build: link controller with system libpq

Unlock LFC mutex when LFC cache is disabled (#10235 )
## Problem See https://github.com/neondatabase/neon/issues/10233 `lfc_containsv` returns with holding lock when LFC was disabled. This bug was introduced in commit 78938d1b59 ## Summary of changes Release lock before return. Co-authored-by: Konstantin Knizhnik <knizhnik@neon.tech>
2026-03-03 16:30:38 +00:00 · 2025-01-02 12:46:51 +00:00 · 2025-01-02 11:28:15 +00:00 · 2025-01-02 09:36:13 +00:00 · 2025-01-02 09:35:28 +00:00 · 2024-12-19 18:40:20 +00:00
282 changed files with 11662 additions and 3473 deletions
--- a/.github/actionlint.yml
+++ b/.github/actionlint.yml
@@ -21,3 +21,7 @@ config-variables:
  - SLACK_UPCOMING_RELEASE_CHANNEL_ID
  - DEV_AWS_OIDC_ROLE_ARN
  - BENCHMARK_INGEST_TARGET_PROJECTID
+  - PGREGRESS_PG16_PROJECT_ID
+  - PGREGRESS_PG17_PROJECT_ID
+  - SLACK_ON_CALL_QA_STAGING_STREAM
+  - DEV_AWS_OIDC_ROLE_MANAGE_BENCHMARK_EC2_VMS_ARN
--- a/.github/actions/allure-report-generate/action.yml
+++ b/.github/actions/allure-report-generate/action.yml
@@ -7,10 +7,9 @@ inputs:
    type: boolean
    required: false
    default: false
-  aws_oicd_role_arn:
-    description: 'the OIDC role arn to (re-)acquire for allure report upload - if not set call must acquire OIDC role'
-    required: false
-    default: ''
+  aws-oicd-role-arn:
+    description: 'OIDC role arn to interract with S3'
+    required: true

 outputs:
  base-url:
@@ -84,12 +83,11 @@ runs:
        ALLURE_VERSION: 2.27.0
        ALLURE_ZIP_SHA256: b071858fb2fa542c65d8f152c5c40d26267b2dfb74df1f1608a589ecca38e777

-    - name: (Re-)configure AWS credentials # necessary to upload reports to S3 after a long-running test
-      if: ${{ !cancelled() && (inputs.aws_oicd_role_arn != '') }}
-      uses: aws-actions/configure-aws-credentials@v4
+    - uses: aws-actions/configure-aws-credentials@v4
+      if: ${{ !cancelled() }}
      with:
        aws-region: eu-central-1
-        role-to-assume: ${{ inputs.aws_oicd_role_arn }}
+        role-to-assume: ${{ inputs.aws-oicd-role-arn }}
        role-duration-seconds: 3600 # 1 hour should be more than enough to upload report

    # Potentially we could have several running build for the same key (for example, for the main branch), so we use improvised lock for this
--- a/.github/actions/allure-report-store/action.yml
+++ b/.github/actions/allure-report-store/action.yml
@@ -8,10 +8,9 @@ inputs:
  unique-key:
    description: 'string to distinguish different results in the same run'
    required: true
-  aws_oicd_role_arn:
-    description: 'the OIDC role arn to (re-)acquire for allure report upload - if not set call must acquire OIDC role'
-    required: false
-    default: ''
+  aws-oicd-role-arn:
+    description: 'OIDC role arn to interract with S3'
+    required: true

 runs:
  using: "composite"
@@ -36,12 +35,11 @@ runs:
      env:
        REPORT_DIR: ${{ inputs.report-dir }}

-    - name: (Re-)configure AWS credentials # necessary to upload reports to S3 after a long-running test
-      if: ${{ !cancelled() && (inputs.aws_oicd_role_arn != '') }}
-      uses: aws-actions/configure-aws-credentials@v4
+    - uses: aws-actions/configure-aws-credentials@v4
+      if: ${{ !cancelled() }}
      with:
        aws-region: eu-central-1
-        role-to-assume: ${{ inputs.aws_oicd_role_arn }}
+        role-to-assume: ${{ inputs.aws-oicd-role-arn }}
        role-duration-seconds: 3600 # 1 hour should be more than enough to upload report

    - name: Upload test results
--- a/.github/actions/download/action.yml
+++ b/.github/actions/download/action.yml
@@ -15,10 +15,19 @@ inputs:
  prefix:
    description: "S3 prefix. Default is '${GITHUB_RUN_ID}/${GITHUB_RUN_ATTEMPT}'"
    required: false
+  aws-oicd-role-arn:
+    description: 'OIDC role arn to interract with S3'
+    required: true

 runs:
  using: "composite"
  steps:
+    - uses: aws-actions/configure-aws-credentials@v4
+      with:
+        aws-region: eu-central-1
+        role-to-assume: ${{ inputs.aws-oicd-role-arn }}
+        role-duration-seconds: 3600
+
    - name: Download artifact
      id: download-artifact
      shell: bash -euxo pipefail {0}
--- a/.github/actions/run-python-test-set/action.yml
+++ b/.github/actions/run-python-test-set/action.yml
@@ -48,10 +48,9 @@ inputs:
    description: 'benchmark durations JSON'
    required: false
    default: '{}'
-  aws_oicd_role_arn:
-    description: 'the OIDC role arn to (re-)acquire for allure report upload - if not set call must acquire OIDC role'
-    required: false
-    default: ''
+  aws-oicd-role-arn:
+    description: 'OIDC role arn to interract with S3'
+    required: true

 runs:
  using: "composite"
@@ -62,6 +61,7 @@ runs:
      with:
        name: neon-${{ runner.os }}-${{ runner.arch }}-${{ inputs.build_type }}-artifact
        path: /tmp/neon
+        aws-oicd-role-arn: ${{ inputs.aws-oicd-role-arn }}

    - name: Download Neon binaries for the previous release
      if: inputs.build_type != 'remote'
@@ -70,6 +70,7 @@ runs:
        name: neon-${{ runner.os }}-${{ runner.arch }}-${{ inputs.build_type }}-artifact
        path: /tmp/neon-previous
        prefix: latest
+        aws-oicd-role-arn: ${{ inputs.aws-oicd-role-arn }}

    - name: Download compatibility snapshot
      if: inputs.build_type != 'remote'
@@ -81,6 +82,7 @@ runs:
        # The lack of compatibility snapshot (for example, for the new Postgres version)
        # shouldn't fail the whole job. Only relevant test should fail.
        skip-if-does-not-exist: true
+        aws-oicd-role-arn: ${{ inputs.aws-oicd-role-arn }}

    - name: Checkout
      if: inputs.needs_postgres_source == 'true'
@@ -218,17 +220,19 @@ runs:
        # The lack of compatibility snapshot shouldn't fail the job
        # (for example if we didn't run the test for non build-and-test workflow)
        skip-if-does-not-exist: true
+        aws-oicd-role-arn: ${{ inputs.aws-oicd-role-arn }}

-    - name: (Re-)configure AWS credentials # necessary to upload reports to S3 after a long-running test
-      if: ${{ !cancelled() && (inputs.aws_oicd_role_arn != '') }}
-      uses: aws-actions/configure-aws-credentials@v4
+    - uses: aws-actions/configure-aws-credentials@v4
+      if: ${{ !cancelled() }}
      with:
        aws-region: eu-central-1
-        role-to-assume: ${{ inputs.aws_oicd_role_arn }}
+        role-to-assume: ${{ inputs.aws-oicd-role-arn }}
        role-duration-seconds: 3600 # 1 hour should be more than enough to upload report
+
    - name: Upload test results
      if: ${{ !cancelled() }}
      uses: ./.github/actions/allure-report-store
      with:
        report-dir: /tmp/test_output/allure/results
        unique-key: ${{ inputs.build_type }}-${{ inputs.pg_version }}
+        aws-oicd-role-arn: ${{ inputs.aws-oicd-role-arn }}
--- a/.github/actions/save-coverage-data/action.yml
+++ b/.github/actions/save-coverage-data/action.yml
@@ -14,9 +14,11 @@ runs:
        name: coverage-data-artifact
        path: /tmp/coverage
        skip-if-does-not-exist: true # skip if there's no previous coverage to download
+        aws-oicd-role-arn: ${{ inputs.aws-oicd-role-arn }}

    - name: Upload coverage data
      uses: ./.github/actions/upload
      with:
        name: coverage-data-artifact
        path: /tmp/coverage
+        aws-oicd-role-arn: ${{ inputs.aws-oicd-role-arn }}
--- a/.github/actions/upload/action.yml
+++ b/.github/actions/upload/action.yml
@@ -14,6 +14,10 @@ inputs:
  prefix:
    description: "S3 prefix. Default is '${GITHUB_SHA}/${GITHUB_RUN_ID}/${GITHUB_RUN_ATTEMPT}'"
    required: false
+  aws-oicd-role-arn:
+    description: "the OIDC role arn for aws auth"
+    required: false
+    default: ""

 runs:
  using: "composite"
@@ -53,6 +57,13 @@ runs:

        echo 'SKIPPED=false' >> $GITHUB_OUTPUT

+    - name: Configure AWS credentials
+      uses: aws-actions/configure-aws-credentials@v4
+      with:
+        aws-region: eu-central-1
+        role-to-assume: ${{ inputs.aws-oicd-role-arn }}
+        role-duration-seconds: 3600
+
    - name: Upload artifact
      if: ${{ steps.prepare-artifact.outputs.SKIPPED == 'false' }}
      shell: bash -euxo pipefail {0}
--- a/.github/workflows/_benchmarking_preparation.yml
+++ b/.github/workflows/_benchmarking_preparation.yml
@@ -70,6 +70,7 @@ jobs:
        name: neon-${{ runner.os }}-${{ runner.arch }}-release-artifact
        path: /tmp/neon/
        prefix: latest
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}

    # we create a table that has one row for each database that we want to restore with the status whether the restore is done
    - name: Create benchmark_restore_status table if it does not exist
--- a/.github/workflows/_build-and-test-locally.yml
+++ b/.github/workflows/_build-and-test-locally.yml
@@ -31,12 +31,13 @@ defaults:
 env:
  RUST_BACKTRACE: 1
  COPT: '-Werror'
-  AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_DEV }}
-  AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_KEY_DEV }}

 jobs:
  build-neon:
    runs-on: ${{ fromJson(format('["self-hosted", "{0}"]', inputs.arch == 'arm64' && 'large-arm64' || 'large')) }}
+    permissions:
+      id-token: write # aws-actions/configure-aws-credentials
+      contents: read
    container:
      image: ${{ inputs.build-tools-image }}
      credentials:
@@ -205,6 +206,13 @@ jobs:
            done
          fi

+      - name: Configure AWS credentials
+        uses: aws-actions/configure-aws-credentials@v4
+        with:
+          aws-region: eu-central-1
+          role-to-assume: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+          role-duration-seconds: 18000 # 5 hours
+
      - name: Run rust tests
        env:
          NEXTEST_RETRIES: 3
@@ -256,6 +264,7 @@ jobs:
        with:
          name: neon-${{ runner.os }}-${{ runner.arch }}-${{ inputs.build-type }}-artifact
          path: /tmp/neon
+          aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}

      # XXX: keep this after the binaries.list is formed, so the coverage can properly work later
      - name: Merge and upload coverage data
@@ -265,6 +274,10 @@ jobs:
  regress-tests:
    # Don't run regression tests on debug arm64 builds
    if: inputs.build-type != 'debug' || inputs.arch != 'arm64'
+    permissions:
+      id-token: write # aws-actions/configure-aws-credentials
+      contents: read
+      statuses: write
    needs: [ build-neon ]
    runs-on: ${{ fromJson(format('["self-hosted", "{0}"]', inputs.arch == 'arm64' && 'large-arm64' || 'large')) }}
    container:
@@ -283,7 +296,7 @@ jobs:
          submodules: true

      - name: Pytest regression tests
-        continue-on-error: ${{ matrix.lfc_state == 'with-lfc' }}
+        continue-on-error: ${{ matrix.lfc_state == 'with-lfc' && inputs.build-type == 'debug' }}
        uses: ./.github/actions/run-python-test-set
        timeout-minutes: 60
        with:
@@ -295,6 +308,7 @@ jobs:
          real_s3_region: eu-central-1
          rerun_failed: true
          pg_version: ${{ matrix.pg_version }}
+          aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
        env:
          TEST_RESULT_CONNSTR: ${{ secrets.REGRESS_TEST_RESULT_CONNSTR_NEW }}
          CHECK_ONDISK_DATA_COMPATIBILITY: nonempty
--- a/.github/workflows/benchmarking.yml
+++ b/.github/workflows/benchmarking.yml
@@ -105,6 +105,7 @@ jobs:
        name: neon-${{ runner.os }}-${{ runner.arch }}-release-artifact
        path: /tmp/neon/
        prefix: latest
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}

    - name: Create Neon Project
      id: create-neon-project
@@ -122,7 +123,7 @@ jobs:
        run_in_parallel: false
        save_perf_report: ${{ env.SAVE_PERF_REPORT }}
        pg_version: ${{ env.DEFAULT_PG_VERSION }}
-        aws_oicd_role_arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
        # Set --sparse-ordering option of pytest-order plugin
        # to ensure tests are running in order of appears in the file.
        # It's important for test_perf_pgbench.py::test_pgbench_remote_* tests
@@ -152,7 +153,7 @@ jobs:
      if: ${{ !cancelled() }}
      uses: ./.github/actions/allure-report-generate
      with:
-        aws_oicd_role_arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}

    - name: Post to a Slack channel
      if: ${{ github.event.schedule && failure() }}
@@ -204,6 +205,7 @@ jobs:
        name: neon-${{ runner.os }}-${{ runner.arch }}-release-artifact
        path: /tmp/neon/
        prefix: latest
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}

    - name: Run Logical Replication benchmarks
      uses: ./.github/actions/run-python-test-set
@@ -214,7 +216,7 @@ jobs:
        save_perf_report: ${{ env.SAVE_PERF_REPORT }}
        extra_params: -m remote_cluster --timeout 5400
        pg_version: ${{ env.DEFAULT_PG_VERSION }}
-        aws_oicd_role_arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
      env:
        VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
        PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
@@ -231,7 +233,7 @@ jobs:
        save_perf_report: ${{ env.SAVE_PERF_REPORT }}
        extra_params: -m remote_cluster --timeout 5400
        pg_version: ${{ env.DEFAULT_PG_VERSION }}
-        aws_oicd_role_arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
      env:
        VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
        PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
@@ -243,7 +245,7 @@ jobs:
      uses: ./.github/actions/allure-report-generate
      with:
        store-test-results-into-db: true
-        aws_oicd_role_arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
      env:
        REGRESS_TEST_RESULT_CONNSTR_NEW: ${{ secrets.REGRESS_TEST_RESULT_CONNSTR_NEW }}

@@ -306,6 +308,7 @@ jobs:
          "image": [ "'"$image_default"'" ],
          "include": [{ "pg_version": 16, "region_id": "'"$region_id_default"'", "platform": "neonvm-captest-freetier",       "db_size": "3gb" ,"runner": '"$runner_default"', "image": "'"$image_default"'" },
                      { "pg_version": 16, "region_id": "'"$region_id_default"'", "platform": "neonvm-captest-new",            "db_size": "10gb","runner": '"$runner_default"', "image": "'"$image_default"'" },
+                      { "pg_version": 16, "region_id": "'"$region_id_default"'", "platform": "neonvm-captest-new-many-tables","db_size": "10gb","runner": '"$runner_default"', "image": "'"$image_default"'" },
                      { "pg_version": 16, "region_id": "'"$region_id_default"'", "platform": "neonvm-captest-new",            "db_size": "50gb","runner": '"$runner_default"', "image": "'"$image_default"'" },
                      { "pg_version": 16, "region_id": "azure-eastus2",          "platform": "neonvm-azure-captest-freetier", "db_size": "3gb" ,"runner": '"$runner_azure"',   "image": "neondatabase/build-tools:pinned-bookworm" },
                      { "pg_version": 16, "region_id": "azure-eastus2",          "platform": "neonvm-azure-captest-new",      "db_size": "10gb","runner": '"$runner_azure"',   "image": "neondatabase/build-tools:pinned-bookworm" },
@@ -405,9 +408,10 @@ jobs:
        name: neon-${{ runner.os }}-${{ runner.arch }}-release-artifact
        path: /tmp/neon/
        prefix: latest
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}

    - name: Create Neon Project
-      if: contains(fromJson('["neonvm-captest-new", "neonvm-captest-freetier", "neonvm-azure-captest-freetier", "neonvm-azure-captest-new"]'), matrix.platform)
+      if: contains(fromJson('["neonvm-captest-new", "neonvm-captest-new-many-tables", "neonvm-captest-freetier", "neonvm-azure-captest-freetier", "neonvm-azure-captest-new"]'), matrix.platform)
      id: create-neon-project
      uses: ./.github/actions/neon-project-create
      with:
@@ -426,7 +430,7 @@ jobs:
          neonvm-captest-sharding-reuse)
            CONNSTR=${{ secrets.BENCHMARK_CAPTEST_SHARDING_CONNSTR }}
            ;;
-          neonvm-captest-new | neonvm-captest-freetier | neonvm-azure-captest-new | neonvm-azure-captest-freetier)
+          neonvm-captest-new | neonvm-captest-new-many-tables | neonvm-captest-freetier | neonvm-azure-captest-new | neonvm-azure-captest-freetier)
            CONNSTR=${{ steps.create-neon-project.outputs.dsn }}
            ;;
          rds-aurora)
@@ -443,6 +447,26 @@ jobs:

        echo "connstr=${CONNSTR}" >> $GITHUB_OUTPUT

+    # we want to compare Neon project OLTP throughput and latency at scale factor 10 GB 
+    # without (neonvm-captest-new)
+    # and with (neonvm-captest-new-many-tables) many relations in the database
+    - name: Create many relations before the run
+      if: contains(fromJson('["neonvm-captest-new-many-tables"]'), matrix.platform)
+      uses: ./.github/actions/run-python-test-set
+      with:
+        build_type: ${{ env.BUILD_TYPE }}
+        test_selection: performance
+        run_in_parallel: false
+        save_perf_report: ${{ env.SAVE_PERF_REPORT }}
+        extra_params: -m remote_cluster --timeout 21600 -k test_perf_many_relations
+        pg_version: ${{ env.DEFAULT_PG_VERSION }}
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+      env:
+        BENCHMARK_CONNSTR: ${{ steps.set-up-connstr.outputs.connstr }}
+        VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
+        PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
+        TEST_NUM_RELATIONS: 10000
+
    - name: Benchmark init
      uses: ./.github/actions/run-python-test-set
      with:
@@ -452,7 +476,7 @@ jobs:
        save_perf_report: ${{ env.SAVE_PERF_REPORT }}
        extra_params: -m remote_cluster --timeout 21600 -k test_pgbench_remote_init
        pg_version: ${{ env.DEFAULT_PG_VERSION }}
-        aws_oicd_role_arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
      env:
        BENCHMARK_CONNSTR: ${{ steps.set-up-connstr.outputs.connstr }}
        VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
@@ -467,7 +491,7 @@ jobs:
        save_perf_report: ${{ env.SAVE_PERF_REPORT }}
        extra_params: -m remote_cluster --timeout 21600 -k test_pgbench_remote_simple_update
        pg_version: ${{ env.DEFAULT_PG_VERSION }}
-        aws_oicd_role_arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
      env:
        BENCHMARK_CONNSTR: ${{ steps.set-up-connstr.outputs.connstr }}
        VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
@@ -482,7 +506,7 @@ jobs:
        save_perf_report: ${{ env.SAVE_PERF_REPORT }}
        extra_params: -m remote_cluster --timeout 21600 -k test_pgbench_remote_select_only
        pg_version: ${{ env.DEFAULT_PG_VERSION }}
-        aws_oicd_role_arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
      env:
        BENCHMARK_CONNSTR: ${{ steps.set-up-connstr.outputs.connstr }}
        VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
@@ -500,7 +524,7 @@ jobs:
      if: ${{ !cancelled() }}
      uses: ./.github/actions/allure-report-generate
      with:
-        aws_oicd_role_arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}

    - name: Post to a Slack channel
      if: ${{ github.event.schedule && failure() }}
@@ -611,7 +635,7 @@ jobs:
        save_perf_report: ${{ env.SAVE_PERF_REPORT }}
        extra_params: -m remote_cluster --timeout 21600 -k test_pgvector_indexing
        pg_version: ${{ env.DEFAULT_PG_VERSION }}
-        aws_oicd_role_arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
      env:
        VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
        PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
@@ -626,7 +650,7 @@ jobs:
        save_perf_report: ${{ env.SAVE_PERF_REPORT }}
        extra_params: -m remote_cluster --timeout 21600
        pg_version: ${{ env.DEFAULT_PG_VERSION }}
-        aws_oicd_role_arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
      env:
        BENCHMARK_CONNSTR: ${{ steps.set-up-connstr.outputs.connstr }}
        VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
@@ -637,7 +661,7 @@ jobs:
      if: ${{ !cancelled() }}
      uses: ./.github/actions/allure-report-generate
      with:
-        aws_oicd_role_arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}

    - name: Post to a Slack channel
      if: ${{ github.event.schedule && failure() }}
@@ -708,6 +732,7 @@ jobs:
        name: neon-${{ runner.os }}-${{ runner.arch }}-release-artifact
        path: /tmp/neon/
        prefix: latest
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}

    - name: Set up Connection String
      id: set-up-connstr
@@ -739,7 +764,7 @@ jobs:
        save_perf_report: ${{ env.SAVE_PERF_REPORT }}
        extra_params: -m remote_cluster --timeout 43200 -k test_clickbench
        pg_version: ${{ env.DEFAULT_PG_VERSION }}
-        aws_oicd_role_arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
      env:
        VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
        PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
@@ -753,7 +778,7 @@ jobs:
      if: ${{ !cancelled() }}
      uses: ./.github/actions/allure-report-generate
      with:
-        aws_oicd_role_arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}

    - name: Post to a Slack channel
      if: ${{ github.event.schedule && failure() }}
@@ -818,6 +843,7 @@ jobs:
        name: neon-${{ runner.os }}-${{ runner.arch }}-release-artifact
        path: /tmp/neon/
        prefix: latest
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}

    - name: Get Connstring Secret Name
      run: |
@@ -856,7 +882,7 @@ jobs:
        save_perf_report: ${{ env.SAVE_PERF_REPORT }}
        extra_params: -m remote_cluster --timeout 21600 -k test_tpch
        pg_version: ${{ env.DEFAULT_PG_VERSION }}
-        aws_oicd_role_arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
      env:
        VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
        PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
@@ -868,7 +894,7 @@ jobs:
      if: ${{ !cancelled() }}
      uses: ./.github/actions/allure-report-generate
      with:
-        aws_oicd_role_arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}

    - name: Post to a Slack channel
      if: ${{ github.event.schedule && failure() }}
@@ -926,6 +952,7 @@ jobs:
        name: neon-${{ runner.os }}-${{ runner.arch }}-release-artifact
        path: /tmp/neon/
        prefix: latest
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}

    - name: Set up Connection String
      id: set-up-connstr
@@ -957,7 +984,7 @@ jobs:
        save_perf_report: ${{ env.SAVE_PERF_REPORT }}
        extra_params: -m remote_cluster --timeout 21600 -k test_user_examples
        pg_version: ${{ env.DEFAULT_PG_VERSION }}
-        aws_oicd_role_arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
      env:
        VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
        PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
@@ -968,7 +995,7 @@ jobs:
      if: ${{ !cancelled() }}
      uses: ./.github/actions/allure-report-generate
      with:
-        aws_oicd_role_arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}

    - name: Post to a Slack channel
      if: ${{ github.event.schedule && failure() }}
--- a/.github/workflows/build_and_test.yml
+++ b/.github/workflows/build_and_test.yml
@@ -21,8 +21,6 @@ concurrency:
 env:
  RUST_BACKTRACE: 1
  COPT: '-Werror'
-  AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_DEV }}
-  AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_KEY_DEV }}
  # A concurrency group that we use for e2e-tests runs, matches `concurrency.group` above with `github.repository` as a prefix
  E2E_CONCURRENCY_GROUP: ${{ github.repository }}-e2e-tests-${{ github.ref_name }}-${{ github.ref_name == 'main' && github.sha || 'anysha' }}

@@ -255,15 +253,15 @@ jobs:
      build-tools-image: ${{ needs.build-build-tools-image.outputs.image }}-bookworm
      build-tag: ${{ needs.tag.outputs.build-tag }}
      build-type: ${{ matrix.build-type }}
-      # Run tests on all Postgres versions in release builds and only on the latest version in debug builds
-      # run without LFC on v17 release only
+      # Run tests on all Postgres versions in release builds and only on the latest version in debug builds.
+      # Run without LFC on v17 release and debug builds only. For all the other cases LFC is enabled.
      test-cfg: |
-        ${{ matrix.build-type == 'release' && '[{"pg_version":"v14", "lfc_state": "without-lfc"},
-                                                {"pg_version":"v15", "lfc_state": "without-lfc"},
-                                                {"pg_version":"v16", "lfc_state": "without-lfc"},
-                                                {"pg_version":"v17", "lfc_state": "without-lfc"},
-                                                {"pg_version":"v17", "lfc_state": "with-lfc"}]'
-                                           || '[{"pg_version":"v17", "lfc_state": "without-lfc"}]' }}
+        ${{ matrix.build-type == 'release' && '[{"pg_version":"v14", "lfc_state": "with-lfc"},
+                                                {"pg_version":"v15", "lfc_state": "with-lfc"},
+                                                {"pg_version":"v16", "lfc_state": "with-lfc"},
+                                                {"pg_version":"v17", "lfc_state": "with-lfc"},
+                                                {"pg_version":"v17", "lfc_state": "without-lfc"}]'
+                                           || '[{"pg_version":"v17", "lfc_state": "without-lfc" }]' }}
    secrets: inherit

  # Keep `benchmarks` job outside of `build-and-test-locally` workflow to make job failures non-blocking
@@ -305,6 +303,11 @@ jobs:
  benchmarks:
    if: github.ref_name == 'main' || contains(github.event.pull_request.labels.*.name, 'run-benchmarks')
    needs: [ check-permissions, build-and-test-locally, build-build-tools-image, get-benchmarks-durations ]
+    permissions:
+      id-token: write # aws-actions/configure-aws-credentials
+      statuses: write
+      contents: write
+      pull-requests: write
    runs-on: [ self-hosted, small ]
    container:
      image: ${{ needs.build-build-tools-image.outputs.image }}-bookworm
@@ -333,6 +336,7 @@ jobs:
          extra_params: --splits 5 --group ${{ matrix.pytest_split_group }}
          benchmark_durations: ${{ needs.get-benchmarks-durations.outputs.json }}
          pg_version: v16
+          aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
        env:
          VIP_VAP_ACCESS_TOKEN: "${{ secrets.VIP_VAP_ACCESS_TOKEN }}"
          PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
@@ -345,6 +349,11 @@ jobs:
  report-benchmarks-failures:
    needs: [ benchmarks, create-test-report ]
    if: github.ref_name == 'main' && failure() && needs.benchmarks.result == 'failure'
+    permissions:
+      id-token: write # aws-actions/configure-aws-credentials
+      statuses: write
+      contents: write
+      pull-requests: write
    runs-on: ubuntu-22.04

    steps:
@@ -360,6 +369,11 @@ jobs:
  create-test-report:
    needs: [ check-permissions, build-and-test-locally, coverage-report, build-build-tools-image, benchmarks ]
    if: ${{ !cancelled() && contains(fromJSON('["skipped", "success"]'), needs.check-permissions.result) }}
+    permissions:
+      id-token: write # aws-actions/configure-aws-credentials
+      statuses: write
+      contents: write
+      pull-requests: write
    outputs:
      report-url: ${{ steps.create-allure-report.outputs.report-url }}

@@ -380,6 +394,7 @@ jobs:
        uses: ./.github/actions/allure-report-generate
        with:
          store-test-results-into-db: true
+          aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
        env:
          REGRESS_TEST_RESULT_CONNSTR_NEW: ${{ secrets.REGRESS_TEST_RESULT_CONNSTR_NEW }}

@@ -411,6 +426,10 @@ jobs:
  coverage-report:
    if: ${{ !startsWith(github.ref_name, 'release') }}
    needs: [ check-permissions, build-build-tools-image, build-and-test-locally ]
+    permissions:
+      id-token: write # aws-actions/configure-aws-credentials
+      statuses: write
+      contents: write
    runs-on: [ self-hosted, small ]
    container:
      image: ${{ needs.build-build-tools-image.outputs.image }}-bookworm
@@ -437,12 +456,14 @@ jobs:
        with:
          name: neon-${{ runner.os }}-${{ runner.arch }}-${{ matrix.build_type }}-artifact
          path: /tmp/neon
+          aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}

      - name: Get coverage artifact
        uses: ./.github/actions/download
        with:
          name: coverage-data-artifact
          path: /tmp/coverage
+          aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}

      - name: Merge coverage data
        run: scripts/coverage "--profraw-prefix=$GITHUB_JOB" --dir=/tmp/coverage merge
@@ -573,6 +594,10 @@ jobs:
  neon-image:
    needs: [ neon-image-arch, tag ]
    runs-on: ubuntu-22.04
+    permissions:
+      id-token: write # aws-actions/configure-aws-credentials
+      statuses: write
+      contents: read

    steps:
      - uses: docker/login-action@v3
@@ -587,11 +612,15 @@ jobs:
                                             neondatabase/neon:${{ needs.tag.outputs.build-tag }}-bookworm-x64 \
                                             neondatabase/neon:${{ needs.tag.outputs.build-tag }}-bookworm-arm64

-      - uses: docker/login-action@v3
+      - name: Configure AWS credentials
+        uses: aws-actions/configure-aws-credentials@v4
        with:
-          registry: 369495373322.dkr.ecr.eu-central-1.amazonaws.com
-          username: ${{ secrets.AWS_ACCESS_KEY_DEV }}
-          password: ${{ secrets.AWS_SECRET_KEY_DEV }}
+          aws-region: eu-central-1
+          role-to-assume: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+          role-duration-seconds: 3600
+
+      - name: Login to Amazon Dev ECR
+        uses: aws-actions/amazon-ecr-login@v2

      - name: Push multi-arch image to ECR
        run: |
@@ -600,6 +629,10 @@ jobs:

  compute-node-image-arch:
    needs: [ check-permissions, build-build-tools-image, tag ]
+    permissions:
+      id-token: write # aws-actions/configure-aws-credentials
+      statuses: write
+      contents: read
    strategy:
      fail-fast: false
      matrix:
@@ -640,11 +673,15 @@ jobs:
          username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
          password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}

-      - uses: docker/login-action@v3
+      - name: Configure AWS credentials
+        uses: aws-actions/configure-aws-credentials@v4
        with:
-          registry: 369495373322.dkr.ecr.eu-central-1.amazonaws.com
-          username: ${{ secrets.AWS_ACCESS_KEY_DEV }}
-          password: ${{ secrets.AWS_SECRET_KEY_DEV }}
+          aws-region: eu-central-1
+          role-to-assume: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+          role-duration-seconds: 3600
+
+      - name: Login to Amazon Dev ECR
+        uses: aws-actions/amazon-ecr-login@v2

