        let prewarm = conf.l0_flush.prewarm_on_write();
+
         Ok(EphemeralFile {
             _tenant_shard_id: tenant_shard_id,
             _timeline_id: timeline_id,
-            rw: page_caching::RW::new(file, conf.l0_flush.prewarm_on_write()),
+            rw: page_caching::RW::new(file, prewarm, gate_guard),
         })
     }
 
@@ -161,7 +164,11 @@ mod tests {
     async fn test_ephemeral_blobs() -> Result<(), io::Error> {
         let (conf, tenant_id, timeline_id, ctx) = harness("ephemeral_blobs")?;
 
-        let mut file = EphemeralFile::create(conf, tenant_id, timeline_id, &ctx).await?;
+        let gate = utils::sync::gate::Gate::default();
+
+        let entered = gate.enter().unwrap();
+
+        let mut file = EphemeralFile::create(conf, tenant_id, timeline_id, entered, &ctx).await?;
 
         let pos_foo = file.write_blob(b"foo", &ctx).await?;
         assert_eq!(
@@ -215,4 +222,38 @@ mod tests {
 
         Ok(())
     }
+
+    #[tokio::test]
+    async fn ephemeral_file_holds_gate_open() {
+        const FOREVER: std::time::Duration = std::time::Duration::from_secs(5);
+
+        let (conf, tenant_id, timeline_id, ctx) =
+            harness("ephemeral_file_holds_gate_open").unwrap();
+
+        let gate = utils::sync::gate::Gate::default();
+
+        let file = EphemeralFile::create(conf, tenant_id, timeline_id, gate.enter().unwrap(), &ctx)
+            .await
+            .unwrap();
+
+        let mut closing = tokio::task::spawn(async move {
+            gate.close().await;
+        });
+
+        // gate is entered until the ephemeral file is dropped
+        // do not start paused tokio-epoll-uring has a sleep loop
+        tokio::time::pause();
+        tokio::time::timeout(FOREVER, &mut closing)
+            .await
+            .expect_err("closing cannot complete before dropping");
+
+        // this is a requirement of the reset_tenant functionality: we have to be able to restart a
+        // tenant fast, and for that, we need all tenant_dir operations be guarded by entering a gate
+        drop(file);
+
+        tokio::time::timeout(FOREVER, &mut closing)
+            .await
+            .expect("closing completes right away")
+            .expect("closing does not panic");
+    }
 }

pageserver/src/tenant/ephemeral_file/page_caching.rs

@@ -18,6 +18,8 @@ use super::zero_padded_read_write;
 pub struct RW {
     page_cache_file_id: page_cache::FileId,
     rw: super::zero_padded_read_write::RW<PreWarmingWriter>,
+    /// Gate guard is held on as long as we need to do operations in the path (delete on drop).
+    _gate_guard: utils::sync::gate::GateGuard,
 }
 
 /// When we flush a block to the underlying [`crate::virtual_file::VirtualFile`],
@@ -29,7 +31,11 @@ pub enum PrewarmOnWrite {
 }
 
 impl RW {
-    pub fn new(file: VirtualFile, prewarm_on_write: PrewarmOnWrite) -> Self {
+    pub fn new(
+        file: VirtualFile,
+        prewarm_on_write: PrewarmOnWrite,
+        _gate_guard: utils::sync::gate::GateGuard,
+    ) -> Self {
         let page_cache_file_id = page_cache::next_file_id();
         Self {
             page_cache_file_id,
@@ -38,6 +44,7 @@ impl RW {
                 file,
                 prewarm_on_write,
             )),
+            _gate_guard,
         }
     }
 
@@ -145,6 +152,7 @@ impl Drop for RW {
         // We leave them there, [`crate::page_cache::PageCache::find_victim`] will evict them when needed.
 
         // unlink the file
+        // we are clear to do this, because we have entered a gate
         let res = std::fs::remove_file(&self.rw.as_writer().file.path);
         if let Err(e) = res {
             if e.kind() != std::io::ErrorKind::NotFound {

pageserver/src/tenant/gc_block.rs

@@ -0,0 +1,213 @@
+use std::collections::HashMap;
+
+use utils::id::TimelineId;
+
+use super::remote_timeline_client::index::GcBlockingReason;
+
+type Storage = HashMap<TimelineId, enumset::EnumSet<GcBlockingReason>>;
+
+#[derive(Default)]
+pub(crate) struct GcBlock {
+    /// The timelines which have current reasons to block gc.
+    ///
+    /// LOCK ORDER: this is held locked while scheduling the next index_part update. This is done
+    /// to keep the this field up to date with RemoteTimelineClient `upload_queue.dirty`.
+    reasons: std::sync::Mutex<Storage>,
+    blocking: tokio::sync::Mutex<()>,
+}
+
+impl GcBlock {
+    /// Start another gc iteration.
+    ///
+    /// Returns a guard to be held for the duration of gc iteration to allow synchronizing with
+    /// it's ending, or if not currently possible, a value describing the reasons why not.
+    ///
+    /// Cancellation safe.
+    pub(super) async fn start(&self) -> Result<Guard<'_>, BlockingReasons> {
+        let reasons = {
+            let g = self.reasons.lock().unwrap();
+
+            // TODO: the assumption is that this method gets called periodically. in prod, we use 1h, in
+            // tests, we use everything. we should warn if the gc has been consecutively blocked
+            // for more than 1h (within single tenant session?).
+            BlockingReasons::clean_and_summarize(g)
+        };
+
+        if let Some(reasons) = reasons {
+            Err(reasons)
+        } else {
+            Ok(Guard {
+                _inner: self.blocking.lock().await,
+            })
+        }
+    }
+
+    pub(crate) fn summary(&self) -> Option<BlockingReasons> {
+        let g = self.reasons.lock().unwrap();
+
+        BlockingReasons::summarize(&g)
+    }
+
+    /// Start blocking gc for this one timeline for the given reason.
+    ///
+    /// This is not a guard based API but instead it mimics set API. The returned future will not
+    /// resolve until an existing gc round has completed.
+    ///
+    /// Returns true if this block was new, false if gc was already blocked for this reason.
+    ///
+    /// Cancellation safe: cancelling after first poll will keep the reason to block gc, but will
+    /// keep the gc blocking reason.
+    pub(crate) async fn insert(
+        &self,
+        timeline: &super::Timeline,
+        reason: GcBlockingReason,
+    ) -> anyhow::Result<bool> {
+        let (added, uploaded) = {
+            let mut g = self.reasons.lock().unwrap();
+            let set = g.entry(timeline.timeline_id).or_default();
+            let added = set.insert(reason);
+
+            // LOCK ORDER: intentionally hold the lock, see self.reasons.
+            let uploaded = timeline
+                .remote_client
+                .schedule_insert_gc_block_reason(reason)?;
+
+            (added, uploaded)
+        };
+
+        uploaded.await?;
+
+        // ensure that any ongoing gc iteration has completed
+        drop(self.blocking.lock().await);
+
+        Ok(added)
+    }
+
+    /// Remove blocking gc for this one timeline and the given reason.
+    pub(crate) async fn remove(
+        &self,
+        timeline: &super::Timeline,
+        reason: GcBlockingReason,
+    ) -> anyhow::Result<()> {
+        use std::collections::hash_map::Entry;
+
+        super::span::debug_assert_current_span_has_tenant_and_timeline_id();
+
+        let (remaining_blocks, uploaded) = {
+            let mut g = self.reasons.lock().unwrap();
+            match g.entry(timeline.timeline_id) {
+                Entry::Occupied(mut oe) => {
+                    let set = oe.get_mut();
+                    set.remove(reason);
+                    if set.is_empty() {
+                        oe.remove();
+                    }
+                }
+                Entry::Vacant(_) => {
+                    // we must still do the index_part.json update regardless, in case we had earlier
+                    // been cancelled
+                }
+            }
+
+            let remaining_blocks = g.len();
+
+            // LOCK ORDER: intentionally hold the lock while scheduling; see self.reasons
+            let uploaded = timeline
+                .remote_client
+                .schedule_remove_gc_block_reason(reason)?;
+
+            (remaining_blocks, uploaded)
+        };
+        uploaded.await?;
+
+        // no need to synchronize with gc iteration again
+
+        if remaining_blocks > 0 {
+            tracing::info!(remaining_blocks, removed=?reason, "gc blocking removed, but gc remains blocked");
+        } else {
+            tracing::info!("gc is now unblocked for the tenant");
+        }
+
+        Ok(())
+    }
+
+    pub(crate) fn before_delete(&self, timeline: &super::Timeline) {
+        let unblocked = {
+            let mut g = self.reasons.lock().unwrap();
+            if g.is_empty() {
+                return;
+            }
+
+            g.remove(&timeline.timeline_id);
+
+            BlockingReasons::clean_and_summarize(g).is_none()
+        };
+
+        if unblocked {
+            tracing::info!("gc is now unblocked following deletion");
+        }
+    }
+
+    /// Initialize with the non-deleted timelines of this tenant.
+    pub(crate) fn set_scanned(&self, scanned: Storage) {
+        let mut g = self.reasons.lock().unwrap();
+        assert!(g.is_empty());
+        g.extend(scanned.into_iter().filter(|(_, v)| !v.is_empty()));
+
+        if let Some(reasons) = BlockingReasons::clean_and_summarize(g) {
+            tracing::info!(summary=?reasons, "initialized with gc blocked");
+        }
+    }
+}
+
+pub(super) struct Guard<'a> {
+    _inner: tokio::sync::MutexGuard<'a, ()>,
+}
+
+#[derive(Debug)]
+pub(crate) struct BlockingReasons {
+    timelines: usize,
+    reasons: enumset::EnumSet<GcBlockingReason>,
+}
+
+impl std::fmt::Display for BlockingReasons {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(
+            f,
+            "{} timelines block for {:?}",
+            self.timelines, self.reasons
+        )
+    }
+}
+
+impl BlockingReasons {
+    fn clean_and_summarize(mut g: std::sync::MutexGuard<'_, Storage>) -> Option<Self> {
+        let mut reasons = enumset::EnumSet::empty();
+        g.retain(|_key, value| {
+            reasons = reasons.union(*value);
+            !value.is_empty()
+        });
+        if !g.is_empty() {
+            Some(BlockingReasons {
+                timelines: g.len(),
+                reasons,
+            })
+        } else {
+            None
+        }
+    }
+
+    fn summarize(g: &std::sync::MutexGuard<'_, Storage>) -> Option<Self> {
+        if g.is_empty() {
+            None
+        } else {
+            let reasons = g
+                .values()
+                .fold(enumset::EnumSet::empty(), |acc, next| acc.union(*next));
+            Some(BlockingReasons {
+                timelines: g.len(),
+                reasons,
+            })
+        }
+    }
+}

pageserver/src/tenant/layer_map.rs

@@ -51,7 +51,8 @@ use crate::keyspace::KeyPartitioning;
 use crate::repository::Key;
 use crate::tenant::storage_layer::InMemoryLayer;
 use anyhow::Result;
-use pageserver_api::keyspace::KeySpaceAccum;
+use pageserver_api::keyspace::{KeySpace, KeySpaceAccum};
+use range_set_blaze::{CheckSortedDisjoint, RangeSetBlaze};
 use std::collections::{HashMap, VecDeque};
 use std::iter::Peekable;
 use std::ops::Range;
@@ -61,7 +62,7 @@ use utils::lsn::Lsn;
 use historic_layer_coverage::BufferedHistoricLayerCoverage;
 pub use historic_layer_coverage::LayerKey;
 
-use super::storage_layer::PersistentLayerDesc;
+use super::storage_layer::{LayerVisibilityHint, PersistentLayerDesc};
 
 ///
 /// LayerMap tracks what layers exist on a timeline.
@@ -845,8 +846,8 @@ impl LayerMap {
     }
 
     /// Return all L0 delta layers
-    pub fn get_level0_deltas(&self) -> Vec<Arc<PersistentLayerDesc>> {
-        self.l0_delta_layers.to_vec()
+    pub fn level0_deltas(&self) -> &Vec<Arc<PersistentLayerDesc>> {
+        &self.l0_delta_layers
     }
 
     /// debugging function to print out the contents of the layer map
@@ -871,11 +872,183 @@ impl LayerMap {
         println!("End dump LayerMap");
         Ok(())
     }
+
+    /// `read_points` represent the tip of a timeline and any branch points, i.e. the places
+    /// where we expect to serve reads.
+    ///
+    /// This function is O(N) and should be called infrequently.  The caller is responsible for
+    /// looking up and updating the Layer objects for these layer descriptors.
+    pub fn get_visibility(
+        &self,
+        mut read_points: Vec<Lsn>,
+    ) -> (
+        Vec<(Arc<PersistentLayerDesc>, LayerVisibilityHint)>,
+        KeySpace,
+    ) {
+        // This is like a KeySpace, but this type is intended for efficient unions with image layer ranges, whereas
+        // KeySpace is intended to be composed statically and iterated over.
+        struct KeyShadow {
+            // Map of range start to range end
+            inner: RangeSetBlaze<i128>,
+        }
+
+        impl KeyShadow {
+            fn new() -> Self {
+                Self {
+                    inner: Default::default(),
+                }
+            }
+
+            fn contains(&self, range: Range<Key>) -> bool {
+                let range_incl = range.start.to_i128()..=range.end.to_i128() - 1;
+                self.inner.is_superset(&RangeSetBlaze::from_sorted_disjoint(
+                    CheckSortedDisjoint::from([range_incl]),
+                ))
+            }
+
+            /// Add the input range to the keys covered by self.
+            ///
+            /// Return true if inserting this range covered some keys that were previously not covered
+            fn cover(&mut self, insert: Range<Key>) -> bool {
+                let range_incl = insert.start.to_i128()..=insert.end.to_i128() - 1;
+                self.inner.ranges_insert(range_incl)
+            }
+
+            fn reset(&mut self) {
+                self.inner = Default::default();
+            }
+
+            fn to_keyspace(&self) -> KeySpace {
+                let mut accum = KeySpaceAccum::new();
+                for range_incl in self.inner.ranges() {
+                    let range = Range {
+                        start: Key::from_i128(*range_incl.start()),
+                        end: Key::from_i128(range_incl.end() + 1),
+                    };
+                    accum.add_range(range)
+                }
+
+                accum.to_keyspace()
+            }
+        }
+
+        // The 'shadow' will be updated as we sweep through the layers: an image layer subtracts from the shadow,
+        // and a ReadPoint
+        read_points.sort_by_key(|rp| rp.0);
+        let mut shadow = KeyShadow::new();
+
+        // We will interleave all our read points and layers into a sorted collection
+        enum Item {
+            ReadPoint { lsn: Lsn },
+            Layer(Arc<PersistentLayerDesc>),
+        }
+
+        let mut items = Vec::with_capacity(self.historic.len() + read_points.len());
+        items.extend(self.iter_historic_layers().map(Item::Layer));
+        items.extend(
+            read_points
+                .into_iter()
+                .map(|rp| Item::ReadPoint { lsn: rp }),
+        );
+
+        // Ordering: we want to iterate like this:
+        // 1. Highest LSNs first
+        // 2. Consider images before deltas if they end at the same LSNs (images cover deltas)
+        // 3. Consider ReadPoints before image layers if they're at the same LSN (readpoints make that image visible)
+        items.sort_by_key(|item| {
+            std::cmp::Reverse(match item {
+                Item::Layer(layer) => {
+                    if layer.is_delta() {
+                        (Lsn(layer.get_lsn_range().end.0 - 1), 0)
+                    } else {
+                        (layer.image_layer_lsn(), 1)
+                    }
+                }
+                Item::ReadPoint { lsn } => (*lsn, 2),
+            })
+        });
+
+        let mut results = Vec::with_capacity(self.historic.len());
+
+        let mut maybe_covered_deltas: Vec<Arc<PersistentLayerDesc>> = Vec::new();
+
+        for item in items {
+            let (reached_lsn, is_readpoint) = match &item {
+                Item::ReadPoint { lsn } => (lsn, true),
+                Item::Layer(layer) => (&layer.lsn_range.start, false),
+            };
+            maybe_covered_deltas.retain(|d| {
+                if *reached_lsn >= d.lsn_range.start && is_readpoint {
+                    // We encountered a readpoint within the delta layer: it is visible
+
+                    results.push((d.clone(), LayerVisibilityHint::Visible));
+                    false
+                } else if *reached_lsn < d.lsn_range.start {
+                    // We passed the layer's range without encountering a read point: it is not visible
+                    results.push((d.clone(), LayerVisibilityHint::Covered));
+                    false
+                } else {
+                    // We're still in the delta layer: continue iterating
+                    true
+                }
+            });
+
+            match item {
+                Item::ReadPoint { lsn: _lsn } => {
+                    // TODO: propagate the child timeline's shadow from their own run of this function, so that we don't have
+                    // to assume that the whole key range is visible at the branch point.
+                    shadow.reset();
+                }
+                Item::Layer(layer) => {
+                    let visibility = if layer.is_delta() {
+                        if shadow.contains(layer.get_key_range()) {
+                            // If a layer isn't visible based on current state, we must defer deciding whether
+                            // it is truly not visible until we have advanced past the delta's range: we might
+                            // encounter another branch point within this delta layer's LSN range.
+                            maybe_covered_deltas.push(layer);
+                            continue;
+                        } else {
+                            LayerVisibilityHint::Visible
+                        }
+                    } else {
+                        let modified = shadow.cover(layer.get_key_range());
+                        if modified {
+                            // An image layer in a region which wasn't fully covered yet: this layer is visible, but layers below it will be covered
+                            LayerVisibilityHint::Visible
+                        } else {
+                            // An image layer in a region that was already covered
+                            LayerVisibilityHint::Covered
+                        }
+                    };
+
+                    results.push((layer, visibility));
+                }
+            }
+        }
+
+        // Drain any remaining maybe_covered deltas
+        results.extend(
+            maybe_covered_deltas
+                .into_iter()
+                .map(|d| (d, LayerVisibilityHint::Covered)),
+        );
+
+        (results, shadow.to_keyspace())
+    }
 }
 
 #[cfg(test)]
 mod tests {
-    use pageserver_api::keyspace::KeySpace;
+    use crate::tenant::{storage_layer::LayerName, IndexPart};
+    use pageserver_api::{
+        key::DBDIR_KEY,
+        keyspace::{KeySpace, KeySpaceRandomAccum},
+    };
+    use std::{collections::HashMap, path::PathBuf};
+    use utils::{
+        id::{TenantId, TimelineId},
+        shard::TenantShardId,
+    };
 
     use super::*;
 
@@ -1002,4 +1175,299 @@ mod tests {
             }
         }
     }
+
+    #[test]
+    fn layer_visibility_basic() {
+        // A simple synthetic input, as a smoke test.
+        let tenant_shard_id = TenantShardId::unsharded(TenantId::generate());
+        let timeline_id = TimelineId::generate();
+        let mut layer_map = LayerMap::default();
+        let mut updates = layer_map.batch_update();
+
+        const FAKE_LAYER_SIZE: u64 = 1024;
+
+        let inject_delta = |updates: &mut BatchedUpdates,
+                            key_start: i128,
+                            key_end: i128,
+                            lsn_start: u64,
+                            lsn_end: u64| {
+            let desc = PersistentLayerDesc::new_delta(
+                tenant_shard_id,
+                timeline_id,
+                Range {
+                    start: Key::from_i128(key_start),
+                    end: Key::from_i128(key_end),
+                },
+                Range {
+                    start: Lsn(lsn_start),
+                    end: Lsn(lsn_end),
+                },
+                1024,
+            );
+            updates.insert_historic(desc.clone());
+            desc
+        };
+
+        let inject_image =
+            |updates: &mut BatchedUpdates, key_start: i128, key_end: i128, lsn: u64| {
+                let desc = PersistentLayerDesc::new_img(
+                    tenant_shard_id,
+                    timeline_id,
+                    Range {
+                        start: Key::from_i128(key_start),
+                        end: Key::from_i128(key_end),
+                    },
+                    Lsn(lsn),
+                    FAKE_LAYER_SIZE,
+                );
+                updates.insert_historic(desc.clone());
+                desc
+            };
+
+        //
+        // Construct our scenario: the following lines go in backward-LSN order, constructing the various scenarios
+        // we expect to handle.  You can follow these examples through in the same order as they would be processed
+        // by the function under test.
+        //
+
+        let mut read_points = vec![Lsn(1000)];
+
+        // A delta ahead of any image layer
+        let ahead_layer = inject_delta(&mut updates, 10, 20, 101, 110);
+
+        // An image layer is visible and covers some layers beneath itself
+        let visible_covering_img = inject_image(&mut updates, 5, 25, 99);
+
+        // A delta layer covered by the image layer: should be covered
+        let covered_delta = inject_delta(&mut updates, 10, 20, 90, 100);
+
+        // A delta layer partially covered by an image layer: should be visible
+        let partially_covered_delta = inject_delta(&mut updates, 1, 7, 90, 100);
+
+        // A delta layer not covered by an image layer: should be visible
+        let not_covered_delta = inject_delta(&mut updates, 1, 4, 90, 100);
+
+        // An image layer covered by the image layer above: should be covered
+        let covered_image = inject_image(&mut updates, 10, 20, 89);
+
+        // An image layer partially covered by an image layer: should be visible
+        let partially_covered_image = inject_image(&mut updates, 1, 7, 89);
+
+        // An image layer not covered by an image layer: should be visible
+        let not_covered_image = inject_image(&mut updates, 1, 4, 89);
+
+        // A read point: this will make subsequent layers below here visible, even if there are
+        // more recent layers covering them.
+        read_points.push(Lsn(80));
+
+        // A delta layer covered by an earlier image layer, but visible to a readpoint below that covering layer
+        let covered_delta_below_read_point = inject_delta(&mut updates, 10, 20, 70, 79);
+
+        // A delta layer whose end LSN is covered, but where a read point is present partway through its LSN range:
+        // the read point should make it visible, even though its end LSN is covered
+        let covering_img_between_read_points = inject_image(&mut updates, 10, 20, 69);
+        let covered_delta_between_read_points = inject_delta(&mut updates, 10, 15, 67, 69);
+        read_points.push(Lsn(65));
+        let covered_delta_intersects_read_point = inject_delta(&mut updates, 15, 20, 60, 69);
+
+        let visible_img_after_last_read_point = inject_image(&mut updates, 10, 20, 65);
+
+        updates.flush();
+
+        let (layer_visibilities, shadow) = layer_map.get_visibility(read_points);
+        let layer_visibilities = layer_visibilities.into_iter().collect::<HashMap<_, _>>();
+
+        assert_eq!(
+            layer_visibilities.get(&ahead_layer),
+            Some(&LayerVisibilityHint::Visible)
+        );
+        assert_eq!(
+            layer_visibilities.get(&visible_covering_img),
+            Some(&LayerVisibilityHint::Visible)
+        );
+        assert_eq!(
+            layer_visibilities.get(&covered_delta),
+            Some(&LayerVisibilityHint::Covered)
+        );
+        assert_eq!(
+            layer_visibilities.get(&partially_covered_delta),
+            Some(&LayerVisibilityHint::Visible)
+        );
+        assert_eq!(
+            layer_visibilities.get(&not_covered_delta),
+            Some(&LayerVisibilityHint::Visible)
+        );
+        assert_eq!(
+            layer_visibilities.get(&covered_image),
+            Some(&LayerVisibilityHint::Covered)
+        );
+        assert_eq!(
+            layer_visibilities.get(&partially_covered_image),
+            Some(&LayerVisibilityHint::Visible)
+        );
+        assert_eq!(
+            layer_visibilities.get(&not_covered_image),
+            Some(&LayerVisibilityHint::Visible)
+        );
+        assert_eq!(
+            layer_visibilities.get(&covered_delta_below_read_point),
+            Some(&LayerVisibilityHint::Visible)
+        );
+        assert_eq!(
+            layer_visibilities.get(&covering_img_between_read_points),
+            Some(&LayerVisibilityHint::Visible)
+        );
+        assert_eq!(
+            layer_visibilities.get(&covered_delta_between_read_points),
+            Some(&LayerVisibilityHint::Covered)
+        );
+        assert_eq!(
+            layer_visibilities.get(&covered_delta_intersects_read_point),
+            Some(&LayerVisibilityHint::Visible)
+        );
+        assert_eq!(
+            layer_visibilities.get(&visible_img_after_last_read_point),
+            Some(&LayerVisibilityHint::Visible)
+        );
+
+        // Shadow should include all the images below the last read point
+        let expected_shadow = KeySpace {
+            ranges: vec![Key::from_i128(10)..Key::from_i128(20)],
+        };
+        assert_eq!(shadow, expected_shadow);
+    }
+
+    fn fixture_path(relative: &str) -> PathBuf {
+        PathBuf::from(env!("CARGO_MANIFEST_DIR")).join(relative)
+    }
+
+    #[test]
+    fn layer_visibility_realistic() {
+        // Load a large example layermap
+        let index_raw = std::fs::read_to_string(fixture_path(
+            "test_data/indices/mixed_workload/index_part.json",
+        ))
+        .unwrap();
+        let index: IndexPart = serde_json::from_str::<IndexPart>(&index_raw).unwrap();
+
+        let tenant_id = TenantId::generate();
+        let tenant_shard_id = TenantShardId::unsharded(tenant_id);
+        let timeline_id = TimelineId::generate();
+
+        let mut layer_map = LayerMap::default();
+        let mut updates = layer_map.batch_update();
+        for (layer_name, layer_metadata) in index.layer_metadata {
+            let layer_desc = match layer_name {
+                LayerName::Image(layer_name) => PersistentLayerDesc {
+                    key_range: layer_name.key_range.clone(),
+                    lsn_range: layer_name.lsn_as_range(),
+                    tenant_shard_id,
+                    timeline_id,
+                    is_delta: false,
+                    file_size: layer_metadata.file_size,
+                },
+                LayerName::Delta(layer_name) => PersistentLayerDesc {
+                    key_range: layer_name.key_range,
+                    lsn_range: layer_name.lsn_range,
+                    tenant_shard_id,
+                    timeline_id,
+                    is_delta: true,
+                    file_size: layer_metadata.file_size,
+                },
+            };
+            updates.insert_historic(layer_desc);
+        }
+        updates.flush();
+
+        let read_points = vec![index.metadata.disk_consistent_lsn()];
+        let (layer_visibilities, shadow) = layer_map.get_visibility(read_points);
+        for (layer_desc, visibility) in &layer_visibilities {
+            tracing::info!("{layer_desc:?}: {visibility:?}");
+            eprintln!("{layer_desc:?}: {visibility:?}");
+        }
+
+        // The shadow should be non-empty, since there were some image layers
+        assert!(!shadow.ranges.is_empty());
+
+        // At least some layers should be marked covered
+        assert!(layer_visibilities
+            .iter()
+            .any(|i| matches!(i.1, LayerVisibilityHint::Covered)));
+
+        let layer_visibilities = layer_visibilities.into_iter().collect::<HashMap<_, _>>();
+
+        // Brute force validation: a layer should be marked covered if and only if there are image layers above it in LSN order which cover it
+        for (layer_desc, visible) in &layer_visibilities {
+            let mut coverage = KeySpaceRandomAccum::new();
+            let mut covered_by = Vec::new();
+
+            for other_layer in layer_map.iter_historic_layers() {
+                if &other_layer == layer_desc {
+                    continue;
+                }
+                if !other_layer.is_delta()
+                    && other_layer.image_layer_lsn() >= Lsn(layer_desc.get_lsn_range().end.0 - 1)
+                    && other_layer.key_range.start <= layer_desc.key_range.end
+                    && layer_desc.key_range.start <= other_layer.key_range.end
+                {
+                    coverage.add_range(other_layer.get_key_range());
+                    covered_by.push((*other_layer).clone());
+                }
+            }
+            let coverage = coverage.to_keyspace();
+
+            let expect_visible = if coverage.ranges.len() == 1
+                && coverage.contains(&layer_desc.key_range.start)
+                && coverage.contains(&Key::from_i128(layer_desc.key_range.end.to_i128() - 1))
+            {
+                LayerVisibilityHint::Covered
+            } else {
+                LayerVisibilityHint::Visible
+            };
+
+            if expect_visible != *visible {
+                eprintln!(
+                    "Layer {}..{} @ {}..{} (delta={}) is {visible:?}, should be {expect_visible:?}",
+                    layer_desc.key_range.start,
+                    layer_desc.key_range.end,
+                    layer_desc.lsn_range.start,
+                    layer_desc.lsn_range.end,
+                    layer_desc.is_delta()
+                );
+                if expect_visible == LayerVisibilityHint::Covered {
+                    eprintln!("Covered by:");
+                    for other in covered_by {
+                        eprintln!(
+                            "  {}..{} @ {}",
+                            other.get_key_range().start,
+                            other.get_key_range().end,
+                            other.image_layer_lsn()
+                        );
+                    }
+                    if let Some(range) = coverage.ranges.first() {
+                        eprintln!(
+                            "Total coverage from contributing layers: {}..{}",
+                            range.start, range.end
+                        );
+                    } else {
+                        eprintln!(
+                            "Total coverage from contributing layers: {:?}",
+                            coverage.ranges
+                        );
+                    }
+                }
+            }
+            assert_eq!(expect_visible, *visible);
+        }
+
+        // Sanity: the layer that holds latest data for the DBDIR key should always be visible
+        // (just using this key as a key that will always exist for any layermap fixture)
+        let dbdir_layer = layer_map
+            .search(DBDIR_KEY, index.metadata.disk_consistent_lsn())
+            .unwrap();
+        assert!(matches!(
+            layer_visibilities.get(&dbdir_layer.layer).unwrap(),
+            LayerVisibilityHint::Visible
+        ));
+    }
 }

pageserver/src/tenant/layer_map/historic_layer_coverage.rs

@@ -521,6 +521,10 @@ impl<Value: Clone> BufferedHistoricLayerCoverage<Value> {
 
         Ok(&self.historic_coverage)
     }
+
+    pub(crate) fn len(&self) -> usize {
+        self.layers.len()
+    }
 }
 
 #[test]

pageserver/src/tenant/metadata.rs

@@ -285,12 +285,15 @@ impl TimelineMetadata {
     }
 
     /// When reparenting, the `ancestor_lsn` does not change.
+    ///
+    /// Returns true if anything was changed.
     pub fn reparent(&mut self, timeline: &TimelineId) {
         assert!(self.body.ancestor_timeline.is_some());
         // no assertion for redoing this: it's fine, we may have to repeat this multiple times over
         self.body.ancestor_timeline = Some(*timeline);
     }
 
+    /// Returns true if anything was changed
     pub fn detach_from_ancestor(&mut self, branchpoint: &(TimelineId, Lsn)) {
         if let Some(ancestor) = self.body.ancestor_timeline {
             assert_eq!(ancestor, branchpoint.0);

pageserver/src/tenant/mgr.rs

@@ -13,7 +13,7 @@ use pageserver_api::upcall_api::ReAttachResponseTenant;
 use rand::{distributions::Alphanumeric, Rng};
 use std::borrow::Cow;
 use std::cmp::Ordering;
-use std::collections::{BTreeMap, HashMap};
+use std::collections::{BTreeMap, HashMap, HashSet};
 use std::ops::Deref;
 use std::sync::Arc;
 use std::time::Duration;
@@ -54,8 +54,8 @@ use utils::id::{TenantId, TimelineId};
 
 use super::remote_timeline_client::remote_tenant_path;
 use super::secondary::SecondaryTenant;
-use super::timeline::detach_ancestor::PreparedTimelineDetach;
-use super::TenantSharedResources;
+use super::timeline::detach_ancestor::{self, PreparedTimelineDetach};
+use super::{GlobalShutDown, TenantSharedResources};
 
 /// For a tenant that appears in TenantsMap, it may either be
 /// - `Attached`: has a full Tenant object, is elegible to service
@@ -116,8 +116,6 @@ pub(crate) enum ShardSelector {
     /// Only return the 0th shard, if it is present.  If a non-0th shard is present,
     /// ignore it.
     Zero,
-    /// Pick the first shard we find for the TenantId
-    First,
     /// Pick the shard that holds this key
     Page(Key),
     /// The shard ID is known: pick the given shard
@@ -226,21 +224,8 @@ async fn safe_rename_tenant_dir(path: impl AsRef<Utf8Path>) -> std::io::Result<U
 }
 
 /// See [`Self::spawn`].
-#[derive(Clone)]
-pub struct BackgroundPurges(Arc<std::sync::Mutex<BackgroundPurgesInner>>);
-enum BackgroundPurgesInner {
-    Open(tokio::task::JoinSet<()>),
-    // we use the async mutex for coalescing
-    ShuttingDown(Arc<tokio::sync::Mutex<tokio::task::JoinSet<()>>>),
-}
-
-impl Default for BackgroundPurges {
-    fn default() -> Self {
-        Self(Arc::new(std::sync::Mutex::new(
-            BackgroundPurgesInner::Open(JoinSet::new()),
-        )))
-    }
-}
+#[derive(Clone, Default)]
+pub struct BackgroundPurges(tokio_util::task::TaskTracker);
 
 impl BackgroundPurges {
     /// When we have moved a tenant's content to a temporary directory, we may delete it lazily in
@@ -249,24 +234,32 @@ impl BackgroundPurges {
     /// Although we are cleaning up the tenant, this task is not meant to be bound by the lifetime of the tenant in memory.
     /// Thus the [`BackgroundPurges`] type to keep track of these tasks.
     pub fn spawn(&self, tmp_path: Utf8PathBuf) {
-        let mut guard = self.0.lock().unwrap();
-        let jset = match &mut *guard {
-            BackgroundPurgesInner::Open(ref mut jset) => jset,
-            BackgroundPurgesInner::ShuttingDown(_) => {
-                warn!("trying to spawn background purge during shutdown, ignoring");
-                return;
+        // because on shutdown we close and wait, we are misusing TaskTracker a bit.
+        //
+        // so first acquire a token, then check if the tracker has been closed. the tracker might get closed
+        // right after, but at least the shutdown will wait for what we are spawning next.
+        let token = self.0.token();
+
+        if self.0.is_closed() {
+            warn!(
+                %tmp_path,
+                "trying to spawn background purge during shutdown, ignoring"
+            );
+            return;
+        }
+
+        let span = info_span!(parent: None, "background_purge", %tmp_path);
+
+        let task = move || {
+            let _token = token;
+            let _entered = span.entered();
+            if let Err(error) = std::fs::remove_dir_all(tmp_path.as_path()) {
+                // should we fatal_io_error here?
+                warn!(%error, "failed to purge tenant directory");
             }
         };
-        jset.spawn_on(
-            async move {
-                if let Err(error) = fs::remove_dir_all(tmp_path.as_path()).await {
-                    // should we fatal_io_error here?
-                    warn!(%error, path=%tmp_path, "failed to purge tenant directory");
-                }
-            }
-            .instrument(info_span!(parent: None, "background_purge")),
-            BACKGROUND_RUNTIME.handle(),
-        );
+
+        BACKGROUND_RUNTIME.spawn_blocking(task);
     }
 
     /// When this future completes, all background purges have completed.
@@ -280,42 +273,9 @@ impl BackgroundPurges {
     /// instances of this future will continue to be correct.
     #[instrument(skip_all)]
     pub async fn shutdown(&self) {
-        let jset = {
-            let mut guard = self.0.lock().unwrap();
-            match &mut *guard {
-                BackgroundPurgesInner::Open(jset) => {
-                    *guard = BackgroundPurgesInner::ShuttingDown(Arc::new(tokio::sync::Mutex::new(
-                        std::mem::take(jset),
-                    )))
-                }
-                BackgroundPurgesInner::ShuttingDown(_) => {
-                    // calling shutdown multiple times is most likely a bug in pageserver shutdown code
-                    warn!("already shutting down");
-                }
-            };
-            match &mut *guard {
-                BackgroundPurgesInner::ShuttingDown(ref mut jset) => jset.clone(),
-                BackgroundPurgesInner::Open(_) => {
-                    unreachable!("above code transitions into shut down state");
-                }
-            }
-        };
-        let mut jset = jset.lock().await; // concurrent callers coalesce here
-        while let Some(res) = jset.join_next().await {
-            match res {
-                Ok(()) => {}
-                Err(e) if e.is_panic() => {
-                    // If it panicked, the error is already logged by the panic hook.
-                }
-                Err(e) if e.is_cancelled() => {
-                    unreachable!("we don't cancel the joinset or runtime")
-                }
-                Err(e) => {
-                    // No idea when this can happen, but let's log it.
-                    warn!(%e, "background purge task failed or panicked");
-                }
-            }
-        }
+        // forbid new tasks (can be called many times)
+        self.0.close();
+        self.0.wait().await;
     }
 }
 
@@ -667,17 +627,20 @@ pub async fn init_tenant_mgr(
         let tenant_dir_path = conf.tenant_path(&tenant_shard_id);
         let shard_identity = location_conf.shard;
         let slot = match location_conf.mode {
-            LocationMode::Attached(attached_conf) => TenantSlot::Attached(tenant_spawn(
-                conf,
-                tenant_shard_id,
-                &tenant_dir_path,
-                resources.clone(),
-                AttachedTenantConf::new(location_conf.tenant_conf, attached_conf),
-                shard_identity,
-                Some(init_order.clone()),
-                SpawnMode::Lazy,
-                &ctx,
-            )),
+            LocationMode::Attached(attached_conf) => TenantSlot::Attached(
+                tenant_spawn(
+                    conf,
+                    tenant_shard_id,
+                    &tenant_dir_path,
+                    resources.clone(),
+                    AttachedTenantConf::new(location_conf.tenant_conf, attached_conf),
+                    shard_identity,
+                    Some(init_order.clone()),
+                    SpawnMode::Lazy,
+                    &ctx,
+                )
+                .expect("global shutdown during init_tenant_mgr cannot happen"),
+            ),
             LocationMode::Secondary(secondary_conf) => {
                 info!(
                     tenant_id = %tenant_shard_id.tenant_id,
@@ -725,7 +688,7 @@ fn tenant_spawn(
     init_order: Option<InitializationOrder>,
     mode: SpawnMode,
     ctx: &RequestContext,
-) -> Arc<Tenant> {
+) -> Result<Arc<Tenant>, GlobalShutDown> {
     // All these conditions should have been satisfied by our caller: the tenant dir exists, is a well formed
     // path, and contains a configuration file.  Assertions that do synchronous I/O are limited to debug mode
     // to avoid impacting prod runtime performance.
@@ -1192,7 +1155,10 @@ impl TenantManager {
                     None,
                     spawn_mode,
                     ctx,
-                );
+                )
+                .map_err(|_: GlobalShutDown| {
+                    UpsertLocationError::Unavailable(TenantMapError::ShuttingDown)
+                })?;
 
