clean up

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

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

You can try:

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

before and after this patch: this patch fixes it.

.github/actionlint.yml

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

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

@@ -14,11 +14,8 @@ inputs:
   api_host:
     description: 'Neon API host'
     default: console-stage.neon.build
-  provisioner:
-    description: 'k8s-pod or k8s-neonvm'
-    default: 'k8s-pod'
   compute_units:
-    description: '[Min, Max] compute units; Min and Max are used for k8s-neonvm with autoscaling, for k8s-pod values Min and Max should be equal'
+    description: '[Min, Max] compute units'
     default: '[1, 1]'
 
 outputs:
@@ -37,10 +34,6 @@ runs:
       # A shell without `set -x` to not to expose password/dsn in logs
       shell: bash -euo pipefail {0}
       run: |
-        if [ "${PROVISIONER}" == "k8s-pod" ] && [ "${MIN_CU}" != "${MAX_CU}" ]; then
-          echo >&2 "For k8s-pod provisioner MIN_CU should be equal to MAX_CU"
-        fi
-
         project=$(curl \
           "https://${API_HOST}/api/v2/projects" \
           --fail \
@@ -52,7 +45,7 @@ runs:
               \"name\": \"Created by actions/neon-project-create; GITHUB_RUN_ID=${GITHUB_RUN_ID}\",
               \"pg_version\": ${POSTGRES_VERSION},
               \"region_id\": \"${REGION_ID}\",
-              \"provisioner\": \"${PROVISIONER}\",
+              \"provisioner\": \"k8s-neonvm\",
               \"autoscaling_limit_min_cu\": ${MIN_CU},
               \"autoscaling_limit_max_cu\": ${MAX_CU},
               \"settings\": { }
@@ -75,6 +68,5 @@ runs:
         API_KEY: ${{ inputs.api_key }}
         REGION_ID: ${{ inputs.region_id }}
         POSTGRES_VERSION: ${{ inputs.postgres_version }}
-        PROVISIONER: ${{ inputs.provisioner }}
         MIN_CU: ${{ fromJSON(inputs.compute_units)[0] }}
         MAX_CU: ${{ fromJSON(inputs.compute_units)[1] }}

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

@@ -131,8 +131,8 @@ runs:
           exit 1
         fi
         if [[ "${{ inputs.run_in_parallel }}" == "true" ]]; then
-          # -n16 uses sixteen processes to run tests via pytest-xdist
-          EXTRA_PARAMS="-n16 $EXTRA_PARAMS"
+          # -n sets the number of parallel processes that pytest-xdist will run
+          EXTRA_PARAMS="-n12 $EXTRA_PARAMS"
 
           # --dist=loadgroup points tests marked with @pytest.mark.xdist_group
           # to the same worker to make @pytest.mark.order work with xdist

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

@@ -19,6 +19,10 @@ on:
         description: 'debug or release'
         required: true
         type: string
+      pg-versions:
+        description: 'a json array of postgres versions to run regression tests on'
+        required: true
+        type: string
 
 defaults:
   run:
@@ -254,7 +258,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        pg_version: [ v14, v15, v16 ]
+        pg_version: ${{ fromJson(inputs.pg-versions) }}
     steps:
       - uses: actions/checkout@v4
         with:
@@ -278,14 +282,11 @@ jobs:
           CHECK_ONDISK_DATA_COMPATIBILITY: nonempty
           BUILD_TAG: ${{ inputs.build-tag }}
           PAGESERVER_VIRTUAL_FILE_IO_ENGINE: tokio-epoll-uring
-          PAGESERVER_GET_VECTORED_IMPL: vectored
-          PAGESERVER_GET_IMPL: vectored
-          PAGESERVER_VALIDATE_VEC_GET: true
 
       # Temporary disable this step until we figure out why it's so flaky
       # Ref https://github.com/neondatabase/neon/issues/4540
       - name: Merge and upload coverage data
         if: |
           false &&
-          inputs.build-type == 'debug' && matrix.pg_version == 'v14'
+          inputs.build-type == 'debug' && matrix.pg_version == 'v16'
         uses: ./.github/actions/save-coverage-data

.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:
@@ -63,11 +67,13 @@ jobs:
           - DEFAULT_PG_VERSION: 16
             PLATFORM: "neon-staging"
             region_id: ${{ github.event.inputs.region_id || 'aws-us-east-2' }}
-            provisioner: 'k8s-pod' 
+            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'
-            provisioner: 'k8s-neonvm'
+            RUNNER: [ self-hosted, eastus2, x64 ]
+            IMAGE: neondatabase/build-tools:pinned
     env:
       TEST_PG_BENCH_DURATIONS_MATRIX: "300"
       TEST_PG_BENCH_SCALES_MATRIX: "10,100"
@@ -78,14 +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:
@@ -100,7 +113,6 @@ jobs:
         region_id: ${{ matrix.region_id }}
         postgres_version: ${{ env.DEFAULT_PG_VERSION }}
         api_key: ${{ secrets.NEON_STAGING_API_KEY }}
-        provisioner: ${{ matrix.provisioner }}
 
     - name: Run benchmark
       uses: ./.github/actions/run-python-test-set
@@ -150,7 +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' ) }}
@@ -164,6 +176,7 @@ jobs:
     steps:
     - uses: actions/checkout@v4
 
+    
     - name: Download Neon artifact
       uses: ./.github/actions/download
       with:
@@ -171,7 +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 }}
@@ -179,12 +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: ${{ env.BUILD_TYPE }}
@@ -216,11 +232,11 @@ jobs:
     # Create matrices for the benchmarking jobs, so we run benchmarks on rds only once a week (on Saturday)
     #
     # Available platforms:
-    # - neon-captest-new: Freshly created project (1 CU)
-    # - neon-captest-freetier: Use freetier-sized compute (0.25 CU)
+    # - neonvm-captest-new: Freshly created project (1 CU)
+    # - neonvm-captest-freetier: Use freetier-sized compute (0.25 CU)
     # - neonvm-captest-azure-new: Freshly created project (1 CU) in azure region
     # - neonvm-captest-azure-freetier: Use freetier-sized compute (0.25 CU) in azure region
-    # - neon-captest-reuse: Reusing existing project
+    # - neonvm-captest-reuse: Reusing existing project
     # - rds-aurora: Aurora Postgres Serverless v2 with autoscaling from 0.5 to 2 ACUs
     # - rds-postgres: RDS Postgres db.m5.large instance (2 vCPU, 8 GiB) with gp3 EBS storage
     env:
@@ -237,6 +253,9 @@ jobs:
       id: pgbench-compare-matrix
       run: |
         region_id_default=${{ 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
@@ -245,24 +264,24 @@ jobs:
             "'"$region_id_default"'"
             ],
           "platform": [
-            "neon-captest-new",
-            "neon-captest-reuse",
+            "neonvm-captest-new",
+            "neonvm-captest-reuse",
             "neonvm-captest-new"
           ],
           "db_size": [ "10gb" ],
-          "include": [{ "pg_version": 16, "region_id": "'"$region_id_default"'", "platform": "neon-captest-freetier",         "db_size": "3gb"  },
-                      { "pg_version": 16, "region_id": "'"$region_id_default"'", "platform": "neon-captest-new",              "db_size": "50gb" },
-                      { "pg_version": 16, "region_id": "'"$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"},
-                                                     { "pg_version": 14, "region_id": "'"$region_id_default"'", "platform": "rds-aurora",   "db_size": "50gb"}]')
+          matrix=$(echo "$matrix" | jq '.include += [{ "pg_version": 14, "region_id": "'"$region_id_default"'", "platform": "rds-postgres", "db_size": "10gb","runner": '"$runner_default"', "image": "'"$image_default"'" }]')
         fi
 
         echo "matrix=$(echo "$matrix" | jq --compact-output '.')" >> $GITHUB_OUTPUT
@@ -272,7 +291,7 @@ jobs:
       run: |
         matrix='{
           "platform": [
-            "neon-captest-reuse"
+            "neonvm-captest-reuse"
           ]
         }'
 
@@ -288,7 +307,7 @@ jobs:
       run: |
         matrix='{
           "platform": [
-            "neon-captest-reuse"
+            "neonvm-captest-reuse"
           ],
           "scale": [
             "10"
@@ -305,6 +324,10 @@ jobs:
   pgbench-compare:
     if: ${{ github.event.inputs.run_only_pgvector_tests == 'false' || github.event.inputs.run_only_pgvector_tests == null }}
     needs: [ generate-matrices ]
+    permissions:
+      contents: write
+      statuses: write
+      id-token: write # Required for OIDC authentication in azure runners
 
     strategy:
       fail-fast: false
@@ -320,9 +343,9 @@ 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
@@ -331,6 +354,13 @@ jobs:
     steps:
     - uses: actions/checkout@v4
 
+    - name: Configure AWS credentials # necessary on Azure runners
+      uses: aws-actions/configure-aws-credentials@v4
+      with:
+        aws-region: eu-central-1
+        role-to-assume: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+        role-duration-seconds: 18000 # 5 hours
+        
     - name: Download Neon artifact
       uses: ./.github/actions/download
       with:
@@ -339,7 +369,7 @@ jobs:
         prefix: latest
 
     - name: Create Neon Project
-      if: contains(fromJson('["neon-captest-new", "neon-captest-freetier", "neonvm-captest-new", "neonvm-captest-freetier", "neonvm-azure-captest-freetier", "neonvm-azure-captest-new"]'), matrix.platform)
+      if: contains(fromJson('["neonvm-captest-new", "neonvm-captest-freetier", "neonvm-azure-captest-freetier", "neonvm-azure-captest-new"]'), matrix.platform)
       id: create-neon-project
       uses: ./.github/actions/neon-project-create
       with:
@@ -347,19 +377,18 @@ jobs:
         postgres_version: ${{ env.DEFAULT_PG_VERSION }}
         api_key: ${{ secrets.NEON_STAGING_API_KEY }}
         compute_units: ${{ (contains(matrix.platform, 'captest-freetier') && '[0.25, 0.25]') || '[1, 1]' }}
-        provisioner: ${{ (contains(matrix.platform, 'neonvm-') && 'k8s-neonvm') || 'k8s-pod' }}
 
     - name: Set up Connection String
       id: set-up-connstr
       run: |
         case "${PLATFORM}" in
-          neon-captest-reuse)
+          neonvm-captest-reuse)
             CONNSTR=${{ secrets.BENCHMARK_CAPTEST_CONNSTR }}
             ;;
           neonvm-captest-sharding-reuse)
             CONNSTR=${{ secrets.BENCHMARK_CAPTEST_SHARDING_CONNSTR }}
             ;;
-          neon-captest-new | neon-captest-freetier | neonvm-captest-new | neonvm-captest-freetier | neonvm-azure-captest-new | neonvm-azure-captest-freetier)
+          neonvm-captest-new | neonvm-captest-freetier | neonvm-azure-captest-new | neonvm-azure-captest-freetier)
             CONNSTR=${{ steps.create-neon-project.outputs.dsn }}
             ;;
           rds-aurora)
@@ -439,13 +468,21 @@ jobs:
         SLACK_BOT_TOKEN: ${{ secrets.SLACK_BOT_TOKEN }}
 
   pgbench-pgvector:
+    permissions:
+      contents: write
+      statuses: write
+      id-token: write # Required for OIDC authentication in azure runners
     strategy:
       fail-fast: false
       matrix:
         include:
-          - PLATFORM: "neon-captest-pgvector"
+          - PLATFORM: "neonvm-captest-pgvector"
+            RUNNER: [ self-hosted, us-east-2, x64 ]
+            IMAGE: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/build-tools:pinned
           - PLATFORM: "azure-captest-pgvector"
-            
+            RUNNER: [ self-hosted, eastus2, x64 ]
+            IMAGE: neondatabase/build-tools:pinned
+
     env:
       TEST_PG_BENCH_DURATIONS_MATRIX: "15m"
       TEST_PG_BENCH_SCALES_MATRIX: "1"
@@ -457,9 +494,9 @@ 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:
@@ -470,12 +507,12 @@ 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  
@@ -487,7 +524,7 @@ jobs:
       id: set-up-connstr
       run: |
         case "${PLATFORM}" in
-          neon-captest-pgvector)
+          neonvm-captest-pgvector)
             CONNSTR=${{ secrets.BENCHMARK_PGVECTOR_CONNSTR }}
             ;;
           azure-captest-pgvector)
@@ -500,6 +537,13 @@ 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
@@ -528,7 +572,7 @@ 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: ${{ !cancelled() }}
       uses: ./.github/actions/allure-report-generate
@@ -586,7 +630,7 @@ jobs:
       id: set-up-connstr
       run: |
         case "${PLATFORM}" in
-          neon-captest-reuse)
+          neonvm-captest-reuse)
             CONNSTR=${{ secrets.BENCHMARK_CAPTEST_CLICKBENCH_10M_CONNSTR }}
             ;;
           rds-aurora)
@@ -596,7 +640,7 @@ jobs:
             CONNSTR=${{ secrets.BENCHMARK_RDS_POSTGRES_CLICKBENCH_10M_CONNSTR }}
             ;;
           *)
-            echo >&2 "Unknown PLATFORM=${PLATFORM}. Allowed only 'neon-captest-reuse', 'rds-aurora', or 'rds-postgres'"
+            echo >&2 "Unknown PLATFORM=${PLATFORM}. Allowed only 'neonvm-captest-reuse', 'rds-aurora', or 'rds-postgres'"
             exit 1
             ;;
         esac
@@ -673,7 +717,7 @@ jobs:
     - name: Get Connstring Secret Name
       run: |
         case "${PLATFORM}" in
-          neon-captest-reuse)
+          neonvm-captest-reuse)
             ENV_PLATFORM=CAPTEST_TPCH
             ;;
           rds-aurora)
@@ -683,7 +727,7 @@ jobs:
             ENV_PLATFORM=RDS_AURORA_TPCH
             ;;
           *)
-            echo >&2 "Unknown PLATFORM=${PLATFORM}. Allowed only 'neon-captest-reuse', 'rds-aurora', or 'rds-postgres'"
+            echo >&2 "Unknown PLATFORM=${PLATFORM}. Allowed only 'neonvm-captest-reuse', 'rds-aurora', or 'rds-postgres'"
             exit 1
             ;;
         esac
@@ -760,7 +804,7 @@ jobs:
       id: set-up-connstr
       run: |
         case "${PLATFORM}" in
-          neon-captest-reuse)
+          neonvm-captest-reuse)
             CONNSTR=${{ secrets.BENCHMARK_USER_EXAMPLE_CAPTEST_CONNSTR }}
             ;;
           rds-aurora)
@@ -770,7 +814,7 @@ jobs:
             CONNSTR=${{ secrets.BENCHMARK_USER_EXAMPLE_RDS_POSTGRES_CONNSTR }}
             ;;
           *)
-            echo >&2 "Unknown PLATFORM=${PLATFORM}. Allowed only 'neon-captest-reuse', 'rds-aurora', or 'rds-postgres'"
+            echo >&2 "Unknown PLATFORM=${PLATFORM}. Allowed only 'neonvm-captest-reuse', 'rds-aurora', or 'rds-postgres'"
             exit 1
             ;;
         esac

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

@@ -72,6 +72,12 @@ jobs:
           username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
           password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
 
