try with unshallow

increase depth
use a version of submodule check with extra debug info
2026-01-25 14:20:38 +00:00 · 2024-11-12 13:00:42 +02:00 · 2024-11-12 11:57:28 +02:00 · 2024-11-12 11:47:42 +02:00 · 2024-11-12 08:25:16 +02:00 · 2024-11-12 08:25:16 +02:00
57 changed files with 1340 additions and 455 deletions
--- a/.github/actionlint.yml
+++ b/.github/actionlint.yml
@@ -20,3 +20,4 @@ config-variables:
  - REMOTE_STORAGE_AZURE_REGION
  - SLACK_UPCOMING_RELEASE_CHANNEL_ID
  - DEV_AWS_OIDC_ROLE_ARN
+  - BENCHMARK_INGEST_TARGET_PROJECTID
--- a/.github/actions/allure-report-generate/action.yml
+++ b/.github/actions/allure-report-generate/action.yml
@@ -221,6 +221,8 @@ runs:
        REPORT_URL: ${{ steps.generate-report.outputs.report-url }}
        COMMIT_SHA: ${{ github.event.pull_request.head.sha || github.sha }}
      with:
+        # Retry script for 5XX server errors: https://github.com/actions/github-script#retries
+        retries: 5
        script: |
          const { REPORT_URL, COMMIT_SHA } = process.env

--- a/.github/pull_request_template.md
+++ b/.github/pull_request_template.md
@@ -1,14 +1,3 @@
 ## Problem

 ## Summary of changes
-
-## Checklist before requesting a review
-
- [ ] I have performed a self-review of my code.
- [ ] If it is a core feature, I have added thorough tests.
- [ ] Do we need to implement analytics? if so did you add the relevant metrics to the dashboard?
- [ ] If this PR requires public announcement, mark it with /release-notes label and add several sentences in this section.
-
-## Checklist before merging
-
- [ ] Do not forget to reformat commit message to not include the above checklist
--- a/.github/workflows/build_and_test.yml
+++ b/.github/workflows/build_and_test.yml
@@ -133,38 +133,38 @@ jobs:

      - name: Check vendor/postgres-v14 submodule reference
        if: steps.check-if-submodules-changed.outputs.vendor == 'true'
-        uses: jtmullen/submodule-branch-check-action@v1
+        uses: hlinnaka/submodule-branch-check-action@main
        with:
          path: "vendor/postgres-v14"
          fetch_depth: "50"
-          sub_fetch_depth: "50"
+          sub_fetch_depth: ""
          pass_if_unchanged: true

      - name: Check vendor/postgres-v15 submodule reference
        if: steps.check-if-submodules-changed.outputs.vendor == 'true'
-        uses: jtmullen/submodule-branch-check-action@v1
+        uses: hlinnaka/submodule-branch-check-action@main
        with:
          path: "vendor/postgres-v15"
          fetch_depth: "50"
-          sub_fetch_depth: "50"
+          sub_fetch_depth: ""
          pass_if_unchanged: true

      - name: Check vendor/postgres-v16 submodule reference
        if: steps.check-if-submodules-changed.outputs.vendor == 'true'
-        uses: jtmullen/submodule-branch-check-action@v1
+        uses: hlinnaka/submodule-branch-check-action@main
        with:
          path: "vendor/postgres-v16"
          fetch_depth: "50"
-          sub_fetch_depth: "50"
+          sub_fetch_depth: ""
          pass_if_unchanged: true

      - name: Check vendor/postgres-v17 submodule reference
        if: steps.check-if-submodules-changed.outputs.vendor == 'true'
-        uses: jtmullen/submodule-branch-check-action@v1
+        uses: hlinnaka/submodule-branch-check-action@main
        with:
          path: "vendor/postgres-v17"
          fetch_depth: "50"
-          sub_fetch_depth: "50"
+          sub_fetch_depth: ""
          pass_if_unchanged: true

  check-codestyle-rust:
@@ -497,6 +497,8 @@ jobs:
          REPORT_URL_NEW: ${{ steps.upload-coverage-report-new.outputs.report-url }}
          COMMIT_SHA: ${{ github.event.pull_request.head.sha || github.sha }}
        with:
+          # Retry script for 5XX server errors: https://github.com/actions/github-script#retries
+          retries: 5
          script: |
            const { REPORT_URL_NEW, COMMIT_SHA } = process.env

--- a/.github/workflows/ingest_benchmark.yml
+++ b/.github/workflows/ingest_benchmark.yml
@@ -0,0 +1,372 @@
+name: Benchmarking
+
+on:
+  # uncomment to run on push for debugging your PR
+  # push:
+  #   branches: [ your branch ]
+  schedule:
+    # * is a special character in YAML so you have to quote this string
+    #          ┌───────────── minute (0 - 59)
+    #          │ ┌───────────── hour (0 - 23)
+    #          │ │ ┌───────────── day of the month (1 - 31)
+    #          │ │ │ ┌───────────── month (1 - 12 or JAN-DEC)
+    #          │ │ │ │ ┌───────────── day of the week (0 - 6 or SUN-SAT)
+    - cron:   '0 9 * * *' # run once a day, timezone is utc
+  workflow_dispatch: # adds ability to run this manually
+    
+defaults:
+  run:
+    shell: bash -euxo pipefail {0}
+
+concurrency:
+  # Allow only one workflow globally because we need dedicated resources which only exist once
+  group: ingest-bench-workflow
+  cancel-in-progress: true
+
+jobs:
+  ingest:
+    strategy:
+      matrix:
+        target_project: [new_empty_project, large_existing_project]  
+    permissions:
+      contents: write
+      statuses: write
+      id-token: write # aws-actions/configure-aws-credentials
+    env:
+      PG_CONFIG: /tmp/neon/pg_install/v16/bin/pg_config
+      PSQL: /tmp/neon/pg_install/v16/bin/psql
+      PG_16_LIB_PATH: /tmp/neon/pg_install/v16/lib
+      PGCOPYDB: /pgcopydb/bin/pgcopydb
+      PGCOPYDB_LIB_PATH: /pgcopydb/lib
+    runs-on: [ self-hosted, us-east-2, x64 ]
+    container:
+      image: neondatabase/build-tools:pinned-bookworm
+      credentials:
+        username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
+        password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
+      options: --init
+    timeout-minutes: 1440
+
+    steps:
+    - uses: actions/checkout@v4
+
+    - name: Configure AWS credentials # necessary to download artefacts
+      uses: aws-actions/configure-aws-credentials@v4
+      with:
+        aws-region: eu-central-1
+        role-to-assume: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+        role-duration-seconds: 18000 # 5 hours is currently max associated with IAM role 
+
+    - name: Download Neon artifact
+      uses: ./.github/actions/download
+      with:
+        name: neon-${{ runner.os }}-${{ runner.arch }}-release-artifact
+        path: /tmp/neon/
+        prefix: latest
+
+    - name: Create Neon Project
+      if: ${{ matrix.target_project == 'new_empty_project' }}
+      id: create-neon-project-ingest-target
+      uses: ./.github/actions/neon-project-create
+      with:
+        region_id: aws-us-east-2
+        postgres_version: 16
+        compute_units: '[7, 7]' # we want to test large compute here to avoid compute-side bottleneck
+        api_key: ${{ secrets.NEON_STAGING_API_KEY }}
+
+    - name: Initialize Neon project and retrieve current backpressure seconds
+      if: ${{ matrix.target_project == 'new_empty_project' }}
+      env:
+          NEW_PROJECT_CONNSTR: ${{ steps.create-neon-project-ingest-target.outputs.dsn }}
+          NEW_PROJECT_ID: ${{ steps.create-neon-project-ingest-target.outputs.project_id }}
+      run: |
+        echo "Initializing Neon project with project_id: ${NEW_PROJECT_ID}"
+        export LD_LIBRARY_PATH=${PG_16_LIB_PATH}
+        ${PSQL} "${NEW_PROJECT_CONNSTR}" -c "CREATE EXTENSION IF NOT EXISTS neon; CREATE EXTENSION IF NOT EXISTS neon_utils;"
+        BACKPRESSURE_TIME_BEFORE_INGEST=$(${PSQL} "${NEW_PROJECT_CONNSTR}" -t -c "select backpressure_throttling_time()/1000000;")
+        echo "BACKPRESSURE_TIME_BEFORE_INGEST=${BACKPRESSURE_TIME_BEFORE_INGEST}" >> $GITHUB_ENV
+        echo "NEW_PROJECT_CONNSTR=${NEW_PROJECT_CONNSTR}" >> $GITHUB_ENV
+
+    - name: Create Neon Branch for large tenant
+      if: ${{ matrix.target_project == 'large_existing_project' }}
+      id: create-neon-branch-ingest-target
+      uses: ./.github/actions/neon-branch-create
+      with:
+        project_id: ${{ vars.BENCHMARK_INGEST_TARGET_PROJECTID }}
+        api_key: ${{ secrets.NEON_STAGING_API_KEY }}
+
+    - name: Initialize Neon project and retrieve current backpressure seconds
+      if: ${{ matrix.target_project == 'large_existing_project' }}
+      env:
+          NEW_PROJECT_CONNSTR: ${{ steps.create-neon-branch-ingest-target.outputs.dsn }}
+          NEW_BRANCH_ID: ${{ steps.create-neon-branch-ingest-target.outputs.branch_id }}
+      run: |
+        echo "Initializing Neon branch with branch_id: ${NEW_BRANCH_ID}"
+        export LD_LIBRARY_PATH=${PG_16_LIB_PATH}
+        # Extract the part before the database name
+        base_connstr="${NEW_PROJECT_CONNSTR%/*}"
+        # Extract the query parameters (if any) after the database name
+        query_params="${NEW_PROJECT_CONNSTR#*\?}"
+        # Reconstruct the new connection string
+        if [ "$query_params" != "$NEW_PROJECT_CONNSTR" ]; then
+          new_connstr="${base_connstr}/neondb?${query_params}"
+        else
+          new_connstr="${base_connstr}/neondb"
+        fi
+        ${PSQL} "${new_connstr}" -c "drop database ludicrous;"
+        ${PSQL} "${new_connstr}" -c "CREATE DATABASE ludicrous;"
+        if [ "$query_params" != "$NEW_PROJECT_CONNSTR" ]; then
+          NEW_PROJECT_CONNSTR="${base_connstr}/ludicrous?${query_params}"
+        else
+          NEW_PROJECT_CONNSTR="${base_connstr}/ludicrous"
+        fi
+        ${PSQL} "${NEW_PROJECT_CONNSTR}" -c "CREATE EXTENSION IF NOT EXISTS neon; CREATE EXTENSION IF NOT EXISTS neon_utils;"
+        BACKPRESSURE_TIME_BEFORE_INGEST=$(${PSQL} "${NEW_PROJECT_CONNSTR}" -t -c "select backpressure_throttling_time()/1000000;")
+        echo "BACKPRESSURE_TIME_BEFORE_INGEST=${BACKPRESSURE_TIME_BEFORE_INGEST}" >> $GITHUB_ENV
+        echo "NEW_PROJECT_CONNSTR=${NEW_PROJECT_CONNSTR}" >> $GITHUB_ENV
+      
+        
+    - name: Create pgcopydb filter file
+      run: |
+        cat << EOF > /tmp/pgcopydb_filter.txt
+          [include-only-table]
+          public.events
+          public.emails
+          public.email_transmissions
+          public.payments
+          public.editions
+          public.edition_modules
+          public.sp_content
+          public.email_broadcasts
+          public.user_collections
+          public.devices
+          public.user_accounts
+          public.lessons
+          public.lesson_users
+          public.payment_methods
+          public.orders
+          public.course_emails
+          public.modules
+          public.users
+          public.module_users
+          public.courses
+          public.payment_gateway_keys
+          public.accounts
+          public.roles
+          public.payment_gateways
+          public.management
+          public.event_names
+        EOF
+
+    - name: Invoke pgcopydb
+      env:
+          BENCHMARK_INGEST_SOURCE_CONNSTR: ${{ secrets.BENCHMARK_INGEST_SOURCE_CONNSTR }}
+      run: |
+        export LD_LIBRARY_PATH=${PGCOPYDB_LIB_PATH}:${PG_16_LIB_PATH}
+        export PGCOPYDB_SOURCE_PGURI="${BENCHMARK_INGEST_SOURCE_CONNSTR}"
+        export PGCOPYDB_TARGET_PGURI="${NEW_PROJECT_CONNSTR}"
+        export PGOPTIONS="-c maintenance_work_mem=8388608 -c max_parallel_maintenance_workers=7"
+        ${PG_CONFIG} --bindir
+        ${PGCOPYDB} --version
+        ${PGCOPYDB} clone --skip-vacuum  --no-owner --no-acl --skip-db-properties --table-jobs 4 \
+          --index-jobs 4 --restore-jobs 4 --split-tables-larger-than 10GB --skip-extensions \
+          --use-copy-binary --filters /tmp/pgcopydb_filter.txt 2>&1 | tee /tmp/pgcopydb_${{ matrix.target_project }}.log
+
+    # create dummy pgcopydb log to test parsing
+    # - name: create dummy log for parser test
+    #   run: |
+    #     cat << EOF > /tmp/pgcopydb_${{ matrix.target_project }}.log
+    #     2024-11-04 18:00:53.433 500861 INFO   main.c:136                Running pgcopydb version 0.17.10.g8361a93 from "/usr/lib/postgresql/17/bin/pgcopydb"
+    #     2024-11-04 18:00:53.434 500861 INFO   cli_common.c:1225         [SOURCE] Copying database from "postgres://neondb_owner@ep-bitter-shape-w2c1ir0a.us-east-2.aws.neon.build/neondb?sslmode=require&keepalives=1&keepalives_idle=10&keepalives_interval=10&keepalives_count=60"
+    #     2024-11-04 18:00:53.434 500861 INFO   cli_common.c:1226         [TARGET] Copying database into "postgres://neondb_owner@ep-icy-union-w25qd5pj.us-east-2.aws.neon.build/ludicrous?sslmode=require&keepalives=1&keepalives_idle=10&keepalives_interval=10&keepalives_count=60"
+    #     2024-11-04 18:00:53.442 500861 INFO   copydb.c:105              Using work dir "/tmp/pgcopydb"
+    #     2024-11-04 18:00:53.541 500861 INFO   snapshot.c:107            Exported snapshot "00000008-00000033-1" from the source database
+    #     2024-11-04 18:00:53.556 500865 INFO   cli_clone_follow.c:543    STEP 1: fetch source database tables, indexes, and sequences
+    #     2024-11-04 18:00:54.570 500865 INFO   copydb_schema.c:716       Splitting source candidate tables larger than 10 GB
+    #     2024-11-04 18:00:54.570 500865 INFO   copydb_schema.c:829       Table public.events is 96 GB large which is larger than --split-tables-larger-than 10 GB, and does not have a unique column of type integer: splitting by CTID
+    #     2024-11-04 18:01:05.538 500865 INFO   copydb_schema.c:905       Table public.events is 96 GB large, 10 COPY processes will be used, partitioning on ctid.
+    #     2024-11-04 18:01:05.564 500865 INFO   copydb_schema.c:905       Table public.email_transmissions is 27 GB large, 4 COPY processes will be used, partitioning on id.
+    #     2024-11-04 18:01:05.584 500865 INFO   copydb_schema.c:905       Table public.lessons is 25 GB large, 4 COPY processes will be used, partitioning on id.
+    #     2024-11-04 18:01:05.605 500865 INFO   copydb_schema.c:905       Table public.lesson_users is 16 GB large, 3 COPY processes will be used, partitioning on id.
+    #     2024-11-04 18:01:05.605 500865 INFO   copydb_schema.c:761       Fetched information for 26 tables (including 4 tables split in 21 partitions total), with an estimated total of 907 million tuples and 175 GB on-disk
+    #     2024-11-04 18:01:05.687 500865 INFO   copydb_schema.c:968       Fetched information for 57 indexes (supporting 25 constraints)
+    #     2024-11-04 18:01:05.753 500865 INFO   sequences.c:78            Fetching information for 24 sequences
+    #     2024-11-04 18:01:05.903 500865 INFO   copydb_schema.c:1122      Fetched information for 4 extensions
+    #     2024-11-04 18:01:06.178 500865 INFO   copydb_schema.c:1538      Found 0 indexes (supporting 0 constraints) in the target database
+    #     2024-11-04 18:01:06.184 500865 INFO   cli_clone_follow.c:584    STEP 2: dump the source database schema (pre/post data)
+    #     2024-11-04 18:01:06.186 500865 INFO   pgcmd.c:468                /usr/lib/postgresql/16/bin/pg_dump -Fc --snapshot 00000008-00000033-1 --section=pre-data --section=post-data --file /tmp/pgcopydb/schema/schema.dump 'postgres://neondb_owner@ep-bitter-shape-w2c1ir0a.us-east-2.aws.neon.build/neondb?sslmode=require&keepalives=1&keepalives_idle=10&keepalives_interval=10&keepalives_count=60'
+    #     2024-11-04 18:01:06.952 500865 INFO   cli_clone_follow.c:592    STEP 3: restore the pre-data section to the target database
+    #     2024-11-04 18:01:07.004 500865 INFO   pgcmd.c:1001               /usr/lib/postgresql/16/bin/pg_restore --dbname 'postgres://neondb_owner@ep-icy-union-w25qd5pj.us-east-2.aws.neon.build/ludicrous?sslmode=require&keepalives=1&keepalives_idle=10&keepalives_interval=10&keepalives_count=60' --section pre-data --jobs 4 --no-owner --no-acl --use-list /tmp/pgcopydb/schema/pre-filtered.list /tmp/pgcopydb/schema/schema.dump
+    #     2024-11-04 18:01:07.438 500874 INFO   table-data.c:656          STEP 4: starting 4 table-data COPY processes
+    #     2024-11-04 18:01:07.451 500877 INFO   vacuum.c:139              STEP 8: skipping VACUUM jobs per --skip-vacuum
+    #     2024-11-04 18:01:07.457 500875 INFO   indexes.c:182             STEP 6: starting 4 CREATE INDEX processes
+    #     2024-11-04 18:01:07.457 500875 INFO   indexes.c:183             STEP 7: constraints are built by the CREATE INDEX processes
+    #     2024-11-04 18:01:07.507 500865 INFO   blobs.c:74                Skipping large objects: none found.
+    #     2024-11-04 18:01:07.509 500865 INFO   sequences.c:194           STEP 9: reset sequences values
+    #     2024-11-04 18:01:07.510 500886 INFO   sequences.c:290           Set sequences values on the target database
+    #     2024-11-04 20:49:00.587 500865 INFO   cli_clone_follow.c:608    STEP 10: restore the post-data section to the target database
+    #     2024-11-04 20:49:00.600 500865 INFO   pgcmd.c:1001               /usr/lib/postgresql/16/bin/pg_restore --dbname 'postgres://neondb_owner@ep-icy-union-w25qd5pj.us-east-2.aws.neon.build/ludicrous?sslmode=require&keepalives=1&keepalives_idle=10&keepalives_interval=10&keepalives_count=60' --section post-data --jobs 4 --no-owner --no-acl --use-list /tmp/pgcopydb/schema/post-filtered.list /tmp/pgcopydb/schema/schema.dump
+    #     2024-11-05 10:50:58.508 500865 INFO   cli_clone_follow.c:639    All step are now done, 16h49m elapsed
+    #     2024-11-05 10:50:58.508 500865 INFO   summary.c:3155            Printing summary for 26 tables and 57 indexes
+
+    #       OID | Schema |                 Name | Parts | copy duration | transmitted bytes | indexes | create index duration 
+    #     ------+--------+----------------------+-------+---------------+-------------------+---------+----------------------
+    #     24654 | public |               events |    10 |         1d11h |            878 GB |       1 |                 1h41m
+    #     24623 | public |  email_transmissions |     4 |         4h46m |             99 GB |       3 |                 2h04m
+    #     24665 | public |              lessons |     4 |         4h42m |            161 GB |       4 |                 1m11s
+    #     24661 | public |         lesson_users |     3 |         2h46m |             49 GB |       3 |                39m35s
+    #     24631 | public |               emails |     1 |        34m07s |             10 GB |       2 |                   17s
+    #     24739 | public |             payments |     1 |         5m47s |           1848 MB |       4 |                 4m40s
+    #     24681 | public |         module_users |     1 |         4m57s |           1610 MB |       3 |                 1m50s
+    #     24694 | public |               orders |     1 |         2m50s |            835 MB |       3 |                 1m05s
+    #     24597 | public |              devices |     1 |         1m45s |            498 MB |       2 |                   40s
+    #     24723 | public |      payment_methods |     1 |         1m24s |            548 MB |       2 |                   31s
+    #     24765 | public |     user_collections |     1 |         2m17s |           1005 MB |       2 |                 968ms
+    #     24774 | public |                users |     1 |           52s |            291 MB |       4 |                   27s
+    #     24760 | public |        user_accounts |     1 |           16s |            172 MB |       3 |                   16s
+    #     24606 | public |      edition_modules |     1 |         8s983 |             46 MB |       3 |                 4s749
+    #     24583 | public |        course_emails |     1 |         8s526 |             26 MB |       2 |                 996ms
+    #     24685 | public |              modules |     1 |         1s592 |             21 MB |       3 |                 1s696
+    #     24610 | public |             editions |     1 |         2s199 |           7483 kB |       2 |                 1s032
+    #     24755 | public |           sp_content |     1 |         1s555 |           4177 kB |       0 |                   0ms
+    #     24619 | public |     email_broadcasts |     1 |         744ms |           2645 kB |       2 |                 677ms
+    #     24590 | public |              courses |     1 |         387ms |           1540 kB |       2 |                 367ms
+    #     24704 | public | payment_gateway_keys |     1 |         1s972 |            164 kB |       2 |                  27ms
+    #     24576 | public |             accounts |     1 |          58ms |             24 kB |       1 |                  14ms
+    #     24647 | public |          event_names |     1 |          32ms |             397 B |       1 |                   8ms
+    #     24716 | public |     payment_gateways |     1 |         1s675 |             117 B |       1 |                  11ms
+    #     24748 | public |                roles |     1 |          71ms |             173 B |       1 |                   8ms
+    #     24676 | public |           management |     1 |          33ms |              40 B |       1 |                  19ms
+
+
+    #                                                   Step   Connection    Duration    Transfer   Concurrency
+    #     --------------------------------------------------   ----------  ----------  ----------  ------------
+    #       Catalog Queries (table ordering, filtering, etc)       source         12s                         1
+    #                                             Dump Schema       source       765ms                         1
+    #                                         Prepare Schema       target       466ms                         1
+    #           COPY, INDEX, CONSTRAINTS, VACUUM (wall clock)         both       2h47m                        12
+    #                                       COPY (cumulative)         both       7h46m     1225 GB             4
+    #                               CREATE INDEX (cumulative)       target       4h36m                         4
+    #                               CONSTRAINTS (cumulative)       target       8s493                         4
+    #                                     VACUUM (cumulative)       target         0ms                         4
+    #                                         Reset Sequences         both        60ms                         1
+    #                             Large Objects (cumulative)       (null)         0ms                         0
+    #                                         Finalize Schema         both      14h01m                         4
+    #     --------------------------------------------------   ----------  ----------  ----------  ------------
+    #                               Total Wall Clock Duration         both      16h49m                        20
+
+
+    #     EOF
+
+
+    - name: show tables sizes and retrieve current backpressure seconds
+      run: |
+        export LD_LIBRARY_PATH=${PG_16_LIB_PATH}
+        ${PSQL} "${NEW_PROJECT_CONNSTR}" -c "\dt+"
+        BACKPRESSURE_TIME_AFTER_INGEST=$(${PSQL} "${NEW_PROJECT_CONNSTR}" -t -c "select backpressure_throttling_time()/1000000;")
+        echo "BACKPRESSURE_TIME_AFTER_INGEST=${BACKPRESSURE_TIME_AFTER_INGEST}" >> $GITHUB_ENV
+
+    - name: Parse pgcopydb log and report performance metrics
+      env:
+        PERF_TEST_RESULT_CONNSTR: ${{ secrets.PERF_TEST_RESULT_CONNSTR }}
+      run: |
+        export LD_LIBRARY_PATH=${PG_16_LIB_PATH}
+
+        # Define the log file path
+        LOG_FILE="/tmp/pgcopydb_${{ matrix.target_project }}.log"
+        
+        # Get the current git commit hash
+        git config --global --add safe.directory /__w/neon/neon
+        COMMIT_HASH=$(git rev-parse --short HEAD)
+        
+        # Define the platform and test suite
+        PLATFORM="pg16-${{ matrix.target_project }}-us-east-2-staging"
+        SUIT="pgcopydb_ingest_bench"
+        
+        # Function to convert time (e.g., "2h47m", "4h36m", "118ms", "8s493") to seconds
+        convert_to_seconds() {
+          local duration=$1
+          local total_seconds=0
+    
+          # Check for hours (h)
+          if [[ "$duration" =~ ([0-9]+)h ]]; then
+            total_seconds=$((total_seconds + ${BASH_REMATCH[1]#0} * 3600))
+          fi
+    
+          # Check for seconds (s)
+          if [[ "$duration" =~ ([0-9]+)s ]]; then
+            total_seconds=$((total_seconds + ${BASH_REMATCH[1]#0}))
+          fi
+    
+          # Check for milliseconds (ms) (if applicable)
+          if [[ "$duration" =~ ([0-9]+)ms ]]; then
+            total_seconds=$((total_seconds + ${BASH_REMATCH[1]#0} / 1000))
+            duration=${duration/${BASH_REMATCH[0]}/} # need to remove it to avoid double counting with m 
+          fi
+
+          # Check for minutes (m) - must be checked after ms because m is contained in ms
+          if [[ "$duration" =~ ([0-9]+)m ]]; then
+            total_seconds=$((total_seconds + ${BASH_REMATCH[1]#0} * 60))
+          fi
+    
+          echo $total_seconds
+        }
+
+        # Calculate the backpressure difference in seconds
+        BACKPRESSURE_TIME_DIFF=$(awk "BEGIN {print $BACKPRESSURE_TIME_AFTER_INGEST - $BACKPRESSURE_TIME_BEFORE_INGEST}")
+
+        # Insert the backpressure time difference into the performance database
+        if [ -n "$BACKPRESSURE_TIME_DIFF" ]; then
+          PSQL_CMD="${PSQL} \"${PERF_TEST_RESULT_CONNSTR}\" -c \"
+          INSERT INTO public.perf_test_results (suit, revision, platform, metric_name, metric_value, metric_unit, metric_report_type, recorded_at_timestamp)
+          VALUES ('${SUIT}', '${COMMIT_HASH}', '${PLATFORM}', 'backpressure_time', ${BACKPRESSURE_TIME_DIFF}, 'seconds', 'lower_is_better', now());
+          \""
+          echo "Inserting backpressure time difference: ${BACKPRESSURE_TIME_DIFF} seconds"
+          eval $PSQL_CMD
+        fi
+
+        # Extract and process log lines
+        while IFS= read -r line; do
+          METRIC_NAME=""
+          # Match each desired line and extract the relevant information
+          if [[ "$line" =~ COPY,\ INDEX,\ CONSTRAINTS,\ VACUUM.* ]]; then
+            METRIC_NAME="COPY, INDEX, CONSTRAINTS, VACUUM (wall clock)"
+          elif [[ "$line" =~ COPY\ \(cumulative\).* ]]; then
+            METRIC_NAME="COPY (cumulative)"
+          elif [[ "$line" =~ CREATE\ INDEX\ \(cumulative\).* ]]; then
+            METRIC_NAME="CREATE INDEX (cumulative)"
+          elif [[ "$line" =~ CONSTRAINTS\ \(cumulative\).* ]]; then
+            METRIC_NAME="CONSTRAINTS (cumulative)"
+          elif [[ "$line" =~ Finalize\ Schema.* ]]; then
+            METRIC_NAME="Finalize Schema"
+          elif [[ "$line" =~ Total\ Wall\ Clock\ Duration.* ]]; then
+            METRIC_NAME="Total Wall Clock Duration"
+          fi
+          
+          # If a metric was matched, insert it into the performance database
+          if [ -n "$METRIC_NAME" ]; then
+            DURATION=$(echo "$line" | grep -oP '\d+h\d+m|\d+s|\d+ms|\d{1,2}h\d{1,2}m|\d+\.\d+s' | head -n 1)
+            METRIC_VALUE=$(convert_to_seconds "$DURATION")
+            PSQL_CMD="${PSQL} \"${PERF_TEST_RESULT_CONNSTR}\" -c \"
+            INSERT INTO public.perf_test_results (suit, revision, platform, metric_name, metric_value, metric_unit, metric_report_type, recorded_at_timestamp)
+            VALUES ('${SUIT}', '${COMMIT_HASH}', '${PLATFORM}', '${METRIC_NAME}', ${METRIC_VALUE}, 'seconds', 'lower_is_better', now());
+            \""
+            echo "Inserting ${METRIC_NAME} with value ${METRIC_VALUE} seconds"
+            eval $PSQL_CMD
+          fi
+        done < "$LOG_FILE"
+      
+    - name: Delete Neon Project
+      if: ${{ always() && matrix.target_project == 'new_empty_project' }}
+      uses: ./.github/actions/neon-project-delete
+      with:
+        project_id: ${{ steps.create-neon-project-ingest-target.outputs.project_id }}
+        api_key: ${{ secrets.NEON_STAGING_API_KEY }}
+
+    - name: Delete Neon Branch for large tenant
+      if: ${{ always() && matrix.target_project == 'large_existing_project' }}
+      uses: ./.github/actions/neon-branch-delete
+      with:
+        project_id: ${{ vars.BENCHMARK_INGEST_TARGET_PROJECTID }}
+        branch_id: ${{ steps.create-neon-branch-ingest-target.outputs.branch_id }}
+        api_key: ${{ secrets.NEON_STAGING_API_KEY }}
--- a/.github/workflows/neon_extra_builds.yml
+++ b/.github/workflows/neon_extra_builds.yml
@@ -201,6 +201,8 @@ jobs:
          REPORT_URL: ${{ steps.upload-stats.outputs.report-url }}
          SHA: ${{ github.event.pull_request.head.sha || github.sha }}
        with:
+          # Retry script for 5XX server errors: https://github.com/actions/github-script#retries
+          retries: 5
          script: |
            const { REPORT_URL, SHA } = process.env

