fixup: update external loop state before handling keyspaces

`ValuesReconstructState::consume_done_keys` may only be called once after a layer visit. The code in the previous commit called it for each keyspace, resulting in keys not being marked done in the split keyspace by floor LSN scenario.
pageserver: extend nested image test with will init keys
2026-05-17 13:10:38 +00:00 · 2024-09-19 10:44:20 +01:00 · 2024-09-19 10:43:55 +01:00 · 2024-09-18 13:35:19 +01:00 · 2024-09-18 13:33:51 +01:00 · 2024-09-18 13:58:51 +02:00
184 changed files with 5462 additions and 2394 deletions
--- a/.github/workflows/_build-and-test-locally.yml
+++ b/.github/workflows/_build-and-test-locally.yml
@@ -62,7 +62,7 @@ jobs:
          #
          git config --global --add safe.directory ${{ github.workspace }}
          git config --global --add safe.directory ${GITHUB_WORKSPACE}
-          for r in 14 15 16; do
+          for r in 14 15 16 17; do
            git config --global --add safe.directory "${{ github.workspace }}/vendor/postgres-v$r"
            git config --global --add safe.directory "${GITHUB_WORKSPACE}/vendor/postgres-v$r"
          done
@@ -83,6 +83,10 @@ jobs:
        id: pg_v16_rev
        run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-v16) >> $GITHUB_OUTPUT

+      - name: Set pg 17 revision for caching
+        id: pg_v17_rev
+        run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-v17) >> $GITHUB_OUTPUT
+
      # Set some environment variables used by all the steps.
      #
      # CARGO_FLAGS is extra options to pass to "cargo build", "cargo test" etc.
@@ -136,6 +140,13 @@ jobs:
          path: pg_install/v16
          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ inputs.build-type }}-pg-${{ steps.pg_v16_rev.outputs.pg_rev }}-${{ hashFiles('Makefile', 'Dockerfile.build-tools') }}

+      - name: Cache postgres v17 build
+        id: cache_pg_17
+        uses: actions/cache@v4
+        with:
+          path: pg_install/v17
+          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ inputs.build-type }}-pg-${{ steps.pg_v17_rev.outputs.pg_rev }}-${{ hashFiles('Makefile', 'Dockerfile.build-tools') }}
+
      - name: Build postgres v14
        if: steps.cache_pg_14.outputs.cache-hit != 'true'
        run: mold -run make postgres-v14 -j$(nproc)
@@ -148,6 +159,10 @@ jobs:
        if: steps.cache_pg_16.outputs.cache-hit != 'true'
        run: mold -run make postgres-v16 -j$(nproc)

+      - name: Build postgres v17
+        if: steps.cache_pg_17.outputs.cache-hit != 'true'
+        run: mold -run make postgres-v17 -j$(nproc)
+
      - name: Build neon extensions
        run: mold -run make neon-pg-ext -j$(nproc)

@@ -210,7 +225,7 @@ jobs:
        run: |
          PQ_LIB_DIR=$(pwd)/pg_install/v16/lib
          export PQ_LIB_DIR
-          LD_LIBRARY_PATH=$(pwd)/pg_install/v16/lib
+          LD_LIBRARY_PATH=$(pwd)/pg_install/v17/lib
          export LD_LIBRARY_PATH

          #nextest does not yet support running doctests
--- a/.github/workflows/_push-to-acr.yml
+++ b/.github/workflows/_push-to-acr.yml
@@ -26,9 +26,15 @@ on:
        description: Azure tenant ID
        required: true
        type: string
+      skip_if:
+        description: Skip the job if this expression is true
+        required: true
+        type: boolean

 jobs:
  push-to-acr:
+    if: ${{ !inputs.skip_if }}
+
    runs-on: ubuntu-22.04
    permissions:
      contents: read  # This is required for actions/checkout
@@ -52,5 +58,5 @@ jobs:
          for image in ${images}; do
            docker buildx imagetools create \
              -t ${{ inputs.registry_name }}.azurecr.io/neondatabase/${image}:${{ inputs.image_tag }} \
-                                        neondatabase/${image}:${{ inputs.image_tag }}
+                                                        neondatabase/${image}:${{ inputs.image_tag }}
          done
--- a/.github/workflows/build_and_test.yml
+++ b/.github/workflows/build_and_test.yml
@@ -54,8 +54,8 @@ jobs:
      build-tag: ${{steps.build-tag.outputs.tag}}

    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
+      # Need `fetch-depth: 0` to count the number of commits in the branch
+      - uses: actions/checkout@v4
        with:
          fetch-depth: 0

@@ -211,7 +211,7 @@ jobs:
      build-tag: ${{ needs.tag.outputs.build-tag }}
      build-type: ${{ matrix.build-type }}
      # Run tests on all Postgres versions in release builds and only on the latest version in debug builds
-      pg-versions: ${{ matrix.build-type == 'release' && '["v14", "v15", "v16"]' || '["v16"]' }}
+      pg-versions: ${{ matrix.build-type == 'release' && '["v14", "v15", "v16", "v17"]' || '["v17"]' }}
    secrets: inherit

  # Keep `benchmarks` job outside of `build-and-test-locally` workflow to make job failures non-blocking
@@ -357,6 +357,7 @@ jobs:
            })

  coverage-report:
+    if: ${{ !startsWith(github.ref_name, 'release') }}
    needs: [ check-permissions, build-build-tools-image, build-and-test-locally ]
    runs-on: [ self-hosted, small ]
    container:
@@ -373,8 +374,8 @@ jobs:
        coverage-html: ${{ steps.upload-coverage-report-new.outputs.report-url }}
        coverage-json: ${{ steps.upload-coverage-report-new.outputs.summary-json }}
    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
+      # Need `fetch-depth: 0` for differential coverage (to get diff between two commits)
+      - uses: actions/checkout@v4
        with:
          submodules: true
          fetch-depth: 0
@@ -475,11 +476,9 @@ jobs:
    runs-on: ${{ fromJson(format('["self-hosted", "{0}"]', matrix.arch == 'arm64' && 'large-arm64' || 'large')) }}

    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
+      - uses: actions/checkout@v4
        with:
          submodules: true
-          fetch-depth: 0

      - uses: ./.github/actions/set-docker-config-dir
      - uses: docker/setup-buildx-action@v3
@@ -548,17 +547,15 @@ jobs:
    strategy:
      fail-fast: false
      matrix:
-        version: [ v14, v15, v16 ]
+        version: [ v14, v15, v16, v17 ]
        arch: [ x64, arm64 ]

    runs-on: ${{ fromJson(format('["self-hosted", "{0}"]', matrix.arch == 'arm64' && 'large-arm64' || 'large')) }}

    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
+      - uses: actions/checkout@v4
        with:
          submodules: true
-          fetch-depth: 0

      - uses: ./.github/actions/set-docker-config-dir
      - uses: docker/setup-buildx-action@v3
@@ -627,7 +624,7 @@ jobs:

      - name: Build compute-tools image
        # compute-tools are Postgres independent, so build it only once
-        if: matrix.version == 'v16'
+        if: matrix.version == 'v17'
        uses: docker/build-push-action@v6
        with:
          target: compute-tools-image
@@ -649,7 +646,7 @@ jobs:

    strategy:
      matrix:
-        version: [ v14, v15, v16 ]
+        version: [ v14, v15, v16, v17 ]

    steps:
      - uses: docker/login-action@v3
@@ -671,7 +668,7 @@ jobs:
                                             neondatabase/neon-test-extensions-${{ matrix.version }}:${{ needs.tag.outputs.build-tag }}-arm64

      - name: Create multi-arch compute-tools image
-        if: matrix.version == 'v16'
+        if: matrix.version == 'v17'
        run: |
          docker buildx imagetools create -t neondatabase/compute-tools:${{ needs.tag.outputs.build-tag }} \
                                             neondatabase/compute-tools:${{ needs.tag.outputs.build-tag }}-x64 \
@@ -689,7 +686,7 @@ jobs:
                                                                                neondatabase/compute-node-${{ matrix.version }}:${{ needs.tag.outputs.build-tag }}

      - name: Push multi-arch compute-tools image to ECR
-        if: matrix.version == 'v16'
+        if: matrix.version == 'v17'
        run: |
          docker buildx imagetools create -t 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-tools:${{ needs.tag.outputs.build-tag }} \
                                                                                neondatabase/compute-tools:${{ needs.tag.outputs.build-tag }}
@@ -700,15 +697,12 @@ jobs:
    strategy:
      fail-fast: false
      matrix:
-        version: [ v14, v15, v16 ]
+        version: [ v14, v15, v16, v17 ]
    env:
      VM_BUILDER_VERSION: v0.29.3

    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
+      - uses: actions/checkout@v4

      - name: Downloading vm-builder
        run: |
@@ -748,10 +742,7 @@ jobs:
    runs-on: ${{ fromJson(format('["self-hosted", "{0}"]', matrix.arch == 'arm64' && 'small-arm64' || 'small')) }}

    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
+      - uses: actions/checkout@v4

      - uses: ./.github/actions/set-docker-config-dir
      - uses: docker/login-action@v3
@@ -798,7 +789,7 @@ jobs:
    runs-on: ubuntu-22.04

    env:
-      VERSIONS: v14 v15 v16
+      VERSIONS: v14 v15 v16 v17

    steps:
      - uses: docker/login-action@v3
@@ -839,7 +830,7 @@ jobs:
            done
          done
          docker buildx imagetools create -t neondatabase/neon-test-extensions-v16:latest \
-                                             neondatabase/neon-test-extensions-v16:${{ needs.tag.outputs.build-tag }}
+                                              neondatabase/neon-test-extensions-v16:${{ needs.tag.outputs.build-tag }}

      - name: Login to prod ECR
        uses: docker/login-action@v3
@@ -852,34 +843,34 @@ jobs:
      - name: Copy all images to prod ECR
        if: github.ref_name == 'release'|| github.ref_name == 'release-proxy'
        run: |
-          for image in neon compute-tools {vm-,}compute-node-{v14,v15,v16}; do
+          for image in neon compute-tools {vm-,}compute-node-{v14,v15,v16,v17}; do
            docker buildx imagetools create -t 093970136003.dkr.ecr.eu-central-1.amazonaws.com/${image}:${{ needs.tag.outputs.build-tag }} \
                                               369495373322.dkr.ecr.eu-central-1.amazonaws.com/${image}:${{ needs.tag.outputs.build-tag }}
          done

  push-to-acr-dev:
-    if: github.ref_name == 'main'
    needs: [ tag, promote-images ]
    uses: ./.github/workflows/_push-to-acr.yml
    with:
      client_id: ${{ vars.AZURE_DEV_CLIENT_ID }}
      image_tag: ${{ needs.tag.outputs.build-tag }}
-      images: neon compute-tools vm-compute-node-v14 vm-compute-node-v15 vm-compute-node-v16 compute-node-v14 compute-node-v15 compute-node-v16
+      images: neon compute-tools vm-compute-node-v14 vm-compute-node-v15 vm-compute-node-v16 vm-compute-node-v17 compute-node-v14 compute-node-v15 compute-node-v16 compute-node-v17
      registry_name: ${{ vars.AZURE_DEV_REGISTRY_NAME }}
      subscription_id: ${{ vars.AZURE_DEV_SUBSCRIPTION_ID }}
      tenant_id: ${{ vars.AZURE_TENANT_ID }}
+      skip_if: ${{ github.ref_name != 'main' }}

  push-to-acr-prod:
-    if: github.ref_name == 'release'|| github.ref_name == 'release-proxy'
    needs: [ tag, promote-images ]
    uses: ./.github/workflows/_push-to-acr.yml
    with:
      client_id: ${{ vars.AZURE_PROD_CLIENT_ID }}
      image_tag: ${{ needs.tag.outputs.build-tag }}
-      images: neon compute-tools vm-compute-node-v14 vm-compute-node-v15 vm-compute-node-v16 compute-node-v14 compute-node-v15 compute-node-v16
+      images: neon compute-tools vm-compute-node-v14 vm-compute-node-v15 vm-compute-node-v16 vm-compute-node-v17 compute-node-v14 compute-node-v15 compute-node-v16 compute-node-v17
      registry_name: ${{ vars.AZURE_PROD_REGISTRY_NAME }}
      subscription_id: ${{ vars.AZURE_PROD_SUBSCRIPTION_ID }}
      tenant_id: ${{ vars.AZURE_TENANT_ID }}
+      skip_if: ${{ !startsWith(github.ref_name, 'release') }}

  trigger-custom-extensions-build-and-wait:
    needs: [ check-permissions, tag ]
@@ -957,7 +948,7 @@ jobs:

  deploy:
    needs: [ check-permissions, promote-images, tag, build-and-test-locally, trigger-custom-extensions-build-and-wait, push-to-acr-dev, push-to-acr-prod ]
-    if: (github.ref_name == 'main' || github.ref_name == 'release' || github.ref_name == 'release-proxy') && !failure() && !cancelled()
+    if: github.ref_name == 'main' || github.ref_name == 'release' || github.ref_name == 'release-proxy'

    runs-on: [ self-hosted, small ]
    container: 369495373322.dkr.ecr.eu-central-1.amazonaws.com/ansible:latest
@@ -971,15 +962,12 @@ jobs:
          #
          git config --global --add safe.directory ${{ github.workspace }}
          git config --global --add safe.directory ${GITHUB_WORKSPACE}
-          for r in 14 15 16; do
+          for r in 14 15 16 17; do
            git config --global --add safe.directory "${{ github.workspace }}/vendor/postgres-v$r"
            git config --global --add safe.directory "${GITHUB_WORKSPACE}/vendor/postgres-v$r"
          done

-      - name: Checkout
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
+      - uses: actions/checkout@v4

      - name: Trigger deploy workflow
        env:
@@ -1117,6 +1105,7 @@ jobs:

              files_to_promote+=("s3://${BUCKET}/${s3_key}")

+              # TODO Add v17
              for pg_version in v14 v15 v16; do
                # We run less tests for debug builds, so we don't need to promote them
                if [ "${build_type}" == "debug" ] && { [ "${arch}" == "ARM64" ] || [ "${pg_version}" != "v16" ] ; }; then
--- a/.github/workflows/neon_extra_builds.yml
+++ b/.github/workflows/neon_extra_builds.yml
@@ -72,6 +72,10 @@ jobs:
        id: pg_v16_rev
        run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-v16) >> $GITHUB_OUTPUT

+      - name: Set pg 17 revision for caching
+        id: pg_v17_rev
+        run: echo pg_rev=$(git rev-parse HEAD:vendor/postgres-v17) >> $GITHUB_OUTPUT
+
      - name: Cache postgres v14 build
        id: cache_pg_14
        uses: actions/cache@v4
@@ -93,6 +97,13 @@ jobs:
          path: pg_install/v16
          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ env.BUILD_TYPE }}-pg-${{ steps.pg_v16_rev.outputs.pg_rev }}-${{ hashFiles('Makefile') }}

+      - name: Cache postgres v17 build
+        id: cache_pg_17
+        uses: actions/cache@v4
+        with:
+          path: pg_install/v17
+          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ env.BUILD_TYPE }}-pg-${{ steps.pg_v17_rev.outputs.pg_rev }}-${{ hashFiles('Makefile') }}
+
      - name: Set extra env for macOS
        run: |
          echo 'LDFLAGS=-L/usr/local/opt/openssl@3/lib' >> $GITHUB_ENV
@@ -120,6 +131,10 @@ jobs:
        if: steps.cache_pg_16.outputs.cache-hit != 'true'
        run: make postgres-v16 -j$(sysctl -n hw.ncpu)

+      - name: Build postgres v17
+        if: steps.cache_pg_17.outputs.cache-hit != 'true'
+        run: make postgres-v17 -j$(sysctl -n hw.ncpu)
+
      - name: Build neon extensions
        run: make neon-pg-ext -j$(sysctl -n hw.ncpu)

@@ -166,7 +181,7 @@ jobs:
        run: make walproposer-lib -j$(nproc)

      - name: Produce the build stats
-        run: PQ_LIB_DIR=$(pwd)/pg_install/v16/lib cargo build --all --release --timings -j$(nproc)
+        run: PQ_LIB_DIR=$(pwd)/pg_install/v17/lib cargo build --all --release --timings -j$(nproc)

      - name: Upload the build stats
        id: upload-stats
--- a/.github/workflows/trigger-e2e-tests.yml
+++ b/.github/workflows/trigger-e2e-tests.yml
@@ -34,8 +34,8 @@ jobs:
      build-tag: ${{ steps.build-tag.outputs.tag }}

    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
+      # Need `fetch-depth: 0` to count the number of commits in the branch
+      - uses: actions/checkout@v4
        with:
          fetch-depth: 0

--- a/.gitmodules
+++ b/.gitmodules
@@ -10,3 +10,7 @@
 	path = vendor/postgres-v16
 	url = https://github.com/neondatabase/postgres.git
 	branch = REL_16_STABLE_neon
+[submodule "vendor/postgres-v17"]
+	path = vendor/postgres-v17
+	url = https://github.com/neondatabase/postgres.git
+	branch = REL_17_STABLE_neon
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -1189,9 +1189,9 @@ dependencies = [

 [[package]]
 name = "comfy-table"
-version = "6.1.4"
+version = "7.1.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "6e7b787b0dc42e8111badfdbe4c3059158ccb2db8780352fa1b01e8ccf45cc4d"
+checksum = "b34115915337defe99b2aff5c2ce6771e5fbc4079f4b506301f5cf394c8452f7"
 dependencies = [
 "crossterm",
 "strum",
@@ -1209,7 +1209,6 @@ dependencies = [
 "remote_storage",
 "serde",
 "serde_json",
- "serde_with",
 "utils",
 ]

@@ -1218,7 +1217,6 @@ name = "compute_tools"
 version = "0.1.0"
 dependencies = [
 "anyhow",
- "async-compression",
 "bytes",
 "cfg-if",
 "chrono",
@@ -1237,7 +1235,6 @@ dependencies = [
 "reqwest 0.12.4",
 "rlimit",
 "rust-ini",
- "serde",
 "serde_json",
 "signal-hook",
 "tar",
@@ -1246,7 +1243,6 @@ dependencies = [
 "tokio-postgres",
 "tokio-stream",
 "tokio-util",
- "toml_edit 0.19.10",
 "tracing",
 "tracing-opentelemetry",
 "tracing-subscriber",
@@ -1317,12 +1313,9 @@ dependencies = [
 name = "consumption_metrics"
 version = "0.1.0"
 dependencies = [
- "anyhow",
 "chrono",
 "rand 0.8.5",
 "serde",
- "serde_with",
- "utils",
 ]

 [[package]]
@@ -1334,9 +1327,7 @@ dependencies = [
 "clap",
 "comfy-table",
 "compute_api",
- "futures",
 "git-version",
- "hex",
 "humantime",
 "humantime-serde",
 "hyper 0.14.26",
@@ -1344,7 +1335,6 @@ dependencies = [
 "once_cell",
 "pageserver_api",
 "pageserver_client",
- "postgres",
 "postgres_backend",
 "postgres_connection",
 "regex",
@@ -1353,15 +1343,13 @@ dependencies = [
 "scopeguard",
 "serde",
 "serde_json",
- "serde_with",
 "storage_broker",
- "tar",
 "thiserror",
 "tokio",
 "tokio-postgres",
 "tokio-util",
- "toml 0.7.4",
- "toml_edit 0.19.10",
+ "toml",
+ "toml_edit",
 "tracing",
 "url",
 "utils",
@@ -1485,25 +1473,22 @@ checksum = "248e3bacc7dc6baa3b21e405ee045c3047101a49145e7e9eca583ab4c2ca5345"

 [[package]]
 name = "crossterm"
-version = "0.25.0"
+version = "0.27.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e64e6c0fbe2c17357405f7c758c1ef960fce08bdfb2c03d88d2a18d7e09c4b67"
+checksum = "f476fe445d41c9e991fd07515a6f463074b782242ccf4a5b7b1d1012e70824df"
 dependencies = [
- "bitflags 1.3.2",
+ "bitflags 2.4.1",
 "crossterm_winapi",
 "libc",
- "mio",
 "parking_lot 0.12.1",
- "signal-hook",
- "signal-hook-mio",
 "winapi",
 ]

 [[package]]
 name = "crossterm_winapi"
-version = "0.9.0"
+version = "0.9.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "2ae1b35a484aa10e07fe0638d02301c5ad24de82d310ccbd2f3693da5f09bf1c"
+checksum = "acdd7c62a3665c7f6830a51635d9ac9b23ed385797f70a83bb8bafe9c572ab2b"
 dependencies = [
 "winapi",
 ]
@@ -1666,7 +1651,6 @@ dependencies = [
 "hex",
 "parking_lot 0.12.1",
 "rand 0.8.5",
- "scopeguard",
 "smallvec",
 "tracing",
 "utils",
@@ -2236,24 +2220,22 @@ checksum = "4271d37baee1b8c7e4b708028c57d816cf9d2434acb33a549475f78c181f6253"

 [[package]]
 name = "git-version"
-version = "0.3.5"
+version = "0.3.9"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "f6b0decc02f4636b9ccad390dcbe77b722a77efedfa393caf8379a51d5c61899"
+checksum = "1ad568aa3db0fcbc81f2f116137f263d7304f512a1209b35b85150d3ef88ad19"
 dependencies = [
 "git-version-macro",
- "proc-macro-hack",
 ]

 [[package]]
 name = "git-version-macro"
-version = "0.3.5"
+version = "0.3.9"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "fe69f1cbdb6e28af2bac214e943b99ce8a0a06b447d15d3e61161b0423139f3f"
+checksum = "53010ccb100b96a67bc32c0175f0ed1426b31b655d562898e57325f81c023ac0"
 dependencies = [
- "proc-macro-hack",
 "proc-macro2",
 "quote",
- "syn 1.0.109",
+ "syn 2.0.52",
 ]

 [[package]]
@@ -2747,19 +2729,6 @@ dependencies = [
 "libc",
 ]

-[[package]]
-name = "inotify"
-version = "0.10.2"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "fdd168d97690d0b8c412d6b6c10360277f4d7ee495c5d0d5d5fe0854923255cc"
-dependencies = [
- "bitflags 1.3.2",
- "futures-core",
- "inotify-sys",
- "libc",
- "tokio",
-]
-
 [[package]]
 name = "inotify-sys"
 version = "0.1.5"
@@ -3144,7 +3113,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "fd01039851e82f8799046eabbb354056283fb265c8ec0996af940f4e85a380ff"
 dependencies = [
 "serde",
- "toml 0.8.14",
+ "toml",
 ]

 [[package]]
@@ -3254,7 +3223,7 @@ dependencies = [
 "crossbeam-channel",
 "filetime",
 "fsevent-sys",
- "inotify 0.9.6",
+ "inotify",
 "kqueue",
 "libc",
 "log",
@@ -3645,7 +3614,6 @@ name = "pagectl"
 version = "0.1.0"
 dependencies = [
 "anyhow",
- "bytes",
 "camino",
 "clap",
 "git-version",
@@ -3654,13 +3622,12 @@ dependencies = [
 "pageserver_api",
 "postgres_ffi",
 "remote_storage",
- "serde",
 "serde_json",
 "svg_fmt",
 "thiserror",
 "tokio",
 "tokio-util",
- "toml_edit 0.19.10",
+ "toml_edit",
 "utils",
 "workspace_hack",
 ]
@@ -3673,7 +3640,6 @@ dependencies = [
 "arc-swap",
 "async-compression",
 "async-stream",
- "async-trait",
 "bit_field",
 "byteorder",
 "bytes",
@@ -3681,16 +3647,13 @@ dependencies = [
 "camino-tempfile",
 "chrono",
 "clap",
- "const_format",
 "consumption_metrics",
 "crc32c",
 "criterion",
- "crossbeam-utils",
 "either",
 "enum-map",
 "enumset",
 "fail",
- "flate2",
 "futures",
 "git-version",
 "hex",
@@ -3729,13 +3692,9 @@ dependencies = [
 "serde_json",
 "serde_path_to_error",
 "serde_with",
- "signal-hook",
- "smallvec",
 "storage_broker",
 "strum",
 "strum_macros",
- "svg_fmt",
- "sync_wrapper",
 "sysinfo",
 "tenant_size_model",
 "thiserror",
@@ -3747,9 +3706,8 @@ dependencies = [
 "tokio-stream",
 "tokio-tar",
 "tokio-util",
- "toml_edit 0.19.10",
+ "toml_edit",
 "tracing",
- "twox-hash",
 "url",
 "utils",
 "walkdir",
@@ -3813,44 +3771,22 @@ name = "pageserver_compaction"
 version = "0.1.0"
 dependencies = [
 "anyhow",
- "async-compression",
 "async-stream",
- "byteorder",
- "bytes",
- "chrono",
 "clap",
- "const_format",
- "consumption_metrics",
 "criterion",
- "crossbeam-utils",
- "either",
- "fail",
- "flate2",
 "futures",
 "git-version",
- "hex",
 "hex-literal",
- "humantime",
- "humantime-serde",
 "itertools 0.10.5",
- "metrics",
 "once_cell",
 "pageserver_api",
 "pin-project-lite",
 "rand 0.8.5",
- "smallvec",
 "svg_fmt",
- "sync_wrapper",
- "thiserror",
 "tokio",
- "tokio-io-timeout",
- "tokio-util",
 "tracing",
- "tracing-error",
 "tracing-subscriber",
- "url",
 "utils",
- "walkdir",
 "workspace_hack",
 ]

@@ -3910,8 +3846,9 @@ dependencies = [

 [[package]]
 name = "parquet"
-version = "51.0.0"
-source = "git+https://github.com/apache/arrow-rs?branch=master#2534976a564be3d2d56312dc88fb1b6ed4cef829"
+version = "53.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f0fbf928021131daaa57d334ca8e3904fe9ae22f73c56244fc7db9b04eedc3d8"
 dependencies = [
 "ahash",
 "bytes",
@@ -3930,8 +3867,9 @@ dependencies = [

 [[package]]
 name = "parquet_derive"
-version = "51.0.0"
-source = "git+https://github.com/apache/arrow-rs?branch=master#2534976a564be3d2d56312dc88fb1b6ed4cef829"
+version = "53.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "86e9fcfae007533a06b580429a3f7e07cb833ec8aa37c041c16563e7918f057e"
 dependencies = [
 "parquet",
 "proc-macro2",
@@ -4121,7 +4059,7 @@ dependencies = [
 [[package]]
 name = "postgres"
 version = "0.19.4"
-source = "git+https://github.com/neondatabase/rust-postgres.git?branch=neon#20031d7a9ee1addeae6e0968e3899ae6bf01cee2"
+source = "git+https://github.com/neondatabase/rust-postgres.git?rev=20031d7a9ee1addeae6e0968e3899ae6bf01cee2#20031d7a9ee1addeae6e0968e3899ae6bf01cee2"
 dependencies = [
 "bytes",
 "fallible-iterator",
@@ -4134,7 +4072,7 @@ dependencies = [
 [[package]]
 name = "postgres-protocol"
 version = "0.6.4"
-source = "git+https://github.com/neondatabase/rust-postgres.git?branch=neon#20031d7a9ee1addeae6e0968e3899ae6bf01cee2"
+source = "git+https://github.com/neondatabase/rust-postgres.git?rev=20031d7a9ee1addeae6e0968e3899ae6bf01cee2#20031d7a9ee1addeae6e0968e3899ae6bf01cee2"
 dependencies = [
 "base64 0.20.0",
 "byteorder",
@@ -4153,7 +4091,7 @@ dependencies = [
 [[package]]
 name = "postgres-types"
 version = "0.2.4"
-source = "git+https://github.com/neondatabase/rust-postgres.git?branch=neon#20031d7a9ee1addeae6e0968e3899ae6bf01cee2"
+source = "git+https://github.com/neondatabase/rust-postgres.git?rev=20031d7a9ee1addeae6e0968e3899ae6bf01cee2#20031d7a9ee1addeae6e0968e3899ae6bf01cee2"
 dependencies = [
 "bytes",
 "fallible-iterator",
@@ -4165,9 +4103,7 @@ name = "postgres_backend"
 version = "0.1.0"
 dependencies = [
 "anyhow",
- "async-trait",
 "bytes",
- "futures",
 "once_cell",
 "pq_proto",
 "rustls 0.22.4",
@@ -4200,16 +4136,13 @@ version = "0.1.0"
 dependencies = [
 "anyhow",
 "bindgen",
- "byteorder",
 "bytes",
 "crc32c",
 "env_logger",
- "hex",
 "log",
 "memoffset 0.8.0",
 "once_cell",
 "postgres",
- "rand 0.8.5",
 "regex",
 "serde",
 "thiserror",
@@ -4244,13 +4177,11 @@ dependencies = [
 "byteorder",
 "bytes",
 "itertools 0.10.5",
- "pin-project-lite",
 "postgres-protocol",
 "rand 0.8.5",
 "serde",
 "thiserror",
 "tokio",
- "tracing",
 ]

 [[package]]
@@ -4282,12 +4213,6 @@ dependencies = [
 "elliptic-curve 0.13.8",
 ]

-[[package]]
-name = "proc-macro-hack"
-version = "0.5.20+deprecated"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "dc375e1527247fe1a97d8b7156678dfe7c1af2fc075c9a4db3690ecd2a148068"
-
 [[package]]
 name = "proc-macro2"
 version = "1.0.78"
@@ -4406,7 +4331,6 @@ dependencies = [
 "aws-config",
 "aws-sdk-iam",
 "aws-sigv4",
- "aws-types",
 "base64 0.13.1",
 "bstr",
 "bytes",
@@ -4415,7 +4339,6 @@ dependencies = [
 "chrono",
 "clap",
 "consumption_metrics",
- "crossbeam-deque",
 "dashmap",
 "ecdsa 0.16.9",
 "env_logger",
@@ -4441,11 +4364,9 @@ dependencies = [
 "jose-jwa",
 "jose-jwk",
 "lasso",
- "md5",
 "measured",
 "metrics",
 "once_cell",
- "opentelemetry",
 "p256 0.13.2",
 "parking_lot 0.12.1",
 "parquet",
@@ -4466,7 +4387,6 @@ dependencies = [
 "reqwest-middleware",
 "reqwest-retry",
 "reqwest-tracing",
- "routerify",
 "rsa",
 "rstest",
 "rustc-hash",
@@ -4482,7 +4402,6 @@ dependencies = [
 "smol_str",
 "socket2 0.5.5",
 "subtle",
- "task-local-extensions",
 "thiserror",
 "tikv-jemalloc-ctl",
 "tikv-jemallocator",
@@ -4492,7 +4411,6 @@ dependencies = [
 "tokio-rustls 0.25.0",
 "tokio-tungstenite",
 "tokio-util",
- "tower-service",
 "tracing",
 "tracing-opentelemetry",
 "tracing-subscriber",
@@ -4782,7 +4700,6 @@ dependencies = [
 "async-stream",
 "async-trait",
 "aws-config",
- "aws-credential-types",
 "aws-sdk-s3",
 "aws-smithy-async",
 "aws-smithy-types",
@@ -4796,7 +4713,6 @@ dependencies = [
 "futures",
 "futures-util",
 "http-types",
- "humantime",
 "humantime-serde",
 "hyper 0.14.26",
 "itertools 0.10.5",
@@ -4812,7 +4728,7 @@ dependencies = [
 "tokio",
 "tokio-stream",
 "tokio-util",
- "toml_edit 0.19.10",
+ "toml_edit",
 "tracing",
 "utils",
 ]
@@ -5276,14 +5192,12 @@ version = "0.1.0"
 dependencies = [
 "anyhow",
 "async-stream",
- "async-trait",
 "byteorder",
 "bytes",
 "camino",
 "camino-tempfile",
 "chrono",
 "clap",
- "const_format",
 "crc32c",
 "desim",
 "fail",
@@ -5309,9 +5223,7 @@ dependencies = [
 "sd-notify",
 "serde",
 "serde_json",
- "serde_with",
 "sha2",
- "signal-hook",
 "storage_broker",
 "strum",
 "strum_macros",
@@ -5322,7 +5234,6 @@ dependencies = [
 "tokio-stream",
 "tokio-tar",
 "tokio-util",
- "toml_edit 0.19.10",
 "tracing",
 "tracing-subscriber",
 "url",
@@ -5337,7 +5248,6 @@ version = "0.1.0"
 dependencies = [
 "const_format",
 "serde",
- "serde_with",
 "utils",
 ]

@@ -5731,17 +5641,6 @@ dependencies = [
 "signal-hook-registry",
 ]

-[[package]]
-name = "signal-hook-mio"
-version = "0.2.3"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "29ad2e15f37ec9a6cc544097b78a1ec90001e9f71b81338ca39f430adaca99af"
-dependencies = [
- "libc",
- "mio",
- "signal-hook",
-]
-
 [[package]]
 name = "signal-hook-registry"
 version = "1.4.1"
@@ -5877,7 +5776,6 @@ version = "0.1.0"
 dependencies = [
 "anyhow",
 "async-stream",
- "bytes",
 "clap",
 "const_format",
 "futures",
@@ -5891,7 +5789,6 @@ dependencies = [
 "parking_lot 0.12.1",
 "prost",
 "tokio",
- "tokio-stream",
 "tonic",
 "tonic-build",
 "tracing",
@@ -5904,9 +5801,7 @@ name = "storage_controller"
 version = "0.1.0"
 dependencies = [
 "anyhow",
- "aws-config",
 "bytes",
- "camino",
 "chrono",
 "clap",
 "control_plane",
@@ -5947,20 +5842,9 @@ dependencies = [
 name = "storage_controller_client"
 version = "0.1.0"
 dependencies = [
- "anyhow",
- "bytes",
- "futures",
- "pageserver_api",
 "pageserver_client",
- "postgres",
 "reqwest 0.12.4",
 "serde",
- "thiserror",
- "tokio",
- "tokio-postgres",
- "tokio-stream",
- "tokio-util",
- "utils",
 "workspace_hack",
 ]

@@ -5972,13 +5856,9 @@ dependencies = [
 "async-stream",
 "aws-config",
 "aws-sdk-s3",
- "aws-smithy-async",
- "bincode",
- "bytes",
 "camino",
 "chrono",
 "clap",
- "crc32c",
 "either",
 "futures",
 "futures-util",
@@ -5990,20 +5870,16 @@ dependencies = [
 "pageserver",
 "pageserver_api",
 "postgres_ffi",
- "rand 0.8.5",
 "remote_storage",
 "reqwest 0.12.4",
 "rustls 0.22.4",
 "rustls-native-certs 0.7.0",
 "serde",
 "serde_json",
- "serde_with",
 "storage_controller_client",
- "thiserror",
 "tokio",
 "tokio-postgres",
 "tokio-postgres-rustls",
- "tokio-rustls 0.25.0",
 "tokio-stream",
 "tokio-util",
 "tracing",
@@ -6022,14 +5898,11 @@ dependencies = [
 "comfy-table",
 "futures",
 "humantime",
- "hyper 0.14.26",
 "pageserver_api",
 "pageserver_client",
 "reqwest 0.12.4",
- "serde",
 "serde_json",
 "storage_controller_client",
- "thiserror",
 "tokio",
 "tracing",
 "utils",
@@ -6054,21 +5927,21 @@ checksum = "73473c0e59e6d5812c5dfe2a064a6444949f089e20eec9a2e5506596494e4623"

 [[package]]
 name = "strum"
-version = "0.24.1"
+version = "0.26.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "063e6045c0e62079840579a7e47a355ae92f60eb74daaf156fb1e84ba164e63f"
+checksum = "8fec0f0aef304996cf250b31b5a10dee7980c85da9d759361292b8bca5a18f06"

 [[package]]
 name = "strum_macros"
-version = "0.24.3"
+version = "0.26.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "1e385be0d24f186b4ce2f9982191e7101bb737312ad61c1f2f984f34bcf85d59"
+checksum = "4c6bee85a5a24955dc440386795aa378cd9cf82acd5f764469152d2270e581be"
 dependencies = [
- "heck 0.4.1",
+ "heck 0.5.0",
 "proc-macro2",
 "quote",
 "rustversion",
- "syn 1.0.109",
+ "syn 2.0.52",
 ]

 [[package]]
@@ -6152,15 +6025,6 @@ dependencies = [
 "xattr",
 ]

-[[package]]
-name = "task-local-extensions"
-version = "0.1.4"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ba323866e5d033818e3240feeb9f7db2c4296674e4d9e16b97b7bf8f490434e8"
-dependencies = [
- "pin-utils",
-]
-
 [[package]]
 name = "tempfile"
 version = "3.9.0"
@@ -6409,7 +6273,7 @@ dependencies = [
 [[package]]
 name = "tokio-postgres"
 version = "0.7.7"
-source = "git+https://github.com/neondatabase/rust-postgres.git?branch=neon#20031d7a9ee1addeae6e0968e3899ae6bf01cee2"
+source = "git+https://github.com/neondatabase/rust-postgres.git?rev=20031d7a9ee1addeae6e0968e3899ae6bf01cee2#20031d7a9ee1addeae6e0968e3899ae6bf01cee2"
 dependencies = [
 "async-trait",
 "byteorder",
@@ -6520,18 +6384,6 @@ dependencies = [
 "tracing",
 ]

-[[package]]
-name = "toml"
-version = "0.7.4"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "d6135d499e69981f9ff0ef2167955a5333c35e36f6937d382974566b3d5b94ec"
-dependencies = [
- "serde",
- "serde_spanned",
- "toml_datetime",
- "toml_edit 0.19.10",
-]
-
 [[package]]
 name = "toml"
 version = "0.8.14"
@@ -6541,7 +6393,7 @@ dependencies = [
 "serde",
 "serde_spanned",
 "toml_datetime",
- "toml_edit 0.22.14",
+ "toml_edit",
 ]

 [[package]]
@@ -6553,19 +6405,6 @@ dependencies = [
 "serde",
 ]

-[[package]]
-name = "toml_edit"
-version = "0.19.10"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "2380d56e8670370eee6566b0bfd4265f65b3f432e8c6d85623f728d4fa31f739"
-dependencies = [
- "indexmap 1.9.3",
- "serde",
- "serde_spanned",
- "toml_datetime",
- "winnow 0.4.6",
-]
-
 [[package]]
 name = "toml_edit"
 version = "0.22.14"
@@ -6576,7 +6415,7 @@ dependencies = [
 "serde",
 "serde_spanned",
 "toml_datetime",
- "winnow 0.6.13",
+ "winnow",
 ]

 [[package]]
@@ -6776,7 +6615,6 @@ dependencies = [
 "opentelemetry",
 "opentelemetry-otlp",
 "opentelemetry-semantic-conventions",
- "reqwest 0.12.4",
 "tokio",
 "tracing",
 "tracing-opentelemetry",
@@ -6980,7 +6818,6 @@ dependencies = [
 "serde_assert",
 "serde_json",
 "serde_path_to_error",
- "serde_with",
 "signal-hook",
 "strum",
 "strum_macros",
@@ -6989,7 +6826,7 @@ dependencies = [
 "tokio-stream",
 "tokio-tar",
 "tokio-util",
- "toml_edit 0.19.10",
+ "toml_edit",
 "tracing",
 "tracing-error",
 "tracing-subscriber",
@@ -7036,13 +6873,11 @@ dependencies = [
 "cgroups-rs",
 "clap",
 "futures",
- "inotify 0.10.2",
 "serde",
 "serde_json",
 "sysinfo",
 "tokio",
 "tokio-postgres",
- "tokio-stream",
 "tokio-util",
 "tracing",
 "tracing-subscriber",
@@ -7069,7 +6904,6 @@ dependencies = [
 "clap",
 "env_logger",
 "log",
- "once_cell",
 "postgres",
 "postgres_ffi",
 "regex",
@@ -7535,15 +7369,6 @@ version = "0.52.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "32b752e52a2da0ddfbdbcc6fceadfeede4c939ed16d13e648833a61dfb611ed8"

-[[package]]
-name = "winnow"
-version = "0.4.6"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "61de7bac303dc551fe038e2b3cef0f571087a47571ea6e79a87692ac99b99699"
-dependencies = [
- "memchr",
-]
-
 [[package]]
 name = "winnow"
 version = "0.6.13"
@@ -7601,6 +7426,7 @@ dependencies = [
 "digest",
 "either",
 "fail",
+ "futures",
 "futures-channel",
 "futures-executor",
 "futures-io",
@@ -7651,10 +7477,13 @@ dependencies = [
 "tokio",
 "tokio-rustls 0.24.0",
 "tokio-util",
+ "toml_edit",
 "tonic",
 "tower",
 "tracing",
 "tracing-core",
+ "tracing-log",
+ "tracing-subscriber",
 "url",
 "uuid",
 "zeroize",
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -73,7 +73,7 @@ camino = "1.1.6"
 cfg-if = "1.0.0"
 chrono = { version = "0.4", default-features = false, features = ["clock"] }
 clap = { version = "4.0", features = ["derive"] }
-comfy-table = "6.1"
+comfy-table = "7.1"
 const_format = "0.2"
 crc32c = "0.6"
 crossbeam-deque = "0.8.5"
@@ -123,8 +123,8 @@ opentelemetry = "0.20.0"
 opentelemetry-otlp = { version = "0.13.0", default-features=false, features = ["http-proto", "trace", "http", "reqwest-client"] }
 opentelemetry-semantic-conventions = "0.12.0"
 parking_lot = "0.12"
-parquet = { version = "51.0.0", default-features = false, features = ["zstd"] }
-parquet_derive = "51.0.0"
+parquet = { version = "53", default-features = false, features = ["zstd"] }
+parquet_derive = "53"
 pbkdf2 = { version = "0.12.1", features = ["simple", "std"] }
 pin-project-lite = "0.2"
 procfs = "0.16"
@@ -158,8 +158,8 @@ signal-hook = "0.3"
 smallvec = "1.11"
 smol_str = { version = "0.2.0", features = ["serde"] }
 socket2 = "0.5"
-strum = "0.24"
-strum_macros = "0.24"
+strum = "0.26"
+strum_macros = "0.26"
 "subtle"  = "2.5.0"
 svg_fmt = "0.4.3"
 sync_wrapper = "0.1.2"
@@ -177,8 +177,8 @@ tokio-rustls = "0.25"
 tokio-stream = "0.1"
 tokio-tar = "0.3"
 tokio-util = { version = "0.7.10", features = ["io", "rt"] }
-toml = "0.7"
-toml_edit = "0.19"
+toml = "0.8"
+toml_edit = "0.22"
 tonic = {version = "0.9", features = ["tls", "tls-roots"]}
 tower-service = "0.3.2"
 tracing = "0.1"
@@ -201,10 +201,21 @@ env_logger = "0.10"
 log = "0.4"

 ## Libraries from neondatabase/ git forks, ideally with changes to be upstreamed
-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", branch="neon" }
-postgres-protocol = { git = "https://github.com/neondatabase/rust-postgres.git", branch="neon" }
-postgres-types = { git = "https://github.com/neondatabase/rust-postgres.git", branch="neon" }
-tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", branch="neon" }
+
+# We want to use the 'neon' branch for these, but there's currently one
+# incompatible change on the branch. See:
+#
+# - PR #8076 which contained changes that depended on the new changes in
+#   the rust-postgres crate, and
+# - PR #8654 which reverted those changes and made the code in proxy incompatible
+#   with the tip of the 'neon' branch again.
+#
+# When those proxy changes are re-applied (see PR #8747), we can switch using
+# the tip of the 'neon' branch again.
+postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev = "20031d7a9ee1addeae6e0968e3899ae6bf01cee2" }
+postgres-protocol = { git = "https://github.com/neondatabase/rust-postgres.git", rev = "20031d7a9ee1addeae6e0968e3899ae6bf01cee2" }
+postgres-types = { git = "https://github.com/neondatabase/rust-postgres.git", rev = "20031d7a9ee1addeae6e0968e3899ae6bf01cee2" }
+tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev = "20031d7a9ee1addeae6e0968e3899ae6bf01cee2" }

 ## Local libraries
 compute_api = { version = "0.1", path = "./libs/compute_api/" }
@@ -241,11 +252,7 @@ tonic-build = "0.9"
 [patch.crates-io]

 # Needed to get `tokio-postgres-rustls` to depend on our fork.
-tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", branch="neon" }
-
-# bug fixes for UUID
-parquet = { git = "https://github.com/apache/arrow-rs", branch = "master" }
-parquet_derive = { git = "https://github.com/apache/arrow-rs", branch = "master" }
+tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", rev = "20031d7a9ee1addeae6e0968e3899ae6bf01cee2" }

 ################# Binary contents sections

--- a/12
+++ b/12
@@ -5,6 +5,8 @@
 ARG REPOSITORY=neondatabase
 ARG IMAGE=build-tools
 ARG TAG=pinned
+ARG DEFAULT_PG_VERSION=17
+ARG STABLE_PG_VERSION=16

 # Build Postgres
 FROM $REPOSITORY/$IMAGE:$TAG AS pg-build
@@ -13,6 +15,7 @@ WORKDIR /home/nonroot
 COPY --chown=nonroot vendor/postgres-v14 vendor/postgres-v14
 COPY --chown=nonroot vendor/postgres-v15 vendor/postgres-v15
 COPY --chown=nonroot vendor/postgres-v16 vendor/postgres-v16
+COPY --chown=nonroot vendor/postgres-v17 vendor/postgres-v17
 COPY --chown=nonroot pgxn pgxn
 COPY --chown=nonroot Makefile Makefile
 COPY --chown=nonroot scripts/ninstall.sh scripts/ninstall.sh
@@ -28,16 +31,19 @@ FROM $REPOSITORY/$IMAGE:$TAG AS build
 WORKDIR /home/nonroot
 ARG GIT_VERSION=local
 ARG BUILD_TAG
+ARG STABLE_PG_VERSION

 COPY --from=pg-build /home/nonroot/pg_install/v14/include/postgresql/server pg_install/v14/include/postgresql/server
 COPY --from=pg-build /home/nonroot/pg_install/v15/include/postgresql/server pg_install/v15/include/postgresql/server
 COPY --from=pg-build /home/nonroot/pg_install/v16/include/postgresql/server pg_install/v16/include/postgresql/server
+COPY --from=pg-build /home/nonroot/pg_install/v17/include/postgresql/server pg_install/v17/include/postgresql/server
 COPY --from=pg-build /home/nonroot/pg_install/v16/lib                       pg_install/v16/lib
+COPY --from=pg-build /home/nonroot/pg_install/v17/lib                       pg_install/v17/lib
 COPY --chown=nonroot . .