+      - uses: docker/login-action@v3
+        with:
+          registry: cache.neon.build
+          username: ${{ secrets.NEON_CI_DOCKERCACHE_USERNAME }}
+          password: ${{ secrets.NEON_CI_DOCKERCACHE_PASSWORD }}
+
       - uses: docker/build-push-action@v6
         with:
           context: .
@@ -79,8 +85,8 @@ jobs:
           push: true
           pull: true
           file: Dockerfile.build-tools
-          cache-from: type=registry,ref=neondatabase/build-tools:cache-${{ matrix.arch }}
-          cache-to: ${{ github.ref_name == 'main' && format('type=registry,ref=neondatabase/build-tools:cache-{0},mode=max', matrix.arch) || '' }}
+          cache-from: type=registry,ref=cache.neon.build/build-tools:cache-${{ matrix.arch }}
+          cache-to: ${{ github.ref_name == 'main' && format('type=registry,ref=cache.neon.build/build-tools:cache-{0},mode=max', matrix.arch) || '' }}
           tags: neondatabase/build-tools:${{ inputs.image-tag }}-${{ matrix.arch }}
 
       - name: Remove custom docker config directory

.github/workflows/build_and_test.yml

@@ -203,7 +203,8 @@ jobs:
       fail-fast: false
       matrix:
         arch: [ x64 ]
-        build-type: [ debug, release ]
+        # Do not build or run tests in debug for release branches
+        build-type: ${{ fromJson((startsWith(github.ref_name, 'release' && github.event_name == 'push')) && '["release"]' || '["debug", "release"]') }}
         include:
           - build-type: release
             arch: arm64
@@ -213,6 +214,8 @@ jobs:
       build-tools-image: ${{ needs.build-build-tools-image.outputs.image }}
       build-tag: ${{ needs.tag.outputs.build-tag }}
       build-type: ${{ matrix.build-type }}
+      # Run tests on all Postgres versions in release builds and only on the latest version in debug builds
+      pg-versions: ${{ matrix.build-type == 'release' && '["v14", "v15", "v16"]' || '["v16"]' }}
     secrets: inherit
 
   # Keep `benchmarks` job outside of `build-and-test-locally` workflow to make job failures non-blocking
@@ -286,9 +289,6 @@ jobs:
           PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
           TEST_RESULT_CONNSTR: "${{ secrets.REGRESS_TEST_RESULT_CONNSTR_NEW }}"
           PAGESERVER_VIRTUAL_FILE_IO_ENGINE: tokio-epoll-uring
-          PAGESERVER_GET_VECTORED_IMPL: vectored
-          PAGESERVER_GET_IMPL: vectored
-          PAGESERVER_VALIDATE_VEC_GET: false
       # XXX: no coverage data handling here, since benchmarks are run on release builds,
       # while coverage is currently collected for the debug ones
 
@@ -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 }}
@@ -499,6 +499,12 @@ jobs:
           username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
           password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
 
+      - uses: docker/login-action@v3
+        with:
+          registry: cache.neon.build
+          username: ${{ secrets.NEON_CI_DOCKERCACHE_USERNAME }}
+          password: ${{ secrets.NEON_CI_DOCKERCACHE_PASSWORD }}
+
       - uses: docker/build-push-action@v6
         with:
           context: .
@@ -510,9 +516,8 @@ jobs:
           push: true
           pull: true
           file: Dockerfile
-          cache-from: type=registry,ref=neondatabase/neon:cache-${{ matrix.arch }}
-          # 23.07.2024 temporarily disable cache saving in the registry as it is very slow
-          # cache-to: ${{ github.ref_name == 'main' && format('type=registry,ref=neondatabase/neon:cache-{0},mode=max', matrix.arch) || '' }}
+          cache-from: type=registry,ref=cache.neon.build/neon:cache-${{ matrix.arch }}
+          cache-to: ${{ github.ref_name == 'main' && format('type=registry,ref=cache.neon.build/neon:cache-{0},mode=max', matrix.arch) || '' }}
           tags: |
             neondatabase/neon:${{ needs.tag.outputs.build-tag }}-${{ matrix.arch }}
 
@@ -591,6 +596,12 @@ jobs:
           username: ${{ secrets.AWS_ACCESS_KEY_DEV }}
           password: ${{ secrets.AWS_SECRET_KEY_DEV }}
 
+      - uses: docker/login-action@v3
+        with:
+          registry: cache.neon.build
+          username: ${{ secrets.NEON_CI_DOCKERCACHE_USERNAME }}
+          password: ${{ secrets.NEON_CI_DOCKERCACHE_PASSWORD }}
+
       - name: Build compute-node image
         uses: docker/build-push-action@v6
         with:
@@ -604,9 +615,8 @@ jobs:
           push: true
           pull: true
           file: Dockerfile.compute-node
-          cache-from: type=registry,ref=neondatabase/compute-node-${{ matrix.version }}:cache-${{ matrix.arch }}
-          # 23.07.2024 temporarily disable cache saving in the registry as it is very slow
-          # cache-to: ${{ github.ref_name == 'main' && format('type=registry,ref=neondatabase/compute-node-{0}:cache-{1},mode=max', matrix.version, matrix.arch) || '' }}
+          cache-from: type=registry,ref=cache.neon.build/compute-node-${{ matrix.version }}:cache-${{ matrix.arch }}
+          cache-to: ${{ github.ref_name == 'main' && format('type=registry,ref=cache.neon.build/compute-node-{0}:cache-{1},mode=max', matrix.version, matrix.arch) || '' }}
           tags: |
             neondatabase/compute-node-${{ matrix.version }}:${{ needs.tag.outputs.build-tag }}-${{ matrix.arch }}
 
@@ -625,9 +635,8 @@ jobs:
           pull: true
           file: Dockerfile.compute-node
           target: neon-pg-ext-test
-          cache-from: type=registry,ref=neondatabase/neon-test-extensions-${{ matrix.version }}:cache-${{ matrix.arch }}
-          # 23.07.2024 temporarily disable cache saving in the registry as it is very slow
-          # cache-to: ${{ github.ref_name == 'main' && format('type=registry,ref=neondatabase/neon-test-extensions-{0}:cache-{1},mode=max', matrix.version, matrix.arch) || '' }}
+          cache-from: type=registry,ref=cache.neon.build/neon-test-extensions-${{ matrix.version }}:cache-${{ matrix.arch }}
+          cache-to: ${{ github.ref_name == 'main' && format('type=registry,ref=cache.neon.build/neon-test-extensions-{0}:cache-{1},mode=max', matrix.version, matrix.arch) || '' }}
           tags: |
             neondatabase/neon-test-extensions-${{ matrix.version }}:${{needs.tag.outputs.build-tag}}-${{ matrix.arch }}
 
@@ -827,6 +836,9 @@ jobs:
           rm -rf .docker-custom
 
   promote-images:
+    permissions:
+      contents: read  # This is required for actions/checkout
+      id-token: write # This is required for Azure Login to work.
     needs: [ check-permissions, tag, test-images, vm-compute-node-image ]
     runs-on: ubuntu-22.04
 
@@ -853,6 +865,28 @@ jobs:
                                                neondatabase/vm-compute-node-${version}:${{ needs.tag.outputs.build-tag }}
           done
 
+      - name: Azure login
+        if: github.ref_name == 'main'
+        uses: azure/login@6c251865b4e6290e7b78be643ea2d005bc51f69a  # @v2.1.1
+        with:
+          client-id: ${{ secrets.AZURE_DEV_CLIENT_ID }}
+          tenant-id: ${{ secrets.AZURE_TENANT_ID }}
+          subscription-id: ${{ secrets.AZURE_DEV_SUBSCRIPTION_ID }}
+
+      - name: Login to ACR
+        if: github.ref_name == 'main'
+        run: |
+          az acr login --name=neoneastus2
+
+      - name: Copy docker images to ACR-dev
+        if: github.ref_name == 'main'
+        run: |
+          for image in neon compute-tools {vm-,}compute-node-{v14,v15,v16}; do
+            docker buildx imagetools create \
+              -t neoneastus2.azurecr.io/neondatabase/${image}:${{ needs.tag.outputs.build-tag }} \
+                                        neondatabase/${image}:${{ needs.tag.outputs.build-tag }}
+          done
+
       - name: Add latest tag to images
         if: github.ref_name == 'main'
         run: |

.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

@@ -13,6 +13,7 @@ on:
     paths:
       - '.github/workflows/pg-clients.yml'
       - 'test_runner/pg_clients/**'
+      - 'test_runner/logical_repl/**'
       - 'poetry.lock'
   workflow_dispatch:
 
@@ -49,6 +50,101 @@ jobs:
       image-tag: ${{ needs.check-build-tools-image.outputs.image-tag }}
     secrets: inherit
 
+  test-logical-replication:
+    needs: [ build-build-tools-image ]
+    runs-on: ubuntu-22.04
+
+    container:
+      image: ${{ needs.build-build-tools-image.outputs.image }}
+      credentials:
+        username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
+        password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
+      options: --init --user root
+    services:
+      clickhouse:
+        image: clickhouse/clickhouse-server:24.6.3.64
+        ports:
+          - 9000:9000
+          - 8123:8123
+      zookeeper:
+        image: quay.io/debezium/zookeeper:2.7
+        ports:
+          - 2181:2181
+      kafka:
+        image: quay.io/debezium/kafka:2.7
+        env:
+          ZOOKEEPER_CONNECT: "zookeeper:2181"
+          KAFKA_ADVERTISED_LISTENERS: PLAINTEXT://kafka:9092
+          KAFKA_BROKER_ID: 1
+          KAFKA_OFFSETS_TOPIC_REPLICATION_FACTOR: 1
+          KAFKA_JMX_PORT: 9991
+        ports:
+          - 9092:9092
+      debezium:
+        image: quay.io/debezium/connect:2.7
+        env:
+          BOOTSTRAP_SERVERS: kafka:9092
+          GROUP_ID: 1
+          CONFIG_STORAGE_TOPIC: debezium-config
+          OFFSET_STORAGE_TOPIC: debezium-offset
+          STATUS_STORAGE_TOPIC: debezium-status
+          DEBEZIUM_CONFIG_CONNECTOR_CLASS: io.debezium.connector.postgresql.PostgresConnector
+        ports:
+          - 8083:8083
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Download Neon artifact
+        uses: ./.github/actions/download
+        with:
+          name: neon-${{ runner.os }}-${{ runner.arch }}-release-artifact
+          path: /tmp/neon/
+          prefix: latest
+
+      - name: Create Neon Project
+        id: create-neon-project
+        uses: ./.github/actions/neon-project-create
+        with:
+          api_key: ${{ secrets.NEON_STAGING_API_KEY }}
+          postgres_version: ${{ env.DEFAULT_PG_VERSION }}
+
+      - name: Run tests
+        uses: ./.github/actions/run-python-test-set
+        with:
+          build_type: remote
+          test_selection: logical_repl
+          run_in_parallel: false
+          extra_params: -m remote_cluster
+          pg_version: ${{ env.DEFAULT_PG_VERSION }}
+        env:
+          BENCHMARK_CONNSTR: ${{ steps.create-neon-project.outputs.dsn }}
+
+      - name: Delete Neon Project
+        if: always()
+        uses: ./.github/actions/neon-project-delete
+        with:
+          project_id: ${{ steps.create-neon-project.outputs.project_id }}
+          api_key: ${{ secrets.NEON_STAGING_API_KEY }}
+
+      - name: Create Allure report
+        if: ${{ !cancelled() }}
+        id: create-allure-report
+        uses: ./.github/actions/allure-report-generate
+        with:
+          store-test-results-into-db: true
+        env:
+          REGRESS_TEST_RESULT_CONNSTR_NEW: ${{ secrets.REGRESS_TEST_RESULT_CONNSTR_NEW }}
+
+      - name: Post to a Slack channel
+        if: github.event.schedule && failure()
+        uses: slackapi/slack-github-action@v1
+        with:
+          channel-id: "C06KHQVQ7U3" # on-call-qa-staging-stream
+          slack-message: |
+            Testing the logical replication: <${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}|${{ job.status }}> (<${{ steps.create-allure-report.outputs.report-url }}|test report>)
+        env:
+          SLACK_BOT_TOKEN: ${{ secrets.SLACK_BOT_TOKEN }}
+
   test-postgres-client-libs:
     needs: [ build-build-tools-image ]
     runs-on: ubuntu-22.04

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

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

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

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

CODEOWNERS

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

Cargo.toml

@@ -9,7 +9,10 @@ members = [
     "pageserver/ctl",
     "pageserver/client",
     "pageserver/pagebench",
-    "proxy",
+    "proxy/core",
+    "proxy/sasl",
+    "proxy/proxy",
+    "proxy/pg_sni_router",
     "safekeeper",
     "storage_broker",
     "storage_controller",
@@ -126,7 +129,7 @@ parquet = { version = "51.0.0", default-features = false, features = ["zstd"] }
 parquet_derive = "51.0.0"
 pbkdf2 = { version = "0.12.1", features = ["simple", "std"] }
 pin-project-lite = "0.2"
-procfs = "0.14"
+procfs = "0.16"
 prometheus = {version = "0.13", default-features=false, features = ["process"]} # removes protobuf dependency
 prost = "0.11"
 rand = "0.8"
@@ -184,6 +187,7 @@ 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=1.79.0
+ENV RUSTC_VERSION=1.80.1
 ENV RUSTUP_HOME="/home/nonroot/.rustup"
 ENV PATH="/home/nonroot/.cargo/bin:${PATH}"
+ARG RUSTFILT_VERSION=0.2.1
+ARG CARGO_HAKARI_VERSION=0.9.30
+ARG CARGO_DENY_VERSION=0.16.1
+ARG CARGO_HACK_VERSION=0.6.31
+ARG CARGO_NEXTEST_VERSION=0.9.72
 RUN curl -sSO https://static.rust-lang.org/rustup/dist/$(uname -m)-unknown-linux-gnu/rustup-init && whoami && \
 	chmod +x rustup-init && \
 	./rustup-init -y --default-toolchain ${RUSTC_VERSION} && \
@@ -203,15 +208,13 @@ RUN curl -sSO https://static.rust-lang.org/rustup/dist/$(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 . && \
@@ -657,7 +657,7 @@ RUN curl -sSO https://static.rust-lang.org/rustup/dist/$(uname -m)-unknown-linux
     chmod +x rustup-init && \
     ./rustup-init -y --no-modify-path --profile minimal --default-toolchain stable && \
     rm rustup-init && \
-    cargo install --locked --version 0.10.2 cargo-pgrx && \
+    cargo install --locked --version 0.11.3 cargo-pgrx && \
     /bin/bash -c 'cargo pgrx init --pg${PG_VERSION:1}=/usr/local/pgsql/bin/pg_config'
 
 USER root
@@ -672,10 +672,15 @@ USER root
 FROM rust-extensions-build AS pg-jsonschema-pg-build
 ARG PG_VERSION
 