--- a/.github/workflows/report-workflow-stats-batch.yml
+++ b/.github/workflows/report-workflow-stats-batch.yml
@@ -0,0 +1,29 @@
+name: Report Workflow Stats Batch
+
+on:
+  schedule:
+    - cron: '*/15 * * * *'
+    - cron: '25 0 * * *'
+
+jobs:
+  gh-workflow-stats-batch:
+    name: GitHub Workflow Stats Batch
+    runs-on: ubuntu-22.04
+    permissions:
+      actions: read
+    steps:
+    - name: Export Workflow Run for the past 2 hours
+      uses: neondatabase/gh-workflow-stats-action@v0.2.1
+      with:
+        db_uri: ${{ secrets.GH_REPORT_STATS_DB_RW_CONNSTR }}
+        db_table: "gh_workflow_stats_batch_neon"
+        gh_token: ${{ secrets.GITHUB_TOKEN }}
+        duration: '2h'
+    - name: Export Workflow Run for the past 24 hours
+      if: github.event.schedule == '25 0 * * *'
+      uses: neondatabase/gh-workflow-stats-action@v0.2.1
+      with:
+        db_uri: ${{ secrets.GH_REPORT_STATS_DB_RW_CONNSTR }}
+        db_table: "gh_workflow_stats_batch_neon"
+        gh_token: ${{ secrets.GITHUB_TOKEN }}
+        duration: '24h'
--- a/compute/compute-node.Dockerfile
+++ b/compute/compute-node.Dockerfile
@@ -624,16 +624,12 @@ FROM build-deps AS pg-cron-pg-build
 ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

-# 1.6.4 available, supports v17
 # This is an experimental extension that we do not support on prod yet.
 # !Do not remove!
 # We set it in shared_preload_libraries and computes will fail to start if library is not found.
 ENV PATH="/usr/local/pgsql/bin/:$PATH"
-RUN case "${PG_VERSION}" in "v17") \
-    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
-    esac && \
-    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 && \
+RUN wget https://github.com/citusdata/pg_cron/archive/refs/tags/v1.6.4.tar.gz -O pg_cron.tar.gz && \
+    echo "52d1850ee7beb85a4cb7185731ef4e5a90d1de216709d8988324b0d02e76af61 pg_cron.tar.gz" | sha256sum --check && \
    mkdir pg_cron-src && cd pg_cron-src && tar xzf ../pg_cron.tar.gz --strip-components=1 -C . && \
    make -j $(getconf _NPROCESSORS_ONLN) && \
    make -j $(getconf _NPROCESSORS_ONLN) install && \
@@ -1475,6 +1471,8 @@ RUN mkdir -p /etc/local_proxy && chown postgres:postgres /etc/local_proxy
 COPY --from=postgres-exporter /bin/postgres_exporter /bin/postgres_exporter
 COPY --from=sql-exporter      /bin/sql_exporter      /bin/sql_exporter

+COPY --chown=postgres compute/etc/postgres_exporter.yml /etc/postgres_exporter.yml
+
 COPY --from=sql_exporter_preprocessor --chmod=0644 /home/nonroot/compute/etc/sql_exporter.yml               /etc/sql_exporter.yml
 COPY --from=sql_exporter_preprocessor --chmod=0644 /home/nonroot/compute/etc/neon_collector.yml             /etc/neon_collector.yml
 COPY --from=sql_exporter_preprocessor --chmod=0644 /home/nonroot/compute/etc/sql_exporter_autoscaling.yml   /etc/sql_exporter_autoscaling.yml
--- a/compute/etc/postgres_exporter.yml
+++ b/compute/etc/postgres_exporter.yml
--- a/compute/vm-image-spec-bookworm.yaml
+++ b/compute/vm-image-spec-bookworm.yaml
@@ -26,7 +26,7 @@ commands:
  - name: postgres-exporter
    user: nobody
    sysvInitAction: respawn
-    shell: 'DATA_SOURCE_NAME="user=cloud_admin sslmode=disable dbname=postgres application_name=postgres-exporter" /bin/postgres_exporter'
+    shell: 'DATA_SOURCE_NAME="user=cloud_admin sslmode=disable dbname=postgres application_name=postgres-exporter" /bin/postgres_exporter --config.file=/etc/postgres_exporter.yml'
  - name: sql-exporter
    user: nobody
    sysvInitAction: respawn
--- a/compute/vm-image-spec-bullseye.yaml
+++ b/compute/vm-image-spec-bullseye.yaml
@@ -26,7 +26,7 @@ commands:
  - name: postgres-exporter
    user: nobody
    sysvInitAction: respawn
-    shell: 'DATA_SOURCE_NAME="user=cloud_admin sslmode=disable dbname=postgres application_name=postgres-exporter" /bin/postgres_exporter'
+    shell: 'DATA_SOURCE_NAME="user=cloud_admin sslmode=disable dbname=postgres application_name=postgres-exporter" /bin/postgres_exporter --config.file=/etc/postgres_exporter.yml'
  - name: sql-exporter
    user: nobody
    sysvInitAction: respawn
--- a/compute_tools/src/config.rs
+++ b/compute_tools/src/config.rs
@@ -74,10 +74,17 @@ pub fn write_postgres_conf(
    }

    // Locales
-    writeln!(file, "lc_messages='C.UTF-8'")?;
-    writeln!(file, "lc_monetary='C.UTF-8'")?;
-    writeln!(file, "lc_time='C.UTF-8'")?;
-    writeln!(file, "lc_numeric='C.UTF-8'")?;
+    if cfg!(target_os = "macos") {
+        writeln!(file, "lc_messages='C'")?;
+        writeln!(file, "lc_monetary='C'")?;
+        writeln!(file, "lc_time='C'")?;
+        writeln!(file, "lc_numeric='C'")?;
+    } else {
+        writeln!(file, "lc_messages='C.UTF-8'")?;
+        writeln!(file, "lc_monetary='C.UTF-8'")?;
+        writeln!(file, "lc_time='C.UTF-8'")?;
+        writeln!(file, "lc_numeric='C.UTF-8'")?;
+    }

    match spec.mode {
        ComputeMode::Primary => {}
--- a/docs/rfcs/038-aux-file-v2.md
+++ b/docs/rfcs/038-aux-file-v2.md
@@ -91,7 +91,7 @@ generating the basebackup by scanning the `REPL_ORIGIN_KEY_PREFIX` keyspace.
 There are two places we need to read the aux files from the pageserver:

 * On the write path, when the compute node adds an aux file to the pageserver, we will retrieve the key from the storage, append the file to the hashed key, and write it back. The current `get` API already supports that.
-*  We use the vectored get API to retrieve all aux files during generating the basebackup. Because we need to scan a sparse keyspace, we slightly modified the vectored get path. The vectorized API will attempt to retrieve every single key within the requested key range, and therefore, we modified it in a way that keys within `NON_INHERITED_SPARSE_RANGE` will not trigger missing key error.
+*  We use the vectored get API to retrieve all aux files during generating the basebackup. Because we need to scan a sparse keyspace, we slightly modified the vectored get path. The vectorized API used to always attempt to retrieve every single key within the requested key range, and therefore, we modified it in a way that keys within `NON_INHERITED_SPARSE_RANGE` will not trigger missing key error. Furthermore, as aux file reads usually need all layer files intersecting with that key range within the branch and cover a big keyspace, it incurs large overhead for tracking keyspaces that have not been read. Therefore, for sparse keyspaces, we [do not track](https://github.com/neondatabase/neon/pull/9631) `ummapped_keyspace`.