 ARG ADDITIONAL_RUSTFLAGS
 RUN set -e \
-    && PQ_LIB_DIR=$(pwd)/pg_install/v16/lib RUSTFLAGS="-Clinker=clang -Clink-arg=-fuse-ld=mold -Clink-arg=-Wl,--no-rosegment ${ADDITIONAL_RUSTFLAGS}" cargo build \
+    && PQ_LIB_DIR=$(pwd)/pg_install/v${STABLE_PG_VERSION}/lib RUSTFLAGS="-Clinker=clang -Clink-arg=-fuse-ld=mold -Clink-arg=-Wl,--no-rosegment ${ADDITIONAL_RUSTFLAGS}" cargo build \
      --bin pg_sni_router  \
      --bin pageserver  \
      --bin pagectl  \
@@ -52,6 +58,7 @@ RUN set -e \
 # Build final image
 #
 FROM debian:bullseye-slim
+ARG DEFAULT_PG_VERSION
 WORKDIR /data

 RUN set -e \
@@ -77,6 +84,7 @@ COPY --from=build --chown=neon:neon /home/nonroot/target/release/storage_scrubbe
 COPY --from=pg-build /home/nonroot/pg_install/v14 /usr/local/v14/
 COPY --from=pg-build /home/nonroot/pg_install/v15 /usr/local/v15/
 COPY --from=pg-build /home/nonroot/pg_install/v16 /usr/local/v16/
+COPY --from=pg-build /home/nonroot/pg_install/v17 /usr/local/v17/
 COPY --from=pg-build /home/nonroot/postgres_install.tar.gz /data/

 # By default, pageserver uses `.neon/` working directory in WORKDIR, so create one and fill it with the dummy config.
@@ -93,7 +101,7 @@ RUN mkdir -p /data/.neon/ && \

 # When running a binary that links with libpq, default to using our most recent postgres version.  Binaries
 # that want a particular postgres version will select it explicitly: this is just a default.
-ENV LD_LIBRARY_PATH=/usr/local/v16/lib
+ENV LD_LIBRARY_PATH=/usr/local/v${DEFAULT_PG_VERSION}/lib


 VOLUME ["/data"]
--- a/Dockerfile.compute-node
+++ b/Dockerfile.compute-node
@@ -55,22 +55,27 @@ RUN cd postgres && \
    # We could add the additional grant statements to the postgres repository but it would be hard to maintain,
    # whenever we need to pick up a new postgres version and we want to limit the changes in our postgres fork,
    # so we do it here.
-    old_list="pg_stat_statements--1.0--1.1.sql pg_stat_statements--1.1--1.2.sql pg_stat_statements--1.2--1.3.sql pg_stat_statements--1.3--1.4.sql pg_stat_statements--1.4--1.5.sql pg_stat_statements--1.4.sql pg_stat_statements--1.5--1.6.sql"; \
-    # the first loop is for pg_stat_statement extension version <= 1.6
    for file in /usr/local/pgsql/share/extension/pg_stat_statements--*.sql; do \
        filename=$(basename "$file"); \
-        if echo "$old_list" | grep -q -F "$filename"; then \
+        # Note that there are no downgrade scripts for pg_stat_statements, so we \
+        # don't have to modify any downgrade paths or (much) older versions: we only \
+        # have to make sure every creation of the pg_stat_statements_reset function \
+        # also adds execute permissions to the neon_superuser.
+        case $filename in \
+          pg_stat_statements--1.4.sql) \
+            # pg_stat_statements_reset is first created with 1.4
            echo 'GRANT EXECUTE ON FUNCTION pg_stat_statements_reset() TO neon_superuser;' >> $file; \
-        fi; \
-    done; \
-    # the second loop is for pg_stat_statement extension versions >= 1.7,
-    # where pg_stat_statement_reset() got 3 additional arguments
-    for file in /usr/local/pgsql/share/extension/pg_stat_statements--*.sql; do \
-        filename=$(basename "$file"); \
-        if ! echo "$old_list" | grep -q -F "$filename"; then \
+            ;; \
+          pg_stat_statements--1.6--1.7.sql) \
+            # Then with the 1.6-1.7 migration it is re-created with a new signature, thus add the permissions back
            echo 'GRANT EXECUTE ON FUNCTION pg_stat_statements_reset(Oid, Oid, bigint) TO neon_superuser;' >> $file; \
-        fi; \
-    done
+            ;; \
+          pg_stat_statements--1.10--1.11.sql) \
+            # Then with the 1.10-1.11 migration it is re-created with a new signature again, thus add the permissions back
+            echo 'GRANT EXECUTE ON FUNCTION pg_stat_statements_reset(Oid, Oid, bigint, boolean) TO neon_superuser;' >> $file; \
+            ;; \
+        esac; \
+    done;

 #########################################################################################
 #
@@ -79,6 +84,7 @@ RUN cd postgres && \
 #
 #########################################################################################
 FROM build-deps AS postgis-build
+ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 RUN apt update && \
    apt install -y cmake gdal-bin libboost-dev libboost-thread-dev libboost-filesystem-dev \
@@ -87,7 +93,11 @@ RUN apt update && \
    protobuf-c-compiler xsltproc

 # SFCGAL > 1.3 requires CGAL > 5.2, Bullseye's libcgal-dev is 5.2
-RUN wget https://gitlab.com/Oslandia/SFCGAL/-/archive/v1.3.10/SFCGAL-v1.3.10.tar.gz -O SFCGAL.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    mkdir -p /sfcgal && \
+    echo "Postgis doensn't yet support PG17 (needs 3.4.3, if not higher)" && exit 0;; \
+    esac && \
+    wget https://gitlab.com/Oslandia/SFCGAL/-/archive/v1.3.10/SFCGAL-v1.3.10.tar.gz -O SFCGAL.tar.gz && \
    echo "4e39b3b2adada6254a7bdba6d297bb28e1a9835a9f879b74f37e2dab70203232 SFCGAL.tar.gz" | sha256sum --check && \
    mkdir sfcgal-src && cd sfcgal-src && tar xzf ../SFCGAL.tar.gz --strip-components=1 -C . && \
    cmake -DCMAKE_BUILD_TYPE=Release . && make -j $(getconf _NPROCESSORS_ONLN) && \
@@ -96,7 +106,10 @@ RUN wget https://gitlab.com/Oslandia/SFCGAL/-/archive/v1.3.10/SFCGAL-v1.3.10.tar

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

-RUN wget https://download.osgeo.org/postgis/source/postgis-3.3.3.tar.gz -O postgis.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "Postgis doensn't yet support PG17 (needs 3.4.3, if not higher)" && exit 0;; \
+    esac && \
+    wget https://download.osgeo.org/postgis/source/postgis-3.3.3.tar.gz -O postgis.tar.gz && \
    echo "74eb356e3f85f14233791013360881b6748f78081cc688ff9d6f0f673a762d13 postgis.tar.gz" | sha256sum --check && \
    mkdir postgis-src && cd postgis-src && tar xzf ../postgis.tar.gz --strip-components=1 -C . && \
    find /usr/local/pgsql -type f | sed 's|^/usr/local/pgsql/||' > /before.txt &&\
@@ -122,7 +135,10 @@ RUN wget https://download.osgeo.org/postgis/source/postgis-3.3.3.tar.gz -O postg
    cp /usr/local/pgsql/share/extension/address_standardizer.control /extensions/postgis && \
    cp /usr/local/pgsql/share/extension/address_standardizer_data_us.control /extensions/postgis

-RUN wget https://github.com/pgRouting/pgrouting/archive/v3.4.2.tar.gz -O pgrouting.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    wget https://github.com/pgRouting/pgrouting/archive/v3.4.2.tar.gz -O pgrouting.tar.gz && \
    echo "cac297c07d34460887c4f3b522b35c470138760fe358e351ad1db4edb6ee306e pgrouting.tar.gz" | sha256sum --check && \
    mkdir pgrouting-src && cd pgrouting-src && tar xzf ../pgrouting.tar.gz --strip-components=1 -C . && \
    mkdir build && cd build && \
@@ -142,12 +158,19 @@ RUN wget https://github.com/pgRouting/pgrouting/archive/v3.4.2.tar.gz -O pgrouti
 #
 #########################################################################################
 FROM build-deps AS plv8-build
+ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

-RUN apt update && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    apt update && \
    apt install -y ninja-build python3-dev libncurses5 binutils clang

-RUN wget https://github.com/plv8/plv8/archive/refs/tags/v3.1.10.tar.gz -O plv8.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    wget https://github.com/plv8/plv8/archive/refs/tags/v3.1.10.tar.gz -O plv8.tar.gz && \
    echo "7096c3290928561f0d4901b7a52794295dc47f6303102fae3f8e42dd575ad97d plv8.tar.gz" | sha256sum --check && \
    mkdir plv8-src && cd plv8-src && tar xzf ../plv8.tar.gz --strip-components=1 -C . && \
    # generate and copy upgrade scripts
@@ -172,9 +195,13 @@ RUN wget https://github.com/plv8/plv8/archive/refs/tags/v3.1.10.tar.gz -O plv8.t
 #
 #########################################################################################
 FROM build-deps AS h3-pg-build
+ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

-RUN case "$(uname -m)" in \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    case "$(uname -m)" in \
      "x86_64") \
        export CMAKE_CHECKSUM=739d372726cb23129d57a539ce1432453448816e345e1545f6127296926b6754 \
        ;; \
@@ -192,7 +219,11 @@ RUN case "$(uname -m)" in \
      && /tmp/cmake-install.sh --skip-license --prefix=/usr/local/ \
      && rm /tmp/cmake-install.sh

-RUN wget https://github.com/uber/h3/archive/refs/tags/v4.1.0.tar.gz -O h3.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+        mkdir -p /h3/usr/ && \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    wget https://github.com/uber/h3/archive/refs/tags/v4.1.0.tar.gz -O h3.tar.gz && \
    echo "ec99f1f5974846bde64f4513cf8d2ea1b8d172d2218ab41803bf6a63532272bc h3.tar.gz" | sha256sum --check && \
    mkdir h3-src && cd h3-src && tar xzf ../h3.tar.gz --strip-components=1 -C . && \
    mkdir build && cd build && \
@@ -202,7 +233,10 @@ RUN wget https://github.com/uber/h3/archive/refs/tags/v4.1.0.tar.gz -O h3.tar.gz
    cp -R /h3/usr / && \
    rm -rf build

-RUN wget https://github.com/zachasme/h3-pg/archive/refs/tags/v4.1.3.tar.gz -O h3-pg.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    wget https://github.com/zachasme/h3-pg/archive/refs/tags/v4.1.3.tar.gz -O h3-pg.tar.gz && \
    echo "5c17f09a820859ffe949f847bebf1be98511fb8f1bd86f94932512c00479e324 h3-pg.tar.gz" | sha256sum --check && \
    mkdir h3-pg-src && cd h3-pg-src && tar xzf ../h3-pg.tar.gz --strip-components=1 -C . && \
    export PATH="/usr/local/pgsql/bin:$PATH" && \
@@ -218,9 +252,13 @@ RUN wget https://github.com/zachasme/h3-pg/archive/refs/tags/v4.1.3.tar.gz -O h3
 #
 #########################################################################################
 FROM build-deps AS unit-pg-build
+ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

-RUN wget https://github.com/df7cb/postgresql-unit/archive/refs/tags/7.7.tar.gz -O postgresql-unit.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    wget https://github.com/df7cb/postgresql-unit/archive/refs/tags/7.7.tar.gz -O postgresql-unit.tar.gz && \
    echo "411d05beeb97e5a4abf17572bfcfbb5a68d98d1018918feff995f6ee3bb03e79 postgresql-unit.tar.gz" | sha256sum --check && \
    mkdir postgresql-unit-src && cd postgresql-unit-src && tar xzf ../postgresql-unit.tar.gz --strip-components=1 -C . && \
    make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
@@ -239,6 +277,7 @@ RUN wget https://github.com/df7cb/postgresql-unit/archive/refs/tags/7.7.tar.gz -
 #
 #########################################################################################
 FROM build-deps AS vector-pg-build
+ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

 COPY patches/pgvector.patch /pgvector.patch
@@ -246,7 +285,10 @@ COPY patches/pgvector.patch /pgvector.patch
 # By default, pgvector Makefile uses `-march=native`. We don't want that,
 # because we build the images on different machines than where we run them.
 # Pass OPTFLAGS="" to remove it.
-RUN wget https://github.com/pgvector/pgvector/archive/refs/tags/v0.7.2.tar.gz -O pgvector.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    wget https://github.com/pgvector/pgvector/archive/refs/tags/v0.7.2.tar.gz -O pgvector.tar.gz && \
    echo "617fba855c9bcb41a2a9bc78a78567fd2e147c72afd5bf9d37b31b9591632b30 pgvector.tar.gz" | sha256sum --check && \
    mkdir pgvector-src && cd pgvector-src && tar xzf ../pgvector.tar.gz --strip-components=1 -C . && \
    patch -p1 < /pgvector.patch && \
@@ -261,10 +303,14 @@ RUN wget https://github.com/pgvector/pgvector/archive/refs/tags/v0.7.2.tar.gz -O
 #
 #########################################################################################
 FROM build-deps AS pgjwt-pg-build
+ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

 # 9742dab1b2f297ad3811120db7b21451bca2d3c9 made on 13/11/2021
-RUN wget https://github.com/michelp/pgjwt/archive/9742dab1b2f297ad3811120db7b21451bca2d3c9.tar.gz -O pgjwt.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    wget https://github.com/michelp/pgjwt/archive/9742dab1b2f297ad3811120db7b21451bca2d3c9.tar.gz -O pgjwt.tar.gz && \
    echo "cfdefb15007286f67d3d45510f04a6a7a495004be5b3aecb12cda667e774203f pgjwt.tar.gz" | sha256sum --check && \
    mkdir pgjwt-src && cd pgjwt-src && tar xzf ../pgjwt.tar.gz --strip-components=1 -C . && \
    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
@@ -277,9 +323,13 @@ RUN wget https://github.com/michelp/pgjwt/archive/9742dab1b2f297ad3811120db7b214
 #
 #########################################################################################
 FROM build-deps AS hypopg-pg-build
+ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

-RUN wget https://github.com/HypoPG/hypopg/archive/refs/tags/1.4.0.tar.gz -O hypopg.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    wget https://github.com/HypoPG/hypopg/archive/refs/tags/1.4.0.tar.gz -O hypopg.tar.gz && \
    echo "0821011743083226fc9b813c1f2ef5897a91901b57b6bea85a78e466187c6819 hypopg.tar.gz" | sha256sum --check && \
    mkdir hypopg-src && cd hypopg-src && tar xzf ../hypopg.tar.gz --strip-components=1 -C . && \
    make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
@@ -293,9 +343,13 @@ RUN wget https://github.com/HypoPG/hypopg/archive/refs/tags/1.4.0.tar.gz -O hypo
 #
 #########################################################################################
 FROM build-deps AS pg-hashids-pg-build
+ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

-RUN wget https://github.com/iCyberon/pg_hashids/archive/refs/tags/v1.2.1.tar.gz -O pg_hashids.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    wget https://github.com/iCyberon/pg_hashids/archive/refs/tags/v1.2.1.tar.gz -O pg_hashids.tar.gz && \
    echo "74576b992d9277c92196dd8d816baa2cc2d8046fe102f3dcd7f3c3febed6822a pg_hashids.tar.gz" | sha256sum --check && \
    mkdir pg_hashids-src && cd pg_hashids-src && tar xzf ../pg_hashids.tar.gz --strip-components=1 -C . && \
    make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config USE_PGXS=1 && \
@@ -309,11 +363,15 @@ RUN wget https://github.com/iCyberon/pg_hashids/archive/refs/tags/v1.2.1.tar.gz
 #
 #########################################################################################
 FROM build-deps AS rum-pg-build
+ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

 COPY patches/rum.patch /rum.patch

-RUN wget https://github.com/postgrespro/rum/archive/refs/tags/1.3.13.tar.gz -O rum.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    wget https://github.com/postgrespro/rum/archive/refs/tags/1.3.13.tar.gz -O rum.tar.gz && \
    echo "6ab370532c965568df6210bd844ac6ba649f53055e48243525b0b7e5c4d69a7d rum.tar.gz" | sha256sum --check && \
    mkdir rum-src && cd rum-src && tar xzf ../rum.tar.gz --strip-components=1 -C . && \
    patch -p1 < /rum.patch && \
@@ -328,9 +386,13 @@ RUN wget https://github.com/postgrespro/rum/archive/refs/tags/1.3.13.tar.gz -O r
 #
 #########################################################################################
 FROM build-deps AS pgtap-pg-build
+ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

-RUN wget https://github.com/theory/pgtap/archive/refs/tags/v1.2.0.tar.gz -O pgtap.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    wget https://github.com/theory/pgtap/archive/refs/tags/v1.2.0.tar.gz -O pgtap.tar.gz && \
    echo "9c7c3de67ea41638e14f06da5da57bac6f5bd03fea05c165a0ec862205a5c052 pgtap.tar.gz" | sha256sum --check && \
    mkdir pgtap-src && cd pgtap-src && tar xzf ../pgtap.tar.gz --strip-components=1 -C . && \
    make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
@@ -344,9 +406,13 @@ RUN wget https://github.com/theory/pgtap/archive/refs/tags/v1.2.0.tar.gz -O pgta
 #
 #########################################################################################
 FROM build-deps AS ip4r-pg-build
+ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

-RUN wget https://github.com/RhodiumToad/ip4r/archive/refs/tags/2.4.2.tar.gz -O ip4r.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    wget https://github.com/RhodiumToad/ip4r/archive/refs/tags/2.4.2.tar.gz -O ip4r.tar.gz && \
    echo "0f7b1f159974f49a47842a8ab6751aecca1ed1142b6d5e38d81b064b2ead1b4b ip4r.tar.gz" | sha256sum --check && \
    mkdir ip4r-src && cd ip4r-src && tar xzf ../ip4r.tar.gz --strip-components=1 -C . && \
    make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
@@ -360,9 +426,13 @@ RUN wget https://github.com/RhodiumToad/ip4r/archive/refs/tags/2.4.2.tar.gz -O i
 #
 #########################################################################################
 FROM build-deps AS prefix-pg-build
+ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

-RUN wget https://github.com/dimitri/prefix/archive/refs/tags/v1.2.10.tar.gz -O prefix.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    wget https://github.com/dimitri/prefix/archive/refs/tags/v1.2.10.tar.gz -O prefix.tar.gz && \
    echo "4342f251432a5f6fb05b8597139d3ccde8dcf87e8ca1498e7ee931ca057a8575 prefix.tar.gz" | sha256sum --check && \
    mkdir prefix-src && cd prefix-src && tar xzf ../prefix.tar.gz --strip-components=1 -C . && \
    make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
@@ -376,9 +446,13 @@ RUN wget https://github.com/dimitri/prefix/archive/refs/tags/v1.2.10.tar.gz -O p
 #
 #########################################################################################
 FROM build-deps AS hll-pg-build
+ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

-RUN wget https://github.com/citusdata/postgresql-hll/archive/refs/tags/v2.18.tar.gz -O hll.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    wget https://github.com/citusdata/postgresql-hll/archive/refs/tags/v2.18.tar.gz -O hll.tar.gz && \
    echo "e2f55a6f4c4ab95ee4f1b4a2b73280258c5136b161fe9d059559556079694f0e hll.tar.gz" | sha256sum --check && \
    mkdir hll-src && cd hll-src && tar xzf ../hll.tar.gz --strip-components=1 -C . && \
    make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
@@ -392,9 +466,13 @@ RUN wget https://github.com/citusdata/postgresql-hll/archive/refs/tags/v2.18.tar
 #
 #########################################################################################
 FROM build-deps AS plpgsql-check-pg-build
+ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

-RUN wget https://github.com/okbob/plpgsql_check/archive/refs/tags/v2.5.3.tar.gz -O plpgsql_check.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    wget https://github.com/okbob/plpgsql_check/archive/refs/tags/v2.5.3.tar.gz -O plpgsql_check.tar.gz && \
    echo "6631ec3e7fb3769eaaf56e3dfedb829aa761abf163d13dba354b4c218508e1c0 plpgsql_check.tar.gz" | sha256sum --check && \
    mkdir plpgsql_check-src && cd plpgsql_check-src && tar xzf ../plpgsql_check.tar.gz --strip-components=1 -C . && \
    make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config USE_PGXS=1 && \
@@ -413,7 +491,10 @@ COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 ARG PG_VERSION
 ENV PATH="/usr/local/pgsql/bin:$PATH"

-RUN case "${PG_VERSION}" in \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    case "${PG_VERSION}" in \
      "v14" | "v15") \
        export TIMESCALEDB_VERSION=2.10.1 \
        export TIMESCALEDB_CHECKSUM=6fca72a6ed0f6d32d2b3523951ede73dc5f9b0077b38450a029a5f411fdb8c73 \
@@ -446,7 +527,10 @@ COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 ARG PG_VERSION
 ENV PATH="/usr/local/pgsql/bin:$PATH"

-RUN case "${PG_VERSION}" in \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    case "${PG_VERSION}" in \
      "v14") \
        export PG_HINT_PLAN_VERSION=14_1_4_1 \
        export PG_HINT_PLAN_CHECKSUM=c3501becf70ead27f70626bce80ea401ceac6a77e2083ee5f3ff1f1444ec1ad1 \
@@ -459,6 +543,9 @@ RUN case "${PG_VERSION}" in \
        export PG_HINT_PLAN_VERSION=16_1_6_0 \
        export PG_HINT_PLAN_CHECKSUM=fc85a9212e7d2819d4ae4ac75817481101833c3cfa9f0fe1f980984e12347d00 \
        ;; \
+      "v17") \
+        echo "TODO: PG17 pg_hint_plan support" && exit 0 \
+        ;; \
      *) \
        echo "Export the valid PG_HINT_PLAN_VERSION variable" && exit 1 \
        ;; \
@@ -478,10 +565,14 @@ RUN case "${PG_VERSION}" in \
 #
 #########################################################################################
 FROM build-deps AS pg-cron-pg-build
+ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

 ENV PATH="/usr/local/pgsql/bin/:$PATH"
-RUN wget https://github.com/citusdata/pg_cron/archive/refs/tags/v1.6.0.tar.gz -O pg_cron.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    wget https://github.com/citusdata/pg_cron/archive/refs/tags/v1.6.0.tar.gz -O pg_cron.tar.gz && \
    echo "383a627867d730222c272bfd25cd5e151c578d73f696d32910c7db8c665cc7db pg_cron.tar.gz" | sha256sum --check && \
    mkdir pg_cron-src && cd pg_cron-src && tar xzf ../pg_cron.tar.gz --strip-components=1 -C . && \
    make -j $(getconf _NPROCESSORS_ONLN) && \
@@ -495,9 +586,13 @@ RUN wget https://github.com/citusdata/pg_cron/archive/refs/tags/v1.6.0.tar.gz -O
 #
 #########################################################################################
 FROM build-deps AS rdkit-pg-build
+ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

-RUN apt-get update && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    apt-get update && \
    apt-get install -y \
        cmake \
        libboost-iostreams1.74-dev \
@@ -507,7 +602,10 @@ RUN apt-get update && \
        libeigen3-dev

 ENV PATH="/usr/local/pgsql/bin/:/usr/local/pgsql/:$PATH"
-RUN wget https://github.com/rdkit/rdkit/archive/refs/tags/Release_2023_03_3.tar.gz -O rdkit.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    wget https://github.com/rdkit/rdkit/archive/refs/tags/Release_2023_03_3.tar.gz -O rdkit.tar.gz && \
    echo "bdbf9a2e6988526bfeb8c56ce3cdfe2998d60ac289078e2215374288185e8c8d rdkit.tar.gz" | sha256sum --check && \
    mkdir rdkit-src && cd rdkit-src && tar xzf ../rdkit.tar.gz --strip-components=1 -C . && \
    cmake \
@@ -544,10 +642,14 @@ RUN wget https://github.com/rdkit/rdkit/archive/refs/tags/Release_2023_03_3.tar.
 #
 #########################################################################################
 FROM build-deps AS pg-uuidv7-pg-build
+ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

 ENV PATH="/usr/local/pgsql/bin/:$PATH"
-RUN wget https://github.com/fboulnois/pg_uuidv7/archive/refs/tags/v1.0.1.tar.gz -O pg_uuidv7.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    wget https://github.com/fboulnois/pg_uuidv7/archive/refs/tags/v1.0.1.tar.gz -O pg_uuidv7.tar.gz && \
    echo "0d0759ab01b7fb23851ecffb0bce27822e1868a4a5819bfd276101c716637a7a pg_uuidv7.tar.gz" | sha256sum --check && \
    mkdir pg_uuidv7-src && cd pg_uuidv7-src && tar xzf ../pg_uuidv7.tar.gz --strip-components=1 -C . && \
    make -j $(getconf _NPROCESSORS_ONLN) && \
@@ -561,10 +663,14 @@ RUN wget https://github.com/fboulnois/pg_uuidv7/archive/refs/tags/v1.0.1.tar.gz
 #
 #########################################################################################
 FROM build-deps AS pg-roaringbitmap-pg-build
+ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

 ENV PATH="/usr/local/pgsql/bin/:$PATH"
-RUN wget https://github.com/ChenHuajun/pg_roaringbitmap/archive/refs/tags/v0.5.4.tar.gz -O pg_roaringbitmap.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions is not supported yet by pg_roaringbitmap. Quit" && exit 0;; \
+    esac && \
+    wget https://github.com/ChenHuajun/pg_roaringbitmap/archive/refs/tags/v0.5.4.tar.gz -O pg_roaringbitmap.tar.gz && \
    echo "b75201efcb1c2d1b014ec4ae6a22769cc7a224e6e406a587f5784a37b6b5a2aa pg_roaringbitmap.tar.gz" | sha256sum --check && \
    mkdir pg_roaringbitmap-src && cd pg_roaringbitmap-src && tar xzf ../pg_roaringbitmap.tar.gz --strip-components=1 -C . && \
    make -j $(getconf _NPROCESSORS_ONLN) && \
@@ -578,10 +684,14 @@ RUN wget https://github.com/ChenHuajun/pg_roaringbitmap/archive/refs/tags/v0.5.4
 #
 #########################################################################################
 FROM build-deps AS pg-semver-pg-build
+ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

 ENV PATH="/usr/local/pgsql/bin/:$PATH"
-RUN wget https://github.com/theory/pg-semver/archive/refs/tags/v0.32.1.tar.gz -O pg_semver.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 is not supported yet by pg_semver. Quit" && exit 0;; \
+    esac && \
+    wget https://github.com/theory/pg-semver/archive/refs/tags/v0.32.1.tar.gz -O pg_semver.tar.gz && \
    echo "fbdaf7512026d62eec03fad8687c15ed509b6ba395bff140acd63d2e4fbe25d7 pg_semver.tar.gz" | sha256sum --check && \
    mkdir pg_semver-src && cd pg_semver-src && tar xzf ../pg_semver.tar.gz --strip-components=1 -C . && \
    make -j $(getconf _NPROCESSORS_ONLN) && \
@@ -620,10 +730,14 @@ RUN case "${PG_VERSION}" in \
 #
 #########################################################################################
 FROM build-deps AS pg-anon-pg-build
+ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

 ENV PATH="/usr/local/pgsql/bin/:$PATH"
-RUN wget  https://github.com/neondatabase/postgresql_anonymizer/archive/refs/tags/neon_1.1.1.tar.gz -O pg_anon.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "postgresql_anonymizer does not yet support PG17" && exit 0;; \
+    esac && \
+    wget  https://github.com/neondatabase/postgresql_anonymizer/archive/refs/tags/neon_1.1.1.tar.gz -O pg_anon.tar.gz && \
    echo "321ea8d5c1648880aafde850a2c576e4a9e7b9933a34ce272efc839328999fa9  pg_anon.tar.gz" | sha256sum --check && \
    mkdir pg_anon-src && cd pg_anon-src && tar xzf ../pg_anon.tar.gz --strip-components=1 -C . && \
    find /usr/local/pgsql -type f | sed 's|^/usr/local/pgsql/||' > /before.txt &&\
@@ -641,6 +755,7 @@ RUN wget  https://github.com/neondatabase/postgresql_anonymizer/archive/refs/tag
 #
 #########################################################################################
 FROM build-deps AS rust-extensions-build
+ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

 RUN apt-get update && \
@@ -651,9 +766,11 @@ ENV HOME=/home/nonroot
 ENV PATH="/home/nonroot/.cargo/bin:/usr/local/pgsql/bin/:$PATH"
 USER nonroot
 WORKDIR /home/nonroot
-ARG PG_VERSION

-RUN curl -sSO https://static.rust-lang.org/rustup/dist/$(uname -m)-unknown-linux-gnu/rustup-init && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 is not supported yet by pgrx. Quit" && exit 0;; \
+    esac && \
+    curl -sSO https://static.rust-lang.org/rustup/dist/$(uname -m)-unknown-linux-gnu/rustup-init && \
    chmod +x rustup-init && \
    ./rustup-init -y --no-modify-path --profile minimal --default-toolchain stable && \
    rm rustup-init && \
@@ -672,7 +789,10 @@ USER root
 FROM rust-extensions-build AS pg-jsonschema-pg-build
 ARG PG_VERSION

-RUN wget https://github.com/supabase/pg_jsonschema/archive/refs/tags/v0.3.1.tar.gz -O pg_jsonschema.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "pg_jsonschema does not yet have a release that supports pg17" && exit 0;; \
+    esac && \
+    wget https://github.com/supabase/pg_jsonschema/archive/refs/tags/v0.3.1.tar.gz -O pg_jsonschema.tar.gz && \
    echo "61df3db1ed83cf24f6aa39c826f8818bfa4f0bd33b587fd6b2b1747985642297 pg_jsonschema.tar.gz" | sha256sum --check && \
    mkdir pg_jsonschema-src && cd pg_jsonschema-src && tar xzf ../pg_jsonschema.tar.gz --strip-components=1 -C . && \
    # see commit 252b3685a27a0f4c31a0f91e983c6314838e89e8
@@ -694,7 +814,10 @@ RUN wget https://github.com/supabase/pg_jsonschema/archive/refs/tags/v0.3.1.tar.
 FROM rust-extensions-build AS pg-graphql-pg-build
 ARG PG_VERSION

-RUN wget https://github.com/supabase/pg_graphql/archive/refs/tags/v1.5.7.tar.gz -O pg_graphql.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "pg_graphql does not yet have a release that supports pg17 as of now" && exit 0;; \
+    esac && \
+    wget https://github.com/supabase/pg_graphql/archive/refs/tags/v1.5.7.tar.gz -O pg_graphql.tar.gz && \
    echo "2b3e567a5b31019cb97ae0e33263c1bcc28580be5a444ac4c8ece5c4be2aea41 pg_graphql.tar.gz" | sha256sum --check && \
    mkdir pg_graphql-src && cd pg_graphql-src && tar xzf ../pg_graphql.tar.gz --strip-components=1 -C . && \
    sed -i 's/pgrx = "=0.11.3"/pgrx = { version = "0.11.3", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
@@ -714,7 +837,10 @@ FROM rust-extensions-build AS pg-tiktoken-pg-build
 ARG PG_VERSION

 # 26806147b17b60763039c6a6878884c41a262318 made on 26/09/2023
-RUN wget https://github.com/kelvich/pg_tiktoken/archive/26806147b17b60763039c6a6878884c41a262318.tar.gz -O pg_tiktoken.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "pg_tiktoken does not have versions, nor support for pg17" && exit 0;; \
+    esac && \
+    wget https://github.com/kelvich/pg_tiktoken/archive/26806147b17b60763039c6a6878884c41a262318.tar.gz -O pg_tiktoken.tar.gz && \
    echo "e64e55aaa38c259512d3e27c572da22c4637418cf124caba904cd50944e5004e pg_tiktoken.tar.gz" | sha256sum --check && \
    mkdir pg_tiktoken-src && cd pg_tiktoken-src && tar xzf ../pg_tiktoken.tar.gz --strip-components=1 -C . && \
    # TODO update pgrx version in the pg_tiktoken repo and remove this line
@@ -733,7 +859,10 @@ RUN wget https://github.com/kelvich/pg_tiktoken/archive/26806147b17b60763039c6a6
 FROM rust-extensions-build AS pg-pgx-ulid-build
 ARG PG_VERSION

-RUN wget https://github.com/pksunkara/pgx_ulid/archive/refs/tags/v0.1.5.tar.gz -O pgx_ulid.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "pgx_ulid does not support pg17 as of the latest version (0.1.5)" && exit 0;; \
+    esac && \
+    wget https://github.com/pksunkara/pgx_ulid/archive/refs/tags/v0.1.5.tar.gz -O pgx_ulid.tar.gz && \
    echo "9d1659a2da65af0133d5451c454de31b37364e3502087dadf579f790bc8bef17 pgx_ulid.tar.gz" | sha256sum --check && \
    mkdir pgx_ulid-src && cd pgx_ulid-src && tar xzf ../pgx_ulid.tar.gz --strip-components=1 -C . && \
    sed -i 's/pgrx       = "^0.11.2"/pgrx = { version = "=0.11.3", features = [ "unsafe-postgres" ] }/g' Cargo.toml && \
@@ -748,10 +877,14 @@ RUN wget https://github.com/pksunkara/pgx_ulid/archive/refs/tags/v0.1.5.tar.gz -
 #########################################################################################

 FROM build-deps AS wal2json-pg-build
+ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

 ENV PATH="/usr/local/pgsql/bin/:$PATH"
-RUN wget https://github.com/eulerto/wal2json/archive/refs/tags/wal2json_2_5.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "We'll need to update wal2json to 2.6+ for pg17 support" && exit 0;; \
+    esac && \
+    wget https://github.com/eulerto/wal2json/archive/refs/tags/wal2json_2_5.tar.gz && \
    echo "b516653575541cf221b99cf3f8be9b6821f6dbcfc125675c85f35090f824f00e wal2json_2_5.tar.gz" | sha256sum --check && \
    mkdir wal2json-src && cd wal2json-src && tar xzf ../wal2json_2_5.tar.gz --strip-components=1 -C . && \
    make -j $(getconf _NPROCESSORS_ONLN) && \
@@ -764,10 +897,14 @@ RUN wget https://github.com/eulerto/wal2json/archive/refs/tags/wal2json_2_5.tar.
 #
 #########################################################################################
 FROM build-deps AS pg-ivm-build
+ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

 ENV PATH="/usr/local/pgsql/bin/:$PATH"
-RUN wget https://github.com/sraoss/pg_ivm/archive/refs/tags/v1.7.tar.gz -O pg_ivm.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "We'll need to update pg_ivm to 1.9+ for pg17 support" && exit 0;; \
+    esac && \
+    wget https://github.com/sraoss/pg_ivm/archive/refs/tags/v1.7.tar.gz -O pg_ivm.tar.gz && \
    echo "ebfde04f99203c7be4b0e873f91104090e2e83e5429c32ac242d00f334224d5e pg_ivm.tar.gz" | sha256sum --check && \
    mkdir pg_ivm-src && cd pg_ivm-src && tar xzf ../pg_ivm.tar.gz --strip-components=1 -C . && \
    make -j $(getconf _NPROCESSORS_ONLN) && \
@@ -781,10 +918,14 @@ RUN wget https://github.com/sraoss/pg_ivm/archive/refs/tags/v1.7.tar.gz -O pg_iv
 #
 #########################################################################################
 FROM build-deps AS pg-partman-build
+ARG PG_VERSION
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/

 ENV PATH="/usr/local/pgsql/bin/:$PATH"
-RUN wget https://github.com/pgpartman/pg_partman/archive/refs/tags/v5.0.1.tar.gz -O pg_partman.tar.gz && \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "pg_partman doesn't support PG17 yet" && exit 0;; \
+    esac && \
+    wget https://github.com/pgpartman/pg_partman/archive/refs/tags/v5.0.1.tar.gz -O pg_partman.tar.gz && \
    echo "75b541733a9659a6c90dbd40fccb904a630a32880a6e3044d0c4c5f4c8a65525 pg_partman.tar.gz" | sha256sum --check && \
    mkdir pg_partman-src && cd pg_partman-src && tar xzf ../pg_partman.tar.gz --strip-components=1 -C . && \
    make -j $(getconf _NPROCESSORS_ONLN) && \
@@ -854,8 +995,8 @@ RUN make -j $(getconf _NPROCESSORS_ONLN) \
    case "${PG_VERSION}" in \
        "v14" | "v15") \
        ;; \
-        "v16") \
-            echo "Skipping HNSW for PostgreSQL 16" && exit 0 \
+        "v16" | "v17") \
+            echo "Skipping HNSW for PostgreSQL ${PG_VERSION}" && exit 0 \
        ;; \
        *) \
            echo "unexpected PostgreSQL version" && exit 1 \
@@ -899,7 +1040,7 @@ FROM neon-pg-ext-build AS postgres-cleanup-layer
 COPY --from=neon-pg-ext-build /usr/local/pgsql /usr/local/pgsql

 # Remove binaries from /bin/ that we won't use (or would manually copy & install otherwise)
-RUN cd /usr/local/pgsql/bin && rm ecpg raster2pgsql shp2pgsql pgtopo_export pgtopo_import pgsql2shp
+RUN cd /usr/local/pgsql/bin && rm -f ecpg raster2pgsql shp2pgsql pgtopo_export pgtopo_import pgsql2shp

 # Remove headers that we won't need anymore - we've completed installation of all extensions
 RUN rm -r /usr/local/pgsql/include
@@ -918,7 +1059,10 @@ RUN rm /usr/local/pgsql/lib/lib*.a

 FROM neon-pg-ext-build AS neon-pg-ext-test
 ARG PG_VERSION
-RUN mkdir /ext-src
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    mkdir /ext-src

 #COPY --from=postgis-build /postgis.tar.gz /ext-src/
 #COPY --from=postgis-build /sfcgal/* /usr
@@ -956,18 +1100,39 @@ COPY --from=pg-anon-pg-build /pg_anon.tar.gz /ext-src
 COPY patches/pg_anon.patch /ext-src
 COPY --from=pg-ivm-build /pg_ivm.tar.gz /ext-src
 COPY --from=pg-partman-build /pg_partman.tar.gz /ext-src
-RUN cd /ext-src/ && for f in *.tar.gz; \
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    cd /ext-src/ && for f in *.tar.gz; \
    do echo $f; dname=$(echo $f | sed 's/\.tar.*//')-src; \
    rm -rf $dname; mkdir $dname; tar xzf $f --strip-components=1 -C $dname \
    || exit 1; rm -f $f; done
-RUN cd /ext-src/pgvector-src && patch -p1 <../pgvector.patch
-RUN cd /ext-src/rum-src && patch -p1 <../rum.patch
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    cd /ext-src/rum-src && patch -p1 <../rum.patch
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    cd /ext-src/pgvector-src && patch -p1 <../pgvector.patch
 # cmake is required for the h3 test
-RUN apt-get update && apt-get install -y cmake
-RUN cd /ext-src/pg_hint_plan-src && patch -p1 < /ext-src/pg_hint_plan.patch
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    apt-get update && apt-get install -y cmake
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    cd /ext-src/pg_hint_plan-src && patch -p1 < /ext-src/pg_hint_plan.patch
 COPY --chmod=755 docker-compose/run-tests.sh /run-tests.sh
-RUN patch -p1 </ext-src/pg_anon.patch
-RUN patch -p1 </ext-src/pg_cron.patch
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    patch -p1 </ext-src/pg_anon.patch
+RUN case "${PG_VERSION}" in "v17") \
+    echo "v17 extensions are not supported yet. Quit" && exit 0;; \
+    esac && \
+    patch -p1 </ext-src/pg_cron.patch
 ENV PATH=/usr/local/pgsql/bin:$PATH
 ENV PGHOST=compute
 ENV PGPORT=55433
--- a/56
+++ b/56
@@ -119,6 +119,8 @@ $(POSTGRES_INSTALL_DIR)/build/%/config.status:
 # I'm not sure why it wouldn't work, but this is the only place (apart from
 # the "build-all-versions" entry points) where direct mention of PostgreSQL
 # versions is used.
+.PHONY: postgres-configure-v17
+postgres-configure-v17: $(POSTGRES_INSTALL_DIR)/build/v17/config.status
 .PHONY: postgres-configure-v16
 postgres-configure-v16: $(POSTGRES_INSTALL_DIR)/build/v16/config.status
 .PHONY: postgres-configure-v15
@@ -215,29 +217,31 @@ neon-pg-clean-ext-%:
 # they depend on openssl and other libraries that are not included in our
 # Rust build.
 .PHONY: walproposer-lib
-walproposer-lib: neon-pg-ext-v16
+walproposer-lib: neon-pg-ext-v17
 	+@echo "Compiling walproposer-lib"
 	mkdir -p $(POSTGRES_INSTALL_DIR)/build/walproposer-lib
-	$(MAKE) PG_CONFIG=$(POSTGRES_INSTALL_DIR)/v16/bin/pg_config CFLAGS='$(PG_CFLAGS) $(COPT)' \
+	$(MAKE) PG_CONFIG=$(POSTGRES_INSTALL_DIR)/v17/bin/pg_config CFLAGS='$(PG_CFLAGS) $(COPT)' \
 		-C $(POSTGRES_INSTALL_DIR)/build/walproposer-lib \
 		-f $(ROOT_PROJECT_DIR)/pgxn/neon/Makefile walproposer-lib
-	cp $(POSTGRES_INSTALL_DIR)/v16/lib/libpgport.a $(POSTGRES_INSTALL_DIR)/build/walproposer-lib
-	cp $(POSTGRES_INSTALL_DIR)/v16/lib/libpgcommon.a $(POSTGRES_INSTALL_DIR)/build/walproposer-lib
-ifeq ($(UNAME_S),Linux)
+	cp $(POSTGRES_INSTALL_DIR)/v17/lib/libpgport.a $(POSTGRES_INSTALL_DIR)/build/walproposer-lib
+	cp $(POSTGRES_INSTALL_DIR)/v17/lib/libpgcommon.a $(POSTGRES_INSTALL_DIR)/build/walproposer-lib
 	$(AR) d $(POSTGRES_INSTALL_DIR)/build/walproposer-lib/libpgport.a \
 		pg_strong_random.o
 	$(AR) d $(POSTGRES_INSTALL_DIR)/build/walproposer-lib/libpgcommon.a \
-		pg_crc32c.o \
-		hmac_openssl.o \
+		checksum_helper.o \
 		cryptohash_openssl.o \
-		scram-common.o \
+		hmac_openssl.o \
 		md5_common.o \
-		checksum_helper.o
+		parse_manifest.o \
+		scram-common.o
+ifeq ($(UNAME_S),Linux)
+	$(AR) d $(POSTGRES_INSTALL_DIR)/build/walproposer-lib/libpgcommon.a \
+		pg_crc32c.o
 endif

 .PHONY: walproposer-lib-clean
 walproposer-lib-clean:
-	$(MAKE) PG_CONFIG=$(POSTGRES_INSTALL_DIR)/v16/bin/pg_config \
+	$(MAKE) PG_CONFIG=$(POSTGRES_INSTALL_DIR)/v17/bin/pg_config \
 		-C $(POSTGRES_INSTALL_DIR)/build/walproposer-lib \
 		-f $(ROOT_PROJECT_DIR)/pgxn/neon/Makefile clean

@@ -245,38 +249,44 @@ walproposer-lib-clean:
 neon-pg-ext: \
 	neon-pg-ext-v14 \
 	neon-pg-ext-v15 \
-	neon-pg-ext-v16
+	neon-pg-ext-v16 \
+	neon-pg-ext-v17

 .PHONY: neon-pg-clean-ext
 neon-pg-clean-ext: \
 	neon-pg-clean-ext-v14 \
 	neon-pg-clean-ext-v15 \
-	neon-pg-clean-ext-v16
+	neon-pg-clean-ext-v16 \
+	neon-pg-clean-ext-v17

 # shorthand to build all Postgres versions
 .PHONY: postgres
 postgres: \
 	postgres-v14 \
 	postgres-v15 \
-	postgres-v16
+	postgres-v16 \
+	postgres-v17

 .PHONY: postgres-headers
 postgres-headers: \
 	postgres-headers-v14 \
 	postgres-headers-v15 \
-	postgres-headers-v16
+	postgres-headers-v16 \
+	postgres-headers-v17

 .PHONY: postgres-clean
 postgres-clean: \
 	postgres-clean-v14 \
 	postgres-clean-v15 \
-	postgres-clean-v16
+	postgres-clean-v16 \
+	postgres-clean-v17

 .PHONY: postgres-check
 postgres-check: \
 	postgres-check-v14 \
 	postgres-check-v15 \
-	postgres-check-v16
+	postgres-check-v16 \
+	postgres-check-v17

 # This doesn't remove the effects of 'configure'.
 .PHONY: clean
@@ -321,13 +331,13 @@ postgres-%-pgindent: postgres-%-pg-bsd-indent postgres-%-typedefs.list
 	rm -f pg*.BAK

 # Indent pxgn/neon.
-.PHONY: pgindent
-neon-pgindent: postgres-v16-pg-bsd-indent neon-pg-ext-v16
-	$(MAKE) PG_CONFIG=$(POSTGRES_INSTALL_DIR)/v16/bin/pg_config CFLAGS='$(PG_CFLAGS) $(COPT)' \
-		FIND_TYPEDEF=$(ROOT_PROJECT_DIR)/vendor/postgres-v16/src/tools/find_typedef \
-		INDENT=$(POSTGRES_INSTALL_DIR)/build/v16/src/tools/pg_bsd_indent/pg_bsd_indent \
-		PGINDENT_SCRIPT=$(ROOT_PROJECT_DIR)/vendor/postgres-v16/src/tools/pgindent/pgindent \
-		-C $(POSTGRES_INSTALL_DIR)/build/neon-v16 \
+.PHONY: neon-pgindent
+neon-pgindent: postgres-v17-pg-bsd-indent neon-pg-ext-v17
+	$(MAKE) PG_CONFIG=$(POSTGRES_INSTALL_DIR)/v17/bin/pg_config CFLAGS='$(PG_CFLAGS) $(COPT)' \
+		FIND_TYPEDEF=$(ROOT_PROJECT_DIR)/vendor/postgres-v17/src/tools/find_typedef \
+		INDENT=$(POSTGRES_INSTALL_DIR)/build/v17/src/tools/pg_bsd_indent/pg_bsd_indent \
+		PGINDENT_SCRIPT=$(ROOT_PROJECT_DIR)/vendor/postgres-v17/src/tools/pgindent/pgindent \
+		-C $(POSTGRES_INSTALL_DIR)/build/neon-v17 \
 		-f $(ROOT_PROJECT_DIR)/pgxn/neon/Makefile pgindent


