hopefully unbroken wip

## Problem

ref https://github.com/neondatabase/neon/issues/4373

## Summary of changes

A step towards immutable layer map. I decided to finish the refactor
with this new approach and apply
https://github.com/neondatabase/neon/pull/4455 on this patch later.

In this PR, we moved all modifications of the layer map to one place
with semantic operations like `initialize_local_layers`,
`finish_compact_l0`, `finish_gc_timeline`, etc, which is now part
of `LayerManager`. This makes it easier to build new features upon
this PR:

* For immutable storage state refactor, we can simply replace the layer
map with `ArcSwap<LayerMap>` and remove the `layers` lock. Moving
towards it requires us to put all layer map changes in a single place as
in https://github.com/neondatabase/neon/pull/4455.
* For manifest, we can write to manifest in each of the semantic
functions.

---------

Signed-off-by: Alex Chi Z <chi@neon.tech>
Co-authored-by: Christian Schwarz <christian@neon.tech>

.github/actions/allure-report-generate/action.yml

@@ -105,7 +105,7 @@ runs:
         # Get previously uploaded data for this run
         ZSTD_NBTHREADS=0
 
-        S3_FILEPATHS=$(aws s3api list-objects-v2 --bucket ${BUCKET} --prefix ${RAW_PREFIX}/ | jq --raw-output  '.Contents[].Key')
+        S3_FILEPATHS=$(aws s3api list-objects-v2 --bucket ${BUCKET} --prefix ${RAW_PREFIX}/ | jq --raw-output '.Contents[]?.Key')
         if [ -z "$S3_FILEPATHS" ]; then
           # There's no previously uploaded data for this $GITHUB_RUN_ID
           exit 0

.github/workflows/benchmarking.yml

@@ -180,7 +180,8 @@ jobs:
       image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:pinned
       options: --init
 
-    timeout-minutes: 360 # 6h
+    # Increase timeout to 8h, default timeout is 6h
+    timeout-minutes: 480
 
     steps:
     - uses: actions/checkout@v3
@@ -321,8 +322,6 @@ jobs:
       image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:pinned
       options: --init
 
-    timeout-minutes: 360 # 6h
-
     steps:
     - uses: actions/checkout@v3
 
@@ -414,8 +413,6 @@ jobs:
       image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:pinned
       options: --init
 
-    timeout-minutes: 360 # 6h
-
     steps:
     - uses: actions/checkout@v3
 
@@ -501,8 +498,6 @@ jobs:
       image: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/rust:pinned
       options: --init
 
-    timeout-minutes: 360 # 6h
-
     steps:
     - uses: actions/checkout@v3
 

.github/workflows/build_and_test.yml

@@ -155,7 +155,7 @@ jobs:
         build_type: [ debug, release ]
     env:
       BUILD_TYPE: ${{ matrix.build_type }}
-      GIT_VERSION: ${{ github.sha }}
+      GIT_VERSION: ${{ github.event.pull_request.head.sha || github.sha }}
 
     steps:
       - name: Fix git ownership
@@ -174,6 +174,27 @@ jobs:
           submodules: true
           fetch-depth: 1
 
+      - name: Check Postgres submodules revision
+        shell: bash -euo pipefail {0}
+        run: |
+          # This is a temporary solution to ensure that the Postgres submodules revision is correct (i.e. the updated intentionally).
+          # Eventually it will be replaced by a regression test https://github.com/neondatabase/neon/pull/4603
+
+          FAILED=false
+          for postgres in postgres-v14 postgres-v15; do
+            expected=$(cat vendor/revisions.json | jq --raw-output '."'"${postgres}"'"')
+            actual=$(git rev-parse "HEAD:vendor/${postgres}")
+            if [ "${expected}" != "${actual}" ]; then
+              echo >&2 "Expected ${postgres} rev to be at '${expected}', but it is at '${actual}'"
+              FAILED=true
+            fi
+          done
+
+          if [ "${FAILED}" = "true" ]; then
+            echo >&2 "Please update vendors/revisions.json if these changes are intentional"
+            exit 1
+          fi
+
       - name: Set pg 14 revision for caching
         id: pg_v14_rev
         run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-v14) >> $GITHUB_OUTPUT
@@ -614,7 +635,7 @@ jobs:
           /kaniko/executor --reproducible --snapshot-mode=redo --skip-unused-stages --cache=true
                            --cache-repo 369495373322.dkr.ecr.eu-central-1.amazonaws.com/cache
                            --context .
-                           --build-arg GIT_VERSION=${{ github.sha }}
+                           --build-arg GIT_VERSION=${{ github.event.pull_request.head.sha || github.sha }}
                            --build-arg REPOSITORY=369495373322.dkr.ecr.eu-central-1.amazonaws.com
                            --destination 369495373322.dkr.ecr.eu-central-1.amazonaws.com/neon:${{needs.tag.outputs.build-tag}}
                            --destination neondatabase/neon:${{needs.tag.outputs.build-tag}}
@@ -658,7 +679,8 @@ jobs:
           /kaniko/executor --reproducible --snapshot-mode=redo --skip-unused-stages --cache=true
                            --cache-repo 369495373322.dkr.ecr.eu-central-1.amazonaws.com/cache
                            --context .
-                           --build-arg GIT_VERSION=${{ github.sha }}
+                           --build-arg GIT_VERSION=${{ github.event.pull_request.head.sha || github.sha }}
+                           --build-arg BUILD_TAG=${{needs.tag.outputs.build-tag}}
                            --build-arg REPOSITORY=369495373322.dkr.ecr.eu-central-1.amazonaws.com
                            --dockerfile Dockerfile.compute-tools
                            --destination 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-tools:${{needs.tag.outputs.build-tag}}
@@ -714,12 +736,42 @@ jobs:
           /kaniko/executor --reproducible --snapshot-mode=redo --skip-unused-stages --cache=true
                            --cache-repo 369495373322.dkr.ecr.eu-central-1.amazonaws.com/cache
                            --context .
-                           --build-arg GIT_VERSION=${{ github.sha }}
+                           --build-arg GIT_VERSION=${{ github.event.pull_request.head.sha || github.sha }}
                            --build-arg PG_VERSION=${{ matrix.version }}
+                           --build-arg BUILD_TAG=${{needs.tag.outputs.build-tag}}
                            --build-arg REPOSITORY=369495373322.dkr.ecr.eu-central-1.amazonaws.com
                            --dockerfile Dockerfile.compute-node
                            --destination 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}}
                            --destination neondatabase/compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}}
+                           --cleanup
+
+      # Due to a kaniko bug, we can't use cache for extensions image, thus it takes about the same amount of time as compute-node image to build (~10 min)
+      # During the transition period we need to have extensions in both places (in S3 and in compute-node image),
+      # so we won't build extension twice, but extract them from compute-node.
+      #
+      # For now we use extensions image only for new custom extensitons
+      - name: Kaniko build extensions only
+        run: |
+          # Kaniko is suposed to clean up after itself if --cleanup flag is set, but it doesn't.
+          # Despite some fixes were made in https://github.com/GoogleContainerTools/kaniko/pull/2504 (in kaniko v1.11.0),
+          # it still fails with error:
+          #   error building image: could not save file: copying file: symlink postgres /kaniko/1/usr/local/pgsql/bin/postmaster: file exists
+          #
+          # Ref https://github.com/GoogleContainerTools/kaniko/issues/1406
+          find /kaniko -maxdepth 1 -mindepth 1 -type d -regex "/kaniko/[0-9]*" -exec rm -rv {} \;
+
+          /kaniko/executor --reproducible --snapshot-mode=redo --skip-unused-stages --cache=true \
+                           --cache-repo 369495373322.dkr.ecr.eu-central-1.amazonaws.com/cache \
+                           --context . \
+                           --build-arg GIT_VERSION=${{ github.event.pull_request.head.sha || github.sha }} \
+                           --build-arg PG_VERSION=${{ matrix.version }} \
+                           --build-arg BUILD_TAG=${{needs.tag.outputs.build-tag}} \
+                           --build-arg REPOSITORY=369495373322.dkr.ecr.eu-central-1.amazonaws.com \
+                           --dockerfile Dockerfile.compute-node \
+                           --destination 369495373322.dkr.ecr.eu-central-1.amazonaws.com/extensions-${{ matrix.version }}:${{needs.tag.outputs.build-tag}} \
+                           --destination neondatabase/extensions-${{ matrix.version }}:${{needs.tag.outputs.build-tag}} \
+                           --cleanup \
+                           --target postgres-extensions
 
       # Cleanup script fails otherwise - rm: cannot remove '/nvme/actions-runner/_work/_temp/_github_home/.ecr': Permission denied
       - name: Cleanup ECR folder
@@ -736,7 +788,7 @@ jobs:
       run:
         shell: sh -eu {0}
     env:
-      VM_BUILDER_VERSION: v0.8.0
+      VM_BUILDER_VERSION: v0.12.1
 
     steps:
       - name: Checkout
@@ -838,8 +890,10 @@ jobs:
           crane tag 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-tools:${{needs.tag.outputs.build-tag}} latest
           crane tag 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-v14:${{needs.tag.outputs.build-tag}} latest
           crane tag 369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-v14:${{needs.tag.outputs.build-tag}} latest
+          crane tag 369495373322.dkr.ecr.eu-central-1.amazonaws.com/extensions-v14:${{needs.tag.outputs.build-tag}} latest
           crane tag 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-v15:${{needs.tag.outputs.build-tag}} latest
           crane tag 369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-v15:${{needs.tag.outputs.build-tag}} latest
+          crane tag 369495373322.dkr.ecr.eu-central-1.amazonaws.com/extensions-v15:${{needs.tag.outputs.build-tag}} latest
 
       - name: Push images to production ECR
         if: |
@@ -850,8 +904,10 @@ jobs:
           crane copy 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-tools:${{needs.tag.outputs.build-tag}} 093970136003.dkr.ecr.eu-central-1.amazonaws.com/compute-tools:latest
           crane copy 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-v14:${{needs.tag.outputs.build-tag}} 093970136003.dkr.ecr.eu-central-1.amazonaws.com/compute-node-v14:latest
           crane copy 369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-v14:${{needs.tag.outputs.build-tag}} 093970136003.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-v14:latest
+          crane copy 369495373322.dkr.ecr.eu-central-1.amazonaws.com/extensions-v14:${{needs.tag.outputs.build-tag}} 093970136003.dkr.ecr.eu-central-1.amazonaws.com/extensions-v14:latest
           crane copy 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-v15:${{needs.tag.outputs.build-tag}} 093970136003.dkr.ecr.eu-central-1.amazonaws.com/compute-node-v15:latest
           crane copy 369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-v15:${{needs.tag.outputs.build-tag}} 093970136003.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-v15:latest
+          crane copy 369495373322.dkr.ecr.eu-central-1.amazonaws.com/extensions-v15:${{needs.tag.outputs.build-tag}} 093970136003.dkr.ecr.eu-central-1.amazonaws.com/extensions-v15:latest
 
       - name: Configure Docker Hub login
         run: |
@@ -873,16 +929,93 @@ jobs:
           crane tag neondatabase/compute-tools:${{needs.tag.outputs.build-tag}} latest
           crane tag neondatabase/compute-node-v14:${{needs.tag.outputs.build-tag}} latest
           crane tag neondatabase/vm-compute-node-v14:${{needs.tag.outputs.build-tag}} latest
+          crane tag neondatabase/extensions-v14:${{needs.tag.outputs.build-tag}} latest
           crane tag neondatabase/compute-node-v15:${{needs.tag.outputs.build-tag}} latest
           crane tag neondatabase/vm-compute-node-v15:${{needs.tag.outputs.build-tag}} latest
+          crane tag neondatabase/extensions-v15:${{needs.tag.outputs.build-tag}} latest
 
       - name: Cleanup ECR folder
         run: rm -rf ~/.ecr
 
+  upload-postgres-extensions-to-s3:
+    if: |
+      (github.ref_name == 'main' || github.ref_name == 'release') &&
+       github.event_name != 'workflow_dispatch'
+    runs-on: ${{ github.ref_name == 'release' && fromJSON('["self-hosted", "prod", "x64"]') || fromJSON('["self-hosted", "gen3", "small"]') }}
+    needs: [ tag, promote-images ]
+    strategy:
+      fail-fast: false
+      matrix:
+        version: [ v14, v15 ]
+
+    env:
+      # While on transition period we extract public extensions from compute-node image and custom extensions from extensions image.
+      # Later all the extensions will be moved to extensions image.
+      EXTENSIONS_IMAGE: ${{ github.ref_name == 'release' && '093970136003' || '369495373322'}}.dkr.ecr.eu-central-1.amazonaws.com/extensions-${{ matrix.version }}:latest
+      COMPUTE_NODE_IMAGE: ${{ github.ref_name == 'release' && '093970136003' || '369495373322'}}.dkr.ecr.eu-central-1.amazonaws.com/compute-node-${{ matrix.version }}:latest
+      AWS_ACCESS_KEY_ID: ${{ github.ref_name == 'release' && secrets.AWS_ACCESS_KEY_PROD || secrets.AWS_ACCESS_KEY_DEV }}
+      AWS_SECRET_ACCESS_KEY: ${{ github.ref_name == 'release' && secrets.AWS_SECRET_KEY_PROD || secrets.AWS_SECRET_KEY_DEV }}
+      S3_BUCKETS: |
+        ${{ github.ref_name == 'release' &&
+          'neon-prod-extensions-ap-southeast-1 neon-prod-extensions-eu-central-1 neon-prod-extensions-us-east-1 neon-prod-extensions-us-east-2 neon-prod-extensions-us-west-2' ||
+          'neon-dev-extensions-eu-central-1 neon-dev-extensions-eu-west-1 neon-dev-extensions-us-east-2' }}
+
+    steps:
+      - name: Pull postgres-extensions image
+        run: |
+          docker pull ${EXTENSIONS_IMAGE}
+          docker pull ${COMPUTE_NODE_IMAGE}
+
+      - name: Create postgres-extensions container
+        id: create-container
+        run: |
+          EID=$(docker create ${EXTENSIONS_IMAGE} true)
+          echo "EID=${EID}" >> $GITHUB_OUTPUT
+
+          CID=$(docker create ${COMPUTE_NODE_IMAGE} true)
+          echo "CID=${CID}" >> $GITHUB_OUTPUT
+
+      - name: Extract postgres-extensions from container
+        run: |
+          rm -rf ./extensions-to-upload ./custom-extensions # Just in case
+
+          # In compute image we have a bit different directory layout
+          mkdir -p extensions-to-upload/share
+          docker cp ${{ steps.create-container.outputs.CID }}:/usr/local/share/extension ./extensions-to-upload/share/extension
+          docker cp ${{ steps.create-container.outputs.CID }}:/usr/local/lib             ./extensions-to-upload/lib
+
+          # Delete Neon extensitons (they always present on compute-node image)
+          rm -rf ./extensions-to-upload/share/extension/neon*
+          rm -rf ./extensions-to-upload/lib/neon*
+
+          # Delete leftovers from the extension build step
+          rm -rf ./extensions-to-upload/lib/pgxs
+          rm -rf ./extensions-to-upload/lib/pkgconfig
+
+          docker cp ${{ steps.create-container.outputs.EID }}:/extensions ./custom-extensions
+          for EXT_NAME in $(ls ./custom-extensions); do
+            mkdir -p ./extensions-to-upload/${EXT_NAME}/share
+
+            mv ./custom-extensions/${EXT_NAME}/share/extension ./extensions-to-upload/${EXT_NAME}/share/extension
+            mv ./custom-extensions/${EXT_NAME}/lib             ./extensions-to-upload/${EXT_NAME}/lib
+          done
+
+      - name: Upload postgres-extensions to S3
+        run: |
+          for BUCKET in $(echo ${S3_BUCKETS}); do
+            aws s3 cp --recursive --only-show-errors ./extensions-to-upload s3://${BUCKET}/${{ needs.tag.outputs.build-tag }}/${{ matrix.version }}
+          done
+
+      - name: Cleanup
+        if: ${{ always() && (steps.create-container.outputs.CID || steps.create-container.outputs.EID) }}
+        run: |
+          docker rm ${{ steps.create-container.outputs.CID }} || true
+          docker rm ${{ steps.create-container.outputs.EID }} || true
+
   deploy:
     runs-on: [ self-hosted, gen3, small ]
     container: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/ansible:latest
-    needs: [ promote-images, tag, regress-tests ]
+    needs: [ upload-postgres-extensions-to-s3, promote-images, tag, regress-tests ]
     if: ( github.ref_name == 'main' || github.ref_name == 'release' ) && github.event_name != 'workflow_dispatch'
     steps:
       - name: Fix git ownership
@@ -914,6 +1047,20 @@ jobs:
             exit 1
           fi
 
+      - name: Create git tag
+        if: github.ref_name == 'release'
+        uses: actions/github-script@v6
+        with:
+          # Retry script for 5XX server errors: https://github.com/actions/github-script#retries
+          retries: 5
+          script: |
+            github.rest.git.createRef({
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              ref: "refs/tags/${{ needs.tag.outputs.build-tag }}",
+              sha: context.sha,
+            })
+
   promote-compatibility-data:
     runs-on: [ self-hosted, gen3, small ]
     container:

.github/workflows/release.yml

@@ -3,6 +3,7 @@ name: Create Release Branch
 on:
   schedule:
     - cron: '0 10 * * 2'
+  workflow_dispatch:
 
 jobs:
   create_release_branch:

Cargo.lock

@@ -158,6 +158,19 @@ dependencies = [
  "syn 1.0.109",
 ]
 
+[[package]]
+name = "async-compression"
+version = "0.4.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5b0122885821398cc923ece939e24d1056a2384ee719432397fa9db87230ff11"
+dependencies = [
+ "flate2",
+ "futures-core",
+ "memchr",
+ "pin-project-lite",
+ "tokio",
+]
+
 [[package]]
 name = "async-stream"
 version = "0.3.5"
@@ -200,17 +213,6 @@ dependencies = [
  "critical-section",
 ]
 
-[[package]]
-name = "atty"
-version = "0.2.14"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "d9b39be18770d11421cdb1b9947a45dd3f37e93092cbf377614828a319d5fee8"
-dependencies = [
- "hermit-abi 0.1.19",
- "libc",
- "winapi",
-]
-
 [[package]]
 name = "autocfg"
 version = "1.1.0"
@@ -604,7 +606,7 @@ dependencies = [
  "cc",
  "cfg-if",
  "libc",
- "miniz_oxide",
+ "miniz_oxide 0.6.2",
  "object",
  "rustc-demangle",
 ]
@@ -805,18 +807,6 @@ dependencies = [
  "libloading",
 ]
 
-[[package]]
-name = "clap"
-version = "3.2.25"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "4ea181bf566f71cb9a5d17a59e1871af638180a18fb0035c92ae62b705207123"
-dependencies = [
- "bitflags",
- "clap_lex 0.2.4",
- "indexmap",
- "textwrap",
-]
-
 [[package]]
 name = "clap"
 version = "4.3.0"
@@ -837,7 +827,7 @@ dependencies = [
  "anstream",
  "anstyle",
  "bitflags",
- "clap_lex 0.5.0",
+ "clap_lex",
  "strsim",
 ]
 
@@ -853,15 +843,6 @@ dependencies = [
  "syn 2.0.16",
 ]
 
-[[package]]
-name = "clap_lex"
-version = "0.2.4"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "2850f2f5a82cbf437dd5af4d49848fbdfc27c157c3d010345776f952765261c5"
-dependencies = [
- "os_str_bytes",
-]
-
 [[package]]
 name = "clap_lex"
 version = "0.5.0"
@@ -914,9 +895,11 @@ name = "compute_tools"
 version = "0.1.0"
 dependencies = [
  "anyhow",
+ "async-compression",
  "chrono",
- "clap 4.3.0",
+ "clap",
  "compute_api",
+ "flate2",
  "futures",
  "hyper",
  "notify",
@@ -924,12 +907,14 @@ dependencies = [
  "opentelemetry",
  "postgres",
  "regex",
+ "remote_storage",
  "reqwest",
  "serde",
  "serde_json",
  "tar",
  "tokio",
  "tokio-postgres",
+ "toml_edit",
  "tracing",
  "tracing-opentelemetry",
  "tracing-subscriber",
@@ -977,7 +962,7 @@ name = "control_plane"
 version = "0.1.0"
 dependencies = [
  "anyhow",
- "clap 4.3.0",
+ "clap",
  "comfy-table",
  "compute_api",
  "git-version",
@@ -997,6 +982,7 @@ dependencies = [
  "tar",
  "thiserror",
  "toml",
+ "tracing",
  "url",
  "utils",
  "workspace_hack",
@@ -1047,19 +1033,19 @@ dependencies = [
 
 [[package]]
 name = "criterion"
-version = "0.4.0"
+version = "0.5.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e7c76e09c1aae2bc52b3d2f29e13c6572553b30c4aa1b8a49fd70de6412654cb"
+checksum = "f2b12d017a929603d80db1831cd3a24082f8137ce19c69e6447f54f5fc8d692f"
 dependencies = [
  "anes",
- "atty",
  "cast",
  "ciborium",
- "clap 3.2.25",
+ "clap",
  "criterion-plot",
+ "is-terminal",
  "itertools",
- "lazy_static",
  "num-traits",
+ "once_cell",
  "oorandom",
  "plotters",
  "rayon",
@@ -1140,7 +1126,7 @@ dependencies = [
  "crossterm_winapi",
  "libc",
  "mio",
- "parking_lot",
+ "parking_lot 0.12.1",
  "signal-hook",
  "signal-hook-mio",
  "winapi",
@@ -1210,7 +1196,7 @@ dependencies = [
  "hashbrown 0.12.3",
  "lock_api",
  "once_cell",
- "parking_lot_core",
+ "parking_lot_core 0.9.7",
 ]
 
 [[package]]
@@ -1399,6 +1385,16 @@ version = "0.4.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "0ce7134b9999ecaf8bcd65542e436736ef32ddca1b3e06094cb6ec5755203b80"
 
+[[package]]
+name = "flate2"
+version = "1.0.26"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3b9429470923de8e8cbd4d2dc513535400b4b3fef0319fb5c4e1f520a7bef743"
+dependencies = [
+ "crc32fast",
+ "miniz_oxide 0.7.1",
+]
+
 [[package]]
 name = "fnv"
 version = "1.0.7"
@@ -1676,15 +1672,6 @@ version = "0.4.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "95505c38b4572b2d910cecb0281560f54b440a19336cbbcb27bf6ce6adc6f5a8"
 
-[[package]]
-name = "hermit-abi"
-version = "0.1.19"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "62b467343b94ba476dcb2500d242dadbb39557df889310ac77c5d99100aaac33"
-dependencies = [
- "libc",
-]
-
 [[package]]
 name = "hermit-abi"
 version = "0.2.6"
@@ -1939,6 +1926,9 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "7a5bbe824c507c5da5956355e86a746d82e0e1464f65d862cc5e71da70e94b2c"
 dependencies = [
  "cfg-if",
+ "js-sys",
+ "wasm-bindgen",
+ "web-sys",
 ]
 
 [[package]]
@@ -2189,6 +2179,15 @@ dependencies = [
  "adler",
 ]
 
+[[package]]
+name = "miniz_oxide"
+version = "0.7.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e7810e0be55b428ada41041c41f32c9f1a42817901b4ccf45fa3d4b6561e74c7"
+dependencies = [
+ "adler",
+]
+
 [[package]]
 name = "mio"
 version = "0.8.6"
@@ -2267,16 +2266,6 @@ dependencies = [
  "windows-sys 0.45.0",
 ]
 
-[[package]]
-name = "nu-ansi-term"
-version = "0.46.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "77a8165726e8236064dbb45459242600304b42a5ea24ee2948e18e023bf7ba84"
-dependencies = [
- "overload",
- "winapi",
-]
-
 [[package]]
 name = "num-bigint"
 version = "0.4.3"
@@ -2349,9 +2338,9 @@ checksum = "0ab1bc2a289d34bd04a330323ac98a1b4bc82c9d9fcb1e66b63caa84da26b575"
 
 [[package]]
 name = "openssl"
-version = "0.10.52"
+version = "0.10.55"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "01b8574602df80f7b85fdfc5392fa884a4e3b3f4f35402c070ab34c3d3f78d56"
+checksum = "345df152bc43501c5eb9e4654ff05f794effb78d4efe3d53abc158baddc0703d"
 dependencies = [
  "bitflags",
  "cfg-if",
@@ -2381,9 +2370,9 @@ checksum = "ff011a302c396a5197692431fc1948019154afc178baf7d8e37367442a4601cf"
 
 [[package]]
 name = "openssl-sys"
-version = "0.9.87"
+version = "0.9.90"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "8e17f59264b2809d77ae94f0e1ebabc434773f370d6ca667bd223ea10e06cc7e"
+checksum = "374533b0e45f3a7ced10fcaeccca020e66656bc03dac384f852e4e5a7a8104a6"
 dependencies = [
  "cc",
  "libc",
@@ -2504,31 +2493,19 @@ dependencies = [
  "winapi",
 ]
 
-[[package]]
-name = "os_str_bytes"
-version = "6.5.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ceedf44fb00f2d1984b0bc98102627ce622e083e49a5bacdb3e514fa4238e267"
-
 [[package]]
 name = "outref"
 version = "0.5.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "4030760ffd992bef45b0ae3f10ce1aba99e33464c90d14dd7c039884963ddc7a"
 
-[[package]]
-name = "overload"
-version = "0.1.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "b15813163c1d831bf4a13c3610c05c0d03b39feb07f7e09fa234dac9b15aaf39"
-
 [[package]]
 name = "pagectl"
 version = "0.1.0"
 dependencies = [
  "anyhow",
  "bytes",
- "clap 4.3.0",
+ "clap",
  "git-version",
  "pageserver",
  "postgres_ffi",
@@ -2542,12 +2519,13 @@ name = "pageserver"
 version = "0.1.0"
 dependencies = [
  "anyhow",
+ "async-compression",
  "async-stream",
  "async-trait",
  "byteorder",
  "bytes",
  "chrono",
- "clap 4.3.0",
+ "clap",
  "close_fds",
  "const_format",
  "consumption_metrics",
@@ -2558,6 +2536,7 @@ dependencies = [
  "enum-map",
  "enumset",
  "fail",
+ "flate2",
  "futures",
  "git-version",
  "hex",
@@ -2629,6 +2608,17 @@ dependencies = [
  "workspace_hack",
 ]
 
+[[package]]
+name = "parking_lot"
+version = "0.11.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7d17b78036a60663b797adeaee46f5c9dfebb86948d1255007a1d6be0271ff99"
+dependencies = [
+ "instant",
+ "lock_api",
+ "parking_lot_core 0.8.6",
+]
+
 [[package]]
 name = "parking_lot"
 version = "0.12.1"
@@ -2636,7 +2626,21 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "3742b2c103b9f06bc9fff0a37ff4912935851bee6d36f3c02bcc755bcfec228f"
 dependencies = [
  "lock_api",
- "parking_lot_core",
+ "parking_lot_core 0.9.7",
+]
+
+[[package]]
+name = "parking_lot_core"
+version = "0.8.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "60a2cfe6f0ad2bfc16aefa463b497d5c7a5ecd44a23efa72aa342d90177356dc"
+dependencies = [
+ "cfg-if",
+ "instant",
+ "libc",
+ "redox_syscall 0.2.16",
+ "smallvec",
+ "winapi",
 ]
 
 [[package]]
@@ -2652,6 +2656,16 @@ dependencies = [
  "windows-sys 0.45.0",
 ]
 
+[[package]]
+name = "pbkdf2"
+version = "0.12.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f0ca0b5a68607598bf3bad68f32227a8164f6254833f84eafaac409cd6746c31"
+dependencies = [
+ "digest",
+ "hmac",
+]
+
 [[package]]
 name = "peeking_take_while"
 version = "0.1.2"
@@ -2957,7 +2971,7 @@ dependencies = [
  "lazy_static",
  "libc",
  "memchr",
- "parking_lot",
+ "parking_lot 0.12.1",
  "procfs",
  "thiserror",
 ]
@@ -3022,12 +3036,11 @@ version = "0.1.0"
 dependencies = [
  "anyhow",
  "async-trait",
- "atty",
  "base64 0.13.1",
  "bstr",
  "bytes",
  "chrono",
- "clap 4.3.0",
+ "clap",
  "consumption_metrics",
  "futures",
  "git-version",
@@ -3045,7 +3058,8 @@ dependencies = [
  "native-tls",
  "once_cell",
  "opentelemetry",
- "parking_lot",
+ "parking_lot 0.12.1",
+ "pbkdf2",
  "pin-project-lite",
  "postgres-native-tls",
  "postgres_backend",
@@ -3056,6 +3070,7 @@ dependencies = [
  "regex",
  "reqwest",
  "reqwest-middleware",
+ "reqwest-retry",
  "reqwest-tracing",
  "routerify",
  "rstest",
@@ -3291,6 +3306,29 @@ dependencies = [
  "thiserror",
 ]
 
+[[package]]
+name = "reqwest-retry"
+version = "0.2.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "48d0fd6ef4c6d23790399fe15efc8d12cd9f3d4133958f9bd7801ee5cbaec6c4"
+dependencies = [
+ "anyhow",
+ "async-trait",
+ "chrono",
+ "futures",
+ "getrandom",
+ "http",
+ "hyper",
+ "parking_lot 0.11.2",
+ "reqwest",
+ "reqwest-middleware",
+ "retry-policies",
+ "task-local-extensions",
+ "tokio",
+ "tracing",
+ "wasm-timer",
+]
+
 [[package]]
 name = "reqwest-tracing"
 version = "0.4.4"
@@ -3309,6 +3347,17 @@ dependencies = [
  "tracing-opentelemetry",
 ]
 
+[[package]]
+name = "retry-policies"
+version = "0.1.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e09bbcb5003282bcb688f0bae741b278e9c7e8f378f561522c9806c58e075d9b"
+dependencies = [
+ "anyhow",
+ "chrono",
+ "rand",
+]
+
 [[package]]
 name = "ring"
 version = "0.16.20"
@@ -3507,7 +3556,7 @@ dependencies = [
  "byteorder",
  "bytes",
  "chrono",
- "clap 4.3.0",
+ "clap",
  "const_format",
  "crc32c",
  "fs2",
@@ -3518,7 +3567,7 @@ dependencies = [
  "hyper",
  "metrics",
  "once_cell",
- "parking_lot",
+ "parking_lot 0.12.1",
  "postgres",
  "postgres-protocol",
  "postgres_backend",
@@ -3937,7 +3986,7 @@ dependencies = [
  "anyhow",
  "async-stream",
  "bytes",
- "clap 4.3.0",
+ "clap",
  "const_format",
  "futures",
  "futures-core",
@@ -3947,7 +3996,7 @@ dependencies = [
  "hyper",
  "metrics",
  "once_cell",
- "parking_lot",
+ "parking_lot 0.12.1",
  "prost",
  "tokio",
  "tokio-stream",
@@ -4118,12 +4167,6 @@ dependencies = [
  "syn 1.0.109",
 ]
 
-[[package]]
-name = "textwrap"
-version = "0.16.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "222a222a5bfe1bba4a77b45ec488a741b3cb8872e5e499451fd7d0129c9c7c3d"
-
 [[package]]
 name = "thiserror"
 version = "1.0.40"
@@ -4281,7 +4324,7 @@ dependencies = [
  "futures-channel",
  "futures-util",
  "log",
- "parking_lot",
+ "parking_lot 0.12.1",
  "percent-encoding",
  "phf",
  "pin-project-lite",
@@ -4539,7 +4582,7 @@ name = "trace"
 version = "0.1.0"
 dependencies = [
  "anyhow",
- "clap 4.3.0",
+ "clap",
  "pageserver_api",
  "utils",
  "workspace_hack",
@@ -4641,7 +4684,6 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "30a651bc37f915e81f087d86e62a18eec5f79550c7faff886f7090b4ea757c77"
 dependencies = [
  "matchers",
- "nu-ansi-term",
  "once_cell",
  "regex",
  "serde",
@@ -4810,11 +4852,11 @@ version = "0.1.0"
 dependencies = [
  "anyhow",
  "async-trait",
- "atty",
  "bincode",
  "byteorder",
  "bytes",
  "chrono",
+ "const_format",
  "criterion",
  "futures",
  "heapless",
@@ -4887,7 +4929,7 @@ name = "wal_craft"
 version = "0.1.0"
 dependencies = [
  "anyhow",
- "clap 4.3.0",
+ "clap",
  "env_logger",
  "log",
  "once_cell",
@@ -4991,6 +5033,21 @@ version = "0.2.86"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "ed9d5b4305409d1fc9482fee2d7f9bcbf24b3972bf59817ef757e23982242a93"
 
+[[package]]
+name = "wasm-timer"
+version = "0.2.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "be0ecb0db480561e9a7642b5d3e4187c128914e58aa84330b9493e3eb68c5e7f"
+dependencies = [
+ "futures",
+ "js-sys",
+ "parking_lot 0.11.2",
+ "pin-utils",
+ "wasm-bindgen",
+ "wasm-bindgen-futures",
+ "web-sys",
+]
+
 [[package]]
 name = "web-sys"
 version = "0.3.63"
@@ -5252,7 +5309,7 @@ dependencies = [
  "anyhow",
  "bytes",
  "chrono",
- "clap 4.3.0",
+ "clap",
  "clap_builder",
  "crossbeam-utils",
  "either",

Cargo.toml

@@ -32,9 +32,10 @@ license = "Apache-2.0"
 ## All dependency versions, used in the project
 [workspace.dependencies]
 anyhow = { version = "1.0", features = ["backtrace"] }
+async-compression = { version = "0.4.0", features = ["tokio", "gzip"] }
+flate2 = "1.0.26"
 async-stream = "0.3"
 async-trait = "0.1"
-atty = "0.2.14"
 aws-config = { version = "0.55", default-features = false, features=["rustls"] }
 aws-sdk-s3 = "0.27"
 aws-smithy-http = "0.55"
@@ -87,6 +88,7 @@ opentelemetry = "0.18.0"
 opentelemetry-otlp = { version = "0.11.0", default_features=false, features = ["http-proto", "trace", "http", "reqwest-client"] }
 opentelemetry-semantic-conventions = "0.10.0"
 parking_lot = "0.12"
+pbkdf2 = "0.12.1"
 pin-project-lite = "0.2"
 prometheus = {version = "0.13", default_features=false, features = ["process"]} # removes protobuf dependency
 prost = "0.11"
@@ -95,6 +97,7 @@ regex = "1.4"
 reqwest = { version = "0.11", default-features = false, features = ["rustls-tls"] }
 reqwest-tracing = { version = "0.4.0", features = ["opentelemetry_0_18"] }
 reqwest-middleware = "0.2.0"
+reqwest-retry = "0.2.2"
 routerify = "3"
 rpds = "0.13"
 rustls = "0.20"
@@ -128,7 +131,7 @@ tonic = {version = "0.9", features = ["tls", "tls-roots"]}
 tracing = "0.1"
 tracing-error = "0.2.0"
 tracing-opentelemetry = "0.18.0"
-tracing-subscriber = { version = "0.3", features = ["env-filter"] }
+tracing-subscriber = { version = "0.3", default_features = false, features = ["smallvec", "fmt", "tracing-log", "std", "env-filter"] }
 url = "2.2"
 uuid = { version = "1.2", features = ["v4", "serde"] }
 walkdir = "2.3.2"
@@ -170,7 +173,7 @@ utils = { version = "0.1", path = "./libs/utils/" }
 workspace_hack = { version = "0.1", path = "./workspace_hack/" }
 
 ## Build dependencies
-criterion = "0.4"
+criterion = "0.5.1"
 rcgen = "0.10"
 rstest = "0.17"
 tempfile = "3.4"

Dockerfile.compute-node

@@ -2,6 +2,7 @@ ARG PG_VERSION
 ARG REPOSITORY=neondatabase
 ARG IMAGE=rust
 ARG TAG=pinned
+ARG BUILD_TAG
 
 #########################################################################################
 #
@@ -131,10 +132,20 @@ RUN wget https://github.com/plv8/plv8/archive/refs/tags/v3.1.5.tar.gz -O plv8.ta
 FROM build-deps AS h3-pg-build
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
-# packaged cmake is too old
-RUN wget https://github.com/Kitware/CMake/releases/download/v3.24.2/cmake-3.24.2-linux-x86_64.sh \
+RUN case "$(uname -m)" in \
+      "x86_64") \
+        export CMAKE_CHECKSUM=739d372726cb23129d57a539ce1432453448816e345e1545f6127296926b6754 \
+        ;; \
+      "aarch64") \
+        export CMAKE_CHECKSUM=281b42627c9a1beed03e29706574d04c6c53fae4994472e90985ef018dd29c02 \
+        ;; \
+      *) \
+        echo "Unsupported architecture '$(uname -m)'. Supported are x86_64 and aarch64" && exit 1 \
+        ;; \
+    esac && \
+    wget https://github.com/Kitware/CMake/releases/download/v3.24.2/cmake-3.24.2-linux-$(uname -m).sh \
       -q -O /tmp/cmake-install.sh \
-      && echo "739d372726cb23129d57a539ce1432453448816e345e1545f6127296926b6754 /tmp/cmake-install.sh" | sha256sum --check \
+      && echo "${CMAKE_CHECKSUM} /tmp/cmake-install.sh" | sha256sum --check \
       && chmod u+x /tmp/cmake-install.sh \
       && /tmp/cmake-install.sh --skip-license --prefix=/usr/local/ \
       && rm /tmp/cmake-install.sh
@@ -188,8 +199,8 @@ RUN wget https://github.com/df7cb/postgresql-unit/archive/refs/tags/7.7.tar.gz -
 FROM build-deps AS vector-pg-build
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
-RUN wget https://github.com/pgvector/pgvector/archive/refs/tags/v0.4.0.tar.gz -O pgvector.tar.gz && \
-    echo "b76cf84ddad452cc880a6c8c661d137ddd8679c000a16332f4f03ecf6e10bcc8 pgvector.tar.gz" | sha256sum --check && \
+RUN wget https://github.com/pgvector/pgvector/archive/refs/tags/v0.4.4.tar.gz -O pgvector.tar.gz && \
+    echo "1cb70a63f8928e396474796c22a20be9f7285a8a013009deb8152445b61b72e6 pgvector.tar.gz" | sha256sum --check && \
     mkdir pgvector-src && cd pgvector-src && tar xvzf ../pgvector.tar.gz --strip-components=1 -C . && \
     make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
     make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
@@ -480,6 +491,79 @@ RUN wget https://github.com/rdkit/rdkit/archive/refs/tags/Release_2023_03_1.tar.
     make -j $(getconf _NPROCESSORS_ONLN) install && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/rdkit.control
 
+#########################################################################################
+#
+# Layer "pg-uuidv7-pg-build"
+# compile pg_uuidv7 extension
+#
+#########################################################################################
+FROM build-deps AS pg-uuidv7-pg-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+ENV PATH "/usr/local/pgsql/bin/:$PATH"
+RUN wget https://github.com/fboulnois/pg_uuidv7/archive/refs/tags/v1.0.1.tar.gz -O pg_uuidv7.tar.gz && \
+    echo "0d0759ab01b7fb23851ecffb0bce27822e1868a4a5819bfd276101c716637a7a pg_uuidv7.tar.gz" | sha256sum --check && \
+    mkdir pg_uuidv7-src && cd pg_uuidv7-src && tar xvzf ../pg_uuidv7.tar.gz --strip-components=1 -C . && \
+    make -j $(getconf _NPROCESSORS_ONLN) && \
+    make -j $(getconf _NPROCESSORS_ONLN) install && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/pg_uuidv7.control
+
+#########################################################################################
+#
+# Layer "pg-roaringbitmap-pg-build"
+# compile pg_roaringbitmap extension
+#
+#########################################################################################
+FROM build-deps AS pg-roaringbitmap-pg-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+ENV PATH "/usr/local/pgsql/bin/:$PATH"
+RUN wget https://github.com/ChenHuajun/pg_roaringbitmap/archive/refs/tags/v0.5.4.tar.gz -O pg_roaringbitmap.tar.gz && \
+    echo "b75201efcb1c2d1b014ec4ae6a22769cc7a224e6e406a587f5784a37b6b5a2aa pg_roaringbitmap.tar.gz" | sha256sum --check && \
+    mkdir pg_roaringbitmap-src && cd pg_roaringbitmap-src && tar xvzf ../pg_roaringbitmap.tar.gz --strip-components=1 -C . && \
+    make -j $(getconf _NPROCESSORS_ONLN) && \
+    make -j $(getconf _NPROCESSORS_ONLN) install && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/roaringbitmap.control
+
+#########################################################################################
+#
+# Layer "pg-embedding-pg-build"
+# compile pg_embedding extension
+#
+#########################################################################################
+FROM build-deps AS pg-embedding-pg-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+ENV PATH "/usr/local/pgsql/bin/:$PATH"
+# 2465f831ea1f8d49c1d74f8959adb7fc277d70cd made on 05/07/2023
+# There is no release tag yet
+RUN wget https://github.com/neondatabase/pg_embedding/archive/2465f831ea1f8d49c1d74f8959adb7fc277d70cd.tar.gz -O pg_embedding.tar.gz && \
+    echo "047af2b1f664a1e6e37867bd4eeaf5934fa27d6ba3d6c4461efa388ddf7cd1d5 pg_embedding.tar.gz" | sha256sum --check && \
+    mkdir pg_embedding-src && cd pg_embedding-src && tar xvzf ../pg_embedding.tar.gz --strip-components=1 -C . && \
+    make -j $(getconf _NPROCESSORS_ONLN) && \
+    make -j $(getconf _NPROCESSORS_ONLN) install && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/embedding.control
+
+#########################################################################################
+#
+# Layer "pg-anon-pg-build"
+# compile anon extension
+#
+#########################################################################################
+FROM build-deps AS pg-anon-pg-build
+COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
+
+# Kaniko doesn't allow to do `${from#/usr/local/pgsql/}`, so we use `${from:17}` instead
+ENV PATH "/usr/local/pgsql/bin/:$PATH"
+RUN wget https://gitlab.com/dalibo/postgresql_anonymizer/-/archive/1.1.0/postgresql_anonymizer-1.1.0.tar.gz -O pg_anon.tar.gz && \
+    echo "08b09d2ff9b962f96c60db7e6f8e79cf7253eb8772516998fc35ece08633d3ad pg_anon.tar.gz" | sha256sum --check && \
+    mkdir pg_anon-src && cd pg_anon-src && tar xvzf ../pg_anon.tar.gz --strip-components=1 -C . && \
+    find /usr/local/pgsql -type f | sort  > /before.txt && \
+    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
+    echo 'trusted = true' >> /usr/local/pgsql/share/extension/anon.control && \
+    find /usr/local/pgsql -type f | sort  > /after.txt && \
+    /bin/bash -c 'for from in $(comm -13 /before.txt /after.txt); do to=/extensions/anon/${from:17} && mkdir -p $(dirname ${to}) && cp -a ${from} ${to}; done'
+
 #########################################################################################
 #
 # Layer "rust extensions"
@@ -588,6 +672,7 @@ RUN wget https://github.com/pksunkara/pgx_ulid/archive/refs/tags/v0.1.0.tar.gz -
 #
 #########################################################################################
 FROM build-deps AS neon-pg-ext-build
+# Public extensions
 COPY --from=postgis-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=postgis-build /sfcgal/* /
 COPY --from=plv8-build /usr/local/pgsql/ /usr/local/pgsql/
@@ -613,6 +698,9 @@ COPY --from=kq-imcx-pg-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg-cron-pg-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg-pgx-ulid-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=rdkit-pg-build /usr/local/pgsql/ /usr/local/pgsql/
+COPY --from=pg-uuidv7-pg-build /usr/local/pgsql/ /usr/local/pgsql/
+COPY --from=pg-roaringbitmap-pg-build /usr/local/pgsql/ /usr/local/pgsql/
+COPY --from=pg-embedding-pg-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY pgxn/ pgxn/
 
 RUN make -j $(getconf _NPROCESSORS_ONLN) \
@@ -634,6 +722,9 @@ RUN make -j $(getconf _NPROCESSORS_ONLN) \
 #
 #########################################################################################
 FROM $REPOSITORY/$IMAGE:$TAG AS compute-tools
+ARG BUILD_TAG
+ENV BUILD_TAG=$BUILD_TAG
+
 USER nonroot
 # Copy entire project to get Cargo.* files with proper dependencies for the whole project
 COPY --chown=nonroot . .
@@ -658,6 +749,22 @@ RUN rm -r /usr/local/pgsql/include
 # if they were to be used by other libraries.
 RUN rm /usr/local/pgsql/lib/lib*.a
 
+#########################################################################################
+#
+# Extenstion only
+#
+#########################################################################################
+FROM scratch AS postgres-extensions
+# After the transition this layer will include all extensitons.
+# As for now, it's only for new custom ones
+#
+# # Default extensions
+# COPY --from=postgres-cleanup-layer /usr/local/pgsql/share/extension /usr/local/pgsql/share/extension
+# COPY --from=postgres-cleanup-layer /usr/local/pgsql/lib             /usr/local/pgsql/lib
+# Custom extensions
+COPY --from=pg-anon-pg-build /extensions/anon/lib/ /extensions/anon/lib
+COPY --from=pg-anon-pg-build /extensions/anon/share/extension /extensions/anon/share/extension
+
 #########################################################################################
 #
 # Final layer

Dockerfile.compute-tools

@@ -3,6 +3,7 @@
 ARG REPOSITORY=neondatabase
 ARG IMAGE=rust
 ARG TAG=pinned
+ARG BUILD_TAG
 
 FROM $REPOSITORY/$IMAGE:$TAG AS rust-build
 WORKDIR /home/nonroot
@@ -16,6 +17,8 @@ ENV CACHEPOT_S3_KEY_PREFIX=cachepot
 ARG CACHEPOT_BUCKET=neon-github-dev
 #ARG AWS_ACCESS_KEY_ID
 #ARG AWS_SECRET_ACCESS_KEY
+ARG BUILD_TAG
+ENV BUILD_TAG=$BUILD_TAG
 
 COPY . .
 