 ## Compaction and Image Layer Generation

--- a/libs/pageserver_api/src/config.rs
+++ b/libs/pageserver_api/src/config.rs
@@ -277,7 +277,11 @@ pub mod defaults {
    pub const DEFAULT_WAL_REDO_TIMEOUT: &str = "60 s";

    pub const DEFAULT_SUPERUSER: &str = "cloud_admin";
-    pub const DEFAULT_LOCALE: &str = "C.UTF-8";
+    pub const DEFAULT_LOCALE: &str = if cfg!(target_os = "macos") {
+        "C"
+    } else {
+        "C.UTF-8"
+    };

    pub const DEFAULT_PAGE_CACHE_SIZE: usize = 8192;
    pub const DEFAULT_MAX_FILE_DESCRIPTORS: usize = 100;
--- a/libs/utils/src/auth.rs
+++ b/libs/utils/src/auth.rs
@@ -40,6 +40,11 @@ pub enum Scope {
    /// Allows access to storage controller APIs used by the scrubber, to interrogate the state
    /// of a tenant & post scrub results.
    Scrubber,
+
+    /// This scope is used for communication with other storage controller instances.
+    /// At the time of writing, this is only used for the step down request.
+    #[serde(rename = "controller_peer")]
+    ControllerPeer,
 }

 /// JWT payload. See docs/authentication.md for the format
--- a/pageserver/compaction/src/helpers.rs
+++ b/pageserver/compaction/src/helpers.rs
@@ -35,6 +35,15 @@ pub fn overlaps_with<T: Ord>(a: &Range<T>, b: &Range<T>) -> bool {
    !(a.end <= b.start || b.end <= a.start)
 }

+/// Whether a fully contains b, example as below
+/// ```plain
+/// |      a       |
+///       |  b  |
+/// ```
+pub fn fully_contains<T: Ord>(a: &Range<T>, b: &Range<T>) -> bool {
+    a.start <= b.start && a.end >= b.end
+}
+
 pub fn union_to_keyspace<K: Ord>(a: &mut CompactionKeySpace<K>, b: CompactionKeySpace<K>) {
    let x = std::mem::take(a);
    let mut all_ranges_iter = [x.into_iter(), b.into_iter()]
--- a/pageserver/src/auth.rs
+++ b/pageserver/src/auth.rs
@@ -19,7 +19,8 @@ pub fn check_permission(claims: &Claims, tenant_id: Option<TenantId>) -> Result<
            | Scope::SafekeeperData
            | Scope::GenerationsApi
            | Scope::Infra
-            | Scope::Scrubber,
+            | Scope::Scrubber
+            | Scope::ControllerPeer,
            _,
        ) => Err(AuthError(
            format!(
--- a/pageserver/src/tenant.rs
+++ b/pageserver/src/tenant.rs
@@ -4786,12 +4786,6 @@ async fn run_initdb(
        .args(["--username", &conf.superuser])
        .args(["--encoding", "utf8"])
        .args(["--locale", &conf.locale])
-        .args(["--lc-collate", &conf.locale])
-        .args(["--lc-ctype", &conf.locale])
-        .args(["--lc-messages", &conf.locale])
-        .args(["--lc-monetary", &conf.locale])
-        .args(["--lc-numeric", &conf.locale])
-        .args(["--lc-time", &conf.locale])
        .arg("--no-instructions")
        .arg("--no-sync")
        .env_clear()
@@ -9229,6 +9223,23 @@ mod tests {
        Ok(())
    }

+    fn sort_layer_key(k1: &PersistentLayerKey, k2: &PersistentLayerKey) -> std::cmp::Ordering {
+        (
+            k1.is_delta,
+            k1.key_range.start,
+            k1.key_range.end,
+            k1.lsn_range.start,
+            k1.lsn_range.end,
+        )
+            .cmp(&(
+                k2.is_delta,
+                k2.key_range.start,
+                k2.key_range.end,
+                k2.lsn_range.start,
+                k2.lsn_range.end,
+            ))
+    }
+
    async fn inspect_and_sort(
        tline: &Arc<Timeline>,
        filter: Option<std::ops::Range<Key>>,
@@ -9237,25 +9248,30 @@ mod tests {
        if let Some(filter) = filter {
            all_layers.retain(|layer| overlaps_with(&layer.key_range, &filter));
        }
-        all_layers.sort_by(|k1, k2| {
-            (
-                k1.is_delta,
-                k1.key_range.start,
-                k1.key_range.end,
-                k1.lsn_range.start,
-                k1.lsn_range.end,
-            )
-                .cmp(&(
-                    k2.is_delta,
-                    k2.key_range.start,
-                    k2.key_range.end,
-                    k2.lsn_range.start,
-                    k2.lsn_range.end,
-                ))
-        });
+        all_layers.sort_by(sort_layer_key);
        all_layers
    }

+    #[cfg(feature = "testing")]
+    fn check_layer_map_key_eq(
+        mut left: Vec<PersistentLayerKey>,
+        mut right: Vec<PersistentLayerKey>,
+    ) {
+        left.sort_by(sort_layer_key);
+        right.sort_by(sort_layer_key);
+        if left != right {
+            eprintln!("---LEFT---");
+            for left in left.iter() {
+                eprintln!("{}", left);
+            }
+            eprintln!("---RIGHT---");
+            for right in right.iter() {
+                eprintln!("{}", right);
+            }
+            assert_eq!(left, right);
+        }
+    }
+
    #[cfg(feature = "testing")]
    #[tokio::test]
    async fn test_simple_partial_bottom_most_compaction() -> anyhow::Result<()> {
@@ -9348,127 +9364,206 @@ mod tests {

        let cancel = CancellationToken::new();

-        // Do a partial compaction on key range 0..4, we should generate a image layer; no other layers
-        // can be removed because they might be used for other key ranges.
+        // Do a partial compaction on key range 0..2
        tline
-            .partial_compact_with_gc(Some(get_key(0)..get_key(4)), &cancel, EnumSet::new(), &ctx)
+            .partial_compact_with_gc(get_key(0)..get_key(2), &cancel, EnumSet::new(), &ctx)
            .await
            .unwrap();
        let all_layers = inspect_and_sort(&tline, Some(get_key(0)..get_key(10))).await;
-        assert_eq!(
+        check_layer_map_key_eq(
            all_layers,
            vec![
+                // newly-generated image layer for the partial compaction range 0-2
                PersistentLayerKey {
-                    key_range: get_key(0)..get_key(4),
+                    key_range: get_key(0)..get_key(2),
                    lsn_range: Lsn(0x20)..Lsn(0x21),
-                    is_delta: false
+                    is_delta: false,
                },
                PersistentLayerKey {
                    key_range: get_key(0)..get_key(10),
                    lsn_range: Lsn(0x10)..Lsn(0x11),
-                    is_delta: false
+                    is_delta: false,
                },
+                // delta1 is split and the second part is rewritten
                PersistentLayerKey {
-                    key_range: get_key(1)..get_key(4),
+                    key_range: get_key(2)..get_key(4),
                    lsn_range: Lsn(0x20)..Lsn(0x48),
-                    is_delta: true
+                    is_delta: true,
                },
                PersistentLayerKey {
                    key_range: get_key(5)..get_key(7),
                    lsn_range: Lsn(0x20)..Lsn(0x48),
-                    is_delta: true
+                    is_delta: true,
                },
                PersistentLayerKey {
                    key_range: get_key(8)..get_key(10),
                    lsn_range: Lsn(0x48)..Lsn(0x50),
-                    is_delta: true
-                }
-            ]
+                    is_delta: true,
+                },
+            ],
        );

-        // Do a partial compaction on key range 4..10
+        // Do a partial compaction on key range 2..4
        tline
-            .partial_compact_with_gc(Some(get_key(4)..get_key(10)), &cancel, EnumSet::new(), &ctx)
+            .partial_compact_with_gc(get_key(2)..get_key(4), &cancel, EnumSet::new(), &ctx)
            .await
            .unwrap();
        let all_layers = inspect_and_sort(&tline, Some(get_key(0)..get_key(10))).await;
-        assert_eq!(
+        check_layer_map_key_eq(
            all_layers,
            vec![
                PersistentLayerKey {
-                    key_range: get_key(0)..get_key(4),
+                    key_range: get_key(0)..get_key(2),
                    lsn_range: Lsn(0x20)..Lsn(0x21),
-                    is_delta: false
+                    is_delta: false,
                },
                PersistentLayerKey {
-                    // if (in the future) GC kicks in, this layer will be removed
                    key_range: get_key(0)..get_key(10),
                    lsn_range: Lsn(0x10)..Lsn(0x11),
-                    is_delta: false
+                    is_delta: false,
                },
+                // image layer generated for the compaction range 2-4
                PersistentLayerKey {
-                    key_range: get_key(4)..get_key(10),
+                    key_range: get_key(2)..get_key(4),
                    lsn_range: Lsn(0x20)..Lsn(0x21),
-                    is_delta: false
+                    is_delta: false,
                },
+                // we have key2/key3 above the retain_lsn, so we still need this delta layer
                PersistentLayerKey {
-                    key_range: get_key(1)..get_key(4),
+                    key_range: get_key(2)..get_key(4),
                    lsn_range: Lsn(0x20)..Lsn(0x48),
-                    is_delta: true
+                    is_delta: true,
                },
                PersistentLayerKey {
                    key_range: get_key(5)..get_key(7),
                    lsn_range: Lsn(0x20)..Lsn(0x48),
-                    is_delta: true
+                    is_delta: true,
                },
                PersistentLayerKey {
                    key_range: get_key(8)..get_key(10),
                    lsn_range: Lsn(0x48)..Lsn(0x50),
-                    is_delta: true
-                }
-            ]
+                    is_delta: true,
+                },
+            ],
+        );
+
+        // Do a partial compaction on key range 4..9
+        tline
+            .partial_compact_with_gc(get_key(4)..get_key(9), &cancel, EnumSet::new(), &ctx)
+            .await
+            .unwrap();
+        let all_layers = inspect_and_sort(&tline, Some(get_key(0)..get_key(10))).await;
+        check_layer_map_key_eq(
+            all_layers,
+            vec![
+                PersistentLayerKey {
+                    key_range: get_key(0)..get_key(2),
+                    lsn_range: Lsn(0x20)..Lsn(0x21),
+                    is_delta: false,
+                },
+                PersistentLayerKey {
+                    key_range: get_key(0)..get_key(10),
+                    lsn_range: Lsn(0x10)..Lsn(0x11),
+                    is_delta: false,
+                },
+                PersistentLayerKey {
+                    key_range: get_key(2)..get_key(4),
+                    lsn_range: Lsn(0x20)..Lsn(0x21),
+                    is_delta: false,
+                },
+                PersistentLayerKey {
+                    key_range: get_key(2)..get_key(4),
+                    lsn_range: Lsn(0x20)..Lsn(0x48),
+                    is_delta: true,
+                },
+                // image layer generated for this compaction range
+                PersistentLayerKey {
+                    key_range: get_key(4)..get_key(9),
+                    lsn_range: Lsn(0x20)..Lsn(0x21),
+                    is_delta: false,
+                },
+                PersistentLayerKey {
+                    key_range: get_key(8)..get_key(10),
+                    lsn_range: Lsn(0x48)..Lsn(0x50),
+                    is_delta: true,
+                },
+            ],
+        );
+
+        // Do a partial compaction on key range 9..10
+        tline
+            .partial_compact_with_gc(get_key(9)..get_key(10), &cancel, EnumSet::new(), &ctx)
+            .await
+            .unwrap();
+        let all_layers = inspect_and_sort(&tline, Some(get_key(0)..get_key(10))).await;
+        check_layer_map_key_eq(
+            all_layers,
+            vec![
+                PersistentLayerKey {
+                    key_range: get_key(0)..get_key(2),
+                    lsn_range: Lsn(0x20)..Lsn(0x21),
+                    is_delta: false,
+                },
+                PersistentLayerKey {
+                    key_range: get_key(0)..get_key(10),
+                    lsn_range: Lsn(0x10)..Lsn(0x11),
+                    is_delta: false,
+                },
+                PersistentLayerKey {
+                    key_range: get_key(2)..get_key(4),
+                    lsn_range: Lsn(0x20)..Lsn(0x21),
+                    is_delta: false,
+                },
+                PersistentLayerKey {
+                    key_range: get_key(2)..get_key(4),
+                    lsn_range: Lsn(0x20)..Lsn(0x48),
+                    is_delta: true,
+                },
+                PersistentLayerKey {
+                    key_range: get_key(4)..get_key(9),
+                    lsn_range: Lsn(0x20)..Lsn(0x21),
+                    is_delta: false,
+                },
+                // image layer generated for the compaction range
+                PersistentLayerKey {
+                    key_range: get_key(9)..get_key(10),
+                    lsn_range: Lsn(0x20)..Lsn(0x21),
+                    is_delta: false,
+                },
+                PersistentLayerKey {
+                    key_range: get_key(8)..get_key(10),
+                    lsn_range: Lsn(0x48)..Lsn(0x50),
+                    is_delta: true,
+                },
+            ],
        );

        // Do a partial compaction on key range 0..10, all image layers below LSN 20 can be replaced with new ones.
        tline
-            .partial_compact_with_gc(Some(get_key(0)..get_key(10)), &cancel, EnumSet::new(), &ctx)
+            .partial_compact_with_gc(get_key(0)..get_key(10), &cancel, EnumSet::new(), &ctx)
            .await
            .unwrap();
        let all_layers = inspect_and_sort(&tline, Some(get_key(0)..get_key(10))).await;
-        assert_eq!(
+        check_layer_map_key_eq(
            all_layers,
            vec![
-                PersistentLayerKey {
-                    key_range: get_key(0)..get_key(4),
-                    lsn_range: Lsn(0x20)..Lsn(0x21),
-                    is_delta: false
-                },
+                // aha, we removed all unnecessary image/delta layers and got a very clean layer map!
                PersistentLayerKey {
                    key_range: get_key(0)..get_key(10),
                    lsn_range: Lsn(0x20)..Lsn(0x21),
-                    is_delta: false
+                    is_delta: false,
                },
                PersistentLayerKey {
-                    key_range: get_key(4)..get_key(10),
-                    lsn_range: Lsn(0x20)..Lsn(0x21),
-                    is_delta: false
-                },
-                PersistentLayerKey {
-                    key_range: get_key(1)..get_key(4),
+                    key_range: get_key(2)..get_key(4),
                    lsn_range: Lsn(0x20)..Lsn(0x48),
-                    is_delta: true
-                },
-                PersistentLayerKey {
-                    key_range: get_key(5)..get_key(7),
-                    lsn_range: Lsn(0x20)..Lsn(0x48),
-                    is_delta: true
+                    is_delta: true,
                },
                PersistentLayerKey {
                    key_range: get_key(8)..get_key(10),
                    lsn_range: Lsn(0x48)..Lsn(0x50),
-                    is_delta: true
-                }
-            ]
+                    is_delta: true,
+                },
+            ],
        );

        Ok(())
--- a/pageserver/src/tenant/storage_layer/delta_layer.rs
+++ b/pageserver/src/tenant/storage_layer/delta_layer.rs
@@ -653,6 +653,10 @@ impl DeltaLayerWriter {
        })
    }

+    pub fn is_empty(&self) -> bool {
+        self.inner.as_ref().unwrap().num_keys == 0
+    }
+
    ///
    /// Append a key-value pair to the file.
    ///
--- a/pageserver/src/tenant/storage_layer/filter_iterator.rs
+++ b/pageserver/src/tenant/storage_layer/filter_iterator.rs
@@ -1,4 +1,4 @@
-use std::ops::Range;
+use std::{ops::Range, sync::Arc};

 use anyhow::bail;
 use pageserver_api::{
@@ -9,7 +9,10 @@ use utils::lsn::Lsn;

 use pageserver_api::value::Value;

-use super::merge_iterator::MergeIterator;
+use super::{
+    merge_iterator::{MergeIterator, MergeIteratorItem},
+    PersistentLayerKey,
+};

 /// A filter iterator over merge iterators (and can be easily extended to other types of iterators).
 ///
@@ -48,10 +51,10 @@ impl<'a> FilterIterator<'a> {
        })
    }

-    pub async fn next(&mut self) -> anyhow::Result<Option<(Key, Lsn, Value)>> {
-        while let Some(item) = self.inner.next().await? {
+    async fn next_inner<R: MergeIteratorItem>(&mut self) -> anyhow::Result<Option<R>> {
+        while let Some(item) = self.inner.next_inner::<R>().await? {
            while self.current_filter_idx < self.retain_key_filters.len()
-                && item.0 >= self.retain_key_filters[self.current_filter_idx].end
+                && item.key_lsn_value().0 >= self.retain_key_filters[self.current_filter_idx].end
            {
                // [filter region]    [filter region]     [filter region]
                //                                     ^ item
@@ -68,7 +71,7 @@ impl<'a> FilterIterator<'a> {
                //                                                 ^ current filter (nothing)
                return Ok(None);
            }
-            if self.retain_key_filters[self.current_filter_idx].contains(&item.0) {
+            if self.retain_key_filters[self.current_filter_idx].contains(&item.key_lsn_value().0) {
                // [filter region]    [filter region]     [filter region]
                //                                              ^ item
                //                                        ^ current filter
@@ -81,6 +84,16 @@ impl<'a> FilterIterator<'a> {
        }
        Ok(None)
    }
+
+    pub async fn next(&mut self) -> anyhow::Result<Option<(Key, Lsn, Value)>> {
+        self.next_inner().await
+    }
+
+    pub async fn next_with_trace(
+        &mut self,
+    ) -> anyhow::Result<Option<((Key, Lsn, Value), Arc<PersistentLayerKey>)>> {
+        self.next_inner().await
+    }
 }