--- a/compute_tools/Cargo.toml
+++ b/compute_tools/Cargo.toml
@@ -11,7 +11,6 @@ testing = []

 [dependencies]
 anyhow.workspace = true
-async-compression.workspace = true
 chrono.workspace = true
 cfg-if.workspace = true
 clap.workspace = true
@@ -24,7 +23,6 @@ num_cpus.workspace = true
 opentelemetry.workspace = true
 postgres.workspace = true
 regex.workspace = true
-serde.workspace = true
 serde_json.workspace = true
 signal-hook.workspace = true
 tar.workspace = true
@@ -43,7 +41,6 @@ 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/" }
 vm_monitor = { version = "0.1", path = "../libs/vm_monitor/" }
 zstd = "0.13"
--- a/compute_tools/src/compute.rs
+++ b/compute_tools/src/compute.rs
@@ -1052,26 +1052,19 @@ impl ComputeNode {
        let pg_process = self.start_postgres(pspec.storage_auth_token.clone())?;

        let config_time = Utc::now();
-        if pspec.spec.mode == ComputeMode::Primary {
-            if !pspec.spec.skip_pg_catalog_updates {
-                let pgdata_path = Path::new(&self.pgdata);
-                // temporarily reset max_cluster_size in config
-                // to avoid the possibility of hitting the limit, while we are applying config:
-                // creating new extensions, roles, etc...
-                config::with_compute_ctl_tmp_override(
-                    pgdata_path,
-                    "neon.max_cluster_size=-1",
-                    || {
-                        self.pg_reload_conf()?;
-
-                        self.apply_config(&compute_state)?;
-
-                        Ok(())
-                    },
-                )?;
+        if pspec.spec.mode == ComputeMode::Primary && !pspec.spec.skip_pg_catalog_updates {
+            let pgdata_path = Path::new(&self.pgdata);
+            // temporarily reset max_cluster_size in config
+            // to avoid the possibility of hitting the limit, while we are applying config:
+            // creating new extensions, roles, etc...
+            config::with_compute_ctl_tmp_override(pgdata_path, "neon.max_cluster_size=-1", || {
                self.pg_reload_conf()?;
-            }
-            self.post_apply_config()?;
+
+                self.apply_config(&compute_state)?;
+
+                Ok(())
+            })?;
+            self.pg_reload_conf()?;
        }

        let startup_end_time = Utc::now();
--- a/compute_tools/src/extension_server.rs
+++ b/compute_tools/src/extension_server.rs
@@ -124,6 +124,7 @@ fn parse_pg_version(human_version: &str) -> &str {
            "14" => return "v14",
            "15" => return "v15",
            "16" => return "v16",
+            "17" => return "v17",
            _ => {}
        },
        _ => {}
--- a/compute_tools/src/migrations/0011-grant_pg_show_replication_origin_status_to_neon_superuser.sql
+++ b/compute_tools/src/migrations/0011-grant_pg_show_replication_origin_status_to_neon_superuser.sql
@@ -0,0 +1 @@
+GRANT EXECUTE ON FUNCTION pg_show_replication_origin_status TO neon_superuser;
--- a/compute_tools/src/spec.rs
+++ b/compute_tools/src/spec.rs
@@ -793,6 +793,9 @@ pub fn handle_migrations(client: &mut Client) -> Result<()> {
        include_str!(
            "./migrations/0010-grant_snapshot_synchronization_funcs_to_neon_superuser.sql"
        ),
+        include_str!(
+            "./migrations/0011-grant_pg_show_replication_origin_status_to_neon_superuser.sql"
+        ),
    ];

    MigrationRunner::new(client, &migrations).run_migrations()?;
--- a/control_plane/Cargo.toml
+++ b/control_plane/Cargo.toml
@@ -9,13 +9,10 @@ anyhow.workspace = true
 camino.workspace = true
 clap.workspace = true
 comfy-table.workspace = true
-futures.workspace = true
 git-version.workspace = true
 humantime.workspace = true
 nix.workspace = true
 once_cell.workspace = true
-postgres.workspace = true
-hex.workspace = true
 humantime-serde.workspace = true
 hyper.workspace = true
 regex.workspace = true
@@ -23,8 +20,6 @@ reqwest = { workspace = true, features = ["blocking", "json"] }
 scopeguard.workspace = true
 serde.workspace = true
 serde_json.workspace = true
-serde_with.workspace = true
-tar.workspace = true
 thiserror.workspace = true
 toml.workspace = true
 toml_edit.workspace = true
--- a/control_plane/src/bin/neon_local.rs
+++ b/control_plane/src/bin/neon_local.rs
@@ -106,6 +106,7 @@ fn main() -> Result<()> {
            "stop" => rt.block_on(handle_stop_all(sub_args, &env)),
            "pageserver" => rt.block_on(handle_pageserver(sub_args, &env)),
            "storage_controller" => rt.block_on(handle_storage_controller(sub_args, &env)),
+            "storage_broker" => rt.block_on(handle_storage_broker(sub_args, &env)),
            "safekeeper" => rt.block_on(handle_safekeeper(sub_args, &env)),
            "endpoint" => rt.block_on(handle_endpoint(sub_args, &env)),
            "mappings" => handle_mappings(sub_args, &mut env),
@@ -640,6 +641,8 @@ async fn handle_timeline(timeline_match: &ArgMatches, env: &mut local_env::Local
        }
        Some(("branch", branch_match)) => {
            let tenant_id = get_tenant_id(branch_match, env)?;
+            let new_timeline_id =
+                parse_timeline_id(branch_match)?.unwrap_or(TimelineId::generate());
            let new_branch_name = branch_match
                .get_one::<String>("branch-name")
                .ok_or_else(|| anyhow!("No branch name provided"))?;
@@ -658,7 +661,6 @@ async fn handle_timeline(timeline_match: &ArgMatches, env: &mut local_env::Local
                .map(|lsn_str| Lsn::from_str(lsn_str))
                .transpose()
                .context("Failed to parse ancestor start Lsn from the request")?;
-            let new_timeline_id = TimelineId::generate();
            let storage_controller = StorageController::from_env(env);
            let create_req = TimelineCreateRequest {
                new_timeline_id,
@@ -1244,6 +1246,32 @@ async fn handle_safekeeper(sub_match: &ArgMatches, env: &local_env::LocalEnv) ->
    Ok(())
 }

+async fn handle_storage_broker(sub_match: &ArgMatches, env: &local_env::LocalEnv) -> Result<()> {
+    let (sub_name, sub_args) = match sub_match.subcommand() {
+        Some(broker_command_data) => broker_command_data,
+        None => bail!("no broker subcommand provided"),
+    };
+
+    match sub_name {
+        "start" => {
+            if let Err(e) = broker::start_broker_process(env, get_start_timeout(sub_args)).await {
+                eprintln!("broker start failed: {e}");
+                exit(1);
+            }
+        }
+
+        "stop" => {
+            if let Err(e) = broker::stop_broker_process(env) {
+                eprintln!("broker stop failed: {e}");
+                exit(1);
+            }
+        }
+
+        _ => bail!("Unexpected broker subcommand '{}'", sub_name),
+    }
+    Ok(())
+}
+
 async fn handle_start_all(
    env: &local_env::LocalEnv,
    retry_timeout: &Duration,
@@ -1570,7 +1598,6 @@ fn cli() -> Command {
                        .value_parser(value_parser!(PathBuf))
                        .value_name("config")
                )
-                .arg(pg_version_arg.clone())
                .arg(force_arg)
        )
        .subcommand(
@@ -1583,6 +1610,7 @@ fn cli() -> Command {
            .subcommand(Command::new("branch")
                .about("Create a new timeline, using another timeline as a base, copying its data")
                .arg(tenant_id_arg.clone())
+                .arg(timeline_id_arg.clone())
                .arg(branch_name_arg.clone())
                .arg(Arg::new("ancestor-branch-name").long("ancestor-branch-name")
                    .help("Use last Lsn of another timeline (and its data) as base when creating the new timeline. The timeline gets resolved by its branch name.").required(false))
@@ -1671,6 +1699,19 @@ fn cli() -> Command {
                            .arg(stop_mode_arg.clone())
                            .arg(instance_id))
        )
+        .subcommand(
+            Command::new("storage_broker")
+                .arg_required_else_help(true)
+                .about("Manage broker")
+                .subcommand(Command::new("start")
+                            .about("Start broker")
+                            .arg(timeout_arg.clone())
+                )
+                .subcommand(Command::new("stop")
+                            .about("Stop broker")
+                            .arg(stop_mode_arg.clone())
+                )
+        )
        .subcommand(
            Command::new("safekeeper")
                .arg_required_else_help(true)
--- a/control_plane/src/local_env.rs
+++ b/control_plane/src/local_env.rs
@@ -342,7 +342,7 @@ impl LocalEnv {

        #[allow(clippy::manual_range_patterns)]
        match pg_version {
-            14 | 15 | 16 => Ok(path.join(format!("v{pg_version}"))),
+            14 | 15 | 16 | 17 => Ok(path.join(format!("v{pg_version}"))),
            _ => bail!("Unsupported postgres version: {}", pg_version),
        }
    }
--- a/control_plane/src/pageserver.rs
+++ b/control_plane/src/pageserver.rs
@@ -75,14 +75,14 @@ impl PageServerNode {
        }
    }

-    fn pageserver_make_identity_toml(&self, node_id: NodeId) -> toml_edit::Document {
-        toml_edit::Document::from_str(&format!("id={node_id}")).unwrap()
+    fn pageserver_make_identity_toml(&self, node_id: NodeId) -> toml_edit::DocumentMut {
+        toml_edit::DocumentMut::from_str(&format!("id={node_id}")).unwrap()
    }

    fn pageserver_init_make_toml(
        &self,
        conf: NeonLocalInitPageserverConf,
-    ) -> anyhow::Result<toml_edit::Document> {
+    ) -> anyhow::Result<toml_edit::DocumentMut> {
        assert_eq!(&PageServerConf::from(&conf), &self.conf, "during neon_local init, we derive the runtime state of ps conf (self.conf) from the --config flag fully");

        // TODO(christian): instead of what we do here, create a pageserver_api::config::ConfigToml (PR #7656)
@@ -137,9 +137,9 @@ impl PageServerNode {

        // Turn `overrides` into a toml document.
        // TODO: above code is legacy code, it should be refactored to use toml_edit directly.
-        let mut config_toml = toml_edit::Document::new();
+        let mut config_toml = toml_edit::DocumentMut::new();
        for fragment_str in overrides {
-            let fragment = toml_edit::Document::from_str(&fragment_str)
+            let fragment = toml_edit::DocumentMut::from_str(&fragment_str)
                .expect("all fragments in `overrides` are valid toml documents, this function controls that");
            for (key, item) in fragment.iter() {
                config_toml.insert(key, item.clone());
--- a/control_plane/src/storage_controller.rs
+++ b/control_plane/src/storage_controller.rs
@@ -28,6 +28,7 @@ use utils::{
    auth::{encode_from_key_file, Claims, Scope},
    id::{NodeId, TenantId},
 };
+use whoami::username;

 pub struct StorageController {
    env: LocalEnv,
@@ -183,7 +184,7 @@ impl StorageController {
    /// to other versions if that one isn't found.  Some automated tests create circumstances
    /// where only one version is available in pg_distrib_dir, such as `test_remote_extensions`.
    async fn get_pg_dir(&self, dir_name: &str) -> anyhow::Result<Utf8PathBuf> {
-        let prefer_versions = [STORAGE_CONTROLLER_POSTGRES_VERSION, 15, 14];
+        let prefer_versions = [STORAGE_CONTROLLER_POSTGRES_VERSION, 16, 15, 14];

        for v in prefer_versions {
            let path = Utf8PathBuf::from_path_buf(self.env.pg_dir(v, dir_name)?).unwrap();
@@ -211,7 +212,16 @@ impl StorageController {
    /// Readiness check for our postgres process
    async fn pg_isready(&self, pg_bin_dir: &Utf8Path, postgres_port: u16) -> anyhow::Result<bool> {
        let bin_path = pg_bin_dir.join("pg_isready");
-        let args = ["-h", "localhost", "-p", &format!("{}", postgres_port)];
+        let args = [
+            "-h",
+            "localhost",
+            "-U",
+            &username(),
+            "-d",
+            DB_NAME,
+            "-p",
+            &format!("{}", postgres_port),
+        ];
        let exitcode = Command::new(bin_path).args(args).spawn()?.wait().await?;

        Ok(exitcode.success())
@@ -225,7 +235,11 @@ impl StorageController {
    ///
    /// Returns the database url
    pub async fn setup_database(&self, postgres_port: u16) -> anyhow::Result<String> {
-        let database_url = format!("postgresql://localhost:{}/{DB_NAME}", postgres_port);
+        let database_url = format!(
+            "postgresql://{}@localhost:{}/{DB_NAME}",
+            &username(),
+            postgres_port
+        );

        let pg_bin_dir = self.get_pg_bin_dir().await?;
        let createdb_path = pg_bin_dir.join("createdb");
@@ -235,6 +249,10 @@ impl StorageController {
                "localhost",
                "-p",
                &format!("{}", postgres_port),
+                "-U",
+                &username(),
+                "-O",
+                &username(),
                DB_NAME,
            ])
            .output()
@@ -271,7 +289,7 @@ impl StorageController {
            // But tokio-postgres fork doesn't have this upstream commit:
            // https://github.com/sfackler/rust-postgres/commit/cb609be758f3fb5af537f04b584a2ee0cebd5e79
            // => we should rebase our fork => TODO https://github.com/neondatabase/neon/issues/8399
-            .user(&whoami::username())
+            .user(&username())
            .dbname(DB_NAME)
            .connect(tokio_postgres::NoTls)
            .await
@@ -328,6 +346,12 @@ impl StorageController {
            let pg_log_path = pg_data_path.join("postgres.log");

            if !tokio::fs::try_exists(&pg_data_path).await? {
+                let initdb_args = ["-D", pg_data_path.as_ref(), "--username", &username()];
+                tracing::info!(
+                    "Initializing storage controller database with args: {:?}",
+                    initdb_args
+                );
+
                // Initialize empty database
                let initdb_path = pg_bin_dir.join("initdb");
                let mut child = Command::new(&initdb_path)
@@ -335,7 +359,7 @@ impl StorageController {
                        ("LD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
                        ("DYLD_LIBRARY_PATH".to_owned(), pg_lib_dir.to_string()),
                    ])
-                    .args(["-D", pg_data_path.as_ref()])
+                    .args(initdb_args)
                    .spawn()
                    .expect("Failed to spawn initdb");
                let status = child.wait().await?;
@@ -364,8 +388,14 @@ impl StorageController {
                pg_data_path.as_ref(),
                "-l",
                pg_log_path.as_ref(),
+                "-U",
+                &username(),
                "start",
            ];
+            tracing::info!(
+                "Starting storage controller database with args: {:?}",
+                db_start_args
+            );

            background_process::start_process(
                "storage_controller_db",
--- a/control_plane/storcon_cli/Cargo.toml
+++ b/control_plane/storcon_cli/Cargo.toml
@@ -11,14 +11,11 @@ clap.workspace = true
 comfy-table.workspace = true
 futures.workspace = true
 humantime.workspace = true
-hyper.workspace = true
 pageserver_api.workspace = true
 pageserver_client.workspace = true
 reqwest.workspace = true
-serde.workspace = true
 serde_json = { workspace = true, features = ["raw_value"] }
 storage_controller_client.workspace = true
-thiserror.workspace = true
 tokio.workspace = true
 tracing.workspace = true
 utils.workspace = true
--- a/libs/compute_api/Cargo.toml
+++ b/libs/compute_api/Cargo.toml
@@ -8,7 +8,6 @@ license.workspace = true
 anyhow.workspace = true
 chrono.workspace = true
 serde.workspace = true
-serde_with.workspace = true
 serde_json.workspace = true
 regex.workspace = true

--- a/libs/consumption_metrics/Cargo.toml
+++ b/libs/consumption_metrics/Cargo.toml
@@ -5,9 +5,6 @@ edition = "2021"
 license = "Apache-2.0"

 [dependencies]
-anyhow.workspace = true
 chrono = { workspace = true, features = ["serde"] }
 rand.workspace = true
 serde.workspace = true
-serde_with.workspace = true
-utils.workspace = true
--- a/libs/desim/Cargo.toml
+++ b/libs/desim/Cargo.toml
@@ -12,5 +12,4 @@ bytes.workspace = true
 utils.workspace = true
 parking_lot.workspace = true
 hex.workspace = true
-scopeguard.workspace = true
 smallvec = { workspace = true, features = ["write"] }
--- a/libs/pageserver_api/src/key.rs
+++ b/libs/pageserver_api/src/key.rs
@@ -1,8 +1,8 @@
 use anyhow::{bail, Result};
 use byteorder::{ByteOrder, BE};
 use postgres_ffi::relfile_utils::{FSM_FORKNUM, VISIBILITYMAP_FORKNUM};
+use postgres_ffi::Oid;
 use postgres_ffi::RepOriginId;
-use postgres_ffi::{Oid, TransactionId};
 use serde::{Deserialize, Serialize};
 use std::{fmt, ops::Range};

@@ -263,15 +263,6 @@ impl Key {
        field5: u8::MAX,
        field6: u32::MAX,
    };
-    /// A key slightly smaller than [`Key::MAX`] for use in layer key ranges to avoid them to be confused with L0 layers
-    pub const NON_L0_MAX: Key = Key {
-        field1: u8::MAX,
-        field2: u32::MAX,
-        field3: u32::MAX,
-        field4: u32::MAX,
-        field5: u8::MAX,
-        field6: u32::MAX - 1,
-    };

    pub fn from_hex(s: &str) -> Result<Self> {
        if s.len() != 36 {
@@ -359,7 +350,17 @@ impl Key {
 // 02 00000000 00000000 00000000 00   00000000
 //
 // TwoPhaseFile:
-// 02 00000000 00000000 00000000 00   XID
+//
+// 02 00000000 00000000 00XXXXXX XX   XXXXXXXX
+//
+//                        \______XID_________/
+//
+// The 64-bit XID is stored a little awkwardly in field6, field5 and
+// field4. PostgreSQL v16 and below only stored a 32-bit XID, which
+// fit completely in field6, but starting with PostgreSQL v17, a full
+// 64-bit XID is used. Most pageserver code that accesses
+// TwoPhaseFiles now deals with 64-bit XIDs even on v16, the high bits
+// are just unused.
 //
 // ControlFile:
 // 03 00000000 00000000 00000000 00   00000000
@@ -591,35 +592,36 @@ pub const TWOPHASEDIR_KEY: Key = Key {
 };

 #[inline(always)]
-pub fn twophase_file_key(xid: TransactionId) -> Key {
+pub fn twophase_file_key(xid: u64) -> Key {
    Key {
        field1: 0x02,
        field2: 0,
        field3: 0,
-        field4: 0,
-        field5: 0,
-        field6: xid,
+        field4: ((xid & 0xFFFFFF0000000000) >> 40) as u32,
+        field5: ((xid & 0x000000FF00000000) >> 32) as u8,
+        field6: (xid & 0x00000000FFFFFFFF) as u32,
    }
 }

 #[inline(always)]
-pub fn twophase_key_range(xid: TransactionId) -> Range<Key> {
+pub fn twophase_key_range(xid: u64) -> Range<Key> {
+    // 64-bit XIDs really should not overflow
    let (next_xid, overflowed) = xid.overflowing_add(1);

    Key {
        field1: 0x02,
        field2: 0,
        field3: 0,
-        field4: 0,
-        field5: 0,
-        field6: xid,
+        field4: ((xid & 0xFFFFFF0000000000) >> 40) as u32,
+        field5: ((xid & 0x000000FF00000000) >> 32) as u8,
+        field6: (xid & 0x00000000FFFFFFFF) as u32,
    }..Key {
        field1: 0x02,
        field2: 0,
-        field3: 0,
-        field4: 0,
-        field5: u8::from(overflowed),
-        field6: next_xid,
+        field3: u32::from(overflowed),
+        field4: ((next_xid & 0xFFFFFF0000000000) >> 40) as u32,
+        field5: ((next_xid & 0x000000FF00000000) >> 32) as u8,
+        field6: (next_xid & 0x00000000FFFFFFFF) as u32,
    }
 }

--- a/libs/pageserver_api/src/models.rs
+++ b/libs/pageserver_api/src/models.rs
@@ -62,7 +62,7 @@ use bytes::{Buf, BufMut, Bytes, BytesMut};
    serde::Serialize,
    serde::Deserialize,
    strum_macros::Display,
-    strum_macros::EnumVariantNames,
+    strum_macros::VariantNames,
    strum_macros::AsRefStr,
    strum_macros::IntoStaticStr,
 )]
--- a/libs/pageserver_api/src/models/utilization.rs
+++ b/libs/pageserver_api/src/models/utilization.rs
@@ -89,8 +89,19 @@ impl PageserverUtilization {

    /// If a node is currently hosting more work than it can comfortably handle.  This does not indicate that
    /// it will fail, but it is a strong signal that more work should not be added unless there is no alternative.
+    ///
+    /// When a node is overloaded, we may override soft affinity preferences and do things like scheduling
+    /// into a node in a less desirable AZ, if all the nodes in the preferred AZ are overloaded.
    pub fn is_overloaded(score: RawScore) -> bool {
-        score >= Self::UTILIZATION_FULL
+        // Why the factor of two?  This is unscientific but reflects behavior of real systems:
+        // - In terms of shard counts, a node's preferred max count is a soft limit intended to keep
+        //   startup and housekeeping jobs nice and responsive.  We can go to double this limit if needed
+        //   until some more nodes are deployed.
+        // - In terms of disk space, the node's utilization heuristic assumes every tenant needs to
+        //   hold its biggest timeline fully on disk, which is tends to be an over estimate when
+        //   some tenants are very idle and have dropped layers from disk.  In practice going up to
+        //   double is generally better than giving up and scheduling in a sub-optimal AZ.
+        score >= 2 * Self::UTILIZATION_FULL
    }

    pub fn adjust_shard_count_max(&mut self, shard_count: u32) {
--- a/libs/postgres_backend/Cargo.toml
+++ b/libs/postgres_backend/Cargo.toml
@@ -5,10 +5,8 @@ edition.workspace = true
 license.workspace = true

 [dependencies]
-async-trait.workspace = true
 anyhow.workspace = true
 bytes.workspace = true
-futures.workspace = true
 rustls.workspace = true
 serde.workspace = true
 thiserror.workspace = true
--- a/libs/postgres_backend/src/lib.rs
+++ b/libs/postgres_backend/src/lib.rs
@@ -81,17 +81,16 @@ pub fn is_expected_io_error(e: &io::Error) -> bool {
    )
 }

-#[async_trait::async_trait]
 pub trait Handler<IO> {
    /// Handle single query.
    /// postgres_backend will issue ReadyForQuery after calling this (this
    /// might be not what we want after CopyData streaming, but currently we don't
    /// care). It will also flush out the output buffer.
-    async fn process_query(
+    fn process_query(
        &mut self,
        pgb: &mut PostgresBackend<IO>,
        query_string: &str,
-    ) -> Result<(), QueryError>;
+    ) -> impl Future<Output = Result<(), QueryError>>;

    /// Called on startup packet receival, allows to process params.
    ///
--- a/libs/postgres_backend/tests/simple_select.rs
+++ b/libs/postgres_backend/tests/simple_select.rs
@@ -23,7 +23,6 @@ async fn make_tcp_pair() -> (TcpStream, TcpStream) {

 struct TestHandler {}

-#[async_trait::async_trait]
 impl<IO: AsyncRead + AsyncWrite + Unpin + Send> Handler<IO> for TestHandler {
    // return single col 'hey' for any query
    async fn process_query(
--- a/libs/postgres_ffi/Cargo.toml
+++ b/libs/postgres_ffi/Cargo.toml
@@ -5,13 +5,10 @@ edition.workspace = true
 license.workspace = true

 [dependencies]
-rand.workspace = true
 regex.workspace = true
 bytes.workspace = true
-byteorder.workspace = true
 anyhow.workspace = true
 crc32c.workspace = true
-hex.workspace = true
 once_cell.workspace = true
 log.workspace = true
 memoffset.workspace = true
--- a/libs/postgres_ffi/build.rs
+++ b/libs/postgres_ffi/build.rs
@@ -56,7 +56,7 @@ fn main() -> anyhow::Result<()> {
        PathBuf::from("pg_install")
    };

-    for pg_version in &["v14", "v15", "v16"] {
+    for pg_version in &["v14", "v15", "v16", "v17"] {
        let mut pg_install_dir_versioned = pg_install_dir.join(pg_version);
        if pg_install_dir_versioned.is_relative() {
            let cwd = env::current_dir().context("Failed to get current_dir")?;
@@ -121,6 +121,7 @@ fn main() -> anyhow::Result<()> {
            .allowlist_type("XLogPageHeaderData")
            .allowlist_type("XLogLongPageHeaderData")
            .allowlist_var("XLOG_PAGE_MAGIC")
+            .allowlist_var("PG_MAJORVERSION_NUM")
            .allowlist_var("PG_CONTROL_FILE_SIZE")
            .allowlist_var("PG_CONTROLFILEDATA_OFFSETOF_CRC")
            .allowlist_type("PageHeaderData")
--- a/libs/postgres_ffi/src/lib.rs
+++ b/libs/postgres_ffi/src/lib.rs
@@ -44,6 +44,9 @@ macro_rules! postgres_ffi {
            // Re-export some symbols from bindings
            pub use bindings::DBState_DB_SHUTDOWNED;
            pub use bindings::{CheckPoint, ControlFileData, XLogRecord};
+
+            pub const ZERO_CHECKPOINT: bytes::Bytes =
+                bytes::Bytes::from_static(&[0u8; xlog_utils::SIZEOF_CHECKPOINT]);
        }
    };
 }
@@ -54,6 +57,7 @@ macro_rules! for_all_postgres_versions {
        $macro!(v14);
        $macro!(v15);
        $macro!(v16);
+        $macro!(v17);
    };
 }

@@ -88,6 +92,7 @@ macro_rules! dispatch_pgversion {
                14 : v14,
                15 : v15,
                16 : v16,
+                17 : v17,
            ]
        )
    };
@@ -106,6 +111,110 @@ macro_rules! dispatch_pgversion {
    };
 }

+#[macro_export]
+macro_rules! enum_pgversion_dispatch {
+    ($name:expr, $typ:ident, $bind:ident, $code:block) => {
+        enum_pgversion_dispatch!(
+            name = $name,
+            bind = $bind,
+            typ = $typ,
+            code = $code,
+            pgversions = [
+                V14 : v14,
+                V15 : v15,
+                V16 : v16,
+                V17 : v17,
+            ]
+        )
+    };
+    (name = $name:expr,
+     bind = $bind:ident,
+     typ = $typ:ident,
+     code = $code:block,
+     pgversions = [$($variant:ident : $md:ident),+ $(,)?]) => {
+        match $name {
+            $(
+            self::$typ::$variant($bind) => {
+                use $crate::$md as pgv;
+                $code
+            }
+            ),+,
+        }
+    };
+}
+
+#[macro_export]
+macro_rules! enum_pgversion {
+    {$name:ident, pgv :: $t:ident} => {
+        enum_pgversion!{
+            name = $name,
+            typ = $t,
+            pgversions = [
+                V14 : v14,
+                V15 : v15,
+                V16 : v16,
+                V17 : v17,
+            ]
+        }
+    };
+    {$name:ident, pgv :: $p:ident :: $t:ident} => {
+        enum_pgversion!{
+            name = $name,
+            path = $p,
+            typ = $t,
+            pgversions = [
+                V14 : v14,
+                V15 : v15,
+                V16 : v16,
+                V17 : v17,
+            ]
+        }
+    };
+    {name = $name:ident,
+     typ = $t:ident,
+     pgversions = [$($variant:ident : $md:ident),+ $(,)?]} => {
+        pub enum $name {
+            $($variant ( $crate::$md::$t )),+
+        }
+        impl self::$name {
+            pub fn pg_version(&self) -> u32 {
+                enum_pgversion_dispatch!(self, $name, _ign, {
+                    pgv::bindings::PG_MAJORVERSION_NUM
+                })
+            }
+        }
+        $(
+        impl Into<self::$name> for $crate::$md::$t {
+            fn into(self) -> self::$name {
+                self::$name::$variant (self)
+            }
+        }
+        )+
+    };
+    {name = $name:ident,
+     path = $p:ident,
+     typ = $t:ident,
+     pgversions = [$($variant:ident : $md:ident),+ $(,)?]} => {
+        pub enum $name {
+            $($variant ($crate::$md::$p::$t)),+
+        }
+        impl $name {
+            pub fn pg_version(&self) -> u32 {
+                enum_pgversion_dispatch!(self, $name, _ign, {
+                    pgv::bindings::PG_MAJORVERSION_NUM
+                })
+            }
+        }
+        $(
+        impl Into<$name> for $crate::$md::$p::$t {
+            fn into(self) -> $name {
+                $name::$variant (self)
+            }
+        }
+        )+
+    };
+}
+
 pub mod pg_constants;
 pub mod relfile_utils;

--- a/libs/postgres_ffi/src/pg_constants.rs
+++ b/libs/postgres_ffi/src/pg_constants.rs
@@ -152,6 +152,9 @@ pub const XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED: u8 = (1 << 0) as u8;
 pub const XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED: u8 = (1 << 1) as u8;
 pub const XLH_DELETE_ALL_VISIBLE_CLEARED: u8 = (1 << 0) as u8;

+// From heapam_xlog.h
+pub const XLOG_HEAP2_REWRITE: u8 = 0x00;
+
 // From replication/message.h
 pub const XLOG_LOGICAL_MESSAGE: u8 = 0x00;

@@ -219,15 +222,20 @@ pub const INVALID_TRANSACTION_ID: u32 = 0;
 pub const FIRST_BOOTSTRAP_OBJECT_ID: u32 = 12000;
 pub const FIRST_NORMAL_OBJECT_ID: u32 = 16384;

+/* pg_control.h */
 pub const XLOG_CHECKPOINT_SHUTDOWN: u8 = 0x00;
 pub const XLOG_CHECKPOINT_ONLINE: u8 = 0x10;
-pub const XLP_FIRST_IS_CONTRECORD: u16 = 0x0001;
-pub const XLP_LONG_HEADER: u16 = 0x0002;
+pub const XLOG_PARAMETER_CHANGE: u8 = 0x60;
+pub const XLOG_END_OF_RECOVERY: u8 = 0x90;

 /* From xlog.h */
 pub const XLOG_REPLORIGIN_SET: u8 = 0x00;
 pub const XLOG_REPLORIGIN_DROP: u8 = 0x10;

+/* xlog_internal.h */
+pub const XLP_FIRST_IS_CONTRECORD: u16 = 0x0001;
+pub const XLP_LONG_HEADER: u16 = 0x0002;
+
 /* From replication/slot.h */
 pub const REPL_SLOT_ON_DISK_OFFSETOF_RESTART_LSN: usize = 4*4  /* offset of `slotdata` in ReplicationSlotOnDisk  */
   + 64 /* NameData */  + 4*4;
@@ -245,33 +253,6 @@ pub const VM_HEAPBLOCKS_PER_PAGE: u32 =
 /* From origin.c */
 pub const REPLICATION_STATE_MAGIC: u32 = 0x1257DADE;

-// List of subdirectories inside pgdata.
-// Copied from src/bin/initdb/initdb.c
-pub const PGDATA_SUBDIRS: [&str; 22] = [
-    "global",
-    "pg_wal/archive_status",
-    "pg_commit_ts",
-    "pg_dynshmem",
-    "pg_notify",
-    "pg_serial",
-    "pg_snapshots",
-    "pg_subtrans",
-    "pg_twophase",
-    "pg_multixact",
-    "pg_multixact/members",
-    "pg_multixact/offsets",
-    "base",
-    "base/1",
-    "pg_replslot",
-    "pg_tblspc",
-    "pg_stat",
-    "pg_stat_tmp",
-    "pg_xact",
-    "pg_logical",
-    "pg_logical/snapshots",
-    "pg_logical/mappings",
-];
-
 // Don't include postgresql.conf as it is inconvenient on node start:
 // we need postgresql.conf before basebackup to synchronize safekeepers
 // so no point in overwriting it during backup restore. Rest of the files
--- a/libs/postgres_ffi/src/pg_constants_v14.rs
+++ b/libs/postgres_ffi/src/pg_constants_v14.rs
@@ -5,6 +5,33 @@ pub const BKPIMAGE_IS_COMPRESSED: u8 = 0x02; /* page image is compressed */
 pub const BKPIMAGE_APPLY: u8 = 0x04; /* page image should be restored during replay */
 pub const SIZEOF_RELMAPFILE: usize = 512; /* sizeof(RelMapFile) in relmapper.c */

+// List of subdirectories inside pgdata.
+// Copied from src/bin/initdb/initdb.c
+pub const PGDATA_SUBDIRS: [&str; 22] = [
+    "global",
+    "pg_wal/archive_status",
+    "pg_commit_ts",
+    "pg_dynshmem",
+    "pg_notify",
+    "pg_serial",
+    "pg_snapshots",
+    "pg_subtrans",
+    "pg_twophase",
+    "pg_multixact",
+    "pg_multixact/members",
+    "pg_multixact/offsets",
+    "base",
+    "base/1",
+    "pg_replslot",
+    "pg_tblspc",
+    "pg_stat",
+    "pg_stat_tmp",
+    "pg_xact",
+    "pg_logical",
+    "pg_logical/snapshots",
+    "pg_logical/mappings",
+];
+
 pub fn bkpimg_is_compressed(bimg_info: u8) -> bool {
    (bimg_info & BKPIMAGE_IS_COMPRESSED) != 0
 }
--- a/libs/postgres_ffi/src/pg_constants_v15.rs
+++ b/libs/postgres_ffi/src/pg_constants_v15.rs
@@ -11,6 +11,8 @@ pub const BKPIMAGE_COMPRESS_ZSTD: u8 = 0x10; /* page image is compressed */

 pub const SIZEOF_RELMAPFILE: usize = 512; /* sizeof(RelMapFile) in relmapper.c */

+pub use super::super::v14::bindings::PGDATA_SUBDIRS;
+
 pub fn bkpimg_is_compressed(bimg_info: u8) -> bool {
    const ANY_COMPRESS_FLAG: u8 = BKPIMAGE_COMPRESS_PGLZ | BKPIMAGE_COMPRESS_LZ4 | BKPIMAGE_COMPRESS_ZSTD;

--- a/libs/postgres_ffi/src/pg_constants_v16.rs
+++ b/libs/postgres_ffi/src/pg_constants_v16.rs
@@ -11,6 +11,8 @@ pub const BKPIMAGE_COMPRESS_ZSTD: u8 = 0x10; /* page image is compressed */

 pub const SIZEOF_RELMAPFILE: usize = 524; /* sizeof(RelMapFile) in relmapper.c */

+pub use super::super::v14::bindings::PGDATA_SUBDIRS;
+
 pub fn bkpimg_is_compressed(bimg_info: u8) -> bool {
    const ANY_COMPRESS_FLAG: u8 = BKPIMAGE_COMPRESS_PGLZ | BKPIMAGE_COMPRESS_LZ4 | BKPIMAGE_COMPRESS_ZSTD;

--- a/libs/postgres_ffi/src/pg_constants_v17.rs
+++ b/libs/postgres_ffi/src/pg_constants_v17.rs
@@ -0,0 +1,55 @@
+pub const XACT_XINFO_HAS_DROPPED_STATS: u32 = 1u32 << 8;
+
+pub const XLOG_DBASE_CREATE_FILE_COPY: u8 = 0x00;
+pub const XLOG_DBASE_CREATE_WAL_LOG: u8 = 0x10;
+pub const XLOG_DBASE_DROP: u8 = 0x20;
+
+pub const BKPIMAGE_APPLY: u8 = 0x02; /* page image should be restored during replay */
+pub const BKPIMAGE_COMPRESS_PGLZ: u8 = 0x04; /* page image is compressed */
+pub const BKPIMAGE_COMPRESS_LZ4: u8 = 0x08; /* page image is compressed */
+pub const BKPIMAGE_COMPRESS_ZSTD: u8 = 0x10; /* page image is compressed */
+
+pub const SIZEOF_RELMAPFILE: usize = 524; /* sizeof(RelMapFile) in relmapper.c */
+
+// List of subdirectories inside pgdata.
+// Copied from src/bin/initdb/initdb.c
+pub const PGDATA_SUBDIRS: [&str; 23] = [
+    "global",
+    "pg_wal/archive_status",
+    "pg_wal/summaries",
+    "pg_commit_ts",
+    "pg_dynshmem",
+    "pg_notify",
+    "pg_serial",
+    "pg_snapshots",
+    "pg_subtrans",
+    "pg_twophase",
+    "pg_multixact",
+    "pg_multixact/members",
+    "pg_multixact/offsets",
+    "base",
+    "base/1",
+    "pg_replslot",
+    "pg_tblspc",
+    "pg_stat",
+    "pg_stat_tmp",
+    "pg_xact",
+    "pg_logical",
+    "pg_logical/snapshots",
+    "pg_logical/mappings",
+];
+
+pub fn bkpimg_is_compressed(bimg_info: u8) -> bool {
+    const ANY_COMPRESS_FLAG: u8 = BKPIMAGE_COMPRESS_PGLZ | BKPIMAGE_COMPRESS_LZ4 | BKPIMAGE_COMPRESS_ZSTD;
+
+    (bimg_info & ANY_COMPRESS_FLAG) != 0
+}
+
+
+pub const XLOG_HEAP2_PRUNE_ON_ACCESS: u8 = 0x10;
+pub const XLOG_HEAP2_PRUNE_VACUUM_SCAN: u8 = 0x20;
+pub const XLOG_HEAP2_PRUNE_VACUUM_CLEANUP: u8 = 0x30;
+
+
+pub const XLOG_OVERWRITE_CONTRECORD: u8 = 0xD0;
+pub const XLOG_CHECKPOINT_REDO: u8 = 0xE0;
--- a/libs/postgres_ffi/src/xlog_utils.rs
+++ b/libs/postgres_ffi/src/xlog_utils.rs
@@ -30,7 +30,7 @@ use std::fs::File;
 use std::io::prelude::*;
 use std::io::ErrorKind;
 use std::io::SeekFrom;
-use std::path::{Path, PathBuf};
+use std::path::Path;
 use std::time::SystemTime;
 use utils::bin_ser::DeserializeError;
 use utils::bin_ser::SerializeError;
@@ -260,13 +260,6 @@ fn open_wal_segment(seg_file_path: &Path) -> anyhow::Result<Option<File>> {
    }
 }

-pub fn main() {
-    let mut data_dir = PathBuf::new();
-    data_dir.push(".");
-    let wal_end = find_end_of_wal(&data_dir, WAL_SEGMENT_SIZE, Lsn(0)).unwrap();
-    println!("wal_end={:?}", wal_end);
-}
-
 impl XLogRecord {
    pub fn from_slice(buf: &[u8]) -> Result<XLogRecord, DeserializeError> {
        use utils::bin_ser::LeSer;
--- a/libs/postgres_ffi/wal_craft/Cargo.toml
+++ b/libs/postgres_ffi/wal_craft/Cargo.toml
@@ -9,7 +9,6 @@ anyhow.workspace = true
 clap.workspace = true
 env_logger.workspace = true
 log.workspace = true
-once_cell.workspace = true
 postgres.workspace = true
 postgres_ffi.workspace = true
 camino-tempfile.workspace = true
--- a/libs/postgres_ffi/wal_craft/src/lib.rs
+++ b/libs/postgres_ffi/wal_craft/src/lib.rs
@@ -53,7 +53,7 @@ impl Conf {

        #[allow(clippy::manual_range_patterns)]
        match self.pg_version {
-            14 | 15 | 16 => Ok(path.join(format!("v{}", self.pg_version))),
+            14 | 15 | 16 | 17 => Ok(path.join(format!("v{}", self.pg_version))),
            _ => bail!("Unsupported postgres version: {}", self.pg_version),
        }
    }
--- a/libs/pq_proto/Cargo.toml
+++ b/libs/pq_proto/Cargo.toml
@@ -8,10 +8,8 @@ license.workspace = true
 bytes.workspace = true
 byteorder.workspace = true
 itertools.workspace = true
-pin-project-lite.workspace = true
 postgres-protocol.workspace = true
 rand.workspace = true
 tokio = { workspace = true, features = ["io-util"] }
-tracing.workspace = true
 thiserror.workspace = true
 serde.workspace = true
--- a/libs/remote_storage/Cargo.toml
+++ b/libs/remote_storage/Cargo.toml
@@ -13,14 +13,11 @@ aws-smithy-async.workspace = true
 aws-smithy-types.workspace = true
 aws-config.workspace = true
 aws-sdk-s3.workspace = true
-aws-credential-types.workspace = true
 bytes.workspace = true
 camino = { workspace = true, features = ["serde1"] }
-humantime.workspace = true
 humantime-serde.workspace = true
 hyper = { workspace = true, features = ["stream"] }
 futures.workspace = true
-rand.workspace = true
 serde.workspace = true
 serde_json.workspace = true
 tokio = { workspace = true, features = ["sync", "fs", "io-util"] }
--- a/libs/remote_storage/src/config.rs
+++ b/libs/remote_storage/src/config.rs
@@ -185,7 +185,7 @@ mod tests {
    use super::*;

    fn parse(input: &str) -> anyhow::Result<RemoteStorageConfig> {
-        let toml = input.parse::<toml_edit::Document>().unwrap();
+        let toml = input.parse::<toml_edit::DocumentMut>().unwrap();
        RemoteStorageConfig::from_toml(toml.as_item())
    }

--- a/libs/safekeeper_api/Cargo.toml
+++ b/libs/safekeeper_api/Cargo.toml
@@ -6,6 +6,5 @@ license.workspace = true

 [dependencies]
 serde.workspace = true
-serde_with.workspace = true
 const_format.workspace = true
 utils.workspace = true
--- a/libs/tracing-utils/Cargo.toml
+++ b/libs/tracing-utils/Cargo.toml
@@ -9,8 +9,9 @@ hyper.workspace = true
 opentelemetry = { workspace = true, features=["rt-tokio"] }
 opentelemetry-otlp = { workspace = true, default-features=false, features = ["http-proto", "trace", "http", "reqwest-client"] }
 opentelemetry-semantic-conventions.workspace = true
-reqwest = { workspace = true, default-features = false, features = ["rustls-tls"] }
 tokio = { workspace = true, features = ["rt", "rt-multi-thread"] }
 tracing.workspace = true
 tracing-opentelemetry.workspace = true
-tracing-subscriber.workspace = true
+
+[dev-dependencies]
+tracing-subscriber.workspace = true    # For examples in docs
--- a/libs/utils/Cargo.toml
+++ b/libs/utils/Cargo.toml
@@ -42,7 +42,6 @@ tracing.workspace = true
 tracing-error.workspace = true
 tracing-subscriber = { workspace = true, features = ["json", "registry"] }
 rand.workspace = true
-serde_with.workspace = true
 strum.workspace = true
 strum_macros.workspace = true
 url.workspace = true
--- a/libs/utils/src/accum.rs
+++ b/libs/utils/src/accum.rs
@@ -1,33 +0,0 @@
-/// A helper to "accumulate" a value similar to `Iterator::reduce`, but lets you
-/// feed the accumulated values by calling the 'accum' function, instead of having an
-/// iterator.
-///
-/// For example, to calculate the smallest value among some integers:
-///
-/// ```
-/// use utils::accum::Accum;
-///
-/// let values = [1, 2, 3];
-///
-/// let mut min_value: Accum<u32> = Accum(None);
-/// for new_value in &values {
-///     min_value.accum(std::cmp::min, *new_value);
-/// }
-///
-/// assert_eq!(min_value.0.unwrap(), 1);
-/// ```
-pub struct Accum<T>(pub Option<T>);
-impl<T: Copy> Accum<T> {
-    pub fn accum<F>(&mut self, func: F, new_value: T)
-    where
-        F: FnOnce(T, T) -> T,
-    {
-        // If there is no previous value, just store the new value.
-        // Otherwise call the function to decide which one to keep.
-        self.0 = Some(if let Some(accum) = self.0 {
-            func(accum, new_value)
-        } else {
-            new_value
-        });
-    }
-}
--- a/libs/utils/src/id.rs
+++ b/libs/utils/src/id.rs
@@ -88,12 +88,6 @@ impl<'de> Deserialize<'de> for Id {
 }

 impl Id {
-    pub fn get_from_buf(buf: &mut impl bytes::Buf) -> Id {
-        let mut arr = [0u8; 16];
-        buf.copy_to_slice(&mut arr);
-        Id::from(arr)
-    }
-
    pub fn from_slice(src: &[u8]) -> Result<Id, IdError> {
        if src.len() != 16 {
            return Err(IdError::SliceParseError(src.len()));
@@ -179,10 +173,6 @@ impl fmt::Debug for Id {
 macro_rules! id_newtype {
    ($t:ident) => {
        impl $t {
-            pub fn get_from_buf(buf: &mut impl bytes::Buf) -> $t {
-                $t(Id::get_from_buf(buf))
-            }
-
            pub fn from_slice(src: &[u8]) -> Result<$t, IdError> {
                Ok($t(Id::from_slice(src)?))
            }
--- a/libs/utils/src/lib.rs
+++ b/libs/utils/src/lib.rs
@@ -43,16 +43,9 @@ pub mod logging;
 pub mod lock_file;
 pub mod pid_file;

-// Misc
-pub mod accum;
-pub mod shutdown;
-
 // Utility for binding TcpListeners with proper socket options.
 pub mod tcp_listener;