README.md

@@ -132,13 +132,13 @@ Python (3.9 or higher), and install python3 packages using `./scripts/pysync` (r
 # Create repository in .neon with proper paths to binaries and data
 # Later that would be responsibility of a package install script
 > cargo neon init
-Starting pageserver at '127.0.0.1:64000' in '.neon'.
+Initializing pageserver node 1 at '127.0.0.1:64000' in ".neon"
 
 # start pageserver, safekeeper, and broker for their intercommunication
 > cargo neon start
-Starting neon broker at 127.0.0.1:50051
+Starting neon broker at 127.0.0.1:50051.
 storage_broker started, pid: 2918372
-Starting pageserver at '127.0.0.1:64000' in '.neon'.
+Starting pageserver node 1 at '127.0.0.1:64000' in ".neon".
 pageserver started, pid: 2918386
 Starting safekeeper at '127.0.0.1:5454' in '.neon/safekeepers/sk1'.
 safekeeper 1 started, pid: 2918437
@@ -152,8 +152,7 @@ Setting tenant 9ef87a5bf0d92544f6fafeeb3239695c as a default one
 # start postgres compute node
 > cargo neon endpoint start main
 Starting new endpoint main (PostgreSQL v14) on timeline de200bd42b49cc1814412c7e592dd6e9 ...
-Extracting base backup to create postgres instance: path=.neon/pgdatadirs/tenants/9ef87a5bf0d92544f6fafeeb3239695c/main port=55432
-Starting postgres at 'host=127.0.0.1 port=55432 user=cloud_admin dbname=postgres'
+Starting postgres at 'postgresql://cloud_admin@127.0.0.1:55432/postgres'
 
 # check list of running postgres instances
 > cargo neon endpoint list
@@ -189,18 +188,17 @@ Created timeline 'b3b863fa45fa9e57e615f9f2d944e601' at Lsn 0/16F9A00 for tenant:
 # start postgres on that branch
 > cargo neon endpoint start migration_check --branch-name migration_check
 Starting new endpoint migration_check (PostgreSQL v14) on timeline b3b863fa45fa9e57e615f9f2d944e601 ...
-Extracting base backup to create postgres instance: path=.neon/pgdatadirs/tenants/9ef87a5bf0d92544f6fafeeb3239695c/migration_check port=55433
-Starting postgres at 'host=127.0.0.1 port=55433 user=cloud_admin dbname=postgres'
+Starting postgres at 'postgresql://cloud_admin@127.0.0.1:55434/postgres'
 
 # check the new list of running postgres instances
 > cargo neon endpoint list
  ENDPOINT         ADDRESS          TIMELINE                          BRANCH NAME      LSN        STATUS
  main             127.0.0.1:55432  de200bd42b49cc1814412c7e592dd6e9  main             0/16F9A38  running
- migration_check  127.0.0.1:55433  b3b863fa45fa9e57e615f9f2d944e601  migration_check  0/16F9A70  running
+ migration_check  127.0.0.1:55434  b3b863fa45fa9e57e615f9f2d944e601  migration_check  0/16F9A70  running
 
 # this new postgres instance will have all the data from 'main' postgres,
 # but all modifications would not affect data in original postgres
-> psql -p55433 -h 127.0.0.1 -U cloud_admin postgres
+> psql -p55434 -h 127.0.0.1 -U cloud_admin postgres
 postgres=# select * from t;
  key | value
 -----+-------

compute_tools/Cargo.toml

@@ -6,8 +6,10 @@ license.workspace = true
 
 [dependencies]
 anyhow.workspace = true
+async-compression.workspace = true
 chrono.workspace = true
 clap.workspace = true
+flate2.workspace = true
 futures.workspace = true
 hyper = { workspace = true, features = ["full"] }
 notify.workspace = true
@@ -30,3 +32,5 @@ url.workspace = true
 compute_api.workspace = true
 utils.workspace = true
 workspace_hack.workspace = true
+toml_edit.workspace = true
+remote_storage = { version = "0.1", path = "../libs/remote_storage/" }

compute_tools/src/bin/compute_ctl.rs

@@ -5,6 +5,8 @@
 //! - `compute_ctl` accepts cluster (compute node) specification as a JSON file.
 //! - Every start is a fresh start, so the data directory is removed and
 //!   initialized again on each run.
+//! - If remote_extension_config is provided, it will be used to fetch extensions list
+//!  and download `shared_preload_libraries` from the remote storage.
 //! - Next it will put configuration files into the `PGDATA` directory.
 //! - Sync safekeepers and get commit LSN.
 //! - Get `basebackup` from pageserver using the returned on the previous step LSN.
@@ -27,7 +29,8 @@
 //! compute_ctl -D /var/db/postgres/compute \
 //!             -C 'postgresql://cloud_admin@localhost/postgres' \
 //!             -S /var/db/postgres/specs/current.json \
-//!             -b /usr/local/bin/postgres
+//!             -b /usr/local/bin/postgres \
+//!             -r {"bucket": "my-bucket", "region": "eu-central-1", "endpoint": "http:://localhost:9000"} \
 //! ```
 //!
 use std::collections::HashMap;
@@ -35,7 +38,7 @@ use std::fs::File;
 use std::panic;
 use std::path::Path;
 use std::process::exit;
-use std::sync::{mpsc, Arc, Condvar, Mutex};
+use std::sync::{mpsc, Arc, Condvar, Mutex, OnceLock};
 use std::{thread, time::Duration};
 
 use anyhow::{Context, Result};
@@ -48,16 +51,51 @@ use compute_api::responses::ComputeStatus;
 
 use compute_tools::compute::{ComputeNode, ComputeState, ParsedSpec};
 use compute_tools::configurator::launch_configurator;
+use compute_tools::extension_server::launch_download_extensions;
+use compute_tools::extension_server::{get_pg_version, init_remote_storage};
 use compute_tools::http::api::launch_http_server;
 use compute_tools::logger::*;
 use compute_tools::monitor::launch_monitor;
 use compute_tools::params::*;
 use compute_tools::spec::*;
 
+const BUILD_TAG_DEFAULT: &str = "local";
+const DEFAULT_REMOTE_EXT_CONFIG: &str = r#"{"bucket": "neon-dev-extensions", "region": "eu-central-1", "endpoint": null, "prefix": "5555"}"#;
+
 fn main() -> Result<()> {
     init_tracing_and_logging(DEFAULT_LOG_LEVEL)?;
 
+    let build_tag = option_env!("BUILD_TAG").unwrap_or(BUILD_TAG_DEFAULT);
+    info!("build_tag: {build_tag}");
+
     let matches = cli().get_matches();
+    let pgbin_default = String::from("postgres");
+    let pgbin = matches.get_one::<String>("pgbin").unwrap_or(&pgbin_default);
+
+    let remote_ext_config = matches
+        .get_one::<String>("remote-ext-config")
+        .map(|x| x.to_string());
+    // let remote_ext_config =
+    //     Some(remote_ext_config.unwrap_or(DEFAULT_REMOTE_EXT_CONFIG.to_string()));
+
+    let ext_remote_storage = remote_ext_config.map(|x| {
+        init_remote_storage(&x, build_tag)
+            .expect("cannot initialize remote extension storage from config")
+    });
+    // creds used to connect to remote extensions bucket
+    // let aws_creds = matches.get_one::<String>("awscreds");
+    // if let Some(aws_creds) = aws_creds {
+    //     // not sure if this is a bad idea?
+    //     let aws_creds_dict: serde_json::Value = serde_json::from_str(aws_creds)?;
+    //     std::env::set_var(
+    //         "AWS_ACCESS_KEY_ID",
+    //         aws_creds_dict["ID"].as_str().expect("config parse error"),
+    //     );
+    //     std::env::set_var(
+    //         "AWS_SECRET_ACCESS_KEY",
+    //         aws_creds_dict["key"].as_str().expect("config parse error"),
+    //     );
+    // }
 
     let http_port = *matches
         .get_one::<u16>("http-port")
@@ -122,9 +160,6 @@ fn main() -> Result<()> {
     let compute_id = matches.get_one::<String>("compute-id");
     let control_plane_uri = matches.get_one::<String>("control-plane-uri");
 
-    // Try to use just 'postgres' if no path is provided
-    let pgbin = matches.get_one::<String>("pgbin").unwrap();
-
     let spec;
     let mut live_config_allowed = false;
     match spec_json {
@@ -162,6 +197,7 @@ fn main() -> Result<()> {
 
     let mut new_state = ComputeState::new();
     let spec_set;
+
     if let Some(spec) = spec {
         let pspec = ParsedSpec::try_from(spec).map_err(|msg| anyhow::anyhow!(msg))?;
         new_state.pspec = Some(pspec);
@@ -173,9 +209,12 @@ fn main() -> Result<()> {
         connstr: Url::parse(connstr).context("cannot parse connstr as a URL")?,
         pgdata: pgdata.to_string(),
         pgbin: pgbin.to_string(),
+        pgversion: get_pg_version(pgbin),
         live_config_allowed,
         state: Mutex::new(new_state),
         state_changed: Condvar::new(),
+        ext_remote_storage,
+        available_extensions: OnceLock::new(),
     };
     let compute = Arc::new(compute_node);
 
@@ -184,6 +223,8 @@ fn main() -> Result<()> {
     let _http_handle =
         launch_http_server(http_port, &compute).expect("cannot launch http endpoint thread");
 
+    let extension_server_port: u16 = http_port;
+
     if !spec_set {
         // No spec provided, hang waiting for it.
         info!("no compute spec provided, waiting");
@@ -221,10 +262,13 @@ fn main() -> Result<()> {
     let _configurator_handle =
         launch_configurator(&compute).expect("cannot launch configurator thread");
 
+    let _download_extensions_handle =
+        launch_download_extensions(&compute).expect("cannot launch download extensions thread");
+
     // Start Postgres
     let mut delay_exit = false;
     let mut exit_code = None;
-    let pg = match compute.start_compute() {
+    let pg = match compute.start_compute(extension_server_port) {
         Ok(pg) => Some(pg),
         Err(err) => {
             error!("could not start the compute node: {:?}", err);
@@ -250,6 +294,16 @@ fn main() -> Result<()> {
         exit_code = ecode.code()
     }
 
+    // Maybe sync safekeepers again, to speed up next startup
+    let compute_state = compute.state.lock().unwrap().clone();
+    let pspec = compute_state.pspec.as_ref().expect("spec must be set");
+    if matches!(pspec.spec.mode, compute_api::spec::ComputeMode::Primary) {
+        info!("syncing safekeepers on shutdown");
+        let storage_auth_token = pspec.storage_auth_token.clone();
+        let lsn = compute.sync_safekeepers(storage_auth_token)?;
+        info!("synced safekeepers at lsn {lsn}");
+    }
+
     if let Err(err) = compute.check_for_core_dumps() {
         error!("error while checking for core dumps: {err:?}");
     }
@@ -343,6 +397,18 @@ fn cli() -> clap::Command {
                 .long("control-plane-uri")
                 .value_name("CONTROL_PLANE_API_BASE_URI"),
         )
+        .arg(
+            Arg::new("remote-ext-config")
+                .short('r')
+                .long("remote-ext-config")
+                .value_name("REMOTE_EXT_CONFIG"),
+        )
+        .arg(
+            Arg::new("awscreds")
+                .short('k')
+                .long("awscreds")
+                .value_name("AWS_CREDENTIALS"),
+        )
 }
 
 #[test]

compute_tools/src/compute.rs

@@ -1,13 +1,17 @@
+use std::collections::HashSet;
 use std::fs;
+use std::io::BufRead;
 use std::os::unix::fs::PermissionsExt;
 use std::path::Path;
 use std::process::{Command, Stdio};
 use std::str::FromStr;
-use std::sync::{Condvar, Mutex};
+use std::sync::{Condvar, Mutex, OnceLock};
 
 use anyhow::{Context, Result};
 use chrono::{DateTime, Utc};
+use futures::future::join_all;
 use postgres::{Client, NoTls};
+use tokio;
 use tokio_postgres;
 use tracing::{info, instrument, warn};
 use utils::id::{TenantId, TimelineId};
@@ -15,10 +19,13 @@ use utils::lsn::Lsn;
 
 use compute_api::responses::{ComputeMetrics, ComputeStatus};
 use compute_api::spec::{ComputeMode, ComputeSpec};
+use utils::measured_stream::MeasuredReader;
+
+use remote_storage::GenericRemoteStorage;
 
-use crate::config;
 use crate::pg_helpers::*;
 use crate::spec::*;
+use crate::{config, extension_server};
 
 /// Compute node info shared across several `compute_ctl` threads.
 pub struct ComputeNode {
@@ -26,6 +33,7 @@ pub struct ComputeNode {
     pub connstr: url::Url,
     pub pgdata: String,
     pub pgbin: String,
+    pub pgversion: String,
     /// We should only allow live re- / configuration of the compute node if
     /// it uses 'pull model', i.e. it can go to control-plane and fetch
     /// the latest configuration. Otherwise, there could be a case:
@@ -45,6 +53,10 @@ pub struct ComputeNode {
     pub state: Mutex<ComputeState>,
     /// `Condvar` to allow notifying waiters about state changes.
     pub state_changed: Condvar,
+    ///  the S3 bucket that we search for extensions in
+    pub ext_remote_storage: Option<GenericRemoteStorage>,
+    // cached lists of available extensions and libraries
+    pub available_extensions: OnceLock<HashSet<String>>,
 }
 
 #[derive(Clone, Debug)]
@@ -133,6 +145,84 @@ impl TryFrom<ComputeSpec> for ParsedSpec {
     }
 }
 
+/// Create special neon_superuser role, that's a slightly nerfed version of a real superuser
+/// that we give to customers
+fn create_neon_superuser(spec: &ComputeSpec, client: &mut Client) -> Result<()> {
+    let roles = spec
+        .cluster
+        .roles
+        .iter()
+        .map(|r| escape_literal(&r.name))
+        .collect::<Vec<_>>();
+
+    let dbs = spec
+        .cluster
+        .databases
+        .iter()
+        .map(|db| escape_literal(&db.name))
+        .collect::<Vec<_>>();
+
+    let roles_decl = if roles.is_empty() {
+        String::from("roles text[] := NULL;")
+    } else {
+        format!(
+            r#"
+               roles text[] := ARRAY(SELECT rolname
+                                     FROM pg_catalog.pg_roles
+                                     WHERE rolname IN ({}));"#,
+            roles.join(", ")
+        )
+    };
+
+    let database_decl = if dbs.is_empty() {
+        String::from("dbs text[] := NULL;")
+    } else {
+        format!(
+            r#"
+               dbs text[] := ARRAY(SELECT datname
+                                   FROM pg_catalog.pg_database
+                                   WHERE datname IN ({}));"#,
+            dbs.join(", ")
+        )
+    };
+
+    // ALL PRIVILEGES grants CREATE, CONNECT, and TEMPORARY on all databases
+    // (see https://www.postgresql.org/docs/current/ddl-priv.html)
+    let query = format!(
+        r#"
+            DO $$
+                DECLARE
+                    r text;
+                    {}
+                    {}
+                BEGIN
+                    IF NOT EXISTS (
+                        SELECT FROM pg_catalog.pg_roles WHERE rolname = 'neon_superuser')
+                    THEN
+                        CREATE ROLE neon_superuser CREATEDB CREATEROLE NOLOGIN IN ROLE pg_read_all_data, pg_write_all_data;
+                        IF array_length(roles, 1) IS NOT NULL THEN
+                            EXECUTE format('GRANT neon_superuser TO %s',
+                                           array_to_string(ARRAY(SELECT quote_ident(x) FROM unnest(roles) as x), ', '));
+                            FOREACH r IN ARRAY roles LOOP
+                                EXECUTE format('ALTER ROLE %s CREATEROLE CREATEDB', quote_ident(r));
+                            END LOOP;
+                        END IF;
+                        IF array_length(dbs, 1) IS NOT NULL THEN
+                            EXECUTE format('GRANT ALL PRIVILEGES ON DATABASE %s TO neon_superuser',
+                                           array_to_string(ARRAY(SELECT quote_ident(x) FROM unnest(dbs) as x), ', '));
+                        END IF;
+                    END IF;
+                END
+            $$;"#,
+        roles_decl, database_decl,
+    );
+    info!("Neon superuser created:\n{}", &query);
+    client
+        .simple_query(&query)
+        .map_err(|e| anyhow::anyhow!(e).context(query))?;
+    Ok(())
+}
+
 impl ComputeNode {
     pub fn set_status(&self, status: ComputeStatus) {
         let mut state = self.state.lock().unwrap();
@@ -157,7 +247,7 @@ impl ComputeNode {
 
     // Get basebackup from the libpq connection to pageserver using `connstr` and
     // unarchive it to `pgdata` directory overriding all its previous content.
-    #[instrument(skip(self, compute_state))]
+    #[instrument(skip_all, fields(%lsn))]
     fn get_basebackup(&self, compute_state: &ComputeState, lsn: Lsn) -> Result<()> {
         let spec = compute_state.pspec.as_ref().expect("spec must be set");
         let start_time = Utc::now();
@@ -175,20 +265,52 @@ impl ComputeNode {
 
         let mut client = config.connect(NoTls)?;
         let basebackup_cmd = match lsn {
-            Lsn(0) => format!("basebackup {} {}", spec.tenant_id, spec.timeline_id), // First start of the compute
-            _ => format!("basebackup {} {} {}", spec.tenant_id, spec.timeline_id, lsn),
+            // HACK We don't use compression on first start (Lsn(0)) because there's no API for it
+            Lsn(0) => format!("basebackup {} {}", spec.tenant_id, spec.timeline_id),
+            _ => format!(
+                "basebackup {} {} {} --gzip",
+                spec.tenant_id, spec.timeline_id, lsn
+            ),
         };
+
         let copyreader = client.copy_out(basebackup_cmd.as_str())?;
+        let mut measured_reader = MeasuredReader::new(copyreader);
+
+        // Check the magic number to see if it's a gzip or not. Even though
+        // we might explicitly ask for gzip, an old pageserver with no implementation
+        // of gzip compression might send us uncompressed data. After some time
+        // passes we can assume all pageservers know how to compress and we can
+        // delete this check.
+        //
+        // If the data is not gzip, it will be tar. It will not be mistakenly
+        // recognized as gzip because tar starts with an ascii encoding of a filename,
+        // and 0x1f and 0x8b are unlikely first characters for any filename. Moreover,
+        // we send the "global" directory first from the pageserver, so it definitely
+        // won't be recognized as gzip.
+        let mut bufreader = std::io::BufReader::new(&mut measured_reader);
+        let gzip = {
+            let peek = bufreader.fill_buf().unwrap();
+            peek[0] == 0x1f && peek[1] == 0x8b
+        };
 
         // Read the archive directly from the `CopyOutReader`
         //
         // Set `ignore_zeros` so that unpack() reads all the Copy data and
         // doesn't stop at the end-of-archive marker. Otherwise, if the server
         // sends an Error after finishing the tarball, we will not notice it.
-        let mut ar = tar::Archive::new(copyreader);
-        ar.set_ignore_zeros(true);
-        ar.unpack(&self.pgdata)?;
+        if gzip {
+            let mut ar = tar::Archive::new(flate2::read::GzDecoder::new(&mut bufreader));
+            ar.set_ignore_zeros(true);
+            ar.unpack(&self.pgdata)?;
+        } else {
+            let mut ar = tar::Archive::new(&mut bufreader);
+            ar.set_ignore_zeros(true);
+            ar.unpack(&self.pgdata)?;
+        };
 
+        // Report metrics
+        self.state.lock().unwrap().metrics.basebackup_bytes =
+            measured_reader.get_byte_count() as u64;
         self.state.lock().unwrap().metrics.basebackup_ms = Utc::now()
             .signed_duration_since(start_time)
             .to_std()
@@ -199,8 +321,8 @@ impl ComputeNode {
 
     // Run `postgres` in a special mode with `--sync-safekeepers` argument
     // and return the reported LSN back to the caller.
-    #[instrument(skip(self, storage_auth_token))]
-    fn sync_safekeepers(&self, storage_auth_token: Option<String>) -> Result<Lsn> {
+    #[instrument(skip_all)]
+    pub fn sync_safekeepers(&self, storage_auth_token: Option<String>) -> Result<Lsn> {
         let start_time = Utc::now();
 
         let sync_handle = Command::new(&self.pgbin)
@@ -244,15 +366,23 @@ impl ComputeNode {
 
     /// Do all the preparations like PGDATA directory creation, configuration,
     /// safekeepers sync, basebackup, etc.
-    #[instrument(skip(self, compute_state))]
-    pub fn prepare_pgdata(&self, compute_state: &ComputeState) -> Result<()> {
+    #[instrument(skip_all)]
+    pub fn prepare_pgdata(
+        &self,
+        compute_state: &ComputeState,
+        extension_server_port: u16,
+    ) -> Result<()> {
         let pspec = compute_state.pspec.as_ref().expect("spec must be set");
         let spec = &pspec.spec;
         let pgdata_path = Path::new(&self.pgdata);
 
         // Remove/create an empty pgdata directory and put configuration there.
         self.create_pgdata()?;
-        config::write_postgres_conf(&pgdata_path.join("postgresql.conf"), &pspec.spec)?;
+        config::write_postgres_conf(
+            &pgdata_path.join("postgresql.conf"),
+            &pspec.spec,
+            Some(extension_server_port),
+        )?;
 
         // Syncing safekeepers is only safe with primary nodes: if a primary
         // is already connected it will be kicked out, so a secondary (standby)
@@ -302,7 +432,7 @@ impl ComputeNode {
 
     /// Start Postgres as a child process and manage DBs/roles.
     /// After that this will hang waiting on the postmaster process to exit.
-    #[instrument(skip(self))]
+    #[instrument(skip_all)]
     pub fn start_postgres(
         &self,
         storage_auth_token: Option<String>,
@@ -326,7 +456,7 @@ impl ComputeNode {
     }
 
     /// Do initial configuration of the already started Postgres.
-    #[instrument(skip(self, compute_state))]
+    #[instrument(skip_all)]
     pub fn apply_config(&self, compute_state: &ComputeState) -> Result<()> {
         // If connection fails,
         // it may be the old node with `zenith_admin` superuser.
@@ -347,6 +477,8 @@ impl ComputeNode {
                     .map_err(|_| anyhow::anyhow!("invalid connstr"))?;
 
                 let mut client = Client::connect(zenith_admin_connstr.as_str(), NoTls)?;
+                // Disable forwarding so that users don't get a cloud_admin role
+                client.simple_query("SET neon.forward_ddl = false")?;
                 client.simple_query("CREATE USER cloud_admin WITH SUPERUSER")?;
                 client.simple_query("GRANT zenith_admin TO cloud_admin")?;
                 drop(client);
@@ -357,14 +489,16 @@ impl ComputeNode {
             Ok(client) => client,
         };
 
-        // Proceed with post-startup configuration. Note, that order of operations is important.
         // Disable DDL forwarding because control plane already knows about these roles/databases.
         client.simple_query("SET neon.forward_ddl = false")?;
+
+        // Proceed with post-startup configuration. Note, that order of operations is important.
         let spec = &compute_state.pspec.as_ref().expect("spec must be set").spec;
+        create_neon_superuser(spec, &mut client)?;
         handle_roles(spec, &mut client)?;
         handle_databases(spec, &mut client)?;
         handle_role_deletions(spec, self.connstr.as_str(), &mut client)?;
-        handle_grants(spec, self.connstr.as_str(), &mut client)?;
+        handle_grants(spec, self.connstr.as_str())?;
         handle_extensions(spec, &mut client)?;
 
         // 'Close' connection
@@ -376,7 +510,7 @@ impl ComputeNode {
     // We could've wrapped this around `pg_ctl reload`, but right now we don't use
     // `pg_ctl` for start / stop, so this just seems much easier to do as we already
     // have opened connection to Postgres and superuser access.
-    #[instrument(skip(self, client))]
+    #[instrument(skip_all)]
     fn pg_reload_conf(&self, client: &mut Client) -> Result<()> {
         client.simple_query("SELECT pg_reload_conf()")?;
         Ok(())
@@ -384,13 +518,13 @@ impl ComputeNode {
 
     /// Similar to `apply_config()`, but does a bit different sequence of operations,
     /// as it's used to reconfigure a previously started and configured Postgres node.
-    #[instrument(skip(self))]
+    #[instrument(skip_all)]
     pub fn reconfigure(&self) -> Result<()> {
         let spec = self.state.lock().unwrap().pspec.clone().unwrap().spec;
 
         // Write new config
         let pgdata_path = Path::new(&self.pgdata);
-        config::write_postgres_conf(&pgdata_path.join("postgresql.conf"), &spec)?;
+        config::write_postgres_conf(&pgdata_path.join("postgresql.conf"), &spec, None)?;
 
         let mut client = Client::connect(self.connstr.as_str(), NoTls)?;
         self.pg_reload_conf(&mut client)?;
@@ -402,7 +536,7 @@ impl ComputeNode {
             handle_roles(&spec, &mut client)?;
             handle_databases(&spec, &mut client)?;
             handle_role_deletions(&spec, self.connstr.as_str(), &mut client)?;
-            handle_grants(&spec, self.connstr.as_str(), &mut client)?;
+            handle_grants(&spec, self.connstr.as_str())?;
             handle_extensions(&spec, &mut client)?;
         }
 
@@ -419,8 +553,8 @@ impl ComputeNode {
         Ok(())
     }
 
-    #[instrument(skip(self))]
-    pub fn start_compute(&self) -> Result<std::process::Child> {
+    #[instrument(skip_all)]
+    pub fn start_compute(&self, extension_server_port: u16) -> Result<std::process::Child> {
         let compute_state = self.state.lock().unwrap().clone();
         let pspec = compute_state.pspec.as_ref().expect("spec must be set");
         info!(
@@ -431,12 +565,31 @@ impl ComputeNode {
             pspec.timeline_id,
         );
 
-        self.prepare_pgdata(&compute_state)?;
+        // This part is sync, because we need to download
+        // remote shared_preload_libraries before postgres start (if any)
+        {
+            let library_load_start_time = Utc::now();
+            self.prepare_preload_libraries(&compute_state)?;
+
+            let library_load_time = Utc::now()
+                .signed_duration_since(library_load_start_time)
+                .to_std()
+                .unwrap()
+                .as_millis() as u64;
+            let mut state = self.state.lock().unwrap();
+            state.metrics.load_libraries_ms = library_load_time;
+            info!(
+                "Loading shared_preload_libraries took {:?}ms",
+                library_load_time
+            );
+        }
+
+        self.prepare_pgdata(&compute_state, extension_server_port)?;
 
         let start_time = Utc::now();
-
         let pg = self.start_postgres(pspec.storage_auth_token.clone())?;
 
+        let config_time = Utc::now();
         if pspec.spec.mode == ComputeMode::Primary && !pspec.spec.skip_pg_catalog_updates {
             self.apply_config(&compute_state)?;
         }
@@ -444,11 +597,16 @@ impl ComputeNode {
         let startup_end_time = Utc::now();
         {
             let mut state = self.state.lock().unwrap();
-            state.metrics.config_ms = startup_end_time
+            state.metrics.start_postgres_ms = config_time
                 .signed_duration_since(start_time)
                 .to_std()
                 .unwrap()
                 .as_millis() as u64;
+            state.metrics.config_ms = startup_end_time
+                .signed_duration_since(config_time)
+                .to_std()
+                .unwrap()
+                .as_millis() as u64;
             state.metrics.total_startup_ms = startup_end_time
                 .signed_duration_since(compute_state.start_time)
                 .to_std()
@@ -462,6 +620,13 @@ impl ComputeNode {
             pspec.spec.cluster.cluster_id.as_deref().unwrap_or("None")
         );
 
+        // Log metrics so that we can search for slow operations in logs
+        let metrics = {
+            let state = self.state.lock().unwrap();
+            state.metrics.clone()
+        };
+        info!(?metrics, "compute start finished");
+
         Ok(pg)
     }
 
@@ -567,4 +732,92 @@ LIMIT 100",
             "{{\"pg_stat_statements\": []}}".to_string()
         }
     }
+
+    // If remote extension storage is configured,
+    // download extension control files
+    #[tokio::main]
+    pub async fn prepare_external_extensions(&self, compute_state: &ComputeState) -> Result<()> {
+        if let Some(ref ext_remote_storage) = self.ext_remote_storage {
+            let pspec = compute_state.pspec.as_ref().expect("spec must be set");
+            let spec = &pspec.spec;
+            let custom_ext_prefixes = spec.custom_extensions.clone().unwrap_or(Vec::new());
+            info!("custom_ext_prefixes: {:?}", &custom_ext_prefixes);
+            let available_extensions = extension_server::get_available_extensions(
+                ext_remote_storage,
+                &self.pgbin,
+                &self.pgversion,
+                &custom_ext_prefixes,
+            )
+            .await?;
+            self.available_extensions
+                .set(available_extensions)
+                .expect("available_extensions.set error");
+        }
+        Ok(())
+    }
+
+    pub async fn download_extension(&self, ext_name: &str) -> Result<()> {
+        match &self.ext_remote_storage {
+            None => anyhow::bail!("No remote extension storage"),
+            Some(remote_storage) => {
+                extension_server::download_extension(
+                    ext_name,
+                    remote_storage,
+                    &self.pgbin,
+                    &self.pgversion,
+                )
+                .await
+            }
+        }
+    }
+
+    #[tokio::main]
+    pub async fn prepare_preload_libraries(&self, compute_state: &ComputeState) -> Result<()> {
+        if self.ext_remote_storage.is_none() {
+            return Ok(());
+        }
+        let pspec = compute_state.pspec.as_ref().expect("spec must be set");
+        let spec = &pspec.spec;
+
+        info!("parse shared_preload_libraries from spec.cluster.settings");
+        let mut libs_vec = Vec::new();
+        if let Some(libs) = spec.cluster.settings.find("shared_preload_libraries") {
+            libs_vec = libs
+                .split(&[',', '\'', ' '])
+                .filter(|s| *s != "neon" && !s.is_empty())
+                .map(str::to_string)
+                .collect();
+        }
+        info!("parse shared_preload_libraries from provided postgresql.conf");
+        // that is used in neon_local and python tests
+        if let Some(conf) = &spec.cluster.postgresql_conf {
+            let conf_lines = conf.split('\n').collect::<Vec<&str>>();
+            let mut shared_preload_libraries_line = "";
+            for line in conf_lines {
+                if line.starts_with("shared_preload_libraries") {
+                    shared_preload_libraries_line = line;
+                }
+            }
+            let mut preload_libs_vec = Vec::new();
+            if let Some(libs) = shared_preload_libraries_line.split("='").nth(1) {
+                preload_libs_vec = libs
+                    .split(&[',', '\'', ' '])
+                    .filter(|s| *s != "neon" && !s.is_empty())
+                    .map(str::to_string)
+                    .collect();
+            }
+            libs_vec.extend(preload_libs_vec);
+        }
+
+        info!("Downloading to shared preload libraries: {:?}", &libs_vec);
+        let mut download_tasks = Vec::new();
+        for library in &libs_vec {
+            download_tasks.push(self.download_extension(library));
+        }
+        let results = join_all(download_tasks).await;
+        for result in results {
+            result?; // propogate any errors
+        }
+        Ok(())
+    }
 }

compute_tools/src/config.rs

@@ -33,7 +33,11 @@ pub fn line_in_file(path: &Path, line: &str) -> Result<bool> {
 }
 
 /// Create or completely rewrite configuration file specified by `path`
-pub fn write_postgres_conf(path: &Path, spec: &ComputeSpec) -> Result<()> {
+pub fn write_postgres_conf(
+    path: &Path,
+    spec: &ComputeSpec,
+    extension_server_port: Option<u16>,
+) -> Result<()> {
     // File::create() destroys the file content if it exists.
     let mut file = File::create(path)?;
 
@@ -47,30 +51,22 @@ pub fn write_postgres_conf(path: &Path, spec: &ComputeSpec) -> Result<()> {
     // Add options for connecting to storage
     writeln!(file, "# Neon storage settings")?;
     if let Some(s) = &spec.pageserver_connstring {
-        writeln!(
-            file,
-            "neon.pageserver_connstring='{}'",
-            escape_conf_value(s)
-        )?;
+        writeln!(file, "neon.pageserver_connstring={}", escape_conf_value(s))?;
     }
     if !spec.safekeeper_connstrings.is_empty() {
         writeln!(
             file,
-            "neon.safekeepers='{}'",
+            "neon.safekeepers={}",
             escape_conf_value(&spec.safekeeper_connstrings.join(","))
         )?;
     }
     if let Some(s) = &spec.tenant_id {
-        writeln!(
-            file,
-            "neon.tenant_id='{}'",
-            escape_conf_value(&s.to_string())
-        )?;
+        writeln!(file, "neon.tenant_id={}", escape_conf_value(&s.to_string()))?;
     }
     if let Some(s) = &spec.timeline_id {
         writeln!(
             file,
-            "neon.timeline_id='{}'",
+            "neon.timeline_id={}",
             escape_conf_value(&s.to_string())
         )?;
     }
@@ -95,5 +91,9 @@ pub fn write_postgres_conf(path: &Path, spec: &ComputeSpec) -> Result<()> {
         writeln!(file, "# Managed by compute_ctl: end")?;
     }
 
+    if let Some(port) = extension_server_port {
+        writeln!(file, "neon.extension_server_port={}", port)?;
+    }
+
     Ok(())
 }

compute_tools/src/configurator.rs

@@ -8,7 +8,7 @@ use compute_api::responses::ComputeStatus;
 
 use crate::compute::ComputeNode;
 
-#[instrument(skip(compute))]
+#[instrument(skip_all)]
 fn configurator_main_loop(compute: &Arc<ComputeNode>) {
     info!("waiting for reconfiguration requests");
     loop {
@@ -42,13 +42,15 @@ fn configurator_main_loop(compute: &Arc<ComputeNode>) {
     }
 }
 
-pub fn launch_configurator(compute: &Arc<ComputeNode>) -> Result<thread::JoinHandle<()>> {
+pub fn launch_configurator(
+    compute: &Arc<ComputeNode>,
+) -> Result<thread::JoinHandle<()>, std::io::Error> {
     let compute = Arc::clone(compute);
 
-    Ok(thread::Builder::new()
+    thread::Builder::new()
         .name("compute-configurator".into())
         .spawn(move || {
             configurator_main_loop(&compute);
             info!("configurator thread is exited");
-        })?)
+        })
 }

compute_tools/src/extension_server.rs

@@ -0,0 +1,237 @@
+// Download extension files from the extension store
+// and put them in the right place in the postgres directory
+/*
+The layout of the S3 bucket is as follows:
+
+v14/ext_index.json
+    -- this contains information necessary to create control files
+v14/extensions/test_ext1.tar.gz
+    -- this contains the library files and sql files necessary to create this extension
+v14/extensions/custom_ext1.tar.gz
+
+The difference between a private and public extensions is determined by who can
+load the extension this is specified in ext_index.json
+
+Speicially, ext_index.json has a list of public extensions, and a list of
+extensions enabled for specific tenant-ids.
+*/
+use crate::compute::ComputeNode;
+use anyhow::Context;
+use anyhow::{self, Result};
+use flate2::read::GzDecoder;
+use remote_storage::*;
+use serde_json::{self, Value};
+use std::collections::HashSet;
+use std::num::{NonZeroU32, NonZeroUsize};
+use std::path::Path;
+use std::str;
+use std::sync::Arc;
+use std::thread;
+use tar::Archive;
+use tokio::io::AsyncReadExt;
+use tracing::info;
+
+fn get_pg_config(argument: &str, pgbin: &str) -> String {
+    // gives the result of `pg_config [argument]`
+    // where argument is a flag like `--version` or `--sharedir`
+    let pgconfig = pgbin
+        .strip_suffix("postgres")
+        .expect("bad pgbin")
+        .to_owned()
+        + "/pg_config";
+    let config_output = std::process::Command::new(pgconfig)
+        .arg(argument)
+        .output()
+        .expect("pg_config error");
+    std::str::from_utf8(&config_output.stdout)
+        .expect("pg_config error")
+        .trim()
+        .to_string()
+}
+
+pub fn get_pg_version(pgbin: &str) -> String {
+    // pg_config --version returns a (platform specific) human readable string
+    // such as "PostgreSQL 15.4". We parse this to v14/v15
+    let human_version = get_pg_config("--version", pgbin);
+    if human_version.contains("15") {
+        return "v15".to_string();
+    } else if human_version.contains("14") {
+        return "v14".to_string();
+    }
+    panic!("Unsuported postgres version {human_version}");
+}
+
+// download extension control files
+// if custom_ext_prefixes is provided - search also in custom extension paths
+pub async fn get_available_extensions(
+    remote_storage: &GenericRemoteStorage,
+    pgbin: &str,
+    pg_version: &str,
+    custom_ext_prefixes: &[String],
+) -> Result<HashSet<String>> {
+    let local_sharedir = Path::new(&get_pg_config("--sharedir", pgbin)).join("extension");
+    let index_path = pg_version.to_owned() + "/ext_index.json";
+    let index_path = RemotePath::new(Path::new(&index_path)).context("error forming path")?;
+    info!("download ext_index.json: {:?}", &index_path);
+
+    // TODO: potential optimization: cache ext_index.json
+    let mut download = remote_storage.download(&index_path).await?;
+    let mut write_data_buffer = Vec::new();
+    download
+        .download_stream
+        .read_to_end(&mut write_data_buffer)
+        .await?;
+    let ext_index_str = match str::from_utf8(&write_data_buffer) {
+        Ok(v) => v,
+        Err(e) => panic!("Invalid UTF-8 sequence: {}", e),
+    };
+
+    let ext_index_full: Value = serde_json::from_str(ext_index_str)?;
+    let ext_index_full = ext_index_full.as_object().context("error parsing json")?;
+    let control_data = ext_index_full["control_data"]
+        .as_object()
+        .context("json parse error")?;
+    let enabled_extensions = ext_index_full["enabled_extensions"]
+        .as_object()
+        .context("json parse error")?;
+    info!("{:?}", control_data.clone());
+    info!("{:?}", enabled_extensions.clone());
+
+    let mut prefixes = vec!["public".to_string()];
+    prefixes.extend(custom_ext_prefixes.to_owned());
+    info!("{:?}", &prefixes);
+    let mut all_extensions = HashSet::new();
+    for prefix in prefixes {
+        let prefix_extensions = match enabled_extensions.get(&prefix) {
+            Some(Value::Array(ext_name)) => ext_name,
+            _ => {
+                info!("prefix {} has no extensions", prefix);
+                continue;
+            }
+        };
+        info!("{:?}", prefix_extensions);
+        for ext_name in prefix_extensions {
+            all_extensions.insert(ext_name.as_str().context("json parse error")?.to_string());
+        }
+    }
+
+    for prefix in &all_extensions {
+        let control_contents = control_data[prefix].as_str().context("json parse error")?;
+        let control_path = local_sharedir.join(prefix.to_owned() + ".control");
+
+        info!("WRITING FILE {:?}{:?}", control_path, control_contents);
+        std::fs::write(control_path, control_contents)?;
+    }
+
+    Ok(all_extensions.into_iter().collect())
+}
+
+// download all sqlfiles (and possibly data files) for a given extension name
+pub async fn download_extension(
+    ext_name: &str,
+    remote_storage: &GenericRemoteStorage,
+    pgbin: &str,
+    pg_version: &str,
+) -> Result<()> {
+    // TODO: potential optimization: only download the extension if it doesn't exist
+    // problem: how would we tell if it exists?
+    let ext_name = ext_name.replace(".so", "");
+    let ext_name_targz = ext_name.to_owned() + ".tar.gz";
+    if Path::new(&ext_name_targz).exists() {
+        info!("extension {:?} already exists", ext_name_targz);
+        return Ok(());
+    }
+    let ext_path = RemotePath::new(
+        &Path::new(pg_version)
+            .join("extensions")
+            .join(ext_name_targz.clone()),
+    )?;
+    info!(
+        "Start downloading extension {:?} from {:?}",
+        ext_name, ext_path
+    );
+    let mut download = remote_storage.download(&ext_path).await?;
+    let mut write_data_buffer = Vec::new();
+    download
+        .download_stream
+        .read_to_end(&mut write_data_buffer)
+        .await?;
+    let unzip_dest = pgbin.strip_suffix("/bin/postgres").expect("bad pgbin");
+    let tar = GzDecoder::new(write_data_buffer.as_slice());
+    let mut archive = Archive::new(tar);
+    archive.unpack(unzip_dest)?;
+    info!("Download + unzip {:?} completed successfully", &ext_path);
+
+    let local_sharedir = Path::new(&get_pg_config("--sharedir", pgbin)).join("extension");
+    let zip_sharedir = format!("{unzip_dest}/extensions/{ext_name}/share/extension");
+    info!("mv {zip_sharedir:?}/* {local_sharedir:?}");
+    for file in std::fs::read_dir(zip_sharedir)? {
+        let old_file = file?.path();
+        let new_file =
+            Path::new(&local_sharedir).join(old_file.file_name().context("error parsing file")?);
+        std::fs::rename(old_file, new_file)?;
+    }
+    let local_libdir = Path::new(&get_pg_config("--libdir", pgbin)).join("postgresql");
+    let zip_libdir = format!("{unzip_dest}/extensions/{ext_name}/lib");
+    info!("mv {zip_libdir:?}/* {local_libdir:?}");
+    for file in std::fs::read_dir(zip_libdir)? {
+        let old_file = file?.path();
+        let new_file =
+            Path::new(&local_libdir).join(old_file.file_name().context("error parsing file")?);
+        std::fs::rename(old_file, new_file)?;
+    }
+    Ok(())
+}
+
+// This function initializes the necessary structs to use remmote storage (should be fairly cheap)
+pub fn init_remote_storage(
+    remote_ext_config: &str,
+    default_prefix: &str,
+) -> anyhow::Result<GenericRemoteStorage> {
+    let remote_ext_config: serde_json::Value = serde_json::from_str(remote_ext_config)?;
+
+    let remote_ext_bucket = remote_ext_config["bucket"]
+        .as_str()
+        .context("config parse error")?;
+    let remote_ext_region = remote_ext_config["region"]
+        .as_str()
+        .context("config parse error")?;
+    let remote_ext_endpoint = remote_ext_config["endpoint"].as_str();
+    let remote_ext_prefix = remote_ext_config["prefix"]
+        .as_str()
+        .unwrap_or(default_prefix)
+        .to_string();
+
+    // TODO: potentially allow modification of other parameters
+    // however, default values should be fine for now
+    let config = S3Config {
+        bucket_name: remote_ext_bucket.to_string(),
+        bucket_region: remote_ext_region.to_string(),
+        prefix_in_bucket: Some(remote_ext_prefix),
+        endpoint: remote_ext_endpoint.map(|x| x.to_string()),
+        concurrency_limit: NonZeroUsize::new(100).expect("100 != 0"),
+        max_keys_per_list_response: None,
+    };
+    let config = RemoteStorageConfig {
+        max_concurrent_syncs: NonZeroUsize::new(100).expect("100 != 0"),
+        max_sync_errors: NonZeroU32::new(100).expect("100 != 0"),
+        storage: RemoteStorageKind::AwsS3(config),
+    };
+    GenericRemoteStorage::from_config(&config)
+}
+
+pub fn launch_download_extensions(
+    compute: &Arc<ComputeNode>,
+) -> Result<thread::JoinHandle<()>, std::io::Error> {
+    let compute = Arc::clone(compute);
+    thread::Builder::new()
+        .name("download-extensions".into())
+        .spawn(move || {
+            info!("start download_extension_files");
+            let compute_state = compute.state.lock().expect("error unlocking compute.state");
+            compute
+                .prepare_external_extensions(&compute_state)
+                .expect("error preparing extensions");
+            info!("download_extension_files done, exiting thread");
+        })
+}

compute_tools/src/http/api.rs

@@ -121,6 +121,27 @@ async fn routes(req: Request<Body>, compute: &Arc<ComputeNode>) -> Response<Body
             }
         }
 