 #[cfg(test)]
--- a/pageserver/src/tenant/storage_layer/merge_iterator.rs
+++ b/pageserver/src/tenant/storage_layer/merge_iterator.rs
@@ -1,6 +1,7 @@
 use std::{
    cmp::Ordering,
    collections::{binary_heap, BinaryHeap},
+    sync::Arc,
 };

 use anyhow::bail;
@@ -13,10 +14,11 @@ use pageserver_api::value::Value;
 use super::{
    delta_layer::{DeltaLayerInner, DeltaLayerIterator},
    image_layer::{ImageLayerInner, ImageLayerIterator},
+    PersistentLayerDesc, PersistentLayerKey,
 };

 #[derive(Clone, Copy)]
-enum LayerRef<'a> {
+pub(crate) enum LayerRef<'a> {
    Image(&'a ImageLayerInner),
    Delta(&'a DeltaLayerInner),
 }
@@ -62,18 +64,20 @@ impl LayerIterRef<'_> {
 /// 1. Unified iterator for image and delta layers.
 /// 2. `Ord` for use in [`MergeIterator::heap`] (for the k-merge).
 /// 3. Lazy creation of the real delta/image iterator.
-enum IteratorWrapper<'a> {
+pub(crate) enum IteratorWrapper<'a> {
    NotLoaded {
        ctx: &'a RequestContext,
        first_key_lower_bound: (Key, Lsn),
        layer: LayerRef<'a>,
+        source_desc: Arc<PersistentLayerKey>,
    },
    Loaded {
        iter: PeekableLayerIterRef<'a>,
+        source_desc: Arc<PersistentLayerKey>,
    },
 }

-struct PeekableLayerIterRef<'a> {
+pub(crate) struct PeekableLayerIterRef<'a> {
    iter: LayerIterRef<'a>,
    peeked: Option<(Key, Lsn, Value)>, // None == end
 }
@@ -151,6 +155,12 @@ impl<'a> IteratorWrapper<'a> {
            layer: LayerRef::Image(image_layer),
            first_key_lower_bound: (image_layer.key_range().start, image_layer.lsn()),
            ctx,
+            source_desc: PersistentLayerKey {
+                key_range: image_layer.key_range().clone(),
+                lsn_range: PersistentLayerDesc::image_layer_lsn_range(image_layer.lsn()),
+                is_delta: false,
+            }
+            .into(),
        }
    }

@@ -162,12 +172,18 @@ impl<'a> IteratorWrapper<'a> {
            layer: LayerRef::Delta(delta_layer),
            first_key_lower_bound: (delta_layer.key_range().start, delta_layer.lsn_range().start),
            ctx,
+            source_desc: PersistentLayerKey {
+                key_range: delta_layer.key_range().clone(),
+                lsn_range: delta_layer.lsn_range().clone(),
+                is_delta: true,
+            }
+            .into(),
        }
    }

    fn peek_next_key_lsn_value(&self) -> Option<(&Key, Lsn, Option<&Value>)> {
        match self {
-            Self::Loaded { iter } => iter
+            Self::Loaded { iter, .. } => iter
                .peek()
                .as_ref()
                .map(|(key, lsn, val)| (key, *lsn, Some(val))),
@@ -191,6 +207,7 @@ impl<'a> IteratorWrapper<'a> {
            ctx,
            first_key_lower_bound,
            layer,
+            source_desc,
        } = self
        else {
            unreachable!()
@@ -206,7 +223,10 @@ impl<'a> IteratorWrapper<'a> {
                );
            }
        }
-        *self = Self::Loaded { iter };
+        *self = Self::Loaded {
+            iter,
+            source_desc: source_desc.clone(),
+        };
        Ok(())
    }

@@ -220,11 +240,19 @@ impl<'a> IteratorWrapper<'a> {
    /// The public interfaces to use are [`crate::tenant::storage_layer::delta_layer::DeltaLayerIterator`] and
    /// [`crate::tenant::storage_layer::image_layer::ImageLayerIterator`].
    async fn next(&mut self) -> anyhow::Result<Option<(Key, Lsn, Value)>> {
-        let Self::Loaded { iter } = self else {
+        let Self::Loaded { iter, .. } = self else {
            panic!("must load the iterator before using")
        };
        iter.next().await
    }
+
+    /// Get the persistent layer key corresponding to this iterator
+    fn trace_source(&self) -> Arc<PersistentLayerKey> {
+        match self {
+            Self::Loaded { source_desc, .. } => source_desc.clone(),
+            Self::NotLoaded { source_desc, .. } => source_desc.clone(),
+        }
+    }
 }

 /// A merge iterator over delta/image layer iterators.
@@ -242,6 +270,32 @@ pub struct MergeIterator<'a> {
    heap: BinaryHeap<IteratorWrapper<'a>>,
 }

+pub(crate) trait MergeIteratorItem {
+    fn new(item: (Key, Lsn, Value), iterator: &IteratorWrapper<'_>) -> Self;
+
+    fn key_lsn_value(&self) -> &(Key, Lsn, Value);
+}
+
+impl MergeIteratorItem for (Key, Lsn, Value) {
+    fn new(item: (Key, Lsn, Value), _: &IteratorWrapper<'_>) -> Self {
+        item
+    }
+
+    fn key_lsn_value(&self) -> &(Key, Lsn, Value) {
+        self
+    }
+}
+
+impl MergeIteratorItem for ((Key, Lsn, Value), Arc<PersistentLayerKey>) {
+    fn new(item: (Key, Lsn, Value), iter: &IteratorWrapper<'_>) -> Self {
+        (item, iter.trace_source().clone())
+    }
+
+    fn key_lsn_value(&self) -> &(Key, Lsn, Value) {
+        &self.0
+    }
+}
+
 impl<'a> MergeIterator<'a> {
    pub fn create(
        deltas: &[&'a DeltaLayerInner],
@@ -260,7 +314,7 @@ impl<'a> MergeIterator<'a> {
        }
    }

-    pub async fn next(&mut self) -> anyhow::Result<Option<(Key, Lsn, Value)>> {
+    pub(crate) async fn next_inner<R: MergeIteratorItem>(&mut self) -> anyhow::Result<Option<R>> {
        while let Some(mut iter) = self.heap.peek_mut() {
            if !iter.is_loaded() {
                // Once we load the iterator, we can know the real first key-value pair in the iterator.
@@ -275,10 +329,22 @@ impl<'a> MergeIterator<'a> {
                binary_heap::PeekMut::pop(iter);
                continue;
            };
-            return Ok(Some(item));
+            return Ok(Some(R::new(item, &iter)));
        }
        Ok(None)
    }
+
+    /// Get the next key-value pair from the iterator.
+    pub async fn next(&mut self) -> anyhow::Result<Option<(Key, Lsn, Value)>> {
+        self.next_inner().await
+    }
+
+    /// Get the next key-value pair from the iterator, and trace where the key comes from.
+    pub async fn next_with_trace(
+        &mut self,
+    ) -> anyhow::Result<Option<((Key, Lsn, Value), Arc<PersistentLayerKey>)>> {
+        self.next_inner().await
+    }
 }

 #[cfg(test)]
--- a/pageserver/src/tenant/timeline/compaction.rs
+++ b/pageserver/src/tenant/timeline/compaction.rs
@@ -4,7 +4,7 @@
 //!
 //! The old legacy algorithm is implemented directly in `timeline.rs`.

-use std::collections::{BinaryHeap, HashSet};
+use std::collections::{BinaryHeap, HashMap, HashSet};
 use std::ops::{Deref, Range};
 use std::sync::Arc;

@@ -56,7 +56,7 @@ use pageserver_api::value::Value;

 use utils::lsn::Lsn;

-use pageserver_compaction::helpers::overlaps_with;
+use pageserver_compaction::helpers::{fully_contains, overlaps_with};
 use pageserver_compaction::interface::*;

 use super::CompactionError;
@@ -64,6 +64,23 @@ use super::CompactionError;
 /// Maximum number of deltas before generating an image layer in bottom-most compaction.
 const COMPACTION_DELTA_THRESHOLD: usize = 5;

+pub struct GcCompactionJobDescription {
+    /// All layers to read in the compaction job
+    selected_layers: Vec<Layer>,
+    /// GC cutoff of the job
+    gc_cutoff: Lsn,
+    /// LSNs to retain for the job
+    retain_lsns_below_horizon: Vec<Lsn>,
+    /// Maximum layer LSN processed in this compaction
+    max_layer_lsn: Lsn,
+    /// Only compact layers overlapping with this range
+    compaction_key_range: Range<Key>,
+    /// When partial compaction is enabled, these layers need to be rewritten to ensure no overlap.
+    /// This field is here solely for debugging. The field will not be read once the compaction
+    /// description is generated.
+    rewrite_layers: Vec<Arc<PersistentLayerDesc>>,
+}
+
 /// The result of bottom-most compaction for a single key at each LSN.
 #[derive(Debug)]
 #[cfg_attr(test, derive(PartialEq))]
@@ -1722,7 +1739,8 @@ impl Timeline {
        flags: EnumSet<CompactFlags>,
        ctx: &RequestContext,
    ) -> anyhow::Result<()> {
-        self.partial_compact_with_gc(None, cancel, flags, ctx).await
+        self.partial_compact_with_gc(Key::MIN..Key::MAX, cancel, flags, ctx)
+            .await
    }

    /// An experimental compaction building block that combines compaction with garbage collection.
@@ -1732,12 +1750,15 @@ impl Timeline {
    /// layers and image layers, which generates image layers on the gc horizon, drop deltas below gc horizon,
    /// and create delta layers with all deltas >= gc horizon.
    ///
-    /// If `key_range`, it will only compact the keys within the range, aka partial compaction. This functionality
-    /// is not complete yet, and if it is set, only image layers will be generated.
-    ///
+    /// If `key_range` is provided, it will only compact the keys within the range, aka partial compaction.
+    /// Partial compaction will read and process all layers overlapping with the key range, even if it might
+    /// contain extra keys. After the gc-compaction phase completes, delta layers that are not fully contained
+    /// within the key range will be rewritten to ensure they do not overlap with the delta layers. Providing
+    /// Key::MIN..Key..MAX to the function indicates a full compaction, though technically, `Key::MAX` is not
+    /// part of the range.
    pub(crate) async fn partial_compact_with_gc(
        self: &Arc<Self>,
-        compaction_key_range: Option<Range<Key>>,
+        compaction_key_range: Range<Key>,
        cancel: &CancellationToken,
        flags: EnumSet<CompactFlags>,
        ctx: &RequestContext,
@@ -1762,9 +1783,8 @@ impl Timeline {
        .await?;

        let dry_run = flags.contains(CompactFlags::DryRun);
-        let partial_compaction = compaction_key_range.is_some();

-        if let Some(ref compaction_key_range) = compaction_key_range {
+        if compaction_key_range == (Key::MIN..Key::MAX) {
            info!("running enhanced gc bottom-most compaction, dry_run={dry_run}, compaction_key_range={}..{}", compaction_key_range.start, compaction_key_range.end);
        } else {
            info!("running enhanced gc bottom-most compaction, dry_run={dry_run}");
@@ -1780,7 +1800,7 @@ impl Timeline {
        // The layer selection has the following properties:
        // 1. If a layer is in the selection, all layers below it are in the selection.
        // 2. Inferred from (1), for each key in the layer selection, the value can be reconstructed only with the layers in the layer selection.
-        let (layer_selection, gc_cutoff, retain_lsns_below_horizon) = if !partial_compaction {
+        let job_desc = {
            let guard = self.layers.read().await;
            let layers = guard.layer_map()?;
            let gc_info = self.gc_info.read().unwrap();
@@ -1810,9 +1830,21 @@ impl Timeline {
            };
            // Then, pick all the layers that are below the max_layer_lsn. This is to ensure we can pick all single-key
            // layers to compact.
+            let mut rewrite_layers = Vec::new();
            for desc in layers.iter_historic_layers() {
-                if desc.get_lsn_range().end <= max_layer_lsn {
+                if desc.get_lsn_range().end <= max_layer_lsn
+                    && overlaps_with(&desc.get_key_range(), &compaction_key_range)
+                {
+                    // If the layer overlaps with the compaction key range, we need to read it to obtain all keys within the range,
+                    // even if it might contain extra keys
                    selected_layers.push(guard.get_from_desc(&desc));
+                    // If the layer is not fully contained within the key range, we need to rewrite it if it's a delta layer (it's fine
+                    // to overlap image layers)
+                    if desc.is_delta()
+                        && !fully_contains(&compaction_key_range, &desc.get_key_range())
+                    {
+                        rewrite_layers.push(desc);
+                    }
                }
            }
            if selected_layers.is_empty() {
@@ -1820,82 +1852,59 @@ impl Timeline {
                return Ok(());
            }
            retain_lsns_below_horizon.sort();
-            (selected_layers, gc_cutoff, retain_lsns_below_horizon)
-        } else {
-            // In case of partial compaction, we currently only support generating image layers, and therefore,
-            // we pick all layers that are below the lowest retain_lsn and does not intersect with any of the layers.
-            let guard = self.layers.read().await;
-            let layers = guard.layer_map()?;
-            let gc_info = self.gc_info.read().unwrap();
-            let mut min_lsn = gc_info.cutoffs.select_min();
-            for (lsn, _, _) in &gc_info.retain_lsns {
-                if lsn < &min_lsn {
-                    min_lsn = *lsn;
-                }
+            GcCompactionJobDescription {
+                selected_layers,
+                gc_cutoff,
+                retain_lsns_below_horizon,
+                max_layer_lsn,
+                compaction_key_range,
+                rewrite_layers,
            }
-            for lsn in gc_info.leases.keys() {
-                if lsn < &min_lsn {
-                    min_lsn = *lsn;
-                }
-            }
-            let mut selected_layers = Vec::new();
-            drop(gc_info);
-            // |-------| |-------| |-------|
-            // | Delta | | Delta | | Delta | -- min_lsn could be intersecting with the layers
-            // |-------| |-------| |-------| <- we want to pick all the layers below min_lsn, so that
-            // | Delta | | Delta | | Delta |    ...we can remove them after compaction
-            // |-------| |-------| |-------|
-            // Pick all the layers intersect or below the min_lsn, get the largest LSN in the selected layers.
-            let Some(compaction_key_range) = compaction_key_range.as_ref() else {
-                unreachable!()
-            };
-            for desc in layers.iter_historic_layers() {
-                if desc.get_lsn_range().end <= min_lsn
-                    && overlaps_with(&desc.key_range, compaction_key_range)
-                {
-                    selected_layers.push(guard.get_from_desc(&desc));
-                }
-            }
-            if selected_layers.is_empty() {
-                info!("no layers to compact with gc");
-                return Ok(());
-            }
-            (selected_layers, min_lsn, Vec::new())
        };
        let lowest_retain_lsn = if self.ancestor_timeline.is_some() {
-            if partial_compaction {
-                warn!("partial compaction cannot run on child branches (for now)");
-                return Ok(());
-            }
            Lsn(self.ancestor_lsn.0 + 1)
        } else {
-            let res = retain_lsns_below_horizon
+            let res = job_desc
+                .retain_lsns_below_horizon
                .first()
                .copied()
-                .unwrap_or(gc_cutoff);
+                .unwrap_or(job_desc.gc_cutoff);
            if cfg!(debug_assertions) {
                assert_eq!(
                    res,
-                    retain_lsns_below_horizon
+                    job_desc
+                        .retain_lsns_below_horizon
                        .iter()
                        .min()
                        .copied()
-                        .unwrap_or(gc_cutoff)
+                        .unwrap_or(job_desc.gc_cutoff)
                );
            }
            res
        };
        info!(
-            "picked {} layers for compaction with gc_cutoff={} lowest_retain_lsn={}",
-            layer_selection.len(),
-            gc_cutoff,
-            lowest_retain_lsn
+            "picked {} layers for compaction ({} layers need rewriting) with max_layer_lsn={} gc_cutoff={} lowest_retain_lsn={}, key_range={}..{}",
+            job_desc.selected_layers.len(),
+            job_desc.rewrite_layers.len(),
+            job_desc.max_layer_lsn,
+            job_desc.gc_cutoff,
+            lowest_retain_lsn,
+            job_desc.compaction_key_range.start,
+            job_desc.compaction_key_range.end
        );

-        self.check_compaction_space(&layer_selection).await?;
+        for layer in &job_desc.selected_layers {
+            debug!("read layer: {}", layer.layer_desc().key());
+        }
+        for layer in &job_desc.rewrite_layers {
+            debug!("rewrite layer: {}", layer.key());
+        }
+
+        self.check_compaction_space(&job_desc.selected_layers)
+            .await?;

        // Generate statistics for the compaction
-        for layer in &layer_selection {
+        for layer in &job_desc.selected_layers {
            let desc = layer.layer_desc();
            if desc.is_delta() {
                stat.visit_delta_layer(desc.file_size());
@@ -1906,25 +1915,25 @@ impl Timeline {

        // Step 1: construct a k-merge iterator over all layers.
        // Also, verify if the layer map can be split by drawing a horizontal line at every LSN start/end split point.
-        let layer_names: Vec<crate::tenant::storage_layer::LayerName> = layer_selection
+        let layer_names = job_desc
+            .selected_layers
            .iter()
            .map(|layer| layer.layer_desc().layer_name())
            .collect_vec();
        if let Some(err) = check_valid_layermap(&layer_names) {
-            bail!("cannot run gc-compaction because {}", err);
+            warn!("gc-compaction layer map check failed because {}, this is normal if partial compaction is not finished yet", err);
        }
        // The maximum LSN we are processing in this compaction loop
-        let end_lsn = layer_selection
+        let end_lsn = job_desc
+            .selected_layers
            .iter()
            .map(|l| l.layer_desc().lsn_range.end)
            .max()
            .unwrap();
-        // We don't want any of the produced layers to cover the full key range (i.e., MIN..MAX) b/c it will then be recognized
-        // as an L0 layer.
        let mut delta_layers = Vec::new();
        let mut image_layers = Vec::new();
        let mut downloaded_layers = Vec::new();
-        for layer in &layer_selection {
+        for layer in &job_desc.selected_layers {
            let resident_layer = layer.download_and_keep_resident().await?;
            downloaded_layers.push(resident_layer);
        }
@@ -1943,8 +1952,8 @@ impl Timeline {
            dense_ks,
            sparse_ks,
        )?;
-        // Step 2: Produce images+deltas. TODO: ensure newly-produced delta does not overlap with other deltas.
-        // Data of the same key.
+
+        // Step 2: Produce images+deltas.
        let mut accumulated_values = Vec::new();
        let mut last_key: Option<Key> = None;

@@ -1956,10 +1965,7 @@ impl Timeline {
                    self.conf,
                    self.timeline_id,
                    self.tenant_shard_id,
-                    compaction_key_range
-                        .as_ref()
-                        .map(|x| x.start)
-                        .unwrap_or(Key::MIN),
+                    job_desc.compaction_key_range.start,
                    lowest_retain_lsn,
                    self.get_compaction_target_size(),
                    ctx,
@@ -1979,6 +1985,13 @@ impl Timeline {
        )
        .await?;

+        #[derive(Default)]
+        struct RewritingLayers {
+            before: Option<DeltaLayerWriter>,
+            after: Option<DeltaLayerWriter>,
+        }
+        let mut delta_layer_rewriters = HashMap::<Arc<PersistentLayerKey>, RewritingLayers>::new();
+
        /// Returns None if there is no ancestor branch. Throw an error when the key is not found.
        ///
        /// Currently, we always get the ancestor image for each key in the child branch no matter whether the image
@@ -2004,10 +2017,51 @@ impl Timeline {
        // the key and LSN range are determined. However, to keep things simple here, we still
        // create this writer, and discard the writer in the end.