-// Utility for putting a raw file descriptor into non-blocking mode
-pub mod nonblock;
-
 // Default signal handling
 pub mod sentry_init;
 pub mod signals;
--- a/libs/utils/src/logging.rs
+++ b/libs/utils/src/logging.rs
@@ -3,11 +3,9 @@ use std::str::FromStr;
 use anyhow::Context;
 use metrics::{IntCounter, IntCounterVec};
 use once_cell::sync::Lazy;
-use strum_macros::{EnumString, EnumVariantNames};
+use strum_macros::{EnumString, VariantNames};

-#[derive(
-    EnumString, strum_macros::Display, EnumVariantNames, Eq, PartialEq, Debug, Clone, Copy,
-)]
+#[derive(EnumString, strum_macros::Display, VariantNames, Eq, PartialEq, Debug, Clone, Copy)]
 #[strum(serialize_all = "snake_case")]
 pub enum LogFormat {
    Plain,
--- a/libs/utils/src/lsn.rs
+++ b/libs/utils/src/lsn.rs
@@ -1,6 +1,5 @@
 #![warn(missing_docs)]

-use camino::Utf8Path;
 use serde::{de::Visitor, Deserialize, Serialize};
 use std::fmt;
 use std::ops::{Add, AddAssign};
@@ -145,14 +144,6 @@ impl Lsn {
        i128::from(self.0) - i128::from(other)
    }

-    /// Parse an LSN from a filename in the form `0000000000000000`
-    pub fn from_filename<F>(filename: F) -> Result<Self, LsnParseError>
-    where
-        F: AsRef<Utf8Path>,
-    {
-        Lsn::from_hex(filename.as_ref().as_str())
-    }
-
    /// Parse an LSN from a string in the form `0000000000000000`
    pub fn from_hex<S>(s: S) -> Result<Self, LsnParseError>
    where
--- a/libs/utils/src/nonblock.rs
+++ b/libs/utils/src/nonblock.rs
@@ -1,17 +0,0 @@
-use nix::fcntl::{fcntl, OFlag, F_GETFL, F_SETFL};
-use std::os::unix::io::RawFd;
-
-/// Put a file descriptor into non-blocking mode
-pub fn set_nonblock(fd: RawFd) -> Result<(), std::io::Error> {
-    let bits = fcntl(fd, F_GETFL)?;
-
-    // If F_GETFL returns some unknown bits, they should be valid
-    // for passing back to F_SETFL, too. If we left them out, the F_SETFL
-    // would effectively clear them, which is not what we want.
-    let mut flags = OFlag::from_bits_retain(bits);
-    flags |= OFlag::O_NONBLOCK;
-
-    fcntl(fd, F_SETFL(flags))?;
-
-    Ok(())
-}
--- a/libs/utils/src/shutdown.rs
+++ b/libs/utils/src/shutdown.rs
@@ -1,7 +0,0 @@
-/// Immediately terminate the calling process without calling
-/// atexit callbacks, C runtime destructors etc. We mainly use
-/// this to protect coverage data from concurrent writes.
-pub fn exit_now(code: u8) -> ! {
-    // SAFETY: exiting is safe, the ffi is not safe
-    unsafe { nix::libc::_exit(code as _) };
-}
--- a/libs/utils/src/toml_edit_ext.rs
+++ b/libs/utils/src/toml_edit_ext.rs
@@ -10,7 +10,7 @@ pub fn deserialize_item<T>(item: &toml_edit::Item) -> Result<T, Error>
 where
    T: serde::de::DeserializeOwned,
 {
-    let document: toml_edit::Document = match item {
+    let document: toml_edit::DocumentMut = match item {
        toml_edit::Item::Table(toml) => toml.clone().into(),
        toml_edit::Item::Value(toml_edit::Value::InlineTable(toml)) => {
            toml.clone().into_table().into()
--- a/libs/vm_monitor/Cargo.toml
+++ b/libs/vm_monitor/Cargo.toml
@@ -15,13 +15,11 @@ anyhow.workspace = true
 axum.workspace = true
 clap.workspace = true
 futures.workspace = true
-inotify.workspace = true
 serde.workspace = true
 serde_json.workspace = true
 sysinfo.workspace = true
 tokio = { workspace = true, features = ["rt-multi-thread"] }
 tokio-postgres.workspace = true
-tokio-stream.workspace = true
 tokio-util.workspace = true
 tracing.workspace = true
 tracing-subscriber.workspace = true
--- a/libs/walproposer/build.rs
+++ b/libs/walproposer/build.rs
@@ -5,6 +5,8 @@ use std::{env, path::PathBuf, process::Command};

 use anyhow::{anyhow, Context};

+const WALPROPOSER_PG_VERSION: &str = "v17";
+
 fn main() -> anyhow::Result<()> {
    // Tell cargo to invalidate the built crate whenever the wrapper changes
    println!("cargo:rerun-if-changed=bindgen_deps.h");
@@ -36,7 +38,10 @@ fn main() -> anyhow::Result<()> {
    // Rebuild crate when libwalproposer.a changes
    println!("cargo:rerun-if-changed={walproposer_lib_search_str}/libwalproposer.a");

-    let pg_config_bin = pg_install_abs.join("v16").join("bin").join("pg_config");
+    let pg_config_bin = pg_install_abs
+        .join(WALPROPOSER_PG_VERSION)
+        .join("bin")
+        .join("pg_config");
    let inc_server_path: String = if pg_config_bin.exists() {
        let output = Command::new(pg_config_bin)
            .arg("--includedir-server")
@@ -53,7 +58,7 @@ fn main() -> anyhow::Result<()> {
            .into()
    } else {
        let server_path = pg_install_abs
-            .join("v16")
+            .join(WALPROPOSER_PG_VERSION)
            .join("include")
            .join("postgresql")
            .join("server")
--- a/pageserver/Cargo.toml
+++ b/pageserver/Cargo.toml
@@ -15,7 +15,6 @@ anyhow.workspace = true
 arc-swap.workspace = true
 async-compression.workspace = true
 async-stream.workspace = true
-async-trait.workspace = true
 bit_field.workspace = true
 byteorder.workspace = true
 bytes.workspace = true
@@ -23,12 +22,9 @@ camino.workspace = true
 camino-tempfile.workspace = true
 chrono = { workspace = true, features = ["serde"] }
 clap = { workspace = true, features = ["string"] }
-const_format.workspace = true
 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
@@ -57,10 +53,6 @@ serde.workspace = true
 serde_json = { workspace = true, features = ["raw_value"] }
 serde_path_to_error.workspace = true
 serde_with.workspace = true
-signal-hook.workspace = true
-smallvec = { workspace = true, features = ["write"] }
-svg_fmt.workspace = true
-sync_wrapper.workspace = true
 sysinfo.workspace = true
 tokio-tar.workspace = true
 thiserror.workspace = true
@@ -73,7 +65,6 @@ tokio-stream.workspace = true
 tokio-util.workspace = true
 toml_edit = { workspace = true, features = [ "serde" ] }
 tracing.workspace = true
-twox-hash.workspace = true
 url.workspace = true
 walkdir.workspace = true
 metrics.workspace = true
--- a/pageserver/compaction/Cargo.toml
+++ b/pageserver/compaction/Cargo.toml
@@ -9,41 +9,19 @@ default = []

 [dependencies]
 anyhow.workspace = true
-async-compression.workspace = true
 async-stream.workspace = true
-byteorder.workspace = true
-bytes.workspace = true
-chrono = { workspace = true, features = ["serde"] }
 clap = { workspace = true, features = ["string"] }
-const_format.workspace = true
-consumption_metrics.workspace = true
-crossbeam-utils.workspace = true
-either.workspace = true
-flate2.workspace = true
-fail.workspace = true
 futures.workspace = true
 git-version.workspace = true
-hex.workspace = true
-humantime.workspace = true
-humantime-serde.workspace = true
 itertools.workspace = true
 once_cell.workspace = true
 pageserver_api.workspace = true
 pin-project-lite.workspace = true
 rand.workspace = true
-smallvec = { workspace = true, features = ["write"] }
 svg_fmt.workspace = true
-sync_wrapper.workspace = true
-thiserror.workspace = true
 tokio = { workspace = true, features = ["process", "sync", "fs", "rt", "io-util", "time"] }
-tokio-io-timeout.workspace = true
-tokio-util.workspace = true
 tracing.workspace = true
-tracing-error.workspace = true
 tracing-subscriber.workspace = true
-url.workspace = true
-walkdir.workspace = true
-metrics.workspace = true
 utils.workspace = true
 workspace_hack.workspace = true

--- a/pageserver/ctl/Cargo.toml
+++ b/pageserver/ctl/Cargo.toml
@@ -8,7 +8,6 @@ license.workspace = true

 [dependencies]
 anyhow.workspace = true
-bytes.workspace = true
 camino.workspace = true
 clap = { workspace = true, features = ["string"] }
 git-version.workspace = true
@@ -24,5 +23,4 @@ toml_edit.workspace = true
 utils.workspace = true
 svg_fmt.workspace = true
 workspace_hack.workspace = true
-serde.workspace = true
 serde_json.workspace = true
--- a/pageserver/ctl/src/layer_map_analyzer.rs
+++ b/pageserver/ctl/src/layer_map_analyzer.rs
@@ -79,16 +79,24 @@ pub(crate) fn parse_filename(name: &str) -> Option<LayerFile> {
        return None;
    }
    let keys: Vec<&str> = split[0].split('-').collect();
-    let mut lsns: Vec<&str> = split[1].split('-').collect();
-    let is_delta = if lsns.len() == 1 {
-        lsns.push(lsns[0]);
+    let lsn_and_opt_generation: Vec<&str> = split[1].split('v').collect();
+    let lsns: Vec<&str> = lsn_and_opt_generation[0].split('-').collect();
+    let the_lsns: [&str; 2];
+
+    /*
+     * Generations add a -vX-XXXXXX postfix, which causes issues when we try to
+     * parse 'vX' as an LSN.
+     */
+    let is_delta = if lsns.len() == 1 || lsns[1].is_empty() {
+        the_lsns = [lsns[0], lsns[0]];
        false
    } else {
+        the_lsns = [lsns[0], lsns[1]];
        true
    };

    let key_range = Key::from_hex(keys[0]).unwrap()..Key::from_hex(keys[1]).unwrap();
-    let lsn_range = Lsn::from_hex(lsns[0]).unwrap()..Lsn::from_hex(lsns[1]).unwrap();
+    let lsn_range = Lsn::from_hex(the_lsns[0]).unwrap()..Lsn::from_hex(the_lsns[1]).unwrap();
    let holes = Vec::new();
    Some(LayerFile {
        key_range,
--- a/pageserver/ctl/src/main.rs
+++ b/pageserver/ctl/src/main.rs
@@ -174,7 +174,7 @@ async fn main() -> anyhow::Result<()> {
                println!("specified prefix '{}' failed validation", cmd.prefix);
                return Ok(());
            };
-            let toml_document = toml_edit::Document::from_str(&cmd.config_toml_str)?;
+            let toml_document = toml_edit::DocumentMut::from_str(&cmd.config_toml_str)?;
            let toml_item = toml_document
                .get("remote_storage")
                .expect("need remote_storage");
--- a/pageserver/src/basebackup.rs
+++ b/pageserver/src/basebackup.rs
@@ -30,9 +30,8 @@ use pageserver_api::reltag::{RelTag, SlruKind};

 use postgres_ffi::dispatch_pgversion;
 use postgres_ffi::pg_constants::{DEFAULTTABLESPACE_OID, GLOBALTABLESPACE_OID};
-use postgres_ffi::pg_constants::{PGDATA_SPECIAL_FILES, PGDATA_SUBDIRS, PG_HBA};
+use postgres_ffi::pg_constants::{PGDATA_SPECIAL_FILES, PG_HBA};
 use postgres_ffi::relfile_utils::{INIT_FORKNUM, MAIN_FORKNUM};
-use postgres_ffi::TransactionId;
 use postgres_ffi::XLogFileName;
 use postgres_ffi::PG_TLI;
 use postgres_ffi::{BLCKSZ, RELSEG_SIZE, WAL_SEGMENT_SIZE};
@@ -255,8 +254,11 @@ where

        let lazy_slru_download = self.timeline.get_lazy_slru_download() && !self.full_backup;

+        let pgversion = self.timeline.pg_version;
+        let subdirs = dispatch_pgversion!(pgversion, &pgv::bindings::PGDATA_SUBDIRS[..]);
+
        // Create pgdata subdirs structure
-        for dir in PGDATA_SUBDIRS.iter() {
+        for dir in subdirs.iter() {
            let header = new_tar_header_dir(dir)?;
            self.ar
                .append(&header, &mut io::empty())
@@ -606,7 +608,7 @@ where
    //
    // Extract twophase state files
    //
-    async fn add_twophase_file(&mut self, xid: TransactionId) -> Result<(), BasebackupError> {
+    async fn add_twophase_file(&mut self, xid: u64) -> Result<(), BasebackupError> {
        let img = self
            .timeline
            .get_twophase_file(xid, self.lsn, self.ctx)
@@ -617,7 +619,11 @@ where
        buf.extend_from_slice(&img[..]);
        let crc = crc32c::crc32c(&img[..]);
        buf.put_u32_le(crc);
-        let path = format!("pg_twophase/{:>08X}", xid);
+        let path = if self.timeline.pg_version < 17 {
+            format!("pg_twophase/{:>08X}", xid)
+        } else {
+            format!("pg_twophase/{:>016X}", xid)
+        };
        let header = new_tar_header(&path, buf.len() as u64)?;
        self.ar
            .append(&header, &buf[..])
--- a/pageserver/src/config.rs
+++ b/pageserver/src/config.rs
@@ -281,7 +281,7 @@ impl PageServerConf {

        #[allow(clippy::manual_range_patterns)]
        match pg_version {
-            14 | 15 | 16 => Ok(path.join(format!("v{pg_version}"))),
+            14 | 15 | 16 | 17 => Ok(path.join(format!("v{pg_version}"))),
            _ => bail!("Unsupported postgres version: {}", pg_version),
        }
    }
--- a/pageserver/src/consumption_metrics.rs
+++ b/pageserver/src/consumption_metrics.rs
@@ -178,7 +178,7 @@ async fn collect_metrics(
                )
                .await;
                if let Err(e) = res {
-                    tracing::error!("failed to upload to S3: {e:#}");
+                    tracing::error!("failed to upload to remote storage: {e:#}");
                }
            }
        };
--- a/pageserver/src/import_datadir.rs
+++ b/pageserver/src/import_datadir.rs
@@ -580,9 +580,11 @@ async fn import_file(
        import_slru(modification, slru, file_path, reader, len, ctx).await?;
        debug!("imported multixact members slru");
    } else if file_path.starts_with("pg_twophase") {
-        let xid = u32::from_str_radix(file_name.as_ref(), 16)?;
-
        let bytes = read_all_bytes(reader).await?;
+
+        // In PostgreSQL v17, this is a 64-bit FullTransactionid. In previous versions,
+        // it's a 32-bit TransactionId, which fits in u64 anyway.
+        let xid = u64::from_str_radix(file_name.as_ref(), 16)?;
        modification
            .put_twophase_file(xid, Bytes::copy_from_slice(&bytes[..]), ctx)
            .await?;
--- a/pageserver/src/metrics.rs
+++ b/pageserver/src/metrics.rs
@@ -9,7 +9,7 @@ use metrics::{
 use once_cell::sync::Lazy;
 use pageserver_api::shard::TenantShardId;
 use strum::{EnumCount, VariantNames};
-use strum_macros::{EnumVariantNames, IntoStaticStr};
+use strum_macros::{IntoStaticStr, VariantNames};
 use tracing::warn;
 use utils::id::TimelineId;

@@ -27,7 +27,7 @@ const CRITICAL_OP_BUCKETS: &[f64] = &[
 ];

 // Metrics collected on operations on the storage repository.
-#[derive(Debug, EnumVariantNames, IntoStaticStr)]
+#[derive(Debug, VariantNames, IntoStaticStr)]
 #[strum(serialize_all = "kebab_case")]
 pub(crate) enum StorageTimeOperation {
    #[strum(serialize = "layer flush")]
@@ -1777,7 +1777,7 @@ pub(crate) static SECONDARY_MODE: Lazy<SecondaryModeMetrics> = Lazy::new(|| {
    .expect("failed to define a metric"),
    upload_heatmap_duration: register_histogram!(
        "pageserver_secondary_upload_heatmap_duration",
-        "Time to build and upload a heatmap, including any waiting inside the S3 client"
+        "Time to build and upload a heatmap, including any waiting inside the remote storage client"
    )
    .expect("failed to define a metric"),
    download_heatmap: register_int_counter!(
--- a/pageserver/src/page_service.rs
+++ b/pageserver/src/page_service.rs
@@ -1199,7 +1199,6 @@ impl PageServerHandler {
    }
 }

-#[async_trait::async_trait]
 impl<IO> postgres_backend::Handler<IO> for PageServerHandler
 where
    IO: AsyncRead + AsyncWrite + Send + Sync + Unpin,
--- a/pageserver/src/pgdatadir_mapping.rs
+++ b/pageserver/src/pgdatadir_mapping.rs
@@ -633,7 +633,7 @@ impl Timeline {

    pub(crate) async fn get_twophase_file(
        &self,
-        xid: TransactionId,
+        xid: u64,
        lsn: Lsn,
        ctx: &RequestContext,
    ) -> Result<Bytes, PageReconstructError> {
@@ -646,11 +646,19 @@ impl Timeline {
        &self,
        lsn: Lsn,
        ctx: &RequestContext,
-    ) -> Result<HashSet<TransactionId>, PageReconstructError> {
+    ) -> Result<HashSet<u64>, PageReconstructError> {
        // fetch directory entry
        let buf = self.get(TWOPHASEDIR_KEY, lsn, ctx).await?;

-        Ok(TwoPhaseDirectory::des(&buf)?.xids)
+        if self.pg_version >= 17 {
+            Ok(TwoPhaseDirectoryV17::des(&buf)?.xids)
+        } else {
+            Ok(TwoPhaseDirectory::des(&buf)?
+                .xids
+                .iter()
+                .map(|x| u64::from(*x))
+                .collect())
+        }
    }

    pub(crate) async fn get_control_file(
@@ -902,9 +910,13 @@ impl Timeline {

        // Then pg_twophase
        result.add_key(TWOPHASEDIR_KEY);
-        let buf = self.get(TWOPHASEDIR_KEY, lsn, ctx).await?;
-        let twophase_dir = TwoPhaseDirectory::des(&buf)?;
-        let mut xids: Vec<TransactionId> = twophase_dir.xids.iter().cloned().collect();
+
+        let mut xids: Vec<u64> = self
+            .list_twophase_files(lsn, ctx)
+            .await?
+            .iter()
+            .cloned()
+            .collect();
        xids.sort_unstable();
        for xid in xids {
            result.add_key(twophase_file_key(xid));
@@ -1127,9 +1139,15 @@ impl<'a> DatadirModification<'a> {
        // Create AuxFilesDirectory
        self.init_aux_dir()?;

-        let buf = TwoPhaseDirectory::ser(&TwoPhaseDirectory {
-            xids: HashSet::new(),
-        })?;
+        let buf = if self.tline.pg_version >= 17 {
+            TwoPhaseDirectoryV17::ser(&TwoPhaseDirectoryV17 {
+                xids: HashSet::new(),
+            })
+        } else {
+            TwoPhaseDirectory::ser(&TwoPhaseDirectory {
+                xids: HashSet::new(),
+            })
+        }?;
        self.pending_directory_entries
            .push((DirectoryKind::TwoPhase, 0));
        self.put(TWOPHASEDIR_KEY, Value::Image(buf.into()));
@@ -1205,6 +1223,13 @@ impl<'a> DatadirModification<'a> {
        img: Bytes,
    ) -> anyhow::Result<()> {
        anyhow::ensure!(rel.relnode != 0, RelationError::InvalidRelnode);
+        let key = rel_block_to_key(rel, blknum);
+        if !key.is_valid_key_on_write_path() {
+            anyhow::bail!(
+                "the request contains data not supported by pageserver at {}",
+                key
+            );
+        }
        self.put(rel_block_to_key(rel, blknum), Value::Image(img));
        Ok(())
    }
@@ -1216,14 +1241,34 @@ impl<'a> DatadirModification<'a> {
        blknum: BlockNumber,
        img: Bytes,
    ) -> anyhow::Result<()> {
-        self.put(slru_block_to_key(kind, segno, blknum), Value::Image(img));
+        let key = slru_block_to_key(kind, segno, blknum);
+        if !key.is_valid_key_on_write_path() {
+            anyhow::bail!(
+                "the request contains data not supported by pageserver at {}",
+                key
+            );
+        }
+        self.put(key, Value::Image(img));
        Ok(())
    }

-    pub(crate) fn put_rel_page_image_zero(&mut self, rel: RelTag, blknum: BlockNumber) {
-        self.pending_zero_data_pages
-            .insert(rel_block_to_key(rel, blknum).to_compact());
+    pub(crate) fn put_rel_page_image_zero(
+        &mut self,
+        rel: RelTag,
+        blknum: BlockNumber,
+    ) -> anyhow::Result<()> {
+        anyhow::ensure!(rel.relnode != 0, RelationError::InvalidRelnode);
+        let key = rel_block_to_key(rel, blknum);
+        if !key.is_valid_key_on_write_path() {
+            anyhow::bail!(
+                "the request contains data not supported by pageserver: {} @ {}",
+                key,
+                self.lsn
+            );
+        }
+        self.pending_zero_data_pages.insert(key.to_compact());
        self.pending_bytes += ZERO_PAGE.len();
+        Ok(())
    }

    pub(crate) fn put_slru_page_image_zero(
@@ -1231,10 +1276,18 @@ impl<'a> DatadirModification<'a> {
        kind: SlruKind,
        segno: u32,
        blknum: BlockNumber,
-    ) {
-        self.pending_zero_data_pages
-            .insert(slru_block_to_key(kind, segno, blknum).to_compact());
+    ) -> anyhow::Result<()> {
+        let key = slru_block_to_key(kind, segno, blknum);
+        if !key.is_valid_key_on_write_path() {
+            anyhow::bail!(
+                "the request contains data not supported by pageserver: {} @ {}",
+                key,
+                self.lsn
+            );
+        }
+        self.pending_zero_data_pages.insert(key.to_compact());
        self.pending_bytes += ZERO_PAGE.len();
+        Ok(())
    }

    /// Call this at the end of each WAL record.
@@ -1286,22 +1339,31 @@ impl<'a> DatadirModification<'a> {

    pub async fn put_twophase_file(
        &mut self,
-        xid: TransactionId,
+        xid: u64,
        img: Bytes,
        ctx: &RequestContext,
    ) -> anyhow::Result<()> {
        // Add it to the directory entry
-        let buf = self.get(TWOPHASEDIR_KEY, ctx).await?;
-        let mut dir = TwoPhaseDirectory::des(&buf)?;
-        if !dir.xids.insert(xid) {
-            anyhow::bail!("twophase file for xid {} already exists", xid);
-        }
-        self.pending_directory_entries
-            .push((DirectoryKind::TwoPhase, dir.xids.len()));
-        self.put(
-            TWOPHASEDIR_KEY,
-            Value::Image(Bytes::from(TwoPhaseDirectory::ser(&dir)?)),
-        );
+        let dirbuf = self.get(TWOPHASEDIR_KEY, ctx).await?;
+        let newdirbuf = if self.tline.pg_version >= 17 {
+            let mut dir = TwoPhaseDirectoryV17::des(&dirbuf)?;
+            if !dir.xids.insert(xid) {
+                anyhow::bail!("twophase file for xid {} already exists", xid);
+            }
+            self.pending_directory_entries
+                .push((DirectoryKind::TwoPhase, dir.xids.len()));
+            Bytes::from(TwoPhaseDirectoryV17::ser(&dir)?)
+        } else {
+            let xid = xid as u32;
+            let mut dir = TwoPhaseDirectory::des(&dirbuf)?;
+            if !dir.xids.insert(xid) {
+                anyhow::bail!("twophase file for xid {} already exists", xid);
+            }
+            self.pending_directory_entries
+                .push((DirectoryKind::TwoPhase, dir.xids.len()));
+            Bytes::from(TwoPhaseDirectory::ser(&dir)?)
+        };
+        self.put(TWOPHASEDIR_KEY, Value::Image(newdirbuf));

        self.put(twophase_file_key(xid), Value::Image(img));
        Ok(())
@@ -1604,22 +1666,32 @@ impl<'a> DatadirModification<'a> {
    /// This method is used for marking truncated SLRU files
    pub async fn drop_twophase_file(
        &mut self,
-        xid: TransactionId,
+        xid: u64,
        ctx: &RequestContext,
    ) -> anyhow::Result<()> {
        // Remove it from the directory entry
        let buf = self.get(TWOPHASEDIR_KEY, ctx).await?;
-        let mut dir = TwoPhaseDirectory::des(&buf)?;
+        let newdirbuf = if self.tline.pg_version >= 17 {
+            let mut dir = TwoPhaseDirectoryV17::des(&buf)?;

-        if !dir.xids.remove(&xid) {
-            warn!("twophase file for xid {} does not exist", xid);
-        }
-        self.pending_directory_entries
-            .push((DirectoryKind::TwoPhase, dir.xids.len()));
-        self.put(
-            TWOPHASEDIR_KEY,
-            Value::Image(Bytes::from(TwoPhaseDirectory::ser(&dir)?)),
-        );
+            if !dir.xids.remove(&xid) {
+                warn!("twophase file for xid {} does not exist", xid);
+            }
+            self.pending_directory_entries
+                .push((DirectoryKind::TwoPhase, dir.xids.len()));
+            Bytes::from(TwoPhaseDirectoryV17::ser(&dir)?)
+        } else {
+            let xid: u32 = u32::try_from(xid)?;
+            let mut dir = TwoPhaseDirectory::des(&buf)?;
+
+            if !dir.xids.remove(&xid) {
+                warn!("twophase file for xid {} does not exist", xid);
+            }
+            self.pending_directory_entries
+                .push((DirectoryKind::TwoPhase, dir.xids.len()));
+            Bytes::from(TwoPhaseDirectory::ser(&dir)?)
+        };
+        self.put(TWOPHASEDIR_KEY, Value::Image(newdirbuf));

        // Delete it
        self.delete(twophase_key_range(xid));
@@ -2089,11 +2161,21 @@ struct DbDirectory {
    dbdirs: HashMap<(Oid, Oid), bool>,
 }

+// The format of TwoPhaseDirectory changed in PostgreSQL v17, because the filenames of
+// pg_twophase files was expanded from 32-bit XIDs to 64-bit XIDs.  Previously, the files
+// were named like "pg_twophase/000002E5", now they're like
+// "pg_twophsae/0000000A000002E4".
+
 #[derive(Debug, Serialize, Deserialize)]
 struct TwoPhaseDirectory {
    xids: HashSet<TransactionId>,
 }

+#[derive(Debug, Serialize, Deserialize)]
+struct TwoPhaseDirectoryV17 {
+    xids: HashSet<u64>,
+}
+
 #[derive(Debug, Serialize, Deserialize, Default)]
 struct RelDirectory {
    // Set of relations that exist. (relfilenode, forknum)
--- a/pageserver/src/tenant.rs
+++ b/pageserver/src/tenant.rs
@@ -4164,9 +4164,18 @@ pub(crate) mod harness {
            let records_neon = records.iter().all(|r| apply_neon::can_apply_in_neon(&r.1));
            if records_neon {
                // For Neon wal records, we can decode without spawning postgres, so do so.
-                let base_img = base_img.expect("Neon WAL redo requires base image").1;
-                let mut page = BytesMut::new();
-                page.extend_from_slice(&base_img);
+                let mut page = match (base_img, records.first()) {
+                    (Some((_lsn, img)), _) => {
+                        let mut page = BytesMut::new();
+                        page.extend_from_slice(&img);
+                        page
+                    }
+                    (_, Some((_lsn, rec))) if rec.will_init() => BytesMut::new(),
+                    _ => {
+                        panic!("Neon WAL redo requires base image or will init record");
+                    }
+                };
+
                for (record_lsn, record) in records {
                    apply_neon::apply_in_neon(&record, record_lsn, key, &mut page)?;
                }
@@ -7091,13 +7100,13 @@ mod tests {
            vec![
                // Image layer at GC horizon
                PersistentLayerKey {
-                    key_range: Key::MIN..Key::NON_L0_MAX,
+                    key_range: Key::MIN..Key::MAX,
                    lsn_range: Lsn(0x30)..Lsn(0x31),
                    is_delta: false
                },
-                // The delta layer covers the full range (with the layer key hack to avoid being recognized as L0)
+                // The delta layer below the horizon
                PersistentLayerKey {
-                    key_range: Key::MIN..Key::NON_L0_MAX,
+                    key_range: get_key(3)..get_key(4),
                    lsn_range: Lsn(0x30)..Lsn(0x48),
                    is_delta: true
                },
@@ -8470,4 +8479,135 @@ mod tests {

        Ok(())
    }
+
+    // Regression test for https://github.com/neondatabase/neon/issues/9012
+    // Create an image arrangement where we have to read at different LSN ranges
+    // from a delta layer. This is achieved by overlapping an image layer on top of
+    // a delta layer. Like so:
+    //
+    //     A      B
+    // +----------------+ -> delta_layer
+    // |                |                           ^ lsn
+    // |       =========|-> nested_image_layer      |
+    // |       C        |                           |
+    // +----------------+                           |
+    // ======== -> baseline_image_layer             +-------> key
+    //
+    //
+    // When querying the key range [A, B) we need to read at different LSN ranges
+    // for [A, C) and [C, B). This test checks that the described edge case is handled correctly.
+    #[tokio::test]
+    async fn test_vectored_read_with_nested_image_layer() -> anyhow::Result<()> {
+        let harness = TenantHarness::create("test_vectored_read_with_nested_image_layer").await?;
+        let (tenant, ctx) = harness.load().await;
+
+        let will_init_keys = [2, 6];
+        fn get_key(id: u32) -> Key {
+            let mut key = Key::from_hex("110000000033333333444444445500000000").unwrap();
+            key.field6 = id;
+            key
+        }
+
+        let mut expected_key_values = HashMap::new();
+
+        let baseline_image_layer_lsn = Lsn(0x10);
+        let mut baseline_img_layer = Vec::new();
+        for i in 0..5 {
+            let key = get_key(i);
+            let value = format!("value {i}@{baseline_image_layer_lsn}");
+
+            let removed = expected_key_values.insert(key, value.clone());
+            assert!(removed.is_none());
+
+            baseline_img_layer.push((key, Bytes::from(value)));
+        }
+
+        let nested_image_layer_lsn = Lsn(0x50);
+        let mut nested_img_layer = Vec::new();
+        for i in 5..10 {
+            let key = get_key(i);
+            let value = format!("value {i}@{nested_image_layer_lsn}");
+
+            let removed = expected_key_values.insert(key, value.clone());
+            assert!(removed.is_none());
+
+            nested_img_layer.push((key, Bytes::from(value)));
+        }
+
+        let mut delta_layer_spec = Vec::default();
+        let delta_layer_start_lsn = Lsn(0x20);
+        let mut delta_layer_end_lsn = delta_layer_start_lsn;
+
+        for i in 0..10 {
+            let key = get_key(i);
+            let key_in_nested = nested_img_layer
+                .iter()
+                .any(|(key_with_img, _)| *key_with_img == key);
+            let lsn = {
+                if key_in_nested {
+                    Lsn(nested_image_layer_lsn.0 + 0x10)
+                } else {
+                    delta_layer_start_lsn
+                }
+            };
+
+            let will_init = will_init_keys.contains(&i);
+            if will_init {
+                delta_layer_spec.push((key, lsn, Value::WalRecord(NeonWalRecord::wal_init())));
+
+                expected_key_values.insert(key, "".to_string());
+            } else {
+                let delta = format!("@{lsn}");
+                delta_layer_spec.push((
+                    key,
+                    lsn,
+                    Value::WalRecord(NeonWalRecord::wal_append(&delta)),
+                ));
+
+                expected_key_values
+                    .get_mut(&key)
+                    .expect("An image exists for each key")
+                    .push_str(delta.as_str());
+            }
+            delta_layer_end_lsn = std::cmp::max(delta_layer_start_lsn, lsn);
+        }
+
+        delta_layer_end_lsn = Lsn(delta_layer_end_lsn.0 + 1);
+
+        assert!(
+            nested_image_layer_lsn > delta_layer_start_lsn
+                && nested_image_layer_lsn < delta_layer_end_lsn
+        );
+
+        let tline = tenant
+            .create_test_timeline_with_layers(
+                TIMELINE_ID,
+                baseline_image_layer_lsn,
+                DEFAULT_PG_VERSION,
+                &ctx,
+                vec![DeltaLayerTestDesc::new_with_inferred_key_range(
+                    delta_layer_start_lsn..delta_layer_end_lsn,
+                    delta_layer_spec,
+                )], // delta layers
+                vec![
+                    (baseline_image_layer_lsn, baseline_img_layer),
+                    (nested_image_layer_lsn, nested_img_layer),
+                ], // image layers
+                delta_layer_end_lsn,
+            )
+            .await?;
+
+        let keyspace = KeySpace::single(get_key(0)..get_key(10));
+        let results = tline
+            .get_vectored(keyspace, delta_layer_end_lsn, &ctx)
+            .await
+            .expect("No vectored errors");
+        for (key, res) in results {
+            let value = res.expect("No key errors");
+            let expected_value = expected_key_values.remove(&key).expect("No unknown keys");
+            assert_eq!(value, Bytes::from(expected_value));
+        }
+
+        Ok(())
+    }
 }
--- a/pageserver/src/tenant/config.rs
+++ b/pageserver/src/tenant/config.rs
@@ -452,7 +452,8 @@ impl TryFrom<toml_edit::Item> for TenantConfOpt {
                    .map_err(|e| anyhow::anyhow!("{}: {}", e.path(), e.inner().message()));
            }
            toml_edit::Item::Table(table) => {
-                let deserializer = toml_edit::de::Deserializer::new(table.into());
+                let deserializer =
+                    toml_edit::de::Deserializer::from(toml_edit::DocumentMut::from(table));
                return serde_path_to_error::deserialize(deserializer)
                    .map_err(|e| anyhow::anyhow!("{}: {}", e.path(), e.inner().message()));
            }
--- a/pageserver/src/tenant/storage_layer.rs
+++ b/pageserver/src/tenant/storage_layer.rs
@@ -287,7 +287,7 @@ pub(crate) enum ReadableLayer {

 /// A partial description of a read to be done.
 #[derive(Debug, Clone)]
-struct ReadDesc {
+struct VisitLocation {
    /// An id used to resolve the readable layer within the fringe
    layer_id: LayerId,
    /// Lsn range for the read, used for selecting the next read
@@ -303,46 +303,442 @@ struct ReadDesc {
 /// a two layer indexing scheme.
 #[derive(Debug)]
 pub(crate) struct LayerFringe {
-    planned_reads_by_lsn: BinaryHeap<ReadDesc>,
-    layers: HashMap<LayerId, LayerKeyspace>,
+    visits_by_lsn_index: BinaryHeap<VisitLocation>,
+    layer_visits: HashMap<LayerId, LayerVisitBuilder>,
 }

 #[derive(Debug)]
-struct LayerKeyspace {
-    layer: ReadableLayer,
-    target_keyspace: KeySpaceRandomAccum,
+pub(crate) enum LayerVisitBuilder {
+    InMemoryLayer(InMemoryLayerVisitBuilder),
+    PersistentLayer(PersistentLayerVisitBuilder),
+}
+
+#[derive(Debug)]
+pub(crate) enum LayerVisit {
+    InMemoryLayer(InMemoryLayerVisit),
+    PersistentLayer(PersistentLayerVisit),
+}
+
+#[derive(Debug)]
+pub(crate) enum PersistentLayerVisitBuilder {
+    DeltaLayer(DeltaLayerVisitBuilder),
+    ImageLayer(ImageLayerVisitBuilder),
+}
+
+#[derive(Debug)]
+pub(crate) enum PersistentLayerVisit {
+    DeltaLayer(DeltaLayerVisit),
+    ImageLayer(ImageLayerVisit),
+}
+
+#[derive(Debug)]
+pub(crate) struct InMemoryLayerVisitBuilder {
+    /// Key space accumulator which will define which keys we are
+    /// interested in for this layer visit.
+    keyspace_accum: KeySpaceRandomAccum,
+    /// Ignore any keys with an LSN greater or equal
+    /// than the specified one.
+    lsn_ceil: Lsn,
+    /// Only consider keys at LSN greater or equal than the specified one
+    lsn_floor: Lsn,
+    // The in-memory layer to visit
+    layer: Arc<InMemoryLayer>,
+}
+
+#[derive(Debug)]
+pub(crate) struct InMemoryLayerVisit {
+    /// Key space of keys considered by the visit
+    keyspace: KeySpace,
+    /// Ignore any keys with an LSN greater or equal
+    /// than the specified one.
+    lsn_ceil: Lsn,
+    /// Only consider keys at LSN greater or equal than the specified one
+    lsn_floor: Lsn,
+}
+
+#[derive(Debug)]
+pub(crate) struct DeltaLayerVisitBuilder {
+    /// List of key spaces accumulators which will define what deltas are read.
+    /// Each accumulator has an associated LSN which specifies
+    /// the LSN floor for it (i.e. do not read below this LSN).
+    keyspace_accums: HashMap<Lsn, KeySpaceRandomAccum>,
+    /// Ignore any keys with an LSN greater or equal
+    /// than the specified one.
+    lsn_ceil: Lsn,
+    /// Handle for the layer to visit (guaranteed to be a delta layer)
+    layer: Layer,
+}
+
+#[derive(Debug)]
+pub(crate) struct DeltaLayerVisit {
+    /// List of key spaces considered during the visit.
+    /// Each keyspace has an associated LSN which specifies
+    /// the LSN floor for it (i.e. do not read below this LSN).
+    keyspaces: Vec<(Lsn, KeySpace)>,
+    /// Ignore any keys with an LSN greater or equal
+    /// than the specified one.
+    lsn_ceil: Lsn,
+}
+
+#[derive(Debug)]
+pub(crate) struct ImageLayerVisitBuilder {
+    /// Key space which defines which keys we are
+    /// interested in for this layer visit.
+    keyspace_accum: KeySpaceRandomAccum,
+    /// Handle for the layer to visit (guaranteed to be an image layer)
+    layer: Layer,
+    /// Only consider keys at LSN greater or equal than the specified one
+    lsn_floor: Lsn,
+}
+
+#[derive(Debug)]
+pub(crate) struct ImageLayerVisit {
+    /// Key space which defines which keys we are
+    /// interested in for this layer visit.
+    keyspace: KeySpace,
+    /// Only consider keys at LSN greater or equal than the specified one
+    lsn_floor: Lsn,
+}
+
+impl LayerVisitBuilder {
+    pub(crate) fn new(layer: ReadableLayer, keyspace: KeySpace, lsn_range: Range<Lsn>) -> Self {
+        match layer {
+            ReadableLayer::InMemoryLayer(in_mem) => {
+                Self::InMemoryLayer(InMemoryLayerVisitBuilder::new(in_mem, keyspace, lsn_range))
+            }
+            ReadableLayer::PersistentLayer(persistent) => Self::PersistentLayer(
+                PersistentLayerVisitBuilder::new(persistent, keyspace, lsn_range),
+            ),
+        }
+    }
+}
+
+impl InMemoryLayerVisitBuilder {
+    fn new(layer: Arc<InMemoryLayer>, keyspace: KeySpace, lsn_range: Range<Lsn>) -> Self {
+        assert_eq!(lsn_range.start, layer.get_lsn_range().start);
+
+        let mut keyspace_accum = KeySpaceRandomAccum::new();
+        keyspace_accum.add_keyspace(keyspace);
+
+        Self {
+            keyspace_accum,
+            lsn_floor: lsn_range.start,
+            lsn_ceil: lsn_range.end,
+            layer,
+        }
+    }
+}
+
+impl PersistentLayerVisitBuilder {
+    fn new(layer: Layer, keyspace: KeySpace, lsn_range: Range<Lsn>) -> Self {
+        let is_delta = layer.layer_desc().is_delta;
+        if is_delta {
+            Self::DeltaLayer(DeltaLayerVisitBuilder::new(layer, keyspace, lsn_range))
+        } else {
+            Self::ImageLayer(ImageLayerVisitBuilder::new(layer, keyspace, lsn_range))
+        }
+    }
+}
+
+impl DeltaLayerVisitBuilder {
+    fn new(layer: Layer, keyspace: KeySpace, lsn_range: Range<Lsn>) -> Self {
+        assert!(layer.layer_desc().is_delta);
+
+        let mut keyspace_accum = KeySpaceRandomAccum::new();
+        keyspace_accum.add_keyspace(keyspace);
+
+        Self {
+            keyspace_accums: HashMap::from([(lsn_range.start, keyspace_accum)]),
+            lsn_ceil: lsn_range.end,
+            layer,
+        }
+    }
+}
+
+impl ImageLayerVisitBuilder {
+    fn new(layer: Layer, keyspace: KeySpace, lsn_range: Range<Lsn>) -> Self {
+        assert!(!layer.layer_desc().is_delta);
+        assert_eq!(lsn_range.start, layer.layer_desc().lsn_range.start);
+
+        let mut keyspace_accum = KeySpaceRandomAccum::new();
+        keyspace_accum.add_keyspace(keyspace);
+
+        Self {
+            keyspace_accum,
+            lsn_floor: lsn_range.start,
+            layer,
+        }
+    }
+}
+
+pub(crate) trait LayerVisitBuilderUpdate {
+    type L;
+    type LV;
+
+    /// Extend an already planned layer visit to also include the keys
+    /// in the provided keyspace and LSN range.
+    fn update(&mut self, keyspace: KeySpace, lsn_range: Range<Lsn>);
+
+    /// Build the visit!
+    fn build(self) -> (Self::L, Self::LV);
+}
+
+impl LayerVisitBuilderUpdate for LayerVisitBuilder {
+    type L = ReadableLayer;
+    type LV = LayerVisit;
+
+    fn update(&mut self, keyspace: KeySpace, lsn_range: Range<Lsn>) {
+        match self {
+            LayerVisitBuilder::InMemoryLayer(v) => v.update(keyspace, lsn_range),
+            LayerVisitBuilder::PersistentLayer(v) => v.update(keyspace, lsn_range),
+        }
+    }
+
+    fn build(self) -> (Self::L, Self::LV) {
+        match self {
+            LayerVisitBuilder::InMemoryLayer(in_mem) => (
+                ReadableLayer::InMemoryLayer(in_mem.layer),
+                LayerVisit::InMemoryLayer(InMemoryLayerVisit {
+                    keyspace: in_mem.keyspace_accum.to_keyspace(),
+                    lsn_ceil: in_mem.lsn_ceil,
+                    lsn_floor: in_mem.lsn_floor,
+                }),
+            ),
+            LayerVisitBuilder::PersistentLayer(pers) => {
+                let (layer, visit) = pers.build();
+                (
+                    ReadableLayer::PersistentLayer(layer),
+                    LayerVisit::PersistentLayer(visit),
+                )
+            }
+        }
+    }
+}
+
+impl LayerVisitBuilderUpdate for PersistentLayerVisitBuilder {
+    type L = Layer;
+    type LV = PersistentLayerVisit;
+
+    fn update(&mut self, keyspace: KeySpace, lsn_range: Range<Lsn>) {
+        match self {
+            PersistentLayerVisitBuilder::DeltaLayer(v) => v.update(keyspace, lsn_range),
+            PersistentLayerVisitBuilder::ImageLayer(v) => v.update(keyspace, lsn_range),
+        }
+    }
+
+    fn build(self) -> (Self::L, Self::LV) {
+        match self {
+            PersistentLayerVisitBuilder::DeltaLayer(delta) => {
+                let (layer, visit) = delta.build();
+                (layer, PersistentLayerVisit::DeltaLayer(visit))
+            }
+            PersistentLayerVisitBuilder::ImageLayer(img) => {
+                let (layer, visit) = img.build();
+                (layer, PersistentLayerVisit::ImageLayer(visit))
+            }
+        }
+    }
+}
+
+impl LayerVisitBuilderUpdate for InMemoryLayerVisitBuilder {
+    type L = Arc<InMemoryLayer>;
+    type LV = InMemoryLayerVisit;
+
+    fn update(&mut self, keyspace: KeySpace, lsn_range: Range<Lsn>) {
+        // Note: I cannot think of any cases when this update should happen,
+        // since in memory layers span the entire key range.
+        assert_eq!(lsn_range.end, self.lsn_ceil);
+        assert_eq!(lsn_range.start, self.lsn_floor);
+        self.keyspace_accum.add_keyspace(keyspace);
+    }
+
+    fn build(self) -> (Self::L, Self::LV) {
+        (
+            self.layer,
+            InMemoryLayerVisit {
+                keyspace: self.keyspace_accum.to_keyspace(),
+                lsn_floor: self.lsn_floor,
+                lsn_ceil: self.lsn_ceil,
+            },
+        )
+    }
+}
+
+impl LayerVisitBuilderUpdate for DeltaLayerVisitBuilder {
+    type L = Layer;
+    type LV = DeltaLayerVisit;
+
+    fn update(&mut self, keyspace: KeySpace, lsn_range: Range<Lsn>) {
+        assert_eq!(lsn_range.end, self.lsn_ceil);
+        self.keyspace_accums
+            .entry(lsn_range.start)
+            .or_default()
+            .add_keyspace(keyspace);
+    }
+
+    fn build(self) -> (Self::L, Self::LV) {
+        use itertools::Itertools;
+
+        let keyspaces = self
+            .keyspace_accums
+            .into_iter()
+            .filter_map(|(lsn_floor, accum)| {
+                let keyspace = accum.to_keyspace();
+                if keyspace.is_empty() {
+                    None
+                } else {
+                    Some((lsn_floor, keyspace))
+                }
+            })
+            .sorted_by_key(|(_lsn_floor, keyspace)| keyspace.start().unwrap())
+            .collect::<Vec<(Lsn, KeySpace)>>();
+
+        if cfg!(debug_assertions) {
+            // Check that the keyspaces we are going to read from
+            // a layer are non-overlapping.
+            //
+            // The keyspaces provided to vectored read initially are non-overlapping.
+            // We may split keyspaces at each step and keyspaces resulting from a split
+            // are non-overlapping as well. One can prove that the property holds by
+            // induction.
+            let mut prev_end: Option<Key> = None;
+            for (_lsn_floor, crnt) in keyspaces.iter() {
+                if let Some(prev_end) = prev_end {
+                    debug_assert!(prev_end <= crnt.start().unwrap())
+                }
+
+                prev_end = Some(crnt.end().unwrap());
+            }
+        }
+
+        (
+            self.layer,
+            DeltaLayerVisit {
+                keyspaces,
+                lsn_ceil: self.lsn_ceil,
+            },
+        )
+    }
+}
+
+impl LayerVisitBuilderUpdate for ImageLayerVisitBuilder {
+    type L = Layer;
+    type LV = ImageLayerVisit;
+
+    fn update(&mut self, keyspace: KeySpace, lsn_range: Range<Lsn>) {
+        assert_eq!(lsn_range.start, self.lsn_floor);
+        self.keyspace_accum.add_keyspace(keyspace);
+    }
+
+    fn build(self) -> (Self::L, Self::LV) {
+        (
+            self.layer,
+            ImageLayerVisit {
+                keyspace: self.keyspace_accum.to_keyspace(),
+                lsn_floor: self.lsn_floor,
+            },
+        )
+    }
+}
+
+pub(crate) trait LayerVisitDetails {
+    /// Returns the key spaces planned for the visit
+    /// and their associated floor LSNs.
+    fn keyspaces(&self) -> Vec<(Lsn, KeySpace)>;
+}
+
+impl LayerVisitDetails for LayerVisit {
+    fn keyspaces(&self) -> Vec<(Lsn, KeySpace)> {
+        match self {
+            LayerVisit::PersistentLayer(pers) => pers.keyspaces(),
+            LayerVisit::InMemoryLayer(in_mem) => in_mem.keyspaces(),
+        }
+    }
+}
+
+impl LayerVisitDetails for PersistentLayerVisit {
+    fn keyspaces(&self) -> Vec<(Lsn, KeySpace)> {
+        match self {
+            PersistentLayerVisit::DeltaLayer(delta) => delta.keyspaces(),
+            PersistentLayerVisit::ImageLayer(image) => image.keyspaces(),
+        }
+    }
+}
+
+impl LayerVisitDetails for DeltaLayerVisit {
+    fn keyspaces(&self) -> Vec<(Lsn, KeySpace)> {
+        self.keyspaces.clone()
+    }
+}
+
+impl LayerVisitDetails for ImageLayerVisit {
+    fn keyspaces(&self) -> Vec<(Lsn, KeySpace)> {
+        vec![(self.lsn_floor, self.keyspace.clone())]
+    }
+}
+
+impl LayerVisitDetails for InMemoryLayerVisit {
+    fn keyspaces(&self) -> Vec<(Lsn, KeySpace)> {
+        vec![(self.lsn_floor, self.keyspace.clone())]
+    }
+}
+
+impl LayerVisit {
+    fn into_persistent_layer_visit(self) -> PersistentLayerVisit {
+        match self {
+            LayerVisit::PersistentLayer(visit) => visit,
+            LayerVisit::InMemoryLayer(visit) => {
+                panic!("Invalid attempt to cast to PersistentLayerVisit: {visit:?}");
+            }
+        }
+    }
+
+    fn into_in_memory_layer_visit(self) -> InMemoryLayerVisit {
+        match self {
+            LayerVisit::InMemoryLayer(visit) => visit,
+            LayerVisit::PersistentLayer(visit) => {
+                panic!("Invalid attempt to cast to InMemoryLayerVisit: {visit:?}");
+            }
+        }
+    }
+}
+
+impl PersistentLayerVisit {
+    fn into_delta_layer_visit(self) -> DeltaLayerVisit {
+        match self {
+            PersistentLayerVisit::DeltaLayer(visit) => visit,
+            PersistentLayerVisit::ImageLayer(visit) => {
+                panic!("Invalid attempt to cast to DeltaLayerVisit: {visit:?}");
+            }
+        }
+    }
+
+    fn into_image_layer_visit(self) -> ImageLayerVisit {
+        match self {
+            PersistentLayerVisit::ImageLayer(visit) => visit,
+            PersistentLayerVisit::DeltaLayer(visit) => {
+                panic!("Invalid attempt to cast to ImageLayerVisit: {visit:?}");
+            }
+        }
+    }
 }

 impl LayerFringe {
    pub(crate) fn new() -> Self {
        LayerFringe {
-            planned_reads_by_lsn: BinaryHeap::new(),
-            layers: HashMap::new(),
+            visits_by_lsn_index: BinaryHeap::new(),
+            layer_visits: HashMap::new(),
        }
    }