+        // download extension files from S3 on demand
+        (&Method::POST, route) if route.starts_with("/extension_server/") => {
+            info!("serving {:?} POST request", route);
+            info!("req.uri {:?}", req.uri());
+            let filename = route.split('/').last().unwrap().to_string();
+            info!(
+                "serving /extension_server POST request, filename: {:?}",
+                &filename
+            );
+
+            match compute.download_extension(&filename).await {
+                Ok(_) => Response::new(Body::from("OK")),
+                Err(e) => {
+                    error!("extension download failed: {}", e);
+                    let mut resp = Response::new(Body::from(e.to_string()));
+                    *resp.status_mut() = StatusCode::INTERNAL_SERVER_ERROR;
+                    resp
+                }
+            }
+        }
+
         // Return the `404 Not Found` for any other routes.
         _ => {
             let mut not_found = Response::new(Body::from("404 Not Found"));

compute_tools/src/http/openapi_spec.yaml

@@ -139,6 +139,34 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/GenericError"
+  /extension_server:
+    post:
+      tags:
+      - Extension
+      summary: Download extension from S3 to local folder.
+      description: ""
+      operationId: downloadExtension
+      responses:
+        200:
+          description: Extension downloaded
+          content:
+            text/plain:
+              schema:
+                type: string
+                description: Error text or 'OK' if download succeeded.
+                example: "OK"
+        400:
+        description: Request is invalid.
+        content:
+          application/json:
+            schema:
+              $ref: "#/components/schemas/GenericError"
+        500:
+        description: Extension download request failed.
+        content:
+          application/json:
+            schema:
+              $ref: "#/components/schemas/GenericError"
 
 components:
   securitySchemes:

compute_tools/src/lib.rs

@@ -9,6 +9,7 @@ pub mod http;
 #[macro_use]
 pub mod logger;
 pub mod compute;
+pub mod extension_server;
 pub mod monitor;
 pub mod params;
 pub mod pg_helpers;

compute_tools/src/logger.rs

@@ -18,6 +18,7 @@ pub fn init_tracing_and_logging(default_log_level: &str) -> anyhow::Result<()> {
         .unwrap_or_else(|_| tracing_subscriber::EnvFilter::new(default_log_level));
 
     let fmt_layer = tracing_subscriber::fmt::layer()
+        .with_ansi(false)
         .with_target(false)
         .with_writer(std::io::stderr);
 

compute_tools/src/monitor.rs

@@ -105,10 +105,10 @@ fn watch_compute_activity(compute: &ComputeNode) {
 }
 
 /// Launch a separate compute monitor thread and return its `JoinHandle`.
-pub fn launch_monitor(state: &Arc<ComputeNode>) -> Result<thread::JoinHandle<()>> {
+pub fn launch_monitor(state: &Arc<ComputeNode>) -> Result<thread::JoinHandle<()>, std::io::Error> {
     let state = Arc::clone(state);
 
-    Ok(thread::Builder::new()
+    thread::Builder::new()
         .name("compute-monitor".into())
-        .spawn(move || watch_compute_activity(&state))?)
+        .spawn(move || watch_compute_activity(&state))
 }

compute_tools/src/pg_helpers.rs

@@ -16,15 +16,26 @@ use compute_api::spec::{Database, GenericOption, GenericOptions, PgIdent, Role};
 
 const POSTGRES_WAIT_TIMEOUT: Duration = Duration::from_millis(60 * 1000); // milliseconds
 
-/// Escape a string for including it in a SQL literal
-fn escape_literal(s: &str) -> String {
-    s.replace('\'', "''").replace('\\', "\\\\")
+/// Escape a string for including it in a SQL literal. Wrapping the result
+/// with `E'{}'` or `'{}'` is not required, as it returns a ready-to-use
+/// SQL string literal, e.g. `'db'''` or `E'db\\'`.
+/// See https://github.com/postgres/postgres/blob/da98d005cdbcd45af563d0c4ac86d0e9772cd15f/src/backend/utils/adt/quote.c#L47
+/// for the original implementation.
+pub fn escape_literal(s: &str) -> String {
+    let res = s.replace('\'', "''").replace('\\', "\\\\");
+
+    if res.contains('\\') {
+        format!("E'{}'", res)
+    } else {
+        format!("'{}'", res)
+    }
 }
 
-/// Escape a string so that it can be used in postgresql.conf.
-/// Same as escape_literal, currently.
+/// Escape a string so that it can be used in postgresql.conf. Wrapping the result
+/// with `'{}'` is not required, as it returns a ready-to-use config string.
 pub fn escape_conf_value(s: &str) -> String {
-    s.replace('\'', "''").replace('\\', "\\\\")
+    let res = s.replace('\'', "''").replace('\\', "\\\\");
+    format!("'{}'", res)
 }
 
 trait GenericOptionExt {
@@ -37,7 +48,7 @@ impl GenericOptionExt for GenericOption {
     fn to_pg_option(&self) -> String {
         if let Some(val) = &self.value {
             match self.vartype.as_ref() {
-                "string" => format!("{} '{}'", self.name, escape_literal(val)),
+                "string" => format!("{} {}", self.name, escape_literal(val)),
                 _ => format!("{} {}", self.name, val),
             }
         } else {
@@ -49,7 +60,7 @@ impl GenericOptionExt for GenericOption {
     fn to_pg_setting(&self) -> String {
         if let Some(val) = &self.value {
             match self.vartype.as_ref() {
-                "string" => format!("{} = '{}'", self.name, escape_conf_value(val)),
+                "string" => format!("{} = {}", self.name, escape_conf_value(val)),
                 _ => format!("{} = {}", self.name, val),
             }
         } else {
@@ -215,7 +226,7 @@ pub fn get_existing_dbs(client: &mut Client) -> Result<Vec<Database>> {
 /// Wait for Postgres to become ready to accept connections. It's ready to
 /// accept connections when the state-field in `pgdata/postmaster.pid` says
 /// 'ready'.
-#[instrument(skip(pg))]
+#[instrument(skip_all, fields(pgdata = %pgdata.display()))]
 pub fn wait_for_postgres(pg: &mut Child, pgdata: &Path) -> Result<()> {
     let pid_path = pgdata.join("postmaster.pid");
 

compute_tools/src/spec.rs

@@ -124,7 +124,7 @@ pub fn get_spec_from_control_plane(
 pub fn handle_configuration(spec: &ComputeSpec, pgdata_path: &Path) -> Result<()> {
     // File `postgresql.conf` is no longer included into `basebackup`, so just
     // always write all config into it creating new file.
-    config::write_postgres_conf(&pgdata_path.join("postgresql.conf"), spec)?;
+    config::write_postgres_conf(&pgdata_path.join("postgresql.conf"), spec, None)?;
 
     update_pg_hba(pgdata_path)?;
 
@@ -269,17 +269,13 @@ pub fn handle_roles(spec: &ComputeSpec, client: &mut Client) -> Result<()> {
                 xact.execute(query.as_str(), &[])?;
             }
             RoleAction::Create => {
-                let mut query: String = format!("CREATE ROLE {} ", name.pg_quote());
+                let mut query: String = format!(
+                    "CREATE ROLE {} CREATEROLE CREATEDB IN ROLE neon_superuser",
+                    name.pg_quote()
+                );
                 info!("role create query: '{}'", &query);
                 query.push_str(&role.to_pg_options());
                 xact.execute(query.as_str(), &[])?;
-
-                let grant_query = format!(
-                    "GRANT pg_read_all_data, pg_write_all_data TO {}",
-                    name.pg_quote()
-                );
-                xact.execute(grant_query.as_str(), &[])?;
-                info!("role grant query: '{}'", &grant_query);
             }
         }
 
@@ -401,10 +397,44 @@ pub fn handle_databases(spec: &ComputeSpec, client: &mut Client) -> Result<()> {
                 // We do not check either DB exists or not,
                 // Postgres will take care of it for us
                 "delete_db" => {
-                    let query: String = format!("DROP DATABASE IF EXISTS {}", &op.name.pg_quote());
+                    // In Postgres we can't drop a database if it is a template.
+                    // So we need to unset the template flag first, but it could
+                    // be a retry, so we could've already dropped the database.
+                    // Check that database exists first to make it idempotent.
+                    let unset_template_query: String = format!(
+                        "
+                        DO $$
+                        BEGIN
+                            IF EXISTS(
+                                SELECT 1
+                                FROM pg_catalog.pg_database
+                                WHERE datname = {}
+                            )
+                            THEN
+                            ALTER DATABASE {} is_template false;
+                            END IF;
+                        END
+                        $$;",
+                        escape_literal(&op.name),
+                        &op.name.pg_quote()
+                    );
+                    // Use FORCE to drop database even if there are active connections.
+                    // We run this from `cloud_admin`, so it should have enough privileges.
+                    // NB: there could be other db states, which prevent us from dropping
+                    // the database. For example, if db is used by any active subscription
+                    // or replication slot.
+                    // TODO: deal with it once we allow logical replication. Proper fix should
+                    // involve returning an error code to the control plane, so it could
+                    // figure out that this is a non-retryable error, return it to the user
+                    // and fail operation permanently.
+                    let drop_db_query: String = format!(
+                        "DROP DATABASE IF EXISTS {} WITH (FORCE)",
+                        &op.name.pg_quote()
+                    );
 
                     warn!("deleting database '{}'", &op.name);
-                    client.execute(query.as_str(), &[])?;
+                    client.execute(unset_template_query.as_str(), &[])?;
+                    client.execute(drop_db_query.as_str(), &[])?;
                 }
                 "rename_db" => {
                     let new_name = op.new_name.as_ref().unwrap();
@@ -476,6 +506,11 @@ pub fn handle_databases(spec: &ComputeSpec, client: &mut Client) -> Result<()> {
                 query.push_str(&db.to_pg_options());
                 let _guard = info_span!("executing", query).entered();
                 client.execute(query.as_str(), &[])?;
+                let grant_query: String = format!(
+                    "GRANT ALL PRIVILEGES ON DATABASE {} TO neon_superuser",
+                    name.pg_quote()
+                );
+                client.execute(grant_query.as_str(), &[])?;
             }
         };
 
@@ -495,35 +530,9 @@ pub fn handle_databases(spec: &ComputeSpec, client: &mut Client) -> Result<()> {
 /// Grant CREATE ON DATABASE to the database owner and do some other alters and grants
 /// to allow users creating trusted extensions and re-creating `public` schema, for example.
 #[instrument(skip_all)]
-pub fn handle_grants(spec: &ComputeSpec, connstr: &str, client: &mut Client) -> Result<()> {
+pub fn handle_grants(spec: &ComputeSpec, connstr: &str) -> Result<()> {
     info!("cluster spec grants:");
 
-    // We now have a separate `web_access` role to connect to the database
-    // via the web interface and proxy link auth. And also we grant a
-    // read / write all data privilege to every role. So also grant
-    // create to everyone.
-    // XXX: later we should stop messing with Postgres ACL in such horrible
-    // ways.
-    let roles = spec
-        .cluster
-        .roles
-        .iter()
-        .map(|r| r.name.pg_quote())
-        .collect::<Vec<_>>();
-
-    for db in &spec.cluster.databases {
-        let dbname = &db.name;
-
-        let query: String = format!(
-            "GRANT CREATE ON DATABASE {} TO {}",
-            dbname.pg_quote(),
-            roles.join(", ")
-        );
-        info!("grant query {}", &query);
-
-        client.execute(query.as_str(), &[])?;
-    }
-
     // Do some per-database access adjustments. We'd better do this at db creation time,
     // but CREATE DATABASE isn't transactional. So we cannot create db + do some grants
     // atomically.

compute_tools/tests/pg_helpers_tests.rs

@@ -89,4 +89,12 @@ test.escaping = 'here''s a backslash \\ and a quote '' and a double-quote " hoor
         assert_eq!(none_generic_options.find("missed_value"), None);
         assert_eq!(none_generic_options.find("invalid_value"), None);
     }
+
+    #[test]
+    fn test_escape_literal() {
+        assert_eq!(escape_literal("test"), "'test'");
+        assert_eq!(escape_literal("test'"), "'test'''");
+        assert_eq!(escape_literal("test\\'"), "E'test\\\\'''");
+        assert_eq!(escape_literal("test\\'\\'"), "E'test\\\\''\\\\'''");
+    }
 }

control_plane/Cargo.toml

@@ -32,3 +32,4 @@ utils.workspace = true
 
 compute_api.workspace = true
 workspace_hack.workspace = true
+tracing.workspace = true

control_plane/src/background_process.rs

@@ -180,6 +180,11 @@ pub fn stop_process(immediate: bool, process_name: &str, pid_file: &Path) -> any
     }
 
     // Wait until process is gone
+    wait_until_stopped(process_name, pid)?;
+    Ok(())
+}
+
+pub fn wait_until_stopped(process_name: &str, pid: Pid) -> anyhow::Result<()> {
     for retries in 0..RETRIES {
         match process_has_stopped(pid) {
             Ok(true) => {

control_plane/src/bin/neon_local.rs

@@ -308,7 +308,8 @@ fn handle_init(init_match: &ArgMatches) -> anyhow::Result<LocalEnv> {
 
     let mut env =
         LocalEnv::parse_config(&toml_file).context("Failed to create neon configuration")?;
-    env.init(pg_version)
+    let force = init_match.get_flag("force");
+    env.init(pg_version, force)
         .context("Failed to initialize neon repository")?;
 
     // Initialize pageserver, create initial tenant and timeline.
@@ -657,6 +658,8 @@ fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<(
                 .get_one::<String>("endpoint_id")
                 .ok_or_else(|| anyhow!("No endpoint ID was provided to start"))?;
 
+            let remote_ext_config = sub_args.get_one::<String>("remote-ext-config");
+
             // If --safekeepers argument is given, use only the listed safekeeper nodes.
             let safekeepers =
                 if let Some(safekeepers_str) = sub_args.get_one::<String>("safekeepers") {
@@ -698,7 +701,7 @@ fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<(
                     _ => {}
                 }
                 println!("Starting existing endpoint {endpoint_id}...");
-                endpoint.start(&auth_token, safekeepers)?;
+                endpoint.start(&auth_token, safekeepers, remote_ext_config)?;
             } else {
                 let branch_name = sub_args
                     .get_one::<String>("branch-name")
@@ -742,7 +745,7 @@ fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<(
                     pg_version,
                     mode,
                 )?;
-                ep.start(&auth_token, safekeepers)?;
+                ep.start(&auth_token, safekeepers, remote_ext_config)?;
             }
         }
         "stop" => {
@@ -1002,6 +1005,12 @@ fn cli() -> Command {
         .help("Additional pageserver's configuration options or overrides, refer to pageserver's 'config-override' CLI parameter docs for more")
         .required(false);
 
+    let remote_ext_config_args = Arg::new("remote-ext-config")
+        .long("remote-ext-config")
+        .num_args(1)
+        .help("Configure the S3 bucket that we search for extensions in.")
+        .required(false);
+
     let lsn_arg = Arg::new("lsn")
         .long("lsn")
         .help("Specify Lsn on the timeline to start from. By default, end of the timeline would be used.")
@@ -1013,6 +1022,13 @@ fn cli() -> Command {
         .help("If set, the node will be a hot replica on the specified timeline")
         .required(false);
 
+    let force_arg = Arg::new("force")
+        .value_parser(value_parser!(bool))
+        .long("force")
+        .action(ArgAction::SetTrue)
+        .help("Force initialization even if the repository is not empty")
+        .required(false);
+
     Command::new("Neon CLI")
         .arg_required_else_help(true)
         .version(GIT_VERSION)
@@ -1028,6 +1044,7 @@ fn cli() -> Command {
                         .value_name("config"),
                 )
                 .arg(pg_version_arg.clone())
+                .arg(force_arg)
         )
         .subcommand(
             Command::new("timeline")
@@ -1152,6 +1169,7 @@ fn cli() -> Command {
                     .arg(pg_version_arg)
                     .arg(hot_standby_arg)
                     .arg(safekeepers_arg)
+                    .arg(remote_ext_config_args)
                 )
                 .subcommand(
                     Command::new("stop")

control_plane/src/endpoint.rs

@@ -67,6 +67,7 @@ pub struct EndpointConf {
     pg_port: u16,
     http_port: u16,
     pg_version: u32,
+    skip_pg_catalog_updates: bool,
 }
 
 //
@@ -135,6 +136,7 @@ impl ComputeControlPlane {
             mode,
             tenant_id,
             pg_version,
+            skip_pg_catalog_updates: false,
         });
 
         ep.create_endpoint_dir()?;
@@ -148,6 +150,7 @@ impl ComputeControlPlane {
                 http_port,
                 pg_port,
                 pg_version,
+                skip_pg_catalog_updates: false,
             })?,
         )?;
         std::fs::write(
@@ -183,6 +186,9 @@ pub struct Endpoint {
     // the endpoint runs in.
     pub env: LocalEnv,
     pageserver: Arc<PageServerNode>,
+
+    // Optimizations
+    skip_pg_catalog_updates: bool,
 }
 
 impl Endpoint {
@@ -216,6 +222,7 @@ impl Endpoint {
             mode: conf.mode,
             tenant_id: conf.tenant_id,
             pg_version: conf.pg_version,
+            skip_pg_catalog_updates: conf.skip_pg_catalog_updates,
         })
     }
 
@@ -304,7 +311,7 @@ impl Endpoint {
 
                 // TODO: use future host field from safekeeper spec
                 // Pass the list of safekeepers to the replica so that it can connect to any of them,
-                // whichever is availiable.
+                // whichever is available.
                 let sk_ports = self
                     .env
                     .safekeepers
@@ -398,10 +405,25 @@ impl Endpoint {
                 String::from_utf8_lossy(&pg_ctl.stderr),
             );
         }
+
+        // Also wait for the compute_ctl process to die. It might have some cleanup
+        // work to do after postgres stops, like syncing safekeepers, etc.
+        //
+        // TODO use background_process::stop_process instead
+        let pidfile_path = self.endpoint_path().join("compute_ctl.pid");
+        let pid: u32 = std::fs::read_to_string(pidfile_path)?.parse()?;
+        let pid = nix::unistd::Pid::from_raw(pid as i32);
+        crate::background_process::wait_until_stopped("compute_ctl", pid)?;
+
         Ok(())
     }
 
-    pub fn start(&self, auth_token: &Option<String>, safekeepers: Vec<NodeId>) -> Result<()> {
+    pub fn start(
+        &self,
+        auth_token: &Option<String>,
+        safekeepers: Vec<NodeId>,
+        remote_ext_config: Option<&String>,
+    ) -> Result<()> {
         if self.status() == "running" {
             anyhow::bail!("The endpoint is already running");
         }
@@ -450,7 +472,7 @@ impl Endpoint {
 
         // Create spec file
         let spec = ComputeSpec {
-            skip_pg_catalog_updates: false,
+            skip_pg_catalog_updates: self.skip_pg_catalog_updates,
             format_version: 1.0,
             operation_uuid: None,
             cluster: Cluster {
@@ -469,6 +491,15 @@ impl Endpoint {
             pageserver_connstring: Some(pageserver_connstring),
             safekeeper_connstrings,
             storage_auth_token: auth_token.clone(),
+            // TODO FIXME: This is a hack to test custom extensions locally.
+            // In test_download_extensions, we assume that the custom extension
+            // prefix is the tenant ID. So we set it here.
+            //
+            // The proper way to implement this is to pass the custom extension
+            // in spec, but we don't have a way to do that yet in the python tests.
+            // NEW HACK: we enable the anon custom extension for everyone! this is of course just for testing
+            // how will we do it for real?
+            custom_extensions: Some(vec!["123454321".to_string(), self.tenant_id.to_string()]),
         };
         let spec_path = self.endpoint_path().join("spec.json");
         std::fs::write(spec_path, serde_json::to_string_pretty(&spec)?)?;
@@ -500,7 +531,18 @@ impl Endpoint {
             .stdin(std::process::Stdio::null())
             .stderr(logfile.try_clone()?)
             .stdout(logfile);
-        let _child = cmd.spawn()?;
+
+        if let Some(remote_ext_config) = remote_ext_config {
+            cmd.args(["--remote-ext-config", remote_ext_config]);
+        }
+
+        let child = cmd.spawn()?;
+
+        // Write down the pid so we can wait for it when we want to stop
+        // TODO use background_process::start_process instead
+        let pid = child.id();
+        let pidfile_path = self.endpoint_path().join("compute_ctl.pid");
+        std::fs::write(pidfile_path, pid.to_string())?;
 
         // Wait for it to start
         let mut attempt = 0;

control_plane/src/local_env.rs

@@ -364,7 +364,7 @@ impl LocalEnv {
     //
     // Initialize a new Neon repository
     //
-    pub fn init(&mut self, pg_version: u32) -> anyhow::Result<()> {
+    pub fn init(&mut self, pg_version: u32, force: bool) -> anyhow::Result<()> {
         // check if config already exists
         let base_path = &self.base_data_dir;
         ensure!(
@@ -372,11 +372,29 @@ impl LocalEnv {
             "repository base path is missing"
         );
 
-        ensure!(
-            !base_path.exists(),
-            "directory '{}' already exists. Perhaps already initialized?",
-            base_path.display()
-        );
+        if base_path.exists() {
+            if force {
+                println!("removing all contents of '{}'", base_path.display());
+                // instead of directly calling `remove_dir_all`, we keep the original dir but removing
+                // all contents inside. This helps if the developer symbol links another directory (i.e.,
+                // S3 local SSD) to the `.neon` base directory.
+                for entry in std::fs::read_dir(base_path)? {
+                    let entry = entry?;
+                    let path = entry.path();
+                    if path.is_dir() {
+                        fs::remove_dir_all(&path)?;
+                    } else {
+                        fs::remove_file(&path)?;
+                    }
+                }
+            } else {
+                bail!(
+                    "directory '{}' already exists. Perhaps already initialized? (Hint: use --force to remove all contents)",
+                    base_path.display()
+                );
+            }
+        }
+
         if !self.pg_bin_dir(pg_version)?.join("postgres").exists() {
             bail!(
                 "Can't find postgres binary at {}",
@@ -392,7 +410,9 @@ impl LocalEnv {
             }
         }
 
-        fs::create_dir(base_path)?;
+        if !base_path.exists() {
+            fs::create_dir(base_path)?;
+        }
 
         // Generate keypair for JWT.
         //

docker-compose/docker-compose.yml

@@ -189,7 +189,7 @@ services:
       - "/bin/bash"
       - "-c"
     command:
-      - "until pg_isready -h compute -p 55433 ; do
+      - "until pg_isready -h compute -p 55433 -U cloud_admin ; do
             echo 'Waiting to start compute...' && sleep 1;
          done"
     depends_on:

docs/docker.md

@@ -48,6 +48,7 @@ Creating docker-compose_storage_broker_1       ... done
 2. connect compute node
 ```
 $ echo "localhost:55433:postgres:cloud_admin:cloud_admin" >> ~/.pgpass
+$ chmod 600 ~/.pgpass
 $ psql -h localhost -p 55433 -U cloud_admin
 postgres=# CREATE TABLE t(key int primary key, value text);
 CREATE TABLE

docs/rfcs/024-extension-loading.md

@@ -0,0 +1,187 @@
+# Supporting custom user Extensions (Dynamic Extension Loading)
+Created 2023-05-03
+
+## Motivation
+
+There are many extensions in the PostgreSQL ecosystem, and not all extensions
+are of a quality that we can confidently support them. Additionally, our
+current extension inclusion mechanism has several problems because we build all
+extensions into the primary Compute image: We build the extensions every time
+we build the compute image regardless of whether we actually need to rebuild
+the image, and the inclusion of these extensions in the image adds a hard
+dependency on all supported extensions - thus increasing the image size, and
+with it the time it takes to download that image - increasing first start
+latency.
+
+This RFC proposes a dynamic loading mechanism that solves most of these
+problems.
+
+## Summary
+
+`compute_ctl` is made responsible for loading extensions on-demand into
+the container's file system for dynamically loaded extensions, and will also
+make sure that the extensions in `shared_preload_libraries` are downloaded
+before the compute node starts.
+
+## Components
+
+compute_ctl, PostgreSQL, neon (extension), Compute Host Node, Extension Store
+
+## Requirements
+
+Compute nodes with no extra extensions should not be negatively impacted by
+the existence of support for many extensions.
+
+Installing an extension into PostgreSQL should be easy.
+
+Non-preloaded extensions shouldn't impact startup latency.
+
+Uninstalled extensions shouldn't impact query latency.
+
+A small latency penalty for dynamically loaded extensions is acceptable in
+the first seconds of compute startup, but not in steady-state operations.
+
+## Proposed implementation
+
+### On-demand, JIT-loading of extensions
+
+Before postgres starts we download 
+- control files for all extensions available to that compute node;
+- all `shared_preload_libraries`;
+
+After postgres is running, `compute_ctl` listens for requests to load files.
+When PostgreSQL requests a file, `compute_ctl` downloads it.
+
+PostgreSQL requests files in the following cases:
+- When loading a preload library set in `local_preload_libraries`
+- When explicitly loading a library with `LOAD`
+- Wnen creating extension with `CREATE EXTENSION` (download sql scripts, (optional) extension data files and (optional) library files)))
+
+
+#### Summary
+
+Pros:
+ - Startup is only as slow as it takes to load all (shared_)preload_libraries
+ - Supports BYO Extension
+
+Cons:
+ - O(sizeof(extensions)) IO requirement for loading all extensions.
+
+### Alternative solutions
+
+1. Allow users to add their extensions to the base image
+   
+   Pros:
+    - Easy to deploy
+
+   Cons:
+    - Doesn't scale - first start size is dependent on image size;
+    - All extensions are shared across all users: It doesn't allow users to
+      bring their own restrictive-licensed extensions
+
+2. Bring Your Own compute image
+   
+   Pros:
+    - Still easy to deploy
+    - User can bring own patched version of PostgreSQL
+
+   Cons:
+    - First start latency is O(sizeof(extensions image))
+    - Warm instance pool for skipping pod schedule latency is not feasible with
+      O(n) custom images
+    - Support channels are difficult to manage
+
+3. Download all user extensions in bulk on compute start
+   
+   Pros:
+    - Easy to deploy
+    - No startup latency issues for "clean" users.
+    - Warm instance pool for skipping pod schedule latency is possible
+
+   Cons:
+    - Downloading all extensions in advance takes a lot of time, thus startup
+      latency issues
+
+4. Store user's extensions in persistent storage
+   
+   Pros:
+    - Easy to deploy
+    - No startup latency issues
+    - Warm instance pool for skipping pod schedule latency is possible
+
+   Cons:
+    - EC2 instances have only limited number of attachments shared between EBS
+      volumes, direct-attached NVMe drives, and ENIs.
+    - Compute instance migration isn't trivially solved for EBS mounts (e.g.
+      the device is unavailable whilst moving the mount between instances).
+    - EBS can only mount on one instance at a time (except the expensive IO2
+      device type).
+
+5. Store user's extensions in network drive
+   
+   Pros:
+    - Easy to deploy
+    - Few startup latency issues
+    - Warm instance pool for skipping pod schedule latency is possible
+
+   Cons:
+    - We'd need networked drives, and a lot of them, which would store many
+      duplicate extensions.
+    - **UNCHECKED:** Compute instance migration may not work nicely with
+      networked IOs
+
+
+### Idea extensions
+
+The extension store does not have to be S3 directly, but could be a Node-local
+caching service on top of S3. This would reduce the load on the network for
+popular extensions.
+
+## Extension Storage implementation
+
+The layout of the S3 bucket is as follows:
+```
+v14/ext_index.json
+    -- this contains information necessary to create control files
+v14/extensions/test_ext1.tar.gz
+    -- this contains the library files and sql files necessary to create this extension
+v14/extensions/custom_ext1.tar.gz
+```
+The difference between private and public extensions is determined by who can
+load the extension. This is specified in `ext_index.json`.
+Speicially, `ext_index.json` has a list of public extensions, and a list of
+extensions enabled for specific tenant-ids. Here is an example `ext_index.json`:
+```
+{
+  "enabled_extensions": {
+    "123454321": [
+      "anon"
+    ],
+    "public": [
+      "embedding"
+    ]
+  },
+  "control_data": {
+    "embedding": "comment = 'hnsw index' \ndefault_version = '0.1.0' \nmodule_pathname = '$libdir/embedding' \nrelocatable = true \ntrusted = true",
+    "anon": "# PostgreSQL Anonymizer (anon) extension \ncomment = 'Data anonymization tools' \ndefault_version = '1.1.0' \ndirectory='extension/anon' \nrelocatable = false \nrequires = 'pgcrypto' \nsuperuser = false \nmodule_pathname = '$libdir/anon' \ntrusted = true \n"
+  }
+}
+```
+
+### How to add new extension to the Extension Storage?
+
+Simply upload build artifacts to the S3 bucket.
+Implement a CI step for that. Splitting it from ompute-node-image build.
+
+### How do we deal with extension versions and updates?
+
+Currently, we rebuild extensions on every compute-node-image build and store them in the <build-version> prefix.
+This is needed to ensure that `/share` and `/lib` files are in sync.
+
+For extension updates, we rely on the PostgreSQL extension versioning mechanism (sql update scripts) and extension authors to not break backwards compatibility within one major version of PostgreSQL.
+
+### Alternatives
+
+For extensions written on trusted languages we can also adopt
+`dbdev` PostgreSQL Package Manager based on `pg_tle` by Supabase.
+This will increase the amount supported extensions and decrease the amount of work required to support them.

docs/rfcs/024-user-mgmt.md

@@ -0,0 +1,84 @@
+# Postgres user and database management
+
+(This supersedes the previous proposal that looked too complicated and desynchronization-prone)
+
+We've accumulated a bunch of problems with our approach to role and database management, namely:
+
+1. we don't allow role and database creation from Postgres, and users are complaining about that
+2. fine-grained role management is not possible both from Postgres and console
+
+Right now, we do store users and databases both in console and Postgres, and there are two main reasons for
+that:
+
+* we want to be able to authenticate users in proxy against the console without Postgres' involvement. Otherwise,
+malicious brute force attempts will wake up Postgres (expensive) and may exhaust the Postgres connections limit (deny of service).
+* it is handy when we can render console UI without waking up compute (e.g., show database list)
+
+This RFC doesn't talk about giving root access to the database, which is blocked by a secure runtime setup.
+
+## Overview
+
+* Add Postgres extension that sends an HTTP request each time transaction that modifies users/databases is about to commit.
+* Add user management API to internal console API. Also, the console should put a JWT token into the compute so that it can access management API.
+
+## Postgres behavior
+
+The default user role (@username) should have `CREATE ROLE`, `CREATE DB`, and `BYPASSRLS` privileges. We expose the Postgres port
+to the open internet, so we need to check password strength. Now console generates strong passwords, so there is no risk of having dumb passwords. With user-provided passwords, such risks exist.
+
+Since we store passwords in the console we should also send unencrypted password when role is created/changed. Hence communication with the console must be encrypted. Postgres also supports creating roles using hashes, in that case, we will not be able to get a raw password. So I can see the following options here:
+  * roles created via SQL will *not* have raw passwords in the console
+  * roles created via SQL will have raw passwords in the console, except ones that were created using hashes
+
+I'm leaning towards the second option here as it is a bit more consistent one -- if raw password storage is enabled then we store passwords in all cases where we can store them.
+
+To send data about roles and databases from Postgres to the console we can create the following Postgres extension:
+
+  * Intercept role/database changes in `ProcessUtility_hook`. Here we have access to the query statement with the raw password. The hook handler itself should not dial the console immediately and rather stash info in some hashmap for later use.
+  * When the transaction is about to commit we execute collected role modifications (all as one -- console should either accept all or reject all, and hence API shouldn't be REST-like). If the console request fails we can roll back the transaction. This way if the transaction is committed we know for sure that console has this information. We can use `XACT_EVENT_PRE_COMMIT` and `XACT_EVENT_PARALLEL_PRE_COMMIT` for that.
+  * Extension should be mindful of the fact that it is possible to create and delete roles within the transaction.
+  * We also need to track who is database owner, some coding around may be needed to get the current user when the database is created.
+
+## Console user management API
+
+The current public API has REST API for role management. We need to have some analog for the internal API (called mgmt API in the console code). But unlike public API here we want to have an atomic way to create several roles/databases (in cases when several roles were created in the same transaction). So something like that may work:
+
+```
+curl -X PATCH /api/v1/roles_and_databases -d '
+[
+    {"op":"create", "type":"role", "name": "kurt", "password":"lYgT3BlbkFJ2vBZrqv"},
+    {"op":"drop", "type":"role", "name": "trout"},
+    {"op":"alter", "type":"role", "name": "kilgore", "password":"3BlbkFJ2vB"},
+    {"op":"create", "type":"database", "name": "db2", "owner": "eliot"},
+]
+'
+```
+
+Makes sense not to error out on duplicated create/delete operations (see failure modes)
+
+## Managing users from the console
+
+Now console puts a spec file with the list of databases/roles and delta operations in all the compute pods. `compute_ctl` then picks up that file and stubbornly executes deltas and checks data in the spec file is the same as in the Postgres. This way if the user creates a role in the UI we restart compute with a new spec file and during the start databases/roles are created. So if Postgres send an HTTP call each time role is created we need to break recursion in that case. We can do that based on application_name or some GUC or user (local == no HTTP hook).
+
+Generally, we have several options when we are creating users via console:
+
+1. restart compute with a new spec file, execute local SQL command; cut recursion in the extension
+2. "push" spec files into running compute, execute local SQL command; cut recursion in the extension
+3. "push" spec files into running compute, execute local SQL command; let extension create those roles in the console
+4. avoid managing roles via spec files, send SQL commands to compute; let extension create those roles in the console
+
+The last option is the most straightforward one, but with the raw password storage opt-out, we will not have the password to establish an SQL connection. Also, we need a spec for provisioning purposes and to address potential desync (but that is quite unlikely). So I think the easiest approach would be:
+
+1. keep role management like it is now and cut the recursion in the extension when SQL is executed by compute_ctl
+2. add "push" endpoint to the compute_ctl to avoid compute restart during the `apply_config` operation -- that can be done as a follow up to avoid increasing scope too much
+
+## Failure modes
+
+* during role creation via SQL role was created in the console but the connection was dropped before Postgres got acknowledgment or some error happened after acknowledgment (out of disk space, deadlock, etc):
+
+  in that case, Postgres won't have a role that exists in the console. Compute restart will heal it (due to the spec file). Also if the console allows repeated creation/deletion user can repeat the transaction.
+
+
+# Scalability
+
+On my laptop, I can create 4200 roles per second. That corresponds to 363 million roles per day. Since each role creation ends up in the console database we can add some limit to the number of roles (could be reasonably big to not run into it often -- like 1k or 10k).

libs/compute_api/src/responses.rs

@@ -71,8 +71,11 @@ pub struct ComputeMetrics {
     pub wait_for_spec_ms: u64,
     pub sync_safekeepers_ms: u64,
     pub basebackup_ms: u64,
+    pub basebackup_bytes: u64,
+    pub start_postgres_ms: u64,
     pub config_ms: u64,
     pub total_startup_ms: u64,
+    pub load_libraries_ms: u64,
 }
 
 /// Response of the `/computes/{compute_id}/spec` control-plane API.

libs/compute_api/src/spec.rs

@@ -60,6 +60,9 @@ pub struct ComputeSpec {
     /// If set, 'storage_auth_token' is used as the password to authenticate to
     /// the pageserver and safekeepers.
     pub storage_auth_token: Option<String>,
+
+    // list of prefixes to search for custom extensions in remote extension storage
+    pub custom_extensions: Option<Vec<String>>,
 }
 
 #[serde_as]

libs/metrics/src/lib.rs

@@ -23,6 +23,7 @@ use prometheus::{Registry, Result};
 pub mod launch_timestamp;
 mod wrappers;
 pub use wrappers::{CountedReader, CountedWriter};
+pub mod metric_vec_duration;
 
 pub type UIntGauge = GenericGauge<AtomicU64>;
 pub type UIntGaugeVec = GenericGaugeVec<AtomicU64>;

libs/metrics/src/metric_vec_duration.rs

@@ -0,0 +1,23 @@
+//! Helpers for observing duration on HistogramVec / CounterVec / GaugeVec / MetricVec<T>.
+
+use std::{future::Future, time::Instant};
+
+pub trait DurationResultObserver {
+    fn observe_result<T, E>(&self, res: &Result<T, E>, duration: std::time::Duration);
+}
+
+pub async fn observe_async_block_duration_by_result<
+    T,
+    E,
+    F: Future<Output = Result<T, E>>,
+    O: DurationResultObserver,
+>(
+    observer: &O,
+    block: F,
+) -> Result<T, E> {
+    let start = Instant::now();
+    let result = block.await;
+    let duration = start.elapsed();
+    observer.observe_result(&result, duration);
+    result
+}

libs/remote_storage/src/lib.rs

@@ -70,6 +70,14 @@ impl RemotePath {
     pub fn join(&self, segment: &Path) -> Self {
         Self(self.0.join(segment))
     }
+
+    pub fn get_path(&self) -> &PathBuf {
+        &self.0
+    }
+
+    pub fn extension(&self) -> Option<&str> {
+        self.0.extension()?.to_str()
+    }
 }
 
 /// Storage (potentially remote) API to manage its state.
@@ -86,6 +94,19 @@ pub trait RemoteStorage: Send + Sync + 'static {
         prefix: Option<&RemotePath>,
     ) -> Result<Vec<RemotePath>, DownloadError>;
 
+    /// Lists all files in directory "recursively"
+    /// (not really recursively, because AWS has a flat namespace)
+    /// Note: This is subtely different than list_prefixes,
+    /// because it is for listing files instead of listing
+    /// names sharing common prefixes.
+    /// For example,
+    /// list_files("foo/bar") = ["foo/bar/cat123.txt",
+    /// "foo/bar/cat567.txt", "foo/bar/dog123.txt", "foo/bar/dog456.txt"]
+    /// whereas,
+    /// list_prefixes("foo/bar/") = ["cat", "dog"]
+    /// See `test_real_s3.rs` for more details.
+    async fn list_files(&self, folder: Option<&RemotePath>) -> anyhow::Result<Vec<RemotePath>>;
+
     /// Streams the local file contents into remote into the remote storage entry.
     async fn upload(
         &self,
@@ -163,6 +184,20 @@ pub enum GenericRemoteStorage {
 }
 
 impl GenericRemoteStorage {
+    // A function for listing all the files in a "directory"
+    // Example:
+    // list_files("foo/bar") = ["foo/bar/a.txt", "foo/bar/b.txt"]
+    pub async fn list_files(&self, folder: Option<&RemotePath>) -> anyhow::Result<Vec<RemotePath>> {
+        match self {
+            Self::LocalFs(s) => s.list_files(folder).await,
+            Self::AwsS3(s) => s.list_files(folder).await,
+            Self::Unreliable(s) => s.list_files(folder).await,
+        }
+    }
+
+    // lists common *prefixes*, if any of files
+    // Example:
+    // list_prefixes("foo123","foo567","bar123","bar432") = ["foo", "bar"]
     pub async fn list_prefixes(
         &self,
         prefix: Option<&RemotePath>,

libs/remote_storage/src/local_fs.rs

@@ -7,6 +7,7 @@
 use std::{
     borrow::Cow,
     future::Future,
+    io::ErrorKind,
     path::{Path, PathBuf},
     pin::Pin,
 };
@@ -48,6 +49,14 @@ impl LocalFs {
         Ok(Self { storage_root })
     }
 
+    // mirrors S3Bucket::s3_object_to_relative_path
+    fn local_file_to_relative_path(&self, key: PathBuf) -> RemotePath {
+        let relative_path = key
+            .strip_prefix(&self.storage_root)
+            .expect("relative path must contain storage_root as prefix");
+        RemotePath(relative_path.into())
+    }
+
     async fn read_storage_metadata(
         &self,
         file_path: &Path,
@@ -132,6 +141,34 @@ impl RemoteStorage for LocalFs {
         Ok(prefixes)
     }
 
+    // recursively lists all files in a directory,
+    // mirroring the `list_files` for `s3_bucket`
+    async fn list_files(&self, folder: Option<&RemotePath>) -> anyhow::Result<Vec<RemotePath>> {
+        let full_path = match folder {
+            Some(folder) => folder.with_base(&self.storage_root),
+            None => self.storage_root.clone(),
+        };
+        let mut files = vec![];
+        let mut directory_queue = vec![full_path.clone()];
+
+        while !directory_queue.is_empty() {
+            let cur_folder = directory_queue
+                .pop()
+                .expect("queue cannot be empty: we just checked");
+            let mut entries = fs::read_dir(cur_folder.clone()).await?;
+            while let Some(entry) = entries.next_entry().await? {
+                let file_name: PathBuf = entry.file_name().into();
+                let full_file_name = cur_folder.clone().join(&file_name);
+                let file_remote_path = self.local_file_to_relative_path(full_file_name.clone());
+                files.push(file_remote_path.clone());
+                if full_file_name.is_dir() {
+                    directory_queue.push(full_file_name);
+                }
+            }
+        }
+        Ok(files)
+    }
+
     async fn upload(
         &self,
         data: impl io::AsyncRead + Unpin + Send + Sync + 'static,
@@ -307,18 +344,14 @@ impl RemoteStorage for LocalFs {
 
     async fn delete(&self, path: &RemotePath) -> anyhow::Result<()> {
         let file_path = path.with_base(&self.storage_root);
-        if !file_path.exists() {
+        match fs::remove_file(&file_path).await {
+            Ok(()) => Ok(()),
+            // The file doesn't exist. This shouldn't yield an error to mirror S3's behaviour.
             // See https://docs.aws.amazon.com/AmazonS3/latest/API/API_DeleteObject.html
             // > If there isn't a null version, Amazon S3 does not remove any objects but will still respond that the command was successful.
-            return Ok(());
+            Err(e) if e.kind() == ErrorKind::NotFound => Ok(()),
+            Err(e) => Err(anyhow::anyhow!(e)),
         }
-
-        if !file_path.is_file() {
-            anyhow::bail!("{file_path:?} is not a file");
-        }
-        Ok(fs::remove_file(file_path)
-            .await
-            .map_err(|e| anyhow::anyhow!(e))?)
     }
 
     async fn delete_objects<'a>(&self, paths: &'a [RemotePath]) -> anyhow::Result<()> {

libs/remote_storage/src/s3_bucket.rs

@@ -347,6 +347,58 @@ impl RemoteStorage for S3Bucket {
         Ok(document_keys)
     }
 
+    /// See the doc for `RemoteStorage::list_files`
+    async fn list_files(&self, folder: Option<&RemotePath>) -> anyhow::Result<Vec<RemotePath>> {
+        let mut folder_name = folder
+            .map(|p| self.relative_path_to_s3_object(p))
+            .or_else(|| self.prefix_in_bucket.clone());
+
+        // remove leading "/" if one exists
+        if let Some(folder_name_slash) = folder_name.clone() {
+            if folder_name_slash.starts_with(REMOTE_STORAGE_PREFIX_SEPARATOR) {
+                folder_name = Some(folder_name_slash[1..].to_string());
+            }
+        }
+
+        // AWS may need to break the response into several parts
+        let mut continuation_token = None;
+        let mut all_files = vec![];
+        loop {
+            let _guard = self
+                .concurrency_limiter
+                .acquire()
+                .await
+                .context("Concurrency limiter semaphore got closed during S3 list_files")?;
+            metrics::inc_list_objects();
+
+            let response = self
+                .client
+                .list_objects_v2()
+                .bucket(self.bucket_name.clone())
+                .set_prefix(folder_name.clone())
+                .set_continuation_token(continuation_token)
+                .set_max_keys(self.max_keys_per_list_response)
+                .send()
+                .await
+                .map_err(|e| {
+                    metrics::inc_list_objects_fail();
+                    e
+                })
+                .context("Failed to list files in S3 bucket")?;
+
+            for object in response.contents().unwrap_or_default() {
+                let object_path = object.key().expect("response does not contain a key");
+                let remote_path = self.s3_object_to_relative_path(object_path);
+                all_files.push(remote_path);
+            }
+            match response.next_continuation_token {
+                Some(new_token) => continuation_token = Some(new_token),
+                None => break,
+            }
+        }
+        Ok(all_files)
+    }
+
     async fn upload(
         &self,
         from: impl io::AsyncRead + Unpin + Send + Sync + 'static,

libs/remote_storage/src/simulate_failures.rs

@@ -83,6 +83,11 @@ impl RemoteStorage for UnreliableWrapper {
         self.inner.list_prefixes(prefix).await
     }
 
+    async fn list_files(&self, folder: Option<&RemotePath>) -> anyhow::Result<Vec<RemotePath>> {
+        self.attempt(RemoteOp::ListPrefixes(folder.cloned()))?;
+        self.inner.list_files(folder).await
+    }
+
     async fn upload(
         &self,
         data: impl tokio::io::AsyncRead + Unpin + Send + Sync + 'static,

libs/remote_storage/tests/test_real_s3.rs

@@ -88,6 +88,58 @@ async fn s3_pagination_should_work(ctx: &mut MaybeEnabledS3WithTestBlobs) -> any
     Ok(())
 }
 
+/// Tests that S3 client can list all files in a folder, even if the response comes paginated and requirees multiple S3 queries.
+/// Uses real S3 and requires [`ENABLE_REAL_S3_REMOTE_STORAGE_ENV_VAR_NAME`] and related S3 cred env vars specified. Test will skip real code and pass if env vars not set.
+/// See `s3_pagination_should_work` for more information.
+///
+/// First, create a set of S3 objects with keys `random_prefix/folder{j}/blob_{i}.txt` in [`upload_s3_data`]
+/// Then performs the following queries:
+///    1. `list_files(None)`. This should return all files `random_prefix/folder{j}/blob_{i}.txt`
+///    2. `list_files("folder1")`.  This  should return all files `random_prefix/folder1/blob_{i}.txt`
+#[test_context(MaybeEnabledS3WithSimpleTestBlobs)]
+#[tokio::test]
+async fn s3_list_files_works(ctx: &mut MaybeEnabledS3WithSimpleTestBlobs) -> anyhow::Result<()> {
+    let ctx = match ctx {
+        MaybeEnabledS3WithSimpleTestBlobs::Enabled(ctx) => ctx,
+        MaybeEnabledS3WithSimpleTestBlobs::Disabled => return Ok(()),
+        MaybeEnabledS3WithSimpleTestBlobs::UploadsFailed(e, _) => {
+            anyhow::bail!("S3 init failed: {e:?}")
+        }
+    };
+    let test_client = Arc::clone(&ctx.enabled.client);
+    let base_prefix =
+        RemotePath::new(Path::new("folder1")).context("common_prefix construction")?;
+    let root_files = test_client
+        .list_files(None)
+        .await
+        .context("client list root files failure")?
+        .into_iter()
+        .collect::<HashSet<_>>();
+    assert_eq!(
+        root_files,
+        ctx.remote_blobs.clone(),
+        "remote storage list_files on root mismatches with the uploads."
+    );
+    let nested_remote_files = test_client
+        .list_files(Some(&base_prefix))
+        .await
+        .context("client list nested files failure")?
+        .into_iter()
+        .collect::<HashSet<_>>();
+    let trim_remote_blobs: HashSet<_> = ctx
+        .remote_blobs
+        .iter()
+        .map(|x| x.get_path().to_str().expect("must be valid name"))
+        .filter(|x| x.starts_with("folder1"))
+        .map(|x| RemotePath::new(Path::new(x)).expect("must be valid name"))
+        .collect();
+    assert_eq!(
+        nested_remote_files, trim_remote_blobs,
+        "remote storage list_files on subdirrectory mismatches with the uploads."
+    );
+    Ok(())
+}
+
 #[test_context(MaybeEnabledS3)]
 #[tokio::test]
 async fn s3_delete_non_exising_works(ctx: &mut MaybeEnabledS3) -> anyhow::Result<()> {
@@ -121,10 +173,15 @@ async fn s3_delete_objects_works(ctx: &mut MaybeEnabledS3) -> anyhow::Result<()>
     let path2 = RemotePath::new(&PathBuf::from(format!("{}/path2", ctx.base_prefix,)))
         .with_context(|| "RemotePath conversion")?;
 
+    let path3 = RemotePath::new(&PathBuf::from(format!("{}/path3", ctx.base_prefix,)))
+        .with_context(|| "RemotePath conversion")?;
+
     let data1 = "remote blob data1".as_bytes();
     let data1_len = data1.len();
     let data2 = "remote blob data2".as_bytes();
     let data2_len = data2.len();
+    let data3 = "remote blob data3".as_bytes();
+    let data3_len = data3.len();
     ctx.client
         .upload(std::io::Cursor::new(data1), data1_len, &path1, None)
         .await?;
@@ -133,8 +190,18 @@ async fn s3_delete_objects_works(ctx: &mut MaybeEnabledS3) -> anyhow::Result<()>
         .upload(std::io::Cursor::new(data2), data2_len, &path2, None)
         .await?;
 
+    ctx.client
+        .upload(std::io::Cursor::new(data3), data3_len, &path3, None)
+        .await?;
+
     ctx.client.delete_objects(&[path1, path2]).await?;
 
+    let prefixes = ctx.client.list_prefixes(None).await?;
+
+    assert_eq!(prefixes.len(), 1);
+
+    ctx.client.delete_objects(&[path3]).await?;
+
     Ok(())
 }
 
@@ -248,6 +315,66 @@ impl AsyncTestContext for MaybeEnabledS3WithTestBlobs {
     }
 }
 
+// NOTE: the setups for the list_prefixes test and the list_files test are very similar
+// However, they are not idential. The list_prefixes function is concerned with listing prefixes,
+// whereas the list_files function is concerned with listing files.
+// See `RemoteStorage::list_files` documentation for more details
+enum MaybeEnabledS3WithSimpleTestBlobs {
+    Enabled(S3WithSimpleTestBlobs),
+    Disabled,
+    UploadsFailed(anyhow::Error, S3WithSimpleTestBlobs),
+}
+struct S3WithSimpleTestBlobs {
+    enabled: EnabledS3,
+    remote_blobs: HashSet<RemotePath>,
+}
+
+#[async_trait::async_trait]
+impl AsyncTestContext for MaybeEnabledS3WithSimpleTestBlobs {
+    async fn setup() -> Self {
+        ensure_logging_ready();
+        if env::var(ENABLE_REAL_S3_REMOTE_STORAGE_ENV_VAR_NAME).is_err() {
+            info!(
+                "`{}` env variable is not set, skipping the test",
+                ENABLE_REAL_S3_REMOTE_STORAGE_ENV_VAR_NAME
+            );
+            return Self::Disabled;
+        }
+
+        let max_keys_in_list_response = 10;
+        let upload_tasks_count = 1 + (2 * usize::try_from(max_keys_in_list_response).unwrap());
+
+        let enabled = EnabledS3::setup(Some(max_keys_in_list_response)).await;
+
+        match upload_simple_s3_data(&enabled.client, upload_tasks_count).await {
+            ControlFlow::Continue(uploads) => {
+                info!("Remote objects created successfully");
+
+                Self::Enabled(S3WithSimpleTestBlobs {
+                    enabled,
+                    remote_blobs: uploads,
+                })
+            }
+            ControlFlow::Break(uploads) => Self::UploadsFailed(
+                anyhow::anyhow!("One or multiple blobs failed to upload to S3"),
+                S3WithSimpleTestBlobs {
+                    enabled,
+                    remote_blobs: uploads,
+                },
+            ),
+        }
+    }
+
+    async fn teardown(self) {
+        match self {
+            Self::Disabled => {}
+            Self::Enabled(ctx) | Self::UploadsFailed(_, ctx) => {
+                cleanup(&ctx.enabled.client, ctx.remote_blobs).await;
+            }
+        }
+    }
+}
+
 fn create_s3_client(
     max_keys_per_list_response: Option<i32>,
 ) -> anyhow::Result<Arc<GenericRemoteStorage>> {
@@ -258,7 +385,7 @@ fn create_s3_client(
     let random_prefix_part = std::time::SystemTime::now()
         .duration_since(UNIX_EPOCH)
         .context("random s3 test prefix part calculation")?
-        .as_millis();
+        .as_nanos();
     let remote_storage_config = RemoteStorageConfig {
         max_concurrent_syncs: NonZeroUsize::new(100).unwrap(),
         max_sync_errors: NonZeroU32::new(5).unwrap(),
@@ -364,3 +491,52 @@ async fn cleanup(client: &Arc<GenericRemoteStorage>, objects_to_delete: HashSet<
         }
     }
 }
+
+// Uploads files `folder{j}/blob{i}.txt`. See test description for more details.
+async fn upload_simple_s3_data(
+    client: &Arc<GenericRemoteStorage>,
+    upload_tasks_count: usize,
+) -> ControlFlow<HashSet<RemotePath>, HashSet<RemotePath>> {
+    info!("Creating {upload_tasks_count} S3 files");
+    let mut upload_tasks = JoinSet::new();
+    for i in 1..upload_tasks_count + 1 {
+        let task_client = Arc::clone(client);
+        upload_tasks.spawn(async move {
+            let blob_path = PathBuf::from(format!("folder{}/blob_{}.txt", i / 7, i));
+            let blob_path = RemotePath::new(&blob_path)
+                .with_context(|| format!("{blob_path:?} to RemotePath conversion"))?;
+            debug!("Creating remote item {i} at path {blob_path:?}");
+
+            let data = format!("remote blob data {i}").into_bytes();
+            let data_len = data.len();
+            task_client
+                .upload(std::io::Cursor::new(data), data_len, &blob_path, None)
+                .await?;
+
+            Ok::<_, anyhow::Error>(blob_path)
+        });
+    }
+
+    let mut upload_tasks_failed = false;
+    let mut uploaded_blobs = HashSet::with_capacity(upload_tasks_count);
+    while let Some(task_run_result) = upload_tasks.join_next().await {
+        match task_run_result
+            .context("task join failed")
+            .and_then(|task_result| task_result.context("upload task failed"))
+        {
+            Ok(upload_path) => {
+                uploaded_blobs.insert(upload_path);
+            }
+            Err(e) => {
+                error!("Upload task failed: {e:?}");
+                upload_tasks_failed = true;
+            }
+        }
+    }
+
+    if upload_tasks_failed {
+        ControlFlow::Break(uploaded_blobs)
+    } else {
+        ControlFlow::Continue(uploaded_blobs)
+    }
+}

libs/utils/Cargo.toml

@@ -5,7 +5,6 @@ edition.workspace = true
 license.workspace = true
 
 [dependencies]
-atty.workspace = true
 sentry.workspace = true
 async-trait.workspace = true
 anyhow.workspace = true
@@ -41,6 +40,8 @@ pq_proto.workspace = true
 metrics.workspace = true
 workspace_hack.workspace = true
 
+const_format.workspace = true
+
 [dev-dependencies]
 byteorder.workspace = true
 bytes.workspace = true

libs/utils/src/http/error.rs

@@ -1,5 +1,6 @@
 use hyper::{header, Body, Response, StatusCode};
 use serde::{Deserialize, Serialize};
+use std::error::Error as StdError;
 use thiserror::Error;
 use tracing::error;
 
@@ -15,7 +16,7 @@ pub enum ApiError {
     Unauthorized(String),
 
     #[error("NotFound: {0}")]
-    NotFound(anyhow::Error),
+    NotFound(Box<dyn StdError + Send + Sync + 'static>),
 
     #[error("Conflict: {0}")]
     Conflict(String),

libs/utils/src/lib.rs

@@ -109,10 +109,16 @@ pub use failpoint_macro_helpers::failpoint_sleep_helper;
 /// * building in docker (either in CI or locally)
 ///
 /// One thing to note is that .git is not available in docker (and it is bad to include it there).
-/// So everything becides docker build is covered by git_version crate, and docker uses a `GIT_VERSION` argument to get the value required.
-/// It takes variable from build process env and puts it to the rustc env. And then we can retrieve it here by using env! macro.
-/// Git version received from environment variable used as a fallback in git_version invocation.
-/// And to avoid running buildscript every recompilation, we use rerun-if-env-changed option.
+/// When building locally, the `git_version` is used to query .git. When building on CI and docker,
+/// we don't build the actual PR branch commits, but always a "phantom" would be merge commit to
+/// the target branch -- the actual PR commit from which we build from is supplied as GIT_VERSION
+/// environment variable.
+///
+/// We ended up with this compromise between phantom would be merge commits vs. pull request branch
+/// heads due to old logs becoming more reliable (github could gc the phantom merge commit
+/// anytime) in #4641.
+///
+/// To avoid running buildscript every recompilation, we use rerun-if-env-changed option.
 /// So the build script will be run only when GIT_VERSION envvar has changed.
 ///
 /// Why not to use buildscript to get git commit sha directly without procmacro from different crate?
@@ -132,17 +138,28 @@ pub use failpoint_macro_helpers::failpoint_sleep_helper;
 #[macro_export]
 macro_rules! project_git_version {
     ($const_identifier:ident) => {
-        const $const_identifier: &str = git_version::git_version!(
-            prefix = "git:",
-            fallback = concat!(
-                "git-env:",
-                env!("GIT_VERSION", "Missing GIT_VERSION envvar")
-            ),
-            args = ["--abbrev=40", "--always", "--dirty=-modified"] // always use full sha
-        );
+        // this should try GIT_VERSION first only then git_version::git_version!
+        const $const_identifier: &::core::primitive::str = {
+            const __COMMIT_FROM_GIT: &::core::primitive::str = git_version::git_version! {
+                prefix = "",
+                fallback = "unknown",
+                args = ["--abbrev=40", "--always", "--dirty=-modified"] // always use full sha
+            };
+
+            const __ARG: &[&::core::primitive::str; 2] = &match ::core::option_env!("GIT_VERSION") {
+                ::core::option::Option::Some(x) => ["git-env:", x],
+                ::core::option::Option::None => ["git:", __COMMIT_FROM_GIT],
+            };
+
+            $crate::__const_format::concatcp!(__ARG[0], __ARG[1])
+        };
     };
 }
 
+/// Re-export for `project_git_version` macro
+#[doc(hidden)]
+pub use const_format as __const_format;
+
 /// Same as `assert!`, but evaluated during compilation and gets optimized out in runtime.
 #[macro_export]
 macro_rules! const_assert {

libs/utils/src/logging.rs

@@ -84,7 +84,7 @@ pub fn init(
     let r = r.with({
         let log_layer = tracing_subscriber::fmt::layer()
             .with_target(false)
-            .with_ansi(atty::is(atty::Stream::Stdout))
+            .with_ansi(false)
             .with_writer(std::io::stdout);
         let log_layer = match log_format {
             LogFormat::Json => log_layer.json().boxed(),

libs/utils/src/measured_stream.rs

@@ -1,4 +1,5 @@
 use pin_project_lite::pin_project;
+use std::io::Read;
 use std::pin::Pin;
 use std::{io, task};
 use tokio::io::{AsyncRead, AsyncWrite, ReadBuf};
@@ -75,3 +76,34 @@ impl<S: AsyncWrite + Unpin, R, W: FnMut(usize)> AsyncWrite for MeasuredStream<S,
         self.project().stream.poll_shutdown(context)
     }
 }
+
+/// Wrapper for a reader that counts bytes read.
+///
+/// Similar to MeasuredStream but it's one way and it's sync
+pub struct MeasuredReader<R: Read> {
+    inner: R,
+    byte_count: usize,
+}
+
+impl<R: Read> MeasuredReader<R> {
+    pub fn new(reader: R) -> Self {
+        Self {
+            inner: reader,
+            byte_count: 0,
+        }
+    }
+
+    pub fn get_byte_count(&self) -> usize {
+        self.byte_count
+    }
+}
+
+impl<R: Read> Read for MeasuredReader<R> {
+    fn read(&mut self, buf: &mut [u8]) -> std::io::Result<usize> {
+        let result = self.inner.read(buf);
+        if let Ok(n_bytes) = result {
+            self.byte_count += n_bytes
+        }
+        result
+    }
+}

pageserver/Cargo.toml

@@ -12,6 +12,7 @@ testing = ["fail/failpoints"]
 
 [dependencies]
 anyhow.workspace = true
+async-compression.workspace = true
 async-stream.workspace = true
 async-trait.workspace = true
 byteorder.workspace = true
@@ -24,6 +25,7 @@ consumption_metrics.workspace = true
 crc32c.workspace = true
 crossbeam-utils.workspace = true
 either.workspace = true
+flate2.workspace = true
 fail.workspace = true
 futures.workspace = true
 git-version.workspace = true

pageserver/benches/bench_layer_map.rs

@@ -1,22 +1,23 @@
 use pageserver::keyspace::{KeyPartitioning, KeySpace};
 use pageserver::repository::Key;
 use pageserver::tenant::layer_map::LayerMap;
-use pageserver::tenant::storage_layer::{Layer, LayerDescriptor, LayerFileName};
+use pageserver::tenant::storage_layer::LayerFileName;
+use pageserver::tenant::storage_layer::PersistentLayerDesc;
 use rand::prelude::{SeedableRng, SliceRandom, StdRng};
 use std::cmp::{max, min};
 use std::fs::File;
 use std::io::{BufRead, BufReader};
 use std::path::PathBuf;
 use std::str::FromStr;
-use std::sync::Arc;
 use std::time::Instant;
+use utils::id::{TenantId, TimelineId};
 
 use utils::lsn::Lsn;
 
 use criterion::{black_box, criterion_group, criterion_main, Criterion};
 
-fn build_layer_map(filename_dump: PathBuf) -> LayerMap<LayerDescriptor> {
-    let mut layer_map = LayerMap::<LayerDescriptor>::default();
+fn build_layer_map(filename_dump: PathBuf) -> LayerMap {
+    let mut layer_map = LayerMap::default();
 
     let mut min_lsn = Lsn(u64::MAX);
     let mut max_lsn = Lsn(0);
@@ -27,13 +28,13 @@ fn build_layer_map(filename_dump: PathBuf) -> LayerMap<LayerDescriptor> {
     for fname in filenames {
         let fname = fname.unwrap();
         let fname = LayerFileName::from_str(&fname).unwrap();
-        let layer = LayerDescriptor::from(fname);
+        let layer = PersistentLayerDesc::from(fname);
 
         let lsn_range = layer.get_lsn_range();
         min_lsn = min(min_lsn, lsn_range.start);
         max_lsn = max(max_lsn, Lsn(lsn_range.end.0 - 1));
 
-        updates.insert_historic(layer.get_persistent_layer_desc(), Arc::new(layer));
+        updates.insert_historic(layer);
     }
 
     println!("min: {min_lsn}, max: {max_lsn}");
@@ -43,7 +44,7 @@ fn build_layer_map(filename_dump: PathBuf) -> LayerMap<LayerDescriptor> {
 }
 
 /// Construct a layer map query pattern for benchmarks
-fn uniform_query_pattern(layer_map: &LayerMap<LayerDescriptor>) -> Vec<(Key, Lsn)> {
+fn uniform_query_pattern(layer_map: &LayerMap) -> Vec<(Key, Lsn)> {
     // For each image layer we query one of the pages contained, at LSN right
     // before the image layer was created. This gives us a somewhat uniform
     // coverage of both the lsn and key space because image layers have
@@ -69,7 +70,7 @@ fn uniform_query_pattern(layer_map: &LayerMap<LayerDescriptor>) -> Vec<(Key, Lsn
 
 // Construct a partitioning for testing get_difficulty map when we
 // don't have an exact result of `collect_keyspace` to work with.
-fn uniform_key_partitioning(layer_map: &LayerMap<LayerDescriptor>, _lsn: Lsn) -> KeyPartitioning {
+fn uniform_key_partitioning(layer_map: &LayerMap, _lsn: Lsn) -> KeyPartitioning {
     let mut parts = Vec::new();
 
     // We add a partition boundary at the start of each image layer,
@@ -209,13 +210,15 @@ fn bench_sequential(c: &mut Criterion) {
     for i in 0..100_000 {
         let i32 = (i as u32) % 100;
         let zero = Key::from_hex("000000000000000000000000000000000000").unwrap();
-        let layer = LayerDescriptor {
-            key: zero.add(10 * i32)..zero.add(10 * i32 + 1),
-            lsn: Lsn(i)..Lsn(i + 1),
-            is_incremental: false,
-            short_id: format!("Layer {}", i),
-        };
-        updates.insert_historic(layer.get_persistent_layer_desc(), Arc::new(layer));
+        let layer = PersistentLayerDesc::new_img(
+            TenantId::generate(),
+            TimelineId::generate(),
+            zero.add(10 * i32)..zero.add(10 * i32 + 1),
+            Lsn(i),
+            false,
+            0,
+        );
+        updates.insert_historic(layer);
     }
     updates.flush();
     println!("Finished layer map init in {:?}", now.elapsed());

pageserver/ctl/src/draw_timeline_dir.rs

@@ -117,7 +117,8 @@ pub fn main() -> Result<()> {
 
         let mut lsn_diff = (lsn_end - lsn_start) as f32;
         let mut fill = Fill::None;
-        let mut margin = 0.05 * lsn_diff; // Height-dependent margin to disambiguate overlapping deltas
+        let mut ymargin = 0.05 * lsn_diff; // Height-dependent margin to disambiguate overlapping deltas
+        let xmargin = 0.05; // Height-dependent margin to disambiguate overlapping deltas
         let mut lsn_offset = 0.0;
 
         // Fill in and thicken rectangle if it's an
@@ -128,7 +129,7 @@ pub fn main() -> Result<()> {
                 num_images += 1;
                 lsn_diff = 0.3;
                 lsn_offset = -lsn_diff / 2.0;
-                margin = 0.05;
+                ymargin = 0.05;
                 fill = Fill::Color(rgb(0, 0, 0));
             }
             Ordering::Greater => panic!("Invalid lsn range {}-{}", lsn_start, lsn_end),
@@ -137,10 +138,10 @@ pub fn main() -> Result<()> {
         println!(
             "    {}",
             rectangle(
-                key_start as f32 + stretch * margin,
-                stretch * (lsn_max as f32 - (lsn_end as f32 - margin - lsn_offset)),
-                key_diff as f32 - stretch * 2.0 * margin,
-                stretch * (lsn_diff - 2.0 * margin)
+                key_start as f32 + stretch * xmargin,
+                stretch * (lsn_max as f32 - (lsn_end as f32 - ymargin - lsn_offset)),
+                key_diff as f32 - stretch * 2.0 * xmargin,
+                stretch * (lsn_diff - 2.0 * ymargin)
             )
             .fill(fill)
             .stroke(Stroke::Color(rgb(0, 0, 0), 0.1))

pageserver/src/bin/pageserver.rs

@@ -495,50 +495,50 @@ fn start_pageserver(
                 Ok(())
             },
         );
+    }
 
-        if let Some(metric_collection_endpoint) = &conf.metric_collection_endpoint {
-            let background_jobs_barrier = background_jobs_barrier;
-            let metrics_ctx = RequestContext::todo_child(
-                TaskKind::MetricsCollection,
-                // This task itself shouldn't download anything.
-                // The actual size calculation does need downloads, and
-                // creates a child context with the right DownloadBehavior.
-                DownloadBehavior::Error,
-            );
-            task_mgr::spawn(
-                MGMT_REQUEST_RUNTIME.handle(),
-                TaskKind::MetricsCollection,
-                None,
-                None,
-                "consumption metrics collection",
-                true,
-                async move {
-                    // first wait until background jobs are cleared to launch.
-                    //
-                    // this is because we only process active tenants and timelines, and the
-                    // Timeline::get_current_logical_size will spawn the logical size calculation,
-                    // which will not be rate-limited.
-                    let cancel = task_mgr::shutdown_token();
+    if let Some(metric_collection_endpoint) = &conf.metric_collection_endpoint {
+        let background_jobs_barrier = background_jobs_barrier;
+        let metrics_ctx = RequestContext::todo_child(
+            TaskKind::MetricsCollection,
+            // This task itself shouldn't download anything.
+            // The actual size calculation does need downloads, and
+            // creates a child context with the right DownloadBehavior.
+            DownloadBehavior::Error,
+        );
+        task_mgr::spawn(
+            crate::BACKGROUND_RUNTIME.handle(),
+            TaskKind::MetricsCollection,
+            None,
+            None,
+            "consumption metrics collection",
+            true,
+            async move {
+                // first wait until background jobs are cleared to launch.
+                //
+                // this is because we only process active tenants and timelines, and the
+                // Timeline::get_current_logical_size will spawn the logical size calculation,
+                // which will not be rate-limited.
+                let cancel = task_mgr::shutdown_token();
 
-                    tokio::select! {
-                        _ = cancel.cancelled() => { return Ok(()); },
-                        _ = background_jobs_barrier.wait() => {}
-                    };
+                tokio::select! {
+                    _ = cancel.cancelled() => { return Ok(()); },
+                    _ = background_jobs_barrier.wait() => {}
+                };
 
-                    pageserver::consumption_metrics::collect_metrics(
-                        metric_collection_endpoint,
-                        conf.metric_collection_interval,
-                        conf.cached_metric_collection_interval,
-                        conf.synthetic_size_calculation_interval,
