Merge pull request #8217 from neondatabase/rc/2024-07-01

Storage & Compute release 2024-07-01
Restore running xacts from CLOG on replica startup (#7288 )
2026-08-18 03:58:19 +00:00 · 2024-07-04 20:22:51 +03:00 · 2024-07-04 18:58:34 +03:00 · 2024-07-04 18:58:28 +03:00 · 2024-07-04 18:58:20 +03:00 · 2024-07-01 01:49:49 +03:00
141 changed files with 5823 additions and 2190 deletions
@@ -183,7 +183,7 @@ runs:
      uses: actions/cache@v4
      with:
        path: ~/.cache/pypoetry/virtualenvs
-        key: v2-${{ runner.os }}-python-deps-${{ hashFiles('poetry.lock') }}
+        key: v2-${{ runner.os }}-${{ runner.arch }}-python-deps-${{ hashFiles('poetry.lock') }}

    - name: Store Allure test stat in the DB (new)
      if: ${{ !cancelled() && inputs.store-test-results-into-db == 'true' }}
@@ -26,7 +26,7 @@ runs:
        TARGET: ${{ inputs.path }}
        ARCHIVE: /tmp/downloads/${{ inputs.name }}.tar.zst
        SKIP_IF_DOES_NOT_EXIST: ${{ inputs.skip-if-does-not-exist }}
-        PREFIX: artifacts/${{ inputs.prefix || format('{0}/{1}', github.run_id, github.run_attempt) }}
+        PREFIX: artifacts/${{ inputs.prefix || format('{0}/{1}/{2}', github.event.pull_request.head.sha || github.sha, github.run_id, github.run_attempt) }}
      run: |
        BUCKET=neon-github-public-dev
        FILENAME=$(basename $ARCHIVE)
@@ -56,14 +56,14 @@ runs:
      if: inputs.build_type != 'remote'
      uses: ./.github/actions/download
      with:
-        name: neon-${{ runner.os }}-${{ inputs.build_type }}-artifact
+        name: neon-${{ runner.os }}-${{ runner.arch }}-${{ inputs.build_type }}-artifact
        path: /tmp/neon

    - name: Download Neon binaries for the previous release
      if: inputs.build_type != 'remote'
      uses: ./.github/actions/download
      with:
-        name: neon-${{ runner.os }}-${{ inputs.build_type }}-artifact
+        name: neon-${{ runner.os }}-${{ runner.arch }}-${{ inputs.build_type }}-artifact
        path: /tmp/neon-previous
        prefix: latest

@@ -89,7 +89,7 @@ runs:
      uses: actions/cache@v4
      with:
        path: ~/.cache/pypoetry/virtualenvs
-        key: v2-${{ runner.os }}-python-deps-${{ hashFiles('poetry.lock') }}
+        key: v2-${{ runner.os }}-${{ runner.arch }}-python-deps-${{ hashFiles('poetry.lock') }}

    - name: Install Python deps
      shell: bash -euxo pipefail {0}
@@ -8,7 +8,7 @@ inputs:
    description: "A directory or file to upload"
    required: true
  prefix:
-    description: "S3 prefix. Default is '${GITHUB_RUN_ID}/${GITHUB_RUN_ATTEMPT}'"
+    description: "S3 prefix. Default is '${GITHUB_SHA}/${GITHUB_RUN_ID}/${GITHUB_RUN_ATTEMPT}'"
    required: false

 runs:
@@ -45,7 +45,7 @@ runs:
      env:
        SOURCE: ${{ inputs.path }}
        ARCHIVE: /tmp/uploads/${{ inputs.name }}.tar.zst
-        PREFIX: artifacts/${{ inputs.prefix || format('{0}/{1}', github.run_id, github.run_attempt) }}
+        PREFIX: artifacts/${{ inputs.prefix || format('{0}/{1}/{2}', github.event.pull_request.head.sha || github.sha, github.run_id , github.run_attempt) }}
      run: |
        BUCKET=neon-github-public-dev
        FILENAME=$(basename $ARCHIVE)
@@ -77,7 +77,7 @@ jobs:
    - name: Download Neon artifact
      uses: ./.github/actions/download
      with:
-        name: neon-${{ runner.os }}-release-artifact
+        name: neon-${{ runner.os }}-${{ runner.arch }}-release-artifact
        path: /tmp/neon/
        prefix: latest

@@ -235,7 +235,7 @@ jobs:
    - name: Download Neon artifact
      uses: ./.github/actions/download
      with:
-        name: neon-${{ runner.os }}-release-artifact
+        name: neon-${{ runner.os }}-${{ runner.arch }}-release-artifact
        path: /tmp/neon/
        prefix: latest

@@ -373,7 +373,7 @@ jobs:
    - name: Download Neon artifact
      uses: ./.github/actions/download
      with:
-        name: neon-${{ runner.os }}-release-artifact
+        name: neon-${{ runner.os }}-${{ runner.arch }}-release-artifact
        path: /tmp/neon/
        prefix: latest

@@ -473,7 +473,7 @@ jobs:
    - name: Download Neon artifact
      uses: ./.github/actions/download
      with:
-        name: neon-${{ runner.os }}-release-artifact
+        name: neon-${{ runner.os }}-${{ runner.arch }}-release-artifact
        path: /tmp/neon/
        prefix: latest

@@ -576,7 +576,7 @@ jobs:
    - name: Download Neon artifact
      uses: ./.github/actions/download
      with:
-        name: neon-${{ runner.os }}-release-artifact
+        name: neon-${{ runner.os }}-${{ runner.arch }}-release-artifact
        path: /tmp/neon/
        prefix: latest

@@ -677,7 +677,7 @@ jobs:
    - name: Download Neon artifact
      uses: ./.github/actions/download
      with:
-        name: neon-${{ runner.os }}-release-artifact
+        name: neon-${{ runner.os }}-${{ runner.arch }}-release-artifact
        path: /tmp/neon/
        prefix: latest

@@ -78,7 +78,7 @@ jobs:
          pull: true
          file: Dockerfile.build-tools
          cache-from: type=registry,ref=neondatabase/build-tools:cache-${{ matrix.arch }}
-          cache-to: type=registry,ref=neondatabase/build-tools:cache-${{ matrix.arch }},mode=max
+          cache-to: ${{ github.ref_name == 'main' && format('type=registry,ref=neondatabase/build-tools:cache-{0},mode=max', matrix.arch) || '' }}
          tags: neondatabase/build-tools:${{ inputs.image-tag }}-${{ matrix.arch }}

      - name: Remove custom docker config directory
@@ -109,7 +109,7 @@ jobs:
        uses: actions/cache@v4
        with:
          path: ~/.cache/pypoetry/virtualenvs
-          key: v2-${{ runner.os }}-python-deps-${{ hashFiles('poetry.lock') }}
+          key: v2-${{ runner.os }}-${{ runner.arch }}-python-deps-${{ hashFiles('poetry.lock') }}

      - name: Install Python deps
        run: ./scripts/pysync
@@ -149,7 +149,7 @@ jobs:
 #            !~/.cargo/registry/src
 #            ~/.cargo/git/
 #            target/
-#          key: v1-${{ runner.os }}-cargo-clippy-${{ hashFiles('rust-toolchain.toml') }}-${{ hashFiles('Cargo.lock') }}
+#          key: v1-${{ runner.os }}-${{ runner.arch }}-cargo-clippy-${{ hashFiles('rust-toolchain.toml') }}-${{ hashFiles('Cargo.lock') }}

      # Some of our rust modules use FFI and need those to be checked
      - name: Get postgres headers
@@ -291,29 +291,29 @@ jobs:
 #            target/
 #          # Fall back to older versions of the key, if no cache for current Cargo.lock was found
 #          key: |
-#            v1-${{ runner.os }}-${{ matrix.build_type }}-cargo-${{ hashFiles('rust-toolchain.toml') }}-${{ hashFiles('Cargo.lock') }}
-#            v1-${{ runner.os }}-${{ matrix.build_type }}-cargo-${{ hashFiles('rust-toolchain.toml') }}-
+#            v1-${{ runner.os }}-${{ runner.arch }}-${{ matrix.build_type }}-cargo-${{ hashFiles('rust-toolchain.toml') }}-${{ hashFiles('Cargo.lock') }}
+#            v1-${{ runner.os }}-${{ runner.arch }}-${{ matrix.build_type }}-cargo-${{ hashFiles('rust-toolchain.toml') }}-

      - name: Cache postgres v14 build
        id: cache_pg_14
        uses: actions/cache@v4
        with:
          path: pg_install/v14
-          key: v1-${{ runner.os }}-${{ matrix.build_type }}-pg-${{ steps.pg_v14_rev.outputs.pg_rev }}-${{ hashFiles('Makefile', 'Dockerfile.build-tools') }}
+          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ matrix.build_type }}-pg-${{ steps.pg_v14_rev.outputs.pg_rev }}-${{ hashFiles('Makefile', 'Dockerfile.build-tools') }}

      - name: Cache postgres v15 build
        id: cache_pg_15
        uses: actions/cache@v4
        with:
          path: pg_install/v15
-          key: v1-${{ runner.os }}-${{ matrix.build_type }}-pg-${{ steps.pg_v15_rev.outputs.pg_rev }}-${{ hashFiles('Makefile', 'Dockerfile.build-tools') }}
+          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ matrix.build_type }}-pg-${{ steps.pg_v15_rev.outputs.pg_rev }}-${{ hashFiles('Makefile', 'Dockerfile.build-tools') }}

      - name: Cache postgres v16 build
        id: cache_pg_16
        uses: actions/cache@v4
        with:
          path: pg_install/v16
-          key: v1-${{ runner.os }}-${{ matrix.build_type }}-pg-${{ steps.pg_v16_rev.outputs.pg_rev }}-${{ hashFiles('Makefile', 'Dockerfile.build-tools') }}
+          key: v1-${{ runner.os }}-${{ runner.arch }}-${{ matrix.build_type }}-pg-${{ steps.pg_v16_rev.outputs.pg_rev }}-${{ hashFiles('Makefile', 'Dockerfile.build-tools') }}

      - name: Build postgres v14
        if: steps.cache_pg_14.outputs.cache-hit != 'true'
@@ -411,7 +411,7 @@ jobs:
      - name: Upload Neon artifact
        uses: ./.github/actions/upload
        with:
-          name: neon-${{ runner.os }}-${{ matrix.build_type }}-artifact
+          name: neon-${{ runner.os }}-${{ runner.arch }}-${{ matrix.build_type }}-artifact
          path: /tmp/neon

      # XXX: keep this after the binaries.list is formed, so the coverage can properly work later
@@ -490,7 +490,7 @@ jobs:
        uses: actions/cache@v4
        with:
          path: ~/.cache/pypoetry/virtualenvs
-          key: v1-${{ runner.os }}-python-deps-${{ hashFiles('poetry.lock') }}
+          key: v1-${{ runner.os }}-${{ runner.arch }}-python-deps-${{ hashFiles('poetry.lock') }}

      - name: Install Python deps
        run: ./scripts/pysync
@@ -639,7 +639,7 @@ jobs:
      - name: Get Neon artifact
        uses: ./.github/actions/download
        with:
-          name: neon-${{ runner.os }}-${{ matrix.build_type }}-artifact
+          name: neon-${{ runner.os }}-${{ runner.arch }}-${{ matrix.build_type }}-artifact
          path: /tmp/neon

      - name: Get coverage artifact
@@ -763,7 +763,7 @@ jobs:
          pull: true
          file: Dockerfile
          cache-from: type=registry,ref=neondatabase/neon:cache-${{ matrix.arch }}
-          cache-to: type=registry,ref=neondatabase/neon:cache-${{ matrix.arch }},mode=max
+          cache-to: ${{ github.ref_name == 'main' && format('type=registry,ref=neondatabase/neon:cache-{0},mode=max', matrix.arch) || '' }}
          tags: |
            neondatabase/neon:${{ needs.tag.outputs.build-tag }}-${{ matrix.arch }}