                 TenantSlot::Attached(tenant)
             }
@@ -1313,7 +1279,7 @@ impl TenantManager {
             None,
             SpawnMode::Eager,
             ctx,
-        );
+        )?;
 
         slot_guard.upsert(TenantSlot::Attached(tenant))?;
 
@@ -1763,14 +1729,9 @@ impl TenantManager {
             let parent_timelines = timelines.keys().cloned().collect::<Vec<_>>();
             for timeline in timelines.values() {
                 tracing::info!(timeline_id=%timeline.timeline_id, "Loading list of layers to hardlink");
-                let timeline_layers = timeline
-                    .layers
-                    .read()
-                    .await
-                    .likely_resident_layers()
-                    .collect::<Vec<_>>();
+                let layers = timeline.layers.read().await;
 
-                for layer in timeline_layers {
+                for layer in layers.likely_resident_layers() {
                     let relative_path = layer
                         .local_path()
                         .strip_prefix(&parent_path)
@@ -1966,8 +1927,11 @@ impl TenantManager {
         tenant_shard_id: TenantShardId,
         timeline_id: TimelineId,
         prepared: PreparedTimelineDetach,
+        mut attempt: detach_ancestor::Attempt,
         ctx: &RequestContext,
-    ) -> Result<Vec<TimelineId>, anyhow::Error> {
+    ) -> Result<HashSet<TimelineId>, anyhow::Error> {
+        use crate::tenant::timeline::detach_ancestor::Error;
+        // FIXME: this is unnecessary, slotguard already has these semantics
         struct RevertOnDropSlot(Option<SlotGuard>);
 
         impl Drop for RevertOnDropSlot {
@@ -2015,43 +1979,98 @@ impl TenantManager {
 
         let timeline = tenant.get_timeline(timeline_id, true)?;
 
-        let reparented = timeline
-            .complete_detaching_timeline_ancestor(&tenant, prepared, ctx)
+        let resp = timeline
+            .detach_from_ancestor_and_reparent(&tenant, prepared, ctx)
             .await?;
 
         let mut slot_guard = slot_guard.into_inner();
 
-        let (_guard, progress) = utils::completion::channel();
-        match tenant.shutdown(progress, ShutdownMode::Hard).await {
-            Ok(()) => {
-                slot_guard.drop_old_value()?;
+        let tenant = if resp.reset_tenant_required() {
+            attempt.before_reset_tenant();
+
+            let (_guard, progress) = utils::completion::channel();
+            match tenant.shutdown(progress, ShutdownMode::Hard).await {
+                Ok(()) => {
+                    slot_guard.drop_old_value()?;
+                }
+                Err(_barrier) => {
+                    slot_guard.revert();
+                    // this really should not happen, at all, unless shutdown was already going?
+                    anyhow::bail!("Cannot restart Tenant, already shutting down");
+                }
             }
-            Err(_barrier) => {
-                slot_guard.revert();
-                // this really should not happen, at all, unless shutdown was already going?
-                anyhow::bail!("Cannot restart Tenant, already shutting down");
+
+            let tenant_path = self.conf.tenant_path(&tenant_shard_id);
+            let config = Tenant::load_tenant_config(self.conf, &tenant_shard_id)?;
+
+            let shard_identity = config.shard;
+            let tenant = tenant_spawn(
+                self.conf,
+                tenant_shard_id,
+                &tenant_path,
+                self.resources.clone(),
+                AttachedTenantConf::try_from(config)?,
+                shard_identity,
+                None,
+                SpawnMode::Eager,
+                ctx,
+            )?;
+
+            {
+                let mut g = tenant.ongoing_timeline_detach.lock().unwrap();
+                assert!(
+                    g.is_none(),
+                    "there cannot be any new timeline detach ancestor on newly created tenant"
+                );
+                *g = Some((attempt.timeline_id, attempt.new_barrier()));
             }
+
+            slot_guard.upsert(TenantSlot::Attached(tenant.clone()))?;
+            tenant
+        } else {
+            tracing::info!("skipping tenant_reset as no changes made required it");
+            tenant
+        };
+
+        if let Some(reparented) = resp.completed() {
+            // finally ask the restarted tenant to complete the detach
+            //
+            // rationale for 9999s: we don't really have a timetable here; if retried, the caller
+            // will get an 503.
+            tenant
+                .wait_to_become_active(std::time::Duration::from_secs(9999))
+                .await
+                .map_err(|e| {
+                    use pageserver_api::models::TenantState;
+                    use GetActiveTenantError::{Cancelled, WillNotBecomeActive};
+                    match e {
+                        Cancelled | WillNotBecomeActive(TenantState::Stopping { .. }) => {
+                            Error::ShuttingDown
+                        }
+                        other => Error::Unexpected(other.into()),
+                    }
+                })?;
+
+            utils::pausable_failpoint!(
+                "timeline-detach-ancestor::after_activating_before_finding-pausable"
+            );
+
+            let timeline = tenant
+                .get_timeline(attempt.timeline_id, true)
+                .map_err(|_| Error::DetachedNotFoundAfterRestart)?;
+
+            timeline
+                .complete_detaching_timeline_ancestor(&tenant, attempt, ctx)
+                .await
+                .map(|()| reparented)
+                .map_err(|e| e.into())
+        } else {
+            // at least the latest versions have now been downloaded and refreshed; be ready to
+            // retry another time.
+            Err(anyhow::anyhow!(
+                "failed to reparent all candidate timelines, please retry"
+            ))
         }
-
-        let tenant_path = self.conf.tenant_path(&tenant_shard_id);
-        let config = Tenant::load_tenant_config(self.conf, &tenant_shard_id)?;
-
-        let shard_identity = config.shard;
-        let tenant = tenant_spawn(
-            self.conf,
-            tenant_shard_id,
-            &tenant_path,
-            self.resources.clone(),
-            AttachedTenantConf::try_from(config)?,
-            shard_identity,
-            None,
-            SpawnMode::Eager,
-            ctx,
-        );
-
-        slot_guard.upsert(TenantSlot::Attached(tenant))?;
-
-        Ok(reparented)
     }
 
     /// A page service client sends a TenantId, and to look up the correct Tenant we must
@@ -2088,7 +2107,6 @@ impl TenantManager {
                     };
 
                     match selector {
-                        ShardSelector::First => return ShardResolveResult::Found(tenant.clone()),
                         ShardSelector::Zero if slot.0.shard_number == ShardNumber(0) => {
                             return ShardResolveResult::Found(tenant.clone())
                         }
@@ -2124,6 +2142,57 @@ impl TenantManager {
             }
         }
     }
+
+    /// Calculate the tenant shards' contributions to this pageserver's utilization metrics.  The
+    /// returned values are:
+    ///  - the number of bytes of local disk space this pageserver's shards are requesting, i.e.
+    ///    how much space they would use if not impacted by disk usage eviction.
+    ///  - the number of tenant shards currently on this pageserver, including attached
+    ///    and secondary.
+    ///
+    /// This function is quite expensive: callers are expected to cache the result and
+    /// limit how often they call it.
+    pub(crate) fn calculate_utilization(&self) -> Result<(u64, u32), TenantMapListError> {
+        let tenants = self.tenants.read().unwrap();
+        let m = match &*tenants {
+            TenantsMap::Initializing => return Err(TenantMapListError::Initializing),
+            TenantsMap::Open(m) | TenantsMap::ShuttingDown(m) => m,
+        };
+        let shard_count = m.len();
+        let mut wanted_bytes = 0;
+
+        for tenant_slot in m.values() {
+            match tenant_slot {
+                TenantSlot::InProgress(_barrier) => {
+                    // While a slot is being changed, we can't know how much storage it wants.  This
+                    // means this function's output can fluctuate if a lot of changes are going on
+                    // (such as transitions from secondary to attached).
+                    //
+                    // We could wait for the barrier and retry, but it's important that the utilization
+                    // API is responsive, and the data quality impact is not very significant.
+                    continue;
+                }
+                TenantSlot::Attached(tenant) => {
+                    wanted_bytes += tenant.local_storage_wanted();
+                }
+                TenantSlot::Secondary(secondary) => {
+                    let progress = secondary.progress.lock().unwrap();
+                    wanted_bytes += if progress.heatmap_mtime.is_some() {
+                        // If we have heatmap info, then we will 'want' the sum
+                        // of the size of layers in the heatmap: this is how much space
+                        // we would use if not doing any eviction.
+                        progress.bytes_total
+                    } else {
+                        // In the absence of heatmap info, assume that the secondary location simply
+                        // needs as much space as it is currently using.
+                        secondary.resident_size_metric.get()
+                    }
+                }
+            }
+        }
+
+        Ok((wanted_bytes, shard_count as u32))
+    }
 }
 
 #[derive(Debug, thiserror::Error)]
@@ -2170,6 +2239,9 @@ pub(crate) enum GetActiveTenantError {
     /// never happen.
     #[error("Tenant is broken: {0}")]
     Broken(String),
+
+    #[error("reconnect to switch tenant id")]
+    SwitchedTenant,
 }
 
 #[derive(Debug, thiserror::Error)]

pageserver/src/tenant/remote_timeline_client.rs

@@ -736,12 +736,13 @@ impl RemoteTimelineClient {
         Ok(())
     }
 
+    /// Reparent this timeline to a new parent.
+    ///
+    /// A retryable step of timeline ancestor detach.
     pub(crate) async fn schedule_reparenting_and_wait(
         self: &Arc<Self>,
         new_parent: &TimelineId,
     ) -> anyhow::Result<()> {
-        // FIXME: because of how Timeline::schedule_uploads works when called from layer flushing
-        // and reads the in-memory part we cannot do the detaching like this
         let receiver = {
             let mut guard = self.upload_queue.lock().unwrap();
             let upload_queue = guard.initialized_mut()?;
@@ -752,17 +753,25 @@ impl RemoteTimelineClient {
                 ));
             };
 
-            upload_queue.dirty.metadata.reparent(new_parent);
-            upload_queue.dirty.lineage.record_previous_ancestor(&prev);
+            let uploaded = &upload_queue.clean.0.metadata;
 
-            self.schedule_index_upload(upload_queue)?;
+            if uploaded.ancestor_timeline().is_none() && !uploaded.ancestor_lsn().is_valid() {
+                // nothing to do
+                None
+            } else {
+                upload_queue.dirty.metadata.reparent(new_parent);
+                upload_queue.dirty.lineage.record_previous_ancestor(&prev);
 
-            self.schedule_barrier0(upload_queue)
+                self.schedule_index_upload(upload_queue)?;
+
+                Some(self.schedule_barrier0(upload_queue))
+            }
         };
 
-        Self::wait_completion0(receiver)
-            .await
-            .context("wait completion")
+        if let Some(receiver) = receiver {
+            Self::wait_completion0(receiver).await?;
+        }
+        Ok(())
     }
 
     /// Schedules uploading a new version of `index_part.json` with the given layers added,
@@ -778,26 +787,142 @@ impl RemoteTimelineClient {
             let mut guard = self.upload_queue.lock().unwrap();
             let upload_queue = guard.initialized_mut()?;
 
-            upload_queue.dirty.metadata.detach_from_ancestor(&adopted);
-            upload_queue.dirty.lineage.record_detaching(&adopted);
+            if upload_queue.clean.0.lineage.detached_previous_ancestor() == Some(adopted) {
+                None
+            } else {
+                upload_queue.dirty.metadata.detach_from_ancestor(&adopted);
+                upload_queue.dirty.lineage.record_detaching(&adopted);
 
-            for layer in layers {
-                upload_queue
-                    .dirty
-                    .layer_metadata
-                    .insert(layer.layer_desc().layer_name(), layer.metadata());
+                for layer in layers {
+                    let prev = upload_queue
+                        .dirty
+                        .layer_metadata
+                        .insert(layer.layer_desc().layer_name(), layer.metadata());
+                    assert!(prev.is_none(), "copied layer existed already {layer}");
+                }
+
+                self.schedule_index_upload(upload_queue)?;
+
+                Some(self.schedule_barrier0(upload_queue))
             }
-
-            self.schedule_index_upload(upload_queue)?;
-
-            let barrier = self.schedule_barrier0(upload_queue);
-            self.launch_queued_tasks(upload_queue);
-            barrier
         };
 
-        Self::wait_completion0(barrier)
-            .await
-            .context("wait completion")
+        if let Some(barrier) = barrier {
+            Self::wait_completion0(barrier).await?;
+        }
+        Ok(())
+    }
+
+    /// Adds a gc blocking reason for this timeline if one does not exist already.
+    ///
+    /// A retryable step of timeline detach ancestor.
+    ///
+    /// Returns a future which waits until the completion of the upload.
+    pub(crate) fn schedule_insert_gc_block_reason(
+        self: &Arc<Self>,
+        reason: index::GcBlockingReason,
+    ) -> Result<impl std::future::Future<Output = Result<(), WaitCompletionError>>, NotInitialized>
+    {
+        let maybe_barrier = {
+            let mut guard = self.upload_queue.lock().unwrap();
+            let upload_queue = guard.initialized_mut()?;
+
+            if let index::GcBlockingReason::DetachAncestor = reason {
+                if upload_queue.dirty.metadata.ancestor_timeline().is_none() {
+                    drop(guard);
+                    panic!("cannot start detach ancestor if there is nothing to detach from");
+                }
+            }
+
+            let wanted = |x: Option<&index::GcBlocking>| x.is_some_and(|x| x.blocked_by(reason));
+
+            let current = upload_queue.dirty.gc_blocking.as_ref();
+            let uploaded = upload_queue.clean.0.gc_blocking.as_ref();
+
+            match (current, uploaded) {
+                (x, y) if wanted(x) && wanted(y) => None,
+                (x, y) if wanted(x) && !wanted(y) => Some(self.schedule_barrier0(upload_queue)),
+                // Usual case: !wanted(x) && !wanted(y)
+                //
+                // Unusual: !wanted(x) && wanted(y) which means we have two processes waiting to
+                // turn on and off some reason.
+                (x, y) => {
+                    if !wanted(x) && wanted(y) {
+                        // this could be avoided by having external in-memory synchronization, like
+                        // timeline detach ancestor
+                        warn!(?reason, op="insert", "unexpected: two racing processes to enable and disable a gc blocking reason");
+                    }
+
+                    // at this point, the metadata must always show that there is a parent
+                    upload_queue.dirty.gc_blocking = current
+                        .map(|x| x.with_reason(reason))
+                        .or_else(|| Some(index::GcBlocking::started_now_for(reason)));
+                    self.schedule_index_upload(upload_queue)?;
+                    Some(self.schedule_barrier0(upload_queue))
+                }
+            }
+        };
+
+        Ok(async move {
+            if let Some(barrier) = maybe_barrier {
+                Self::wait_completion0(barrier).await?;
+            }
+            Ok(())
+        })
+    }
+
+    /// Removes a gc blocking reason for this timeline if one exists.
+    ///
+    /// A retryable step of timeline detach ancestor.
+    ///
+    /// Returns a future which waits until the completion of the upload.
+    pub(crate) fn schedule_remove_gc_block_reason(
+        self: &Arc<Self>,
+        reason: index::GcBlockingReason,
+    ) -> Result<impl std::future::Future<Output = Result<(), WaitCompletionError>>, NotInitialized>
+    {
+        let maybe_barrier = {
+            let mut guard = self.upload_queue.lock().unwrap();
+            let upload_queue = guard.initialized_mut()?;
+
+            if let index::GcBlockingReason::DetachAncestor = reason {
+                if !upload_queue.clean.0.lineage.is_detached_from_ancestor() {
+                    drop(guard);
+                    panic!("cannot complete timeline_ancestor_detach while not detached");
+                }
+            }
+
+            let wanted = |x: Option<&index::GcBlocking>| {
+                x.is_none() || x.is_some_and(|b| !b.blocked_by(reason))
+            };
+
+            let current = upload_queue.dirty.gc_blocking.as_ref();
+            let uploaded = upload_queue.clean.0.gc_blocking.as_ref();
+
+            match (current, uploaded) {
+                (x, y) if wanted(x) && wanted(y) => None,
+                (x, y) if wanted(x) && !wanted(y) => Some(self.schedule_barrier0(upload_queue)),
+                (x, y) => {
+                    if !wanted(x) && wanted(y) {
+                        warn!(?reason, op="remove", "unexpected: two racing processes to enable and disable a gc blocking reason (remove)");
+                    }
+
+                    upload_queue.dirty.gc_blocking =
+                        current.as_ref().and_then(|x| x.without_reason(reason));
+                    assert!(wanted(upload_queue.dirty.gc_blocking.as_ref()));
+                    // FIXME: bogus ?
+                    self.schedule_index_upload(upload_queue)?;
+                    Some(self.schedule_barrier0(upload_queue))
+                }
+            }
+        };
+
+        Ok(async move {
+            if let Some(barrier) = maybe_barrier {
+                Self::wait_completion0(barrier).await?;
+            }
+            Ok(())
+        })
     }
 
     /// Launch an upload operation in the background; the file is added to be included in next
@@ -868,7 +993,10 @@ impl RemoteTimelineClient {
     ///
     /// The files will be leaked in remote storage unless [`Self::schedule_deletion_of_unlinked`]
     /// is invoked on them.
-    pub(crate) fn schedule_gc_update(self: &Arc<Self>, gc_layers: &[Layer]) -> anyhow::Result<()> {
+    pub(crate) fn schedule_gc_update(
+        self: &Arc<Self>,
+        gc_layers: &[Layer],
+    ) -> Result<(), NotInitialized> {
         let mut guard = self.upload_queue.lock().unwrap();
         let upload_queue = guard.initialized_mut()?;
 
@@ -1378,6 +1506,18 @@ impl RemoteTimelineClient {
                 .dirty
                 .layer_metadata
                 .drain()
+                .filter(|(_file_name, meta)| {
+                    // Filter out layers that belonged to an ancestor shard.  Since we are deleting the whole timeline from
+                    // all shards anyway, we _could_ delete these, but
+                    // - it creates a potential race if other shards are still
+                    //   using the layers while this shard deletes them.
+                    // - it means that if we rolled back the shard split, the ancestor shards would be in a state where
+                    //   these timelines are present but corrupt (their index exists but some layers don't)
+                    //
+                    // These layers will eventually be cleaned up by the scrubber when it does physical GC.
+                    meta.shard.shard_number == self.tenant_shard_id.shard_number
+                        && meta.shard.shard_count == self.tenant_shard_id.shard_count
+                })
                 .map(|(file_name, meta)| {
                     remote_layer_path(
                         &self.tenant_shard_id.tenant_id,

pageserver/src/tenant/remote_timeline_client/index.rs

@@ -60,6 +60,9 @@ pub struct IndexPart {
     #[serde(default)]
     pub(crate) lineage: Lineage,
 
+    #[serde(skip_serializing_if = "Option::is_none", default)]
+    pub(crate) gc_blocking: Option<GcBlocking>,
+
     /// Describes the kind of aux files stored in the timeline.
     ///
     /// The value is modified during file ingestion when the latest wanted value communicated via tenant config is applied if it is acceptable.
@@ -85,10 +88,11 @@ impl IndexPart {
     /// - 6: last_aux_file_policy is added.
     /// - 7: metadata_bytes is no longer written, but still read
     /// - 8: added `archived_at`
-    const LATEST_VERSION: usize = 8;
+    /// - 9: +gc_blocking
+    const LATEST_VERSION: usize = 9;
 
     // Versions we may see when reading from a bucket.
-    pub const KNOWN_VERSIONS: &'static [usize] = &[1, 2, 3, 4, 5, 6, 7, 8];
+    pub const KNOWN_VERSIONS: &'static [usize] = &[1, 2, 3, 4, 5, 6, 7, 8, 9];
 
     pub const FILE_NAME: &'static str = "index_part.json";
 
@@ -101,6 +105,7 @@ impl IndexPart {
             deleted_at: None,
             archived_at: None,
             lineage: Default::default(),
+            gc_blocking: None,
             last_aux_file_policy: None,
         }
     }
@@ -211,26 +216,47 @@ fn is_false(b: &bool) -> bool {
 impl Lineage {
     const REMEMBER_AT_MOST: usize = 100;
 
-    pub(crate) fn record_previous_ancestor(&mut self, old_ancestor: &TimelineId) {
+    pub(crate) fn record_previous_ancestor(&mut self, old_ancestor: &TimelineId) -> bool {
         if self.reparenting_history.last() == Some(old_ancestor) {
             // do not re-record it
-            return;
-        }
+            false
+        } else {
+            #[cfg(feature = "testing")]
+            {
+                let existing = self
+                    .reparenting_history
+                    .iter()
+                    .position(|x| x == old_ancestor);
+                assert_eq!(
+                    existing, None,
+                    "we cannot reparent onto and off and onto the same timeline twice"
+                );
+            }
+            let drop_oldest = self.reparenting_history.len() + 1 >= Self::REMEMBER_AT_MOST;
 
-        let drop_oldest = self.reparenting_history.len() + 1 >= Self::REMEMBER_AT_MOST;
-
-        self.reparenting_history_truncated |= drop_oldest;
-        if drop_oldest {
-            self.reparenting_history.remove(0);
+            self.reparenting_history_truncated |= drop_oldest;
+            if drop_oldest {
+                self.reparenting_history.remove(0);
+            }
+            self.reparenting_history.push(*old_ancestor);
+            true
         }
-        self.reparenting_history.push(*old_ancestor);
     }
 
-    pub(crate) fn record_detaching(&mut self, branchpoint: &(TimelineId, Lsn)) {
-        assert!(self.original_ancestor.is_none());
-
-        self.original_ancestor =
-            Some((branchpoint.0, branchpoint.1, chrono::Utc::now().naive_utc()));
+    /// Returns true if anything changed.
+    pub(crate) fn record_detaching(&mut self, branchpoint: &(TimelineId, Lsn)) -> bool {
+        if let Some((id, lsn, _)) = self.original_ancestor {
+            assert_eq!(
+                &(id, lsn),
+                branchpoint,
+                "detaching attempt has to be for the same ancestor we are already detached from"
+            );
+            false
+        } else {
+            self.original_ancestor =
+                Some((branchpoint.0, branchpoint.1, chrono::Utc::now().naive_utc()));
+            true
+        }
     }
 
     /// The queried lsn is most likely the basebackup lsn, and this answers question "is it allowed
@@ -242,15 +268,79 @@ impl Lineage {
             .is_some_and(|(_, ancestor_lsn, _)| ancestor_lsn == lsn)
     }
 
-    pub(crate) fn is_detached_from_original_ancestor(&self) -> bool {
+    /// Returns true if the timeline originally had an ancestor, and no longer has one.
+    pub(crate) fn is_detached_from_ancestor(&self) -> bool {
         self.original_ancestor.is_some()
     }
 
+    /// Returns original ancestor timeline id and lsn that this timeline has been detached from.
+    pub(crate) fn detached_previous_ancestor(&self) -> Option<(TimelineId, Lsn)> {
+        self.original_ancestor.map(|(id, lsn, _)| (id, lsn))
+    }
+
     pub(crate) fn is_reparented(&self) -> bool {
         !self.reparenting_history.is_empty()
     }
 }
 
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
+pub(crate) struct GcBlocking {
+    pub(crate) started_at: NaiveDateTime,
+    pub(crate) reasons: enumset::EnumSet<GcBlockingReason>,
+}
+
+#[derive(Debug, enumset::EnumSetType, serde::Serialize, serde::Deserialize)]
+#[enumset(serialize_repr = "list")]
+pub(crate) enum GcBlockingReason {
+    Manual,
+    DetachAncestor,
+}
+
+impl GcBlocking {
+    pub(super) fn started_now_for(reason: GcBlockingReason) -> Self {
+        GcBlocking {
+            started_at: chrono::Utc::now().naive_utc(),
+            reasons: enumset::EnumSet::only(reason),
+        }
+    }
+
+    /// Returns true if the given reason is one of the reasons why the gc is blocked.
+    pub(crate) fn blocked_by(&self, reason: GcBlockingReason) -> bool {
+        self.reasons.contains(reason)
+    }
+
+    /// Returns a version of self with the given reason.
+    pub(super) fn with_reason(&self, reason: GcBlockingReason) -> Self {
+        assert!(!self.blocked_by(reason));
+        let mut reasons = self.reasons;
+        reasons.insert(reason);
+
+        Self {
+            started_at: self.started_at,
+            reasons,
+        }
+    }
+
+    /// Returns a version of self without the given reason. Assumption is that if
+    /// there are no more reasons, we can unblock the gc by returning `None`.
+    pub(super) fn without_reason(&self, reason: GcBlockingReason) -> Option<Self> {
+        assert!(self.blocked_by(reason));
+
+        if self.reasons.len() == 1 {
+            None
+        } else {
+            let mut reasons = self.reasons;
+            assert!(reasons.remove(reason));
+            assert!(!reasons.is_empty());
+
+            Some(Self {
+                started_at: self.started_at,
+                reasons,
+            })
+        }
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
@@ -292,6 +382,7 @@ mod tests {
             deleted_at: None,
             archived_at: None,
             lineage: Lineage::default(),
+            gc_blocking: None,
             last_aux_file_policy: None,
         };
 
@@ -335,6 +426,7 @@ mod tests {
             deleted_at: None,
             archived_at: None,
             lineage: Lineage::default(),
+            gc_blocking: None,
             last_aux_file_policy: None,
         };
 
@@ -379,6 +471,7 @@ mod tests {
             deleted_at: Some(parse_naive_datetime("2023-07-31T09:00:00.123000000")),
             archived_at: None,
             lineage: Lineage::default(),
+            gc_blocking: None,
             last_aux_file_policy: None,
         };
 
@@ -426,6 +519,7 @@ mod tests {
             deleted_at: None,
             archived_at: None,
             lineage: Lineage::default(),
+            gc_blocking: None,
             last_aux_file_policy: None,
         };
 
@@ -468,6 +562,7 @@ mod tests {
             deleted_at: Some(parse_naive_datetime("2023-07-31T09:00:00.123000000")),
             archived_at: None,
             lineage: Lineage::default(),
+            gc_blocking: None,
             last_aux_file_policy: None,
         };
 
@@ -513,6 +608,7 @@ mod tests {
                 reparenting_history: vec![TimelineId::from_str("e1bfd8c633d713d279e6fcd2bcc15b6d").unwrap()],
                 original_ancestor: Some((TimelineId::from_str("e2bfd8c633d713d279e6fcd2bcc15b6d").unwrap(), Lsn::from_str("0/15A7618").unwrap(), parse_naive_datetime("2024-05-07T18:52:36.322426563"))),
             },
+            gc_blocking: None,
             last_aux_file_policy: None,
         };
 
@@ -563,6 +659,7 @@ mod tests {
                 reparenting_history: vec![TimelineId::from_str("e1bfd8c633d713d279e6fcd2bcc15b6d").unwrap()],
                 original_ancestor: Some((TimelineId::from_str("e2bfd8c633d713d279e6fcd2bcc15b6d").unwrap(), Lsn::from_str("0/15A7618").unwrap(), parse_naive_datetime("2024-05-07T18:52:36.322426563"))),
             },
+            gc_blocking: None,
             last_aux_file_policy: Some(AuxFilePolicy::V2),
         };
 
@@ -618,6 +715,7 @@ mod tests {
             deleted_at: Some(parse_naive_datetime("2023-07-31T09:00:00.123000000")),
             archived_at: None,
             lineage: Default::default(),
+            gc_blocking: None,
             last_aux_file_policy: Default::default(),
         };
 
@@ -674,6 +772,7 @@ mod tests {
             deleted_at: Some(parse_naive_datetime("2023-07-31T09:00:00.123000000")),
             archived_at: Some(parse_naive_datetime("2023-04-29T09:00:00.123000000")),
             lineage: Default::default(),
+            gc_blocking: None,
             last_aux_file_policy: Default::default(),
         };
 
@@ -681,6 +780,68 @@ mod tests {
         assert_eq!(part, expected);
     }
 
+    #[test]
+    fn v9_indexpart_is_parsed() {
+        let example = r#"{
+            "version": 9,
+            "layer_metadata":{
+                "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__0000000001696070-00000000016960E9": { "file_size": 25600000 },
+                "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51": { "file_size": 9007199254741001 }
+            },
+            "disk_consistent_lsn":"0/16960E8",
+            "metadata": {
+                "disk_consistent_lsn": "0/16960E8",
+                "prev_record_lsn": "0/1696070",
+                "ancestor_timeline": "e45a7f37d3ee2ff17dc14bf4f4e3f52e",
+                "ancestor_lsn": "0/0",
+                "latest_gc_cutoff_lsn": "0/1696070",
+                "initdb_lsn": "0/1696070",
+                "pg_version": 14
+            },
+            "gc_blocking": {
+                "started_at": "2024-07-19T09:00:00.123",
+                "reasons": ["DetachAncestor"]
+            }
+        }"#;
+
+        let expected = IndexPart {
+            version: 9,
+            layer_metadata: HashMap::from([
+                ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__0000000001696070-00000000016960E9".parse().unwrap(), LayerFileMetadata {
+                    file_size: 25600000,
+                    generation: Generation::none(),
+                    shard: ShardIndex::unsharded()
+                }),
+                ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap(), LayerFileMetadata {
+                    file_size: 9007199254741001,
+                    generation: Generation::none(),
+                    shard: ShardIndex::unsharded()
+                })
+            ]),
+            disk_consistent_lsn: "0/16960E8".parse::<Lsn>().unwrap(),
+            metadata: TimelineMetadata::new(
+                Lsn::from_str("0/16960E8").unwrap(),
+                Some(Lsn::from_str("0/1696070").unwrap()),
+                Some(TimelineId::from_str("e45a7f37d3ee2ff17dc14bf4f4e3f52e").unwrap()),
+                Lsn::INVALID,
+                Lsn::from_str("0/1696070").unwrap(),
+                Lsn::from_str("0/1696070").unwrap(),
+                14,
+            ).with_recalculated_checksum().unwrap(),
+            deleted_at: None,
+            lineage: Default::default(),
+            gc_blocking: Some(GcBlocking {
+                started_at: parse_naive_datetime("2024-07-19T09:00:00.123000000"),
+                reasons: enumset::EnumSet::from_iter([GcBlockingReason::DetachAncestor]),
+            }),
+            last_aux_file_policy: Default::default(),
+            archived_at: None,
+        };
+
+        let part = IndexPart::from_s3_bytes(example.as_bytes()).unwrap();
+        assert_eq!(part, expected);
+    }
+
     fn parse_naive_datetime(s: &str) -> NaiveDateTime {
         chrono::NaiveDateTime::parse_from_str(s, "%Y-%m-%dT%H:%M:%S.%f").unwrap()
     }

pageserver/src/tenant/secondary/downloader.rs

@@ -55,7 +55,7 @@ use tokio_util::sync::CancellationToken;
 use tracing::{info_span, instrument, warn, Instrument};
 use utils::{
     backoff, completion::Barrier, crashsafe::path_with_suffix_extension, failpoint_support, fs_ext,
-    id::TimelineId, serde_system_time,
+    id::TimelineId, pausable_failpoint, serde_system_time,
 };
 
 use super::{
@@ -1146,12 +1146,14 @@ impl<'a> TenantDownloader<'a> {
         layer: HeatMapLayer,
         ctx: &RequestContext,
     ) -> Result<Option<HeatMapLayer>, UpdateError> {
-        // Failpoint for simulating slow remote storage
+        // Failpoints for simulating slow remote storage
         failpoint_support::sleep_millis_async!(
             "secondary-layer-download-sleep",
             &self.secondary_state.cancel
         );
 
+        pausable_failpoint!("secondary-layer-download-pausable");
+
         let local_path = local_layer_path(
             self.conf,
             tenant_shard_id,

pageserver/src/tenant/storage_layer.rs

@@ -8,6 +8,9 @@ mod layer_desc;
 mod layer_name;
 pub mod merge_iterator;
 
+#[cfg(test)]
+pub mod split_writer;
+
 use crate::context::{AccessStatsBehavior, RequestContext};
 use crate::repository::Value;
 use crate::walrecord::NeonWalRecord;
@@ -432,39 +435,18 @@ impl ReadableLayer {
     }
 }
 
-/// Return value from [`Layer::get_value_reconstruct_data`]
-#[derive(Clone, Copy, Debug)]
-pub enum ValueReconstructResult {
-    /// Got all the data needed to reconstruct the requested page
-    Complete,
-    /// This layer didn't contain all the required data, the caller should look up
-    /// the predecessor layer at the returned LSN and collect more data from there.
-    Continue,
-
-    /// This layer didn't contain data needed to reconstruct the page version at
-    /// the returned LSN. This is usually considered an error, but might be OK
-    /// in some circumstances.
-    Missing,
-}
-
 /// Layers contain a hint indicating whether they are likely to be used for reads.  This is a hint rather
 /// than an authoritative value, so that we do not have to update it synchronously when changing the visibility
 /// of layers (for example when creating a branch that makes some previously covered layers visible).  It should
 /// be used for cache management but not for correctness-critical checks.
-#[derive(Default, Debug, Clone, PartialEq, Eq)]
-pub(crate) enum LayerVisibilityHint {
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub enum LayerVisibilityHint {
     /// A Visible layer might be read while serving a read, because there is not an image layer between it
     /// and a readable LSN (the tip of the branch or a child's branch point)
     Visible,
     /// A Covered layer probably won't be read right now, but _can_ be read in future if someone creates
     /// a branch or ephemeral endpoint at an LSN below the layer that covers this.
-    #[allow(unused)]
     Covered,
-    /// Calculating layer visibilty requires I/O, so until this has happened layers are loaded
-    /// in this state.  Note that newly written layers may be called Visible immediately, this uninitialized
-    /// state is for when existing layers are constructed while loading a timeline.
-    #[default]
-    Uninitialized,
 }
 
 pub(crate) struct LayerAccessStats(std::sync::atomic::AtomicU64);
@@ -557,19 +539,25 @@ impl LayerAccessStats {
         self.record_residence_event_at(SystemTime::now())
     }
 
-    pub(crate) fn record_access_at(&self, now: SystemTime) {
+    fn record_access_at(&self, now: SystemTime) -> bool {
         let (mut mask, mut value) = Self::to_low_res_timestamp(Self::ATIME_SHIFT, now);
 
         // A layer which is accessed must be visible.
         mask |= 0x1 << Self::VISIBILITY_SHIFT;
         value |= 0x1 << Self::VISIBILITY_SHIFT;
 
-        self.write_bits(mask, value);
+        let old_bits = self.write_bits(mask, value);
+        !matches!(
+            self.decode_visibility(old_bits),
+            LayerVisibilityHint::Visible
+        )
     }
 
-    pub(crate) fn record_access(&self, ctx: &RequestContext) {
+    /// Returns true if we modified the layer's visibility to set it to Visible implicitly
+    /// as a result of this access
+    pub(crate) fn record_access(&self, ctx: &RequestContext) -> bool {
         if ctx.access_stats_behavior() == AccessStatsBehavior::Skip {
-            return;
+            return false;
         }
 
         self.record_access_at(SystemTime::now())
@@ -626,23 +614,30 @@ impl LayerAccessStats {
         }
     }
 
-    pub(crate) fn set_visibility(&self, visibility: LayerVisibilityHint) {
-        let value = match visibility {
-            LayerVisibilityHint::Visible => 0x1 << Self::VISIBILITY_SHIFT,
-            LayerVisibilityHint::Covered | LayerVisibilityHint::Uninitialized => 0x0,
-        };
-
-        self.write_bits(0x1 << Self::VISIBILITY_SHIFT, value);
-    }
-
-    pub(crate) fn visibility(&self) -> LayerVisibilityHint {
-        let read = self.0.load(std::sync::atomic::Ordering::Relaxed);
-        match (read >> Self::VISIBILITY_SHIFT) & 0x1 {
+    /// Helper for extracting the visibility hint from the literal value of our inner u64
+    fn decode_visibility(&self, bits: u64) -> LayerVisibilityHint {
+        match (bits >> Self::VISIBILITY_SHIFT) & 0x1 {
             1 => LayerVisibilityHint::Visible,
             0 => LayerVisibilityHint::Covered,
             _ => unreachable!(),
         }
     }
+
+    /// Returns the old value which has been replaced
+    pub(crate) fn set_visibility(&self, visibility: LayerVisibilityHint) -> LayerVisibilityHint {
+        let value = match visibility {
+            LayerVisibilityHint::Visible => 0x1 << Self::VISIBILITY_SHIFT,
+            LayerVisibilityHint::Covered => 0x0,
+        };
+
+        let old_bits = self.write_bits(0x1 << Self::VISIBILITY_SHIFT, value);
+        self.decode_visibility(old_bits)
+    }
+
+    pub(crate) fn visibility(&self) -> LayerVisibilityHint {
+        let read = self.0.load(std::sync::atomic::Ordering::Relaxed);
+        self.decode_visibility(read)
+    }
 }
 
 /// Get a layer descriptor from a layer.