-RUN wget https://github.com/supabase/pg_jsonschema/archive/refs/tags/v0.2.0.tar.gz -O pg_jsonschema.tar.gz && \
-    echo "9118fc508a6e231e7a39acaa6f066fcd79af17a5db757b47d2eefbe14f7794f0 pg_jsonschema.tar.gz" | sha256sum --check && \
+RUN wget https://github.com/supabase/pg_jsonschema/archive/refs/tags/v0.3.1.tar.gz -O pg_jsonschema.tar.gz && \
+    echo "61df3db1ed83cf24f6aa39c826f8818bfa4f0bd33b587fd6b2b1747985642297 pg_jsonschema.tar.gz" | sha256sum --check && \
     mkdir pg_jsonschema-src && cd pg_jsonschema-src && tar xzf ../pg_jsonschema.tar.gz --strip-components=1 -C . && \
-    sed -i 's/pgrx = "0.10.2"/pgrx = { version = "0.10.2", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    # see commit 252b3685a27a0f4c31a0f91e983c6314838e89e8
+    # `unsafe-postgres` feature allows to build pgx extensions
+    # against postgres forks that decided to change their ABI name (like us).
+    # With that we can build extensions without forking them and using stock
+    # pgx. As this feature is new few manual version bumps were required.
+    sed -i 's/pgrx = "0.11.3"/pgrx = { version = "0.11.3", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
     cargo pgrx install --release && \
     echo "trusted = true" >> /usr/local/pgsql/share/extension/pg_jsonschema.control
 
@@ -689,10 +694,10 @@ RUN wget https://github.com/supabase/pg_jsonschema/archive/refs/tags/v0.2.0.tar.
 FROM rust-extensions-build AS pg-graphql-pg-build
 ARG PG_VERSION
 
-RUN wget https://github.com/supabase/pg_graphql/archive/refs/tags/v1.4.0.tar.gz -O pg_graphql.tar.gz && \
-    echo "bd8dc7230282b3efa9ae5baf053a54151ed0e66881c7c53750e2d0c765776edc pg_graphql.tar.gz" | sha256sum --check && \
+RUN wget https://github.com/supabase/pg_graphql/archive/refs/tags/v1.5.7.tar.gz -O pg_graphql.tar.gz && \
+    echo "2b3e567a5b31019cb97ae0e33263c1bcc28580be5a444ac4c8ece5c4be2aea41 pg_graphql.tar.gz" | sha256sum --check && \
     mkdir pg_graphql-src && cd pg_graphql-src && tar xzf ../pg_graphql.tar.gz --strip-components=1 -C . && \
-    sed -i 's/pgrx = "=0.10.2"/pgrx = { version = "0.10.2", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    sed -i 's/pgrx = "=0.11.3"/pgrx = { version = "0.11.3", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
     cargo pgrx install --release && \
     # it's needed to enable extension because it uses untrusted C language
     sed -i 's/superuser = false/superuser = true/g' /usr/local/pgsql/share/extension/pg_graphql.control && \
@@ -712,6 +717,9 @@ ARG PG_VERSION
 RUN wget https://github.com/kelvich/pg_tiktoken/archive/26806147b17b60763039c6a6878884c41a262318.tar.gz -O pg_tiktoken.tar.gz && \
     echo "e64e55aaa38c259512d3e27c572da22c4637418cf124caba904cd50944e5004e pg_tiktoken.tar.gz" | sha256sum --check && \
     mkdir pg_tiktoken-src && cd pg_tiktoken-src && tar xzf ../pg_tiktoken.tar.gz --strip-components=1 -C . && \
+    # TODO update pgrx version in the pg_tiktoken repo and remove this line
+    sed -i 's/pgrx = { version = "=0.10.2",/pgrx = { version = "0.11.3",/g' Cargo.toml && \
+    sed -i 's/pgrx-tests = "=0.10.2"/pgrx-tests = "0.11.3"/g' Cargo.toml && \
     cargo pgrx install --release && \
     echo "trusted = true" >> /usr/local/pgsql/share/extension/pg_tiktoken.control
 
@@ -725,14 +733,10 @@ RUN wget https://github.com/kelvich/pg_tiktoken/archive/26806147b17b60763039c6a6
 FROM rust-extensions-build AS pg-pgx-ulid-build
 ARG PG_VERSION
 
-RUN wget https://github.com/pksunkara/pgx_ulid/archive/refs/tags/v0.1.3.tar.gz -O pgx_ulid.tar.gz && \
-    echo "ee5db82945d2d9f2d15597a80cf32de9dca67b897f605beb830561705f12683c pgx_ulid.tar.gz" | sha256sum --check && \
+RUN wget https://github.com/pksunkara/pgx_ulid/archive/refs/tags/v0.1.5.tar.gz -O pgx_ulid.tar.gz && \
+    echo "9d1659a2da65af0133d5451c454de31b37364e3502087dadf579f790bc8bef17 pgx_ulid.tar.gz" | sha256sum --check && \
     mkdir pgx_ulid-src && cd pgx_ulid-src && tar xzf ../pgx_ulid.tar.gz --strip-components=1 -C . && \
-    echo "******************* Apply a patch for Postgres 16 support; delete in the next release ******************" && \
-    wget https://github.com/pksunkara/pgx_ulid/commit/f84954cf63fc8c80d964ac970d9eceed3c791196.patch && \
-    patch -p1 < f84954cf63fc8c80d964ac970d9eceed3c791196.patch && \
-    echo "********************************************************************************************************" && \
-    sed -i 's/pgrx       = "=0.10.2"/pgrx = { version = "=0.10.2", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
+    sed -i 's/pgrx       = "^0.11.2"/pgrx = { version = "=0.11.3", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
     cargo pgrx install --release && \
     echo "trusted = true" >> /usr/local/pgsql/share/extension/ulid.control
 
@@ -746,7 +750,7 @@ RUN wget https://github.com/pksunkara/pgx_ulid/archive/refs/tags/v0.1.3.tar.gz -
 FROM build-deps AS wal2json-pg-build
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
-ENV PATH "/usr/local/pgsql/bin/:$PATH"
+ENV PATH="/usr/local/pgsql/bin/:$PATH"
 RUN wget https://github.com/eulerto/wal2json/archive/refs/tags/wal2json_2_5.tar.gz && \
     echo "b516653575541cf221b99cf3f8be9b6821f6dbcfc125675c85f35090f824f00e wal2json_2_5.tar.gz" | sha256sum --check && \
     mkdir wal2json-src && cd wal2json-src && tar xzf ../wal2json_2_5.tar.gz --strip-components=1 -C . && \
@@ -762,7 +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 . && \
@@ -779,7 +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 . && \
@@ -929,7 +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
@@ -941,7 +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
@@ -956,6 +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
@@ -1028,6 +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"]

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/bin/neon_local.rs

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

control_plane/src/local_env.rs

@@ -151,23 +151,31 @@ pub struct NeonBroker {
 pub struct NeonStorageControllerConf {
     /// Heartbeat timeout before marking a node offline
     #[serde(with = "humantime_serde")]
-    pub max_unavailable: Duration,
+    pub max_offline: Duration,
+
+    #[serde(with = "humantime_serde")]
+    pub max_warming_up: Duration,
 
     /// Threshold for auto-splitting a tenant into shards
     pub split_threshold: Option<u64>,
+
+    pub max_secondary_lag_bytes: Option<u64>,
 }
 
 impl NeonStorageControllerConf {
     // Use a shorter pageserver unavailability interval than the default to speed up tests.
-    const DEFAULT_MAX_UNAVAILABLE_INTERVAL: std::time::Duration =
-        std::time::Duration::from_secs(10);
+    const DEFAULT_MAX_OFFLINE_INTERVAL: std::time::Duration = std::time::Duration::from_secs(10);
+
+    const DEFAULT_MAX_WARMING_UP_INTERVAL: std::time::Duration = std::time::Duration::from_secs(30);
 }
 
 impl Default for NeonStorageControllerConf {
     fn default() -> Self {
         Self {
-            max_unavailable: Self::DEFAULT_MAX_UNAVAILABLE_INTERVAL,
+            max_offline: Self::DEFAULT_MAX_OFFLINE_INTERVAL,
+            max_warming_up: Self::DEFAULT_MAX_WARMING_UP_INTERVAL,
             split_threshold: None,
+            max_secondary_lag_bytes: None,
         }
     }
 }
@@ -509,7 +517,6 @@ impl LocalEnv {
                 #[derive(serde::Serialize, serde::Deserialize)]
                 // (allow unknown fields, unlike PageServerConf)
                 struct PageserverConfigTomlSubset {
-                    id: NodeId,
                     listen_pg_addr: String,
                     listen_http_addr: String,
                     pg_auth_type: AuthType,
@@ -521,18 +528,30 @@ impl LocalEnv {
                         .with_context(|| format!("read {:?}", config_toml_path))?,
                 )
                 .context("parse pageserver.toml")?;
+                let identity_toml_path = dentry.path().join("identity.toml");
+                #[derive(serde::Serialize, serde::Deserialize)]
+                struct IdentityTomlSubset {
+                    id: NodeId,
+                }
+                let identity_toml: IdentityTomlSubset = toml_edit::de::from_str(
+                    &std::fs::read_to_string(&identity_toml_path)
+                        .with_context(|| format!("read {:?}", identity_toml_path))?,
+                )
+                .context("parse identity.toml")?;
                 let PageserverConfigTomlSubset {
-                    id: config_toml_id,
                     listen_pg_addr,
                     listen_http_addr,
                     pg_auth_type,
                     http_auth_type,
                 } = config_toml;
+                let IdentityTomlSubset {
+                    id: identity_toml_id,
+                } = identity_toml;
                 let conf = PageServerConf {
                     id: {
                         anyhow::ensure!(
-                            config_toml_id == id,
-                            "id mismatch: config_toml.id={config_toml_id} id={id}",
+                            identity_toml_id == id,
+                            "id mismatch: identity.toml:id={identity_toml_id} pageserver_(.*) id={id}",
                         );
                         id
                     },

control_plane/src/pageserver.rs

@@ -127,10 +127,13 @@ impl PageServerNode {
         }
 
         // Apply the user-provided overrides
-        overrides.push(
-            toml_edit::ser::to_string_pretty(&conf)
-                .expect("we deserialized this from toml earlier"),
-        );
+        overrides.push({
+            let mut doc =
+                toml_edit::ser::to_document(&conf).expect("we deserialized this from toml earlier");
+            // `id` is written out to `identity.toml` instead of `pageserver.toml`
+            doc.remove("id").expect("it's part of the struct");
+            doc.to_string()
+        });
 
         // Turn `overrides` into a toml document.
         // TODO: above code is legacy code, it should be refactored to use toml_edit directly.

control_plane/src/storage_controller.rs

@@ -5,8 +5,9 @@ use crate::{
 use camino::{Utf8Path, Utf8PathBuf};
 use pageserver_api::{
     controller_api::{
-        NodeConfigureRequest, NodeRegisterRequest, TenantCreateRequest, TenantCreateResponse,
-        TenantLocateResponse, TenantShardMigrateRequest, TenantShardMigrateResponse,
+        NodeConfigureRequest, NodeDescribeResponse, NodeRegisterRequest, TenantCreateRequest,
+        TenantCreateResponse, TenantLocateResponse, TenantShardMigrateRequest,
+        TenantShardMigrateResponse,
     },
     models::{
         TenantShardSplitRequest, TenantShardSplitResponse, TimelineCreateRequest, TimelineInfo,
@@ -353,8 +354,10 @@ impl StorageController {
             "--dev",
             "--database-url",
             &database_url,
-            "--max-unavailable-interval",
-            &humantime::Duration::from(self.config.max_unavailable).to_string(),
+            "--max-offline-interval",
+            &humantime::Duration::from(self.config.max_offline).to_string(),
+            "--max-warming-up-interval",
+            &humantime::Duration::from(self.config.max_warming_up).to_string(),
         ]
         .into_iter()
         .map(|s| s.to_string())
@@ -380,6 +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()
@@ -625,6 +632,15 @@ impl StorageController {
         .await
     }
 
+    pub async fn node_list(&self) -> anyhow::Result<Vec<NodeDescribeResponse>> {
+        self.dispatch::<(), Vec<NodeDescribeResponse>>(
+            Method::GET,
+            "control/v1/node".to_string(),
+            None,
+        )
+        .await
+    }
+
     #[instrument(skip(self))]
     pub async fn ready(&self) -> anyhow::Result<()> {
         self.dispatch::<(), ()>(Method::GET, "ready".to_string(), None)

deny.toml

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

docker-compose/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/synthetic-size.md

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

libs/pageserver_api/src/controller_api.rs

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

libs/pageserver_api/src/key.rs

@@ -107,7 +107,10 @@ impl Key {
     /// As long as Neon does not support tablespace (because of lack of access to local file system),
     /// we can assume that only some predefined namespace OIDs are used which can fit in u16
     pub fn to_i128(&self) -> i128 {
-        assert!(self.field2 <= 0xFFFF || self.field2 == 0xFFFFFFFF || self.field2 == 0x22222222);
+        assert!(
+            self.field2 <= 0xFFFF || self.field2 == 0xFFFFFFFF || self.field2 == 0x22222222,
+            "invalid key: {self}",
+        );
         (((self.field1 & 0x7F) as i128) << 120)
             | (((self.field2 & 0xFFFF) as i128) << 104)
             | ((self.field3 as i128) << 72)

libs/pageserver_api/src/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/postgres_ffi/src/controlfile_utils.rs

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

libs/postgres_ffi/src/pg_constants.rs

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

libs/postgres_ffi/src/xlog_utils.rs

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

libs/postgres_ffi/wal_craft/src/xlog_utils_test.rs

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

libs/remote_storage/src/azure_blob.rs

@@ -33,6 +33,7 @@ use tracing::debug;
 use utils::backoff;
 
 use crate::metrics::{start_measuring_requests, AttemptOutcome, RequestKind};
+use crate::ListingObject;
 use crate::{
     config::AzureConfig, error::Cancelled, ConcurrencyLimiter, Download, DownloadError, Listing,
     ListingMode, RemotePath, RemoteStorage, StorageMetadata, TimeTravelError, TimeoutOrCancel,
@@ -352,7 +353,12 @@ impl RemoteStorage for AzureBlobStorage {
                 let blob_iter = entry
                     .blobs
                     .blobs()
-                    .map(|k| self.name_to_relative_path(&k.name));
+                    .map(|k| ListingObject{
+                        key: self.name_to_relative_path(&k.name),
+                        last_modified: k.properties.last_modified.into(),
+                        size: k.properties.content_length,
+                    }
+                    );
 
                 for key in blob_iter {
                     res.keys.push(key);

libs/remote_storage/src/lib.rs

@@ -144,15 +144,23 @@ impl RemotePath {
 ///
 /// The WithDelimiter mode will populate `prefixes` and `keys` in the result.  The
 /// NoDelimiter mode will only populate `keys`.
+#[derive(Copy, Clone)]
 pub enum ListingMode {
     WithDelimiter,
     NoDelimiter,
 }
 
+#[derive(PartialEq, Eq, Debug)]
+pub struct ListingObject {
+    pub key: RemotePath,
+    pub last_modified: SystemTime,
+    pub size: u64,
+}
+
 #[derive(Default)]
 pub struct Listing {
     pub prefixes: Vec<RemotePath>,
-    pub keys: Vec<RemotePath>,
+    pub keys: Vec<ListingObject>,
 }
 