@@ -855,7 +855,7 @@ jobs:
          pull: true
          file: Dockerfile.compute-node
          cache-from: type=registry,ref=neondatabase/compute-node-${{ matrix.version }}:cache-${{ matrix.arch }}
-          cache-to: type=registry,ref=neondatabase/compute-node-${{ matrix.version }}:cache-${{ matrix.arch }},mode=max
+          cache-to: ${{ github.ref_name == 'main' && format('type=registry,ref=neondatabase/compute-node-{0}:cache-{1},mode=max', matrix.version, matrix.arch) || '' }}
          tags: |
            neondatabase/compute-node-${{ matrix.version }}:${{ needs.tag.outputs.build-tag }}-${{ matrix.arch }}

@@ -875,7 +875,7 @@ jobs:
          file: Dockerfile.compute-node
          target: neon-pg-ext-test
          cache-from: type=registry,ref=neondatabase/neon-test-extensions-${{ matrix.version }}:cache-${{ matrix.arch }}
-          cache-to: type=registry,ref=neondatabase/neon-test-extensions-${{ matrix.version }}:cache-${{ matrix.arch }},mode=max
+          cache-to: ${{ github.ref_name == 'main' && format('type=registry,ref=neondatabase/neon-test-extensions-{0}:cache-{1},mode=max', matrix.version, matrix.arch) || '' }}
          tags: |
            neondatabase/neon-test-extensions-${{ matrix.version }}:${{needs.tag.outputs.build-tag}}-${{ matrix.arch }}

@@ -1245,6 +1245,7 @@ jobs:
        run: |
          if [[ "$GITHUB_REF_NAME" == "main" ]]; then
            gh workflow --repo neondatabase/aws run deploy-dev.yml --ref main -f branch=main -f dockerTag=${{needs.tag.outputs.build-tag}} -f deployPreprodRegion=false
+            gh workflow --repo neondatabase/azure run deploy.yml -f dockerTag=${{needs.tag.outputs.build-tag}}
          elif [[ "$GITHUB_REF_NAME" == "release" ]]; then
            gh workflow --repo neondatabase/aws run deploy-dev.yml --ref main \
              -f deployPgSniRouter=false \
@@ -1339,7 +1340,7 @@ jobs:
          # Update Neon artifact for the release (reuse already uploaded artifact)
          for build_type in debug release; do
            OLD_PREFIX=artifacts/${GITHUB_RUN_ID}
-            FILENAME=neon-${{ runner.os }}-${build_type}-artifact.tar.zst
+            FILENAME=neon-${{ runner.os }}-${{ runner.arch }}-${build_type}-artifact.tar.zst

            S3_KEY=$(aws s3api list-objects-v2 --bucket ${BUCKET} --prefix ${OLD_PREFIX} | jq -r '.Contents[]?.Key' | grep ${FILENAME} | sort --version-sort | tail -1 || true)
            if [ -z "${S3_KEY}" ]; then
@@ -41,7 +41,7 @@ jobs:
      uses: actions/cache@v4
      with:
        path: ~/.cache/pypoetry/virtualenvs
-        key: v2-${{ runner.os }}-python-deps-ubunutu-latest-${{ hashFiles('poetry.lock') }}
+        key: v2-${{ runner.os }}-${{ runner.arch }}-python-deps-ubunutu-latest-${{ hashFiles('poetry.lock') }}

    - name: Install Python deps
      shell: bash -euxo pipefail {0}
@@ -85,7 +85,7 @@ jobs:
      uses: actions/upload-artifact@v4
      with:
        retention-days: 7
-        name: python-test-pg_clients-${{ runner.os }}-stage-logs
+        name: python-test-pg_clients-${{ runner.os }}-${{ runner.arch }}-stage-logs
        path: ${{ env.TEST_OUTPUT }}

    - name: Post to a Slack channel
@@ -1,4 +1,5 @@
 # * `-A unknown_lints` – do not warn about unknown lint suppressions
 #                        that people with newer toolchains might use
 # * `-D warnings`      - fail on any warnings (`cargo` returns non-zero exit status)
-export CLIPPY_COMMON_ARGS="--locked --workspace --all-targets -- -A unknown_lints -D warnings"
+# * `-D clippy::todo`  - don't let `todo!()` slip into `main`
+export CLIPPY_COMMON_ARGS="--locked --workspace --all-targets -- -A unknown_lints -D warnings -D clippy::todo"
@@ -1246,7 +1246,7 @@ dependencies = [
 "tokio-postgres",
 "tokio-stream",
 "tokio-util",
- "toml_edit",
+ "toml_edit 0.19.10",
 "tracing",
 "tracing-opentelemetry",
 "tracing-subscriber",
@@ -1362,8 +1362,8 @@ dependencies = [
 "tokio",
 "tokio-postgres",
 "tokio-util",
- "toml",
- "toml_edit",
+ "toml 0.7.4",
+ "toml_edit 0.19.10",
 "tracing",
 "url",
 "utils",
@@ -1669,9 +1669,9 @@ dependencies = [

 [[package]]
 name = "diesel"
-version = "2.1.4"
+version = "2.2.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "62c6fcf842f17f8c78ecf7c81d75c5ce84436b41ee07e03f490fbb5f5a8731d8"
+checksum = "62d6dcd069e7b5fe49a302411f759d4cf1cf2c27fe798ef46fb8baefc053dd2b"
 dependencies = [
 "bitflags 2.4.1",
 "byteorder",
@@ -1684,11 +1684,12 @@ dependencies = [

 [[package]]
 name = "diesel_derives"
-version = "2.1.2"
+version = "2.2.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ef8337737574f55a468005a83499da720f20c65586241ffea339db9ecdfd2b44"
+checksum = "59de76a222c2b8059f789cbe07afbfd8deb8c31dd0bc2a21f85e256c1def8259"
 dependencies = [
 "diesel_table_macro_syntax",
+ "dsl_auto_type",
 "proc-macro2",
 "quote",
 "syn 2.0.52",
@@ -1696,9 +1697,9 @@ dependencies = [

 [[package]]
 name = "diesel_migrations"
-version = "2.1.0"
+version = "2.2.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "6036b3f0120c5961381b570ee20a02432d7e2d27ea60de9578799cf9156914ac"
+checksum = "8a73ce704bad4231f001bff3314d91dce4aba0770cee8b233991859abc15c1f6"
 dependencies = [
 "diesel",
 "migrations_internals",
@@ -1707,9 +1708,9 @@ dependencies = [

 [[package]]
 name = "diesel_table_macro_syntax"
-version = "0.1.0"
+version = "0.2.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "fc5557efc453706fed5e4fa85006fe9817c224c3f480a34c7e5959fd700921c5"
+checksum = "209c735641a413bc68c4923a9d6ad4bcb3ca306b794edaa7eb0b3228a99ffb25"
 dependencies = [
 "syn 2.0.52",
 ]
@@ -1745,6 +1746,20 @@ dependencies = [
 "const-random",
 ]

+[[package]]
+name = "dsl_auto_type"
+version = "0.1.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0892a17df262a24294c382f0d5997571006e7a4348b4327557c4ff1cd4a8bccc"
+dependencies = [
+ "darling",
+ "either",
+ "heck 0.5.0",
+ "proc-macro2",
+ "quote",
+ "syn 2.0.52",
+]
+
 [[package]]
 name = "dyn-clone"
 version = "1.0.14"
@@ -3084,19 +3099,19 @@ dependencies = [

 [[package]]
 name = "migrations_internals"
-version = "2.1.0"
+version = "2.2.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "0f23f71580015254b020e856feac3df5878c2c7a8812297edd6c0a485ac9dada"
+checksum = "fd01039851e82f8799046eabbb354056283fb265c8ec0996af940f4e85a380ff"
 dependencies = [
 "serde",
- "toml",
+ "toml 0.8.14",
 ]

 [[package]]
 name = "migrations_macros"
-version = "2.1.0"
+version = "2.2.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "cce3325ac70e67bbab5bd837a31cae01f1a6db64e0e744a33cb03a543469ef08"
+checksum = "ffb161cc72176cb37aa47f1fc520d3ef02263d67d661f44f05d05a079e1237fd"
 dependencies = [
 "migrations_internals",
 "proc-macro2",
@@ -3576,7 +3591,7 @@ dependencies = [
 "thiserror",
 "tokio",
 "tokio-util",
- "toml_edit",
+ "toml_edit 0.19.10",
 "utils",
 "workspace_hack",
 ]
@@ -3659,7 +3674,7 @@ dependencies = [
 "tokio-stream",
 "tokio-tar",
 "tokio-util",
- "toml_edit",
+ "toml_edit 0.19.10",
 "tracing",
 "twox-hash",
 "url",
@@ -4665,7 +4680,7 @@ dependencies = [
 "tokio",
 "tokio-stream",
 "tokio-util",
- "toml_edit",
+ "toml_edit 0.19.10",
 "tracing",
 "utils",
 "workspace_hack",
@@ -5164,7 +5179,7 @@ dependencies = [
 "tokio-stream",
 "tokio-tar",
 "tokio-util",
- "toml_edit",
+ "toml_edit 0.19.10",
 "tracing",
 "tracing-subscriber",
 "url",
@@ -5443,9 +5458,9 @@ dependencies = [

 [[package]]
 name = "serde_spanned"
-version = "0.6.2"
+version = "0.6.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "93107647184f6027e3b7dcb2e11034cf95ffa1e3a682c67951963ac69c1c007d"
+checksum = "79e674e01f999af37c49f70a6ede167a8a60b2503e56c5599532a65baa5969a0"
 dependencies = [
 "serde",
 ]
@@ -6330,14 +6345,26 @@ dependencies = [
 "serde",
 "serde_spanned",
 "toml_datetime",
- "toml_edit",
+ "toml_edit 0.19.10",
+]
+
+[[package]]
+name = "toml"
+version = "0.8.14"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6f49eb2ab21d2f26bd6db7bf383edc527a7ebaee412d17af4d40fdccd442f335"
+dependencies = [
+ "serde",
+ "serde_spanned",
+ "toml_datetime",
+ "toml_edit 0.22.14",
 ]

 [[package]]
 name = "toml_datetime"
-version = "0.6.2"
+version = "0.6.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "5a76a9312f5ba4c2dec6b9161fdf25d87ad8a09256ccea5a556fef03c706a10f"
+checksum = "4badfd56924ae69bcc9039335b2e017639ce3f9b001c393c1b2d1ef846ce2cbf"
 dependencies = [
 "serde",
 ]
@@ -6352,7 +6379,20 @@ dependencies = [
 "serde",
 "serde_spanned",
 "toml_datetime",
- "winnow",
+ "winnow 0.4.6",
+]
+
+[[package]]
+name = "toml_edit"
+version = "0.22.14"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f21c7aaf97f1bd9ca9d4f9e73b0a6c74bd5afef56f2bc931943a6e1c37e04e38"
+dependencies = [
+ "indexmap 2.0.1",
+ "serde",
+ "serde_spanned",
+ "toml_datetime",
+ "winnow 0.6.13",
 ]

 [[package]]
@@ -7335,6 +7375,15 @@ dependencies = [
 "memchr",
 ]

+[[package]]
+name = "winnow"
+version = "0.6.13"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "59b5e5f6c299a3c7890b876a2a587f3115162487e704907d9b6cd29473052ba1"
+dependencies = [
+ "memchr",
+]
+
 [[package]]
 name = "winreg"
 version = "0.50.0"
@@ -7424,7 +7473,7 @@ dependencies = [
 "tokio-rustls 0.24.0",
 "tokio-util",
 "toml_datetime",
- "toml_edit",
+ "toml_edit 0.19.10",
 "tonic",
 "tower",
 "tracing",
@@ -73,13 +73,6 @@ RUN curl -fsSL 'https://apt.llvm.org/llvm-snapshot.gpg.key' | apt-key add - \
    && bash -c 'for f in /usr/bin/clang*-${LLVM_VERSION} /usr/bin/llvm*-${LLVM_VERSION}; do ln -s "${f}" "${f%-${LLVM_VERSION}}"; done' \
    && rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/*