-                        conf.id,
-                        metrics_ctx,
-                    )
-                    .instrument(info_span!("metrics_collection"))
-                    .await?;
-                    Ok(())
-                },
-            );
-        }
+                pageserver::consumption_metrics::collect_metrics(
+                    metric_collection_endpoint,
+                    conf.metric_collection_interval,
+                    conf.cached_metric_collection_interval,
+                    conf.synthetic_size_calculation_interval,
+                    conf.id,
+                    metrics_ctx,
+                )
+                .instrument(info_span!("metrics_collection"))
+                .await?;
+                Ok(())
+            },
+        );
     }
 
     // Spawn a task to listen for libpq connections. It will spawn further tasks

pageserver/src/config.rs

@@ -96,12 +96,12 @@ pub mod defaults {
 
 #background_task_maximum_delay = '{DEFAULT_BACKGROUND_TASK_MAXIMUM_DELAY}'
 
-# [tenant_config]
+[tenant_config]
 #checkpoint_distance = {DEFAULT_CHECKPOINT_DISTANCE} # in bytes
 #checkpoint_timeout = {DEFAULT_CHECKPOINT_TIMEOUT}
 #compaction_target_size = {DEFAULT_COMPACTION_TARGET_SIZE} # in bytes
 #compaction_period = '{DEFAULT_COMPACTION_PERIOD}'
-#compaction_threshold = '{DEFAULT_COMPACTION_THRESHOLD}'
+#compaction_threshold = {DEFAULT_COMPACTION_THRESHOLD}
 
 #gc_period = '{DEFAULT_GC_PERIOD}'
 #gc_horizon = {DEFAULT_GC_HORIZON}
@@ -111,7 +111,8 @@ pub mod defaults {
 #min_resident_size_override = .. # in bytes
 #evictions_low_residence_duration_metric_threshold = '{DEFAULT_EVICTIONS_LOW_RESIDENCE_DURATION_METRIC_THRESHOLD}'
 #gc_feedback = false
-# [remote_storage]
+
+[remote_storage]
 
 "###
     );
@@ -569,21 +570,21 @@ impl PageServerConf {
             .join(TENANT_ATTACHING_MARKER_FILENAME)
     }
 
-    pub fn tenant_ignore_mark_file_path(&self, tenant_id: TenantId) -> PathBuf {
-        self.tenant_path(&tenant_id).join(IGNORED_TENANT_FILE_NAME)
+    pub fn tenant_ignore_mark_file_path(&self, tenant_id: &TenantId) -> PathBuf {
+        self.tenant_path(tenant_id).join(IGNORED_TENANT_FILE_NAME)
     }
 
     /// Points to a place in pageserver's local directory,
     /// where certain tenant's tenantconf file should be located.
-    pub fn tenant_config_path(&self, tenant_id: TenantId) -> PathBuf {
-        self.tenant_path(&tenant_id).join(TENANT_CONFIG_NAME)
+    pub fn tenant_config_path(&self, tenant_id: &TenantId) -> PathBuf {
+        self.tenant_path(tenant_id).join(TENANT_CONFIG_NAME)
     }
 
     pub fn timelines_path(&self, tenant_id: &TenantId) -> PathBuf {
         self.tenant_path(tenant_id).join(TIMELINES_SEGMENT_NAME)
     }
 
-    pub fn timeline_path(&self, timeline_id: &TimelineId, tenant_id: &TenantId) -> PathBuf {
+    pub fn timeline_path(&self, tenant_id: &TenantId, timeline_id: &TimelineId) -> PathBuf {
         self.timelines_path(tenant_id).join(timeline_id.to_string())
     }
 
@@ -593,7 +594,7 @@ impl PageServerConf {
         timeline_id: TimelineId,
     ) -> PathBuf {
         path_with_suffix_extension(
-            self.timeline_path(&timeline_id, &tenant_id),
+            self.timeline_path(&tenant_id, &timeline_id),
             TIMELINE_UNINIT_MARK_SUFFIX,
         )
     }
@@ -616,8 +617,8 @@ impl PageServerConf {
 
     /// Points to a place in pageserver's local directory,
     /// where certain timeline's metadata file should be located.
-    pub fn metadata_path(&self, timeline_id: TimelineId, tenant_id: TenantId) -> PathBuf {
-        self.timeline_path(&timeline_id, &tenant_id)
+    pub fn metadata_path(&self, tenant_id: &TenantId, timeline_id: &TimelineId) -> PathBuf {
+        self.timeline_path(tenant_id, timeline_id)
             .join(METADATA_FILE_NAME)
     }
 

pageserver/src/consumption_metrics.rs

@@ -24,6 +24,8 @@ const RESIDENT_SIZE: &str = "resident_size";
 const REMOTE_STORAGE_SIZE: &str = "remote_storage_size";
 const TIMELINE_LOGICAL_SIZE: &str = "timeline_logical_size";
 
+const DEFAULT_HTTP_REPORTING_TIMEOUT: Duration = Duration::from_secs(60);
+
 #[serde_as]
 #[derive(Serialize, Debug)]
 struct Ids {
@@ -73,7 +75,10 @@ pub async fn collect_metrics(
     );
 
     // define client here to reuse it for all requests
-    let client = reqwest::Client::new();
+    let client = reqwest::ClientBuilder::new()
+        .timeout(DEFAULT_HTTP_REPORTING_TIMEOUT)
+        .build()
+        .expect("Failed to create http client with timeout");
     let mut cached_metrics: HashMap<PageserverConsumptionMetricsKey, u64> = HashMap::new();
     let mut prev_iteration_time: std::time::Instant = std::time::Instant::now();
 
@@ -83,7 +88,7 @@ pub async fn collect_metrics(
                 info!("collect_metrics received cancellation request");
                 return Ok(());
             },
-            _ = ticker.tick() => {
+            tick_at = ticker.tick() => {
 
                 // send cached metrics every cached_metric_collection_interval
                 let send_cached = prev_iteration_time.elapsed() >= cached_metric_collection_interval;
@@ -93,6 +98,12 @@ pub async fn collect_metrics(
                 }
 
                 collect_metrics_iteration(&client, &mut cached_metrics, metric_collection_endpoint, node_id, &ctx, send_cached).await;
+
+                crate::tenant::tasks::warn_when_period_overrun(
+                    tick_at.elapsed(),
+                    metric_collection_interval,
+                    "consumption_metrics_collect_metrics",
+                );
             }
         }
     }
@@ -223,14 +234,18 @@ pub async fn collect_metrics_iteration(
         // Note that this metric is calculated in a separate bgworker
         // Here we only use cached value, which may lag behind the real latest one
         let tenant_synthetic_size = tenant.get_cached_synthetic_size();
-        current_metrics.push((
-            PageserverConsumptionMetricsKey {
-                tenant_id,
-                timeline_id: None,
-                metric: SYNTHETIC_STORAGE_SIZE,
-            },
-            tenant_synthetic_size,
-        ));
+
+        if tenant_synthetic_size != 0 {
+            // only send non-zeroes because otherwise these show up as errors in logs
+            current_metrics.push((
+                PageserverConsumptionMetricsKey {
+                    tenant_id,
+                    timeline_id: None,
+                    metric: SYNTHETIC_STORAGE_SIZE,
+                },
+                tenant_synthetic_size,
+            ));
+        }
     }
 
     // Filter metrics, unless we want to send all metrics, including cached ones.
@@ -273,31 +288,42 @@ pub async fn collect_metrics_iteration(
         })
         .expect("PageserverConsumptionMetric should not fail serialization");
 
-        let res = client
-            .post(metric_collection_endpoint.clone())
-            .json(&chunk_json)
-            .send()
-            .await;
+        const MAX_RETRIES: u32 = 3;
 
-        match res {
-            Ok(res) => {
-                if res.status().is_success() {
-                    // update cached metrics after they were sent successfully
-                    for (curr_key, curr_val) in chunk.iter() {
-                        cached_metrics.insert(curr_key.clone(), *curr_val);
-                    }
-                } else {
-                    error!("metrics endpoint refused the sent metrics: {:?}", res);
-                    for metric in chunk_to_send.iter() {
-                        // Report if the metric value is suspiciously large
-                        if metric.value > (1u64 << 40) {
+        for attempt in 0..MAX_RETRIES {
+            let res = client
+                .post(metric_collection_endpoint.clone())
+                .json(&chunk_json)
+                .send()
+                .await;
+
+            match res {
+                Ok(res) => {
+                    if res.status().is_success() {
+                        // update cached metrics after they were sent successfully
+                        for (curr_key, curr_val) in chunk.iter() {
+                            cached_metrics.insert(curr_key.clone(), *curr_val);
+                        }
+                    } else {
+                        error!("metrics endpoint refused the sent metrics: {:?}", res);
+                        for metric in chunk_to_send
+                            .iter()
+                            .filter(|metric| metric.value > (1u64 << 40))
+                        {
+                            // Report if the metric value is suspiciously large
                             error!("potentially abnormal metric value: {:?}", metric);
                         }
                     }
+                    break;
+                }
+                Err(err) if err.is_timeout() => {
+                    error!(attempt, "timeout sending metrics, retrying immediately");
+                    continue;
+                }
+                Err(err) => {
+                    error!(attempt, ?err, "failed to send metrics");
+                    break;
                 }
-            }
-            Err(err) => {
-                error!("failed to send metrics: {:?}", err);
             }
         }
     }
@@ -317,7 +343,7 @@ pub async fn calculate_synthetic_size_worker(
             _ = task_mgr::shutdown_watcher() => {
                 return Ok(());
             },
-        _ = ticker.tick() => {
+        tick_at = ticker.tick() => {
 
                 let tenants = match mgr::list_tenants().await {
                     Ok(tenants) => tenants,
@@ -343,6 +369,12 @@ pub async fn calculate_synthetic_size_worker(
                     }
 
                 }
+
+                crate::tenant::tasks::warn_when_period_overrun(
+                    tick_at.elapsed(),
+                    synthetic_size_calculation_interval,
+                    "consumption_metrics_synthetic_size_worker",
+                );
             }
         }
     }

pageserver/src/disk_usage_eviction_task.rs

@@ -110,7 +110,6 @@ pub fn launch_disk_usage_global_eviction_task(
 
             disk_usage_eviction_task(&state, task_config, storage, &conf.tenants_path(), cancel)
                 .await;
-            info!("disk usage based eviction task finishing");
             Ok(())
         },
     );
@@ -126,13 +125,16 @@ async fn disk_usage_eviction_task(
     tenants_dir: &Path,
     cancel: CancellationToken,
 ) {
+    scopeguard::defer! {
+        info!("disk usage based eviction task finishing");
+    };
+
     use crate::tenant::tasks::random_init_delay;
     {
         if random_init_delay(task_config.period, &cancel)
             .await
             .is_err()
         {
-            info!("shutting down");
             return;
         }
     }
@@ -167,7 +169,6 @@ async fn disk_usage_eviction_task(
         tokio::select! {
             _ = tokio::time::sleep_until(sleep_until) => {},
             _ = cancel.cancelled() => {
-                info!("shutting down");
                 break
             }
         }
@@ -304,7 +305,7 @@ pub async fn disk_usage_eviction_task_iteration_impl<U: Usage>(
     let now = SystemTime::now();
     for (i, (partition, candidate)) in candidates.iter().enumerate() {
         debug!(
-            "cand {}/{}: size={}, no_access_for={}us, parition={:?}, tenant={} timeline={} layer={}",
+            "cand {}/{}: size={}, no_access_for={}us, partition={:?}, {}/{}/{}",
             i + 1,
             candidates.len(),
             candidate.layer.file_size(),
@@ -314,7 +315,7 @@ pub async fn disk_usage_eviction_task_iteration_impl<U: Usage>(
             partition,
             candidate.layer.get_tenant_id(),
             candidate.layer.get_timeline_id(),
-            candidate.layer.filename().file_name(),
+            candidate.layer,
         );
     }
 

pageserver/src/http/openapi_spec.yml

@@ -186,10 +186,8 @@ paths:
               schema:
                 $ref: "#/components/schemas/Error"
     delete:
-      description: "Attempts to delete specified timeline. On 500 errors should be retried"
+      description: "Attempts to delete specified timeline. 500 and 409 errors should be retried"
       responses:
-        "200":
-          description: Ok
         "400":
           description: Error when no tenant id found in path or no timeline id
           content:
@@ -214,6 +212,12 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/NotFoundError"
+        "409":
+          description: Deletion is already in progress, continue polling
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/ConflictError"
         "412":
           description: Tenant is missing, or timeline has children
           content:
@@ -718,6 +722,12 @@ paths:
             application/json:
               schema:
                 $ref: "#/components/schemas/ForbiddenError"
+        "406":
+          description: Permanently unsatisfiable request, don't retry.
+          content:
+            application/json:
+              schema:
+                $ref: "#/components/schemas/Error"
         "409":
           description: Timeline already exists, creation skipped
           content:

pageserver/src/http/routes.rs

@@ -23,7 +23,6 @@ use super::models::{
     TimelineCreateRequest, TimelineGcRequest, TimelineInfo,
 };
 use crate::context::{DownloadBehavior, RequestContext};
-use crate::disk_usage_eviction_task;
 use crate::metrics::{StorageTimeOperation, STORAGE_TIME_GLOBAL};
 use crate::pgdatadir_mapping::LsnForTimestamp;
 use crate::task_mgr::TaskKind;
@@ -35,6 +34,7 @@ use crate::tenant::size::ModelInputs;
 use crate::tenant::storage_layer::LayerAccessStatsReset;
 use crate::tenant::{LogicalSizeCalculationCause, PageReconstructError, Timeline};
 use crate::{config::PageServerConf, tenant::mgr};
+use crate::{disk_usage_eviction_task, tenant};
 use utils::{
     auth::JwtAuth,
     http::{
@@ -142,7 +142,7 @@ impl From<TenantMapInsertError> for ApiError {
 impl From<TenantStateError> for ApiError {
     fn from(tse: TenantStateError) -> ApiError {
         match tse {
-            TenantStateError::NotFound(tid) => ApiError::NotFound(anyhow!("tenant {}", tid)),
+            TenantStateError::NotFound(tid) => ApiError::NotFound(anyhow!("tenant {}", tid).into()),
             _ => ApiError::InternalServerError(anyhow::Error::new(tse)),
         }
     }
@@ -151,7 +151,7 @@ impl From<TenantStateError> for ApiError {
 impl From<GetTenantError> for ApiError {
     fn from(tse: GetTenantError) -> ApiError {
         match tse {
-            GetTenantError::NotFound(tid) => ApiError::NotFound(anyhow!("tenant {}", tid)),
+            GetTenantError::NotFound(tid) => ApiError::NotFound(anyhow!("tenant {}", tid).into()),
             e @ GetTenantError::NotActive(_) => {
                 // Why is this not `ApiError::NotFound`?
                 // Because we must be careful to never return 404 for a tenant if it does
@@ -169,7 +169,7 @@ impl From<SetNewTenantConfigError> for ApiError {
     fn from(e: SetNewTenantConfigError) -> ApiError {
         match e {
             SetNewTenantConfigError::GetTenant(tid) => {
-                ApiError::NotFound(anyhow!("tenant {}", tid))
+                ApiError::NotFound(anyhow!("tenant {}", tid).into())
             }
             e @ SetNewTenantConfigError::Persist(_) => {
                 ApiError::InternalServerError(anyhow::Error::new(e))
@@ -182,11 +182,12 @@ impl From<crate::tenant::DeleteTimelineError> for ApiError {
     fn from(value: crate::tenant::DeleteTimelineError) -> Self {
         use crate::tenant::DeleteTimelineError::*;
         match value {
-            NotFound => ApiError::NotFound(anyhow::anyhow!("timeline not found")),
+            NotFound => ApiError::NotFound(anyhow::anyhow!("timeline not found").into()),
             HasChildren(children) => ApiError::PreconditionFailed(
                 format!("Cannot delete timeline which has child timelines: {children:?}")
                     .into_boxed_str(),
             ),
+            a @ AlreadyInProgress => ApiError::Conflict(a.to_string()),
             Other(e) => ApiError::InternalServerError(e),
         }
     }
@@ -327,18 +328,25 @@ async fn timeline_create_handler(
             &ctx,
         )
         .await {
-            Ok(Some(new_timeline)) => {
+            Ok(new_timeline) => {
                 // Created. Construct a TimelineInfo for it.
                 let timeline_info = build_timeline_info_common(&new_timeline, &ctx)
                     .await
                     .map_err(ApiError::InternalServerError)?;
                 json_response(StatusCode::CREATED, timeline_info)
             }
-            Ok(None) => json_response(StatusCode::CONFLICT, ()), // timeline already exists
-            Err(err) => Err(ApiError::InternalServerError(err)),
+            Err(tenant::CreateTimelineError::AlreadyExists) => {
+                json_response(StatusCode::CONFLICT, ())
+            }
+            Err(tenant::CreateTimelineError::AncestorLsn(err)) => {
+                json_response(StatusCode::NOT_ACCEPTABLE, HttpErrorBody::from_msg(
+                    format!("{err:#}")
+                ))
+            }
+            Err(tenant::CreateTimelineError::Other(err)) => Err(ApiError::InternalServerError(err)),
         }
     }
-    .instrument(info_span!("timeline_create", tenant = %tenant_id, timeline_id = %new_timeline_id, lsn=?request_data.ancestor_start_lsn, pg_version=?request_data.pg_version))
+    .instrument(info_span!("timeline_create", %tenant_id, timeline_id = %new_timeline_id, lsn=?request_data.ancestor_start_lsn, pg_version=?request_data.pg_version))
     .await
 }
 
@@ -373,7 +381,7 @@ async fn timeline_list_handler(
         }
         Ok::<Vec<TimelineInfo>, ApiError>(response_data)
     }
-    .instrument(info_span!("timeline_list", tenant = %tenant_id))
+    .instrument(info_span!("timeline_list", %tenant_id))
     .await?;
 
     json_response(StatusCode::OK, response_data)
@@ -397,7 +405,7 @@ async fn timeline_detail_handler(
 
         let timeline = tenant
             .get_timeline(timeline_id, false)
-            .map_err(ApiError::NotFound)?;
+            .map_err(|e| ApiError::NotFound(e.into()))?;
 
         let timeline_info = build_timeline_info(
             &timeline,
@@ -410,7 +418,7 @@ async fn timeline_detail_handler(
 
         Ok::<_, ApiError>(timeline_info)
     }
-    .instrument(info_span!("timeline_detail", tenant = %tenant_id, timeline = %timeline_id))
+    .instrument(info_span!("timeline_detail", %tenant_id, %timeline_id))
     .await?;
 
     json_response(StatusCode::OK, timeline_info)
@@ -471,7 +479,7 @@ async fn tenant_attach_handler(
             remote_storage.clone(),
             &ctx,
         )
-        .instrument(info_span!("tenant_attach", tenant = %tenant_id))
+        .instrument(info_span!("tenant_attach", %tenant_id))
         .await?;
     } else {
         return Err(ApiError::BadRequest(anyhow!(
@@ -493,7 +501,7 @@ async fn timeline_delete_handler(
     let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
 
     mgr::delete_timeline(tenant_id, timeline_id, &ctx)
-        .instrument(info_span!("timeline_delete", tenant = %tenant_id, timeline = %timeline_id))
+        .instrument(info_span!("timeline_delete", %tenant_id, %timeline_id))
         .await?;
 
     // FIXME: needs to be an error for console to retry it. Ideally Accepted should be used and retried until 404.
@@ -511,7 +519,7 @@ async fn tenant_detach_handler(
     let state = get_state(&request);
     let conf = state.conf;
     mgr::detach_tenant(conf, tenant_id, detach_ignored.unwrap_or(false))
-        .instrument(info_span!("tenant_detach", tenant = %tenant_id))
+        .instrument(info_span!("tenant_detach", %tenant_id))
         .await?;
 
     json_response(StatusCode::OK, ())
@@ -534,7 +542,7 @@ async fn tenant_load_handler(
         state.remote_storage.clone(),
         &ctx,
     )
-    .instrument(info_span!("load", tenant = %tenant_id))
+    .instrument(info_span!("load", %tenant_id))
     .await?;
 
     json_response(StatusCode::ACCEPTED, ())
@@ -550,7 +558,7 @@ async fn tenant_ignore_handler(
     let state = get_state(&request);
     let conf = state.conf;
     mgr::ignore_tenant(conf, tenant_id)
-        .instrument(info_span!("ignore_tenant", tenant = %tenant_id))
+        .instrument(info_span!("ignore_tenant", %tenant_id))
         .await?;
 
     json_response(StatusCode::OK, ())
@@ -603,7 +611,7 @@ async fn tenant_status(
             attachment_status: state.attachment_status(),
         })
     }
-    .instrument(info_span!("tenant_status_handler", tenant = %tenant_id))
+    .instrument(info_span!("tenant_status_handler", %tenant_id))
     .await?;
 
     json_response(StatusCode::OK, tenant_info)
@@ -842,7 +850,7 @@ async fn tenant_create_handler(
         state.remote_storage.clone(),
         &ctx,
     )
-    .instrument(info_span!("tenant_create", tenant = ?target_tenant_id))
+    .instrument(info_span!("tenant_create", tenant_id = %target_tenant_id))
     .await?;
 
     // We created the tenant. Existing API semantics are that the tenant
@@ -904,7 +912,7 @@ async fn update_tenant_config_handler(
 
     let state = get_state(&request);
     mgr::set_new_tenant_config(state.conf, tenant_conf, tenant_id)
-        .instrument(info_span!("tenant_config", tenant = ?tenant_id))
+        .instrument(info_span!("tenant_config", %tenant_id))
         .await?;
 
     json_response(StatusCode::OK, ())
@@ -1061,7 +1069,7 @@ async fn timeline_download_remote_layers_handler_get(
     let info = timeline
         .get_download_all_remote_layers_task_info()
         .context("task never started since last pageserver process start")
-        .map_err(ApiError::NotFound)?;
+        .map_err(|e| ApiError::NotFound(e.into()))?;
     json_response(StatusCode::OK, info)
 }
 
@@ -1072,7 +1080,7 @@ async fn active_timeline_of_active_tenant(
     let tenant = mgr::get_tenant(tenant_id, true).await?;
     tenant
         .get_timeline(timeline_id, true)
-        .map_err(ApiError::NotFound)
+        .map_err(|e| ApiError::NotFound(e.into()))
 }
 
 async fn always_panic_handler(
@@ -1128,8 +1136,6 @@ async fn disk_usage_eviction_run(
         freed_bytes: 0,
     };
 
-    use crate::task_mgr::MGMT_REQUEST_RUNTIME;
-
     let (tx, rx) = tokio::sync::oneshot::channel();
 
     let state = get_state(&r);
@@ -1147,7 +1153,7 @@ async fn disk_usage_eviction_run(
     let _g = cancel.drop_guard();
 
     crate::task_mgr::spawn(
-        MGMT_REQUEST_RUNTIME.handle(),
+        crate::task_mgr::BACKGROUND_RUNTIME.handle(),
         TaskKind::DiskUsageEviction,
         None,
         None,

pageserver/src/import_datadir.rs

@@ -148,17 +148,17 @@ async fn import_rel(
     // because there is no guarantee about the order in which we are processing segments.
     // ignore "relation already exists" error
     //
-    // FIXME: use proper error type for this, instead of parsing the error message.
-    // Or better yet, keep track of which relations we've already created
+    // FIXME: Keep track of which relations we've already created?
     // https://github.com/neondatabase/neon/issues/3309
     if let Err(e) = modification
         .put_rel_creation(rel, nblocks as u32, ctx)
         .await
     {
-        if e.to_string().contains("already exists") {
-            debug!("relation {} already exists. we must be extending it", rel);
-        } else {
-            return Err(e);
+        match e {
+            RelationError::AlreadyExists => {
+                debug!("Relation {} already exist. We must be extending it.", rel)
+            }
+            _ => return Err(e.into()),
         }
     }
 

pageserver/src/metrics.rs

@@ -1,8 +1,9 @@
+use metrics::metric_vec_duration::DurationResultObserver;
 use metrics::{
     register_counter_vec, register_histogram, register_histogram_vec, register_int_counter,
-    register_int_counter_vec, register_int_gauge, register_int_gauge_vec, register_uint_gauge_vec,
-    Counter, CounterVec, Histogram, HistogramVec, IntCounter, IntCounterVec, IntGauge, IntGaugeVec,
-    UIntGauge, UIntGaugeVec,
+    register_int_counter_vec, register_int_gauge, register_int_gauge_vec, register_uint_gauge,
+    register_uint_gauge_vec, Counter, CounterVec, Histogram, HistogramVec, IntCounter,
+    IntCounterVec, IntGauge, IntGaugeVec, UIntGauge, UIntGaugeVec,
 };
 use once_cell::sync::Lazy;
 use pageserver_api::models::TenantState;
@@ -129,6 +130,122 @@ pub static MATERIALIZED_PAGE_CACHE_HIT: Lazy<IntCounter> = Lazy::new(|| {
     .expect("failed to define a metric")
 });
 
+pub struct PageCacheMetrics {
+    pub read_accesses_materialized_page: IntCounter,
+    pub read_accesses_ephemeral: IntCounter,
+    pub read_accesses_immutable: IntCounter,
+
+    pub read_hits_ephemeral: IntCounter,
+    pub read_hits_immutable: IntCounter,
+    pub read_hits_materialized_page_exact: IntCounter,
+    pub read_hits_materialized_page_older_lsn: IntCounter,
+}
+
+static PAGE_CACHE_READ_HITS: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
+        "pageserver_page_cache_read_hits_total",
+        "Number of read accesses to the page cache that hit",
+        &["key_kind", "hit_kind"]
+    )
+    .expect("failed to define a metric")
+});
+
+static PAGE_CACHE_READ_ACCESSES: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
+        "pageserver_page_cache_read_accesses_total",
+        "Number of read accesses to the page cache",
+        &["key_kind"]
+    )
+    .expect("failed to define a metric")
+});
+
+pub static PAGE_CACHE: Lazy<PageCacheMetrics> = Lazy::new(|| PageCacheMetrics {
+    read_accesses_materialized_page: {
+        PAGE_CACHE_READ_ACCESSES
+            .get_metric_with_label_values(&["materialized_page"])
+            .unwrap()
+    },
+
+    read_accesses_ephemeral: {
+        PAGE_CACHE_READ_ACCESSES
+            .get_metric_with_label_values(&["ephemeral"])
+            .unwrap()
+    },
+
+    read_accesses_immutable: {
+        PAGE_CACHE_READ_ACCESSES
+            .get_metric_with_label_values(&["immutable"])
+            .unwrap()
+    },
+
+    read_hits_ephemeral: {
+        PAGE_CACHE_READ_HITS
+            .get_metric_with_label_values(&["ephemeral", "-"])
+            .unwrap()
+    },
+
+    read_hits_immutable: {
+        PAGE_CACHE_READ_HITS
+            .get_metric_with_label_values(&["immutable", "-"])
+            .unwrap()
+    },
+
+    read_hits_materialized_page_exact: {
+        PAGE_CACHE_READ_HITS
+            .get_metric_with_label_values(&["materialized_page", "exact"])
+            .unwrap()
+    },
+
+    read_hits_materialized_page_older_lsn: {
+        PAGE_CACHE_READ_HITS
+            .get_metric_with_label_values(&["materialized_page", "older_lsn"])
+            .unwrap()
+    },
+});
+
+pub struct PageCacheSizeMetrics {
+    pub max_bytes: UIntGauge,
+
+    pub current_bytes_ephemeral: UIntGauge,
+    pub current_bytes_immutable: UIntGauge,
+    pub current_bytes_materialized_page: UIntGauge,
+}
+
+static PAGE_CACHE_SIZE_CURRENT_BYTES: Lazy<UIntGaugeVec> = Lazy::new(|| {
+    register_uint_gauge_vec!(
+        "pageserver_page_cache_size_current_bytes",
+        "Current size of the page cache in bytes, by key kind",
+        &["key_kind"]
+    )
+    .expect("failed to define a metric")
+});
+
+pub static PAGE_CACHE_SIZE: Lazy<PageCacheSizeMetrics> = Lazy::new(|| PageCacheSizeMetrics {
+    max_bytes: {
+        register_uint_gauge!(
+            "pageserver_page_cache_size_max_bytes",
+            "Maximum size of the page cache in bytes"
+        )
+        .expect("failed to define a metric")
+    },
+
+    current_bytes_ephemeral: {
+        PAGE_CACHE_SIZE_CURRENT_BYTES
+            .get_metric_with_label_values(&["ephemeral"])
+            .unwrap()
+    },
+    current_bytes_immutable: {
+        PAGE_CACHE_SIZE_CURRENT_BYTES
+            .get_metric_with_label_values(&["immutable"])
+            .unwrap()
+    },
+    current_bytes_materialized_page: {
+        PAGE_CACHE_SIZE_CURRENT_BYTES
+            .get_metric_with_label_values(&["materialized_page"])
+            .unwrap()
+    },
+});
+
 static WAIT_LSN_TIME: Lazy<HistogramVec> = Lazy::new(|| {
     register_histogram_vec!(
         "pageserver_wait_lsn_seconds",
@@ -203,11 +320,11 @@ pub static TENANT_STATE_METRIC: Lazy<UIntGaugeVec> = Lazy::new(|| {
 
 pub static TENANT_SYNTHETIC_SIZE_METRIC: Lazy<UIntGaugeVec> = Lazy::new(|| {
     register_uint_gauge_vec!(
-        "pageserver_tenant_synthetic_size",
-        "Synthetic size of each tenant",
+        "pageserver_tenant_synthetic_cached_size_bytes",
+        "Synthetic size of each tenant in bytes",
         &["tenant_id"]
     )
-    .expect("Failed to register pageserver_tenant_synthetic_size metric")
+    .expect("Failed to register pageserver_tenant_synthetic_cached_size_bytes metric")
 });
 
 // Metrics for cloud upload. These metrics reflect data uploaded to cloud storage,
@@ -424,6 +541,27 @@ pub static SMGR_QUERY_TIME: Lazy<HistogramVec> = Lazy::new(|| {
     .expect("failed to define a metric")
 });
 
+pub struct BasebackupQueryTime(HistogramVec);
+pub static BASEBACKUP_QUERY_TIME: Lazy<BasebackupQueryTime> = Lazy::new(|| {
+    BasebackupQueryTime({
+        register_histogram_vec!(
+            "pageserver_basebackup_query_seconds",
+            "Histogram of basebackup queries durations, by result type",
+            &["result"],
+            CRITICAL_OP_BUCKETS.into(),
+        )
+        .expect("failed to define a metric")
+    })
+});
+
+impl DurationResultObserver for BasebackupQueryTime {
+    fn observe_result<T, E>(&self, res: &Result<T, E>, duration: std::time::Duration) {
+        let label_value = if res.is_ok() { "ok" } else { "error" };
+        let metric = self.0.get_metric_with_label_values(&[label_value]).unwrap();
+        metric.observe(duration.as_secs_f64());
+    }
+}
+
 pub static LIVE_CONNECTIONS_COUNT: Lazy<IntGaugeVec> = Lazy::new(|| {
     register_int_gauge_vec!(
         "pageserver_live_connections",
@@ -823,11 +961,6 @@ impl TimelineMetrics {
         let evictions_with_low_residence_duration =
             evictions_with_low_residence_duration_builder.build(&tenant_id, &timeline_id);
 
-        // TODO(chi): remove this once we remove Lazy for all metrics. Otherwise this will not appear in the exporter
-        // and integration test will error.
-        MATERIALIZED_PAGE_CACHE_HIT_DIRECT.get();
-        MATERIALIZED_PAGE_CACHE_HIT.get();
-
         TimelineMetrics {
             tenant_id,
             timeline_id,
@@ -951,7 +1084,6 @@ impl RemoteTimelineClientMetrics {
         op_kind: &RemoteOpKind,
         status: &'static str,
     ) -> Histogram {
-        // XXX would be nice to have an upgradable RwLock
         let mut guard = self.remote_operation_time.lock().unwrap();
         let key = (file_kind.as_str(), op_kind.as_str(), status);
         let metric = guard.entry(key).or_insert_with(move || {
@@ -973,7 +1105,6 @@ impl RemoteTimelineClientMetrics {
         file_kind: &RemoteOpFileKind,
         op_kind: &RemoteOpKind,
     ) -> IntGauge {
-        // XXX would be nice to have an upgradable RwLock
         let mut guard = self.calls_unfinished_gauge.lock().unwrap();
         let key = (file_kind.as_str(), op_kind.as_str());
         let metric = guard.entry(key).or_insert_with(move || {
@@ -994,7 +1125,6 @@ impl RemoteTimelineClientMetrics {
         file_kind: &RemoteOpFileKind,
         op_kind: &RemoteOpKind,
     ) -> Histogram {
-        // XXX would be nice to have an upgradable RwLock
         let mut guard = self.calls_started_hist.lock().unwrap();
         let key = (file_kind.as_str(), op_kind.as_str());
         let metric = guard.entry(key).or_insert_with(move || {
@@ -1015,7 +1145,6 @@ impl RemoteTimelineClientMetrics {
         file_kind: &RemoteOpFileKind,
         op_kind: &RemoteOpKind,
     ) -> IntCounter {
-        // XXX would be nice to have an upgradable RwLock
         let mut guard = self.bytes_started_counter.lock().unwrap();
         let key = (file_kind.as_str(), op_kind.as_str());
         let metric = guard.entry(key).or_insert_with(move || {
@@ -1036,7 +1165,6 @@ impl RemoteTimelineClientMetrics {
         file_kind: &RemoteOpFileKind,
         op_kind: &RemoteOpKind,
     ) -> IntCounter {
-        // XXX would be nice to have an upgradable RwLock
         let mut guard = self.bytes_finished_counter.lock().unwrap();
         let key = (file_kind.as_str(), op_kind.as_str());
         let metric = guard.entry(key).or_insert_with(move || {
@@ -1302,4 +1430,8 @@ pub fn preinitialize_metrics() {
 
     // Same as above for this metric, but, it's a Vec-type metric for which we don't know all the labels.
     BACKGROUND_LOOP_PERIOD_OVERRUN_COUNT.reset();
+
+    // Python tests need these.
+    MATERIALIZED_PAGE_CACHE_HIT_DIRECT.get();
+    MATERIALIZED_PAGE_CACHE_HIT.get();
 }

pageserver/src/page_cache.rs

@@ -53,8 +53,8 @@ use utils::{
     lsn::Lsn,
 };
 
-use crate::repository::Key;
 use crate::tenant::writeback_ephemeral_file;
+use crate::{metrics::PageCacheSizeMetrics, repository::Key};
 
 static PAGE_CACHE: OnceCell<PageCache> = OnceCell::new();
 const TEST_PAGE_CACHE_SIZE: usize = 50;
@@ -187,6 +187,8 @@ pub struct PageCache {
     /// Index of the next candidate to evict, for the Clock replacement algorithm.
     /// This is interpreted modulo the page cache size.
     next_evict_slot: AtomicUsize,
+
+    size_metrics: &'static PageCacheSizeMetrics,
 }
 
 ///
@@ -313,6 +315,10 @@ impl PageCache {
         key: &Key,
         lsn: Lsn,
     ) -> Option<(Lsn, PageReadGuard)> {
+        crate::metrics::PAGE_CACHE
+            .read_accesses_materialized_page
+            .inc();
+
         let mut cache_key = CacheKey::MaterializedPage {
             hash_key: MaterializedPageHashKey {
                 tenant_id,
@@ -323,8 +329,21 @@ impl PageCache {
         };
 
         if let Some(guard) = self.try_lock_for_read(&mut cache_key) {
-            if let CacheKey::MaterializedPage { hash_key: _, lsn } = cache_key {
-                Some((lsn, guard))
+            if let CacheKey::MaterializedPage {
+                hash_key: _,
+                lsn: available_lsn,
+            } = cache_key
+            {
+                if available_lsn == lsn {
+                    crate::metrics::PAGE_CACHE
+                        .read_hits_materialized_page_exact
+                        .inc();
+                } else {
+                    crate::metrics::PAGE_CACHE
+                        .read_hits_materialized_page_older_lsn
+                        .inc();
+                }
+                Some((available_lsn, guard))
             } else {
                 panic!("unexpected key type in slot");
             }
@@ -499,11 +518,31 @@ impl PageCache {
     /// ```
     ///
     fn lock_for_read(&self, cache_key: &mut CacheKey) -> anyhow::Result<ReadBufResult> {
+        let (read_access, hit) = match cache_key {
+            CacheKey::MaterializedPage { .. } => {
+                unreachable!("Materialized pages use lookup_materialized_page")
+            }
+            CacheKey::EphemeralPage { .. } => (
+                &crate::metrics::PAGE_CACHE.read_accesses_ephemeral,
+                &crate::metrics::PAGE_CACHE.read_hits_ephemeral,
+            ),
+            CacheKey::ImmutableFilePage { .. } => (
+                &crate::metrics::PAGE_CACHE.read_accesses_immutable,
+                &crate::metrics::PAGE_CACHE.read_hits_immutable,
+            ),
+        };
+        read_access.inc();
+
+        let mut is_first_iteration = true;
         loop {
             // First check if the key already exists in the cache.
             if let Some(read_guard) = self.try_lock_for_read(cache_key) {
+                if is_first_iteration {
+                    hit.inc();
+                }
                 return Ok(ReadBufResult::Found(read_guard));
             }
+            is_first_iteration = false;
 
             // Not found. Find a victim buffer
             let (slot_idx, mut inner) =
@@ -681,6 +720,9 @@ impl PageCache {
 
                     if let Ok(version_idx) = versions.binary_search_by_key(old_lsn, |v| v.lsn) {
                         versions.remove(version_idx);
+                        self.size_metrics
+                            .current_bytes_materialized_page
+                            .sub_page_sz(1);
                         if versions.is_empty() {
                             old_entry.remove_entry();
                         }
@@ -693,11 +735,13 @@ impl PageCache {
                 let mut map = self.ephemeral_page_map.write().unwrap();
                 map.remove(&(*file_id, *blkno))
                     .expect("could not find old key in mapping");
+                self.size_metrics.current_bytes_ephemeral.sub_page_sz(1);
             }
             CacheKey::ImmutableFilePage { file_id, blkno } => {
                 let mut map = self.immutable_page_map.write().unwrap();
                 map.remove(&(*file_id, *blkno))
                     .expect("could not find old key in mapping");
+                self.size_metrics.current_bytes_immutable.sub_page_sz(1);
             }
         }
     }
@@ -725,6 +769,9 @@ impl PageCache {
                                 slot_idx,
                             },
                         );
+                        self.size_metrics
+                            .current_bytes_materialized_page
+                            .add_page_sz(1);
                         None
                     }
                 }
@@ -735,6 +782,7 @@ impl PageCache {
                     Entry::Occupied(entry) => Some(*entry.get()),
                     Entry::Vacant(entry) => {
                         entry.insert(slot_idx);
+                        self.size_metrics.current_bytes_ephemeral.add_page_sz(1);
                         None
                     }
                 }
@@ -745,6 +793,7 @@ impl PageCache {
                     Entry::Occupied(entry) => Some(*entry.get()),
                     Entry::Vacant(entry) => {
                         entry.insert(slot_idx);
+                        self.size_metrics.current_bytes_immutable.add_page_sz(1);
                         None
                     }
                 }
@@ -844,6 +893,12 @@ impl PageCache {
 
         let page_buffer = Box::leak(vec![0u8; num_pages * PAGE_SZ].into_boxed_slice());
 
+        let size_metrics = &crate::metrics::PAGE_CACHE_SIZE;
+        size_metrics.max_bytes.set_page_sz(num_pages);
+        size_metrics.current_bytes_ephemeral.set_page_sz(0);
+        size_metrics.current_bytes_immutable.set_page_sz(0);
+        size_metrics.current_bytes_materialized_page.set_page_sz(0);
+
         let slots = page_buffer
             .chunks_exact_mut(PAGE_SZ)
             .map(|chunk| {
@@ -866,6 +921,30 @@ impl PageCache {
             immutable_page_map: Default::default(),
             slots,
             next_evict_slot: AtomicUsize::new(0),
+            size_metrics,
         }
     }
 }
+
+trait PageSzBytesMetric {
+    fn set_page_sz(&self, count: usize);
+    fn add_page_sz(&self, count: usize);
+    fn sub_page_sz(&self, count: usize);
+}
+
+#[inline(always)]
+fn count_times_page_sz(count: usize) -> u64 {
+    u64::try_from(count).unwrap() * u64::try_from(PAGE_SZ).unwrap()
+}
+
+impl PageSzBytesMetric for metrics::UIntGauge {
+    fn set_page_sz(&self, count: usize) {
+        self.set(count_times_page_sz(count));
+    }
+    fn add_page_sz(&self, count: usize) {
+        self.add(count_times_page_sz(count));
+    }
+    fn sub_page_sz(&self, count: usize) {
+        self.sub(count_times_page_sz(count));
+    }
+}

pageserver/src/page_service.rs

@@ -10,6 +10,7 @@
 //
 
 use anyhow::Context;
+use async_compression::tokio::write::GzipEncoder;
 use bytes::Buf;
 use bytes::Bytes;
 use futures::Stream;
@@ -31,8 +32,10 @@ use std::str;
 use std::str::FromStr;
 use std::sync::Arc;
 use std::time::Duration;
+use tokio::io::AsyncWriteExt;
 use tokio::io::{AsyncRead, AsyncWrite};
 use tokio_util::io::StreamReader;
+use tracing::field;
 use tracing::*;
 use utils::id::ConnectionId;
 use utils::{
@@ -51,6 +54,7 @@ use crate::metrics::{LIVE_CONNECTIONS_COUNT, SMGR_QUERY_TIME};
 use crate::task_mgr;
 use crate::task_mgr::TaskKind;
 use crate::tenant;
+use crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id;
 use crate::tenant::mgr;
 use crate::tenant::mgr::GetTenantError;
 use crate::tenant::{Tenant, Timeline};
@@ -238,6 +242,7 @@ pub async fn libpq_listener_main(
     Ok(())
 }
 
+#[instrument(skip_all, fields(peer_addr))]
 async fn page_service_conn_main(
     conf: &'static PageServerConf,
     broker_client: storage_broker::BrokerClientChannel,
@@ -260,6 +265,7 @@ async fn page_service_conn_main(
         .context("could not set TCP_NODELAY")?;
 
     let peer_addr = socket.peer_addr().context("get peer address")?;
+    tracing::Span::current().record("peer_addr", field::display(peer_addr));
 
     // setup read timeout of 10 minutes. the timeout is rather arbitrary for requirements:
     // - long enough for most valid compute connections
@@ -362,7 +368,7 @@ impl PageServerHandler {
         }
     }
 
-    #[instrument(skip(self, pgb, ctx))]
+    #[instrument(skip_all)]
     async fn handle_pagerequests<IO>(
         &self,
         pgb: &mut PostgresBackend<IO>,
@@ -373,6 +379,8 @@ impl PageServerHandler {
     where
         IO: AsyncRead + AsyncWrite + Send + Sync + Unpin,
     {
+        debug_assert_current_span_has_tenant_and_timeline_id();
+
         // NOTE: pagerequests handler exits when connection is closed,
         //       so there is no need to reset the association
         task_mgr::associate_with(Some(tenant_id), Some(timeline_id));
@@ -390,7 +398,9 @@ impl PageServerHandler {
         };
 
         // Check that the timeline exists
-        let timeline = tenant.get_timeline(timeline_id, true)?;
+        let timeline = tenant
+            .get_timeline(timeline_id, true)
+            .map_err(|e| anyhow::anyhow!(e))?;
 
         // switch client to COPYBOTH
         pgb.write_message_noflush(&BeMessage::CopyBothResponse)?;
@@ -471,7 +481,7 @@ impl PageServerHandler {
     }
 
     #[allow(clippy::too_many_arguments)]
-    #[instrument(skip(self, pgb, ctx))]
+    #[instrument(skip_all, fields(%base_lsn, end_lsn=%_end_lsn, %pg_version))]
     async fn handle_import_basebackup<IO>(
         &self,
         pgb: &mut PostgresBackend<IO>,
@@ -485,6 +495,8 @@ impl PageServerHandler {
     where
         IO: AsyncRead + AsyncWrite + Send + Sync + Unpin,
     {
+        debug_assert_current_span_has_tenant_and_timeline_id();
+
         task_mgr::associate_with(Some(tenant_id), Some(timeline_id));
         // Create empty timeline
         info!("creating new timeline");
@@ -529,7 +541,7 @@ impl PageServerHandler {
         Ok(())
     }
 
-    #[instrument(skip(self, pgb, ctx))]
+    #[instrument(skip_all, fields(%start_lsn, %end_lsn))]
     async fn handle_import_wal<IO>(
         &self,
         pgb: &mut PostgresBackend<IO>,
@@ -542,6 +554,7 @@ impl PageServerHandler {
     where
         IO: AsyncRead + AsyncWrite + Send + Sync + Unpin,
     {
+        debug_assert_current_span_has_tenant_and_timeline_id();
         task_mgr::associate_with(Some(tenant_id), Some(timeline_id));
 
         let timeline = get_active_tenant_timeline(tenant_id, timeline_id, &ctx).await?;
@@ -736,7 +749,7 @@ impl PageServerHandler {
     }
 
     #[allow(clippy::too_many_arguments)]
-    #[instrument(skip(self, pgb, ctx))]
+    #[instrument(skip_all, fields(?lsn, ?prev_lsn, %full_backup))]
     async fn handle_basebackup_request<IO>(
         &mut self,
         pgb: &mut PostgresBackend<IO>,
@@ -745,11 +758,14 @@ impl PageServerHandler {
         lsn: Option<Lsn>,
         prev_lsn: Option<Lsn>,
         full_backup: bool,
+        gzip: bool,
         ctx: RequestContext,
     ) -> anyhow::Result<()>
     where
         IO: AsyncRead + AsyncWrite + Send + Sync + Unpin,
     {
+        debug_assert_current_span_has_tenant_and_timeline_id();
+
         let started = std::time::Instant::now();
 
         // check that the timeline exists
@@ -770,8 +786,9 @@ impl PageServerHandler {
         pgb.write_message_noflush(&BeMessage::CopyOutResponse)?;
         pgb.flush().await?;
 
-        // Send a tarball of the latest layer on the timeline
-        {
+        // Send a tarball of the latest layer on the timeline. Compress if not
+        // fullbackup. TODO Compress in that case too (tests need to be updated)
+        if full_backup {
             let mut writer = pgb.copyout_writer();
             basebackup::send_basebackup_tarball(
                 &mut writer,
@@ -782,6 +799,40 @@ impl PageServerHandler {
                 &ctx,
             )
             .await?;
+        } else {
+            let mut writer = pgb.copyout_writer();
+            if gzip {
+                let mut encoder = GzipEncoder::with_quality(
+                    writer,
+                    // NOTE using fast compression because it's on the critical path
+                    //      for compute startup. For an empty database, we get
+                    //      <100KB with this method. The Level::Best compression method
+                    //      gives us <20KB, but maybe we should add basebackup caching
+                    //      on compute shutdown first.
+                    async_compression::Level::Fastest,
+                );
+                basebackup::send_basebackup_tarball(
+                    &mut encoder,
+                    &timeline,
+                    lsn,
+                    prev_lsn,
+                    full_backup,
+                    &ctx,
+                )
+                .await?;
+                // shutdown the encoder to ensure the gzip footer is written
+                encoder.shutdown().await?;
+            } else {
+                basebackup::send_basebackup_tarball(
+                    &mut writer,
+                    &timeline,
+                    lsn,
+                    prev_lsn,
+                    full_backup,
+                    &ctx,
+                )
+                .await?;
+            }
         }
 
         pgb.write_message_noflush(&BeMessage::CopyDone)?;
@@ -860,6 +911,7 @@ where
         Ok(())
     }
 
+    #[instrument(skip_all, fields(tenant_id, timeline_id))]
     async fn process_query(
         &mut self,
         pgb: &mut PostgresBackend<IO>,
@@ -881,6 +933,10 @@ where
             let timeline_id = TimelineId::from_str(params[1])
                 .with_context(|| format!("Failed to parse timeline id from {}", params[1]))?;
 
+            tracing::Span::current()
+                .record("tenant_id", field::display(tenant_id))
+                .record("timeline_id", field::display(timeline_id));
+
             self.check_permission(Some(tenant_id))?;
 
             self.handle_pagerequests(pgb, tenant_id, timeline_id, ctx)
@@ -900,9 +956,13 @@ where
             let timeline_id = TimelineId::from_str(params[1])
                 .with_context(|| format!("Failed to parse timeline id from {}", params[1]))?;
 