-        while let Some((key, lsn, val)) = merge_iter.next().await? {
+        while let Some(((key, lsn, val), desc)) = merge_iter.next_with_trace().await? {
            if cancel.is_cancelled() {
                return Err(anyhow!("cancelled")); // TODO: refactor to CompactionError and pass cancel error
            }
+            if !job_desc.compaction_key_range.contains(&key) {
+                if !desc.is_delta {
+                    continue;
+                }
+                let rewriter = delta_layer_rewriters.entry(desc.clone()).or_default();
+                let rewriter = if key < job_desc.compaction_key_range.start {
+                    if rewriter.before.is_none() {
+                        rewriter.before = Some(
+                            DeltaLayerWriter::new(
+                                self.conf,
+                                self.timeline_id,
+                                self.tenant_shard_id,
+                                desc.key_range.start,
+                                desc.lsn_range.clone(),
+                                ctx,
+                            )
+                            .await?,
+                        );
+                    }
+                    rewriter.before.as_mut().unwrap()
+                } else if key >= job_desc.compaction_key_range.end {
+                    if rewriter.after.is_none() {
+                        rewriter.after = Some(
+                            DeltaLayerWriter::new(
+                                self.conf,
+                                self.timeline_id,
+                                self.tenant_shard_id,
+                                job_desc.compaction_key_range.end,
+                                desc.lsn_range.clone(),
+                                ctx,
+                            )
+                            .await?,
+                        );
+                    }
+                    rewriter.after.as_mut().unwrap()
+                } else {
+                    unreachable!()
+                };
+                rewriter.put_value(key, lsn, val, ctx).await?;
+                continue;
+            }
            match val {
                Value::Image(_) => stat.visit_image_key(&val),
                Value::WalRecord(_) => stat.visit_wal_key(&val),
@@ -2018,35 +2072,27 @@ impl Timeline {
                }
                accumulated_values.push((key, lsn, val));
            } else {
-                let last_key = last_key.as_mut().unwrap();
-                stat.on_unique_key_visited();
-                let skip_adding_key = if let Some(ref compaction_key_range) = compaction_key_range {
-                    !compaction_key_range.contains(last_key)
-                } else {
-                    false
-                };
-                if !skip_adding_key {
-                    let retention = self
-                        .generate_key_retention(
-                            *last_key,
-                            &accumulated_values,
-                            gc_cutoff,
-                            &retain_lsns_below_horizon,
-                            COMPACTION_DELTA_THRESHOLD,
-                            get_ancestor_image(self, *last_key, ctx).await?,
-                        )
-                        .await?;
-                    // Put the image into the image layer. Currently we have a single big layer for the compaction.
-                    retention
-                        .pipe_to(
-                            *last_key,
-                            &mut delta_layer_writer,
-                            image_layer_writer.as_mut(),
-                            &mut stat,
-                            ctx,
-                        )
-                        .await?;
-                }
+                let last_key: &mut Key = last_key.as_mut().unwrap();
+                stat.on_unique_key_visited(); // TODO: adjust statistics for partial compaction
+                let retention = self
+                    .generate_key_retention(
+                        *last_key,
+                        &accumulated_values,
+                        job_desc.gc_cutoff,
+                        &job_desc.retain_lsns_below_horizon,
+                        COMPACTION_DELTA_THRESHOLD,
+                        get_ancestor_image(self, *last_key, ctx).await?,
+                    )
+                    .await?;
+                retention
+                    .pipe_to(
+                        *last_key,
+                        &mut delta_layer_writer,
+                        image_layer_writer.as_mut(),
+                        &mut stat,
+                        ctx,
+                    )
+                    .await?;
                accumulated_values.clear();
                *last_key = key;
                accumulated_values.push((key, lsn, val));
@@ -2057,35 +2103,43 @@ impl Timeline {
        let last_key = last_key.expect("no keys produced during compaction");
        stat.on_unique_key_visited();

-        let skip_adding_key = if let Some(ref compaction_key_range) = compaction_key_range {
-            !compaction_key_range.contains(&last_key)
-        } else {
-            false
-        };
-        if !skip_adding_key {
-            let retention = self
-                .generate_key_retention(
-                    last_key,
-                    &accumulated_values,
-                    gc_cutoff,
-                    &retain_lsns_below_horizon,
-                    COMPACTION_DELTA_THRESHOLD,
-                    get_ancestor_image(self, last_key, ctx).await?,
-                )
-                .await?;
-            // Put the image into the image layer. Currently we have a single big layer for the compaction.
-            retention
-                .pipe_to(
-                    last_key,
-                    &mut delta_layer_writer,
-                    image_layer_writer.as_mut(),
-                    &mut stat,
-                    ctx,
-                )
-                .await?;
-        }
+        let retention = self
+            .generate_key_retention(
+                last_key,
+                &accumulated_values,
+                job_desc.gc_cutoff,
+                &job_desc.retain_lsns_below_horizon,
+                COMPACTION_DELTA_THRESHOLD,
+                get_ancestor_image(self, last_key, ctx).await?,
+            )
+            .await?;
+        retention
+            .pipe_to(
+                last_key,
+                &mut delta_layer_writer,
+                image_layer_writer.as_mut(),
+                &mut stat,
+                ctx,
+            )
+            .await?;
        // end: move the above part to the loop body

+        let mut rewrote_delta_layers = Vec::new();
+        for (key, writers) in delta_layer_rewriters {
+            if let Some(delta_writer_before) = writers.before {
+                let (desc, path) = delta_writer_before
+                    .finish(job_desc.compaction_key_range.start, ctx)
+                    .await?;
+                let layer = Layer::finish_creating(self.conf, self, desc, &path)?;
+                rewrote_delta_layers.push(layer);
+            }
+            if let Some(delta_writer_after) = writers.after {
+                let (desc, path) = delta_writer_after.finish(key.key_range.end, ctx).await?;
+                let layer = Layer::finish_creating(self.conf, self, desc, &path)?;
+                rewrote_delta_layers.push(layer);
+            }
+        }
+
        let discard = |key: &PersistentLayerKey| {
            let key = key.clone();
            async move { KeyHistoryRetention::discard_key(&key, self, dry_run).await }
@@ -2093,10 +2147,7 @@ impl Timeline {

        let produced_image_layers = if let Some(writer) = image_layer_writer {
            if !dry_run {
-                let end_key = compaction_key_range
-                    .as_ref()
-                    .map(|x| x.end)
-                    .unwrap_or(Key::MAX);
+                let end_key = job_desc.compaction_key_range.end;
                writer
                    .finish_with_discard_fn(self, ctx, end_key, discard)
                    .await?
@@ -2117,10 +2168,8 @@ impl Timeline {
            Vec::new()
        };

-        if partial_compaction && !produced_delta_layers.is_empty() {
-            bail!("implementation error: partial compaction should not be producing delta layers (for now)");
-        }
-
+        // TODO: make image/delta/rewrote_delta layers generation atomic. At this point, we already generated resident layers, and if
+        // compaction is cancelled at this point, we might have some layers that are not cleaned up.
        let mut compact_to = Vec::new();
        let mut keep_layers = HashSet::new();
        let produced_delta_layers_len = produced_delta_layers.len();
@@ -2128,52 +2177,84 @@ impl Timeline {
        for action in produced_delta_layers {
            match action {
                BatchWriterResult::Produced(layer) => {
+                    if cfg!(debug_assertions) {
+                        info!("produced delta layer: {}", layer.layer_desc().key());
+                    }
                    stat.produce_delta_layer(layer.layer_desc().file_size());
                    compact_to.push(layer);
                }
                BatchWriterResult::Discarded(l) => {
+                    if cfg!(debug_assertions) {
+                        info!("discarded delta layer: {}", l);
+                    }
                    keep_layers.insert(l);
                    stat.discard_delta_layer();
                }
            }
        }
+        for layer in &rewrote_delta_layers {
+            debug!(
+                "produced rewritten delta layer: {}",
+                layer.layer_desc().key()
+            );
+        }
+        compact_to.extend(rewrote_delta_layers);
        for action in produced_image_layers {
            match action {
                BatchWriterResult::Produced(layer) => {
+                    debug!("produced image layer: {}", layer.layer_desc().key());
                    stat.produce_image_layer(layer.layer_desc().file_size());
                    compact_to.push(layer);
                }
                BatchWriterResult::Discarded(l) => {
+                    debug!("discarded image layer: {}", l);
                    keep_layers.insert(l);
                    stat.discard_image_layer();
                }
            }
        }
-        let mut layer_selection = layer_selection;
-        layer_selection.retain(|x| !keep_layers.contains(&x.layer_desc().key()));
-        if let Some(ref compaction_key_range) = compaction_key_range {
-            // Partial compaction might select more data than it processes, e.g., if
-            // the compaction_key_range only partially overlaps:
-            //
-            //         [---compaction_key_range---]
-            //   [---A----][----B----][----C----][----D----]
-            //
-            // A,B,C,D are all in the `layer_selection`. The created image layers contain
-            // whatever is needed from B, C, and from `----]` of A, and from  `[--` of D.
-            //
-            // In contrast, `[--A-` and `--D----]` have not been processed, so, we must
-            // keep that data.
-            //
-            // The solution for now is to keep A and D completely.
-            // (layer_selection is what we'll remove from the layer map, so,
-            //  retain what is _not_ fully covered by compaction_key_range).
-            layer_selection.retain(|x| {
-                let key_range = &x.layer_desc().key_range;
-                key_range.start >= compaction_key_range.start
-                    && key_range.end <= compaction_key_range.end
-            });
+
+        let mut layer_selection = job_desc.selected_layers;
+
+        // Partial compaction might select more data than it processes, e.g., if
+        // the compaction_key_range only partially overlaps:
+        //
+        //         [---compaction_key_range---]
+        //   [---A----][----B----][----C----][----D----]
+        //
+        // For delta layers, we will rewrite the layers so that it is cut exactly at
+        // the compaction key range, so we can always discard them. However, for image
+        // layers, as we do not rewrite them for now, we need to handle them differently.
+        // Assume image layers  A, B, C, D are all in the `layer_selection`.
+        //
+        // The created image layers contain whatever is needed from B, C, and from
+        // `----]` of A, and from  `[---` of D.
+        //
+        // In contrast, `[---A` and `D----]` have not been processed, so, we must
+        // keep that data.
+        //
+        // The solution for now is to keep A and D completely if they are image layers.
+        // (layer_selection is what we'll remove from the layer map, so, retain what
+        // is _not_ fully covered by compaction_key_range).
+        for layer in &layer_selection {
+            if !layer.layer_desc().is_delta() {
+                if !overlaps_with(
+                    &layer.layer_desc().key_range,
+                    &job_desc.compaction_key_range,
+                ) {
+                    bail!("violated constraint: image layer outside of compaction key range");
+                }
+                if !fully_contains(
+                    &job_desc.compaction_key_range,
+                    &layer.layer_desc().key_range,
+                ) {
+                    keep_layers.insert(layer.layer_desc().key());
+                }
+            }
        }

+        layer_selection.retain(|x| !keep_layers.contains(&x.layer_desc().key()));
+
        info!(
            "gc-compaction statistics: {}",
            serde_json::to_string(&stat)?
@@ -2192,6 +2273,7 @@ impl Timeline {

        // Step 3: Place back to the layer map.
        {
+            // TODO: sanity check if the layer map is valid (i.e., should not have overlaps)
            let mut guard = self.layers.write().await;
            guard
                .open_mut()?
--- a/safekeeper/src/auth.rs
+++ b/safekeeper/src/auth.rs
@@ -20,7 +20,8 @@ pub fn check_permission(claims: &Claims, tenant_id: Option<TenantId>) -> Result<
            | Scope::PageServerApi
            | Scope::GenerationsApi
            | Scope::Infra
-            | Scope::Scrubber,
+            | Scope::Scrubber
+            | Scope::ControllerPeer,
            _,
        ) => Err(AuthError(
            format!(
--- a/safekeeper/src/metrics.rs
+++ b/safekeeper/src/metrics.rs
@@ -55,7 +55,7 @@ pub static WRITE_WAL_SECONDS: Lazy<Histogram> = Lazy::new(|| {
 pub static FLUSH_WAL_SECONDS: Lazy<Histogram> = Lazy::new(|| {
    register_histogram!(
        "safekeeper_flush_wal_seconds",
-        "Seconds spent syncing WAL to a disk",
+        "Seconds spent syncing WAL to a disk (excluding segment initialization)",
        DISK_FSYNC_SECONDS_BUCKETS.to_vec()
    )
    .expect("Failed to register safekeeper_flush_wal_seconds histogram")
--- a/safekeeper/src/wal_storage.rs
+++ b/safekeeper/src/wal_storage.rs
@@ -257,6 +257,9 @@ impl PhysicalStorage {
            // Try to open existing partial file
            Ok((file, true))
        } else {
+            let _timer = WAL_STORAGE_OPERATION_SECONDS
+                .with_label_values(&["initialize_segment"])
+                .start_timer();
            // Create and fill new partial file
            //
            // We're using fdatasync during WAL writing, so file size must not
@@ -274,8 +277,6 @@ impl PhysicalStorage {
            });
            file.set_len(self.wal_seg_size as u64).await?;

-            // Note: this doesn't get into observe_flush_seconds metric. But
-            // segment init should be separate metric, if any.
            if let Err(e) = durable_rename(&tmp_path, &wal_file_partial_path, !self.no_sync).await {
                // Probably rename succeeded, but fsync of it failed. Remove
                // the file then to avoid using it.
--- a/storage_controller/src/http.rs
+++ b/storage_controller/src/http.rs
@@ -1033,7 +1033,7 @@ async fn handle_update_preferred_azs(req: Request<Body>) -> Result<Response<Body
 }

 async fn handle_step_down(req: Request<Body>) -> Result<Response<Body>, ApiError> {
-    check_permissions(&req, Scope::Admin)?;
+    check_permissions(&req, Scope::ControllerPeer)?;

    let req = match maybe_forward(req).await {
        ForwardOutcome::Forwarded(res) => {
--- a/test_runner/fixtures/neon_fixtures.py
+++ b/test_runner/fixtures/neon_fixtures.py
@@ -286,7 +286,7 @@ class PgProtocol:
        return self.safe_psql_many([query], **kwargs)[0]

    def safe_psql_many(
-        self, queries: Iterable[str], log_query=True, **kwargs: Any
+        self, queries: Iterable[str], log_query: bool = True, **kwargs: Any
    ) -> list[list[tuple[Any, ...]]]:
        """
        Execute queries against the node and return all rows.
@@ -306,7 +306,7 @@ class PgProtocol:
                        result.append(cur.fetchall())
        return result

-    def safe_psql_scalar(self, query, log_query=True) -> Any:
+    def safe_psql_scalar(self, query: str, log_query: bool = True) -> Any:
        """
        Execute query returning single row with single column.
        """
--- a/test_runner/fixtures/pg_version.py
+++ b/test_runner/fixtures/pg_version.py
@@ -1,10 +1,8 @@
 from __future__ import annotations

 import enum
-import os
 from typing import TYPE_CHECKING

-import pytest
 from typing_extensions import override

 if TYPE_CHECKING:
@@ -18,12 +16,15 @@ This fixture is used to determine which version of Postgres to use for tests.