-# PostgreSQL 14
-RUN curl -fsSL 'https://www.postgresql.org/media/keys/ACCC4CF8.asc' | apt-key add - \
-    && echo 'deb http://apt.postgresql.org/pub/repos/apt bullseye-pgdg main' > /etc/apt/sources.list.d/pgdg.list \
-    && apt update \
-    && apt install -y postgresql-client-14 \
-    && rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/*
-
 # AWS CLI
 RUN curl "https://awscli.amazonaws.com/awscli-exe-linux-$(uname -m).zip" -o "awscliv2.zip" \
    && unzip -q awscliv2.zip \
@@ -113,10 +106,10 @@ RUN for package in Capture::Tiny DateTime Devel::Cover Digest::MD5 File::Spec JS
    && rm -rf ../lcov.tar.gz

 # Compile and install the static OpenSSL library
-ENV OPENSSL_VERSION=3.2.2
+ENV OPENSSL_VERSION=1.1.1w
 ENV OPENSSL_PREFIX=/usr/local/openssl
 RUN wget -O /tmp/openssl-${OPENSSL_VERSION}.tar.gz https://www.openssl.org/source/openssl-${OPENSSL_VERSION}.tar.gz && \
-    echo "197149c18d9e9f292c43f0400acaba12e5f52cacfe050f3d199277ea738ec2e7 /tmp/openssl-${OPENSSL_VERSION}.tar.gz" | sha256sum --check && \
+    echo "cf3098950cb4d853ad95c0841f1f9c6d3dc102dccfcacd521d93925208b76ac8 /tmp/openssl-${OPENSSL_VERSION}.tar.gz" | sha256sum --check && \
    cd /tmp && \
    tar xzvf /tmp/openssl-${OPENSSL_VERSION}.tar.gz && \
    rm /tmp/openssl-${OPENSSL_VERSION}.tar.gz && \
@@ -83,12 +83,6 @@ pub fn write_postgres_conf(
        ComputeMode::Replica => {
            // hot_standby is 'on' by default, but let's be explicit
            writeln!(file, "hot_standby=on")?;
-
-            // Inform the replica about the primary state
-            // Default is 'false'
-            if let Some(primary_is_running) = spec.primary_is_running {
-                writeln!(file, "neon.primary_is_running={}", primary_is_running)?;
-            }
        }
    }

@@ -17,7 +17,11 @@ const MONITOR_CHECK_INTERVAL: Duration = Duration::from_millis(500);
 // should be handled gracefully.
 fn watch_compute_activity(compute: &ComputeNode) {
    // Suppose that `connstr` doesn't change
-    let connstr = compute.connstr.as_str();
+    let mut connstr = compute.connstr.clone();
+    connstr
+        .query_pairs_mut()
+        .append_pair("application_name", "compute_activity_monitor");
+    let connstr = connstr.as_str();

    // During startup and configuration we connect to every Postgres database,
    // but we don't want to count this as some user activity. So wait until
@@ -21,10 +21,8 @@ use pageserver_api::config::{
    DEFAULT_HTTP_LISTEN_PORT as DEFAULT_PAGESERVER_HTTP_PORT,
    DEFAULT_PG_LISTEN_PORT as DEFAULT_PAGESERVER_PG_PORT,
 };
-use pageserver_api::controller_api::PlacementPolicy;
-use pageserver_api::models::{
-    ShardParameters, TenantCreateRequest, TimelineCreateRequest, TimelineInfo,
-};
+use pageserver_api::controller_api::{PlacementPolicy, TenantCreateRequest};
+use pageserver_api::models::{ShardParameters, TimelineCreateRequest, TimelineInfo};
 use pageserver_api::shard::{ShardCount, ShardStripeSize, TenantShardId};
 use postgres_backend::AuthType;
 use postgres_connection::parse_host_port;
@@ -600,13 +598,9 @@ async fn handle_timeline(timeline_match: &ArgMatches, env: &mut local_env::Local
        Some(("import", import_match)) => {
            let tenant_id = get_tenant_id(import_match, env)?;
            let timeline_id = parse_timeline_id(import_match)?.expect("No timeline id provided");
-            let name = import_match
-                .get_one::<String>("node-name")
-                .ok_or_else(|| anyhow!("No node name provided"))?;
-            let update_catalog = import_match
-                .get_one::<bool>("update-catalog")
-                .cloned()
-                .unwrap_or_default();
+            let branch_name = import_match
+                .get_one::<String>("branch-name")
+                .ok_or_else(|| anyhow!("No branch name provided"))?;

            // Parse base inputs
            let base_tarfile = import_match
@@ -633,24 +627,11 @@ async fn handle_timeline(timeline_match: &ArgMatches, env: &mut local_env::Local
                .copied()
                .context("Failed to parse postgres version from the argument string")?;

-            let mut cplane = ComputeControlPlane::load(env.clone())?;
            println!("Importing timeline into pageserver ...");
            pageserver
                .timeline_import(tenant_id, timeline_id, base, pg_wal, pg_version)
                .await?;
-            env.register_branch_mapping(name.to_string(), tenant_id, timeline_id)?;
-
-            println!("Creating endpoint for imported timeline ...");
-            cplane.new_endpoint(
-                name,
-                tenant_id,
-                timeline_id,
-                None,
-                None,
-                pg_version,
-                ComputeMode::Primary,
-                !update_catalog,
-            )?;
+            env.register_branch_mapping(branch_name.to_string(), tenant_id, timeline_id)?;
            println!("Done");
        }
        Some(("branch", branch_match)) => {
@@ -865,20 +846,13 @@ async fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Re

            let allow_multiple = sub_args.get_flag("allow-multiple");

-            // If --safekeepers argument is given, use only the listed safekeeper nodes.
-            let safekeepers =
-                if let Some(safekeepers_str) = sub_args.get_one::<String>("safekeepers") {
-                    let mut safekeepers: Vec<NodeId> = Vec::new();
-                    for sk_id in safekeepers_str.split(',').map(str::trim) {
-                        let sk_id = NodeId(u64::from_str(sk_id).map_err(|_| {
-                            anyhow!("invalid node ID \"{sk_id}\" in --safekeepers list")
-                        })?);
-                        safekeepers.push(sk_id);
-                    }
-                    safekeepers
-                } else {
-                    env.safekeepers.iter().map(|sk| sk.id).collect()
-                };
+            // If --safekeepers argument is given, use only the listed
+            // safekeeper nodes; otherwise all from the env.
+            let safekeepers = if let Some(safekeepers) = parse_safekeepers(sub_args)? {
+                safekeepers
+            } else {
+                env.safekeepers.iter().map(|sk| sk.id).collect()
+            };

            let endpoint = cplane
                .endpoints
@@ -982,7 +956,10 @@ async fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Re
                        })
                        .collect::<Vec<_>>()
                };
-            endpoint.reconfigure(pageservers, None).await?;
+            // If --safekeepers argument is given, use only the listed
+            // safekeeper nodes; otherwise all from the env.
+            let safekeepers = parse_safekeepers(sub_args)?;
+            endpoint.reconfigure(pageservers, None, safekeepers).await?;
        }
        "stop" => {
            let endpoint_id = sub_args
@@ -1004,6 +981,23 @@ async fn handle_endpoint(ep_match: &ArgMatches, env: &local_env::LocalEnv) -> Re
    Ok(())
 }

+/// Parse --safekeepers as list of safekeeper ids.
+fn parse_safekeepers(sub_args: &ArgMatches) -> Result<Option<Vec<NodeId>>> {
+    if let Some(safekeepers_str) = sub_args.get_one::<String>("safekeepers") {
+        let mut safekeepers: Vec<NodeId> = Vec::new();
+        for sk_id in safekeepers_str.split(',').map(str::trim) {
+            let sk_id = NodeId(
+                u64::from_str(sk_id)
+                    .map_err(|_| anyhow!("invalid node ID \"{sk_id}\" in --safekeepers list"))?,
+            );
+            safekeepers.push(sk_id);
+        }
+        Ok(Some(safekeepers))
+    } else {
+        Ok(None)
+    }
+}
+
 fn handle_mappings(sub_match: &ArgMatches, env: &mut local_env::LocalEnv) -> Result<()> {
    let (sub_name, sub_args) = match sub_match.subcommand() {
        Some(ep_subcommand_data) => ep_subcommand_data,
@@ -1487,8 +1481,7 @@ fn cli() -> Command {
                .about("Import timeline from basebackup directory")
                .arg(tenant_id_arg.clone())
                .arg(timeline_id_arg.clone())
-                .arg(Arg::new("node-name").long("node-name")
-                    .help("Name to assign to the imported timeline"))
+                .arg(branch_name_arg.clone())
                .arg(Arg::new("base-tarfile")
                    .long("base-tarfile")
                    .value_parser(value_parser!(PathBuf))
@@ -1504,7 +1497,6 @@ fn cli() -> Command {
                .arg(Arg::new("end-lsn").long("end-lsn")
                    .help("Lsn the basebackup ends at"))
                .arg(pg_version_arg.clone())
-                .arg(update_catalog.clone())
            )
        ).subcommand(
            Command::new("tenant")
@@ -1609,7 +1601,7 @@ fn cli() -> Command {
                    .about("Start postgres.\n If the endpoint doesn't exist yet, it is created.")
                    .arg(endpoint_id_arg.clone())
                    .arg(endpoint_pageserver_id_arg.clone())
-                    .arg(safekeepers_arg)
+                    .arg(safekeepers_arg.clone())
                    .arg(remote_ext_config_args)
                    .arg(create_test_user)
                    .arg(allow_multiple.clone())
@@ -1618,6 +1610,7 @@ fn cli() -> Command {
                .subcommand(Command::new("reconfigure")
                            .about("Reconfigure the endpoint")
                            .arg(endpoint_pageserver_id_arg)
+                            .arg(safekeepers_arg)
                            .arg(endpoint_id_arg.clone())
                            .arg(tenant_id_arg.clone())
                )
@@ -499,6 +499,23 @@ impl Endpoint {
            .join(",")
    }

+    /// Map safekeepers ids to the actual connection strings.
+    fn build_safekeepers_connstrs(&self, sk_ids: Vec<NodeId>) -> Result<Vec<String>> {
+        let mut safekeeper_connstrings = Vec::new();
+        if self.mode == ComputeMode::Primary {
+            for sk_id in sk_ids {
+                let sk = self
+                    .env
+                    .safekeepers
+                    .iter()
+                    .find(|node| node.id == sk_id)
+                    .ok_or_else(|| anyhow!("safekeeper {sk_id} does not exist"))?;
+                safekeeper_connstrings.push(format!("127.0.0.1:{}", sk.get_compute_port()));
+            }
+        }
+        Ok(safekeeper_connstrings)
+    }
+
    pub async fn start(
        &self,
        auth_token: &Option<String>,
@@ -523,18 +540,7 @@ impl Endpoint {
        let pageserver_connstring = Self::build_pageserver_connstr(&pageservers);
        assert!(!pageserver_connstring.is_empty());

-        let mut safekeeper_connstrings = Vec::new();
-        if self.mode == ComputeMode::Primary {
-            for sk_id in safekeepers {
-                let sk = self
-                    .env
-                    .safekeepers
-                    .iter()
-                    .find(|node| node.id == sk_id)
-                    .ok_or_else(|| anyhow!("safekeeper {sk_id} does not exist"))?;
-                safekeeper_connstrings.push(format!("127.0.0.1:{}", sk.get_compute_port()));
-            }
-        }
+        let safekeeper_connstrings = self.build_safekeepers_connstrs(safekeepers)?;

        // check for file remote_extensions_spec.json
        // if it is present, read it and pass to compute_ctl
@@ -592,7 +598,6 @@ impl Endpoint {
            remote_extensions,
            pgbouncer_settings: None,
            shard_stripe_size: Some(shard_stripe_size),
-            primary_is_running: None,
        };
        let spec_path = self.endpoint_path().join("spec.json");
        std::fs::write(spec_path, serde_json::to_string_pretty(&spec)?)?;
@@ -741,6 +746,7 @@ impl Endpoint {
        &self,
        mut pageservers: Vec<(Host, u16)>,
        stripe_size: Option<ShardStripeSize>,
+        safekeepers: Option<Vec<NodeId>>,
    ) -> Result<()> {
        let mut spec: ComputeSpec = {
            let spec_path = self.endpoint_path().join("spec.json");
@@ -775,6 +781,12 @@ impl Endpoint {
            spec.shard_stripe_size = stripe_size.map(|s| s.0 as usize);
        }