pageserver/src/tenant/storage_layer/delta_layer.rs

@@ -36,13 +36,12 @@ use crate::tenant::block_io::{BlockBuf, BlockCursor, BlockLease, BlockReader, Fi
 use crate::tenant::disk_btree::{
     DiskBtreeBuilder, DiskBtreeIterator, DiskBtreeReader, VisitDirection,
 };
-use crate::tenant::storage_layer::{Layer, ValueReconstructResult, ValueReconstructState};
 use crate::tenant::timeline::GetVectoredError;
 use crate::tenant::vectored_blob_io::{
     BlobFlag, MaxVectoredReadBytes, StreamingVectoredReadPlanner, VectoredBlobReader, VectoredRead,
     VectoredReadPlanner,
 };
-use crate::tenant::{PageReconstructError, Timeline};
+use crate::tenant::PageReconstructError;
 use crate::virtual_file::{self, VirtualFile};
 use crate::{walrecord, TEMP_FILE_SUFFIX};
 use crate::{DELTA_FILE_MAGIC, STORAGE_FORMAT_VERSION};
@@ -72,10 +71,7 @@ use utils::{
     lsn::Lsn,
 };
 
-use super::{
-    AsLayerDesc, LayerAccessStats, LayerName, PersistentLayerDesc, ResidentLayer,
-    ValuesReconstructState,
-};
+use super::{AsLayerDesc, LayerName, PersistentLayerDesc, ValuesReconstructState};
 
 ///
 /// Header stored in the beginning of the file
@@ -200,7 +196,6 @@ impl DeltaKey {
 pub struct DeltaLayer {
     path: Utf8PathBuf,
     pub desc: PersistentLayerDesc,
-    access_stats: LayerAccessStats,
     inner: OnceCell<Arc<DeltaLayerInner>>,
 }
 
@@ -299,7 +294,6 @@ impl DeltaLayer {
     /// not loaded already.
     ///
     async fn load(&self, ctx: &RequestContext) -> Result<&Arc<DeltaLayerInner>> {
-        self.access_stats.record_access(ctx);
         // Quick exit if already loaded
         self.inner
             .get_or_try_init(|| self.load_inner(ctx))
@@ -350,7 +344,6 @@ impl DeltaLayer {
                 summary.lsn_range,
                 metadata.len(),
             ),
-            access_stats: Default::default(),
             inner: OnceCell::new(),
         })
     }
@@ -373,7 +366,6 @@ impl DeltaLayer {
 /// 3. Call `finish`.
 ///
 struct DeltaLayerWriterInner {
-    conf: &'static PageServerConf,
     pub path: Utf8PathBuf,
     timeline_id: TimelineId,
     tenant_shard_id: TenantShardId,
@@ -384,6 +376,9 @@ struct DeltaLayerWriterInner {
     tree: DiskBtreeBuilder<BlockBuf, DELTA_KEY_SIZE>,
 
     blob_writer: BlobWriter<true>,
+
+    // Number of key-lsns in the layer.
+    num_keys: usize,
 }
 
 impl DeltaLayerWriterInner {
@@ -417,7 +412,6 @@ impl DeltaLayerWriterInner {
         let tree_builder = DiskBtreeBuilder::new(block_buf);
 
         Ok(Self {
-            conf,
             path,
             timeline_id,
             tenant_shard_id,
@@ -425,6 +419,7 @@ impl DeltaLayerWriterInner {
             lsn_range,
             tree: tree_builder,
             blob_writer,
+            num_keys: 0,
         })
     }
 
@@ -475,6 +470,9 @@ impl DeltaLayerWriterInner {
 
         let delta_key = DeltaKey::from_key_lsn(&key, lsn);
         let res = self.tree.append(&delta_key.0, blob_ref.0);
+
+        self.num_keys += 1;
+
         (val, res.map_err(|e| anyhow::anyhow!(e)))
     }
 
@@ -488,11 +486,10 @@ impl DeltaLayerWriterInner {
     async fn finish(
         self,
         key_end: Key,
-        timeline: &Arc<Timeline>,
         ctx: &RequestContext,
-    ) -> anyhow::Result<ResidentLayer> {
+    ) -> anyhow::Result<(PersistentLayerDesc, Utf8PathBuf)> {
         let temp_path = self.path.clone();
-        let result = self.finish0(key_end, timeline, ctx).await;
+        let result = self.finish0(key_end, ctx).await;
         if result.is_err() {
             tracing::info!(%temp_path, "cleaning up temporary file after error during writing");
             if let Err(e) = std::fs::remove_file(&temp_path) {
@@ -505,9 +502,8 @@ impl DeltaLayerWriterInner {
     async fn finish0(
         self,
         key_end: Key,
-        timeline: &Arc<Timeline>,
         ctx: &RequestContext,
-    ) -> anyhow::Result<ResidentLayer> {
+    ) -> anyhow::Result<(PersistentLayerDesc, Utf8PathBuf)> {
         let index_start_blk =
             ((self.blob_writer.size() + PAGE_SZ as u64 - 1) / PAGE_SZ as u64) as u32;
 
@@ -572,11 +568,9 @@ impl DeltaLayerWriterInner {
         // fsync the file
         file.sync_all().await?;
 
-        let layer = Layer::finish_creating(self.conf, timeline, desc, &self.path)?;
+        trace!("created delta layer {}", self.path);
 
-        trace!("created delta layer {}", layer.local_path());
-
-        Ok(layer)
+        Ok((desc, self.path))
     }
 }
 
@@ -677,14 +671,20 @@ impl DeltaLayerWriter {
     pub(crate) async fn finish(
         mut self,
         key_end: Key,
-        timeline: &Arc<Timeline>,
         ctx: &RequestContext,
-    ) -> anyhow::Result<ResidentLayer> {
-        self.inner
-            .take()
-            .unwrap()
-            .finish(key_end, timeline, ctx)
-            .await
+    ) -> anyhow::Result<(PersistentLayerDesc, Utf8PathBuf)> {
+        self.inner.take().unwrap().finish(key_end, ctx).await
+    }
+
+    #[cfg(test)]
+    pub(crate) fn num_keys(&self) -> usize {
+        self.inner.as_ref().unwrap().num_keys
+    }
+
+    #[cfg(test)]
+    pub(crate) fn estimated_size(&self) -> u64 {
+        let inner = self.inner.as_ref().unwrap();
+        inner.blob_writer.size() + inner.tree.borrow_writer().size() + PAGE_SZ as u64
     }
 }
 
@@ -808,95 +808,6 @@ impl DeltaLayerInner {
         })
     }
 
-    pub(super) async fn get_value_reconstruct_data(
-        &self,
-        key: Key,
-        lsn_range: Range<Lsn>,
-        reconstruct_state: &mut ValueReconstructState,
-        ctx: &RequestContext,
-    ) -> anyhow::Result<ValueReconstructResult> {
-        let mut need_image = true;
-        // Scan the page versions backwards, starting from `lsn`.
-        let block_reader = FileBlockReader::new(&self.file, self.file_id);
-        let tree_reader = DiskBtreeReader::<_, DELTA_KEY_SIZE>::new(
-            self.index_start_blk,
-            self.index_root_blk,
-            &block_reader,
-        );
-        let search_key = DeltaKey::from_key_lsn(&key, Lsn(lsn_range.end.0 - 1));
-
-        let mut offsets: Vec<(Lsn, u64)> = Vec::new();
-
-        tree_reader
-            .visit(
-                &search_key.0,
-                VisitDirection::Backwards,
-                |key, value| {
-                    let blob_ref = BlobRef(value);
-                    if key[..KEY_SIZE] != search_key.0[..KEY_SIZE] {
-                        return false;
-                    }
-                    let entry_lsn = DeltaKey::extract_lsn_from_buf(key);
-                    if entry_lsn < lsn_range.start {
-                        return false;
-                    }
-                    offsets.push((entry_lsn, blob_ref.pos()));
-
-                    !blob_ref.will_init()
-                },
-                &RequestContextBuilder::extend(ctx)
-                    .page_content_kind(PageContentKind::DeltaLayerBtreeNode)
-                    .build(),
-            )
-            .await?;
-
-        let ctx = &RequestContextBuilder::extend(ctx)
-            .page_content_kind(PageContentKind::DeltaLayerValue)
-            .build();
-
-        // Ok, 'offsets' now contains the offsets of all the entries we need to read
-        let cursor = block_reader.block_cursor();
-        let mut buf = Vec::new();
-        for (entry_lsn, pos) in offsets {
-            cursor
-                .read_blob_into_buf(pos, &mut buf, ctx)
-                .await
-                .with_context(|| {
-                    format!("Failed to read blob from virtual file {}", self.file.path)
-                })?;
-            let val = Value::des(&buf).with_context(|| {
-                format!(
-                    "Failed to deserialize file blob from virtual file {}",
-                    self.file.path
-                )
-            })?;
-            match val {
-                Value::Image(img) => {
-                    reconstruct_state.img = Some((entry_lsn, img));
-                    need_image = false;
-                    break;
-                }
-                Value::WalRecord(rec) => {
-                    let will_init = rec.will_init();
-                    reconstruct_state.records.push((entry_lsn, rec));
-                    if will_init {
-                        // This WAL record initializes the page, so no need to go further back
-                        need_image = false;
-                        break;
-                    }
-                }
-            }
-        }
-
-        // If an older page image is needed to reconstruct the page, let the
-        // caller know.
-        if need_image {
-            Ok(ValueReconstructResult::Continue)
-        } else {
-            Ok(ValueReconstructResult::Complete)
-        }
-    }
-
     // Look up the keys in the provided keyspace and update
     // the reconstruct state with whatever is found.
     //
@@ -1669,8 +1580,9 @@ pub(crate) mod test {
     use super::*;
     use crate::repository::Value;
     use crate::tenant::harness::TIMELINE_ID;
+    use crate::tenant::storage_layer::{Layer, ResidentLayer};
     use crate::tenant::vectored_blob_io::StreamingVectoredReadPlanner;
-    use crate::tenant::Tenant;
+    use crate::tenant::{Tenant, Timeline};
     use crate::{
         context::DownloadBehavior,
         task_mgr::TaskKind,
@@ -1964,9 +1876,8 @@ pub(crate) mod test {
             res?;
         }
 
-        let resident = writer
-            .finish(entries_meta.key_range.end, &timeline, &ctx)
-            .await?;
+        let (desc, path) = writer.finish(entries_meta.key_range.end, &ctx).await?;
+        let resident = Layer::finish_creating(harness.conf, &timeline, desc, &path)?;
 
         let inner = resident.get_as_delta(&ctx).await?;
 
@@ -2046,6 +1957,7 @@ pub(crate) mod test {
             .await
             .likely_resident_layers()
             .next()
+            .cloned()
             .unwrap();
 
         {
@@ -2120,7 +2032,8 @@ pub(crate) mod test {
             .read()
             .await
             .likely_resident_layers()
-            .find(|x| x != &initdb_layer)
+            .find(|&x| x != &initdb_layer)
+            .cloned()
             .unwrap();
 
         // create a copy for the timeline, so we don't overwrite the file
@@ -2155,7 +2068,8 @@ pub(crate) mod test {
                 .await
                 .unwrap();
 
-            let copied_layer = writer.finish(Key::MAX, &branch, ctx).await.unwrap();
+            let (desc, path) = writer.finish(Key::MAX, ctx).await.unwrap();
+            let copied_layer = Layer::finish_creating(tenant.conf, &branch, desc, &path).unwrap();
 
             copied_layer.get_as_delta(ctx).await.unwrap();
 
@@ -2283,7 +2197,9 @@ pub(crate) mod test {
         for (key, lsn, value) in deltas {
             writer.put_value(key, lsn, value, ctx).await?;
         }
-        let delta_layer = writer.finish(key_end, tline, ctx).await?;
+
+        let (desc, path) = writer.finish(key_end, ctx).await?;
+        let delta_layer = Layer::finish_creating(tenant.conf, tline, desc, &path)?;
 
         Ok::<_, anyhow::Error>(delta_layer)
     }

pageserver/src/tenant/storage_layer/image_layer.rs

@@ -32,9 +32,6 @@ use crate::tenant::block_io::{BlockBuf, BlockReader, FileBlockReader};
 use crate::tenant::disk_btree::{
     DiskBtreeBuilder, DiskBtreeIterator, DiskBtreeReader, VisitDirection,
 };
-use crate::tenant::storage_layer::{
-    LayerAccessStats, ValueReconstructResult, ValueReconstructState,
-};
 use crate::tenant::timeline::GetVectoredError;
 use crate::tenant::vectored_blob_io::{
     BlobFlag, MaxVectoredReadBytes, StreamingVectoredReadPlanner, VectoredBlobReader, VectoredRead,
@@ -137,7 +134,6 @@ pub struct ImageLayer {
     pub desc: PersistentLayerDesc,
     // This entry contains an image of all pages as of this LSN, should be the same as desc.lsn
     pub lsn: Lsn,
-    access_stats: LayerAccessStats,
     inner: OnceCell<ImageLayerInner>,
 }
 
@@ -255,7 +251,6 @@ impl ImageLayer {
     /// not loaded already.
     ///
     async fn load(&self, ctx: &RequestContext) -> Result<&ImageLayerInner> {
-        self.access_stats.record_access(ctx);
         self.inner
             .get_or_try_init(|| self.load_inner(ctx))
             .await
@@ -306,7 +301,6 @@ impl ImageLayer {
                 metadata.len(),
             ), // Now we assume image layer ALWAYS covers the full range. This may change in the future.
             lsn: summary.lsn,
-            access_stats: Default::default(),
             inner: OnceCell::new(),
         })
     }
@@ -375,9 +369,6 @@ impl ImageLayerInner {
         self.lsn
     }
 
-    /// Returns nested result following Result<Result<_, OpErr>, Critical>:
-    /// - inner has the success or transient failure
-    /// - outer has the permanent failure
     pub(super) async fn load(
         path: &Utf8Path,
         lsn: Lsn,
@@ -429,46 +420,6 @@ impl ImageLayerInner {
         })
     }
 
-    pub(super) async fn get_value_reconstruct_data(
-        &self,
-        key: Key,
-        reconstruct_state: &mut ValueReconstructState,
-        ctx: &RequestContext,
-    ) -> anyhow::Result<ValueReconstructResult> {
-        let block_reader = FileBlockReader::new(&self.file, self.file_id);
-        let tree_reader =
-            DiskBtreeReader::new(self.index_start_blk, self.index_root_blk, &block_reader);
-
-        let mut keybuf: [u8; KEY_SIZE] = [0u8; KEY_SIZE];
-        key.write_to_byte_slice(&mut keybuf);
-        if let Some(offset) = tree_reader
-            .get(
-                &keybuf,
-                &RequestContextBuilder::extend(ctx)
-                    .page_content_kind(PageContentKind::ImageLayerBtreeNode)
-                    .build(),
-            )
-            .await?
-        {
-            let blob = block_reader
-                .block_cursor()
-                .read_blob(
-                    offset,
-                    &RequestContextBuilder::extend(ctx)
-                        .page_content_kind(PageContentKind::ImageLayerValue)
-                        .build(),
-                )
-                .await
-                .with_context(|| format!("failed to read value from offset {}", offset))?;
-            let value = Bytes::from(blob);
-
-            reconstruct_state.img = Some((self.lsn, value));
-            Ok(ValueReconstructResult::Complete)
-        } else {
-            Ok(ValueReconstructResult::Missing)
-        }
-    }
-
     // Look up the keys in the provided keyspace and update
     // the reconstruct state with whatever is found.
     pub(super) async fn get_values_reconstruct_data(
@@ -742,11 +693,21 @@ struct ImageLayerWriterInner {
     // where we have chosen their compressed form
     uncompressed_bytes_chosen: u64,
 
+    // Number of keys in the layer.
+    num_keys: usize,
+
     blob_writer: BlobWriter<false>,
     tree: DiskBtreeBuilder<BlockBuf, KEY_SIZE>,
+
+    #[cfg_attr(not(feature = "testing"), allow(dead_code))]
+    last_written_key: Key,
 }
 
 impl ImageLayerWriterInner {
+    fn size(&self) -> u64 {
+        self.tree.borrow_writer().size() + self.blob_writer.size()
+    }
+
     ///
     /// Start building a new image layer.
     ///
@@ -800,6 +761,8 @@ impl ImageLayerWriterInner {
             uncompressed_bytes: 0,
             uncompressed_bytes_eligible: 0,
             uncompressed_bytes_chosen: 0,
+            num_keys: 0,
+            last_written_key: Key::MIN,
         };
 
         Ok(writer)
@@ -820,6 +783,7 @@ impl ImageLayerWriterInner {
         let compression = self.conf.image_compression;
         let uncompressed_len = img.len() as u64;
         self.uncompressed_bytes += uncompressed_len;
+        self.num_keys += 1;
         let (_img, res) = self
             .blob_writer
             .write_blob_maybe_compressed(img, ctx, compression)
@@ -839,6 +803,11 @@ impl ImageLayerWriterInner {
         key.write_to_byte_slice(&mut keybuf);
         self.tree.append(&keybuf, off)?;
 
+        #[cfg(feature = "testing")]
+        {
+            self.last_written_key = key;
+        }
+
         Ok(())
     }
 
@@ -849,6 +818,7 @@ impl ImageLayerWriterInner {
         self,
         timeline: &Arc<Timeline>,
         ctx: &RequestContext,
+        end_key: Option<Key>,
     ) -> anyhow::Result<ResidentLayer> {
         let index_start_blk =
             ((self.blob_writer.size() + PAGE_SZ as u64 - 1) / PAGE_SZ as u64) as u32;
@@ -899,11 +869,23 @@ impl ImageLayerWriterInner {
         let desc = PersistentLayerDesc::new_img(
             self.tenant_shard_id,
             self.timeline_id,
-            self.key_range.clone(),
+            if let Some(end_key) = end_key {
+                self.key_range.start..end_key
+            } else {
+                self.key_range.clone()
+            },
             self.lsn,
             metadata.len(),
         );
 
+        #[cfg(feature = "testing")]
+        if let Some(end_key) = end_key {
+            assert!(
+                self.last_written_key < end_key,
+                "written key violates end_key range"
+            );
+        }
+
         // Note: Because we open the file in write-only mode, we cannot
         // reuse the same VirtualFile for reading later. That's why we don't
         // set inner.file here. The first read will have to re-open it.
@@ -980,6 +962,18 @@ impl ImageLayerWriter {
         self.inner.as_mut().unwrap().put_image(key, img, ctx).await
     }
 
+    #[cfg(test)]
+    /// Estimated size of the image layer.
+    pub(crate) fn estimated_size(&self) -> u64 {
+        let inner = self.inner.as_ref().unwrap();
+        inner.blob_writer.size() + inner.tree.borrow_writer().size() + PAGE_SZ as u64
+    }
+
+    #[cfg(test)]
+    pub(crate) fn num_keys(&self) -> usize {
+        self.inner.as_ref().unwrap().num_keys
+    }
+
     ///
     /// Finish writing the image layer.
     ///
@@ -988,7 +982,26 @@ impl ImageLayerWriter {
         timeline: &Arc<Timeline>,
         ctx: &RequestContext,
     ) -> anyhow::Result<super::ResidentLayer> {
-        self.inner.take().unwrap().finish(timeline, ctx).await
+        self.inner.take().unwrap().finish(timeline, ctx, None).await
+    }
+
+    #[cfg(test)]
+    /// Finish writing the image layer with an end key, used in [`super::split_writer::SplitImageLayerWriter`]. The end key determines the end of the image layer's covered range and is exclusive.
+    pub(super) async fn finish_with_end_key(
+        mut self,
+        timeline: &Arc<Timeline>,
+        end_key: Key,
+        ctx: &RequestContext,
+    ) -> anyhow::Result<super::ResidentLayer> {
+        self.inner
+            .take()
+            .unwrap()
+            .finish(timeline, ctx, Some(end_key))
+            .await
+    }
+
+    pub(crate) fn size(&self) -> u64 {
+        self.inner.as_ref().unwrap().size()
     }
 }
 

pageserver/src/tenant/storage_layer/inmemory_layer.rs

@@ -10,11 +10,12 @@ use crate::page_cache::PAGE_SZ;
 use crate::repository::{Key, Value};
 use crate::tenant::block_io::{BlockCursor, BlockReader, BlockReaderRef};
 use crate::tenant::ephemeral_file::EphemeralFile;
-use crate::tenant::storage_layer::ValueReconstructResult;
 use crate::tenant::timeline::GetVectoredError;
-use crate::tenant::{PageReconstructError, Timeline};
+use crate::tenant::PageReconstructError;
 use crate::{l0_flush, page_cache, walrecord};
-use anyhow::{anyhow, ensure, Result};
+use anyhow::{anyhow, Result};
+use camino::Utf8PathBuf;
+use pageserver_api::key::CompactKey;
 use pageserver_api::keyspace::KeySpace;
 use pageserver_api::models::InMemoryLayerInfo;
 use pageserver_api::shard::TenantShardId;
@@ -34,8 +35,7 @@ use std::sync::atomic::{AtomicU64, AtomicUsize};
 use tokio::sync::{RwLock, RwLockWriteGuard};
 
 use super::{
-    DeltaLayerWriter, ResidentLayer, ValueReconstructSituation, ValueReconstructState,
-    ValuesReconstructState,
+    DeltaLayerWriter, PersistentLayerDesc, ValueReconstructSituation, ValuesReconstructState,
 };
 
 #[derive(Debug, PartialEq, Eq, Clone, Copy, Hash)]
@@ -55,9 +55,6 @@ pub struct InMemoryLayer {
     /// Writes are only allowed when this is `None`.
     pub(crate) end_lsn: OnceLock<Lsn>,
 
-    /// Used for traversal path. Cached representation of the in-memory layer before frozen.
-    local_path_str: Arc<str>,
-
     /// Used for traversal path. Cached representation of the in-memory layer after frozen.
     frozen_local_path_str: OnceLock<Arc<str>>,
 
@@ -82,7 +79,7 @@ pub struct InMemoryLayerInner {
     /// All versions of all pages in the layer are kept here. Indexed
     /// by block number and LSN. The value is an offset into the
     /// ephemeral file where the page version is stored.
-    index: BTreeMap<Key, VecMap<Lsn, u64>>,
+    index: BTreeMap<CompactKey, VecMap<Lsn, u64>>,
 
     /// The values are stored in a serialized format in this file.
     /// Each serialized Value is preceded by a 'u32' length field.
@@ -248,12 +245,6 @@ impl InMemoryLayer {
         self.start_lsn..self.end_lsn_or_max()
     }
 
-    pub(crate) fn local_path_str(&self) -> &Arc<str> {
-        self.frozen_local_path_str
-            .get()
-            .unwrap_or(&self.local_path_str)
-    }
-
     /// debugging function to print out the contents of the layer
     ///
     /// this is likely completly unused
@@ -303,60 +294,6 @@ impl InMemoryLayer {
         Ok(())
     }
 
-    /// Look up given value in the layer.
-    pub(crate) async fn get_value_reconstruct_data(
-        &self,
-        key: Key,
-        lsn_range: Range<Lsn>,
-        reconstruct_state: &mut ValueReconstructState,
-        ctx: &RequestContext,
-    ) -> anyhow::Result<ValueReconstructResult> {
-        ensure!(lsn_range.start >= self.start_lsn);
-        let mut need_image = true;
-
-        let ctx = RequestContextBuilder::extend(ctx)
-            .page_content_kind(PageContentKind::InMemoryLayer)
-            .build();
-
-        let inner = self.inner.read().await;
-
-        let reader = inner.file.block_cursor();
-
-        // Scan the page versions backwards, starting from `lsn`.
-        if let Some(vec_map) = inner.index.get(&key) {
-            let slice = vec_map.slice_range(lsn_range);
-            for (entry_lsn, pos) in slice.iter().rev() {
-                let buf = reader.read_blob(*pos, &ctx).await?;
-                let value = Value::des(&buf)?;
-                match value {
-                    Value::Image(img) => {
-                        reconstruct_state.img = Some((*entry_lsn, img));
-                        return Ok(ValueReconstructResult::Complete);
-                    }
-                    Value::WalRecord(rec) => {
-                        let will_init = rec.will_init();
-                        reconstruct_state.records.push((*entry_lsn, rec));
-                        if will_init {
-                            // This WAL record initializes the page, so no need to go further back
-                            need_image = false;
-                            break;
-                        }
-                    }
-                }
-            }
-        }
-
-        // release lock on 'inner'
-
-        // If an older page image is needed to reconstruct the page, let the
-        // caller know.
-        if need_image {
-            Ok(ValueReconstructResult::Continue)
-        } else {
-            Ok(ValueReconstructResult::Complete)
-        }
-    }
-
     // Look up the keys in the provided keyspace and update
     // the reconstruct state with whatever is found.
     //
@@ -376,8 +313,12 @@ impl InMemoryLayer {
         let reader = inner.file.block_cursor();
 
         for range in keyspace.ranges.iter() {
-            for (key, vec_map) in inner.index.range(range.start..range.end) {
-                let lsn_range = match reconstruct_state.get_cached_lsn(key) {
+            for (key, vec_map) in inner
+                .index
+                .range(range.start.to_compact()..range.end.to_compact())
+            {
+                let key = Key::from_compact(*key);
+                let lsn_range = match reconstruct_state.get_cached_lsn(&key) {
                     Some(cached_lsn) => (cached_lsn + 1)..end_lsn,
                     None => self.start_lsn..end_lsn,
                 };
@@ -388,20 +329,18 @@ impl InMemoryLayer {
                     // TODO: this uses the page cache => https://github.com/neondatabase/neon/issues/8183
                     let buf = reader.read_blob(*pos, &ctx).await;
                     if let Err(e) = buf {
-                        reconstruct_state
-                            .on_key_error(*key, PageReconstructError::from(anyhow!(e)));
+                        reconstruct_state.on_key_error(key, PageReconstructError::from(anyhow!(e)));
                         break;
                     }
 
                     let value = Value::des(&buf.unwrap());
                     if let Err(e) = value {
-                        reconstruct_state
-                            .on_key_error(*key, PageReconstructError::from(anyhow!(e)));
+                        reconstruct_state.on_key_error(key, PageReconstructError::from(anyhow!(e)));
                         break;
                     }
 
                     let key_situation =
-                        reconstruct_state.update_key(key, *entry_lsn, value.unwrap());
+                        reconstruct_state.update_key(&key, *entry_lsn, value.unwrap());
                     if key_situation == ValueReconstructSituation::Complete {
                         break;
                     }
@@ -449,20 +388,17 @@ impl InMemoryLayer {
         timeline_id: TimelineId,
         tenant_shard_id: TenantShardId,
         start_lsn: Lsn,
+        gate_guard: utils::sync::gate::GateGuard,
         ctx: &RequestContext,
     ) -> Result<InMemoryLayer> {
         trace!("initializing new empty InMemoryLayer for writing on timeline {timeline_id} at {start_lsn}");
 
-        let file = EphemeralFile::create(conf, tenant_shard_id, timeline_id, ctx).await?;
+        let file =
+            EphemeralFile::create(conf, tenant_shard_id, timeline_id, gate_guard, ctx).await?;
         let key = InMemoryLayerFileId(file.page_cache_file_id());
 
         Ok(InMemoryLayer {
             file_id: key,
-            local_path_str: {
-                let mut buf = String::new();
-                inmem_layer_log_display(&mut buf, timeline_id, start_lsn, Lsn::MAX).unwrap();
-                buf.into()
-            },
             frozen_local_path_str: OnceLock::new(),
             conf,
             timeline_id,
@@ -482,10 +418,9 @@ impl InMemoryLayer {
 
     /// Common subroutine of the public put_wal_record() and put_page_image() functions.
     /// Adds the page version to the in-memory tree
-
-    pub(crate) async fn put_value(
+    pub async fn put_value(
         &self,
-        key: Key,
+        key: CompactKey,
         lsn: Lsn,
         buf: &[u8],
         ctx: &RequestContext,
@@ -498,7 +433,7 @@ impl InMemoryLayer {
     async fn put_value_locked(
         &self,
         locked_inner: &mut RwLockWriteGuard<'_, InMemoryLayerInner>,
-        key: Key,
+        key: CompactKey,
         lsn: Lsn,
         buf: &[u8],
         ctx: &RequestContext,
@@ -548,8 +483,6 @@ impl InMemoryLayer {
     /// Records the end_lsn for non-dropped layers.
     /// `end_lsn` is exclusive
     pub async fn freeze(&self, end_lsn: Lsn) {
-        let inner = self.inner.write().await;
-
         assert!(
             self.start_lsn < end_lsn,
             "{} >= {}",
@@ -567,9 +500,13 @@ impl InMemoryLayer {
             })
             .expect("frozen_local_path_str set only once");
 
-        for vec_map in inner.index.values() {
-            for (lsn, _pos) in vec_map.as_slice() {
-                assert!(*lsn < end_lsn);
+        #[cfg(debug_assertions)]
+        {
+            let inner = self.inner.write().await;
+            for vec_map in inner.index.values() {
+                for (lsn, _pos) in vec_map.as_slice() {
+                    assert!(*lsn < end_lsn);
+                }
             }
         }
     }
@@ -579,12 +516,12 @@ impl InMemoryLayer {
     /// if there are no matching keys.
     ///
     /// Returns a new delta layer with all the same data as this in-memory layer
-    pub(crate) async fn write_to_disk(
+    pub async fn write_to_disk(
         &self,
-        timeline: &Arc<Timeline>,
         ctx: &RequestContext,
         key_range: Option<Range<Key>>,
-    ) -> Result<Option<ResidentLayer>> {
+        l0_flush_global_state: &l0_flush::Inner,
+    ) -> Result<Option<(PersistentLayerDesc, Utf8PathBuf)>> {
         // Grab the lock in read-mode. We hold it over the I/O, but because this
         // layer is not writeable anymore, no one should be trying to acquire the
         // write lock on it, so we shouldn't block anyone. There's one exception
@@ -596,9 +533,8 @@ impl InMemoryLayer {
         // rare though, so we just accept the potential latency hit for now.
         let inner = self.inner.read().await;
 
-        let l0_flush_global_state = timeline.l0_flush_global_state.inner().clone();
         use l0_flush::Inner;
-        let _concurrency_permit = match &*l0_flush_global_state {
+        let _concurrency_permit = match l0_flush_global_state {
             Inner::PageCached => None,
             Inner::Direct { semaphore, .. } => Some(semaphore.acquire().await),
         };
@@ -606,6 +542,8 @@ impl InMemoryLayer {
         let end_lsn = *self.end_lsn.get().unwrap();
 
         let key_count = if let Some(key_range) = key_range {
+            let key_range = key_range.start.to_compact()..key_range.end.to_compact();
+
             inner
                 .index
                 .iter()
@@ -628,7 +566,7 @@ impl InMemoryLayer {
         )
         .await?;
 
-        match &*l0_flush_global_state {
+        match l0_flush_global_state {
             l0_flush::Inner::PageCached => {
                 let ctx = RequestContextBuilder::extend(ctx)
                     .page_content_kind(PageContentKind::InMemoryLayer)
@@ -645,7 +583,7 @@ impl InMemoryLayer {
                         let will_init = Value::des(&buf)?.will_init();
                         let res;
                         (buf, res) = delta_layer_writer
-                            .put_value_bytes(*key, *lsn, buf, will_init, &ctx)
+                            .put_value_bytes(Key::from_compact(*key), *lsn, buf, will_init, &ctx)
                             .await;
                         res?;
                     }
@@ -684,7 +622,7 @@ impl InMemoryLayer {
                         let will_init = Value::des(&buf)?.will_init();
                         let res;
                         (buf, res) = delta_layer_writer
-                            .put_value_bytes(*key, *lsn, buf, will_init, ctx)
+                            .put_value_bytes(Key::from_compact(*key), *lsn, buf, will_init, ctx)
                             .await;
                         res?;
                     }
@@ -693,7 +631,7 @@ impl InMemoryLayer {
         }
 
         // MAX is used here because we identify L0 layers by full key range
-        let delta_layer = delta_layer_writer.finish(Key::MAX, timeline, ctx).await?;
+        let (desc, path) = delta_layer_writer.finish(Key::MAX, ctx).await?;
 
         // Hold the permit until all the IO is done, including the fsync in `delta_layer_writer.finish()``.
         //
@@ -705,6 +643,6 @@ impl InMemoryLayer {
         // we dirtied when writing to the filesystem have been flushed and marked !dirty.
         drop(_concurrency_permit);
 
-        Ok(Some(delta_layer))
+        Ok(Some((desc, path)))
     }
 }

pageserver/src/tenant/storage_layer/layer.rs

@@ -24,7 +24,7 @@ use super::delta_layer::{self, DeltaEntry};
 use super::image_layer::{self};
 use super::{
     AsLayerDesc, ImageLayerWriter, LayerAccessStats, LayerAccessStatsReset, LayerName,
-    PersistentLayerDesc, ValueReconstructResult, ValueReconstructState, ValuesReconstructState,
+    LayerVisibilityHint, PersistentLayerDesc, ValuesReconstructState,
 };
 
 use utils::generation::Generation;
@@ -246,7 +246,7 @@ impl Layer {
                 &timeline.generation,
             );
 
-            let layer = LayerInner::new(
+            LayerInner::new(
                 conf,
                 timeline,
                 local_path,
@@ -254,14 +254,7 @@ impl Layer {
                 Some(inner),
                 timeline.generation,
                 timeline.get_shard_index(),
-            );
-
-            // Newly created layers are marked visible by default: the usual case is that they were created to be read.
-            layer
-                .access_stats
-                .set_visibility(super::LayerVisibilityHint::Visible);
-
-            layer
+            )
         }));
 
         let downloaded = resident.expect("just initialized");
@@ -307,42 +300,6 @@ impl Layer {
         self.0.delete_on_drop();
     }
 
-    /// Return data needed to reconstruct given page at LSN.
-    ///
-    /// It is up to the caller to collect more data from the previous layer and
-    /// perform WAL redo, if necessary.
-    ///
-    /// # Cancellation-Safety
-    ///
-    /// This method is cancellation-safe.
-    pub(crate) async fn get_value_reconstruct_data(
-        &self,
-        key: Key,
-        lsn_range: Range<Lsn>,
-        reconstruct_data: &mut ValueReconstructState,
-        ctx: &RequestContext,
-    ) -> anyhow::Result<ValueReconstructResult> {
-        use anyhow::ensure;
-
-        let layer = self.0.get_or_maybe_download(true, Some(ctx)).await?;
-        self.0.access_stats.record_access(ctx);
-
-        if self.layer_desc().is_delta {
-            ensure!(lsn_range.start >= self.layer_desc().lsn_range.start);
-            ensure!(self.layer_desc().key_range.contains(&key));
-        } else {
-            ensure!(self.layer_desc().key_range.contains(&key));
-            ensure!(lsn_range.start >= self.layer_desc().image_layer_lsn());
-            ensure!(lsn_range.end >= self.layer_desc().image_layer_lsn());
-        }
-
-        layer
-            .get_value_reconstruct_data(key, lsn_range, reconstruct_data, &self.0, ctx)
-            .instrument(tracing::debug_span!("get_value_reconstruct_data", layer=%self))
-            .await
-            .with_context(|| format!("get_value_reconstruct_data for layer {self}"))
-    }
-
     pub(crate) async fn get_values_reconstruct_data(
         &self,
         keyspace: KeySpace,
@@ -359,7 +316,7 @@ impl Layer {
                 other => GetVectoredError::Other(anyhow::anyhow!(other)),
             })?;
 
-        self.0.access_stats.record_access(ctx);
+        self.record_access(ctx);
 
         layer
             .get_values_reconstruct_data(keyspace, lsn_range, reconstruct_data, &self.0, ctx)
@@ -439,18 +396,18 @@ impl Layer {
         self.0.info(reset)
     }
 
-    pub(crate) fn access_stats(&self) -> &LayerAccessStats {
-        &self.0.access_stats
+    pub(crate) fn latest_activity(&self) -> SystemTime {
+        self.0.access_stats.latest_activity()
+    }
+
+    pub(crate) fn visibility(&self) -> LayerVisibilityHint {
+        self.0.access_stats.visibility()
     }
 
     pub(crate) fn local_path(&self) -> &Utf8Path {
         &self.0.path
     }
 
-    pub(crate) fn debug_str(&self) -> &Arc<str> {
-        &self.0.debug_str
-    }
-
     pub(crate) fn metadata(&self) -> LayerFileMetadata {
         self.0.metadata()
     }
@@ -493,13 +450,57 @@ impl Layer {
             }
         }
     }
+
+    fn record_access(&self, ctx: &RequestContext) {
+        if self.0.access_stats.record_access(ctx) {
+            // Visibility was modified to Visible
+            tracing::info!(
+                "Layer {} became visible as a result of access",
+                self.0.desc.key()
+            );
+            if let Some(tl) = self.0.timeline.upgrade() {
+                tl.metrics
+                    .visible_physical_size_gauge
+                    .add(self.0.desc.file_size)
+            }
+        }
+    }
+
+    pub(crate) fn set_visibility(&self, visibility: LayerVisibilityHint) {
+        let old_visibility = self.0.access_stats.set_visibility(visibility.clone());
+        use LayerVisibilityHint::*;
+        match (old_visibility, visibility) {
+            (Visible, Covered) => {
+                // Subtract this layer's contribution to the visible size metric
+                if let Some(tl) = self.0.timeline.upgrade() {
+                    debug_assert!(
+                        tl.metrics.visible_physical_size_gauge.get() >= self.0.desc.file_size
+                    );
+                    tl.metrics
+                        .visible_physical_size_gauge
+                        .sub(self.0.desc.file_size)
+                }
+            }
+            (Covered, Visible) => {
+                // Add this layer's contribution to the visible size metric
+                if let Some(tl) = self.0.timeline.upgrade() {
+                    tl.metrics
+                        .visible_physical_size_gauge
+                        .add(self.0.desc.file_size)
+                }
+            }
+            (Covered, Covered) | (Visible, Visible) => {
+                // no change
+            }
+        }
+    }
 }
 
 /// The download-ness ([`DownloadedLayer`]) can be either resident or wanted evicted.
 ///
 /// However when we want something evicted, we cannot evict it right away as there might be current
 /// reads happening on it. For example: it has been searched from [`LayerMap::search`] but not yet
-/// read with [`Layer::get_value_reconstruct_data`].
+/// read with [`Layer::get_values_reconstruct_data`].
 ///
 /// [`LayerMap::search`]: crate::tenant::layer_map::LayerMap::search
 #[derive(Debug)]
@@ -580,9 +581,6 @@ struct LayerInner {
     /// Full path to the file; unclear if this should exist anymore.
     path: Utf8PathBuf,
 
-    /// String representation of the layer, used for traversal id.
-    debug_str: Arc<str>,
-
     desc: PersistentLayerDesc,
 
     /// Timeline access is needed for remote timeline client and metrics.
@@ -693,6 +691,16 @@ impl Drop for LayerInner {
                 timeline.metrics.layer_count_image.dec();
                 timeline.metrics.layer_size_image.sub(self.desc.file_size);
             }
+
+            if matches!(self.access_stats.visibility(), LayerVisibilityHint::Visible) {
+                debug_assert!(
+                    timeline.metrics.visible_physical_size_gauge.get() >= self.desc.file_size
+                );
+                timeline
+                    .metrics
+                    .visible_physical_size_gauge
+                    .sub(self.desc.file_size);
+            }
         }
 
         if !*self.wanted_deleted.get_mut() {
@@ -801,11 +809,14 @@ impl LayerInner {
             timeline.metrics.layer_size_image.add(desc.file_size);
         }
 
+        // New layers are visible by default. This metric is later updated on drop or in set_visibility
+        timeline
+            .metrics
+            .visible_physical_size_gauge
+            .add(desc.file_size);
+
         LayerInner {
             conf,
-            debug_str: {
-                format!("timelines/{}/{}", timeline.timeline_id, desc.layer_name()).into()
-            },
             path: local_path,
             desc,
             timeline: Arc::downgrade(timeline),
@@ -1726,28 +1737,6 @@ impl DownloadedLayer {
             .map_err(|e| anyhow::anyhow!("layer load failed earlier: {e}"))
     }
 
-    async fn get_value_reconstruct_data(
-        &self,
-        key: Key,
-        lsn_range: Range<Lsn>,
-        reconstruct_data: &mut ValueReconstructState,
-        owner: &Arc<LayerInner>,
-        ctx: &RequestContext,
-    ) -> anyhow::Result<ValueReconstructResult> {
-        use LayerKind::*;
-
-        match self.get(owner, ctx).await? {
-            Delta(d) => {
-                d.get_value_reconstruct_data(key, lsn_range, reconstruct_data, ctx)
-                    .await
-            }
-            Image(i) => {
-                i.get_value_reconstruct_data(key, reconstruct_data, ctx)
-                    .await
-            }
-        }
-    }
-
     async fn get_values_reconstruct_data(
         &self,
         keyspace: KeySpace,
@@ -1846,7 +1835,7 @@ impl ResidentLayer {
                 // this is valid because the DownloadedLayer::kind is a OnceCell, not a
                 // Mutex<OnceCell>, so we cannot go and deinitialize the value with OnceCell::take
                 // while it's being held.
-                owner.access_stats.record_access(ctx);
+                self.owner.record_access(ctx);
 
                 delta_layer::DeltaLayerInner::load_keys(d, ctx)
                     .await
@@ -1859,8 +1848,8 @@ impl ResidentLayer {
     /// Read all they keys in this layer which match the ShardIdentity, and write them all to
     /// the provided writer.  Return the number of keys written.
     #[tracing::instrument(level = tracing::Level::DEBUG, skip_all, fields(layer=%self))]
-    pub(crate) async fn filter<'a>(
-        &'a self,
+    pub(crate) async fn filter(
+        &self,
         shard_identity: &ShardIdentity,
         writer: &mut ImageLayerWriter,
         ctx: &RequestContext,

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

@@ -39,7 +39,7 @@ async fn smoke_test() {
     let layer = {
         let mut layers = {
             let layers = timeline.layers.read().await;
-            layers.likely_resident_layers().collect::<Vec<_>>()
+            layers.likely_resident_layers().cloned().collect::<Vec<_>>()
         };
 
         assert_eq!(layers.len(), 1);
@@ -50,13 +50,26 @@ async fn smoke_test() {
     // all layers created at pageserver are like `layer`, initialized with strong
     // Arc<DownloadedLayer>.
 