 /// Storage (potentially remote) API to manage its state.
@@ -188,7 +196,7 @@ pub trait RemoteStorage: Send + Sync + 'static {
         mode: ListingMode,
         max_keys: Option<NonZeroU32>,
         cancel: &CancellationToken,
-    ) -> impl Stream<Item = Result<Listing, DownloadError>>;
+    ) -> impl Stream<Item = Result<Listing, DownloadError>> + Send;
 
     async fn list(
         &self,
@@ -201,7 +209,7 @@ pub trait RemoteStorage: Send + Sync + 'static {
         let mut combined = stream.next().await.expect("At least one item required")?;
         while let Some(list) = stream.next().await {
             let list = list?;
-            combined.keys.extend_from_slice(&list.keys);
+            combined.keys.extend(list.keys.into_iter());
             combined.prefixes.extend_from_slice(&list.prefixes);
         }
         Ok(combined)
@@ -345,10 +353,10 @@ impl<Other: RemoteStorage> GenericRemoteStorage<Arc<Other>> {
         mode: ListingMode,
         max_keys: Option<NonZeroU32>,
         cancel: &'a CancellationToken,
-    ) -> impl Stream<Item = Result<Listing, DownloadError>> + 'a {
+    ) -> impl Stream<Item = Result<Listing, DownloadError>> + 'a + Send {
         match self {
             Self::LocalFs(s) => Box::pin(s.list_streaming(prefix, mode, max_keys, cancel))
-                as Pin<Box<dyn Stream<Item = Result<Listing, DownloadError>>>>,
+                as Pin<Box<dyn Stream<Item = Result<Listing, DownloadError>> + Send>>,
             Self::AwsS3(s) => Box::pin(s.list_streaming(prefix, mode, max_keys, cancel)),
             Self::AzureBlob(s) => Box::pin(s.list_streaming(prefix, mode, max_keys, cancel)),
             Self::Unreliable(s) => Box::pin(s.list_streaming(prefix, mode, max_keys, cancel)),

libs/remote_storage/src/local_fs.rs

@@ -23,8 +23,8 @@ use tokio_util::{io::ReaderStream, sync::CancellationToken};
 use utils::crashsafe::path_with_suffix_extension;
 
 use crate::{
-    Download, DownloadError, Listing, ListingMode, RemotePath, TimeTravelError, TimeoutOrCancel,
-    REMOTE_STORAGE_PREFIX_SEPARATOR,
+    Download, DownloadError, Listing, ListingMode, ListingObject, RemotePath, TimeTravelError,
+    TimeoutOrCancel, REMOTE_STORAGE_PREFIX_SEPARATOR,
 };
 
 use super::{RemoteStorage, StorageMetadata};
@@ -357,19 +357,29 @@ impl RemoteStorage for LocalFs {
                 .list_recursive(prefix)
                 .await
                 .map_err(DownloadError::Other)?;
-            let keys = keys
+            let objects = keys
                 .into_iter()
-                .filter(|k| {
+                .filter_map(|k| {
                     let path = k.with_base(&self.storage_root);
-                    !path.is_dir()
+                    if path.is_dir() {
+                        None
+                    } else {
+                        Some(ListingObject {
+                            key: k.clone(),
+                            // LocalFs is just for testing, so just specify a dummy time
+                            last_modified: SystemTime::now(),
+                            size: 0,
+                        })
+                    }
                 })
                 .collect();
 
             if let ListingMode::NoDelimiter = mode {
-                result.keys = keys;
+                result.keys = objects;
             } else {
                 let mut prefixes = HashSet::new();
-                for key in keys {
+                for object in objects {
+                    let key = object.key;
                     // If the part after the prefix includes a "/", take only the first part and put it in `prefixes`.
                     let relative_key = if let Some(prefix) = prefix {
                         let mut prefix = prefix.clone();
@@ -398,9 +408,12 @@ impl RemoteStorage for LocalFs {
                             .to_owned();
                         prefixes.insert(first_part);
                     } else {
-                        result
-                            .keys
-                            .push(RemotePath::from_string(&relative_key).unwrap());
+                        result.keys.push(ListingObject {
+                            key: RemotePath::from_string(&relative_key).unwrap(),
+                            // LocalFs is just for testing
+                            last_modified: SystemTime::now(),
+                            size: 0,
+                        });
                     }
                 }
                 result.prefixes = prefixes
@@ -950,7 +963,11 @@ mod fs_tests {
             .await?;
         assert!(listing.prefixes.is_empty());
         assert_eq!(
-            listing.keys.into_iter().collect::<HashSet<_>>(),
+            listing
+                .keys
+                .into_iter()
+                .map(|o| o.key)
+                .collect::<HashSet<_>>(),
             HashSet::from([uncle.clone(), child.clone(), child_sibling.clone()])
         );
 
@@ -975,7 +992,7 @@ mod fs_tests {
             )
             .await?;
         assert_eq!(
-            listing.keys,
+            listing.keys.into_iter().map(|o| o.key).collect::<Vec<_>>(),
             [RemotePath::from_string("uncle").unwrap()].to_vec()
         );
         assert_eq!(
@@ -992,7 +1009,7 @@ mod fs_tests {
                 &cancel,
             )
             .await?;
-        assert_eq!(listing.keys, [].to_vec());
+        assert_eq!(listing.keys, vec![]);
         assert_eq!(
             listing.prefixes,
             [RemotePath::from_string("grandparent").unwrap()].to_vec()
@@ -1007,7 +1024,7 @@ mod fs_tests {
                 &cancel,
             )
             .await?;
-        assert_eq!(listing.keys, [].to_vec());
+        assert_eq!(listing.keys, vec![]);
         assert_eq!(
             listing.prefixes,
             [RemotePath::from_string("grandparent").unwrap()].to_vec()
@@ -1040,7 +1057,7 @@ mod fs_tests {
                 &cancel,
             )
             .await?;
-        assert_eq!(listing.keys, [].to_vec());
+        assert_eq!(listing.keys, vec![]);
 
         let mut found_prefixes = listing.prefixes.clone();
         found_prefixes.sort();

libs/remote_storage/src/s3_bucket.rs

@@ -44,8 +44,9 @@ use crate::{
     error::Cancelled,
     metrics::{start_counting_cancelled_wait, start_measuring_requests},
     support::PermitCarrying,
-    ConcurrencyLimiter, Download, DownloadError, Listing, ListingMode, RemotePath, RemoteStorage,
-    TimeTravelError, TimeoutOrCancel, MAX_KEYS_PER_DELETE, REMOTE_STORAGE_PREFIX_SEPARATOR,
+    ConcurrencyLimiter, Download, DownloadError, Listing, ListingMode, ListingObject, RemotePath,
+    RemoteStorage, TimeTravelError, TimeoutOrCancel, MAX_KEYS_PER_DELETE,
+    REMOTE_STORAGE_PREFIX_SEPARATOR,
 };
 
 use crate::metrics::AttemptOutcome;
@@ -548,9 +549,29 @@ impl RemoteStorage for S3Bucket {
                 let mut result = Listing::default();
 
                 for object in keys {
-                    let object_path = object.key().expect("response does not contain a key");
-                    let remote_path = self.s3_object_to_relative_path(object_path);
-                    result.keys.push(remote_path);
+                    let key = object.key().expect("response does not contain a key");
+                    let key = self.s3_object_to_relative_path(key);
+
+                    let last_modified = match object.last_modified.map(SystemTime::try_from) {
+                        Some(Ok(t)) => t,
+                        Some(Err(_)) => {
+                            tracing::warn!("Remote storage last_modified {:?} for {} is out of bounds",
+                                object.last_modified, key
+                        );
+                            SystemTime::now()
+                        },
+                        None => {
+                            SystemTime::now()
+                        }
+                    };
+
+                    let size = object.size.unwrap_or(0) as u64;
+
+                    result.keys.push(ListingObject{
+                        key,
+                        last_modified,
+                        size,
+                    });
                     if let Some(mut mk) = max_keys {
                         assert!(mk > 0);
                         mk -= 1;

libs/remote_storage/src/simulate_failures.rs

@@ -114,7 +114,7 @@ impl RemoteStorage for UnreliableWrapper {
         mode: ListingMode,
         max_keys: Option<NonZeroU32>,
         cancel: &CancellationToken,
-    ) -> impl Stream<Item = Result<Listing, DownloadError>> {
+    ) -> impl Stream<Item = Result<Listing, DownloadError>> + Send {
         async_stream::stream! {
             self.attempt(RemoteOp::ListPrefixes(prefix.cloned()))
                 .map_err(DownloadError::Other)?;

libs/remote_storage/tests/common/tests.rs

@@ -156,6 +156,7 @@ async fn list_no_delimiter_works(
         .context("client list root files failure")?
         .keys
         .into_iter()
+        .map(|o| o.key)
         .collect::<HashSet<_>>();
     assert_eq!(
         root_files,
@@ -182,6 +183,7 @@ async fn list_no_delimiter_works(
         .context("client list nested files failure")?
         .keys
         .into_iter()
+        .map(|o| o.key)
         .collect::<HashSet<_>>();
     let trim_remote_blobs: HashSet<_> = ctx
         .remote_blobs

libs/remote_storage/tests/test_real_s3.rs

@@ -81,6 +81,7 @@ async fn s3_time_travel_recovery_works(ctx: &mut MaybeEnabledStorage) -> anyhow:
                 .context("list root files failure")?
                 .keys
                 .into_iter()
+                .map(|o| o.key)
                 .collect::<HashSet<_>>(),
         )
     }

libs/utils/src/auth.rs

@@ -18,20 +18,20 @@ const STORAGE_TOKEN_ALGORITHM: Algorithm = Algorithm::EdDSA;
 #[derive(Debug, Serialize, Deserialize, Clone, Copy, PartialEq)]
 #[serde(rename_all = "lowercase")]
 pub enum Scope {
-    // Provides access to all data for a specific tenant (specified in `struct Claims` below)
+    /// Provides access to all data for a specific tenant (specified in `struct Claims` below)
     // TODO: join these two?
     Tenant,
-    // Provides blanket access to all tenants on the pageserver plus pageserver-wide APIs.
-    // Should only be used e.g. for status check/tenant creation/list.
+    /// Provides blanket access to all tenants on the pageserver plus pageserver-wide APIs.
+    /// Should only be used e.g. for status check/tenant creation/list.
     PageServerApi,
-    // Provides blanket access to all data on the safekeeper plus safekeeper-wide APIs.
-    // Should only be used e.g. for status check.
-    // Currently also used for connection from any pageserver to any safekeeper.
+    /// Provides blanket access to all data on the safekeeper plus safekeeper-wide APIs.
+    /// Should only be used e.g. for status check.
+    /// Currently also used for connection from any pageserver to any safekeeper.
     SafekeeperData,
-    // The scope used by pageservers in upcalls to storage controller and cloud control plane
+    /// The scope used by pageservers in upcalls to storage controller and cloud control plane
     #[serde(rename = "generations_api")]
     GenerationsApi,
-    // Allows access to control plane managment API and some storage controller endpoints.
+    /// Allows access to control plane managment API and some storage controller endpoints.
     Admin,
 
     /// Allows access to storage controller APIs used by the scrubber, to interrogate the state

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, 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 +123,10 @@ 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.virtual_file_direct_io, "starting with virtual_file Direct IO settings");
     info!(?conf.get_impl, "starting with get page implementation");
     info!(?conf.get_vectored_impl, "starting with vectored get page implementation");
+    info!(?conf.compact_level0_phase1_value_access, "starting with setting for compact_level0_phase1_value_access");
 
     let tenants_path = conf.tenants_path();
     if !tenants_path.exists() {
@@ -593,30 +591,13 @@ 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());
 
@@ -644,7 +625,7 @@ 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

@@ -29,6 +29,7 @@ use utils::{
     logging::LogFormat,
 };
 
+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};
@@ -52,7 +53,7 @@ pub mod defaults {
     use pageserver_api::models::ImageCompressionAlgorithm;
     pub use storage_broker::DEFAULT_ENDPOINT as BROKER_DEFAULT_ENDPOINT;
 
-    pub const DEFAULT_WAIT_LSN_TIMEOUT: &str = "60 s";
+    pub const DEFAULT_WAIT_LSN_TIMEOUT: &str = "300 s";
     pub const DEFAULT_WAL_REDO_TIMEOUT: &str = "60 s";
 
     pub const DEFAULT_SUPERUSER: &str = "cloud_admin";
@@ -83,16 +84,16 @@ pub mod defaults {
     #[cfg(not(target_os = "linux"))]
     pub const DEFAULT_VIRTUAL_FILE_IO_ENGINE: &str = "std-fs";
 
-    pub const DEFAULT_GET_VECTORED_IMPL: &str = "sequential";
+    pub const DEFAULT_GET_VECTORED_IMPL: &str = "vectored";
 
-    pub const DEFAULT_GET_IMPL: &str = "legacy";
+    pub const DEFAULT_GET_IMPL: &str = "vectored";
 
     pub const DEFAULT_MAX_VECTORED_READ_BYTES: usize = 128 * 1024; // 128 KiB
 
     pub const DEFAULT_IMAGE_COMPRESSION: ImageCompressionAlgorithm =
         ImageCompressionAlgorithm::Disabled;
 
-    pub const DEFAULT_VALIDATE_VECTORED_GET: bool = true;
+    pub const DEFAULT_VALIDATE_VECTORED_GET: bool = false;
 
     pub const DEFAULT_EPHEMERAL_BYTES_PER_MEMORY_KB: usize = 0;
 
@@ -295,6 +296,13 @@ pub struct PageServerConf {
     pub ephemeral_bytes_per_memory_kb: usize,
 
     pub l0_flush: L0FlushConfig,
+
+    /// 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
@@ -356,8 +364,6 @@ struct PageServerConfigBuilder {
     auth_validation_public_key_path: BuilderValue<Option<Utf8PathBuf>>,
     remote_storage_config: BuilderValue<Option<RemoteStorageConfig>>,
 
-    id: BuilderValue<NodeId>,
-
     broker_endpoint: BuilderValue<Uri>,
     broker_keepalive_interval: BuilderValue<Duration>,
 
@@ -403,14 +409,15 @@ struct PageServerConfigBuilder {
     ephemeral_bytes_per_memory_kb: BuilderValue<usize>,
 
     l0_flush: BuilderValue<L0FlushConfig>,
+
+    compact_level0_phase1_value_access: BuilderValue<CompactL0Phase1ValueAccess>,
+
+    virtual_file_direct_io: BuilderValue<virtual_file::DirectIoMode>,
 }
 
 impl PageServerConfigBuilder {
-    fn new(node_id: NodeId) -> Self {
-        let mut this = Self::default();
-        this.id(node_id);
-
-        this
+    fn new() -> Self {
+        Self::default()
     }
 
     #[inline(always)]
@@ -438,7 +445,6 @@ impl PageServerConfigBuilder {
             pg_auth_type: Set(AuthType::Trust),
             auth_validation_public_key_path: Set(None),
             remote_storage_config: Set(None),
-            id: NotSet,
             broker_endpoint: Set(storage_broker::DEFAULT_ENDPOINT
                 .parse()
                 .expect("failed to parse default broker endpoint")),
@@ -496,6 +502,8 @@ impl PageServerConfigBuilder {
             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()),
         }
     }
 }
@@ -568,10 +576,6 @@ impl PageServerConfigBuilder {
         self.broker_keepalive_interval = BuilderValue::Set(broker_keepalive_interval)
     }
 