+        // If safekeepers are not specified, don't change them.
+        if let Some(safekeepers) = safekeepers {
+            let safekeeper_connstrings = self.build_safekeepers_connstrs(safekeepers)?;
+            spec.safekeeper_connstrings = safekeeper_connstrings;
+        }
+
        let client = reqwest::Client::builder()
            .timeout(Duration::from_secs(30))
            .build()
@@ -17,8 +17,7 @@ use anyhow::{bail, Context};
 use camino::Utf8PathBuf;
 use futures::SinkExt;
 use pageserver_api::models::{
-    self, AuxFilePolicy, LocationConfig, ShardParameters, TenantHistorySize, TenantInfo,
-    TimelineInfo,
+    self, AuxFilePolicy, LocationConfig, TenantHistorySize, TenantInfo, TimelineInfo,
 };
 use pageserver_api::shard::TenantShardId;
 use pageserver_client::mgmt_api;
@@ -397,28 +396,6 @@ impl PageServerNode {
        }
    }

-    pub async fn tenant_create(
-        &self,
-        new_tenant_id: TenantId,
-        generation: Option<u32>,
-        settings: HashMap<&str, &str>,
-    ) -> anyhow::Result<TenantId> {
-        let config = Self::parse_config(settings.clone())?;
-
-        let request = models::TenantCreateRequest {
-            new_tenant_id: TenantShardId::unsharded(new_tenant_id),
-            generation,
-            config,
-            shard_parameters: ShardParameters::default(),
-            // Placement policy is not meaningful for creations not done via storage controller
-            placement_policy: None,
-        };
-        if !settings.is_empty() {
-            bail!("Unrecognized tenant settings: {settings:?}")
-        }
-        Ok(self.http_client.tenant_create(&request).await?)
-    }
-
    pub async fn tenant_config(
        &self,
        tenant_id: TenantId,
@@ -5,12 +5,11 @@ use crate::{
 use camino::{Utf8Path, Utf8PathBuf};
 use pageserver_api::{
    controller_api::{
-        NodeConfigureRequest, NodeRegisterRequest, TenantCreateResponse, TenantLocateResponse,
-        TenantShardMigrateRequest, TenantShardMigrateResponse,
+        NodeConfigureRequest, NodeRegisterRequest, TenantCreateRequest, TenantCreateResponse,
+        TenantLocateResponse, TenantShardMigrateRequest, TenantShardMigrateResponse,
    },
    models::{
-        TenantCreateRequest, TenantShardSplitRequest, TenantShardSplitResponse,
-        TimelineCreateRequest, TimelineInfo,
+        TenantShardSplitRequest, TenantShardSplitResponse, TimelineCreateRequest, TimelineInfo,
    },
    shard::{ShardStripeSize, TenantShardId},
 };
@@ -1,16 +1,16 @@
 use futures::StreamExt;
-use std::{collections::HashMap, str::FromStr, time::Duration};
+use std::{str::FromStr, time::Duration};

 use clap::{Parser, Subcommand};
 use pageserver_api::{
    controller_api::{
-        NodeAvailabilityWrapper, NodeDescribeResponse, ShardSchedulingPolicy,
+        NodeAvailabilityWrapper, NodeDescribeResponse, ShardSchedulingPolicy, TenantCreateRequest,
        TenantDescribeResponse, TenantPolicyRequest,
    },
    models::{
        EvictionPolicy, EvictionPolicyLayerAccessThreshold, LocationConfigSecondary,
-        ShardParameters, TenantConfig, TenantConfigRequest, TenantCreateRequest,
-        TenantShardSplitRequest, TenantShardSplitResponse,
+        ShardParameters, TenantConfig, TenantConfigRequest, TenantShardSplitRequest,
+        TenantShardSplitResponse,
    },
    shard::{ShardStripeSize, TenantShardId},
 };
@@ -21,7 +21,7 @@ use utils::id::{NodeId, TenantId};

 use pageserver_api::controller_api::{
    NodeConfigureRequest, NodeRegisterRequest, NodeSchedulingPolicy, PlacementPolicy,
-    TenantLocateResponse, TenantShardMigrateRequest, TenantShardMigrateResponse,
+    TenantShardMigrateRequest, TenantShardMigrateResponse,
 };

 #[derive(Subcommand, Debug)]
@@ -110,12 +110,6 @@ enum Command {
        #[arg(long)]
        config: String,
    },
-    /// Attempt to balance the locations for a tenant across pageservers.  This is a client-side
-    /// alternative to the storage controller's scheduling optimization behavior.
-    TenantScatter {
-        #[arg(long)]
-        tenant_id: TenantId,
-    },
    /// Print details about a particular tenant, including all its shards' states.
    TenantDescribe {
        #[arg(long)]
@@ -342,14 +336,18 @@ async fn main() -> anyhow::Result<()> {
                .await?;
        }
        Command::TenantCreate { tenant_id } => {
-            vps_client
-                .tenant_create(&TenantCreateRequest {
-                    new_tenant_id: TenantShardId::unsharded(tenant_id),
-                    generation: None,
-                    shard_parameters: ShardParameters::default(),
-                    placement_policy: Some(PlacementPolicy::Attached(1)),
-                    config: TenantConfig::default(),
-                })
+            storcon_client
+                .dispatch(
+                    Method::POST,
+                    "v1/tenant".to_string(),
+                    Some(TenantCreateRequest {
+                        new_tenant_id: TenantShardId::unsharded(tenant_id),
+                        generation: None,
+                        shard_parameters: ShardParameters::default(),
+                        placement_policy: Some(PlacementPolicy::Attached(1)),
+                        config: TenantConfig::default(),
+                    }),
+                )
                .await?;
        }
        Command::TenantDelete { tenant_id } => {
@@ -498,88 +496,6 @@ async fn main() -> anyhow::Result<()> {
                })
                .await?;
        }
-        Command::TenantScatter { tenant_id } => {
-            // Find the shards
-            let locate_response = storcon_client
-                .dispatch::<(), TenantLocateResponse>(
-                    Method::GET,
-                    format!("control/v1/tenant/{tenant_id}/locate"),
-                    None,
-                )
-                .await?;
-            let shards = locate_response.shards;
-
-            let mut node_to_shards: HashMap<NodeId, Vec<TenantShardId>> = HashMap::new();
-            let shard_count = shards.len();
-            for s in shards {
-                let entry = node_to_shards.entry(s.node_id).or_default();
-                entry.push(s.shard_id);
-            }
-
-            // Load list of available nodes
-            let nodes_resp = storcon_client
-                .dispatch::<(), Vec<NodeDescribeResponse>>(
-                    Method::GET,
-                    "control/v1/node".to_string(),
-                    None,
-                )
-                .await?;
-
-            for node in nodes_resp {
-                if matches!(node.availability, NodeAvailabilityWrapper::Active) {
-                    node_to_shards.entry(node.id).or_default();
-                }
-            }
-
-            let max_shard_per_node = shard_count / node_to_shards.len();
-
-            loop {
-                let mut migrate_shard = None;
-                for shards in node_to_shards.values_mut() {
-                    if shards.len() > max_shard_per_node {
-                        // Pick the emptiest
-                        migrate_shard = Some(shards.pop().unwrap());
-                    }
-                }
-                let Some(migrate_shard) = migrate_shard else {
-                    break;
-                };
-
-                // Pick the emptiest node to migrate to
-                let mut destinations = node_to_shards
-                    .iter()
-                    .map(|(k, v)| (k, v.len()))
-                    .collect::<Vec<_>>();
-                destinations.sort_by_key(|i| i.1);
-                let (destination_node, destination_count) = *destinations.first().unwrap();
-                if destination_count + 1 > max_shard_per_node {
-                    // Even the emptiest destination doesn't have space: we're done
-                    break;
-                }
-                let destination_node = *destination_node;
-
-                node_to_shards
-                    .get_mut(&destination_node)
-                    .unwrap()
-                    .push(migrate_shard);
-
-                println!("Migrate {} -> {} ...", migrate_shard, destination_node);
-
-                storcon_client
-                    .dispatch::<TenantShardMigrateRequest, TenantShardMigrateResponse>(
-                        Method::PUT,
-                        format!("control/v1/tenant/{migrate_shard}/migrate"),
-                        Some(TenantShardMigrateRequest {
-                            tenant_shard_id: migrate_shard,
-                            node_id: destination_node,
-                        }),
-                    )
-                    .await?;
-                println!("Migrate {} -> {} OK", migrate_shard, destination_node);
-            }
-
-            // Spread the shards across the nodes
-        }
        Command::TenantDescribe { tenant_id } => {
            let describe_response = storcon_client
                .dispatch::<(), TenantDescribeResponse>(
@@ -0,0 +1,10 @@
+
+# Example docker compose configuration
+
+The configuration in this directory is used for testing Neon docker images: it is
+not intended for deploying a usable system.  To run a development environment where
+you can experiment with a minature Neon system, use `cargo neon` rather than container images.
+
+This configuration does not start the storage controller, because the controller
+needs a way to reconfigure running computes, and no such thing exists in this setup.
+
@@ -23,11 +23,10 @@ echo "Page server is ready."
 echo "Create a tenant and timeline"
 generate_id tenant_id
 PARAMS=(
-     -sb 
-     -X POST
+     -X PUT
     -H "Content-Type: application/json"
-     -d "{\"new_tenant_id\": \"${tenant_id}\"}"
-     http://pageserver:9898/v1/tenant/
+     -d "{\"mode\": \"AttachedSingle\", \"generation\": 1, \"tenant_conf\": {}}"
+     "http://pageserver:9898/v1/tenant/${tenant_id}/location_config"
 )
 result=$(curl "${PARAMS[@]}")
 echo $result | jq .
@@ -96,12 +96,6 @@ pub struct ComputeSpec {
    // Stripe size for pageserver sharding, in pages
    #[serde(default)]
    pub shard_stripe_size: Option<usize>,
-
-    // When we are starting a new replica in hot standby mode,
-    // we need to know if the primary is running.
-    // This is used to determine if replica should wait for
-    // RUNNING_XACTS from primary or not.
-    pub primary_is_running: Option<bool>,
 }

 /// Feature flag to signal `compute_ctl` to enable certain experimental functionality.
@@ -103,9 +103,10 @@ static MAXRSS_KB: Lazy<IntGauge> = Lazy::new(|| {
    .expect("Failed to register maxrss_kb int gauge")
 });

-pub const DISK_WRITE_SECONDS_BUCKETS: &[f64] = &[
-    0.000_050, 0.000_100, 0.000_500, 0.001, 0.003, 0.005, 0.01, 0.05, 0.1, 0.3, 0.5,
-];
+/// Most common fsync latency is 50 µs - 100 µs, but it can be much higher,
+/// especially during many concurrent disk operations.
+pub const DISK_FSYNC_SECONDS_BUCKETS: &[f64] =
+    &[0.001, 0.005, 0.01, 0.05, 0.1, 0.5, 1.0, 5.0, 10.0, 30.0];

 pub struct BuildInfo {
    pub revision: &'static str,
@@ -11,6 +11,27 @@ use crate::{
    shard::{ShardStripeSize, TenantShardId},
 };