+    let controlfile_keyspace = KeySpace {
+        ranges: vec![CONTROLFILE_KEY..CONTROLFILE_KEY.next()],
+    };
+
     let img_before = {
-        let mut data = ValueReconstructState::default();
+        let mut data = ValuesReconstructState::default();
         layer
-            .get_value_reconstruct_data(CONTROLFILE_KEY, Lsn(0x10)..Lsn(0x11), &mut data, &ctx)
+            .get_values_reconstruct_data(
+                controlfile_keyspace.clone(),
+                Lsn(0x10)..Lsn(0x11),
+                &mut data,
+                &ctx,
+            )
             .await
             .unwrap();
-        data.img
+        data.keys
+            .remove(&CONTROLFILE_KEY)
+            .expect("must be present")
+            .expect("should not error")
+            .img
             .take()
             .expect("tenant harness writes the control file")
     };
@@ -74,13 +87,24 @@ async fn smoke_test() {
 
     // on accesses when the layer is evicted, it will automatically be downloaded.
     let img_after = {
-        let mut data = ValueReconstructState::default();
+        let mut data = ValuesReconstructState::default();
         layer
-            .get_value_reconstruct_data(CONTROLFILE_KEY, Lsn(0x10)..Lsn(0x11), &mut data, &ctx)
+            .get_values_reconstruct_data(
+                controlfile_keyspace.clone(),
+                Lsn(0x10)..Lsn(0x11),
+                &mut data,
+                &ctx,
+            )
             .instrument(download_span.clone())
             .await
             .unwrap();
-        data.img.take().unwrap()
+        data.keys
+            .remove(&CONTROLFILE_KEY)
+            .expect("must be present")
+            .expect("should not error")
+            .img
+            .take()
+            .expect("tenant harness writes the control file")
     };
 
     assert_eq!(img_before, img_after);
@@ -152,7 +176,7 @@ async fn smoke_test() {
     {
         let layers = &[layer];
         let mut g = timeline.layers.write().await;
-        g.finish_gc_timeline(layers);
+        g.open_mut().unwrap().finish_gc_timeline(layers);
         // this just updates the remote_physical_size for demonstration purposes
         rtc.schedule_gc_update(layers).unwrap();
     }
@@ -192,7 +216,7 @@ async fn evict_and_wait_on_wanted_deleted() {
     let layer = {
         let mut layers = {
             let layers = timeline.layers.read().await;
-            layers.likely_resident_layers().collect::<Vec<_>>()
+            layers.likely_resident_layers().cloned().collect::<Vec<_>>()
         };
 
         assert_eq!(layers.len(), 1);
@@ -236,7 +260,7 @@ async fn evict_and_wait_on_wanted_deleted() {
     // the deletion of the layer in remote_storage happens.
     {
         let mut layers = timeline.layers.write().await;
-        layers.finish_gc_timeline(&[layer]);
+        layers.open_mut().unwrap().finish_gc_timeline(&[layer]);
     }
 
     SpawnBlockingPoolHelper::consume_and_release_all_of_spawn_blocking_threads(&handle).await;
@@ -277,7 +301,7 @@ fn read_wins_pending_eviction() {
         let layer = {
             let mut layers = {
                 let layers = timeline.layers.read().await;
-                layers.likely_resident_layers().collect::<Vec<_>>()
+                layers.likely_resident_layers().cloned().collect::<Vec<_>>()
             };
 
             assert_eq!(layers.len(), 1);
@@ -409,7 +433,7 @@ fn multiple_pending_evictions_scenario(name: &'static str, in_order: bool) {
         let layer = {
             let mut layers = {
                 let layers = timeline.layers.read().await;
-                layers.likely_resident_layers().collect::<Vec<_>>()
+                layers.likely_resident_layers().cloned().collect::<Vec<_>>()
             };
 
             assert_eq!(layers.len(), 1);
@@ -578,7 +602,7 @@ async fn cancelled_get_or_maybe_download_does_not_cancel_eviction() {
     let layer = {
         let mut layers = {
             let layers = timeline.layers.read().await;
-            layers.likely_resident_layers().collect::<Vec<_>>()
+            layers.likely_resident_layers().cloned().collect::<Vec<_>>()
         };
 
         assert_eq!(layers.len(), 1);
@@ -658,7 +682,7 @@ async fn evict_and_wait_does_not_wait_for_download() {
     let layer = {
         let mut layers = {
             let layers = timeline.layers.read().await;
-            layers.likely_resident_layers().collect::<Vec<_>>()
+            layers.likely_resident_layers().cloned().collect::<Vec<_>>()
         };
 
         assert_eq!(layers.len(), 1);
@@ -777,9 +801,9 @@ async fn eviction_cancellation_on_drop() {
     let (evicted_layer, not_evicted) = {
         let mut layers = {
             let mut guard = timeline.layers.write().await;
-            let layers = guard.likely_resident_layers().collect::<Vec<_>>();
+            let layers = guard.likely_resident_layers().cloned().collect::<Vec<_>>();
             // remove the layers from layermap
-            guard.finish_gc_timeline(&layers);
+            guard.open_mut().unwrap().finish_gc_timeline(&layers);
 
             layers
         };
@@ -830,7 +854,7 @@ async fn eviction_cancellation_on_drop() {
 fn layer_size() {
     assert_eq!(size_of::<LayerAccessStats>(), 8);
     assert_eq!(size_of::<PersistentLayerDesc>(), 104);
-    assert_eq!(size_of::<LayerInner>(), 312);
+    assert_eq!(size_of::<LayerInner>(), 296);
     // it also has the utf8 path
 }
 

pageserver/src/tenant/storage_layer/layer_desc.rs

@@ -41,6 +41,20 @@ pub struct PersistentLayerKey {
     pub is_delta: bool,
 }
 
+impl std::fmt::Display for PersistentLayerKey {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(
+            f,
+            "{}..{} {}..{} is_delta={}",
+            self.key_range.start,
+            self.key_range.end,
+            self.lsn_range.start,
+            self.lsn_range.end,
+            self.is_delta
+        )
+    }
+}
+
 impl PersistentLayerDesc {
     pub fn key(&self) -> PersistentLayerKey {
         PersistentLayerKey {

pageserver/src/tenant/storage_layer/split_writer.rs

@@ -0,0 +1,454 @@
+use std::{ops::Range, sync::Arc};
+
+use bytes::Bytes;
+use pageserver_api::key::{Key, KEY_SIZE};
+use utils::{id::TimelineId, lsn::Lsn, shard::TenantShardId};
+
+use crate::tenant::storage_layer::Layer;
+use crate::{config::PageServerConf, context::RequestContext, repository::Value, tenant::Timeline};
+
+use super::{DeltaLayerWriter, ImageLayerWriter, ResidentLayer};
+
+/// An image writer that takes images and produces multiple image layers. The interface does not
+/// guarantee atomicity (i.e., if the image layer generation fails, there might be leftover files
+/// to be cleaned up)
+#[must_use]
+pub struct SplitImageLayerWriter {
+    inner: ImageLayerWriter,
+    target_layer_size: u64,
+    generated_layers: Vec<ResidentLayer>,
+    conf: &'static PageServerConf,
+    timeline_id: TimelineId,
+    tenant_shard_id: TenantShardId,
+    lsn: Lsn,
+}
+
+impl SplitImageLayerWriter {
+    pub async fn new(
+        conf: &'static PageServerConf,
+        timeline_id: TimelineId,
+        tenant_shard_id: TenantShardId,
+        start_key: Key,
+        lsn: Lsn,
+        target_layer_size: u64,
+        ctx: &RequestContext,
+    ) -> anyhow::Result<Self> {
+        Ok(Self {
+            target_layer_size,
+            inner: ImageLayerWriter::new(
+                conf,
+                timeline_id,
+                tenant_shard_id,
+                &(start_key..Key::MAX),
+                lsn,
+                ctx,
+            )
+            .await?,
+            generated_layers: Vec::new(),
+            conf,
+            timeline_id,
+            tenant_shard_id,
+            lsn,
+        })
+    }
+
+    pub async fn put_image(
+        &mut self,
+        key: Key,
+        img: Bytes,
+        tline: &Arc<Timeline>,
+        ctx: &RequestContext,
+    ) -> anyhow::Result<()> {
+        // The current estimation is an upper bound of the space that the key/image could take
+        // because we did not consider compression in this estimation. The resulting image layer
+        // could be smaller than the target size.
+        let addition_size_estimation = KEY_SIZE as u64 + img.len() as u64;
+        if self.inner.num_keys() >= 1
+            && self.inner.estimated_size() + addition_size_estimation >= self.target_layer_size
+        {
+            let next_image_writer = ImageLayerWriter::new(
+                self.conf,
+                self.timeline_id,
+                self.tenant_shard_id,
+                &(key..Key::MAX),
+                self.lsn,
+                ctx,
+            )
+            .await?;
+            let prev_image_writer = std::mem::replace(&mut self.inner, next_image_writer);
+            self.generated_layers.push(
+                prev_image_writer
+                    .finish_with_end_key(tline, key, ctx)
+                    .await?,
+            );
+        }
+        self.inner.put_image(key, img, ctx).await
+    }
+
+    pub(crate) async fn finish(
+        self,
+        tline: &Arc<Timeline>,
+        ctx: &RequestContext,
+        end_key: Key,
+    ) -> anyhow::Result<Vec<ResidentLayer>> {
+        let Self {
+            mut generated_layers,
+            inner,
+            ..
+        } = self;
+        generated_layers.push(inner.finish_with_end_key(tline, end_key, ctx).await?);
+        Ok(generated_layers)
+    }
+
+    /// When split writer fails, the caller should call this function and handle partially generated layers.
+    #[allow(dead_code)]
+    pub(crate) async fn take(self) -> anyhow::Result<(Vec<ResidentLayer>, ImageLayerWriter)> {
+        Ok((self.generated_layers, self.inner))
+    }
+}
+
+/// A delta writer that takes key-lsn-values and produces multiple delta layers. The interface does not
+/// guarantee atomicity (i.e., if the delta layer generation fails, there might be leftover files
+/// to be cleaned up).
+#[must_use]
+pub struct SplitDeltaLayerWriter {
+    inner: DeltaLayerWriter,
+    target_layer_size: u64,
+    generated_layers: Vec<ResidentLayer>,
+    conf: &'static PageServerConf,
+    timeline_id: TimelineId,
+    tenant_shard_id: TenantShardId,
+    lsn_range: Range<Lsn>,
+}
+
+impl SplitDeltaLayerWriter {
+    pub async fn new(
+        conf: &'static PageServerConf,
+        timeline_id: TimelineId,
+        tenant_shard_id: TenantShardId,
+        start_key: Key,
+        lsn_range: Range<Lsn>,
+        target_layer_size: u64,
+        ctx: &RequestContext,
+    ) -> anyhow::Result<Self> {
+        Ok(Self {
+            target_layer_size,
+            inner: DeltaLayerWriter::new(
+                conf,
+                timeline_id,
+                tenant_shard_id,
+                start_key,
+                lsn_range.clone(),
+                ctx,
+            )
+            .await?,
+            generated_layers: Vec::new(),
+            conf,
+            timeline_id,
+            tenant_shard_id,
+            lsn_range,
+        })
+    }
+
+    pub async fn put_value(
+        &mut self,
+        key: Key,
+        lsn: Lsn,
+        val: Value,
+        tline: &Arc<Timeline>,
+        ctx: &RequestContext,
+    ) -> anyhow::Result<()> {
+        // The current estimation is key size plus LSN size plus value size estimation. This is not an accurate
+        // number, and therefore the final layer size could be a little bit larger or smaller than the target.
+        let addition_size_estimation = KEY_SIZE as u64 + 8 /* LSN u64 size */ + 80 /* value size estimation */;
+        if self.inner.num_keys() >= 1
+            && self.inner.estimated_size() + addition_size_estimation >= self.target_layer_size
+        {
+            let next_delta_writer = DeltaLayerWriter::new(
+                self.conf,
+                self.timeline_id,
+                self.tenant_shard_id,
+                key,
+                self.lsn_range.clone(),
+                ctx,
+            )
+            .await?;
+            let prev_delta_writer = std::mem::replace(&mut self.inner, next_delta_writer);
+            let (desc, path) = prev_delta_writer.finish(key, ctx).await?;
+            let delta_layer = Layer::finish_creating(self.conf, tline, desc, &path)?;
+            self.generated_layers.push(delta_layer);
+        }
+        self.inner.put_value(key, lsn, val, ctx).await
+    }
+
+    pub(crate) async fn finish(
+        self,
+        tline: &Arc<Timeline>,
+        ctx: &RequestContext,
+        end_key: Key,
+    ) -> anyhow::Result<Vec<ResidentLayer>> {
+        let Self {
+            mut generated_layers,
+            inner,
+            ..
+        } = self;
+
+        let (desc, path) = inner.finish(end_key, ctx).await?;
+        let delta_layer = Layer::finish_creating(self.conf, tline, desc, &path)?;
+        generated_layers.push(delta_layer);
+        Ok(generated_layers)
+    }
+
+    /// When split writer fails, the caller should call this function and handle partially generated layers.
+    #[allow(dead_code)]
+    pub(crate) async fn take(self) -> anyhow::Result<(Vec<ResidentLayer>, DeltaLayerWriter)> {
+        Ok((self.generated_layers, self.inner))
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use crate::{
+        tenant::{
+            harness::{TenantHarness, TIMELINE_ID},
+            storage_layer::AsLayerDesc,
+        },
+        DEFAULT_PG_VERSION,
+    };
+
+    use super::*;
+
+    fn get_key(id: u32) -> Key {
+        let mut key = Key::from_hex("000000000033333333444444445500000000").unwrap();
+        key.field6 = id;
+        key
+    }
+
+    fn get_img(id: u32) -> Bytes {
+        format!("{id:064}").into()
+    }
+
+    fn get_large_img() -> Bytes {
+        vec![0; 8192].into()
+    }
+
+    #[tokio::test]
+    async fn write_one_image() {
+        let harness = TenantHarness::create("split_writer_write_one_image")
+            .await
+            .unwrap();
+        let (tenant, ctx) = harness.load().await;
+
+        let tline = tenant
+            .create_test_timeline(TIMELINE_ID, Lsn(0x10), DEFAULT_PG_VERSION, &ctx)
+            .await
+            .unwrap();
+
+        let mut image_writer = SplitImageLayerWriter::new(
+            tenant.conf,
+            tline.timeline_id,
+            tenant.tenant_shard_id,
+            get_key(0),
+            Lsn(0x18),
+            4 * 1024 * 1024,
+            &ctx,
+        )
+        .await
+        .unwrap();
+
+        let mut delta_writer = SplitDeltaLayerWriter::new(
+            tenant.conf,
+            tline.timeline_id,
+            tenant.tenant_shard_id,
+            get_key(0),
+            Lsn(0x18)..Lsn(0x20),
+            4 * 1024 * 1024,
+            &ctx,
+        )
+        .await
+        .unwrap();
+
+        image_writer
+            .put_image(get_key(0), get_img(0), &tline, &ctx)
+            .await
+            .unwrap();
+        let layers = image_writer
+            .finish(&tline, &ctx, get_key(10))
+            .await
+            .unwrap();
+        assert_eq!(layers.len(), 1);
+
+        delta_writer
+            .put_value(
+                get_key(0),
+                Lsn(0x18),
+                Value::Image(get_img(0)),
+                &tline,
+                &ctx,
+            )
+            .await
+            .unwrap();
+        let layers = delta_writer
+            .finish(&tline, &ctx, get_key(10))
+            .await
+            .unwrap();
+        assert_eq!(layers.len(), 1);
+    }
+
+    #[tokio::test]
+    async fn write_split() {
+        let harness = TenantHarness::create("split_writer_write_split")
+            .await
+            .unwrap();
+        let (tenant, ctx) = harness.load().await;
+
+        let tline = tenant
+            .create_test_timeline(TIMELINE_ID, Lsn(0x10), DEFAULT_PG_VERSION, &ctx)
+            .await
+            .unwrap();
+
+        let mut image_writer = SplitImageLayerWriter::new(
+            tenant.conf,
+            tline.timeline_id,
+            tenant.tenant_shard_id,
+            get_key(0),
+            Lsn(0x18),
+            4 * 1024 * 1024,
+            &ctx,
+        )
+        .await
+        .unwrap();
+        let mut delta_writer = SplitDeltaLayerWriter::new(
+            tenant.conf,
+            tline.timeline_id,
+            tenant.tenant_shard_id,
+            get_key(0),
+            Lsn(0x18)..Lsn(0x20),
+            4 * 1024 * 1024,
+            &ctx,
+        )
+        .await
+        .unwrap();
+        const N: usize = 2000;
+        for i in 0..N {
+            let i = i as u32;
+            image_writer
+                .put_image(get_key(i), get_large_img(), &tline, &ctx)
+                .await
+                .unwrap();
+            delta_writer
+                .put_value(
+                    get_key(i),
+                    Lsn(0x20),
+                    Value::Image(get_large_img()),
+                    &tline,
+                    &ctx,
+                )
+                .await
+                .unwrap();
+        }
+        let image_layers = image_writer
+            .finish(&tline, &ctx, get_key(N as u32))
+            .await
+            .unwrap();
+        let delta_layers = delta_writer
+            .finish(&tline, &ctx, get_key(N as u32))
+            .await
+            .unwrap();
+        assert_eq!(image_layers.len(), N / 512 + 1);
+        assert_eq!(delta_layers.len(), N / 512 + 1);
+        for idx in 0..image_layers.len() {
+            assert_ne!(image_layers[idx].layer_desc().key_range.start, Key::MIN);
+            assert_ne!(image_layers[idx].layer_desc().key_range.end, Key::MAX);
+            assert_ne!(delta_layers[idx].layer_desc().key_range.start, Key::MIN);
+            assert_ne!(delta_layers[idx].layer_desc().key_range.end, Key::MAX);
+            if idx > 0 {
+                assert_eq!(
+                    image_layers[idx - 1].layer_desc().key_range.end,
+                    image_layers[idx].layer_desc().key_range.start
+                );
+                assert_eq!(
+                    delta_layers[idx - 1].layer_desc().key_range.end,
+                    delta_layers[idx].layer_desc().key_range.start
+                );
+            }
+        }
+    }
+
+    #[tokio::test]
+    async fn write_large_img() {
+        let harness = TenantHarness::create("split_writer_write_large_img")
+            .await
+            .unwrap();
+        let (tenant, ctx) = harness.load().await;
+
+        let tline = tenant
+            .create_test_timeline(TIMELINE_ID, Lsn(0x10), DEFAULT_PG_VERSION, &ctx)
+            .await
+            .unwrap();
+
+        let mut image_writer = SplitImageLayerWriter::new(
+            tenant.conf,
+            tline.timeline_id,
+            tenant.tenant_shard_id,
+            get_key(0),
+            Lsn(0x18),
+            4 * 1024,
+            &ctx,
+        )
+        .await
+        .unwrap();
+
+        let mut delta_writer = SplitDeltaLayerWriter::new(
+            tenant.conf,
+            tline.timeline_id,
+            tenant.tenant_shard_id,
+            get_key(0),
+            Lsn(0x18)..Lsn(0x20),
+            4 * 1024,
+            &ctx,
+        )
+        .await
+        .unwrap();
+
+        image_writer
+            .put_image(get_key(0), get_img(0), &tline, &ctx)
+            .await
+            .unwrap();
+        image_writer
+            .put_image(get_key(1), get_large_img(), &tline, &ctx)
+            .await
+            .unwrap();
+        let layers = image_writer
+            .finish(&tline, &ctx, get_key(10))
+            .await
+            .unwrap();
+        assert_eq!(layers.len(), 2);
+
+        delta_writer
+            .put_value(
+                get_key(0),
+                Lsn(0x18),
+                Value::Image(get_img(0)),
+                &tline,
+                &ctx,
+            )
+            .await
+            .unwrap();
+        delta_writer
+            .put_value(
+                get_key(1),
+                Lsn(0x1A),
+                Value::Image(get_large_img()),
+                &tline,
+                &ctx,
+            )
+            .await
+            .unwrap();
+        let layers = delta_writer
+            .finish(&tline, &ctx, get_key(10))
+            .await
+            .unwrap();
+        assert_eq!(layers.len(), 2);
+    }
+}

pageserver/src/tenant/tasks.rs

@@ -211,6 +211,11 @@ async fn compaction_loop(tenant: Arc<Tenant>, cancel: CancellationToken) {
             } else {
                 // Run compaction
                 match tenant.compaction_iteration(&cancel, &ctx).await {
+                    Ok(has_pending_task) => {
+                        error_run_count = 0;
+                        // schedule the next compaction immediately in case there is a pending compaction task
+                        if has_pending_task { Duration::ZERO } else { period }
+                    }
                     Err(e) => {
                         let wait_duration = backoff::exponential_backoff_duration_seconds(
                             error_run_count + 1,
@@ -227,11 +232,6 @@ async fn compaction_loop(tenant: Arc<Tenant>, cancel: CancellationToken) {
                         );
                         wait_duration
                     }
-                    Ok(has_pending_task) => {
-                        error_run_count = 0;
-                        // schedule the next compaction immediately in case there is a pending compaction task
-                        if has_pending_task { Duration::from_secs(0) } else { period }
-                    }
                 }
             };
 
@@ -265,7 +265,8 @@ async fn compaction_loop(tenant: Arc<Tenant>, cancel: CancellationToken) {
                     count_throttled,
                     sum_throttled_usecs,
                     allowed_rps=%format_args!("{allowed_rps:.0}"),
-                    "shard was throttled in the last n_seconds")
+                    "shard was throttled in the last n_seconds"
+                );
             });
 
             // Sleep
@@ -365,14 +366,13 @@ async fn gc_loop(tenant: Arc<Tenant>, cancel: CancellationToken) {
             if first {
                 first = false;
 
-                if delay_by_lease_length(tenant.get_lsn_lease_length(), &cancel)
-                    .await
-                    .is_err()
-                {
-                    break;
-                }
+                let delays = async {
+                    delay_by_lease_length(tenant.get_lsn_lease_length(), &cancel).await?;
+                    random_init_delay(period, &cancel).await?;
+                    Ok::<_, Cancelled>(())
+                };
 
-                if random_init_delay(period, &cancel).await.is_err() {
+                if delays.await.is_err() {
                     break;
                 }
             }
@@ -407,9 +407,16 @@ async fn gc_loop(tenant: Arc<Tenant>, cancel: CancellationToken) {
                         error_run_count += 1;
                         let wait_duration = Duration::from_secs_f64(wait_duration);
 
-                        error!(
-                        "Gc failed {error_run_count} times, retrying in {wait_duration:?}: {e:?}",
-                    );
+                        if matches!(e, crate::tenant::GcError::TimelineCancelled) {
+                            // Timeline was cancelled during gc. We might either be in an event
+                            // that affects the entire tenant (tenant deletion, pageserver shutdown),
+                            // or in one that affects the timeline only (timeline deletion).
+                            // Therefore, don't exit the loop.
+                            info!("Gc failed {error_run_count} times, retrying in {wait_duration:?}: {e:?}");
+                        } else {
+                            error!("Gc failed {error_run_count} times, retrying in {wait_duration:?}: {e:?}");
+                        }
+
                         wait_duration
                     }
                 }
@@ -417,7 +424,6 @@ async fn gc_loop(tenant: Arc<Tenant>, cancel: CancellationToken) {
 
             warn_when_period_overrun(started_at.elapsed(), period, BackgroundLoopKind::Gc);
 
-            // Sleep
             if tokio::time::timeout(sleep_duration, cancel.cancelled())
                 .await
                 .is_ok()

pageserver/src/tenant/timeline/delete.rs

@@ -63,10 +63,19 @@ pub(super) async fn delete_local_timeline_directory(
     tenant_shard_id: TenantShardId,
     timeline: &Timeline,
 ) -> anyhow::Result<()> {
-    let guards = async { tokio::join!(timeline.gc_lock.lock(), timeline.compaction_lock.lock()) };
-    let guards = crate::timed(
-        guards,
-        "acquire gc and compaction locks",
+    // Always ensure the lock order is compaction -> gc.
+    let compaction_lock = timeline.compaction_lock.lock();
+    let compaction_lock = crate::timed(
+        compaction_lock,
+        "acquires compaction lock",
+        std::time::Duration::from_secs(5),
+    )
+    .await;
+
+    let gc_lock = timeline.gc_lock.lock();
+    let gc_lock = crate::timed(
+        gc_lock,
+        "acquires gc lock",
         std::time::Duration::from_secs(5),
     )
     .await;
@@ -107,7 +116,8 @@ pub(super) async fn delete_local_timeline_directory(
         .context("fsync_pre_mark_remove")?;
 
     info!("finished deleting layer files, releasing locks");
-    drop(guards);
+    drop(gc_lock);
+    drop(compaction_lock);
 
     fail::fail_point!("timeline-delete-after-rm", |_| {
         Err(anyhow::anyhow!("failpoint: timeline-delete-after-rm"))?
@@ -206,11 +216,10 @@ impl DeleteTimelineFlow {
     // NB: If this fails half-way through, and is retried, the retry will go through
     // all the same steps again. Make sure the code here is idempotent, and don't
     // error out if some of the shutdown tasks have already been completed!
-    #[instrument(skip_all, fields(%inplace))]
+    #[instrument(skip_all)]
     pub async fn run(
         tenant: &Arc<Tenant>,
         timeline_id: TimelineId,
-        inplace: bool,
     ) -> Result<(), DeleteTimelineError> {
         super::debug_assert_current_span_has_tenant_and_timeline_id();
 
@@ -221,6 +230,8 @@ impl DeleteTimelineFlow {
         // Now that the Timeline is in Stopping state, request all the related tasks to shut down.
         timeline.shutdown(super::ShutdownMode::Hard).await;
 
+        tenant.gc_block.before_delete(&timeline);
+
         fail::fail_point!("timeline-delete-before-index-deleted-at", |_| {
             Err(anyhow::anyhow!(
                 "failpoint: timeline-delete-before-index-deleted-at"
@@ -235,11 +246,7 @@ impl DeleteTimelineFlow {
             ))?
         });
 
-        if inplace {
-            Self::background(guard, tenant.conf, tenant, &timeline).await?
-        } else {
-            Self::schedule_background(guard, tenant.conf, Arc::clone(tenant), timeline);
-        }
+        Self::schedule_background(guard, tenant.conf, Arc::clone(tenant), timeline);
 
         Ok(())
     }

pageserver/src/tenant/timeline/detach_ancestor.rs

@@ -1,16 +1,20 @@
-use std::sync::Arc;
+use std::{collections::HashSet, sync::Arc};
 
 use super::{layer_manager::LayerManager, FlushLayerError, Timeline};
 use crate::{
     context::{DownloadBehavior, RequestContext},
     task_mgr::TaskKind,
     tenant::{
+        mgr::GetActiveTenantError,
+        remote_timeline_client::index::GcBlockingReason::DetachAncestor,
         storage_layer::{AsLayerDesc as _, DeltaLayerWriter, Layer, ResidentLayer},
         Tenant,
     },
     virtual_file::{MaybeFatalIo, VirtualFile},
 };
+use anyhow::Context;
 use pageserver_api::models::detach_ancestor::AncestorDetached;
+use tokio::sync::Semaphore;
 use tokio_util::sync::CancellationToken;
 use tracing::Instrument;
 use utils::{completion, generation::Generation, http::error::ApiError, id::TimelineId, lsn::Lsn};
@@ -38,6 +42,12 @@ pub(crate) enum Error {
     #[error("remote copying layer failed")]
     CopyFailed(#[source] anyhow::Error),
 
+    #[error("wait for tenant to activate after restarting")]
+    WaitToActivate(#[source] GetActiveTenantError),
+
+    #[error("detached timeline was not found after restart")]
+    DetachedNotFoundAfterRestart,
+
     #[error("unexpected error")]
     Unexpected(#[source] anyhow::Error),
 
@@ -55,6 +65,10 @@ impl From<Error> for ApiError {
             Error::OtherTimelineDetachOngoing(_) => {
                 ApiError::ResourceUnavailable("other timeline detach is already ongoing".into())
             }
+            e @ Error::WaitToActivate(_) => {
+                let s = utils::error::report_compact_sources(&e).to_string();
+                ApiError::ResourceUnavailable(s.into())
+            }
             // All of these contain shutdown errors, in fact, it's the most common
             e @ Error::FlushAncestor(_)
             | e @ Error::RewrittenDeltaDownloadFailed(_)
@@ -63,6 +77,7 @@ impl From<Error> for ApiError {
             | e @ Error::CopyFailed(_)
             | e @ Error::Unexpected(_)
             | e @ Error::Failpoint(_) => ApiError::InternalServerError(e.into()),
+            Error::DetachedNotFoundAfterRestart => ApiError::NotFound(value.into()),
         }
     }
 }
@@ -74,6 +89,11 @@ impl From<crate::tenant::upload_queue::NotInitialized> for Error {
         Error::ShuttingDown
     }
 }
+impl From<super::layer_manager::Shutdown> for Error {
+    fn from(_: super::layer_manager::Shutdown) -> Self {
+        Error::ShuttingDown
+    }
+}
 
 impl From<FlushLayerError> for Error {
     fn from(value: FlushLayerError) -> Self {
@@ -91,8 +111,25 @@ impl From<FlushLayerError> for Error {
     }
 }
 
+impl From<GetActiveTenantError> for Error {
+    fn from(value: GetActiveTenantError) -> Self {
+        use pageserver_api::models::TenantState;
+        use GetActiveTenantError::*;
+
+        match value {
+            Cancelled | WillNotBecomeActive(TenantState::Stopping { .. }) | SwitchedTenant => {
+                Error::ShuttingDown
+            }
+            WaitForActiveTimeout { .. } | NotFound(_) | Broken(_) | WillNotBecomeActive(_) => {
+                // NotFound seems out-of-place
+                Error::WaitToActivate(value)
+            }
+        }
+    }
+}
+
 pub(crate) enum Progress {
-    Prepared(completion::Completion, PreparedTimelineDetach),
+    Prepared(Attempt, PreparedTimelineDetach),
     Done(AncestorDetached),
 }
 
@@ -116,6 +153,26 @@ impl Default for Options {
     }
 }
 
+/// Represents an across tenant reset exclusive single attempt to detach ancestor.
+#[derive(Debug)]
+pub(crate) struct Attempt {
+    pub(crate) timeline_id: TimelineId,
+
+    _guard: completion::Completion,
+    gate_entered: Option<utils::sync::gate::GateGuard>,
+}
+
+impl Attempt {
+    pub(crate) fn before_reset_tenant(&mut self) {
+        let taken = self.gate_entered.take();
+        assert!(taken.is_some());
+    }
+
+    pub(crate) fn new_barrier(&self) -> completion::Barrier {
+        self._guard.barrier()
+    }
+}
+
 /// See [`Timeline::prepare_to_detach_from_ancestor`]
 pub(super) async fn prepare(
     detached: &Arc<Timeline>,
@@ -130,61 +187,38 @@ pub(super) async fn prepare(
         .as_ref()
         .map(|tl| (tl.clone(), detached.ancestor_lsn))
     else {
-        {
+        let still_in_progress = {
             let accessor = detached.remote_client.initialized_upload_queue()?;
 
             // we are safe to inspect the latest uploaded, because we can only witness this after
             // restart is complete and ancestor is no more.
             let latest = accessor.latest_uploaded_index_part();
-            if !latest.lineage.is_detached_from_original_ancestor() {
+            if latest.lineage.detached_previous_ancestor().is_none() {
                 return Err(NoAncestor);
-            }
+            };
+
+            latest
+                .gc_blocking
+                .as_ref()
+                .is_some_and(|b| b.blocked_by(DetachAncestor))
+        };
+
+        if still_in_progress {
+            // gc is still blocked, we can still reparent and complete.
+            // we are safe to reparent remaining, because they were locked in in the beginning.
+            let attempt = continue_with_blocked_gc(detached, tenant).await?;
+
+            // because the ancestor of detached is already set to none, we have published all
+            // of the layers, so we are still "prepared."
+            return Ok(Progress::Prepared(
+                attempt,
+                PreparedTimelineDetach { layers: Vec::new() },
+            ));
         }
 
-        // detached has previously been detached; let's inspect each of the current timelines and
-        // report back the timelines which have been reparented by our detach
-        let mut all_direct_children = tenant
-            .timelines
-            .lock()
-            .unwrap()
-            .values()
-            .filter(|tl| matches!(tl.ancestor_timeline.as_ref(), Some(ancestor) if Arc::ptr_eq(ancestor, detached)))
-            .map(|tl| (tl.ancestor_lsn, tl.clone()))
-            .collect::<Vec<_>>();
-
-        let mut any_shutdown = false;
-
-        all_direct_children.retain(
-            |(_, tl)| match tl.remote_client.initialized_upload_queue() {
-                Ok(accessor) => accessor
-                    .latest_uploaded_index_part()
-                    .lineage
-                    .is_reparented(),
-                Err(_shutdownalike) => {
-                    // not 100% a shutdown, but let's bail early not to give inconsistent results in
-                    // sharded enviroment.
-                    any_shutdown = true;
-                    true
-                }
-            },
-        );
-
-        if any_shutdown {
-            // it could be one or many being deleted; have client retry
-            return Err(Error::ShuttingDown);
-        }
-
-        let mut reparented = all_direct_children;
-        // why this instead of hashset? there is a reason, but I've forgotten it many times.
-        //
-        // maybe if this was a hashset we would not be able to distinguish some race condition.
-        reparented.sort_unstable_by_key(|(lsn, tl)| (*lsn, tl.timeline_id));
-
+        let reparented_timelines = reparented_direct_children(detached, tenant)?;
         return Ok(Progress::Done(AncestorDetached {
-            reparented_timelines: reparented
-                .into_iter()
-                .map(|(_, tl)| tl.timeline_id)
-                .collect(),
+            reparented_timelines,
         }));
     };
 