-    pub fn id(&mut self, node_id: NodeId) {
-        self.id = BuilderValue::Set(node_id)
-    }
-
     pub fn log_format(&mut self, log_format: LogFormat) {
         self.log_format = BuilderValue::Set(log_format)
     }
@@ -683,7 +687,15 @@ impl PageServerConfigBuilder {
         self.l0_flush = BuilderValue::Set(value);
     }
 
-    pub fn build(self) -> anyhow::Result<PageServerConf> {
+    pub fn compact_level0_phase1_value_access(&mut self, value: CompactL0Phase1ValueAccess) {
+        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();
 
         macro_rules! conf {
@@ -716,7 +728,6 @@ impl PageServerConfigBuilder {
                 pg_auth_type,
                 auth_validation_public_key_path,
                 remote_storage_config,
-                id,
                 broker_endpoint,
                 broker_keepalive_interval,
                 log_format,
@@ -741,9 +752,12 @@ impl PageServerConfigBuilder {
                 image_compression,
                 ephemeral_bytes_per_memory_kb,
                 l0_flush,
+                compact_level0_phase1_value_access,
+                virtual_file_direct_io,
             }
             CUSTOM LOGIC
             {
+                id: id,
                 // TenantConf is handled separately
                 default_tenant_conf: TenantConf::default(),
                 concurrent_tenant_warmup: ConfigurableSemaphore::new({
@@ -893,7 +907,7 @@ impl PageServerConf {
         toml: &Document,
         workdir: &Utf8Path,
     ) -> anyhow::Result<Self> {
-        let mut builder = PageServerConfigBuilder::new(node_id);
+        let mut builder = PageServerConfigBuilder::new();
         builder.workdir(workdir.to_owned());
 
         let mut t_conf = TenantConfOpt::default();
@@ -924,8 +938,6 @@ impl PageServerConf {
                 "tenant_config" => {
                     t_conf = TenantConfOpt::try_from(item.to_owned()).context(format!("failed to parse: '{key}'"))?;
                 }
-                "id" => {}, // Ignoring `id` field in pageserver.toml - using identity.toml as the source of truth
-                            // Logging is not set up yet, so we can't do it.
                 "broker_endpoint" => builder.broker_endpoint(parse_toml_string(key, item)?.parse().context("failed to parse broker endpoint")?),
                 "broker_keepalive_interval" => builder.broker_keepalive_interval(parse_toml_duration(key, item)?),
                 "log_format" => builder.log_format(
@@ -1014,11 +1026,17 @@ impl PageServerConf {
                 "l0_flush" => {
                     builder.l0_flush(utils::toml_edit_ext::deserialize_item(item).context("l0_flush")?)
                 }
+                "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}'"),
             }
         }
 
-        let mut conf = builder.build().context("invalid config")?;
+        let mut conf = builder.build(node_id).context("invalid config")?;
 
         if conf.http_auth_type == AuthType::NeonJWT || conf.pg_auth_type == AuthType::NeonJWT {
             let auth_validation_public_key_path = conf
@@ -1098,6 +1116,8 @@ impl PageServerConf {
             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(),
         }
     }
 }
@@ -1255,7 +1275,6 @@ max_file_descriptors = 333
 
 # initial superuser role name to use when creating a new tenant
 initial_superuser_name = 'zzzz'
-id = 10
 
 metric_collection_interval = '222 s'
 metric_collection_endpoint = 'http://localhost:80/metrics'
@@ -1272,9 +1291,8 @@ background_task_maximum_delay = '334 s'
         let (workdir, pg_distrib_dir) = prepare_fs(&tempdir)?;
         let broker_endpoint = storage_broker::DEFAULT_ENDPOINT;
         // we have to create dummy values to overcome the validation errors
-        let config_string = format!(
-            "pg_distrib_dir='{pg_distrib_dir}'\nid=10\nbroker_endpoint = '{broker_endpoint}'",
-        );
+        let config_string =
+            format!("pg_distrib_dir='{pg_distrib_dir}'\nbroker_endpoint = '{broker_endpoint}'",);
         let toml = config_string.parse()?;
 
         let parsed_config = PageServerConf::parse_and_validate(NodeId(10), &toml, &workdir)
@@ -1341,6 +1359,8 @@ background_task_maximum_delay = '334 s'
                 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"
         );
@@ -1415,6 +1435,8 @@ background_task_maximum_delay = '334 s'
                 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"
         );
@@ -1579,7 +1601,6 @@ broker_endpoint = '{broker_endpoint}'
             r#"pg_distrib_dir = "{pg_distrib_dir}"
 metric_collection_endpoint = "http://sample.url"
 metric_collection_interval = "10min"
-id = 222
 
 [disk_usage_based_eviction]
 max_usage_pct = 80
@@ -1649,7 +1670,6 @@ threshold = "20m"
             r#"pg_distrib_dir = "{pg_distrib_dir}"
 metric_collection_endpoint = "http://sample.url"
 metric_collection_interval = "10min"
-id = 222
 
 [tenant_config]
 evictions_low_residence_duration_metric_threshold = "20m"

pageserver/src/control_plane_client.rs

@@ -171,14 +171,14 @@ impl ControlPlaneGenerationsApi for ControlPlaneClient {
             register,
         };
 
-        fail::fail_point!("control-plane-client-re-attach");
-
         let response: ReAttachResponse = self.retry_http_forever(&re_attach_path, request).await?;
         tracing::info!(
             "Received re-attach response with {} tenants",
             response.tenants.len()
         );
 
+        failpoint_support::sleep_millis_async!("control-plane-client-re-attach");
+
         Ok(response
             .tenants
             .into_iter()

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,
@@ -1650,7 +1726,9 @@ async fn timeline_compact_handler(
             .await
             .map_err(|e| ApiError::InternalServerError(e.into()))?;
         if wait_until_uploaded {
-            timeline.remote_client.wait_completion().await.map_err(ApiError::InternalServerError)?;
+            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)?;
         }
         json_response(StatusCode::OK, ())
     }
@@ -1709,7 +1787,9 @@ async fn timeline_checkpoint_handler(
         }
 
         if wait_until_uploaded {
-            timeline.remote_client.wait_completion().await.map_err(ApiError::InternalServerError)?;
+            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)?;
         }
 
         json_response(StatusCode::OK, ())
@@ -2125,14 +2205,24 @@ async fn secondary_download_handler(
 
     let timeout = wait.unwrap_or(Duration::MAX);
 
-    let status = match tokio::time::timeout(
+    let result = tokio::time::timeout(
         timeout,
         state.secondary_controller.download_tenant(tenant_shard_id),
     )
-    .await
-    {
-        // Download job ran to completion.
-        Ok(Ok(())) => StatusCode::OK,
+    .await;
+
+    let progress = secondary_tenant.progress.lock().unwrap().clone();
+
+    let status = match result {
+        Ok(Ok(())) => {
+            if progress.layers_downloaded >= progress.layers_total {
+                // Download job ran to completion
+                StatusCode::OK
+            } else {
+                // Download dropped out without errors because it ran out of time budget
+                StatusCode::ACCEPTED
+            }
+        }
         // Edge case: downloads aren't usually fallible: things like a missing heatmap are considered
         // okay.  We could get an error here in the unlikely edge case that the tenant
         // was detached between our check above and executing the download job.
@@ -2142,8 +2232,6 @@ async fn secondary_download_handler(
         Err(_) => StatusCode::ACCEPTED,
     };
 
-    let progress = secondary_tenant.progress.lock().unwrap().clone();
-
     json_response(status, progress)
 }
 
@@ -2887,6 +2975,14 @@ 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

@@ -2,19 +2,29 @@ use std::{num::NonZeroUsize, sync::Arc};
 
 use crate::tenant::ephemeral_file;
 
-#[derive(Default, Debug, PartialEq, Eq, Clone, serde::Deserialize)]
+#[derive(Debug, PartialEq, Eq, Clone, serde::Deserialize)]
 #[serde(tag = "mode", rename_all = "kebab-case", deny_unknown_fields)]
 pub enum L0FlushConfig {
-    #[default]
     PageCached,
     #[serde(rename_all = "snake_case")]
-    Direct { max_concurrency: NonZeroUsize },
+    Direct {
+        max_concurrency: NonZeroUsize,
+    },
+}
+
+impl Default for L0FlushConfig {
+    fn default() -> Self {
+        Self::Direct {
+            // TODO: using num_cpus results in different peak memory usage on different instance types.
+            max_concurrency: NonZeroUsize::new(usize::max(1, num_cpus::get())).unwrap(),
+        }
+    }
 }
 
 #[derive(Clone)]
 pub struct L0FlushGlobalState(Arc<Inner>);
 
-pub(crate) enum Inner {
+pub enum Inner {
     PageCached,
     Direct { semaphore: tokio::sync::Semaphore },
 }
@@ -30,7 +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",
@@ -613,7 +622,23 @@ pub(crate) static CIRCUIT_BREAKERS_UNBROKEN: Lazy<IntCounter> = Lazy::new(|| {
 pub(crate) static COMPRESSION_IMAGE_INPUT_BYTES: Lazy<IntCounter> = Lazy::new(|| {
     register_int_counter!(
         "pageserver_compression_image_in_bytes_total",
-        "Size of uncompressed data written into image layers"
+        "Size of data written into image layers before compression"
+    )
+    .expect("failed to define a metric")
+});
+
+pub(crate) static COMPRESSION_IMAGE_INPUT_BYTES_CONSIDERED: Lazy<IntCounter> = Lazy::new(|| {
+    register_int_counter!(
+        "pageserver_compression_image_in_bytes_considered",
+        "Size of potentially compressible data written into image layers before compression"
+    )
+    .expect("failed to define a metric")
+});
+
+pub(crate) static COMPRESSION_IMAGE_INPUT_BYTES_CHOSEN: Lazy<IntCounter> = Lazy::new(|| {
+    register_int_counter!(
+        "pageserver_compression_image_in_bytes_chosen",
+        "Size of data whose compressed form was written into image layers"
     )
     .expect("failed to define a metric")
 });
@@ -2188,6 +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,
@@ -2310,6 +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])
@@ -2364,6 +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,
@@ -2415,6 +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]);
@@ -3104,6 +3135,8 @@ pub fn preinitialize_metrics() {
         &tokio_epoll_uring::THREAD_LOCAL_LAUNCH_SUCCESSES,
         &REMOTE_ONDEMAND_DOWNLOADED_LAYERS,
         &REMOTE_ONDEMAND_DOWNLOADED_BYTES,
+        &CIRCUIT_BREAKERS_BROKEN,
+        &CIRCUIT_BREAKERS_UNBROKEN,
     ]
     .into_iter()
     .for_each(|c| {

pageserver/src/repository.rs

@@ -8,8 +8,7 @@ use std::time::Duration;
 pub use pageserver_api::key::{Key, KEY_SIZE};
 
 /// A 'value' stored for a one Key.
-#[derive(Debug, Clone, Serialize, Deserialize)]
-#[cfg_attr(test, derive(PartialEq))]
+#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
 pub enum Value {
     /// An Image value contains a full copy of the value
     Image(Bytes),

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/blob_io.rs

@@ -28,6 +28,12 @@ use crate::virtual_file::VirtualFile;
 use std::cmp::min;
 use std::io::{Error, ErrorKind};
 
+#[derive(Copy, Clone, Debug)]
+pub struct CompressionInfo {
+    pub written_compressed: bool,
+    pub compressed_size: Option<usize>,
+}
+
 impl<'a> BlockCursor<'a> {
     /// Read a blob into a new buffer.
     pub async fn read_blob(
@@ -273,8 +279,10 @@ impl<const BUFFERED: bool> BlobWriter<BUFFERED> {
         srcbuf: B,
         ctx: &RequestContext,
     ) -> (B::Buf, Result<u64, Error>) {
-        self.write_blob_maybe_compressed(srcbuf, ctx, ImageCompressionAlgorithm::Disabled)
-            .await
+        let (buf, res) = self
+            .write_blob_maybe_compressed(srcbuf, ctx, ImageCompressionAlgorithm::Disabled)
+            .await;
+        (buf, res.map(|(off, _compression_info)| off))
     }
 
     /// Write a blob of data. Returns the offset that it was written to,
@@ -284,8 +292,12 @@ impl<const BUFFERED: bool> BlobWriter<BUFFERED> {
         srcbuf: B,
         ctx: &RequestContext,
         algorithm: ImageCompressionAlgorithm,
-    ) -> (B::Buf, Result<u64, Error>) {
+    ) -> (B::Buf, Result<(u64, CompressionInfo), Error>) {
         let offset = self.offset;
+        let mut compression_info = CompressionInfo {
+            written_compressed: false,
+            compressed_size: None,
+        };
 
         let len = srcbuf.bytes_init();
 
@@ -328,7 +340,9 @@ impl<const BUFFERED: bool> BlobWriter<BUFFERED> {
                         encoder.write_all(&slice[..]).await.unwrap();
                         encoder.shutdown().await.unwrap();
                         let compressed = encoder.into_inner();
+                        compression_info.compressed_size = Some(compressed.len());
                         if compressed.len() < len {
+                            compression_info.written_compressed = true;
                             let compressed_len = compressed.len();
                             compressed_buf = Some(compressed);
                             (BYTE_ZSTD, compressed_len, slice.into_inner())
@@ -359,7 +373,7 @@ impl<const BUFFERED: bool> BlobWriter<BUFFERED> {
         } else {
             self.write_all(srcbuf, ctx).await
         };
-        (srcbuf, res.map(|_| offset))
+        (srcbuf, res.map(|_| (offset, compression_info)))
     }
 }
 
@@ -416,12 +430,14 @@ pub(crate) mod tests {
             let mut wtr = BlobWriter::<BUFFERED>::new(file, 0);
             for blob in blobs.iter() {
                 let (_, res) = if compression {
-                    wtr.write_blob_maybe_compressed(
-                        blob.clone(),
-                        ctx,
-                        ImageCompressionAlgorithm::Zstd { level: Some(1) },
-                    )
-                    .await
+                    let res = wtr
+                        .write_blob_maybe_compressed(
+                            blob.clone(),
+                            ctx,
+                            ImageCompressionAlgorithm::Zstd { level: Some(1) },
+                        )
+                        .await;
+                    (res.0, res.1.map(|(off, _)| off))
                 } else {
                     wtr.write_blob(blob.clone(), ctx).await
                 };

pageserver/src/tenant/disk_btree.rs

@@ -296,13 +296,19 @@ where
             let mut stack = Vec::new();
             stack.push((self.root_blk, None));
             let block_cursor = self.reader.block_cursor();
+            let mut node_buf = [0_u8; PAGE_SZ];
             while let Some((node_blknum, opt_iter)) = stack.pop() {
-                // Locate the node.
-                let node_buf = block_cursor
+                // Read the node, through the PS PageCache, into local variable `node_buf`.
+                // We could keep the page cache read guard alive, but, at the time of writing,
+                // we run quite small PS PageCache s => can't risk running out of
+                // PageCache space because this stream isn't consumed fast enough.
+                let page_read_guard = block_cursor
                     .read_blk(self.start_blk + node_blknum, ctx)
                     .await?;
+                node_buf.copy_from_slice(page_read_guard.as_ref());
+                drop(page_read_guard); // drop page cache read guard early
 
-                let node = OnDiskNode::deparse(node_buf.as_ref())?;
+                let node = OnDiskNode::deparse(&node_buf)?;
                 let prefix_len = node.prefix_len as usize;
                 let suffix_len = node.suffix_len as usize;
 