+#[derive(Serialize, Deserialize, Debug)]
+#[serde(deny_unknown_fields)]
+pub struct TenantCreateRequest {
+    pub new_tenant_id: TenantShardId,
+    #[serde(default)]
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub generation: Option<u32>,
+
+    // If omitted, create a single shard with TenantShardId::unsharded()
+    #[serde(default)]
+    #[serde(skip_serializing_if = "ShardParameters::is_unsharded")]
+    pub shard_parameters: ShardParameters,
+
+    #[serde(default)]
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub placement_policy: Option<PlacementPolicy>,
+
+    #[serde(flatten)]
+    pub config: TenantConfig, // as we have a flattened field, we should reject all unknown fields in it
+}
+
 #[derive(Serialize, Deserialize)]
 pub struct TenantCreateResponseShard {
    pub shard_id: TenantShardId,
@@ -280,4 +301,19 @@ mod test {
        assert_eq!(serde_json::from_str::<PlacementPolicy>(&encoded)?, v);
        Ok(())
    }
+
+    #[test]
+    fn test_reject_unknown_field() {
+        let id = TenantId::generate();
+        let create_request = serde_json::json!({
+            "new_tenant_id": id.to_string(),
+            "unknown_field": "unknown_value".to_string(),
+        });
+        let err = serde_json::from_value::<TenantCreateRequest>(create_request).unwrap_err();
+        assert!(
+            err.to_string().contains("unknown field `unknown_field`"),
+            "expect unknown field `unknown_field` error, got: {}",
+            err
+        );
+    }
 }
@@ -160,8 +160,9 @@ impl Key {
        key
    }

-    /// Convert a 18B slice to a key. This function should not be used for metadata keys because field2 is handled differently.
-    /// Use [`Key::from_i128`] instead if you want to handle 16B keys (i.e., metadata keys).
+    /// Convert a 18B slice to a key. This function should not be used for 16B metadata keys because `field2` is handled differently.
+    /// Use [`Key::from_i128`] instead if you want to handle 16B keys (i.e., metadata keys). There are some restrictions on `field2`,
+    /// and therefore not all 18B slices are valid page server keys.
    pub fn from_slice(b: &[u8]) -> Self {
        Key {
            field1: b[0],
@@ -173,7 +174,7 @@ impl Key {
        }
    }

-    /// Convert a key to a 18B slice. This function should not be used for metadata keys because field2 is handled differently.
+    /// Convert a key to a 18B slice. This function should not be used for getting a 16B metadata key because `field2` is handled differently.
    /// Use [`Key::to_i128`] instead if you want to get a 16B key (i.e., metadata keys).
    pub fn write_to_byte_slice(&self, buf: &mut [u8]) {
        buf[0] = self.field1;
@@ -25,7 +25,6 @@ use utils::{
    serde_system_time,
 };

-use crate::controller_api::PlacementPolicy;
 use crate::{
    reltag::RelTag,
    shard::{ShardCount, ShardStripeSize, TenantShardId},
@@ -271,28 +270,6 @@ impl Default for ShardParameters {
    }
 }

-#[derive(Serialize, Deserialize, Debug)]
-#[serde(deny_unknown_fields)]
-pub struct TenantCreateRequest {
-    pub new_tenant_id: TenantShardId,
-    #[serde(default)]
-    #[serde(skip_serializing_if = "Option::is_none")]
-    pub generation: Option<u32>,
-
-    // If omitted, create a single shard with TenantShardId::unsharded()
-    #[serde(default)]
-    #[serde(skip_serializing_if = "ShardParameters::is_unsharded")]
-    pub shard_parameters: ShardParameters,
-
-    // This parameter is only meaningful in requests sent to the storage controller
-    #[serde(default)]
-    #[serde(skip_serializing_if = "Option::is_none")]
-    pub placement_policy: Option<PlacementPolicy>,
-
-    #[serde(flatten)]
-    pub config: TenantConfig, // as we have a flattened field, we should reject all unknown fields in it
-}
-
 /// An alternative representation of `pageserver::tenant::TenantConf` with
 /// simpler types.
 #[derive(Serialize, Deserialize, Debug, Default, Clone, Eq, PartialEq)]
@@ -547,10 +524,6 @@ pub struct LocationConfigListResponse {
    pub tenant_shards: Vec<(TenantShardId, Option<LocationConfig>)>,
 }

-#[derive(Serialize, Deserialize)]
-#[serde(transparent)]
-pub struct TenantCreateResponse(pub TenantId);
-
 #[derive(Serialize)]
 pub struct StatusResponse {
    pub id: NodeId,
@@ -607,31 +580,6 @@ impl TenantConfigRequest {
    }
 }

-#[derive(Debug, Deserialize)]
-pub struct TenantAttachRequest {
-    #[serde(default)]
-    pub config: TenantAttachConfig,
-    #[serde(default)]
-    pub generation: Option<u32>,
-}
-
-/// Newtype to enforce deny_unknown_fields on TenantConfig for
-/// its usage inside `TenantAttachRequest`.
-#[derive(Debug, Serialize, Deserialize, Default)]
-#[serde(deny_unknown_fields)]
-pub struct TenantAttachConfig {
-    #[serde(flatten)]
-    allowing_unknown_fields: TenantConfig,
-}
-
-impl std::ops::Deref for TenantAttachConfig {
-    type Target = TenantConfig;
-
-    fn deref(&self) -> &Self::Target {
-        &self.allowing_unknown_fields
-    }
-}
-
 /// See [`TenantState::attachment_status`] and the OpenAPI docs for context.
 #[derive(Serialize, Deserialize, Clone)]
 #[serde(tag = "slug", content = "data", rename_all = "snake_case")]
@@ -650,8 +598,7 @@ pub struct TenantInfo {
    /// If a layer is present in both local FS and S3, it counts only once.
    pub current_physical_size: Option<u64>, // physical size is only included in `tenant_status` endpoint
    pub attachment_status: TenantAttachmentStatus,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    pub generation: Option<u32>,
+    pub generation: u32,
 }

 #[derive(Serialize, Deserialize, Clone)]
@@ -1478,7 +1425,7 @@ mod tests {
            state: TenantState::Active,
            current_physical_size: Some(42),
            attachment_status: TenantAttachmentStatus::Attached,
-            generation: None,
+            generation: 1,
        };
        let expected_active = json!({
            "id": original_active.id.to_string(),
@@ -1488,7 +1435,8 @@ mod tests {
            "current_physical_size": 42,
            "attachment_status": {
                "slug":"attached",
-            }
+            },
+            "generation" : 1
        });

        let original_broken = TenantInfo {
@@ -1499,7 +1447,7 @@ mod tests {
            },
            current_physical_size: Some(42),
            attachment_status: TenantAttachmentStatus::Attached,
-            generation: None,
+            generation: 1,
        };
        let expected_broken = json!({
            "id": original_broken.id.to_string(),
@@ -1513,7 +1461,8 @@ mod tests {
            "current_physical_size": 42,
            "attachment_status": {
                "slug":"attached",
-            }
+            },
+            "generation" : 1
        });

        assert_eq!(
@@ -1531,18 +1480,6 @@ mod tests {

    #[test]
    fn test_reject_unknown_field() {
-        let id = TenantId::generate();
-        let create_request = json!({
-            "new_tenant_id": id.to_string(),
-            "unknown_field": "unknown_value".to_string(),
-        });
-        let err = serde_json::from_value::<TenantCreateRequest>(create_request).unwrap_err();
-        assert!(
-            err.to_string().contains("unknown field `unknown_field`"),
-            "expect unknown field `unknown_field` error, got: {}",
-            err
-        );
-
        let id = TenantId::generate();
        let config_request = json!({
            "tenant_id": id.to_string(),
@@ -1554,18 +1491,6 @@ mod tests {
            "expect unknown field `unknown_field` error, got: {}",
            err
        );
-
-        let attach_request = json!({
-            "config": {
-                "unknown_field": "unknown_value".to_string(),
-            },
-        });
-        let err = serde_json::from_value::<TenantAttachRequest>(attach_request).unwrap_err();
-        assert!(
-            err.to_string().contains("unknown field `unknown_field`"),
-            "expect unknown field `unknown_field` error, got: {}",
-            err
-        );
    }

    #[test]
@@ -356,6 +356,28 @@ impl CheckPoint {
        }
        false
    }
+
+    /// Advance next multi-XID/offset to those given in arguments.
+    ///
+    /// It's important that this handles wraparound correctly. This should match the
+    /// MultiXactAdvanceNextMXact() logic in PostgreSQL's xlog_redo() function.
+    ///
+    /// Returns 'true' if the Checkpoint was updated.
+    pub fn update_next_multixid(&mut self, multi_xid: u32, multi_offset: u32) -> bool {
+        let mut modified = false;
+
+        if multi_xid.wrapping_sub(self.nextMulti) as i32 > 0 {
+            self.nextMulti = multi_xid;
+            modified = true;
+        }
+
+        if multi_offset.wrapping_sub(self.nextMultiOffset) as i32 > 0 {
+            self.nextMultiOffset = multi_offset;
+            modified = true;
+        }
+
+        modified
+    }
 }

 /// Generate new, empty WAL segment, with correct block headers at the first
@@ -202,6 +202,53 @@ pub fn test_update_next_xid() {
    assert_eq!(checkpoint.nextXid.value, 2048);
 }

+#[test]
+pub fn test_update_next_multixid() {
+    let checkpoint_buf = [0u8; std::mem::size_of::<CheckPoint>()];
+    let mut checkpoint = CheckPoint::decode(&checkpoint_buf).unwrap();
+
+    // simple case
+    checkpoint.nextMulti = 20;
+    checkpoint.nextMultiOffset = 20;
+    checkpoint.update_next_multixid(1000, 2000);
+    assert_eq!(checkpoint.nextMulti, 1000);
+    assert_eq!(checkpoint.nextMultiOffset, 2000);
+
+    // No change
+    checkpoint.update_next_multixid(500, 900);
+    assert_eq!(checkpoint.nextMulti, 1000);
+    assert_eq!(checkpoint.nextMultiOffset, 2000);
+
+    // Close to wraparound, but not wrapped around yet
+    checkpoint.nextMulti = 0xffff0000;
+    checkpoint.nextMultiOffset = 0xfffe0000;
+    checkpoint.update_next_multixid(0xffff00ff, 0xfffe00ff);
+    assert_eq!(checkpoint.nextMulti, 0xffff00ff);
+    assert_eq!(checkpoint.nextMultiOffset, 0xfffe00ff);
+
+    // Wraparound
+    checkpoint.update_next_multixid(1, 900);
+    assert_eq!(checkpoint.nextMulti, 1);
+    assert_eq!(checkpoint.nextMultiOffset, 900);
+
+    // Wraparound nextMulti to 0.
+    //
+    // It's a bit surprising that nextMulti can be 0, because that's a special value
+    // (InvalidMultiXactId). However, that's how Postgres does it at multi-xid wraparound:
+    // nextMulti wraps around to 0, but then when the next multi-xid is assigned, it skips
+    // the 0 and the next multi-xid actually assigned is 1.
+    checkpoint.nextMulti = 0xffff0000;
+    checkpoint.nextMultiOffset = 0xfffe0000;
+    checkpoint.update_next_multixid(0, 0xfffe00ff);
+    assert_eq!(checkpoint.nextMulti, 0);
+    assert_eq!(checkpoint.nextMultiOffset, 0xfffe00ff);
+
+    // Wraparound nextMultiOffset to 0
+    checkpoint.update_next_multixid(0, 0);
+    assert_eq!(checkpoint.nextMulti, 0);
+    assert_eq!(checkpoint.nextMultiOffset, 0);
+}
+
 #[test]
 pub fn test_encode_logical_message() {
    let expected = [
@@ -9,20 +9,11 @@ use serde::{Deserialize, Serialize};
 /// numbers are used.
 #[derive(Copy, Clone, Eq, PartialEq, PartialOrd, Ord, Hash)]
 pub enum Generation {
-    // Generations with this magic value will not add a suffix to S3 keys, and will not
-    // be included in persisted index_part.json.  This value is only to be used
-    // during migration from pre-generation metadata to generation-aware metadata,
-    // and should eventually go away.
-    //
-    // A special Generation is used rather than always wrapping Generation in an Option,
-    // so that code handling generations doesn't have to be aware of the legacy
-    // case everywhere it touches a generation.
+    // The None Generation is used in the metadata of layers written before generations were
+    // introduced.  A running Tenant always has a valid generation, but the layer metadata may
+    // include None generations.
    None,
-    // Generations with this magic value may never be used to construct S3 keys:
-    // we will panic if someone tries to.  This is for Tenants in the "Broken" state,
-    // so that we can satisfy their constructor with a Generation without risking
-    // a code bug using it in an S3 write (broken tenants should never write)
-    Broken,
+
    Valid(u32),
 }