+            tracing::Span::current()
+                .record("tenant_id", field::display(tenant_id))
+                .record("timeline_id", field::display(timeline_id));
+
             self.check_permission(Some(tenant_id))?;
 
-            let lsn = if params.len() == 3 {
+            let lsn = if params.len() >= 3 {
                 Some(
                     Lsn::from_str(params[2])
                         .with_context(|| format!("Failed to parse Lsn from {}", params[2]))?,
@@ -911,10 +971,38 @@ where
                 None
             };
 
-            // Check that the timeline exists
-            self.handle_basebackup_request(pgb, tenant_id, timeline_id, lsn, None, false, ctx)
-                .await?;
-            pgb.write_message_noflush(&BeMessage::CommandComplete(b"SELECT 1"))?;
+            let gzip = if params.len() >= 4 {
+                if params[3] == "--gzip" {
+                    true
+                } else {
+                    return Err(QueryError::Other(anyhow::anyhow!(
+                        "Parameter in position 3 unknown {}",
+                        params[3],
+                    )));
+                }
+            } else {
+                false
+            };
+
+            metrics::metric_vec_duration::observe_async_block_duration_by_result(
+                &*crate::metrics::BASEBACKUP_QUERY_TIME,
+                async move {
+                    self.handle_basebackup_request(
+                        pgb,
+                        tenant_id,
+                        timeline_id,
+                        lsn,
+                        None,
+                        false,
+                        gzip,
+                        ctx,
+                    )
+                    .await?;
+                    pgb.write_message_noflush(&BeMessage::CommandComplete(b"SELECT 1"))?;
+                    anyhow::Ok(())
+                },
+            )
+            .await?;
         }
         // return pair of prev_lsn and last_lsn
         else if query_string.starts_with("get_last_record_rlsn ") {
@@ -932,6 +1020,10 @@ where
             let timeline_id = TimelineId::from_str(params[1])
                 .with_context(|| format!("Failed to parse timeline id from {}", params[1]))?;
 
+            tracing::Span::current()
+                .record("tenant_id", field::display(tenant_id))
+                .record("timeline_id", field::display(timeline_id));
+
             self.check_permission(Some(tenant_id))?;
             let timeline = get_active_tenant_timeline(tenant_id, timeline_id, &ctx).await?;
 
@@ -963,6 +1055,10 @@ where
             let timeline_id = TimelineId::from_str(params[1])
                 .with_context(|| format!("Failed to parse timeline id from {}", params[1]))?;
 
+            tracing::Span::current()
+                .record("tenant_id", field::display(tenant_id))
+                .record("timeline_id", field::display(timeline_id));
+
             // The caller is responsible for providing correct lsn and prev_lsn.
             let lsn = if params.len() > 2 {
                 Some(
@@ -984,8 +1080,17 @@ where
             self.check_permission(Some(tenant_id))?;
 
             // Check that the timeline exists
-            self.handle_basebackup_request(pgb, tenant_id, timeline_id, lsn, prev_lsn, true, ctx)
-                .await?;
+            self.handle_basebackup_request(
+                pgb,
+                tenant_id,
+                timeline_id,
+                lsn,
+                prev_lsn,
+                true,
+                false,
+                ctx,
+            )
+            .await?;
             pgb.write_message_noflush(&BeMessage::CommandComplete(b"SELECT 1"))?;
         } else if query_string.starts_with("import basebackup ") {
             // Import the `base` section (everything but the wal) of a basebackup.
@@ -1017,6 +1122,10 @@ where
             let pg_version = u32::from_str(params[4])
                 .with_context(|| format!("Failed to parse pg_version from {}", params[4]))?;
 
+            tracing::Span::current()
+                .record("tenant_id", field::display(tenant_id))
+                .record("timeline_id", field::display(timeline_id));
+
             self.check_permission(Some(tenant_id))?;
 
             match self
@@ -1061,6 +1170,10 @@ where
             let end_lsn = Lsn::from_str(params[3])
                 .with_context(|| format!("Failed to parse Lsn from {}", params[3]))?;
 
+            tracing::Span::current()
+                .record("tenant_id", field::display(tenant_id))
+                .record("timeline_id", field::display(timeline_id));
+
             self.check_permission(Some(tenant_id))?;
 
             match self
@@ -1092,6 +1205,8 @@ where
             let tenant_id = TenantId::from_str(params[0])
                 .with_context(|| format!("Failed to parse tenant id from {}", params[0]))?;
 
+            tracing::Span::current().record("tenant_id", field::display(tenant_id));
+
             self.check_permission(Some(tenant_id))?;
 
             let tenant = get_active_tenant_with_timeout(tenant_id, &ctx).await?;
@@ -1230,6 +1345,6 @@ async fn get_active_tenant_timeline(
         .map_err(GetActiveTimelineError::Tenant)?;
     let timeline = tenant
         .get_timeline(timeline_id, true)
-        .map_err(GetActiveTimelineError::Timeline)?;
+        .map_err(|e| GetActiveTimelineError::Timeline(anyhow::anyhow!(e)))?;
     Ok(timeline)
 }

pageserver/src/pgdatadir_mapping.rs

@@ -43,6 +43,16 @@ pub enum CalculateLogicalSizeError {
     Other(#[from] anyhow::Error),
 }
 
+#[derive(Debug, thiserror::Error)]
+pub enum RelationError {
+    #[error("Relation Already Exists")]
+    AlreadyExists,
+    #[error("invalid relnode")]
+    InvalidRelnode,
+    #[error(transparent)]
+    Other(#[from] anyhow::Error),
+}
+
 ///
 /// This impl provides all the functionality to store PostgreSQL relations, SLRUs,
 /// and other special kinds of files, in a versioned key-value store. The
@@ -101,9 +111,9 @@ impl Timeline {
         ctx: &RequestContext,
     ) -> Result<Bytes, PageReconstructError> {
         if tag.relnode == 0 {
-            return Err(PageReconstructError::Other(anyhow::anyhow!(
-                "invalid relnode"
-            )));
+            return Err(PageReconstructError::Other(
+                RelationError::InvalidRelnode.into(),
+            ));
         }
 
         let nblocks = self.get_rel_size(tag, lsn, latest, ctx).await?;
@@ -148,9 +158,9 @@ impl Timeline {
         ctx: &RequestContext,
     ) -> Result<BlockNumber, PageReconstructError> {
         if tag.relnode == 0 {
-            return Err(PageReconstructError::Other(anyhow::anyhow!(
-                "invalid relnode"
-            )));
+            return Err(PageReconstructError::Other(
+                RelationError::InvalidRelnode.into(),
+            ));
         }
 
         if let Some(nblocks) = self.get_cached_rel_size(&tag, lsn) {
@@ -193,9 +203,9 @@ impl Timeline {
         ctx: &RequestContext,
     ) -> Result<bool, PageReconstructError> {
         if tag.relnode == 0 {
-            return Err(PageReconstructError::Other(anyhow::anyhow!(
-                "invalid relnode"
-            )));
+            return Err(PageReconstructError::Other(
+                RelationError::InvalidRelnode.into(),
+            ));
         }
 
         // first try to lookup relation in cache
@@ -724,7 +734,7 @@ impl<'a> DatadirModification<'a> {
         blknum: BlockNumber,
         rec: NeonWalRecord,
     ) -> anyhow::Result<()> {
-        anyhow::ensure!(rel.relnode != 0, "invalid relnode");
+        anyhow::ensure!(rel.relnode != 0, RelationError::InvalidRelnode);
         self.put(rel_block_to_key(rel, blknum), Value::WalRecord(rec));
         Ok(())
     }
@@ -751,7 +761,7 @@ impl<'a> DatadirModification<'a> {
         blknum: BlockNumber,
         img: Bytes,
     ) -> anyhow::Result<()> {
-        anyhow::ensure!(rel.relnode != 0, "invalid relnode");
+        anyhow::ensure!(rel.relnode != 0, RelationError::InvalidRelnode);
         self.put(rel_block_to_key(rel, blknum), Value::Image(img));
         Ok(())
     }
@@ -875,32 +885,38 @@ impl<'a> DatadirModification<'a> {
         rel: RelTag,
         nblocks: BlockNumber,
         ctx: &RequestContext,
-    ) -> anyhow::Result<()> {
-        anyhow::ensure!(rel.relnode != 0, "invalid relnode");
+    ) -> Result<(), RelationError> {
+        if rel.relnode == 0 {
+            return Err(RelationError::InvalidRelnode);
+        }
         // It's possible that this is the first rel for this db in this
         // tablespace.  Create the reldir entry for it if so.
-        let mut dbdir = DbDirectory::des(&self.get(DBDIR_KEY, ctx).await?)?;
+        let mut dbdir = DbDirectory::des(&self.get(DBDIR_KEY, ctx).await.context("read db")?)
+            .context("deserialize db")?;
         let rel_dir_key = rel_dir_to_key(rel.spcnode, rel.dbnode);
         let mut rel_dir = if dbdir.dbdirs.get(&(rel.spcnode, rel.dbnode)).is_none() {
             // Didn't exist. Update dbdir
             dbdir.dbdirs.insert((rel.spcnode, rel.dbnode), false);
-            let buf = DbDirectory::ser(&dbdir)?;
+            let buf = DbDirectory::ser(&dbdir).context("serialize db")?;
             self.put(DBDIR_KEY, Value::Image(buf.into()));
 
             // and create the RelDirectory
             RelDirectory::default()
         } else {
             // reldir already exists, fetch it
-            RelDirectory::des(&self.get(rel_dir_key, ctx).await?)?
+            RelDirectory::des(&self.get(rel_dir_key, ctx).await.context("read db")?)
+                .context("deserialize db")?
         };
 
         // Add the new relation to the rel directory entry, and write it back
         if !rel_dir.rels.insert((rel.relnode, rel.forknum)) {
-            anyhow::bail!("rel {rel} already exists");
+            return Err(RelationError::AlreadyExists);
         }
         self.put(
             rel_dir_key,
-            Value::Image(Bytes::from(RelDirectory::ser(&rel_dir)?)),
+            Value::Image(Bytes::from(
+                RelDirectory::ser(&rel_dir).context("serialize")?,
+            )),
         );
 
         // Put size
@@ -925,7 +941,7 @@ impl<'a> DatadirModification<'a> {
         nblocks: BlockNumber,
         ctx: &RequestContext,
     ) -> anyhow::Result<()> {
-        anyhow::ensure!(rel.relnode != 0, "invalid relnode");
+        anyhow::ensure!(rel.relnode != 0, RelationError::InvalidRelnode);
         let last_lsn = self.tline.get_last_record_lsn();
         if self.tline.get_rel_exists(rel, last_lsn, true, ctx).await? {
             let size_key = rel_size_to_key(rel);
@@ -956,7 +972,7 @@ impl<'a> DatadirModification<'a> {
         nblocks: BlockNumber,
         ctx: &RequestContext,
     ) -> anyhow::Result<()> {
-        anyhow::ensure!(rel.relnode != 0, "invalid relnode");
+        anyhow::ensure!(rel.relnode != 0, RelationError::InvalidRelnode);
 
         // Put size
         let size_key = rel_size_to_key(rel);
@@ -977,7 +993,7 @@ impl<'a> DatadirModification<'a> {
 
     /// Drop a relation.
     pub async fn put_rel_drop(&mut self, rel: RelTag, ctx: &RequestContext) -> anyhow::Result<()> {
-        anyhow::ensure!(rel.relnode != 0, "invalid relnode");
+        anyhow::ensure!(rel.relnode != 0, RelationError::InvalidRelnode);
 
         // Remove it from the directory entry
         let dir_key = rel_dir_to_key(rel.spcnode, rel.dbnode);

pageserver/src/task_mgr.rs

@@ -506,17 +506,17 @@ pub async fn shutdown_tasks(
                     warn!(name = task.name, tenant_id = ?tenant_id, timeline_id = ?timeline_id, kind = ?task_kind, "stopping left-over");
                 }
             }
-            let completed = tokio::select! {
+            let join_handle = tokio::select! {
                 biased;
-                _ = &mut join_handle => { true },
+                _ = &mut join_handle => { None },
                 _ = tokio::time::sleep(std::time::Duration::from_secs(1)) => {
                     // allow some time to elapse before logging to cut down the number of log
                     // lines.
                     info!("waiting for {} to shut down", task.name);
-                    false
+                    Some(join_handle)
                 }
             };
-            if !completed {
+            if let Some(join_handle) = join_handle {
                 // we never handled this return value, but:
                 // - we don't deschedule which would lead to is_cancelled
                 // - panics are already logged (is_panicked)

pageserver/src/tenant/config.rs

@@ -38,8 +38,8 @@ pub mod defaults {
     pub const DEFAULT_GC_PERIOD: &str = "1 hr";
     pub const DEFAULT_IMAGE_CREATION_THRESHOLD: usize = 3;
     pub const DEFAULT_PITR_INTERVAL: &str = "7 days";
-    pub const DEFAULT_WALRECEIVER_CONNECT_TIMEOUT: &str = "2 seconds";
-    pub const DEFAULT_WALRECEIVER_LAGGING_WAL_TIMEOUT: &str = "3 seconds";
+    pub const DEFAULT_WALRECEIVER_CONNECT_TIMEOUT: &str = "10 seconds";
+    pub const DEFAULT_WALRECEIVER_LAGGING_WAL_TIMEOUT: &str = "10 seconds";
     pub const DEFAULT_MAX_WALRECEIVER_LSN_WAL_LAG: u64 = 10 * 1024 * 1024;
     pub const DEFAULT_EVICTIONS_LOW_RESIDENCE_DURATION_METRIC_THRESHOLD: &str = "24 hour";
 }

pageserver/src/tenant/ephemeral_file.rs

@@ -55,7 +55,7 @@ impl EphemeralFile {
         l.next_file_id += 1;
 
         let filename = conf
-            .timeline_path(&timeline_id, &tenant_id)
+            .timeline_path(&tenant_id, &timeline_id)
             .join(PathBuf::from(format!("ephemeral-{}", file_id)));
 
         let file = VirtualFile::open_with_options(
@@ -346,7 +346,7 @@ mod tests {
 
         let tenant_id = TenantId::from_str("11000000000000000000000000000000").unwrap();
         let timeline_id = TimelineId::from_str("22000000000000000000000000000000").unwrap();
-        fs::create_dir_all(conf.timeline_path(&timeline_id, &tenant_id))?;
+        fs::create_dir_all(conf.timeline_path(&tenant_id, &timeline_id))?;
 
         Ok((conf, tenant_id, timeline_id))
     }

pageserver/src/tenant/layer_map.rs

@@ -51,25 +51,22 @@ use crate::keyspace::KeyPartitioning;
 use crate::repository::Key;
 use crate::tenant::storage_layer::InMemoryLayer;
 use crate::tenant::storage_layer::Layer;
-use anyhow::Context;
 use anyhow::Result;
-use std::collections::HashMap;
 use std::collections::VecDeque;
 use std::ops::Range;
 use std::sync::Arc;
 use utils::lsn::Lsn;
 
 use historic_layer_coverage::BufferedHistoricLayerCoverage;
-pub use historic_layer_coverage::Replacement;
+pub use historic_layer_coverage::LayerKey;
 
-use super::storage_layer::range_eq;
 use super::storage_layer::PersistentLayerDesc;
-use super::storage_layer::PersistentLayerKey;
 
 ///
 /// LayerMap tracks what layers exist on a timeline.
 ///
-pub struct LayerMap<L: ?Sized> {
+#[derive(Default)]
+pub struct LayerMap {
     //
     // 'open_layer' holds the current InMemoryLayer that is accepting new
     // records. If it is None, 'next_open_layer_at' will be set instead, indicating
@@ -95,24 +92,6 @@ pub struct LayerMap<L: ?Sized> {
     /// L0 layers have key range Key::MIN..Key::MAX, and locating them using R-Tree search is very inefficient.
     /// So L0 layers are held in l0_delta_layers vector, in addition to the R-tree.
     l0_delta_layers: Vec<Arc<PersistentLayerDesc>>,
-
-    /// Mapping from persistent layer key to the actual layer object. Currently, it stores delta, image, and
-    /// remote layers. In future refactors, this will be eventually moved out of LayerMap into Timeline, and
-    /// RemoteLayer will be removed.
-    mapping: HashMap<PersistentLayerKey, Arc<L>>,
-}
-
-impl<L: ?Sized> Default for LayerMap<L> {
-    fn default() -> Self {
-        Self {
-            open_layer: None,
-            next_open_layer_at: None,
-            frozen_layers: VecDeque::default(),
-            l0_delta_layers: Vec::default(),
-            historic: BufferedHistoricLayerCoverage::default(),
-            mapping: HashMap::default(),
-        }
-    }
 }
 
 /// The primary update API for the layer map.
@@ -120,24 +99,21 @@ impl<L: ?Sized> Default for LayerMap<L> {
 /// Batching historic layer insertions and removals is good for
 /// performance and this struct helps us do that correctly.
 #[must_use]
-pub struct BatchedUpdates<'a, L: ?Sized + Layer> {
+pub struct BatchedUpdates<'a> {
     // While we hold this exclusive reference to the layer map the type checker
     // will prevent us from accidentally reading any unflushed updates.
-    layer_map: &'a mut LayerMap<L>,
+    layer_map: &'a mut LayerMap,
 }
 
 /// Provide ability to batch more updates while hiding the read
 /// API so we don't accidentally read without flushing.
-impl<L> BatchedUpdates<'_, L>
-where
-    L: ?Sized + Layer,
-{
+impl BatchedUpdates<'_> {
     ///
     /// Insert an on-disk layer.
     ///
     // TODO remove the `layer` argument when `mapping` is refactored out of `LayerMap`
-    pub fn insert_historic(&mut self, layer_desc: PersistentLayerDesc, layer: Arc<L>) {
-        self.layer_map.insert_historic_noflush(layer_desc, layer)
+    pub fn insert_historic(&mut self, layer_desc: PersistentLayerDesc) {
+        self.layer_map.insert_historic_noflush(layer_desc)
     }
 
     ///
@@ -145,31 +121,8 @@ where
     ///
     /// This should be called when the corresponding file on disk has been deleted.
     ///
-    pub fn remove_historic(&mut self, layer_desc: PersistentLayerDesc, layer: Arc<L>) {
-        self.layer_map.remove_historic_noflush(layer_desc, layer)
-    }
-
-    /// Replaces existing layer iff it is the `expected`.
-    ///
-    /// If the expected layer has been removed it will not be inserted by this function.
-    ///
-    /// Returned `Replacement` describes succeeding in replacement or the reason why it could not
-    /// be done.
-    ///
-    /// TODO replacement can be done without buffering and rebuilding layer map updates.
-    ///      One way to do that is to add a layer of indirection for returned values, so
-    ///      that we can replace values only by updating a hashmap.
-    pub fn replace_historic(
-        &mut self,
-        expected_desc: PersistentLayerDesc,
-        expected: &Arc<L>,
-        new_desc: PersistentLayerDesc,
-        new: Arc<L>,
-    ) -> anyhow::Result<Replacement<Arc<L>>> {
-        fail::fail_point!("layermap-replace-notfound", |_| Ok(Replacement::NotFound));
-
-        self.layer_map
-            .replace_historic_noflush(expected_desc, expected, new_desc, new)
+    pub fn remove_historic(&mut self, layer_desc: PersistentLayerDesc) {
+        self.layer_map.remove_historic_noflush(layer_desc)
     }
 
     // We will flush on drop anyway, but this method makes it
@@ -185,25 +138,19 @@ where
 // than panic later or read without flushing.
 //
 // TODO maybe warn if flush hasn't explicitly been called
-impl<L> Drop for BatchedUpdates<'_, L>
-where
-    L: ?Sized + Layer,
-{
+impl Drop for BatchedUpdates<'_> {
     fn drop(&mut self) {
         self.layer_map.flush_updates();
     }
 }
 
 /// Return value of LayerMap::search
-pub struct SearchResult<L: ?Sized> {
-    pub layer: Arc<L>,
+pub struct SearchResult {
+    pub layer: Arc<PersistentLayerDesc>,
     pub lsn_floor: Lsn,
 }
 
-impl<L> LayerMap<L>
-where
-    L: ?Sized + Layer,
-{
+impl LayerMap {
     ///
     /// Find the latest layer (by lsn.end) that covers the given
     /// 'key', with lsn.start < 'end_lsn'.
@@ -235,7 +182,7 @@ where
     /// NOTE: This only searches the 'historic' layers, *not* the
     /// 'open' and 'frozen' layers!
     ///
-    pub fn search(&self, key: Key, end_lsn: Lsn) -> Option<SearchResult<L>> {
+    pub fn search(&self, key: Key, end_lsn: Lsn) -> Option<SearchResult> {
         let version = self.historic.get().unwrap().get_version(end_lsn.0 - 1)?;
         let latest_delta = version.delta_coverage.query(key.to_i128());
         let latest_image = version.image_coverage.query(key.to_i128());
@@ -244,7 +191,6 @@ where
             (None, None) => None,
             (None, Some(image)) => {
                 let lsn_floor = image.get_lsn_range().start;
-                let image = self.get_layer_from_mapping(&image.key()).clone();
                 Some(SearchResult {
                     layer: image,
                     lsn_floor,
@@ -252,7 +198,6 @@ where
             }
             (Some(delta), None) => {
                 let lsn_floor = delta.get_lsn_range().start;
-                let delta = self.get_layer_from_mapping(&delta.key()).clone();
                 Some(SearchResult {
                     layer: delta,
                     lsn_floor,
@@ -263,7 +208,6 @@ where
                 let image_is_newer = image.get_lsn_range().end >= delta.get_lsn_range().end;
                 let image_exact_match = img_lsn + 1 == end_lsn;
                 if image_is_newer || image_exact_match {
-                    let image = self.get_layer_from_mapping(&image.key()).clone();
                     Some(SearchResult {
                         layer: image,
                         lsn_floor: img_lsn,
@@ -271,7 +215,6 @@ where
                 } else {
                     let lsn_floor =
                         std::cmp::max(delta.get_lsn_range().start, image.get_lsn_range().start + 1);
-                    let delta = self.get_layer_from_mapping(&delta.key()).clone();
                     Some(SearchResult {
                         layer: delta,
                         lsn_floor,
@@ -282,7 +225,7 @@ where
     }
 
     /// Start a batch of updates, applied on drop
-    pub fn batch_update(&mut self) -> BatchedUpdates<'_, L> {
+    pub fn batch_update(&mut self) -> BatchedUpdates<'_> {
         BatchedUpdates { layer_map: self }
     }
 
@@ -292,48 +235,32 @@ where
     /// Helper function for BatchedUpdates::insert_historic
     ///
     /// TODO(chi): remove L generic so that we do not need to pass layer object.
-    pub(self) fn insert_historic_noflush(
-        &mut self,
-        layer_desc: PersistentLayerDesc,
-        layer: Arc<L>,
-    ) {
-        self.mapping.insert(layer_desc.key(), layer.clone());
-
+    pub(self) fn insert_historic_noflush(&mut self, layer_desc: PersistentLayerDesc) {
         // TODO: See #3869, resulting #4088, attempted fix and repro #4094
 
-        if Self::is_l0(&layer) {
+        if Self::is_l0(&layer_desc) {
             self.l0_delta_layers.push(layer_desc.clone().into());
         }
 
         self.historic.insert(
-            historic_layer_coverage::LayerKey::from(&*layer),
+            historic_layer_coverage::LayerKey::from(&layer_desc),
             layer_desc.into(),
         );
     }
 
-    fn get_layer_from_mapping(&self, key: &PersistentLayerKey) -> &Arc<L> {
-        let layer = self
-            .mapping
-            .get(key)
-            .with_context(|| format!("{key:?}"))
-            .expect("inconsistent layer mapping");
-        layer
-    }
-
     ///
     /// Remove an on-disk layer from the map.
     ///
     /// Helper function for BatchedUpdates::remove_historic
     ///
-    pub fn remove_historic_noflush(&mut self, layer_desc: PersistentLayerDesc, layer: Arc<L>) {
+    pub fn remove_historic_noflush(&mut self, layer_desc: PersistentLayerDesc) {
         self.historic
-            .remove(historic_layer_coverage::LayerKey::from(&*layer));
-        if Self::is_l0(&layer) {
+            .remove(historic_layer_coverage::LayerKey::from(&layer_desc));
+        let layer_key = layer_desc.key();
+        if Self::is_l0(&layer_desc) {
             let len_before = self.l0_delta_layers.len();
             let mut l0_delta_layers = std::mem::take(&mut self.l0_delta_layers);
-            l0_delta_layers.retain(|other| {
-                !Self::compare_arced_layers(self.get_layer_from_mapping(&other.key()), &layer)
-            });
+            l0_delta_layers.retain(|other| other.key() != layer_key);
             self.l0_delta_layers = l0_delta_layers;
             // this assertion is related to use of Arc::ptr_eq in Self::compare_arced_layers,
             // there's a chance that the comparison fails at runtime due to it comparing (pointer,
@@ -344,69 +271,6 @@ where
                 "failed to locate removed historic layer from l0_delta_layers"
             );
         }
-        self.mapping.remove(&layer_desc.key());
-    }
-
-    pub(self) fn replace_historic_noflush(
-        &mut self,
-        expected_desc: PersistentLayerDesc,
-        expected: &Arc<L>,
-        new_desc: PersistentLayerDesc,
-        new: Arc<L>,
-    ) -> anyhow::Result<Replacement<Arc<L>>> {
-        let key = historic_layer_coverage::LayerKey::from(&**expected);
-        let other = historic_layer_coverage::LayerKey::from(&*new);
-
-        let expected_l0 = Self::is_l0(expected);
-        let new_l0 = Self::is_l0(&new);
-
-        anyhow::ensure!(
-            key == other,
-            "expected and new must have equal LayerKeys: {key:?} != {other:?}"
-        );
-
-        anyhow::ensure!(
-            expected_l0 == new_l0,
-            "expected and new must both be l0 deltas or neither should be: {expected_l0} != {new_l0}"
-        );
-
-        let l0_index = if expected_l0 {
-            // find the index in case replace worked, we need to replace that as well
-            let pos = self.l0_delta_layers.iter().position(|slot| {
-                Self::compare_arced_layers(self.get_layer_from_mapping(&slot.key()), expected)
-            });
-
-            if pos.is_none() {
-                return Ok(Replacement::NotFound);
-            }
-            pos
-        } else {
-            None
-        };
-
-        let new_desc = Arc::new(new_desc);
-        let replaced = self.historic.replace(&key, new_desc.clone(), |existing| {
-            **existing == expected_desc
-        });
-
-        if let Replacement::Replaced { .. } = &replaced {
-            self.mapping.remove(&expected_desc.key());
-            self.mapping.insert(new_desc.key(), new);
-            if let Some(index) = l0_index {
-                self.l0_delta_layers[index] = new_desc;
-            }
-        }
-
-        let replaced = match replaced {
-            Replacement::Replaced { in_buffered } => Replacement::Replaced { in_buffered },
-            Replacement::NotFound => Replacement::NotFound,
-            Replacement::RemovalBuffered => Replacement::RemovalBuffered,
-            Replacement::Unexpected(x) => {
-                Replacement::Unexpected(self.get_layer_from_mapping(&x.key()).clone())
-            }
-        };
-
-        Ok(replaced)
     }
 
     /// Helper function for BatchedUpdates::drop.
@@ -454,10 +318,8 @@ where
         Ok(true)
     }
 
-    pub fn iter_historic_layers(&self) -> impl '_ + Iterator<Item = Arc<L>> {
-        self.historic
-            .iter()
-            .map(|x| self.get_layer_from_mapping(&x.key()).clone())
+    pub fn iter_historic_layers(&self) -> impl '_ + Iterator<Item = Arc<PersistentLayerDesc>> {
+        self.historic.iter()
     }
 
     ///
@@ -472,7 +334,7 @@ where
         &self,
         key_range: &Range<Key>,
         lsn: Lsn,
-    ) -> Result<Vec<(Range<Key>, Option<Arc<L>>)>> {
+    ) -> Result<Vec<(Range<Key>, Option<Arc<PersistentLayerDesc>>)>> {
         let version = match self.historic.get().unwrap().get_version(lsn.0) {
             Some(v) => v,
             None => return Ok(vec![]),
@@ -482,37 +344,27 @@ where
         let end = key_range.end.to_i128();
 
         // Initialize loop variables
-        let mut coverage: Vec<(Range<Key>, Option<Arc<L>>)> = vec![];
+        let mut coverage: Vec<(Range<Key>, Option<Arc<PersistentLayerDesc>>)> = vec![];
         let mut current_key = start;
         let mut current_val = version.image_coverage.query(start);
 
         // Loop through the change events and push intervals
         for (change_key, change_val) in version.image_coverage.range(start..end) {
             let kr = Key::from_i128(current_key)..Key::from_i128(change_key);
-            coverage.push((
-                kr,
-                current_val
-                    .take()
-                    .map(|l| self.get_layer_from_mapping(&l.key()).clone()),
-            ));
+            coverage.push((kr, current_val.take()));
             current_key = change_key;
             current_val = change_val.clone();
         }
 
         // Add the final interval
         let kr = Key::from_i128(current_key)..Key::from_i128(end);
-        coverage.push((
-            kr,
-            current_val
-                .take()
-                .map(|l| self.get_layer_from_mapping(&l.key()).clone()),
-        ));
+        coverage.push((kr, current_val.take()));
 
         Ok(coverage)
     }
 
-    pub fn is_l0(layer: &L) -> bool {
-        range_eq(&layer.get_key_range(), &(Key::MIN..Key::MAX))
+    pub fn is_l0(layer: &PersistentLayerDesc) -> bool {
+        layer.get_key_range() == (Key::MIN..Key::MAX)
     }
 
     /// This function determines which layers are counted in `count_deltas`:
@@ -537,14 +389,14 @@ where
     /// TODO The optimal number should probably be slightly higher than 1, but to
     ///      implement that we need to plumb a lot more context into this function
     ///      than just the current partition_range.
-    pub fn is_reimage_worthy(layer: &L, partition_range: &Range<Key>) -> bool {
+    pub fn is_reimage_worthy(layer: &PersistentLayerDesc, partition_range: &Range<Key>) -> bool {
         // Case 1
         if !Self::is_l0(layer) {
             return true;
         }
 
         // Case 2
-        if range_eq(partition_range, &(Key::MIN..Key::MAX)) {
+        if partition_range == &(Key::MIN..Key::MAX) {
             return true;
         }
 
@@ -595,9 +447,7 @@ where
                     let kr = Key::from_i128(current_key)..Key::from_i128(change_key);
                     let lr = lsn.start..val.get_lsn_range().start;
                     if !kr.is_empty() {
-                        let base_count =
-                            Self::is_reimage_worthy(self.get_layer_from_mapping(&val.key()), key)
-                                as usize;
+                        let base_count = Self::is_reimage_worthy(&val, key) as usize;
                         let new_limit = limit.map(|l| l - base_count);
                         let max_stacked_deltas_underneath =
                             self.count_deltas(&kr, &lr, new_limit)?;
@@ -620,9 +470,7 @@ where
                 let lr = lsn.start..val.get_lsn_range().start;
 
                 if !kr.is_empty() {
-                    let base_count =
-                        Self::is_reimage_worthy(self.get_layer_from_mapping(&val.key()), key)
-                            as usize;
+                    let base_count = Self::is_reimage_worthy(&val, key) as usize;
                     let new_limit = limit.map(|l| l - base_count);
                     let max_stacked_deltas_underneath = self.count_deltas(&kr, &lr, new_limit)?;
                     max_stacked_deltas = std::cmp::max(
@@ -772,12 +620,8 @@ where
     }
 
     /// Return all L0 delta layers
-    pub fn get_level0_deltas(&self) -> Result<Vec<Arc<L>>> {
-        Ok(self
-            .l0_delta_layers
-            .iter()
-            .map(|x| self.get_layer_from_mapping(&x.key()).clone())
-            .collect())
+    pub fn get_level0_deltas(&self) -> Result<Vec<Arc<PersistentLayerDesc>>> {
+        Ok(self.l0_delta_layers.to_vec())
     }
 
     /// debugging function to print out the contents of the layer map
@@ -802,72 +646,67 @@ where
         println!("End dump LayerMap");
         Ok(())
     }
-
-    /// Similar to `Arc::ptr_eq`, but only compares the object pointers, not vtables.
-    ///
-    /// Returns `true` if the two `Arc` point to the same layer, false otherwise.
-    #[inline(always)]
-    pub fn compare_arced_layers(left: &Arc<L>, right: &Arc<L>) -> bool {
-        // "dyn Trait" objects are "fat pointers" in that they have two components:
-        // - pointer to the object
-        // - pointer to the vtable
-        //
-        // rust does not provide a guarantee that these vtables are unique, but however
-        // `Arc::ptr_eq` as of writing (at least up to 1.67) uses a comparison where both the
-        // pointer and the vtable need to be equal.
-        //
-        // See: https://github.com/rust-lang/rust/issues/103763
-        //
-        // A future version of rust will most likely use this form below, where we cast each
-        // pointer into a pointer to unit, which drops the inaccessible vtable pointer, making it
-        // not affect the comparison.
-        //
-        // See: https://github.com/rust-lang/rust/pull/106450
-        let left = Arc::as_ptr(left) as *const ();
-        let right = Arc::as_ptr(right) as *const ();
-
-        left == right
-    }
 }
 
 #[cfg(test)]
 mod tests {
-    use super::{LayerMap, Replacement};
-    use crate::tenant::storage_layer::{Layer, LayerDescriptor, LayerFileName};
+    use super::LayerMap;
+    use crate::tenant::storage_layer::LayerFileName;
     use std::str::FromStr;
     use std::sync::Arc;
 
     mod l0_delta_layers_updated {
 
+        use crate::tenant::{
+            storage_layer::{AsLayerDesc, PersistentLayerDesc},
+            timeline::layer_manager::LayerFileManager,
+        };
+
         use super::*;
 
+        struct LayerObject(PersistentLayerDesc);
+
+        impl AsLayerDesc for LayerObject {
+            fn layer_desc(&self) -> &PersistentLayerDesc {
+                &self.0
+            }
+        }
+
+        impl LayerObject {
+            fn new(desc: PersistentLayerDesc) -> Self {
+                LayerObject(desc)
+            }
+        }
+
+        type TestLayerFileManager = LayerFileManager<LayerObject>;
+
         #[test]
         fn for_full_range_delta() {
             // l0_delta_layers are used by compaction, and should observe all buffered updates
             l0_delta_layers_updated_scenario(
-                "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__0000000053423C21-0000000053424D69",
-                true
-            )
+                 "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__0000000053423C21-0000000053424D69",
+                 true
+             )
         }
 
         #[test]
         fn for_non_full_range_delta() {
             // has minimal uncovered areas compared to l0_delta_layers_updated_on_insert_replace_remove_for_full_range_delta
             l0_delta_layers_updated_scenario(
-                "000000000000000000000000000000000001-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFE__0000000053423C21-0000000053424D69",
-                // because not full range
-                false
-            )
+                 "000000000000000000000000000000000001-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFE__0000000053423C21-0000000053424D69",
+                 // because not full range
+                 false
+             )
         }
 
         #[test]
         fn for_image() {
             l0_delta_layers_updated_scenario(
-                "000000000000000000000000000000000000-000000000000000000000000000000010000__0000000053424D69",
-                // code only checks if it is a full range layer, doesn't care about images, which must
-                // mean we should in practice never have full range images
-                false
-            )
+                 "000000000000000000000000000000000000-000000000000000000000000000000010000__0000000053424D69",
+                 // code only checks if it is a full range layer, doesn't care about images, which must
+                 // mean we should in practice never have full range images
+                 false
+             )
         }
 
         #[test]
@@ -877,75 +716,70 @@ mod tests {
 
             let layer = "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__0000000053423C21-0000000053424D69";
             let layer = LayerFileName::from_str(layer).unwrap();
-            let layer = LayerDescriptor::from(layer);
+            let layer = PersistentLayerDesc::from(layer);
 
             // same skeletan construction; see scenario below
-            let not_found = Arc::new(layer.clone());
-            let new_version = Arc::new(layer);
+            let not_found = Arc::new(LayerObject::new(layer.clone()));
+            let new_version = Arc::new(LayerObject::new(layer));
 
-            let mut map = LayerMap::default();
+            // after the immutable storage state refactor, the replace operation
+            // will not use layer map any more. We keep it here for consistency in test cases
+            // and can remove it in the future.
+            let _map = LayerMap::default();
 
-            let res = map.batch_update().replace_historic(
-                not_found.get_persistent_layer_desc(),
-                &not_found,
-                new_version.get_persistent_layer_desc(),
-                new_version,
-            );
+            let mut mapping = TestLayerFileManager::new();
 
-            assert!(matches!(res, Ok(Replacement::NotFound)), "{res:?}");
+            mapping
+                .replace_and_verify(not_found, new_version)
+                .unwrap_err();
         }
 
         fn l0_delta_layers_updated_scenario(layer_name: &str, expected_l0: bool) {
             let name = LayerFileName::from_str(layer_name).unwrap();
-            let skeleton = LayerDescriptor::from(name);
+            let skeleton = PersistentLayerDesc::from(name);
 
-            let remote = Arc::new(skeleton.clone());
-            let downloaded = Arc::new(skeleton);
+            let remote = Arc::new(LayerObject::new(skeleton.clone()));
+            let downloaded = Arc::new(LayerObject::new(skeleton));
 
             let mut map = LayerMap::default();
+            let mut mapping = LayerFileManager::new();
 
             // two disjoint Arcs in different lifecycle phases. even if it seems they must be the
             // same layer, we use LayerMap::compare_arced_layers as the identity of layers.
-            assert!(!LayerMap::compare_arced_layers(&remote, &downloaded));
+            assert_eq!(remote.layer_desc(), downloaded.layer_desc());
 
             let expected_in_counts = (1, usize::from(expected_l0));
 
             map.batch_update()
-                .insert_historic(remote.get_persistent_layer_desc(), remote.clone());
-            assert_eq!(count_layer_in(&map, &remote), expected_in_counts);
-
-            let replaced = map
-                .batch_update()
-                .replace_historic(
-                    remote.get_persistent_layer_desc(),
-                    &remote,
-                    downloaded.get_persistent_layer_desc(),
-                    downloaded.clone(),
-                )
-                .expect("name derived attributes are the same");
-            assert!(
-                matches!(replaced, Replacement::Replaced { .. }),
-                "{replaced:?}"
+                .insert_historic(remote.layer_desc().clone());
+            mapping.insert(remote.clone());
+            assert_eq!(
+                count_layer_in(&map, remote.layer_desc()),
+                expected_in_counts
+            );
+
+            mapping
+                .replace_and_verify(remote, downloaded.clone())
+                .expect("name derived attributes are the same");
+            assert_eq!(
+                count_layer_in(&map, downloaded.layer_desc()),
+                expected_in_counts
             );
-            assert_eq!(count_layer_in(&map, &downloaded), expected_in_counts);
 
             map.batch_update()
-                .remove_historic(downloaded.get_persistent_layer_desc(), downloaded.clone());
-            assert_eq!(count_layer_in(&map, &downloaded), (0, 0));
+                .remove_historic(downloaded.layer_desc().clone());
+            assert_eq!(count_layer_in(&map, downloaded.layer_desc()), (0, 0));
         }
 
-        fn count_layer_in<L: Layer + ?Sized>(map: &LayerMap<L>, layer: &Arc<L>) -> (usize, usize) {
+        fn count_layer_in(map: &LayerMap, layer: &PersistentLayerDesc) -> (usize, usize) {
             let historic = map
                 .iter_historic_layers()
-                .filter(|x| LayerMap::compare_arced_layers(x, layer))
+                .filter(|x| x.key() == layer.key())
                 .count();
             let l0s = map
                 .get_level0_deltas()
                 .expect("why does this return a result");
-            let l0 = l0s
-                .iter()
-                .filter(|x| LayerMap::compare_arced_layers(x, layer))
-                .count();
+            let l0 = l0s.iter().filter(|x| x.key() == layer.key()).count();
 
             (historic, l0)
         }

pageserver/src/tenant/layer_map/historic_layer_coverage.rs

@@ -3,6 +3,8 @@ use std::ops::Range;
 
 use tracing::info;
 
+use crate::tenant::storage_layer::PersistentLayerDesc;
+
 use super::layer_coverage::LayerCoverageTuple;
 
 /// Layers in this module are identified and indexed by this data.
@@ -41,8 +43,8 @@ impl Ord for LayerKey {
     }
 }
 
-impl<'a, L: crate::tenant::storage_layer::Layer + ?Sized> From<&'a L> for LayerKey {
-    fn from(layer: &'a L) -> Self {
+impl From<&PersistentLayerDesc> for LayerKey {
+    fn from(layer: &PersistentLayerDesc) -> Self {
         let kr = layer.get_key_range();
         let lr = layer.get_lsn_range();
         LayerKey {
@@ -454,59 +456,6 @@ impl<Value: Clone> BufferedHistoricLayerCoverage<Value> {
         self.buffer.insert(layer_key, None);
     }
 
-    /// Replaces a previous layer with a new layer value.
-    ///
-    /// The replacement is conditional on:
-    /// - there is an existing `LayerKey` record
-    /// - there is no buffered removal for the given `LayerKey`
-    /// - the given closure returns true for the current `Value`
-    ///
-    /// The closure is used to compare the latest value (buffered insert, or existing layer)
-    /// against some expectation. This allows to use `Arc::ptr_eq` or similar which would be
-    /// inaccessible via `PartialEq` trait.
-    ///
-    /// Returns a `Replacement` value describing the outcome; only the case of
-    /// `Replacement::Replaced` modifies the map and requires a rebuild.
-    pub fn replace<F>(
-        &mut self,
-        layer_key: &LayerKey,
-        new: Value,
-        check_expected: F,
-    ) -> Replacement<Value>
-    where
-        F: FnOnce(&Value) -> bool,
-    {
-        let (slot, in_buffered) = match self.buffer.get(layer_key) {
-            Some(inner @ Some(_)) => {
-                // we compare against the buffered version, because there will be a later
-                // rebuild before querying
-                (inner.as_ref(), true)
-            }
-            Some(None) => {
-                // buffer has removal for this key; it will not be equivalent by any check_expected.
-                return Replacement::RemovalBuffered;
-            }
-            None => {
-                // no pending modification for the key, check layers
-                (self.layers.get(layer_key), false)
-            }
-        };
-
-        match slot {
-            Some(existing) if !check_expected(existing) => {
-                // unfortunate clone here, but otherwise the nll borrowck grows the region of
-                // 'a to cover the whole function, and we could not mutate in the other
-                // Some(existing) branch
-                Replacement::Unexpected(existing.clone())
-            }
-            None => Replacement::NotFound,
-            Some(_existing) => {
-                self.insert(layer_key.to_owned(), new);
-                Replacement::Replaced { in_buffered }
-            }
-        }
-    }
-
     pub fn rebuild(&mut self) {
         // Find the first LSN that needs to be rebuilt
         let rebuild_since: u64 = match self.buffer.iter().next() {
@@ -575,22 +524,6 @@ impl<Value: Clone> BufferedHistoricLayerCoverage<Value> {
     }
 }
 
-/// Outcome of the replace operation.
-#[derive(Debug)]
-pub enum Replacement<Value> {
-    /// Previous value was replaced with the new value.
-    Replaced {
-        /// Replacement happened for a scheduled insert.
-        in_buffered: bool,
-    },
-    /// Key was not found buffered updates or existing layers.
-    NotFound,
-    /// Key has been scheduled for removal, it was not replaced.
-    RemovalBuffered,
-    /// Previous value was rejected by the closure.
-    Unexpected(Value),
-}
-
 #[test]
 fn test_retroactive_regression_1() {
     let mut map = BufferedHistoricLayerCoverage::new();
@@ -699,139 +632,3 @@ fn test_retroactive_simple() {
         assert_eq!(version.image_coverage.query(8), Some("Image 4".to_string()));
     }
 }
-
-#[test]
-fn test_retroactive_replacement() {
-    let mut map = BufferedHistoricLayerCoverage::new();
-
-    let keys = [
-        LayerKey {
-            key: 0..5,
-            lsn: 100..101,
-            is_image: true,
-        },
-        LayerKey {
-            key: 3..9,
-            lsn: 110..111,
-            is_image: true,
-        },
-        LayerKey {
-            key: 4..6,
-            lsn: 120..121,
-            is_image: true,
-        },
-    ];
-
-    let layers = [
-        "Image 1".to_string(),
-        "Image 2".to_string(),
-        "Image 3".to_string(),
-    ];
-
-    for (key, layer) in keys.iter().zip(layers.iter()) {
-        map.insert(key.to_owned(), layer.to_owned());
-    }
-
-    // rebuild is not necessary here, because replace works for both buffered updates and existing
-    // layers.
-
-    for (key, orig_layer) in keys.iter().zip(layers.iter()) {
-        let replacement = format!("Remote {orig_layer}");
-
-        // evict
-        let ret = map.replace(key, replacement.clone(), |l| l == orig_layer);
-        assert!(
-            matches!(ret, Replacement::Replaced { .. }),
-            "replace {orig_layer}: {ret:?}"
-        );
-        map.rebuild();
-
-        let at = key.lsn.end + 1;
-
-        let version = map.get().expect("rebuilt").get_version(at).unwrap();
-        assert_eq!(
-            version.image_coverage.query(4).as_deref(),
-            Some(replacement.as_str()),
-            "query for 4 at version {at} after eviction",
-        );
-
-        // download
-        let ret = map.replace(key, orig_layer.clone(), |l| l == &replacement);
-        assert!(
-            matches!(ret, Replacement::Replaced { .. }),
-            "replace {orig_layer} back: {ret:?}"
-        );
-        map.rebuild();
-        let version = map.get().expect("rebuilt").get_version(at).unwrap();
-        assert_eq!(
-            version.image_coverage.query(4).as_deref(),
-            Some(orig_layer.as_str()),
-            "query for 4 at version {at} after download",
-        );
-    }
-}
-
-#[test]
-fn missing_key_is_not_inserted_with_replace() {
-    let mut map = BufferedHistoricLayerCoverage::new();
-    let key = LayerKey {
-        key: 0..5,
-        lsn: 100..101,
-        is_image: true,
-    };
-
-    let ret = map.replace(&key, "should not replace", |_| true);