 # Inherit PgVersion from str rather than int to make it easier to pass as a command-line argument
 # TODO: use enum.StrEnum for Python >= 3.11
-@enum.unique
 class PgVersion(str, enum.Enum):
    V14 = "14"
    V15 = "15"
    V16 = "16"
    V17 = "17"
+
+    # Default Postgres Version for tests that don't really depend on Postgres itself
+    DEFAULT = V16
+
    # Instead of making version an optional parameter in methods, we can use this fake entry
    # to explicitly rely on the default server version (could be different from pg_version fixture value)
    NOT_SET = "<-POSTRGRES VERSION IS NOT SET->"
@@ -59,27 +60,3 @@ class PgVersion(str, enum.Enum):
        # Make mypy happy
        # See https://github.com/python/mypy/issues/3974
        return None
-
-
-DEFAULT_VERSION: PgVersion = PgVersion.V16
-
-
-def skip_on_postgres(version: PgVersion, reason: str):
-    return pytest.mark.skipif(
-        PgVersion(os.environ.get("DEFAULT_PG_VERSION", DEFAULT_VERSION)) is version,
-        reason=reason,
-    )
-
-
-def xfail_on_postgres(version: PgVersion, reason: str):
-    return pytest.mark.xfail(
-        PgVersion(os.environ.get("DEFAULT_PG_VERSION", DEFAULT_VERSION)) is version,
-        reason=reason,
-    )
-
-
-def run_only_on_default_postgres(reason: str):
-    return pytest.mark.skipif(
-        PgVersion(os.environ.get("DEFAULT_PG_VERSION", DEFAULT_VERSION)) is not DEFAULT_VERSION,
-        reason=reason,
-    )
--- a/test_runner/fixtures/utils.py
+++ b/test_runner/fixtures/utils.py
@@ -25,6 +25,7 @@ from fixtures.pageserver.common_types import (
    parse_delta_layer,
    parse_image_layer,
 )
+from fixtures.pg_version import PgVersion

 if TYPE_CHECKING:
    from collections.abc import Iterable
@@ -37,6 +38,7 @@ if TYPE_CHECKING:


 Fn = TypeVar("Fn", bound=Callable[..., Any])
+
 COMPONENT_BINARIES = {
    "storage_controller": ("storage_controller",),
    "storage_broker": ("storage_broker",),
@@ -519,7 +521,7 @@ def assert_pageserver_backups_equal(left: Path, right: Path, skip_files: set[str
    This is essentially:

    lines=$(comm -3 \
-        <(mkdir left && cd left && tar xf "$left" && find . -type f -print0 | xargs sha256sum | sort -k2) \
+        <(mkdir left  && cd left  && tar xf "$left"  && find . -type f -print0 | xargs sha256sum | sort -k2) \
        <(mkdir right && cd right && tar xf "$right" && find . -type f -print0 | xargs sha256sum | sort -k2) \
        | wc -l)
    [ "$lines" = "0" ]
@@ -643,3 +645,40 @@ def allpairs_versions():
        )
        ids.append(f"combination_{''.join(cur_id)}")
    return {"argnames": "combination", "argvalues": tuple(argvalues), "ids": ids}
+
+
+def skip_on_postgres(version: PgVersion, reason: str):
+    return pytest.mark.skipif(
+        PgVersion(os.getenv("DEFAULT_PG_VERSION", PgVersion.DEFAULT)) is version,
+        reason=reason,
+    )
+
+
+def xfail_on_postgres(version: PgVersion, reason: str):
+    return pytest.mark.xfail(
+        PgVersion(os.getenv("DEFAULT_PG_VERSION", PgVersion.DEFAULT)) is version,
+        reason=reason,
+    )
+
+
+def run_only_on_default_postgres(reason: str):
+    return pytest.mark.skipif(
+        PgVersion(os.getenv("DEFAULT_PG_VERSION", PgVersion.DEFAULT)) is not PgVersion.DEFAULT,
+        reason=reason,
+    )
+
+
+def skip_in_debug_build(reason: str):
+    return pytest.mark.skipif(
+        os.getenv("BUILD_TYPE", "debug") == "debug",
+        reason=reason,
+    )
+
+
+def skip_on_ci(reason: str):
+    # `CI` variable is always set to `true` on GitHub
+    # Ref: https://docs.github.com/en/actions/writing-workflows/choosing-what-your-workflow-does/store-information-in-variables#default-environment-variables
+    return pytest.mark.skipif(
+        os.getenv("CI", "false") == "true",
+        reason=reason,
+    )
--- a/test_runner/performance/pageserver/pagebench/test_pageserver_max_throughput_getpage_at_latest_lsn.py
+++ b/test_runner/performance/pageserver/pagebench/test_pageserver_max_throughput_getpage_at_latest_lsn.py
@@ -1,7 +1,6 @@
 from __future__ import annotations

 import json
-import os
 from pathlib import Path
 from typing import TYPE_CHECKING

@@ -14,7 +13,7 @@ from fixtures.neon_fixtures import (
    PgBin,
    wait_for_last_flush_lsn,
 )
-from fixtures.utils import get_scale_for_db, humantime_to_ms
+from fixtures.utils import get_scale_for_db, humantime_to_ms, skip_on_ci

 from performance.pageserver.util import (
    setup_pageserver_with_tenants,
@@ -38,9 +37,8 @@ if TYPE_CHECKING:
@pytest.mark.parametrize("pgbench_scale", [get_scale_for_db(200)])
@pytest.mark.parametrize("n_tenants", [500])
@pytest.mark.timeout(10000)
-@pytest.mark.skipif(
-    os.getenv("CI", "false") == "true",
-    reason="This test needs lot of resources and should run on dedicated HW, not in github action runners as part of CI",
+@skip_on_ci(
+    "This test needs lot of resources and should run on dedicated HW, not in github action runners as part of CI"
 )
 def test_pageserver_characterize_throughput_with_n_tenants(
    neon_env_builder: NeonEnvBuilder,
@@ -66,9 +64,8 @@ def test_pageserver_characterize_throughput_with_n_tenants(
@pytest.mark.parametrize("n_clients", [1, 64])
@pytest.mark.parametrize("n_tenants", [1])
@pytest.mark.timeout(2400)
-@pytest.mark.skipif(
-    os.getenv("CI", "false") == "true",
-    reason="This test needs lot of resources and should run on dedicated HW, not in github action runners as part of CI",
+@skip_on_ci(
+    "This test needs lot of resources and should run on dedicated HW, not in github action runners as part of CI"
 )
 def test_pageserver_characterize_latencies_with_1_client_and_throughput_with_many_clients_one_tenant(
    neon_env_builder: NeonEnvBuilder,
--- a/test_runner/regress/test_branch_and_gc.py
+++ b/test_runner/regress/test_branch_and_gc.py
@@ -8,7 +8,7 @@ from fixtures.common_types import Lsn, TimelineId
 from fixtures.log_helper import log
 from fixtures.neon_fixtures import NeonEnv
 from fixtures.pageserver.http import TimelineCreate406
-from fixtures.utils import query_scalar
+from fixtures.utils import query_scalar, skip_in_debug_build


 # Test the GC implementation when running with branching.
@@ -48,10 +48,8 @@ from fixtures.utils import query_scalar
 # Because the delta layer D covering lsn1 is corrupted, creating a branch
 # starting from lsn1 should return an error as follows:
 #     could not find data for key ... at LSN ..., for request at LSN ...
-def test_branch_and_gc(neon_simple_env: NeonEnv, build_type: str):
-    if build_type == "debug":
-        pytest.skip("times out in debug builds")
-
+@skip_in_debug_build("times out in debug builds")
+def test_branch_and_gc(neon_simple_env: NeonEnv):
    env = neon_simple_env
    pageserver_http_client = env.pageserver.http_client()

--- a/test_runner/regress/test_compaction.py
+++ b/test_runner/regress/test_compaction.py
@@ -2,7 +2,6 @@ from __future__ import annotations

 import enum
 import json
-import os
 import time
 from typing import TYPE_CHECKING

@@ -13,7 +12,7 @@ from fixtures.neon_fixtures import (
    generate_uploads_and_deletions,
 )
 from fixtures.pageserver.http import PageserverApiException
-from fixtures.utils import wait_until
+from fixtures.utils import skip_in_debug_build, wait_until
 from fixtures.workload import Workload

 if TYPE_CHECKING:
@@ -32,7 +31,7 @@ AGGRESIVE_COMPACTION_TENANT_CONF = {
 }


-@pytest.mark.skipif(os.environ.get("BUILD_TYPE") == "debug", reason="only run with release build")
+@skip_in_debug_build("only run with release build")
 def test_pageserver_compaction_smoke(neon_env_builder: NeonEnvBuilder):
    """
    This is a smoke test that compaction kicks in. The workload repeatedly churns
--- a/test_runner/regress/test_download_extensions.py
+++ b/test_runner/regress/test_download_extensions.py
@@ -12,6 +12,7 @@ from fixtures.neon_fixtures import (
    NeonEnvBuilder,
 )
 from fixtures.pg_version import PgVersion
+from fixtures.utils import skip_on_postgres
 from pytest_httpserver import HTTPServer
 from werkzeug.wrappers.request import Request
 from werkzeug.wrappers.response import Response
@@ -41,17 +42,14 @@ def neon_env_builder_local(
    return neon_env_builder


+@skip_on_postgres(PgVersion.V16, reason="TODO: PG16 extension building")
+@skip_on_postgres(PgVersion.V17, reason="TODO: PG17 extension building")
 def test_remote_extensions(
    httpserver: HTTPServer,
    neon_env_builder_local: NeonEnvBuilder,
    httpserver_listen_address,
    pg_version,
 ):
-    if pg_version == PgVersion.V16:
-        pytest.skip("TODO: PG16 extension building")
-    if pg_version == PgVersion.V17:
-        pytest.skip("TODO: PG17 extension building")
-
    # setup mock http server
    # that expects request for anon.tar.zst
    # and returns the requested file
--- a/test_runner/regress/test_ingestion_layer_size.py
+++ b/test_runner/regress/test_ingestion_layer_size.py
@@ -4,25 +4,22 @@ from collections.abc import Iterable
 from dataclasses import dataclass
 from typing import TYPE_CHECKING

-import pytest
 from fixtures.log_helper import log
 from fixtures.neon_fixtures import NeonEnvBuilder, wait_for_last_flush_lsn
 from fixtures.pageserver.http import HistoricLayerInfo, LayerMapInfo
-from fixtures.utils import human_bytes
+from fixtures.utils import human_bytes, skip_in_debug_build

 if TYPE_CHECKING:
    from typing import Union


-def test_ingesting_large_batches_of_images(neon_env_builder: NeonEnvBuilder, build_type: str):
+@skip_in_debug_build("debug run is unnecessarily slow")
+def test_ingesting_large_batches_of_images(neon_env_builder: NeonEnvBuilder):
    """
    Build a non-small GIN index which includes similarly batched up images in WAL stream as does pgvector
    to show that we no longer create oversized layers.
    """

-    if build_type == "debug":
-        pytest.skip("debug run is unnecessarily slow")
-
    minimum_initdb_size = 20 * 1024**2
    checkpoint_distance = 32 * 1024**2
    minimum_good_layer_size = checkpoint_distance * 0.9
--- a/test_runner/regress/test_layer_bloating.py
+++ b/test_runner/regress/test_layer_bloating.py
@@ -2,7 +2,6 @@ from __future__ import annotations

 import os

-import pytest
 from fixtures.log_helper import log
 from fixtures.neon_fixtures import (
    NeonEnvBuilder,
@@ -10,12 +9,18 @@ from fixtures.neon_fixtures import (
    wait_for_last_flush_lsn,
 )
 from fixtures.pg_version import PgVersion
+from fixtures.utils import skip_on_postgres


+@skip_on_postgres(
+    PgVersion.V14,
+    reason="pg_log_standby_snapshot() function is available since Postgres 16",
+)
+@skip_on_postgres(
+    PgVersion.V15,
+    reason="pg_log_standby_snapshot() function is available since Postgres 16",
+)
 def test_layer_bloating(neon_env_builder: NeonEnvBuilder, vanilla_pg):
-    if neon_env_builder.pg_version != PgVersion.V16:
-        pytest.skip("pg_log_standby_snapshot() function is available only in PG16")
-
    env = neon_env_builder.init_start(
        initial_tenant_conf={
            "gc_period": "0s",
--- a/test_runner/regress/test_layer_eviction.py
+++ b/test_runner/regress/test_layer_eviction.py
@@ -2,7 +2,6 @@ from __future__ import annotations

 import time

-import pytest
 from fixtures.log_helper import log
 from fixtures.neon_fixtures import (
    NeonEnvBuilder,
@@ -12,17 +11,13 @@ from fixtures.neon_fixtures import (
 from fixtures.pageserver.common_types import parse_layer_file_name
 from fixtures.pageserver.utils import wait_for_upload
 from fixtures.remote_storage import RemoteStorageKind
+from fixtures.utils import skip_in_debug_build


 # Crates a few layers, ensures that we can evict them (removing locally but keeping track of them anyway)
 # and then download them back.
-def test_basic_eviction(
-    neon_env_builder: NeonEnvBuilder,
-    build_type: str,
-):
-    if build_type == "debug":
-        pytest.skip("times out in debug builds")
-
+@skip_in_debug_build("times out in debug builds")
+def test_basic_eviction(neon_env_builder: NeonEnvBuilder):
    neon_env_builder.enable_pageserver_remote_storage(RemoteStorageKind.LOCAL_FS)

    env = neon_env_builder.init_start(
--- a/test_runner/regress/test_logging.py
+++ b/test_runner/regress/test_logging.py
@@ -5,8 +5,7 @@ import uuid
 import pytest
 from fixtures.log_helper import log
 from fixtures.neon_fixtures import NeonEnvBuilder
-from fixtures.pg_version import run_only_on_default_postgres
-from fixtures.utils import wait_until
+from fixtures.utils import run_only_on_default_postgres, wait_until


@pytest.mark.parametrize("level", ["trace", "debug", "info", "warn", "error"])
--- a/test_runner/regress/test_logical_replication.py
+++ b/test_runner/regress/test_logical_replication.py
@@ -4,24 +4,31 @@ import time
 from functools import partial
 from random import choice
 from string import ascii_lowercase
+from typing import TYPE_CHECKING, cast

-from fixtures.common_types import Lsn
+from fixtures.common_types import Lsn, TenantId, TimelineId
 from fixtures.log_helper import log
 from fixtures.neon_fixtures import (
-    NeonEnv,
-    NeonEnvBuilder,
-    PgProtocol,
    logical_replication_sync,
    wait_for_last_flush_lsn,
 )
 from fixtures.utils import wait_until

+if TYPE_CHECKING:
+    from fixtures.neon_fixtures import (
+        Endpoint,
+        NeonEnv,
+        NeonEnvBuilder,
+        PgProtocol,
+        VanillaPostgres,
+    )
+

 def random_string(n: int):
    return "".join([choice(ascii_lowercase) for _ in range(n)])


-def test_logical_replication(neon_simple_env: NeonEnv, vanilla_pg):
+def test_logical_replication(neon_simple_env: NeonEnv, vanilla_pg: VanillaPostgres):
    env = neon_simple_env

    tenant_id = env.initial_tenant
@@ -160,10 +167,10 @@ COMMIT;


 # Test that neon.logical_replication_max_snap_files works
-def test_obsolete_slot_drop(neon_simple_env: NeonEnv, vanilla_pg):
-    def slot_removed(ep):
+def test_obsolete_slot_drop(neon_simple_env: NeonEnv, vanilla_pg: VanillaPostgres):
+    def slot_removed(ep: Endpoint):
        assert (
-            endpoint.safe_psql(
+            ep.safe_psql(
                "select count(*) from pg_replication_slots where slot_name = 'stale_slot'"
            )[0][0]
            == 0
@@ -254,7 +261,7 @@ FROM generate_series(1, 16384) AS seq; -- Inserts enough rows to exceed 16MB of


 # Tests that walsender correctly blocks until WAL is downloaded from safekeepers
-def test_lr_with_slow_safekeeper(neon_env_builder: NeonEnvBuilder, vanilla_pg):
+def test_lr_with_slow_safekeeper(neon_env_builder: NeonEnvBuilder, vanilla_pg: VanillaPostgres):
    neon_env_builder.num_safekeepers = 3
    env = neon_env_builder.init_start()

@@ -336,13 +343,13 @@ FROM generate_series(1, 16384) AS seq; -- Inserts enough rows to exceed 16MB of
 #
 # Most pages start with a contrecord, so we don't do anything special
 # to ensure that.
-def test_restart_endpoint(neon_simple_env: NeonEnv, vanilla_pg):
+def test_restart_endpoint(neon_simple_env: NeonEnv, vanilla_pg: VanillaPostgres):
    env = neon_simple_env

    env.create_branch("init")
    endpoint = env.endpoints.create_start("init")
-    tenant_id = endpoint.safe_psql("show neon.tenant_id")[0][0]
-    timeline_id = endpoint.safe_psql("show neon.timeline_id")[0][0]
+    tenant_id = TenantId(cast("str", endpoint.safe_psql("show neon.tenant_id")[0][0]))
+    timeline_id = TimelineId(cast("str", endpoint.safe_psql("show neon.timeline_id")[0][0]))

    cur = endpoint.connect().cursor()
    cur.execute("create table t(key int, value text)")
@@ -380,7 +387,7 @@ def test_restart_endpoint(neon_simple_env: NeonEnv, vanilla_pg):
 # logical replication bug as such, but without logical replication,
 # records passed ot the WAL redo process are never large enough to hit
 # the bug.
-def test_large_records(neon_simple_env: NeonEnv, vanilla_pg):
+def test_large_records(neon_simple_env: NeonEnv, vanilla_pg: VanillaPostgres):
    env = neon_simple_env

    env.create_branch("init")
@@ -522,15 +529,20 @@ def logical_replication_wait_flush_lsn_sync(publisher: PgProtocol) -> Lsn:
    because for some WAL records like vacuum subscriber won't get any data at
    all.
    """
-    publisher_flush_lsn = Lsn(publisher.safe_psql("SELECT pg_current_wal_flush_lsn()")[0][0])
+    publisher_flush_lsn = Lsn(
+        cast("str", publisher.safe_psql("SELECT pg_current_wal_flush_lsn()")[0][0])
+    )

    def check_caughtup():
-        res = publisher.safe_psql(
-            """
+        res = cast(
+            "tuple[str, str, str]",
+            publisher.safe_psql(
+                """
 select sent_lsn, flush_lsn, pg_current_wal_flush_lsn() from pg_stat_replication sr, pg_replication_slots s
   where s.active_pid = sr.pid and s.slot_type = 'logical';
                                  """
-        )[0]
+            )[0],
+        )
        sent_lsn, flush_lsn, curr_publisher_flush_lsn = Lsn(res[0]), Lsn(res[1]), Lsn(res[2])
        log.info(
            f"sent_lsn={sent_lsn}, flush_lsn={flush_lsn}, publisher_flush_lsn={curr_publisher_flush_lsn}, waiting flush_lsn to reach {publisher_flush_lsn}"
@@ -545,7 +557,7 @@ select sent_lsn, flush_lsn, pg_current_wal_flush_lsn() from pg_stat_replication
 # flush_lsn reporting to publisher. Without this, subscriber may ack too far,
 # losing data on restart because publisher implicitly advances positition given
 # in START_REPLICATION to the confirmed_flush_lsn of the slot.
-def test_subscriber_synchronous_commit(neon_simple_env: NeonEnv, vanilla_pg):
+def test_subscriber_synchronous_commit(neon_simple_env: NeonEnv, vanilla_pg: VanillaPostgres):
    env = neon_simple_env
    # use vanilla as publisher to allow writes on it when safekeeper is down
    vanilla_pg.configure(
@@ -593,7 +605,7 @@ def test_subscriber_synchronous_commit(neon_simple_env: NeonEnv, vanilla_pg):
    # logical_replication_wait_flush_lsn_sync is expected to hang while
    # safekeeper is down.
    vanilla_pg.safe_psql("checkpoint;")
-    assert sub.safe_psql_scalar("SELECT count(*) FROM t") == 1000
+    assert cast("int", sub.safe_psql_scalar("SELECT count(*) FROM t")) == 1000

    # restart subscriber and ensure it can catch up lost tail again
    sub.stop(mode="immediate")
--- a/test_runner/regress/test_neon_cli.py
+++ b/test_runner/regress/test_neon_cli.py
@@ -1,6 +1,5 @@
 from __future__ import annotations

-import os
 import subprocess
 from pathlib import Path
 from typing import cast
@@ -15,7 +14,7 @@ from fixtures.neon_fixtures import (
    parse_project_git_version_output,
 )
 from fixtures.pageserver.http import PageserverHttpClient
-from fixtures.pg_version import PgVersion, skip_on_postgres
+from fixtures.utils import run_only_on_default_postgres, skip_in_debug_build


 def helper_compare_timeline_list(
@@ -195,10 +194,8 @@ def test_cli_start_stop_multi(neon_env_builder: NeonEnvBuilder):
    res.check_returncode()