@@ -42,11 +33,6 @@ impl Generation {
        Self::None
    }

-    // Create a new generation that will panic if you try to use get_suffix
-    pub fn broken() -> Self {
-        Self::Broken
-    }
-
    pub const fn new(v: u32) -> Self {
        Self::Valid(v)
    }
@@ -60,9 +46,6 @@ impl Generation {
        match self {
            Self::Valid(v) => GenerationFileSuffix(Some(*v)),
            Self::None => GenerationFileSuffix(None),
-            Self::Broken => {
-                panic!("Tried to use a broken generation");
-            }
        }
    }

@@ -86,7 +69,6 @@ impl Generation {
                }
            }
            Self::None => Self::None,
-            Self::Broken => panic!("Attempted to use a broken generation"),
        }
    }

@@ -95,7 +77,6 @@ impl Generation {
        match self {
            Self::Valid(n) => Self::Valid(*n + 1),
            Self::None => Self::Valid(1),
-            Self::Broken => panic!("Attempted to use a broken generation"),
        }
    }

@@ -128,7 +109,7 @@ impl Serialize for Generation {
        if let Self::Valid(v) = self {
            v.serialize(serializer)
        } else {
-            // We should never be asked to serialize a None or Broken.  Structures
+            // We should never be asked to serialize a None. Structures
            // that include an optional generation should convert None to an
            // Option<Generation>::None
            Err(serde::ser::Error::custom(
@@ -159,9 +140,6 @@ impl Debug for Generation {
            Self::None => {
                write!(f, "<none>")
            }
-            Self::Broken => {
-                write!(f, "<broken>")
-            }
        }
    }
 }
@@ -8,22 +8,15 @@ use super::error::ApiError;
 pub async fn json_request<T: for<'de> Deserialize<'de>>(
    request: &mut Request<Body>,
 ) -> Result<T, ApiError> {
-    json_request_or_empty_body(request)
-        .await?
-        .context("missing request body")
-        .map_err(ApiError::BadRequest)
-}
-
-/// Will be removed as part of <https://github.com/neondatabase/neon/issues/4282>
-pub async fn json_request_or_empty_body<T: for<'de> Deserialize<'de>>(
-    request: &mut Request<Body>,
-) -> Result<Option<T>, ApiError> {
    let body = hyper::body::aggregate(request.body_mut())
        .await
        .context("Failed to read request body")
        .map_err(ApiError::BadRequest)?;
+
    if body.remaining() == 0 {
-        return Ok(None);
+        return Err(ApiError::BadRequest(anyhow::anyhow!(
+            "missing request body"
+        )));
    }

    let mut deser = serde_json::de::Deserializer::from_reader(body.reader());
@@ -31,7 +24,6 @@ pub async fn json_request_or_empty_body<T: for<'de> Deserialize<'de>>(
    serde_path_to_error::deserialize(&mut deser)
        // intentionally stringify because the debug version is not helpful in python logs
        .map_err(|e| anyhow::anyhow!("Failed to parse json request: {e}"))
-        .map(Some)
        .map_err(ApiError::BadRequest)
 }

@@ -1,3 +1,5 @@
+#![allow(clippy::todo)]
+
 use std::ffi::CString;

 use crate::{
@@ -48,6 +48,7 @@
 //! medium/128        time:   [8.8311 ms 8.9849 ms 9.1263 ms]
 //! ```

+use anyhow::Context;
 use bytes::{Buf, Bytes};
 use criterion::{BenchmarkId, Criterion};
 use pageserver::{config::PageServerConf, walrecord::NeonWalRecord, walredo::PostgresRedoManager};
@@ -188,6 +189,7 @@ impl Request {
        manager
            .request_redo(*key, *lsn, base_img.clone(), records.clone(), *pg_version)
            .await
+            .context("request_redo")
    }

    fn pg_record(will_init: bool, bytes: &'static [u8]) -> NeonWalRecord {
@@ -205,15 +205,6 @@ impl Client {
        Ok(())
    }

-    pub async fn tenant_create(&self, req: &TenantCreateRequest) -> Result<TenantId> {
-        let uri = format!("{}/v1/tenant", self.mgmt_api_endpoint);
-        self.request(Method::POST, &uri, req)
-            .await?
-            .json()
-            .await
-            .map_err(Error::ReceiveBody)
-    }
-
    /// The tenant deletion API can return 202 if deletion is incomplete, or
    /// 404 if it is complete.  Callers are responsible for checking the status
    /// code and retrying.  Error codes other than 404 will return Err().
@@ -83,10 +83,18 @@ fn parse_filename(name: &str) -> (Range<Key>, Range<Lsn>) {
    let keys: Vec<&str> = split[0].split('-').collect();
    let mut lsns: Vec<&str> = split[1].split('-').collect();

+    // The current format of the layer file name: 000000067F0000000400000B150100000000-000000067F0000000400000D350100000000__00000000014B7AC8-v1-00000001
+
+    // Handle generation number `-00000001` part
    if lsns.last().expect("should").len() == 8 {
        lsns.pop();
    }

+    // Handle version number `-v1` part
+    if lsns.last().expect("should").starts_with('v') {
+        lsns.pop();
+    }
+
    if lsns.len() == 1 {
        lsns.push(lsns[0]);
    }
@@ -36,10 +36,7 @@ use crate::tenant::{config::TenantConfOpt, timeline::GetImpl};
 use crate::tenant::{TENANTS_SEGMENT_NAME, TIMELINES_SEGMENT_NAME};
 use crate::{disk_usage_eviction_task::DiskUsageEvictionTaskConfig, virtual_file::io_engine};
 use crate::{tenant::config::TenantConf, virtual_file};
-use crate::{
-    TENANT_CONFIG_NAME, TENANT_HEATMAP_BASENAME, TENANT_LOCATION_CONFIG_NAME,
-    TIMELINE_DELETE_MARK_SUFFIX,
-};
+use crate::{TENANT_HEATMAP_BASENAME, TENANT_LOCATION_CONFIG_NAME, TIMELINE_DELETE_MARK_SUFFIX};

 use self::defaults::DEFAULT_CONCURRENT_TENANT_WARMUP;

@@ -810,15 +807,11 @@ impl PageServerConf {
    }

    /// Points to a place in pageserver's local directory,
-    /// where certain tenant's tenantconf file should be located.
-    ///
-    /// Legacy: superseded by tenant_location_config_path.  Eventually
-    /// remove this function.
-    pub fn tenant_config_path(&self, tenant_shard_id: &TenantShardId) -> Utf8PathBuf {
-        self.tenant_path(tenant_shard_id).join(TENANT_CONFIG_NAME)
-    }
-
-    pub fn tenant_location_config_path(&self, tenant_shard_id: &TenantShardId) -> Utf8PathBuf {
+    /// where certain tenant's LocationConf be stored.
+    pub(crate) fn tenant_location_config_path(
+        &self,
+        tenant_shard_id: &TenantShardId,
+    ) -> Utf8PathBuf {
        self.tenant_path(tenant_shard_id)
            .join(TENANT_LOCATION_CONFIG_NAME)
    }
@@ -382,17 +382,6 @@ pub enum DeletionQueueError {
 }

 impl DeletionQueueClient {
-    pub(crate) fn broken() -> Self {
-        // Channels whose receivers are immediately dropped.
-        let (tx, _rx) = tokio::sync::mpsc::unbounded_channel();
-        let (executor_tx, _executor_rx) = tokio::sync::mpsc::channel(1);
-        Self {
-            tx,
-            executor_tx,
-            lsn_table: Arc::default(),
-        }
-    }
-
    /// This is cancel-safe.  If you drop the future before it completes, the message
    /// is not pushed, although in the context of the deletion queue it doesn't matter: once
    /// we decide to do a deletion the decision is always final.
@@ -236,6 +236,13 @@ paths:
            type: string
            format: date-time
          description: A timestamp to get the LSN
+        - name: with_lease
+          in: query
+          required: false
+          schema:
+            type: boolean
+          description: Whether to grant a lease to the corresponding LSN. Default to false.
+
      responses:
        "200":
          description: OK
@@ -360,16 +367,7 @@ paths:
                $ref: "#/components/schemas/TenantLocationConfigResponse"
        "409":
          description: |
-            The tenant is already known to Pageserver in some way,
-            and hence this `/attach` call has been rejected.
-
-            Some examples of how this can happen:
-            - tenant was created on this pageserver
-            - tenant attachment was started by an earlier call to `/attach`.
-
-            Callers should poll the tenant status's `attachment_status` field,
-            like for status 202. See the longer description for `POST /attach`
-            for details.
+            The tenant is already being modified, perhaps by a concurrent call to this API
          content:
            application/json:
              schema:
@@ -755,8 +753,6 @@ components:
              For example this can be caused by s3 being unreachable. The retry may be implemented
              with call to detach, though it would be better to not automate it and inspec failed state
              manually before proceeding with a retry.
-
-            See the tenant `/attach` endpoint for more information.
          type: object
          required:
            - slug
@@ -1029,6 +1025,10 @@ components:
        kind:
          type: string
          enum: [past, present, future, nodata]
+        valid_until:
+          type: string
+          format: date-time
+          description: The expiration time of the granted lease.

    LsnLease:
      type: object
@@ -21,6 +21,7 @@ use pageserver_api::models::IngestAuxFilesRequest;
 use pageserver_api::models::ListAuxFilesRequest;
 use pageserver_api::models::LocationConfig;
 use pageserver_api::models::LocationConfigListResponse;
+use pageserver_api::models::LsnLease;
 use pageserver_api::models::ShardParameters;
 use pageserver_api::models::TenantDetails;
 use pageserver_api::models::TenantLocationConfigResponse;
@@ -30,13 +31,11 @@ use pageserver_api::models::TenantShardLocation;
 use pageserver_api::models::TenantShardSplitRequest;
 use pageserver_api::models::TenantShardSplitResponse;
 use pageserver_api::models::TenantSorting;
-use pageserver_api::models::TenantState;
 use pageserver_api::models::TopTenantShardItem;
 use pageserver_api::models::TopTenantShardsRequest;
 use pageserver_api::models::TopTenantShardsResponse;
 use pageserver_api::models::{
-    DownloadRemoteLayersTaskSpawnRequest, LocationConfigMode, TenantAttachRequest,
-    TenantLocationConfigRequest,
+    DownloadRemoteLayersTaskSpawnRequest, LocationConfigMode, TenantLocationConfigRequest,
 };
 use pageserver_api::shard::ShardCount;
 use pageserver_api::shard::TenantShardId;
@@ -50,12 +49,10 @@ use utils::auth::JwtAuth;
 use utils::failpoint_support::failpoints_handler;
 use utils::http::endpoint::prometheus_metrics_handler;
 use utils::http::endpoint::request_span;
-use utils::http::json::json_request_or_empty_body;
 use utils::http::request::{get_request_param, must_get_query_param, parse_query_param};

 use crate::context::{DownloadBehavior, RequestContext};
 use crate::deletion_queue::DeletionQueueClient;
-use crate::metrics::{StorageTimeOperation, STORAGE_TIME_GLOBAL};
 use crate::pgdatadir_mapping::LsnForTimestamp;
 use crate::task_mgr::TaskKind;
 use crate::tenant::config::{LocationConf, TenantConfOpt};
@@ -77,13 +74,12 @@ use crate::tenant::timeline::CompactFlags;
 use crate::tenant::timeline::CompactionError;
 use crate::tenant::timeline::Timeline;
 use crate::tenant::GetTimelineError;
-use crate::tenant::SpawnMode;
 use crate::tenant::{LogicalSizeCalculationCause, PageReconstructError};
 use crate::{config::PageServerConf, tenant::mgr};
 use crate::{disk_usage_eviction_task, tenant};
 use pageserver_api::models::{
-    StatusResponse, TenantConfigRequest, TenantCreateRequest, TenantCreateResponse, TenantInfo,
-    TimelineCreateRequest, TimelineGcRequest, TimelineInfo,
+    StatusResponse, TenantConfigRequest, TenantInfo, TimelineCreateRequest, TimelineGcRequest,
+    TimelineInfo,
 };
 use utils::{
    auth::SwappableJwtAuth,
@@ -728,6 +724,8 @@ async fn get_lsn_by_timestamp_handler(
        .map_err(ApiError::BadRequest)?;
    let timestamp_pg = postgres_ffi::to_pg_timestamp(timestamp);