      - uses: docker/login-action@v3
        with:
@@ -717,6 +754,10 @@ jobs:

  compute-node-image:
    needs: [ compute-node-image-arch, tag ]
+    permissions:
+      id-token: write # aws-actions/configure-aws-credentials
+      statuses: write
+      contents: read
    runs-on: ubuntu-22.04

    strategy:
@@ -761,11 +802,15 @@ jobs:
                                             neondatabase/compute-tools:${{ needs.tag.outputs.build-tag }}-${{ matrix.version.debian }}-x64 \
                                             neondatabase/compute-tools:${{ needs.tag.outputs.build-tag }}-${{ matrix.version.debian }}-arm64

-      - uses: docker/login-action@v3
+      - name: Configure AWS credentials
+        uses: aws-actions/configure-aws-credentials@v4
        with:
-          registry: 369495373322.dkr.ecr.eu-central-1.amazonaws.com
-          username: ${{ secrets.AWS_ACCESS_KEY_DEV }}
-          password: ${{ secrets.AWS_SECRET_KEY_DEV }}
+          aws-region: eu-central-1
+          role-to-assume: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+          role-duration-seconds: 3600
+
+      - name: Login to Amazon Dev ECR
+        uses: aws-actions/amazon-ecr-login@v2

      - name: Push multi-arch compute-node-${{ matrix.version.pg }} image to ECR
        run: |
@@ -795,7 +840,7 @@ jobs:
          - pg: v17
            debian: bookworm
    env:
-      VM_BUILDER_VERSION: v0.35.0
+      VM_BUILDER_VERSION: v0.37.1

    steps:
      - uses: actions/checkout@v4
@@ -890,7 +935,9 @@ jobs:
    runs-on: ubuntu-22.04

    permissions:
-      id-token: write # for `aws-actions/configure-aws-credentials`
+      id-token: write # aws-actions/configure-aws-credentials
+      statuses: write
+      contents: read

    env:
      VERSIONS: v14 v15 v16 v17
@@ -901,12 +948,15 @@ jobs:
          username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
          password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}

-      - name: Login to dev ECR
-        uses: docker/login-action@v3
+      - name: Configure AWS credentials
+        uses: aws-actions/configure-aws-credentials@v4
        with:
-          registry: 369495373322.dkr.ecr.eu-central-1.amazonaws.com
-          username: ${{ secrets.AWS_ACCESS_KEY_DEV }}
-          password: ${{ secrets.AWS_SECRET_KEY_DEV }}
+          aws-region: eu-central-1
+          role-to-assume: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+          role-duration-seconds: 3600
+
+      - name: Login to Amazon Dev ECR
+        uses: aws-actions/amazon-ecr-login@v2

      - name: Copy vm-compute-node images to ECR
        run: |
@@ -985,6 +1035,11 @@ jobs:
  trigger-custom-extensions-build-and-wait:
    needs: [ check-permissions, tag ]
    runs-on: ubuntu-22.04
+    permissions:
+      id-token: write # aws-actions/configure-aws-credentials
+      statuses: write
+      contents: write
+      pull-requests: write
    steps:
      - name: Set PR's status to pending and request a remote CI test
        run: |
@@ -1060,12 +1115,79 @@ jobs:
    needs: [ check-permissions, promote-images, tag, build-and-test-locally, trigger-custom-extensions-build-and-wait, push-to-acr-dev, push-to-acr-prod ]
    # `!failure() && !cancelled()` is required because the workflow depends on the job that can be skipped: `push-to-acr-dev` and `push-to-acr-prod`
    if: (github.ref_name == 'main' || github.ref_name == 'release' || github.ref_name == 'release-proxy' || github.ref_name == 'release-compute') && !failure() && !cancelled()
-
+    permissions:
+      id-token: write # aws-actions/configure-aws-credentials
+      statuses: write
+      contents: write
    runs-on: [ self-hosted, small ]
    container: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/ansible:latest
    steps:
      - uses: actions/checkout@v4

+      - name: Create git tag and GitHub release
+        if: github.ref_name == 'release' || github.ref_name == 'release-proxy' || github.ref_name == 'release-compute'
+        uses: actions/github-script@v7
+        with:
+          retries: 5
+          script: |
+            const tag = "${{ needs.tag.outputs.build-tag }}";
+
+            try {
+              const existingRef = await github.rest.git.getRef({
+                owner: context.repo.owner,
+                repo: context.repo.repo,
+                ref: `tags/${tag}`,
+              });
+
+              if (existingRef.data.object.sha !== context.sha) {
+                throw new Error(`Tag ${tag} already exists but points to a different commit (expected: ${context.sha}, actual: ${existingRef.data.object.sha}).`);
+              }
+
+              console.log(`Tag ${tag} already exists and points to ${context.sha} as expected.`);
+            } catch (error) {
+              if (error.status !== 404) {
+                throw error;
+              }
+
+              console.log(`Tag ${tag} does not exist. Creating it...`);
+              await github.rest.git.createRef({
+                owner: context.repo.owner,
+                repo: context.repo.repo,
+                ref: `refs/tags/${tag}`,
+                sha: context.sha,
+              });
+              console.log(`Tag ${tag} created successfully.`);
+            }
+
+            // TODO: check how GitHub releases looks for proxy/compute releases and enable them if they're ok
+            if (context.ref !== 'refs/heads/release') {
+              console.log(`GitHub release skipped for ${context.ref}.`);
+              return;
+            }
+
+            try {
+              const existingRelease = await github.rest.repos.getReleaseByTag({
+                owner: context.repo.owner,
+                repo: context.repo.repo,
+                tag: tag,
+              });
+
+              console.log(`Release for tag ${tag} already exists (ID: ${existingRelease.data.id}).`);
+            } catch (error) {
+              if (error.status !== 404) {
+                throw error;
+              }
+
+              console.log(`Release for tag ${tag} does not exist. Creating it...`);
+              await github.rest.repos.createRelease({
+                owner: context.repo.owner,
+                repo: context.repo.repo,
+                tag_name: tag,
+                generate_release_notes: true,
+              });
+              console.log(`Release for tag ${tag} created successfully.`);
+            }
+
      - name: Trigger deploy workflow
        env:
          GH_TOKEN: ${{ secrets.CI_ACCESS_TOKEN }}
@@ -1115,38 +1237,13 @@ jobs:
            exit 1
          fi

-      - name: Create git tag
-        if: github.ref_name == 'release' || github.ref_name == 'release-proxy' || github.ref_name == 'release-compute'
-        uses: actions/github-script@v7
-        with:
-          # Retry script for 5XX server errors: https://github.com/actions/github-script#retries
-          retries: 5
-          script: |
-            await github.rest.git.createRef({
-              owner: context.repo.owner,
-              repo: context.repo.repo,
-              ref: "refs/tags/${{ needs.tag.outputs.build-tag }}",
-              sha: context.sha,
-            })
-
-      # TODO: check how GitHub releases looks for proxy releases and enable it if it's ok
-      - name: Create GitHub release
-        if: github.ref_name == 'release'
-        uses: actions/github-script@v7
-        with:
-          # Retry script for 5XX server errors: https://github.com/actions/github-script#retries
-          retries: 5
-          script: |
-            await github.rest.repos.createRelease({
-              owner: context.repo.owner,
-              repo: context.repo.repo,
-              tag_name: "${{ needs.tag.outputs.build-tag }}",
-              generate_release_notes: true,
-            })
-
  # The job runs on `release` branch and copies compatibility data and Neon artifact from the last *release PR* to the latest directory
  promote-compatibility-data:
    needs: [ deploy ]
+    permissions:
+      id-token: write # aws-actions/configure-aws-credentials
+      statuses: write
+      contents: read
    # `!failure() && !cancelled()` is required because the workflow transitively depends on the job that can be skipped: `push-to-acr-dev` and `push-to-acr-prod`
    if: github.ref_name == 'release' && !failure() && !cancelled()

@@ -1183,6 +1280,12 @@ jobs:
          echo "run-id=${run_id}" | tee -a ${GITHUB_OUTPUT}
          echo "commit-sha=${last_commit_sha}" | tee -a ${GITHUB_OUTPUT}

+      - 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: 3600
+
      - name: Promote compatibility snapshot and Neon artifact
        env:
          BUCKET: neon-github-public-dev
--- a/.github/workflows/cloud-regress.yml
+++ b/.github/workflows/cloud-regress.yml
@@ -19,15 +19,21 @@ concurrency:
  group: ${{ github.workflow }}
  cancel-in-progress: true

+permissions:
+  id-token: write # aws-actions/configure-aws-credentials
+  statuses: write
+  contents: write
+
 jobs:
  regress:
    env:
      POSTGRES_DISTRIB_DIR: /tmp/neon/pg_install
-      DEFAULT_PG_VERSION: 16
      TEST_OUTPUT: /tmp/test_output
      BUILD_TYPE: remote
-      AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_DEV }}
-      AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_KEY_DEV }}
+    strategy:
+      fail-fast: false
+      matrix:
+        pg-version: [16, 17]

    runs-on: us-east-2
    container:
@@ -40,9 +46,11 @@ jobs:
          submodules: true

      - name: Patch the test
+        env:
+          PG_VERSION: ${{matrix.pg-version}}
        run: |
-          cd "vendor/postgres-v${DEFAULT_PG_VERSION}"
-          patch -p1 < "../../compute/patches/cloud_regress_pg${DEFAULT_PG_VERSION}.patch"
+          cd "vendor/postgres-v${PG_VERSION}"
+          patch -p1 < "../../compute/patches/cloud_regress_pg${PG_VERSION}.patch"

      - name: Generate a random password
        id: pwgen
@@ -55,8 +63,9 @@ jobs:
      - name: Change tests according to the generated password
        env:
          DBPASS: ${{ steps.pwgen.outputs.DBPASS }}
+          PG_VERSION: ${{matrix.pg-version}}
        run: |
-          cd vendor/postgres-v"${DEFAULT_PG_VERSION}"/src/test/regress
+          cd vendor/postgres-v"${PG_VERSION}"/src/test/regress
          for fname in sql/*.sql expected/*.out; do
            sed -i.bak s/NEON_PASSWORD_PLACEHOLDER/"'${DBPASS}'"/ "${fname}"
          done
@@ -72,27 +81,46 @@ jobs:
          name: neon-${{ runner.os }}-${{ runner.arch }}-release-artifact
          path: /tmp/neon/
          prefix: latest
+          aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+
+      - name: Create a new branch
+        id: create-branch
+        uses: ./.github/actions/neon-branch-create
+        with:
+          api_key: ${{ secrets.NEON_STAGING_API_KEY }}
+          project_id: ${{ vars[format('PGREGRESS_PG{0}_PROJECT_ID', matrix.pg-version)] }}

      - name: Run the regression tests
        uses: ./.github/actions/run-python-test-set
        with:
          build_type: ${{ env.BUILD_TYPE }}
          test_selection: cloud_regress
-          pg_version: ${{ env.DEFAULT_PG_VERSION }}
+          pg_version: ${{matrix.pg-version}}
          extra_params: -m remote_cluster
+          aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
        env:
-          BENCHMARK_CONNSTR: ${{ secrets.PG_REGRESS_CONNSTR }}
+          BENCHMARK_CONNSTR: ${{steps.create-branch.outputs.dsn}}
+
+      - name: Delete branch
+        if: always()
+        uses: ./.github/actions/neon-branch-delete
+        with:
+          api_key: ${{ secrets.NEON_STAGING_API_KEY }}
+          project_id: ${{ vars[format('PGREGRESS_PG{0}_PROJECT_ID', matrix.pg-version)] }}
+          branch_id: ${{steps.create-branch.outputs.branch_id}}

      - name: Create Allure report
        id: create-allure-report
        if: ${{ !cancelled() }}
        uses: ./.github/actions/allure-report-generate
+        with:
+          aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}

      - name: Post to a Slack channel
        if: ${{ github.event.schedule && failure() }}
        uses: slackapi/slack-github-action@v1
        with:
-          channel-id: "C033QLM5P7D" # on-call-staging-stream
+          channel-id: ${{ vars.SLACK_ON_CALL_QA_STAGING_STREAM }}
          slack-message: |
            Periodic pg_regress on staging: ${{ job.status }}
            <${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}|GitHub Run>
--- a/.github/workflows/ingest_benchmark.yml
+++ b/.github/workflows/ingest_benchmark.yml
@@ -13,7 +13,7 @@ on:
    #          │ │ │ │ ┌───────────── 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}
@@ -28,7 +28,7 @@ jobs:
    strategy:
      fail-fast: false # allow other variants to continue even if one fails
      matrix:
-        target_project: [new_empty_project, large_existing_project]  
+        target_project: [new_empty_project, large_existing_project]
    permissions:
      contents: write
      statuses: write
@@ -56,7 +56,7 @@ jobs:
      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 
+        role-duration-seconds: 18000 # 5 hours is currently max associated with IAM role

    - name: Download Neon artifact
      uses: ./.github/actions/download
@@ -64,6 +64,7 @@ jobs:
        name: neon-${{ runner.os }}-${{ runner.arch }}-release-artifact
        path: /tmp/neon/
        prefix: latest
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}

    - name: Create Neon Project
      if: ${{ matrix.target_project == 'new_empty_project' }}
@@ -94,7 +95,7 @@ jobs:
        project_id: ${{ vars.BENCHMARK_INGEST_TARGET_PROJECTID }}
        api_key: ${{ secrets.NEON_STAGING_API_KEY }}

-    - name: Initialize Neon project 
+    - name: Initialize Neon project
      if: ${{ matrix.target_project == 'large_existing_project' }}
      env:
          BENCHMARK_INGEST_TARGET_CONNSTR: ${{ steps.create-neon-branch-ingest-target.outputs.dsn }}
@@ -122,7 +123,7 @@ jobs:
        ${PSQL} "${BENCHMARK_INGEST_TARGET_CONNSTR}" -c "CREATE EXTENSION IF NOT EXISTS neon; CREATE EXTENSION IF NOT EXISTS neon_utils;"
        echo "BENCHMARK_INGEST_TARGET_CONNSTR=${BENCHMARK_INGEST_TARGET_CONNSTR}" >> $GITHUB_ENV

-    - name: Invoke pgcopydb  
+    - name: Invoke pgcopydb
      uses: ./.github/actions/run-python-test-set
      with:
        build_type: remote
@@ -131,7 +132,7 @@ jobs:
        extra_params: -s -m remote_cluster --timeout 86400 -k test_ingest_performance_using_pgcopydb
        pg_version: v16
        save_perf_report: true
-        aws_oicd_role_arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
      env:
        BENCHMARK_INGEST_SOURCE_CONNSTR: ${{ secrets.BENCHMARK_INGEST_SOURCE_CONNSTR }}
        TARGET_PROJECT_TYPE: ${{ matrix.target_project }}
@@ -143,7 +144,7 @@ jobs:
      run: |
        export LD_LIBRARY_PATH=${PG_16_LIB_PATH}
        ${PSQL} "${BENCHMARK_INGEST_TARGET_CONNSTR}" -c "\dt+"
-      
+
    - name: Delete Neon Project
      if: ${{ always() && matrix.target_project == 'new_empty_project' }}
      uses: ./.github/actions/neon-project-delete
--- a/.github/workflows/neon_extra_builds.yml
+++ b/.github/workflows/neon_extra_builds.yml
@@ -143,6 +143,10 @@ jobs:

  gather-rust-build-stats:
    needs: [ check-permissions, build-build-tools-image ]
+    permissions:
+      id-token: write # aws-actions/configure-aws-credentials
+      statuses: write
+      contents: write
    if: |
      contains(github.event.pull_request.labels.*.name, 'run-extra-build-stats') ||
      contains(github.event.pull_request.labels.*.name, 'run-extra-build-*') ||
@@ -177,13 +181,18 @@ jobs:
      - name: Produce the build stats
        run: PQ_LIB_DIR=$(pwd)/pg_install/v17/lib cargo build --all --release --timings -j$(nproc)

+      - name: Configure AWS credentials
+        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: 3600
+
      - name: Upload the build stats
        id: upload-stats
        env:
          BUCKET: neon-github-public-dev
          SHA: ${{ github.event.pull_request.head.sha || github.sha }}
-          AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_DEV }}
-          AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_KEY_DEV }}
        run: |
          REPORT_URL=https://${BUCKET}.s3.amazonaws.com/build-stats/${SHA}/${GITHUB_RUN_ID}/cargo-timing.html
          aws s3 cp --only-show-errors ./target/cargo-timings/cargo-timing.html "s3://${BUCKET}/build-stats/${SHA}/${GITHUB_RUN_ID}/"
--- a/.github/workflows/periodic_pagebench.yml
+++ b/.github/workflows/periodic_pagebench.yml
@@ -27,6 +27,11 @@ concurrency:

 jobs:
  trigger_bench_on_ec2_machine_in_eu_central_1:
+    permissions:
+      id-token: write # aws-actions/configure-aws-credentials
+      statuses: write
+      contents: write
+      pull-requests: write
    runs-on: [ self-hosted, small ]
    container:
      image: neondatabase/build-tools:pinned-bookworm
@@ -38,8 +43,6 @@ jobs:
    env:
      API_KEY: ${{ secrets.PERIODIC_PAGEBENCH_EC2_RUNNER_API_KEY }}
      RUN_ID: ${{ github.run_id }}
-      AWS_ACCESS_KEY_ID: ${{ secrets.AWS_EC2_US_TEST_RUNNER_ACCESS_KEY_ID }}
-      AWS_SECRET_ACCESS_KEY : ${{ secrets.AWS_EC2_US_TEST_RUNNER_ACCESS_KEY_SECRET }}
      AWS_DEFAULT_REGION : "eu-central-1"
      AWS_INSTANCE_ID : "i-02a59a3bf86bc7e74"
    steps:
@@ -50,6 +53,13 @@ jobs:
    - name: Show my own (github runner) external IP address - usefull for IP allowlisting
      run: curl https://ifconfig.me

+    - name: Assume AWS OIDC role that allows to manage (start/stop/describe... EC machine)
+      uses: aws-actions/configure-aws-credentials@v4
+      with:
+        aws-region: eu-central-1
+        role-to-assume: ${{ vars.DEV_AWS_OIDC_ROLE_MANAGE_BENCHMARK_EC2_VMS_ARN }}
+        role-duration-seconds: 3600
+
    - name: Start EC2 instance and wait for the instance to boot up
      run: |
        aws ec2 start-instances --instance-ids $AWS_INSTANCE_ID
@@ -124,11 +134,10 @@ jobs:
        cat "test_log_${GITHUB_RUN_ID}"

    - name: Create Allure report
-      env:
-        AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_DEV }}
-        AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_KEY_DEV }}
      if: ${{ !cancelled() }}
      uses: ./.github/actions/allure-report-generate
+      with:
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}

    - name: Post to a Slack channel
      if: ${{ github.event.schedule && failure() }}
@@ -148,6 +157,14 @@ jobs:
        -H "Authorization: Bearer $API_KEY" \
        -d ''

+    - name: Assume AWS OIDC role that allows to manage (start/stop/describe... EC machine)
+      if: always() && steps.poll_step.outputs.too_many_runs != 'true'
+      uses: aws-actions/configure-aws-credentials@v4
+      with:
+        aws-region: eu-central-1
+        role-to-assume: ${{ vars.DEV_AWS_OIDC_ROLE_MANAGE_BENCHMARK_EC2_VMS_ARN }}
+        role-duration-seconds: 3600
+
    - name: Stop EC2 instance and wait for the instance to be stopped
      if: always() && steps.poll_step.outputs.too_many_runs != 'true'
      run: |
--- a/.github/workflows/pg-clients.yml
+++ b/.github/workflows/pg-clients.yml
@@ -25,11 +25,13 @@ defaults:
  run:
    shell: bash -euxo pipefail {0}

+permissions:
+  id-token: write # aws-actions/configure-aws-credentials
+  statuses: write # require for posting a status update
+
 env:
  DEFAULT_PG_VERSION: 16
  PLATFORM: neon-captest-new
-  AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_DEV }}
-  AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_KEY_DEV }}
  AWS_DEFAULT_REGION: eu-central-1

 jobs:
@@ -94,6 +96,7 @@ jobs:
          name: neon-${{ runner.os }}-${{ runner.arch }}-release-artifact
          path: /tmp/neon/
          prefix: latest
+          aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}

      - name: Create Neon Project
        id: create-neon-project
@@ -110,6 +113,7 @@ jobs:
          run_in_parallel: false
          extra_params: -m remote_cluster
          pg_version: ${{ env.DEFAULT_PG_VERSION }}
+          aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
        env:
          BENCHMARK_CONNSTR: ${{ steps.create-neon-project.outputs.dsn }}

@@ -126,6 +130,7 @@ jobs:
        uses: ./.github/actions/allure-report-generate
        with:
          store-test-results-into-db: true
+          aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
        env:
          REGRESS_TEST_RESULT_CONNSTR_NEW: ${{ secrets.REGRESS_TEST_RESULT_CONNSTR_NEW }}

@@ -159,6 +164,7 @@ jobs:
        name: neon-${{ runner.os }}-${{ runner.arch }}-release-artifact
        path: /tmp/neon/
        prefix: latest
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}

    - name: Create Neon Project
      id: create-neon-project
@@ -175,6 +181,7 @@ jobs:
        run_in_parallel: false
        extra_params: -m remote_cluster
        pg_version: ${{ env.DEFAULT_PG_VERSION }}
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
      env:
        BENCHMARK_CONNSTR: ${{ steps.create-neon-project.outputs.dsn }}

@@ -191,6 +198,7 @@ jobs:
      uses: ./.github/actions/allure-report-generate
      with:
        store-test-results-into-db: true
+        aws-oicd-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
      env:
        REGRESS_TEST_RESULT_CONNSTR_NEW: ${{ secrets.REGRESS_TEST_RESULT_CONNSTR_NEW }}

--- a/.github/workflows/pin-build-tools-image.yml
+++ b/.github/workflows/pin-build-tools-image.yml
@@ -67,7 +67,7 @@ jobs:
    runs-on: ubuntu-22.04

    permissions:
-      id-token: write # for `azure/login`
+      id-token: write # for `azure/login` and aws auth

    steps:
      - uses: docker/login-action@v3
@@ -75,11 +75,15 @@ jobs:
          username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
          password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}

-      - uses: docker/login-action@v3
+      - name: Configure AWS credentials
+        uses: aws-actions/configure-aws-credentials@v4
        with:
-          registry: 369495373322.dkr.ecr.eu-central-1.amazonaws.com
-          username: ${{ secrets.AWS_ACCESS_KEY_DEV }}
-          password: ${{ secrets.AWS_SECRET_KEY_DEV }}
+          aws-region: eu-central-1
+          role-to-assume: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+          role-duration-seconds: 3600
+
+      - name: Login to Amazon Dev ECR
+        uses: aws-actions/amazon-ecr-login@v2

      - name: Azure login
        uses: azure/login@6c251865b4e6290e7b78be643ea2d005bc51f69a  # @v2.1.1
--- a/.github/workflows/pre-merge-checks.yml
+++ b/.github/workflows/pre-merge-checks.yml
@@ -63,6 +63,7 @@ jobs:
    if: always()
    permissions:
      statuses: write # for `github.repos.createCommitStatus(...)`
+      contents: write
    needs:
      - get-changed-files
      - check-codestyle-python
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -3,7 +3,7 @@ name: Create Release Branch
 on:
  schedule:
    # It should be kept in sync with if-condition in jobs
-    - cron: '0 6 * * MON' # Storage release
+    - cron: '0 6 * * FRI' # Storage release
    - cron: '0 6 * * THU' # Proxy release
  workflow_dispatch:
    inputs:
@@ -29,7 +29,7 @@ defaults:

 jobs:
  create-storage-release-branch:
-    if: ${{ github.event.schedule == '0 6 * * MON' || inputs.create-storage-release-branch }}
+    if: ${{ github.event.schedule == '0 6 * * FRI' || inputs.create-storage-release-branch }}

    permissions:
      contents: write
--- a/33
+++ b/33
@@ -1,16 +1,29 @@
-/.github/ @neondatabase/developer-productivity
-/compute_tools/ @neondatabase/control-plane @neondatabase/compute
-/libs/pageserver_api/ @neondatabase/storage
-/libs/postgres_ffi/ @neondatabase/compute @neondatabase/storage
-/libs/proxy/ @neondatabase/proxy
-/libs/remote_storage/ @neondatabase/storage
-/libs/safekeeper_api/ @neondatabase/storage
+# Autoscaling
 /libs/vm_monitor/ @neondatabase/autoscaling
-/pageserver/ @neondatabase/storage
+
+# DevProd
+/.github/ @neondatabase/developer-productivity
+
+# Compute
 /pgxn/ @neondatabase/compute
-/pgxn/neon/ @neondatabase/compute @neondatabase/storage
+/vendor/ @neondatabase/compute
+/compute/ @neondatabase/compute
+/compute_tools/ @neondatabase/compute
+
+# Proxy
+/libs/proxy/ @neondatabase/proxy
 /proxy/ @neondatabase/proxy
+
+# Storage
+/pageserver/ @neondatabase/storage
 /safekeeper/ @neondatabase/storage
 /storage_controller @neondatabase/storage
 /storage_scrubber @neondatabase/storage
-/vendor/ @neondatabase/compute
+/libs/pageserver_api/ @neondatabase/storage
+/libs/remote_storage/ @neondatabase/storage
+/libs/safekeeper_api/ @neondatabase/storage
+
+# Shared
+/pgxn/neon/ @neondatabase/compute @neondatabase/storage
+/libs/compute_api/ @neondatabase/compute @neondatabase/control-plane
+/libs/postgres_ffi/ @neondatabase/compute @neondatabase/storage
--- a/Cargo.lock
+++ b/Cargo.lock
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -11,6 +11,7 @@ members = [
    "pageserver/pagebench",
    "proxy",
    "safekeeper",
+    "safekeeper/client",
    "storage_broker",
    "storage_controller",
    "storage_controller/client",
@@ -51,10 +52,7 @@ anyhow = { version = "1.0", features = ["backtrace"] }
 arc-swap = "1.6"
 async-compression = { version = "0.4.0", features = ["tokio", "gzip", "zstd"] }
 atomic-take = "1.1.0"
-azure_core = { version = "0.19", default-features = false, features = ["enable_reqwest_rustls", "hmac_rust"] }
-azure_identity = { version = "0.19", default-features = false, features = ["enable_reqwest_rustls"] }
-azure_storage = { version = "0.19", default-features = false, features = ["enable_reqwest_rustls"] }
-azure_storage_blobs = { version = "0.19", default-features = false, features = ["enable_reqwest_rustls"] }
+backtrace = "0.3.74"
 flate2 = "1.0.26"
 async-stream = "0.3"
 async-trait = "0.1"
@@ -216,6 +214,12 @@ postgres-protocol = { git = "https://github.com/neondatabase/rust-postgres.git",
 postgres-types = { git = "https://github.com/neondatabase/rust-postgres.git", branch = "neon" }
 tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", branch = "neon" }

+## Azure SDK crates
+azure_core = { git = "https://github.com/neondatabase/azure-sdk-for-rust.git", branch = "neon", default-features = false, features = ["enable_reqwest_rustls", "hmac_rust"] }
+azure_identity = { git = "https://github.com/neondatabase/azure-sdk-for-rust.git", branch = "neon", default-features = false, features = ["enable_reqwest_rustls"] }
+azure_storage = { git = "https://github.com/neondatabase/azure-sdk-for-rust.git", branch = "neon", default-features = false, features = ["enable_reqwest_rustls"] }
+azure_storage_blobs = { git = "https://github.com/neondatabase/azure-sdk-for-rust.git", branch = "neon", default-features = false, features = ["enable_reqwest_rustls"] }
+
 ## Local libraries
 compute_api = { version = "0.1", path = "./libs/compute_api/" }
 consumption_metrics = { version = "0.1", path = "./libs/consumption_metrics/" }
@@ -231,6 +235,7 @@ postgres_initdb = { path = "./libs/postgres_initdb" }
 pq_proto = { version = "0.1", path = "./libs/pq_proto/" }
 remote_storage = { version = "0.1", path = "./libs/remote_storage/" }
 safekeeper_api = { version = "0.1", path = "./libs/safekeeper_api" }
+safekeeper_client = { path = "./safekeeper/client" }
 desim = { version = "0.1", path = "./libs/desim" }
 storage_broker = { version = "0.1", path = "./storage_broker/" } # Note: main broker code is inside the binary crate, so linking with the library shouldn't be heavy.
 storage_controller_client = { path = "./storage_controller/client" }
--- a/2
+++ b/2
@@ -69,6 +69,8 @@ RUN set -e \
        libreadline-dev \
        libseccomp-dev \
        ca-certificates \
+	# System postgres for use with client libraries (e.g. in storage controller)
+        postgresql-15 \
    && rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/* \
    && useradd -d /data neon \
    && chown -R neon:neon /data
--- a/2
+++ b/2
@@ -67,8 +67,6 @@ CARGO_BUILD_FLAGS += $(filter -j1,$(MAKEFLAGS))
 CARGO_CMD_PREFIX += $(if $(filter n,$(MAKEFLAGS)),,+)
 # Force cargo not to print progress bar
 CARGO_CMD_PREFIX += CARGO_TERM_PROGRESS_WHEN=never CI=1
-# Set PQ_LIB_DIR to make sure `storage_controller` get linked with bundled libpq (through diesel)
-CARGO_CMD_PREFIX += PQ_LIB_DIR=$(POSTGRES_INSTALL_DIR)/v16/lib

 CACHEDIR_TAG_CONTENTS := "Signature: 8a477f597d28d172789f06886806bc55"

--- a/build-tools.Dockerfile
+++ b/build-tools.Dockerfile
@@ -115,7 +115,7 @@ RUN set -e \

 # Keep the version the same as in compute/compute-node.Dockerfile and
 # test_runner/regress/test_compute_metrics.py.
-ENV SQL_EXPORTER_VERSION=0.13.1
+ENV SQL_EXPORTER_VERSION=0.16.0
 RUN curl -fsSL \
    "https://github.com/burningalchemist/sql_exporter/releases/download/${SQL_EXPORTER_VERSION}/sql_exporter-${SQL_EXPORTER_VERSION}.linux-$(case "$(uname -m)" in x86_64) echo amd64;; aarch64) echo arm64;; esac).tar.gz" \
    --output sql_exporter.tar.gz \
--- a/compute/compute-node.Dockerfile
+++ b/compute/compute-node.Dockerfile
@@ -35,10 +35,12 @@ RUN case $DEBIAN_VERSION in \
      ;; \
    esac && \
    apt update &&  \
-    apt install --no-install-recommends -y git autoconf automake libtool build-essential bison flex libreadline-dev \
+    apt install --no-install-recommends --no-install-suggests -y \
+    ninja-build git autoconf automake libtool build-essential bison flex libreadline-dev \
    zlib1g-dev libxml2-dev libcurl4-openssl-dev libossp-uuid-dev wget ca-certificates pkg-config libssl-dev \
    libicu-dev libxslt1-dev liblz4-dev libzstd-dev zstd \
-    $VERSION_INSTALLS
+    $VERSION_INSTALLS \
+    && apt clean && rm -rf /var/lib/apt/lists/*