@@ -200,22 +234,7 @@ pub(super) async fn prepare(
         return Err(TooManyAncestors);
     }
 
-    // before we acquire the gate, we must mark the ancestor as having a detach operation
-    // ongoing which will block other concurrent detach operations so we don't get to ackward
-    // situations where there would be two branches trying to reparent earlier branches.
-    let (guard, barrier) = completion::channel();
-
-    {
-        let mut guard = tenant.ongoing_timeline_detach.lock().unwrap();
-        if let Some((tl, other)) = guard.as_ref() {
-            if !other.is_ready() {
-                return Err(OtherTimelineDetachOngoing(*tl));
-            }
-        }
-        *guard = Some((detached.timeline_id, barrier));
-    }
-
-    let _gate_entered = detached.gate.enter().map_err(|_| ShuttingDown)?;
+    let attempt = start_new_attempt(detached, tenant).await?;
 
     utils::pausable_failpoint!("timeline-detach-ancestor::before_starting_after_locking_pausable");
 
@@ -277,11 +296,12 @@ pub(super) async fn prepare(
 
         // between retries, these can change if compaction or gc ran in between. this will mean
         // we have to redo work.
-        partition_work(ancestor_lsn, &layers)
+        partition_work(ancestor_lsn, &layers)?
     };
 
     // TODO: layers are already sorted by something: use that to determine how much of remote
-    // copies are already done.
+    // copies are already done -- gc is blocked, but a compaction could had happened on ancestor,
+    // which is something to keep in mind if copy skipping is implemented.
     tracing::info!(filtered=%filtered_layers, to_rewrite = straddling_branchpoint.len(), historic=%rest_of_historic.len(), "collected layers");
 
     // TODO: copying and lsn prefix copying could be done at the same time with a single fsync after
@@ -295,29 +315,33 @@ pub(super) async fn prepare(
 
         let mut wrote_any = false;
 
-        let limiter = Arc::new(tokio::sync::Semaphore::new(
-            options.rewrite_concurrency.get(),
-        ));
+        let limiter = Arc::new(Semaphore::new(options.rewrite_concurrency.get()));
 
         for layer in straddling_branchpoint {
             let limiter = limiter.clone();
             let timeline = detached.clone();
             let ctx = ctx.detached_child(TaskKind::DetachAncestor, DownloadBehavior::Download);
 
-            tasks.spawn(async move {
-                let _permit = limiter.acquire().await;
-                let copied =
-                    upload_rewritten_layer(end_lsn, &layer, &timeline, &timeline.cancel, &ctx)
-                        .await?;
-                Ok(copied)
-            });
+            let span = tracing::info_span!("upload_rewritten_layer", %layer);
+            tasks.spawn(
+                async move {
+                    let _permit = limiter.acquire().await;
+                    let copied =
+                        upload_rewritten_layer(end_lsn, &layer, &timeline, &timeline.cancel, &ctx)
+                            .await?;
+                    if let Some(copied) = copied.as_ref() {
+                        tracing::info!(%copied, "rewrote and uploaded");
+                    }
+                    Ok(copied)
+                }
+                .instrument(span),
+            );
         }
 
         while let Some(res) = tasks.join_next().await {
             match res {
                 Ok(Ok(Some(copied))) => {
                     wrote_any = true;
-                    tracing::info!(layer=%copied, "rewrote and uploaded");
                     new_layers.push(copied);
                 }
                 Ok(Ok(None)) => {}
@@ -344,7 +368,7 @@ pub(super) async fn prepare(
     }
 
     let mut tasks = tokio::task::JoinSet::new();
-    let limiter = Arc::new(tokio::sync::Semaphore::new(options.copy_concurrency.get()));
+    let limiter = Arc::new(Semaphore::new(options.copy_concurrency.get()));
 
     for adopted in rest_of_historic {
         let limiter = limiter.clone();
@@ -378,19 +402,119 @@ pub(super) async fn prepare(
 
     let prepared = PreparedTimelineDetach { layers: new_layers };
 
-    Ok(Progress::Prepared(guard, prepared))
+    Ok(Progress::Prepared(attempt, prepared))
+}
+
+async fn start_new_attempt(detached: &Timeline, tenant: &Tenant) -> Result<Attempt, Error> {
+    let attempt = obtain_exclusive_attempt(detached, tenant)?;
+
+    // insert the block in the index_part.json, if not already there.
+    let _dont_care = tenant
+        .gc_block
+        .insert(
+            detached,
+            crate::tenant::remote_timeline_client::index::GcBlockingReason::DetachAncestor,
+        )
+        .await
+        // FIXME: better error
+        .map_err(Error::Unexpected)?;
+
+    Ok(attempt)
+}
+
+async fn continue_with_blocked_gc(detached: &Timeline, tenant: &Tenant) -> Result<Attempt, Error> {
+    // FIXME: it would be nice to confirm that there is an in-memory version, since we've just
+    // verified there is a persistent one?
+    obtain_exclusive_attempt(detached, tenant)
+}
+
+fn obtain_exclusive_attempt(detached: &Timeline, tenant: &Tenant) -> Result<Attempt, Error> {
+    use Error::{OtherTimelineDetachOngoing, ShuttingDown};
+
+    // ensure we are the only active attempt for this tenant
+    let (guard, barrier) = completion::channel();
+    {
+        let mut guard = tenant.ongoing_timeline_detach.lock().unwrap();
+        if let Some((tl, other)) = guard.as_ref() {
+            if !other.is_ready() {
+                return Err(OtherTimelineDetachOngoing(*tl));
+            }
+            // FIXME: no test enters here
+        }
+        *guard = Some((detached.timeline_id, barrier));
+    }
+
+    // ensure the gate is still open
+    let _gate_entered = detached.gate.enter().map_err(|_| ShuttingDown)?;
+
+    Ok(Attempt {
+        timeline_id: detached.timeline_id,
+        _guard: guard,
+        gate_entered: Some(_gate_entered),
+    })
+}
+
+fn reparented_direct_children(
+    detached: &Arc<Timeline>,
+    tenant: &Tenant,
+) -> Result<HashSet<TimelineId>, Error> {
+    let mut all_direct_children = tenant
+        .timelines
+        .lock()
+        .unwrap()
+        .values()
+        .filter_map(|tl| {
+            let is_direct_child = matches!(tl.ancestor_timeline.as_ref(), Some(ancestor) if Arc::ptr_eq(ancestor, detached));
+
+            if is_direct_child {
+                Some(tl.clone())
+            } else {
+                if let Some(timeline) = tl.ancestor_timeline.as_ref() {
+                    assert_ne!(timeline.timeline_id, detached.timeline_id, "we cannot have two timelines with the same timeline_id live");
+                }
+                None
+            }
+        })
+        // Collect to avoid lock taking order problem with Tenant::timelines and
+        // Timeline::remote_client
+        .collect::<Vec<_>>();
+
+    let mut any_shutdown = false;
+
+    all_direct_children.retain(|tl| match tl.remote_client.initialized_upload_queue() {
+        Ok(accessor) => accessor
+            .latest_uploaded_index_part()
+            .lineage
+            .is_reparented(),
+        Err(_shutdownalike) => {
+            // not 100% a shutdown, but let's bail early not to give inconsistent results in
+            // sharded enviroment.
+            any_shutdown = true;
+            true
+        }
+    });
+
+    if any_shutdown {
+        // it could be one or many being deleted; have client retry
+        return Err(Error::ShuttingDown);
+    }
+
+    Ok(all_direct_children
+        .into_iter()
+        .map(|tl| tl.timeline_id)
+        .collect())
 }
 
 fn partition_work(
     ancestor_lsn: Lsn,
-    source_layermap: &LayerManager,
-) -> (usize, Vec<Layer>, Vec<Layer>) {
+    source: &LayerManager,
+) -> Result<(usize, Vec<Layer>, Vec<Layer>), Error> {
     let mut straddling_branchpoint = vec![];
     let mut rest_of_historic = vec![];
 
     let mut later_by_lsn = 0;
 
-    for desc in source_layermap.layer_map().iter_historic_layers() {
+    for desc in source.layer_map()?.iter_historic_layers() {
         // off by one chances here:
         // - start is inclusive
         // - end is exclusive
@@ -409,10 +533,10 @@ fn partition_work(
             &mut rest_of_historic
         };
 
-        target.push(source_layermap.get_from_desc(&desc));
+        target.push(source.get_from_desc(&desc));
     }
 
-    (later_by_lsn, straddling_branchpoint, rest_of_historic)
+    Ok((later_by_lsn, straddling_branchpoint, rest_of_historic))
 }
 
 async fn upload_rewritten_layer(
@@ -488,10 +612,12 @@ async fn copy_lsn_prefix(
         // reuse the key instead of adding more holes between layers by using the real
         // highest key in the layer.
         let reused_highest_key = layer.layer_desc().key_range.end;
-        let copied = writer
-            .finish(reused_highest_key, target_timeline, ctx)
+        let (desc, path) = writer
+            .finish(reused_highest_key, ctx)
             .await
             .map_err(CopyDeltaPrefix)?;
+        let copied = Layer::finish_creating(target_timeline.conf, target_timeline, desc, &path)
+            .map_err(CopyDeltaPrefix)?;
 
         tracing::debug!(%layer, %copied, "new layer produced");
 
@@ -531,131 +657,311 @@ async fn remote_copy(
         .map_err(CopyFailed)
 }
 
-/// See [`Timeline::complete_detaching_timeline_ancestor`].
-pub(super) async fn complete(
+pub(crate) enum DetachingAndReparenting {
+    /// All of the following timeline ids were reparented and the timeline ancestor detach must be
+    /// marked as completed.
+    Reparented(HashSet<TimelineId>),
+
+    /// Some of the reparentings failed. The timeline ancestor detach must **not** be marked as
+    /// completed.
+    ///
+    /// Nested `must_reset_tenant` is set to true when any restart requiring changes were made.
+    SomeReparentingFailed { must_reset_tenant: bool },
+
+    /// Detaching and reparentings were completed in a previous attempt. Timeline ancestor detach
+    /// must be marked as completed.
+    AlreadyDone(HashSet<TimelineId>),
+}
+
+impl DetachingAndReparenting {
+    pub(crate) fn reset_tenant_required(&self) -> bool {
+        use DetachingAndReparenting::*;
+        match self {
+            Reparented(_) => true,
+            SomeReparentingFailed { must_reset_tenant } => *must_reset_tenant,
+            AlreadyDone(_) => false,
+        }
+    }
+
+    pub(crate) fn completed(self) -> Option<HashSet<TimelineId>> {
+        use DetachingAndReparenting::*;
+        match self {
+            Reparented(x) | AlreadyDone(x) => Some(x),
+            SomeReparentingFailed { .. } => None,
+        }
+    }
+}
+
+/// See [`Timeline::detach_from_ancestor_and_reparent`].
+pub(super) async fn detach_and_reparent(
     detached: &Arc<Timeline>,
     tenant: &Tenant,
     prepared: PreparedTimelineDetach,
     _ctx: &RequestContext,
-) -> Result<Vec<TimelineId>, anyhow::Error> {
+) -> Result<DetachingAndReparenting, anyhow::Error> {
     let PreparedTimelineDetach { layers } = prepared;
 
-    let ancestor = detached
-        .get_ancestor_timeline()
-        .expect("must still have a ancestor");
-    let ancestor_lsn = detached.get_ancestor_lsn();
+    #[derive(Debug)]
+    enum Ancestor {
+        NotDetached(Arc<Timeline>, Lsn),
+        Detached(Arc<Timeline>, Lsn),
+    }
+
+    let (recorded_branchpoint, still_ongoing) = {
+        let access = detached.remote_client.initialized_upload_queue()?;
+        let latest = access.latest_uploaded_index_part();
+
+        (
+            latest.lineage.detached_previous_ancestor(),
+            latest
+                .gc_blocking
+                .as_ref()
+                .is_some_and(|b| b.blocked_by(DetachAncestor)),
+        )
+    };
+    assert!(
+        still_ongoing,
+        "cannot (detach? reparent)? complete if the operation is not still ongoing"
+    );
+
+    let ancestor = match (detached.ancestor_timeline.as_ref(), recorded_branchpoint) {
+        (Some(ancestor), None) => {
+            assert!(
+                !layers.is_empty(),
+                "there should always be at least one layer to inherit"
+            );
+            Ancestor::NotDetached(ancestor.clone(), detached.ancestor_lsn)
+        }
+        (Some(_), Some(_)) => {
+            panic!(
+                "it should be impossible to get to here without having gone through the tenant reset; if the tenant was reset, then the ancestor_timeline would be None"
+            );
+        }
+        (None, Some((ancestor_id, ancestor_lsn))) => {
+            // it has been either:
+            // - detached but still exists => we can try reparenting
+            // - detached and deleted
+            //
+            // either way, we must complete
+            assert!(
+                layers.is_empty(),
+                "no layers should had been copied as detach is done"
+            );
+
+            let existing = tenant.timelines.lock().unwrap().get(&ancestor_id).cloned();
+
+            if let Some(ancestor) = existing {
+                Ancestor::Detached(ancestor, ancestor_lsn)
+            } else {
+                let direct_children = reparented_direct_children(detached, tenant)?;
+                return Ok(DetachingAndReparenting::AlreadyDone(direct_children));
+            }
+        }
+        (None, None) => {
+            // TODO: make sure there are no `?` before tenant_reset from after a questionmark from
+            // here.
+            panic!(
+            "bug: detach_and_reparent called on a timeline which has not been detached or which has no live ancestor"
+            );
+        }
+    };
 
     // publish the prepared layers before we reparent any of the timelines, so that on restart
     // reparented timelines find layers. also do the actual detaching.
     //
-    // if we crash after this operation, we will at least come up having detached a timeline, but
-    // we cannot go back and reparent the timelines which would had been reparented in normal
-    // execution.
-    //
-    // this is not perfect, but it avoids us a retry happening after a compaction or gc on restart
-    // which could give us a completely wrong layer combination.
-    detached
-        .remote_client
-        .schedule_adding_existing_layers_to_index_detach_and_wait(
-            &layers,
-            (ancestor.timeline_id, ancestor_lsn),
-        )
-        .await?;
+    // if we crash after this operation, a retry will allow reparenting the remaining timelines as
+    // gc is blocked.
+
+    let (ancestor, ancestor_lsn, was_detached) = match ancestor {
+        Ancestor::NotDetached(ancestor, ancestor_lsn) => {
+            // this has to complete before any reparentings because otherwise they would not have
+            // layers on the new parent.
+            detached
+                .remote_client
+                .schedule_adding_existing_layers_to_index_detach_and_wait(
+                    &layers,
+                    (ancestor.timeline_id, ancestor_lsn),
+                )
+                .await
+                .context("publish layers and detach ancestor")?;
+
+            tracing::info!(
+                ancestor=%ancestor.timeline_id,
+                %ancestor_lsn,
+                inherited_layers=%layers.len(),
+                "detached from ancestor"
+            );
+            (ancestor, ancestor_lsn, true)
+        }
+        Ancestor::Detached(ancestor, ancestor_lsn) => (ancestor, ancestor_lsn, false),
+    };
 
     let mut tasks = tokio::task::JoinSet::new();
 
+    // Returns a single permit semaphore which will be used to make one reparenting succeed,
+    // others will fail as if those timelines had been stopped for whatever reason.
+    #[cfg(feature = "testing")]
+    let failpoint_sem = || -> Option<Arc<Semaphore>> {
+        fail::fail_point!("timeline-detach-ancestor::allow_one_reparented", |_| Some(
+            Arc::new(Semaphore::new(1))
+        ));
+        None
+    }();
+
     // because we are now keeping the slot in progress, it is unlikely that there will be any
     // timeline deletions during this time. if we raced one, then we'll just ignore it.
-    tenant
-        .timelines
-        .lock()
-        .unwrap()
-        .values()
-        .filter_map(|tl| {
-            if Arc::ptr_eq(tl, detached) {
-                return None;
-            }
+    {
+        let g = tenant.timelines.lock().unwrap();
+        reparentable_timelines(g.values(), detached, &ancestor, ancestor_lsn)
+            .cloned()
+            .for_each(|timeline| {
+                // important in this scope: we are holding the Tenant::timelines lock
+                let span = tracing::info_span!("reparent", reparented=%timeline.timeline_id);
+                let new_parent = detached.timeline_id;
+                #[cfg(feature = "testing")]
+                let failpoint_sem = failpoint_sem.clone();
 
-            if !tl.is_active() {
-                return None;
-            }
+                tasks.spawn(
+                    async move {
+                        let res = async {
+                            #[cfg(feature = "testing")]
+                            if let Some(failpoint_sem) = failpoint_sem {
+                                let _permit = failpoint_sem.acquire().await.map_err(|_| {
+                                    anyhow::anyhow!(
+                                        "failpoint: timeline-detach-ancestor::allow_one_reparented",
+                                    )
+                                })?;
+                                failpoint_sem.close();
+                            }
 
-            let tl_ancestor = tl.ancestor_timeline.as_ref()?;
-            let is_same = Arc::ptr_eq(&ancestor, tl_ancestor);
-            let is_earlier = tl.get_ancestor_lsn() <= ancestor_lsn;
-
-            let is_deleting = tl
-                .delete_progress
-                .try_lock()
-                .map(|flow| !flow.is_not_started())
-                .unwrap_or(true);
-
-            if is_same && is_earlier && !is_deleting {
-                Some(tl.clone())
-            } else {
-                None
-            }
-        })
-        .for_each(|timeline| {
-            // important in this scope: we are holding the Tenant::timelines lock
-            let span = tracing::info_span!("reparent", reparented=%timeline.timeline_id);
-            let new_parent = detached.timeline_id;
-
-            tasks.spawn(
-                async move {
-                    let res = timeline
-                        .remote_client
-                        .schedule_reparenting_and_wait(&new_parent)
+                            timeline
+                                .remote_client
+                                .schedule_reparenting_and_wait(&new_parent)
+                                .await
+                        }
                         .await;
 
-                    match res {
-                        Ok(()) => Some(timeline),
-                        Err(e) => {
-                            // with the use of tenant slot, we no longer expect these.
-                            tracing::warn!("reparenting failed: {e:#}");
-                            None
+                        match res {
+                            Ok(()) => {
+                                tracing::info!("reparented");
+                                Some(timeline)
+                            }
+                            Err(e) => {
+                                // with the use of tenant slot, raced timeline deletion is the most
+                                // likely reason.
+                                tracing::warn!("reparenting failed: {e:#}");
+                                None
+                            }
                         }
                     }
-                }
-                .instrument(span),
-            );
-        });
+                    .instrument(span),
+                );
+            });
+    }
 
     let reparenting_candidates = tasks.len();
-    let mut reparented = Vec::with_capacity(tasks.len());
+    let mut reparented = HashSet::with_capacity(tasks.len());
 
     while let Some(res) = tasks.join_next().await {
         match res {
             Ok(Some(timeline)) => {
-                tracing::info!(reparented=%timeline.timeline_id, "reparenting done");
-                reparented.push((timeline.ancestor_lsn, timeline.timeline_id));
-            }
-            Ok(None) => {
-                // lets just ignore this for now. one or all reparented timelines could had
-                // started deletion, and that is fine.
+                assert!(
+                    reparented.insert(timeline.timeline_id),
+                    "duplicate reparenting? timeline_id={}",
+                    timeline.timeline_id
+                );
             }
             Err(je) if je.is_cancelled() => unreachable!("not used"),
-            Err(je) if je.is_panic() => {
-                // ignore; it's better to continue with a single reparenting failing (or even
-                // all of them) in order to get to the goal state.
-                //
-                // these timelines will never be reparentable, but they can be always detached as
-                // separate tree roots.
-            }
+            // just ignore failures now, we can retry
+            Ok(None) => {}
+            Err(je) if je.is_panic() => {}
             Err(je) => tracing::error!("unexpected join error: {je:?}"),
         }
     }
 
-    if reparenting_candidates != reparented.len() {
-        tracing::info!("failed to reparent some candidates");
+    let reparented_all = reparenting_candidates == reparented.len();
+
+    if reparented_all {
+        Ok(DetachingAndReparenting::Reparented(reparented))
+    } else {
+        tracing::info!(
+            reparented = reparented.len(),
+            candidates = reparenting_candidates,
+            "failed to reparent all candidates; they can be retried after the tenant_reset",
+        );
+
+        let must_reset_tenant = !reparented.is_empty() || was_detached;
+        Ok(DetachingAndReparenting::SomeReparentingFailed { must_reset_tenant })
+    }
+}
+
+pub(super) async fn complete(
+    detached: &Arc<Timeline>,
+    tenant: &Tenant,
+    mut attempt: Attempt,
+    _ctx: &RequestContext,
+) -> Result<(), Error> {
+    assert_eq!(detached.timeline_id, attempt.timeline_id);
+
+    if attempt.gate_entered.is_none() {
+        let entered = detached.gate.enter().map_err(|_| Error::ShuttingDown)?;
+        attempt.gate_entered = Some(entered);
+    } else {
+        // Some(gate_entered) means the tenant was not restarted, as is not required
     }
 
-    reparented.sort_unstable();
+    assert!(detached.ancestor_timeline.is_none());
 
-    let reparented = reparented
-        .into_iter()
-        .map(|(_, timeline_id)| timeline_id)
-        .collect();
+    // this should be an 503 at least...?
+    fail::fail_point!(
+        "timeline-detach-ancestor::complete_before_uploading",
+        |_| Err(Error::Failpoint(
+            "timeline-detach-ancestor::complete_before_uploading"
+        ))
+    );
 
-    Ok(reparented)
+    tenant
+        .gc_block
+        .remove(
+            detached,
+            crate::tenant::remote_timeline_client::index::GcBlockingReason::DetachAncestor,
+        )
+        .await
+        // FIXME: better error
+        .map_err(Error::Unexpected)?;
+
+    Ok(())
+}
+
+/// Query against a locked `Tenant::timelines`.
+fn reparentable_timelines<'a, I>(
+    timelines: I,
+    detached: &'a Arc<Timeline>,
+    ancestor: &'a Arc<Timeline>,
+    ancestor_lsn: Lsn,
+) -> impl Iterator<Item = &'a Arc<Timeline>> + 'a
+where
+    I: Iterator<Item = &'a Arc<Timeline>> + 'a,
+{
+    timelines.filter_map(move |tl| {
+        if Arc::ptr_eq(tl, detached) {
+            return None;
+        }
+
+        let tl_ancestor = tl.ancestor_timeline.as_ref()?;
+        let is_same = Arc::ptr_eq(ancestor, tl_ancestor);
+        let is_earlier = tl.get_ancestor_lsn() <= ancestor_lsn;
+
+        let is_deleting = tl
+            .delete_progress
+            .try_lock()
+            .map(|flow| !flow.is_not_started())
+            .unwrap_or(true);
+
+        if is_same && is_earlier && !is_deleting {
+            Some(tl)
+        } else {
+            None
+        }
+    })
 }

pageserver/src/tenant/timeline/eviction_task.rs

@@ -213,51 +213,45 @@ impl Timeline {
         let mut js = tokio::task::JoinSet::new();
         {
             let guard = self.layers.read().await;
-            let layers = guard.layer_map();
-            for layer in layers.iter_historic_layers() {
-                let layer = guard.get_from_desc(&layer);
 
-                // guard against eviction while we inspect it; it might be that eviction_task and
-                // disk_usage_eviction_task both select the same layers to be evicted, and
-                // seemingly free up double the space. both succeeding is of no consequence.
+            guard
+                .likely_resident_layers()
+                .filter(|layer| {
+                    let last_activity_ts = layer.latest_activity();
 
-                if !layer.is_likely_resident() {
-                    continue;
-                }
+                    let no_activity_for = match now.duration_since(last_activity_ts) {
+                        Ok(d) => d,
+                        Err(_e) => {
+                            // We reach here if `now` < `last_activity_ts`, which can legitimately
+                            // happen if there is an access between us getting `now`, and us getting
+                            // the access stats from the layer.
+                            //
+                            // The other reason why it can happen is system clock skew because
+                            // SystemTime::now() is not monotonic, so, even if there is no access
+                            // to the layer after we get `now` at the beginning of this function,
+                            // it could be that `now`  < `last_activity_ts`.
+                            //
+                            // To distinguish the cases, we would need to record `Instant`s in the
+                            // access stats (i.e., monotonic timestamps), but then, the timestamps
+                            // values in the access stats would need to be `Instant`'s, and hence
+                            // they would be meaningless outside of the pageserver process.
+                            // At the time of writing, the trade-off is that access stats are more
+                            // valuable than detecting clock skew.
+                            return false;
+                        }
+                    };
 
-                let last_activity_ts = layer.access_stats().latest_activity();
-
-                let no_activity_for = match now.duration_since(last_activity_ts) {
-                    Ok(d) => d,
-                    Err(_e) => {
-                        // We reach here if `now` < `last_activity_ts`, which can legitimately
-                        // happen if there is an access between us getting `now`, and us getting
-                        // the access stats from the layer.
-                        //
-                        // The other reason why it can happen is system clock skew because
-                        // SystemTime::now() is not monotonic, so, even if there is no access
-                        // to the layer after we get `now` at the beginning of this function,
-                        // it could be that `now`  < `last_activity_ts`.
-                        //
-                        // To distinguish the cases, we would need to record `Instant`s in the
-                        // access stats (i.e., monotonic timestamps), but then, the timestamps
-                        // values in the access stats would need to be `Instant`'s, and hence
-                        // they would be meaningless outside of the pageserver process.
-                        // At the time of writing, the trade-off is that access stats are more
-                        // valuable than detecting clock skew.
-                        continue;
-                    }
-                };
-
-                if no_activity_for > p.threshold {
+                    no_activity_for > p.threshold
+                })
+                .cloned()
+                .for_each(|layer| {
                     js.spawn(async move {
                         layer
                             .evict_and_wait(std::time::Duration::from_secs(5))
                             .await
                     });
                     stats.candidates += 1;
-                }
-            }
+                });
         };
 