-    pub(crate) fn next_layer(&mut self) -> Option<(ReadableLayer, KeySpace, Range<Lsn>)> {
-        let read_desc = match self.planned_reads_by_lsn.pop() {
+    pub(crate) fn next_layer(&mut self) -> Option<(ReadableLayer, LayerVisit)> {
+        let read_desc = match self.visits_by_lsn_index.pop() {
            Some(desc) => desc,
            None => return None,
        };

-        let removed = self.layers.remove_entry(&read_desc.layer_id);
-
-        match removed {
-            Some((
-                _,
-                LayerKeyspace {
-                    layer,
-                    mut target_keyspace,
-                },
-            )) => Some((
-                layer,
-                target_keyspace.consume_keyspace(),
-                read_desc.lsn_range,
-            )),
-            None => unreachable!("fringe internals are always consistent"),
-        }
+        let removed = self.layer_visits.remove(&read_desc.layer_id)?;
+        Some(removed.build())
    }

    pub(crate) fn update(
@@ -352,22 +748,18 @@ impl LayerFringe {
        lsn_range: Range<Lsn>,
    ) {
        let layer_id = layer.id();
-        let entry = self.layers.entry(layer_id.clone());
+        let entry = self.layer_visits.entry(layer_id.clone());
        match entry {
            Entry::Occupied(mut entry) => {
-                entry.get_mut().target_keyspace.add_keyspace(keyspace);
+                entry.get_mut().update(keyspace, lsn_range);
            }
            Entry::Vacant(entry) => {
-                self.planned_reads_by_lsn.push(ReadDesc {
-                    lsn_range,
+                self.visits_by_lsn_index.push(VisitLocation {
+                    lsn_range: lsn_range.clone(),
                    layer_id: layer_id.clone(),
                });
-                let mut accum = KeySpaceRandomAccum::new();
-                accum.add_keyspace(keyspace);
-                entry.insert(LayerKeyspace {
-                    layer,
-                    target_keyspace: accum,
-                });
+
+                entry.insert(LayerVisitBuilder::new(layer, keyspace, lsn_range));
            }
        }
    }
@@ -379,7 +771,7 @@ impl Default for LayerFringe {
    }
 }

-impl Ord for ReadDesc {
+impl Ord for VisitLocation {
    fn cmp(&self, other: &Self) -> Ordering {
        let ord = self.lsn_range.end.cmp(&other.lsn_range.end);
        if ord == std::cmp::Ordering::Equal {
@@ -390,19 +782,19 @@ impl Ord for ReadDesc {
    }
 }

-impl PartialOrd for ReadDesc {
+impl PartialOrd for VisitLocation {
    fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
        Some(self.cmp(other))
    }
 }

-impl PartialEq for ReadDesc {
+impl PartialEq for VisitLocation {
    fn eq(&self, other: &Self) -> bool {
        self.lsn_range == other.lsn_range
    }
 }

-impl Eq for ReadDesc {}
+impl Eq for VisitLocation {}

 impl ReadableLayer {
    pub(crate) fn id(&self) -> LayerId {
@@ -414,20 +806,27 @@ impl ReadableLayer {

    pub(crate) async fn get_values_reconstruct_data(
        &self,
-        keyspace: KeySpace,
-        lsn_range: Range<Lsn>,
+        visit: LayerVisit,
        reconstruct_state: &mut ValuesReconstructState,
        ctx: &RequestContext,
    ) -> Result<(), GetVectoredError> {
        match self {
            ReadableLayer::PersistentLayer(layer) => {
                layer
-                    .get_values_reconstruct_data(keyspace, lsn_range, reconstruct_state, ctx)
+                    .get_values_reconstruct_data(
+                        visit.into_persistent_layer_visit(),
+                        reconstruct_state,
+                        ctx,
+                    )
                    .await
            }
            ReadableLayer::InMemoryLayer(layer) => {
                layer
-                    .get_values_reconstruct_data(keyspace, lsn_range.end, reconstruct_state, ctx)
+                    .get_values_reconstruct_data(
+                        visit.into_in_memory_layer_visit(),
+                        reconstruct_state,
+                        ctx,
+                    )
                    .await
            }
        }
--- a/pageserver/src/tenant/storage_layer/delta_layer.rs
+++ b/pageserver/src/tenant/storage_layer/delta_layer.rs
@@ -75,7 +75,7 @@ use utils::{
    lsn::Lsn,
 };

-use super::{AsLayerDesc, LayerName, PersistentLayerDesc, ValuesReconstructState};
+use super::{AsLayerDesc, DeltaLayerVisit, LayerName, PersistentLayerDesc, ValuesReconstructState};

 ///
 /// Header stored in the beginning of the file
@@ -841,8 +841,7 @@ impl DeltaLayerInner {
    // can be further optimised to visit the index only once.
    pub(super) async fn get_values_reconstruct_data(
        &self,
-        keyspace: KeySpace,
-        lsn_range: Range<Lsn>,
+        visit: DeltaLayerVisit,
        reconstruct_state: &mut ValuesReconstructState,
        ctx: &RequestContext,
    ) -> Result<(), GetVectoredError> {
@@ -863,8 +862,8 @@ impl DeltaLayerInner {
        let data_end_offset = self.index_start_offset();

        let reads = Self::plan_reads(
-            &keyspace,
-            lsn_range.clone(),
+            &visit.keyspaces,
+            visit.lsn_ceil,
            data_end_offset,
            index_reader,
            planner,
@@ -877,14 +876,16 @@ impl DeltaLayerInner {
        self.do_reads_and_update_state(reads, reconstruct_state, ctx)
            .await;

-        reconstruct_state.on_lsn_advanced(&keyspace, lsn_range.start);
+        for (lsn_floor, keyspace) in visit.keyspaces {
+            reconstruct_state.on_lsn_advanced(&keyspace, lsn_floor);
+        }

        Ok(())
    }

    async fn plan_reads<Reader>(
-        keyspace: &KeySpace,
-        lsn_range: Range<Lsn>,
+        keyspaces: &Vec<(Lsn, KeySpace)>,
+        lsn_ceil: Lsn,
        data_end_offset: u64,
        index_reader: DiskBtreeReader<Reader, DELTA_KEY_SIZE>,
        mut planner: VectoredReadPlanner,
@@ -898,48 +899,52 @@ impl DeltaLayerInner {
            .page_content_kind(PageContentKind::DeltaLayerBtreeNode)
            .build();

-        for range in keyspace.ranges.iter() {
-            let mut range_end_handled = false;
+        for (lsn_floor, keyspace) in keyspaces {
+            let lsn_range = *lsn_floor..lsn_ceil;

-            let start_key = DeltaKey::from_key_lsn(&range.start, lsn_range.start);
-            let index_stream = index_reader.clone().into_stream(&start_key.0, &ctx);
-            let mut index_stream = std::pin::pin!(index_stream);
+            for range in keyspace.ranges.iter() {
+                let mut range_end_handled = false;

-            while let Some(index_entry) = index_stream.next().await {
-                let (raw_key, value) = index_entry?;
-                let key = Key::from_slice(&raw_key[..KEY_SIZE]);
-                let lsn = DeltaKey::extract_lsn_from_buf(&raw_key);
-                let blob_ref = BlobRef(value);
+                let start_key = DeltaKey::from_key_lsn(&range.start, lsn_range.start);
+                let index_stream = index_reader.clone().into_stream(&start_key.0, &ctx);
+                let mut index_stream = std::pin::pin!(index_stream);

-                // Lsns are not monotonically increasing across keys, so we don't assert on them.
-                assert!(key >= range.start);
+                while let Some(index_entry) = index_stream.next().await {
+                    let (raw_key, value) = index_entry?;
+                    let key = Key::from_slice(&raw_key[..KEY_SIZE]);
+                    let lsn = DeltaKey::extract_lsn_from_buf(&raw_key);
+                    let blob_ref = BlobRef(value);

-                let outside_lsn_range = !lsn_range.contains(&lsn);
-                let below_cached_lsn = reconstruct_state.get_cached_lsn(&key) >= Some(lsn);
+                    // Lsns are not monotonically increasing across keys, so we don't assert on them.
+                    assert!(key >= range.start);

-                let flag = {
-                    if outside_lsn_range || below_cached_lsn {
-                        BlobFlag::Ignore
-                    } else if blob_ref.will_init() {
-                        BlobFlag::ReplaceAll
+                    let outside_lsn_range = !lsn_range.contains(&lsn);
+                    let below_cached_lsn = reconstruct_state.get_cached_lsn(&key) >= Some(lsn);
+
+                    let flag = {
+                        if outside_lsn_range || below_cached_lsn {
+                            BlobFlag::Ignore
+                        } else if blob_ref.will_init() {
+                            BlobFlag::ReplaceAll
+                        } else {
+                            // Usual path: add blob to the read
+                            BlobFlag::None
+                        }
+                    };
+
+                    if key >= range.end || (key.next() == range.end && lsn >= lsn_range.end) {
+                        planner.handle_range_end(blob_ref.pos());
+                        range_end_handled = true;
+                        break;
                    } else {
-                        // Usual path: add blob to the read
-                        BlobFlag::None
+                        planner.handle(key, lsn, blob_ref.pos(), flag);
                    }
-                };
-
-                if key >= range.end || (key.next() == range.end && lsn >= lsn_range.end) {
-                    planner.handle_range_end(blob_ref.pos());
-                    range_end_handled = true;
-                    break;
-                } else {
-                    planner.handle(key, lsn, blob_ref.pos(), flag);
                }
-            }

-            if !range_end_handled {
-                tracing::debug!("Handling range end fallback at {}", data_end_offset);
-                planner.handle_range_end(data_end_offset);
+                if !range_end_handled {
+                    tracing::debug!("Handling range end fallback at {}", data_end_offset);
+                    planner.handle_range_end(data_end_offset);
+                }
            }
        }

@@ -1641,8 +1646,8 @@ pub(crate) mod test {

        // Plan and validate
        let vectored_reads = DeltaLayerInner::plan_reads(
-            &keyspace,
-            lsn_range.clone(),
+            &vec![(lsn_range.start, keyspace.clone())],
+            lsn_range.end,
            disk_offset,
            reader,
            planner,
@@ -1895,8 +1900,8 @@ pub(crate) mod test {
            let data_end_offset = inner.index_start_blk as u64 * PAGE_SZ as u64;

            let vectored_reads = DeltaLayerInner::plan_reads(
-                &keyspace,
-                entries_meta.lsn_range.clone(),
+                &vec![(entries_meta.lsn_range.start, keyspace)],
+                entries_meta.lsn_range.end,
                data_end_offset,
                index_reader,
                planner,
--- a/pageserver/src/tenant/storage_layer/image_layer.rs
+++ b/pageserver/src/tenant/storage_layer/image_layer.rs
@@ -38,7 +38,7 @@ use crate::tenant::timeline::GetVectoredError;
 use crate::tenant::vectored_blob_io::{
    BlobFlag, StreamingVectoredReadPlanner, VectoredBlobReader, VectoredRead, VectoredReadPlanner,
 };
-use crate::tenant::{PageReconstructError, Timeline};
+use crate::tenant::PageReconstructError;
 use crate::virtual_file::owned_buffers_io::io_buf_ext::IoBufExt;
 use crate::virtual_file::{self, VirtualFile};
 use crate::{IMAGE_FILE_MAGIC, STORAGE_FORMAT_VERSION, TEMP_FILE_SUFFIX};
@@ -58,7 +58,6 @@ use std::io::SeekFrom;
 use std::ops::Range;
 use std::os::unix::prelude::FileExt;
 use std::str::FromStr;
-use std::sync::Arc;
 use tokio::sync::OnceCell;
 use tokio_stream::StreamExt;
 use tracing::*;
@@ -70,9 +69,7 @@ use utils::{
 };

 use super::layer_name::ImageLayerName;
-use super::{
-    AsLayerDesc, Layer, LayerName, PersistentLayerDesc, ResidentLayer, ValuesReconstructState,
-};
+use super::{AsLayerDesc, ImageLayerVisit, LayerName, PersistentLayerDesc, ValuesReconstructState};

 ///
 /// Header stored in the beginning of the file
@@ -438,12 +435,12 @@ impl ImageLayerInner {
    // the reconstruct state with whatever is found.
    pub(super) async fn get_values_reconstruct_data(
        &self,
-        keyspace: KeySpace,
+        visit: ImageLayerVisit,
        reconstruct_state: &mut ValuesReconstructState,
        ctx: &RequestContext,
    ) -> Result<(), GetVectoredError> {
        let reads = self
-            .plan_reads(keyspace, None, ctx)
+            .plan_reads(visit.keyspace, None, ctx)
            .await
            .map_err(GetVectoredError::Other)?;

@@ -800,10 +797,9 @@ impl ImageLayerWriterInner {
    ///
    async fn finish(
        self,
-        timeline: &Arc<Timeline>,
        ctx: &RequestContext,
        end_key: Option<Key>,
-    ) -> anyhow::Result<ResidentLayer> {
+    ) -> anyhow::Result<(PersistentLayerDesc, Utf8PathBuf)> {
        let index_start_blk =
            ((self.blob_writer.size() + PAGE_SZ as u64 - 1) / PAGE_SZ as u64) as u32;

@@ -879,12 +875,9 @@ impl ImageLayerWriterInner {
        // fsync the file
        file.sync_all().await?;

-        // FIXME: why not carry the virtualfile here, it supports renaming?
-        let layer = Layer::finish_creating(self.conf, timeline, desc, &self.path)?;
+        trace!("created image layer {}", self.path);

-        info!("created image layer {}", layer.local_path());
-
-        Ok(layer)
+        Ok((desc, self.path))
    }
 }

@@ -963,24 +956,18 @@ impl ImageLayerWriter {
    ///
    pub(crate) async fn finish(
        mut self,
-        timeline: &Arc<Timeline>,
        ctx: &RequestContext,
-    ) -> anyhow::Result<super::ResidentLayer> {
-        self.inner.take().unwrap().finish(timeline, ctx, None).await
+    ) -> anyhow::Result<(PersistentLayerDesc, Utf8PathBuf)> {
+        self.inner.take().unwrap().finish(ctx, None).await
    }

    /// Finish writing the image layer with an end key, used in [`super::split_writer::SplitImageLayerWriter`]. The end key determines the end of the image layer's covered range and is exclusive.
    pub(super) async fn finish_with_end_key(
        mut self,
-        timeline: &Arc<Timeline>,
        end_key: Key,
        ctx: &RequestContext,
-    ) -> anyhow::Result<super::ResidentLayer> {
-        self.inner
-            .take()
-            .unwrap()
-            .finish(timeline, ctx, Some(end_key))
-            .await
+    ) -> anyhow::Result<(PersistentLayerDesc, Utf8PathBuf)> {
+        self.inner.take().unwrap().finish(ctx, Some(end_key)).await
    }
 }

@@ -1084,7 +1071,7 @@ mod test {
        tenant::{
            config::TenantConf,
            harness::{TenantHarness, TIMELINE_ID},
-            storage_layer::ResidentLayer,
+            storage_layer::{Layer, ResidentLayer},
            vectored_blob_io::StreamingVectoredReadPlanner,
            Tenant, Timeline,
        },
@@ -1155,7 +1142,8 @@ mod test {

                key = key.next();
            }
-            writer.finish(&timeline, &ctx).await.unwrap()
+            let (desc, path) = writer.finish(&ctx).await.unwrap();
+            Layer::finish_creating(tenant.conf, &timeline, desc, &path).unwrap()
        };
        let original_size = resident.metadata().file_size;

@@ -1217,7 +1205,9 @@ mod test {
                .await
                .unwrap();
            let replacement = if wrote_keys > 0 {
-                Some(filtered_writer.finish(&timeline, &ctx).await.unwrap())
+                let (desc, path) = filtered_writer.finish(&ctx).await.unwrap();
+                let resident = Layer::finish_creating(tenant.conf, &timeline, desc, &path).unwrap();
+                Some(resident)
            } else {
                None
            };
@@ -1290,7 +1280,8 @@ mod test {
        for (key, img) in images {
            writer.put_image(key, img, ctx).await?;
        }
-        let img_layer = writer.finish(tline, ctx).await?;
+        let (desc, path) = writer.finish(ctx).await?;
+        let img_layer = Layer::finish_creating(tenant.conf, tline, desc, &path)?;

        Ok::<_, anyhow::Error>(img_layer)
    }
--- a/pageserver/src/tenant/storage_layer/inmemory_layer.rs
+++ b/pageserver/src/tenant/storage_layer/inmemory_layer.rs
@@ -17,7 +17,6 @@ use anyhow::{anyhow, Context, Result};
 use bytes::Bytes;
 use camino::Utf8PathBuf;
 use pageserver_api::key::CompactKey;
-use pageserver_api::keyspace::KeySpace;
 use pageserver_api::models::InMemoryLayerInfo;
 use pageserver_api::shard::TenantShardId;
 use std::collections::{BTreeMap, HashMap};
@@ -36,7 +35,8 @@ use std::sync::atomic::{AtomicU64, AtomicUsize};
 use tokio::sync::RwLock;

 use super::{
-    DeltaLayerWriter, PersistentLayerDesc, ValueReconstructSituation, ValuesReconstructState,
+    DeltaLayerWriter, InMemoryLayerVisit, PersistentLayerDesc, ValueReconstructSituation,
+    ValuesReconstructState,
 };

 pub(crate) mod vectored_dio_read;
@@ -416,11 +416,14 @@ impl InMemoryLayer {
    // If the key is cached, go no further than the cached Lsn.
    pub(crate) async fn get_values_reconstruct_data(
        &self,
-        keyspace: KeySpace,
-        end_lsn: Lsn,
+        visit: InMemoryLayerVisit,
        reconstruct_state: &mut ValuesReconstructState,
        ctx: &RequestContext,
    ) -> Result<(), GetVectoredError> {
+        let InMemoryLayerVisit {
+            keyspace, lsn_ceil, ..
+        } = visit;
+
        let ctx = RequestContextBuilder::extend(ctx)
            .page_content_kind(PageContentKind::InMemoryLayer)
            .build();
@@ -440,8 +443,8 @@ impl InMemoryLayer {
            {
                let key = Key::from_compact(*key);
                let lsn_range = match reconstruct_state.get_cached_lsn(&key) {
-                    Some(cached_lsn) => (cached_lsn + 1)..end_lsn,
-                    None => self.start_lsn..end_lsn,
+                    Some(cached_lsn) => (cached_lsn + 1)..lsn_ceil,
+                    None => self.start_lsn..lsn_ceil,
                };

                let slice = vec_map.slice_range(lsn_range);
--- a/pageserver/src/tenant/storage_layer/layer.rs
+++ b/pageserver/src/tenant/storage_layer/layer.rs
@@ -1,9 +1,7 @@
 use anyhow::Context;
 use camino::{Utf8Path, Utf8PathBuf};
-use pageserver_api::keyspace::KeySpace;
 use pageserver_api::models::HistoricLayerInfo;
 use pageserver_api::shard::{ShardIdentity, ShardIndex, TenantShardId};
-use std::ops::Range;
 use std::sync::atomic::{AtomicBool, AtomicUsize, Ordering};
 use std::sync::{Arc, Weak};
 use std::time::{Duration, SystemTime};
@@ -23,7 +21,7 @@ use super::delta_layer::{self, DeltaEntry};
 use super::image_layer::{self};
 use super::{
    AsLayerDesc, ImageLayerWriter, LayerAccessStats, LayerAccessStatsReset, LayerName,
-    LayerVisibilityHint, PersistentLayerDesc, ValuesReconstructState,
+    LayerVisibilityHint, PersistentLayerDesc, PersistentLayerVisit, ValuesReconstructState,
 };

 use utils::generation::Generation;
@@ -303,8 +301,7 @@ impl Layer {

    pub(crate) async fn get_values_reconstruct_data(
        &self,
-        keyspace: KeySpace,
-        lsn_range: Range<Lsn>,
+        visit: PersistentLayerVisit,
        reconstruct_data: &mut ValuesReconstructState,
        ctx: &RequestContext,
    ) -> Result<(), GetVectoredError> {
@@ -322,7 +319,7 @@ impl Layer {
        self.record_access(ctx);

        layer
-            .get_values_reconstruct_data(keyspace, lsn_range, reconstruct_data, &self.0, ctx)
+            .get_values_reconstruct_data(visit, reconstruct_data, &self.0, ctx)
            .instrument(tracing::debug_span!("get_values_reconstruct_data", layer=%self))
            .await
            .map_err(|err| match err {
@@ -1741,8 +1738,7 @@ impl DownloadedLayer {

    async fn get_values_reconstruct_data(
        &self,
-        keyspace: KeySpace,
-        lsn_range: Range<Lsn>,
+        visit: PersistentLayerVisit,
        reconstruct_data: &mut ValuesReconstructState,
        owner: &Arc<LayerInner>,
        ctx: &RequestContext,
@@ -1755,11 +1751,11 @@ impl DownloadedLayer {
            .map_err(GetVectoredError::Other)?
        {
            Delta(d) => {
-                d.get_values_reconstruct_data(keyspace, lsn_range, reconstruct_data, ctx)
+                d.get_values_reconstruct_data(visit.into_delta_layer_visit(), reconstruct_data, ctx)
                    .await
            }
            Image(i) => {
-                i.get_values_reconstruct_data(keyspace, reconstruct_data, ctx)
+                i.get_values_reconstruct_data(visit.into_image_layer_visit(), reconstruct_data, ctx)
                    .await
            }
        }
--- a/pageserver/src/tenant/storage_layer/layer/tests.rs
+++ b/pageserver/src/tenant/storage_layer/layer/tests.rs
@@ -1,6 +1,6 @@
 use std::time::UNIX_EPOCH;

-use pageserver_api::key::CONTROLFILE_KEY;
+use pageserver_api::{key::CONTROLFILE_KEY, keyspace::KeySpace};
 use tokio::task::JoinSet;
 use utils::{
    completion::{self, Completion},
@@ -9,7 +9,10 @@ use utils::{

 use super::failpoints::{Failpoint, FailpointKind};
 use super::*;
-use crate::{context::DownloadBehavior, tenant::storage_layer::LayerVisibilityHint};
+use crate::{
+    context::DownloadBehavior,
+    tenant::storage_layer::{ImageLayerVisit, LayerVisibilityHint},
+};
 use crate::{task_mgr::TaskKind, tenant::harness::TenantHarness};

 /// Used in tests to advance a future to wanted await point, and not futher.
@@ -56,13 +59,14 @@ async fn smoke_test() {

    let img_before = {
        let mut data = ValuesReconstructState::default();
+        let visit = ImageLayerVisit {
+            keyspace: controlfile_keyspace.clone(),
+            lsn_floor: Lsn(0x10),
+        };
+        let visit = PersistentLayerVisit::ImageLayer(visit);
+
        layer
-            .get_values_reconstruct_data(
-                controlfile_keyspace.clone(),
-                Lsn(0x10)..Lsn(0x11),
-                &mut data,
-                &ctx,
-            )
+            .get_values_reconstruct_data(visit, &mut data, &ctx)
            .await
            .unwrap();
        data.keys
@@ -88,13 +92,14 @@ async fn smoke_test() {
    // on accesses when the layer is evicted, it will automatically be downloaded.
    let img_after = {
        let mut data = ValuesReconstructState::default();
+        let visit = ImageLayerVisit {
+            keyspace: controlfile_keyspace.clone(),
+            lsn_floor: Lsn(0x10),
+        };
+        let visit = PersistentLayerVisit::ImageLayer(visit);
+
        layer
-            .get_values_reconstruct_data(
-                controlfile_keyspace.clone(),
-                Lsn(0x10)..Lsn(0x11),
-                &mut data,
-                &ctx,
-            )
+            .get_values_reconstruct_data(visit, &mut data, &ctx)
            .instrument(download_span.clone())
            .await
            .unwrap();
--- a/pageserver/src/tenant/storage_layer/split_writer.rs
+++ b/pageserver/src/tenant/storage_layer/split_writer.rs
@@ -121,11 +121,11 @@ impl SplitImageLayerWriter {
                self.generated_layers
                    .push(SplitWriterResult::Discarded(layer_key));
            } else {
-                self.generated_layers.push(SplitWriterResult::Produced(
-                    prev_image_writer
-                        .finish_with_end_key(tline, key, ctx)
-                        .await?,
-                ));
+                let (desc, path) = prev_image_writer.finish_with_end_key(key, ctx).await?;
+
+                let layer = Layer::finish_creating(self.conf, tline, desc, &path)?;
+                self.generated_layers
+                    .push(SplitWriterResult::Produced(layer));
            }
        }
        self.inner.put_image(key, img, ctx).await
@@ -170,9 +170,9 @@ impl SplitImageLayerWriter {
        if discard(&layer_key).await {
            generated_layers.push(SplitWriterResult::Discarded(layer_key));
        } else {
-            generated_layers.push(SplitWriterResult::Produced(
-                inner.finish_with_end_key(tline, end_key, ctx).await?,
-            ));
+            let (desc, path) = inner.finish_with_end_key(end_key, ctx).await?;
+            let layer = Layer::finish_creating(self.conf, tline, desc, &path)?;
+            generated_layers.push(SplitWriterResult::Produced(layer));
        }
        Ok(generated_layers)
    }
@@ -188,7 +188,7 @@ impl SplitImageLayerWriter {
            .await
    }

-    /// When split writer fails, the caller should call this function and handle partially generated layers.
+    /// This function will be deprecated with #8841.
    pub(crate) fn take(self) -> anyhow::Result<(Vec<SplitWriterResult>, ImageLayerWriter)> {
        Ok((self.generated_layers, self.inner))
    }
@@ -204,7 +204,7 @@ impl SplitImageLayerWriter {
 /// will split them into multiple files based on size.
 #[must_use]
 pub struct SplitDeltaLayerWriter {
-    inner: DeltaLayerWriter,
+    inner: Option<(Key, DeltaLayerWriter)>,
    target_layer_size: u64,
    generated_layers: Vec<SplitWriterResult>,
    conf: &'static PageServerConf,
@@ -212,7 +212,6 @@ pub struct SplitDeltaLayerWriter {
    tenant_shard_id: TenantShardId,
    lsn_range: Range<Lsn>,
    last_key_written: Key,
-    start_key: Key,
 }

 impl SplitDeltaLayerWriter {
@@ -220,29 +219,18 @@ impl SplitDeltaLayerWriter {
        conf: &'static PageServerConf,
        timeline_id: TimelineId,
        tenant_shard_id: TenantShardId,
-        start_key: Key,
        lsn_range: Range<Lsn>,
        target_layer_size: u64,
-        ctx: &RequestContext,
    ) -> anyhow::Result<Self> {
        Ok(Self {
            target_layer_size,
-            inner: DeltaLayerWriter::new(
-                conf,
-                timeline_id,
-                tenant_shard_id,
-                start_key,
-                lsn_range.clone(),
-                ctx,
-            )
-            .await?,
+            inner: None,
            generated_layers: Vec::new(),
            conf,
            timeline_id,
            tenant_shard_id,
            lsn_range,
            last_key_written: Key::MIN,
-            start_key,
        })
    }

@@ -265,9 +253,26 @@ impl SplitDeltaLayerWriter {
        //
        // Also, keep all updates of a single key in a single file. TODO: split them using the legacy compaction
        // strategy. https://github.com/neondatabase/neon/issues/8837
+
+        if self.inner.is_none() {
+            self.inner = Some((
+                key,
+                DeltaLayerWriter::new(
+                    self.conf,
+                    self.timeline_id,
+                    self.tenant_shard_id,
+                    key,
+                    self.lsn_range.clone(),
+                    ctx,
+                )
+                .await?,
+            ));
+        }
+        let (_, inner) = self.inner.as_mut().unwrap();
+
        let addition_size_estimation = KEY_SIZE as u64 + 8 /* LSN u64 size */ + 80 /* value size estimation */;
-        if self.inner.num_keys() >= 1
-            && self.inner.estimated_size() + addition_size_estimation >= self.target_layer_size
+        if inner.num_keys() >= 1
+            && inner.estimated_size() + addition_size_estimation >= self.target_layer_size
        {
            if key != self.last_key_written {
                let next_delta_writer = DeltaLayerWriter::new(
@@ -279,13 +284,13 @@ impl SplitDeltaLayerWriter {
                    ctx,
                )
                .await?;
-                let prev_delta_writer = std::mem::replace(&mut self.inner, next_delta_writer);
+                let (start_key, prev_delta_writer) =
+                    std::mem::replace(&mut self.inner, Some((key, next_delta_writer))).unwrap();
                let layer_key = PersistentLayerKey {
-                    key_range: self.start_key..key,
+                    key_range: start_key..key,
                    lsn_range: self.lsn_range.clone(),
                    is_delta: true,
                };
-                self.start_key = key;
                if discard(&layer_key).await {
                    drop(prev_delta_writer);
                    self.generated_layers
@@ -296,17 +301,18 @@ impl SplitDeltaLayerWriter {
                    self.generated_layers
                        .push(SplitWriterResult::Produced(delta_layer));
                }
-            } else if self.inner.estimated_size() >= S3_UPLOAD_LIMIT {
+            } else if inner.estimated_size() >= S3_UPLOAD_LIMIT {
                // We have to produce a very large file b/c a key is updated too often.
                anyhow::bail!(
                    "a single key is updated too often: key={}, estimated_size={}, and the layer file cannot be produced",
                    key,
-                    self.inner.estimated_size()
+                    inner.estimated_size()
                );
            }
        }
        self.last_key_written = key;
-        self.inner.put_value(key, lsn, val, ctx).await
+        let (_, inner) = self.inner.as_mut().unwrap();
+        inner.put_value(key, lsn, val, ctx).await
    }

    pub async fn put_value(
@@ -325,7 +331,6 @@ impl SplitDeltaLayerWriter {
        self,
        tline: &Arc<Timeline>,
        ctx: &RequestContext,
-        end_key: Key,
        discard: D,
    ) -> anyhow::Result<Vec<SplitWriterResult>>
    where
@@ -337,11 +342,15 @@ impl SplitDeltaLayerWriter {
            inner,
            ..
        } = self;
+        let Some((start_key, inner)) = inner else {
+            return Ok(generated_layers);
+        };
        if inner.num_keys() == 0 {
            return Ok(generated_layers);
        }
+        let end_key = self.last_key_written.next();
        let layer_key = PersistentLayerKey {
-            key_range: self.start_key..end_key,
+            key_range: start_key..end_key,
            lsn_range: self.lsn_range.clone(),
            is_delta: true,
        };
@@ -360,15 +369,14 @@ impl SplitDeltaLayerWriter {
        self,
        tline: &Arc<Timeline>,
        ctx: &RequestContext,
-        end_key: Key,
    ) -> anyhow::Result<Vec<SplitWriterResult>> {
-        self.finish_with_discard_fn(tline, ctx, end_key, |_| async { false })
+        self.finish_with_discard_fn(tline, ctx, |_| async { false })
            .await
    }

-    /// When split writer fails, the caller should call this function and handle partially generated layers.
-    pub(crate) fn take(self) -> anyhow::Result<(Vec<SplitWriterResult>, DeltaLayerWriter)> {
-        Ok((self.generated_layers, self.inner))
+    /// This function will be deprecated with #8841.
+    pub(crate) fn take(self) -> anyhow::Result<(Vec<SplitWriterResult>, Option<DeltaLayerWriter>)> {
+        Ok((self.generated_layers, self.inner.map(|x| x.1)))
    }
 }

@@ -432,10 +440,8 @@ mod tests {
            tenant.conf,
            tline.timeline_id,
            tenant.tenant_shard_id,
-            get_key(0),
            Lsn(0x18)..Lsn(0x20),
            4 * 1024 * 1024,
-            &ctx,
        )
        .await
        .unwrap();
@@ -460,11 +466,22 @@ mod tests {
            )
            .await
            .unwrap();
-        let layers = delta_writer
-            .finish(&tline, &ctx, get_key(10))
-            .await
-            .unwrap();
+        let layers = delta_writer.finish(&tline, &ctx).await.unwrap();
        assert_eq!(layers.len(), 1);
+        assert_eq!(
+            layers
+                .into_iter()
+                .next()
+                .unwrap()
+                .into_resident_layer()
+                .layer_desc()
+                .key(),
+            PersistentLayerKey {
+                key_range: get_key(0)..get_key(1),
+                lsn_range: Lsn(0x18)..Lsn(0x20),
+                is_delta: true
+            }
+        );
    }

    #[tokio::test]
@@ -501,10 +518,8 @@ mod tests {
            tenant.conf,
            tline.timeline_id,
            tenant.tenant_shard_id,
-            get_key(0),
            Lsn(0x18)..Lsn(0x20),
            4 * 1024 * 1024,
-            &ctx,
        )
        .await
        .unwrap();
@@ -533,10 +548,7 @@ mod tests {
            .finish(&tline, &ctx, get_key(N as u32))
            .await
            .unwrap();
-        let delta_layers = delta_writer
-            .finish(&tline, &ctx, get_key(N as u32))
-            .await
-            .unwrap();
+        let delta_layers = delta_writer.finish(&tline, &ctx).await.unwrap();
        if discard {
            for layer in image_layers {
                layer.into_discarded_layer();
@@ -555,6 +567,14 @@ mod tests {
                .collect_vec();
            assert_eq!(image_layers.len(), N / 512 + 1);
            assert_eq!(delta_layers.len(), N / 512 + 1);
+            assert_eq!(
+                delta_layers.first().unwrap().layer_desc().key_range.start,
+                get_key(0)
+            );
+            assert_eq!(
+                delta_layers.last().unwrap().layer_desc().key_range.end,
+                get_key(N as u32)
+            );
            for idx in 0..image_layers.len() {
                assert_ne!(image_layers[idx].layer_desc().key_range.start, Key::MIN);
                assert_ne!(image_layers[idx].layer_desc().key_range.end, Key::MAX);
@@ -602,10 +622,8 @@ mod tests {
            tenant.conf,
            tline.timeline_id,
            tenant.tenant_shard_id,
-            get_key(0),
            Lsn(0x18)..Lsn(0x20),
            4 * 1024,
-            &ctx,
        )
        .await
        .unwrap();
@@ -644,11 +662,35 @@ mod tests {
            )
            .await
            .unwrap();
-        let layers = delta_writer
-            .finish(&tline, &ctx, get_key(10))
-            .await
-            .unwrap();
+        let layers = delta_writer.finish(&tline, &ctx).await.unwrap();
        assert_eq!(layers.len(), 2);
+        let mut layers_iter = layers.into_iter();
+        assert_eq!(
+            layers_iter
+                .next()
+                .unwrap()
+                .into_resident_layer()
+                .layer_desc()
+                .key(),
+            PersistentLayerKey {
+                key_range: get_key(0)..get_key(1),
+                lsn_range: Lsn(0x18)..Lsn(0x20),
+                is_delta: true
+            }
+        );
+        assert_eq!(
+            layers_iter
+                .next()
+                .unwrap()
+                .into_resident_layer()
+                .layer_desc()
+                .key(),
+            PersistentLayerKey {
+                key_range: get_key(1)..get_key(2),
+                lsn_range: Lsn(0x18)..Lsn(0x20),
+                is_delta: true
+            }
+        );
    }

    #[tokio::test]
@@ -668,10 +710,8 @@ mod tests {
            tenant.conf,
            tline.timeline_id,
            tenant.tenant_shard_id,
-            get_key(0),
            Lsn(0x10)..Lsn(N as u64 * 16 + 0x10),
            4 * 1024 * 1024,
-            &ctx,
        )
        .await
        .unwrap();
@@ -689,10 +729,20 @@ mod tests {
                .await
                .unwrap();
        }
-        let delta_layers = delta_writer
-            .finish(&tline, &ctx, get_key(N as u32))
-            .await
-            .unwrap();
+        let delta_layers = delta_writer.finish(&tline, &ctx).await.unwrap();
        assert_eq!(delta_layers.len(), 1);
+        let delta_layer = delta_layers
+            .into_iter()
+            .next()
+            .unwrap()
+            .into_resident_layer();
+        assert_eq!(
+            delta_layer.layer_desc().key(),
+            PersistentLayerKey {
+                key_range: get_key(0)..get_key(1),
+                lsn_range: Lsn(0x10)..Lsn(N as u64 * 16 + 0x10),
+                is_delta: true
+            }
+        );
    }
 }
--- a/pageserver/src/tenant/timeline.rs
+++ b/pageserver/src/tenant/timeline.rs
@@ -136,7 +136,8 @@ use self::logical_size::LogicalSize;
 use self::walreceiver::{WalReceiver, WalReceiverConf};

 use super::{
-    config::TenantConf, storage_layer::inmemory_layer, storage_layer::LayerVisibilityHint,
+    config::TenantConf,
+    storage_layer::{inmemory_layer, LayerVisibilityHint, LayerVisitDetails},
    upload_queue::NotInitialized,
 };
 use super::{debug_assert_current_span_has_tenant_and_timeline_id, AttachedTenantConf};
@@ -3153,12 +3154,12 @@ impl Timeline {
    async fn get_vectored_reconstruct_data_timeline(
        timeline: &Timeline,
        keyspace: KeySpace,
-        mut cont_lsn: Lsn,
+        cont_lsn: Lsn,
        reconstruct_state: &mut ValuesReconstructState,
        cancel: &CancellationToken,
        ctx: &RequestContext,
    ) -> Result<TimelineVisitOutcome, GetVectoredError> {
-        let mut unmapped_keyspace = keyspace.clone();
+        let mut unmapped_keyspaces = vec![(cont_lsn, keyspace.clone())];
        let mut fringe = LayerFringe::new();

        let mut completed_keyspace = KeySpace::default();
@@ -3171,84 +3172,86 @@ impl Timeline {

            let (keys_done_last_step, keys_with_image_coverage) =
                reconstruct_state.consume_done_keys();
-            unmapped_keyspace.remove_overlapping_with(&keys_done_last_step);
+
+            // Update state that is external to the loop.
            completed_keyspace.merge(&keys_done_last_step);
-            if let Some(keys_with_image_coverage) = keys_with_image_coverage {
-                unmapped_keyspace
-                    .remove_overlapping_with(&KeySpace::single(keys_with_image_coverage.clone()));
+            if let Some(keys_with_image_coverage) = keys_with_image_coverage.clone() {
                image_covered_keyspace.add_range(keys_with_image_coverage);
            }

-            // Do not descent any further if the last layer we visited
-            // completed all keys in the keyspace it inspected. This is not
-            // required for correctness, but avoids visiting extra layers
-            // which turns out to be a perf bottleneck in some cases.
-            if !unmapped_keyspace.is_empty() {
-                let guard = timeline.layers.read().await;
-                let layers = guard.layer_map()?;
-
-                let in_memory_layer = layers.find_in_memory_layer(|l| {
-                    let start_lsn = l.get_lsn_range().start;
-                    cont_lsn > start_lsn
-                });
-
-                match in_memory_layer {
-                    Some(l) => {
-                        let lsn_range = l.get_lsn_range().start..cont_lsn;
-                        fringe.update(
-                            ReadableLayer::InMemoryLayer(l),
-                            unmapped_keyspace.clone(),
-                            lsn_range,
-                        );
-                    }
-                    None => {
-                        for range in unmapped_keyspace.ranges.iter() {
-                            let results = layers.range_search(range.clone(), cont_lsn);
-
-                            results
-                                .found
-                                .into_iter()
-                                .map(|(SearchResult { layer, lsn_floor }, keyspace_accum)| {
-                                    (
-                                        ReadableLayer::PersistentLayer(guard.get_from_desc(&layer)),
-                                        keyspace_accum.to_keyspace(),
-                                        lsn_floor..cont_lsn,
-                                    )
-                                })
-                                .for_each(|(layer, keyspace, lsn_range)| {
-                                    fringe.update(layer, keyspace, lsn_range)
-                                });
-                        }
-                    }
+            for (cont_lsn, unmapped_keyspace) in unmapped_keyspaces.iter_mut() {
+                // Remove any completed keys from the currently inspected keyspace.
+                unmapped_keyspace.remove_overlapping_with(&keys_done_last_step);
+                if let Some(keys_with_image_coverage) = keys_with_image_coverage.clone() {
+                    unmapped_keyspace
+                        .remove_overlapping_with(&KeySpace::single(keys_with_image_coverage));
                }

-                // It's safe to drop the layer map lock after planning the next round of reads.
-                // The fringe keeps readable handles for the layers which are safe to read even
-                // if layers were compacted or flushed.
-                //
-                // The more interesting consideration is: "Why is the read algorithm still correct
-                // if the layer map changes while it is operating?". Doing a vectored read on a
-                // timeline boils down to pushing an imaginary lsn boundary downwards for each range
-                // covered by the read. The layer map tells us how to move the lsn downwards for a
-                // range at *a particular point in time*. It is fine for the answer to be different
-                // at two different time points.
-                drop(guard);
+                // Do not descent any further if the last layer we visited
+                // completed all keys in the keyspace it inspected. This is not
+                // required for correctness, but avoids visiting extra layers
+                // which turns out to be a perf bottleneck in some cases.
+                if !unmapped_keyspace.is_empty() {
+                    let guard = timeline.layers.read().await;
+                    let layers = guard.layer_map()?;
+
+                    let in_memory_layer = layers.find_in_memory_layer(|l| {
+                        let start_lsn = l.get_lsn_range().start;
+                        *cont_lsn > start_lsn
+                    });
+
+                    match in_memory_layer {
+                        Some(l) => {
+                            let lsn_range = l.get_lsn_range().start..*cont_lsn;
+                            fringe.update(
+                                ReadableLayer::InMemoryLayer(l),
+                                unmapped_keyspace.clone(),
+                                lsn_range,
+                            );
+                        }
+                        None => {
+                            for range in unmapped_keyspace.ranges.iter() {
+                                let results = layers.range_search(range.clone(), *cont_lsn);
+
+                                results
+                                    .found
+                                    .into_iter()
+                                    .map(|(SearchResult { layer, lsn_floor }, keyspace_accum)| {
+                                        (
+                                            ReadableLayer::PersistentLayer(
+                                                guard.get_from_desc(&layer),
+                                            ),
+                                            keyspace_accum.to_keyspace(),
+                                            lsn_floor..*cont_lsn,
+                                        )
+                                    })
+                                    .for_each(|(layer, keyspace, lsn_range)| {
+                                        fringe.update(layer, keyspace, lsn_range)
+                                    });
+                            }
+                        }
+                    }
+
+                    // It's safe to drop the layer map lock after planning the next round of reads.
+                    // The fringe keeps readable handles for the layers which are safe to read even
+                    // if layers were compacted or flushed.
+                    //
+                    // The more interesting consideration is: "Why is the read algorithm still correct
+                    // if the layer map changes while it is operating?". Doing a vectored read on a
+                    // timeline boils down to pushing an imaginary lsn boundary downwards for each range
+                    // covered by the read. The layer map tells us how to move the lsn downwards for a
+                    // range at *a particular point in time*. It is fine for the answer to be different
+                    // at two different time points.
+                    drop(guard);
+                }
            }