 #########################################################################################
 #
@@ -113,10 +115,12 @@ ARG DEBIAN_VERSION
 ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 RUN apt update && \
-    apt install --no-install-recommends -y gdal-bin libboost-dev libboost-thread-dev libboost-filesystem-dev \
+    apt install --no-install-recommends --no-install-suggests -y \
+    gdal-bin libboost-dev libboost-thread-dev libboost-filesystem-dev \
    libboost-system-dev libboost-iostreams-dev libboost-program-options-dev libboost-timer-dev \
    libcgal-dev libgdal-dev libgmp-dev libmpfr-dev libopenscenegraph-dev libprotobuf-c-dev \
-    protobuf-c-compiler xsltproc
+    protobuf-c-compiler xsltproc \
+    && apt clean && rm -rf /var/lib/apt/lists/*


 # Postgis 3.5.0 requires SFCGAL 1.4+
@@ -143,9 +147,9 @@ RUN case "${DEBIAN_VERSION}" in \
    wget https://gitlab.com/sfcgal/SFCGAL/-/archive/v${SFCGAL_VERSION}/SFCGAL-v${SFCGAL_VERSION}.tar.gz -O SFCGAL.tar.gz && \
    echo "${SFCGAL_CHECKSUM} SFCGAL.tar.gz" | sha256sum --check && \
    mkdir sfcgal-src && cd sfcgal-src && tar xzf ../SFCGAL.tar.gz --strip-components=1 -C . && \
-    cmake -DCMAKE_BUILD_TYPE=Release . && make -j $(getconf _NPROCESSORS_ONLN) && \
-    DESTDIR=/sfcgal make install -j $(getconf _NPROCESSORS_ONLN) && \
-    make clean && cp -R /sfcgal/* /
+    cmake -DCMAKE_BUILD_TYPE=Release -GNinja . && ninja -j $(getconf _NPROCESSORS_ONLN) && \
+    DESTDIR=/sfcgal ninja install -j $(getconf _NPROCESSORS_ONLN) && \
+    ninja clean && cp -R /sfcgal/* /

 ENV PATH="/usr/local/pgsql/bin:$PATH"

@@ -213,9 +217,9 @@ RUN case "${PG_VERSION}" in \
    echo "${PGROUTING_CHECKSUM} pgrouting.tar.gz" | sha256sum --check && \
    mkdir pgrouting-src && cd pgrouting-src && tar xzf ../pgrouting.tar.gz --strip-components=1 -C . && \
    mkdir build && cd build && \
-    cmake -DCMAKE_BUILD_TYPE=Release .. && \
-    make -j $(getconf _NPROCESSORS_ONLN) && \
-    make -j $(getconf _NPROCESSORS_ONLN) install && \
+    cmake -GNinja -DCMAKE_BUILD_TYPE=Release .. && \
+    ninja -j $(getconf _NPROCESSORS_ONLN) && \
+    ninja -j $(getconf _NPROCESSORS_ONLN) install && \
    echo 'trusted = true' >> /usr/local/pgsql/share/extension/pgrouting.control && \
    find /usr/local/pgsql -type f | sed 's|^/usr/local/pgsql/||' > /after.txt &&\
    cp /usr/local/pgsql/share/extension/pgrouting.control /extensions/postgis && \
@@ -235,7 +239,9 @@ COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY compute/patches/plv8-3.1.10.patch /plv8-3.1.10.patch

 RUN apt update && \
-    apt install --no-install-recommends -y ninja-build python3-dev libncurses5 binutils clang
+    apt install --no-install-recommends --no-install-suggests -y \
+    ninja-build python3-dev libncurses5 binutils clang \
+    && apt clean && rm -rf /var/lib/apt/lists/*

 # plv8 3.2.3 supports v17
 # last release v3.2.3 - Sep 7, 2024
@@ -301,9 +307,10 @@ RUN mkdir -p /h3/usr/ && \
    echo "ec99f1f5974846bde64f4513cf8d2ea1b8d172d2218ab41803bf6a63532272bc h3.tar.gz" | sha256sum --check && \
    mkdir h3-src && cd h3-src && tar xzf ../h3.tar.gz --strip-components=1 -C . && \
    mkdir build && cd build && \
-    cmake .. -DCMAKE_BUILD_TYPE=Release && \
-    make -j $(getconf _NPROCESSORS_ONLN) && \
-    DESTDIR=/h3 make install && \
+    cmake .. -GNinja -DBUILD_BENCHMARKS=0 -DCMAKE_BUILD_TYPE=Release \
+        -DBUILD_FUZZERS=0 -DBUILD_FILTERS=0 -DBUILD_GENERATORS=0 -DBUILD_TESTING=0 \
+    && ninja -j $(getconf _NPROCESSORS_ONLN) && \
+    DESTDIR=/h3 ninja install && \
    cp -R /h3/usr / && \
    rm -rf build

@@ -650,14 +657,15 @@ FROM build-deps AS rdkit-pg-build
 ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

-RUN apt-get update && \
-    apt-get install --no-install-recommends -y \
+RUN apt update && \
+    apt install --no-install-recommends --no-install-suggests -y \
        libboost-iostreams1.74-dev \
        libboost-regex1.74-dev \
        libboost-serialization1.74-dev \
        libboost-system1.74-dev \
        libeigen3-dev \
-        libboost-all-dev
+        libboost-all-dev \
+    && apt clean && rm -rf /var/lib/apt/lists/*

 # rdkit Release_2024_09_1 supports v17
 # last release Release_2024_09_1 - Sep 27, 2024
@@ -693,6 +701,8 @@ RUN case "${PG_VERSION}" in \
        -D RDK_BUILD_MOLINTERCHANGE_SUPPORT=OFF \
        -D RDK_BUILD_YAEHMOP_SUPPORT=OFF \
        -D RDK_BUILD_STRUCTCHECKER_SUPPORT=OFF \
+        -D RDK_TEST_MULTITHREADED=OFF \
+        -D RDK_BUILD_CPP_TESTS=OFF \
        -D RDK_USE_URF=OFF \
        -D RDK_BUILD_PGSQL=ON \
        -D RDK_PGSQL_STATIC=ON \
@@ -704,9 +714,10 @@ RUN case "${PG_VERSION}" in \
        -D RDK_INSTALL_COMIC_FONTS=OFF \
        -D RDK_BUILD_FREETYPE_SUPPORT=OFF \
        -D CMAKE_BUILD_TYPE=Release \
+        -GNinja \
        . && \
-    make -j $(getconf _NPROCESSORS_ONLN) && \
-    make -j $(getconf _NPROCESSORS_ONLN) install && \
+    ninja -j $(getconf _NPROCESSORS_ONLN) && \
+    ninja -j $(getconf _NPROCESSORS_ONLN) install && \
    echo 'trusted = true' >> /usr/local/pgsql/share/extension/rdkit.control

 #########################################################################################
@@ -849,8 +860,9 @@ FROM build-deps AS rust-extensions-build
 ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

-RUN apt-get update && \
-    apt-get install --no-install-recommends -y curl libclang-dev && \
+RUN apt update && \
+    apt install --no-install-recommends --no-install-suggests -y curl libclang-dev && \
+    apt clean && rm -rf /var/lib/apt/lists/* && \
    useradd -ms /bin/bash nonroot -b /home

 ENV HOME=/home/nonroot
@@ -885,8 +897,9 @@ FROM build-deps AS rust-extensions-build-pgrx12
 ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

-RUN apt-get update && \
-    apt-get install --no-install-recommends -y curl libclang-dev && \
+RUN apt update && \
+    apt install --no-install-recommends --no-install-suggests -y curl libclang-dev && \
+    apt clean && rm -rf /var/lib/apt/lists/* && \
    useradd -ms /bin/bash nonroot -b /home

 ENV HOME=/home/nonroot
@@ -914,18 +927,22 @@ FROM rust-extensions-build-pgrx12 AS pg-onnx-build

 # cmake 3.26 or higher is required, so installing it using pip (bullseye-backports has cmake 3.25).
 # Install it using virtual environment, because Python 3.11 (the default version on Debian 12 (Bookworm)) complains otherwise
-RUN apt-get update && apt-get install -y python3 python3-pip python3-venv && \
+RUN apt update && apt install --no-install-recommends --no-install-suggests -y \
+    python3 python3-pip python3-venv && \
+    apt clean && rm -rf /var/lib/apt/lists/* && \
    python3 -m venv venv && \
    . venv/bin/activate && \
    python3 -m pip install cmake==3.30.5 && \
    wget https://github.com/microsoft/onnxruntime/archive/refs/tags/v1.18.1.tar.gz -O onnxruntime.tar.gz && \
    mkdir onnxruntime-src && cd onnxruntime-src && tar xzf ../onnxruntime.tar.gz --strip-components=1 -C . && \
-    ./build.sh --config Release --parallel --skip_submodule_sync --skip_tests --allow_running_as_root
+    ./build.sh --config Release --parallel --cmake_generator Ninja \
+    --skip_submodule_sync --skip_tests --allow_running_as_root


 FROM pg-onnx-build AS pgrag-pg-build

-RUN apt-get install -y protobuf-compiler && \
+RUN apt update && apt install --no-install-recommends --no-install-suggests -y protobuf-compiler \
+    && apt clean && rm -rf /var/lib/apt/lists/* && \
    wget https://github.com/neondatabase-labs/pgrag/archive/refs/tags/v0.0.0.tar.gz -O pgrag.tar.gz &&  \
    echo "2cbe394c1e74fc8bcad9b52d5fbbfb783aef834ca3ce44626cfd770573700bb4 pgrag.tar.gz" | sha256sum --check && \
    mkdir pgrag-src && cd pgrag-src && tar xzf ../pgrag.tar.gz --strip-components=1 -C . && \
@@ -1168,6 +1185,25 @@ RUN case "${PG_VERSION}" in \
    make BUILD_TYPE=release -j $(getconf _NPROCESSORS_ONLN) install && \
    echo 'trusted = true' >> /usr/local/pgsql/share/extension/pg_mooncake.control

+#########################################################################################
+#
+# Layer "pg_repack"
+# compile pg_repack extension
+#
+#########################################################################################
+
+FROM build-deps AS pg-repack-build
+ARG PG_VERSION
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+ENV PATH="/usr/local/pgsql/bin/:$PATH"
+
+RUN wget https://github.com/reorg/pg_repack/archive/refs/tags/ver_1.5.2.tar.gz -O pg_repack.tar.gz && \
+    echo '4516cad42251ed3ad53ff619733004db47d5755acac83f75924cd94d1c4fb681 pg_repack.tar.gz' | sha256sum --check && \
+    mkdir pg_repack-src && cd pg_repack-src && tar xzf ../pg_repack.tar.gz --strip-components=1 -C . && \
+    make -j $(getconf _NPROCESSORS_ONLN) && \
+    make -j $(getconf _NPROCESSORS_ONLN) install
+
 #########################################################################################
 #
 # Layer "neon-pg-ext-build"
@@ -1213,6 +1249,7 @@ COPY --from=pg-anon-pg-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg-ivm-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg-partman-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg-mooncake-build /usr/local/pgsql/ /usr/local/pgsql/
+COPY --from=pg-repack-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY pgxn/ pgxn/

 RUN make -j $(getconf _NPROCESSORS_ONLN) \
@@ -1279,8 +1316,8 @@ COPY --from=compute-tools /home/nonroot/target/release-line-debug-size-lto/fast_

 FROM debian:$DEBIAN_FLAVOR AS pgbouncer
 RUN set -e \
-    && apt-get update \
-    && apt-get install --no-install-recommends -y \
+    && apt update \
+    && apt install --no-install-suggests --no-install-recommends -y \
        build-essential \
        git \
        ca-certificates \
@@ -1288,7 +1325,8 @@ RUN set -e \
        automake \
        libevent-dev \
        libtool \
-        pkg-config
+        pkg-config \
+    && apt clean && rm -rf /var/lib/apt/lists/*

 # Use `dist_man_MANS=` to skip manpage generation (which requires python3/pandoc)
 ENV PGBOUNCER_TAG=pgbouncer_1_22_1
@@ -1324,7 +1362,7 @@ FROM quay.io/prometheuscommunity/postgres-exporter:v0.12.1 AS postgres-exporter

 # Keep the version the same as in build-tools.Dockerfile and
 # test_runner/regress/test_compute_metrics.py.
-FROM burningalchemist/sql_exporter:0.13.1 AS sql-exporter
+FROM burningalchemist/sql_exporter:0.16.0 AS sql-exporter

 #########################################################################################
 #
@@ -1518,28 +1556,30 @@ RUN apt update && \
        locales \
        procps \
        ca-certificates \
+        curl \
+        unzip \
        $VERSION_INSTALLS && \
-    rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/* && \
+    apt clean && rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/* && \
    localedef -i en_US -c -f UTF-8 -A /usr/share/locale/locale.alias en_US.UTF-8

-# s5cmd 2.2.2 from https://github.com/peak/s5cmd/releases/tag/v2.2.2
-# used by fast_import
+# aws cli is used by fast_import (curl and unzip above are at this time only used for this installation step)
 ARG TARGETARCH
-ADD https://github.com/peak/s5cmd/releases/download/v2.2.2/s5cmd_2.2.2_linux_$TARGETARCH.deb /tmp/s5cmd.deb
 RUN set -ex; \
-    \
-    # Determine the expected checksum based on TARGETARCH
    if [ "${TARGETARCH}" = "amd64" ]; then \
-        CHECKSUM="392c385320cd5ffa435759a95af77c215553d967e4b1c0fffe52e4f14c29cf85"; \
+        TARGETARCH_ALT="x86_64"; \
+        CHECKSUM="c9a9df3770a3ff9259cb469b6179e02829687a464e0824d5c32d378820b53a00"; \
    elif [ "${TARGETARCH}" = "arm64" ]; then \
-        CHECKSUM="939bee3cf4b5604ddb00e67f8c157b91d7c7a5b553d1fbb6890fad32894b7b46"; \
+        TARGETARCH_ALT="aarch64"; \
+        CHECKSUM="8181730be7891582b38b028112e81b4899ca817e8c616aad807c9e9d1289223a"; \
    else \
        echo "Unsupported architecture: ${TARGETARCH}"; exit 1; \
    fi; \
-    \
-    # Compute and validate the checksum
-    echo "${CHECKSUM}  /tmp/s5cmd.deb" | sha256sum -c -
-RUN dpkg -i /tmp/s5cmd.deb && rm /tmp/s5cmd.deb
+    curl -L "https://awscli.amazonaws.com/awscli-exe-linux-${TARGETARCH_ALT}-2.17.5.zip" -o /tmp/awscliv2.zip; \
+    echo "${CHECKSUM}  /tmp/awscliv2.zip" | sha256sum -c -; \
+    unzip /tmp/awscliv2.zip -d /tmp/awscliv2; \
+    /tmp/awscliv2/aws/install; \
+    rm -rf /tmp/awscliv2.zip /tmp/awscliv2; \
+    true

 ENV LANG=en_US.utf8
 USER postgres
--- a/compute/etc/neon_collector.jsonnet
+++ b/compute/etc/neon_collector.jsonnet
@@ -3,7 +3,7 @@
  metrics: [
    import 'sql_exporter/checkpoints_req.libsonnet',
    import 'sql_exporter/checkpoints_timed.libsonnet',
-    import 'sql_exporter/compute_backpressure_throttling_seconds.libsonnet',
+    import 'sql_exporter/compute_backpressure_throttling_seconds_total.libsonnet',
    import 'sql_exporter/compute_current_lsn.libsonnet',
    import 'sql_exporter/compute_logical_snapshot_files.libsonnet',
    import 'sql_exporter/compute_logical_snapshots_bytes.libsonnet',
--- a/compute/etc/pgbouncer.ini
+++ b/compute/etc/pgbouncer.ini
@@ -19,3 +19,10 @@ max_prepared_statements=0
 admin_users=postgres
 unix_socket_dir=/tmp/
 unix_socket_mode=0777
+
+;; Disable connection logging. It produces a lot of logs that no one looks at,
+;; and we can get similar log entries from the proxy too. We had incidents in
+;; the past where the logging significantly stressed the log device or pgbouncer
+;; itself.
+log_connections=0
+log_disconnections=0
--- a/compute/etc/sql_exporter/compute_backpressure_throttling_seconds_total.libsonnet
+++ b/compute/etc/sql_exporter/compute_backpressure_throttling_seconds_total.libsonnet
@@ -1,10 +1,10 @@
 {
-  metric_name: 'compute_backpressure_throttling_seconds',
-  type: 'gauge',
+  metric_name: 'compute_backpressure_throttling_seconds_total',
+  type: 'counter',
  help: 'Time compute has spent throttled',
  key_labels: null,
  values: [
    'throttled',
  ],
-  query: importstr 'sql_exporter/compute_backpressure_throttling_seconds.sql',
+  query: importstr 'sql_exporter/compute_backpressure_throttling_seconds_total.sql',
 }
--- a/compute/etc/sql_exporter/compute_backpressure_throttling_seconds_total.sql
+++ b/compute/etc/sql_exporter/compute_backpressure_throttling_seconds_total.sql
--- a/compute/patches/cloud_regress_pg16.patch
+++ b/compute/patches/cloud_regress_pg16.patch
@@ -981,7 +981,7 @@ index fc42d418bf..e38f517574 100644
 CREATE SCHEMA addr_nsp;
 SET search_path TO 'addr_nsp';
 diff --git a/src/test/regress/expected/password.out b/src/test/regress/expected/password.out
-index 8475231735..1afae5395f 100644
+index 8475231735..0653946337 100644
 --- a/src/test/regress/expected/password.out
 +++ b/src/test/regress/expected/password.out
@@ -12,11 +12,11 @@ SET password_encryption = 'md5'; -- ok
@@ -1006,65 +1006,63 @@ index 8475231735..1afae5395f 100644
 -----------------+---------------------------------------------------
 - regress_passwd1 | md5783277baca28003b33453252be4dbb34
 - regress_passwd2 | md54044304ba511dd062133eb5b4b84a2a3
-+ regress_passwd1 | NEON_MD5_PLACEHOLDER_regress_passwd1
-+ regress_passwd2 | NEON_MD5_PLACEHOLDER_regress_passwd2
+ regress_passwd1 | NEON_MD5_PLACEHOLDER:regress_passwd1
+ regress_passwd2 | NEON_MD5_PLACEHOLDER:regress_passwd2
  regress_passwd3 | SCRAM-SHA-256$4096:<salt>$<storedkey>:<serverkey>
 - regress_passwd4 | 
 + regress_passwd4 | SCRAM-SHA-256$4096:<salt>$<storedkey>:<serverkey>
 (4 rows)
 
 -- Rename a role
-@@ -54,24 +54,30 @@ ALTER ROLE regress_passwd2_new RENAME TO regress_passwd2;
+@@ -54,24 +54,16 @@ ALTER ROLE regress_passwd2_new RENAME TO regress_passwd2;
 -- passwords.
 SET password_encryption = 'md5';
 -- encrypt with MD5
 -ALTER ROLE regress_passwd2 PASSWORD 'foo';
+--- already encrypted, use as they are
+-ALTER ROLE regress_passwd1 PASSWORD 'md5cd3578025fe2c3d7ed1b9a9b26238b70';
+-ALTER ROLE regress_passwd3 PASSWORD 'SCRAM-SHA-256$4096:VLK4RMaQLCvNtQ==$6YtlR4t69SguDiwFvbVgVZtuz6gpJQQqUMZ7IQJK5yI=:ps75jrHeYU4lXCcXI4O8oIdJ3eO8o2jirjruw9phBTo=';
 +ALTER ROLE regress_passwd2 PASSWORD NEON_PASSWORD_PLACEHOLDER;
- -- already encrypted, use as they are
- ALTER ROLE regress_passwd1 PASSWORD 'md5cd3578025fe2c3d7ed1b9a9b26238b70';
-+ERROR:  Received HTTP code 400 from control plane: {"error":"Neon only supports being given plaintext passwords"}
- ALTER ROLE regress_passwd3 PASSWORD 'SCRAM-SHA-256$4096:VLK4RMaQLCvNtQ==$6YtlR4t69SguDiwFvbVgVZtuz6gpJQQqUMZ7IQJK5yI=:ps75jrHeYU4lXCcXI4O8oIdJ3eO8o2jirjruw9phBTo=';
-+ERROR:  Received HTTP code 400 from control plane: {"error":"Neon only supports being given plaintext passwords"}
 SET password_encryption = 'scram-sha-256';
 -- create SCRAM secret
 -ALTER ROLE  regress_passwd4 PASSWORD 'foo';
+--- already encrypted with MD5, use as it is
+-CREATE ROLE regress_passwd5 PASSWORD 'md5e73a4b11df52a6068f8b39f90be36023';
+--- This looks like a valid SCRAM-SHA-256 secret, but it is not
+--- so it should be hashed with SCRAM-SHA-256.
+-CREATE ROLE regress_passwd6 PASSWORD 'SCRAM-SHA-256$1234';
+--- These may look like valid MD5 secrets, but they are not, so they
+--- should be hashed with SCRAM-SHA-256.
+--- trailing garbage at the end
+-CREATE ROLE regress_passwd7 PASSWORD 'md5012345678901234567890123456789zz';
+--- invalid length
+-CREATE ROLE regress_passwd8 PASSWORD 'md501234567890123456789012345678901zz';
 +ALTER ROLE  regress_passwd4 PASSWORD NEON_PASSWORD_PLACEHOLDER;
- -- already encrypted with MD5, use as it is
- CREATE ROLE regress_passwd5 PASSWORD 'md5e73a4b11df52a6068f8b39f90be36023';
-+ERROR:  Received HTTP code 400 from control plane: {"error":"Neon only supports being given plaintext passwords"}
- -- This looks like a valid SCRAM-SHA-256 secret, but it is not
- -- so it should be hashed with SCRAM-SHA-256.
- CREATE ROLE regress_passwd6 PASSWORD 'SCRAM-SHA-256$1234';
-+ERROR:  Received HTTP code 400 from control plane: {"error":"Neon only supports being given plaintext passwords"}
- -- These may look like valid MD5 secrets, but they are not, so they
- -- should be hashed with SCRAM-SHA-256.
- -- trailing garbage at the end
- CREATE ROLE regress_passwd7 PASSWORD 'md5012345678901234567890123456789zz';
-+ERROR:  Received HTTP code 400 from control plane: {"error":"Neon only supports being given plaintext passwords"}
- -- invalid length
- CREATE ROLE regress_passwd8 PASSWORD 'md501234567890123456789012345678901zz';
-+ERROR:  Received HTTP code 400 from control plane: {"error":"Neon only supports being given plaintext passwords"}
+-- Neon does not support encrypted passwords, use unencrypted instead
+CREATE ROLE regress_passwd5 PASSWORD NEON_PASSWORD_PLACEHOLDER;
+-- Neon does not support encrypted passwords, use unencrypted instead
+CREATE ROLE regress_passwd6 PASSWORD NEON_PASSWORD_PLACEHOLDER;
+CREATE ROLE regress_passwd7 PASSWORD NEON_PASSWORD_PLACEHOLDER;
+CREATE ROLE regress_passwd8 PASSWORD NEON_PASSWORD_PLACEHOLDER;
 -- Changing the SCRAM iteration count
 SET scram_iterations = 1024;
 CREATE ROLE regress_passwd9 PASSWORD 'alterediterationcount';
-@@ -81,63 +87,67 @@ SELECT rolname, regexp_replace(rolpassword, '(SCRAM-SHA-256)\$(\d+):([a-zA-Z0-9+
+@@ -81,11 +73,11 @@ SELECT rolname, regexp_replace(rolpassword, '(SCRAM-SHA-256)\$(\d+):([a-zA-Z0-9+
     ORDER BY rolname, rolpassword;
      rolname     |                rolpassword_masked                 
 -----------------+---------------------------------------------------
 - regress_passwd1 | md5cd3578025fe2c3d7ed1b9a9b26238b70
 - regress_passwd2 | md5dfa155cadd5f4ad57860162f3fab9cdb
-+ regress_passwd1 | NEON_MD5_PLACEHOLDER_regress_passwd1
-+ regress_passwd2 | NEON_MD5_PLACEHOLDER_regress_passwd2
+ regress_passwd1 | NEON_MD5_PLACEHOLDER:regress_passwd1
+ regress_passwd2 | NEON_MD5_PLACEHOLDER:regress_passwd2
  regress_passwd3 | SCRAM-SHA-256$4096:<salt>$<storedkey>:<serverkey>
  regress_passwd4 | SCRAM-SHA-256$4096:<salt>$<storedkey>:<serverkey>
 - regress_passwd5 | md5e73a4b11df52a6068f8b39f90be36023
- regress_passwd6 | SCRAM-SHA-256$4096:<salt>$<storedkey>:<serverkey>
- regress_passwd7 | SCRAM-SHA-256$4096:<salt>$<storedkey>:<serverkey>
- regress_passwd8 | SCRAM-SHA-256$4096:<salt>$<storedkey>:<serverkey>
-  regress_passwd9 | SCRAM-SHA-256$1024:<salt>$<storedkey>:<serverkey>
-(9 rows)
-+(5 rows)
- 
+ regress_passwd5 | SCRAM-SHA-256$4096:<salt>$<storedkey>:<serverkey>
+  regress_passwd6 | SCRAM-SHA-256$4096:<salt>$<storedkey>:<serverkey>
+  regress_passwd7 | SCRAM-SHA-256$4096:<salt>$<storedkey>:<serverkey>
+  regress_passwd8 | SCRAM-SHA-256$4096:<salt>$<storedkey>:<serverkey>
+@@ -95,23 +87,20 @@ SELECT rolname, regexp_replace(rolpassword, '(SCRAM-SHA-256)\$(\d+):([a-zA-Z0-9+
 -- An empty password is not allowed, in any form
 CREATE ROLE regress_passwd_empty PASSWORD '';
 NOTICE:  empty string is not a valid password, clearing password
@@ -1082,56 +1080,37 @@ index 8475231735..1afae5395f 100644
 -(1 row)
 +(0 rows)
 
- -- Test with invalid stored and server keys.
- --
- -- The first is valid, to act as a control. The others have too long
- -- stored/server keys. They will be re-hashed.
- CREATE ROLE regress_passwd_sha_len0 PASSWORD 'SCRAM-SHA-256$4096:A6xHKoH/494E941doaPOYg==$Ky+A30sewHIH3VHQLRN9vYsuzlgNyGNKCh37dy96Rqw=:COPdlNiIkrsacU5QoxydEuOH6e/KfiipeETb/bPw8ZI=';
-+ERROR:  Received HTTP code 400 from control plane: {"error":"Neon only supports being given plaintext passwords"}
- CREATE ROLE regress_passwd_sha_len1 PASSWORD 'SCRAM-SHA-256$4096:A6xHKoH/494E941doaPOYg==$Ky+A30sewHIH3VHQLRN9vYsuzlgNyGNKCh37dy96RqwAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=:COPdlNiIkrsacU5QoxydEuOH6e/KfiipeETb/bPw8ZI=';
-+ERROR:  Received HTTP code 400 from control plane: {"error":"Neon only supports being given plaintext passwords"}
- CREATE ROLE regress_passwd_sha_len2 PASSWORD 'SCRAM-SHA-256$4096:A6xHKoH/494E941doaPOYg==$Ky+A30sewHIH3VHQLRN9vYsuzlgNyGNKCh37dy96Rqw=:COPdlNiIkrsacU5QoxydEuOH6e/KfiipeETb/bPw8ZIAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=';
-+ERROR:  Received HTTP code 400 from control plane: {"error":"Neon only supports being given plaintext passwords"}
+--- Test with invalid stored and server keys.
+---
+--- The first is valid, to act as a control. The others have too long
+--- stored/server keys. They will be re-hashed.
+-CREATE ROLE regress_passwd_sha_len0 PASSWORD 'SCRAM-SHA-256$4096:A6xHKoH/494E941doaPOYg==$Ky+A30sewHIH3VHQLRN9vYsuzlgNyGNKCh37dy96Rqw=:COPdlNiIkrsacU5QoxydEuOH6e/KfiipeETb/bPw8ZI=';
+-CREATE ROLE regress_passwd_sha_len1 PASSWORD 'SCRAM-SHA-256$4096:A6xHKoH/494E941doaPOYg==$Ky+A30sewHIH3VHQLRN9vYsuzlgNyGNKCh37dy96RqwAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=:COPdlNiIkrsacU5QoxydEuOH6e/KfiipeETb/bPw8ZI=';
+-CREATE ROLE regress_passwd_sha_len2 PASSWORD 'SCRAM-SHA-256$4096:A6xHKoH/494E941doaPOYg==$Ky+A30sewHIH3VHQLRN9vYsuzlgNyGNKCh37dy96Rqw=:COPdlNiIkrsacU5QoxydEuOH6e/KfiipeETb/bPw8ZIAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=';
+-- Neon does not support encrypted passwords, use unencrypted instead
+CREATE ROLE regress_passwd_sha_len0 PASSWORD NEON_PASSWORD_PLACEHOLDER;
+CREATE ROLE regress_passwd_sha_len1 PASSWORD NEON_PASSWORD_PLACEHOLDER;
+CREATE ROLE regress_passwd_sha_len2 PASSWORD NEON_PASSWORD_PLACEHOLDER;
 -- Check that the invalid secrets were re-hashed. A re-hashed secret
 -- should not contain the original salt.
 SELECT rolname, rolpassword not like '%A6xHKoH/494E941doaPOYg==%' as is_rolpassword_rehashed
-     FROM pg_authid
-     WHERE rolname LIKE 'regress_passwd_sha_len%'
+@@ -120,7 +109,7 @@ SELECT rolname, rolpassword not like '%A6xHKoH/494E941doaPOYg==%' as is_rolpassw
     ORDER BY rolname;
-         rolname         | is_rolpassword_rehashed 
--------------------------+-------------------------
+          rolname         | is_rolpassword_rehashed 
+ -------------------------+-------------------------
 - regress_passwd_sha_len0 | f
- regress_passwd_sha_len1 | t
- regress_passwd_sha_len2 | t
-(3 rows)
-+ rolname | is_rolpassword_rehashed 
-+---------+-------------------------
-+(0 rows)
- 
- DROP ROLE regress_passwd1;
- DROP ROLE regress_passwd2;
- DROP ROLE regress_passwd3;
- DROP ROLE regress_passwd4;
- DROP ROLE regress_passwd5;
-+ERROR:  role "regress_passwd5" does not exist
- DROP ROLE regress_passwd6;
-+ERROR:  role "regress_passwd6" does not exist
- DROP ROLE regress_passwd7;
-+ERROR:  role "regress_passwd7" does not exist
+ regress_passwd_sha_len0 | t
+  regress_passwd_sha_len1 | t
+  regress_passwd_sha_len2 | t
+ (3 rows)
+@@ -135,6 +124,7 @@ DROP ROLE regress_passwd7;
 DROP ROLE regress_passwd8;
-+ERROR:  role "regress_passwd8" does not exist
 DROP ROLE regress_passwd9;
 DROP ROLE regress_passwd_empty;
 +ERROR:  role "regress_passwd_empty" does not exist
 DROP ROLE regress_passwd_sha_len0;
-+ERROR:  role "regress_passwd_sha_len0" does not exist
 DROP ROLE regress_passwd_sha_len1;
-+ERROR:  role "regress_passwd_sha_len1" does not exist
 DROP ROLE regress_passwd_sha_len2;
-+ERROR:  role "regress_passwd_sha_len2" does not exist
- -- all entries should have been removed
- SELECT rolname, rolpassword
-     FROM pg_authid
 diff --git a/src/test/regress/expected/privileges.out b/src/test/regress/expected/privileges.out
 index 5b9dba7b32..cc408dad42 100644
 --- a/src/test/regress/expected/privileges.out
@@ -3194,7 +3173,7 @@ index 1a6c61f49d..1c31ac6a53 100644
 -- Test generic object addressing/identification functions
 CREATE SCHEMA addr_nsp;
 diff --git a/src/test/regress/sql/password.sql b/src/test/regress/sql/password.sql
-index 53e86b0b6c..f07cf1ec54 100644
+index 53e86b0b6c..0303fdfe96 100644
 --- a/src/test/regress/sql/password.sql
 +++ b/src/test/regress/sql/password.sql
@@ -10,11 +10,11 @@ SET password_encryption = 'scram-sha-256'; -- ok
@@ -3213,23 +3192,59 @@ index 53e86b0b6c..f07cf1ec54 100644
 