         let join_all = async move {

pageserver/src/tenant/timeline/handle.rs

@@ -0,0 +1,967 @@
+//! An efficient way to keep the timeline gate open without preventing
+//! timeline shutdown for longer than a single call to a timeline method.
+//!
+//! # Motivation
+//!
+//! On a single page service connection, we're typically serving a single TenantTimelineId.
+//!
+//! Without sharding, there is a single Timeline object to which we dispatch
+//! all requests. For example, a getpage request gets dispatched to the
+//! Timeline::get method of the Timeline object that represents the
+//! (tenant,timeline) of that connection.
+//!
+//! With sharding, for each request that comes in on the connection,
+//! we first have to perform shard routing based on the requested key (=~ page number).
+//! The result of shard routing is a Timeline object.
+//! We then dispatch the request to that Timeline object.
+//!
+//! Regardless of whether the tenant is sharded or not, we want to ensure that
+//! we hold the Timeline gate open while we're invoking the method on the
+//! Timeline object.
+//!
+//! However, we want to avoid the overhead of entering the gate for every
+//! method invocation.
+//!
+//! Further, for shard routing, we want to avoid calling the tenant manager to
+//! resolve the shard for every request. Instead, we want to cache the
+//! routing result so we can bypass the tenant manager for all subsequent requests
+//! that get routed to that shard.
+//!
+//! Regardless of how we accomplish the above, it should not
+//! prevent the Timeline from shutting down promptly.
+//!
+//! # Design
+//!
+//! There are three user-facing data structures:
+//! - `PerTimelineState`: a struct embedded into each Timeline struct. Lifetime == Timeline lifetime.
+//! - `Cache`: a struct private to each connection handler; Lifetime == connection lifetime.
+//! - `Handle`: a smart pointer that holds the Timeline gate open and derefs to `&Timeline`.
+//!   Lifetime: for a single request dispatch on the Timeline (i.e., one getpage request)
+//!
+//! The `Handle` is just a wrapper around an `Arc<HandleInner>`.
+//!
+//! There is one long-lived `Arc<HandleInner>`, which is stored in the `PerTimelineState`.
+//! The `Cache` stores a `Weak<HandleInner>` for each cached Timeline.
+//!
+//! To dispatch a request, the page service connection calls `Cache::get`.
+//!
+//! A cache miss means we consult the tenant manager for shard routing,
+//! resulting in an `Arc<Timeline>`. We enter its gate _once_ and construct an
+//! `Arc<HandleInner>`. We store a `Weak<HandleInner>` in the cache
+//! and the `Arc<HandleInner>` in the `PerTimelineState`.
+//!
+//! For subsequent requests, `Cache::get` will perform a "fast path" shard routing
+//! and find the `Weak<HandleInner>` in the cache.
+//! We upgrade the `Weak<HandleInner>` to an `Arc<HandleInner>` and wrap it in the user-facing `Handle` type.
+//!
+//! The request handler dispatches the request to the right `<Handle as Deref<Target = Timeline>>::$request_method`.
+//! It then drops the `Handle`, which drops the `Arc<HandleInner>`.
+//!
+//! # Memory Management / How The Reference Cycle Is Broken
+//!
+//! The attentive reader may have noticed the strong reference cycle
+//! from `Arc<HandleInner>` to `PerTimelineState` to `Arc<Timeline>`.
+//!
+//! This cycle is intentional: while it exists, the `Cache` can upgrade its
+//! `Weak<HandleInner>` to an `Arc<HandleInner>` in a single atomic operation.
+//!
+//! The cycle is broken by either
+//! - `PerTimelineState::shutdown` or
+//! - dropping the `Cache`.
+//!
+//! Concurrently existing `Handle`s will extend the existence of the cycle.
+//! However, since `Handle`s are short-lived and new `Handle`s are not
+//! handed out after either `PerTimelineState::shutdown` or `Cache` drop,
+//! that extension of the cycle is bounded.
+//!
+//! # Fast Path for Shard Routing
+//!
+//! The `Cache` has a fast path for shard routing to avoid calling into
+//! the tenant manager for every request.
+//!
+//! The `Cache` maintains a hash map of `ShardTimelineId` to `Weak<HandleInner>`.
+//!
+//! The current implementation uses the first entry in the hash map
+//! to determine the `ShardParameters` and derive the correct
+//! `ShardIndex` for the requested key.
+//!
+//! It then looks up the hash map for that `ShardTimelineId := {ShardIndex,TimelineId}`.
+//!
+//! If the lookup is successful and the `Weak<HandleInner>` can be upgraded,
+//! it's a hit.
+//!
+//! ## Cache invalidation
+//!
+//! The insight is that cache invalidation is sufficient and most efficiently done lazily.
+//! The only reasons why an entry in the cache can become stale are:
+//! 1. The `PerTimelineState` / Timeline is shutting down e.g. because the shard is
+//!    being detached, timeline or shard deleted, or pageserver is shutting down.
+//! 2. We're doing a shard split and new traffic should be routed to the child shards.
+//!
+//! Regarding (1), we will eventually fail to upgrade the `Weak<HandleInner>` once the
+//! timeline has shut down, and when that happens, we remove the entry from the cache.
+//!
+//! Regarding (2), the insight is that it is toally fine to keep dispatching requests
+//! to the parent shard during a shard split. Eventually, the shard split task will
+//! shut down the parent => case (1).
+
+use std::collections::hash_map;
+use std::collections::HashMap;
+use std::sync::atomic::AtomicBool;
+use std::sync::atomic::Ordering;
+use std::sync::Arc;
+use std::sync::Mutex;
+use std::sync::Weak;
+
+use pageserver_api::shard::ShardIdentity;
+use tracing::instrument;
+use tracing::trace;
+use utils::id::TimelineId;
+use utils::shard::ShardIndex;
+use utils::shard::ShardNumber;
+
+use crate::tenant::mgr::ShardSelector;
+
+/// The requirement for Debug is so that #[derive(Debug)] works in some places.
+pub(crate) trait Types: Sized + std::fmt::Debug {
+    type TenantManagerError: Sized + std::fmt::Debug;
+    type TenantManager: TenantManager<Self> + Sized;
+    type Timeline: ArcTimeline<Self> + Sized;
+}
+
+/// Uniquely identifies a [`Cache`] instance over the lifetime of the process.
+/// Required so [`Cache::drop`] can take out the handles from the [`PerTimelineState`].
+/// Alternative to this would be to allocate [`Cache`] in a `Box` and identify it by the pointer.
+#[derive(Debug, Hash, PartialEq, Eq, Clone, Copy)]
+struct CacheId(u64);
+
+impl CacheId {
+    fn next() -> Self {
+        static NEXT_ID: std::sync::atomic::AtomicU64 = std::sync::atomic::AtomicU64::new(1);
+        let id = NEXT_ID.fetch_add(1, std::sync::atomic::Ordering::Relaxed);
+        if id == 0 {
+            panic!("CacheId::new() returned 0, overflow");
+        }
+        Self(id)
+    }
+}
+
+/// See module-level comment.
+pub(crate) struct Cache<T: Types> {
+    id: CacheId,
+    map: Map<T>,
+}
+
+type Map<T> = HashMap<ShardTimelineId, Weak<HandleInner<T>>>;
+
+impl<T: Types> Default for Cache<T> {
+    fn default() -> Self {
+        Self {
+            id: CacheId::next(),
+            map: Default::default(),
+        }
+    }
+}
+
+#[derive(PartialEq, Eq, Debug, Hash, Clone, Copy)]
+pub(crate) struct ShardTimelineId {
+    pub(crate) shard_index: ShardIndex,
+    pub(crate) timeline_id: TimelineId,
+}
+
+/// See module-level comment.
+pub(crate) struct Handle<T: Types>(Arc<HandleInner<T>>);
+struct HandleInner<T: Types> {
+    shut_down: AtomicBool,
+    timeline: T::Timeline,
+    // The timeline's gate held open.
+    _gate_guard: utils::sync::gate::GateGuard,
+}
+
+/// Embedded in each [`Types::Timeline`] as the anchor for the only long-lived strong ref to `HandleInner`.
+///
+/// See module-level comment for details.
+pub struct PerTimelineState<T: Types> {
+    // None = shutting down
+    handles: Mutex<Option<HashMap<CacheId, Arc<HandleInner<T>>>>>,
+}
+
+impl<T: Types> Default for PerTimelineState<T> {
+    fn default() -> Self {
+        Self {
+            handles: Mutex::new(Some(Default::default())),
+        }
+    }
+}
+
+/// Abstract view of [`crate::tenant::mgr`], for testability.
+pub(crate) trait TenantManager<T: Types> {
+    /// Invoked by [`Cache::get`] to resolve a [`ShardTimelineId`] to a [`Types::Timeline`].
+    /// Errors are returned as [`GetError::TenantManager`].
+    async fn resolve(
+        &self,
+        timeline_id: TimelineId,
+        shard_selector: ShardSelector,
+    ) -> Result<T::Timeline, T::TenantManagerError>;
+}
+
+/// Abstract view of an [`Arc<Timeline>`], for testability.
+pub(crate) trait ArcTimeline<T: Types>: Clone {
+    fn gate(&self) -> &utils::sync::gate::Gate;
+    fn shard_timeline_id(&self) -> ShardTimelineId;
+    fn get_shard_identity(&self) -> &ShardIdentity;
+    fn per_timeline_state(&self) -> &PerTimelineState<T>;
+}
+
+/// Errors returned by [`Cache::get`].
+#[derive(Debug)]
+pub(crate) enum GetError<T: Types> {
+    TenantManager(T::TenantManagerError),
+    TimelineGateClosed,
+    PerTimelineStateShutDown,
+}
+
+/// Internal type used in [`Cache::get`].
+enum RoutingResult<T: Types> {
+    FastPath(Handle<T>),
+    SlowPath(ShardTimelineId),
+    NeedConsultTenantManager,
+}
+
+impl<T: Types> Cache<T> {
+    /// See module-level comment for details.
+    ///
+    /// Does NOT check for the shutdown state of [`Types::Timeline`].
+    /// Instead, the methods of [`Types::Timeline`] that are invoked through
+    /// the [`Handle`] are responsible for checking these conditions
+    /// and if so, return an error that causes the page service to
+    /// close the connection.
+    #[instrument(level = "trace", skip_all)]
+    pub(crate) async fn get(
+        &mut self,
+        timeline_id: TimelineId,
+        shard_selector: ShardSelector,
+        tenant_manager: &T::TenantManager,
+    ) -> Result<Handle<T>, GetError<T>> {
+        // terminates because each iteration removes an element from the map
+        loop {
+            let handle = self
+                .get_impl(timeline_id, shard_selector, tenant_manager)
+                .await?;
+            if handle.0.shut_down.load(Ordering::Relaxed) {
+                let removed = self
+                    .map
+                    .remove(&handle.0.timeline.shard_timeline_id())
+                    .expect("invariant of get_impl is that the returned handle is in the map");
+                assert!(
+                    Weak::ptr_eq(&removed, &Arc::downgrade(&handle.0)),
+                    "shard_timeline_id() incorrect?"
+                );
+            } else {
+                return Ok(handle);
+            }
+        }
+    }
+
+    #[instrument(level = "trace", skip_all)]
+    async fn get_impl(
+        &mut self,
+        timeline_id: TimelineId,
+        shard_selector: ShardSelector,
+        tenant_manager: &T::TenantManager,
+    ) -> Result<Handle<T>, GetError<T>> {
+        let miss: ShardSelector = {
+            let routing_state = self.shard_routing(timeline_id, shard_selector);
+            match routing_state {
+                RoutingResult::FastPath(handle) => return Ok(handle),
+                RoutingResult::SlowPath(key) => match self.map.get(&key) {
+                    Some(cached) => match cached.upgrade() {
+                        Some(upgraded) => return Ok(Handle(upgraded)),
+                        None => {
+                            trace!("handle cache stale");
+                            self.map.remove(&key).unwrap();
+                            ShardSelector::Known(key.shard_index)
+                        }
+                    },
+                    None => ShardSelector::Known(key.shard_index),
+                },
+                RoutingResult::NeedConsultTenantManager => shard_selector,
+            }
+        };
+        self.get_miss(timeline_id, miss, tenant_manager).await
+    }
+
+    #[inline(always)]
+    fn shard_routing(
+        &mut self,
+        timeline_id: TimelineId,
+        shard_selector: ShardSelector,
+    ) -> RoutingResult<T> {
+        loop {
+            // terminates because when every iteration we remove an element from the map
+            let Some((first_key, first_handle)) = self.map.iter().next() else {
+                return RoutingResult::NeedConsultTenantManager;
+            };
+            let Some(first_handle) = first_handle.upgrade() else {
+                // TODO: dedup with get()
+                trace!("handle cache stale");
+                let first_key_owned = *first_key;
+                self.map.remove(&first_key_owned).unwrap();
+                continue;
+            };
+
+            let first_handle_shard_identity = first_handle.timeline.get_shard_identity();
+            let make_shard_index = |shard_num: ShardNumber| ShardIndex {
+                shard_number: shard_num,
+                shard_count: first_handle_shard_identity.count,
+            };
+
+            let need_idx = match shard_selector {
+                ShardSelector::Page(key) => {
+                    make_shard_index(first_handle_shard_identity.get_shard_number(&key))
+                }
+                ShardSelector::Zero => make_shard_index(ShardNumber(0)),
+                ShardSelector::Known(shard_idx) => shard_idx,
+            };
+            let need_shard_timeline_id = ShardTimelineId {
+                shard_index: need_idx,
+                timeline_id,
+            };
+            let first_handle_shard_timeline_id = ShardTimelineId {
+                shard_index: first_handle_shard_identity.shard_index(),
+                timeline_id: first_handle.timeline.shard_timeline_id().timeline_id,
+            };
+
+            if need_shard_timeline_id == first_handle_shard_timeline_id {
+                return RoutingResult::FastPath(Handle(first_handle));
+            } else {
+                return RoutingResult::SlowPath(need_shard_timeline_id);
+            }
+        }
+    }
+
+    #[instrument(level = "trace", skip_all)]
+    #[inline(always)]
+    async fn get_miss(
+        &mut self,
+        timeline_id: TimelineId,
+        shard_selector: ShardSelector,
+        tenant_manager: &T::TenantManager,
+    ) -> Result<Handle<T>, GetError<T>> {
+        match tenant_manager.resolve(timeline_id, shard_selector).await {
+            Ok(timeline) => {
+                let key = timeline.shard_timeline_id();
+                match &shard_selector {
+                    ShardSelector::Zero => assert_eq!(key.shard_index.shard_number, ShardNumber(0)),
+                    ShardSelector::Page(_) => (), // gotta trust tenant_manager
+                    ShardSelector::Known(idx) => assert_eq!(idx, &key.shard_index),
+                }
+
+                let gate_guard = match timeline.gate().enter() {
+                    Ok(guard) => guard,
+                    Err(_) => {
+                        return Err(GetError::TimelineGateClosed);
+                    }
+                };
+                trace!("creating new HandleInner");
+                let handle = Arc::new(
+                    // TODO: global metric that keeps track of the number of live HandlerTimeline instances
+                    // so we can identify reference cycle bugs.
+                    HandleInner {
+                        shut_down: AtomicBool::new(false),
+                        _gate_guard: gate_guard,
+                        timeline: timeline.clone(),
+                    },
+                );
+                let handle = {
+                    let mut lock_guard = timeline
+                        .per_timeline_state()
+                        .handles
+                        .lock()
+                        .expect("mutex poisoned");
+                    match &mut *lock_guard {
+                        Some(per_timeline_state) => {
+                            let replaced = per_timeline_state.insert(self.id, Arc::clone(&handle));
+                            assert!(replaced.is_none(), "some earlier code left a stale handle");
+                            match self.map.entry(key) {
+                                hash_map::Entry::Occupied(_o) => {
+                                    // This cannot not happen because
+                                    // 1. we're the _miss_ handle, i.e., `self.map` didn't contain an entry and
+                                    // 2. we were holding &mut self during .resolve().await above, so, no other thread can have inserted a handle
+                                    //    while we were waiting for the tenant manager.
+                                    unreachable!()
+                                }
+                                hash_map::Entry::Vacant(v) => {
+                                    v.insert(Arc::downgrade(&handle));
+                                    handle
+                                }
+                            }
+                        }
+                        None => {
+                            return Err(GetError::PerTimelineStateShutDown);
+                        }
+                    }
+                };
+                Ok(Handle(handle))
+            }
+            Err(e) => Err(GetError::TenantManager(e)),
+        }
+    }
+}
+
+impl<T: Types> PerTimelineState<T> {
+    /// After this method returns, [`Cache::get`] will never again return a [`Handle`]
+    /// to the [`Types::Timeline`] that embeds this per-timeline state.
+    /// Even if [`TenantManager::resolve`] would still resolve to it.
+    ///
+    /// Already-alive [`Handle`]s for will remain open, usable, and keeping the [`ArcTimeline`] alive.
+    /// That's ok because they're short-lived. See module-level comment for details.
+    #[instrument(level = "trace", skip_all)]
+    pub(super) fn shutdown(&self) {
+        let handles = self
+            .handles
+            .lock()
+            .expect("mutex poisoned")
+            // NB: this .take() sets locked to None.
+            // That's what makes future `Cache::get` misses fail.
+            // Cache hits are taken care of below.
+            .take();
+        let Some(handles) = handles else {
+            trace!("already shut down");
+            return;
+        };
+        for handle in handles.values() {
+            // Make hits fail.
+            handle.shut_down.store(true, Ordering::Relaxed);
+        }
+        drop(handles);
+    }
+}
+
+impl<T: Types> std::ops::Deref for Handle<T> {
+    type Target = T::Timeline;
+    fn deref(&self) -> &Self::Target {
+        &self.0.timeline
+    }
+}
+
+#[cfg(test)]
+impl<T: Types> Drop for HandleInner<T> {
+    fn drop(&mut self) {
+        trace!("HandleInner dropped");
+    }
+}
+
+// When dropping a [`Cache`], prune its handles in the [`PerTimelineState`] to break the reference cycle.
+impl<T: Types> Drop for Cache<T> {
+    fn drop(&mut self) {
+        for (_, weak) in self.map.drain() {
+            if let Some(strong) = weak.upgrade() {
+                // handle is still being kept alive in PerTimelineState
+                let timeline = strong.timeline.per_timeline_state();
+                let mut handles = timeline.handles.lock().expect("mutex poisoned");
+                if let Some(handles) = &mut *handles {
+                    let Some(removed) = handles.remove(&self.id) else {
+                        // There could have been a shutdown inbetween us upgrading the weak and locking the mutex.
+                        continue;
+                    };
+                    assert!(Arc::ptr_eq(&removed, &strong));
+                }
+            }
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use pageserver_api::{
+        key::{rel_block_to_key, Key, DBDIR_KEY},
+        models::ShardParameters,
+        reltag::RelTag,
+        shard::ShardStripeSize,
+    };
+    use utils::shard::ShardCount;
+
+    use super::*;
+
+    const FOREVER: std::time::Duration = std::time::Duration::from_secs(u64::MAX);
+
+    #[derive(Debug)]
+    struct TestTypes;
+    impl Types for TestTypes {
+        type TenantManagerError = anyhow::Error;
+        type TenantManager = StubManager;
+        type Timeline = Arc<StubTimeline>;
+    }
+
+    struct StubManager {
+        shards: Vec<Arc<StubTimeline>>,
+    }
+
+    struct StubTimeline {
+        gate: utils::sync::gate::Gate,
+        id: TimelineId,
+        shard: ShardIdentity,
+        per_timeline_state: PerTimelineState<TestTypes>,
+        myself: Weak<StubTimeline>,
+    }
+
+    impl StubTimeline {
+        fn getpage(&self) {
+            // do nothing
+        }
+    }
+
+    impl ArcTimeline<TestTypes> for Arc<StubTimeline> {
+        fn gate(&self) -> &utils::sync::gate::Gate {
+            &self.gate
+        }
+
+        fn shard_timeline_id(&self) -> ShardTimelineId {
+            ShardTimelineId {
+                shard_index: self.shard.shard_index(),
+                timeline_id: self.id,
+            }
+        }
+
+        fn get_shard_identity(&self) -> &ShardIdentity {
+            &self.shard
+        }
+
+        fn per_timeline_state(&self) -> &PerTimelineState<TestTypes> {
+            &self.per_timeline_state
+        }
+    }
+
+    impl TenantManager<TestTypes> for StubManager {
+        async fn resolve(
+            &self,
+            timeline_id: TimelineId,
+            shard_selector: ShardSelector,
+        ) -> anyhow::Result<Arc<StubTimeline>> {
+            for timeline in &self.shards {
+                if timeline.id == timeline_id {
+                    match &shard_selector {
+                        ShardSelector::Zero if timeline.shard.is_shard_zero() => {
+                            return Ok(Arc::clone(timeline));
+                        }
+                        ShardSelector::Zero => continue,
+                        ShardSelector::Page(key) if timeline.shard.is_key_local(key) => {
+                            return Ok(Arc::clone(timeline));
+                        }
+                        ShardSelector::Page(_) => continue,
+                        ShardSelector::Known(idx) if idx == &timeline.shard.shard_index() => {
+                            return Ok(Arc::clone(timeline));
+                        }
+                        ShardSelector::Known(_) => continue,
+                    }
+                }
+            }
+            anyhow::bail!("not found")
+        }
+    }
+
+    #[tokio::test(start_paused = true)]
+    async fn test_timeline_shutdown() {
+        crate::tenant::harness::setup_logging();
+
+        let timeline_id = TimelineId::generate();
+        let shard0 = Arc::new_cyclic(|myself| StubTimeline {
+            gate: Default::default(),
+            id: timeline_id,
+            shard: ShardIdentity::unsharded(),
+            per_timeline_state: PerTimelineState::default(),
+            myself: myself.clone(),
+        });
+        let mgr = StubManager {
+            shards: vec![shard0.clone()],
+        };
+        let key = DBDIR_KEY;
+
+        let mut cache = Cache::<TestTypes>::default();
+
+        //
+        // fill the cache
+        //
+        assert_eq!(
+            (Arc::strong_count(&shard0), Arc::weak_count(&shard0)),
+            (2, 1),
+            "strong: shard0, mgr; weak: myself"
+        );
+
+        let handle: Handle<_> = cache
+            .get(timeline_id, ShardSelector::Page(key), &mgr)
+            .await
+            .expect("we have the timeline");
+        let handle_inner_weak = Arc::downgrade(&handle.0);
+        assert!(Weak::ptr_eq(&handle.myself, &shard0.myself));
+        assert_eq!(
+            (
+                Weak::strong_count(&handle_inner_weak),
+                Weak::weak_count(&handle_inner_weak)
+            ),
+            (2, 2),
+            "strong: handle, per_timeline_state, weak: handle_inner_weak, cache"
+        );
+        assert_eq!(cache.map.len(), 1);
+
+        assert_eq!(
+            (Arc::strong_count(&shard0), Arc::weak_count(&shard0)),
+            (3, 1),
+            "strong: handleinner(per_timeline_state), shard0, mgr; weak: myself"
+        );
+        drop(handle);
+        assert_eq!(
+            (Arc::strong_count(&shard0), Arc::weak_count(&shard0)),
+            (3, 1),
+            "strong: handleinner(per_timeline_state), shard0, mgr; weak: myself"
+        );
+
+        //
+        // demonstrate that Handle holds up gate closure
+        // but shutdown prevents new handles from being handed out
+        //
+
+        tokio::select! {
+            _ = shard0.gate.close() => {
+                panic!("cache and per-timeline handler state keep cache open");
+            }
+            _ = tokio::time::sleep(FOREVER) => {
+                // NB: first poll of close() makes it enter closing state
+            }
+        }
+
+        let handle = cache
+            .get(timeline_id, ShardSelector::Page(key), &mgr)
+            .await
+            .expect("we have the timeline");
+        assert!(Weak::ptr_eq(&handle.myself, &shard0.myself));
+
+        // SHUTDOWN
+        shard0.per_timeline_state.shutdown(); // keeping handle alive across shutdown
+
+        assert_eq!(
+            1,
+            Weak::strong_count(&handle_inner_weak),
+            "through local var handle"
+        );
+        assert_eq!(
+            cache.map.len(),
+            1,
+            "this is an implementation detail but worth pointing out: we can't clear the cache from shutdown(), it's cleared on first access after"
+        );
+        assert_eq!(
+            (Arc::strong_count(&shard0), Arc::weak_count(&shard0)),
+            (3, 1),
+            "strong: handleinner(via handle), shard0, mgr; weak: myself"
+        );
+
+        // this handle is perfectly usable
+        handle.getpage();
+
+        cache
+            .get(timeline_id, ShardSelector::Page(key), &mgr)
+            .await
+            .err()
+            .expect("documented behavior: can't get new handle after shutdown, even if there is an alive Handle");
+        assert_eq!(
+            cache.map.len(),
+            0,
+            "first access after shutdown cleans up the Weak's from the cache"
+        );
+
+        tokio::select! {
+            _ = shard0.gate.close() => {
+                panic!("handle is keeping gate open");
+            }
+            _ = tokio::time::sleep(FOREVER) => { }
+        }
+
+        drop(handle);
+        assert_eq!(
+            0,
+            Weak::strong_count(&handle_inner_weak),
+            "the HandleInner destructor already ran"
+        );
+        assert_eq!(
+            (Arc::strong_count(&shard0), Arc::weak_count(&shard0)),
+            (2, 1),
+            "strong: shard0, mgr; weak: myself"
+        );
+
+        // closing gate succeeds after dropping handle
+        tokio::select! {
+            _ = shard0.gate.close() => { }
+            _ = tokio::time::sleep(FOREVER) => {
+                panic!("handle is dropped, no other gate holders exist")
+            }
+        }
+
+        // map gets cleaned on next lookup
+        cache
+            .get(timeline_id, ShardSelector::Page(key), &mgr)
+            .await
+            .err()
+            .expect("documented behavior: can't get new handle after shutdown");
+        assert_eq!(cache.map.len(), 0);
+
+        // ensure all refs to shard0 are gone and we're not leaking anything
+        let myself = Weak::clone(&shard0.myself);
+        drop(shard0);
+        drop(mgr);
+        assert_eq!(Weak::strong_count(&myself), 0);
+    }
+
+    #[tokio::test]
+    async fn test_multiple_timelines_and_deletion() {
+        crate::tenant::harness::setup_logging();
+
+        let timeline_a = TimelineId::generate();
+        let timeline_b = TimelineId::generate();
+        assert_ne!(timeline_a, timeline_b);
+        let timeline_a = Arc::new_cyclic(|myself| StubTimeline {
+            gate: Default::default(),
+            id: timeline_a,
+            shard: ShardIdentity::unsharded(),
+            per_timeline_state: PerTimelineState::default(),
+            myself: myself.clone(),
+        });
+        let timeline_b = Arc::new_cyclic(|myself| StubTimeline {
+            gate: Default::default(),
+            id: timeline_b,
+            shard: ShardIdentity::unsharded(),
+            per_timeline_state: PerTimelineState::default(),
+            myself: myself.clone(),
+        });
+        let mut mgr = StubManager {
+            shards: vec![timeline_a.clone(), timeline_b.clone()],
+        };
+        let key = DBDIR_KEY;
+
+        let mut cache = Cache::<TestTypes>::default();
+
+        cache
+            .get(timeline_a.id, ShardSelector::Page(key), &mgr)
+            .await
+            .expect("we have it");
+        cache
+            .get(timeline_b.id, ShardSelector::Page(key), &mgr)
+            .await
+            .expect("we have it");
+        assert_eq!(cache.map.len(), 2);
+
+        // delete timeline A
+        timeline_a.per_timeline_state.shutdown();
+        mgr.shards.retain(|t| t.id != timeline_a.id);
+        assert!(
+            mgr.resolve(timeline_a.id, ShardSelector::Page(key))
+                .await
+                .is_err(),
+            "broken StubManager implementation"
+        );
+
+        assert_eq!(
+            cache.map.len(),
+            2,
+            "cache still has a Weak handle to Timeline A"
+        );
+        cache
+            .get(timeline_a.id, ShardSelector::Page(key), &mgr)
+            .await
+            .err()
+            .expect("documented behavior: can't get new handle after shutdown");
+        assert_eq!(cache.map.len(), 1, "next access cleans up the cache");
+
+        cache
+            .get(timeline_b.id, ShardSelector::Page(key), &mgr)
+            .await
+            .expect("we still have it");
+    }
+
+    fn make_relation_key_for_shard(shard: ShardNumber, params: &ShardParameters) -> Key {
+        rel_block_to_key(
+            RelTag {
+                spcnode: 1663,
+                dbnode: 208101,
+                relnode: 2620,
+                forknum: 0,
+            },
+            shard.0 as u32 * params.stripe_size.0,
+        )
+    }
+
+    #[tokio::test(start_paused = true)]
+    async fn test_shard_split() {
+        crate::tenant::harness::setup_logging();
+        let timeline_id = TimelineId::generate();
+        let parent = Arc::new_cyclic(|myself| StubTimeline {
+            gate: Default::default(),
+            id: timeline_id,
+            shard: ShardIdentity::unsharded(),
+            per_timeline_state: PerTimelineState::default(),
+            myself: myself.clone(),
+        });
+        let child_params = ShardParameters {
+            count: ShardCount(2),
+            stripe_size: ShardStripeSize::default(),
+        };
+        let child0 = Arc::new_cyclic(|myself| StubTimeline {
+            gate: Default::default(),
+            id: timeline_id,
+            shard: ShardIdentity::from_params(ShardNumber(0), &child_params),
+            per_timeline_state: PerTimelineState::default(),
+            myself: myself.clone(),
+        });
+        let child1 = Arc::new_cyclic(|myself| StubTimeline {
+            gate: Default::default(),
+            id: timeline_id,
+            shard: ShardIdentity::from_params(ShardNumber(1), &child_params),
+            per_timeline_state: PerTimelineState::default(),
+            myself: myself.clone(),
+        });
+        let child_shards_by_shard_number = [child0.clone(), child1.clone()];
+
+        let mut cache = Cache::<TestTypes>::default();
+
+        // fill the cache with the parent
+        for i in 0..2 {
+            let handle = cache
+                .get(
+                    timeline_id,
+                    ShardSelector::Page(make_relation_key_for_shard(ShardNumber(i), &child_params)),
+                    &StubManager {
+                        shards: vec![parent.clone()],
+                    },
+                )
+                .await
+                .expect("we have it");
+            assert!(
+                Weak::ptr_eq(&handle.myself, &parent.myself),
+                "mgr returns parent first"
+            );
+            drop(handle);
+        }
+
+        //
+        // SHARD SPLIT: tenant manager changes, but the cache isn't informed
+        //
+
+        // while we haven't shut down the parent, the cache will return the cached parent, even
+        // if the tenant manager returns the child
+        for i in 0..2 {
+            let handle = cache
+                .get(
+                    timeline_id,
+                    ShardSelector::Page(make_relation_key_for_shard(ShardNumber(i), &child_params)),
+                    &StubManager {
+                        shards: vec![], // doesn't matter what's in here, the cache is fully loaded
+                    },
+                )
+                .await
+                .expect("we have it");
+            assert!(
+                Weak::ptr_eq(&handle.myself, &parent.myself),
+                "mgr returns parent"
+            );
+            drop(handle);
+        }
+
+        let parent_handle = cache
+            .get(
+                timeline_id,
+                ShardSelector::Page(make_relation_key_for_shard(ShardNumber(0), &child_params)),
+                &StubManager {
+                    shards: vec![parent.clone()],
+                },
+            )
+            .await
+            .expect("we have it");
+        assert!(Weak::ptr_eq(&parent_handle.myself, &parent.myself));
+
+        // invalidate the cache
+        parent.per_timeline_state.shutdown();
+
+        // the cache will now return the child, even though the parent handle still exists
+        for i in 0..2 {
+            let handle = cache
+                .get(
+                    timeline_id,
+                    ShardSelector::Page(make_relation_key_for_shard(ShardNumber(i), &child_params)),
+                    &StubManager {
+                        shards: vec![child0.clone(), child1.clone()], // <====== this changed compared to previous loop
+                    },
+                )
+                .await
+                .expect("we have it");
+            assert!(
+                Weak::ptr_eq(
+                    &handle.myself,
+                    &child_shards_by_shard_number[i as usize].myself
+                ),
+                "mgr returns child"
+            );
+            drop(handle);
+        }
+
+        // all the while the parent handle kept the parent gate open
+        tokio::select! {
+            _ = parent_handle.gate.close() => {
+                panic!("parent handle is keeping gate open");
+            }
+            _ = tokio::time::sleep(FOREVER) => { }
+        }
+        drop(parent_handle);
+        tokio::select! {
+            _ = parent.gate.close() => { }
+            _ = tokio::time::sleep(FOREVER) => {
+                panic!("parent handle is dropped, no other gate holders exist")
+            }
+        }
+    }
+
+    #[tokio::test(start_paused = true)]
+    async fn test_connection_handler_exit() {
+        crate::tenant::harness::setup_logging();
+        let timeline_id = TimelineId::generate();
+        let shard0 = Arc::new_cyclic(|myself| StubTimeline {
+            gate: Default::default(),
+            id: timeline_id,
+            shard: ShardIdentity::unsharded(),
+            per_timeline_state: PerTimelineState::default(),
+            myself: myself.clone(),
+        });
+        let mgr = StubManager {
+            shards: vec![shard0.clone()],
+        };
+        let key = DBDIR_KEY;
+
+        // Simulate 10 connections that's opened, used, and closed
+        let mut used_handles = vec![];
+        for _ in 0..10 {
+            let mut cache = Cache::<TestTypes>::default();
+            let handle = {
+                let handle = cache
+                    .get(timeline_id, ShardSelector::Page(key), &mgr)
+                    .await
+                    .expect("we have the timeline");
+                assert!(Weak::ptr_eq(&handle.myself, &shard0.myself));
+                handle
+            };
+            handle.getpage();
+            used_handles.push(Arc::downgrade(&handle.0));
+        }
+
+        // No handles exist, thus gates are closed and don't require shutdown
+        assert!(used_handles
+            .iter()
+            .all(|weak| Weak::strong_count(weak) == 0));
+
+        // ... thus the gate should close immediately, even without shutdown
+        tokio::select! {
+            _ = shard0.gate.close() => { }
+            _ = tokio::time::sleep(FOREVER) => {
+                panic!("handle is dropped, no other gate holders exist")
+            }
+        }
+    }
+}

pageserver/src/tenant/timeline/layer_manager.rs

@@ -1,4 +1,4 @@
-use anyhow::{bail, ensure, Context, Result};
+use anyhow::{bail, ensure, Context};
 use itertools::Itertools;
 use pageserver_api::shard::TenantShardId;
 use std::{collections::HashMap, sync::Arc};
@@ -24,35 +24,142 @@ use crate::{
 use super::TimelineWriterState;
 
 /// Provides semantic APIs to manipulate the layer map.
-#[derive(Default)]
-pub(crate) struct LayerManager {
-    layer_map: LayerMap,
-    layer_fmgr: LayerFileManager<Layer>,
+pub(crate) enum LayerManager {
+    /// Open as in not shutdown layer manager; we still have in-memory layers and we can manipulate
+    /// the layers.
+    Open(OpenLayerManager),
+    /// Shutdown layer manager where there are no more in-memory layers and persistent layers are
+    /// read-only.
+    Closed {
+        layers: HashMap<PersistentLayerKey, Layer>,
+    },
+}
+
+impl Default for LayerManager {
+    fn default() -> Self {
+        LayerManager::Open(OpenLayerManager::default())
+    }
 }
 
 impl LayerManager {
+    pub(crate) fn get_from_key(&self, key: &PersistentLayerKey) -> Layer {
+        // The assumption for the `expect()` is that all code maintains the following invariant:
+        // A layer's descriptor is present in the LayerMap => the LayerFileManager contains a layer for the descriptor.
+        self.layers()
+            .get(key)
+            .with_context(|| format!("get layer from key: {key}"))
+            .expect("not found")
+            .clone()
+    }
+
     pub(crate) fn get_from_desc(&self, desc: &PersistentLayerDesc) -> Layer {
-        self.layer_fmgr.get_from_desc(desc)
+        self.get_from_key(&desc.key())
     }
 
     /// Get an immutable reference to the layer map.
     ///
     /// We expect users only to be able to get an immutable layer map. If users want to make modifications,
     /// they should use the below semantic APIs. This design makes us step closer to immutable storage state.
-    pub(crate) fn layer_map(&self) -> &LayerMap {
-        &self.layer_map
+    pub(crate) fn layer_map(&self) -> Result<&LayerMap, Shutdown> {
+        use LayerManager::*;
+        match self {
+            Open(OpenLayerManager { layer_map, .. }) => Ok(layer_map),
+            Closed { .. } => Err(Shutdown),
+        }
     }
 
+    pub(crate) fn open_mut(&mut self) -> Result<&mut OpenLayerManager, Shutdown> {
+        use LayerManager::*;
+
+        match self {
+            Open(open) => Ok(open),
+            Closed { .. } => Err(Shutdown),
+        }
+    }
+
+    /// LayerManager shutdown. The in-memory layers do cleanup on drop, so we must drop them in
+    /// order to allow shutdown to complete.
+    ///
+    /// If there was a want to flush in-memory layers, it must have happened earlier.
+    pub(crate) fn shutdown(&mut self, writer_state: &mut Option<TimelineWriterState>) {
+        use LayerManager::*;
+        match self {
+            Open(OpenLayerManager {
+                layer_map,
+                layer_fmgr: LayerFileManager(hashmap),
+            }) => {
+                let open = layer_map.open_layer.take();
+                let frozen = layer_map.frozen_layers.len();
+                let taken_writer_state = writer_state.take();
+                tracing::info!(open = open.is_some(), frozen, "dropped inmemory layers");
+                let layers = std::mem::take(hashmap);
+                *self = Closed { layers };
+                assert_eq!(open.is_some(), taken_writer_state.is_some());
+            }
+            Closed { .. } => {
+                tracing::debug!("ignoring multiple shutdowns on layer manager")
+            }
+        }
+    }
+
+    /// Sum up the historic layer sizes
+    pub(crate) fn layer_size_sum(&self) -> u64 {
+        self.layers()
+            .values()
+            .map(|l| l.layer_desc().file_size)
+            .sum()
+    }
+
+    pub(crate) fn likely_resident_layers(&self) -> impl Iterator<Item = &'_ Layer> + '_ {
+        self.layers().values().filter(|l| l.is_likely_resident())
+    }
+
+    pub(crate) fn contains(&self, layer: &Layer) -> bool {
+        self.contains_key(&layer.layer_desc().key())
+    }
+
+    pub(crate) fn contains_key(&self, key: &PersistentLayerKey) -> bool {
+        self.layers().contains_key(key)
+    }
+
+    pub(crate) fn all_persistent_layers(&self) -> Vec<PersistentLayerKey> {
+        self.layers().keys().cloned().collect_vec()
+    }
+
+    fn layers(&self) -> &HashMap<PersistentLayerKey, Layer> {
+        use LayerManager::*;
+        match self {
+            Open(OpenLayerManager { layer_fmgr, .. }) => &layer_fmgr.0,
+            Closed { layers } => layers,
+        }
+    }
+}
+
+#[derive(Default)]
+pub(crate) struct OpenLayerManager {
+    layer_map: LayerMap,
+    layer_fmgr: LayerFileManager<Layer>,
+}
+
+impl std::fmt::Debug for OpenLayerManager {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.debug_struct("OpenLayerManager")
+            .field("layer_count", &self.layer_fmgr.0.len())
+            .finish()
+    }
+}
+
+#[derive(Debug, thiserror::Error)]
+#[error("layer manager has been shutdown")]
+pub(crate) struct Shutdown;
+
+impl OpenLayerManager {
     /// Called from `load_layer_map`. Initialize the layer manager with:
     /// 1. all on-disk layers
     /// 2. next open layer (with disk disk_consistent_lsn LSN)
-    pub(crate) fn initialize_local_layers(
-        &mut self,
-        on_disk_layers: Vec<Layer>,
-        next_open_layer_at: Lsn,
-    ) {
+    pub(crate) fn initialize_local_layers(&mut self, layers: Vec<Layer>, next_open_layer_at: Lsn) {
         let mut updates = self.layer_map.batch_update();
-        for layer in on_disk_layers {
+        for layer in layers {
             Self::insert_historic_layer(layer, &mut updates, &mut self.layer_fmgr);
         }
         updates.flush();
@@ -64,26 +171,19 @@ impl LayerManager {
         self.layer_map.next_open_layer_at = Some(next_open_layer_at);
     }
 
-    /// Open a new writable layer to append data if there is no open layer, otherwise return the current open layer,
-    /// called within `get_layer_for_write`.
+    /// Open a new writable layer to append data if there is no open layer, otherwise return the
+    /// current open layer, called within `get_layer_for_write`.
     pub(crate) async fn get_layer_for_write(
         &mut self,
         lsn: Lsn,
-        last_record_lsn: Lsn,
         conf: &'static PageServerConf,
         timeline_id: TimelineId,
         tenant_shard_id: TenantShardId,
+        gate_guard: utils::sync::gate::GateGuard,
         ctx: &RequestContext,
-    ) -> Result<Arc<InMemoryLayer>> {
+    ) -> anyhow::Result<Arc<InMemoryLayer>> {
         ensure!(lsn.is_aligned());
 
-        ensure!(
-            lsn > last_record_lsn,
-            "cannot modify relation after advancing last_record_lsn (incoming_lsn={}, last_record_lsn={})",
-            lsn,
-            last_record_lsn,
-        );
-
         // Do we have a layer open for writing already?
         let layer = if let Some(open_layer) = &self.layer_map.open_layer {
             if open_layer.get_lsn_range().start > lsn {
@@ -109,8 +209,15 @@ impl LayerManager {
                 lsn
             );
 
-            let new_layer =
-                InMemoryLayer::create(conf, timeline_id, tenant_shard_id, start_lsn, ctx).await?;
+            let new_layer = InMemoryLayer::create(
+                conf,
+                timeline_id,
+                tenant_shard_id,
+                start_lsn,
+                gate_guard,
+                ctx,
+            )
+            .await?;
             let layer = Arc::new(new_layer);
 
             self.layer_map.open_layer = Some(layer.clone());
@@ -164,7 +271,7 @@ impl LayerManager {
         froze
     }
 
-    /// Add image layers to the layer map, called from `create_image_layers`.
+    /// Add image layers to the layer map, called from [`super::Timeline::create_image_layers`].
     pub(crate) fn track_new_image_layers(
         &mut self,
         image_layers: &[ResidentLayer],
@@ -237,7 +344,7 @@ impl LayerManager {
         self.finish_compact_l0(compact_from, compact_to, metrics)
     }
 
-    /// Called when compaction is completed.
+    /// Called post-compaction when some previous generation image layers were trimmed.
     pub(crate) fn rewrite_layers(
         &mut self,
         rewrite_layers: &[(Layer, ResidentLayer)],
@@ -255,13 +362,10 @@ impl LayerManager {
                 new_layer.layer_desc().lsn_range
             );
 
-            // Transfer visibilty hint from old to new layer, since the new layer covers the same key space.  This is not guaranteed to
+            // Transfer visibility hint from old to new layer, since the new layer covers the same key space.  This is not guaranteed to
             // be accurate (as the new layer may cover a different subset of the key range), but is a sensible default, and prevents
             // always marking rewritten layers as visible.
-            new_layer
-                .as_ref()
-                .access_stats()
-                .set_visibility(old_layer.access_stats().visibility());
+            new_layer.as_ref().set_visibility(old_layer.visibility());
 
             // Safety: we may never rewrite the same file in-place.  Callers are responsible
             // for ensuring that they only rewrite layers after something changes the path,
@@ -329,31 +433,6 @@ impl LayerManager {
         mapping.remove(layer);
         layer.delete_on_drop();
     }
-
-    pub(crate) fn likely_resident_layers(&self) -> impl Iterator<Item = Layer> + '_ {
-        // for small layer maps, we most likely have all resident, but for larger more are likely
-        // to be evicted assuming lots of layers correlated with longer lifespan.
-
-        self.layer_map().iter_historic_layers().filter_map(|desc| {
-            self.layer_fmgr
-                .0
-                .get(&desc.key())
-                .filter(|l| l.is_likely_resident())
-                .cloned()
-        })
-    }
-
-    pub(crate) fn contains(&self, layer: &Layer) -> bool {
-        self.layer_fmgr.contains(layer)
-    }
-
-    pub(crate) fn contains_key(&self, key: &PersistentLayerKey) -> bool {
-        self.layer_fmgr.contains_key(key)
-    }
-
-    pub(crate) fn all_persistent_layers(&self) -> Vec<PersistentLayerKey> {
-        self.layer_fmgr.0.keys().cloned().collect_vec()
-    }
 }
 
 pub(crate) struct LayerFileManager<T>(HashMap<PersistentLayerKey, T>);
@@ -365,20 +444,6 @@ impl<T> Default for LayerFileManager<T> {
 }
 
 impl<T: AsLayerDesc + Clone> LayerFileManager<T> {
-    fn get_from_desc(&self, desc: &PersistentLayerDesc) -> T {
-        // The assumption for the `expect()` is that all code maintains the following invariant:
-        // A layer's descriptor is present in the LayerMap => the LayerFileManager contains a layer for the descriptor.
-        self.0
-            .get(&desc.key())
-            .with_context(|| format!("get layer from desc: {}", desc.layer_name()))
-            .expect("not found")
-            .clone()
-    }
-
-    fn contains_key(&self, key: &PersistentLayerKey) -> bool {
-        self.0.contains_key(key)
-    }
-
     pub(crate) fn insert(&mut self, layer: T) {
         let present = self.0.insert(layer.layer_desc().key(), layer.clone());
         if present.is_some() && cfg!(debug_assertions) {
@@ -386,10 +451,6 @@ impl<T: AsLayerDesc + Clone> LayerFileManager<T> {
         }
     }
 
-    pub(crate) fn contains(&self, layer: &T) -> bool {
-        self.0.contains_key(&layer.layer_desc().key())
-    }
-
     pub(crate) fn remove(&mut self, layer: &T) {
         let present = self.0.remove(&layer.layer_desc().key());
         if present.is_none() && cfg!(debug_assertions) {

pageserver/src/tenant/timeline/logical_size.rs

@@ -122,6 +122,10 @@ impl CurrentLogicalSize {
             Self::Exact(_) => Accuracy::Exact,
         }
     }
+
+    pub(crate) fn is_exact(&self) -> bool {
+        matches!(self, Self::Exact(_))
+    }
 }
 
 impl LogicalSize {

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

@@ -335,6 +335,9 @@ pub(super) async fn handle_walreceiver_connection(
                             filtered_records += 1;
                         }
 
+                        // FIXME: this cannot be made pausable_failpoint without fixing the
+                        // failpoint library; in tests, the added amount of debugging will cause us
+                        // to timeout the tests.
                         fail_point!("walreceiver-after-ingest");
 
                         last_rec_lsn = lsn;

pageserver/src/utilization.rs

@@ -5,12 +5,17 @@
 
 use anyhow::Context;
 use std::path::Path;
+use utils::serde_percent::Percent;
 
 use pageserver_api::models::PageserverUtilization;
 
-pub(crate) fn regenerate(tenants_path: &Path) -> anyhow::Result<PageserverUtilization> {
-    // TODO: currently the http api ratelimits this to 1Hz at most, which is probably good enough
+use crate::{config::PageServerConf, tenant::mgr::TenantManager};
 
+pub(crate) fn regenerate(
+    conf: &PageServerConf,
+    tenants_path: &Path,
+    tenant_manager: &TenantManager,
+) -> anyhow::Result<PageserverUtilization> {
     let statvfs = nix::sys::statvfs::statvfs(tenants_path)
         .map_err(std::io::Error::from)
         .context("statvfs tenants directory")?;
@@ -34,16 +39,31 @@ pub(crate) fn regenerate(tenants_path: &Path) -> anyhow::Result<PageserverUtiliz
 
     let captured_at = std::time::SystemTime::now();
 
-    let doc = PageserverUtilization {
+    // Calculate aggregate utilization from tenants on this pageserver
+    let (disk_wanted_bytes, shard_count) = tenant_manager.calculate_utilization()?;
+
+    // Fetch the fraction of disk space which may be used
+    let disk_usable_pct = match conf.disk_usage_based_eviction.clone() {
+        Some(e) => e.max_usage_pct,
+        None => Percent::new(100).unwrap(),
+    };
+
+    // Express a static value for how many shards we may schedule on one node
+    const MAX_SHARDS: u32 = 20000;
+
+    let mut doc = PageserverUtilization {
         disk_usage_bytes: used,
         free_space_bytes: free,
-        // lower is better; start with a constant
-        //
-        // note that u64::MAX will be output as i64::MAX as u64, but that should not matter
-        utilization_score: u64::MAX,
+        disk_wanted_bytes,
+        disk_usable_pct,
+        shard_count,
+        max_shard_count: MAX_SHARDS,
+        utilization_score: 0,
         captured_at: utils::serde_system_time::SystemTime(captured_at),
     };
 