-            if let Some((layer_to_read, keyspace_to_read, lsn_range)) = fringe.next_layer() {
-                let next_cont_lsn = lsn_range.start;
+            if let Some((layer_to_read, layer_visit)) = fringe.next_layer() {
+                unmapped_keyspaces = layer_visit.keyspaces();
                layer_to_read
-                    .get_values_reconstruct_data(
-                        keyspace_to_read.clone(),
-                        lsn_range,
-                        reconstruct_state,
-                        ctx,
-                    )
+                    .get_values_reconstruct_data(layer_visit, reconstruct_state, ctx)
                    .await?;

-                unmapped_keyspace = keyspace_to_read;
-                cont_lsn = next_cont_lsn;
-
                reconstruct_state.on_layer_visited(&layer_to_read);
            } else {
                break;
@@ -4013,7 +4016,8 @@ impl Timeline {
        if wrote_keys {
            // Normal path: we have written some data into the new image layer for this
            // partition, so flush it to disk.
-            let image_layer = image_layer_writer.finish(self, ctx).await?;
+            let (desc, path) = image_layer_writer.finish(ctx).await?;
+            let image_layer = Layer::finish_creating(self.conf, self, desc, &path)?;
            Ok(ImageLayerCreationOutcome {
                image: Some(image_layer),
                next_start_key: img_range.end,
@@ -4101,7 +4105,8 @@ impl Timeline {
        if wrote_any_image {
            // Normal path: we have written some data into the new image layer for this
            // partition, so flush it to disk.
-            let image_layer = image_layer_writer.finish(self, ctx).await?;
+            let (desc, path) = image_layer_writer.finish(ctx).await?;
+            let image_layer = Layer::finish_creating(self.conf, self, desc, &path)?;
            Ok(ImageLayerCreationOutcome {
                image: Some(image_layer),
                next_start_key: img_range.end,
@@ -5403,7 +5408,8 @@ impl Timeline {
        for (key, img) in images {
            image_layer_writer.put_image(key, img, ctx).await?;
        }
-        let image_layer = image_layer_writer.finish(self, ctx).await?;
+        let (desc, path) = image_layer_writer.finish(ctx).await?;
+        let image_layer = Layer::finish_creating(self.conf, self, desc, &path)?;

        {
            let mut guard = self.layers.write().await;
@@ -5499,19 +5505,24 @@ impl Timeline {
        lsn: Lsn,
        ctx: &RequestContext,
    ) -> anyhow::Result<Vec<(Key, Bytes)>> {
+        use super::storage_layer::{LayerVisitBuilder, LayerVisitBuilderUpdate};
+
        let mut all_data = Vec::new();
        let guard = self.layers.read().await;
        for layer in guard.layer_map()?.iter_historic_layers() {
            if !layer.is_delta() && layer.image_layer_lsn() == lsn {
                let layer = guard.get_from_desc(&layer);
                let mut reconstruct_data = ValuesReconstructState::default();
+
+                let (layer, visit) = LayerVisitBuilder::new(
+                    ReadableLayer::PersistentLayer(layer),
+                    KeySpace::single(Key::MIN..Key::MAX),
+                    lsn..Lsn(lsn.0 + 1),
+                )
+                .build();
+
                layer
-                    .get_values_reconstruct_data(
-                        KeySpace::single(Key::MIN..Key::MAX),
-                        lsn..Lsn(lsn.0 + 1),
-                        &mut reconstruct_data,
-                        ctx,
-                    )
+                    .get_values_reconstruct_data(visit, &mut reconstruct_data, ctx)
                    .await?;
                for (k, v) in reconstruct_data.keys {
                    all_data.push((k, v?.img.unwrap().1));
--- a/pageserver/src/tenant/timeline/compaction.rs
+++ b/pageserver/src/tenant/timeline/compaction.rs
@@ -563,10 +563,12 @@ impl Timeline {
                .await?;

            if keys_written > 0 {
-                let new_layer = image_layer_writer
-                    .finish(self, ctx)
+                let (desc, path) = image_layer_writer
+                    .finish(ctx)
                    .await
                    .map_err(CompactionError::Other)?;
+                let new_layer = Layer::finish_creating(self.conf, self, desc, &path)
+                    .map_err(CompactionError::Other)?;
                tracing::info!(layer=%new_layer, "Rewrote layer, {} -> {} bytes",
                    layer.metadata().file_size,
                    new_layer.metadata().file_size);
@@ -1809,7 +1811,6 @@ impl Timeline {
            .unwrap();
        // We don't want any of the produced layers to cover the full key range (i.e., MIN..MAX) b/c it will then be recognized
        // as an L0 layer.
-        let hack_end_key = Key::NON_L0_MAX;
        let mut delta_layers = Vec::new();
        let mut image_layers = Vec::new();
        let mut downloaded_layers = Vec::new();
@@ -1855,10 +1856,8 @@ impl Timeline {
            self.conf,
            self.timeline_id,
            self.tenant_shard_id,
-            Key::MIN,
            lowest_retain_lsn..end_lsn,
            self.get_compaction_target_size(),
-            ctx,
        )
        .await?;

@@ -1965,7 +1964,7 @@ impl Timeline {
        let produced_image_layers = if let Some(writer) = image_layer_writer {
            if !dry_run {
                writer
-                    .finish_with_discard_fn(self, ctx, hack_end_key, discard)
+                    .finish_with_discard_fn(self, ctx, Key::MAX, discard)
                    .await?
            } else {
                let (layers, _) = writer.take()?;
@@ -1978,7 +1977,7 @@ impl Timeline {

        let produced_delta_layers = if !dry_run {
            delta_layer_writer
-                .finish_with_discard_fn(self, ctx, hack_end_key, discard)
+                .finish_with_discard_fn(self, ctx, discard)
                .await?
        } else {
            let (layers, _) = delta_layer_writer.take()?;
--- a/pageserver/src/walingest.rs
+++ b/pageserver/src/walingest.rs
@@ -25,9 +25,7 @@ use std::time::Duration;
 use std::time::SystemTime;

 use pageserver_api::shard::ShardIdentity;
-use postgres_ffi::v14::nonrelfile_utils::clogpage_precedes;
-use postgres_ffi::v14::nonrelfile_utils::slru_may_delete_clogsegment;
-use postgres_ffi::TimestampTz;
+use postgres_ffi::{dispatch_pgversion, enum_pgversion, enum_pgversion_dispatch, TimestampTz};
 use postgres_ffi::{fsm_logical_to_physical, page_is_new, page_set_lsn};

 use anyhow::{bail, Context, Result};
@@ -48,16 +46,31 @@ use pageserver_api::key::rel_block_to_key;
 use pageserver_api::reltag::{BlockNumber, RelTag, SlruKind};
 use postgres_ffi::pg_constants;
 use postgres_ffi::relfile_utils::{FSM_FORKNUM, INIT_FORKNUM, MAIN_FORKNUM, VISIBILITYMAP_FORKNUM};
-use postgres_ffi::v14::nonrelfile_utils::mx_offset_to_member_segment;
-use postgres_ffi::v14::xlog_utils::*;
-use postgres_ffi::v14::CheckPoint;
 use postgres_ffi::TransactionId;
 use postgres_ffi::BLCKSZ;
+use utils::bin_ser::SerializeError;
 use utils::lsn::Lsn;

+enum_pgversion! {CheckPoint, pgv::CheckPoint}
+
+impl CheckPoint {
+    fn encode(&self) -> Result<Bytes, SerializeError> {
+        enum_pgversion_dispatch!(self, CheckPoint, cp, { cp.encode() })
+    }
+
+    fn update_next_xid(&mut self, xid: u32) -> bool {
+        enum_pgversion_dispatch!(self, CheckPoint, cp, { cp.update_next_xid(xid) })
+    }
+
+    pub fn update_next_multixid(&mut self, multi_xid: u32, multi_offset: u32) -> bool {
+        enum_pgversion_dispatch!(self, CheckPoint, cp, {
+            cp.update_next_multixid(multi_xid, multi_offset)
+        })
+    }
+}
+
 pub struct WalIngest {
    shard: ShardIdentity,
-    pg_version: u32,
    checkpoint: CheckPoint,
    checkpoint_modified: bool,
    warn_ingest_lag: WarnIngestLag,
@@ -78,12 +91,16 @@ impl WalIngest {
        // Fetch the latest checkpoint into memory, so that we can compare with it
        // quickly in `ingest_record` and update it when it changes.
        let checkpoint_bytes = timeline.get_checkpoint(startpoint, ctx).await?;
-        let checkpoint = CheckPoint::decode(&checkpoint_bytes)?;
-        trace!("CheckPoint.nextXid = {}", checkpoint.nextXid.value);
+        let pgversion = timeline.pg_version;
+
+        let checkpoint = dispatch_pgversion!(pgversion, {
+            let checkpoint = pgv::CheckPoint::decode(&checkpoint_bytes)?;
+            trace!("CheckPoint.nextXid = {}", checkpoint.nextXid.value);
+            <pgv::CheckPoint as Into<CheckPoint>>::into(checkpoint)
+        });

        Ok(WalIngest {
            shard: *timeline.get_shard_identity(),
-            pg_version: timeline.pg_version,
            checkpoint,
            checkpoint_modified: false,
            warn_ingest_lag: WarnIngestLag {
@@ -117,7 +134,7 @@ impl WalIngest {

        modification.set_lsn(lsn)?;

-        if decoded.is_dbase_create_copy(self.pg_version) {
+        if decoded.is_dbase_create_copy(pg_version) {
            // Records of this type should always be preceded by a commit(), as they
            // rely on reading data pages back from the Timeline.
            assert!(!modification.has_dirty_data_pages());
@@ -220,6 +237,26 @@ impl WalIngest {
                                .await?;
                        }
                    }
+                } else if pg_version == 17 {
+                    if info == postgres_ffi::v17::bindings::XLOG_DBASE_CREATE_WAL_LOG {
+                        debug!("XLOG_DBASE_CREATE_WAL_LOG: noop");
+                    } else if info == postgres_ffi::v17::bindings::XLOG_DBASE_CREATE_FILE_COPY {
+                        // The XLOG record was renamed between v14 and v15,
+                        // but the record format is the same.
+                        // So we can reuse XlCreateDatabase here.
+                        debug!("XLOG_DBASE_CREATE_FILE_COPY");
+                        let createdb = XlCreateDatabase::decode(&mut buf);
+                        self.ingest_xlog_dbase_create(modification, &createdb, ctx)
+                            .await?;
+                    } else if info == postgres_ffi::v17::bindings::XLOG_DBASE_DROP {
+                        let dropdb = XlDropDatabase::decode(&mut buf);
+                        for tablespace_id in dropdb.tablespace_ids {
+                            trace!("Drop db {}, {}", tablespace_id, dropdb.db_id);
+                            modification
+                                .drop_dbdir(tablespace_id, dropdb.db_id, ctx)
+                                .await?;
+                        }
+                    }
                }
            }
            pg_constants::RM_TBLSPC_ID => {
@@ -229,7 +266,11 @@ impl WalIngest {
                let info = decoded.xl_info & !pg_constants::XLR_INFO_MASK;

                if info == pg_constants::CLOG_ZEROPAGE {
-                    let pageno = buf.get_u32_le();
+                    let pageno = if pg_version < 17 {
+                        buf.get_u32_le()
+                    } else {
+                        buf.get_u64_le() as u32
+                    };
                    let segno = pageno / pg_constants::SLRU_PAGES_PER_SEGMENT;
                    let rpageno = pageno % pg_constants::SLRU_PAGES_PER_SEGMENT;
                    self.put_slru_page_image(
@@ -243,7 +284,7 @@ impl WalIngest {
                    .await?;
                } else {
                    assert!(info == pg_constants::CLOG_TRUNCATE);
-                    let xlrec = XlClogTruncate::decode(&mut buf);
+                    let xlrec = XlClogTruncate::decode(&mut buf, pg_version);
                    self.ingest_clog_truncate_record(modification, &xlrec, ctx)
                        .await?;
                }
@@ -282,12 +323,21 @@ impl WalIngest {
                        parsed_xact.xid,
                        lsn,
                    );
-                    modification
-                        .drop_twophase_file(parsed_xact.xid, ctx)
-                        .await?;
+
+                    let xid: u64 = if pg_version >= 17 {
+                        self.adjust_to_full_transaction_id(parsed_xact.xid)?
+                    } else {
+                        parsed_xact.xid as u64
+                    };
+                    modification.drop_twophase_file(xid, ctx).await?;
                } else if info == pg_constants::XLOG_XACT_PREPARE {
+                    let xid: u64 = if pg_version >= 17 {
+                        self.adjust_to_full_transaction_id(decoded.xl_xid)?
+                    } else {
+                        decoded.xl_xid as u64
+                    };
                    modification
-                        .put_twophase_file(decoded.xl_xid, Bytes::copy_from_slice(&buf[..]), ctx)
+                        .put_twophase_file(xid, Bytes::copy_from_slice(&buf[..]), ctx)
                        .await?;
                }
            }
@@ -295,7 +345,11 @@ impl WalIngest {
                let info = decoded.xl_info & pg_constants::XLR_RMGR_INFO_MASK;

                if info == pg_constants::XLOG_MULTIXACT_ZERO_OFF_PAGE {
-                    let pageno = buf.get_u32_le();
+                    let pageno = if pg_version < 17 {
+                        buf.get_u32_le()
+                    } else {
+                        buf.get_u64_le() as u32
+                    };
                    let segno = pageno / pg_constants::SLRU_PAGES_PER_SEGMENT;
                    let rpageno = pageno % pg_constants::SLRU_PAGES_PER_SEGMENT;
                    self.put_slru_page_image(
@@ -308,7 +362,11 @@ impl WalIngest {
                    )
                    .await?;
                } else if info == pg_constants::XLOG_MULTIXACT_ZERO_MEM_PAGE {
-                    let pageno = buf.get_u32_le();
+                    let pageno = if pg_version < 17 {
+                        buf.get_u32_le()
+                    } else {
+                        buf.get_u64_le() as u32
+                    };
                    let segno = pageno / pg_constants::SLRU_PAGES_PER_SEGMENT;
                    let rpageno = pageno % pg_constants::SLRU_PAGES_PER_SEGMENT;
                    self.put_slru_page_image(
@@ -337,70 +395,93 @@ impl WalIngest {
            pg_constants::RM_XLOG_ID => {
                let info = decoded.xl_info & pg_constants::XLR_RMGR_INFO_MASK;

-                if info == pg_constants::XLOG_NEXTOID {
-                    let next_oid = buf.get_u32_le();
-                    if self.checkpoint.nextOid != next_oid {
-                        self.checkpoint.nextOid = next_oid;
+                if info == pg_constants::XLOG_PARAMETER_CHANGE {
+                    if let CheckPoint::V17(cp) = &mut self.checkpoint {
+                        let rec = v17::XlParameterChange::decode(&mut buf);
+                        cp.wal_level = rec.wal_level;
                        self.checkpoint_modified = true;
                    }
-                } else if info == pg_constants::XLOG_CHECKPOINT_ONLINE
-                    || info == pg_constants::XLOG_CHECKPOINT_SHUTDOWN
-                {
-                    let mut checkpoint_bytes = [0u8; SIZEOF_CHECKPOINT];
-                    buf.copy_to_slice(&mut checkpoint_bytes);
-                    let xlog_checkpoint = CheckPoint::decode(&checkpoint_bytes)?;
-                    trace!(
-                        "xlog_checkpoint.oldestXid={}, checkpoint.oldestXid={}",
-                        xlog_checkpoint.oldestXid,
-                        self.checkpoint.oldestXid
-                    );
-                    if (self
-                        .checkpoint
-                        .oldestXid
-                        .wrapping_sub(xlog_checkpoint.oldestXid) as i32)
-                        < 0
-                    {
-                        self.checkpoint.oldestXid = xlog_checkpoint.oldestXid;
+                } else if info == pg_constants::XLOG_END_OF_RECOVERY {
+                    if let CheckPoint::V17(cp) = &mut self.checkpoint {
+                        let rec = v17::XlEndOfRecovery::decode(&mut buf);
+                        cp.wal_level = rec.wal_level;
+                        self.checkpoint_modified = true;
                    }
-                    trace!(
-                        "xlog_checkpoint.oldestActiveXid={}, checkpoint.oldestActiveXid={}",
-                        xlog_checkpoint.oldestActiveXid,
-                        self.checkpoint.oldestActiveXid
-                    );
-
-                    // A shutdown checkpoint has `oldestActiveXid == InvalidTransactionid`,
-                    // because at shutdown, all in-progress transactions will implicitly
-                    // end. Postgres startup code knows that, and allows hot standby to start
-                    // immediately from a shutdown checkpoint.
-                    //
-                    // In Neon, Postgres hot standby startup always behaves as if starting from
-                    // an online checkpoint. It needs a valid `oldestActiveXid` value, so
-                    // instead of overwriting self.checkpoint.oldestActiveXid with
-                    // InvalidTransactionid from the checkpoint WAL record, update it to a
-                    // proper value, knowing that there are no in-progress transactions at this
-                    // point, except for prepared transactions.
-                    //
-                    // See also the neon code changes in the InitWalRecovery() function.
-                    if xlog_checkpoint.oldestActiveXid == pg_constants::INVALID_TRANSACTION_ID
-                        && info == pg_constants::XLOG_CHECKPOINT_SHUTDOWN
-                    {
-                        let mut oldest_active_xid = self.checkpoint.nextXid.value as u32;
-                        for xid in modification.tline.list_twophase_files(lsn, ctx).await? {
-                            if (xid.wrapping_sub(oldest_active_xid) as i32) < 0 {
-                                oldest_active_xid = xid;
-                            }
-                        }
-                        self.checkpoint.oldestActiveXid = oldest_active_xid;
-                    } else {
-                        self.checkpoint.oldestActiveXid = xlog_checkpoint.oldestActiveXid;
-                    }
-
-                    // Write a new checkpoint key-value pair on every checkpoint record, even
-                    // if nothing really changed. Not strictly required, but it seems nice to
-                    // have some trace of the checkpoint records in the layer files at the same
-                    // LSNs.
-                    self.checkpoint_modified = true;
                }
+
+                enum_pgversion_dispatch!(&mut self.checkpoint, CheckPoint, cp, {
+                    if info == pg_constants::XLOG_NEXTOID {
+                        let next_oid = buf.get_u32_le();
+                        if cp.nextOid != next_oid {
+                            cp.nextOid = next_oid;
+                            self.checkpoint_modified = true;
+                        }
+                    } else if info == pg_constants::XLOG_CHECKPOINT_ONLINE
+                        || info == pg_constants::XLOG_CHECKPOINT_SHUTDOWN
+                    {
+                        let mut checkpoint_bytes = [0u8; pgv::xlog_utils::SIZEOF_CHECKPOINT];
+                        buf.copy_to_slice(&mut checkpoint_bytes);
+                        let xlog_checkpoint = pgv::CheckPoint::decode(&checkpoint_bytes)?;
+                        trace!(
+                            "xlog_checkpoint.oldestXid={}, checkpoint.oldestXid={}",
+                            xlog_checkpoint.oldestXid,
+                            cp.oldestXid
+                        );
+                        if (cp.oldestXid.wrapping_sub(xlog_checkpoint.oldestXid) as i32) < 0 {
+                            cp.oldestXid = xlog_checkpoint.oldestXid;
+                        }
+                        trace!(
+                            "xlog_checkpoint.oldestActiveXid={}, checkpoint.oldestActiveXid={}",
+                            xlog_checkpoint.oldestActiveXid,
+                            cp.oldestActiveXid
+                        );
+
+                        // A shutdown checkpoint has `oldestActiveXid == InvalidTransactionid`,
+                        // because at shutdown, all in-progress transactions will implicitly
+                        // end. Postgres startup code knows that, and allows hot standby to start
+                        // immediately from a shutdown checkpoint.
+                        //
+                        // In Neon, Postgres hot standby startup always behaves as if starting from
+                        // an online checkpoint. It needs a valid `oldestActiveXid` value, so
+                        // instead of overwriting self.checkpoint.oldestActiveXid with
+                        // InvalidTransactionid from the checkpoint WAL record, update it to a
+                        // proper value, knowing that there are no in-progress transactions at this
+                        // point, except for prepared transactions.
+                        //
+                        // See also the neon code changes in the InitWalRecovery() function.
+                        if xlog_checkpoint.oldestActiveXid == pg_constants::INVALID_TRANSACTION_ID
+                            && info == pg_constants::XLOG_CHECKPOINT_SHUTDOWN
+                        {
+                            let oldest_active_xid = if pg_version >= 17 {
+                                let mut oldest_active_full_xid = cp.nextXid.value;
+                                for xid in modification.tline.list_twophase_files(lsn, ctx).await? {
+                                    if xid < oldest_active_full_xid {
+                                        oldest_active_full_xid = xid;
+                                    }
+                                }
+                                oldest_active_full_xid as u32
+                            } else {
+                                let mut oldest_active_xid = cp.nextXid.value as u32;
+                                for xid in modification.tline.list_twophase_files(lsn, ctx).await? {
+                                    let narrow_xid = xid as u32;
+                                    if (narrow_xid.wrapping_sub(oldest_active_xid) as i32) < 0 {
+                                        oldest_active_xid = narrow_xid;
+                                    }
+                                }
+                                oldest_active_xid
+                            };
+                            cp.oldestActiveXid = oldest_active_xid;
+                        } else {
+                            cp.oldestActiveXid = xlog_checkpoint.oldestActiveXid;
+                        }
+
+                        // Write a new checkpoint key-value pair on every checkpoint record, even
+                        // if nothing really changed. Not strictly required, but it seems nice to
+                        // have some trace of the checkpoint records in the layer files at the same
+                        // LSNs.
+                        self.checkpoint_modified = true;
+                    }
+                });
            }
            pg_constants::RM_LOGICALMSG_ID => {
                let info = decoded.xl_info & pg_constants::XLR_RMGR_INFO_MASK;
@@ -424,7 +505,11 @@ impl WalIngest {
                let info = decoded.xl_info & pg_constants::XLR_RMGR_INFO_MASK;
                if info == pg_constants::XLOG_RUNNING_XACTS {
                    let xlrec = crate::walrecord::XlRunningXacts::decode(&mut buf);
-                    self.checkpoint.oldestActiveXid = xlrec.oldest_running_xid;
+
+                    enum_pgversion_dispatch!(&mut self.checkpoint, CheckPoint, cp, {
+                        cp.oldestActiveXid = xlrec.oldest_running_xid;
+                    });
+
                    self.checkpoint_modified = true;
                }
            }
@@ -497,6 +582,25 @@ impl WalIngest {
        Ok(modification.len() > prev_len)
    }

+    /// This is the same as AdjustToFullTransactionId(xid) in PostgreSQL
+    fn adjust_to_full_transaction_id(&self, xid: TransactionId) -> Result<u64> {
+        let next_full_xid =
+            enum_pgversion_dispatch!(&self.checkpoint, CheckPoint, cp, { cp.nextXid.value });
+
+        let next_xid = (next_full_xid) as u32;
+        let mut epoch = (next_full_xid >> 32) as u32;
+
+        if xid > next_xid {
+            // Wraparound occurred, must be from a prev epoch.
+            if epoch == 0 {
+                bail!("apparent XID wraparound with prepared transaction XID {xid}, nextXid is {next_full_xid}");
+            }
+            epoch -= 1;
+        }
+
+        Ok((epoch as u64) << 32 | xid as u64)
+    }
+
    /// Do not store this block, but observe it for the purposes of updating our relation size state.
    async fn observe_decoded_block(
        &mut self,
@@ -539,7 +643,7 @@ impl WalIngest {
            && blk.has_image
            && decoded.xl_rmid == pg_constants::RM_XLOG_ID
            && (decoded.xl_info == pg_constants::XLOG_FPI
-                || decoded.xl_info == pg_constants::XLOG_FPI_FOR_HINT)
+            || decoded.xl_info == pg_constants::XLOG_FPI_FOR_HINT)
            // compression of WAL is not yet supported: fall back to storing the original WAL record
            && !postgres_ffi::bkpimage_is_compressed(blk.bimg_info, modification.tline.pg_version)
            // do not materialize null pages because them most likely be soon replaced with real data
@@ -797,6 +901,73 @@ impl WalIngest {
                    bail!("Unknown RMGR {} for Heap decoding", decoded.xl_rmid);
                }
            }
+            17 => {
+                if decoded.xl_rmid == pg_constants::RM_HEAP_ID {
+                    let info = decoded.xl_info & pg_constants::XLOG_HEAP_OPMASK;
+
+                    if info == pg_constants::XLOG_HEAP_INSERT {
+                        let xlrec = v17::XlHeapInsert::decode(buf);
+                        assert_eq!(0, buf.remaining());
+                        if (xlrec.flags & pg_constants::XLH_INSERT_ALL_VISIBLE_CLEARED) != 0 {
+                            new_heap_blkno = Some(decoded.blocks[0].blkno);
+                        }
+                    } else if info == pg_constants::XLOG_HEAP_DELETE {
+                        let xlrec = v17::XlHeapDelete::decode(buf);
+                        if (xlrec.flags & pg_constants::XLH_DELETE_ALL_VISIBLE_CLEARED) != 0 {
+                            new_heap_blkno = Some(decoded.blocks[0].blkno);
+                        }
+                    } else if info == pg_constants::XLOG_HEAP_UPDATE
+                        || info == pg_constants::XLOG_HEAP_HOT_UPDATE
+                    {
+                        let xlrec = v17::XlHeapUpdate::decode(buf);
+                        // the size of tuple data is inferred from the size of the record.
+                        // we can't validate the remaining number of bytes without parsing
+                        // the tuple data.
+                        if (xlrec.flags & pg_constants::XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED) != 0 {
+                            old_heap_blkno = Some(decoded.blocks.last().unwrap().blkno);
+                        }
+                        if (xlrec.flags & pg_constants::XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED) != 0 {
+                            // PostgreSQL only uses XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED on a
+                            // non-HOT update where the new tuple goes to different page than
+                            // the old one. Otherwise, only XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED is
+                            // set.
+                            new_heap_blkno = Some(decoded.blocks[0].blkno);
+                        }
+                    } else if info == pg_constants::XLOG_HEAP_LOCK {
+                        let xlrec = v17::XlHeapLock::decode(buf);
+                        if (xlrec.flags & pg_constants::XLH_LOCK_ALL_FROZEN_CLEARED) != 0 {
+                            old_heap_blkno = Some(decoded.blocks[0].blkno);
+                            flags = pg_constants::VISIBILITYMAP_ALL_FROZEN;
+                        }
+                    }
+                } else if decoded.xl_rmid == pg_constants::RM_HEAP2_ID {
+                    let info = decoded.xl_info & pg_constants::XLOG_HEAP_OPMASK;
+                    if info == pg_constants::XLOG_HEAP2_MULTI_INSERT {
+                        let xlrec = v17::XlHeapMultiInsert::decode(buf);
+
+                        let offset_array_len =
+                            if decoded.xl_info & pg_constants::XLOG_HEAP_INIT_PAGE > 0 {
+                                // the offsets array is omitted if XLOG_HEAP_INIT_PAGE is set
+                                0
+                            } else {
+                                size_of::<u16>() * xlrec.ntuples as usize
+                            };
+                        assert_eq!(offset_array_len, buf.remaining());
+
+                        if (xlrec.flags & pg_constants::XLH_INSERT_ALL_VISIBLE_CLEARED) != 0 {
+                            new_heap_blkno = Some(decoded.blocks[0].blkno);
+                        }
+                    } else if info == pg_constants::XLOG_HEAP2_LOCK_UPDATED {
+                        let xlrec = v17::XlHeapLockUpdated::decode(buf);
+                        if (xlrec.flags & pg_constants::XLH_LOCK_ALL_FROZEN_CLEARED) != 0 {
+                            old_heap_blkno = Some(decoded.blocks[0].blkno);
+                            flags = pg_constants::VISIBILITYMAP_ALL_FROZEN;
+                        }
+                    }
+                } else {
+                    bail!("Unknown RMGR {} for Heap decoding", decoded.xl_rmid);
+                }
+            }
            _ => {}
        }

@@ -905,26 +1076,26 @@ impl WalIngest {
        assert_eq!(decoded.xl_rmid, pg_constants::RM_NEON_ID);

        match pg_version {
-            16 => {
+            16 | 17 => {
                let info = decoded.xl_info & pg_constants::XLOG_HEAP_OPMASK;

                match info {
                    pg_constants::XLOG_NEON_HEAP_INSERT => {
-                        let xlrec = v16::rm_neon::XlNeonHeapInsert::decode(buf);
+                        let xlrec = v17::rm_neon::XlNeonHeapInsert::decode(buf);
                        assert_eq!(0, buf.remaining());
                        if (xlrec.flags & pg_constants::XLH_INSERT_ALL_VISIBLE_CLEARED) != 0 {
                            new_heap_blkno = Some(decoded.blocks[0].blkno);
                        }
                    }
                    pg_constants::XLOG_NEON_HEAP_DELETE => {
-                        let xlrec = v16::rm_neon::XlNeonHeapDelete::decode(buf);
+                        let xlrec = v17::rm_neon::XlNeonHeapDelete::decode(buf);
                        if (xlrec.flags & pg_constants::XLH_DELETE_ALL_VISIBLE_CLEARED) != 0 {
                            new_heap_blkno = Some(decoded.blocks[0].blkno);
                        }
                    }
                    pg_constants::XLOG_NEON_HEAP_UPDATE
                    | pg_constants::XLOG_NEON_HEAP_HOT_UPDATE => {
-                        let xlrec = v16::rm_neon::XlNeonHeapUpdate::decode(buf);
+                        let xlrec = v17::rm_neon::XlNeonHeapUpdate::decode(buf);
                        // the size of tuple data is inferred from the size of the record.
                        // we can't validate the remaining number of bytes without parsing
                        // the tuple data.
@@ -940,7 +1111,7 @@ impl WalIngest {
                        }
                    }
                    pg_constants::XLOG_NEON_HEAP_MULTI_INSERT => {
-                        let xlrec = v16::rm_neon::XlNeonHeapMultiInsert::decode(buf);
+                        let xlrec = v17::rm_neon::XlNeonHeapMultiInsert::decode(buf);

                        let offset_array_len =
                            if decoded.xl_info & pg_constants::XLOG_HEAP_INIT_PAGE > 0 {
@@ -956,7 +1127,7 @@ impl WalIngest {
                        }
                    }
                    pg_constants::XLOG_NEON_HEAP_LOCK => {
-                        let xlrec = v16::rm_neon::XlNeonHeapLock::decode(buf);
+                        let xlrec = v17::rm_neon::XlNeonHeapLock::decode(buf);
                        if (xlrec.flags & pg_constants::XLH_LOCK_ALL_FROZEN_CLEARED) != 0 {
                            old_heap_blkno = Some(decoded.blocks[0].blkno);
                            flags = pg_constants::VISIBILITYMAP_ALL_FROZEN;
@@ -1204,7 +1375,7 @@ impl WalIngest {
            if rec.blkno % pg_constants::SLOTS_PER_FSM_PAGE != 0 {
                // Tail of last remaining FSM page has to be zeroed.
                // We are not precise here and instead of digging in FSM bitmap format just clear the whole page.
-                modification.put_rel_page_image_zero(rel, fsm_physical_page_no);
+                modification.put_rel_page_image_zero(rel, fsm_physical_page_no)?;
                fsm_physical_page_no += 1;
            }
            let nblocks = get_relsize(modification, rel, ctx).await?;
@@ -1226,7 +1397,7 @@ impl WalIngest {
            if rec.blkno % pg_constants::VM_HEAPBLOCKS_PER_PAGE != 0 {
                // Tail of last remaining vm page has to be zeroed.
                // We are not precise here and instead of digging in VM bitmap format just clear the whole page.
-                modification.put_rel_page_image_zero(rel, vm_page_no);
+                modification.put_rel_page_image_zero(rel, vm_page_no)?;
                vm_page_no += 1;
            }
            let nblocks = get_relsize(modification, rel, ctx).await?;
@@ -1242,12 +1413,17 @@ impl WalIngest {
    fn warn_on_ingest_lag(
        &mut self,
        conf: &crate::config::PageServerConf,
-        wal_timestmap: TimestampTz,
+        wal_timestamp: TimestampTz,
    ) {
        debug_assert_current_span_has_tenant_and_timeline_id();
        let now = SystemTime::now();
        let rate_limits = &mut self.warn_ingest_lag;
-        match try_from_pg_timestamp(wal_timestmap) {
+
+        let ts = enum_pgversion_dispatch!(&self.checkpoint, CheckPoint, _cp, {
+            pgv::xlog_utils::try_from_pg_timestamp(wal_timestamp)
+        });
+
+        match ts {
            Ok(ts) => {
                match now.duration_since(ts) {
                    Ok(lag) => {
@@ -1257,7 +1433,7 @@ impl WalIngest {
                                warn!(%rate_limit_stats, %lag, "ingesting record with timestamp lagging more than wait_lsn_timeout");
                            })
                        }
-                    },
+                    }
                    Err(e) => {
                        let delta_t = e.duration();
                        // determined by prod victoriametrics query: 1000 * (timestamp(node_time_seconds{neon_service="pageserver"}) - node_time_seconds)
@@ -1271,7 +1447,6 @@ impl WalIngest {
                        }
                    }
                };
-
            }
            Err(error) => {
                rate_limits.timestamp_invalid_msg_ratelimit.call2(|rate_limit_stats| {
@@ -1379,14 +1554,17 @@ impl WalIngest {
        // truncated, but a checkpoint record with the updated values isn't written until
        // later. In Neon, a server can start at any LSN, not just on a checkpoint record,
        // so we keep the oldestXid and oldestXidDB up-to-date.
-        self.checkpoint.oldestXid = xlrec.oldest_xid;
-        self.checkpoint.oldestXidDB = xlrec.oldest_xid_db;
+        enum_pgversion_dispatch!(&mut self.checkpoint, CheckPoint, cp, {
+            cp.oldestXid = xlrec.oldest_xid;
+            cp.oldestXidDB = xlrec.oldest_xid_db;
+        });
        self.checkpoint_modified = true;

        // TODO Treat AdvanceOldestClogXid() or write a comment why we don't need it

        let latest_page_number =
-            self.checkpoint.nextXid.value as u32 / pg_constants::CLOG_XACTS_PER_PAGE;
+            enum_pgversion_dispatch!(self.checkpoint, CheckPoint, cp, { cp.nextXid.value }) as u32
+                / pg_constants::CLOG_XACTS_PER_PAGE;

        // Now delete all segments containing pages between xlrec.pageno
        // and latest_page_number.
@@ -1394,7 +1572,9 @@ impl WalIngest {
        // First, make an important safety check:
        // the current endpoint page must not be eligible for removal.
        // See SimpleLruTruncate() in slru.c
-        if clogpage_precedes(latest_page_number, xlrec.pageno) {
+        if dispatch_pgversion!(modification.tline.pg_version, {
+            pgv::nonrelfile_utils::clogpage_precedes(latest_page_number, xlrec.pageno)
+        }) {
            info!("could not truncate directory pg_xact apparent wraparound");
            return Ok(());
        }
@@ -1411,7 +1591,12 @@ impl WalIngest {
            .await?
        {
            let segpage = segno * pg_constants::SLRU_PAGES_PER_SEGMENT;
-            if slru_may_delete_clogsegment(segpage, xlrec.pageno) {
+
+            let may_delete = dispatch_pgversion!(modification.tline.pg_version, {
+                pgv::nonrelfile_utils::slru_may_delete_clogsegment(segpage, xlrec.pageno)
+            });
+
+            if may_delete {
                modification
                    .drop_slru_segment(SlruKind::Clog, segno, ctx)
                    .await?;
@@ -1530,14 +1715,23 @@ impl WalIngest {
        xlrec: &XlMultiXactTruncate,
        ctx: &RequestContext,
    ) -> Result<()> {
-        self.checkpoint.oldestMulti = xlrec.end_trunc_off;
-        self.checkpoint.oldestMultiDB = xlrec.oldest_multi_db;
+        let (maxsegment, startsegment, endsegment) =
+            enum_pgversion_dispatch!(&mut self.checkpoint, CheckPoint, cp, {
+                cp.oldestMulti = xlrec.end_trunc_off;
+                cp.oldestMultiDB = xlrec.oldest_multi_db;
+                let maxsegment: i32 = pgv::nonrelfile_utils::mx_offset_to_member_segment(
+                    pg_constants::MAX_MULTIXACT_OFFSET,
+                );
+                let startsegment: i32 =
+                    pgv::nonrelfile_utils::mx_offset_to_member_segment(xlrec.start_trunc_memb);
+                let endsegment: i32 =
+                    pgv::nonrelfile_utils::mx_offset_to_member_segment(xlrec.end_trunc_memb);
+                (maxsegment, startsegment, endsegment)
+            });
+
        self.checkpoint_modified = true;

        // PerformMembersTruncation
-        let maxsegment: i32 = mx_offset_to_member_segment(pg_constants::MAX_MULTIXACT_OFFSET);
-        let startsegment: i32 = mx_offset_to_member_segment(xlrec.start_trunc_memb);
-        let endsegment: i32 = mx_offset_to_member_segment(xlrec.end_trunc_memb);
        let mut segment: i32 = startsegment;

        // Delete all the segments except the last one. The last segment can still
@@ -1696,7 +1890,7 @@ impl WalIngest {
                    continue;
                }

-                modification.put_rel_page_image_zero(rel, gap_blknum);
+                modification.put_rel_page_image_zero(rel, gap_blknum)?;
            }
        }
        Ok(())
@@ -1762,7 +1956,7 @@ impl WalIngest {

            // fill the gap with zeros
            for gap_blknum in old_nblocks..blknum {
-                modification.put_slru_page_image_zero(kind, segno, gap_blknum);
+                modification.put_slru_page_image_zero(kind, segno, gap_blknum)?;
            }
        }
        Ok(())
@@ -1811,11 +2005,23 @@ mod tests {
        // TODO
    }

-    static ZERO_CHECKPOINT: Bytes = Bytes::from_static(&[0u8; SIZEOF_CHECKPOINT]);
+    #[tokio::test]
+    async fn test_zeroed_checkpoint_decodes_correctly() -> Result<()> {
+        for i in 14..=16 {
+            dispatch_pgversion!(i, {
+                pgv::CheckPoint::decode(&pgv::ZERO_CHECKPOINT)?;
+            });
+        }
+
+        Ok(())
+    }

    async fn init_walingest_test(tline: &Timeline, ctx: &RequestContext) -> Result<WalIngest> {
        let mut m = tline.begin_modification(Lsn(0x10));
-        m.put_checkpoint(ZERO_CHECKPOINT.clone())?;
+        m.put_checkpoint(dispatch_pgversion!(
+            tline.pg_version,
+            pgv::ZERO_CHECKPOINT.clone()
+        ))?;
        m.put_relmap_file(0, 111, Bytes::from(""), ctx).await?; // dummy relmapper file
        m.commit(ctx).await?;
        let walingest = WalIngest::new(tline, Lsn(0x10), ctx).await?;
--- a/pageserver/src/walrecord.rs
+++ b/pageserver/src/walrecord.rs
@@ -174,6 +174,7 @@ impl DecodedWALRecord {
                }
                15 => info == postgres_ffi::v15::bindings::XLOG_DBASE_CREATE_FILE_COPY,
                16 => info == postgres_ffi::v16::bindings::XLOG_DBASE_CREATE_FILE_COPY,
+                17 => info == postgres_ffi::v17::bindings::XLOG_DBASE_CREATE_FILE_COPY,
                _ => {
                    panic!("Unsupported postgres version {pg_version}")
                }
@@ -341,16 +342,47 @@ pub mod v14 {
            }
        }
    }
+
+    #[repr(C)]
+    #[derive(Debug)]
+    pub struct XlParameterChange {
+        pub max_connections: i32,
+        pub max_worker_processes: i32,
+        pub max_wal_senders: i32,
+        pub max_prepared_xacts: i32,
+        pub max_locks_per_xact: i32,
+        pub wal_level: i32,
+        pub wal_log_hints: bool,
+        pub track_commit_timestamp: bool,
+        pub _padding: [u8; 2],
+    }
+
+    impl XlParameterChange {
+        pub fn decode(buf: &mut Bytes) -> XlParameterChange {
+            XlParameterChange {
+                max_connections: buf.get_i32_le(),
+                max_worker_processes: buf.get_i32_le(),
+                max_wal_senders: buf.get_i32_le(),
+                max_prepared_xacts: buf.get_i32_le(),
+                max_locks_per_xact: buf.get_i32_le(),
+                wal_level: buf.get_i32_le(),
+                wal_log_hints: buf.get_u8() != 0,
+                track_commit_timestamp: buf.get_u8() != 0,
+                _padding: [buf.get_u8(), buf.get_u8()],
+            }
+        }
+    }
 }

 pub mod v15 {
    pub use super::v14::{
        XlHeapDelete, XlHeapInsert, XlHeapLock, XlHeapLockUpdated, XlHeapMultiInsert, XlHeapUpdate,
+        XlParameterChange,
    };
 }

 pub mod v16 {
-    pub use super::v14::{XlHeapInsert, XlHeapLockUpdated, XlHeapMultiInsert};
+    pub use super::v14::{XlHeapInsert, XlHeapLockUpdated, XlHeapMultiInsert, XlParameterChange};
    use bytes::{Buf, Bytes};
    use postgres_ffi::{OffsetNumber, TransactionId};

@@ -529,6 +561,37 @@ pub mod v16 {
    }
 }

+pub mod v17 {
+    pub use super::v14::XlHeapLockUpdated;
+    use bytes::{Buf, Bytes};
+    pub use postgres_ffi::{TimeLineID, TimestampTz};
+
+    pub use super::v16::rm_neon;
+    pub use super::v16::{
+        XlHeapDelete, XlHeapInsert, XlHeapLock, XlHeapMultiInsert, XlHeapUpdate, XlParameterChange,
+    };
+
+    #[repr(C)]
+    #[derive(Debug)]
+    pub struct XlEndOfRecovery {
+        pub end_time: TimestampTz,
+        pub this_time_line_id: TimeLineID,
+        pub prev_time_line_id: TimeLineID,
+        pub wal_level: i32,
+    }
+
+    impl XlEndOfRecovery {
+        pub fn decode(buf: &mut Bytes) -> XlEndOfRecovery {
+            XlEndOfRecovery {
+                end_time: buf.get_i64_le(),
+                this_time_line_id: buf.get_u32_le(),
+                prev_time_line_id: buf.get_u32_le(),
+                wal_level: buf.get_i32_le(),
+            }
+        }
+    }
+}
+
 #[repr(C)]
 #[derive(Debug)]
 pub struct XlSmgrCreate {
@@ -746,9 +809,13 @@ pub struct XlClogTruncate {
 }

 impl XlClogTruncate {
-    pub fn decode(buf: &mut Bytes) -> XlClogTruncate {
+    pub fn decode(buf: &mut Bytes, pg_version: u32) -> XlClogTruncate {
        XlClogTruncate {
-            pageno: buf.get_u32_le(),
+            pageno: if pg_version < 17 {
+                buf.get_u32_le()
+            } else {
+                buf.get_u64_le() as u32
+            },
            oldest_xid: buf.get_u32_le(),
            oldest_xid_db: buf.get_u32_le(),
        }
--- a/pageserver/src/walredo.rs
+++ b/pageserver/src/walredo.rs
@@ -35,6 +35,7 @@ use anyhow::Context;
 use bytes::{Bytes, BytesMut};
 use pageserver_api::models::{WalRedoManagerProcessStatus, WalRedoManagerStatus};
 use pageserver_api::shard::TenantShardId;
+use std::future::Future;
 use std::sync::Arc;
 use std::time::Duration;
 use std::time::Instant;
@@ -296,6 +297,97 @@ impl PostgresRedoManager {
        }
    }