 -- check list of created entries
 --
-@@ -42,14 +42,14 @@ ALTER ROLE regress_passwd2_new RENAME TO regress_passwd2;
+@@ -42,26 +42,18 @@ ALTER ROLE regress_passwd2_new RENAME TO regress_passwd2;
 SET password_encryption = 'md5';
 
 -- encrypt with MD5
 -ALTER ROLE regress_passwd2 PASSWORD 'foo';
+--- already encrypted, use as they are
+-ALTER ROLE regress_passwd1 PASSWORD 'md5cd3578025fe2c3d7ed1b9a9b26238b70';
+-ALTER ROLE regress_passwd3 PASSWORD 'SCRAM-SHA-256$4096:VLK4RMaQLCvNtQ==$6YtlR4t69SguDiwFvbVgVZtuz6gpJQQqUMZ7IQJK5yI=:ps75jrHeYU4lXCcXI4O8oIdJ3eO8o2jirjruw9phBTo=';
 +ALTER ROLE regress_passwd2 PASSWORD NEON_PASSWORD_PLACEHOLDER;
- -- already encrypted, use as they are
- ALTER ROLE regress_passwd1 PASSWORD 'md5cd3578025fe2c3d7ed1b9a9b26238b70';
- ALTER ROLE regress_passwd3 PASSWORD 'SCRAM-SHA-256$4096:VLK4RMaQLCvNtQ==$6YtlR4t69SguDiwFvbVgVZtuz6gpJQQqUMZ7IQJK5yI=:ps75jrHeYU4lXCcXI4O8oIdJ3eO8o2jirjruw9phBTo=';
 
 SET password_encryption = 'scram-sha-256';
 -- create SCRAM secret
 -ALTER ROLE  regress_passwd4 PASSWORD 'foo';
+--- already encrypted with MD5, use as it is
+-CREATE ROLE regress_passwd5 PASSWORD 'md5e73a4b11df52a6068f8b39f90be36023';
 +ALTER ROLE  regress_passwd4 PASSWORD NEON_PASSWORD_PLACEHOLDER;
- -- already encrypted with MD5, use as it is
- CREATE ROLE regress_passwd5 PASSWORD 'md5e73a4b11df52a6068f8b39f90be36023';
+-- Neon does not support encrypted passwords, use unencrypted instead
+CREATE ROLE regress_passwd5 PASSWORD NEON_PASSWORD_PLACEHOLDER;
 
+--- This looks like a valid SCRAM-SHA-256 secret, but it is not
+--- so it should be hashed with SCRAM-SHA-256.
+-CREATE ROLE regress_passwd6 PASSWORD 'SCRAM-SHA-256$1234';
+--- These may look like valid MD5 secrets, but they are not, so they
+--- should be hashed with SCRAM-SHA-256.
+--- trailing garbage at the end
+-CREATE ROLE regress_passwd7 PASSWORD 'md5012345678901234567890123456789zz';
+--- invalid length
+-CREATE ROLE regress_passwd8 PASSWORD 'md501234567890123456789012345678901zz';
+-- Neon does not support encrypted passwords, use unencrypted instead
+CREATE ROLE regress_passwd6 PASSWORD NEON_PASSWORD_PLACEHOLDER;
+CREATE ROLE regress_passwd7 PASSWORD NEON_PASSWORD_PLACEHOLDER;
+CREATE ROLE regress_passwd8 PASSWORD NEON_PASSWORD_PLACEHOLDER;
+ 
+ -- Changing the SCRAM iteration count
+ SET scram_iterations = 1024;
+@@ -78,13 +70,10 @@ ALTER ROLE regress_passwd_empty PASSWORD 'md585939a5ce845f1a1b620742e3c659e0a';
+ ALTER ROLE regress_passwd_empty PASSWORD 'SCRAM-SHA-256$4096:hpFyHTUsSWcR7O9P$LgZFIt6Oqdo27ZFKbZ2nV+vtnYM995pDh9ca6WSi120=:qVV5NeluNfUPkwm7Vqat25RjSPLkGeoZBQs6wVv+um4=';
+ SELECT rolpassword FROM pg_authid WHERE rolname='regress_passwd_empty';
+ 
+--- Test with invalid stored and server keys.
+---
+--- The first is valid, to act as a control. The others have too long
+--- stored/server keys. They will be re-hashed.
+-CREATE ROLE regress_passwd_sha_len0 PASSWORD 'SCRAM-SHA-256$4096:A6xHKoH/494E941doaPOYg==$Ky+A30sewHIH3VHQLRN9vYsuzlgNyGNKCh37dy96Rqw=:COPdlNiIkrsacU5QoxydEuOH6e/KfiipeETb/bPw8ZI=';
+-CREATE ROLE regress_passwd_sha_len1 PASSWORD 'SCRAM-SHA-256$4096:A6xHKoH/494E941doaPOYg==$Ky+A30sewHIH3VHQLRN9vYsuzlgNyGNKCh37dy96RqwAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=:COPdlNiIkrsacU5QoxydEuOH6e/KfiipeETb/bPw8ZI=';
+-CREATE ROLE regress_passwd_sha_len2 PASSWORD 'SCRAM-SHA-256$4096:A6xHKoH/494E941doaPOYg==$Ky+A30sewHIH3VHQLRN9vYsuzlgNyGNKCh37dy96Rqw=:COPdlNiIkrsacU5QoxydEuOH6e/KfiipeETb/bPw8ZIAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA=';
+-- Neon does not support encrypted passwords, use unencrypted instead
+CREATE ROLE regress_passwd_sha_len0 PASSWORD NEON_PASSWORD_PLACEHOLDER;
+CREATE ROLE regress_passwd_sha_len1 PASSWORD NEON_PASSWORD_PLACEHOLDER;
+CREATE ROLE regress_passwd_sha_len2 PASSWORD NEON_PASSWORD_PLACEHOLDER;
+ 
+ -- Check that the invalid secrets were re-hashed. A re-hashed secret
+ -- should not contain the original salt.
 diff --git a/src/test/regress/sql/privileges.sql b/src/test/regress/sql/privileges.sql
 index 249df17a58..b258e7f26a 100644
 --- a/src/test/regress/sql/privileges.sql
--- a/compute/patches/cloud_regress_pg17.patch
+++ b/compute/patches/cloud_regress_pg17.patch
--- a/compute_tools/src/bin/compute_ctl.rs
+++ b/compute_tools/src/bin/compute_ctl.rs
@@ -246,47 +246,48 @@ fn try_spec_from_cli(
    let compute_id = matches.get_one::<String>("compute-id");
    let control_plane_uri = matches.get_one::<String>("control-plane-uri");

-    let spec;
-    let mut live_config_allowed = false;
-    match spec_json {
-        // First, try to get cluster spec from the cli argument
-        Some(json) => {
-            info!("got spec from cli argument {}", json);
-            spec = Some(serde_json::from_str(json)?);
-        }
-        None => {
-            // Second, try to read it from the file if path is provided
-            if let Some(sp) = spec_path {
-                let path = Path::new(sp);
-                let file = File::open(path)?;
-                spec = Some(serde_json::from_reader(file)?);
-                live_config_allowed = true;
-            } else if let Some(id) = compute_id {
-                if let Some(cp_base) = control_plane_uri {
-                    live_config_allowed = true;
-                    spec = match get_spec_from_control_plane(cp_base, id) {
-                        Ok(s) => s,
-                        Err(e) => {
-                            error!("cannot get response from control plane: {}", e);
-                            panic!("neither spec nor confirmation that compute is in the Empty state was received");
-                        }
-                    };
-                } else {
-                    panic!("must specify both --control-plane-uri and --compute-id or none");
-                }
-            } else {
-                panic!(
-                    "compute spec should be provided by one of the following ways: \
-                    --spec OR --spec-path OR --control-plane-uri and --compute-id"
-                );
-            }
-        }
+    // First, try to get cluster spec from the cli argument
+    if let Some(spec_json) = spec_json {
+        info!("got spec from cli argument {}", spec_json);
+        return Ok(CliSpecParams {
+            spec: Some(serde_json::from_str(spec_json)?),
+            live_config_allowed: false,
+        });
+    }
+
+    // Second, try to read it from the file if path is provided
+    if let Some(spec_path) = spec_path {
+        let file = File::open(Path::new(spec_path))?;
+        return Ok(CliSpecParams {
+            spec: Some(serde_json::from_reader(file)?),
+            live_config_allowed: true,
+        });
+    }
+
+    let Some(compute_id) = compute_id else {
+        panic!(
+            "compute spec should be provided by one of the following ways: \
+                --spec OR --spec-path OR --control-plane-uri and --compute-id"
+        );
+    };
+    let Some(control_plane_uri) = control_plane_uri else {
+        panic!("must specify both --control-plane-uri and --compute-id or none");
    };

-    Ok(CliSpecParams {
-        spec,
-        live_config_allowed,
-    })
+    match get_spec_from_control_plane(control_plane_uri, compute_id) {
+        Ok(spec) => Ok(CliSpecParams {
+            spec,
+            live_config_allowed: true,
+        }),
+        Err(e) => {
+            error!(
+                "cannot get response from control plane: {}\n\
+                neither spec nor confirmation that compute is in the Empty state was received",
+                e
+            );
+            Err(e)
+        }
+    }
 }

 struct CliSpecParams {
--- a/compute_tools/src/bin/fast_import.rs
+++ b/compute_tools/src/bin/fast_import.rs
@@ -34,12 +34,12 @@ use nix::unistd::Pid;
 use tracing::{info, info_span, warn, Instrument};
 use utils::fs_ext::is_directory_empty;

+#[path = "fast_import/aws_s3_sync.rs"]
+mod aws_s3_sync;
 #[path = "fast_import/child_stdio_to_log.rs"]
 mod child_stdio_to_log;
 #[path = "fast_import/s3_uri.rs"]
 mod s3_uri;
-#[path = "fast_import/s5cmd.rs"]
-mod s5cmd;

 #[derive(clap::Parser)]
 struct Args {
@@ -326,7 +326,7 @@ pub(crate) async fn main() -> anyhow::Result<()> {
    }

    info!("upload pgdata");
-    s5cmd::sync(Utf8Path::new(&pgdata_dir), &s3_prefix.append("/"))
+    aws_s3_sync::sync(Utf8Path::new(&pgdata_dir), &s3_prefix.append("/pgdata/"))
        .await
        .context("sync dump directory to destination")?;

@@ -334,10 +334,10 @@ pub(crate) async fn main() -> anyhow::Result<()> {
    {
        let status_dir = working_directory.join("status");
        std::fs::create_dir(&status_dir).context("create status directory")?;
-        let status_file = status_dir.join("status");
+        let status_file = status_dir.join("pgdata");
        std::fs::write(&status_file, serde_json::json!({"done": true}).to_string())
            .context("write status file")?;
-        s5cmd::sync(&status_file, &s3_prefix.append("/status/pgdata"))
+        aws_s3_sync::sync(&status_dir, &s3_prefix.append("/status/"))
            .await
            .context("sync status directory to destination")?;
    }
--- a/compute_tools/src/bin/fast_import/aws_s3_sync.rs
+++ b/compute_tools/src/bin/fast_import/aws_s3_sync.rs
@@ -4,24 +4,21 @@ use camino::Utf8Path;
 use super::s3_uri::S3Uri;

 pub(crate) async fn sync(local: &Utf8Path, remote: &S3Uri) -> anyhow::Result<()> {
-    let mut builder = tokio::process::Command::new("s5cmd");
-    // s5cmd uses aws-sdk-go v1, hence doesn't support AWS_ENDPOINT_URL
-    if let Some(val) = std::env::var_os("AWS_ENDPOINT_URL") {
-        builder.arg("--endpoint-url").arg(val);
-    }
+    let mut builder = tokio::process::Command::new("aws");
    builder
+        .arg("s3")
        .arg("sync")
        .arg(local.as_str())
        .arg(remote.to_string());
    let st = builder
        .spawn()
-        .context("spawn s5cmd")?
+        .context("spawn aws s3 sync")?
        .wait()
        .await
-        .context("wait for s5cmd")?;
+        .context("wait for aws s3 sync")?;
    if st.success() {
        Ok(())
    } else {
-        Err(anyhow::anyhow!("s5cmd failed"))
+        Err(anyhow::anyhow!("aws s3 sync failed"))
    }
 }
--- a/compute_tools/src/http/openapi_spec.yaml
+++ b/compute_tools/src/http/openapi_spec.yaml
@@ -537,12 +537,14 @@ components:
            properties:
              extname:
                type: string
-              versions:
-                type: array
+              version:
+                type: string
                items:
                  type: string
              n_databases:
                type: integer
+              owned_by_superuser:
+                type: integer

    SetRoleGrantsRequest:
      type: object
--- a/compute_tools/src/installed_extensions.rs
+++ b/compute_tools/src/installed_extensions.rs
@@ -1,7 +1,6 @@
 use compute_api::responses::{InstalledExtension, InstalledExtensions};
 use metrics::proto::MetricFamily;
 use std::collections::HashMap;
-use std::collections::HashSet;

 use anyhow::Result;
 use postgres::{Client, NoTls};
@@ -38,61 +37,77 @@ fn list_dbs(client: &mut Client) -> Result<Vec<String>> {
 /// Connect to every database (see list_dbs above) and get the list of installed extensions.
 ///
 /// Same extension can be installed in multiple databases with different versions,
-/// we only keep the highest and lowest version across all databases.
+/// so we report a separate metric (number of databases where it is installed)
+/// for each extension version.
 pub fn get_installed_extensions(mut conf: postgres::config::Config) -> Result<InstalledExtensions> {
    conf.application_name("compute_ctl:get_installed_extensions");
    let mut client = conf.connect(NoTls)?;
-
    let databases: Vec<String> = list_dbs(&mut client)?;

-    let mut extensions_map: HashMap<String, InstalledExtension> = HashMap::new();
+    let mut extensions_map: HashMap<(String, String, String), InstalledExtension> = HashMap::new();
    for db in databases.iter() {
        conf.dbname(db);
        let mut db_client = conf.connect(NoTls)?;
-        let extensions: Vec<(String, String)> = db_client
+        let extensions: Vec<(String, String, i32)> = db_client
            .query(
-                "SELECT extname, extversion FROM pg_catalog.pg_extension;",
+                "SELECT extname, extversion, extowner::integer FROM pg_catalog.pg_extension",
                &[],
            )?
            .iter()
-            .map(|row| (row.get("extname"), row.get("extversion")))
+            .map(|row| {
+                (
+                    row.get("extname"),
+                    row.get("extversion"),
+                    row.get("extowner"),
+                )
+            })
            .collect();

-        for (extname, v) in extensions.iter() {
+        for (extname, v, extowner) in extensions.iter() {
            let version = v.to_string();

-            // increment the number of databases where the version of extension is installed
-            INSTALLED_EXTENSIONS
-                .with_label_values(&[extname, &version])
-                .inc();
+            // check if the extension is owned by superuser
+            // 10 is the oid of superuser
+            let owned_by_superuser = if *extowner == 10 { "1" } else { "0" };

            extensions_map
-                .entry(extname.to_string())
+                .entry((
+                    extname.to_string(),
+                    version.clone(),
+                    owned_by_superuser.to_string(),
+                ))
                .and_modify(|e| {
-                    e.versions.insert(version.clone());
                    // count the number of databases where the extension is installed
                    e.n_databases += 1;
                })
                .or_insert(InstalledExtension {
                    extname: extname.to_string(),
-                    versions: HashSet::from([version.clone()]),
+                    version: version.clone(),
                    n_databases: 1,
+                    owned_by_superuser: owned_by_superuser.to_string(),
                });
        }
    }

-    let res = InstalledExtensions {
-        extensions: extensions_map.into_values().collect(),
-    };
+    for (key, ext) in extensions_map.iter() {
+        let (extname, version, owned_by_superuser) = key;
+        let n_databases = ext.n_databases as u64;

-    Ok(res)
+        INSTALLED_EXTENSIONS
+            .with_label_values(&[extname, version, owned_by_superuser])
+            .set(n_databases);
+    }
+
+    Ok(InstalledExtensions {
+        extensions: extensions_map.into_values().collect(),
+    })
 }

 static INSTALLED_EXTENSIONS: Lazy<UIntGaugeVec> = Lazy::new(|| {
    register_uint_gauge_vec!(
        "compute_installed_extensions",
        "Number of databases where the version of extension is installed",
-        &["extension_name", "version"]
+        &["extension_name", "version", "owned_by_superuser"]
    )
    .expect("failed to define a metric")
 });
--- a/control_plane/src/background_process.rs
+++ b/control_plane/src/background_process.rs
@@ -274,6 +274,7 @@ fn fill_remote_storage_secrets_vars(mut cmd: &mut Command) -> &mut Command {
    for env_key in [
        "AWS_ACCESS_KEY_ID",
        "AWS_SECRET_ACCESS_KEY",
+        "AWS_SESSION_TOKEN",
        "AWS_PROFILE",
        // HOME is needed in combination with `AWS_PROFILE` to pick up the SSO sessions.
        "HOME",
--- a/control_plane/src/bin/neon_local.rs
+++ b/control_plane/src/bin/neon_local.rs
@@ -19,6 +19,7 @@ use control_plane::storage_controller::{
    NeonStorageControllerStartArgs, NeonStorageControllerStopArgs, StorageController,
 };
 use control_plane::{broker, local_env};
+use nix::fcntl::{flock, FlockArg};
 use pageserver_api::config::{
    DEFAULT_HTTP_LISTEN_PORT as DEFAULT_PAGESERVER_HTTP_PORT,
    DEFAULT_PG_LISTEN_PORT as DEFAULT_PAGESERVER_PG_PORT,
@@ -36,6 +37,8 @@ use safekeeper_api::{
 };
 use std::borrow::Cow;
 use std::collections::{BTreeSet, HashMap};
+use std::fs::File;
+use std::os::fd::AsRawFd;
 use std::path::PathBuf;
 use std::process::exit;
 use std::str::FromStr;
@@ -689,6 +692,21 @@ struct TimelineTreeEl {
    pub children: BTreeSet<TimelineId>,
 }

+/// A flock-based guard over the neon_local repository directory
+struct RepoLock {
+    _file: File,
+}
+
+impl RepoLock {
+    fn new() -> Result<Self> {
+        let repo_dir = File::open(local_env::base_path())?;
+        let repo_dir_fd = repo_dir.as_raw_fd();
+        flock(repo_dir_fd, FlockArg::LockExclusive)?;
+
+        Ok(Self { _file: repo_dir })
+    }
+}
+
 // Main entry point for the 'neon_local' CLI utility
 //
 // This utility helps to manage neon installation. That includes following:
@@ -700,9 +718,14 @@ fn main() -> Result<()> {
    let cli = Cli::parse();

    // Check for 'neon init' command first.
-    let subcommand_result = if let NeonLocalCmd::Init(args) = cli.command {
-        handle_init(&args).map(|env| Some(Cow::Owned(env)))
+    let (subcommand_result, _lock) = if let NeonLocalCmd::Init(args) = cli.command {
+        (handle_init(&args).map(|env| Some(Cow::Owned(env))), None)
    } else {
+        // This tool uses a collection of simple files to store its state, and consequently
+        // it is not generally safe to run multiple commands concurrently.  Rather than expect
+        // all callers to know this, use a lock file to protect against concurrent execution.
+        let _repo_lock = RepoLock::new().unwrap();
+
        // all other commands need an existing config
        let env = LocalEnv::load_config(&local_env::base_path()).context("Error loading config")?;
        let original_env = env.clone();
@@ -728,11 +751,12 @@ fn main() -> Result<()> {
            NeonLocalCmd::Mappings(subcmd) => handle_mappings(&subcmd, env),
        };

-        if &original_env != env {
+        let subcommand_result = if &original_env != env {
            subcommand_result.map(|()| Some(Cow::Borrowed(env)))
        } else {
            subcommand_result.map(|()| None)
-        }
+        };
+        (subcommand_result, Some(_repo_lock))
    };

    match subcommand_result {
@@ -922,7 +946,7 @@ fn handle_init(args: &InitCmdArgs) -> anyhow::Result<LocalEnv> {
    } else {
        // User (likely interactive) did not provide a description of the environment, give them the default
        NeonLocalInitConf {
-            control_plane_api: Some(Some(DEFAULT_PAGESERVER_CONTROL_PLANE_API.parse().unwrap())),
+            control_plane_api: Some(DEFAULT_PAGESERVER_CONTROL_PLANE_API.parse().unwrap()),
            broker: NeonBroker {
                listen_addr: DEFAULT_BROKER_ADDR.parse().unwrap(),
            },
@@ -1718,18 +1742,15 @@ async fn handle_start_all_impl(
            broker::start_broker_process(env, &retry_timeout).await
        });

-        // Only start the storage controller if the pageserver is configured to need it
-        if env.control_plane_api.is_some() {
-            js.spawn(async move {
-                let storage_controller = StorageController::from_env(env);
-                storage_controller
-                    .start(NeonStorageControllerStartArgs::with_default_instance_id(
-                        retry_timeout,
-                    ))
-                    .await
-                    .map_err(|e| e.context("start storage_controller"))
-            });
-        }
+        js.spawn(async move {
+            let storage_controller = StorageController::from_env(env);
+            storage_controller
+                .start(NeonStorageControllerStartArgs::with_default_instance_id(
+                    retry_timeout,
+                ))
+                .await
+                .map_err(|e| e.context("start storage_controller"))
+        });

        for ps_conf in &env.pageservers {
            js.spawn(async move {
@@ -1774,10 +1795,6 @@ async fn neon_start_status_check(
    const RETRY_INTERVAL: Duration = Duration::from_millis(100);
    const NOTICE_AFTER_RETRIES: Duration = Duration::from_secs(5);

-    if env.control_plane_api.is_none() {
-        return Ok(());
-    }
-
    let storcon = StorageController::from_env(env);

    let retries = retry_timeout.as_millis() / RETRY_INTERVAL.as_millis();
--- a/control_plane/src/endpoint.rs
+++ b/control_plane/src/endpoint.rs
@@ -316,6 +316,10 @@ impl Endpoint {
        // and can cause errors like 'no unpinned buffers available', see
        // <https://github.com/neondatabase/neon/issues/9956>
        conf.append("shared_buffers", "1MB");
+        // Postgres defaults to effective_io_concurrency=1, which does not exercise the pageserver's
+        // batching logic.  Set this to 2 so that we exercise the code a bit without letting
+        // individual tests do a lot of concurrent work on underpowered test machines
+        conf.append("effective_io_concurrency", "2");
        conf.append("fsync", "off");
        conf.append("max_connections", "100");
        conf.append("wal_level", "logical");
@@ -810,7 +814,7 @@ impl Endpoint {
        }

        let client = reqwest::Client::builder()
-            .timeout(Duration::from_secs(30))
+            .timeout(Duration::from_secs(120))
            .build()
            .unwrap();
        let response = client
--- a/control_plane/src/local_env.rs
+++ b/control_plane/src/local_env.rs
@@ -76,7 +76,7 @@ pub struct LocalEnv {

    // Control plane upcall API for pageserver: if None, we will not run storage_controller  If set, this will
    // be propagated into each pageserver's configuration.
-    pub control_plane_api: Option<Url>,
+    pub control_plane_api: Url,

    // Control plane upcall API for storage controller.  If set, this will be propagated into the
    // storage controller's configuration.
@@ -133,7 +133,7 @@ pub struct NeonLocalInitConf {
    pub storage_controller: Option<NeonStorageControllerConf>,
    pub pageservers: Vec<NeonLocalInitPageserverConf>,
    pub safekeepers: Vec<SafekeeperConf>,
-    pub control_plane_api: Option<Option<Url>>,
+    pub control_plane_api: Option<Url>,
    pub control_plane_compute_hook_api: Option<Option<Url>>,
 }

@@ -180,7 +180,7 @@ impl NeonStorageControllerConf {
    const DEFAULT_MAX_WARMING_UP_INTERVAL: std::time::Duration = std::time::Duration::from_secs(30);

    // Very tight heartbeat interval to speed up tests
-    const DEFAULT_HEARTBEAT_INTERVAL: std::time::Duration = std::time::Duration::from_millis(100);
+    const DEFAULT_HEARTBEAT_INTERVAL: std::time::Duration = std::time::Duration::from_millis(1000);
 }

 impl Default for NeonStorageControllerConf {
@@ -535,7 +535,7 @@ impl LocalEnv {
                storage_controller,
                pageservers,
                safekeepers,
-                control_plane_api,
+                control_plane_api: control_plane_api.unwrap(),
                control_plane_compute_hook_api,
                branch_name_mappings,
            }
@@ -638,7 +638,7 @@ impl LocalEnv {
                storage_controller: self.storage_controller.clone(),
                pageservers: vec![], // it's skip_serializing anyway
                safekeepers: self.safekeepers.clone(),
-                control_plane_api: self.control_plane_api.clone(),
+                control_plane_api: Some(self.control_plane_api.clone()),
                control_plane_compute_hook_api: self.control_plane_compute_hook_api.clone(),
                branch_name_mappings: self.branch_name_mappings.clone(),
            },
@@ -768,7 +768,7 @@ impl LocalEnv {
            storage_controller: storage_controller.unwrap_or_default(),
            pageservers: pageservers.iter().map(Into::into).collect(),
            safekeepers,
-            control_plane_api: control_plane_api.unwrap_or_default(),
+            control_plane_api: control_plane_api.unwrap(),
            control_plane_compute_hook_api: control_plane_compute_hook_api.unwrap_or_default(),
            branch_name_mappings: Default::default(),
        };
--- a/control_plane/src/pageserver.rs
+++ b/control_plane/src/pageserver.rs
@@ -95,21 +95,19 @@ impl PageServerNode {

        let mut overrides = vec![pg_distrib_dir_param, broker_endpoint_param];

-        if let Some(control_plane_api) = &self.env.control_plane_api {
-            overrides.push(format!(
-                "control_plane_api='{}'",
-                control_plane_api.as_str()
-            ));
+        overrides.push(format!(
+            "control_plane_api='{}'",
+            self.env.control_plane_api.as_str()
+        ));

-            // Storage controller uses the same auth as pageserver: if JWT is enabled
-            // for us, we will also need it to talk to them.
-            if matches!(conf.http_auth_type, AuthType::NeonJWT) {
-                let jwt_token = self
-                    .env
-                    .generate_auth_token(&Claims::new(None, Scope::GenerationsApi))
-                    .unwrap();
-                overrides.push(format!("control_plane_api_token='{}'", jwt_token));
-            }
+        // Storage controller uses the same auth as pageserver: if JWT is enabled
+        // for us, we will also need it to talk to them.
+        if matches!(conf.http_auth_type, AuthType::NeonJWT) {
+            let jwt_token = self
+                .env
+                .generate_auth_token(&Claims::new(None, Scope::GenerationsApi))
+                .unwrap();
+            overrides.push(format!("control_plane_api_token='{}'", jwt_token));
        }

        if !conf.other.contains_key("remote_storage") {
@@ -435,7 +433,7 @@ impl PageServerNode {
    ) -> anyhow::Result<()> {
        let config = Self::parse_config(settings)?;
        self.http_client
-            .tenant_config(&models::TenantConfigRequest { tenant_id, config })
+            .set_tenant_config(&models::TenantConfigRequest { tenant_id, config })
            .await?;

        Ok(())
--- a/control_plane/src/storage_controller.rs
+++ b/control_plane/src/storage_controller.rs
@@ -338,7 +338,7 @@ impl StorageController {
                        .port(),
                )
            } else {
-                let listen_url = self.env.control_plane_api.clone().unwrap();
+                let listen_url = self.env.control_plane_api.clone();

                let listen = format!(
                    "{}:{}",
@@ -708,7 +708,7 @@ impl StorageController {
        } else {
            // The configured URL has the /upcall path prefix for pageservers to use: we will strip that out
            // for general purpose API access.
-            let listen_url = self.env.control_plane_api.clone().unwrap();
+            let listen_url = self.env.control_plane_api.clone();
            Url::from_str(&format!(
                "http://{}:{}/{path}",
                listen_url.host_str().unwrap(),
--- a/control_plane/storcon_cli/src/main.rs
+++ b/control_plane/storcon_cli/src/main.rs
@@ -5,12 +5,13 @@ use clap::{Parser, Subcommand};
 use pageserver_api::{
    controller_api::{
        AvailabilityZone, NodeAvailabilityWrapper, NodeDescribeResponse, NodeShardResponse,
-        ShardSchedulingPolicy, TenantCreateRequest, TenantDescribeResponse, TenantPolicyRequest,
+        SafekeeperDescribeResponse, ShardSchedulingPolicy, TenantCreateRequest,
+        TenantDescribeResponse, TenantPolicyRequest,
    },
    models::{
        EvictionPolicy, EvictionPolicyLayerAccessThreshold, LocationConfigSecondary,
-        ShardParameters, TenantConfig, TenantConfigRequest, TenantShardSplitRequest,
-        TenantShardSplitResponse,
+        ShardParameters, TenantConfig, TenantConfigPatchRequest, TenantConfigRequest,
+        TenantShardSplitRequest, TenantShardSplitResponse,
    },
    shard::{ShardStripeSize, TenantShardId},
 };
@@ -116,9 +117,19 @@ enum Command {
        #[arg(long)]
        tenant_shard_id: TenantShardId,
    },
-    /// Modify the pageserver tenant configuration of a tenant: this is the configuration structure
+    /// Set the pageserver tenant configuration of a tenant: this is the configuration structure
    /// that is passed through to pageservers, and does not affect storage controller behavior.
-    TenantConfig {
+    /// Any previous tenant configs are overwritten.
+    SetTenantConfig {
+        #[arg(long)]
+        tenant_id: TenantId,
+        #[arg(long)]
+        config: String,
+    },
+    /// Patch the pageserver tenant configuration of a tenant. Any fields with null values in the
+    /// provided JSON are unset from the tenant config and all fields with non-null values are set.
+    /// Unspecified fields are not changed.
+    PatchTenantConfig {
        #[arg(long)]
        tenant_id: TenantId,
        #[arg(long)]
@@ -201,6 +212,8 @@ enum Command {
        #[arg(long)]
        timeout: humantime::Duration,
    },
+    /// List safekeepers known to the storage controller
+    Safekeepers {},
 }