-@skip_on_postgres(PgVersion.V14, reason="does not use postgres")
-@pytest.mark.skipif(
-    os.environ.get("BUILD_TYPE") == "debug", reason="unit test for test support, either build works"
-)
+@run_only_on_default_postgres(reason="does not use postgres")
+@skip_in_debug_build("unit test for test support, either build works")
 def test_parse_project_git_version_output_positive():
    commit = "b6f77b5816cf1dba12a3bc8747941182ce220846"

@@ -217,10 +214,8 @@ def test_parse_project_git_version_output_positive():
        assert parse_project_git_version_output(example) == commit


-@skip_on_postgres(PgVersion.V14, reason="does not use postgres")
-@pytest.mark.skipif(
-    os.environ.get("BUILD_TYPE") == "debug", reason="unit test for test support, either build works"
-)
+@run_only_on_default_postgres(reason="does not use postgres")
+@skip_in_debug_build("unit test for test support, either build works")
 def test_parse_project_git_version_output_local_docker():
    """
    Makes sure the tests don't accept the default version in Dockerfile one gets without providing
@@ -234,10 +229,8 @@ def test_parse_project_git_version_output_local_docker():
    assert input in str(e)


-@skip_on_postgres(PgVersion.V14, reason="does not use postgres")
-@pytest.mark.skipif(
-    os.environ.get("BUILD_TYPE") == "debug", reason="cli api sanity, either build works"
-)
+@run_only_on_default_postgres(reason="does not use postgres")
+@skip_in_debug_build("unit test for test support, either build works")
 def test_binaries_version_parses(neon_binpath: Path):
    """
    Ensures that we can parse the actual outputs of --version from a set of binaries.
--- a/test_runner/regress/test_pageserver_layer_rolling.py
+++ b/test_runner/regress/test_pageserver_layer_rolling.py
@@ -1,7 +1,6 @@
 from __future__ import annotations

 import asyncio
-import os
 import time
 from typing import TYPE_CHECKING

@@ -16,7 +15,7 @@ from fixtures.neon_fixtures import (
 )
 from fixtures.pageserver.http import PageserverHttpClient
 from fixtures.pageserver.utils import wait_for_last_record_lsn, wait_for_upload
-from fixtures.utils import wait_until
+from fixtures.utils import skip_in_debug_build, wait_until

 if TYPE_CHECKING:
    from typing import Optional
@@ -227,12 +226,9 @@ def test_idle_checkpoints(neon_env_builder: NeonEnvBuilder):
        assert get_dirty_bytes(env) >= dirty_after_write


-@pytest.mark.skipif(
-    # We have to use at least ~100MB of data to hit the lowest limit we can configure, which is
-    # prohibitively slow in debug mode
-    os.getenv("BUILD_TYPE") == "debug",
-    reason="Avoid running bulkier ingest tests in debug mode",
-)
+# We have to use at least ~100MB of data to hit the lowest limit we can configure, which is
+# prohibitively slow in debug mode
+@skip_in_debug_build("Avoid running bulkier ingest tests in debug mode")
 def test_total_size_limit(neon_env_builder: NeonEnvBuilder):
    """
    Test that checkpoints are done based on total ephemeral layer size, even if no one timeline is
--- a/test_runner/regress/test_pageserver_restart.py
+++ b/test_runner/regress/test_pageserver_restart.py
@@ -8,7 +8,7 @@ import pytest
 from fixtures.log_helper import log
 from fixtures.neon_fixtures import NeonEnvBuilder
 from fixtures.remote_storage import s3_storage
-from fixtures.utils import wait_until
+from fixtures.utils import skip_in_debug_build, wait_until


 # Test restarting page server, while safekeeper and compute node keep
@@ -155,12 +155,8 @@ def test_pageserver_restart(neon_env_builder: NeonEnvBuilder):
 # safekeeper and compute node keep running.
@pytest.mark.timeout(540)
@pytest.mark.parametrize("shard_count", [None, 4])
-def test_pageserver_chaos(
-    neon_env_builder: NeonEnvBuilder, build_type: str, shard_count: Optional[int]
-):
-    if build_type == "debug":
-        pytest.skip("times out in debug builds")
-
+@skip_in_debug_build("times out in debug builds")
+def test_pageserver_chaos(neon_env_builder: NeonEnvBuilder, shard_count: Optional[int]):
    # same rationale as with the immediate stop; we might leave orphan layers behind.
    neon_env_builder.disable_scrub_on_exit()
    neon_env_builder.enable_pageserver_remote_storage(s3_storage())
--- a/test_runner/regress/test_pageserver_secondary.py
+++ b/test_runner/regress/test_pageserver_secondary.py
@@ -17,7 +17,7 @@ from fixtures.pageserver.utils import (
    wait_for_upload_queue_empty,
 )
 from fixtures.remote_storage import LocalFsStorage, RemoteStorageKind, S3Storage, s3_storage
-from fixtures.utils import wait_until
+from fixtures.utils import skip_in_debug_build, wait_until
 from fixtures.workload import Workload
 from werkzeug.wrappers.request import Request
 from werkzeug.wrappers.response import Response
@@ -765,7 +765,7 @@ def test_secondary_background_downloads(neon_env_builder: NeonEnvBuilder):
    assert download_rate < expect_download_rate * 2


-@pytest.mark.skipif(os.environ.get("BUILD_TYPE") == "debug", reason="only run with release build")
+@skip_in_debug_build("only run with release build")
@pytest.mark.parametrize("via_controller", [True, False])
 def test_slow_secondary_downloads(neon_env_builder: NeonEnvBuilder, via_controller: bool):
    """
--- a/test_runner/regress/test_pg_regress.py
+++ b/test_runner/regress/test_pg_regress.py
@@ -3,7 +3,6 @@
 #
 from __future__ import annotations

-import os
 from concurrent.futures import ThreadPoolExecutor
 from pathlib import Path
 from typing import TYPE_CHECKING, cast
@@ -19,6 +18,7 @@ from fixtures.neon_fixtures import (
 )
 from fixtures.pg_version import PgVersion
 from fixtures.remote_storage import s3_storage
+from fixtures.utils import skip_in_debug_build

 if TYPE_CHECKING:
    from typing import Optional
@@ -329,7 +329,7 @@ def test_sql_regress(
    post_checks(env, test_output_dir, DBNAME, endpoint)


-@pytest.mark.skipif(os.environ.get("BUILD_TYPE") == "debug", reason="only run with release build")
+@skip_in_debug_build("only run with release build")
 def test_tx_abort_with_many_relations(
    neon_env_builder: NeonEnvBuilder,
 ):
--- a/test_runner/regress/test_physical_and_logical_replicaiton.py
+++ b/test_runner/regress/test_physical_and_logical_replicaiton.py
@@ -5,7 +5,8 @@ import time
 from fixtures.neon_fixtures import NeonEnv, logical_replication_sync


-def test_physical_and_logical_replication(neon_simple_env: NeonEnv, vanilla_pg):
+def test_physical_and_logical_replication_slot_not_copied(neon_simple_env: NeonEnv, vanilla_pg):
+    """Test read replica of a primary which has a logical replication publication"""
    env = neon_simple_env

    n_records = 100000
@@ -13,7 +14,6 @@ def test_physical_and_logical_replication(neon_simple_env: NeonEnv, vanilla_pg):
    primary = env.endpoints.create_start(
        branch_name="main",
        endpoint_id="primary",
-        config_lines=["min_wal_size=32MB", "max_wal_size=64MB"],
    )
    p_con = primary.connect()
    p_cur = p_con.cursor()
@@ -30,7 +30,6 @@ def test_physical_and_logical_replication(neon_simple_env: NeonEnv, vanilla_pg):
    secondary = env.endpoints.new_replica_start(
        origin=primary,
        endpoint_id="secondary",
-        config_lines=["min_wal_size=32MB", "max_wal_size=64MB"],
    )

    s_con = secondary.connect()
@@ -48,3 +47,51 @@ def test_physical_and_logical_replication(neon_simple_env: NeonEnv, vanilla_pg):
    # Check that LR slot is not copied to replica
    s_cur.execute("select count(*) from pg_replication_slots")
    assert s_cur.fetchall()[0][0] == 0
+
+
+def test_aux_not_logged_at_replica(neon_simple_env: NeonEnv, vanilla_pg):
+    """Test that AUX files are not saved at replica"""
+    env = neon_simple_env
+
+    n_records = 200000
+
+    primary = env.endpoints.create_start(
+        branch_name="main",
+        endpoint_id="primary",
+    )
+    p_con = primary.connect()
+    p_cur = p_con.cursor()
+    p_cur.execute("CREATE TABLE t(pk bigint primary key, payload text default repeat('?',200))")
+    p_cur.execute("create publication pub1 for table t")
+
+    # start subscriber
+    vanilla_pg.start()
+    vanilla_pg.safe_psql("CREATE TABLE t(pk bigint primary key, payload text)")
+    connstr = primary.connstr().replace("'", "''")
+    vanilla_pg.safe_psql(f"create subscription sub1 connection '{connstr}' publication pub1")
+
+    for pk in range(n_records):
+        p_cur.execute("insert into t (pk) values (%s)", (pk,))
+
+    # LR snapshot is stored each 15 seconds
+    time.sleep(16)
+
+    # start replica
+    secondary = env.endpoints.new_replica_start(
+        origin=primary,
+        endpoint_id="secondary",
+    )
+
+    s_con = secondary.connect()
+    s_cur = s_con.cursor()
+
+    logical_replication_sync(vanilla_pg, primary)
+
+    assert vanilla_pg.safe_psql("select count(*) from t")[0][0] == n_records
+    s_cur.execute("select count(*) from t")
+    assert s_cur.fetchall()[0][0] == n_records
+
+    primary.stop()
+    time.sleep(1)
+    secondary.stop()
+    assert not secondary.log_contains("cannot make new WAL entries during recovery")
--- a/test_runner/regress/test_replica_start.py
+++ b/test_runner/regress/test_replica_start.py
@@ -30,7 +30,7 @@ import pytest
 from fixtures.log_helper import log
 from fixtures.neon_fixtures import NeonEnv, wait_for_last_flush_lsn, wait_replica_caughtup
 from fixtures.pg_version import PgVersion
-from fixtures.utils import query_scalar, wait_until
+from fixtures.utils import query_scalar, skip_on_postgres, wait_until

 CREATE_SUBXACTS_FUNC = """
 create or replace function create_subxacts(n integer) returns void as $$
@@ -137,6 +137,12 @@ def test_replica_start_scan_clog_crashed_xids(neon_simple_env: NeonEnv):
    assert secondary_cur.fetchone() == (1,)


+@skip_on_postgres(
+    PgVersion.V14, reason="pg_log_standby_snapshot() function is available since Postgres 16"
+)
+@skip_on_postgres(
+    PgVersion.V15, reason="pg_log_standby_snapshot() function is available since Postgres 16"
+)
 def test_replica_start_at_running_xacts(neon_simple_env: NeonEnv, pg_version):
    """
    Test that starting a replica works right after the primary has
@@ -149,9 +155,6 @@ def test_replica_start_at_running_xacts(neon_simple_env: NeonEnv, pg_version):
    """
    env = neon_simple_env

-    if env.pg_version == PgVersion.V14 or env.pg_version == PgVersion.V15:
-        pytest.skip("pg_log_standby_snapshot() function is available only in PG16")
-
    primary = env.endpoints.create_start(branch_name="main", endpoint_id="primary")
    primary_conn = primary.connect()
    primary_cur = primary_conn.cursor()
--- a/test_runner/regress/test_sharding.py
+++ b/test_runner/regress/test_sharding.py
@@ -20,7 +20,7 @@ from fixtures.neon_fixtures import (
 )
 from fixtures.pageserver.utils import assert_prefix_empty, assert_prefix_not_empty
 from fixtures.remote_storage import s3_storage
-from fixtures.utils import wait_until
+from fixtures.utils import skip_in_debug_build, wait_until
 from fixtures.workload import Workload
 from pytest_httpserver import HTTPServer
 from typing_extensions import override
@@ -853,12 +853,9 @@ def test_sharding_split_stripe_size(
    wait_until(10, 1, assert_restart_notification)


-@pytest.mark.skipif(
-    # The quantity of data isn't huge, but debug can be _very_ slow, and the things we're
-    # validating in this test don't benefit much from debug assertions.
-    os.getenv("BUILD_TYPE") == "debug",
-    reason="Avoid running bulkier ingest tests in debug mode",
-)
+# The quantity of data isn't huge, but debug can be _very_ slow, and the things we're
+# validating in this test don't benefit much from debug assertions.
+@skip_in_debug_build("Avoid running bulkier ingest tests in debug mode")
 def test_sharding_ingest_layer_sizes(
    neon_env_builder: NeonEnvBuilder,
 ):
--- a/test_runner/regress/test_storage_controller.py
+++ b/test_runner/regress/test_storage_controller.py
@@ -36,11 +36,12 @@ from fixtures.pageserver.utils import (
    remote_storage_delete_key,
    timeline_delete_wait_completed,
 )
-from fixtures.pg_version import PgVersion, run_only_on_default_postgres
+from fixtures.pg_version import PgVersion
 from fixtures.port_distributor import PortDistributor
 from fixtures.remote_storage import RemoteStorageKind, s3_storage
 from fixtures.storage_controller_proxy import StorageControllerProxy
 from fixtures.utils import (
+    run_only_on_default_postgres,
    run_pg_bench_small,
    subprocess_capture,
    wait_until,
--- a/test_runner/regress/test_tenant_size.py
+++ b/test_runner/regress/test_tenant_size.py
@@ -1,6 +1,5 @@
 from __future__ import annotations

-import os
 from concurrent.futures import ThreadPoolExecutor
 from pathlib import Path

@@ -21,7 +20,7 @@ from fixtures.pageserver.utils import (
    wait_until_tenant_active,
 )
 from fixtures.pg_version import PgVersion
-from fixtures.utils import wait_until
+from fixtures.utils import skip_in_debug_build, wait_until


 def test_empty_tenant_size(neon_env_builder: NeonEnvBuilder):
@@ -279,7 +278,7 @@ def test_only_heads_within_horizon(neon_simple_env: NeonEnv, test_output_dir: Pa
    size_debug_file.write(size_debug)


-@pytest.mark.skipif(os.environ.get("BUILD_TYPE") == "debug", reason="only run with release build")
+@skip_in_debug_build("only run with release build")
 def test_single_branch_get_tenant_size_grows(
    neon_env_builder: NeonEnvBuilder, test_output_dir: Path, pg_version: PgVersion
 ):
--- a/test_runner/regress/test_timeline_archive.py
+++ b/test_runner/regress/test_timeline_archive.py
@@ -1,15 +1,22 @@
 from __future__ import annotations

+import json
+from typing import Optional
+
 import pytest
 from fixtures.common_types import TenantId, TimelineArchivalState, TimelineId
+from fixtures.log_helper import log
 from fixtures.neon_fixtures import (
    NeonEnvBuilder,
    last_flush_lsn_upload,
 )
 from fixtures.pageserver.http import PageserverApiException
-from fixtures.pageserver.utils import assert_prefix_empty, assert_prefix_not_empty
-from fixtures.remote_storage import s3_storage
+from fixtures.pageserver.utils import assert_prefix_empty, assert_prefix_not_empty, list_prefix
+from fixtures.remote_storage import S3Storage, s3_storage
 from fixtures.utils import wait_until
+from mypy_boto3_s3.type_defs import (
+    ObjectTypeDef,
+)