+    async fn do_with_walredo_process<
+        F: FnOnce(Arc<Process>) -> Fut,
+        Fut: Future<Output = Result<O, Error>>,
+        O,
+    >(
+        &self,
+        pg_version: u32,
+        closure: F,
+    ) -> Result<O, Error> {
+        let proc: Arc<Process> = match self.redo_process.get_or_init_detached().await {
+            Ok(guard) => match &*guard {
+                ProcessOnceCell::Spawned(proc) => Arc::clone(proc),
+                ProcessOnceCell::ManagerShutDown => {
+                    return Err(Error::Cancelled);
+                }
+            },
+            Err(permit) => {
+                let start = Instant::now();
+                // acquire guard before spawning process, so that we don't spawn new processes
+                // if the gate is already closed.
+                let _launched_processes_guard = match self.launched_processes.enter() {
+                    Ok(guard) => guard,
+                    Err(GateError::GateClosed) => unreachable!(
+                        "shutdown sets the once cell to `ManagerShutDown` state before closing the gate"
+                    ),
+                };
+                let proc = Arc::new(Process {
+                    process: process::WalRedoProcess::launch(
+                        self.conf,
+                        self.tenant_shard_id,
+                        pg_version,
+                    )
+                    .context("launch walredo process")?,
+                    _launched_processes_guard,
+                });
+                let duration = start.elapsed();
+                WAL_REDO_PROCESS_LAUNCH_DURATION_HISTOGRAM.observe(duration.as_secs_f64());
+                info!(
+                    elapsed_ms = duration.as_millis(),
+                    pid = proc.id(),
+                    "launched walredo process"
+                );
+                self.redo_process
+                    .set(ProcessOnceCell::Spawned(Arc::clone(&proc)), permit);
+                proc
+            }
+        };
+
+        // async closures are unstable, would support &Process
+        let result = closure(proc.clone()).await;
+
+        if result.is_err() {
+            // Avoid concurrent callers hitting the same issue by taking `proc` out of the rotation.
+            // Note that there may be other tasks concurrent with us that also hold `proc`.
+            // We have to deal with that here.
+            // Also read the doc comment on field `self.redo_process`.
+            //
+            // NB: there may still be other concurrent threads using `proc`.
+            // The last one will send SIGKILL when the underlying Arc reaches refcount 0.
+            //
+            // NB: the drop impl blocks the dropping thread with a wait() system call for
+            // the child process. In some ways the blocking is actually good: if we
+            // deferred the waiting into the background / to tokio if we used `tokio::process`,
+            // it could happen that if walredo always fails immediately, we spawn processes faster
+            // than we can SIGKILL & `wait` for them to exit. By doing it the way we do here,
+            // we limit this risk of run-away to at most $num_runtimes * $num_executor_threads.
+            // This probably needs revisiting at some later point.
+            match self.redo_process.get() {
+                None => (),
+                Some(guard) => {
+                    match &*guard {
+                        ProcessOnceCell::ManagerShutDown => {}
+                        ProcessOnceCell::Spawned(guard_proc) => {
+                            if Arc::ptr_eq(&proc, guard_proc) {
+                                // We're the first to observe an error from `proc`, it's our job to take it out of rotation.
+                                guard.take_and_deinit();
+                            } else {
+                                // Another task already spawned another redo process (further up in this method)
+                                // and put it into `redo_process`. Do nothing, our view of the world is behind.
+                            }
+                        }
+                    }
+                }
+            }
+            // The last task that does this `drop()` of `proc` will do a blocking `wait()` syscall.
+            drop(proc);
+        }
+
+        result
+    }
+
    ///
    /// Process one request for WAL redo using wal-redo postgres
    ///
@@ -319,130 +411,63 @@ impl PostgresRedoManager {
        const MAX_RETRY_ATTEMPTS: u32 = 1;
        let mut n_attempts = 0u32;
        loop {
-            let proc: Arc<Process> = match self.redo_process.get_or_init_detached().await {
-                Ok(guard) => match &*guard {
-                    ProcessOnceCell::Spawned(proc) => Arc::clone(proc),
-                    ProcessOnceCell::ManagerShutDown => {
-                        return Err(Error::Cancelled);
-                    }
-                },
-                Err(permit) => {
-                    let start = Instant::now();
-                    // acquire guard before spawning process, so that we don't spawn new processes
-                    // if the gate is already closed.
-                    let _launched_processes_guard = match self.launched_processes.enter() {
-                                Ok(guard) => guard,
-                                Err(GateError::GateClosed) => unreachable!(
-                                    "shutdown sets the once cell to `ManagerShutDown` state before closing the gate"
-                                ),
-                            };
-                    let proc = Arc::new(Process {
-                        process: process::WalRedoProcess::launch(
-                            self.conf,
-                            self.tenant_shard_id,
-                            pg_version,
-                        )
-                        .context("launch walredo process")?,
-                        _launched_processes_guard,
-                    });
-                    let duration = start.elapsed();
-                    WAL_REDO_PROCESS_LAUNCH_DURATION_HISTOGRAM.observe(duration.as_secs_f64());
-                    info!(
-                        duration_ms = duration.as_millis(),
-                        pid = proc.id(),
-                        "launched walredo process"
-                    );
-                    self.redo_process
-                        .set(ProcessOnceCell::Spawned(Arc::clone(&proc)), permit);
-                    proc
-                }
-            };
+            let base_img = &base_img;
+            let closure = |proc: Arc<Process>| async move {
+                let started_at = std::time::Instant::now();

-            let started_at = std::time::Instant::now();
+                // Relational WAL records are applied using wal-redo-postgres
+                let result = proc
+                    .apply_wal_records(rel, blknum, base_img, records, wal_redo_timeout)
+                    .await
+                    .context("apply_wal_records");

-            // Relational WAL records are applied using wal-redo-postgres
-            let result = proc
-                .apply_wal_records(rel, blknum, &base_img, records, wal_redo_timeout)
-                .await
-                .context("apply_wal_records");
+                let duration = started_at.elapsed();

-            let duration = started_at.elapsed();
-
-            let len = records.len();
-            let nbytes = records.iter().fold(0, |acumulator, record| {
-                acumulator
-                    + match &record.1 {
-                        NeonWalRecord::Postgres { rec, .. } => rec.len(),
-                        _ => unreachable!("Only PostgreSQL records are accepted in this batch"),
-                    }
-            });
-
-            WAL_REDO_TIME.observe(duration.as_secs_f64());
-            WAL_REDO_RECORDS_HISTOGRAM.observe(len as f64);
-            WAL_REDO_BYTES_HISTOGRAM.observe(nbytes as f64);
-
-            debug!(
-                "postgres applied {} WAL records ({} bytes) in {} us to reconstruct page image at LSN {}",
-                len,
-                nbytes,
-                duration.as_micros(),
-                lsn
-            );
-
-            // If something went wrong, don't try to reuse the process. Kill it, and
-            // next request will launch a new one.
-            if let Err(e) = result.as_ref() {
-                error!(
-                    "error applying {} WAL records {}..{} ({} bytes) to key {key}, from base image with LSN {} to reconstruct page image at LSN {} n_attempts={}: {:?}",
-                    records.len(),
-                    records.first().map(|p| p.0).unwrap_or(Lsn(0)),
-                    records.last().map(|p| p.0).unwrap_or(Lsn(0)),
-                    nbytes,
-                    base_img_lsn,
-                    lsn,
-                    n_attempts,
-                    e,
-                );
-                // Avoid concurrent callers hitting the same issue by taking `proc` out of the rotation.
-                // Note that there may be other tasks concurrent with us that also hold `proc`.
-                // We have to deal with that here.
-                // Also read the doc comment on field `self.redo_process`.
-                //
-                // NB: there may still be other concurrent threads using `proc`.
-                // The last one will send SIGKILL when the underlying Arc reaches refcount 0.
-                //
-                // NB: the drop impl blocks the dropping thread with a wait() system call for
-                // the child process. In some ways the blocking is actually good: if we
-                // deferred the waiting into the background / to tokio if we used `tokio::process`,
-                // it could happen that if walredo always fails immediately, we spawn processes faster
-                // than we can SIGKILL & `wait` for them to exit. By doing it the way we do here,
-                // we limit this risk of run-away to at most $num_runtimes * $num_executor_threads.
-                // This probably needs revisiting at some later point.
-                match self.redo_process.get() {
-                    None => (),
-                    Some(guard) => {
-                        match &*guard {
-                            ProcessOnceCell::ManagerShutDown => {}
-                            ProcessOnceCell::Spawned(guard_proc) => {
-                                if Arc::ptr_eq(&proc, guard_proc) {
-                                    // We're the first to observe an error from `proc`, it's our job to take it out of rotation.
-                                    guard.take_and_deinit();
-                                } else {
-                                    // Another task already spawned another redo process (further up in this method)
-                                    // and put it into `redo_process`. Do nothing, our view of the world is behind.
-                                }
-                            }
+                let len = records.len();
+                let nbytes = records.iter().fold(0, |acumulator, record| {
+                    acumulator
+                        + match &record.1 {
+                            NeonWalRecord::Postgres { rec, .. } => rec.len(),
+                            _ => unreachable!("Only PostgreSQL records are accepted in this batch"),
                        }
-                    }
+                });
+
+                WAL_REDO_TIME.observe(duration.as_secs_f64());
+                WAL_REDO_RECORDS_HISTOGRAM.observe(len as f64);
+                WAL_REDO_BYTES_HISTOGRAM.observe(nbytes as f64);
+
+                debug!(
+                    "postgres applied {} WAL records ({} bytes) in {} us to reconstruct page image at LSN {}",
+                    len,
+                    nbytes,
+                    duration.as_micros(),
+                    lsn
+                );
+
+                if let Err(e) = result.as_ref() {
+                    error!(
+                        "error applying {} WAL records {}..{} ({} bytes) to key {key}, from base image with LSN {} to reconstruct page image at LSN {} n_attempts={}: {:?}",
+                        records.len(),
+                        records.first().map(|p| p.0).unwrap_or(Lsn(0)),
+                        records.last().map(|p| p.0).unwrap_or(Lsn(0)),
+                        nbytes,
+                        base_img_lsn,
+                        lsn,
+                        n_attempts,
+                        e,
+                    );
                }
-                // The last task that does this `drop()` of `proc` will do a blocking `wait()` syscall.
-                drop(proc);
-            } else if n_attempts != 0 {
+
+                result.map_err(Error::Other)
+            };
+            let result = self.do_with_walredo_process(pg_version, closure).await;
+
+            if result.is_ok() && n_attempts != 0 {
                info!(n_attempts, "retried walredo succeeded");
            }
            n_attempts += 1;
            if n_attempts > MAX_RETRY_ATTEMPTS || result.is_ok() {
-                return result.map_err(Error::Other);
+                return result;
            }
        }
    }
--- a/pgxn/neon/bitmap.h
+++ b/pgxn/neon/bitmap.h
@@ -0,0 +1,12 @@
+#ifndef NEON_BITMAP_H
+#define NEON_BITMAP_H
+
+/*
+ * Utilities for manipulating bits8* as bitmaps.
+ */
+
+#define BITMAP_ISSET(bm, bit) ((bm)[(bit) >> 3] & (1 << ((bit) & 7)))
+#define BITMAP_SET(bm, bit) (bm)[(bit) >> 3] |= (1 << ((bit) & 7))
+#define BITMAP_CLR(bm, bit) (bm)[(bit) >> 3] &= ~(1 << ((bit) & 7))
+
+#endif //NEON_BITMAP_H
--- a/pgxn/neon/file_cache.c
+++ b/pgxn/neon/file_cache.c
@@ -27,6 +27,7 @@
 #include "pagestore_client.h"
 #include "common/hashfn.h"
 #include "pgstat.h"
+#include "port/pg_iovec.h"
 #include "postmaster/bgworker.h"
 #include RELFILEINFO_HDR
 #include "storage/buf_internals.h"
@@ -40,6 +41,7 @@
 #include "utils/guc.h"

 #include "hll.h"
+#include "bitmap.h"

 #define CriticalAssert(cond) do if (!(cond)) elog(PANIC, "Assertion %s failed at %s:%d: ", #cond, __FILE__, __LINE__); while (0)

@@ -469,6 +471,99 @@ lfc_cache_contains(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno)
 	return found;
 }

+/*
+ * Check if page is present in the cache.
+ * Returns true if page is found in local cache.
+ */
+int
+lfc_cache_containsv(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
+					int nblocks, bits8 *bitmap)
+{
+	BufferTag	tag;
+	FileCacheEntry *entry;
+	uint32		chunk_offs;
+	int			found = 0;
+	uint32		hash;
+	int			i = 0;
+
+	if (lfc_maybe_disabled())	/* fast exit if file cache is disabled */
+		return 0;
+
+	CopyNRelFileInfoToBufTag(tag, rinfo);
+	tag.forkNum = forkNum;
+
+	CriticalAssert(BufTagGetRelNumber(&tag) != InvalidRelFileNumber);
+
+	tag.blockNum = (blkno + i) & ~(BLOCKS_PER_CHUNK - 1);
+	hash = get_hash_value(lfc_hash, &tag);
+	chunk_offs = (blkno + i) & (BLOCKS_PER_CHUNK - 1);
+
+	LWLockAcquire(lfc_lock, LW_SHARED);
+
+	while (true)
+	{
+		int		this_chunk = Min(nblocks, BLOCKS_PER_CHUNK - chunk_offs);
+		if (LFC_ENABLED())
+		{
+			entry = hash_search_with_hash_value(lfc_hash, &tag, hash, HASH_FIND, NULL);
+
+			if (entry != NULL)
+			{
+				for (; chunk_offs < BLOCKS_PER_CHUNK && i < nblocks; chunk_offs++, i++)
+				{
+					if ((entry->bitmap[chunk_offs >> 5] & 
+						(1 << (chunk_offs & 31))) != 0)
+					{
+						BITMAP_SET(bitmap, i);
+						found++;
+					}
+				}
+			}
+			else
+			{
+				i += this_chunk;
+			}
+		}
+		else
+		{
+			return found;
+		}
+
+		/*
+		 * Break out of the iteration before doing expensive stuff for
+		 * a next iteration
+		 */
+		if (i + 1 >= nblocks)
+			break;
+
+		/*
+		 * Prepare for the next iteration. We don't unlock here, as that'd
+		 * probably be more expensive than the gains it'd get us.
+		 */
+		tag.blockNum = (blkno + i) & ~(BLOCKS_PER_CHUNK - 1);
+		hash = get_hash_value(lfc_hash, &tag);
+		chunk_offs = (blkno + i) & (BLOCKS_PER_CHUNK - 1);
+	}
+
+	LWLockRelease(lfc_lock);
+
+#if USE_ASSERT_CHECKING
+	do {
+		int count = 0;
+
+		for (int j = 0; j < nblocks; j++)
+		{
+			if (BITMAP_ISSET(bitmap, j))
+				count++;
+		}
+
+		Assert(count == found);
+	} while (false);
+#endif
+
+	return found;
+}
+
 /*
 * Evict a page (if present) from the local file cache
 */
@@ -548,91 +643,171 @@ lfc_evict(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno)
 }

 /*
- * Try to read page from local cache.
- * Returns true if page is found in local cache.
- * In case of error local file cache is disabled (lfc->limit is set to zero).
+ * Try to read pages from local cache.
+ * Returns the number of pages read from the local cache, and sets bits in
+ * 'read' for the pages which were read. This may scribble over buffers not
+ * marked in 'read', so be careful with operation ordering.
+ *
+ * In case of error local file cache is disabled (lfc->limit is set to zero),
+ * and -1 is returned. Note that 'read' and the buffers may be touched and in
+ * an otherwise invalid state.
+ *
+ * If the mask argument is supplied, bits will be set at the offsets of pages
+ * that were present and read from the LFC.
 */
-bool
-lfc_read(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
-		 char *buffer)
+int
+lfc_readv_select(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
+				 void **buffers, BlockNumber nblocks, bits8 *mask)
 {
 	BufferTag	tag;
 	FileCacheEntry *entry;
 	ssize_t		rc;
-	int			chunk_offs = blkno & (BLOCKS_PER_CHUNK - 1);
 	bool		result = true;
 	uint32		hash;
 	uint64		generation;
 	uint32		entry_offset;
+	int			blocks_read = 0;
+	int			buf_offset = 0;

 	if (lfc_maybe_disabled())	/* fast exit if file cache is disabled */
-		return false;
+		return 0;

 	if (!lfc_ensure_opened())
-		return false;
+		return 0;

 	CopyNRelFileInfoToBufTag(tag, rinfo);
 	tag.forkNum = forkNum;
-	tag.blockNum = blkno & ~(BLOCKS_PER_CHUNK - 1);

 	CriticalAssert(BufTagGetRelNumber(&tag) != InvalidRelFileNumber);
-	hash = get_hash_value(lfc_hash, &tag);

-	LWLockAcquire(lfc_lock, LW_EXCLUSIVE);
-
-	if (!LFC_ENABLED())
+	/* 
+	 * For every chunk that has blocks we're interested in, we
+	 * 1. get the chunk header
+	 * 2. Check if the chunk actually has the blocks we're interested in
+	 * 3. Read the blocks we're looking for (in one preadv), assuming they exist
+	 * 4. Update the statistics for the read call.
+	 *
+	 * If there is an error, we do an early return.
+	 */
+	while (nblocks > 0)
 	{
+		struct iovec iov[PG_IOV_MAX];
+		int		chunk_offs = blkno & (BLOCKS_PER_CHUNK - 1);
+		int		blocks_in_chunk = Min(nblocks, BLOCKS_PER_CHUNK - (blkno % BLOCKS_PER_CHUNK));
+		int		iteration_hits = 0;
+		int		iteration_misses = 0;
+		Assert(blocks_in_chunk > 0);
+
+		for (int i = 0; i < blocks_in_chunk; i++)
+		{
+			iov[i].iov_base = buffers[buf_offset + i];
+			iov[i].iov_len = BLCKSZ;
+		}
+
+		tag.blockNum = blkno - chunk_offs;
+		hash = get_hash_value(lfc_hash, &tag);
+
+		LWLockAcquire(lfc_lock, LW_EXCLUSIVE);
+
+		/* We can return the blocks we've read before LFC got disabled;
+		 * assuming we read any. */
+		if (!LFC_ENABLED())
+		{
+			LWLockRelease(lfc_lock);
+			return blocks_read;
+		}
+
+		entry = hash_search_with_hash_value(lfc_hash, &tag, hash, HASH_FIND, NULL);
+
+		/* Approximate working set for the blocks assumed in this entry */
+		for (int i = 0; i < blocks_in_chunk; i++)
+		{
+			tag.blockNum = blkno + i;
+			addSHLL(&lfc_ctl->wss_estimation, hash_bytes((uint8_t const*)&tag, sizeof(tag)));
+		}
+
+		if (entry == NULL)
+		{
+			/* Pages are not cached */
+			lfc_ctl->misses += blocks_in_chunk;
+			pgBufferUsage.file_cache.misses += blocks_in_chunk;
+			LWLockRelease(lfc_lock);
+
+			buf_offset += blocks_in_chunk;
+			nblocks -= blocks_in_chunk;
+			blkno += blocks_in_chunk;
+
+			continue;
+		}
+
+		/* Unlink entry from LRU list to pin it for the duration of IO operation */
+		if (entry->access_count++ == 0)
+			dlist_delete(&entry->list_node);
+
+		generation = lfc_ctl->generation;
+		entry_offset = entry->offset;
+
 		LWLockRelease(lfc_lock);
-		return false;
-	}

-	entry = hash_search_with_hash_value(lfc_hash, &tag, hash, HASH_FIND, NULL);
+		for (int i = 0; i < blocks_in_chunk; i++)
+		{
+			/*
+			 * If the page is valid, we consider it "read".
+			 * All other pages will be fetched separately by the next cache
+			 */
+			if (entry->bitmap[(chunk_offs + i) / 32] & (1 << ((chunk_offs + i) % 32)))
+			{
+				BITMAP_SET(mask, buf_offset + i);
+				iteration_hits++;
+			}
+			else
+				iteration_misses++;
+		}

-	/* Approximate working set */
-	tag.blockNum = blkno;
-	addSHLL(&lfc_ctl->wss_estimation, hash_bytes((uint8_t const*)&tag, sizeof(tag)));
+		Assert(iteration_hits + iteration_misses > 0);
+
+		if (iteration_hits != 0)
+		{
+			rc = preadv(lfc_desc, iov, blocks_in_chunk,
+						((off_t) entry_offset * BLOCKS_PER_CHUNK + chunk_offs) * BLCKSZ);
+
+			if (rc != (BLCKSZ * blocks_in_chunk))
+			{
+				lfc_disable("read");
+				return -1;
+			}
+		}
+
+		/* Place entry to the head of LRU list */
+		LWLockAcquire(lfc_lock, LW_EXCLUSIVE);
+
+		if (lfc_ctl->generation == generation)
+		{
+			CriticalAssert(LFC_ENABLED());
+			lfc_ctl->hits += iteration_hits;
+			lfc_ctl->misses += iteration_misses;
+			pgBufferUsage.file_cache.hits += iteration_hits;
+			pgBufferUsage.file_cache.misses += iteration_misses;
+			CriticalAssert(entry->access_count > 0);
+			if (--entry->access_count == 0)
+				dlist_push_tail(&lfc_ctl->lru, &entry->list_node);
+		}
+		else
+		{
+			/* generation mismatch, assume error condition */
+			LWLockRelease(lfc_lock);
+			return -1;
+		}

-	if (entry == NULL || (entry->bitmap[chunk_offs >> 5] & (1 << (chunk_offs & 31))) == 0)
-	{
-		/* Page is not cached */
-		lfc_ctl->misses += 1;
-		pgBufferUsage.file_cache.misses += 1;
 		LWLockRelease(lfc_lock);
-		return false;
-	}
-	/* Unlink entry from LRU list to pin it for the duration of IO operation */
-	if (entry->access_count++ == 0)
-		dlist_delete(&entry->list_node);
-	generation = lfc_ctl->generation;
-	entry_offset = entry->offset;

-	LWLockRelease(lfc_lock);
-
-	rc = pread(lfc_desc, buffer, BLCKSZ, ((off_t) entry_offset * BLOCKS_PER_CHUNK + chunk_offs) * BLCKSZ);
-	if (rc != BLCKSZ)
-	{
-		lfc_disable("read");
-		return false;
+		buf_offset += blocks_in_chunk;
+		nblocks -= blocks_in_chunk;
+		blkno += blocks_in_chunk;
+		blocks_read += iteration_hits;
 	}

-	/* Place entry to the head of LRU list */
-	LWLockAcquire(lfc_lock, LW_EXCLUSIVE);
-
-	if (lfc_ctl->generation == generation)
-	{
-		CriticalAssert(LFC_ENABLED());
-		lfc_ctl->hits += 1;
-		pgBufferUsage.file_cache.hits += 1;
-		CriticalAssert(entry->access_count > 0);
-		if (--entry->access_count == 0)
-			dlist_push_tail(&lfc_ctl->lru, &entry->list_node);
-	}
-	else
-		result = false;
-
-	LWLockRelease(lfc_lock);
-
-	return result;
+	return blocks_read;
 }

 /*
@@ -640,20 +815,17 @@ lfc_read(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
 * If cache is full then evict some other page.
 */
 void
-#if PG_MAJORVERSION_NUM < 16
-lfc_write(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno, char *buffer)
-#else
-lfc_write(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno, const void *buffer)
-#endif
+lfc_writev(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
+		   const void *const *buffers, BlockNumber nblocks)
 {
 	BufferTag	tag;
 	FileCacheEntry *entry;
 	ssize_t		rc;
 	bool		found;
-	int			chunk_offs = blkno & (BLOCKS_PER_CHUNK - 1);
 	uint32		hash;
 	uint64		generation;
 	uint32		entry_offset;
+	int			buf_offset = 0;

 	if (lfc_maybe_disabled())	/* fast exit if file cache is disabled */
 		return;
@@ -661,110 +833,142 @@ lfc_write(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno, const void
 	if (!lfc_ensure_opened())
 		return;

-	tag.forkNum = forkNum;
-	tag.blockNum = blkno & ~(BLOCKS_PER_CHUNK - 1);
 	CopyNRelFileInfoToBufTag(tag, rinfo);
+	tag.forkNum = forkNum;

 	CriticalAssert(BufTagGetRelNumber(&tag) != InvalidRelFileNumber);
-	hash = get_hash_value(lfc_hash, &tag);

-	LWLockAcquire(lfc_lock, LW_EXCLUSIVE);
-
-	if (!LFC_ENABLED())
+	/* 
+	 * For every chunk that has blocks we're interested in, we
+	 * 1. get the chunk header
+	 * 2. Check if the chunk actually has the blocks we're interested in
+	 * 3. Read the blocks we're looking for (in one preadv), assuming they exist
+	 * 4. Update the statistics for the read call.
+	 *
+	 * If there is an error, we do an early return.
+	 */
+	while (nblocks > 0)
 	{
-		LWLockRelease(lfc_lock);
-		return;
-	}
+		struct iovec iov[PG_IOV_MAX];
+		int		chunk_offs = blkno & (BLOCKS_PER_CHUNK - 1);
+		int		blocks_in_chunk = Min(nblocks, BLOCKS_PER_CHUNK - (blkno % BLOCKS_PER_CHUNK));
+		Assert(blocks_in_chunk > 0);

-	entry = hash_search_with_hash_value(lfc_hash, &tag, hash, HASH_ENTER, &found);
-
-	if (found)
-	{
-		/*
-		 * Unlink entry from LRU list to pin it for the duration of IO
-		 * operation
-		 */
-		if (entry->access_count++ == 0)
-			dlist_delete(&entry->list_node);
-	}
-	else
-	{
-		/*
-		 * We have two choices if all cache pages are pinned (i.e. used in IO
-		 * operations):
-		 *
-		 * 1) Wait until some of this operation is completed and pages is
-		 * unpinned.
-		 *
-		 * 2) Allocate one more chunk, so that specified cache size is more
-		 * recommendation than hard limit.
-		 *
-		 * As far as probability of such event (that all pages are pinned) is
-		 * considered to be very very small: there are should be very large
-		 * number of concurrent IO operations and them are limited by
-		 * max_connections, we prefer not to complicate code and use second
-		 * approach.
-		 */
-		if (lfc_ctl->used >= lfc_ctl->limit && !dlist_is_empty(&lfc_ctl->lru))
+		for (int i = 0; i < blocks_in_chunk; i++)
 		{
-			/* Cache overflow: evict least recently used chunk */
-			FileCacheEntry *victim = dlist_container(FileCacheEntry, list_node, dlist_pop_head_node(&lfc_ctl->lru));
-
-			CriticalAssert(victim->access_count == 0);
-			entry->offset = victim->offset; /* grab victim's chunk */
-			hash_search_with_hash_value(lfc_hash, &victim->key, victim->hash, HASH_REMOVE, NULL);
-			neon_log(DEBUG2, "Swap file cache page");
+			iov[i].iov_base = unconstify(void *, buffers[buf_offset + i]);
+			iov[i].iov_len = BLCKSZ;
 		}
-		else if (!dlist_is_empty(&lfc_ctl->holes))
+
+		tag.blockNum = blkno & ~(BLOCKS_PER_CHUNK - 1);
+		hash = get_hash_value(lfc_hash, &tag);
+
+		LWLockAcquire(lfc_lock, LW_EXCLUSIVE);
+
+		if (!LFC_ENABLED())
 		{
-			/* We can reuse a hole that was left behind when the LFC was shrunk previously */
-			FileCacheEntry *hole = dlist_container(FileCacheEntry, list_node, dlist_pop_head_node(&lfc_ctl->holes));
-			uint32		offset = hole->offset;
-			bool		found;
+			LWLockRelease(lfc_lock);
+			return;
+		}

-			hash_search_with_hash_value(lfc_hash, &hole->key, hole->hash, HASH_REMOVE, &found);
-			CriticalAssert(found);
+		entry = hash_search_with_hash_value(lfc_hash, &tag, hash, HASH_ENTER, &found);

-			lfc_ctl->used += 1;
-			entry->offset = offset;	/* reuse the hole */
+		if (found)
+		{
+			/*
+			 * Unlink entry from LRU list to pin it for the duration of IO
+			 * operation
+			 */
+			if (entry->access_count++ == 0)
+				dlist_delete(&entry->list_node);
 		}
 		else
 		{
-			lfc_ctl->used += 1;
-			entry->offset = lfc_ctl->size++;	/* allocate new chunk at end
-												 * of file */
-		}
-		entry->access_count = 1;
-		entry->hash = hash;
-		memset(entry->bitmap, 0, sizeof entry->bitmap);
-	}
-
-	generation = lfc_ctl->generation;
-	entry_offset = entry->offset;
-	lfc_ctl->writes += 1;
-	LWLockRelease(lfc_lock);
-
-	rc = pwrite(lfc_desc, buffer, BLCKSZ, ((off_t) entry_offset * BLOCKS_PER_CHUNK + chunk_offs) * BLCKSZ);
-	if (rc != BLCKSZ)
-	{
-		lfc_disable("write");
-	}
-	else
-	{
-		LWLockAcquire(lfc_lock, LW_EXCLUSIVE);
-
-		if (lfc_ctl->generation == generation)
-		{
-			CriticalAssert(LFC_ENABLED());
-			/* Place entry to the head of LRU list */
-			CriticalAssert(entry->access_count > 0);
-			if (--entry->access_count == 0)
-				dlist_push_tail(&lfc_ctl->lru, &entry->list_node);
-
-			entry->bitmap[chunk_offs >> 5] |= (1 << (chunk_offs & 31));
+			/*
+			 * We have two choices if all cache pages are pinned (i.e. used in IO
+			 * operations):
+			 *
+			 * 1) Wait until some of this operation is completed and pages is
+			 * unpinned.
+			 *
+			 * 2) Allocate one more chunk, so that specified cache size is more
+			 * recommendation than hard limit.
+			 *
+			 * As far as probability of such event (that all pages are pinned) is
+			 * considered to be very very small: there are should be very large
+			 * number of concurrent IO operations and them are limited by
+			 * max_connections, we prefer not to complicate code and use second
+			 * approach.
+			 */
+			if (lfc_ctl->used >= lfc_ctl->limit && !dlist_is_empty(&lfc_ctl->lru))
+			{
+				/* Cache overflow: evict least recently used chunk */
+				FileCacheEntry *victim = dlist_container(FileCacheEntry, list_node, dlist_pop_head_node(&lfc_ctl->lru));
+	
+				CriticalAssert(victim->access_count == 0);
+				entry->offset = victim->offset; /* grab victim's chunk */
+				hash_search_with_hash_value(lfc_hash, &victim->key, victim->hash, HASH_REMOVE, NULL);
+				neon_log(DEBUG2, "Swap file cache page");
+			}
+			else if (!dlist_is_empty(&lfc_ctl->holes))
+			{
+				/* We can reuse a hole that was left behind when the LFC was shrunk previously */
+				FileCacheEntry *hole = dlist_container(FileCacheEntry, list_node, dlist_pop_head_node(&lfc_ctl->holes));
+				uint32		offset = hole->offset;
+				bool		found;
+	
+				hash_search_with_hash_value(lfc_hash, &hole->key, hole->hash, HASH_REMOVE, &found);
+				CriticalAssert(found);
+	
+				lfc_ctl->used += 1;
+				entry->offset = offset;	/* reuse the hole */
+			}
+			else
+			{
+				lfc_ctl->used += 1;
+				entry->offset = lfc_ctl->size++;	/* allocate new chunk at end
+													 * of file */
+			}
+			entry->access_count = 1;
+			entry->hash = hash;
+			memset(entry->bitmap, 0, sizeof entry->bitmap);
 		}

+		generation = lfc_ctl->generation;
+		entry_offset = entry->offset;
+		lfc_ctl->writes += blocks_in_chunk;
 		LWLockRelease(lfc_lock);
+
+		rc = pwritev(lfc_desc, iov, blocks_in_chunk,
+					 ((off_t) entry_offset * BLOCKS_PER_CHUNK + chunk_offs) * BLCKSZ);
+		if (rc != BLCKSZ * blocks_in_chunk)
+		{
+			lfc_disable("write");
+		}
+		else
+		{
+			LWLockAcquire(lfc_lock, LW_EXCLUSIVE);
+
+			if (lfc_ctl->generation == generation)
+			{
+				CriticalAssert(LFC_ENABLED());
+				/* Place entry to the head of LRU list */
+				CriticalAssert(entry->access_count > 0);
+				if (--entry->access_count == 0)
+					dlist_push_tail(&lfc_ctl->lru, &entry->list_node);
+
+				for (int i = 0; i < blocks_in_chunk; i++)
+				{
+					entry->bitmap[(chunk_offs + i) >> 5] |=
+						(1 << ((chunk_offs + i) & 31));
+				}
+			}
+
+			LWLockRelease(lfc_lock);
+		}
+		blkno += blocks_in_chunk;
+		buf_offset += blocks_in_chunk;
+		nblocks -= blocks_in_chunk;
 	}
 }

--- a/pgxn/neon/libpagestore.c
+++ b/pgxn/neon/libpagestore.c
@@ -537,7 +537,11 @@ pageserver_connect(shardno_t shard_no, int elevel)
 		/* No more polling needed; connection succeeded */
 		shard->last_connect_time = GetCurrentTimestamp();

+#if PG_MAJORVERSION_NUM >= 17
+		shard->wes_read = CreateWaitEventSet(NULL, 3);
+#else
 		shard->wes_read = CreateWaitEventSet(TopMemoryContext, 3);
+#endif
 		AddWaitEventToSet(shard->wes_read, WL_LATCH_SET, PGINVALID_SOCKET,
 						  MyLatch, NULL);
 		AddWaitEventToSet(shard->wes_read, WL_EXIT_ON_PM_DEATH, PGINVALID_SOCKET,
--- a/pgxn/neon/neon_pgversioncompat.h
+++ b/pgxn/neon/neon_pgversioncompat.h
@@ -6,7 +6,11 @@
 #ifndef NEON_PGVERSIONCOMPAT_H
 #define NEON_PGVERSIONCOMPAT_H

+#if PG_MAJORVERSION_NUM < 17
 #define NRelFileInfoBackendIsTemp(rinfo) (rinfo.backend != InvalidBackendId)
+#else
+#define NRelFileInfoBackendIsTemp(rinfo) (rinfo.backend != INVALID_PROC_NUMBER)
+#endif

 #define RelFileInfoEquals(a, b) ( \
 	NInfoGetSpcOid(a) == NInfoGetSpcOid(b) && \
@@ -50,7 +54,7 @@
 #define CopyNRelFileInfoToBufTag(tag, rinfo) \
 	do { \
 		(tag).rnode = (rinfo); \
-	} while (false);
+	} while (false)

 #define BufTagGetNRelFileInfo(tag) tag.rnode

@@ -98,7 +102,7 @@
 		(tag).spcOid = (rinfo).spcOid; \
 		(tag).dbOid = (rinfo).dbOid; \
 		(tag).relNumber = (rinfo).relNumber; \
-	} while (false);
+	} while (false)

 #define BufTagGetNRelFileInfo(tag) \
 	((RelFileLocator) { \
@@ -113,4 +117,10 @@
 #define DropRelationAllLocalBuffers DropRelationAllLocalBuffers
 #endif

+#if PG_MAJORVERSION_NUM < 17
+#define ProcNumber BackendId
+#define INVALID_PROC_NUMBER InvalidBackendId
+#define AmAutoVacuumWorkerProcess() (IsAutoVacuumWorkerProcess())
+#endif
+
 #endif							/* NEON_PGVERSIONCOMPAT_H */
--- a/pgxn/neon/pagestore_client.h
+++ b/pgxn/neon/pagestore_client.h
@@ -6,8 +6,6 @@
 * Portions Copyright (c) 1996-2021, PostgreSQL Global Development Group
 * Portions Copyright (c) 1994, Regents of the University of California
 *
- * contrib/neon/pagestore_client.h
- *
 *-------------------------------------------------------------------------
 */
 #ifndef pageserver_h
@@ -187,7 +185,7 @@ extern char *nm_to_string(NeonMessage *msg);
 * API
 */

-typedef unsigned shardno_t;
+typedef uint16 shardno_t;

 typedef struct
 {
@@ -211,7 +209,7 @@ extern int  neon_protocol_version;

 extern shardno_t get_shard_number(BufferTag* tag);

-extern const f_smgr *smgr_neon(BackendId backend, NRelFileInfo rinfo);
+extern const f_smgr *smgr_neon(ProcNumber backend, NRelFileInfo rinfo);
 extern void smgr_init_neon(void);
 extern void readahead_buffer_resize(int newsize, void *extra);

@@ -233,8 +231,13 @@ extern void neon_zeroextend(SMgrRelation reln, ForkNumber forknum,
 							BlockNumber blocknum, int nbuffers, bool skipFsync);
 #endif

+#if PG_MAJORVERSION_NUM >=17
+extern bool neon_prefetch(SMgrRelation reln, ForkNumber forknum,
+						  BlockNumber blocknum, int nblocks);
+#else
 extern bool neon_prefetch(SMgrRelation reln, ForkNumber forknum,
 						  BlockNumber blocknum);
+#endif

 /*
 * LSN values associated with each request to the pageserver
@@ -269,19 +272,11 @@ typedef struct
 } neon_request_lsns;

 #if PG_MAJORVERSION_NUM < 16
-extern void neon_read(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum,
-					  char *buffer);
 extern PGDLLEXPORT void neon_read_at_lsn(NRelFileInfo rnode, ForkNumber forkNum, BlockNumber blkno,
 										 neon_request_lsns request_lsns, char *buffer);
-extern void neon_write(SMgrRelation reln, ForkNumber forknum,
-					   BlockNumber blocknum, char *buffer, bool skipFsync);
 #else
-extern void neon_read(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum,
-					  void *buffer);
 extern PGDLLEXPORT void neon_read_at_lsn(NRelFileInfo rnode, ForkNumber forkNum, BlockNumber blkno,
 										 neon_request_lsns request_lsns, void *buffer);
-extern void neon_write(SMgrRelation reln, ForkNumber forknum,
-					   BlockNumber blocknum, const void *buffer, bool skipFsync);
 #endif
 extern void neon_writeback(SMgrRelation reln, ForkNumber forknum,
 						   BlockNumber blocknum, BlockNumber nblocks);
@@ -299,17 +294,34 @@ extern void update_cached_relsize(NRelFileInfo rinfo, ForkNumber forknum, BlockN
 extern void forget_cached_relsize(NRelFileInfo rinfo, ForkNumber forknum);

 /* functions for local file cache */
-#if PG_MAJORVERSION_NUM < 16
-extern void lfc_write(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
-					  char *buffer);
-#else
-extern void lfc_write(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
-					  const void *buffer);
-#endif
-extern bool lfc_read(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno, char *buffer);
-extern bool lfc_cache_contains(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno);
+extern void lfc_writev(NRelFileInfo rinfo, ForkNumber forkNum,
+					   BlockNumber blkno, const void *const *buffers,
+					   BlockNumber nblocks);
+/* returns number of blocks read, with one bit set in *read for each  */
+extern int lfc_readv_select(NRelFileInfo rinfo, ForkNumber forkNum,
+							BlockNumber blkno, void **buffers,
+							BlockNumber nblocks, bits8 *mask);
+
+extern bool lfc_cache_contains(NRelFileInfo rinfo, ForkNumber forkNum,
+							   BlockNumber blkno);
+extern int lfc_cache_containsv(NRelFileInfo rinfo, ForkNumber forkNum,
+							   BlockNumber blkno, int nblocks, bits8 *bitmap);
 extern void lfc_evict(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno);
 extern void lfc_init(void);

+static inline bool
+lfc_read(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
+		 void *buffer)
+{
+	bits8		rv = 0;
+	return lfc_readv_select(rinfo, forkNum, blkno, &buffer, 1, &rv) == 1;
+}
+
+static inline void
+lfc_write(NRelFileInfo rinfo, ForkNumber forkNum, BlockNumber blkno,
+		  const void *buffer)
+{
+	return lfc_writev(rinfo, forkNum, blkno, &buffer, 1);
+}

 #endif
--- a/pgxn/neon/pagestore_smgr.c
+++ b/pgxn/neon/pagestore_smgr.c
--- a/pgxn/neon/walproposer_pg.c
+++ b/pgxn/neon/walproposer_pg.c
@@ -81,6 +81,7 @@ static void nwp_register_gucs(void);
 static void assign_neon_safekeepers(const char *newval, void *extra);
 static void nwp_prepare_shmem(void);
 static uint64 backpressure_lag_impl(void);
+static uint64 startup_backpressure_wrap(void);
 static bool backpressure_throttling_impl(void);
 static void walprop_register_bgworker(void);

@@ -90,7 +91,7 @@ static void walprop_pg_init_bgworker(void);
 static TimestampTz walprop_pg_get_current_timestamp(WalProposer *wp);
 static void walprop_pg_load_libpqwalreceiver(void);

-static process_interrupts_callback_t PrevProcessInterruptsCallback;
+static process_interrupts_callback_t PrevProcessInterruptsCallback = NULL;
 static shmem_startup_hook_type prev_shmem_startup_hook_type;
 #if PG_VERSION_NUM >= 150000
 static shmem_request_hook_type prev_shmem_request_hook = NULL;
@@ -178,7 +179,7 @@ pg_init_walproposer(void)

 	nwp_prepare_shmem();

-	delay_backend_us = &backpressure_lag_impl;
+	delay_backend_us = &startup_backpressure_wrap;
 	PrevProcessInterruptsCallback = ProcessInterruptsCallback;
 	ProcessInterruptsCallback = backpressure_throttling_impl;

@@ -352,6 +353,22 @@ backpressure_lag_impl(void)
 	return 0;
 }

+/*
+ * We don't apply backpressure when we're the postmaster, or the startup
+ * process, because in postmaster we can't apply backpressure, and in
+ * the startup process we can't afford to slow down.
+ */
+static uint64
+startup_backpressure_wrap(void)
+{
+	if (AmStartupProcess() || !IsUnderPostmaster)
+		return 0;
+
+	delay_backend_us = &backpressure_lag_impl;
+
+	return backpressure_lag_impl();
+}
+
 /*
 * WalproposerShmemSize --- report amount of shared memory space needed
 */
@@ -401,12 +418,13 @@ WalproposerShmemInit_SyncSafekeeper(void)
 static bool
 backpressure_throttling_impl(void)
 {
-	int64		lag;
+	uint64		lag;
 	TimestampTz start,
 				stop;
-	bool		retry = PrevProcessInterruptsCallback
-		? PrevProcessInterruptsCallback()
-		: false;
+	bool		retry = false;
+
+	if (PointerIsValid(PrevProcessInterruptsCallback))
+		retry = PrevProcessInterruptsCallback();

 	/*
 	 * Don't throttle read only transactions or wal sender. Do throttle CREATE
@@ -602,7 +620,12 @@ walprop_pg_init_walsender(void)
 	/* Create replication slot for WAL proposer if not exists */
 	if (SearchNamedReplicationSlot(WAL_PROPOSER_SLOT_NAME, false) == NULL)
 	{
+#if PG_MAJORVERSION_NUM >= 17
+		ReplicationSlotCreate(WAL_PROPOSER_SLOT_NAME, false, RS_PERSISTENT,
+							  false, false, false);
+#else
 		ReplicationSlotCreate(WAL_PROPOSER_SLOT_NAME, false, RS_PERSISTENT, false);
+#endif
 		ReplicationSlotReserveWal();
 		/* Write this slot to disk */
 		ReplicationSlotMarkDirty();
@@ -1509,7 +1532,11 @@ walprop_pg_init_event_set(WalProposer *wp)
 		wpg_log(FATAL, "double-initialization of event set");

 	/* for each sk, we have socket plus potentially socket for neon walreader */
+#if PG_MAJORVERSION_NUM >= 17
+	waitEvents = CreateWaitEventSet(NULL, 2 + 2 * wp->n_safekeepers);
+#else
 	waitEvents = CreateWaitEventSet(TopMemoryContext, 2 + 2 * wp->n_safekeepers);
+#endif
 	AddWaitEventToSet(waitEvents, WL_LATCH_SET, PGINVALID_SOCKET,
 					  MyLatch, NULL);
 	AddWaitEventToSet(waitEvents, WL_EXIT_ON_PM_DEATH, PGINVALID_SOCKET,
--- a/pgxn/neon_rmgr/neon_rmgr_decode.c
+++ b/pgxn/neon_rmgr/neon_rmgr_decode.c
@@ -1,6 +1,7 @@
 #include "postgres.h"

 #if PG_MAJORVERSION_NUM >= 16
+
 #include "access/heapam_xlog.h"
 #include "access/neon_xlog.h"
 #include "replication/decode.h"
@@ -9,6 +10,10 @@

 #include "neon_rmgr.h"

+#endif /* PG >= 16 */
+
+#if PG_MAJORVERSION_NUM == 16
+
 /* individual record(group)'s handlers */
 static void DecodeNeonInsert(LogicalDecodingContext *ctx, XLogRecordBuffer *buf);
 static void DecodeNeonUpdate(LogicalDecodingContext *ctx, XLogRecordBuffer *buf);
@@ -399,6 +404,398 @@ DecodeXLogTuple(char *data, Size len, ReorderBufferTupleBuf *tuple)
 	header->t_infomask2 = xlhdr.t_infomask2;
 	header->t_hoff = xlhdr.t_hoff;
 }
+#endif
+
+#if PG_MAJORVERSION_NUM == 17
+
+/* individual record(group)'s handlers */
+static void DecodeNeonInsert(LogicalDecodingContext *ctx, XLogRecordBuffer *buf);
+static void DecodeNeonUpdate(LogicalDecodingContext *ctx, XLogRecordBuffer *buf);
+static void DecodeNeonDelete(LogicalDecodingContext *ctx, XLogRecordBuffer *buf);
+static void DecodeNeonMultiInsert(LogicalDecodingContext *ctx, XLogRecordBuffer *buf);
+
+/* common function to decode tuples */
+static void DecodeXLogTuple(char *data, Size len, HeapTuple tuple);