 #[derive(Parser)]
@@ -549,11 +562,21 @@ async fn main() -> anyhow::Result<()> {
                )
                .await?;
        }
-        Command::TenantConfig { tenant_id, config } => {
+        Command::SetTenantConfig { tenant_id, config } => {
            let tenant_conf = serde_json::from_str(&config)?;

            vps_client
-                .tenant_config(&TenantConfigRequest {
+                .set_tenant_config(&TenantConfigRequest {
+                    tenant_id,
+                    config: tenant_conf,
+                })
+                .await?;
+        }
+        Command::PatchTenantConfig { tenant_id, config } => {
+            let tenant_conf = serde_json::from_str(&config)?;
+
+            vps_client
+                .patch_tenant_config(&TenantConfigPatchRequest {
                    tenant_id,
                    config: tenant_conf,
                })
@@ -736,7 +759,7 @@ async fn main() -> anyhow::Result<()> {
            threshold,
        } => {
            vps_client
-                .tenant_config(&TenantConfigRequest {
+                .set_tenant_config(&TenantConfigRequest {
                    tenant_id,
                    config: TenantConfig {
                        eviction_policy: Some(EvictionPolicy::LayerAccessThreshold(
@@ -1000,6 +1023,31 @@ async fn main() -> anyhow::Result<()> {
                "Fill was cancelled for node {node_id}. Schedulling policy is now {final_policy:?}"
            );
        }
+        Command::Safekeepers {} => {
+            let mut resp = storcon_client
+                .dispatch::<(), Vec<SafekeeperDescribeResponse>>(
+                    Method::GET,
+                    "control/v1/safekeeper".to_string(),
+                    None,
+                )
+                .await?;
+
+            resp.sort_by(|a, b| a.id.cmp(&b.id));
+
+            let mut table = comfy_table::Table::new();
+            table.set_header(["Id", "Version", "Host", "Port", "Http Port", "AZ Id"]);
+            for sk in resp {
+                table.add_row([
+                    format!("{}", sk.id),
+                    format!("{}", sk.version),
+                    sk.host,
+                    format!("{}", sk.port),
+                    format!("{}", sk.http_port),
+                    sk.availability_zone_id.to_string(),
+                ]);
+            }
+            println!("{table}");
+        }
    }

    Ok(())
--- a/deny.toml
+++ b/deny.toml
@@ -42,6 +42,7 @@ allow = [
    "MPL-2.0",
    "OpenSSL",
    "Unicode-DFS-2016",
+    "Unicode-3.0",
 ]
 confidence-threshold = 0.8
 exceptions = [
--- a/docker-compose/compute_wrapper/var/db/postgres/specs/spec.json
+++ b/docker-compose/compute_wrapper/var/db/postgres/specs/spec.json
@@ -132,11 +132,6 @@
                "name": "cron.database",
                "value": "postgres",
                "vartype": "string"
-            },
-            {
-                "name": "session_preload_libraries",
-                "value": "anon",
-                "vartype": "string"
            }
        ]
    },
--- a/docker-compose/docker_compose_test.sh
+++ b/docker-compose/docker_compose_test.sh
@@ -35,11 +35,11 @@ for pg_version in ${TEST_VERSION_ONLY-14 15 16 17}; do
    echo "clean up containers if exists"
    cleanup
    PG_TEST_VERSION=$((pg_version < 16 ? 16 : pg_version))
-    # The support of pg_anon not yet added to PG17, so we have to remove the corresponding option
-    if [ $pg_version -eq 17 ]; then
+    # The support of pg_anon not yet added to PG17, so we have to add the corresponding option for other PG versions
+    if [ "${pg_version}" -ne 17 ]; then
      SPEC_PATH="compute_wrapper/var/db/postgres/specs"
      mv $SPEC_PATH/spec.json $SPEC_PATH/spec.bak
-      jq 'del(.cluster.settings[] | select (.name == "session_preload_libraries"))' $SPEC_PATH/spec.bak > $SPEC_PATH/spec.json
+      jq '.cluster.settings += [{"name": "session_preload_libraries","value": "anon","vartype": "string"}]' "${SPEC_PATH}/spec.bak" > "${SPEC_PATH}/spec.json"
    fi
    PG_VERSION=$pg_version PG_TEST_VERSION=$PG_TEST_VERSION docker compose --profile test-extensions -f $COMPOSE_FILE up --build -d

@@ -106,8 +106,8 @@ for pg_version in ${TEST_VERSION_ONLY-14 15 16 17}; do
        fi
    fi
    cleanup
-    # The support of pg_anon not yet added to PG17, so we have to remove the corresponding option
-    if [ $pg_version -eq 17 ]; then
-      mv $SPEC_PATH/spec.bak $SPEC_PATH/spec.json
+    # Restore the original spec.json
+    if [ "$pg_version" -ne 17 ]; then
+      mv "$SPEC_PATH/spec.bak" "$SPEC_PATH/spec.json"
    fi
 done
--- a/libs/compute_api/src/responses.rs
+++ b/libs/compute_api/src/responses.rs
@@ -1,6 +1,5 @@
 //! Structs representing the JSON formats used in the compute_ctl's HTTP API.

-use std::collections::HashSet;
 use std::fmt::Display;

 use chrono::{DateTime, Utc};
@@ -163,8 +162,9 @@ pub enum ControlPlaneComputeStatus {
 #[derive(Clone, Debug, Default, Serialize)]
 pub struct InstalledExtension {
    pub extname: String,
-    pub versions: HashSet<String>,
+    pub version: String,
    pub n_databases: u32, // Number of databases using this extension
+    pub owned_by_superuser: String,
 }

 #[derive(Clone, Debug, Default, Serialize)]
--- a/libs/desim/src/time.rs
+++ b/libs/desim/src/time.rs
@@ -91,7 +91,7 @@ impl Timing {

    /// Return true if there is a ready event.
    fn is_event_ready(&self, queue: &mut BinaryHeap<Pending>) -> bool {
-        queue.peek().map_or(false, |x| x.time <= self.now())
+        queue.peek().is_some_and(|x| x.time <= self.now())
    }

    /// Clear all pending events.
--- a/libs/pageserver_api/src/controller_api.rs
+++ b/libs/pageserver_api/src/controller_api.rs
@@ -75,7 +75,7 @@ pub struct TenantPolicyRequest {
    pub scheduling: Option<ShardSchedulingPolicy>,
 }

-#[derive(Clone, Serialize, Deserialize, PartialEq, Eq, Hash, Debug)]
+#[derive(Clone, Serialize, Deserialize, PartialEq, Eq, Hash, Debug, PartialOrd, Ord)]
 pub struct AvailabilityZone(pub String);

 impl Display for AvailabilityZone {
@@ -245,6 +245,17 @@ impl From<NodeAvailability> for NodeAvailabilityWrapper {
    }
 }

+/// Scheduling policy enables us to selectively disable some automatic actions that the
+/// controller performs on a tenant shard. This is only set to a non-default value by
+/// human intervention, and it is reset to the default value (Active) when the tenant's
+/// placement policy is modified away from Attached.
+///
+/// The typical use of a non-Active scheduling policy is one of:
+/// - Pinnning a shard to a node (i.e. migrating it there & setting a non-Active scheduling policy)
+/// - Working around a bug (e.g. if something is flapping and we need to stop it until the bug is fixed)
+///
+/// If you're not sure which policy to use to pin a shard to its current location, you probably
+/// want Pause.
 #[derive(Serialize, Deserialize, Clone, Copy, Eq, PartialEq, Debug)]
 pub enum ShardSchedulingPolicy {
    // Normal mode: the tenant's scheduled locations may be updated at will, including
@@ -361,6 +372,23 @@ pub struct MetadataHealthListOutdatedResponse {
    pub health_records: Vec<MetadataHealthRecord>,
 }

+/// Publicly exposed safekeeper description
+///
+/// The `active` flag which we have in the DB is not included on purpose: it is deprecated.
+#[derive(Serialize, Deserialize, Clone)]
+pub struct SafekeeperDescribeResponse {
+    pub id: NodeId,
+    pub region_id: String,
+    /// 1 is special, it means just created (not currently posted to storcon).
+    /// Zero or negative is not really expected.
+    /// Otherwise the number from `release-$(number_of_commits_on_branch)` tag.
+    pub version: i64,
+    pub host: String,
+    pub port: i32,
+    pub http_port: i32,
+    pub availability_zone_id: String,
+}
+
 #[cfg(test)]
 mod test {
    use super::*;
--- a/libs/pageserver_api/src/key.rs
+++ b/libs/pageserver_api/src/key.rs
@@ -24,7 +24,7 @@ pub struct Key {

 /// When working with large numbers of Keys in-memory, it is more efficient to handle them as i128 than as
 /// a struct of fields.
-#[derive(Clone, Copy, Hash, PartialEq, Eq, Ord, PartialOrd, Serialize, Deserialize)]
+#[derive(Clone, Copy, Hash, PartialEq, Eq, Ord, PartialOrd, Serialize, Deserialize, Debug)]
 pub struct CompactKey(i128);

 /// The storage key size.
@@ -565,6 +565,10 @@ impl Key {
            && self.field5 == 0
            && self.field6 == u32::MAX
    }
+
+    pub fn is_slru_dir_key(&self) -> bool {
+        slru_dir_kind(self).is_some()
+    }
 }

 #[inline(always)]
--- a/libs/pageserver_api/src/models.rs
+++ b/libs/pageserver_api/src/models.rs
@@ -6,6 +6,7 @@ pub mod utilization;
 use camino::Utf8PathBuf;
 pub use utilization::PageserverUtilization;

+use core::ops::Range;
 use std::{
    collections::HashMap,
    fmt::Display,
@@ -17,7 +18,7 @@ use std::{

 use byteorder::{BigEndian, ReadBytesExt};
 use postgres_ffi::BLCKSZ;
-use serde::{Deserialize, Serialize};
+use serde::{Deserialize, Deserializer, Serialize, Serializer};
 use serde_with::serde_as;
 use utils::{
    completion,
@@ -28,6 +29,7 @@ use utils::{
 };

 use crate::{
+    key::Key,
    reltag::RelTag,
    shard::{ShardCount, ShardStripeSize, TenantShardId},
 };
@@ -210,6 +212,68 @@ pub enum TimelineState {
    Broken { reason: String, backtrace: String },
 }

+#[serde_with::serde_as]
+#[derive(Debug, Clone, serde::Deserialize, serde::Serialize)]
+pub struct CompactLsnRange {
+    pub start: Lsn,
+    pub end: Lsn,
+}
+
+#[serde_with::serde_as]
+#[derive(Debug, Clone, serde::Deserialize, serde::Serialize)]
+pub struct CompactKeyRange {
+    #[serde_as(as = "serde_with::DisplayFromStr")]
+    pub start: Key,
+    #[serde_as(as = "serde_with::DisplayFromStr")]
+    pub end: Key,
+}
+
+impl From<Range<Lsn>> for CompactLsnRange {
+    fn from(range: Range<Lsn>) -> Self {
+        Self {
+            start: range.start,
+            end: range.end,
+        }
+    }
+}
+
+impl From<Range<Key>> for CompactKeyRange {
+    fn from(range: Range<Key>) -> Self {
+        Self {
+            start: range.start,
+            end: range.end,
+        }
+    }
+}
+
+impl From<CompactLsnRange> for Range<Lsn> {
+    fn from(range: CompactLsnRange) -> Self {
+        range.start..range.end
+    }
+}
+
+impl From<CompactKeyRange> for Range<Key> {
+    fn from(range: CompactKeyRange) -> Self {
+        range.start..range.end
+    }
+}
+
+impl CompactLsnRange {
+    pub fn above(lsn: Lsn) -> Self {
+        Self {
+            start: lsn,
+            end: Lsn::MAX,
+        }
+    }
+}
+
+#[derive(Debug, Clone, Serialize)]
+pub struct CompactInfoResponse {
+    pub compact_key_range: Option<CompactKeyRange>,
+    pub compact_lsn_range: Option<CompactLsnRange>,
+    pub sub_compaction: bool,
+}
+
 #[derive(Serialize, Deserialize, Clone)]
 pub struct TimelineCreateRequest {
    pub new_timeline_id: TimelineId,
@@ -325,6 +389,115 @@ impl Default for ShardParameters {
    }
 }

+#[derive(Debug, Default, Clone, Eq, PartialEq)]
+pub enum FieldPatch<T> {
+    Upsert(T),
+    Remove,
+    #[default]
+    Noop,
+}
+
+impl<T> FieldPatch<T> {
+    fn is_noop(&self) -> bool {
+        matches!(self, FieldPatch::Noop)
+    }
+
+    pub fn apply(self, target: &mut Option<T>) {
+        match self {
+            Self::Upsert(v) => *target = Some(v),
+            Self::Remove => *target = None,
+            Self::Noop => {}
+        }
+    }
+
+    pub fn map<U, E, F: FnOnce(T) -> Result<U, E>>(self, map: F) -> Result<FieldPatch<U>, E> {
+        match self {
+            Self::Upsert(v) => Ok(FieldPatch::<U>::Upsert(map(v)?)),
+            Self::Remove => Ok(FieldPatch::<U>::Remove),
+            Self::Noop => Ok(FieldPatch::<U>::Noop),
+        }
+    }
+}
+
+impl<'de, T: Deserialize<'de>> Deserialize<'de> for FieldPatch<T> {
+    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
+    where
+        D: Deserializer<'de>,
+    {
+        Option::deserialize(deserializer).map(|opt| match opt {
+            None => FieldPatch::Remove,
+            Some(val) => FieldPatch::Upsert(val),
+        })
+    }
+}
+
+impl<T: Serialize> Serialize for FieldPatch<T> {
+    fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
+    where
+        S: Serializer,
+    {
+        match self {
+            FieldPatch::Upsert(val) => serializer.serialize_some(val),
+            FieldPatch::Remove => serializer.serialize_none(),
+            FieldPatch::Noop => unreachable!(),
+        }
+    }
+}
+
+#[derive(Serialize, Deserialize, Debug, Default, Clone, Eq, PartialEq)]
+#[serde(default)]
+pub struct TenantConfigPatch {
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub checkpoint_distance: FieldPatch<u64>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub checkpoint_timeout: FieldPatch<String>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub compaction_target_size: FieldPatch<u64>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub compaction_period: FieldPatch<String>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub compaction_threshold: FieldPatch<usize>,
+    // defer parsing compaction_algorithm, like eviction_policy
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub compaction_algorithm: FieldPatch<CompactionAlgorithmSettings>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub gc_horizon: FieldPatch<u64>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub gc_period: FieldPatch<String>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub image_creation_threshold: FieldPatch<usize>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub pitr_interval: FieldPatch<String>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub walreceiver_connect_timeout: FieldPatch<String>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub lagging_wal_timeout: FieldPatch<String>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub max_lsn_wal_lag: FieldPatch<NonZeroU64>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub eviction_policy: FieldPatch<EvictionPolicy>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub min_resident_size_override: FieldPatch<u64>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub evictions_low_residence_duration_metric_threshold: FieldPatch<String>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub heatmap_period: FieldPatch<String>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub lazy_slru_download: FieldPatch<bool>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub timeline_get_throttle: FieldPatch<ThrottleConfig>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub image_layer_creation_check_threshold: FieldPatch<u8>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub lsn_lease_length: FieldPatch<String>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub lsn_lease_length_for_ts: FieldPatch<String>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub timeline_offloading: FieldPatch<bool>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub wal_receiver_protocol_override: FieldPatch<PostgresClientProtocol>,
+}
+
 /// An alternative representation of `pageserver::tenant::TenantConf` with
 /// simpler types.
 #[derive(Serialize, Deserialize, Debug, Default, Clone, Eq, PartialEq)]
@@ -356,6 +529,107 @@ pub struct TenantConfig {
    pub wal_receiver_protocol_override: Option<PostgresClientProtocol>,
 }

+impl TenantConfig {
+    pub fn apply_patch(self, patch: TenantConfigPatch) -> TenantConfig {
+        let Self {
+            mut checkpoint_distance,
+            mut checkpoint_timeout,
+            mut compaction_target_size,
+            mut compaction_period,
+            mut compaction_threshold,
+            mut compaction_algorithm,
+            mut gc_horizon,
+            mut gc_period,
+            mut image_creation_threshold,
+            mut pitr_interval,
+            mut walreceiver_connect_timeout,
+            mut lagging_wal_timeout,
+            mut max_lsn_wal_lag,
+            mut eviction_policy,
+            mut min_resident_size_override,
+            mut evictions_low_residence_duration_metric_threshold,
+            mut heatmap_period,
+            mut lazy_slru_download,
+            mut timeline_get_throttle,
+            mut image_layer_creation_check_threshold,
+            mut lsn_lease_length,
+            mut lsn_lease_length_for_ts,
+            mut timeline_offloading,
+            mut wal_receiver_protocol_override,
+        } = self;
+
+        patch.checkpoint_distance.apply(&mut checkpoint_distance);
+        patch.checkpoint_timeout.apply(&mut checkpoint_timeout);
+        patch
+            .compaction_target_size
+            .apply(&mut compaction_target_size);
+        patch.compaction_period.apply(&mut compaction_period);
+        patch.compaction_threshold.apply(&mut compaction_threshold);
+        patch.compaction_algorithm.apply(&mut compaction_algorithm);
+        patch.gc_horizon.apply(&mut gc_horizon);
+        patch.gc_period.apply(&mut gc_period);
+        patch
+            .image_creation_threshold
+            .apply(&mut image_creation_threshold);
+        patch.pitr_interval.apply(&mut pitr_interval);
+        patch
+            .walreceiver_connect_timeout
+            .apply(&mut walreceiver_connect_timeout);
+        patch.lagging_wal_timeout.apply(&mut lagging_wal_timeout);
+        patch.max_lsn_wal_lag.apply(&mut max_lsn_wal_lag);
+        patch.eviction_policy.apply(&mut eviction_policy);
+        patch
+            .min_resident_size_override
+            .apply(&mut min_resident_size_override);
+        patch
+            .evictions_low_residence_duration_metric_threshold
+            .apply(&mut evictions_low_residence_duration_metric_threshold);
+        patch.heatmap_period.apply(&mut heatmap_period);
+        patch.lazy_slru_download.apply(&mut lazy_slru_download);
+        patch
+            .timeline_get_throttle
+            .apply(&mut timeline_get_throttle);
+        patch
+            .image_layer_creation_check_threshold
+            .apply(&mut image_layer_creation_check_threshold);
+        patch.lsn_lease_length.apply(&mut lsn_lease_length);
+        patch
+            .lsn_lease_length_for_ts
+            .apply(&mut lsn_lease_length_for_ts);
+        patch.timeline_offloading.apply(&mut timeline_offloading);
+        patch
+            .wal_receiver_protocol_override
+            .apply(&mut wal_receiver_protocol_override);
+
+        Self {
+            checkpoint_distance,
+            checkpoint_timeout,
+            compaction_target_size,
+            compaction_period,
+            compaction_threshold,
+            compaction_algorithm,
+            gc_horizon,
+            gc_period,
+            image_creation_threshold,
+            pitr_interval,
+            walreceiver_connect_timeout,
+            lagging_wal_timeout,
+            max_lsn_wal_lag,
+            eviction_policy,
+            min_resident_size_override,
+            evictions_low_residence_duration_metric_threshold,
+            heatmap_period,
+            lazy_slru_download,
+            timeline_get_throttle,
+            image_layer_creation_check_threshold,
+            lsn_lease_length,
+            lsn_lease_length_for_ts,
+            timeline_offloading,
+            wal_receiver_protocol_override,
+        }
+    }
+}
+
 /// The policy for the aux file storage.
 ///
 /// It can be switched through `switch_aux_file_policy` tenant config.
@@ -686,6 +960,14 @@ impl TenantConfigRequest {
    }
 }

+#[derive(Serialize, Deserialize, Debug)]
+#[serde(deny_unknown_fields)]
+pub struct TenantConfigPatchRequest {
+    pub tenant_id: TenantId,
+    #[serde(flatten)]
+    pub config: TenantConfigPatch, // as we have a flattened field, we should reject all unknown fields in it
+}
+
 /// See [`TenantState::attachment_status`] and the OpenAPI docs for context.
 #[derive(Serialize, Deserialize, Clone)]
 #[serde(tag = "slug", content = "data", rename_all = "snake_case")]
@@ -1699,4 +1981,45 @@ mod tests {
            );
        }
    }
+
+    #[test]
+    fn test_tenant_config_patch_request_serde() {
+        let patch_request = TenantConfigPatchRequest {
+            tenant_id: TenantId::from_str("17c6d121946a61e5ab0fe5a2fd4d8215").unwrap(),
+            config: TenantConfigPatch {
+                checkpoint_distance: FieldPatch::Upsert(42),
+                gc_horizon: FieldPatch::Remove,
+                compaction_threshold: FieldPatch::Noop,
+                ..TenantConfigPatch::default()
+            },
+        };
+
+        let json = serde_json::to_string(&patch_request).unwrap();
+
+        let expected = r#"{"tenant_id":"17c6d121946a61e5ab0fe5a2fd4d8215","checkpoint_distance":42,"gc_horizon":null}"#;
+        assert_eq!(json, expected);
+
+        let decoded: TenantConfigPatchRequest = serde_json::from_str(&json).unwrap();
+        assert_eq!(decoded.tenant_id, patch_request.tenant_id);
+        assert_eq!(decoded.config, patch_request.config);
+
+        // Now apply the patch to a config to demonstrate semantics
+
+        let base = TenantConfig {
+            checkpoint_distance: Some(28),
+            gc_horizon: Some(100),
+            compaction_target_size: Some(1024),
+            ..Default::default()
+        };
+
+        let expected = TenantConfig {
+            checkpoint_distance: Some(42),
+            gc_horizon: None,
+            ..base.clone()
+        };
+
+        let patched = base.apply_patch(decoded.config);
+
+        assert_eq!(patched, expected);
+    }
 }
--- a/libs/pageserver_api/src/shard.rs
+++ b/libs/pageserver_api/src/shard.rs
@@ -173,7 +173,11 @@ impl ShardIdentity {

    /// Return true if the key should be stored on all shards, not just one.
    pub fn is_key_global(&self, key: &Key) -> bool {
-        if key.is_slru_block_key() || key.is_slru_segment_size_key() || key.is_aux_file_key() {
+        if key.is_slru_block_key()
+            || key.is_slru_segment_size_key()
+            || key.is_aux_file_key()
+            || key.is_slru_dir_key()
+        {
            // Special keys that are only stored on shard 0
            false
        } else if key.is_rel_block_key() {
--- a/libs/postgres_ffi/Cargo.toml
+++ b/libs/postgres_ffi/Cargo.toml
@@ -9,9 +9,11 @@ regex.workspace = true
 bytes.workspace = true
 anyhow.workspace = true
 crc32c.workspace = true
+criterion.workspace = true
 once_cell.workspace = true
 log.workspace = true
 memoffset.workspace = true
+pprof.workspace = true
 thiserror.workspace = true
 serde.workspace = true
 utils.workspace = true
@@ -24,3 +26,7 @@ postgres.workspace = true
 [build-dependencies]
 anyhow.workspace = true
 bindgen.workspace = true
+
+[[bench]]
+name = "waldecoder"
+harness = false
--- a/libs/postgres_ffi/benches/README.md
+++ b/libs/postgres_ffi/benches/README.md
@@ -0,0 +1,26 @@
+## Benchmarks
+
+To run benchmarks:
+
+```sh
+# All benchmarks.
+cargo bench --package postgres_ffi
+
+# Specific file.
+cargo bench --package postgres_ffi --bench waldecoder
+
+# Specific benchmark.
+cargo bench --package postgres_ffi --bench waldecoder complete_record/size=1024
+
+# List available benchmarks.
+cargo bench --package postgres_ffi --benches -- --list
+
+# Generate flamegraph profiles using pprof-rs, profiling for 10 seconds.
+# Output in target/criterion/*/profile/flamegraph.svg.
+cargo bench --package postgres_ffi --bench waldecoder complete_record/size=1024 -- --profile-time 10
+```
+
+Additional charts and statistics are available in `target/criterion/report/index.html`.
+
+Benchmarks are automatically compared against the previous run. To compare against other runs, see
+`--baseline` and `--save-baseline`.
--- a/libs/postgres_ffi/benches/waldecoder.rs
+++ b/libs/postgres_ffi/benches/waldecoder.rs
@@ -0,0 +1,49 @@
+use std::ffi::CStr;
+
+use criterion::{criterion_group, criterion_main, Bencher, Criterion};
+use postgres_ffi::v17::wal_generator::LogicalMessageGenerator;
+use postgres_ffi::v17::waldecoder_handler::WalStreamDecoderHandler;
+use postgres_ffi::waldecoder::WalStreamDecoder;
+use pprof::criterion::{Output, PProfProfiler};
+use utils::lsn::Lsn;
+
+const KB: usize = 1024;
+
+// Register benchmarks with Criterion.
+criterion_group!(
+    name = benches;
+    config = Criterion::default().with_profiler(PProfProfiler::new(100, Output::Flamegraph(None)));
+    targets = bench_complete_record,
+);
+criterion_main!(benches);
+
+/// Benchmarks WalStreamDecoder::complete_record() for a logical message of varying size.
+fn bench_complete_record(c: &mut Criterion) {
+    let mut g = c.benchmark_group("complete_record");
+    for size in [64, KB, 8 * KB, 128 * KB] {
+        // Kind of weird to change the group throughput per benchmark, but it's the only way
+        // to vary it per benchmark. It works.
+        g.throughput(criterion::Throughput::Bytes(size as u64));
+        g.bench_function(format!("size={size}"), |b| run_bench(b, size).unwrap());
+    }
+
+    fn run_bench(b: &mut Bencher, size: usize) -> anyhow::Result<()> {
+        const PREFIX: &CStr = c"";
+        let value_size = LogicalMessageGenerator::make_value_size(size, PREFIX);
+        let value = vec![1; value_size];
+
+        let mut decoder = WalStreamDecoder::new(Lsn(0), 170000);
+        let msg = LogicalMessageGenerator::new(PREFIX, &value)
+            .next()
+            .unwrap()
+            .encode(Lsn(0));
+        assert_eq!(msg.len(), size);
+
+        b.iter(|| {
+            let msg = msg.clone(); // Bytes::clone() is cheap
+            decoder.complete_record(msg).unwrap();
+        });
+
+        Ok(())
+    }
+}
--- a/libs/postgres_ffi/src/wal_generator.rs
+++ b/libs/postgres_ffi/src/wal_generator.rs
@@ -106,11 +106,11 @@ impl<R: RecordGenerator> WalGenerator<R> {
    const TIMELINE_ID: u32 = 1;

    /// Creates a new WAL generator with the given record generator.
-    pub fn new(record_generator: R) -> WalGenerator<R> {
+    pub fn new(record_generator: R, start_lsn: Lsn) -> WalGenerator<R> {
        Self {
            record_generator,
-            lsn: Lsn(0),
-            prev_lsn: Lsn(0),
+            lsn: start_lsn,
+            prev_lsn: start_lsn,
        }
    }

@@ -231,6 +231,22 @@ impl LogicalMessageGenerator {
        };
        [&header.encode(), prefix, message].concat().into()
    }
+
+    /// Computes how large a value must be to get a record of the given size. Convenience method to
+    /// construct records of pre-determined size. Panics if the record size is too small.
+    pub fn make_value_size(record_size: usize, prefix: &CStr) -> usize {
+        let xlog_header_size = XLOG_SIZE_OF_XLOG_RECORD;
+        let lm_header_size = size_of::<XlLogicalMessage>();
+        let prefix_size = prefix.to_bytes_with_nul().len();
+        let data_header_size = match record_size - xlog_header_size - 2 {
+            0..=255 => 2,
+            256..=258 => panic!("impossible record_size {record_size}"),
+            259.. => 5,
+        };
+        record_size
+            .checked_sub(xlog_header_size + lm_header_size + prefix_size + data_header_size)
+            .expect("record_size too small")
+    }
 }

 impl Iterator for LogicalMessageGenerator {
--- a/libs/postgres_ffi/wal_craft/src/xlog_utils_test.rs
+++ b/libs/postgres_ffi/wal_craft/src/xlog_utils_test.rs
@@ -81,7 +81,7 @@ fn test_end_of_wal<C: crate::Crafter>(test_name: &str) {
                continue;
            }
            let mut f = File::options().write(true).open(file.path()).unwrap();
-            const ZEROS: [u8; WAL_SEGMENT_SIZE] = [0u8; WAL_SEGMENT_SIZE];
+            static ZEROS: [u8; WAL_SEGMENT_SIZE] = [0u8; WAL_SEGMENT_SIZE];
            f.write_all(
                &ZEROS[0..min(
                    WAL_SEGMENT_SIZE,
--- a/libs/proxy/postgres-protocol2/Cargo.toml
+++ b/libs/proxy/postgres-protocol2/Cargo.toml
@@ -1,7 +1,7 @@
 [package]
 name = "postgres-protocol2"
 version = "0.1.0"
-edition = "2018"
+edition = "2021"
 license = "MIT/Apache-2.0"

 [dependencies]
--- a/libs/proxy/postgres-protocol2/src/lib.rs
+++ b/libs/proxy/postgres-protocol2/src/lib.rs
@@ -9,8 +9,7 @@
 //!
 //! This library assumes that the `client_encoding` backend parameter has been
 //! set to `UTF8`. It will most likely not behave properly if that is not the case.
-#![doc(html_root_url = "https://docs.rs/postgres-protocol/0.6")]
-#![warn(missing_docs, rust_2018_idioms, clippy::all)]
+#![warn(missing_docs, clippy::all)]

 use byteorder::{BigEndian, ByteOrder};
 use bytes::{BufMut, BytesMut};
--- a/libs/proxy/postgres-protocol2/src/message/frontend.rs
+++ b/libs/proxy/postgres-protocol2/src/message/frontend.rs
@@ -3,7 +3,6 @@

 use byteorder::{BigEndian, ByteOrder};
 use bytes::{Buf, BufMut, BytesMut};
-use std::convert::TryFrom;
 use std::error::Error;
 use std::io;
 use std::marker;
--- a/libs/proxy/postgres-types2/Cargo.toml
+++ b/libs/proxy/postgres-types2/Cargo.toml
@@ -1,7 +1,7 @@
 [package]
 name = "postgres-types2"
 version = "0.1.0"
-edition = "2018"
+edition = "2021"
 license = "MIT/Apache-2.0"