+    let with_lease = parse_query_param(&request, "with_lease")?.unwrap_or(false);
+
    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);

    let timeline =
@@ -736,10 +734,15 @@ async fn get_lsn_by_timestamp_handler(
    let result = timeline
        .find_lsn_for_timestamp(timestamp_pg, &cancel, &ctx)
        .await?;
+
    #[derive(serde::Serialize, Debug)]
    struct Result {
        lsn: Lsn,
        kind: &'static str,
+        #[serde(default)]
+        #[serde(skip_serializing_if = "Option::is_none")]
+        #[serde(flatten)]
+        lease: Option<LsnLease>,
    }
    let (lsn, kind) = match result {
        LsnForTimestamp::Present(lsn) => (lsn, "present"),
@@ -747,11 +750,28 @@ async fn get_lsn_by_timestamp_handler(
        LsnForTimestamp::Past(lsn) => (lsn, "past"),
        LsnForTimestamp::NoData(lsn) => (lsn, "nodata"),
    };
-    let result = Result { lsn, kind };
+
+    let lease = if with_lease {
+        timeline
+            .make_lsn_lease(lsn, timeline.get_lsn_lease_length_for_ts(), &ctx)
+            .inspect_err(|_| {
+                warn!("fail to grant a lease to {}", lsn);
+            })
+            .ok()
+    } else {
+        None
+    };
+
+    let result = Result { lsn, kind, lease };
+    let valid_until = result
+        .lease
+        .as_ref()
+        .map(|l| humantime::format_rfc3339_millis(l.valid_until).to_string());
    tracing::info!(
        lsn=?result.lsn,
        kind=%result.kind,
        timestamp=%timestamp_raw,
+        valid_until=?valid_until,
        "lsn_by_timestamp finished"
    );
    json_response(StatusCode::OK, result)
@@ -796,58 +816,6 @@ async fn get_timestamp_of_lsn_handler(
    }
 }

-async fn tenant_attach_handler(
-    mut request: Request<Body>,
-    _cancel: CancellationToken,
-) -> Result<Response<Body>, ApiError> {
-    let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
-    check_permission(&request, Some(tenant_id))?;
-
-    let maybe_body: Option<TenantAttachRequest> = json_request_or_empty_body(&mut request).await?;
-    let tenant_conf = match &maybe_body {
-        Some(request) => TenantConfOpt::try_from(&*request.config).map_err(ApiError::BadRequest)?,
-        None => TenantConfOpt::default(),
-    };
-
-    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
-
-    info!("Handling tenant attach {tenant_id}");
-
-    let state = get_state(&request);
-
-    let generation = get_request_generation(state, maybe_body.as_ref().and_then(|r| r.generation))?;
-
-    let tenant_shard_id = TenantShardId::unsharded(tenant_id);
-    let shard_params = ShardParameters::default();
-    let location_conf = LocationConf::attached_single(tenant_conf, generation, &shard_params);
-
-    let tenant = state
-        .tenant_manager
-        .upsert_location(tenant_shard_id, location_conf, None, SpawnMode::Eager, &ctx)
-        .await?;
-
-    let Some(tenant) = tenant else {
-        // This should never happen: indicates a bug in upsert_location
-        return Err(ApiError::InternalServerError(anyhow::anyhow!(
-            "Upsert succeeded but didn't return tenant!"
-        )));
-    };
-
-    // We might have successfully constructed a Tenant, but it could still
-    // end up in a broken state:
-    if let TenantState::Broken {
-        reason,
-        backtrace: _,
-    } = tenant.current_state()
-    {
-        return Err(ApiError::InternalServerError(anyhow::anyhow!(
-            "Tenant state is Broken: {reason}"
-        )));
-    }
-
-    json_response(StatusCode::ACCEPTED, ())
-}
-
 async fn timeline_delete_handler(
    request: Request<Body>,
    _cancel: CancellationToken,
@@ -878,26 +846,6 @@ async fn timeline_delete_handler(
    json_response(StatusCode::ACCEPTED, ())
 }

-async fn tenant_detach_handler(
-    request: Request<Body>,
-    _cancel: CancellationToken,
-) -> Result<Response<Body>, ApiError> {
-    let tenant_id: TenantId = parse_request_param(&request, "tenant_id")?;
-    check_permission(&request, Some(tenant_id))?;
-    // This is a legacy API (`/location_conf` is the replacement).  It only supports unsharded tenants
-    let tenant_shard_id = TenantShardId::unsharded(tenant_id);
-
-    let state = get_state(&request);
-    let conf = state.conf;
-    state
-        .tenant_manager
-        .detach_tenant(conf, tenant_shard_id, &state.deletion_queue_client)
-        .instrument(info_span!("tenant_detach", %tenant_id, shard_id=%tenant_shard_id.shard_slug()))
-        .await?;
-
-    json_response(StatusCode::OK, ())
-}
-
 async fn tenant_reset_handler(
    request: Request<Body>,
    _cancel: CancellationToken,
@@ -937,7 +885,9 @@ async fn tenant_list_handler(
            state: state.clone(),
            current_physical_size: None,
            attachment_status: state.attachment_status(),
-            generation: (*gen).into(),
+            generation: (*gen)
+                .into()
+                .expect("Tenants are always attached with a generation"),
        })
        .collect::<Vec<TenantInfo>>();

@@ -985,7 +935,10 @@ async fn tenant_status(
                state: state.clone(),
                current_physical_size: Some(current_physical_size),
                attachment_status: state.attachment_status(),
-                generation: tenant.generation().into(),
+                generation: tenant
+                    .generation()
+                    .into()
+                    .expect("Tenants are always attached with a generation"),
            },
            walredo: tenant.wal_redo_manager_status(),
            timelines: tenant.list_timeline_ids(),
@@ -1282,75 +1235,6 @@ pub fn html_response(status: StatusCode, data: String) -> Result<Response<Body>,
    Ok(response)
 }

-/// Helper for requests that may take a generation, which is mandatory
-/// when control_plane_api is set, but otherwise defaults to Generation::none()
-fn get_request_generation(state: &State, req_gen: Option<u32>) -> Result<Generation, ApiError> {
-    if state.conf.control_plane_api.is_some() {
-        req_gen
-            .map(Generation::new)
-            .ok_or(ApiError::BadRequest(anyhow!(
-                "generation attribute missing"
-            )))
-    } else {
-        // Legacy mode: all tenants operate with no generation
-        Ok(Generation::none())
-    }
-}
-
-async fn tenant_create_handler(
-    mut request: Request<Body>,
-    _cancel: CancellationToken,
-) -> Result<Response<Body>, ApiError> {
-    let request_data: TenantCreateRequest = json_request(&mut request).await?;
-    let target_tenant_id = request_data.new_tenant_id;
-    check_permission(&request, None)?;
-
-    let _timer = STORAGE_TIME_GLOBAL
-        .get_metric_with_label_values(&[StorageTimeOperation::CreateTenant.into()])
-        .expect("bug")
-        .start_timer();
-
-    let tenant_conf =
-        TenantConfOpt::try_from(&request_data.config).map_err(ApiError::BadRequest)?;
-
-    let state = get_state(&request);
-
-    let generation = get_request_generation(state, request_data.generation)?;
-
-    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Warn);
-
-    let location_conf =
-        LocationConf::attached_single(tenant_conf, generation, &request_data.shard_parameters);
-
-    let new_tenant = state
-        .tenant_manager
-        .upsert_location(
-            target_tenant_id,
-            location_conf,
-            None,
-            SpawnMode::Create,
-            &ctx,
-        )
-        .await?;
-
-    let Some(new_tenant) = new_tenant else {
-        // This should never happen: indicates a bug in upsert_location
-        return Err(ApiError::InternalServerError(anyhow::anyhow!(
-            "Upsert succeeded but didn't return tenant!"
-        )));
-    };
-    // We created the tenant. Existing API semantics are that the tenant
-    // is Active when this function returns.
-    new_tenant
-        .wait_to_become_active(ACTIVE_TENANT_TIMEOUT)
-        .await?;
-
-    json_response(
-        StatusCode::CREATED,
-        TenantCreateResponse(new_tenant.tenant_shard_id().tenant_id),
-    )
-}
-
 async fn get_tenant_config_handler(
    request: Request<Body>,
    _cancel: CancellationToken,
@@ -1702,6 +1586,14 @@ async fn timeline_compact_handler(
    if Some(true) == parse_query_param::<_, bool>(&request, "force_image_layer_creation")? {
        flags |= CompactFlags::ForceImageLayerCreation;
    }
+    if Some(true) == parse_query_param::<_, bool>(&request, "enhanced_gc_bottom_most_compaction")? {
+        if !cfg!(feature = "testing") {
+            return Err(ApiError::InternalServerError(anyhow!(
+                "enhanced_gc_bottom_most_compaction is only available in testing mode"
+            )));
+        }
+        flags |= CompactFlags::EnhancedGcBottomMostCompaction;
+    }
    let wait_until_uploaded =
        parse_query_param::<_, bool>(&request, "wait_until_uploaded")?.unwrap_or(false);

@@ -2648,7 +2540,6 @@ pub fn make_router(
            api_handler(r, reload_auth_validation_keys_handler)
        })
        .get("/v1/tenant", |r| api_handler(r, tenant_list_handler))
-        .post("/v1/tenant", |r| api_handler(r, tenant_create_handler))
        .get("/v1/tenant/:tenant_shard_id", |r| {
            api_handler(r, tenant_status)
        })
@@ -2686,12 +2577,6 @@ pub fn make_router(
        .post("/v1/tenant/:tenant_shard_id/timeline", |r| {
            api_handler(r, timeline_create_handler)
        })
-        .post("/v1/tenant/:tenant_id/attach", |r| {
-            api_handler(r, tenant_attach_handler)
-        })
-        .post("/v1/tenant/:tenant_id/detach", |r| {
-            api_handler(r, tenant_detach_handler)
-        })
        .post("/v1/tenant/:tenant_shard_id/reset", |r| {
            api_handler(r, tenant_reset_handler)
        })
@@ -113,11 +113,7 @@ pub async fn shutdown_pageserver(
 }

 /// Per-tenant configuration file.
-/// Full path: `tenants/<tenant_id>/config`.
-pub(crate) const TENANT_CONFIG_NAME: &str = "config";
-
-/// Per-tenant configuration file.
-/// Full path: `tenants/<tenant_id>/config`.
+/// Full path: `tenants/<tenant_id>/config-v1`.
 pub(crate) const TENANT_LOCATION_CONFIG_NAME: &str = "config-v1";

 /// Per-tenant copy of their remote heatmap, downloaded into the local
@@ -53,9 +53,6 @@ pub(crate) enum StorageTimeOperation {

    #[strum(serialize = "find gc cutoffs")]
    FindGcCutoffs,
-
-    #[strum(serialize = "create tenant")]
-    CreateTenant,
 }

 pub(crate) static STORAGE_TIME_SUM_PER_TIMELINE: Lazy<CounterVec> = Lazy::new(|| {
@@ -545,6 +542,15 @@ static AUX_FILE_SIZE: Lazy<IntGaugeVec> = Lazy::new(|| {
    .expect("failed to define a metric")
 });