@@ -345,6 +351,7 @@ where
                     Either::Left(idx..node.num_children.into())
                 };
 
+
                 // idx points to the first match now. Keep going from there
                 while let Some(idx) = iter.next() {
                     let key_off = idx * suffix_len;

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) -> Result<Vec<Arc<PersistentLayerDesc>>> {
-        Ok(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

@@ -111,7 +111,7 @@ impl TryFrom<&TimelineMetadataBodyV2> for TimelineMetadataHeader {
 #[error("re-serializing for crc32 failed")]
 struct Crc32CalculationFailed(#[source] utils::bin_ser::SerializeError);
 
-const METADATA_HDR_SIZE: usize = std::mem::size_of::<TimelineMetadataHeader>();
+const METADATA_HDR_SIZE: usize = size_of::<TimelineMetadataHeader>();
 
 #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
 struct TimelineMetadataBodyV2 {

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;
@@ -55,7 +55,7 @@ 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::{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))?;
 
@@ -1384,33 +1350,32 @@ impl TenantManager {
         tenant_shard_id: TenantShardId,
     ) -> Result<(), DeleteTenantError> {
         let remote_path = remote_tenant_path(&tenant_shard_id);
-        let keys = match self
-            .resources
-            .remote_storage
-            .list(
-                Some(&remote_path),
-                remote_storage::ListingMode::NoDelimiter,
-                None,
-                &self.cancel,
-            )
-            .await
-        {
-            Ok(listing) => listing.keys,
-            Err(remote_storage::DownloadError::Cancelled) => {
-                return Err(DeleteTenantError::Cancelled)
-            }
-            Err(remote_storage::DownloadError::NotFound) => return Ok(()),
-            Err(other) => return Err(DeleteTenantError::Other(anyhow::anyhow!(other))),
-        };
+        let mut keys_stream = self.resources.remote_storage.list_streaming(
+            Some(&remote_path),
+            remote_storage::ListingMode::NoDelimiter,
+            None,
+            &self.cancel,
+        );
+        while let Some(chunk) = keys_stream.next().await {
+            let keys = match chunk {
+                Ok(listing) => listing.keys,
+                Err(remote_storage::DownloadError::Cancelled) => {
+                    return Err(DeleteTenantError::Cancelled)
+                }
+                Err(remote_storage::DownloadError::NotFound) => return Ok(()),
+                Err(other) => return Err(DeleteTenantError::Other(anyhow::anyhow!(other))),
+            };
 
-        if keys.is_empty() {
-            tracing::info!("Remote storage already deleted");
-        } else {
-            tracing::info!("Deleting {} keys from remote storage", keys.len());
-            self.resources
-                .remote_storage
-                .delete_objects(&keys, &self.cancel)
-                .await?;
+            if keys.is_empty() {
+                tracing::info!("Remote storage already deleted");
+            } else {
+                tracing::info!("Deleting {} keys from remote storage", keys.len());
+                let keys = keys.into_iter().map(|o| o.key).collect::<Vec<_>>();
+                self.resources
+                    .remote_storage
+                    .delete_objects(&keys, &self.cancel)
+                    .await?;
+            }
         }
 
         Ok(())
@@ -1764,14 +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)
@@ -1968,7 +1928,8 @@ impl TenantManager {
         timeline_id: TimelineId,
         prepared: PreparedTimelineDetach,
         ctx: &RequestContext,
-    ) -> Result<Vec<TimelineId>, anyhow::Error> {
+    ) -> Result<HashSet<TimelineId>, anyhow::Error> {
+        // FIXME: this is unnecessary, slotguard already has these semantics
         struct RevertOnDropSlot(Option<SlotGuard>);
 
         impl Drop for RevertOnDropSlot {
@@ -2048,7 +2009,7 @@ impl TenantManager {
             None,
             SpawnMode::Eager,
             ctx,
-        );
+        )?;
 
         slot_guard.upsert(TenantSlot::Attached(tenant))?;
 
@@ -2089,7 +2050,6 @@ 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())
                         }
@@ -2171,6 +2131,9 @@ 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

@@ -187,7 +187,7 @@ use camino::Utf8Path;
 use chrono::{NaiveDateTime, Utc};
 
 pub(crate) use download::download_initdb_tar_zst;
-use pageserver_api::models::AuxFilePolicy;
+use pageserver_api::models::{AuxFilePolicy, TimelineArchivalState};
 use pageserver_api::shard::{ShardIndex, TenantShardId};
 use scopeguard::ScopeGuard;
 use tokio_util::sync::CancellationToken;
@@ -287,6 +287,14 @@ pub enum PersistIndexPartWithDeletedFlagError {
     Other(#[from] anyhow::Error),
 }
 
+#[derive(Debug, thiserror::Error)]
+pub enum WaitCompletionError {
+    #[error(transparent)]
+    NotInitialized(NotInitialized),
+    #[error("wait_completion aborted because upload queue was stopped")]
+    UploadQueueShutDownOrStopped,
+}
+
 /// A client for accessing a timeline's data in remote storage.
 ///
 /// This takes care of managing the number of connections, and balancing them
@@ -449,6 +457,17 @@ impl RemoteTimelineClient {
             .unwrap_or(false)
     }
 
+    /// Returns whether the timeline is archived.
+    /// Return None if the remote index_part hasn't been downloaded yet.
+    pub(crate) fn is_archived(&self) -> Option<bool> {
+        self.upload_queue
+            .lock()
+            .unwrap()
+            .initialized_mut()
+            .map(|q| q.clean.0.archived_at.is_some())
+            .ok()
+    }
+
     fn update_remote_physical_size_gauge(&self, current_remote_index_part: Option<&IndexPart>) {
         let size: u64 = if let Some(current_remote_index_part) = current_remote_index_part {
             current_remote_index_part
@@ -609,7 +628,7 @@ impl RemoteTimelineClient {
         Ok(())
     }
 
-    /// Launch an index-file upload operation in the background, with only aux_file_policy flag updated.
+    /// Launch an index-file upload operation in the background, with only the `aux_file_policy` flag updated.
     pub(crate) fn schedule_index_upload_for_aux_file_policy_update(
         self: &Arc<Self>,
         last_aux_file_policy: Option<AuxFilePolicy>,
@@ -620,6 +639,48 @@ impl RemoteTimelineClient {
         self.schedule_index_upload(upload_queue)?;
         Ok(())
     }
+
+    /// Launch an index-file upload operation in the background, with only the `archived_at` field updated.
+    ///
+    /// Returns whether it is required to wait for the queue to be empty to ensure that the change is uploaded,
+    /// so either if the change is already sitting in the queue, but not commited yet, or the change has not
+    /// been in the queue yet.
+    pub(crate) fn schedule_index_upload_for_timeline_archival_state(
+        self: &Arc<Self>,
+        state: TimelineArchivalState,
+    ) -> anyhow::Result<bool> {
+        let mut guard = self.upload_queue.lock().unwrap();
+        let upload_queue = guard.initialized_mut()?;
+
+        /// Returns Some(_) if a change is needed, and Some(true) if it's a
+        /// change needed to set archived_at.
+        fn need_change(
+            archived_at: &Option<NaiveDateTime>,
+            state: TimelineArchivalState,
+        ) -> Option<bool> {
+            match (archived_at, state) {
+                (Some(_), TimelineArchivalState::Archived)
+                | (None, TimelineArchivalState::Unarchived) => {
+                    // Nothing to do
+                    tracing::info!("intended state matches present state");
+                    None
+                }
+                (None, TimelineArchivalState::Archived) => Some(true),
+                (Some(_), TimelineArchivalState::Unarchived) => Some(false),
+            }
+        }
+        let need_upload_scheduled = need_change(&upload_queue.dirty.archived_at, state);
+
+        if let Some(archived_at_set) = need_upload_scheduled {
+            let intended_archived_at = archived_at_set.then(|| Utc::now().naive_utc());
+            upload_queue.dirty.archived_at = intended_archived_at;
+            self.schedule_index_upload(upload_queue)?;
+        }
+
+        let need_wait = need_change(&upload_queue.clean.0.archived_at, state).is_some();
+        Ok(need_wait)
+    }
+
     ///
     /// Launch an index-file upload operation in the background, if necessary.
     ///
@@ -630,7 +691,7 @@ impl RemoteTimelineClient {
     ///
     /// Like schedule_index_upload_for_metadata_update(), this merely adds
     /// the upload to the upload queue and returns quickly.
-    pub fn schedule_index_upload_for_file_changes(self: &Arc<Self>) -> anyhow::Result<()> {
+    pub fn schedule_index_upload_for_file_changes(self: &Arc<Self>) -> Result<(), NotInitialized> {
         let mut guard = self.upload_queue.lock().unwrap();
         let upload_queue = guard.initialized_mut()?;
 
@@ -645,7 +706,7 @@ impl RemoteTimelineClient {
     fn schedule_index_upload(
         self: &Arc<Self>,
         upload_queue: &mut UploadQueueInitialized,
-    ) -> anyhow::Result<()> {
+    ) -> Result<(), NotInitialized> {
         let disk_consistent_lsn = upload_queue.dirty.metadata.disk_consistent_lsn();
         // fix up the duplicated field
         upload_queue.dirty.disk_consistent_lsn = disk_consistent_lsn;
@@ -653,7 +714,7 @@ impl RemoteTimelineClient {
         // make sure it serializes before doing it in perform_upload_task so that it doesn't
         // look like a retryable error
         let void = std::io::sink();
-        serde_json::to_writer(void, &upload_queue.dirty).context("serialize index_part.json")?;
+        serde_json::to_writer(void, &upload_queue.dirty).expect("serialize index_part.json");
 
         let index_part = &upload_queue.dirty;
 
@@ -699,7 +760,9 @@ impl RemoteTimelineClient {
             self.schedule_barrier0(upload_queue)
         };
 
-        Self::wait_completion0(receiver).await
+        Self::wait_completion0(receiver)
+            .await
+            .context("wait completion")
     }
 
     /// Schedules uploading a new version of `index_part.json` with the given layers added,
@@ -732,7 +795,126 @@ impl RemoteTimelineClient {
             barrier
         };
 
-        Self::wait_completion0(barrier).await
+        Self::wait_completion0(barrier)
+            .await
+            .context("wait completion")
+    }
+
+    /// 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_original_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
@@ -740,7 +922,7 @@ impl RemoteTimelineClient {
     pub(crate) fn schedule_layer_file_upload(
         self: &Arc<Self>,
         layer: ResidentLayer,
-    ) -> anyhow::Result<()> {
+    ) -> Result<(), NotInitialized> {
         let mut guard = self.upload_queue.lock().unwrap();
         let upload_queue = guard.initialized_mut()?;
 
@@ -826,7 +1008,7 @@ impl RemoteTimelineClient {
         self: &Arc<Self>,
         upload_queue: &mut UploadQueueInitialized,
         names: I,
-    ) -> anyhow::Result<Vec<(LayerName, LayerFileMetadata)>>
+    ) -> Result<Vec<(LayerName, LayerFileMetadata)>, NotInitialized>
     where
         I: IntoIterator<Item = LayerName>,
     {
@@ -952,7 +1134,7 @@ impl RemoteTimelineClient {
         self: &Arc<Self>,
         compacted_from: &[Layer],
         compacted_to: &[ResidentLayer],
-    ) -> anyhow::Result<()> {
+    ) -> Result<(), NotInitialized> {
         let mut guard = self.upload_queue.lock().unwrap();
         let upload_queue = guard.initialized_mut()?;
 
@@ -969,10 +1151,12 @@ impl RemoteTimelineClient {
     }
 
     /// Wait for all previously scheduled uploads/deletions to complete
-    pub(crate) async fn wait_completion(self: &Arc<Self>) -> anyhow::Result<()> {
+    pub(crate) async fn wait_completion(self: &Arc<Self>) -> Result<(), WaitCompletionError> {
         let receiver = {
             let mut guard = self.upload_queue.lock().unwrap();
-            let upload_queue = guard.initialized_mut()?;
+            let upload_queue = guard
+                .initialized_mut()
+                .map_err(WaitCompletionError::NotInitialized)?;
             self.schedule_barrier0(upload_queue)
         };
 
@@ -981,9 +1165,9 @@ impl RemoteTimelineClient {
 
     async fn wait_completion0(
         mut receiver: tokio::sync::watch::Receiver<()>,
-    ) -> anyhow::Result<()> {
+    ) -> Result<(), WaitCompletionError> {
         if receiver.changed().await.is_err() {
-            anyhow::bail!("wait_completion aborted because upload queue was stopped");
+            return Err(WaitCompletionError::UploadQueueShutDownOrStopped);
         }
 
         Ok(())
@@ -1311,6 +1495,18 @@ 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,
@@ -1366,12 +1562,13 @@ impl RemoteTimelineClient {
         // marker via its deleted_at attribute
         let latest_index = remaining
             .iter()
-            .filter(|p| {
-                p.object_name()
+            .filter(|o| {
+                o.key
+                    .object_name()
                     .map(|n| n.starts_with(IndexPart::FILE_NAME))
                     .unwrap_or(false)
             })
-            .filter_map(|path| parse_remote_index_path(path.clone()).map(|gen| (path, gen)))
+            .filter_map(|o| parse_remote_index_path(o.key.clone()).map(|gen| (o.key.clone(), gen)))
             .max_by_key(|i| i.1)
             .map(|i| i.0.clone())
             .unwrap_or(
@@ -1382,14 +1579,12 @@ impl RemoteTimelineClient {
 
         let remaining_layers: Vec<RemotePath> = remaining
             .into_iter()
-            .filter(|p| {
-                if p == &latest_index {
-                    return false;
+            .filter_map(|o| {
+                if o.key == latest_index || o.key.object_name() == Some(INITDB_PRESERVED_PATH) {
+                    None
+                } else {
+                    Some(o.key)
                 }
-                if p.object_name() == Some(INITDB_PRESERVED_PATH) {
-                    return false;
-                }
-                true
             })
             .inspect(|path| {
                 if let Some(name) = path.object_name() {

pageserver/src/tenant/remote_timeline_client/download.rs

@@ -295,10 +295,11 @@ where
         };
     }
 
-    for key in listing.keys {
-        let object_name = key
+    for object in listing.keys {
+        let object_name = object
+            .key
             .object_name()
-            .ok_or_else(|| anyhow::anyhow!("object name for key {key}"))?;
+            .ok_or_else(|| anyhow::anyhow!("object name for key {}", object.key))?;
         other_prefixes.insert(object_name.to_string());
     }
 
@@ -459,7 +460,7 @@ pub(crate) async fn download_index_part(
     // is <= our own.  See "Finding the remote indices for timelines" in docs/rfcs/025-generation-numbers.md
     let max_previous_generation = indices
         .into_iter()
-        .filter_map(parse_remote_index_path)
+        .filter_map(|o| parse_remote_index_path(o.key))
         .filter(|g| g <= &my_generation)
         .max();
 

pageserver/src/tenant/remote_timeline_client/index.rs

@@ -32,6 +32,10 @@ pub struct IndexPart {
     #[serde(skip_serializing_if = "Option::is_none")]
     pub deleted_at: Option<NaiveDateTime>,
 