 [dependencies]
--- a/libs/proxy/postgres-types2/src/lib.rs
+++ b/libs/proxy/postgres-types2/src/lib.rs
@@ -2,8 +2,7 @@
 //!
 //! This crate is used by the `tokio-postgres` and `postgres` crates. You normally don't need to depend directly on it
 //! unless you want to define your own `ToSql` or `FromSql` definitions.
-#![doc(html_root_url = "https://docs.rs/postgres-types/0.2")]
-#![warn(clippy::all, rust_2018_idioms, missing_docs)]
+#![warn(clippy::all, missing_docs)]

 use fallible_iterator::FallibleIterator;
 use postgres_protocol2::types;
--- a/libs/proxy/tokio-postgres2/Cargo.toml
+++ b/libs/proxy/tokio-postgres2/Cargo.toml
@@ -1,7 +1,7 @@
 [package]
 name = "tokio-postgres2"
 version = "0.1.0"
-edition = "2018"
+edition = "2021"
 license = "MIT/Apache-2.0"

 [dependencies]
--- a/libs/proxy/tokio-postgres2/src/connection.rs
+++ b/libs/proxy/tokio-postgres2/src/connection.rs
@@ -33,10 +33,14 @@ pub struct Response {
 #[derive(PartialEq, Debug)]
 enum State {
    Active,
-    Terminating,
    Closing,
 }

+enum WriteReady {
+    Terminating,
+    WaitingOnRead,
+}
+
 /// A connection to a PostgreSQL database.
 ///
 /// This is one half of what is returned when a new connection is established. It performs the actual IO with the
@@ -51,7 +55,6 @@ pub struct Connection<S, T> {
    /// HACK: we need this in the Neon Proxy to forward params.
    pub parameters: HashMap<String, String>,
    receiver: mpsc::UnboundedReceiver<Request>,
-    pending_request: Option<RequestMessages>,
    pending_responses: VecDeque<BackendMessage>,
    responses: VecDeque<Response>,
    state: State,
@@ -72,7 +75,6 @@ where
            stream,
            parameters,
            receiver,
-            pending_request: None,
            pending_responses,
            responses: VecDeque::new(),
            state: State::Active,
@@ -93,26 +95,23 @@ where
            .map(|o| o.map(|r| r.map_err(Error::io)))
    }

-    fn poll_read(&mut self, cx: &mut Context<'_>) -> Result<Option<AsyncMessage>, Error> {
-        if self.state != State::Active {
-            trace!("poll_read: done");
-            return Ok(None);
-        }
-
+    /// Read and process messages from the connection to postgres.
+    /// client <- postgres
+    fn poll_read(&mut self, cx: &mut Context<'_>) -> Poll<Result<AsyncMessage, Error>> {
        loop {
            let message = match self.poll_response(cx)? {
                Poll::Ready(Some(message)) => message,
-                Poll::Ready(None) => return Err(Error::closed()),
+                Poll::Ready(None) => return Poll::Ready(Err(Error::closed())),
                Poll::Pending => {
                    trace!("poll_read: waiting on response");
-                    return Ok(None);
+                    return Poll::Pending;
                }
            };

            let (mut messages, request_complete) = match message {
                BackendMessage::Async(Message::NoticeResponse(body)) => {
                    let error = DbError::parse(&mut body.fields()).map_err(Error::parse)?;
-                    return Ok(Some(AsyncMessage::Notice(error)));
+                    return Poll::Ready(Ok(AsyncMessage::Notice(error)));
                }
                BackendMessage::Async(Message::NotificationResponse(body)) => {
                    let notification = Notification {
@@ -120,7 +119,7 @@ where
                        channel: body.channel().map_err(Error::parse)?.to_string(),
                        payload: body.message().map_err(Error::parse)?.to_string(),
                    };
-                    return Ok(Some(AsyncMessage::Notification(notification)));
+                    return Poll::Ready(Ok(AsyncMessage::Notification(notification)));
                }
                BackendMessage::Async(Message::ParameterStatus(body)) => {
                    self.parameters.insert(
@@ -139,8 +138,10 @@ where
            let mut response = match self.responses.pop_front() {
                Some(response) => response,
                None => match messages.next().map_err(Error::parse)? {
-                    Some(Message::ErrorResponse(error)) => return Err(Error::db(error)),
-                    _ => return Err(Error::unexpected_message()),
+                    Some(Message::ErrorResponse(error)) => {
+                        return Poll::Ready(Err(Error::db(error)))
+                    }
+                    _ => return Poll::Ready(Err(Error::unexpected_message())),
                },
            };

@@ -164,18 +165,14 @@ where
                        request_complete,
                    });
                    trace!("poll_read: waiting on sender");
-                    return Ok(None);
+                    return Poll::Pending;
                }
            }
        }
    }

+    /// Fetch the next client request and enqueue the response sender.
    fn poll_request(&mut self, cx: &mut Context<'_>) -> Poll<Option<RequestMessages>> {
-        if let Some(messages) = self.pending_request.take() {
-            trace!("retrying pending request");
-            return Poll::Ready(Some(messages));
-        }
-
        if self.receiver.is_closed() {
            return Poll::Ready(None);
        }
@@ -193,74 +190,80 @@ where
        }
    }

-    fn poll_write(&mut self, cx: &mut Context<'_>) -> Result<bool, Error> {
+    /// Process client requests and write them to the postgres connection, flushing if necessary.
+    /// client -> postgres
+    fn poll_write(&mut self, cx: &mut Context<'_>) -> Poll<Result<WriteReady, Error>> {
        loop {
-            if self.state == State::Closing {
-                trace!("poll_write: done");
-                return Ok(false);
-            }
-
            if Pin::new(&mut self.stream)
                .poll_ready(cx)
                .map_err(Error::io)?
                .is_pending()
            {
                trace!("poll_write: waiting on socket");
-                return Ok(false);
+
+                // poll_ready is self-flushing.
+                return Poll::Pending;
            }

-            let request = match self.poll_request(cx) {
-                Poll::Ready(Some(request)) => request,
-                Poll::Ready(None) if self.responses.is_empty() && self.state == State::Active => {
+            match self.poll_request(cx) {
+                // send the message to postgres
+                Poll::Ready(Some(RequestMessages::Single(request))) => {
+                    Pin::new(&mut self.stream)
+                        .start_send(request)
+                        .map_err(Error::io)?;
+                }
+                // No more messages from the client, and no more responses to wait for.
+                // Send a terminate message to postgres
+                Poll::Ready(None) if self.responses.is_empty() => {
                    trace!("poll_write: at eof, terminating");
-                    self.state = State::Terminating;
                    let mut request = BytesMut::new();
                    frontend::terminate(&mut request);
-                    RequestMessages::Single(FrontendMessage::Raw(request.freeze()))
+                    let request = FrontendMessage::Raw(request.freeze());
+
+                    Pin::new(&mut self.stream)
+                        .start_send(request)
+                        .map_err(Error::io)?;
+
+                    trace!("poll_write: sent eof, closing");
+                    trace!("poll_write: done");
+                    return Poll::Ready(Ok(WriteReady::Terminating));
                }
+                // No more messages from the client, but there are still some responses to wait for.
                Poll::Ready(None) => {
                    trace!(
                        "poll_write: at eof, pending responses {}",
                        self.responses.len()
                    );
-                    return Ok(true);
+                    ready!(self.poll_flush(cx))?;
+                    return Poll::Ready(Ok(WriteReady::WaitingOnRead));
                }
+                // Still waiting for a message from the client.
                Poll::Pending => {
                    trace!("poll_write: waiting on request");
-                    return Ok(true);
-                }
-            };
-
-            match request {
-                RequestMessages::Single(request) => {
-                    Pin::new(&mut self.stream)
-                        .start_send(request)
-                        .map_err(Error::io)?;
-                    if self.state == State::Terminating {
-                        trace!("poll_write: sent eof, closing");
-                        self.state = State::Closing;
-                    }
+                    ready!(self.poll_flush(cx))?;
+                    return Poll::Pending;
                }
            }
        }
    }

-    fn poll_flush(&mut self, cx: &mut Context<'_>) -> Result<(), Error> {
+    fn poll_flush(&mut self, cx: &mut Context<'_>) -> Poll<Result<(), Error>> {
        match Pin::new(&mut self.stream)
            .poll_flush(cx)
            .map_err(Error::io)?
        {
-            Poll::Ready(()) => trace!("poll_flush: flushed"),
-            Poll::Pending => trace!("poll_flush: waiting on socket"),
+            Poll::Ready(()) => {
+                trace!("poll_flush: flushed");
+                Poll::Ready(Ok(()))
+            }
+            Poll::Pending => {
+                trace!("poll_flush: waiting on socket");
+                Poll::Pending
+            }
        }
-        Ok(())
    }

    fn poll_shutdown(&mut self, cx: &mut Context<'_>) -> Poll<Result<(), Error>> {
-        if self.state != State::Closing {
-            return Poll::Pending;
-        }
-
        match Pin::new(&mut self.stream)
            .poll_close(cx)
            .map_err(Error::io)?
@@ -289,18 +292,30 @@ where
        &mut self,
        cx: &mut Context<'_>,
    ) -> Poll<Option<Result<AsyncMessage, Error>>> {
-        let message = self.poll_read(cx)?;
-        let want_flush = self.poll_write(cx)?;
-        if want_flush {
-            self.poll_flush(cx)?;
+        if self.state != State::Closing {
+            // if the state is still active, try read from and write to postgres.
+            let message = self.poll_read(cx)?;
+            let closing = self.poll_write(cx)?;
+            if let Poll::Ready(WriteReady::Terminating) = closing {
+                self.state = State::Closing;
+            }
+
+            if let Poll::Ready(message) = message {
+                return Poll::Ready(Some(Ok(message)));
+            }
+
+            // poll_read returned Pending.
+            // poll_write returned Pending or Ready(WriteReady::WaitingOnRead).
+            // if poll_write returned Ready(WriteReady::WaitingOnRead), then we are waiting to read more data from postgres.
+            if self.state != State::Closing {
+                return Poll::Pending;
+            }
        }
-        match message {
-            Some(message) => Poll::Ready(Some(Ok(message))),
-            None => match self.poll_shutdown(cx) {
-                Poll::Ready(Ok(())) => Poll::Ready(None),
-                Poll::Ready(Err(e)) => Poll::Ready(Some(Err(e))),
-                Poll::Pending => Poll::Pending,
-            },
+
+        match self.poll_shutdown(cx) {
+            Poll::Ready(Ok(())) => Poll::Ready(None),
+            Poll::Ready(Err(e)) => Poll::Ready(Some(Err(e))),
+            Poll::Pending => Poll::Pending,
        }
    }
 }
--- a/libs/proxy/tokio-postgres2/src/lib.rs
+++ b/libs/proxy/tokio-postgres2/src/lib.rs
@@ -1,5 +1,5 @@
 //! An asynchronous, pipelined, PostgreSQL client.
-#![warn(rust_2018_idioms, clippy::all)]
+#![warn(clippy::all)]

 pub use crate::cancel_token::CancelToken;
 pub use crate::client::{Client, SocketConfig};
--- a/libs/proxy/tokio-postgres2/src/to_statement.rs
+++ b/libs/proxy/tokio-postgres2/src/to_statement.rs
@@ -11,7 +11,7 @@ mod private {
        Query(&'a str),
    }

-    impl<'a> ToStatementType<'a> {
+    impl ToStatementType<'_> {
        pub async fn into_statement(self, client: &Client) -> Result<Statement, Error> {
            match self {
                ToStatementType::Statement(s) => Ok(s.clone()),
--- a/libs/remote_storage/Cargo.toml
+++ b/libs/remote_storage/Cargo.toml
@@ -18,6 +18,7 @@ camino = { workspace = true, features = ["serde1"] }
 humantime-serde.workspace = true
 hyper = { workspace = true, features = ["client"] }
 futures.workspace = true
+reqwest.workspace = true
 serde.workspace = true
 serde_json.workspace = true
 tokio = { workspace = true, features = ["sync", "fs", "io-util"] }
--- a/libs/remote_storage/src/azure_blob.rs
+++ b/libs/remote_storage/src/azure_blob.rs
@@ -13,10 +13,12 @@ use std::time::Duration;
 use std::time::SystemTime;

 use super::REMOTE_STORAGE_PREFIX_SEPARATOR;
+use anyhow::Context;
 use anyhow::Result;
 use azure_core::request_options::{IfMatchCondition, MaxResults, Metadata, Range};
+use azure_core::HttpClient;
+use azure_core::TransportOptions;
 use azure_core::{Continuable, RetryOptions};
-use azure_identity::DefaultAzureCredential;
 use azure_storage::StorageCredentials;
 use azure_storage_blobs::blob::CopyStatus;
 use azure_storage_blobs::prelude::ClientBuilder;
@@ -76,12 +78,18 @@ impl AzureBlobStorage {
        let credentials = if let Ok(access_key) = env::var("AZURE_STORAGE_ACCESS_KEY") {
            StorageCredentials::access_key(account.clone(), access_key)
        } else {
-            let token_credential = DefaultAzureCredential::default();
-            StorageCredentials::token_credential(Arc::new(token_credential))
+            let token_credential = azure_identity::create_default_credential()
+                .context("trying to obtain Azure default credentials")?;
+            StorageCredentials::token_credential(token_credential)
        };

-        // we have an outer retry
-        let builder = ClientBuilder::new(account, credentials).retry(RetryOptions::none());
+        let builder = ClientBuilder::new(account, credentials)
+            // we have an outer retry
+            .retry(RetryOptions::none())
+            // Customize transport to configure conneciton pooling
+            .transport(TransportOptions::new(Self::reqwest_client(
+                azure_config.conn_pool_size,
+            )));

        let client = builder.container_client(azure_config.container_name.to_owned());

@@ -106,6 +114,14 @@ impl AzureBlobStorage {
        })
    }

+    fn reqwest_client(conn_pool_size: usize) -> Arc<dyn HttpClient> {
+        let client = reqwest::ClientBuilder::new()
+            .pool_max_idle_per_host(conn_pool_size)
+            .build()
+            .expect("failed to build `reqwest` client");
+        Arc::new(client)
+    }
+
    pub fn relative_path_to_name(&self, path: &RemotePath) -> String {
        assert_eq!(std::path::MAIN_SEPARATOR, REMOTE_STORAGE_PREFIX_SEPARATOR);
        let path_string = path.get_path().as_str();
@@ -544,9 +560,9 @@ impl RemoteStorage for AzureBlobStorage {
            .await
    }

-    async fn delete_objects<'a>(
+    async fn delete_objects(
        &self,
-        paths: &'a [RemotePath],
+        paths: &[RemotePath],
        cancel: &CancellationToken,
    ) -> anyhow::Result<()> {
        let kind = RequestKind::Delete;
@@ -624,6 +640,10 @@ impl RemoteStorage for AzureBlobStorage {
        res
    }

+    fn max_keys_per_delete(&self) -> usize {
+        super::MAX_KEYS_PER_DELETE_AZURE
+    }
+
    async fn copy(
        &self,
        from: &RemotePath,
--- a/libs/remote_storage/src/config.rs
+++ b/libs/remote_storage/src/config.rs
@@ -114,6 +114,16 @@ fn default_max_keys_per_list_response() -> Option<i32> {
    DEFAULT_MAX_KEYS_PER_LIST_RESPONSE
 }

+fn default_azure_conn_pool_size() -> usize {
+    // Conservative default: no connection pooling.  At time of writing this is the Azure
+    // SDK's default as well, due to historic reports of hard-to-reproduce issues
+    // (https://github.com/hyperium/hyper/issues/2312)
+    //
+    // However, using connection pooling is important to avoid exhausting client ports when
+    // doing huge numbers of requests (https://github.com/neondatabase/cloud/issues/20971)
+    0
+}
+
 impl Debug for S3Config {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        f.debug_struct("S3Config")
@@ -146,6 +156,8 @@ pub struct AzureConfig {
    pub concurrency_limit: NonZeroUsize,
    #[serde(default = "default_max_keys_per_list_response")]
    pub max_keys_per_list_response: Option<i32>,
+    #[serde(default = "default_azure_conn_pool_size")]
+    pub conn_pool_size: usize,
 }

 fn default_remote_storage_azure_concurrency_limit() -> NonZeroUsize {
@@ -302,6 +314,7 @@ timeout = '5s'";
    container_region = 'westeurope'
    upload_storage_class = 'INTELLIGENT_TIERING'
    timeout = '7s'
+    conn_pool_size = 8
    ";

        let config = parse(toml).unwrap();
@@ -316,6 +329,7 @@ timeout = '5s'";
                    prefix_in_container: None,
                    concurrency_limit: default_remote_storage_azure_concurrency_limit(),
                    max_keys_per_list_response: DEFAULT_MAX_KEYS_PER_LIST_RESPONSE,
+                    conn_pool_size: 8,
                }),
                timeout: Duration::from_secs(7),
                small_timeout: RemoteStorageConfig::DEFAULT_SMALL_TIMEOUT
--- a/libs/remote_storage/src/lib.rs
+++ b/libs/remote_storage/src/lib.rs
@@ -70,7 +70,14 @@ pub const DEFAULT_REMOTE_STORAGE_AZURE_CONCURRENCY_LIMIT: usize = 100;
 pub const DEFAULT_MAX_KEYS_PER_LIST_RESPONSE: Option<i32> = None;

 /// As defined in S3 docs
-pub const MAX_KEYS_PER_DELETE: usize = 1000;
+///
+/// <https://docs.aws.amazon.com/AmazonS3/latest/API/API_DeleteObjects.html>
+pub const MAX_KEYS_PER_DELETE_S3: usize = 1000;
+
+/// As defined in Azure docs
+///
+/// <https://learn.microsoft.com/en-us/rest/api/storageservices/blob-batch>
+pub const MAX_KEYS_PER_DELETE_AZURE: usize = 256;

 const REMOTE_STORAGE_PREFIX_SEPARATOR: char = '/';

@@ -334,12 +341,20 @@ pub trait RemoteStorage: Send + Sync + 'static {
    /// If the operation fails because of timeout or cancellation, the root cause of the error will be
    /// set to `TimeoutOrCancel`. In such situation it is unknown which deletions, if any, went
    /// through.
-    async fn delete_objects<'a>(
+    async fn delete_objects(
        &self,
-        paths: &'a [RemotePath],
+        paths: &[RemotePath],
        cancel: &CancellationToken,
    ) -> anyhow::Result<()>;

+    /// Returns the maximum number of keys that a call to [`Self::delete_objects`] can delete without chunking
+    ///
+    /// The value returned is only an optimization hint, One can pass larger number of objects to
+    /// `delete_objects` as well.
+    ///
+    /// The value is guaranteed to be >= 1.
+    fn max_keys_per_delete(&self) -> usize;
+
    /// Deletes all objects matching the given prefix.
    ///
    /// NB: this uses NoDelimiter and will match partial prefixes. For example, the prefix /a/b will
@@ -533,6 +548,16 @@ impl<Other: RemoteStorage> GenericRemoteStorage<Arc<Other>> {
        }
    }

+    /// [`RemoteStorage::max_keys_per_delete`]
+    pub fn max_keys_per_delete(&self) -> usize {
+        match self {
+            Self::LocalFs(s) => s.max_keys_per_delete(),
+            Self::AwsS3(s) => s.max_keys_per_delete(),
+            Self::AzureBlob(s) => s.max_keys_per_delete(),
+            Self::Unreliable(s) => s.max_keys_per_delete(),
+        }
+    }
+
    /// See [`RemoteStorage::delete_prefix`]
    pub async fn delete_prefix(
        &self,
--- a/libs/remote_storage/src/local_fs.rs
+++ b/libs/remote_storage/src/local_fs.rs
@@ -562,9 +562,9 @@ impl RemoteStorage for LocalFs {
        }
    }

-    async fn delete_objects<'a>(
+    async fn delete_objects(
        &self,
-        paths: &'a [RemotePath],
+        paths: &[RemotePath],
        cancel: &CancellationToken,
    ) -> anyhow::Result<()> {
        for path in paths {
@@ -573,6 +573,10 @@ impl RemoteStorage for LocalFs {
        Ok(())
    }

+    fn max_keys_per_delete(&self) -> usize {
+        super::MAX_KEYS_PER_DELETE_S3
+    }
+
    async fn copy(
        &self,
        from: &RemotePath,
--- a/libs/remote_storage/src/s3_bucket.rs
+++ b/libs/remote_storage/src/s3_bucket.rs
@@ -48,7 +48,7 @@ use crate::{
    metrics::{start_counting_cancelled_wait, start_measuring_requests},
    support::PermitCarrying,
    ConcurrencyLimiter, Download, DownloadError, DownloadOpts, Listing, ListingMode, ListingObject,
-    RemotePath, RemoteStorage, TimeTravelError, TimeoutOrCancel, MAX_KEYS_PER_DELETE,
+    RemotePath, RemoteStorage, TimeTravelError, TimeoutOrCancel, MAX_KEYS_PER_DELETE_S3,
    REMOTE_STORAGE_PREFIX_SEPARATOR,
 };

@@ -355,7 +355,7 @@ impl S3Bucket {
        let kind = RequestKind::Delete;
        let mut cancel = std::pin::pin!(cancel.cancelled());

-        for chunk in delete_objects.chunks(MAX_KEYS_PER_DELETE) {
+        for chunk in delete_objects.chunks(MAX_KEYS_PER_DELETE_S3) {
            let started_at = start_measuring_requests(kind);

            let req = self
@@ -813,9 +813,9 @@ impl RemoteStorage for S3Bucket {
        .await
    }

-    async fn delete_objects<'a>(
+    async fn delete_objects(
        &self,
-        paths: &'a [RemotePath],
+        paths: &[RemotePath],
        cancel: &CancellationToken,
    ) -> anyhow::Result<()> {
        let kind = RequestKind::Delete;
@@ -832,6 +832,10 @@ impl RemoteStorage for S3Bucket {
        self.delete_oids(&permit, &delete_objects, cancel).await
    }

+    fn max_keys_per_delete(&self) -> usize {
+        MAX_KEYS_PER_DELETE_S3
+    }
+
    async fn delete(&self, path: &RemotePath, cancel: &CancellationToken) -> anyhow::Result<()> {
        let paths = std::array::from_ref(path);
        self.delete_objects(paths, cancel).await
--- a/libs/remote_storage/src/simulate_failures.rs
+++ b/libs/remote_storage/src/simulate_failures.rs
@@ -181,9 +181,9 @@ impl RemoteStorage for UnreliableWrapper {
        self.delete_inner(path, true, cancel).await
    }

-    async fn delete_objects<'a>(
+    async fn delete_objects(
        &self,
-        paths: &'a [RemotePath],
+        paths: &[RemotePath],
        cancel: &CancellationToken,
    ) -> anyhow::Result<()> {
        self.attempt(RemoteOp::DeleteObjects(paths.to_vec()))?;
@@ -203,6 +203,10 @@ impl RemoteStorage for UnreliableWrapper {
        Ok(())
    }

+    fn max_keys_per_delete(&self) -> usize {
+        self.inner.max_keys_per_delete()
+    }
+
    async fn copy(
        &self,
        from: &RemotePath,
--- a/libs/remote_storage/tests/test_real_azure.rs
+++ b/libs/remote_storage/tests/test_real_azure.rs
@@ -218,6 +218,7 @@ async fn create_azure_client(
            prefix_in_container: Some(format!("test_{millis}_{random:08x}/")),
            concurrency_limit: NonZeroUsize::new(100).unwrap(),
            max_keys_per_list_response,
+            conn_pool_size: 8,
        }),
        timeout: RemoteStorageConfig::DEFAULT_TIMEOUT,
        small_timeout: RemoteStorageConfig::DEFAULT_SMALL_TIMEOUT,
--- a/libs/safekeeper_api/Cargo.toml
+++ b/libs/safekeeper_api/Cargo.toml
@@ -5,6 +5,9 @@ edition.workspace = true
 license.workspace = true

 [dependencies]
-serde.workspace = true
 const_format.workspace = true
+serde.workspace = true
+postgres_ffi.workspace = true
+pq_proto.workspace = true
+tokio.workspace = true
 utils.workspace = true
--- a/libs/safekeeper_api/src/lib.rs
+++ b/libs/safekeeper_api/src/lib.rs
@@ -1,10 +1,27 @@
 #![deny(unsafe_code)]
 #![deny(clippy::undocumented_unsafe_blocks)]
 use const_format::formatcp;
+use pq_proto::SystemId;
+use serde::{Deserialize, Serialize};

 /// Public API types
 pub mod models;

+/// Consensus logical timestamp. Note: it is a part of sk control file.
+pub type Term = u64;
+pub const INVALID_TERM: Term = 0;
+
+/// Information about Postgres. Safekeeper gets it once and then verifies all
+/// further connections from computes match. Note: it is a part of sk control
+/// file.
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
+pub struct ServerInfo {
+    /// Postgres server version
+    pub pg_version: u32,
+    pub system_id: SystemId,
+    pub wal_seg_size: u32,
+}
+
 pub const DEFAULT_PG_LISTEN_PORT: u16 = 5454;
 pub const DEFAULT_PG_LISTEN_ADDR: &str = formatcp!("127.0.0.1:{DEFAULT_PG_LISTEN_PORT}");

--- a/libs/safekeeper_api/src/models.rs
+++ b/libs/safekeeper_api/src/models.rs
@@ -1,10 +1,23 @@
+//! Types used in safekeeper http API. Many of them are also reused internally.
+
+use postgres_ffi::TimestampTz;
 use serde::{Deserialize, Serialize};
+use std::net::SocketAddr;
+use tokio::time::Instant;

 use utils::{
-    id::{NodeId, TenantId, TimelineId},
+    id::{NodeId, TenantId, TenantTimelineId, TimelineId},
    lsn::Lsn,
+    pageserver_feedback::PageserverFeedback,
 };

+use crate::{ServerInfo, Term};
+
+#[derive(Debug, Serialize)]
+pub struct SafekeeperStatus {
+    pub id: NodeId,
+}
+
 #[derive(Serialize, Deserialize)]
 pub struct TimelineCreateRequest {
    pub tenant_id: TenantId,
@@ -18,6 +31,161 @@ pub struct TimelineCreateRequest {
    pub local_start_lsn: Option<Lsn>,
 }

+/// Same as TermLsn, but serializes LSN using display serializer
+/// in Postgres format, i.e. 0/FFFFFFFF. Used only for the API response.
+#[derive(Debug, Clone, Copy, Serialize, Deserialize)]
+pub struct TermSwitchApiEntry {
+    pub term: Term,
+    pub lsn: Lsn,
+}
+
+/// Augment AcceptorState with last_log_term for convenience
+#[derive(Debug, Serialize, Deserialize)]
+pub struct AcceptorStateStatus {
+    pub term: Term,
+    pub epoch: Term, // aka last_log_term, old `epoch` name is left for compatibility
+    pub term_history: Vec<TermSwitchApiEntry>,
+}
+
+/// Things safekeeper should know about timeline state on peers.
+/// Used as both model and internally.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct PeerInfo {
+    pub sk_id: NodeId,
+    pub term: Term,
+    /// Term of the last entry.
+    pub last_log_term: Term,
+    /// LSN of the last record.
+    pub flush_lsn: Lsn,
+    pub commit_lsn: Lsn,
+    /// Since which LSN safekeeper has WAL.
+    pub local_start_lsn: Lsn,
+    /// When info was received. Serde annotations are not very useful but make
+    /// the code compile -- we don't rely on this field externally.
+    #[serde(skip)]
+    #[serde(default = "Instant::now")]
+    pub ts: Instant,
+    pub pg_connstr: String,
+    pub http_connstr: String,
+}
+
+pub type FullTransactionId = u64;
+
+/// Hot standby feedback received from replica
+#[derive(Debug, Clone, Copy, Serialize, Deserialize)]
+pub struct HotStandbyFeedback {
+    pub ts: TimestampTz,
+    pub xmin: FullTransactionId,
+    pub catalog_xmin: FullTransactionId,
+}
+
+pub const INVALID_FULL_TRANSACTION_ID: FullTransactionId = 0;
+
+impl HotStandbyFeedback {
+    pub fn empty() -> HotStandbyFeedback {
+        HotStandbyFeedback {
+            ts: 0,
+            xmin: 0,
+            catalog_xmin: 0,
+        }
+    }
+}
+
+/// Standby status update
+#[derive(Debug, Clone, Copy, Serialize, Deserialize)]
+pub struct StandbyReply {
+    pub write_lsn: Lsn, // The location of the last WAL byte + 1 received and written to disk in the standby.
+    pub flush_lsn: Lsn, // The location of the last WAL byte + 1 flushed to disk in the standby.
+    pub apply_lsn: Lsn, // The location of the last WAL byte + 1 applied in the standby.
+    pub reply_ts: TimestampTz, // The client's system clock at the time of transmission, as microseconds since midnight on 2000-01-01.
+    pub reply_requested: bool,
+}
+
+impl StandbyReply {
+    pub fn empty() -> Self {
+        StandbyReply {
+            write_lsn: Lsn::INVALID,
+            flush_lsn: Lsn::INVALID,
+            apply_lsn: Lsn::INVALID,
+            reply_ts: 0,
+            reply_requested: false,
+        }
+    }
+}
+
+#[derive(Debug, Clone, Copy, Serialize, Deserialize)]
+pub struct StandbyFeedback {
+    pub reply: StandbyReply,
+    pub hs_feedback: HotStandbyFeedback,
+}
+
+impl StandbyFeedback {
+    pub fn empty() -> Self {
+        StandbyFeedback {
+            reply: StandbyReply::empty(),
+            hs_feedback: HotStandbyFeedback::empty(),
+        }
+    }
+}
+
+/// Receiver is either pageserver or regular standby, which have different
+/// feedbacks.
+/// Used as both model and internally.
+#[derive(Debug, Clone, Copy, Serialize, Deserialize)]
+pub enum ReplicationFeedback {
+    Pageserver(PageserverFeedback),
+    Standby(StandbyFeedback),
+}
+
+/// Uniquely identifies a WAL service connection. Logged in spans for
+/// observability.
+pub type ConnectionId = u32;
+
+/// Serialize is used only for json'ing in API response. Also used internally.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct WalSenderState {
+    pub ttid: TenantTimelineId,
+    pub addr: SocketAddr,
+    pub conn_id: ConnectionId,
+    // postgres application_name
+    pub appname: Option<String>,
+    pub feedback: ReplicationFeedback,
+}
+
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct WalReceiverState {
+    /// None means it is recovery initiated by us (this safekeeper).
+    pub conn_id: Option<ConnectionId>,
+    pub status: WalReceiverStatus,
+}
+
+/// Walreceiver status. Currently only whether it passed voting stage and
+/// started receiving the stream, but it is easy to add more if needed.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub enum WalReceiverStatus {
+    Voting,
+    Streaming,
+}
+
+/// Info about timeline on safekeeper ready for reporting.
+#[derive(Debug, Serialize, Deserialize)]
+pub struct TimelineStatus {
+    pub tenant_id: TenantId,
+    pub timeline_id: TimelineId,
+    pub acceptor_state: AcceptorStateStatus,
+    pub pg_info: ServerInfo,
+    pub flush_lsn: Lsn,
+    pub timeline_start_lsn: Lsn,
+    pub local_start_lsn: Lsn,
+    pub commit_lsn: Lsn,
+    pub backup_lsn: Lsn,
+    pub peer_horizon_lsn: Lsn,
+    pub remote_consistent_lsn: Lsn,
+    pub peers: Vec<PeerInfo>,
+    pub walsenders: Vec<WalSenderState>,
+    pub walreceivers: Vec<WalReceiverState>,
+}
+
 fn lsn_invalid() -> Lsn {
    Lsn::INVALID
 }
--- a/libs/utils/Cargo.toml
+++ b/libs/utils/Cargo.toml
@@ -15,17 +15,20 @@ arc-swap.workspace = true
 sentry.workspace = true
 async-compression.workspace = true
 anyhow.workspace = true
+backtrace.workspace = true
 bincode.workspace = true
 bytes.workspace = true
 camino.workspace = true
 chrono.workspace = true
 diatomic-waker.workspace = true
+flate2.workspace = true
 git-version.workspace = true
 hex = { workspace = true, features = ["serde"] }
 humantime.workspace = true
 hyper0 = { workspace = true, features = ["full"] }
+itertools.workspace = true
 fail.workspace = true
-futures = { workspace = true}
+futures = { workspace = true }
 jemalloc_pprof.workspace = true
 jsonwebtoken.workspace = true
 nix.workspace = true
--- a/libs/utils/src/http/endpoint.rs
+++ b/libs/utils/src/http/endpoint.rs
@@ -1,15 +1,22 @@
 use crate::auth::{AuthError, Claims, SwappableJwtAuth};
 use crate::http::error::{api_error_handler, route_error_handler, ApiError};
 use crate::http::request::{get_query_param, parse_query_param};
+use crate::pprof;
+use ::pprof::protos::Message as _;
+use ::pprof::ProfilerGuardBuilder;
 use anyhow::{anyhow, Context};
+use bytes::{Bytes, BytesMut};
 use hyper::header::{HeaderName, AUTHORIZATION, CONTENT_DISPOSITION};
 use hyper::http::HeaderValue;
 use hyper::Method;
 use hyper::{header::CONTENT_TYPE, Body, Request, Response};
 use metrics::{register_int_counter, Encoder, IntCounter, TextEncoder};
 use once_cell::sync::Lazy;
+use regex::Regex;
 use routerify::ext::RequestExt;
 use routerify::{Middleware, RequestInfo, Router, RouterBuilder};
+use tokio::sync::{mpsc, Mutex};
+use tokio_stream::wrappers::ReceiverStream;
 use tokio_util::io::ReaderStream;
 use tracing::{debug, info, info_span, warn, Instrument};