+    doc.refresh_score();
+
     // TODO: make utilization_score into a metric
 
     Ok(doc)

pageserver/src/virtual_file.rs

@@ -30,10 +30,12 @@ use tokio::time::Instant;
 pub use pageserver_api::models::virtual_file as api;
 pub(crate) mod io_engine;
 pub use io_engine::feature_test as io_engine_feature_test;
+pub use io_engine::io_engine_for_bench;
 pub use io_engine::FeatureTestResult as IoEngineFeatureTestResult;
 mod metadata;
 mod open_options;
 use self::owned_buffers_io::write::OwnedAsyncWriter;
+pub(crate) use api::DirectIoMode;
 pub(crate) use io_engine::IoEngineKind;
 pub(crate) use metadata::Metadata;
 pub(crate) use open_options::*;

pageserver/src/virtual_file/io_engine.rs

@@ -328,3 +328,29 @@ pub fn feature_test() -> anyhow::Result<FeatureTestResult> {
     .join()
     .unwrap()
 }
+
+/// For use in benchmark binaries only.
+///
+/// Benchmarks which initialize `virtual_file` need to know what engine to use, but we also
+/// don't want to silently fall back to slower I/O engines in a benchmark: this could waste
+/// developer time trying to figure out why it's slow.
+///
+/// In practice, this method will either return IoEngineKind::TokioEpollUring, or panic.
+pub fn io_engine_for_bench() -> IoEngineKind {
+    #[cfg(not(target_os = "linux"))]
+    {
+        panic!("This benchmark does I/O and can only give a representative result on Linux");
+    }
+    #[cfg(target_os = "linux")]
+    {
+        match feature_test().unwrap() {
+            FeatureTestResult::PlatformPreferred(engine) => engine,
+            FeatureTestResult::Worse {
+                engine: _engine,
+                remark,
+            } => {
+                panic!("This benchmark does I/O can requires the preferred I/O engine: {remark}");
+            }
+        }
+    }
+}

pageserver/src/walredo.rs

@@ -107,8 +107,10 @@ enum ProcessOnceCell {
 }
 
 struct Process {
-    _launched_processes_guard: utils::sync::gate::GateGuard,
     process: process::WalRedoProcess,
+    /// This field is last in this struct so the guard gets dropped _after_ [`Self::process`].
+    /// (Reminder: dropping [`Self::process`] synchronously sends SIGKILL and then `wait()`s for it to exit).
+    _launched_processes_guard: utils::sync::gate::GateGuard,
 }
 
 impl std::ops::Deref for Process {
@@ -241,6 +243,9 @@ impl PostgresRedoManager {
 
     /// Shut down the WAL redo manager.
     ///
+    /// Returns `true` if this call was the one that initiated shutdown.
+    /// `true` may be observed by no caller if the first caller stops polling.
+    ///
     /// After this future completes
     /// - no redo process is running
     /// - no new redo process will be spawned
@@ -250,22 +255,32 @@ impl PostgresRedoManager {
     /// # Cancel-Safety
     ///
     /// This method is cancellation-safe.
-    pub async fn shutdown(&self) {
+    pub async fn shutdown(&self) -> bool {
         // prevent new processes from being spawned
-        let permit = match self.redo_process.get_or_init_detached().await {
+        let maybe_permit = match self.redo_process.get_or_init_detached().await {
             Ok(guard) => {
-                let (proc, permit) = guard.take_and_deinit();
-                drop(proc); // this just drops the Arc, its refcount may not be zero yet
-                permit
+                if matches!(&*guard, ProcessOnceCell::ManagerShutDown) {
+                    None
+                } else {
+                    let (proc, permit) = guard.take_and_deinit();
+                    drop(proc); // this just drops the Arc, its refcount may not be zero yet
+                    Some(permit)
+                }
             }
-            Err(permit) => permit,
+            Err(permit) => Some(permit),
+        };
+        let it_was_us = if let Some(permit) = maybe_permit {
+            self.redo_process
+                .set(ProcessOnceCell::ManagerShutDown, permit);
+            true
+        } else {
+            false
         };
-        self.redo_process
-            .set(ProcessOnceCell::ManagerShutDown, permit);
         // wait for ongoing requests to drain and the refcounts of all Arc<WalRedoProcess> that
         // we ever launched to drop to zero, which when it happens synchronously kill()s & wait()s
         // for the underlying process.
         self.launched_processes.close().await;
+        it_was_us
     }
 
     /// This type doesn't have its own background task to check for idleness: we
@@ -314,20 +329,23 @@ impl PostgresRedoManager {
                 },
                 Err(permit) => {
                     let start = Instant::now();
-                    let proc = Arc::new(Process {
-                            _launched_processes_guard: match self.launched_processes.enter() {
+                    // acquire guard before spawning process, so that we don't spawn new processes
+                    // if the gate is already closed.
+                    let _launched_processes_guard = match self.launched_processes.enter() {
                                 Ok(guard) => guard,
                                 Err(GateError::GateClosed) => unreachable!(
                                     "shutdown sets the once cell to `ManagerShutDown` state before closing the gate"
                                 ),
-                            },
-                            process: process::WalRedoProcess::launch(
-                                self.conf,
-                                self.tenant_shard_id,
-                                pg_version,
-                            )
-                            .context("launch walredo process")?,
-                        });
+                            };
+                    let proc = Arc::new(Process {
+                        process: process::WalRedoProcess::launch(
+                            self.conf,
+                            self.tenant_shard_id,
+                            pg_version,
+                        )
+                        .context("launch walredo process")?,
+                        _launched_processes_guard,
+                    });
                     let duration = start.elapsed();
                     WAL_REDO_PROCESS_LAUNCH_DURATION_HISTOGRAM.observe(duration.as_secs_f64());
                     info!(

pageserver/test_data/indices/mixed_workload/README.md

@@ -0,0 +1,7 @@
+
+# This was captured from one shard of a large tenant in staging.
+
+# It has a mixture of deltas and image layers, >1000 layers in total.
+
+# This is suitable for general smoke tests that want an index which is not
+# trivially small, but doesn't contain weird/pathological cases.

pgxn/neon/control_plane_connector.c

@@ -45,6 +45,7 @@ static const char *jwt_token = NULL;
 /* GUCs */
 static char *ConsoleURL = NULL;
 static bool ForwardDDL = true;
+static bool RegressTestMode = false;
 
 /*
  * CURL docs say that this buffer must exist until we call curl_easy_cleanup
@@ -802,6 +803,14 @@ NeonProcessUtility(
 		case T_DropRoleStmt:
 			HandleDropRole(castNode(DropRoleStmt, parseTree));
 			break;
+		case T_CreateTableSpaceStmt:
+			if (!RegressTestMode)
+			{
+				ereport(ERROR,
+					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+					errmsg("CREATE TABLESPACE is not supported on Neon")));
+			}
+   			break;
 		default:
 			break;
 	}
@@ -864,6 +873,18 @@ InitControlPlaneConnector()
 							 NULL,
 							 NULL);
 
+	DefineCustomBoolVariable(
+							 "neon.regress_test_mode",
+							 "Controls whether we are running in the regression test mode",
+							 NULL,
+							 &RegressTestMode,
+							 false,
+							 PGC_SUSET,
+							 0,
+							 NULL,
+							 NULL,
+							 NULL);
+
 	jwt_token = getenv("NEON_CONTROL_PLANE_TOKEN");
 	if (!jwt_token)
 	{

pgxn/neon/neon.c

@@ -32,6 +32,7 @@
 #include "utils/builtins.h"
 #include "utils/pg_lsn.h"
 #include "utils/guc.h"
+#include "utils/guc_tables.h"
 #include "utils/wait_event.h"
 
 #include "extension_server.h"
@@ -68,10 +69,10 @@ InitLogicalReplicationMonitor(void)
 
 	DefineCustomIntVariable(
 							"neon.logical_replication_max_snap_files",
-							"Maximum allowed logical replication .snap files",
+							"Maximum allowed logical replication .snap files. When exceeded, slots are dropped until the limit is met. -1 disables the limit.",
 							NULL,
 							&logical_replication_max_snap_files,
-							300, 0, INT_MAX,
+							300, -1, INT_MAX,
 							PGC_SIGHUP,
 							0,
 							NULL, NULL, NULL);
@@ -584,6 +585,40 @@ RestoreRunningXactsFromClog(CheckPoint *checkpoint, TransactionId **xids, int *n
 	return false;
 }
 
+
+/*
+ * pgbouncer is able to track GUCs reported by Postgres.
+ * But most parameters cannot be tracked this way. The only parameters that can be tracked are ones
+ * that Postgres reports to the client. Unfortunately `search_path` is not reported by Postgres:
+ * https://www.postgresql.org/message-id/flat/CAGECzQQ6xFcgrg%2Be0p9mCumtK362TiA6vTiiZKoYbS8OXggwuQ%40mail.gmail.com#be4bfd7a9cf1f0633bdb2d1790a0a1be
+ * This code sets GUC_REPORT flag for `search_path`making it possible to include it in
+ * pgbouncer's `track_extra_parameters` list.
+ *
+ * This code is inspired by how the Citus extension does this, see
+ * https://github.com/citusdata/citus/blob/2a263fe69a707d16ef24378f7650742386b0968f/src/backend/distributed/shared_library_init.c#L2694
+ */
+static void
+ReportSearchPath(void)
+{
+#if PG_VERSION_NUM >= 160000
+	int nGucs = 0;
+	struct config_generic **gucs = get_guc_variables(&nGucs);
+#else
+	struct config_generic **gucs = get_guc_variables();
+	int nGucs = GetNumConfigOptions();
+#endif
+
+	for (int i = 0; i < nGucs; i++)
+	{
+		struct config_generic *guc = (struct config_generic *) gucs[i];
+
+		if (strcmp(guc->name, "search_path") == 0)
+		{
+			guc->flags |= GUC_REPORT;
+		}
+	}
+}
+
 void
 _PG_init(void)
 {
@@ -599,6 +634,7 @@ _PG_init(void)
 	pg_init_walproposer();
 	WalSender_Custom_XLogReaderRoutines = NeonOnDemandXLogReaderRoutines;
 	LogicalFuncs_Custom_XLogReaderRoutines = NeonOnDemandXLogReaderRoutines;
+	SlotFuncs_Custom_XLogReaderRoutines = NeonOnDemandXLogReaderRoutines;
 
 	InitLogicalReplicationMonitor();
 
@@ -626,6 +662,8 @@ _PG_init(void)
 	 * extension was loaded will be removed.
 	 */
 	EmitWarningsOnPlaceholders("neon");
+
+	ReportSearchPath();
 }
 
 PG_FUNCTION_INFO_V1(pg_cluster_size);

pgxn/neon/walproposer_pg.c

@@ -512,7 +512,7 @@ replication_feedback_get_lsns(XLogRecPtr *writeLsn, XLogRecPtr *flushLsn, XLogRe
 }
 
 /*
- * Start walsender streaming replication
+ * Start walproposer streaming replication
  */
 static void
 walprop_pg_start_streaming(WalProposer *wp, XLogRecPtr startpos)

pgxn/neon/walsender_hooks.c

@@ -20,6 +20,7 @@
 #include "utils/guc.h"
 #include "postmaster/interrupt.h"
 
+#include "neon.h"
 #include "neon_walreader.h"
 #include "walproposer.h"
 
@@ -181,6 +182,13 @@ NeonWALReadSegmentClose(XLogReaderState *xlogreader)
 void
 NeonOnDemandXLogReaderRoutines(XLogReaderRoutine *xlr)
 {
+	/*
+	 * If safekeepers are not configured, assume we don't need neon_walreader,
+	 * i.e. running neon fork locally.
+	 */
+	if (wal_acceptors_list[0] == '\0')
+		return;
+
 	if (!wal_reader)
 	{
 		XLogRecPtr	epochStartLsn = pg_atomic_read_u64(&GetWalpropShmemState()->propEpochStartLsn);

pgxn/neon_rmgr/neon_rmgr.c

@@ -186,7 +186,7 @@ static void
 fix_infomask_from_infobits(uint8 infobits, uint16 *infomask, uint16 *infomask2)
 {
 	*infomask &= ~(HEAP_XMAX_IS_MULTI | HEAP_XMAX_LOCK_ONLY |
-				   HEAP_XMAX_KEYSHR_LOCK | HEAP_XMAX_EXCL_LOCK);
+				   HEAP_XMAX_KEYSHR_LOCK | HEAP_XMAX_EXCL_LOCK | HEAP_COMBOCID);
 	*infomask2 &= ~HEAP_KEYS_UPDATED;
 
 	if (infobits & XLHL_XMAX_IS_MULTI)
@@ -195,6 +195,8 @@ fix_infomask_from_infobits(uint8 infobits, uint16 *infomask, uint16 *infomask2)
 		*infomask |= HEAP_XMAX_LOCK_ONLY;
 	if (infobits & XLHL_XMAX_EXCL_LOCK)
 		*infomask |= HEAP_XMAX_EXCL_LOCK;
+	if (infobits & XLHL_COMBOCID)
+		*infomask |= HEAP_COMBOCID;
 	/* note HEAP_XMAX_SHR_LOCK isn't considered here */
 	if (infobits & XLHL_XMAX_KEYSHR_LOCK)
 		*infomask |= HEAP_XMAX_KEYSHR_LOCK;
@@ -284,7 +286,7 @@ redo_neon_heap_insert(XLogReaderState *record)
 		htup->t_infomask = xlhdr.t_infomask;
 		htup->t_hoff = xlhdr.t_hoff;
 		HeapTupleHeaderSetXmin(htup, XLogRecGetXid(record));
-		HeapTupleHeaderSetCmin(htup, xlhdr.t_cid);
+		htup->t_choice.t_heap.t_field3.t_cid = xlhdr.t_cid;
 		htup->t_ctid = target_tid;
 
 		if (PageAddItem(page, (Item) htup, newlen, xlrec->offnum,
@@ -373,7 +375,7 @@ redo_neon_heap_delete(XLogReaderState *record)
 			HeapTupleHeaderSetXmax(htup, xlrec->xmax);
 		else
 			HeapTupleHeaderSetXmin(htup, InvalidTransactionId);
-		HeapTupleHeaderSetCmax(htup, xlrec->t_cid, false);
+		htup->t_choice.t_heap.t_field3.t_cid = xlrec->t_cid;
 
 		/* Mark the page as a candidate for pruning */
 		PageSetPrunable(page, XLogRecGetXid(record));
@@ -490,7 +492,7 @@ redo_neon_heap_update(XLogReaderState *record, bool hot_update)
 		fix_infomask_from_infobits(xlrec->old_infobits_set, &htup->t_infomask,
 								   &htup->t_infomask2);
 		HeapTupleHeaderSetXmax(htup, xlrec->old_xmax);
-		HeapTupleHeaderSetCmax(htup, xlrec->t_cid, false);
+		htup->t_choice.t_heap.t_field3.t_cid = xlrec->t_cid;
 		/* Set forward chain link in t_ctid */
 		htup->t_ctid = newtid;
 
@@ -623,7 +625,7 @@ redo_neon_heap_update(XLogReaderState *record, bool hot_update)
 		htup->t_hoff = xlhdr.t_hoff;
 
 		HeapTupleHeaderSetXmin(htup, XLogRecGetXid(record));
-		HeapTupleHeaderSetCmin(htup, xlhdr.t_cid);
+		htup->t_choice.t_heap.t_field3.t_cid = xlhdr.t_cid;
 		HeapTupleHeaderSetXmax(htup, xlrec->new_xmax);
 		/* Make sure there is no forward chain link in t_ctid */
 		htup->t_ctid = newtid;
@@ -728,7 +730,7 @@ redo_neon_heap_lock(XLogReaderState *record)
 						   offnum);
 		}
 		HeapTupleHeaderSetXmax(htup, xlrec->xmax);
-		HeapTupleHeaderSetCmax(htup, xlrec->t_cid, false);
+		htup->t_choice.t_heap.t_field3.t_cid = xlrec->t_cid;
 		PageSetLSN(page, lsn);
 		MarkBufferDirty(buffer);
 	}
@@ -840,7 +842,7 @@ redo_neon_heap_multi_insert(XLogReaderState *record)
 			htup->t_infomask = xlhdr->t_infomask;
 			htup->t_hoff = xlhdr->t_hoff;
 			HeapTupleHeaderSetXmin(htup, XLogRecGetXid(record));
-			HeapTupleHeaderSetCmin(htup, xlrec->t_cid);
+			htup->t_choice.t_heap.t_field3.t_cid = xlrec->t_cid;
 			ItemPointerSetBlockNumber(&htup->t_ctid, blkno);
 			ItemPointerSetOffsetNumber(&htup->t_ctid, offnum);
 

poetry.lock

@@ -1,91 +1,103 @@
 # This file is automatically @generated by Poetry 1.8.3 and should not be changed by hand.
 
+[[package]]
+name = "aiohappyeyeballs"
+version = "2.3.5"
+description = "Happy Eyeballs for asyncio"
+optional = false
+python-versions = ">=3.8"
+files = [
+    {file = "aiohappyeyeballs-2.3.5-py3-none-any.whl", hash = "sha256:4d6dea59215537dbc746e93e779caea8178c866856a721c9c660d7a5a7b8be03"},
+    {file = "aiohappyeyeballs-2.3.5.tar.gz", hash = "sha256:6fa48b9f1317254f122a07a131a86b71ca6946ca989ce6326fff54a99a920105"},
+]
+
 [[package]]
 name = "aiohttp"
-version = "3.9.4"
+version = "3.10.2"
 description = "Async http client/server framework (asyncio)"
 optional = false
 python-versions = ">=3.8"
 files = [
-    {file = "aiohttp-3.9.4-cp310-cp310-macosx_10_9_universal2.whl", hash = "sha256:76d32588ef7e4a3f3adff1956a0ba96faabbdee58f2407c122dd45aa6e34f372"},
-    {file = "aiohttp-3.9.4-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:56181093c10dbc6ceb8a29dfeea1e815e1dfdc020169203d87fd8d37616f73f9"},
-    {file = "aiohttp-3.9.4-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:c7a5b676d3c65e88b3aca41816bf72831898fcd73f0cbb2680e9d88e819d1e4d"},
-    {file = "aiohttp-3.9.4-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:d1df528a85fb404899d4207a8d9934cfd6be626e30e5d3a5544a83dbae6d8a7e"},
-    {file = "aiohttp-3.9.4-cp310-cp310-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:f595db1bceabd71c82e92df212dd9525a8a2c6947d39e3c994c4f27d2fe15b11"},
-    {file = "aiohttp-3.9.4-cp310-cp310-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:9c0b09d76e5a4caac3d27752027fbd43dc987b95f3748fad2b924a03fe8632ad"},
-    {file = "aiohttp-3.9.4-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:689eb4356649ec9535b3686200b231876fb4cab4aca54e3bece71d37f50c1d13"},
-    {file = "aiohttp-3.9.4-cp310-cp310-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:a3666cf4182efdb44d73602379a66f5fdfd5da0db5e4520f0ac0dcca644a3497"},
-    {file = "aiohttp-3.9.4-cp310-cp310-musllinux_1_1_aarch64.whl", hash = "sha256:b65b0f8747b013570eea2f75726046fa54fa8e0c5db60f3b98dd5d161052004a"},
-    {file = "aiohttp-3.9.4-cp310-cp310-musllinux_1_1_i686.whl", hash = "sha256:a1885d2470955f70dfdd33a02e1749613c5a9c5ab855f6db38e0b9389453dce7"},
-    {file = "aiohttp-3.9.4-cp310-cp310-musllinux_1_1_ppc64le.whl", hash = "sha256:0593822dcdb9483d41f12041ff7c90d4d1033ec0e880bcfaf102919b715f47f1"},
-    {file = "aiohttp-3.9.4-cp310-cp310-musllinux_1_1_s390x.whl", hash = "sha256:47f6eb74e1ecb5e19a78f4a4228aa24df7fbab3b62d4a625d3f41194a08bd54f"},
-    {file = "aiohttp-3.9.4-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:c8b04a3dbd54de6ccb7604242fe3ad67f2f3ca558f2d33fe19d4b08d90701a89"},
-    {file = "aiohttp-3.9.4-cp310-cp310-win32.whl", hash = "sha256:8a78dfb198a328bfb38e4308ca8167028920fb747ddcf086ce706fbdd23b2926"},
-    {file = "aiohttp-3.9.4-cp310-cp310-win_amd64.whl", hash = "sha256:e78da6b55275987cbc89141a1d8e75f5070e577c482dd48bd9123a76a96f0bbb"},
-    {file = "aiohttp-3.9.4-cp311-cp311-macosx_10_9_universal2.whl", hash = "sha256:c111b3c69060d2bafc446917534150fd049e7aedd6cbf21ba526a5a97b4402a5"},
-    {file = "aiohttp-3.9.4-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:efbdd51872cf170093998c87ccdf3cb5993add3559341a8e5708bcb311934c94"},
-    {file = "aiohttp-3.9.4-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:7bfdb41dc6e85d8535b00d73947548a748e9534e8e4fddd2638109ff3fb081df"},
-    {file = "aiohttp-3.9.4-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:2bd9d334412961125e9f68d5b73c1d0ab9ea3f74a58a475e6b119f5293eee7ba"},
-    {file = "aiohttp-3.9.4-cp311-cp311-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:35d78076736f4a668d57ade00c65d30a8ce28719d8a42471b2a06ccd1a2e3063"},
-    {file = "aiohttp-3.9.4-cp311-cp311-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:824dff4f9f4d0f59d0fa3577932ee9a20e09edec8a2f813e1d6b9f89ced8293f"},
-    {file = "aiohttp-3.9.4-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:52b8b4e06fc15519019e128abedaeb56412b106ab88b3c452188ca47a25c4093"},
-    {file = "aiohttp-3.9.4-cp311-cp311-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:eae569fb1e7559d4f3919965617bb39f9e753967fae55ce13454bec2d1c54f09"},
-    {file = "aiohttp-3.9.4-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:69b97aa5792428f321f72aeb2f118e56893371f27e0b7d05750bcad06fc42ca1"},
-    {file = "aiohttp-3.9.4-cp311-cp311-musllinux_1_1_i686.whl", hash = "sha256:4d79aad0ad4b980663316f26d9a492e8fab2af77c69c0f33780a56843ad2f89e"},
-    {file = "aiohttp-3.9.4-cp311-cp311-musllinux_1_1_ppc64le.whl", hash = "sha256:d6577140cd7db19e430661e4b2653680194ea8c22c994bc65b7a19d8ec834403"},
-    {file = "aiohttp-3.9.4-cp311-cp311-musllinux_1_1_s390x.whl", hash = "sha256:9860d455847cd98eb67897f5957b7cd69fbcb436dd3f06099230f16a66e66f79"},
-    {file = "aiohttp-3.9.4-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:69ff36d3f8f5652994e08bd22f093e11cfd0444cea310f92e01b45a4e46b624e"},
-    {file = "aiohttp-3.9.4-cp311-cp311-win32.whl", hash = "sha256:e27d3b5ed2c2013bce66ad67ee57cbf614288bda8cdf426c8d8fe548316f1b5f"},
-    {file = "aiohttp-3.9.4-cp311-cp311-win_amd64.whl", hash = "sha256:d6a67e26daa686a6fbdb600a9af8619c80a332556245fa8e86c747d226ab1a1e"},
-    {file = "aiohttp-3.9.4-cp312-cp312-macosx_10_9_universal2.whl", hash = "sha256:c5ff8ff44825736a4065d8544b43b43ee4c6dd1530f3a08e6c0578a813b0aa35"},
-    {file = "aiohttp-3.9.4-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:d12a244627eba4e9dc52cbf924edef905ddd6cafc6513849b4876076a6f38b0e"},
-    {file = "aiohttp-3.9.4-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:dcad56c8d8348e7e468899d2fb3b309b9bc59d94e6db08710555f7436156097f"},
-    {file = "aiohttp-3.9.4-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:4f7e69a7fd4b5ce419238388e55abd220336bd32212c673ceabc57ccf3d05b55"},
-    {file = "aiohttp-3.9.4-cp312-cp312-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:c4870cb049f10d7680c239b55428916d84158798eb8f353e74fa2c98980dcc0b"},
-    {file = "aiohttp-3.9.4-cp312-cp312-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:3b2feaf1b7031ede1bc0880cec4b0776fd347259a723d625357bb4b82f62687b"},
-    {file = "aiohttp-3.9.4-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:939393e8c3f0a5bcd33ef7ace67680c318dc2ae406f15e381c0054dd658397de"},
-    {file = "aiohttp-3.9.4-cp312-cp312-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:7d2334e387b2adcc944680bebcf412743f2caf4eeebd550f67249c1c3696be04"},
-    {file = "aiohttp-3.9.4-cp312-cp312-musllinux_1_1_aarch64.whl", hash = "sha256:e0198ea897680e480845ec0ffc5a14e8b694e25b3f104f63676d55bf76a82f1a"},
-    {file = "aiohttp-3.9.4-cp312-cp312-musllinux_1_1_i686.whl", hash = "sha256:e40d2cd22914d67c84824045861a5bb0fb46586b15dfe4f046c7495bf08306b2"},
-    {file = "aiohttp-3.9.4-cp312-cp312-musllinux_1_1_ppc64le.whl", hash = "sha256:aba80e77c227f4234aa34a5ff2b6ff30c5d6a827a91d22ff6b999de9175d71bd"},
-    {file = "aiohttp-3.9.4-cp312-cp312-musllinux_1_1_s390x.whl", hash = "sha256:fb68dc73bc8ac322d2e392a59a9e396c4f35cb6fdbdd749e139d1d6c985f2527"},
-    {file = "aiohttp-3.9.4-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:f3460a92638dce7e47062cf088d6e7663adb135e936cb117be88d5e6c48c9d53"},
-    {file = "aiohttp-3.9.4-cp312-cp312-win32.whl", hash = "sha256:32dc814ddbb254f6170bca198fe307920f6c1308a5492f049f7f63554b88ef36"},
-    {file = "aiohttp-3.9.4-cp312-cp312-win_amd64.whl", hash = "sha256:63f41a909d182d2b78fe3abef557fcc14da50c7852f70ae3be60e83ff64edba5"},
-    {file = "aiohttp-3.9.4-cp38-cp38-macosx_10_9_universal2.whl", hash = "sha256:c3770365675f6be220032f6609a8fbad994d6dcf3ef7dbcf295c7ee70884c9af"},
-    {file = "aiohttp-3.9.4-cp38-cp38-macosx_10_9_x86_64.whl", hash = "sha256:305edae1dea368ce09bcb858cf5a63a064f3bff4767dec6fa60a0cc0e805a1d3"},
-    {file = "aiohttp-3.9.4-cp38-cp38-macosx_11_0_arm64.whl", hash = "sha256:6f121900131d116e4a93b55ab0d12ad72573f967b100e49086e496a9b24523ea"},
-    {file = "aiohttp-3.9.4-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:b71e614c1ae35c3d62a293b19eface83d5e4d194e3eb2fabb10059d33e6e8cbf"},
-    {file = "aiohttp-3.9.4-cp38-cp38-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:419f009fa4cfde4d16a7fc070d64f36d70a8d35a90d71aa27670bba2be4fd039"},
-    {file = "aiohttp-3.9.4-cp38-cp38-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:7b39476ee69cfe64061fd77a73bf692c40021f8547cda617a3466530ef63f947"},
-    {file = "aiohttp-3.9.4-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:b33f34c9c7decdb2ab99c74be6443942b730b56d9c5ee48fb7df2c86492f293c"},
-    {file = "aiohttp-3.9.4-cp38-cp38-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:c78700130ce2dcebb1a8103202ae795be2fa8c9351d0dd22338fe3dac74847d9"},
-    {file = "aiohttp-3.9.4-cp38-cp38-musllinux_1_1_aarch64.whl", hash = "sha256:268ba22d917655d1259af2d5659072b7dc11b4e1dc2cb9662fdd867d75afc6a4"},
-    {file = "aiohttp-3.9.4-cp38-cp38-musllinux_1_1_i686.whl", hash = "sha256:17e7c051f53a0d2ebf33013a9cbf020bb4e098c4bc5bce6f7b0c962108d97eab"},
-    {file = "aiohttp-3.9.4-cp38-cp38-musllinux_1_1_ppc64le.whl", hash = "sha256:7be99f4abb008cb38e144f85f515598f4c2c8932bf11b65add0ff59c9c876d99"},
-    {file = "aiohttp-3.9.4-cp38-cp38-musllinux_1_1_s390x.whl", hash = "sha256:d58a54d6ff08d2547656356eea8572b224e6f9bbc0cf55fa9966bcaac4ddfb10"},
-    {file = "aiohttp-3.9.4-cp38-cp38-musllinux_1_1_x86_64.whl", hash = "sha256:7673a76772bda15d0d10d1aa881b7911d0580c980dbd16e59d7ba1422b2d83cd"},
-    {file = "aiohttp-3.9.4-cp38-cp38-win32.whl", hash = "sha256:e4370dda04dc8951012f30e1ce7956a0a226ac0714a7b6c389fb2f43f22a250e"},
-    {file = "aiohttp-3.9.4-cp38-cp38-win_amd64.whl", hash = "sha256:eb30c4510a691bb87081192a394fb661860e75ca3896c01c6d186febe7c88530"},
-    {file = "aiohttp-3.9.4-cp39-cp39-macosx_10_9_universal2.whl", hash = "sha256:84e90494db7df3be5e056f91412f9fa9e611fbe8ce4aaef70647297f5943b276"},
-    {file = "aiohttp-3.9.4-cp39-cp39-macosx_10_9_x86_64.whl", hash = "sha256:7d4845f8501ab28ebfdbeab980a50a273b415cf69e96e4e674d43d86a464df9d"},
-    {file = "aiohttp-3.9.4-cp39-cp39-macosx_11_0_arm64.whl", hash = "sha256:69046cd9a2a17245c4ce3c1f1a4ff8c70c7701ef222fce3d1d8435f09042bba1"},
-    {file = "aiohttp-3.9.4-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:8b73a06bafc8dcc508420db43b4dd5850e41e69de99009d0351c4f3007960019"},
-    {file = "aiohttp-3.9.4-cp39-cp39-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:418bb0038dfafeac923823c2e63226179976c76f981a2aaad0ad5d51f2229bca"},
-    {file = "aiohttp-3.9.4-cp39-cp39-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:71a8f241456b6c2668374d5d28398f8e8cdae4cce568aaea54e0f39359cd928d"},
-    {file = "aiohttp-3.9.4-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:935c369bf8acc2dc26f6eeb5222768aa7c62917c3554f7215f2ead7386b33748"},
-    {file = "aiohttp-3.9.4-cp39-cp39-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:74e4e48c8752d14ecfb36d2ebb3d76d614320570e14de0a3aa7a726ff150a03c"},
-    {file = "aiohttp-3.9.4-cp39-cp39-musllinux_1_1_aarch64.whl", hash = "sha256:916b0417aeddf2c8c61291238ce25286f391a6acb6f28005dd9ce282bd6311b6"},
-    {file = "aiohttp-3.9.4-cp39-cp39-musllinux_1_1_i686.whl", hash = "sha256:9b6787b6d0b3518b2ee4cbeadd24a507756ee703adbac1ab6dc7c4434b8c572a"},
-    {file = "aiohttp-3.9.4-cp39-cp39-musllinux_1_1_ppc64le.whl", hash = "sha256:221204dbda5ef350e8db6287937621cf75e85778b296c9c52260b522231940ed"},
-    {file = "aiohttp-3.9.4-cp39-cp39-musllinux_1_1_s390x.whl", hash = "sha256:10afd99b8251022ddf81eaed1d90f5a988e349ee7d779eb429fb07b670751e8c"},
-    {file = "aiohttp-3.9.4-cp39-cp39-musllinux_1_1_x86_64.whl", hash = "sha256:2506d9f7a9b91033201be9ffe7d89c6a54150b0578803cce5cb84a943d075bc3"},
-    {file = "aiohttp-3.9.4-cp39-cp39-win32.whl", hash = "sha256:e571fdd9efd65e86c6af2f332e0e95dad259bfe6beb5d15b3c3eca3a6eb5d87b"},
-    {file = "aiohttp-3.9.4-cp39-cp39-win_amd64.whl", hash = "sha256:7d29dd5319d20aa3b7749719ac9685fbd926f71ac8c77b2477272725f882072d"},
-    {file = "aiohttp-3.9.4.tar.gz", hash = "sha256:6ff71ede6d9a5a58cfb7b6fffc83ab5d4a63138276c771ac91ceaaddf5459644"},
+    {file = "aiohttp-3.10.2-cp310-cp310-macosx_10_9_universal2.whl", hash = "sha256:95213b3d79c7e387144e9cb7b9d2809092d6ff2c044cb59033aedc612f38fb6d"},
+    {file = "aiohttp-3.10.2-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:1aa005f060aff7124cfadaa2493f00a4e28ed41b232add5869e129a2e395935a"},
+    {file = "aiohttp-3.10.2-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:eabe6bf4c199687592f5de4ccd383945f485779c7ffb62a9b9f1f8a3f9756df8"},
+    {file = "aiohttp-3.10.2-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:96e010736fc16d21125c7e2dc5c350cd43c528b85085c04bf73a77be328fe944"},
+    {file = "aiohttp-3.10.2-cp310-cp310-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:99f81f9c1529fd8e03be4a7bd7df32d14b4f856e90ef6e9cbad3415dbfa9166c"},
+    {file = "aiohttp-3.10.2-cp310-cp310-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:d611d1a01c25277bcdea06879afbc11472e33ce842322496b211319aa95441bb"},
+    {file = "aiohttp-3.10.2-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:e00191d38156e09e8c81ef3d75c0d70d4f209b8381e71622165f22ef7da6f101"},
+    {file = "aiohttp-3.10.2-cp310-cp310-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:74c091a5ded6cb81785de2d7a8ab703731f26de910dbe0f3934eabef4ae417cc"},
+    {file = "aiohttp-3.10.2-cp310-cp310-musllinux_1_2_aarch64.whl", hash = "sha256:18186a80ec5a701816adbf1d779926e1069392cf18504528d6e52e14b5920525"},
+    {file = "aiohttp-3.10.2-cp310-cp310-musllinux_1_2_i686.whl", hash = "sha256:5a7ceb2a0d2280f23a02c64cd0afdc922079bb950400c3dd13a1ab2988428aac"},
+    {file = "aiohttp-3.10.2-cp310-cp310-musllinux_1_2_ppc64le.whl", hash = "sha256:8bd7be6ff6c162a60cb8fce65ee879a684fbb63d5466aba3fa5b9288eb04aefa"},
+    {file = "aiohttp-3.10.2-cp310-cp310-musllinux_1_2_s390x.whl", hash = "sha256:fae962b62944eaebff4f4fddcf1a69de919e7b967136a318533d82d93c3c6bd1"},
+    {file = "aiohttp-3.10.2-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:a0fde16d284efcacbe15fb0c1013f0967b6c3e379649239d783868230bf1db42"},
+    {file = "aiohttp-3.10.2-cp310-cp310-win32.whl", hash = "sha256:f81cd85a0e76ec7b8e2b6636fe02952d35befda4196b8c88f3cec5b4fb512839"},
+    {file = "aiohttp-3.10.2-cp310-cp310-win_amd64.whl", hash = "sha256:54ba10eb5a3481c28282eb6afb5f709aedf53cf9c3a31875ffbdc9fc719ffd67"},
+    {file = "aiohttp-3.10.2-cp311-cp311-macosx_10_9_universal2.whl", hash = "sha256:87fab7f948e407444c2f57088286e00e2ed0003ceaf3d8f8cc0f60544ba61d91"},
+    {file = "aiohttp-3.10.2-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:ec6ad66ed660d46503243cbec7b2b3d8ddfa020f984209b3b8ef7d98ce69c3f2"},
+    {file = "aiohttp-3.10.2-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:a4be88807283bd96ae7b8e401abde4ca0bab597ba73b5e9a2d98f36d451e9aac"},
+    {file = "aiohttp-3.10.2-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:01c98041f90927c2cbd72c22a164bb816fa3010a047d264969cf82e1d4bcf8d1"},
+    {file = "aiohttp-3.10.2-cp311-cp311-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:54e36c67e1a9273ecafab18d6693da0fb5ac48fd48417e4548ac24a918c20998"},
+    {file = "aiohttp-3.10.2-cp311-cp311-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:7de3ddb6f424af54535424082a1b5d1ae8caf8256ebd445be68c31c662354720"},
+    {file = "aiohttp-3.10.2-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:7dd9c7db94b4692b827ce51dcee597d61a0e4f4661162424faf65106775b40e7"},
+    {file = "aiohttp-3.10.2-cp311-cp311-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:e57e21e1167705f8482ca29cc5d02702208d8bf4aff58f766d94bcd6ead838cd"},
+    {file = "aiohttp-3.10.2-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:a1a50e59b720060c29e2951fd9f13c01e1ea9492e5a527b92cfe04dd64453c16"},
+    {file = "aiohttp-3.10.2-cp311-cp311-musllinux_1_2_i686.whl", hash = "sha256:686c87782481fda5ee6ba572d912a5c26d9f98cc5c243ebd03f95222af3f1b0f"},
+    {file = "aiohttp-3.10.2-cp311-cp311-musllinux_1_2_ppc64le.whl", hash = "sha256:dafb4abb257c0ed56dc36f4e928a7341b34b1379bd87e5a15ce5d883c2c90574"},
+    {file = "aiohttp-3.10.2-cp311-cp311-musllinux_1_2_s390x.whl", hash = "sha256:494a6f77560e02bd7d1ab579fdf8192390567fc96a603f21370f6e63690b7f3d"},
+    {file = "aiohttp-3.10.2-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:6fe8503b1b917508cc68bf44dae28823ac05e9f091021e0c41f806ebbb23f92f"},
+    {file = "aiohttp-3.10.2-cp311-cp311-win32.whl", hash = "sha256:4ddb43d06ce786221c0dfd3c91b4892c318eaa36b903f7c4278e7e2fa0dd5102"},
+    {file = "aiohttp-3.10.2-cp311-cp311-win_amd64.whl", hash = "sha256:ca2f5abcb0a9a47e56bac173c01e9f6c6e7f27534d91451c5f22e6a35a5a2093"},
+    {file = "aiohttp-3.10.2-cp312-cp312-macosx_10_9_universal2.whl", hash = "sha256:14eb6b17f6246959fb0b035d4f4ae52caa870c4edfb6170aad14c0de5bfbf478"},
+    {file = "aiohttp-3.10.2-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:465e445ec348d4e4bd349edd8b22db75f025da9d7b6dc1369c48e7935b85581e"},
+    {file = "aiohttp-3.10.2-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:341f8ece0276a828d95b70cd265d20e257f5132b46bf77d759d7f4e0443f2906"},
+    {file = "aiohttp-3.10.2-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:c01fbb87b5426381cd9418b3ddcf4fc107e296fa2d3446c18ce6c76642f340a3"},
+    {file = "aiohttp-3.10.2-cp312-cp312-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:2c474af073e1a6763e1c5522bbb2d85ff8318197e4c6c919b8d7886e16213345"},
+    {file = "aiohttp-3.10.2-cp312-cp312-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:d9076810a5621236e29b2204e67a68e1fe317c8727ee4c9abbfbb1083b442c38"},
+    {file = "aiohttp-3.10.2-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:e8f515d6859e673940e08de3922b9c4a2249653b0ac181169313bd6e4b1978ac"},
+    {file = "aiohttp-3.10.2-cp312-cp312-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:655e583afc639bef06f3b2446972c1726007a21003cd0ef57116a123e44601bc"},
+    {file = "aiohttp-3.10.2-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:8da9449a575133828cc99985536552ea2dcd690e848f9d41b48d8853a149a959"},
+    {file = "aiohttp-3.10.2-cp312-cp312-musllinux_1_2_i686.whl", hash = "sha256:19073d57d0feb1865d12361e2a1f5a49cb764bf81a4024a3b608ab521568093a"},
+    {file = "aiohttp-3.10.2-cp312-cp312-musllinux_1_2_ppc64le.whl", hash = "sha256:c8e98e1845805f184d91fda6f9ab93d7c7b0dddf1c07e0255924bfdb151a8d05"},
+    {file = "aiohttp-3.10.2-cp312-cp312-musllinux_1_2_s390x.whl", hash = "sha256:377220a5efde6f9497c5b74649b8c261d3cce8a84cb661be2ed8099a2196400a"},
+    {file = "aiohttp-3.10.2-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:92f7f4a4dc9cdb5980973a74d43cdbb16286dacf8d1896b6c3023b8ba8436f8e"},
+    {file = "aiohttp-3.10.2-cp312-cp312-win32.whl", hash = "sha256:9bb2834a6f11d65374ce97d366d6311a9155ef92c4f0cee543b2155d06dc921f"},
+    {file = "aiohttp-3.10.2-cp312-cp312-win_amd64.whl", hash = "sha256:518dc3cb37365255708283d1c1c54485bbacccd84f0a0fb87ed8917ba45eda5b"},
+    {file = "aiohttp-3.10.2-cp38-cp38-macosx_10_9_universal2.whl", hash = "sha256:7f98e70bbbf693086efe4b86d381efad8edac040b8ad02821453083d15ec315f"},
+    {file = "aiohttp-3.10.2-cp38-cp38-macosx_10_9_x86_64.whl", hash = "sha256:9f6f0b252a009e98fe84028a4ec48396a948e7a65b8be06ccfc6ef68cf1f614d"},
+    {file = "aiohttp-3.10.2-cp38-cp38-macosx_11_0_arm64.whl", hash = "sha256:9360e3ffc7b23565600e729e8c639c3c50d5520e05fdf94aa2bd859eef12c407"},
+    {file = "aiohttp-3.10.2-cp38-cp38-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:3988044d1635c7821dd44f0edfbe47e9875427464e59d548aece447f8c22800a"},
+    {file = "aiohttp-3.10.2-cp38-cp38-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:30a9d59da1543a6f1478c3436fd49ec59be3868bca561a33778b4391005e499d"},
+    {file = "aiohttp-3.10.2-cp38-cp38-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:f9f49bdb94809ac56e09a310a62f33e5f22973d6fd351aac72a39cd551e98194"},
+    {file = "aiohttp-3.10.2-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:ddfd2dca3f11c365d6857a07e7d12985afc59798458a2fdb2ffa4a0332a3fd43"},
+    {file = "aiohttp-3.10.2-cp38-cp38-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:685c1508ec97b2cd3e120bfe309a4ff8e852e8a7460f1ef1de00c2c0ed01e33c"},
+    {file = "aiohttp-3.10.2-cp38-cp38-musllinux_1_2_aarch64.whl", hash = "sha256:49904f38667c44c041a0b44c474b3ae36948d16a0398a8f8cd84e2bb3c42a069"},
+    {file = "aiohttp-3.10.2-cp38-cp38-musllinux_1_2_i686.whl", hash = "sha256:352f3a4e5f11f3241a49b6a48bc5b935fabc35d1165fa0d87f3ca99c1fcca98b"},
+    {file = "aiohttp-3.10.2-cp38-cp38-musllinux_1_2_ppc64le.whl", hash = "sha256:fc61f39b534c5d5903490478a0dd349df397d2284a939aa3cbaa2fb7a19b8397"},
+    {file = "aiohttp-3.10.2-cp38-cp38-musllinux_1_2_s390x.whl", hash = "sha256:ad2274e707be37420d0b6c3d26a8115295fe9d8e6e530fa6a42487a8ca3ad052"},
+    {file = "aiohttp-3.10.2-cp38-cp38-musllinux_1_2_x86_64.whl", hash = "sha256:c836bf3c7512100219fe1123743fd8dd9a2b50dd7cfb0c3bb10d041309acab4b"},
+    {file = "aiohttp-3.10.2-cp38-cp38-win32.whl", hash = "sha256:53e8898adda402be03ff164b0878abe2d884e3ea03a4701e6ad55399d84b92dc"},
+    {file = "aiohttp-3.10.2-cp38-cp38-win_amd64.whl", hash = "sha256:7cc8f65f5b22304693de05a245b6736b14cb5bc9c8a03da6e2ae9ef15f8b458f"},
+    {file = "aiohttp-3.10.2-cp39-cp39-macosx_10_9_universal2.whl", hash = "sha256:9dfc906d656e14004c5bc672399c1cccc10db38df2b62a13fb2b6e165a81c316"},
+    {file = "aiohttp-3.10.2-cp39-cp39-macosx_10_9_x86_64.whl", hash = "sha256:91b10208b222ddf655c3a3d5b727879d7163db12b634492df41a9182a76edaae"},
+    {file = "aiohttp-3.10.2-cp39-cp39-macosx_11_0_arm64.whl", hash = "sha256:9fd16b5e1a7bdd14668cd6bde60a2a29b49147a535c74f50d8177d11b38433a7"},
+    {file = "aiohttp-3.10.2-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:b2bfdda4971bd79201f59adbad24ec2728875237e1c83bba5221284dbbf57bda"},
+    {file = "aiohttp-3.10.2-cp39-cp39-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:69d73f869cf29e8a373127fc378014e2b17bcfbe8d89134bc6fb06a2f67f3cb3"},
+    {file = "aiohttp-3.10.2-cp39-cp39-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:df59f8486507c421c0620a2c3dce81fbf1d54018dc20ff4fecdb2c106d6e6abc"},
+    {file = "aiohttp-3.10.2-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:0df930015db36b460aa9badbf35eccbc383f00d52d4b6f3de2ccb57d064a6ade"},
+    {file = "aiohttp-3.10.2-cp39-cp39-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:562b1153ab7f766ee6b8b357ec777a302770ad017cf18505d34f1c088fccc448"},
+    {file = "aiohttp-3.10.2-cp39-cp39-musllinux_1_2_aarch64.whl", hash = "sha256:d984db6d855de58e0fde1ef908d48fe9a634cadb3cf715962722b4da1c40619d"},
+    {file = "aiohttp-3.10.2-cp39-cp39-musllinux_1_2_i686.whl", hash = "sha256:14dc3fcb0d877911d775d511eb617a486a8c48afca0a887276e63db04d3ee920"},
+    {file = "aiohttp-3.10.2-cp39-cp39-musllinux_1_2_ppc64le.whl", hash = "sha256:b52a27a5c97275e254704e1049f4b96a81e67d6205f52fa37a4777d55b0e98ef"},
+    {file = "aiohttp-3.10.2-cp39-cp39-musllinux_1_2_s390x.whl", hash = "sha256:cd33d9de8cfd006a0d0fe85f49b4183c57e91d18ffb7e9004ce855e81928f704"},
+    {file = "aiohttp-3.10.2-cp39-cp39-musllinux_1_2_x86_64.whl", hash = "sha256:1238fc979160bc03a92fff9ad021375ff1c8799c6aacb0d8ea1b357ea40932bb"},
+    {file = "aiohttp-3.10.2-cp39-cp39-win32.whl", hash = "sha256:e2f43d238eae4f0b04f58d4c0df4615697d4ca3e9f9b1963d49555a94f0f5a04"},
+    {file = "aiohttp-3.10.2-cp39-cp39-win_amd64.whl", hash = "sha256:947847f07a8f81d7b39b2d0202fd73e61962ebe17ac2d8566f260679e467da7b"},
+    {file = "aiohttp-3.10.2.tar.gz", hash = "sha256:4d1f694b5d6e459352e5e925a42e05bac66655bfde44d81c59992463d2897014"},
 ]
 