@pytest.mark.parametrize("shard_count", [0, 4])
@@ -369,3 +376,146 @@ def test_timeline_offload_persist(neon_env_builder: NeonEnvBuilder, delete_timel
        neon_env_builder.pageserver_remote_storage,
        prefix=f"tenants/{str(tenant_id)}/",
    )
+
+
+@pytest.mark.parametrize("offload_child", ["offload", "offload-corrupt", "archive", None])
+def test_timeline_retain_lsn(neon_env_builder: NeonEnvBuilder, offload_child: Optional[str]):
+    """
+    Ensure that retain_lsn functionality for timelines works, both for offloaded and non-offloaded ones
+    """
+    if offload_child == "offload-corrupt":
+        # Our corruption code only works with S3 compatible storage
+        neon_env_builder.enable_pageserver_remote_storage(s3_storage())
+
+    env = neon_env_builder.init_start()
+    ps_http = env.pageserver.http_client()
+
+    # Turn off gc and compaction loops: we want to issue them manually for better reliability
+    tenant_id, root_timeline_id = env.create_tenant(
+        conf={
+            # small checkpointing and compaction targets to ensure we generate many upload operations
+            "checkpoint_distance": 128 * 1024,
+            "compaction_threshold": 1,
+            "compaction_target_size": 128 * 1024,
+            # set small image creation thresholds so that gc deletes data
+            "image_creation_threshold": 2,
+            # disable background compaction and GC. We invoke it manually when we want it to happen.
+            "gc_period": "0s",
+            "compaction_period": "0s",
+            # Disable pitr, we only want the latest lsn
+            "pitr_interval": "0s",
+            # Don't rely on endpoint lsn leases
+            "lsn_lease_length": "0s",
+        }
+    )
+
+    with env.endpoints.create_start("main", tenant_id=tenant_id) as endpoint:
+        endpoint.safe_psql_many(
+            [
+                "CREATE TABLE foo(v int, key serial primary key, t text default 'data_content')",
+                "SELECT setseed(0.4321)",
+                "INSERT INTO foo SELECT v FROM (SELECT generate_series(1,2048), (random() * 409600)::int as v) as random_numbers",
+            ]
+        )
+        pre_branch_sum = endpoint.safe_psql("SELECT sum(key) from foo where v < 51200")
+        log.info(f"Pre branch sum: {pre_branch_sum}")
+        last_flush_lsn_upload(env, endpoint, tenant_id, root_timeline_id)
+
+    # Create a branch and write some additional data to the parent
+    child_timeline_id = env.create_branch("test_archived_branch", tenant_id)
+
+    with env.endpoints.create_start("main", tenant_id=tenant_id) as endpoint:
+        # Do some churn of the data. This is important so that we can overwrite image layers.
+        for i in range(10):
+            endpoint.safe_psql_many(
+                [
+                    f"SELECT setseed(0.23{i})",
+                    "UPDATE foo SET v=(random() * 409600)::int WHERE v % 3 = 2",
+                    "UPDATE foo SET v=(random() * 409600)::int WHERE v % 3 = 1",
+                    "UPDATE foo SET v=(random() * 409600)::int WHERE v % 3 = 0",
+                ]
+            )
+        post_branch_sum = endpoint.safe_psql("SELECT sum(key) from foo where v < 51200")
+        log.info(f"Post branch sum: {post_branch_sum}")
+        last_flush_lsn_upload(env, endpoint, tenant_id, root_timeline_id)
+
+    if offload_child is not None:
+        ps_http.timeline_archival_config(
+            tenant_id,
+            child_timeline_id,
+            state=TimelineArchivalState.ARCHIVED,
+        )
+        leaf_detail = ps_http.timeline_detail(
+            tenant_id,
+            child_timeline_id,
+        )
+        assert leaf_detail["is_archived"] is True
+        if "offload" in offload_child:
+            ps_http.timeline_offload(tenant_id, child_timeline_id)
+
+    # Do a restart to get rid of any in-memory objects (we only init gc info once, at attach)
+    env.pageserver.stop()
+    if offload_child == "offload-corrupt":
+        assert isinstance(env.pageserver_remote_storage, S3Storage)
+        listing = list_prefix(
+            env.pageserver_remote_storage, f"tenants/{str(tenant_id)}/tenant-manifest"
+        )
+        objects: list[ObjectTypeDef] = listing.get("Contents", [])
+        assert len(objects) > 0
+        remote_key: str = str(objects[0].get("Key", []))
+        local_path = str(env.repo_dir / "tenant-manifest.json")
+
+        log.info(f"Downloading {remote_key} -> {local_path}")
+        env.pageserver_remote_storage.client.download_file(
+            env.pageserver_remote_storage.bucket_name, remote_key, local_path
+        )
+
+        log.info(f"Corrupting {local_path}")
+        with open(local_path) as manifest_json_file:
+            manifest_json = json.load(manifest_json_file)
+        for offloaded_timeline in manifest_json["offloaded_timelines"]:
+            offloaded_timeline["ancestor_retain_lsn"] = None
+        with open(local_path, "w") as manifest_json_file:
+            json.dump(manifest_json, manifest_json_file)
+
+        log.info(f"Uploading {local_path} -> {remote_key}")
+        env.pageserver_remote_storage.client.upload_file(
+            local_path, env.pageserver_remote_storage.bucket_name, remote_key
+        )
+        # The point of our earlier efforts was to provoke these
+        env.pageserver.allowed_errors.extend(
+            [
+                ".*initial size calculation failed: PageRead.MissingKey.could not find data for key.*",
+                ".*page_service_conn_main.*could not find data for key.*",
+            ]
+        )
+    env.pageserver.start()
+
+    # Do an agressive gc and compaction of the parent branch
+    ps_http.timeline_gc(tenant_id=tenant_id, timeline_id=root_timeline_id, gc_horizon=0)
+    ps_http.timeline_checkpoint(
+        tenant_id,
+        root_timeline_id,
+        force_l0_compaction=True,
+        force_repartition=True,
+        wait_until_uploaded=True,
+        compact=True,
+    )
+
+    if offload_child is not None:
+        ps_http.timeline_archival_config(
+            tenant_id,
+            child_timeline_id,
+            state=TimelineArchivalState.UNARCHIVED,
+        )
+
+    # Now, after unarchival, the child timeline should still have its data accessible (or corrupted)
+    if offload_child == "offload-corrupt":
+        with pytest.raises(RuntimeError, match=".*failed to get basebackup.*"):
+            env.endpoints.create_start(
+                "test_archived_branch", tenant_id=tenant_id, basebackup_request_tries=1
+            )
+    else:
+        with env.endpoints.create_start("test_archived_branch", tenant_id=tenant_id) as endpoint:
+            sum = endpoint.safe_psql("SELECT sum(key) from foo where v < 51200")
+            assert sum == pre_branch_sum
--- a/test_runner/regress/test_timeline_detach_ancestor.py
+++ b/test_runner/regress/test_timeline_detach_ancestor.py
@@ -869,8 +869,17 @@ def test_sharded_timeline_detach_ancestor(neon_env_builder: NeonEnvBuilder):
        assert count == 10000


-@pytest.mark.parametrize("mode", ["delete_timeline", "delete_tenant"])
-@pytest.mark.parametrize("sharded", [False, True])
+@pytest.mark.parametrize(
+    "mode, sharded",
+    [
+        ("delete_timeline", False),
+        ("delete_timeline", True),
+        ("delete_tenant", False),
+        # the shared/exclusive lock for tenant is blocking this:
+        # timeline detach ancestor takes shared, delete tenant takes exclusive
+        # ("delete_tenant", True)
+    ],
+)
 def test_timeline_detach_ancestor_interrupted_by_deletion(
    neon_env_builder: NeonEnvBuilder, mode: str, sharded: bool
 ):
@@ -885,11 +894,6 @@ def test_timeline_detach_ancestor_interrupted_by_deletion(
    - shutdown winning over complete, see test_timeline_is_deleted_before_timeline_detach_ancestor_completes
    """

-    if sharded and mode == "delete_tenant":
-        # the shared/exclusive lock for tenant is blocking this:
-        # timeline detach ancestor takes shared, delete tenant takes exclusive
-        pytest.skip("tenant deletion while timeline ancestor detach is underway cannot happen")
-
    shard_count = 2 if sharded else 1

    neon_env_builder.num_pageservers = shard_count
--- a/test_runner/regress/test_wal_acceptor.py
+++ b/test_runner/regress/test_wal_acceptor.py
@@ -54,6 +54,8 @@ from fixtures.utils import (
    PropagatingThread,
    get_dir_size,
    query_scalar,
+    run_only_on_default_postgres,
+    skip_in_debug_build,
    start_in_background,
    wait_until,
 )
@@ -2104,10 +2106,9 @@ def test_pull_timeline_while_evicted(neon_env_builder: NeonEnvBuilder):
 # The only way to verify this without manipulating time is to sleep for a while.
 # In this test we sleep for 60 seconds, so this test takes at least 1 minute to run.
 # This is longer than most other tests, we run it only for v16 to save CI resources.
+@run_only_on_default_postgres("run only on release build to save CI resources")
+@skip_in_debug_build("run only on release build to save CI resources")
 def test_idle_reconnections(neon_env_builder: NeonEnvBuilder):
-    if os.environ.get("PYTEST_CURRENT_TEST", "").find("[debug-pg16]") == -1:
-        pytest.skip("run only on debug postgres v16 to save CI resources")
-
    neon_env_builder.num_safekeepers = 3
    env = neon_env_builder.init_start()

--- a/test_runner/regress/test_wal_acceptor_async.py
+++ b/test_runner/regress/test_wal_acceptor_async.py
@@ -14,6 +14,7 @@ from fixtures.common_types import Lsn, TenantId, TimelineId
 from fixtures.log_helper import getLogger
 from fixtures.neon_fixtures import Endpoint, NeonEnv, NeonEnvBuilder, Safekeeper
 from fixtures.remote_storage import RemoteStorageKind
+from fixtures.utils import skip_in_debug_build

 if TYPE_CHECKING:
    from typing import Optional
@@ -760,10 +761,8 @@ async def run_wal_lagging(env: NeonEnv, endpoint: Endpoint, test_output_dir: Pat
 # The test takes more than default 5 minutes on Postgres 16,
 # see https://github.com/neondatabase/neon/issues/5305
@pytest.mark.timeout(600)
+@skip_in_debug_build("times out in debug builds")
 def test_wal_lagging(neon_env_builder: NeonEnvBuilder, test_output_dir: Path, build_type: str):
-    if build_type == "debug":
-        pytest.skip("times out in debug builds")
-
    neon_env_builder.num_safekeepers = 3
    env = neon_env_builder.init_start()

--- a/vendor/postgres-v14
+++ b/vendor/postgres-v14
--- a/vendor/postgres-v15
+++ b/vendor/postgres-v15
--- a/vendor/postgres-v16
+++ b/vendor/postgres-v16
--- a/vendor/postgres-v17
+++ b/vendor/postgres-v17
--- a/vendor/revisions.json
+++ b/vendor/revisions.json
@@ -1,18 +1,18 @@
 {
  "v17": [
    "17.0",
-    "9ad2f3c5c37c08069a01c1e3f6b7cf275437e0cb"
+    "ae4cc30dba24f3910533e5a48e8103c3f2fff300"
  ],
  "v16": [
    "16.4",
-    "e131a9c027b202ce92bd7b9cf2569d48a6f9948e"
+    "03b43900edc5d8d6eecec460bfc89aec7174bd84"
  ],
  "v15": [
    "15.8",
-    "22e580fe9ffcea7e02592110b1c9bf426d83cada"
+    "fd631a959049dfe2b82f67409c8b8b0d3e0016d1"
  ],
  "v14": [
    "14.13",
-    "2199b83fb72680001ce0f43bf6187a21dfb8f45d"
+    "de0a000dafc2e66ce2e39282d3aa1c704fe0390e"
  ]
 }
Author	SHA1	Message	Date
Heikki Linnakangas	679b302c06	try with unshallow	2024-11-12 13:00:42 +02:00
Heikki Linnakangas	9a07960e83	increase depth	2024-11-12 11:57:28 +02:00
Heikki Linnakangas	6ddaca25dc	use a version of submodule check with extra debug info	2024-11-12 11:47:42 +02:00
Konstantin Knizhnik	f9dca3ad8a	Bump Postgres versions	2024-11-12 08:25:16 +02:00
Konstantin Knizhnik	84cbab7ada	Undo occasional changes in test_physical_and_logical_replicaiton.py test	2024-11-12 08:25:16 +02:00
Konstantin Knizhnik	e252a4208b	Bump postgres version	2024-11-12 08:25:15 +02:00
Konstantin Knizhnik	26f5bafb99	Update test_physical_and_logical_replicaiton.py to reproduce the problem	2024-11-12 08:25:15 +02:00
Konstantin Knizhnik	1eddf30ff3	Update test_runner/regress/test_physical_and_logical_replicaiton.py Co-authored-by: Heikki Linnakangas <heikki@neon.tech>	2024-11-12 08:25:15 +02:00
Konstantin Knizhnik	25532eed33	Bump Postgres version	2024-11-12 08:25:15 +02:00
Konstantin Knizhnik	637da7719e	Add test for combination of physical and logical replication	2024-11-12 08:25:15 +02:00
Konstantin Knizhnik	301edc753c	Do not wallog AUX files at replica	2024-11-12 08:25:15 +02:00
Tristan Partin	cc8029c4c8	Update pg_cron to 1.6.4 This comes with PG 17 support. Signed-off-by: Tristan Partin <tristan@neon.tech>	2024-11-11 20:10:53 -06:00
Tristan Partin	5be6b07cf1	Improve typing related to regress/test_logical_replication.py (#9725 ) Signed-off-by: Tristan Partin <tristan@neon.tech>	2024-11-11 17:36:45 -06:00
Arpad Müller	b018bc7da8	Add a retain_lsn test (#9599 ) Add a test that ensures the `retain_lsn` functionality works. Right now, there is not a single test that is broken if offloaded or non-offloaded timelines don't get registered at their parents, preventing gc from discarding the ancestor_lsns of the children. This PR fills that gap. The test has four modes: * `offloaded`: offload the child timeline, run compaction on the parent timeline, unarchive the child timeline, then try reading from it. hopefully the data is still there. * `offloaded-corrupted`: offload the child timeline, corrupts the manifest in a way that the pageserver believes the timeline was flattened. This is the closest we can get to pretend the `retain_lsn` mechanism doesn't exist for offloaded timelines, so we can avoid adding endpoints to the pageserver that do this manually for tests. The test then checks that indeed data is corrupted and the endpoint can't be started. That way we know that the test is actually working, and actually tests the `retain_lsn` mechanism, instead of say the lsn lease mechanism, or one of the many other mechanisms that impede gc. * `archived`: the child timeline gets archived but doesn't get offloaded. this currently matches the `None` case but we might have refactors in the future that make archived timelines sufficiently different from non-archived ones. * `None`: the child timeline doesn't even get archived. this tests that normal timelines participate in `retain_lsn`. I've made them locally not participate in `retain_lsn` (via commenting out the respective `ancestor_children.push` statement in tenant.rs) and ran the testsuite, and not a single test failed. So this test is first of its kind. Part of #8088.	2024-11-11 22:29:21 +00:00
Tristan Partin	4b075db7ea	Add a postgres_exporter config file This exporter logs an ERROR if a file called `postgres_exporter.yml` is not located in its current working directory. We can silence it by adding an empty config file and pointing the exporter at it. Signed-off-by: Tristan Partin <tristan@neon.tech>	2024-11-11 14:49:37 -06:00
Fedor Dikarev	fde16f8614	use batch gh-workflow-stats-action with separate table (#9722 ) We found that exporting GH Workflow Runs in batch is more efficient due to - better utilisation of Github API - and gh runners usage is rounded to minutes, so even when ad-hoc export is done in 5-10 seconds, we billed for one minute usage So now we introduce batch exporting, with version v0.2.x of github workflow stats exporter. How it's expected to work now: - every 15 minutes we query for the workflow runs, created in last 2 hours - to avoid missing workflows that ran for more than 2 hours, every night (00:25) we will query workflows created in past 24 hours and export them as well - should we have query for even longer periods? - lets see how it works with current schedule - for longer periods like for days or weeks, it may require to adjust logic and concurrency of querying data, so lets for now use simpler version	2024-11-11 20:33:29 +00:00
Alex Chi Z.	5a138d08a3	feat(pageserver): support partial gc-compaction for delta layers (#9611 ) The final patch for partial compaction, part of https://github.com/neondatabase/neon/issues/9114, close https://github.com/neondatabase/neon/issues/8921 (note that we didn't implement parallel compaction or compaction scheduler for partial compaction -- currently this needs to be scheduled by using a Python script to split the keyspace, and in the future, automatically split based on the key partitioning when the pageserver wants to trigger a gc-compaction) ## Summary of changes * Update the layer selection algorithm to use the same selection as full compaction (everything intersect/below gc horizon) * Update the layer selection algorithm to also generate a list of delta layers that need to be rewritten * Add the logic to rewrite delta layers and add them back to the layer map * Update test case to do partial compaction on deltas --------- Signed-off-by: Alex Chi Z <chi@neon.tech>	2024-11-11 20:30:32 +00:00
Tristan Partin	2d9652c434	Clean up C.UTF-8 locale changes Removes some unnecessary initdb arguments, and fixes Neon for MacOS since it doesn't seem to ship a C.UTF-8 locale. Signed-off-by: Tristan Partin <tristan@neon.tech>	2024-11-11 13:53:12 -06:00
Alexander Bayandin	e9dcfa2eb2	test_runner: skip more tests using decorator instead of pytest.skip (#9704 ) ## Problem Running `pytest.skip(...)` in a test body instead of marking the test with `@pytest.mark.skipif(...)` makes all fixtures to be initialised, which is not necessary if the test is going to be skipped anyway. Also, some tests are unnecessarily skipped (e.g. `test_layer_bloating` on Postgres 17, or `test_idle_reconnections` at all) or run (e.g. `test_parse_project_git_version_output_positive` more than on once configuration) according to comments. ## Summary of changes - Move `skip_on_postgres` / `xfail_on_postgres` / `run_only_on_default_postgres` decorators to `fixture.utils` - Add new `skip_in_debug_build` and `skip_on_ci` decorators - Replace `pytest.skip(...)` calls with decorators where possible	2024-11-11 18:07:01 +00:00
Peter Bendel	8db84d9964	new ingest benchmark (#9711 ) ## Problem We have no specific benchmark testing project migration of postgresql project with existing data into Neon. Typical steps of such a project migration are - schema creation in the neon project - initial COPY of relations - creation of indexes and constraints - vacuum analyze ## Summary of changes Add a periodic benchmark running 9 AM UTC every day. In each run: - copy a 200 GiB project that has realistic schema, data, tables, indexes and constraints from another project into - a new Neon project (7 CU fixed) - an existing tenant, (but new branch and new database) that already has 4 TiB of data - use pgcopydb tool to automate all steps and parallelize COPY and index creation - parse pgcopydb output and report performance metrics in Neon performance test database ## Logs This benchmark has been tested first manually and then as part of benchmarking.yml workflow, example run see https://github.com/neondatabase/neon/actions/runs/11757679870	2024-11-11 17:51:15 +00:00
Alexander Bayandin	1aab34715a	Remove checklist from the PR template (#9702 ) ## Problem Once we enable the merge queue for the `main` branch, it won't be possible to adjust the commit message right after pressing the "Squash and merge" button and the PR title + description will be used as is. To avoid extra noise in the commits in the `main` with the checklist leftovers, I propose removing the checklist from the PR template and keeping only the Problem / Summary of changes. ## Summary of changes - Remove the checklist from the PR template	2024-11-11 17:01:02 +00:00
Erik Grinaker	f63de5f527	safekeeper: add `initialize_segment` variant of `safekeeper_wal_storage_operation_seconds` (#9691 ) ## Problem We don't have a metric capturing the latency of segment initialization. This can be significant due to fsyncs. ## Summary of changes Add an `initialize_segment` variant of `safekeeper_wal_storage_operation_seconds`.	2024-11-11 17:55:50 +01:00
Alex Chi Z.	54a1676680	rfc: update aux file rfc to reflect latest optimizations (#9681 ) Reflects https://github.com/neondatabase/neon/pull/9631 in the RFC. Signed-off-by: Alex Chi Z <chi@neon.tech>	2024-11-11 09:19:03 -05:00
Alex Chi Z.	48c06d9f7b	fix(pageserver): increase frozen layer warning threshold; ignore in tests (#9705 ) Perf benchmarks produce a lot of layers. ## Summary of changes Bumping the threshold and ignore the warning. --------- Signed-off-by: Alex Chi Z <chi@neon.tech>	2024-11-11 09:13:46 -05:00
Alexander Bayandin	f510647c7e	CI: retry `actions/github-script` for 5XX errors (#9703 ) ## Problem GitHub API can return error 500, and it fails jobs that use `actions/github-script` action. ## Summary of changes - Add `retry: 500` to all `actions/github-script` usage	2024-11-11 12:42:32 +00:00
Vlad Lazar	ceaa80ffeb	storcon: add peer token for peer to peer communication (#9695 ) ## Problem We wish to stop using admin tokens in the infra repo, but step down requests use the admin token. ## Summary of Changes Introduce a new "ControllerPeer" scope and use it for step-down requests.	2024-11-11 09:58:41 +00:00