@@ -18,11 +25,6 @@ use std::io::Write as _;
 use std::str::FromStr;
 use std::time::Duration;

-use bytes::{Bytes, BytesMut};
-use pprof::protos::Message as _;
-use tokio::sync::{mpsc, Mutex};
-use tokio_stream::wrappers::ReceiverStream;
-
 static SERVE_METRICS_COUNT: Lazy<IntCounter> = Lazy::new(|| {
    register_int_counter!(
        "libmetrics_metric_handler_requests_total",
@@ -365,7 +367,7 @@ pub async fn profile_cpu_handler(req: Request<Body>) -> Result<Response<Body>, A

    // Take the profile.
    let report = tokio::task::spawn_blocking(move || {
-        let guard = pprof::ProfilerGuardBuilder::default()
+        let guard = ProfilerGuardBuilder::default()
            .frequency(frequency_hz)
            .blocklist(&["libc", "libgcc", "pthread", "vdso"])
            .build()?;
@@ -457,10 +459,34 @@ pub async fn profile_heap_handler(req: Request<Body>) -> Result<Response<Body>,
        }

        Format::Pprof => {
-            let data = tokio::task::spawn_blocking(move || prof_ctl.dump_pprof())
-                .await
-                .map_err(|join_err| ApiError::InternalServerError(join_err.into()))?
-                .map_err(ApiError::InternalServerError)?;
+            let data = tokio::task::spawn_blocking(move || {
+                let bytes = prof_ctl.dump_pprof()?;
+                // Symbolize the profile.
+                // TODO: consider moving this upstream to jemalloc_pprof and avoiding the
+                // serialization roundtrip.
+                static STRIP_FUNCTIONS: Lazy<Vec<(Regex, bool)>> = Lazy::new(|| {
+                    // Functions to strip from profiles. If true, also remove child frames.
+                    vec![
+                        (Regex::new("^__rust").unwrap(), false),
+                        (Regex::new("^_start$").unwrap(), false),
+                        (Regex::new("^irallocx_prof").unwrap(), true),
+                        (Regex::new("^prof_alloc_prep").unwrap(), true),
+                        (Regex::new("^std::rt::lang_start").unwrap(), false),
+                        (Regex::new("^std::sys::backtrace::__rust").unwrap(), false),
+                    ]
+                });
+                let profile = pprof::decode(&bytes)?;
+                let profile = pprof::symbolize(profile)?;
+                let profile = pprof::strip_locations(
+                    profile,
+                    &["libc", "libgcc", "pthread", "vdso"],
+                    &STRIP_FUNCTIONS,
+                );
+                pprof::encode(&profile)
+            })
+            .await
+            .map_err(|join_err| ApiError::InternalServerError(join_err.into()))?
+            .map_err(ApiError::InternalServerError)?;
            Response::builder()
                .status(200)
                .header(CONTENT_TYPE, "application/octet-stream")
--- a/libs/utils/src/lib.rs
+++ b/libs/utils/src/lib.rs
@@ -94,6 +94,10 @@ pub mod toml_edit_ext;

 pub mod circuit_breaker;

+pub mod try_rcu;
+
+pub mod pprof;
+
 // Re-export used in macro. Avoids adding git-version as dep in target crates.
 #[doc(hidden)]
 pub use git_version;
--- a/libs/utils/src/pprof.rs
+++ b/libs/utils/src/pprof.rs
@@ -0,0 +1,190 @@
+use flate2::write::{GzDecoder, GzEncoder};
+use flate2::Compression;
+use itertools::Itertools as _;
+use once_cell::sync::Lazy;
+use pprof::protos::{Function, Line, Message as _, Profile};
+use regex::Regex;
+
+use std::borrow::Cow;
+use std::collections::{HashMap, HashSet};
+use std::ffi::c_void;
+use std::io::Write as _;
+
+/// Decodes a gzip-compressed Protobuf-encoded pprof profile.
+pub fn decode(bytes: &[u8]) -> anyhow::Result<Profile> {
+    let mut gz = GzDecoder::new(Vec::new());
+    gz.write_all(bytes)?;
+    Ok(Profile::parse_from_bytes(&gz.finish()?)?)
+}
+
+/// Encodes a pprof profile as gzip-compressed Protobuf.
+pub fn encode(profile: &Profile) -> anyhow::Result<Vec<u8>> {
+    let mut gz = GzEncoder::new(Vec::new(), Compression::default());
+    profile.write_to_writer(&mut gz)?;
+    Ok(gz.finish()?)
+}
+
+/// Symbolizes a pprof profile using the current binary.
+pub fn symbolize(mut profile: Profile) -> anyhow::Result<Profile> {
+    if !profile.function.is_empty() {
+        return Ok(profile); // already symbolized
+    }
+
+    // Collect function names.
+    let mut functions: HashMap<String, Function> = HashMap::new();
+    let mut strings: HashMap<String, i64> = profile
+        .string_table
+        .into_iter()
+        .enumerate()
+        .map(|(i, s)| (s, i as i64))
+        .collect();
+
+    // Helper to look up or register a string.
+    let mut string_id = |s: &str| -> i64 {
+        // Don't use .entry() to avoid unnecessary allocations.
+        if let Some(id) = strings.get(s) {
+            return *id;
+        }
+        let id = strings.len() as i64;
+        strings.insert(s.to_string(), id);
+        id
+    };
+
+    for loc in &mut profile.location {
+        if !loc.line.is_empty() {
+            continue;
+        }
+
+        // Resolve the line and function for each location.
+        backtrace::resolve(loc.address as *mut c_void, |symbol| {
+            let Some(symname) = symbol.name() else {
+                return;
+            };
+            let mut name = symname.to_string();
+
+            // Strip the Rust monomorphization suffix from the symbol name.
+            static SUFFIX_REGEX: Lazy<Regex> =
+                Lazy::new(|| Regex::new("::h[0-9a-f]{16}$").expect("invalid regex"));
+            if let Some(m) = SUFFIX_REGEX.find(&name) {
+                name.truncate(m.start());
+            }
+
+            let function_id = match functions.get(&name) {
+                Some(function) => function.id,
+                None => {
+                    let id = functions.len() as u64 + 1;
+                    let system_name = String::from_utf8_lossy(symname.as_bytes());
+                    let filename = symbol
+                        .filename()
+                        .map(|path| path.to_string_lossy())
+                        .unwrap_or(Cow::Borrowed(""));
+                    let function = Function {
+                        id,
+                        name: string_id(&name),
+                        system_name: string_id(&system_name),
+                        filename: string_id(&filename),
+                        ..Default::default()
+                    };
+                    functions.insert(name, function);
+                    id
+                }
+            };
+            loc.line.push(Line {
+                function_id,
+                line: symbol.lineno().unwrap_or(0) as i64,
+                ..Default::default()
+            });
+        });
+    }
+
+    // Store the resolved functions, and mark the mapping as resolved.
+    profile.function = functions.into_values().sorted_by_key(|f| f.id).collect();
+    profile.string_table = strings
+        .into_iter()
+        .sorted_by_key(|(_, i)| *i)
+        .map(|(s, _)| s)
+        .collect();
+
+    for mapping in &mut profile.mapping {
+        mapping.has_functions = true;
+        mapping.has_filenames = true;
+    }
+
+    Ok(profile)
+}
+
+/// Strips locations (stack frames) matching the given mappings (substring) or function names
+/// (regex). The function bool specifies whether child frames should be stripped as well.
+///
+/// The string definitions are left behind in the profile for simplicity, to avoid rewriting all
+/// string references.
+pub fn strip_locations(
+    mut profile: Profile,
+    mappings: &[&str],
+    functions: &[(Regex, bool)],
+) -> Profile {
+    // Strip mappings.
+    let mut strip_mappings: HashSet<u64> = HashSet::new();
+
+    profile.mapping.retain(|mapping| {
+        let Some(name) = profile.string_table.get(mapping.filename as usize) else {
+            return true;
+        };
+        if mappings.iter().any(|substr| name.contains(substr)) {
+            strip_mappings.insert(mapping.id);
+            return false;
+        }
+        true
+    });
+
+    // Strip functions.
+    let mut strip_functions: HashMap<u64, bool> = HashMap::new();
+
+    profile.function.retain(|function| {
+        let Some(name) = profile.string_table.get(function.name as usize) else {
+            return true;
+        };
+        for (regex, strip_children) in functions {
+            if regex.is_match(name) {
+                strip_functions.insert(function.id, *strip_children);
+                return false;
+            }
+        }
+        true
+    });
+
+    // Strip locations. The bool specifies whether child frames should be stripped too.
+    let mut strip_locations: HashMap<u64, bool> = HashMap::new();
+
+    profile.location.retain(|location| {
+        for line in &location.line {
+            if let Some(strip_children) = strip_functions.get(&line.function_id) {
+                strip_locations.insert(location.id, *strip_children);
+                return false;
+            }
+        }
+        if strip_mappings.contains(&location.mapping_id) {
+            strip_locations.insert(location.id, false);
+            return false;
+        }
+        true
+    });
+
+    // Strip sample locations.
+    for sample in &mut profile.sample {
+        // First, find the uppermost function with child removal and truncate the stack.
+        if let Some(truncate) = sample
+            .location_id
+            .iter()
+            .rposition(|id| strip_locations.get(id) == Some(&true))
+        {
+            sample.location_id.drain(..=truncate);
+        }
+        // Next, strip any individual frames without child removal.
+        sample
+            .location_id
+            .retain(|id| !strip_locations.contains_key(id));
+    }
+
+    profile
+}
--- a/libs/utils/src/try_rcu.rs
+++ b/libs/utils/src/try_rcu.rs
@@ -0,0 +1,77 @@
+//! Try RCU extension lifted from <https://github.com/vorner/arc-swap/issues/94#issuecomment-1987154023>
+
+pub trait ArcSwapExt<T> {
+    /// [`ArcSwap::rcu`](arc_swap::ArcSwap::rcu), but with Result that short-circuits on error.
+    fn try_rcu<R, F, E>(&self, f: F) -> Result<T, E>
+    where
+        F: FnMut(&T) -> Result<R, E>,
+        R: Into<T>;
+}
+
+impl<T, S> ArcSwapExt<T> for arc_swap::ArcSwapAny<T, S>
+where
+    T: arc_swap::RefCnt,
+    S: arc_swap::strategy::CaS<T>,
+{
+    fn try_rcu<R, F, E>(&self, mut f: F) -> Result<T, E>
+    where
+        F: FnMut(&T) -> Result<R, E>,
+        R: Into<T>,
+    {
+        fn ptr_eq<Base, A, B>(a: A, b: B) -> bool
+        where
+            A: arc_swap::AsRaw<Base>,
+            B: arc_swap::AsRaw<Base>,
+        {
+            let a = a.as_raw();
+            let b = b.as_raw();
+            std::ptr::eq(a, b)
+        }
+
+        let mut cur = self.load();
+        loop {
+            let new = f(&cur)?.into();
+            let prev = self.compare_and_swap(&*cur, new);
+            let swapped = ptr_eq(&*cur, &*prev);
+            if swapped {
+                return Ok(arc_swap::Guard::into_inner(prev));
+            } else {
+                cur = prev;
+            }
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use arc_swap::ArcSwap;
+    use std::sync::Arc;
+
+    #[test]
+    fn test_try_rcu_success() {
+        let swap = ArcSwap::from(Arc::new(42));
+
+        let result = swap.try_rcu(|value| -> Result<_, String> { Ok(**value + 1) });
+
+        assert!(result.is_ok());
+        assert_eq!(**swap.load(), 43);
+    }
+
+    #[test]
+    fn test_try_rcu_error() {
+        let swap = ArcSwap::from(Arc::new(42));
+
+        let result = swap.try_rcu(|value| -> Result<i32, _> {
+            if **value == 42 {
+                Err("err")
+            } else {
+                Ok(**value + 1)
+            }
+        });
+
+        assert!(result.is_err());
+        assert_eq!(result.unwrap_err(), "err");
+        assert_eq!(**swap.load(), 42);
+    }
+}
--- a/libs/wal_decoder/proto/interpreted_wal.proto
+++ b/libs/wal_decoder/proto/interpreted_wal.proto
@@ -37,7 +37,7 @@ message ValueMeta {
 }

 message CompactKey {
-  int64 high = 1;
-  int64 low = 2;
+  uint64 high = 1;
+  uint64 low = 2;
 }

--- a/libs/wal_decoder/src/wire_format.rs
+++ b/libs/wal_decoder/src/wire_format.rs
@@ -236,8 +236,8 @@ impl From<ValueMeta> for proto::ValueMeta {
 impl From<CompactKey> for proto::CompactKey {
    fn from(value: CompactKey) -> Self {
        proto::CompactKey {
-            high: (value.raw() >> 64) as i64,
-            low: value.raw() as i64,
+            high: (value.raw() >> 64) as u64,
+            low: value.raw() as u64,
        }
    }
 }
@@ -354,3 +354,64 @@ impl From<proto::CompactKey> for CompactKey {
        (((value.high as i128) << 64) | (value.low as i128)).into()
    }
 }
+
+#[test]
+fn test_compact_key_with_large_relnode() {
+    use pageserver_api::key::Key;
+
+    let inputs = vec![
+        Key {
+            field1: 0,
+            field2: 0x100,
+            field3: 0x200,
+            field4: 0,
+            field5: 0x10,
+            field6: 0x5,
+        },
+        Key {
+            field1: 0,
+            field2: 0x100,
+            field3: 0x200,
+            field4: 0x007FFFFF,
+            field5: 0x10,
+            field6: 0x5,
+        },
+        Key {
+            field1: 0,
+            field2: 0x100,
+            field3: 0x200,
+            field4: 0x00800000,
+            field5: 0x10,
+            field6: 0x5,
+        },
+        Key {
+            field1: 0,
+            field2: 0x100,
+            field3: 0x200,
+            field4: 0x00800001,
+            field5: 0x10,
+            field6: 0x5,
+        },
+        Key {
+            field1: 0,
+            field2: 0xFFFFFFFF,
+            field3: 0xFFFFFFFF,
+            field4: 0xFFFFFFFF,
+            field5: 0x0,
+            field6: 0x0,
+        },
+    ];
+
+    for input in inputs {
+        assert!(input.is_valid_key_on_write_path());
+        let compact = input.to_compact();
+        let proto: proto::CompactKey = compact.into();
+        let from_proto: CompactKey = proto.into();
+
+        assert_eq!(
+            compact, from_proto,
+            "Round trip failed for key with relnode={:#x}",
+            input.field4
+        );
+    }
+}
--- a/libs/walproposer/build.rs
+++ b/libs/walproposer/build.rs
@@ -30,9 +30,9 @@ fn main() -> anyhow::Result<()> {
    let pgxn_neon = std::fs::canonicalize(pgxn_neon)?;
    let pgxn_neon = pgxn_neon.to_str().ok_or(anyhow!("Bad non-UTF path"))?;

+    println!("cargo:rustc-link-lib=static=walproposer");
    println!("cargo:rustc-link-lib=static=pgport");
    println!("cargo:rustc-link-lib=static=pgcommon");
-    println!("cargo:rustc-link-lib=static=walproposer");
    println!("cargo:rustc-link-search={walproposer_lib_search_str}");

    // Rebuild crate when libwalproposer.a changes
--- a/pageserver/client/src/mgmt_api.rs
+++ b/pageserver/client/src/mgmt_api.rs
@@ -270,12 +270,18 @@ impl Client {
        Ok(body)
    }

-    pub async fn tenant_config(&self, req: &TenantConfigRequest) -> Result<()> {
+    pub async fn set_tenant_config(&self, req: &TenantConfigRequest) -> Result<()> {
        let uri = format!("{}/v1/tenant/config", self.mgmt_api_endpoint);
        self.request(Method::PUT, &uri, req).await?;
        Ok(())
    }

+    pub async fn patch_tenant_config(&self, req: &TenantConfigPatchRequest) -> Result<()> {
+        let uri = format!("{}/v1/tenant/config", self.mgmt_api_endpoint);
+        self.request(Method::PATCH, &uri, req).await?;
+        Ok(())
+    }
+
    pub async fn tenant_secondary_download(
        &self,
        tenant_id: TenantShardId,
--- a/pageserver/compaction/src/compact_tiered.rs
+++ b/pageserver/compaction/src/compact_tiered.rs
@@ -272,7 +272,7 @@ struct CompactionJob<E: CompactionJobExecutor> {
    completed: bool,
 }

-impl<'a, E> LevelCompactionState<'a, E>
+impl<E> LevelCompactionState<'_, E>
 where
    E: CompactionJobExecutor,
 {
--- a/pageserver/compaction/src/identify_levels.rs
+++ b/pageserver/compaction/src/identify_levels.rs
@@ -224,9 +224,8 @@ impl<L> Level<L> {
            }

            // recalculate depth if this was the last event at this point
-            let more_events_at_this_key = events_iter
-                .peek()
-                .map_or(false, |next_e| next_e.key == e.key);
+            let more_events_at_this_key =
+                events_iter.peek().is_some_and(|next_e| next_e.key == e.key);
            if !more_events_at_this_key {
                let mut active_depth = 0;
                for (_end_lsn, is_image, _idx) in active_set.iter().rev() {
--- a/pageserver/compaction/src/interface.rs
+++ b/pageserver/compaction/src/interface.rs
@@ -148,7 +148,7 @@ pub trait CompactionDeltaLayer<E: CompactionJobExecutor + ?Sized>: CompactionLay
        Self: 'a;

    /// Return all keys in this delta layer.
-    fn load_keys<'a>(
+    fn load_keys(
        &self,
        ctx: &E::RequestContext,
    ) -> impl Future<Output = anyhow::Result<Vec<Self::DeltaEntry<'_>>>> + Send;
--- a/pageserver/compaction/src/simulator.rs
+++ b/pageserver/compaction/src/simulator.rs
@@ -143,7 +143,7 @@ impl interface::CompactionLayer<Key> for Arc<MockDeltaLayer> {
 impl interface::CompactionDeltaLayer<MockTimeline> for Arc<MockDeltaLayer> {
    type DeltaEntry<'a> = MockRecord;

-    async fn load_keys<'a>(&self, _ctx: &MockRequestContext) -> anyhow::Result<Vec<MockRecord>> {
+    async fn load_keys(&self, _ctx: &MockRequestContext) -> anyhow::Result<Vec<MockRecord>> {
        Ok(self.records.clone())
    }
 }
--- a/pageserver/pagebench/src/cmd/aux_files.rs
+++ b/pageserver/pagebench/src/cmd/aux_files.rs
@@ -64,7 +64,7 @@ async fn main_impl(args: Args) -> anyhow::Result<()> {
    println!("operating on timeline {}", timeline);

    mgmt_api_client
-        .tenant_config(&TenantConfigRequest {
+        .set_tenant_config(&TenantConfigRequest {
            tenant_id: timeline.tenant_id,
            config: TenantConfig::default(),
        })
--- a/pageserver/src/basebackup.rs
+++ b/pageserver/src/basebackup.rs
@@ -248,7 +248,7 @@ where
    }
 }

-impl<'a, W> Basebackup<'a, W>
+impl<W> Basebackup<'_, W>
 where
    W: AsyncWrite + Send + Sync + Unpin,
 {
--- a/pageserver/src/deletion_queue/deleter.rs
+++ b/pageserver/src/deletion_queue/deleter.rs
@@ -9,7 +9,6 @@
 use remote_storage::GenericRemoteStorage;
 use remote_storage::RemotePath;
 use remote_storage::TimeoutOrCancel;
-use remote_storage::MAX_KEYS_PER_DELETE;
 use std::time::Duration;
 use tokio_util::sync::CancellationToken;
 use tracing::info;
@@ -131,7 +130,8 @@ impl Deleter {
    }

    pub(super) async fn background(&mut self) -> Result<(), DeletionQueueError> {
-        self.accumulator.reserve(MAX_KEYS_PER_DELETE);
+        let max_keys_per_delete = self.remote_storage.max_keys_per_delete();
+        self.accumulator.reserve(max_keys_per_delete);

        loop {
            if self.cancel.is_cancelled() {
@@ -156,14 +156,14 @@ impl Deleter {

            match msg {
                DeleterMessage::Delete(mut list) => {
-                    while !list.is_empty() || self.accumulator.len() == MAX_KEYS_PER_DELETE {
-                        if self.accumulator.len() == MAX_KEYS_PER_DELETE {
+                    while !list.is_empty() || self.accumulator.len() == max_keys_per_delete {
+                        if self.accumulator.len() == max_keys_per_delete {
                            self.flush().await?;
                            // If we have received this number of keys, proceed with attempting to execute
                            assert_eq!(self.accumulator.len(), 0);
                        }

-                        let available_slots = MAX_KEYS_PER_DELETE - self.accumulator.len();
+                        let available_slots = max_keys_per_delete - self.accumulator.len();
                        let take_count = std::cmp::min(available_slots, list.len());
                        for path in list.drain(list.len() - take_count..) {
                            self.accumulator.push(path);
--- a/pageserver/src/http/openapi_spec.yml
+++ b/pageserver/src/http/openapi_spec.yml
@@ -767,7 +767,27 @@ paths:
  /v1/tenant/config:
    put:
      description: |
-        Update tenant's config.
+        Update tenant's config by setting it to the provided value
+
+        Invalid fields in the tenant config will cause the request to be rejected with status 400.
+      requestBody:
+        content:
+          application/json:
+            schema:
+              $ref: "#/components/schemas/TenantConfigRequest"
+      responses:
+        "200":
+          description: OK
+          content:
+            application/json:
+              schema:
+                type: array
+                items:
+                  $ref: "#/components/schemas/TenantInfo"
+    patch:
+      description: |
+        Update tenant's config additively by patching the updated fields provided.
+        Null values unset the field and non-null values upsert it.

        Invalid fields in the tenant config will cause the request to be rejected with status 400.
      requestBody:
--- a/pageserver/src/http/routes.rs
+++ b/pageserver/src/http/routes.rs
@@ -28,6 +28,7 @@ use pageserver_api::models::LsnLease;
 use pageserver_api::models::LsnLeaseRequest;
 use pageserver_api::models::OffloadedTimelineInfo;
 use pageserver_api::models::ShardParameters;
+use pageserver_api::models::TenantConfigPatchRequest;
 use pageserver_api::models::TenantDetails;
 use pageserver_api::models::TenantLocationConfigRequest;
 use pageserver_api::models::TenantLocationConfigResponse;
@@ -96,8 +97,8 @@ use crate::tenant::{LogicalSizeCalculationCause, PageReconstructError};
 use crate::DEFAULT_PG_VERSION;
 use crate::{disk_usage_eviction_task, tenant};
 use pageserver_api::models::{
-    StatusResponse, TenantConfigRequest, TenantInfo, TimelineCreateRequest, TimelineGcRequest,
-    TimelineInfo,
+    CompactInfoResponse, StatusResponse, TenantConfigRequest, TenantInfo, TimelineCreateRequest,
+    TimelineGcRequest, TimelineInfo,
 };
 use utils::{
    auth::SwappableJwtAuth,
@@ -1695,7 +1696,47 @@ async fn update_tenant_config_handler(
    crate::tenant::Tenant::persist_tenant_config(state.conf, &tenant_shard_id, &location_conf)
        .await
        .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;
-    tenant.set_new_tenant_config(new_tenant_conf);
+
+    let _ = tenant
+        .update_tenant_config(|_crnt| Ok(new_tenant_conf.clone()))
+        .expect("Closure returns Ok()");
+
+    json_response(StatusCode::OK, ())
+}
+
+async fn patch_tenant_config_handler(
+    mut request: Request<Body>,
+    _cancel: CancellationToken,
+) -> Result<Response<Body>, ApiError> {
+    let request_data: TenantConfigPatchRequest = json_request(&mut request).await?;
+    let tenant_id = request_data.tenant_id;
+    check_permission(&request, Some(tenant_id))?;
+
+    let state = get_state(&request);
+
+    let tenant_shard_id = TenantShardId::unsharded(tenant_id);
+
+    let tenant = state
+        .tenant_manager
+        .get_attached_tenant_shard(tenant_shard_id)?;
+    tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
+
+    let updated = tenant
+        .update_tenant_config(|crnt| crnt.apply_patch(request_data.config.clone()))
+        .map_err(ApiError::BadRequest)?;
+
+    // This is a legacy API that only operates on attached tenants: the preferred
+    // API to use is the location_config/ endpoint, which lets the caller provide
+    // the full LocationConf.
+    let location_conf = LocationConf::attached_single(
+        updated,
+        tenant.get_generation(),
+        &ShardParameters::default(),
+    );
+
+    crate::tenant::Tenant::persist_tenant_config(state.conf, &tenant_shard_id, &location_conf)
+        .await
+        .map_err(|e| ApiError::InternalServerError(anyhow::anyhow!(e)))?;

    json_response(StatusCode::OK, ())
 }
@@ -1998,6 +2039,34 @@ async fn timeline_cancel_compact_handler(
    .await
 }

+// Get compact info of a timeline
+async fn timeline_compact_info_handler(
+    request: Request<Body>,
+    _cancel: CancellationToken,
+) -> Result<Response<Body>, ApiError> {
+    let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
+    let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
+    check_permission(&request, Some(tenant_shard_id.tenant_id))?;
+    let state = get_state(&request);
+    async {
+        let tenant = state
+            .tenant_manager
+            .get_attached_tenant_shard(tenant_shard_id)?;
+        let res = tenant.get_scheduled_compaction_tasks(timeline_id);
+        let mut resp = Vec::new();
+        for item in res {
+            resp.push(CompactInfoResponse {
+                compact_key_range: item.compact_key_range,
+                compact_lsn_range: item.compact_lsn_range,
+                sub_compaction: item.sub_compaction,
+            });
+        }
+        json_response(StatusCode::OK, resp)
+    }
+    .instrument(info_span!("timeline_compact_info", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
+    .await
+}
+
 // Run compaction immediately on given timeline.
 async fn timeline_compact_handler(
    mut request: Request<Body>,
@@ -2036,15 +2105,30 @@ async fn timeline_compact_handler(
        parse_query_param::<_, bool>(&request, "wait_until_scheduled_compaction_done")?
            .unwrap_or(false);

+    let sub_compaction = compact_request
+        .as_ref()
+        .map(|r| r.sub_compaction)
+        .unwrap_or(false);
+    let sub_compaction_max_job_size_mb = compact_request
+        .as_ref()
+        .and_then(|r| r.sub_compaction_max_job_size_mb);
+
    let options = CompactOptions {
-        compact_range: compact_request
+        compact_key_range: compact_request
            .as_ref()
-            .and_then(|r| r.compact_range.clone()),
-        compact_below_lsn: compact_request.as_ref().and_then(|r| r.compact_below_lsn),
+            .and_then(|r| r.compact_key_range.clone()),
+        compact_lsn_range: compact_request
+            .as_ref()
+            .and_then(|r| r.compact_lsn_range.clone()),
        flags,
+        sub_compaction,
+        sub_compaction_max_job_size_mb,
    };