 [package.dependencies]
+aiohappyeyeballs = ">=2.3.0"
 aiosignal = ">=1.1.2"
 async-timeout = {version = ">=4.0,<5.0", markers = "python_version < \"3.11\""}
 attrs = ">=17.3.0"
@@ -94,7 +106,7 @@ multidict = ">=4.5,<7.0"
 yarl = ">=1.0,<2.0"
 
 [package.extras]
-speedups = ["Brotli", "aiodns", "brotlicffi"]
+speedups = ["Brotli", "aiodns (>=3.2.0)", "brotlicffi"]
 
 [[package]]
 name = "aiopg"
@@ -1514,6 +1526,20 @@ files = [
 [package.dependencies]
 six = "*"
 
+[[package]]
+name = "kafka-python"
+version = "2.0.2"
+description = "Pure Python client for Apache Kafka"
+optional = false
+python-versions = "*"
+files = [
+    {file = "kafka-python-2.0.2.tar.gz", hash = "sha256:04dfe7fea2b63726cd6f3e79a2d86e709d608d74406638c5da33a01d45a9d7e3"},
+    {file = "kafka_python-2.0.2-py2.py3-none-any.whl", hash = "sha256:2d92418c7cb1c298fa6c7f0fb3519b520d0d7526ac6cb7ae2a4fc65a51a94b6e"},
+]
+
+[package.extras]
+crc32c = ["crc32c"]
+
 [[package]]
 name = "lazy-object-proxy"
 version = "1.10.0"
@@ -3357,4 +3383,4 @@ cffi = ["cffi (>=1.11)"]
 [metadata]
 lock-version = "2.0"
 python-versions = "^3.9"
-content-hash = "7cee6a8c30bc7f4bfb0a87c6bad3952dfb4da127fad853d2710a93ac3eab8a00"
+content-hash = "c09bcb333ab550958b33dbf4fec968c500d8e701fd4c96402cddbd9bb8048055"

proxy/Cargo.toml

@@ -92,6 +92,7 @@ tracing-opentelemetry.workspace = true
 tracing-subscriber.workspace = true
 tracing-utils.workspace = true
 tracing.workspace = true
+try-lock.workspace = true
 typed-json.workspace = true
 url.workspace = true
 urlencoding.workspace = true

proxy/src/auth/backend.rs

@@ -218,7 +218,7 @@ impl RateBucketInfo {
 impl AuthenticationConfig {
     pub fn check_rate_limit(
         &self,
-        ctx: &mut RequestMonitoring,
+        ctx: &RequestMonitoring,
         config: &AuthenticationConfig,
         secret: AuthSecret,
         endpoint: &EndpointId,
@@ -243,7 +243,7 @@ impl AuthenticationConfig {
         let limit_not_exceeded = self.rate_limiter.check(
             (
                 endpoint_int,
-                MaskedIp::new(ctx.peer_addr, config.rate_limit_ip_subnet),
+                MaskedIp::new(ctx.peer_addr(), config.rate_limit_ip_subnet),
             ),
             password_weight,
         );
@@ -274,7 +274,7 @@ impl AuthenticationConfig {
 ///
 /// All authentication flows will emit an AuthenticationOk message if successful.
 async fn auth_quirks(
-    ctx: &mut RequestMonitoring,
+    ctx: &RequestMonitoring,
     api: &impl console::Api,
     user_info: ComputeUserInfoMaybeEndpoint,
     client: &mut stream::PqStream<Stream<impl AsyncRead + AsyncWrite + Unpin>>,
@@ -303,8 +303,8 @@ async fn auth_quirks(
     let (allowed_ips, maybe_secret) = api.get_allowed_ips_and_secret(ctx, &info).await?;
 
     // check allowed list
-    if !check_peer_addr_is_in_list(&ctx.peer_addr, &allowed_ips) {
-        return Err(auth::AuthError::ip_address_not_allowed(ctx.peer_addr));
+    if !check_peer_addr_is_in_list(&ctx.peer_addr(), &allowed_ips) {
+        return Err(auth::AuthError::ip_address_not_allowed(ctx.peer_addr()));
     }
 
     if !endpoint_rate_limiter.check(info.endpoint.clone().into(), 1) {
@@ -356,7 +356,7 @@ async fn auth_quirks(
 }
 
 async fn authenticate_with_secret(
-    ctx: &mut RequestMonitoring,
+    ctx: &RequestMonitoring,
     secret: AuthSecret,
     info: ComputeUserInfo,
     client: &mut stream::PqStream<Stream<impl AsyncRead + AsyncWrite + Unpin>>,
@@ -421,7 +421,7 @@ impl<'a> BackendType<'a, ComputeUserInfoMaybeEndpoint, &()> {
     #[tracing::instrument(fields(allow_cleartext = allow_cleartext), skip_all)]
     pub async fn authenticate(
         self,
-        ctx: &mut RequestMonitoring,
+        ctx: &RequestMonitoring,
         client: &mut stream::PqStream<Stream<impl AsyncRead + AsyncWrite + Unpin>>,
         allow_cleartext: bool,
         config: &'static AuthenticationConfig,
@@ -467,7 +467,7 @@ impl<'a> BackendType<'a, ComputeUserInfoMaybeEndpoint, &()> {
 impl BackendType<'_, ComputeUserInfo, &()> {
     pub async fn get_role_secret(
         &self,
-        ctx: &mut RequestMonitoring,
+        ctx: &RequestMonitoring,
     ) -> Result<CachedRoleSecret, GetAuthInfoError> {
         use BackendType::*;
         match self {
@@ -478,7 +478,7 @@ impl BackendType<'_, ComputeUserInfo, &()> {
 
     pub async fn get_allowed_ips_and_secret(
         &self,
-        ctx: &mut RequestMonitoring,
+        ctx: &RequestMonitoring,
     ) -> Result<(CachedAllowedIps, Option<CachedRoleSecret>), GetAuthInfoError> {
         use BackendType::*;
         match self {
@@ -492,7 +492,7 @@ impl BackendType<'_, ComputeUserInfo, &()> {
 impl ComputeConnectBackend for BackendType<'_, ComputeCredentials, NodeInfo> {
     async fn wake_compute(
         &self,
-        ctx: &mut RequestMonitoring,
+        ctx: &RequestMonitoring,
     ) -> Result<CachedNodeInfo, console::errors::WakeComputeError> {
         use BackendType::*;
 
@@ -514,7 +514,7 @@ impl ComputeConnectBackend for BackendType<'_, ComputeCredentials, NodeInfo> {
 impl ComputeConnectBackend for BackendType<'_, ComputeCredentials, &()> {
     async fn wake_compute(
         &self,
-        ctx: &mut RequestMonitoring,
+        ctx: &RequestMonitoring,
     ) -> Result<CachedNodeInfo, console::errors::WakeComputeError> {
         use BackendType::*;
 
@@ -571,7 +571,7 @@ mod tests {
     impl console::Api for Auth {
         async fn get_role_secret(
             &self,
-            _ctx: &mut RequestMonitoring,
+            _ctx: &RequestMonitoring,
             _user_info: &super::ComputeUserInfo,
         ) -> Result<CachedRoleSecret, console::errors::GetAuthInfoError> {
             Ok(CachedRoleSecret::new_uncached(Some(self.secret.clone())))
@@ -579,7 +579,7 @@ mod tests {
 
         async fn get_allowed_ips_and_secret(
             &self,
-            _ctx: &mut RequestMonitoring,
+            _ctx: &RequestMonitoring,
             _user_info: &super::ComputeUserInfo,
         ) -> Result<(CachedAllowedIps, Option<CachedRoleSecret>), console::errors::GetAuthInfoError>
         {
@@ -591,7 +591,7 @@ mod tests {
 
         async fn wake_compute(
             &self,
-            _ctx: &mut RequestMonitoring,
+            _ctx: &RequestMonitoring,
             _user_info: &super::ComputeUserInfo,
         ) -> Result<CachedNodeInfo, console::errors::WakeComputeError> {
             unimplemented!()
@@ -665,7 +665,7 @@ mod tests {
         let (mut client, server) = tokio::io::duplex(1024);
         let mut stream = PqStream::new(Stream::from_raw(server));
 
-        let mut ctx = RequestMonitoring::test();
+        let ctx = RequestMonitoring::test();
         let api = Auth {
             ips: vec![],
             secret: AuthSecret::Scram(ServerSecret::build("my-secret-password").await.unwrap()),
@@ -723,7 +723,7 @@ mod tests {
         ));
 
         let _creds = auth_quirks(
-            &mut ctx,
+            &ctx,
             &api,
             user_info,
             &mut stream,
@@ -742,7 +742,7 @@ mod tests {
         let (mut client, server) = tokio::io::duplex(1024);
         let mut stream = PqStream::new(Stream::from_raw(server));
 
-        let mut ctx = RequestMonitoring::test();
+        let ctx = RequestMonitoring::test();
         let api = Auth {
             ips: vec![],
             secret: AuthSecret::Scram(ServerSecret::build("my-secret-password").await.unwrap()),
@@ -775,7 +775,7 @@ mod tests {
         ));
 
         let _creds = auth_quirks(
-            &mut ctx,
+            &ctx,
             &api,
             user_info,
             &mut stream,
@@ -794,7 +794,7 @@ mod tests {
         let (mut client, server) = tokio::io::duplex(1024);
         let mut stream = PqStream::new(Stream::from_raw(server));
 
-        let mut ctx = RequestMonitoring::test();
+        let ctx = RequestMonitoring::test();
         let api = Auth {
             ips: vec![],
             secret: AuthSecret::Scram(ServerSecret::build("my-secret-password").await.unwrap()),
@@ -828,7 +828,7 @@ mod tests {
         ));
 
         let creds = auth_quirks(
-            &mut ctx,
+            &ctx,
             &api,
             user_info,
             &mut stream,

proxy/src/auth/backend/classic.rs

@@ -12,7 +12,7 @@ use tokio::io::{AsyncRead, AsyncWrite};
 use tracing::{info, warn};
 
 pub(super) async fn authenticate(
-    ctx: &mut RequestMonitoring,
+    ctx: &RequestMonitoring,
     creds: ComputeUserInfo,
     client: &mut PqStream<Stream<impl AsyncRead + AsyncWrite + Unpin>>,
     config: &'static AuthenticationConfig,
@@ -27,7 +27,7 @@ pub(super) async fn authenticate(
         }
         AuthSecret::Scram(secret) => {
             info!("auth endpoint chooses SCRAM");
-            let scram = auth::Scram(&secret, &mut *ctx);
+            let scram = auth::Scram(&secret, ctx);
 
             let auth_outcome = tokio::time::timeout(
                 config.scram_protocol_timeout,

proxy/src/auth/backend/hacks.rs

@@ -18,7 +18,7 @@ use tracing::{info, warn};
 /// These properties are benefical for serverless JS workers, so we
 /// use this mechanism for websocket connections.
 pub async fn authenticate_cleartext(
-    ctx: &mut RequestMonitoring,
+    ctx: &RequestMonitoring,
     info: ComputeUserInfo,
     client: &mut stream::PqStream<Stream<impl AsyncRead + AsyncWrite + Unpin>>,
     secret: AuthSecret,
@@ -28,7 +28,7 @@ pub async fn authenticate_cleartext(
     ctx.set_auth_method(crate::context::AuthMethod::Cleartext);
 
     // pause the timer while we communicate with the client
-    let paused = ctx.latency_timer.pause(crate::metrics::Waiting::Client);
+    let paused = ctx.latency_timer_pause(crate::metrics::Waiting::Client);
 
     let ep = EndpointIdInt::from(&info.endpoint);
 
@@ -60,7 +60,7 @@ pub async fn authenticate_cleartext(
 /// Similar to [`authenticate_cleartext`], but there's a specific password format,
 /// and passwords are not yet validated (we don't know how to validate them!)
 pub async fn password_hack_no_authentication(
-    ctx: &mut RequestMonitoring,
+    ctx: &RequestMonitoring,
     info: ComputeUserInfoNoEndpoint,
     client: &mut stream::PqStream<Stream<impl AsyncRead + AsyncWrite + Unpin>>,
 ) -> auth::Result<ComputeCredentials> {
@@ -68,7 +68,7 @@ pub async fn password_hack_no_authentication(
     ctx.set_auth_method(crate::context::AuthMethod::Cleartext);
 
     // pause the timer while we communicate with the client
-    let _paused = ctx.latency_timer.pause(crate::metrics::Waiting::Client);
+    let _paused = ctx.latency_timer_pause(crate::metrics::Waiting::Client);
 
     let payload = AuthFlow::new(client)
         .begin(auth::PasswordHack)

proxy/src/auth/backend/link.rs

@@ -57,7 +57,7 @@ pub fn new_psql_session_id() -> String {
 }
 
 pub(super) async fn authenticate(
-    ctx: &mut RequestMonitoring,
+    ctx: &RequestMonitoring,
     link_uri: &reqwest::Url,
     client: &mut PqStream<impl AsyncRead + AsyncWrite + Unpin>,
 ) -> auth::Result<NodeInfo> {

proxy/src/auth/credentials.rs

@@ -84,7 +84,7 @@ pub fn endpoint_sni(
 
 impl ComputeUserInfoMaybeEndpoint {
     pub fn parse(
-        ctx: &mut RequestMonitoring,
+        ctx: &RequestMonitoring,
         params: &StartupMessageParams,
         sni: Option<&str>,
         common_names: Option<&HashSet<String>>,
@@ -249,8 +249,8 @@ mod tests {
     fn parse_bare_minimum() -> anyhow::Result<()> {
         // According to postgresql, only `user` should be required.
         let options = StartupMessageParams::new([("user", "john_doe")]);
-        let mut ctx = RequestMonitoring::test();
-        let user_info = ComputeUserInfoMaybeEndpoint::parse(&mut ctx, &options, None, None)?;
+        let ctx = RequestMonitoring::test();
+        let user_info = ComputeUserInfoMaybeEndpoint::parse(&ctx, &options, None, None)?;
         assert_eq!(user_info.user, "john_doe");
         assert_eq!(user_info.endpoint_id, None);
 
@@ -264,8 +264,8 @@ mod tests {
             ("database", "world"), // should be ignored
             ("foo", "bar"),        // should be ignored
         ]);
-        let mut ctx = RequestMonitoring::test();
-        let user_info = ComputeUserInfoMaybeEndpoint::parse(&mut ctx, &options, None, None)?;
+        let ctx = RequestMonitoring::test();
+        let user_info = ComputeUserInfoMaybeEndpoint::parse(&ctx, &options, None, None)?;
         assert_eq!(user_info.user, "john_doe");
         assert_eq!(user_info.endpoint_id, None);
 
@@ -279,9 +279,9 @@ mod tests {
         let sni = Some("foo.localhost");
         let common_names = Some(["localhost".into()].into());
 
-        let mut ctx = RequestMonitoring::test();
+        let ctx = RequestMonitoring::test();
         let user_info =
-            ComputeUserInfoMaybeEndpoint::parse(&mut ctx, &options, sni, common_names.as_ref())?;
+            ComputeUserInfoMaybeEndpoint::parse(&ctx, &options, sni, common_names.as_ref())?;
         assert_eq!(user_info.user, "john_doe");
         assert_eq!(user_info.endpoint_id.as_deref(), Some("foo"));
         assert_eq!(user_info.options.get_cache_key("foo"), "foo");
@@ -296,8 +296,8 @@ mod tests {
             ("options", "-ckey=1 project=bar -c geqo=off"),
         ]);
 
-        let mut ctx = RequestMonitoring::test();
-        let user_info = ComputeUserInfoMaybeEndpoint::parse(&mut ctx, &options, None, None)?;
+        let ctx = RequestMonitoring::test();
+        let user_info = ComputeUserInfoMaybeEndpoint::parse(&ctx, &options, None, None)?;
         assert_eq!(user_info.user, "john_doe");
         assert_eq!(user_info.endpoint_id.as_deref(), Some("bar"));
 
@@ -311,8 +311,8 @@ mod tests {
             ("options", "-ckey=1 endpoint=bar -c geqo=off"),
         ]);
 
-        let mut ctx = RequestMonitoring::test();
-        let user_info = ComputeUserInfoMaybeEndpoint::parse(&mut ctx, &options, None, None)?;
+        let ctx = RequestMonitoring::test();
+        let user_info = ComputeUserInfoMaybeEndpoint::parse(&ctx, &options, None, None)?;
         assert_eq!(user_info.user, "john_doe");
         assert_eq!(user_info.endpoint_id.as_deref(), Some("bar"));
 
@@ -329,8 +329,8 @@ mod tests {
             ),
         ]);
 
-        let mut ctx = RequestMonitoring::test();
-        let user_info = ComputeUserInfoMaybeEndpoint::parse(&mut ctx, &options, None, None)?;
+        let ctx = RequestMonitoring::test();
+        let user_info = ComputeUserInfoMaybeEndpoint::parse(&ctx, &options, None, None)?;
         assert_eq!(user_info.user, "john_doe");
         assert!(user_info.endpoint_id.is_none());
 
@@ -344,8 +344,8 @@ mod tests {
             ("options", "-ckey=1 endpoint=bar project=foo -c geqo=off"),
         ]);
 
-        let mut ctx = RequestMonitoring::test();
-        let user_info = ComputeUserInfoMaybeEndpoint::parse(&mut ctx, &options, None, None)?;
+        let ctx = RequestMonitoring::test();
+        let user_info = ComputeUserInfoMaybeEndpoint::parse(&ctx, &options, None, None)?;
         assert_eq!(user_info.user, "john_doe");
         assert!(user_info.endpoint_id.is_none());
 
@@ -359,9 +359,9 @@ mod tests {
         let sni = Some("baz.localhost");
         let common_names = Some(["localhost".into()].into());
 
-        let mut ctx = RequestMonitoring::test();
+        let ctx = RequestMonitoring::test();
         let user_info =
-            ComputeUserInfoMaybeEndpoint::parse(&mut ctx, &options, sni, common_names.as_ref())?;
+            ComputeUserInfoMaybeEndpoint::parse(&ctx, &options, sni, common_names.as_ref())?;
         assert_eq!(user_info.user, "john_doe");
         assert_eq!(user_info.endpoint_id.as_deref(), Some("baz"));
 
@@ -374,16 +374,16 @@ mod tests {
 
         let common_names = Some(["a.com".into(), "b.com".into()].into());
         let sni = Some("p1.a.com");
-        let mut ctx = RequestMonitoring::test();
+        let ctx = RequestMonitoring::test();
         let user_info =
-            ComputeUserInfoMaybeEndpoint::parse(&mut ctx, &options, sni, common_names.as_ref())?;
+            ComputeUserInfoMaybeEndpoint::parse(&ctx, &options, sni, common_names.as_ref())?;
         assert_eq!(user_info.endpoint_id.as_deref(), Some("p1"));
 
         let common_names = Some(["a.com".into(), "b.com".into()].into());
         let sni = Some("p1.b.com");
-        let mut ctx = RequestMonitoring::test();
+        let ctx = RequestMonitoring::test();
         let user_info =
-            ComputeUserInfoMaybeEndpoint::parse(&mut ctx, &options, sni, common_names.as_ref())?;
+            ComputeUserInfoMaybeEndpoint::parse(&ctx, &options, sni, common_names.as_ref())?;
         assert_eq!(user_info.endpoint_id.as_deref(), Some("p1"));
 
         Ok(())
@@ -397,10 +397,9 @@ mod tests {
         let sni = Some("second.localhost");
         let common_names = Some(["localhost".into()].into());
 
-        let mut ctx = RequestMonitoring::test();
-        let err =
-            ComputeUserInfoMaybeEndpoint::parse(&mut ctx, &options, sni, common_names.as_ref())
-                .expect_err("should fail");
+        let ctx = RequestMonitoring::test();
+        let err = ComputeUserInfoMaybeEndpoint::parse(&ctx, &options, sni, common_names.as_ref())
+            .expect_err("should fail");
         match err {
             InconsistentProjectNames { domain, option } => {
                 assert_eq!(option, "first");
@@ -417,10 +416,9 @@ mod tests {
         let sni = Some("project.localhost");
         let common_names = Some(["example.com".into()].into());
 
-        let mut ctx = RequestMonitoring::test();
-        let err =
-            ComputeUserInfoMaybeEndpoint::parse(&mut ctx, &options, sni, common_names.as_ref())
-                .expect_err("should fail");
+        let ctx = RequestMonitoring::test();
+        let err = ComputeUserInfoMaybeEndpoint::parse(&ctx, &options, sni, common_names.as_ref())
+            .expect_err("should fail");
         match err {
             UnknownCommonName { cn } => {
                 assert_eq!(cn, "localhost");
@@ -438,9 +436,9 @@ mod tests {
 
         let sni = Some("project.localhost");
         let common_names = Some(["localhost".into()].into());
-        let mut ctx = RequestMonitoring::test();
+        let ctx = RequestMonitoring::test();
         let user_info =
-            ComputeUserInfoMaybeEndpoint::parse(&mut ctx, &options, sni, common_names.as_ref())?;
+            ComputeUserInfoMaybeEndpoint::parse(&ctx, &options, sni, common_names.as_ref())?;
         assert_eq!(user_info.endpoint_id.as_deref(), Some("project"));
         assert_eq!(
             user_info.options.get_cache_key("project"),

proxy/src/auth/flow.rs

@@ -27,7 +27,7 @@ pub trait AuthMethod {
 pub struct Begin;
 
 /// Use [SCRAM](crate::scram)-based auth in [`AuthFlow`].
-pub struct Scram<'a>(pub &'a scram::ServerSecret, pub &'a mut RequestMonitoring);
+pub struct Scram<'a>(pub &'a scram::ServerSecret, pub &'a RequestMonitoring);
 
 impl AuthMethod for Scram<'_> {
     #[inline(always)]
@@ -155,7 +155,7 @@ impl<S: AsyncRead + AsyncWrite + Unpin> AuthFlow<'_, S, Scram<'_>> {
         let Scram(secret, ctx) = self.state;
 
         // pause the timer while we communicate with the client
-        let _paused = ctx.latency_timer.pause(crate::metrics::Waiting::Client);
+        let _paused = ctx.latency_timer_pause(crate::metrics::Waiting::Client);
 
         // Initial client message contains the chosen auth method's name.
         let msg = self.stream.read_password_message().await?;
@@ -168,10 +168,8 @@ impl<S: AsyncRead + AsyncWrite + Unpin> AuthFlow<'_, S, Scram<'_>> {
         }
 
         match sasl.method {
-            SCRAM_SHA_256 => ctx.auth_method = Some(crate::context::AuthMethod::ScramSha256),
-            SCRAM_SHA_256_PLUS => {
-                ctx.auth_method = Some(crate::context::AuthMethod::ScramSha256Plus)
-            }
+            SCRAM_SHA_256 => ctx.set_auth_method(crate::context::AuthMethod::ScramSha256),
+            SCRAM_SHA_256_PLUS => ctx.set_auth_method(crate::context::AuthMethod::ScramSha256Plus),
             _ => {}
         }
         info!("client chooses {}", sasl.method);

proxy/src/bin/pg_sni_router.rs

@@ -205,7 +205,7 @@ async fn task_main(
 const ERR_INSECURE_CONNECTION: &str = "connection is insecure (try using `sslmode=require`)";
 
 async fn ssl_handshake<S: AsyncRead + AsyncWrite + Unpin>(
-    ctx: &mut RequestMonitoring,
+    ctx: &RequestMonitoring,
     raw_stream: S,
     tls_config: Arc<rustls::ServerConfig>,
     tls_server_end_point: TlsServerEndPoint,
@@ -256,13 +256,13 @@ async fn ssl_handshake<S: AsyncRead + AsyncWrite + Unpin>(
 }
 
 async fn handle_client(
-    mut ctx: RequestMonitoring,
+    ctx: RequestMonitoring,
     dest_suffix: Arc<String>,
     tls_config: Arc<rustls::ServerConfig>,
     tls_server_end_point: TlsServerEndPoint,
     stream: impl AsyncRead + AsyncWrite + Unpin,
 ) -> anyhow::Result<()> {
-    let mut tls_stream = ssl_handshake(&mut ctx, stream, tls_config, tls_server_end_point).await?;
+    let mut tls_stream = ssl_handshake(&ctx, stream, tls_config, tls_server_end_point).await?;
 
     // Cut off first part of the SNI domain
     // We receive required destination details in the format of

proxy/src/bin/proxy.rs

@@ -5,6 +5,7 @@ use aws_config::meta::region::RegionProviderChain;
 use aws_config::profile::ProfileFileCredentialsProvider;
 use aws_config::provider_config::ProviderConfig;
 use aws_config::web_identity_token::WebIdentityTokenCredentialsProvider;
+use aws_config::Region;
 use futures::future::Either;
 use proxy::auth;
 use proxy::auth::backend::AuthRateLimiter;
@@ -290,9 +291,10 @@ async fn main() -> anyhow::Result<()> {
     let config = build_config(&args)?;
 
     info!("Authentication backend: {}", config.auth_backend);
-    info!("Using region: {}", config.aws_region);
+    info!("Using region: {}", args.aws_region);
 
-    let region_provider = RegionProviderChain::default_provider().or_else(&*config.aws_region); // Replace with your Redis region if needed
+    let region_provider =
+        RegionProviderChain::default_provider().or_else(Region::new(args.aws_region.clone()));
     let provider_conf =
         ProviderConfig::without_region().with_region(region_provider.region().await);
     let aws_credentials_provider = {
@@ -318,7 +320,7 @@ async fn main() -> anyhow::Result<()> {
     };
     let elasticache_credentials_provider = Arc::new(elasticache::CredentialsProvider::new(
         elasticache::AWSIRSAConfig::new(
-            config.aws_region.clone(),
+            args.aws_region.clone(),
             args.redis_cluster_name,
             args.redis_user_id,
         ),
@@ -376,11 +378,14 @@ async fn main() -> anyhow::Result<()> {
 
     let cancel_map = CancelMap::default();
 
+    let redis_rps_limit = Vec::leak(args.redis_rps_limit.clone());
+    RateBucketInfo::validate(redis_rps_limit)?;
+
     let redis_publisher = match &regional_redis_client {
         Some(redis_publisher) => Some(Arc::new(Mutex::new(RedisPublisherClient::new(
             redis_publisher.clone(),
             args.region.clone(),
-            &config.redis_rps_limit,
+            redis_rps_limit,
         )?))),
         None => None,
     };
@@ -656,7 +661,6 @@ fn build_config(args: &ProxyCliArgs) -> anyhow::Result<&'static ProxyConfig> {
     )?;
 
     let http_config = HttpConfig {
-        request_timeout: args.sql_over_http.sql_over_http_timeout,
         pool_options: GlobalConnPoolOptions {
             max_conns_per_endpoint: args.sql_over_http.sql_over_http_pool_max_conns_per_endpoint,
             gc_epoch: args.sql_over_http.sql_over_http_pool_gc_epoch,
@@ -676,9 +680,6 @@ fn build_config(args: &ProxyCliArgs) -> anyhow::Result<&'static ProxyConfig> {
         rate_limit_ip_subnet: args.auth_rate_limit_ip_subnet,
     };
 
-    let mut redis_rps_limit = args.redis_rps_limit.clone();
-    RateBucketInfo::validate(&mut redis_rps_limit)?;
-
     let config = Box::leak(Box::new(ProxyConfig {
         tls_config,
         auth_backend,
@@ -687,11 +688,8 @@ fn build_config(args: &ProxyCliArgs) -> anyhow::Result<&'static ProxyConfig> {
         http_config,
         authentication_config,
         require_client_ip: args.require_client_ip,
-        disable_ip_check_for_http: args.disable_ip_check_for_http,
-        redis_rps_limit,
         handshake_timeout: args.handshake_timeout,
         region: args.region.clone(),
-        aws_region: args.aws_region.clone(),
         wake_compute_retry_config: config::RetryConfig::parse(&args.wake_compute_retry)?,
         connect_compute_locks,
         connect_to_compute_retry_config: config::RetryConfig::parse(