-    assert!(matches!(ret, Replacement::NotFound), "{ret:?}");
-    map.rebuild();
-    assert!(map
-        .get()
-        .expect("no changes to rebuild")
-        .get_version(102)
-        .is_none());
-}
-
-#[test]
-fn replacing_buffered_insert_and_remove() {
-    let mut map = BufferedHistoricLayerCoverage::new();
-    let key = LayerKey {
-        key: 0..5,
-        lsn: 100..101,
-        is_image: true,
-    };
-
-    map.insert(key.clone(), "Image 1");
-    let ret = map.replace(&key, "Remote Image 1", |&l| l == "Image 1");
-    assert!(
-        matches!(ret, Replacement::Replaced { in_buffered: true }),
-        "{ret:?}"
-    );
-    map.rebuild();
-
-    assert_eq!(
-        map.get()
-            .expect("rebuilt")
-            .get_version(102)
-            .unwrap()
-            .image_coverage
-            .query(4),
-        Some("Remote Image 1")
-    );
-
-    map.remove(key.clone());
-    let ret = map.replace(&key, "should not replace", |_| true);
-    assert!(
-        matches!(ret, Replacement::RemovalBuffered),
-        "cannot replace after scheduled remove: {ret:?}"
-    );
-
-    map.rebuild();
-
-    let ret = map.replace(&key, "should not replace", |_| true);
-    assert!(
-        matches!(ret, Replacement::NotFound),
-        "cannot replace after remove + rebuild: {ret:?}"
-    );
-
-    let at_version = map.get().expect("rebuilt").get_version(102);
-    assert!(at_version.is_none());
-}

pageserver/src/tenant/metadata.rs

@@ -232,13 +232,13 @@ impl TimelineMetadata {
 /// Save timeline metadata to file
 pub fn save_metadata(
     conf: &'static PageServerConf,
-    timeline_id: TimelineId,
-    tenant_id: TenantId,
+    tenant_id: &TenantId,
+    timeline_id: &TimelineId,
     data: &TimelineMetadata,
     first_save: bool,
 ) -> anyhow::Result<()> {
     let _enter = info_span!("saving metadata").entered();
-    let path = conf.metadata_path(timeline_id, tenant_id);
+    let path = conf.metadata_path(tenant_id, timeline_id);
     // use OpenOptions to ensure file presence is consistent with first_save
     let mut file = VirtualFile::open_with_options(
         &path,
@@ -267,10 +267,10 @@ pub fn save_metadata(
 
 pub fn load_metadata(
     conf: &'static PageServerConf,
-    timeline_id: TimelineId,
-    tenant_id: TenantId,
+    tenant_id: &TenantId,
+    timeline_id: &TimelineId,
 ) -> anyhow::Result<TimelineMetadata> {
-    let metadata_path = conf.metadata_path(timeline_id, tenant_id);
+    let metadata_path = conf.metadata_path(tenant_id, timeline_id);
     let metadata_bytes = std::fs::read(&metadata_path).with_context(|| {
         format!(
             "Failed to read metadata bytes from path {}",

pageserver/src/tenant/mgr.rs

@@ -184,9 +184,9 @@ pub fn schedule_local_tenant_processing(
             format!("Could not parse tenant id out of the tenant dir name in path {tenant_path:?}")
         })?;
 
-    let tenant_ignore_mark = conf.tenant_ignore_mark_file_path(tenant_id);
+    let tenant_ignore_mark = conf.tenant_ignore_mark_file_path(&tenant_id);
     anyhow::ensure!(
-        !conf.tenant_ignore_mark_file_path(tenant_id).exists(),
+        !conf.tenant_ignore_mark_file_path(&tenant_id).exists(),
         "Cannot load tenant, ignore mark found at {tenant_ignore_mark:?}"
     );
 
@@ -310,7 +310,7 @@ pub async fn create_tenant(
         // We're holding the tenants lock in write mode while doing local IO.
         // If this section ever becomes contentious, introduce a new `TenantState::Creating`
         // and do the work in that state.
-        let tenant_directory = super::create_tenant_files(conf, tenant_conf, tenant_id, CreateTenantFilesMode::Create)?;
+        let tenant_directory = super::create_tenant_files(conf, tenant_conf, &tenant_id, CreateTenantFilesMode::Create)?;
         // TODO: tenant directory remains on disk if we bail out from here on.
         //       See https://github.com/neondatabase/neon/issues/4233
 
@@ -344,14 +344,9 @@ pub async fn set_new_tenant_config(
     info!("configuring tenant {tenant_id}");
     let tenant = get_tenant(tenant_id, true).await?;
 
-    let tenant_config_path = conf.tenant_config_path(tenant_id);
-    Tenant::persist_tenant_config(
-        &tenant.tenant_id(),
-        &tenant_config_path,
-        new_tenant_conf,
-        false,
-    )
-    .map_err(SetNewTenantConfigError::Persist)?;
+    let tenant_config_path = conf.tenant_config_path(&tenant_id);
+    Tenant::persist_tenant_config(&tenant_id, &tenant_config_path, new_tenant_conf, false)
+        .map_err(SetNewTenantConfigError::Persist)?;
     tenant.set_new_tenant_config(new_tenant_conf);
     Ok(())
 }
@@ -435,7 +430,7 @@ pub async fn detach_tenant(
     // Ignored tenants are not present in memory and will bail the removal from memory operation.
     // Before returning the error, check for ignored tenant removal case — we only need to clean its local files then.
     if detach_ignored && matches!(removal_result, Err(TenantStateError::NotFound(_))) {
-        let tenant_ignore_mark = conf.tenant_ignore_mark_file_path(tenant_id);
+        let tenant_ignore_mark = conf.tenant_ignore_mark_file_path(&tenant_id);
         if tenant_ignore_mark.exists() {
             info!("Detaching an ignored tenant");
             local_files_cleanup_operation(tenant_id)
@@ -457,7 +452,7 @@ pub async fn load_tenant(
 ) -> Result<(), TenantMapInsertError> {
     tenant_map_insert(tenant_id, || {
         let tenant_path = conf.tenant_path(&tenant_id);
-        let tenant_ignore_mark = conf.tenant_ignore_mark_file_path(tenant_id);
+        let tenant_ignore_mark = conf.tenant_ignore_mark_file_path(&tenant_id);
         if tenant_ignore_mark.exists() {
             std::fs::remove_file(&tenant_ignore_mark)
                 .with_context(|| format!("Failed to remove tenant ignore mark {tenant_ignore_mark:?} during tenant loading"))?;
@@ -478,7 +473,7 @@ pub async fn ignore_tenant(
     tenant_id: TenantId,
 ) -> Result<(), TenantStateError> {
     remove_tenant_from_memory(tenant_id, async {
-        let ignore_mark_file = conf.tenant_ignore_mark_file_path(tenant_id);
+        let ignore_mark_file = conf.tenant_ignore_mark_file_path(&tenant_id);
         fs::File::create(&ignore_mark_file)
             .await
             .context("Failed to create ignore mark file")
@@ -525,7 +520,7 @@ pub async fn attach_tenant(
     ctx: &RequestContext,
 ) -> Result<(), TenantMapInsertError> {
     tenant_map_insert(tenant_id, || {
-        let tenant_dir = create_tenant_files(conf, tenant_conf, tenant_id, CreateTenantFilesMode::Attach)?;
+        let tenant_dir = create_tenant_files(conf, tenant_conf, &tenant_id, CreateTenantFilesMode::Attach)?;
         // TODO: tenant directory remains on disk if we bail out from here on.
         //       See https://github.com/neondatabase/neon/issues/4233
 
@@ -675,7 +670,7 @@ pub async fn immediate_gc(
         .get(&tenant_id)
         .map(Arc::clone)
         .with_context(|| format!("tenant {tenant_id}"))
-        .map_err(ApiError::NotFound)?;
+        .map_err(|e| ApiError::NotFound(e.into()))?;
 
     let gc_horizon = gc_req.gc_horizon.unwrap_or_else(|| tenant.get_gc_horizon());
     // Use tenant's pitr setting
@@ -695,7 +690,7 @@ pub async fn immediate_gc(
             fail::fail_point!("immediate_gc_task_pre");
             let result = tenant
                 .gc_iteration(Some(timeline_id), gc_horizon, pitr, &ctx)
-                .instrument(info_span!("manual_gc", tenant = %tenant_id, timeline = %timeline_id))
+                .instrument(info_span!("manual_gc", %tenant_id, %timeline_id))
                 .await;
                 // FIXME: `gc_iteration` can return an error for multiple reasons; we should handle it
                 // better once the types support it.
@@ -724,11 +719,11 @@ pub async fn immediate_compact(
         .get(&tenant_id)
         .map(Arc::clone)
         .with_context(|| format!("tenant {tenant_id}"))
-        .map_err(ApiError::NotFound)?;
+        .map_err(|e| ApiError::NotFound(e.into()))?;
 
     let timeline = tenant
         .get_timeline(timeline_id, true)
-        .map_err(ApiError::NotFound)?;
+        .map_err(|e| ApiError::NotFound(e.into()))?;
 
     // Run in task_mgr to avoid race with tenant_detach operation
     let ctx = ctx.detached_child(TaskKind::Compaction, DownloadBehavior::Download);
@@ -745,9 +740,7 @@ pub async fn immediate_compact(
         async move {
             let result = timeline
                 .compact(&ctx)
-                .instrument(
-                    info_span!("manual_compact", tenant = %tenant_id, timeline = %timeline_id),
-                )
+                .instrument(info_span!("manual_compact", %tenant_id, %timeline_id))
                 .await;
 
             match task_done.send(result) {

pageserver/src/tenant/remote_timeline_client.rs

@@ -442,8 +442,8 @@ impl RemoteTimelineClient {
         let index_part = download::download_index_part(
             self.conf,
             &self.storage_impl,
-            self.tenant_id,
-            self.timeline_id,
+            &self.tenant_id,
+            &self.timeline_id,
         )
         .measure_remote_op(
             self.tenant_id,
@@ -608,10 +608,7 @@ impl RemoteTimelineClient {
         self.calls_unfinished_metric_begin(&op);
         upload_queue.queued_operations.push_back(op);
 
-        info!(
-            "scheduled layer file upload {}",
-            layer_file_name.file_name()
-        );
+        info!("scheduled layer file upload {layer_file_name}");
 
         // Launch the task immediately, if possible
         self.launch_queued_tasks(upload_queue);
@@ -664,7 +661,7 @@ impl RemoteTimelineClient {
                 });
                 self.calls_unfinished_metric_begin(&op);
                 upload_queue.queued_operations.push_back(op);
-                info!("scheduled layer file deletion {}", name.file_name());
+                info!("scheduled layer file deletion {name}");
             }
 
             // Launch the tasks immediately, if possible
@@ -751,29 +748,13 @@ impl RemoteTimelineClient {
             stopped.deleted_at = SetDeletedFlagProgress::NotRunning;
         });
 
-        // Have a failpoint that can use the `pause` failpoint action.
-        // We don't want to block the executor thread, hence, spawn_blocking + await.
-        #[cfg(feature = "testing")]
-        tokio::task::spawn_blocking({
-            let current = tracing::Span::current();
-            move || {
-                let _entered = current.entered();
-                tracing::info!(
-                    "at failpoint persist_index_part_with_deleted_flag_after_set_before_upload_pause"
-                );
-                fail::fail_point!(
-                    "persist_index_part_with_deleted_flag_after_set_before_upload_pause"
-                );
-            }
-        })
-        .await
-        .expect("spawn_blocking");
+        pausable_failpoint!("persist_deleted_index_part");
 
         upload::upload_index_part(
             self.conf,
             &self.storage_impl,
-            self.tenant_id,
-            self.timeline_id,
+            &self.tenant_id,
+            &self.timeline_id,
             &index_part_with_deleted_at,
         )
         .await?;
@@ -832,7 +813,7 @@ impl RemoteTimelineClient {
                     .queued_operations
                     .push_back(op);
 
-                info!("scheduled layer file deletion {}", name.file_name());
+                info!("scheduled layer file deletion {name}");
                 deletions_queued += 1;
             }
 
@@ -848,7 +829,7 @@ impl RemoteTimelineClient {
 
         // Do not delete index part yet, it is needed for possible retry. If we remove it first
         // and retry will arrive to different pageserver there wont be any traces of it on remote storage
-        let timeline_path = self.conf.timeline_path(&self.timeline_id, &self.tenant_id);
+        let timeline_path = self.conf.timeline_path(&self.tenant_id, &self.timeline_id);
         let timeline_storage_path = self.conf.remote_path(&timeline_path)?;
 
         let remaining = self
@@ -866,10 +847,8 @@ impl RemoteTimelineClient {
                 "Found {} files not bound to index_file.json, proceeding with their deletion",
                 remaining.len()
             );
-            for file in remaining {
-                warn!("Removing {}", file.object_name().unwrap_or_default());
-                self.storage_impl.delete(&file).await?;
-            }
+            warn!("About to remove {} files", remaining.len());
+            self.storage_impl.delete_objects(&remaining).await?;
         }
 
         let index_file_path = timeline_storage_path.join(Path::new(IndexPart::FILE_NAME));
@@ -942,11 +921,11 @@ impl RemoteTimelineClient {
 
             // Assign unique ID to this task
             upload_queue.task_counter += 1;
-            let task_id = upload_queue.task_counter;
+            let upload_task_id = upload_queue.task_counter;
 
             // Add it to the in-progress map
             let task = Arc::new(UploadTask {
-                task_id,
+                task_id: upload_task_id,
                 op: next_op,
                 retries: AtomicU32::new(0),
             });
@@ -956,6 +935,8 @@ impl RemoteTimelineClient {
 
             // Spawn task to perform the task
             let self_rc = Arc::clone(self);
+            let tenant_id = self.tenant_id;
+            let timeline_id = self.timeline_id;
             task_mgr::spawn(
                 self.runtime.handle(),
                 TaskKind::RemoteUploadTask,
@@ -967,7 +948,7 @@ impl RemoteTimelineClient {
                     self_rc.perform_upload_task(task).await;
                     Ok(())
                 }
-                .instrument(info_span!(parent: None, "remote_upload", tenant = %self.tenant_id, timeline = %self.timeline_id, upload_task_id = %task_id)),
+                .instrument(info_span!(parent: None, "remote_upload", %tenant_id, %timeline_id, %upload_task_id)),
             );
 
             // Loop back to process next task
@@ -1012,7 +993,7 @@ impl RemoteTimelineClient {
                 UploadOp::UploadLayer(ref layer_file_name, ref layer_metadata) => {
                     let path = &self
                         .conf
-                        .timeline_path(&self.timeline_id, &self.tenant_id)
+                        .timeline_path(&self.tenant_id, &self.timeline_id)
                         .join(layer_file_name.file_name());
                     upload::upload_timeline_layer(
                         self.conf,
@@ -1033,8 +1014,8 @@ impl RemoteTimelineClient {
                     let res = upload::upload_index_part(
                         self.conf,
                         &self.storage_impl,
-                        self.tenant_id,
-                        self.timeline_id,
+                        &self.tenant_id,
+                        &self.timeline_id,
                         index_part,
                     )
                     .measure_remote_op(
@@ -1053,7 +1034,7 @@ impl RemoteTimelineClient {
                 UploadOp::Delete(delete) => {
                     let path = &self
                         .conf
-                        .timeline_path(&self.timeline_id, &self.tenant_id)
+                        .timeline_path(&self.tenant_id, &self.timeline_id)
                         .join(delete.layer_file_name.file_name());
                     delete::delete_layer(self.conf, &self.storage_impl, path)
                         .measure_remote_op(
@@ -1371,7 +1352,7 @@ mod tests {
     struct TestSetup {
         runtime: &'static tokio::runtime::Runtime,
         entered_runtime: EnterGuard<'static>,
-        harness: TenantHarness<'static>,
+        harness: TenantHarness,
         tenant: Arc<Tenant>,
         tenant_ctx: RequestContext,
         remote_fs_dir: PathBuf,

pageserver/src/tenant/remote_timeline_client/delete.rs

@@ -19,9 +19,10 @@ pub(super) async fn delete_layer<'a>(
 
     let path_to_delete = conf.remote_path(local_layer_path)?;
 
-    // XXX: If the deletion fails because the object already didn't exist,
-    // it would be good to just issue a warning but consider it success.
-    // https://github.com/neondatabase/neon/issues/2934
+    // We don't want to print an error if the delete failed if the file has
+    // already been deleted. Thankfully, in this situation S3 already
+    // does not yield an error. While OS-provided local file system APIs do yield
+    // errors, we avoid them in the `LocalFs` wrapper.
     storage.delete(&path_to_delete).await.with_context(|| {
         format!("Failed to delete remote layer from storage at {path_to_delete:?}")
     })

pageserver/src/tenant/remote_timeline_client/download.rs

@@ -16,7 +16,7 @@ use tracing::{info, warn};
 
 use crate::config::PageServerConf;
 use crate::tenant::storage_layer::LayerFileName;
-use crate::tenant::timeline::debug_assert_current_span_has_tenant_and_timeline_id;
+use crate::tenant::timeline::span::debug_assert_current_span_has_tenant_and_timeline_id;
 use crate::{exponential_backoff, DEFAULT_BASE_BACKOFF_SECONDS, DEFAULT_MAX_BACKOFF_SECONDS};
 use remote_storage::{DownloadError, GenericRemoteStorage};
 use utils::crashsafe::path_with_suffix_extension;
@@ -46,7 +46,7 @@ pub async fn download_layer_file<'a>(
 ) -> Result<u64, DownloadError> {
     debug_assert_current_span_has_tenant_and_timeline_id();
 
-    let timeline_path = conf.timeline_path(&timeline_id, &tenant_id);
+    let timeline_path = conf.timeline_path(&tenant_id, &timeline_id);
 
     let local_path = timeline_path.join(layer_file_name.file_name());
 
@@ -229,11 +229,11 @@ pub async fn list_remote_timelines<'a>(
 pub(super) async fn download_index_part(
     conf: &'static PageServerConf,
     storage: &GenericRemoteStorage,
-    tenant_id: TenantId,
-    timeline_id: TimelineId,
+    tenant_id: &TenantId,
+    timeline_id: &TimelineId,
 ) -> Result<IndexPart, DownloadError> {
     let index_part_path = conf
-        .metadata_path(timeline_id, tenant_id)
+        .metadata_path(tenant_id, timeline_id)
         .with_file_name(IndexPart::FILE_NAME);
     let part_storage_path = conf
         .remote_path(&index_part_path)

pageserver/src/tenant/remote_timeline_client/upload.rs

@@ -2,7 +2,7 @@
 
 use anyhow::{bail, Context};
 use fail::fail_point;
-use std::path::Path;
+use std::{io::ErrorKind, path::Path};
 use tokio::fs;
 
 use crate::{config::PageServerConf, tenant::remote_timeline_client::index::IndexPart};
@@ -11,12 +11,14 @@ use utils::id::{TenantId, TimelineId};
 
 use super::index::LayerFileMetadata;
 
+use tracing::info;
+
 /// Serializes and uploads the given index part data to the remote storage.
 pub(super) async fn upload_index_part<'a>(
     conf: &'static PageServerConf,
     storage: &'a GenericRemoteStorage,
-    tenant_id: TenantId,
-    timeline_id: TimelineId,
+    tenant_id: &TenantId,
+    timeline_id: &TimelineId,
     index_part: &'a IndexPart,
 ) -> anyhow::Result<()> {
     tracing::trace!("uploading new index part");
@@ -31,7 +33,7 @@ pub(super) async fn upload_index_part<'a>(
     let index_part_bytes = tokio::io::BufReader::new(std::io::Cursor::new(index_part_bytes));
 
     let index_part_path = conf
-        .metadata_path(timeline_id, tenant_id)
+        .metadata_path(tenant_id, timeline_id)
         .with_file_name(IndexPart::FILE_NAME);
     let storage_path = conf.remote_path(&index_part_path)?;
 
@@ -56,9 +58,22 @@ pub(super) async fn upload_timeline_layer<'a>(
     });
     let storage_path = conf.remote_path(source_path)?;
 
-    let source_file = fs::File::open(&source_path)
-        .await
-        .with_context(|| format!("Failed to open a source file for layer {source_path:?}"))?;
+    let source_file_res = fs::File::open(&source_path).await;
+    let source_file = match source_file_res {
+        Ok(source_file) => source_file,
+        Err(e) if e.kind() == ErrorKind::NotFound => {
+            // In some situations we might run into the underlying file being deleted by
+            // e.g. compaction before the uploader gets to it. In that instance, we don't
+            // want to retry the error: a deleted file won't come back. In theory, the
+            // file might not have been written in the first place, which also indicates
+            // a bug. Still log the situation so that we can keep an eye on it.
+            // See https://github.com/neondatabase/neon/issues/4526
+            info!(path = %source_path.display(), "File to upload doesn't exist. Likely the file has been deleted and an upload is not required any more.");
+            return Ok(());
+        }
+        Err(e) => Err(e)
+            .with_context(|| format!("Failed to open a source file for layer {source_path:?}"))?,
+    };
 
     let fs_size = source_file
         .metadata()

pageserver/src/tenant/span.rs

@@ -0,0 +1,20 @@
+#[cfg(debug_assertions)]
+use utils::tracing_span_assert::{check_fields_present, MultiNameExtractor};
+
+#[cfg(not(debug_assertions))]
+pub(crate) fn debug_assert_current_span_has_tenant_id() {}
+
+#[cfg(debug_assertions)]
+pub(crate) static TENANT_ID_EXTRACTOR: once_cell::sync::Lazy<MultiNameExtractor<1>> =
+    once_cell::sync::Lazy::new(|| MultiNameExtractor::new("TenantId", ["tenant_id"]));
+
+#[cfg(debug_assertions)]
+#[track_caller]
+pub(crate) fn debug_assert_current_span_has_tenant_id() {
+    if let Err(missing) = check_fields_present([&*TENANT_ID_EXTRACTOR]) {
+        panic!(
+            "missing extractors: {:?}",
+            missing.into_iter().map(|e| e.name()).collect::<Vec<_>>()
+        )
+    }
+}

pageserver/src/tenant/storage_layer.rs

@@ -41,7 +41,7 @@ pub use inmemory_layer::InMemoryLayer;
 pub use layer_desc::{PersistentLayerDesc, PersistentLayerKey};
 pub use remote_layer::RemoteLayer;
 
-use super::layer_map::BatchedUpdates;
+use super::timeline::layer_manager::LayerManager;
 
 pub fn range_overlaps<T>(a: &Range<T>, b: &Range<T>) -> bool
 where
@@ -54,13 +54,6 @@ where
     }
 }
 
-pub fn range_eq<T>(a: &Range<T>, b: &Range<T>) -> bool
-where
-    T: PartialEq<T>,
-{
-    a.start == b.start && a.end == b.end
-}
-
 /// Struct used to communicate across calls to 'get_value_reconstruct_data'.
 ///
 /// Before first call, you can fill in 'page_img' if you have an older cached
@@ -176,13 +169,10 @@ impl LayerAccessStats {
     /// Create an empty stats object and record a [`LayerLoad`] event with the given residence status.
     ///
     /// See [`record_residence_event`] for why you need to do this while holding the layer map lock.
-    pub(crate) fn for_loading_layer<L>(
-        layer_map_lock_held_witness: &BatchedUpdates<'_, L>,
+    pub(crate) fn for_loading_layer(
+        layer_map_lock_held_witness: &LayerManager,
         status: LayerResidenceStatus,
-    ) -> Self
-    where
-        L: ?Sized + Layer,
-    {
+    ) -> Self {
         let new = LayerAccessStats(Mutex::new(LayerAccessStatsLocked::default()));
         new.record_residence_event(
             layer_map_lock_held_witness,
@@ -197,14 +187,11 @@ impl LayerAccessStats {
     /// The `new_status` is not recorded in `self`.
     ///
     /// See [`record_residence_event`] for why you need to do this while holding the layer map lock.
-    pub(crate) fn clone_for_residence_change<L>(
+    pub(crate) fn clone_for_residence_change(
         &self,
-        layer_map_lock_held_witness: &BatchedUpdates<'_, L>,
+        layer_map_lock_held_witness: &LayerManager,
         new_status: LayerResidenceStatus,
-    ) -> LayerAccessStats
-    where
-        L: ?Sized + Layer,
-    {
+    ) -> LayerAccessStats {
         let clone = {
             let inner = self.0.lock().unwrap();
             inner.clone()
@@ -232,14 +219,12 @@ impl LayerAccessStats {
     /// - Compact: Grab layer map lock, add the new L1 to layer map and remove the L0s, release layer map lock.
     /// - Eviction: observes the new L1 layer whose only activity timestamp is the LayerCreate event.
     ///
-    pub(crate) fn record_residence_event<L>(
+    pub(crate) fn record_residence_event(
         &self,
-        _layer_map_lock_held_witness: &BatchedUpdates<'_, L>,
+        _layer_map_lock_held_witness: &LayerManager,
         status: LayerResidenceStatus,
         reason: LayerResidenceEventReason,
-    ) where
-        L: ?Sized + Layer,
-    {
+    ) {
         let mut locked = self.0.lock().unwrap();
         locked.iter_mut().for_each(|inner| {
             inner
@@ -343,7 +328,7 @@ impl LayerAccessStats {
 /// All layers should implement a minimal `std::fmt::Debug` without tenant or
 /// timeline names, because those are known in the context of which the layers
 /// are used in (timeline).
-pub trait Layer: std::fmt::Debug + Send + Sync {
+pub trait Layer: std::fmt::Debug + std::fmt::Display + Send + Sync {
     /// Range of keys that this layer covers
     fn get_key_range(&self) -> Range<Key>;
 
@@ -381,9 +366,6 @@ pub trait Layer: std::fmt::Debug + Send + Sync {
         ctx: &RequestContext,
     ) -> Result<ValueReconstructResult>;
 
-    /// A short ID string that uniquely identifies the given layer within a [`LayerMap`].
-    fn short_id(&self) -> String;
-
     /// Dump summary of the contents of the layer to stdout
     fn dump(&self, verbose: bool, ctx: &RequestContext) -> Result<()>;
 }
@@ -394,6 +376,12 @@ pub type LayerIter<'i> = Box<dyn Iterator<Item = Result<(Key, Lsn, Value)>> + 'i
 /// Returned by [`Layer::key_iter`]
 pub type LayerKeyIter<'i> = Box<dyn Iterator<Item = (Key, Lsn, u64)> + 'i + Send>;
 
+/// Get a layer descriptor from a layer.
+pub trait AsLayerDesc {
+    /// Get the layer descriptor.
+    fn layer_desc(&self) -> &PersistentLayerDesc;
+}
+
 /// A Layer contains all data in a "rectangle" consisting of a range of keys and
 /// range of LSNs.
 ///
@@ -407,10 +395,8 @@ pub type LayerKeyIter<'i> = Box<dyn Iterator<Item = (Key, Lsn, u64)> + 'i + Send
 /// A delta layer contains all modifications within a range of LSNs and keys.
 /// An image layer is a snapshot of all the data in a key-range, at a single
 /// LSN.
-pub trait PersistentLayer: Layer {
-    /// Get the layer descriptor.
-    fn layer_desc(&self) -> &PersistentLayerDesc;
-
+pub trait PersistentLayer: Layer + AsLayerDesc {
+    /// Identify the tenant this layer belongs to
     fn get_tenant_id(&self) -> TenantId {
         self.layer_desc().tenant_id
     }
@@ -473,94 +459,40 @@ pub fn downcast_remote_layer(
     }
 }
 
-/// Holds metadata about a layer without any content. Used mostly for testing.
-///
-/// To use filenames as fixtures, parse them as [`LayerFileName`] then convert from that to a
-/// LayerDescriptor.
-#[derive(Clone, Debug)]
-pub struct LayerDescriptor {
-    pub key: Range<Key>,
-    pub lsn: Range<Lsn>,
-    pub is_incremental: bool,
-    pub short_id: String,
-}
+pub mod tests {
+    use super::*;
 
-impl LayerDescriptor {
-    /// `LayerDescriptor` is only used for testing purpose so it does not matter whether it is image / delta,
-    /// and the tenant / timeline id does not matter.
-    pub fn get_persistent_layer_desc(&self) -> PersistentLayerDesc {
-        PersistentLayerDesc::new_delta(
-            TenantId::from_array([0; 16]),
-            TimelineId::from_array([0; 16]),
-            self.key.clone(),
-            self.lsn.clone(),
-            233,
-        )
-    }
-}
-
-impl Layer for LayerDescriptor {
-    fn get_key_range(&self) -> Range<Key> {
-        self.key.clone()
-    }
-
-    fn get_lsn_range(&self) -> Range<Lsn> {
-        self.lsn.clone()
-    }
-
-    fn is_incremental(&self) -> bool {
-        self.is_incremental
-    }
-
-    fn get_value_reconstruct_data(
-        &self,
-        _key: Key,
-        _lsn_range: Range<Lsn>,
-        _reconstruct_data: &mut ValueReconstructState,
-        _ctx: &RequestContext,
-    ) -> Result<ValueReconstructResult> {
-        todo!("This method shouldn't be part of the Layer trait")
-    }
-
-    fn short_id(&self) -> String {
-        self.short_id.clone()
-    }
-
-    fn dump(&self, _verbose: bool, _ctx: &RequestContext) -> Result<()> {
-        todo!()
-    }
-}
-
-impl From<DeltaFileName> for LayerDescriptor {
-    fn from(value: DeltaFileName) -> Self {
-        let short_id = value.to_string();
-        LayerDescriptor {
-            key: value.key_range,
-            lsn: value.lsn_range,
-            is_incremental: true,
-            short_id,
+    impl From<DeltaFileName> for PersistentLayerDesc {
+        fn from(value: DeltaFileName) -> Self {
+            PersistentLayerDesc::new_delta(
+                TenantId::from_array([0; 16]),
+                TimelineId::from_array([0; 16]),
+                value.key_range,
+                value.lsn_range,
+                233,
+            )
         }
     }
-}
 
-impl From<ImageFileName> for LayerDescriptor {
-    fn from(value: ImageFileName) -> Self {
-        let short_id = value.to_string();
-        let lsn = value.lsn_as_range();
-        LayerDescriptor {
-            key: value.key_range,
-            lsn,
-            is_incremental: false,
-            short_id,
+    impl From<ImageFileName> for PersistentLayerDesc {
+        fn from(value: ImageFileName) -> Self {
+            PersistentLayerDesc::new_img(
+                TenantId::from_array([0; 16]),
+                TimelineId::from_array([0; 16]),
+                value.key_range,
+                value.lsn,
+                false,
+                233,
+            )
         }
     }
-}
 
-impl From<LayerFileName> for LayerDescriptor {
-    fn from(value: LayerFileName) -> Self {
-        match value {
-            LayerFileName::Delta(d) => Self::from(d),
-            LayerFileName::Image(i) => Self::from(i),
+    impl From<LayerFileName> for PersistentLayerDesc {
+        fn from(value: LayerFileName) -> Self {
+            match value {
+                LayerFileName::Delta(d) => Self::from(d),
+                LayerFileName::Image(i) => Self::from(i),
+            }
         }
     }
 }

pageserver/src/tenant/storage_layer/delta_layer.rs

@@ -10,7 +10,7 @@
 //! The delta files are stored in timelines/<timeline_id> directory.  Currently,
 //! there are no subdirectories, and each delta file is named like this:
 //!
-//!    <key start>-<key end>__<start LSN>-<end LSN
+//!    <key start>-<key end>__<start LSN>-<end LSN>
 //!
 //! For example:
 //!
@@ -56,8 +56,8 @@ use utils::{
 };
 
 use super::{
-    DeltaFileName, Layer, LayerAccessStats, LayerAccessStatsReset, LayerIter, LayerKeyIter,
-    PathOrConf, PersistentLayerDesc,
+    AsLayerDesc, DeltaFileName, Layer, LayerAccessStats, LayerAccessStatsReset, LayerIter,
+    LayerKeyIter, PathOrConf, PersistentLayerDesc,
 };
 
 ///
@@ -222,13 +222,14 @@ impl Layer for DeltaLayer {
     /// debugging function to print out the contents of the layer
     fn dump(&self, verbose: bool, ctx: &RequestContext) -> Result<()> {
         println!(
-            "----- delta layer for ten {} tli {} keys {}-{} lsn {}-{} ----",
+            "----- delta layer for ten {} tli {} keys {}-{} lsn {}-{} size {} ----",
             self.desc.tenant_id,
             self.desc.timeline_id,
             self.desc.key_range.start,
             self.desc.key_range.end,
             self.desc.lsn_range.start,
-            self.desc.lsn_range.end
+            self.desc.lsn_range.end,
+            self.desc.file_size,
         );
 
         if !verbose {
@@ -394,18 +395,21 @@ impl Layer for DeltaLayer {
     fn is_incremental(&self) -> bool {
         self.layer_desc().is_incremental
     }
+}
+/// Boilerplate to implement the Layer trait, always use layer_desc for persistent layers.
+impl std::fmt::Display for DeltaLayer {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(f, "{}", self.layer_desc().short_id())
+    }
+}
 
-    /// Boilerplate to implement the Layer trait, always use layer_desc for persistent layers.
-    fn short_id(&self) -> String {
-        self.layer_desc().short_id()
+impl AsLayerDesc for DeltaLayer {
+    fn layer_desc(&self) -> &PersistentLayerDesc {
+        &self.desc
     }
 }
 
 impl PersistentLayer for DeltaLayer {
-    fn layer_desc(&self) -> &PersistentLayerDesc {
-        &self.desc
-    }
-
     fn local_path(&self) -> Option<PathBuf> {
         Some(self.path())
     }
@@ -457,22 +461,22 @@ impl PersistentLayer for DeltaLayer {
 impl DeltaLayer {
     fn path_for(
         path_or_conf: &PathOrConf,
-        timeline_id: TimelineId,
-        tenant_id: TenantId,
+        tenant_id: &TenantId,
+        timeline_id: &TimelineId,
         fname: &DeltaFileName,
     ) -> PathBuf {
         match path_or_conf {
             PathOrConf::Path(path) => path.clone(),
             PathOrConf::Conf(conf) => conf
-                .timeline_path(&timeline_id, &tenant_id)
+                .timeline_path(tenant_id, timeline_id)
                 .join(fname.to_string()),
         }
     }
 
     fn temp_path_for(
         conf: &PageServerConf,
-        timeline_id: TimelineId,
-        tenant_id: TenantId,
+        tenant_id: &TenantId,
+        timeline_id: &TimelineId,
         key_start: Key,
         lsn_range: &Range<Lsn>,
     ) -> PathBuf {
@@ -482,7 +486,7 @@ impl DeltaLayer {
             .map(char::from)
             .collect();
 
-        conf.timeline_path(&timeline_id, &tenant_id).join(format!(
+        conf.timeline_path(tenant_id, timeline_id).join(format!(
             "{}-XXX__{:016X}-{:016X}.{}.{}",
             key_start,
             u64::from(lsn_range.start),
@@ -604,8 +608,8 @@ impl DeltaLayer {
     pub fn path(&self) -> PathBuf {
         Self::path_for(
             &self.path_or_conf,
-            self.desc.timeline_id,
-            self.desc.tenant_id,
+            &self.desc.tenant_id,
+            &self.desc.timeline_id,
             &self.layer_name(),
         )
     }
@@ -653,7 +657,7 @@ impl DeltaLayerWriterInner {
         //
         // Note: This overwrites any existing file. There shouldn't be any.
         // FIXME: throw an error instead?
-        let path = DeltaLayer::temp_path_for(conf, timeline_id, tenant_id, key_start, &lsn_range);
+        let path = DeltaLayer::temp_path_for(conf, &tenant_id, &timeline_id, key_start, &lsn_range);
 
         let mut file = VirtualFile::create(&path)?;
         // make room for the header block
@@ -768,8 +772,8 @@ impl DeltaLayerWriterInner {
         // FIXME: throw an error instead?
         let final_path = DeltaLayer::path_for(
             &PathOrConf::Conf(self.conf),
-            self.timeline_id,
-            self.tenant_id,
+            &self.tenant_id,
+            &self.timeline_id,
             &DeltaFileName {
                 key_range: self.key_start..key_end,
                 lsn_range: self.lsn_range,

pageserver/src/tenant/storage_layer/filename.rs

@@ -210,9 +210,15 @@ pub enum LayerFileName {
 
 impl LayerFileName {
     pub fn file_name(&self) -> String {
+        self.to_string()
+    }
+}
+
+impl fmt::Display for LayerFileName {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
         match self {
-            Self::Image(fname) => fname.to_string(),
-            Self::Delta(fname) => fname.to_string(),
+            Self::Image(fname) => write!(f, "{fname}"),
+            Self::Delta(fname) => write!(f, "{fname}"),
         }
     }
 }

pageserver/src/tenant/storage_layer/image_layer.rs

@@ -53,7 +53,9 @@ use utils::{
 };
 
 use super::filename::ImageFileName;
-use super::{Layer, LayerAccessStatsReset, LayerIter, PathOrConf, PersistentLayerDesc};
+use super::{
+    AsLayerDesc, Layer, LayerAccessStatsReset, LayerIter, PathOrConf, PersistentLayerDesc,
+};
 
 ///
 /// Header stored in the beginning of the file
@@ -153,12 +155,14 @@ impl Layer for ImageLayer {
     /// debugging function to print out the contents of the layer
     fn dump(&self, verbose: bool, ctx: &RequestContext) -> Result<()> {
         println!(
-            "----- image layer for ten {} tli {} key {}-{} at {} ----",
+            "----- image layer for ten {} tli {} key {}-{} at {} is_incremental {} size {} ----",
             self.desc.tenant_id,
             self.desc.timeline_id,
             self.desc.key_range.start,
             self.desc.key_range.end,
-            self.lsn
+            self.lsn,
+            self.desc.is_incremental,
+            self.desc.file_size
         );
 
         if !verbose {
@@ -230,18 +234,22 @@ impl Layer for ImageLayer {
     fn is_incremental(&self) -> bool {
         self.layer_desc().is_incremental
     }
+}
 
-    /// Boilerplate to implement the Layer trait, always use layer_desc for persistent layers.
-    fn short_id(&self) -> String {
-        self.layer_desc().short_id()
+/// Boilerplate to implement the Layer trait, always use layer_desc for persistent layers.
+impl std::fmt::Display for ImageLayer {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(f, "{}", self.layer_desc().short_id())
+    }
+}
+
+impl AsLayerDesc for ImageLayer {
+    fn layer_desc(&self) -> &PersistentLayerDesc {
+        &self.desc
     }
 }
 
 impl PersistentLayer for ImageLayer {
-    fn layer_desc(&self) -> &PersistentLayerDesc {
-        &self.desc
-    }
-
     fn local_path(&self) -> Option<PathBuf> {
         Some(self.path())
     }
@@ -284,7 +292,7 @@ impl ImageLayer {
         match path_or_conf {
             PathOrConf::Path(path) => path.to_path_buf(),
             PathOrConf::Conf(conf) => conf
-                .timeline_path(&timeline_id, &tenant_id)
+                .timeline_path(&tenant_id, &timeline_id)
                 .join(fname.to_string()),
         }
     }
@@ -301,7 +309,7 @@ impl ImageLayer {
             .map(char::from)
             .collect();
 
-        conf.timeline_path(&timeline_id, &tenant_id)
+        conf.timeline_path(&tenant_id, &timeline_id)
             .join(format!("{fname}.{rand_string}.{TEMP_FILE_SUFFIX}"))
     }
 

pageserver/src/tenant/storage_layer/inmemory_layer.rs

@@ -131,13 +131,6 @@ impl Layer for InMemoryLayer {
         true
     }
 
-    fn short_id(&self) -> String {
-        let inner = self.inner.read().unwrap();
-
-        let end_lsn = inner.end_lsn.unwrap_or(Lsn(u64::MAX));
-        format!("inmem-{:016X}-{:016X}", self.start_lsn.0, end_lsn.0)
-    }
-
     /// debugging function to print out the contents of the layer
     fn dump(&self, verbose: bool, _ctx: &RequestContext) -> Result<()> {
         let inner = self.inner.read().unwrap();
@@ -240,6 +233,15 @@ impl Layer for InMemoryLayer {
     }
 }
 
+impl std::fmt::Display for InMemoryLayer {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        let inner = self.inner.read().unwrap();
+
+        let end_lsn = inner.end_lsn.unwrap_or(Lsn(u64::MAX));
+        write!(f, "inmem-{:016X}-{:016X}", self.start_lsn.0, end_lsn.0)
+    }
+}
+
 impl InMemoryLayer {
     ///
     /// Get layer size on the disk

pageserver/src/tenant/storage_layer/layer_desc.rs

@@ -1,4 +1,5 @@
 use anyhow::Result;
+use core::fmt::Display;
 use std::ops::Range;
 use utils::{
     id::{TenantId, TimelineId},
@@ -48,8 +49,8 @@ impl PersistentLayerDesc {
         }
     }
 
-    pub fn short_id(&self) -> String {
-        self.filename().file_name()
+    pub fn short_id(&self) -> impl Display {
+        self.filename()
     }
 
     #[cfg(test)]
@@ -173,13 +174,16 @@ impl PersistentLayerDesc {
 
     pub fn dump(&self, _verbose: bool, _ctx: &RequestContext) -> Result<()> {
         println!(
-            "----- layer for ten {} tli {} keys {}-{} lsn {}-{} ----",
+            "----- layer for ten {} tli {} keys {}-{} lsn {}-{} is_delta {} is_incremental {} size {} ----",
             self.tenant_id,
             self.timeline_id,
             self.key_range.start,
             self.key_range.end,
             self.lsn_range.start,
-            self.lsn_range.end
+            self.lsn_range.end,
+            self.is_delta,
+            self.is_incremental,
+            self.file_size,
         );
 
         Ok(())

pageserver/src/tenant/storage_layer/remote_layer.rs

@@ -4,9 +4,9 @@
 use crate::config::PageServerConf;
 use crate::context::RequestContext;
 use crate::repository::Key;
-use crate::tenant::layer_map::BatchedUpdates;
 use crate::tenant::remote_timeline_client::index::LayerFileMetadata;
 use crate::tenant::storage_layer::{Layer, ValueReconstructResult, ValueReconstructState};
+use crate::tenant::timeline::layer_manager::LayerManager;
 use anyhow::{bail, Result};
 use pageserver_api::models::HistoricLayerInfo;
 use std::ops::Range;
@@ -20,8 +20,8 @@ use utils::{
 
 use super::filename::{DeltaFileName, ImageFileName};
 use super::{
-    DeltaLayer, ImageLayer, LayerAccessStats, LayerAccessStatsReset, LayerIter, LayerKeyIter,
-    LayerResidenceStatus, PersistentLayer, PersistentLayerDesc,
+    AsLayerDesc, DeltaLayer, ImageLayer, LayerAccessStats, LayerAccessStatsReset, LayerIter,
+    LayerKeyIter, LayerResidenceStatus, PersistentLayer, PersistentLayerDesc,
 };
 
 /// RemoteLayer is a not yet downloaded [`ImageLayer`] or
@@ -71,22 +71,22 @@ impl Layer for RemoteLayer {
         _reconstruct_state: &mut ValueReconstructState,
         _ctx: &RequestContext,
     ) -> Result<ValueReconstructResult> {
-        bail!(
-            "layer {} needs to be downloaded",
-            self.filename().file_name()
-        );
+        bail!("layer {self} needs to be downloaded");
     }
 
     /// debugging function to print out the contents of the layer
     fn dump(&self, _verbose: bool, _ctx: &RequestContext) -> Result<()> {
         println!(
-            "----- remote layer for ten {} tli {} keys {}-{} lsn {}-{} ----",
+            "----- remote layer for ten {} tli {} keys {}-{} lsn {}-{} is_delta {} is_incremental {} size {} ----",
             self.desc.tenant_id,
             self.desc.timeline_id,
             self.desc.key_range.start,
             self.desc.key_range.end,
             self.desc.lsn_range.start,
-            self.desc.lsn_range.end
+            self.desc.lsn_range.end,
+            self.desc.is_delta,
+            self.desc.is_incremental,
+            self.desc.file_size,
         );
 
         Ok(())
@@ -106,18 +106,22 @@ impl Layer for RemoteLayer {
     fn is_incremental(&self) -> bool {
         self.layer_desc().is_incremental
     }
+}
 
-    /// Boilerplate to implement the Layer trait, always use layer_desc for persistent layers.
-    fn short_id(&self) -> String {
-        self.layer_desc().short_id()
+/// Boilerplate to implement the Layer trait, always use layer_desc for persistent layers.
+impl std::fmt::Display for RemoteLayer {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(f, "{}", self.layer_desc().short_id())
+    }
+}
+
+impl AsLayerDesc for RemoteLayer {
+    fn layer_desc(&self) -> &PersistentLayerDesc {
+        &self.desc
     }
 }
 
 impl PersistentLayer for RemoteLayer {
-    fn layer_desc(&self) -> &PersistentLayerDesc {
-        &self.desc
-    }
-
     fn local_path(&self) -> Option<PathBuf> {
         None
     }
@@ -218,15 +222,12 @@ impl RemoteLayer {
     }
 
     /// Create a Layer struct representing this layer, after it has been downloaded.
-    pub fn create_downloaded_layer<L>(
+    pub fn create_downloaded_layer(
         &self,
-        layer_map_lock_held_witness: &BatchedUpdates<'_, L>,
+        layer_map_lock_held_witness: &LayerManager,
         conf: &'static PageServerConf,
         file_size: u64,
-    ) -> Arc<dyn PersistentLayer>
-    where
-        L: ?Sized + Layer,
-    {
+    ) -> Arc<dyn PersistentLayer> {
         if self.desc.is_delta {
             let fname = self.desc.delta_file_name();
             Arc::new(DeltaLayer::new(

pageserver/src/tenant/timeline/eviction_task.rs

@@ -70,7 +70,6 @@ impl Timeline {
                 };
 
                 self_clone.eviction_task(cancel).await;
-                info!("eviction task finishing");
                 Ok(())
             },
         );
@@ -78,6 +77,9 @@ impl Timeline {
 
     #[instrument(skip_all, fields(tenant_id = %self.tenant_id, timeline_id = %self.timeline_id))]
     async fn eviction_task(self: Arc<Self>, cancel: CancellationToken) {
+        scopeguard::defer! {
+            info!("eviction task finishing");
+        }
         use crate::tenant::tasks::random_init_delay;
         {
             let policy = self.get_eviction_policy();
@@ -86,7 +88,6 @@ impl Timeline {
                 EvictionPolicy::NoEviction => Duration::from_secs(10),
             };
             if random_init_delay(period, &cancel).await.is_err() {
-                info!("shutting down");
                 return;
             }
         }
@@ -101,7 +102,6 @@ impl Timeline {
                 ControlFlow::Continue(sleep_until) => {
                     tokio::select! {
                         _ = cancel.cancelled() => {
-                            info!("shutting down");
                             break;
                         }
                         _ = tokio::time::sleep_until(sleep_until) => { }
@@ -197,9 +197,11 @@ impl Timeline {
         // We don't want to hold the layer map lock during eviction.
         // So, we just need to deal with this.
         let candidates: Vec<Arc<dyn PersistentLayer>> = {
-            let layers = self.layers.read().await;
+            let guard = self.layers.read().await;
+            let layers = guard.layer_map();
             let mut candidates = Vec::new();
             for hist_layer in layers.iter_historic_layers() {
+                let hist_layer = guard.get_from_desc(&hist_layer);
                 if hist_layer.is_remote_layer() {
                     continue;
                 }
@@ -207,7 +209,7 @@ impl Timeline {
                 let last_activity_ts = hist_layer.access_stats().latest_activity().unwrap_or_else(|| {
                     // We only use this fallback if there's an implementation error.
                     // `latest_activity` already does rate-limited warn!() log.
-                    debug!(layer=%hist_layer.filename().file_name(), "last_activity returns None, using SystemTime::now");
+                    debug!(layer=%hist_layer, "last_activity returns None, using SystemTime::now");
                     SystemTime::now()
                 });
 

pageserver/src/tenant/timeline/layer_manager.rs

@@ -0,0 +1,370 @@
+use anyhow::{bail, ensure, Context, Result};
+use std::{collections::HashMap, sync::Arc};
+use tracing::trace;
+use utils::{
+    id::{TenantId, TimelineId},
+    lsn::{AtomicLsn, Lsn},
+};
+
+use crate::{
+    config::PageServerConf,
+    metrics::TimelineMetrics,
+    tenant::{
+        layer_map::{BatchedUpdates, LayerMap},
+        storage_layer::{
+            AsLayerDesc, DeltaLayer, ImageLayer, InMemoryLayer, Layer, PersistentLayer,
+            PersistentLayerDesc, PersistentLayerKey, RemoteLayer,
+        },
+        timeline::compare_arced_layers,
+    },
+};
+
+/// Provides semantic APIs to manipulate the layer map.
+pub struct LayerManager {
+    layer_map: LayerMap,
+    layer_fmgr: LayerFileManager,
+}
+
+/// After GC, the layer map changes will not be applied immediately. Users should manually apply the changes after
+/// scheduling deletes in remote client.
+pub struct ApplyGcResultGuard<'a>(BatchedUpdates<'a>);