-    let scheduled = compact_request.map(|r| r.scheduled).unwrap_or(false);
+    let scheduled = compact_request
+        .as_ref()
+        .map(|r| r.scheduled)
+        .unwrap_or(false);

    async {
        let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
@@ -2053,7 +2137,7 @@ async fn timeline_compact_handler(
            let tenant = state
                .tenant_manager
                .get_attached_tenant_shard(tenant_shard_id)?;
-            let rx = tenant.schedule_compaction(timeline_id, options).await;
+            let rx = tenant.schedule_compaction(timeline_id, options).await.map_err(ApiError::InternalServerError)?;
            if wait_until_scheduled_compaction_done {
                // It is possible that this will take a long time, dropping the HTTP request will not cancel the compaction.
                rx.await.ok();
@@ -3280,6 +3364,9 @@ pub fn make_router(
        .get("/v1/tenant/:tenant_shard_id/synthetic_size", |r| {
            api_handler(r, tenant_size_handler)
        })
+        .patch("/v1/tenant/config", |r| {
+            api_handler(r, patch_tenant_config_handler)
+        })
        .put("/v1/tenant/config", |r| {
            api_handler(r, update_tenant_config_handler)
        })
@@ -3341,6 +3428,10 @@ pub fn make_router(
            "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/do_gc",
            |r| api_handler(r, timeline_gc_handler),
        )
+        .get(
+            "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/compact",
+            |r| api_handler(r, timeline_compact_info_handler),
+        )
        .put(
            "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/compact",
            |r| api_handler(r, timeline_compact_handler),
--- a/pageserver/src/metrics.rs
+++ b/pageserver/src/metrics.rs
@@ -3,7 +3,7 @@ use metrics::{
    register_counter_vec, register_gauge_vec, register_histogram, register_histogram_vec,
    register_int_counter, register_int_counter_pair_vec, register_int_counter_vec,
    register_int_gauge, register_int_gauge_vec, register_uint_gauge, register_uint_gauge_vec,
-    Counter, CounterVec, Gauge, GaugeVec, Histogram, HistogramVec, IntCounter, IntCounterPair,
+    Counter, CounterVec, GaugeVec, Histogram, HistogramVec, IntCounter, IntCounterPair,
    IntCounterPairVec, IntCounterVec, IntGauge, IntGaugeVec, UIntGauge, UIntGaugeVec,
 };
 use once_cell::sync::Lazy;
@@ -16,7 +16,6 @@ use postgres_backend::{is_expected_io_error, QueryError};
 use pq_proto::framed::ConnectionError;
 use strum::{EnumCount, VariantNames};
 use strum_macros::{IntoStaticStr, VariantNames};
-use tracing::warn;
 use utils::id::TimelineId;

 /// Prometheus histogram buckets (in seconds) for operations in the critical
@@ -446,15 +445,6 @@ pub(crate) static WAIT_LSN_TIME: Lazy<Histogram> = Lazy::new(|| {
    .expect("failed to define a metric")
 });

-static FLUSH_WAIT_UPLOAD_TIME: Lazy<GaugeVec> = Lazy::new(|| {
-    register_gauge_vec!(
-        "pageserver_flush_wait_upload_seconds",
-        "Time spent waiting for preceding uploads during layer flush",
-        &["tenant_id", "shard_id", "timeline_id"]
-    )
-    .expect("failed to define a metric")
-});
-
 static LAST_RECORD_LSN: Lazy<IntGaugeVec> = Lazy::new(|| {
    register_int_gauge_vec!(
        "pageserver_last_record_lsn",
@@ -1223,54 +1213,163 @@ pub(crate) mod virtual_file_io_engine {
    });
 }

-pub(crate) struct SmgrOpTimer {
-    global_latency_histo: Histogram,
+pub(crate) struct SmgrOpTimer(Option<SmgrOpTimerInner>);
+pub(crate) struct SmgrOpTimerInner {
+    global_execution_latency_histo: Histogram,
+    per_timeline_execution_latency_histo: Option<Histogram>,

-    // Optional because not all op types are tracked per-timeline
-    per_timeline_latency_histo: Option<Histogram>,
+    global_batch_wait_time: Histogram,
+    per_timeline_batch_wait_time: Histogram,

-    start: Instant,
-    throttled: Duration,
-    op: SmgrQueryType,
+    global_flush_in_progress_micros: IntCounter,
+    per_timeline_flush_in_progress_micros: IntCounter,
+
+    timings: SmgrOpTimerState,
+}
+
+#[derive(Debug)]
+enum SmgrOpTimerState {
+    Received {
+        received_at: Instant,
+    },
+    ThrottleDoneExecutionStarting {
+        received_at: Instant,
+        throttle_started_at: Instant,
+        started_execution_at: Instant,
+    },
+}
+
+pub(crate) struct SmgrOpFlushInProgress {
+    flush_started_at: Instant,
+    global_micros: IntCounter,
+    per_timeline_micros: IntCounter,
 }

 impl SmgrOpTimer {
-    pub(crate) fn deduct_throttle(&mut self, throttle: &Option<Duration>) {
-        let Some(throttle) = throttle else {
-            return;
-        };
-        self.throttled += *throttle;
+    pub(crate) fn observe_throttle_done_execution_starting(&mut self, throttle: &ThrottleResult) {
+        let inner = self.0.as_mut().expect("other public methods consume self");
+        match (&mut inner.timings, throttle) {
+            (SmgrOpTimerState::Received { received_at }, throttle) => match throttle {
+                ThrottleResult::NotThrottled { start } => {
+                    inner.timings = SmgrOpTimerState::ThrottleDoneExecutionStarting {
+                        received_at: *received_at,
+                        throttle_started_at: *start,
+                        started_execution_at: *start,
+                    };
+                }
+                ThrottleResult::Throttled { start, end } => {
+                    inner.timings = SmgrOpTimerState::ThrottleDoneExecutionStarting {
+                        received_at: *start,
+                        throttle_started_at: *start,
+                        started_execution_at: *end,
+                    };
+                }
+            },
+            (x, _) => panic!("called in unexpected state: {x:?}"),
+        }
+    }
+
+    pub(crate) fn observe_smgr_op_completion_and_start_flushing(mut self) -> SmgrOpFlushInProgress {
+        let (flush_start, inner) = self
+            .smgr_op_end()
+            .expect("this method consume self, and the only other caller is drop handler");
+        let SmgrOpTimerInner {
+            global_flush_in_progress_micros,
+            per_timeline_flush_in_progress_micros,
+            ..
+        } = inner;
+        SmgrOpFlushInProgress {
+            flush_started_at: flush_start,
+            global_micros: global_flush_in_progress_micros,
+            per_timeline_micros: per_timeline_flush_in_progress_micros,
+        }
+    }
+
+    /// Returns `None`` if this method has already been called, `Some` otherwise.
+    fn smgr_op_end(&mut self) -> Option<(Instant, SmgrOpTimerInner)> {
+        let inner = self.0.take()?;
+
+        let now = Instant::now();
+
+        let batch;
+        let execution;
+        let throttle;
+        match inner.timings {
+            SmgrOpTimerState::Received { received_at } => {
+                batch = (now - received_at).as_secs_f64();
+                // TODO: use label for dropped requests.
+                // This is quite rare in practice, only during tenant/pageservers shutdown.
+                throttle = Duration::ZERO;
+                execution = Duration::ZERO.as_secs_f64();
+            }
+            SmgrOpTimerState::ThrottleDoneExecutionStarting {
+                received_at,
+                throttle_started_at,
+                started_execution_at,
+            } => {
+                batch = (throttle_started_at - received_at).as_secs_f64();
+                throttle = started_execution_at - throttle_started_at;
+                execution = (now - started_execution_at).as_secs_f64();
+            }
+        }
+
+        // update time spent in batching
+        inner.global_batch_wait_time.observe(batch);
+        inner.per_timeline_batch_wait_time.observe(batch);
+
+        // time spent in throttle metric is updated by throttle impl
+        let _ = throttle;
+
+        // update metrics for execution latency
+        inner.global_execution_latency_histo.observe(execution);
+        if let Some(per_timeline_execution_latency_histo) =
+            &inner.per_timeline_execution_latency_histo
+        {
+            per_timeline_execution_latency_histo.observe(execution);
+        }
+
+        Some((now, inner))
    }
 }

 impl Drop for SmgrOpTimer {
    fn drop(&mut self) {
-        let elapsed = self.start.elapsed();
+        self.smgr_op_end();
+    }
+}

-        let elapsed = match elapsed.checked_sub(self.throttled) {
-            Some(elapsed) => elapsed,
-            None => {
-                use utils::rate_limit::RateLimit;
-                static LOGGED: Lazy<Mutex<enum_map::EnumMap<SmgrQueryType, RateLimit>>> =
-                    Lazy::new(|| {
-                        Mutex::new(enum_map::EnumMap::from_array(std::array::from_fn(|_| {
-                            RateLimit::new(Duration::from_secs(10))
-                        })))
-                    });
-                let mut guard = LOGGED.lock().unwrap();
-                let rate_limit = &mut guard[self.op];
-                rate_limit.call(|| {
-                    warn!(op=?self.op, ?elapsed, ?self.throttled, "implementation error: time spent throttled exceeds total request wall clock time");
-                });
-                elapsed // un-throttled time, more info than just saturating to 0
+impl SmgrOpFlushInProgress {
+    pub(crate) async fn measure<Fut, O>(mut self, mut fut: Fut) -> O
+    where
+        Fut: std::future::Future<Output = O>,
+    {
+        let mut fut = std::pin::pin!(fut);
+
+        let now = Instant::now();
+        // Whenever observe_guard gets called, or dropped,
+        // it adds the time elapsed since its last call to metrics.
+        // Last call is tracked in `now`.
+        let mut observe_guard = scopeguard::guard(
+            || {
+                let elapsed = now - self.flush_started_at;
+                self.global_micros
+                    .inc_by(u64::try_from(elapsed.as_micros()).unwrap());
+                self.per_timeline_micros
+                    .inc_by(u64::try_from(elapsed.as_micros()).unwrap());
+                self.flush_started_at = now;
+            },
+            |mut observe| {
+                observe();
+            },
+        );
+
+        loop {
+            match tokio::time::timeout(Duration::from_secs(10), &mut fut).await {
+                Ok(v) => return v,
+                Err(_timeout) => {
+                    (*observe_guard)();
+                }
            }
-        };
-
-        let elapsed = elapsed.as_secs_f64();
-
-        self.global_latency_histo.observe(elapsed);
-        if let Some(per_timeline_getpage_histo) = &self.per_timeline_latency_histo {
-            per_timeline_getpage_histo.observe(elapsed);
        }
    }
 }
@@ -1302,6 +1401,10 @@ pub(crate) struct SmgrQueryTimePerTimeline {
    per_timeline_getpage_latency: Histogram,
    global_batch_size: Histogram,
    per_timeline_batch_size: Histogram,
+    global_flush_in_progress_micros: IntCounter,
+    per_timeline_flush_in_progress_micros: IntCounter,
+    global_batch_wait_time: Histogram,
+    per_timeline_batch_wait_time: Histogram,
 }

 static SMGR_QUERY_STARTED_GLOBAL: Lazy<IntCounterVec> = Lazy::new(|| {
@@ -1324,12 +1427,15 @@ static SMGR_QUERY_STARTED_PER_TENANT_TIMELINE: Lazy<IntCounterVec> = Lazy::new(|
    .expect("failed to define a metric")
 });

+// Alias so all histograms recording per-timeline smgr timings use the same buckets.
+static SMGR_QUERY_TIME_PER_TENANT_TIMELINE_BUCKETS: &[f64] = CRITICAL_OP_BUCKETS;
+
 static SMGR_QUERY_TIME_PER_TENANT_TIMELINE: Lazy<HistogramVec> = Lazy::new(|| {
    register_histogram_vec!(
        "pageserver_smgr_query_seconds",
-        "Time spent on smgr query handling, aggegated by query type and tenant/timeline.",
+        "Time spent _executing_ smgr query handling, excluding batch and throttle delays.",
        &["smgr_query_type", "tenant_id", "shard_id", "timeline_id"],
-        CRITICAL_OP_BUCKETS.into(),
+        SMGR_QUERY_TIME_PER_TENANT_TIMELINE_BUCKETS.into(),
    )
    .expect("failed to define a metric")
 });
@@ -1387,7 +1493,7 @@ static SMGR_QUERY_TIME_GLOBAL_BUCKETS: Lazy<Vec<f64>> = Lazy::new(|| {
 static SMGR_QUERY_TIME_GLOBAL: Lazy<HistogramVec> = Lazy::new(|| {
    register_histogram_vec!(
        "pageserver_smgr_query_seconds_global",
-        "Time spent on smgr query handling, aggregated by query type.",
+        "Like pageserver_smgr_query_seconds, but aggregated to instance level.",
        &["smgr_query_type"],
        SMGR_QUERY_TIME_GLOBAL_BUCKETS.clone(),
    )
@@ -1464,6 +1570,45 @@ fn set_page_service_config_max_batch_size(conf: &PageServicePipeliningConfig) {
        .set(value.try_into().unwrap());
 }

+static PAGE_SERVICE_SMGR_FLUSH_INPROGRESS_MICROS: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
+        "pageserver_page_service_pagestream_flush_in_progress_micros",
+        "Counter that sums up the microseconds that a pagestream response was being flushed into the TCP connection. \
+         If the flush is particularly slow, this counter will be updated periodically to make slow flushes \
+         easily discoverable in monitoring. \
+         Hence, this is NOT a completion latency historgram.",
+        &["tenant_id", "shard_id", "timeline_id"],
+    )
+    .expect("failed to define a metric")
+});
+
+static PAGE_SERVICE_SMGR_FLUSH_INPROGRESS_MICROS_GLOBAL: Lazy<IntCounter> = Lazy::new(|| {
+    register_int_counter!(
+        "pageserver_page_service_pagestream_flush_in_progress_micros_global",
+        "Like pageserver_page_service_pagestream_flush_in_progress_seconds, but instance-wide.",
+    )
+    .expect("failed to define a metric")
+});
+
+static PAGE_SERVICE_SMGR_BATCH_WAIT_TIME: Lazy<HistogramVec> = Lazy::new(|| {
+    register_histogram_vec!(
+        "pageserver_page_service_pagestream_batch_wait_time_seconds",
+        "Time a request spent waiting in its batch until the batch moved to throttle&execution.",
+        &["tenant_id", "shard_id", "timeline_id"],
+        SMGR_QUERY_TIME_PER_TENANT_TIMELINE_BUCKETS.into(),
+    )
+    .expect("failed to define a metric")
+});
+
+static PAGE_SERVICE_SMGR_BATCH_WAIT_TIME_GLOBAL: Lazy<Histogram> = Lazy::new(|| {
+    register_histogram!(
+        "pageserver_page_service_pagestream_batch_wait_time_seconds_global",
+        "Like pageserver_page_service_pagestream_batch_wait_time_seconds, but aggregated to instance level.",
+        SMGR_QUERY_TIME_GLOBAL_BUCKETS.to_vec(),
+    )
+    .expect("failed to define a metric")
+});
+
 impl SmgrQueryTimePerTimeline {
    pub(crate) fn new(tenant_shard_id: &TenantShardId, timeline_id: &TimelineId) -> Self {
        let tenant_id = tenant_shard_id.tenant_id.to_string();
@@ -1504,6 +1649,17 @@ impl SmgrQueryTimePerTimeline {
            .get_metric_with_label_values(&[&tenant_id, &shard_slug, &timeline_id])
            .unwrap();

+        let global_batch_wait_time = PAGE_SERVICE_SMGR_BATCH_WAIT_TIME_GLOBAL.clone();
+        let per_timeline_batch_wait_time = PAGE_SERVICE_SMGR_BATCH_WAIT_TIME
+            .get_metric_with_label_values(&[&tenant_id, &shard_slug, &timeline_id])
+            .unwrap();
+
+        let global_flush_in_progress_micros =
+            PAGE_SERVICE_SMGR_FLUSH_INPROGRESS_MICROS_GLOBAL.clone();
+        let per_timeline_flush_in_progress_micros = PAGE_SERVICE_SMGR_FLUSH_INPROGRESS_MICROS
+            .get_metric_with_label_values(&[&tenant_id, &shard_slug, &timeline_id])
+            .unwrap();
+
        Self {
            global_started,
            global_latency,
@@ -1511,9 +1667,13 @@ impl SmgrQueryTimePerTimeline {
            per_timeline_getpage_started,
            global_batch_size,
            per_timeline_batch_size,
+            global_flush_in_progress_micros,
+            per_timeline_flush_in_progress_micros,
+            global_batch_wait_time,
+            per_timeline_batch_wait_time,
        }
    }
-    pub(crate) fn start_smgr_op(&self, op: SmgrQueryType, started_at: Instant) -> SmgrOpTimer {
+    pub(crate) fn start_smgr_op(&self, op: SmgrQueryType, received_at: Instant) -> SmgrOpTimer {
        self.global_started[op as usize].inc();

        let per_timeline_latency_histo = if matches!(op, SmgrQueryType::GetPageAtLsn) {
@@ -1523,13 +1683,17 @@ impl SmgrQueryTimePerTimeline {
            None
        };

-        SmgrOpTimer {
-            global_latency_histo: self.global_latency[op as usize].clone(),
-            per_timeline_latency_histo,
-            start: started_at,
-            op,
-            throttled: Duration::ZERO,
-        }
+        SmgrOpTimer(Some(SmgrOpTimerInner {
+            global_execution_latency_histo: self.global_latency[op as usize].clone(),
+            per_timeline_execution_latency_histo: per_timeline_latency_histo,
+            timings: SmgrOpTimerState::Received { received_at },
+            global_flush_in_progress_micros: self.global_flush_in_progress_micros.clone(),
+            per_timeline_flush_in_progress_micros: self
+                .per_timeline_flush_in_progress_micros
+                .clone(),
+            global_batch_wait_time: self.global_batch_wait_time.clone(),
+            per_timeline_batch_wait_time: self.per_timeline_batch_wait_time.clone(),
+        }))
    }

    pub(crate) fn observe_getpage_batch_start(&self, batch_size: usize) {
@@ -2204,6 +2368,15 @@ pub(crate) static WAL_INGEST: Lazy<WalIngestMetrics> = Lazy::new(|| WalIngestMet
    .expect("failed to define a metric"),
 });

+pub(crate) static PAGESERVER_TIMELINE_WAL_RECORDS_RECEIVED: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
+        "pageserver_timeline_wal_records_received",
+        "Number of WAL records received per shard",
+        &["tenant_id", "shard_id", "timeline_id"]
+    )
+    .expect("failed to define a metric")
+});
+
 pub(crate) static WAL_REDO_TIME: Lazy<Histogram> = Lazy::new(|| {
    register_histogram!(
        "pageserver_wal_redo_seconds",
@@ -2404,7 +2577,6 @@ pub(crate) struct TimelineMetrics {
    shard_id: String,
    timeline_id: String,
    pub flush_time_histo: StorageTimeMetrics,
-    pub flush_wait_upload_time_gauge: Gauge,
    pub compact_time_histo: StorageTimeMetrics,
    pub create_images_time_histo: StorageTimeMetrics,
    pub logical_size_histo: StorageTimeMetrics,
@@ -2431,6 +2603,7 @@ pub(crate) struct TimelineMetrics {
    pub evictions_with_low_residence_duration: std::sync::RwLock<EvictionsWithLowResidenceDuration>,
    /// Number of valid LSN leases.
    pub valid_lsn_lease_count_gauge: UIntGauge,
+    pub wal_records_received: IntCounter,
    shutdown: std::sync::atomic::AtomicBool,
 }

@@ -2449,9 +2622,6 @@ impl TimelineMetrics {
            &shard_id,
            &timeline_id,
        );
-        let flush_wait_upload_time_gauge = FLUSH_WAIT_UPLOAD_TIME
-            .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
-            .unwrap();
        let compact_time_histo = StorageTimeMetrics::new(
            StorageTimeOperation::Compact,
            &tenant_id,
@@ -2588,12 +2758,15 @@ impl TimelineMetrics {
            .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
            .unwrap();

+        let wal_records_received = PAGESERVER_TIMELINE_WAL_RECORDS_RECEIVED
+            .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
+            .unwrap();
+
        TimelineMetrics {
            tenant_id,
            shard_id,
            timeline_id,
            flush_time_histo,
-            flush_wait_upload_time_gauge,
            compact_time_histo,
            create_images_time_histo,
            logical_size_histo,
@@ -2620,6 +2793,7 @@ impl TimelineMetrics {
                evictions_with_low_residence_duration,
            ),
            valid_lsn_lease_count_gauge,
+            wal_records_received,
            shutdown: std::sync::atomic::AtomicBool::default(),
        }
    }
@@ -2642,14 +2816,6 @@ impl TimelineMetrics {
        self.resident_physical_size_gauge.get()
    }

-    pub(crate) fn flush_wait_upload_time_gauge_add(&self, duration: f64) {
-        self.flush_wait_upload_time_gauge.add(duration);
-        crate::metrics::FLUSH_WAIT_UPLOAD_TIME
-            .get_metric_with_label_values(&[&self.tenant_id, &self.shard_id, &self.timeline_id])
-            .unwrap()
-            .add(duration);
-    }
-
    pub(crate) fn shutdown(&self) {
        let was_shutdown = self
            .shutdown
@@ -2667,7 +2833,6 @@ impl TimelineMetrics {
        let shard_id = &self.shard_id;
        let _ = LAST_RECORD_LSN.remove_label_values(&[tenant_id, shard_id, timeline_id]);
        let _ = DISK_CONSISTENT_LSN.remove_label_values(&[tenant_id, shard_id, timeline_id]);
-        let _ = FLUSH_WAIT_UPLOAD_TIME.remove_label_values(&[tenant_id, shard_id, timeline_id]);
        let _ = STANDBY_HORIZON.remove_label_values(&[tenant_id, shard_id, timeline_id]);
        {
            RESIDENT_PHYSICAL_SIZE_GLOBAL.sub(self.resident_physical_size_get());
@@ -2757,6 +2922,21 @@ impl TimelineMetrics {
            shard_id,
            timeline_id,
        ]);
+        let _ = PAGESERVER_TIMELINE_WAL_RECORDS_RECEIVED.remove_label_values(&[
+            tenant_id,
+            shard_id,
+            timeline_id,
+        ]);
+        let _ = PAGE_SERVICE_SMGR_FLUSH_INPROGRESS_MICROS.remove_label_values(&[
+            tenant_id,
+            shard_id,
+            timeline_id,
+        ]);
+        let _ = PAGE_SERVICE_SMGR_BATCH_WAIT_TIME.remove_label_values(&[
+            tenant_id,
+            shard_id,
+            timeline_id,
+        ]);
    }
 }

@@ -2787,6 +2967,7 @@ use crate::context::{PageContentKind, RequestContext};
 use crate::task_mgr::TaskKind;
 use crate::tenant::mgr::TenantSlot;
 use crate::tenant::tasks::BackgroundLoopKind;
+use crate::tenant::throttle::ThrottleResult;
 use crate::tenant::Timeline;

 /// Maintain a per timeline gauge in addition to the global gauge.
@@ -3641,6 +3822,7 @@ pub fn preinitialize_metrics(conf: &'static PageServerConf) {
        &REMOTE_ONDEMAND_DOWNLOADED_BYTES,
        &CIRCUIT_BREAKERS_BROKEN,
        &CIRCUIT_BREAKERS_UNBROKEN,
+        &PAGE_SERVICE_SMGR_FLUSH_INPROGRESS_MICROS_GLOBAL,
    ]
    .into_iter()
    .for_each(|c| {
@@ -3688,6 +3870,7 @@ pub fn preinitialize_metrics(conf: &'static PageServerConf) {
        &WAL_REDO_BYTES_HISTOGRAM,
        &WAL_REDO_PROCESS_LAUNCH_DURATION_HISTOGRAM,
        &PAGE_SERVICE_BATCH_SIZE_GLOBAL,
+        &PAGE_SERVICE_SMGR_BATCH_WAIT_TIME_GLOBAL,
    ]
    .into_iter()
    .for_each(|h| {
--- a/pageserver/src/page_service.rs
+++ b/pageserver/src/page_service.rs
@@ -575,7 +575,10 @@ enum BatchedFeMessage {
 }

 impl BatchedFeMessage {
-    async fn throttle(&mut self, cancel: &CancellationToken) -> Result<(), QueryError> {
+    async fn throttle_and_record_start_processing(
+        &mut self,
+        cancel: &CancellationToken,
+    ) -> Result<(), QueryError> {
        let (shard, tokens, timers) = match self {
            BatchedFeMessage::Exists { shard, timer, .. }
            | BatchedFeMessage::Nblocks { shard, timer, .. }
@@ -603,7 +606,7 @@ impl BatchedFeMessage {
            }
        };
        for timer in timers {
-            timer.deduct_throttle(&throttled);
+            timer.observe_throttle_done_execution_starting(&throttled);
        }
        Ok(())
    }
@@ -1017,10 +1020,8 @@ impl PageServerHandler {
        // Map handler result to protocol behavior.
        // Some handler errors cause exit from pagestream protocol.
        // Other handler errors are sent back as an error message and we stay in pagestream protocol.
-        let mut timers: smallvec::SmallVec<[_; 1]> =
-            smallvec::SmallVec::with_capacity(handler_results.len());
        for handler_result in handler_results {
-            let response_msg = match handler_result {
+            let (response_msg, timer) = match handler_result {
                Err(e) => match &e {
                    PageStreamError::Shutdown => {
                        // If we fail to fulfil a request during shutdown, which may be _because_ of
@@ -1044,34 +1045,66 @@ impl PageServerHandler {
                        span.in_scope(|| {
                            error!("error reading relation or page version: {full:#}")
                        });
-                        PagestreamBeMessage::Error(PagestreamErrorResponse {
-                            message: e.to_string(),
-                        })
+                        (
+                            PagestreamBeMessage::Error(PagestreamErrorResponse {
+                                message: e.to_string(),
+                            }),
+                            None, // TODO: measure errors
+                        )
                    }
                },
-                Ok((response_msg, timer)) => {
-                    // Extending the lifetime of the timers so observations on drop
-                    // include the flush time.
-                    timers.push(timer);
-                    response_msg
-                }
+                Ok((response_msg, timer)) => (response_msg, Some(timer)),
            };

+            //
            // marshal & transmit response message
+            //
+
            pgb_writer.write_message_noflush(&BeMessage::CopyData(&response_msg.serialize()))?;
-        }
-        tokio::select! {
-            biased;
-            _ = cancel.cancelled() => {
-                // We were requested to shut down.
-                info!("shutdown request received in page handler");
-                return Err(QueryError::Shutdown)
-            }
-            res = pgb_writer.flush() => {
-                res?;
+
+            // We purposefully don't count flush time into the timer.
+            //
+            // The reason is that current compute client will not perform protocol processing
+            // if the postgres backend process is doing things other than `->smgr_read()`.
+            // This is especially the case for prefetch.
+            //
+            // If the compute doesn't read from the connection, eventually TCP will backpressure
+            // all the way into our flush call below.
+            //
+            // The timer's underlying metric is used for a storage-internal latency SLO and
+            // we don't want to include latency in it that we can't control.
+            // And as pointed out above, in this case, we don't control the time that flush will take.
+            let flushing_timer =
+                timer.map(|timer| timer.observe_smgr_op_completion_and_start_flushing());
+
+            // what we want to do
+            let flush_fut = pgb_writer.flush();
+            // metric for how long flushing takes
+            let flush_fut = match flushing_timer {
+                Some(flushing_timer) => {
+                    futures::future::Either::Left(flushing_timer.measure(flush_fut))
+                }
+                None => futures::future::Either::Right(flush_fut),
+            };
+            // do it while respecting cancellation
+            let _: () = async move {
+                tokio::select! {
+                    biased;
+                    _ = cancel.cancelled() => {
+                        // We were requested to shut down.
+                        info!("shutdown request received in page handler");
+                        return Err(QueryError::Shutdown)
+                    }
+                    res = flush_fut => {
+                        res?;
+                    }
+                }
+                Ok(())
            }
+            // and log the info! line inside the request span
+            .instrument(span.clone())
+            .await?;
        }
-        drop(timers);
        Ok(())
    }

@@ -1200,7 +1233,7 @@ impl PageServerHandler {
                }
            };

-            if let Err(cancelled) = msg.throttle(&self.cancel).await {
+            if let Err(cancelled) = msg.throttle_and_record_start_processing(&self.cancel).await {
                break cancelled;
            }

@@ -1367,7 +1400,9 @@ impl PageServerHandler {
                            return Err(e);
                        }
                    };
-                    batch.throttle(&self.cancel).await?;
+                    batch
+                        .throttle_and_record_start_processing(&self.cancel)
+                        .await?;
                    self.pagesteam_handle_batched_message(pgb_writer, batch, &cancel, &ctx)
                        .await?;
                }
--- a/pageserver/src/pgdatadir_mapping.rs
+++ b/pageserver/src/pgdatadir_mapping.rs
@@ -1242,7 +1242,7 @@ pub struct DatadirModification<'a> {
    pending_metadata_bytes: usize,
 }

-impl<'a> DatadirModification<'a> {
+impl DatadirModification<'_> {
    // When a DatadirModification is committed, we do a monolithic serialization of all its contents.  WAL records can
    // contain multiple pages, so the pageserver's record-based batch size isn't sufficient to bound this allocation: we
    // additionally specify a limit on how much payload a DatadirModification may contain before it should be committed.
@@ -1263,7 +1263,7 @@ impl<'a> DatadirModification<'a> {
    pub(crate) fn has_dirty_data(&self) -> bool {
        self.pending_data_batch
            .as_ref()
-            .map_or(false, |b| b.has_data())
+            .is_some_and(|b| b.has_data())
    }

    /// Set the current lsn
@@ -1319,18 +1319,23 @@ impl<'a> DatadirModification<'a> {

        let buf: Bytes = SlruSegmentDirectory::ser(&SlruSegmentDirectory::default())?.into();
        let empty_dir = Value::Image(buf);
-        self.put(slru_dir_to_key(SlruKind::Clog), empty_dir.clone());
-        self.pending_directory_entries
-            .push((DirectoryKind::SlruSegment(SlruKind::Clog), 0));
-        self.put(
-            slru_dir_to_key(SlruKind::MultiXactMembers),
-            empty_dir.clone(),
-        );
-        self.pending_directory_entries
-            .push((DirectoryKind::SlruSegment(SlruKind::Clog), 0));
-        self.put(slru_dir_to_key(SlruKind::MultiXactOffsets), empty_dir);
-        self.pending_directory_entries
-            .push((DirectoryKind::SlruSegment(SlruKind::MultiXactOffsets), 0));
+
+        // Initialize SLRUs on shard 0 only: creating these on other shards would be
+        // harmless but they'd just be dropped on later compaction.
+        if self.tline.tenant_shard_id.is_shard_zero() {
+            self.put(slru_dir_to_key(SlruKind::Clog), empty_dir.clone());
+            self.pending_directory_entries
+                .push((DirectoryKind::SlruSegment(SlruKind::Clog), 0));
+            self.put(
+                slru_dir_to_key(SlruKind::MultiXactMembers),
+                empty_dir.clone(),
+            );
+            self.pending_directory_entries
+                .push((DirectoryKind::SlruSegment(SlruKind::Clog), 0));
+            self.put(slru_dir_to_key(SlruKind::MultiXactOffsets), empty_dir);
+            self.pending_directory_entries
+                .push((DirectoryKind::SlruSegment(SlruKind::MultiXactOffsets), 0));
+        }

        Ok(())
    }
@@ -2225,7 +2230,7 @@ impl<'a> DatadirModification<'a> {
                assert!(!self
                    .pending_data_batch
                    .as_ref()
-                    .map_or(false, |b| b.updates_key(&key)));
+                    .is_some_and(|b| b.updates_key(&key)));
            }
        }

@@ -2294,7 +2299,7 @@ pub enum Version<'a> {
    Modified(&'a DatadirModification<'a>),
 }

-impl<'a> Version<'a> {
+impl Version<'_> {
    async fn get(
        &self,
        timeline: &Timeline,
--- a/pageserver/src/tenant.rs
+++ b/pageserver/src/tenant.rs
--- a/Show More
+++ b/Show More