-#endif
+void
+neon_rm_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
+{
+	uint8		info = XLogRecGetInfo(buf->record) & XLOG_NEON_OPMASK;
+	TransactionId xid = XLogRecGetXid(buf->record);
+	SnapBuild  *builder = ctx->snapshot_builder;
+
+	ReorderBufferProcessXid(ctx->reorder, xid, buf->origptr);
+
+	/*
+	 * If we don't have snapshot or we are just fast-forwarding, there is no
+	 * point in decoding data changes.
+	 */
+	if (SnapBuildCurrentState(builder) < SNAPBUILD_FULL_SNAPSHOT ||
+		ctx->fast_forward)
+		return;
+
+	switch (info)
+	{
+		case XLOG_NEON_HEAP_INSERT:
+			if (SnapBuildProcessChange(builder, xid, buf->origptr))
+				DecodeNeonInsert(ctx, buf);
+			break;
+		case XLOG_NEON_HEAP_DELETE:
+			if (SnapBuildProcessChange(builder, xid, buf->origptr))
+				DecodeNeonDelete(ctx, buf);
+			break;
+		case XLOG_NEON_HEAP_UPDATE:
+		case XLOG_NEON_HEAP_HOT_UPDATE:
+			if (SnapBuildProcessChange(builder, xid, buf->origptr))
+				DecodeNeonUpdate(ctx, buf);
+			break;
+		case XLOG_NEON_HEAP_LOCK:
+			break;
+		case XLOG_NEON_HEAP_MULTI_INSERT:
+			if (SnapBuildProcessChange(builder, xid, buf->origptr))
+				DecodeNeonMultiInsert(ctx, buf);
+			break;
+		default:
+			elog(ERROR, "unexpected RM_HEAP_ID record type: %u", info);
+			break;
+	}
+}
+
+static inline bool
+FilterByOrigin(LogicalDecodingContext *ctx, RepOriginId origin_id)
+{
+	if (ctx->callbacks.filter_by_origin_cb == NULL)
+		return false;
+
+	return filter_by_origin_cb_wrapper(ctx, origin_id);
+}
+
+/*
+ * Parse XLOG_HEAP_INSERT (not MULTI_INSERT!) records into tuplebufs.
+ *
+ * Deletes can contain the new tuple.
+ */
+static void
+DecodeNeonInsert(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
+{
+	Size		datalen;
+	char	   *tupledata;
+	Size		tuplelen;
+	XLogReaderState *r = buf->record;
+	xl_neon_heap_insert *xlrec;
+	ReorderBufferChange *change;
+	RelFileLocator target_locator;
+
+	xlrec = (xl_neon_heap_insert *) XLogRecGetData(r);
+
+	/*
+	 * Ignore insert records without new tuples (this does happen when
+	 * raw_heap_insert marks the TOAST record as HEAP_INSERT_NO_LOGICAL).
+	 */
+	if (!(xlrec->flags & XLH_INSERT_CONTAINS_NEW_TUPLE))
+		return;
+
+	/* only interested in our database */
+	XLogRecGetBlockTag(r, 0, &target_locator, NULL, NULL);
+	if (target_locator.dbOid != ctx->slot->data.database)
+		return;
+
+	/* output plugin doesn't look for this origin, no need to queue */
+	if (FilterByOrigin(ctx, XLogRecGetOrigin(r)))
+		return;
+
+	change = ReorderBufferGetChange(ctx->reorder);
+	if (!(xlrec->flags & XLH_INSERT_IS_SPECULATIVE))
+		change->action = REORDER_BUFFER_CHANGE_INSERT;
+	else
+		change->action = REORDER_BUFFER_CHANGE_INTERNAL_SPEC_INSERT;
+	change->origin_id = XLogRecGetOrigin(r);
+
+	memcpy(&change->data.tp.rlocator, &target_locator, sizeof(RelFileLocator));
+
+	tupledata = XLogRecGetBlockData(r, 0, &datalen);
+	tuplelen = datalen - SizeOfHeapHeader;
+
+	change->data.tp.newtuple =
+		ReorderBufferGetTupleBuf(ctx->reorder, tuplelen);
+
+	DecodeXLogTuple(tupledata, datalen, change->data.tp.newtuple);
+
+	change->data.tp.clear_toast_afterwards = true;
+
+	ReorderBufferQueueChange(ctx->reorder, XLogRecGetXid(r), buf->origptr,
+							 change,
+							 xlrec->flags & XLH_INSERT_ON_TOAST_RELATION);
+}
+
+/*
+ * Parse XLOG_HEAP_DELETE from wal into proper tuplebufs.
+ *
+ * Deletes can possibly contain the old primary key.
+ */
+static void
+DecodeNeonDelete(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
+{
+	XLogReaderState *r = buf->record;
+	xl_neon_heap_delete *xlrec;
+	ReorderBufferChange *change;
+	RelFileLocator target_locator;
+
+	xlrec = (xl_neon_heap_delete *) XLogRecGetData(r);
+
+	/* only interested in our database */
+	XLogRecGetBlockTag(r, 0, &target_locator, NULL, NULL);
+	if (target_locator.dbOid != ctx->slot->data.database)
+		return;
+
+	/* output plugin doesn't look for this origin, no need to queue */
+	if (FilterByOrigin(ctx, XLogRecGetOrigin(r)))
+		return;
+
+	change = ReorderBufferGetChange(ctx->reorder);
+
+	if (xlrec->flags & XLH_DELETE_IS_SUPER)
+		change->action = REORDER_BUFFER_CHANGE_INTERNAL_SPEC_ABORT;
+	else
+		change->action = REORDER_BUFFER_CHANGE_DELETE;
+
+	change->origin_id = XLogRecGetOrigin(r);
+
+	memcpy(&change->data.tp.rlocator, &target_locator, sizeof(RelFileLocator));
+
+	/* old primary key stored */
+	if (xlrec->flags & XLH_DELETE_CONTAINS_OLD)
+	{
+		Size		datalen = XLogRecGetDataLen(r) - SizeOfNeonHeapHeader;
+		Size		tuplelen = datalen - SizeOfNeonHeapHeader;
+
+		Assert(XLogRecGetDataLen(r) > (SizeOfNeonHeapDelete + SizeOfNeonHeapHeader));
+
+		change->data.tp.oldtuple =
+			ReorderBufferGetTupleBuf(ctx->reorder, tuplelen);
+
+		DecodeXLogTuple((char *) xlrec + SizeOfNeonHeapDelete,
+						datalen, change->data.tp.oldtuple);
+	}
+
+	change->data.tp.clear_toast_afterwards = true;
+
+	ReorderBufferQueueChange(ctx->reorder, XLogRecGetXid(r), buf->origptr,
+							 change, false);
+}
+
+/*
+ * Parse XLOG_HEAP_UPDATE and XLOG_HEAP_HOT_UPDATE, which have the same layout
+ * in the record, from wal into proper tuplebufs.
+ *
+ * Updates can possibly contain a new tuple and the old primary key.
+ */
+static void
+DecodeNeonUpdate(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
+{
+	XLogReaderState *r = buf->record;
+	xl_neon_heap_update *xlrec;
+	ReorderBufferChange *change;
+	char	   *data;
+	RelFileLocator target_locator;
+
+	xlrec = (xl_neon_heap_update *) XLogRecGetData(r);
+
+	/* only interested in our database */
+	XLogRecGetBlockTag(r, 0, &target_locator, NULL, NULL);
+	if (target_locator.dbOid != ctx->slot->data.database)
+		return;
+
+	/* output plugin doesn't look for this origin, no need to queue */
+	if (FilterByOrigin(ctx, XLogRecGetOrigin(r)))
+		return;
+
+	change = ReorderBufferGetChange(ctx->reorder);
+	change->action = REORDER_BUFFER_CHANGE_UPDATE;
+	change->origin_id = XLogRecGetOrigin(r);
+	memcpy(&change->data.tp.rlocator, &target_locator, sizeof(RelFileLocator));
+
+	if (xlrec->flags & XLH_UPDATE_CONTAINS_NEW_TUPLE)
+	{
+		Size		datalen;
+		Size		tuplelen;
+
+		data = XLogRecGetBlockData(r, 0, &datalen);
+
+		tuplelen = datalen - SizeOfNeonHeapHeader;
+
+		change->data.tp.newtuple =
+			ReorderBufferGetTupleBuf(ctx->reorder, tuplelen);
+
+		DecodeXLogTuple(data, datalen, change->data.tp.newtuple);
+	}
+
+	if (xlrec->flags & XLH_UPDATE_CONTAINS_OLD)
+	{
+		Size		datalen;
+		Size		tuplelen;
+
+		/* caution, remaining data in record is not aligned */
+		data = XLogRecGetData(r) + SizeOfNeonHeapUpdate;
+		datalen = XLogRecGetDataLen(r) - SizeOfNeonHeapUpdate;
+		tuplelen = datalen - SizeOfNeonHeapHeader;
+
+		change->data.tp.oldtuple =
+			ReorderBufferGetTupleBuf(ctx->reorder, tuplelen);
+
+		DecodeXLogTuple(data, datalen, change->data.tp.oldtuple);
+	}
+
+	change->data.tp.clear_toast_afterwards = true;
+
+	ReorderBufferQueueChange(ctx->reorder, XLogRecGetXid(r), buf->origptr,
+							 change, false);
+}
+
+/*
+ * Decode XLOG_HEAP2_MULTI_INSERT_insert record into multiple tuplebufs.
+ *
+ * Currently MULTI_INSERT will always contain the full tuples.
+ */
+static void
+DecodeNeonMultiInsert(LogicalDecodingContext *ctx, XLogRecordBuffer *buf)
+{
+	XLogReaderState *r = buf->record;
+	xl_neon_heap_multi_insert *xlrec;
+	int			i;
+	char	   *data;
+	char	   *tupledata;
+	Size		tuplelen;
+	RelFileLocator rlocator;
+
+	xlrec = (xl_neon_heap_multi_insert *) XLogRecGetData(r);
+
+	/*
+	 * Ignore insert records without new tuples.  This happens when a
+	 * multi_insert is done on a catalog or on a non-persistent relation.
+	 */
+	if (!(xlrec->flags & XLH_INSERT_CONTAINS_NEW_TUPLE))
+		return;
+
+	/* only interested in our database */
+	XLogRecGetBlockTag(r, 0, &rlocator, NULL, NULL);
+	if (rlocator.dbOid != ctx->slot->data.database)
+		return;
+
+	/* output plugin doesn't look for this origin, no need to queue */
+	if (FilterByOrigin(ctx, XLogRecGetOrigin(r)))
+		return;
+
+	/*
+	 * We know that this multi_insert isn't for a catalog, so the block should
+	 * always have data even if a full-page write of it is taken.
+	 */
+	tupledata = XLogRecGetBlockData(r, 0, &tuplelen);
+	Assert(tupledata != NULL);
+
+	data = tupledata;
+	for (i = 0; i < xlrec->ntuples; i++)
+	{
+		ReorderBufferChange *change;
+		xl_neon_multi_insert_tuple *xlhdr;
+		int			datalen;
+		HeapTuple	tuple;
+		HeapTupleHeader header;
+
+		change = ReorderBufferGetChange(ctx->reorder);
+		change->action = REORDER_BUFFER_CHANGE_INSERT;
+		change->origin_id = XLogRecGetOrigin(r);
+
+		memcpy(&change->data.tp.rlocator, &rlocator, sizeof(RelFileLocator));
+
+		xlhdr = (xl_neon_multi_insert_tuple *) SHORTALIGN(data);
+		data = ((char *) xlhdr) + SizeOfNeonMultiInsertTuple;
+		datalen = xlhdr->datalen;
+
+		change->data.tp.newtuple =
+			ReorderBufferGetTupleBuf(ctx->reorder, datalen);
+
+		tuple = change->data.tp.newtuple;
+		header = tuple->t_data;
+
+		/* not a disk based tuple */
+		ItemPointerSetInvalid(&tuple->t_self);
+
+		/*
+		 * We can only figure this out after reassembling the transactions.
+		 */
+		tuple->t_tableOid = InvalidOid;
+
+		tuple->t_len = datalen + SizeofHeapTupleHeader;
+
+		memset(header, 0, SizeofHeapTupleHeader);
+
+		memcpy((char *) tuple->t_data + SizeofHeapTupleHeader,
+			   (char *) data,
+			   datalen);
+		header->t_infomask = xlhdr->t_infomask;
+		header->t_infomask2 = xlhdr->t_infomask2;
+		header->t_hoff = xlhdr->t_hoff;
+
+		/*
+		 * Reset toast reassembly state only after the last row in the last
+		 * xl_multi_insert_tuple record emitted by one heap_multi_insert()
+		 * call.
+		 */
+		if (xlrec->flags & XLH_INSERT_LAST_IN_MULTI &&
+			(i + 1) == xlrec->ntuples)
+			change->data.tp.clear_toast_afterwards = true;
+		else
+			change->data.tp.clear_toast_afterwards = false;
+
+		ReorderBufferQueueChange(ctx->reorder, XLogRecGetXid(r),
+								 buf->origptr, change, false);
+
+		/* move to the next xl_neon_multi_insert_tuple entry */
+		data += datalen;
+	}
+	Assert(data == tupledata + tuplelen);
+}
+
+/*
+ * Read a HeapTuple as WAL logged by heap_insert, heap_update and heap_delete
+ * (but not by heap_multi_insert) into a tuplebuf.
+ *
+ * The size 'len' and the pointer 'data' in the record need to be
+ * computed outside as they are record specific.
+ */
+static void
+DecodeXLogTuple(char *data, Size len, HeapTuple tuple)
+{
+	xl_neon_heap_header xlhdr;
+	int			datalen = len - SizeOfNeonHeapHeader;
+	HeapTupleHeader header;
+
+	Assert(datalen >= 0);
+
+	tuple->t_len = datalen + SizeofHeapTupleHeader;
+	header = tuple->t_data;
+
+	/* not a disk based tuple */
+	ItemPointerSetInvalid(&tuple->t_self);
+
+	/* we can only figure this out after reassembling the transactions */
+	tuple->t_tableOid = InvalidOid;
+
+	/* data is not stored aligned, copy to aligned storage */
+	memcpy((char *) &xlhdr,
+		   data,
+		   SizeOfNeonHeapHeader);
+
+	memset(header, 0, SizeofHeapTupleHeader);
+
+	memcpy(((char *) tuple->t_data) + SizeofHeapTupleHeader,
+		   data + SizeOfNeonHeapHeader,
+		   datalen);
+
+	header->t_infomask = xlhdr.t_infomask;
+	header->t_infomask2 = xlhdr.t_infomask2;
+	header->t_hoff = xlhdr.t_hoff;
+}
+#endif
--- a/pgxn/neon_walredo/inmem_smgr.c
+++ b/pgxn/neon_walredo/inmem_smgr.c
@@ -68,8 +68,13 @@ static void inmem_close(SMgrRelation reln, ForkNumber forknum);
 static void inmem_create(SMgrRelation reln, ForkNumber forknum, bool isRedo);
 static bool inmem_exists(SMgrRelation reln, ForkNumber forknum);
 static void inmem_unlink(NRelFileInfoBackend rinfo, ForkNumber forknum, bool isRedo);
+#if PG_MAJORVERSION_NUM >= 17
+static bool inmem_prefetch(SMgrRelation reln, ForkNumber forknum,
+						   BlockNumber blocknum, int nblocks);
+#else
 static bool inmem_prefetch(SMgrRelation reln, ForkNumber forknum,
 						   BlockNumber blocknum);
+#endif
 #if PG_MAJORVERSION_NUM < 16
 static void inmem_extend(SMgrRelation reln, ForkNumber forknum,
 						 BlockNumber blocknum, char *buffer, bool skipFsync);
@@ -93,7 +98,9 @@ static BlockNumber inmem_nblocks(SMgrRelation reln, ForkNumber forknum);
 static void inmem_truncate(SMgrRelation reln, ForkNumber forknum,
 						   BlockNumber nblocks);
 static void inmem_immedsync(SMgrRelation reln, ForkNumber forknum);
-
+#if PG_MAJORVERSION_NUM >= 17
+static void inmem_registersync(SMgrRelation reln, ForkNumber forknum);
+#endif

 /*
 *	inmem_init() -- Initialize private state
@@ -190,6 +197,14 @@ inmem_close(SMgrRelation reln, ForkNumber forknum)
 {
 }

+#if PG_MAJORVERSION_NUM >= 17
+static bool
+inmem_prefetch(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum,
+			   int nblocks)
+{
+	return true;
+}
+#else
 /*
 *	inmem_prefetch() -- Initiate asynchronous read of the specified block of a relation
 */
@@ -198,6 +213,7 @@ inmem_prefetch(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum)
 {
 	return true;
 }
+#endif

 /*
 * inmem_writeback() -- Tell the kernel to write pages back to storage.
@@ -211,11 +227,13 @@ inmem_writeback(SMgrRelation reln, ForkNumber forknum,
 /*
 *	inmem_read() -- Read the specified block from a relation.
 */
+#if PG_MAJORVERSION_NUM < 16
 static void
 inmem_read(SMgrRelation reln, ForkNumber forknum, BlockNumber blkno,
-#if PG_MAJORVERSION_NUM < 16
 		   char *buffer)
 #else
+static void
+inmem_read(SMgrRelation reln, ForkNumber forknum, BlockNumber blkno,
 		   void *buffer)
 #endif
 {
@@ -228,6 +246,18 @@ inmem_read(SMgrRelation reln, ForkNumber forknum, BlockNumber blkno,
 		memcpy(buffer, page_body[pg], BLCKSZ);
 }

+#if PG_MAJORVERSION_NUM >= 17
+static void
+inmem_readv(SMgrRelation reln, ForkNumber forknum, BlockNumber blkno,
+			void **buffers, BlockNumber nblocks)
+{
+	for (int i = 0; i < nblocks; i++)
+	{
+		inmem_read(reln, forknum, blkno, buffers[i]);
+	}
+}
+#endif
+
 /*
 *	inmem_write() -- Write the supplied block at the appropriate location.
 *
@@ -280,6 +310,18 @@ inmem_write(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum,
 	memcpy(page_body[pg], buffer, BLCKSZ);
 }

+#if PG_MAJORVERSION_NUM >= 17
+static void
+inmem_writev(SMgrRelation reln, ForkNumber forknum, BlockNumber blkno,
+			 const void **buffers, BlockNumber nblocks, bool skipFsync)
+{
+	for (int i = 0; i < nblocks; i++)
+	{
+		inmem_write(reln, forknum, blkno, buffers[i], skipFsync);
+	}
+}
+#endif
+
 /*
 *	inmem_nblocks() -- Get the number of blocks stored in a relation.
 */
@@ -315,6 +357,13 @@ inmem_immedsync(SMgrRelation reln, ForkNumber forknum)
 {
 }

+#if PG_MAJORVERSION_NUM >= 17
+static void
+inmem_registersync(SMgrRelation reln, ForkNumber forknum)
+{
+}
+#endif
+
 static const struct f_smgr inmem_smgr =
 {
 	.smgr_init = inmem_init,
@@ -328,23 +377,39 @@ static const struct f_smgr inmem_smgr =
 #if PG_MAJORVERSION_NUM >= 16
 	.smgr_zeroextend = inmem_zeroextend,
 #endif
+#if PG_MAJORVERSION_NUM >= 17
+	.smgr_prefetch = inmem_prefetch,
+	.smgr_readv = inmem_readv,
+	.smgr_writev = inmem_writev,
+#else
 	.smgr_prefetch = inmem_prefetch,
 	.smgr_read = inmem_read,
 	.smgr_write = inmem_write,
+#endif
 	.smgr_writeback = inmem_writeback,
 	.smgr_nblocks = inmem_nblocks,
 	.smgr_truncate = inmem_truncate,
 	.smgr_immedsync = inmem_immedsync,
+
+#if PG_MAJORVERSION_NUM >= 17
+	.smgr_registersync = inmem_registersync,
+#endif
+
+	.smgr_start_unlogged_build = NULL,
+	.smgr_finish_unlogged_build_phase_1 = NULL,
+	.smgr_end_unlogged_build = NULL,
+	.smgr_read_slru_segment = NULL,
 };

 const f_smgr *
-smgr_inmem(BackendId backend, NRelFileInfo rinfo)
+smgr_inmem(ProcNumber backend, NRelFileInfo rinfo)
 {
 	Assert(InRecovery);
-	if (backend != InvalidBackendId)
-		return smgr_standard(backend, rinfo);
-	else
-		return &inmem_smgr;
+	// // What does this code do?
+	// if (backend != INVALID_PROC_NUMBER)
+	// 	return smgr_standard(backend, rinfo);
+	// else
+	return &inmem_smgr;
 }

 void
--- a/pgxn/neon_walredo/inmem_smgr.h
+++ b/pgxn/neon_walredo/inmem_smgr.h
@@ -11,7 +11,7 @@
 #ifndef INMEM_SMGR_H
 #define INMEM_SMGR_H

-extern const f_smgr *smgr_inmem(BackendId backend, NRelFileInfo rinfo);
+extern const f_smgr *smgr_inmem(ProcNumber backend, NRelFileInfo rinfo);
 extern void smgr_init_inmem(void);

 #endif /* INMEM_SMGR_H */
--- a/pgxn/neon_walredo/walredoproc.c
+++ b/pgxn/neon_walredo/walredoproc.c
@@ -100,6 +100,9 @@
 #include "storage/buf_internals.h"
 #include "storage/bufmgr.h"
 #include "storage/dsm.h"
+#if PG_MAJORVERSION_NUM >= 17
+#include "storage/dsm_registry.h"
+#endif
 #include "storage/ipc.h"
 #include "storage/pg_shmem.h"
 #include "storage/pmsignal.h"
@@ -137,7 +140,7 @@ static BufferTag target_redo_tag;

 static XLogReaderState *reader_state;

-#define TRACE DEBUG5
+#define TRACE LOG

 #ifdef HAVE_LIBSECCOMP

@@ -517,6 +520,10 @@ CreateFakeSharedMemoryAndSemaphores()
 	/*
 	 * Set up xlog, clog, and buffers
 	 */
+#if PG_MAJORVERSION_NUM >= 17
+	DSMRegistryShmemInit();
+	VarsupShmemInit();
+#endif
 	XLOGShmemInit();
 	CLOGShmemInit();
 	CommitTsShmemInit();
@@ -566,7 +573,10 @@ CreateFakeSharedMemoryAndSemaphores()
 	/*
 	 * Set up other modules that need some shared memory space
 	 */
+#if PG_MAJORVERSION_NUM < 17
+	/* "snapshot too old" was removed in PG17, and with it the SnapMgr */
 	SnapMgrInit();
+#endif
 	BTreeShmemInit();
 	SyncScanShmemInit();
 	/* Skip due to the 'pg_notify' directory check */
@@ -742,7 +752,7 @@ BeginRedoForBlock(StringInfo input_message)
 		 target_redo_tag.forkNum,
 		 target_redo_tag.blockNum);

-	reln = smgropen(rinfo, InvalidBackendId, RELPERSISTENCE_PERMANENT);
+	reln = smgropen(rinfo, INVALID_PROC_NUMBER, RELPERSISTENCE_PERMANENT);
 	if (reln->smgr_cached_nblocks[forknum] == InvalidBlockNumber ||
 		reln->smgr_cached_nblocks[forknum] < blknum + 1)
 	{
--- a/proxy/Cargo.toml
+++ b/proxy/Cargo.toml
@@ -18,7 +18,6 @@ atomic-take.workspace = true
 aws-config.workspace = true
 aws-sdk-iam.workspace = true
 aws-sigv4.workspace = true
-aws-types.workspace = true
 base64.workspace = true
 bstr.workspace = true
 bytes = { workspace = true, features = ["serde"] }
@@ -26,7 +25,6 @@ camino.workspace = true
 chrono.workspace = true
 clap.workspace = true
 consumption_metrics.workspace = true
-crossbeam-deque.workspace = true
 dashmap.workspace = true
 env_logger.workspace = true
 framed-websockets.workspace = true
@@ -48,11 +46,9 @@ indexmap.workspace = true
 ipnet.workspace = true
 itertools.workspace = true
 lasso = { workspace = true, features = ["multi-threaded"] }
-md5.workspace = true
 measured = { workspace = true, features = ["lasso"] }
 metrics.workspace = true
 once_cell.workspace = true
-opentelemetry.workspace = true
 parking_lot.workspace = true
 parquet.workspace = true
 parquet_derive.workspace = true
@@ -67,7 +63,6 @@ reqwest.workspace = true
 reqwest-middleware = { workspace = true, features = ["json"] }
 reqwest-retry.workspace = true
 reqwest-tracing.workspace = true
-routerify.workspace = true
 rustc-hash.workspace = true
 rustls-pemfile.workspace = true
 rustls.workspace = true
@@ -79,7 +74,6 @@ smol_str.workspace = true
 smallvec.workspace = true
 socket2.workspace = true
 subtle.workspace = true
-task-local-extensions.workspace = true
 thiserror.workspace = true
 tikv-jemallocator.workspace = true
 tikv-jemalloc-ctl = { workspace = true, features = ["use_std"] }
@@ -88,7 +82,6 @@ tokio-postgres-rustls.workspace = true
 tokio-rustls.workspace = true
 tokio-util.workspace = true
 tokio = { workspace = true, features = ["signal"] }
-tower-service.workspace = true
 tracing-opentelemetry.workspace = true
 tracing-subscriber.workspace = true
 tracing-utils.workspace = true
--- a/proxy/src/auth/backend.rs
+++ b/proxy/src/auth/backend.rs
@@ -311,7 +311,9 @@ async fn auth_quirks(
    let (allowed_ips, maybe_secret) = api.get_allowed_ips_and_secret(ctx, &info).await?;

    // check allowed list
-    if !check_peer_addr_is_in_list(&ctx.peer_addr(), &allowed_ips) {
+    if config.ip_allowlist_check_enabled
+        && !check_peer_addr_is_in_list(&ctx.peer_addr(), &allowed_ips)
+    {
        return Err(auth::AuthError::ip_address_not_allowed(ctx.peer_addr()));
    }

@@ -603,6 +605,7 @@ mod tests {
        rate_limiter_enabled: true,
        rate_limiter: AuthRateLimiter::new(&RateBucketInfo::DEFAULT_AUTH_SET),
        rate_limit_ip_subnet: 64,
+        ip_allowlist_check_enabled: true,
    });

    async fn read_message(r: &mut (impl AsyncRead + Unpin), b: &mut BytesMut) -> PgMessage {
--- a/proxy/src/auth/credentials.rs
+++ b/proxy/src/auth/credentials.rs
@@ -538,4 +538,17 @@ mod tests {
        ));
        Ok(())
    }
+
+    #[test]
+    fn test_connection_blocker() {
+        fn check(v: serde_json::Value) -> bool {
+            let peer_addr = IpAddr::from([127, 0, 0, 1]);
+            let ip_list: Vec<IpPattern> = serde_json::from_value(v).unwrap();
+            check_peer_addr_is_in_list(&peer_addr, &ip_list)
+        }
+
+        assert!(check(json!([])));
+        assert!(check(json!(["127.0.0.1"])));
+        assert!(!check(json!(["255.255.255.255"])));
+    }
 }
--- a/proxy/src/bin/local_proxy.rs
+++ b/proxy/src/bin/local_proxy.rs
@@ -92,6 +92,12 @@ struct SqlOverHttpArgs {

    #[clap(long, default_value_t = 16)]
    sql_over_http_cancel_set_shards: usize,
+
+    #[clap(long, default_value_t = 10 * 1024 * 1024)] // 10 MiB
+    sql_over_http_max_request_size_bytes: u64,
+
+    #[clap(long, default_value_t = 10 * 1024 * 1024)] // 10 MiB
+    sql_over_http_max_response_size_bytes: usize,
 }

 #[tokio::main]
@@ -208,6 +214,8 @@ fn build_config(args: &LocalProxyCliArgs) -> anyhow::Result<&'static ProxyConfig
        },
        cancel_set: CancelSet::new(args.sql_over_http.sql_over_http_cancel_set_shards),
        client_conn_threshold: args.sql_over_http.sql_over_http_client_conn_threshold,
+        max_request_size_bytes: args.sql_over_http.sql_over_http_max_request_size_bytes,
+        max_response_size_bytes: args.sql_over_http.sql_over_http_max_response_size_bytes,
    };

    Ok(Box::leak(Box::new(ProxyConfig {
@@ -224,6 +232,7 @@ fn build_config(args: &LocalProxyCliArgs) -> anyhow::Result<&'static ProxyConfig
            rate_limiter_enabled: false,
            rate_limiter: BucketRateLimiter::new(vec![]),
            rate_limit_ip_subnet: 64,
+            ip_allowlist_check_enabled: true,
        },
        require_client_ip: false,
        handshake_timeout: Duration::from_secs(10),
--- a/proxy/src/bin/proxy.rs
+++ b/proxy/src/bin/proxy.rs
@@ -224,6 +224,10 @@ struct ProxyCliArgs {
    /// Whether to retry the wake_compute request
    #[clap(long, default_value = config::RetryConfig::WAKE_COMPUTE_DEFAULT_VALUES)]
    wake_compute_retry: String,
+
+    /// Configure if this is a private access proxy for the POC: In that case the proxy will ignore the IP allowlist
+    #[clap(long, default_value_t = false, value_parser = clap::builder::BoolishValueParser::new(), action = clap::ArgAction::Set)]
+    is_private_access_proxy: bool,
 }

 #[derive(clap::Args, Clone, Copy, Debug)]
@@ -264,6 +268,12 @@ struct SqlOverHttpArgs {

    #[clap(long, default_value_t = 64)]
    sql_over_http_cancel_set_shards: usize,
+
+    #[clap(long, default_value_t = 10 * 1024 * 1024)] // 10 MiB
+    sql_over_http_max_request_size_bytes: u64,
+
+    #[clap(long, default_value_t = 10 * 1024 * 1024)] // 10 MiB
+    sql_over_http_max_response_size_bytes: usize,
 }

 #[tokio::main]
@@ -675,6 +685,8 @@ fn build_config(args: &ProxyCliArgs) -> anyhow::Result<&'static ProxyConfig> {
        },
        cancel_set: CancelSet::new(args.sql_over_http.sql_over_http_cancel_set_shards),
        client_conn_threshold: args.sql_over_http.sql_over_http_client_conn_threshold,
+        max_request_size_bytes: args.sql_over_http.sql_over_http_max_request_size_bytes,
+        max_response_size_bytes: args.sql_over_http.sql_over_http_max_response_size_bytes,
    };
    let authentication_config = AuthenticationConfig {
        thread_pool,
@@ -682,6 +694,7 @@ fn build_config(args: &ProxyCliArgs) -> anyhow::Result<&'static ProxyConfig> {
        rate_limiter_enabled: args.auth_rate_limit_enabled,
        rate_limiter: AuthRateLimiter::new(args.auth_rate_limit.clone()),
        rate_limit_ip_subnet: args.auth_rate_limit_ip_subnet,
+        ip_allowlist_check_enabled: !args.is_private_access_proxy,
    };

    let config = Box::leak(Box::new(ProxyConfig {
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Vlad Lazar	84d1af736e	fixup: update external loop state before handling keyspaces `ValuesReconstructState::consume_done_keys` may only be called once after a layer visit. The code in the previous commit called it for each keyspace, resulting in keys not being marked done in the split keyspace by floor LSN scenario.	2024-09-19 10:44:20 +01:00
Vlad Lazar	3a218d7525	pageserver: extend nested image test with will init keys	2024-09-19 10:43:55 +01:00
Vlad Lazar	f0ad90f3ee	pageserver: handle reading up to different floor LSNs in deltas Problem Different keyspaces may require different floor LSNs in vectored delta layer visits. This patch adds support for such cases. Summary of Changes * Rework layer visit collection. Each layer type has a separate visit type which is aware of the requirements. For delta layers we track the floor LSN of keyspaces and merge only when that's matching. Other layer types do not have this requirement so they merge everything. * Thread the new visit types into the `get_values_reconstruct_data` calls. For delta layers, the code was adapted such that it may merge reads across keyspaces with different LSN floor requirements. * Tweak the fringe update code in `get_vectored_reconstruct_data_timeline` to handle different cont LSNs for different keyspaces. In practice, we will only update the fringe from one keyspace, since this "keyspace split" only happens when an image layer overlaps a delta layer (and image layers always complete all their keys). * Update tests with the new interfaces	2024-09-18 13:35:19 +01:00
Vlad Lazar	a8ad678574	tests: add unit test for vec read with overlapped images	2024-09-18 13:33:51 +01:00
Folke Behrens	c5cd8577ff	proxy: make sql-over-http max request/response sizes configurable (#9029 )	2024-09-18 13:58:51 +02:00
Heikki Linnakangas	3454ef7507	Refactor ImageLayerWriter to avoid passing a Timeline to finish() (#9028 ) Commit `ca5390a89d` made a similar change to DeltaLayerWriter. We bumped into this with Stas with our hackathon project, to create a standalong program to create image layers directly from a Postgres data directory. It needs to create image layers without having a Timeline and other pageserver machinery. This downgrades the "created image layer {}" message from INFO to TRACE level. TRACE is used for the corresponding message on delta layer creation too. The path logged in the message is now the temporary path, before the file is renamed to its final name. Again commit `ca5390a89d` made the same change for the message on delta layer creation.	2024-09-18 13:16:51 +03:00
Christian Schwarz	135e7e4306	add `neon_local` subcommand for the broker & use that from regression tests (#8948 ) There's currently no way to just start/stop broker from `neon_local`. This PR * adds a sub-command * uses that sub-command from the test suite instead of the pre-existing Python `subprocess` based approach. Found this useful during investigation https://github.com/neondatabase/cloud/issues/16886.	2024-09-18 09:10:27 +02:00
Christian Schwarz	3cd2a3f931	refactor(walredo): process launch & kill-on-error machinery (#8951 ) Immediate benefit: easier to spot what's going on. Later benefit: use the extracted method in PR - https://github.com/neondatabase/neon/pull/8952 which adds a `ping` command to walredo. Found this useful during investigation https://github.com/neondatabase/cloud/issues/16886.	2024-09-17 19:16:33 +00:00
Alexander Bayandin	d78f5ce6da	CI: don't fetch the whole git history if it's not required (#9021 ) ## Problem We do use `actions/checkout` with `fetch-depth: 0` when it's not required ## Summary of changes - Remove unneeded `fetch-depth: 0` - Add a comment if `fetch-depth: 0` is required	2024-09-17 18:40:05 +01:00
Arpad Müller	a1b71b73fe	Rename some S3 usages to "remote storage" in exposed messages (#8999 ) In exposed messages like log messages we mentioned "S3", which is not entirely accurate as we support Azure blob storage now as well.	2024-09-17 19:15:01 +02:00
Tristan Partin	6138eb50e9	Fix test code related to migrations We added another migration in `5876c441ab`, but didn't bump this value. This had no effect, but best to fix it anyway. Signed-off-by: Tristan Partin <tristan@neon.tech>	2024-09-17 15:56:05 +01:00
Heikki Linnakangas	d211f00f05	Remove unnecessary dependencies (#9000 ) Found by "cargo machete"	2024-09-17 17:55:45 +03:00
Alexander Bayandin	cd4276fd65	CI: fix release pipeline (#9017 ) ## Problem We've got 2 non-blocking failures on the release pipeline: - `promote-compatibility-data` job got skipped _presumably_ because one of the dependencies of `deploy` job (`push-to-acr-dev`) got skipped (https://github.com/neondatabase/neon/pull/8940) - `coverage-report` job fails because we don't build debug artifacts in the release branch (https://github.com/neondatabase/neon/pull/8561) ## Summary of changes - Always run `push-to-acr-dev` / `push-to-acr-prod` jobs, but add `skip_if` parameter to the reusable workflow, which can skip the job internally, without skipping externally - Do not run `coverage-report` on release branches	2024-09-17 10:17:48 +01:00
Vlad Lazar	b719d58863	storcon: forward requests from stepped down instance to the current leader (#8954 ) ## Problem It turns out that we can't rely on external orchestration to promptly route trafic to the new leader. This is downtime inducing. Forwarding provides a safe way out. ## Safety We forward when: 1. Request is not one of ["/control/v1/step_down", "/status", "/ready", "/metrics"] 2. Current instance is in [`LeadershipStatus::SteppedDown`] state 3. There is a leader in the database to forward to 4. Leader from step (3) is not the current instance If a storcon instance is persisted in the database, then we know that it is the current leader. There's one exception: time between handling step-down request and the new leader updating the database. Let's treat the happy case first. The stepped down node does not produce any side effects, since all request handling happens on the leader. As for the edge case, we are guaranteed to always have a maximum of two running instances. Hence, if we are in the edge case scenario the leader persisted in the database is the stepped down instance that received the request. Condition (4) above covers this scenario. ## Summary of changes * Conversion utilities for reqwest <-> hyper. I'm not happy with these, but I don't see a better way. Open to suggestions. * Add request forwarding logic * Update each request handler. Again, not happy with this. If anyone knows a nice to wrap the handlers, lmk. Me and Joonas tried :/ * Update each handler to maybe forward * Tweak tests to showcase new behaviour	2024-09-17 09:25:42 +01:00
Heikki Linnakangas	2db840d8b8	Move a few test functions related to auth tokens to separate file (#9018 ) For readability. neon_fixtures.py is huge.	2024-09-17 06:53:18 +03:00
Heikki Linnakangas	4295ff0f07	Mark a couple of test fixtures as session-scoped (#9018 ) pg_distrib_dir doesn't include the Postgres version and only depends on env variables which cannot change during a test run, so it can be marked as session-scoped. Similarly, the platform cannot change during a test run.	2024-09-17 06:53:18 +03:00
Heikki Linnakangas	c6f56b8462	Remove redundant get_dir_size() function (#9018 ) There was another copy of it in utils.py. The only difference is that the version in utils.py tolerates files that are concurrently removed. That seems fine for the few callers in neon_fixtures.py too.	2024-09-17 06:53:18 +03:00
Heikki Linnakangas	fec9321fc0	Use Path type in a few more places in neon_fixtures.py (#9018 ) This is in preparation of replacing neon_fixtures.get_dir_size with neon_fixtures.utils.get_dir_size() in next commit.	2024-09-17 06:53:18 +03:00
Heikki Linnakangas	3a52e356c1	Remove unused function (#9018 )	2024-09-17 06:53:18 +03:00
Tristan Partin	5e16c7bb0b	Generate pgbench data on the server for most tests This should generally be faster when running tests, especially those that run with higher scales. Ignoring test_lfc_resize since it seems like we are hitting a query timeout for some reason that I have yet to investigate. A little bit of improvemnt is better than none. Signed-off-by: Tristan Partin <tristan@neon.tech>	2024-09-16 23:37:36 +01:00
Heikki Linnakangas	2bbb4d3e1c	Remove misc unused code (#9014 )	2024-09-16 18:45:19 +00:00
Matthias van de Meent	c8bedca582	Fix PG17's extension modifications (#9010 ) This also reduces the GRANT statements to one per created _reset function	2024-09-16 17:06:31 +01:00
Tristan Partin	5876c441ab	Grant access to pg_show_replication_origin_status for neon_superuser Signed-off-by: Tristan Partin <tristan@neon.tech>	2024-09-16 16:38:55 +01:00
Alexander Bayandin	b2c83db54d	CI(gather-rust-build-stats): set PQ_LIB_DIR to Postgres 17 (#9001 ) ## Problem `gather-rust-build-stats` extra CI job fails with ``` "PQ_LIB_DIR" doesn't exist in the configured path: "/__w/neon/neon/pg_install/v16/lib" ``` ## Summary of changes - Use the path to Postgres 17 for the `gather-rust-build-stats` job. The job uses Postgres built by `make walproposer-lib`	2024-09-16 12:44:26 +01:00
Matthias van de Meent	0a8c5e1214	Fix broken image for PG17 (#8998 ) Most extensions are not required to run Neon-based PostgreSQL, but the Neon extension is _quite_ critical, so let's make sure we include it. ## Problem Staging doesn't have working compute images for PG17 ## Summary of changes Disable some PG17 filters so that we get the critical components into the PG17 image	2024-09-13 15:10:52 +01:00
Matthias van de Meent	78938d1b59	[compute/postgres] feature: PostgreSQL 17 (#8573 ) This adds preliminary PG17 support to Neon, based on RC1 / 2024-09-04 `07b828e9d4` NOTICE: The data produced by the included version of the PostgreSQL fork may not be compatible with the future full release of PostgreSQL 17 due to expected or unexpected future changes in magic numbers and internals. DO NOT EXPECT DATA IN V17-TENANTS TO BE COMPATIBLE WITH THE 17.0 RELEASE! Co-authored-by: Anastasia Lubennikova <anastasia@neon.tech> Co-authored-by: Alexander Bayandin <alexander@neon.tech> Co-authored-by: Konstantin Knizhnik <knizhnik@neon.tech> Co-authored-by: Heikki Linnakangas <heikki@neon.tech>	2024-09-12 23:18:41 +01:00
Stefan Radig	fcab61bdcd	Prototype implementation for private access poc (#8976 ) ## Problem For the Private Access POC we want users to be able to disable access from the public proxy. To limit the number of changes this can be done by configuring an IP allowlist [ "255.255.255.255" ]. For the Private Access proxy a new commandline flag allows to disable IP allowlist completely. See https://www.notion.so/neondatabase/Neon-Private-Access-POC-Proposal-8f707754e1ab4190ad5709da7832f020?d=887495c15e884aa4973f973a8a0a582a#7ac6ec249b524a74adbeddc4b84b8f5f for details about the POC., ## Summary of changes - Adding the commandline flag is_private_access_proxy=true will disable IP allowlist	2024-09-12 15:55:12 +01:00
Tristan Partin	9e3ead3689	Collect the last of on-demand WAL download in CreateReplicationSlot reverts Signed-off-by: Tristan Partin <tristan@neon.tech>	2024-09-12 11:31:38 +01:00
Heikki Linnakangas	8dc069037b	Remove NeonEnvBuilder.start() function It feels wrong to me to start() from the builder object. Surely the thing you start is the environment itself, not its configuration.	2024-09-12 01:28:56 +03:00
Heikki Linnakangas	0a363c3dce	Add --timeline-id option to "neon_local timeline branch" command Makes it consistent with the "timeline create" and "timeline import" commands, which allowed you to pass the timeline id as argument. This also makes it unnecessary to parse the timeline ID from the output in the python function that calls it.	2024-09-12 01:28:56 +03:00
Heikki Linnakangas	aeca15008c	Remove obsolete and misleading comment The tenant ID was not actually generated here but in NeonEnvBuilder. And the "neon_local init" command hasn't been able to generate the initial tenant since `8712e1899e` anyway.	2024-09-12 01:28:56 +03:00
Heikki Linnakangas	43846b72fa	Remove unused "neon_local init --pg-version" arg It has been unused since commit `8712e1899e`, when it stopped creating the initial timeline.	2024-09-12 01:28:56 +03:00
John Spray	cb060548fb	libs: tweak PageserverUtilization::is_overloaded (#8946 ) ## Problem Having run in production for a while, we see that nodes are generally safely oversubscribed by about a factor of 2. ## Summary of changes Tweak the is_overloaded method to check for utililzation over 200% rather than over 100%	2024-09-11 18:45:34 +01:00
Folke Behrens	bae793ffcd	proxy: Handle all let underscore instances (#8898 ) * Most can be simply replaced * One instance renamed to _rtchk (return-type check)	2024-09-10 15:36:08 +02:00
John Spray	26b5fcdc50	reinstate write-path key check (#8973 ) ## Problem In https://github.com/neondatabase/neon/pull/8621, validation of keys during ingest was removed because the places where we actually store keys are now past the point where we have already converted them to CompactKey (i128) representation. ## Summary of changes Reinstate validation at an earlier stage in ingest. This doesn't cover literally every place we write a key, but it covers most cases where we're trusting postgres to give us a valid key (i.e. one that doesn't try and use a custom spacenode).	2024-09-10 12:54:25 +01:00
Arpad Müller	97582178cb	Remove async_trait from the Handler trait (#8958 ) Newest attempt to remove `async_trait` from the Handler trait. Earlier attempts were in #7301 and #8296 .	2024-09-10 02:40:00 +02:00
Matthias van de Meent	842be0ba74	Specialize WalIngest on PostgreSQL version (#8904 ) The current code assumes that most of this functionality is version-independent, which is only true up to v16 - PostgreSQL 17 has a new field in CheckPoint that we need to keep track of. This basically removes the file-level dependency on v14, and replaces it with switches that load the correct version dependencies where required.	2024-09-09 23:01:52 +01:00
Heikki Linnakangas	982b376ea2	Update parquet crate to a released version (#8961 ) PR #7782 set the dependency in Cargo.toml to 'master', and locked the version to commit that contained a specific fix, because we needed the fix before it was included in a versioned release. The fix was later included in parquet crate version 52.0.0, so we can now switch back to using a released version. The latest release is 53.0.0, switch straight to that. --------- Co-authored-by: Conrad Ludgate <conradludgate@gmail.com>	2024-09-10 00:04:00 +03:00
Alex Chi Z.	e158df4e86	feat(pageserver): split delta writer automatically determines key range (#8850 ) close https://github.com/neondatabase/neon/issues/8838 ## Summary of changes This patch modifies the split delta layer writer to avoid taking start_key and end_key when creating/finishing the layer writer. The start_key for the delta layers will be the first key provided to the layer writer, and the end_key would be the `last_key.next()`. This simplifies the delta layer writer API. On that, the layer key hack is removed. Image layers now use the full key range, and delta layers use the first/last key provided by the user. --------- Signed-off-by: Alex Chi Z <chi@neon.tech>	2024-09-09 22:03:27 +01:00
Heikki Linnakangas	723c0971e8	Don't create 'empty' branch in neon_simple_env (#8965 ) Now that we've given up hope on sharing the neon_simple_env between tests, there's no reason to not use the 'main' branch directly.	2024-09-09 12:38:34 +03:00
Heikki Linnakangas	c8f67eed8f	Remove TEST_SHARED_FIXTURES (#8965 ) I wish it worked, but it's been broken for a long time, so let's admit defeat and remove it. The idea of sharing the same pageserver and safekeeper environment between tests is still sound, and it could save a lot of time in our CI. We should perhaps put some time into doing that, but we're better off starting from scratch than trying to make TEST_SHARED_FIXTURES work in its current form.	2024-09-09 12:38:34 +03:00
Heikki Linnakangas	2d885ac07a	Update strum (#8962 ) I wanted to use some features from the newer version. The PR that needed the new version is not ready yet (and might never be), but seems nice to stay up in any case.	2024-09-08 21:47:57 +03:00
Heikki Linnakangas	89c5e80b3f	Update toml and toml_edit crates (#8963 ) Eliminates a few duplicate versions from the dependency tree.	2024-09-08 21:47:23 +03:00
Heikki Linnakangas	93ec7503e0	Lock the correct revision of rust-postgres crates (#8960 ) We modified the crate in an incompatible way and upgraded to the new version in PR #8076. However, it was reverted in #8654. The revert reverted the Cargo.lock reference to it, but since Cargo.toml still points to the (tip of the) 'neon' branch, every time you make any other unrelated changes to Cargo.toml, it also tries to update the rust-postgres crates to the tip of the 'neon' branch again, which doesn't work. To fix, lock the crates to the exact commit SHA that works.	2024-09-07 14:11:36 +01:00
				`@@ -0,0 +1 @@`
				`GRANT EXECUTE ON FUNCTION pg_show_replication_origin_status TO neon_superuser;`