+    #[serde(default)]
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub archived_at: Option<NaiveDateTime>,
+
     /// Per layer file name metadata, which can be present for a present or missing layer file.
     ///
     /// Older versions of `IndexPart` will not have this property or have only a part of metadata
@@ -56,6 +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.
@@ -80,10 +87,12 @@ impl IndexPart {
     /// - 5: lineage was added
     /// - 6: last_aux_file_policy is added.
     /// - 7: metadata_bytes is no longer written, but still read
-    const LATEST_VERSION: usize = 7;
+    /// - 8: added `archived_at`
+    /// - 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];
+    pub const KNOWN_VERSIONS: &'static [usize] = &[1, 2, 3, 4, 5, 6, 7, 8, 9];
 
     pub const FILE_NAME: &'static str = "index_part.json";
 
@@ -94,7 +103,9 @@ impl IndexPart {
             disk_consistent_lsn: metadata.disk_consistent_lsn(),
             metadata,
             deleted_at: None,
+            archived_at: None,
             lineage: Default::default(),
+            gc_blocking: None,
             last_aux_file_policy: None,
         }
     }
@@ -245,6 +256,64 @@ impl Lineage {
     }
 }
 
+#[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::*;
@@ -284,7 +353,9 @@ mod tests {
             disk_consistent_lsn: "0/16960E8".parse::<Lsn>().unwrap(),
             metadata: TimelineMetadata::from_bytes(&[113,11,159,210,0,54,0,4,0,0,0,0,1,105,96,232,1,0,0,0,0,1,105,96,112,0,0,0,0,0,0,0,0,0,0,0,0,0,1,105,96,112,0,0,0,0,1,105,96,112,0,0,0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]).unwrap(),
             deleted_at: None,
+            archived_at: None,
             lineage: Lineage::default(),
+            gc_blocking: None,
             last_aux_file_policy: None,
         };
 
@@ -326,7 +397,9 @@ mod tests {
             disk_consistent_lsn: "0/16960E8".parse::<Lsn>().unwrap(),
             metadata: TimelineMetadata::from_bytes(&[113,11,159,210,0,54,0,4,0,0,0,0,1,105,96,232,1,0,0,0,0,1,105,96,112,0,0,0,0,0,0,0,0,0,0,0,0,0,1,105,96,112,0,0,0,0,1,105,96,112,0,0,0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]).unwrap(),
             deleted_at: None,
+            archived_at: None,
             lineage: Lineage::default(),
+            gc_blocking: None,
             last_aux_file_policy: None,
         };
 
@@ -369,7 +442,9 @@ mod tests {
             disk_consistent_lsn: "0/16960E8".parse::<Lsn>().unwrap(),
             metadata: TimelineMetadata::from_bytes(&[113,11,159,210,0,54,0,4,0,0,0,0,1,105,96,232,1,0,0,0,0,1,105,96,112,0,0,0,0,0,0,0,0,0,0,0,0,0,1,105,96,112,0,0,0,0,1,105,96,112,0,0,0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]).unwrap(),
             deleted_at: Some(parse_naive_datetime("2023-07-31T09:00:00.123000000")),
+            archived_at: None,
             lineage: Lineage::default(),
+            gc_blocking: None,
             last_aux_file_policy: None,
         };
 
@@ -415,7 +490,9 @@ mod tests {
             ])
             .unwrap(),
             deleted_at: None,
+            archived_at: None,
             lineage: Lineage::default(),
+            gc_blocking: None,
             last_aux_file_policy: None,
         };
 
@@ -456,7 +533,9 @@ mod tests {
             disk_consistent_lsn: "0/16960E8".parse::<Lsn>().unwrap(),
             metadata: TimelineMetadata::from_bytes(&[113,11,159,210,0,54,0,4,0,0,0,0,1,105,96,232,1,0,0,0,0,1,105,96,112,0,0,0,0,0,0,0,0,0,0,0,0,0,1,105,96,112,0,0,0,0,1,105,96,112,0,0,0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]).unwrap(),
             deleted_at: Some(parse_naive_datetime("2023-07-31T09:00:00.123000000")),
+            archived_at: None,
             lineage: Lineage::default(),
+            gc_blocking: None,
             last_aux_file_policy: None,
         };
 
@@ -496,11 +575,13 @@ mod tests {
             disk_consistent_lsn: Lsn::from_str("0/15A7618").unwrap(),
             metadata: TimelineMetadata::from_bytes(&[226,88,25,241,0,46,0,4,0,0,0,0,1,90,118,24,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,78,244,32,0,0,0,0,1,78,244,32,0,0,0,16,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]).unwrap(),
             deleted_at: None,
+            archived_at: None,
             lineage: Lineage {
                 reparenting_history_truncated: false,
                 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,
         };
 
@@ -545,11 +626,13 @@ mod tests {
             disk_consistent_lsn: "0/16960E8".parse::<Lsn>().unwrap(),
             metadata: TimelineMetadata::from_bytes(&[113,11,159,210,0,54,0,4,0,0,0,0,1,105,96,232,1,0,0,0,0,1,105,96,112,0,0,0,0,0,0,0,0,0,0,0,0,0,1,105,96,112,0,0,0,0,1,105,96,112,0,0,0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]).unwrap(),
             deleted_at: Some(parse_naive_datetime("2023-07-31T09:00:00.123000000")),
+            archived_at: None,
             lineage: Lineage {
                 reparenting_history_truncated: false,
                 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),
         };
 
@@ -603,7 +686,9 @@ mod tests {
                 14,
             ).with_recalculated_checksum().unwrap(),
             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(),
         };
 
@@ -611,6 +696,125 @@ mod tests {
         assert_eq!(part, expected);
     }
 
+    #[test]
+    fn v8_indexpart_is_parsed() {
+        let example = r#"{
+            "version": 8,
+            "layer_metadata":{
+                "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__0000000001696070-00000000016960E9": { "file_size": 25600000 },
+                "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51": { "file_size": 9007199254741001 }
+            },
+            "disk_consistent_lsn":"0/16960E8",
+            "metadata": {
+                "disk_consistent_lsn": "0/16960E8",
+                "prev_record_lsn": "0/1696070",
+                "ancestor_timeline": "e45a7f37d3ee2ff17dc14bf4f4e3f52e",
+                "ancestor_lsn": "0/0",
+                "latest_gc_cutoff_lsn": "0/1696070",
+                "initdb_lsn": "0/1696070",
+                "pg_version": 14
+            },
+            "deleted_at": "2023-07-31T09:00:00.123",
+            "archived_at": "2023-04-29T09:00:00.123"
+        }"#;
+
+        let expected = IndexPart {
+            version: 8,
+            layer_metadata: HashMap::from([
+                ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__0000000001696070-00000000016960E9".parse().unwrap(), LayerFileMetadata {
+                    file_size: 25600000,
+                    generation: Generation::none(),
+                    shard: ShardIndex::unsharded()
+                }),
+                ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap(), LayerFileMetadata {
+                    file_size: 9007199254741001,
+                    generation: Generation::none(),
+                    shard: ShardIndex::unsharded()
+                })
+            ]),
+            disk_consistent_lsn: "0/16960E8".parse::<Lsn>().unwrap(),
+            metadata: TimelineMetadata::new(
+                Lsn::from_str("0/16960E8").unwrap(),
+                Some(Lsn::from_str("0/1696070").unwrap()),
+                Some(TimelineId::from_str("e45a7f37d3ee2ff17dc14bf4f4e3f52e").unwrap()),
+                Lsn::INVALID,
+                Lsn::from_str("0/1696070").unwrap(),
+                Lsn::from_str("0/1696070").unwrap(),
+                14,
+            ).with_recalculated_checksum().unwrap(),
+            deleted_at: Some(parse_naive_datetime("2023-07-31T09:00:00.123000000")),
+            archived_at: Some(parse_naive_datetime("2023-04-29T09:00:00.123000000")),
+            lineage: Default::default(),
+            gc_blocking: None,
+            last_aux_file_policy: Default::default(),
+        };
+
+        let part = IndexPart::from_s3_bytes(example.as_bytes()).unwrap();
+        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))
@@ -307,12 +301,10 @@ impl DeltaLayer {
             .with_context(|| format!("Failed to load delta layer {}", self.path()))
     }
 
-    async fn load_inner(&self, ctx: &RequestContext) -> Result<Arc<DeltaLayerInner>> {
+    async fn load_inner(&self, ctx: &RequestContext) -> anyhow::Result<Arc<DeltaLayerInner>> {
         let path = self.path();
 
-        let loaded = DeltaLayerInner::load(&path, None, None, ctx)
-            .await
-            .and_then(|res| res)?;
+        let loaded = DeltaLayerInner::load(&path, None, None, ctx).await?;
 
         // not production code
         let actual_layer_name = LayerName::from_str(path.file_name().unwrap()).unwrap();
@@ -352,7 +344,6 @@ impl DeltaLayer {
                 summary.lsn_range,
                 metadata.len(),
             ),
-            access_stats: Default::default(),
             inner: OnceCell::new(),
         })
     }
@@ -375,7 +366,6 @@ impl DeltaLayer {
 /// 3. Call `finish`.
 ///
 struct DeltaLayerWriterInner {
-    conf: &'static PageServerConf,
     pub path: Utf8PathBuf,
     timeline_id: TimelineId,
     tenant_shard_id: TenantShardId,
@@ -386,6 +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 {
@@ -419,7 +412,6 @@ impl DeltaLayerWriterInner {
         let tree_builder = DiskBtreeBuilder::new(block_buf);
 
         Ok(Self {
-            conf,
             path,
             timeline_id,
             tenant_shard_id,
@@ -427,6 +419,7 @@ impl DeltaLayerWriterInner {
             lsn_range,
             tree: tree_builder,
             blob_writer,
+            num_keys: 0,
         })
     }
 
@@ -469,7 +462,7 @@ impl DeltaLayerWriterInner {
             .write_blob_maybe_compressed(val, ctx, compression)
             .await;
         let off = match res {
-            Ok(off) => off,
+            Ok((off, _)) => off,
             Err(e) => return (val, Err(anyhow::anyhow!(e))),
         };
 
@@ -477,6 +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)))
     }
 
@@ -490,11 +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) {
@@ -507,9 +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;
 
@@ -574,11 +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))
     }
 }
 
@@ -679,14 +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
     }
 }
 
@@ -760,27 +758,24 @@ impl DeltaLayerInner {
         &self.layer_lsn_range
     }
 
-    /// 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,
         summary: Option<Summary>,
         max_vectored_read_bytes: Option<MaxVectoredReadBytes>,
         ctx: &RequestContext,
-    ) -> Result<Result<Self, anyhow::Error>, anyhow::Error> {
-        let file = match VirtualFile::open(path, ctx).await {
-            Ok(file) => file,
-            Err(e) => return Ok(Err(anyhow::Error::new(e).context("open layer file"))),
-        };
+    ) -> anyhow::Result<Self> {
+        let file = VirtualFile::open(path, ctx)
+            .await
+            .context("open layer file")?;
+
         let file_id = page_cache::next_file_id();
 
         let block_reader = FileBlockReader::new(&file, file_id);
 
-        let summary_blk = match block_reader.read_blk(0, ctx).await {
-            Ok(blk) => blk,
-            Err(e) => return Ok(Err(anyhow::Error::new(e).context("read first block"))),
-        };
+        let summary_blk = block_reader
+            .read_blk(0, ctx)
+            .await
+            .context("read first block")?;
 
         // TODO: this should be an assertion instead; see ImageLayerInner::load
         let actual_summary =
@@ -802,7 +797,7 @@ impl DeltaLayerInner {
             }
         }
 
-        Ok(Ok(DeltaLayerInner {
+        Ok(DeltaLayerInner {
             file,
             file_id,
             index_start_blk: actual_summary.index_start_blk,
@@ -810,96 +805,7 @@ impl DeltaLayerInner {
             max_vectored_read_bytes,
             layer_key_range: actual_summary.key_range,
             layer_lsn_range: actual_summary.lsn_range,
-        }))
-    }
-
-    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
@@ -1674,8 +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,
@@ -1969,9 +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?;
 
@@ -2051,6 +1957,7 @@ pub(crate) mod test {
             .await
             .likely_resident_layers()
             .next()
+            .cloned()
             .unwrap();
 
         {
@@ -2125,7 +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
@@ -2160,7 +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();
 
@@ -2288,7 +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
@@ -265,9 +260,8 @@ impl ImageLayer {
     async fn load_inner(&self, ctx: &RequestContext) -> Result<ImageLayerInner> {
         let path = self.path();
 
-        let loaded = ImageLayerInner::load(&path, self.desc.image_layer_lsn(), None, None, ctx)
-            .await
-            .and_then(|res| res)?;
+        let loaded =
+            ImageLayerInner::load(&path, self.desc.image_layer_lsn(), None, None, ctx).await?;
 
         // not production code
         let actual_layer_name = LayerName::from_str(path.file_name().unwrap()).unwrap();
@@ -307,7 +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(),
         })
     }
@@ -385,17 +378,16 @@ impl ImageLayerInner {
         summary: Option<Summary>,
         max_vectored_read_bytes: Option<MaxVectoredReadBytes>,
         ctx: &RequestContext,
-    ) -> Result<Result<Self, anyhow::Error>, anyhow::Error> {
-        let file = match VirtualFile::open(path, ctx).await {
-            Ok(file) => file,
-            Err(e) => return Ok(Err(anyhow::Error::new(e).context("open layer file"))),
-        };
+    ) -> anyhow::Result<Self> {
+        let file = VirtualFile::open(path, ctx)
+            .await
+            .context("open layer file")?;
         let file_id = page_cache::next_file_id();
         let block_reader = FileBlockReader::new(&file, file_id);
-        let summary_blk = match block_reader.read_blk(0, ctx).await {
-            Ok(blk) => blk,
-            Err(e) => return Ok(Err(anyhow::Error::new(e).context("read first block"))),
-        };
+        let summary_blk = block_reader
+            .read_blk(0, ctx)
+            .await
+            .context("read first block")?;
 
         // length is the only way how this could fail, so it's not actually likely at all unless
         // read_blk returns wrong sized block.
@@ -420,7 +412,7 @@ impl ImageLayerInner {
             }
         }
 
-        Ok(Ok(ImageLayerInner {
+        Ok(ImageLayerInner {
             index_start_blk: actual_summary.index_start_blk,
             index_root_blk: actual_summary.index_root_blk,
             lsn,
@@ -428,47 +420,7 @@ impl ImageLayerInner {
             file_id,
             max_vectored_read_bytes,
             key_range: actual_summary.key_range,
-        }))
-    }
-
-    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
@@ -736,11 +688,29 @@ struct ImageLayerWriterInner {
     // Total uncompressed bytes passed into put_image
     uncompressed_bytes: u64,
 
+    // Like `uncompressed_bytes`,
+    // but only of images we might consider for compression
+    uncompressed_bytes_eligible: u64,
+
+    // Like `uncompressed_bytes`, but only of images
+    // 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.
     ///
@@ -792,6 +762,10 @@ impl ImageLayerWriterInner {
             tree: tree_builder,
             blob_writer,
             uncompressed_bytes: 0,
+            uncompressed_bytes_eligible: 0,
+            uncompressed_bytes_chosen: 0,
+            num_keys: 0,
+            last_written_key: Key::MIN,
         };
 