+
+impl ApplyGcResultGuard<'_> {
+    pub fn flush(self) {
+        self.0.flush();
+    }
+}
+
+impl LayerManager {
+    pub fn create() -> Self {
+        Self {
+            layer_map: LayerMap::default(),
+            layer_fmgr: LayerFileManager::new(),
+        }
+    }
+
+    pub fn get_from_desc(&self, desc: &PersistentLayerDesc) -> Arc<dyn PersistentLayer> {
+        self.layer_fmgr.get_from_desc(desc)
+    }
+
+    /// Get an immutable reference to the layer map.
+    ///
+    /// We expect users only to be able to get an immutable layer map. If users want to make modifications,
+    /// they should use the below semantic APIs. This design makes us step closer to immutable storage state.
+    pub fn layer_map(&self) -> &LayerMap {
+        &self.layer_map
+    }
+
+    /// Get a mutable reference to the layer map. This function will be removed once `flush_frozen_layer`
+    /// gets a refactor.
+    pub fn layer_map_mut(&mut self) -> &mut LayerMap {
+        &mut self.layer_map
+    }
+
+    /// Replace layers in the layer file manager, used in evictions and layer downloads.
+    pub fn replace_and_verify(
+        &mut self,
+        expected: Arc<dyn PersistentLayer>,
+        new: Arc<dyn PersistentLayer>,
+    ) -> Result<()> {
+        self.layer_fmgr.replace_and_verify(expected, new)
+    }
+
+    /// Called from `load_layer_map`. Initialize the layer manager with:
+    /// 1. all on-disk layers
+    /// 2. next open layer (with disk disk_consistent_lsn LSN)
+    pub fn initialize_local_layers(
+        &mut self,
+        on_disk_layers: Vec<Arc<dyn PersistentLayer>>,
+        next_open_layer_at: Lsn,
+    ) {
+        let mut updates = self.layer_map.batch_update();
+        for layer in on_disk_layers {
+            Self::insert_historic_layer(layer, &mut updates, &mut self.layer_fmgr);
+        }
+        updates.flush();
+        self.layer_map.next_open_layer_at = Some(next_open_layer_at);
+    }
+
+    /// Initialize when creating a new timeline, called in `init_empty_layer_map`.
+    pub fn initialize_empty(&mut self, next_open_layer_at: Lsn) {
+        self.layer_map.next_open_layer_at = Some(next_open_layer_at);
+    }
+
+    pub fn initialize_remote_layers(
+        &mut self,
+        corrupted_local_layers: Vec<Arc<dyn PersistentLayer>>,
+        remote_layers: Vec<Arc<RemoteLayer>>,
+    ) {
+        let mut updates = self.layer_map.batch_update();
+        for layer in corrupted_local_layers {
+            Self::remove_historic_layer(layer, &mut updates, &mut self.layer_fmgr);
+        }
+        for layer in remote_layers {
+            Self::insert_historic_layer(layer, &mut updates, &mut self.layer_fmgr);
+        }
+        updates.flush();
+    }
+
+    /// Open a new writable layer to append data if there is no open layer, otherwise return the current open layer,
+    /// called within `get_layer_for_write`.
+    pub fn get_layer_for_write(
+        &mut self,
+        lsn: Lsn,
+        last_record_lsn: Lsn,
+        conf: &'static PageServerConf,
+        timeline_id: TimelineId,
+        tenant_id: TenantId,
+    ) -> Result<Arc<InMemoryLayer>> {
+        ensure!(lsn.is_aligned());
+
+        ensure!(
+            lsn > last_record_lsn,
+            "cannot modify relation after advancing last_record_lsn (incoming_lsn={}, last_record_lsn={})\n{}",
+            lsn,
+            last_record_lsn,
+            std::backtrace::Backtrace::force_capture(),
+        );
+
+        // Do we have a layer open for writing already?
+        let layer = if let Some(open_layer) = &self.layer_map.open_layer {
+            if open_layer.get_lsn_range().start > lsn {
+                bail!(
+                    "unexpected open layer in the future: open layers starts at {}, write lsn {}",
+                    open_layer.get_lsn_range().start,
+                    lsn
+                );
+            }
+
+            Arc::clone(open_layer)
+        } else {
+            // No writeable layer yet. Create one.
+            let start_lsn = self
+                .layer_map
+                .next_open_layer_at
+                .context("No next open layer found")?;
+
+            trace!(
+                "creating in-memory layer at {}/{} for record at {}",
+                timeline_id,
+                start_lsn,
+                lsn
+            );
+
+            let new_layer = InMemoryLayer::create(conf, timeline_id, tenant_id, start_lsn)?;
+            let layer = Arc::new(new_layer);
+
+            self.layer_map.open_layer = Some(layer.clone());
+            self.layer_map.next_open_layer_at = None;
+
+            layer
+        };
+
+        Ok(layer)
+    }
+
+    /// Called from `freeze_inmem_layer`, returns true if successfully frozen.
+    pub fn try_freeze_in_memory_layer(
+        &mut self,
+        Lsn(last_record_lsn): Lsn,
+        last_freeze_at: &AtomicLsn,
+    ) {
+        let end_lsn = Lsn(last_record_lsn + 1);
+
+        if let Some(open_layer) = &self.layer_map.open_layer {
+            let open_layer_rc = Arc::clone(open_layer);
+            // Does this layer need freezing?
+            open_layer.freeze(end_lsn);
+
+            // The layer is no longer open, update the layer map to reflect this.
+            // We will replace it with on-disk historics below.
+            self.layer_map.frozen_layers.push_back(open_layer_rc);
+            self.layer_map.open_layer = None;
+            self.layer_map.next_open_layer_at = Some(end_lsn);
+            last_freeze_at.store(end_lsn);
+        }
+    }
+
+    /// Add image layers to the layer map, called from `create_image_layers`.
+    pub fn track_new_image_layers(&mut self, image_layers: Vec<ImageLayer>) {
+        let mut updates = self.layer_map.batch_update();
+        for layer in image_layers {
+            Self::insert_historic_layer(Arc::new(layer), &mut updates, &mut self.layer_fmgr);
+        }
+        updates.flush();
+    }
+
+    /// Insert into the layer map when a new delta layer is created, called from `create_delta_layer`.
+    pub fn track_new_l0_delta_layer(&mut self, delta_layer: Arc<DeltaLayer>) {
+        let mut updates = self.layer_map.batch_update();
+        Self::insert_historic_layer(delta_layer, &mut updates, &mut self.layer_fmgr);
+        updates.flush();
+    }
+
+    /// Called when compaction is completed.
+    pub fn finish_compact_l0(
+        &mut self,
+        layer_removal_cs: Arc<tokio::sync::OwnedMutexGuard<()>>,
+        compact_from: Vec<Arc<dyn PersistentLayer>>,
+        compact_to: Vec<Arc<dyn PersistentLayer>>,
+        metrics: &TimelineMetrics,
+    ) -> Result<()> {
+        let mut updates = self.layer_map.batch_update();
+        for l in compact_to {
+            Self::insert_historic_layer(l, &mut updates, &mut self.layer_fmgr);
+        }
+        for l in compact_from {
+            // NB: the layer file identified by descriptor `l` is guaranteed to be present
+            // in the LayerFileManager because compaction kept holding `layer_removal_cs` the entire
+            // time, even though we dropped `Timeline::layers` inbetween.
+            Self::delete_historic_layer(
+                layer_removal_cs.clone(),
+                l,
+                &mut updates,
+                metrics,
+                &mut self.layer_fmgr,
+            )?;
+        }
+        updates.flush();
+        Ok(())
+    }
+
+    /// Called when garbage collect the timeline. Returns a guard that will apply the updates to the layer map.
+    pub fn finish_gc_timeline(
+        &mut self,
+        layer_removal_cs: Arc<tokio::sync::OwnedMutexGuard<()>>,
+        gc_layers: Vec<Arc<dyn PersistentLayer>>,
+        metrics: &TimelineMetrics,
+    ) -> Result<ApplyGcResultGuard> {
+        let mut updates = self.layer_map.batch_update();
+        for doomed_layer in gc_layers {
+            Self::delete_historic_layer(
+                layer_removal_cs.clone(),
+                doomed_layer,
+                &mut updates,
+                metrics,
+                &mut self.layer_fmgr,
+            )?; // FIXME: schedule succeeded deletions in timeline.rs `gc_timeline` instead of in batch?
+        }
+        Ok(ApplyGcResultGuard(updates))
+    }
+
+    /// Helper function to insert a layer into the layer map and file manager.
+    fn insert_historic_layer(
+        layer: Arc<dyn PersistentLayer>,
+        updates: &mut BatchedUpdates<'_>,
+        mapping: &mut LayerFileManager,
+    ) {
+        updates.insert_historic(layer.layer_desc().clone());
+        mapping.insert(layer);
+    }
+
+    /// Helper function to remove a layer into the layer map and file manager
+    fn remove_historic_layer(
+        layer: Arc<dyn PersistentLayer>,
+        updates: &mut BatchedUpdates<'_>,
+        mapping: &mut LayerFileManager,
+    ) {
+        updates.remove_historic(layer.layer_desc().clone());
+        mapping.remove(layer);
+    }
+
+    /// Removes the layer from local FS (if present) and from memory.
+    /// Remote storage is not affected by this operation.
+    fn delete_historic_layer(
+        // we cannot remove layers otherwise, since gc and compaction will race
+        _layer_removal_cs: Arc<tokio::sync::OwnedMutexGuard<()>>,
+        layer: Arc<dyn PersistentLayer>,
+        updates: &mut BatchedUpdates<'_>,
+        metrics: &TimelineMetrics,
+        mapping: &mut LayerFileManager,
+    ) -> anyhow::Result<()> {
+        if !layer.is_remote_layer() {
+            layer.delete_resident_layer_file()?;
+            let layer_file_size = layer.file_size();
+            metrics.resident_physical_size_gauge.sub(layer_file_size);
+        }
+
+        // TODO Removing from the bottom of the layer map is expensive.
+        //      Maybe instead discard all layer map historic versions that
+        //      won't be needed for page reconstruction for this timeline,
+        //      and mark what we can't delete yet as deleted from the layer
+        //      map index without actually rebuilding the index.
+        updates.remove_historic(layer.layer_desc().clone());
+        mapping.remove(layer);
+
+        Ok(())
+    }
+}
+
+pub struct LayerFileManager<T: AsLayerDesc + ?Sized = dyn PersistentLayer>(
+    HashMap<PersistentLayerKey, Arc<T>>,
+);
+
+impl<T: AsLayerDesc + ?Sized> LayerFileManager<T> {
+    fn get_from_desc(&self, desc: &PersistentLayerDesc) -> Arc<T> {
+        // The assumption for the `expect()` is that all code maintains the following invariant:
+        // A layer's descriptor is present in the LayerMap => the LayerFileManager contains a layer for the descriptor.
+        self.0
+            .get(&desc.key())
+            .with_context(|| format!("get layer from desc: {}", desc.filename()))
+            .expect("not found")
+            .clone()
+    }
+
+    pub(crate) fn insert(&mut self, layer: Arc<T>) {
+        let present = self.0.insert(layer.layer_desc().key(), layer.clone());
+        if present.is_some() && cfg!(debug_assertions) {
+            panic!("overwriting a layer: {:?}", layer.layer_desc())
+        }
+    }
+
+    pub(crate) fn new() -> Self {
+        Self(HashMap::new())
+    }
+
+    pub(crate) fn remove(&mut self, layer: Arc<T>) {
+        let present = self.0.remove(&layer.layer_desc().key());
+        if present.is_none() && cfg!(debug_assertions) {
+            panic!(
+                "removing layer that is not present in layer mapping: {:?}",
+                layer.layer_desc()
+            )
+        }
+    }
+
+    pub(crate) fn replace_and_verify(&mut self, expected: Arc<T>, new: Arc<T>) -> Result<()> {
+        let key = expected.layer_desc().key();
+        let other = new.layer_desc().key();
+
+        let expected_l0 = LayerMap::is_l0(expected.layer_desc());
+        let new_l0 = LayerMap::is_l0(new.layer_desc());
+
+        fail::fail_point!("layermap-replace-notfound", |_| anyhow::bail!(
+            "layermap-replace-notfound"
+        ));
+
+        anyhow::ensure!(
+            key == other,
+            "expected and new layer have different keys: {key:?} != {other:?}"
+        );
+
+        anyhow::ensure!(
+            expected_l0 == new_l0,
+            "one layer is l0 while the other is not: {expected_l0} != {new_l0}"
+        );
+
+        if let Some(layer) = self.0.get_mut(&key) {
+            anyhow::ensure!(
+                compare_arced_layers(&expected, layer),
+                "another layer was found instead of expected, expected={expected:?}, new={new:?}",
+                expected = Arc::as_ptr(&expected),
+                new = Arc::as_ptr(layer),
+            );
+            *layer = new;
+            Ok(())
+        } else {
+            anyhow::bail!("layer was not found");
+        }
+    }
+}

pageserver/src/tenant/timeline/logical_size.rs

@@ -0,0 +1,128 @@
+use anyhow::Context;
+use once_cell::sync::OnceCell;
+
+use tokio::sync::Semaphore;
+use utils::lsn::Lsn;
+
+use std::sync::atomic::{AtomicI64, Ordering as AtomicOrdering};
+use std::sync::Arc;
+
+/// Internal structure to hold all data needed for logical size calculation.
+///
+/// Calculation consists of two stages:
+///
+/// 1. Initial size calculation. That might take a long time, because it requires
+/// reading all layers containing relation sizes at `initial_part_end`.
+///
+/// 2. Collecting an incremental part and adding that to the initial size.
+/// Increments are appended on walreceiver writing new timeline data,
+/// which result in increase or decrease of the logical size.
+pub(super) struct LogicalSize {
+    /// Size, potentially slow to compute. Calculating this might require reading multiple
+    /// layers, and even ancestor's layers.
+    ///
+    /// NOTE: size at a given LSN is constant, but after a restart we will calculate
+    /// the initial size at a different LSN.
+    pub initial_logical_size: OnceCell<u64>,
+
+    /// Semaphore to track ongoing calculation of `initial_logical_size`.
+    pub initial_size_computation: Arc<tokio::sync::Semaphore>,
+
+    /// Latest Lsn that has its size uncalculated, could be absent for freshly created timelines.
+    pub initial_part_end: Option<Lsn>,
+
+    /// All other size changes after startup, combined together.
+    ///
+    /// Size shouldn't ever be negative, but this is signed for two reasons:
+    ///
+    /// 1. If we initialized the "baseline" size lazily, while we already
+    /// process incoming WAL, the incoming WAL records could decrement the
+    /// variable and temporarily make it negative. (This is just future-proofing;
+    /// the initialization is currently not done lazily.)
+    ///
+    /// 2. If there is a bug and we e.g. forget to increment it in some cases
+    /// when size grows, but remember to decrement it when it shrinks again, the
+    /// variable could go negative. In that case, it seems better to at least
+    /// try to keep tracking it, rather than clamp or overflow it. Note that
+    /// get_current_logical_size() will clamp the returned value to zero if it's
+    /// negative, and log an error. Could set it permanently to zero or some
+    /// special value to indicate "broken" instead, but this will do for now.
+    ///
+    /// Note that we also expose a copy of this value as a prometheus metric,
+    /// see `current_logical_size_gauge`. Use the `update_current_logical_size`
+    /// to modify this, it will also keep the prometheus metric in sync.
+    pub size_added_after_initial: AtomicI64,
+}
+
+/// Normalized current size, that the data in pageserver occupies.
+#[derive(Debug, Clone, Copy)]
+pub(super) enum CurrentLogicalSize {
+    /// The size is not yet calculated to the end, this is an intermediate result,
+    /// constructed from walreceiver increments and normalized: logical data could delete some objects, hence be negative,
+    /// yet total logical size cannot be below 0.
+    Approximate(u64),
+    // Fully calculated logical size, only other future walreceiver increments are changing it, and those changes are
+    // available for observation without any calculations.
+    Exact(u64),
+}
+
+impl CurrentLogicalSize {
+    pub(super) fn size(&self) -> u64 {
+        *match self {
+            Self::Approximate(size) => size,
+            Self::Exact(size) => size,
+        }
+    }
+}
+
+impl LogicalSize {
+    pub(super) fn empty_initial() -> Self {
+        Self {
+            initial_logical_size: OnceCell::with_value(0),
+            //  initial_logical_size already computed, so, don't admit any calculations
+            initial_size_computation: Arc::new(Semaphore::new(0)),
+            initial_part_end: None,
+            size_added_after_initial: AtomicI64::new(0),
+        }
+    }
+
+    pub(super) fn deferred_initial(compute_to: Lsn) -> Self {
+        Self {
+            initial_logical_size: OnceCell::new(),
+            initial_size_computation: Arc::new(Semaphore::new(1)),
+            initial_part_end: Some(compute_to),
+            size_added_after_initial: AtomicI64::new(0),
+        }
+    }
+
+    pub(super) fn current_size(&self) -> anyhow::Result<CurrentLogicalSize> {
+        let size_increment: i64 = self.size_added_after_initial.load(AtomicOrdering::Acquire);
+        //                  ^^^ keep this type explicit so that the casts in this function break if
+        //                  we change the type.
+        match self.initial_logical_size.get() {
+            Some(initial_size) => {
+                initial_size.checked_add_signed(size_increment)
+                    .with_context(|| format!("Overflow during logical size calculation, initial_size: {initial_size}, size_increment: {size_increment}"))
+                    .map(CurrentLogicalSize::Exact)
+            }
+            None => {
+                let non_negative_size_increment = u64::try_from(size_increment).unwrap_or(0);
+                Ok(CurrentLogicalSize::Approximate(non_negative_size_increment))
+            }
+        }
+    }
+
+    pub(super) fn increment_size(&self, delta: i64) {
+        self.size_added_after_initial
+            .fetch_add(delta, AtomicOrdering::SeqCst);
+    }
+
+    /// Make the value computed by initial logical size computation
+    /// available for re-use. This doesn't contain the incremental part.
+    pub(super) fn initialized_size(&self, lsn: Lsn) -> Option<u64> {
+        match self.initial_part_end {
+            Some(v) if v == lsn => self.initial_logical_size.get().copied(),
+            _ => None,
+        }
+    }
+}

pageserver/src/tenant/timeline/span.rs

@@ -0,0 +1,23 @@
+#[cfg(debug_assertions)]
+use utils::tracing_span_assert::{check_fields_present, Extractor, MultiNameExtractor};
+
+#[cfg(not(debug_assertions))]
+pub(crate) fn debug_assert_current_span_has_tenant_and_timeline_id() {}
+
+#[cfg(debug_assertions)]
+#[track_caller]
+pub(crate) fn debug_assert_current_span_has_tenant_and_timeline_id() {
+    static TIMELINE_ID_EXTRACTOR: once_cell::sync::Lazy<MultiNameExtractor<1>> =
+        once_cell::sync::Lazy::new(|| MultiNameExtractor::new("TimelineId", ["timeline_id"]));
+
+    let fields: [&dyn Extractor; 2] = [
+        &*crate::tenant::span::TENANT_ID_EXTRACTOR,
+        &*TIMELINE_ID_EXTRACTOR,
+    ];
+    if let Err(missing) = check_fields_present(fields) {
+        panic!(
+            "missing extractors: {:?}",
+            missing.into_iter().map(|e| e.name()).collect::<Vec<_>>()
+        )
+    }
+}

pageserver/src/tenant/timeline/uninit.rs

@@ -0,0 +1,219 @@
+use std::{collections::hash_map::Entry, fs, path::PathBuf, sync::Arc};
+
+use anyhow::Context;
+use tracing::{error, info, info_span, warn};
+use utils::{crashsafe, id::TimelineId, lsn::Lsn};
+
+use crate::{
+    context::RequestContext,
+    import_datadir,
+    tenant::{ignore_absent_files, Tenant},
+};
+
+use super::Timeline;
+
+/// A timeline with some of its files on disk, being initialized.
+/// This struct ensures the atomicity of the timeline init: it's either properly created and inserted into pageserver's memory, or
+/// its local files are removed. In the worst case of a crash, an uninit mark file is left behind, which causes the directory
+/// to be removed on next restart.
+///
+/// The caller is responsible for proper timeline data filling before the final init.
+#[must_use]
+pub struct UninitializedTimeline<'t> {
+    pub(crate) owning_tenant: &'t Tenant,
+    timeline_id: TimelineId,
+    raw_timeline: Option<(Arc<Timeline>, TimelineUninitMark)>,
+}
+
+impl<'t> UninitializedTimeline<'t> {
+    pub(crate) fn new(
+        owning_tenant: &'t Tenant,
+        timeline_id: TimelineId,
+        raw_timeline: Option<(Arc<Timeline>, TimelineUninitMark)>,
+    ) -> Self {
+        Self {
+            owning_tenant,
+            timeline_id,
+            raw_timeline,
+        }
+    }
+
+    /// Finish timeline creation: insert it into the Tenant's timelines map and remove the
+    /// uninit mark file.
+    ///
+    /// This function launches the flush loop if not already done.
+    ///
+    /// The caller is responsible for activating the timeline (function `.activate()`).
+    pub(crate) fn finish_creation(mut self) -> anyhow::Result<Arc<Timeline>> {
+        let timeline_id = self.timeline_id;
+        let tenant_id = self.owning_tenant.tenant_id;
+
+        let (new_timeline, uninit_mark) = self.raw_timeline.take().with_context(|| {
+            format!("No timeline for initalization found for {tenant_id}/{timeline_id}")
+        })?;
+
+        // Check that the caller initialized disk_consistent_lsn
+        let new_disk_consistent_lsn = new_timeline.get_disk_consistent_lsn();
+        anyhow::ensure!(
+            new_disk_consistent_lsn.is_valid(),
+            "new timeline {tenant_id}/{timeline_id} has invalid disk_consistent_lsn"
+        );
+
+        let mut timelines = self.owning_tenant.timelines.lock().unwrap();
+        match timelines.entry(timeline_id) {
+            Entry::Occupied(_) => anyhow::bail!(
+                "Found freshly initialized timeline {tenant_id}/{timeline_id} in the tenant map"
+            ),
+            Entry::Vacant(v) => {
+                uninit_mark.remove_uninit_mark().with_context(|| {
+                    format!(
+                        "Failed to remove uninit mark file for timeline {tenant_id}/{timeline_id}"
+                    )
+                })?;
+                v.insert(Arc::clone(&new_timeline));
+
+                new_timeline.maybe_spawn_flush_loop();
+            }
+        }
+
+        Ok(new_timeline)
+    }
+
+    /// Prepares timeline data by loading it from the basebackup archive.
+    pub(crate) async fn import_basebackup_from_tar(
+        self,
+        copyin_read: &mut (impl tokio::io::AsyncRead + Send + Sync + Unpin),
+        base_lsn: Lsn,
+        broker_client: storage_broker::BrokerClientChannel,
+        ctx: &RequestContext,
+    ) -> anyhow::Result<Arc<Timeline>> {
+        let raw_timeline = self.raw_timeline()?;
+
+        import_datadir::import_basebackup_from_tar(raw_timeline, copyin_read, base_lsn, ctx)
+            .await
+            .context("Failed to import basebackup")?;
+
+        // Flush the new layer files to disk, before we make the timeline as available to
+        // the outside world.
+        //
+        // Flush loop needs to be spawned in order to be able to flush.
+        raw_timeline.maybe_spawn_flush_loop();
+
+        fail::fail_point!("before-checkpoint-new-timeline", |_| {
+            anyhow::bail!("failpoint before-checkpoint-new-timeline");
+        });
+
+        raw_timeline
+            .freeze_and_flush()
+            .await
+            .context("Failed to flush after basebackup import")?;
+
+        // All the data has been imported. Insert the Timeline into the tenant's timelines
+        // map and remove the uninit mark file.
+        let tl = self.finish_creation()?;
+        tl.activate(broker_client, None, ctx);
+        Ok(tl)
+    }
+
+    pub(crate) fn raw_timeline(&self) -> anyhow::Result<&Arc<Timeline>> {
+        Ok(&self
+            .raw_timeline
+            .as_ref()
+            .with_context(|| {
+                format!(
+                    "No raw timeline {}/{} found",
+                    self.owning_tenant.tenant_id, self.timeline_id
+                )
+            })?
+            .0)
+    }
+}
+
+impl Drop for UninitializedTimeline<'_> {
+    fn drop(&mut self) {
+        if let Some((_, uninit_mark)) = self.raw_timeline.take() {
+            let _entered = info_span!("drop_uninitialized_timeline", tenant_id = %self.owning_tenant.tenant_id, timeline_id = %self.timeline_id).entered();
+            error!("Timeline got dropped without initializing, cleaning its files");
+            cleanup_timeline_directory(uninit_mark);
+        }
+    }
+}
+
+pub(crate) fn cleanup_timeline_directory(uninit_mark: TimelineUninitMark) {
+    let timeline_path = &uninit_mark.timeline_path;
+    match ignore_absent_files(|| fs::remove_dir_all(timeline_path)) {
+        Ok(()) => {
+            info!("Timeline dir {timeline_path:?} removed successfully, removing the uninit mark")
+        }
+        Err(e) => {
+            error!("Failed to clean up uninitialized timeline directory {timeline_path:?}: {e:?}")
+        }
+    }
+    drop(uninit_mark); // mark handles its deletion on drop, gets retained if timeline dir exists
+}
+
+/// An uninit mark file, created along the timeline dir to ensure the timeline either gets fully initialized and loaded into pageserver's memory,
+/// or gets removed eventually.
+///
+/// XXX: it's important to create it near the timeline dir, not inside it to ensure timeline dir gets removed first.
+#[must_use]
+pub(crate) struct TimelineUninitMark {
+    uninit_mark_deleted: bool,
+    uninit_mark_path: PathBuf,
+    pub(crate) timeline_path: PathBuf,
+}
+
+impl TimelineUninitMark {
+    pub(crate) fn new(uninit_mark_path: PathBuf, timeline_path: PathBuf) -> Self {
+        Self {
+            uninit_mark_deleted: false,
+            uninit_mark_path,
+            timeline_path,
+        }
+    }
+
+    fn remove_uninit_mark(mut self) -> anyhow::Result<()> {
+        if !self.uninit_mark_deleted {
+            self.delete_mark_file_if_present()?;
+        }
+
+        Ok(())
+    }
+
+    fn delete_mark_file_if_present(&mut self) -> anyhow::Result<()> {
+        let uninit_mark_file = &self.uninit_mark_path;
+        let uninit_mark_parent = uninit_mark_file
+            .parent()
+            .with_context(|| format!("Uninit mark file {uninit_mark_file:?} has no parent"))?;
+        ignore_absent_files(|| fs::remove_file(uninit_mark_file)).with_context(|| {
+            format!("Failed to remove uninit mark file at path {uninit_mark_file:?}")
+        })?;
+        crashsafe::fsync(uninit_mark_parent).context("Failed to fsync uninit mark parent")?;
+        self.uninit_mark_deleted = true;
+
+        Ok(())
+    }
+}
+
+impl Drop for TimelineUninitMark {
+    fn drop(&mut self) {
+        if !self.uninit_mark_deleted {
+            if self.timeline_path.exists() {
+                error!(
+                    "Uninit mark {} is not removed, timeline {} stays uninitialized",
+                    self.uninit_mark_path.display(),
+                    self.timeline_path.display()
+                )
+            } else {
+                // unblock later timeline creation attempts
+                warn!(
+                    "Removing intermediate uninit mark file {}",
+                    self.uninit_mark_path.display()
+                );
+                if let Err(e) = self.delete_mark_file_if_present() {
+                    error!("Failed to remove the uninit mark file: {e}")
+                }
+            }
+        }
+    }
+}

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

@@ -266,7 +266,7 @@ pub struct ConnectionManagerStatus {
 impl ConnectionManagerStatus {
     /// Generates a string, describing current connection status in a form, suitable for logging.
     pub fn to_human_readable_string(&self) -> String {
-        let mut resulting_string = "WalReceiver status".to_string();
+        let mut resulting_string = String::new();
         match &self.existing_connection {
             Some(connection) => {
                 if connection.has_processed_wal {
@@ -1321,7 +1321,7 @@ mod tests {
 
     const DUMMY_SAFEKEEPER_HOST: &str = "safekeeper_connstr";
 
-    async fn dummy_state(harness: &TenantHarness<'_>) -> ConnectionManagerState {
+    async fn dummy_state(harness: &TenantHarness) -> ConnectionManagerState {
         let (tenant, ctx) = harness.load().await;
         let timeline = tenant
             .create_test_timeline(TIMELINE_ID, Lsn(0x8), crate::DEFAULT_PG_VERSION, &ctx)

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

@@ -71,6 +71,8 @@ pub(super) async fn handle_walreceiver_connection(
     ctx: RequestContext,
     node: NodeId,
 ) -> anyhow::Result<()> {
+    debug_assert_current_span_has_tenant_and_timeline_id();
+
     WALRECEIVER_STARTED_CONNECTIONS.inc();
 
     // Connect to the database in replication mode.
@@ -140,6 +142,9 @@ pub(super) async fn handle_walreceiver_connection(
             }
             Ok(())
         }
+        // Enrich the log lines emitted by this closure with meaningful context.
+        // TODO: technically, this task outlives the surrounding function, so, the
+        // spans won't be properly nested.
         .instrument(tracing::info_span!("poller")),
     );
 

pageserver/src/virtual_file.rs

@@ -302,15 +302,6 @@ impl VirtualFile {
             .observe_closure_duration(|| self.open_options.open(&self.path))?;
 
         // Perform the requested operation on it
-        //
-        // TODO: We could downgrade the locks to read mode before calling
-        // 'func', to allow a little bit more concurrency, but the standard
-        // library RwLock doesn't allow downgrading without releasing the lock,
-        // and that doesn't seem worth the trouble.
-        //
-        // XXX: `parking_lot::RwLock` can enable such downgrades, yet its implementation is fair and
-        // may deadlock on subsequent read calls.
-        // Simply replacing all `RwLock` in project causes deadlocks, so use it sparingly.
         let result = STORAGE_IO_TIME
             .with_label_values(&[op, &self.tenant_id, &self.timeline_id])
             .observe_closure_duration(|| func(&file));

pageserver/src/walingest.rs

@@ -25,7 +25,7 @@ use postgres_ffi::v14::nonrelfile_utils::clogpage_precedes;
 use postgres_ffi::v14::nonrelfile_utils::slru_may_delete_clogsegment;
 use postgres_ffi::{fsm_logical_to_physical, page_is_new, page_set_lsn};
 
-use anyhow::Result;
+use anyhow::{Context, Result};
 use bytes::{Buf, Bytes, BytesMut};
 use tracing::*;
 
@@ -1082,7 +1082,10 @@ impl<'a> WalIngest<'a> {
             .await?
         {
             // create it with 0 size initially, the logic below will extend it
-            modification.put_rel_creation(rel, 0, ctx).await?;
+            modification
+                .put_rel_creation(rel, 0, ctx)
+                .await
+                .context("Relation Error")?;
             0
         } else {
             self.timeline.get_rel_size(rel, last_lsn, true, ctx).await?

pageserver/src/walredo.rs

@@ -175,8 +175,8 @@ impl WalRedoManager for PostgresRedoManager {
         let mut img = base_img.map(|p| p.1);
         let mut batch_neon = can_apply_in_neon(&records[0].1);
         let mut batch_start = 0;
-        for i in 1..records.len() {
-            let rec_neon = can_apply_in_neon(&records[i].1);
+        for (i, record) in records.iter().enumerate().skip(1) {
+            let rec_neon = can_apply_in_neon(&record.1);
 
             if rec_neon != batch_neon {
                 let result = if batch_neon {
@@ -685,7 +685,7 @@ impl PostgresRedoManager {
             // as close-on-exec by default, but that's not enough, since we use
             // libraries that directly call libc open without setting that flag.
             .close_fds()
-            .spawn_no_leak_child()
+            .spawn_no_leak_child(self.tenant_id)
             .map_err(|e| {
                 Error::new(
                     e.kind(),
@@ -989,6 +989,7 @@ impl PostgresRedoManager {
 /// Wrapper type around `std::process::Child` which guarantees that the child
 /// will be killed and waited-for by this process before being dropped.
 struct NoLeakChild {
+    tenant_id: TenantId,
     child: Option<Child>,
 }
 
@@ -1007,9 +1008,12 @@ impl DerefMut for NoLeakChild {
 }
 
 impl NoLeakChild {
-    fn spawn(command: &mut Command) -> io::Result<Self> {
+    fn spawn(tenant_id: TenantId, command: &mut Command) -> io::Result<Self> {
         let child = command.spawn()?;
-        Ok(NoLeakChild { child: Some(child) })
+        Ok(NoLeakChild {
+            tenant_id,
+            child: Some(child),
+        })
     }
 
     fn kill_and_wait(mut self) {
@@ -1056,11 +1060,16 @@ impl Drop for NoLeakChild {
             Some(child) => child,
             None => return,
         };
+        let tenant_id = self.tenant_id;
         // Offload the kill+wait of the child process into the background.
         // If someone stops the runtime, we'll leak the child process.
         // We can ignore that case because we only stop the runtime on pageserver exit.
         BACKGROUND_RUNTIME.spawn(async move {
             tokio::task::spawn_blocking(move || {
+                // Intentionally don't inherit the tracing context from whoever is dropping us.
+                // This thread here is going to outlive of our dropper.
+                let span = tracing::info_span!("walredo", %tenant_id);
+                let _entered = span.enter();
                 Self::kill_and_wait_impl(child);
             })
             .await
@@ -1069,12 +1078,12 @@ impl Drop for NoLeakChild {
 }
 
 trait NoLeakChildCommandExt {
-    fn spawn_no_leak_child(&mut self) -> io::Result<NoLeakChild>;
+    fn spawn_no_leak_child(&mut self, tenant_id: TenantId) -> io::Result<NoLeakChild>;
 }
 
 impl NoLeakChildCommandExt for Command {
-    fn spawn_no_leak_child(&mut self) -> io::Result<NoLeakChild> {
-        NoLeakChild::spawn(self)
+    fn spawn_no_leak_child(&mut self, tenant_id: TenantId) -> io::Result<NoLeakChild> {
+        NoLeakChild::spawn(tenant_id, self)
     }
 }
 

pgxn/hnsw/hnsw.c

@@ -122,6 +122,43 @@ hnsw_populate(HierarchicalNSW* hnsw, Relation indexRel, Relation heapRel)
 						   true, true, hnsw_build_callback, (void *) hnsw, NULL);
 }
 
+#ifdef __APPLE__
+
+#include <sys/types.h>
+#include <sys/sysctl.h>
+
+static void
+hnsw_check_available_memory(Size requested)
+{
+	size_t total;
+	if (sysctlbyname("hw.memsize", NULL, &total, NULL, 0) < 0)
+		elog(ERROR, "Failed to get amount of RAM: %m");
+
+	if ((Size)NBuffers*BLCKSZ + requested >= total)
+		elog(ERROR, "HNSW index requeries %ld bytes while only %ld are available",
+			requested, total - (Size)NBuffers*BLCKSZ);
+}
+
+#else
+
+#include <sys/sysinfo.h>
+
+static void
+hnsw_check_available_memory(Size requested)
+{
+	struct sysinfo si;
+	Size total;
+	if (sysinfo(&si) < 0)
+		elog(ERROR, "Failed to get amount of RAM: %n");
+
+	total = si.totalram*si.mem_unit;
+	if ((Size)NBuffers*BLCKSZ + requested >= total)
+		elog(ERROR, "HNSW index requeries %ld bytes while only %ld are available",
+			requested, total - (Size)NBuffers*BLCKSZ);
+}
+
+#endif
+
 static HierarchicalNSW*
 hnsw_get_index(Relation indexRel, Relation heapRel)
 {
@@ -156,6 +193,8 @@ hnsw_get_index(Relation indexRel, Relation heapRel)
 		size_data_per_element = size_links_level0 + data_size + sizeof(label_t);
 		shmem_size =  hnsw_sizeof() + maxelements * size_data_per_element;
 
+		hnsw_check_available_memory(shmem_size);
+
 		/* first try to attach to existed index */
 		if (!dsm_impl_op(DSM_OP_ATTACH, handle, 0, &impl_private,
 						 &mapped_address, &mapped_size, DEBUG1))

pgxn/hnsw/hnsw.control

@@ -1,4 +1,4 @@
-comment = 'hNsw index'
+comment = 'hnsw index'
 default_version = '0.1.0'
 module_pathname = '$libdir/hnsw'
 relocatable = true

pgxn/neon/Makefile

@@ -4,6 +4,7 @@
 MODULE_big = neon
 OBJS = \
 	$(WIN32RES) \
+	extension_server.o \
 	file_cache.o \
 	libpagestore.o \
 	libpqwalproposer.o \

pgxn/neon/control_plane_connector.c

@@ -32,6 +32,7 @@
 #include "port.h"
 #include <curl/curl.h>
 #include "utils/jsonb.h"
+#include "libpq/crypt.h"
 
 static ProcessUtility_hook_type PreviousProcessUtilityHook = NULL;
 
@@ -161,7 +162,22 @@ ConstructDeltaMessage()
 			PushKeyValue(&state, "name", entry->name);
 			if (entry->password)
 			{
+#if PG_MAJORVERSION_NUM == 14
+				char	   *logdetail;
+#else
+				const char *logdetail;
+#endif
 				PushKeyValue(&state, "password", (char *) entry->password);
+				char	   *encrypted_password = get_role_password(entry->name, &logdetail);
+
+				if (encrypted_password)
+				{
+					PushKeyValue(&state, "encrypted_password", encrypted_password);
+				}
+				else
+				{
+					elog(ERROR, "Failed to get encrypted password: %s", logdetail);
+				}
 			}
 			if (entry->old_name[0] != '\0')
 			{

pgxn/neon/extension_server.c

@@ -0,0 +1,104 @@
+
+/*-------------------------------------------------------------------------
+ *
+ * extension_server.c
+ *	  Request compute_ctl to download extension files.
+ *
+ * IDENTIFICATION
+ *	 contrib/neon/extension_server.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+#include "tcop/pquery.h"
+#include "tcop/utility.h"
+#include "access/xact.h"
+#include "utils/hsearch.h"
+#include "utils/memutils.h"
+#include "commands/defrem.h"
+#include "miscadmin.h"
+#include "utils/acl.h"
+#include "fmgr.h"
+#include "utils/guc.h"
+#include "port.h"
+#include "fmgr.h"
+
+#include <curl/curl.h>
+
+static int extension_server_port = 0;
+
+static download_extension_file_hook_type prev_download_extension_file_hook = NULL;
+
+// to download all SQL (and data) files for an extension:
+// curl -X POST http://localhost:8080/extension_server/postgis
+// it covers two possible extension files layouts:
+// 1. extension_name--version--platform.sql
+// 2. extension_name/extension_name--version.sql
+//    extension_name/extra_files.csv
+//
+// to download specific library file:
+// curl -X POST http://localhost:8080/extension_server/postgis-3.so?is_library=true
+static bool
+neon_download_extension_file_http(const char *filename, bool is_library)
+{
+    CURL *curl;
+    CURLcode res;
+    char *compute_ctl_url;
+    char *postdata;
+    bool ret = false;
+
+    if ((curl = curl_easy_init()) == NULL)
+    {
+        elog(ERROR, "Failed to initialize curl handle");
+    }
+
+    compute_ctl_url = psprintf("http://localhost:%d/extension_server/%s%s",
+                               extension_server_port, filename, is_library ? "?is_library=true" : "");
+
+    elog(LOG, "Sending request to compute_ctl: %s", compute_ctl_url);
+
+    curl_easy_setopt(curl, CURLOPT_CUSTOMREQUEST, "POST");
+    curl_easy_setopt(curl, CURLOPT_URL, compute_ctl_url);
+    // NOTE: 15L may be insufficient time for large extensions like postgis
+    curl_easy_setopt(curl, CURLOPT_TIMEOUT, 15L /* seconds */);
+
+    if (curl)
+    {
+        /* Perform the request, res will get the return code */
+        res = curl_easy_perform(curl);
+        /* Check for errors */
+        if (res == CURLE_OK)
+        {
+            ret = true;
+        }
+        else
+        {
+            // Don't error here because postgres will try to find the file
+            // and will fail with some proper error message if it's not found.
+            elog(WARNING, "neon_download_extension_file_http failed: %s\n", curl_easy_strerror(res));
+        }
+
+        /* always cleanup */
+        curl_easy_cleanup(curl);
+    }
+
+    return ret;
+}
+
+void pg_init_extension_server()
+{
+    // Port to connect to compute_ctl on localhost
+    // to request extension files.
+    DefineCustomIntVariable("neon.extension_server_port",
+                            "connection string to the compute_ctl",
+                            NULL,
+                            &extension_server_port,
+                            0, 0, INT_MAX,
+                            PGC_POSTMASTER,
+                            0, /* no flags required */
+                            NULL, NULL, NULL);
+
+    // set download_extension_file_hook
+    prev_download_extension_file_hook = download_extension_file_hook;
+    download_extension_file_hook = neon_download_extension_file_http;
+}

pgxn/neon/extension_server.h

@@ -0,0 +1 @@
+

pgxn/neon/file_cache.c

@@ -190,7 +190,7 @@ lfc_change_limit_hook(int newval, void *extra)
 		hash_search(lfc_hash, &victim->key, HASH_REMOVE, NULL);
 		lfc_ctl->used -= 1;
 	}
-	elog(LOG, "set local file cache limit to %d", new_size);
+	elog(DEBUG1, "set local file cache limit to %d", new_size);
 	LWLockRelease(lfc_lock);
 }
 

pgxn/neon/libpagestore.c

@@ -34,7 +34,6 @@
 
 #define PageStoreTrace DEBUG5
 
-#define MAX_RECONNECT_ATTEMPTS 5
 #define RECONNECT_INTERVAL_USEC 1000000
 
 bool		connected = false;
@@ -55,13 +54,15 @@ int32		max_cluster_size;
 char	   *page_server_connstring;
 char	   *neon_auth_token;
 
-int			n_unflushed_requests = 0;
-int			flush_every_n_requests = 8;
 int			readahead_buffer_size = 128;
+int			flush_every_n_requests = 8;
+
+int			n_reconnect_attempts = 0;
+int			max_reconnect_attempts = 60;
 
 bool	(*old_redo_read_buffer_filter) (XLogReaderState *record, uint8 block_id) = NULL;
 
-static void pageserver_flush(void);
+static bool pageserver_flush(void);
 
 static bool
 pageserver_connect(int elevel)
@@ -232,16 +233,17 @@ pageserver_disconnect(void)
 	}
 }
 
-static void
+static bool
 pageserver_send(NeonRequest * request)
 {
 	StringInfoData req_buff;
-	int n_reconnect_attempts = 0;
 
 	/* If the connection was lost for some reason, reconnect */
 	if (connected && PQstatus(pageserver_conn) == CONNECTION_BAD)
+	{
+		neon_log(LOG, "pageserver_send disconnect bad connection");
 		pageserver_disconnect();
-
+	}
 
 	req_buff = nm_pack_request(request);
 
@@ -252,53 +254,36 @@ pageserver_send(NeonRequest * request)
 	 * See https://github.com/neondatabase/neon/issues/1138
 	 * So try to reestablish connection in case of failure.
 	 */
-	while (true)
+	if (!connected)
 	{
-		if (!connected)
+		while (!pageserver_connect(n_reconnect_attempts < max_reconnect_attempts ? LOG : ERROR))
 		{
-			if (!pageserver_connect(n_reconnect_attempts < MAX_RECONNECT_ATTEMPTS ? LOG : ERROR))
-			{
-				n_reconnect_attempts += 1;
-				pg_usleep(RECONNECT_INTERVAL_USEC);
-				continue;
-			}
+			n_reconnect_attempts += 1;
+			pg_usleep(RECONNECT_INTERVAL_USEC);
 		}
+		n_reconnect_attempts = 0;
+	}
 
-		/*
-		 * Send request.
-		 *
-		 * In principle, this could block if the output buffer is full, and we
-		 * should use async mode and check for interrupts while waiting. In
-		 * practice, our requests are small enough to always fit in the output and
-		 * TCP buffer.
-		 */
-		if (PQputCopyData(pageserver_conn, req_buff.data, req_buff.len) <= 0)
-		{
-			char	   *msg = pchomp(PQerrorMessage(pageserver_conn));
-			if (n_reconnect_attempts < MAX_RECONNECT_ATTEMPTS)
-			{
-				neon_log(LOG, "failed to send page request (try to reconnect): %s", msg);
-				if (n_reconnect_attempts != 0) /* do not sleep before first reconnect attempt, assuming that pageserver is already restarted */
-					pg_usleep(RECONNECT_INTERVAL_USEC);
-				n_reconnect_attempts += 1;
-				continue;
-			}
-			else
-			{
-				pageserver_disconnect();
-				neon_log(ERROR, "failed to send page request: %s", msg);
-			}
-		}
-		break;
+	/*
+	 * Send request.
+	 *
+	 * In principle, this could block if the output buffer is full, and we
+	 * should use async mode and check for interrupts while waiting. In
+	 * practice, our requests are small enough to always fit in the output and
+	 * TCP buffer.
+	 */
+	if (PQputCopyData(pageserver_conn, req_buff.data, req_buff.len) <= 0)
+	{
+		char	   *msg = pchomp(PQerrorMessage(pageserver_conn));
+		pageserver_disconnect();
+		neon_log(LOG, "pageserver_send disconnect because failed to send page request (try to reconnect): %s", msg);
+		pfree(msg);
+		pfree(req_buff.data);
+		return false;
 	}
 
 	pfree(req_buff.data);
 
-	n_unflushed_requests++;
-
-	if (flush_every_n_requests > 0 && n_unflushed_requests >= flush_every_n_requests)
-		pageserver_flush();
-
 	if (message_level_is_interesting(PageStoreTrace))
 	{
 		char	   *msg = nm_to_string((NeonMessage *) request);
@@ -306,6 +291,7 @@ pageserver_send(NeonRequest * request)
 		neon_log(PageStoreTrace, "sent request: %s", msg);
 		pfree(msg);
 	}
+	return true;
 }
 
 static NeonResponse *
@@ -340,16 +326,25 @@ pageserver_receive(void)
 		}
 		else if (rc == -1)
 		{
+			neon_log(LOG, "pageserver_receive disconnect because call_PQgetCopyData returns -1: %s", pchomp(PQerrorMessage(pageserver_conn)));
 			pageserver_disconnect();
 			resp = NULL;
 		}
 		else if (rc == -2)
-			neon_log(ERROR, "could not read COPY data: %s", pchomp(PQerrorMessage(pageserver_conn)));
+		{
+			char* msg = pchomp(PQerrorMessage(pageserver_conn));
+			pageserver_disconnect();
+			neon_log(ERROR, "pageserver_receive disconnect because could not read COPY data: %s", msg);
+		}
 		else
-			neon_log(ERROR, "unexpected PQgetCopyData return value: %d", rc);
+		{
+			pageserver_disconnect();
+			neon_log(ERROR, "pageserver_receive disconnect because unexpected PQgetCopyData return value: %d", rc);
+		}
 	}
 	PG_CATCH();
 	{
+		neon_log(LOG, "pageserver_receive disconnect due to caught exception");
 		pageserver_disconnect();
 		PG_RE_THROW();
 	}
@@ -359,21 +354,25 @@ pageserver_receive(void)
 }
 
 
-static void
+static bool
 pageserver_flush(void)
 {
 	if (!connected)
 	{
 		neon_log(WARNING, "Tried to flush while disconnected");
 	}
-	else if (PQflush(pageserver_conn))
+	else
 	{
-		char	   *msg = pchomp(PQerrorMessage(pageserver_conn));
-
-		pageserver_disconnect();
-		neon_log(ERROR, "failed to flush page requests: %s", msg);
+		if (PQflush(pageserver_conn))
+		{
+			char	   *msg = pchomp(PQerrorMessage(pageserver_conn));
+			pageserver_disconnect();
+			neon_log(LOG, "pageserver_flush disconnect because failed to flush page requests: %s", msg);
+			pfree(msg);
+			return false;
+		}
 	}
-	n_unflushed_requests = 0;
+	return true;
 }
 
 page_server_api api = {
@@ -439,6 +438,14 @@ pg_init_libpagestore(void)
 							PGC_USERSET,
 							0,	/* no flags required */
 							NULL, NULL, NULL);
+	DefineCustomIntVariable("neon.max_reconnect_attempts",
+							"Maximal attempts to reconnect to pages server (with 1 second timeout)",
+							NULL,
+							&max_reconnect_attempts,
+							10, 0, INT_MAX,
+							PGC_USERSET,
+							0,
+							NULL, NULL, NULL);
 	DefineCustomIntVariable("neon.readahead_buffer_size",
 							"number of prefetches to buffer",
 							"This buffer is used to hold and manage prefetched "

pgxn/neon/neon.c

@@ -35,8 +35,11 @@ _PG_init(void)
 {
 	pg_init_libpagestore();
 	pg_init_walproposer();
+
 	InitControlPlaneConnector();
 
+	pg_init_extension_server();
+
         // Important: This must happen after other parts of the extension
         // are loaded, otherwise any settings to GUCs that were set before
         // the extension was loaded will be removed.