         Ok(writer)
@@ -810,18 +784,33 @@ impl ImageLayerWriterInner {
     ) -> anyhow::Result<()> {
         ensure!(self.key_range.contains(&key));
         let compression = self.conf.image_compression;
-        self.uncompressed_bytes += img.len() as u64;
+        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)
             .await;
         // TODO: re-use the buffer for `img` further upstack
-        let off = res?;
+        let (off, compression_info) = res?;
+        if compression_info.compressed_size.is_some() {
+            // The image has been considered for compression at least
+            self.uncompressed_bytes_eligible += uncompressed_len;
+        }
+        if compression_info.written_compressed {
+            // The image has been compressed
+            self.uncompressed_bytes_chosen += uncompressed_len;
+        }
 
         let mut keybuf: [u8; KEY_SIZE] = [0u8; KEY_SIZE];
         key.write_to_byte_slice(&mut keybuf);
         self.tree.append(&keybuf, off)?;
 
+        #[cfg(feature = "testing")]
+        {
+            self.last_written_key = key;
+        }
+
         Ok(())
     }
 
@@ -832,6 +821,7 @@ 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;
@@ -839,6 +829,9 @@ impl ImageLayerWriterInner {
         // Calculate compression ratio
         let compressed_size = self.blob_writer.size() - PAGE_SZ as u64; // Subtract PAGE_SZ for header
         crate::metrics::COMPRESSION_IMAGE_INPUT_BYTES.inc_by(self.uncompressed_bytes);
+        crate::metrics::COMPRESSION_IMAGE_INPUT_BYTES_CONSIDERED
+            .inc_by(self.uncompressed_bytes_eligible);
+        crate::metrics::COMPRESSION_IMAGE_INPUT_BYTES_CHOSEN.inc_by(self.uncompressed_bytes_chosen);
         crate::metrics::COMPRESSION_IMAGE_OUTPUT_BYTES.inc_by(compressed_size);
 
         let mut file = self.blob_writer.into_inner();
@@ -879,11 +872,23 @@ 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.
@@ -960,6 +965,18 @@ 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.
     ///
@@ -968,7 +985,26 @@ 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 +10,11 @@ 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::keyspace::KeySpace;
 use pageserver_api::models::InMemoryLayerInfo;
 use pageserver_api::shard::TenantShardId;
@@ -34,8 +34,7 @@ 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 +54,6 @@ 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>>,
 
@@ -248,12 +244,6 @@ 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 +293,6 @@ 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.
     //
@@ -449,20 +385,17 @@ 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,
@@ -482,8 +415,7 @@ 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,
         lsn: Lsn,
@@ -548,8 +480,6 @@ 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 +497,13 @@ 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 +513,12 @@ 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 +530,8 @@ 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),
         };
@@ -628,7 +561,7 @@ 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)
@@ -693,7 +626,7 @@ 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 +638,6 @@ 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

@@ -17,14 +17,14 @@ use crate::context::{DownloadBehavior, RequestContext};
 use crate::repository::Key;
 use crate::span::debug_assert_current_span_has_tenant_and_timeline_id;
 use crate::task_mgr::TaskKind;
-use crate::tenant::timeline::GetVectoredError;
+use crate::tenant::timeline::{CompactionError, GetVectoredError};
 use crate::tenant::{remote_timeline_client::LayerFileMetadata, Timeline};
 
 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)
@@ -426,7 +383,7 @@ impl Layer {
     }
 
     /// Downloads if necessary and creates a guard, which will keep this layer from being evicted.
-    pub(crate) async fn download_and_keep_resident(&self) -> anyhow::Result<ResidentLayer> {
+    pub(crate) async fn download_and_keep_resident(&self) -> Result<ResidentLayer, DownloadError> {
         let downloaded = self.0.get_or_maybe_download(true, None).await?;
 
         Ok(ResidentLayer {
@@ -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),
@@ -1651,8 +1662,9 @@ impl Drop for DownloadedLayer {
 }
 
 impl DownloadedLayer {
-    /// Initializes the `DeltaLayerInner` or `ImageLayerInner` within [`LayerKind`], or fails to
-    /// initialize it permanently.
+    /// Initializes the `DeltaLayerInner` or `ImageLayerInner` within [`LayerKind`].
+    /// Failure to load the layer is sticky, i.e., future `get()` calls will return
+    /// the initial load failure immediately.
     ///
     /// `owner` parameter is a strong reference at the same `LayerInner` as the
     /// `DownloadedLayer::owner` would be when upgraded. Given how this method ends up called,
@@ -1683,7 +1695,7 @@ impl DownloadedLayer {
                     ctx,
                 )
                 .await
-                .map(|res| res.map(LayerKind::Delta))
+                .map(LayerKind::Delta)
             } else {
                 let lsn = owner.desc.image_layer_lsn();
                 let summary = Some(image_layer::Summary::expected(
@@ -1700,54 +1712,29 @@ impl DownloadedLayer {
                     ctx,
                 )
                 .await
-                .map(|res| res.map(LayerKind::Image))
+                .map(LayerKind::Image)
             };
 
             match res {
-                Ok(Ok(layer)) => Ok(Ok(layer)),
-                Ok(Err(transient)) => Err(transient),
-                Err(permanent) => {
+                Ok(layer) => Ok(layer),
+                Err(err) => {
                     LAYER_IMPL_METRICS.inc_permanent_loading_failures();
-                    // TODO(#5815): we are not logging all errors, so temporarily log them **once**
-                    // here as well
-                    let permanent = permanent.context("load layer");
-                    tracing::error!("layer loading failed permanently: {permanent:#}");
-                    Ok(Err(permanent))
+                    // We log this message once over the lifetime of `Self`
+                    // => Ok and good to log backtrace and path here.
+                    tracing::error!(
+                        "layer load failed, assuming permanent failure: {}: {err:?}",
+                        owner.path
+                    );
+                    Err(err)
                 }
             }
         };
         self.kind
-            .get_or_try_init(init)
-            // return transient errors using `?`
-            .await?
+            .get_or_init(init)
+            .await
             .as_ref()
-            .map_err(|e| {
-                // errors are not clonabled, cannot but stringify
-                // test_broken_timeline matches this string
-                anyhow::anyhow!("layer loading failed: {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
-            }
-        }
+            // We already logged the full backtrace above, once. Don't repeat that here.
+            .map_err(|e| anyhow::anyhow!("layer load failed earlier: {e}"))
     }
 
     async fn get_values_reconstruct_data(
@@ -1760,7 +1747,11 @@ impl DownloadedLayer {
     ) -> Result<(), GetVectoredError> {
         use LayerKind::*;
 
-        match self.get(owner, ctx).await.map_err(GetVectoredError::from)? {
+        match self
+            .get(owner, ctx)
+            .await
+            .map_err(GetVectoredError::Other)?
+        {
             Delta(d) => {
                 d.get_values_reconstruct_data(keyspace, lsn_range, reconstruct_data, ctx)
                     .await
@@ -1844,7 +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
@@ -1862,12 +1853,24 @@ impl ResidentLayer {
         shard_identity: &ShardIdentity,
         writer: &mut ImageLayerWriter,
         ctx: &RequestContext,
-    ) -> anyhow::Result<usize> {
+    ) -> Result<usize, CompactionError> {
         use LayerKind::*;
 
-        match self.downloaded.get(&self.owner.0, ctx).await? {
-            Delta(_) => anyhow::bail!(format!("cannot filter() on a delta layer {self}")),
-            Image(i) => i.filter(shard_identity, writer, ctx).await,
+        match self
+            .downloaded
+            .get(&self.owner.0, ctx)
+            .await
+            .map_err(CompactionError::Other)?
+        {
+            Delta(_) => {
+                return Err(CompactionError::Other(anyhow::anyhow!(format!(
+                    "cannot filter() on a delta layer {self}"
+                ))));
+            }
+            Image(i) => i
+                .filter(shard_identity, writer, ctx)
+                .await
+                .map_err(CompactionError::Other),
         }
     }
 

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
         };
@@ -828,9 +852,9 @@ async fn eviction_cancellation_on_drop() {
 #[test]
 #[cfg(target_arch = "x86_64")]
 fn layer_size() {
-    assert_eq!(std::mem::size_of::<LayerAccessStats>(), 8);
-    assert_eq!(std::mem::size_of::<PersistentLayerDesc>(), 104);
-    assert_eq!(std::mem::size_of::<LayerInner>(), 312);
+    assert_eq!(size_of::<LayerAccessStats>(), 8);
+    assert_eq!(size_of::<PersistentLayerDesc>(), 104);
+    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/merge_iterator.rs

@@ -204,9 +204,11 @@ impl<'a> IteratorWrapper<'a> {
 /// A merge iterator over delta/image layer iterators. When duplicated records are
 /// found, the iterator will not perform any deduplication, and the caller should handle
 /// these situation. By saying duplicated records, there are many possibilities:
+///
 /// * Two same delta at the same LSN.
 /// * Two same image at the same LSN.
 /// * Delta/image at the same LSN where the image has already applied the delta.
+///
 /// The iterator will always put the image before the delta.
 pub struct MergeIterator<'a> {
     heap: BinaryHeap<IteratorWrapper<'a>>,

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

@@ -210,24 +210,28 @@ async fn compaction_loop(tenant: Arc<Tenant>, cancel: CancellationToken) {
                 Duration::from_secs(10)
             } else {
                 // Run compaction
-                if let Err(e) = tenant.compaction_iteration(&cancel, &ctx).await {
-                    let wait_duration = backoff::exponential_backoff_duration_seconds(
-                        error_run_count + 1,
-                        1.0,
-                        MAX_BACKOFF_SECS,
-                    );
-                    error_run_count += 1;
-                    let wait_duration = Duration::from_secs_f64(wait_duration);
-                    log_compaction_error(
-                        &e,
-                        error_run_count,
-                        &wait_duration,
-                        cancel.is_cancelled(),
-                    );
-                    wait_duration
-                } else {
-                    error_run_count = 0;
-                    period
+                match tenant.compaction_iteration(&cancel, &ctx).await {
+                    Err(e) => {
+                        let wait_duration = backoff::exponential_backoff_duration_seconds(
+                            error_run_count + 1,
+                            1.0,
+                            MAX_BACKOFF_SECS,
+                        );
+                        error_run_count += 1;
+                        let wait_duration = Duration::from_secs_f64(wait_duration);
+                        log_compaction_error(
+                            &e,
+                            error_run_count,
+                            &wait_duration,
+                            cancel.is_cancelled(),
+                        );
+                        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 }
+                    }
                 }
             };
 
@@ -403,9 +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
                     }
                 }

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

@@ -1,4 +1,4 @@
-use std::sync::Arc;
+use std::{collections::HashSet, sync::Arc};
 
 use super::{layer_manager::LayerManager, FlushLayerError, Timeline};
 use crate::{
@@ -26,7 +26,7 @@ pub(crate) enum Error {
     #[error("flushing failed")]
     FlushAncestor(#[source] FlushLayerError),
     #[error("layer download failed")]
-    RewrittenDeltaDownloadFailed(#[source] anyhow::Error),
+    RewrittenDeltaDownloadFailed(#[source] crate::tenant::storage_layer::layer::DownloadError),
     #[error("copying LSN prefix locally failed")]
     CopyDeltaPrefix(#[source] anyhow::Error),
     #[error("upload rewritten layer")]
@@ -74,6 +74,11 @@ 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 {
@@ -141,50 +146,9 @@ pub(super) async fn prepare(
             }
         }
 
-        // 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,
         }));
     };
 
@@ -277,7 +241,7 @@ 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
@@ -381,16 +345,67 @@ pub(super) async fn prepare(
     Ok(Progress::Prepared(guard, prepared))
 }
 
+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 +424,10 @@ 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 +503,12 @@ 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");
 
@@ -537,11 +554,12 @@ pub(super) async fn complete(
     tenant: &Tenant,
     prepared: PreparedTimelineDetach,
     _ctx: &RequestContext,
-) -> Result<Vec<TimelineId>, anyhow::Error> {
+) -> Result<HashSet<TimelineId>, anyhow::Error> {
     let PreparedTimelineDetach { layers } = prepared;
 
     let ancestor = detached
-        .get_ancestor_timeline()
+        .ancestor_timeline
+        .as_ref()
         .expect("must still have a ancestor");
     let ancestor_lsn = detached.get_ancestor_lsn();
 
@@ -581,7 +599,7 @@ pub(super) async fn complete(
             }
 
             let tl_ancestor = tl.ancestor_timeline.as_ref()?;
-            let is_same = Arc::ptr_eq(&ancestor, tl_ancestor);
+            let is_same = Arc::ptr_eq(ancestor, tl_ancestor);
             let is_earlier = tl.get_ancestor_lsn() <= ancestor_lsn;
 
             let is_deleting = tl
@@ -622,13 +640,18 @@ pub(super) async fn complete(
         });
 
     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));
+
+                assert!(
+                    reparented.insert(timeline.timeline_id),
+                    "duplicate reparenting? timeline_id={}",
+                    timeline.timeline_id
+                );
             }
             Ok(None) => {
                 // lets just ignore this for now. one or all reparented timelines could had
@@ -650,12 +673,5 @@ pub(super) async fn complete(
         tracing::info!("failed to reparent some candidates");
     }
 
-    reparented.sort_unstable();
-
-    let reparented = reparented
-        .into_iter()
-        .map(|(_, timeline_id)| timeline_id)
-        .collect();
-
     Ok(reparented)
 }

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/upload_queue.rs

@@ -130,7 +130,7 @@ pub(super) enum UploadQueueStopped {
 }
 
 #[derive(thiserror::Error, Debug)]
-pub(crate) enum NotInitialized {
+pub enum NotInitialized {
     #[error("queue is in state Uninitialized")]
     Uninitialized,
     #[error("queue is in state Stopped")]

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/walingest.rs

@@ -618,7 +618,7 @@ impl WalIngest {
                                 // the offsets array is omitted if XLOG_HEAP_INIT_PAGE is set
                                 0
                             } else {
-                                std::mem::size_of::<u16>() * xlrec.ntuples as usize
+                                size_of::<u16>() * xlrec.ntuples as usize
                             };
                         assert_eq!(offset_array_len, buf.remaining());
 
@@ -685,7 +685,7 @@ impl WalIngest {
                                 // the offsets array is omitted if XLOG_HEAP_INIT_PAGE is set
                                 0
                             } else {
-                                std::mem::size_of::<u16>() * xlrec.ntuples as usize
+                                size_of::<u16>() * xlrec.ntuples as usize
                             };
                         assert_eq!(offset_array_len, buf.remaining());
 
@@ -752,7 +752,7 @@ impl WalIngest {
                                 // the offsets array is omitted if XLOG_HEAP_INIT_PAGE is set
                                 0
                             } else {
-                                std::mem::size_of::<u16>() * xlrec.ntuples as usize
+                                size_of::<u16>() * xlrec.ntuples as usize
                             };
                         assert_eq!(offset_array_len, buf.remaining());
 
@@ -920,7 +920,7 @@ impl WalIngest {
                                 // the offsets array is omitted if XLOG_HEAP_INIT_PAGE is set
                                 0
                             } else {
-                                std::mem::size_of::<u16>() * xlrec.ntuples as usize
+                                size_of::<u16>() * xlrec.ntuples as usize
                             };
                         assert_eq!(offset_array_len, buf.remaining());
 

pageserver/src/walredo.rs

@@ -241,6 +241,9 @@ 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 +253,32 @@ 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

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.