+static VALID_LSN_LEASE_COUNT: Lazy<UIntGaugeVec> = Lazy::new(|| {
+    register_uint_gauge_vec!(
+        "pageserver_valid_lsn_lease_count",
+        "The number of valid leases after refreshing gc info.",
+        &["tenant_id", "shard_id", "timeline_id"],
+    )
+    .expect("failed to define a metric")
+});
+
 pub(crate) mod initial_logical_size {
    use metrics::{register_int_counter, register_int_counter_vec, IntCounter, IntCounterVec};
    use once_cell::sync::Lazy;
@@ -1436,6 +1442,46 @@ pub(crate) static LIVE_CONNECTIONS_COUNT: Lazy<IntGaugeVec> = Lazy::new(|| {
    .expect("failed to define a metric")
 });

+#[derive(Clone, Copy, enum_map::Enum, IntoStaticStr)]
+pub(crate) enum ComputeCommandKind {
+    PageStreamV2,
+    PageStream,
+    Basebackup,
+    GetLastRecordRlsn,
+    Fullbackup,
+    ImportBasebackup,
+    ImportWal,
+    LeaseLsn,
+    Show,
+}
+
+pub(crate) struct ComputeCommandCounters {
+    map: EnumMap<ComputeCommandKind, IntCounter>,
+}
+
+pub(crate) static COMPUTE_COMMANDS_COUNTERS: Lazy<ComputeCommandCounters> = Lazy::new(|| {
+    let inner = register_int_counter_vec!(
+        "pageserver_compute_commands",
+        "Number of compute -> pageserver commands processed",
+        &["command"]
+    )
+    .expect("failed to define a metric");
+
+    ComputeCommandCounters {
+        map: EnumMap::from_array(std::array::from_fn(|i| {
+            let command = <ComputeCommandKind as enum_map::Enum>::from_usize(i);
+            let command_str: &'static str = command.into();
+            inner.with_label_values(&[command_str])
+        })),
+    }
+});
+
+impl ComputeCommandCounters {
+    pub(crate) fn for_command(&self, command: ComputeCommandKind) -> &IntCounter {
+        &self.map[command]
+    }
+}
+
 // remote storage metrics

 static REMOTE_TIMELINE_CLIENT_CALLS: Lazy<IntCounterPairVec> = Lazy::new(|| {
@@ -2055,6 +2101,8 @@ pub(crate) struct TimelineMetrics {
    pub directory_entries_count_gauge: Lazy<UIntGauge, Box<dyn Send + Fn() -> UIntGauge>>,
    pub evictions: IntCounter,
    pub evictions_with_low_residence_duration: std::sync::RwLock<EvictionsWithLowResidenceDuration>,
+    /// Number of valid LSN leases.
+    pub valid_lsn_lease_count_gauge: UIntGauge,
    shutdown: std::sync::atomic::AtomicBool,
 }

@@ -2153,6 +2201,10 @@ impl TimelineMetrics {
        let evictions_with_low_residence_duration = evictions_with_low_residence_duration_builder
            .build(&tenant_id, &shard_id, &timeline_id);

+        let valid_lsn_lease_count_gauge = VALID_LSN_LEASE_COUNT
+            .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
+            .unwrap();
+
        TimelineMetrics {
            tenant_id,
            shard_id,
@@ -2175,6 +2227,7 @@ impl TimelineMetrics {
            evictions_with_low_residence_duration: std::sync::RwLock::new(
                evictions_with_low_residence_duration,
            ),
+            valid_lsn_lease_count_gauge,
            shutdown: std::sync::atomic::AtomicBool::default(),
        }
    }
@@ -2224,6 +2277,7 @@ impl TimelineMetrics {
        }
        let _ = EVICTIONS.remove_label_values(&[tenant_id, shard_id, timeline_id]);
        let _ = AUX_FILE_SIZE.remove_label_values(&[tenant_id, shard_id, timeline_id]);
+        let _ = VALID_LSN_LEASE_COUNT.remove_label_values(&[tenant_id, shard_id, timeline_id]);

        self.evictions_with_low_residence_duration
            .write()
@@ -2932,4 +2986,5 @@ pub fn preinitialize_metrics() {
    Lazy::force(&RECONSTRUCT_TIME);
    Lazy::force(&tenant_throttling::TIMELINE_GET);
    Lazy::force(&BASEBACKUP_QUERY_TIME);
+    Lazy::force(&COMPUTE_COMMANDS_COUNTERS);
 }
@@ -55,7 +55,7 @@ use crate::basebackup::BasebackupError;
 use crate::context::{DownloadBehavior, RequestContext};
 use crate::import_datadir::import_wal_from_tar;
 use crate::metrics;
-use crate::metrics::LIVE_CONNECTIONS_COUNT;
+use crate::metrics::{ComputeCommandKind, COMPUTE_COMMANDS_COUNTERS, LIVE_CONNECTIONS_COUNT};
 use crate::pgdatadir_mapping::Version;
 use crate::span::debug_assert_current_span_has_tenant_and_timeline_id;
 use crate::span::debug_assert_current_span_has_tenant_and_timeline_id_no_shard_id;
@@ -1554,6 +1554,10 @@ where

            self.check_permission(Some(tenant_id))?;

+            COMPUTE_COMMANDS_COUNTERS
+                .for_command(ComputeCommandKind::PageStreamV2)
+                .inc();
+
            self.handle_pagerequests(
                pgb,
                tenant_id,
@@ -1579,6 +1583,10 @@ where

            self.check_permission(Some(tenant_id))?;

+            COMPUTE_COMMANDS_COUNTERS
+                .for_command(ComputeCommandKind::PageStream)
+                .inc();
+
            self.handle_pagerequests(
                pgb,
                tenant_id,
@@ -1605,6 +1613,10 @@ where

            self.check_permission(Some(tenant_id))?;

+            COMPUTE_COMMANDS_COUNTERS
+                .for_command(ComputeCommandKind::Basebackup)
+                .inc();
+
            let lsn = if let Some(lsn_str) = params.get(2) {
                Some(
                    Lsn::from_str(lsn_str)
@@ -1662,6 +1674,11 @@ where
                .record("timeline_id", field::display(timeline_id));

            self.check_permission(Some(tenant_id))?;
+
+            COMPUTE_COMMANDS_COUNTERS
+                .for_command(ComputeCommandKind::GetLastRecordRlsn)
+                .inc();
+
            async {
                let timeline = self
                    .get_active_tenant_timeline(tenant_id, timeline_id, ShardSelector::Zero)
@@ -1723,6 +1740,10 @@ where

            self.check_permission(Some(tenant_id))?;

+            COMPUTE_COMMANDS_COUNTERS
+                .for_command(ComputeCommandKind::Fullbackup)
+                .inc();
+
            // Check that the timeline exists
            self.handle_basebackup_request(
                pgb,
@@ -1771,6 +1792,10 @@ where

            self.check_permission(Some(tenant_id))?;

+            COMPUTE_COMMANDS_COUNTERS
+                .for_command(ComputeCommandKind::ImportBasebackup)
+                .inc();
+
            match self
                .handle_import_basebackup(
                    pgb,
@@ -1818,6 +1843,10 @@ where

            self.check_permission(Some(tenant_id))?;

+            COMPUTE_COMMANDS_COUNTERS
+                .for_command(ComputeCommandKind::ImportWal)
+                .inc();
+
            match self
                .handle_import_wal(pgb, tenant_id, timeline_id, start_lsn, end_lsn, ctx)
                .await
@@ -1855,6 +1884,10 @@ where

            self.check_permission(Some(tenant_shard_id.tenant_id))?;

+            COMPUTE_COMMANDS_COUNTERS
+                .for_command(ComputeCommandKind::LeaseLsn)
+                .inc();
+
            // The caller is responsible for providing correct lsn.
            let lsn = Lsn::from_str(params[2])
                .with_context(|| format!("Failed to parse Lsn from {}", params[2]))?;
@@ -1886,6 +1919,10 @@ where

            self.check_permission(Some(tenant_id))?;

+            COMPUTE_COMMANDS_COUNTERS
+                .for_command(ComputeCommandKind::Show)
+                .inc();
+
            let tenant = self
                .get_active_tenant_with_timeout(
                    tenant_id,
@@ -88,6 +88,7 @@ use crate::tenant::remote_timeline_client::MaybeDeletedIndexPart;
 use crate::tenant::remote_timeline_client::INITDB_PATH;
 use crate::tenant::storage_layer::DeltaLayer;
 use crate::tenant::storage_layer::ImageLayer;
+use crate::walredo;
 use crate::InitializationOrder;
 use std::collections::hash_map::Entry;
 use std::collections::BTreeSet;
@@ -212,8 +213,6 @@ pub(crate) enum SpawnMode {
    Eager,
    /// Lazy activation in the background, with the option to skip the queue if the need comes up
    Lazy,
-    /// Tenant has been created during the lifetime of this process
-    Create,
 }

 ///
@@ -323,6 +322,16 @@ impl From<harness::TestRedoManager> for WalRedoManager {
 }

 impl WalRedoManager {
+    pub(crate) async fn shutdown(&self) {
+        match self {
+            Self::Prod(mgr) => mgr.shutdown().await,
+            #[cfg(test)]
+            Self::Test(_) => {
+                // Not applicable to test redo manager
+            }
+        }
+    }
+
    pub(crate) fn maybe_quiesce(&self, idle_timeout: Duration) {
        match self {
            Self::Prod(mgr) => mgr.maybe_quiesce(idle_timeout),
@@ -343,7 +352,7 @@ impl WalRedoManager {
        base_img: Option<(Lsn, bytes::Bytes)>,
        records: Vec<(Lsn, crate::walrecord::NeonWalRecord)>,
        pg_version: u32,
-    ) -> anyhow::Result<bytes::Bytes> {
+    ) -> Result<bytes::Bytes, walredo::Error> {
        match self {
            Self::Prod(mgr) => {
                mgr.request_redo(key, lsn, base_img, records, pg_version)
@@ -648,7 +657,7 @@ impl Tenant {
        init_order: Option<InitializationOrder>,
        mode: SpawnMode,
        ctx: &RequestContext,
-    ) -> anyhow::Result<Arc<Tenant>> {
+    ) -> Arc<Tenant> {
        let wal_redo_manager = Arc::new(WalRedoManager::from(PostgresRedoManager::new(
            conf,
            tenant_shard_id,
@@ -797,9 +806,6 @@ impl Tenant {
                };

                let preload = match &mode {
-                    SpawnMode::Create => {
-                        None
-                    },
                    SpawnMode::Eager | SpawnMode::Lazy => {
                        let _preload_timer = TENANT.preload.start_timer();
                        let res = tenant_clone
@@ -821,11 +827,8 @@ impl Tenant {

                // We will time the duration of the attach phase unless this is a creation (attach will do no work)
                let attached = {
-                    let _attach_timer = match mode {
-                        SpawnMode::Create => None,
-                        SpawnMode::Eager | SpawnMode::Lazy => Some(TENANT.attach.start_timer()),
-                    };
-                    tenant_clone.attach(preload, mode, &ctx).await
+                    let _attach_timer = Some(TENANT.attach.start_timer());
+                    tenant_clone.attach(preload, &ctx).await
                };

                match attached {
@@ -856,7 +859,7 @@ impl Tenant {
            }
            .instrument(tracing::info_span!(parent: None, "attach", tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(), gen=?generation)),
        );
-        Ok(tenant)
+        tenant
    }

    #[instrument(skip_all)]
@@ -901,21 +904,14 @@ impl Tenant {
    async fn attach(
        self: &Arc<Tenant>,
        preload: Option<TenantPreload>,
-        mode: SpawnMode,
        ctx: &RequestContext,
    ) -> anyhow::Result<()> {
        span::debug_assert_current_span_has_tenant_id();

        failpoint_support::sleep_millis_async!("before-attaching-tenant");

-        let preload = match (preload, mode) {
-            (Some(p), _) => p,
-            (None, SpawnMode::Create) => TenantPreload {
-                timelines: HashMap::new(),
-            },
-            (None, _) => {
-                anyhow::bail!("local-only deployment is no longer supported, https://github.com/neondatabase/neon/issues/5624");
-            }
+        let Some(preload) = preload else {
+            anyhow::bail!("local-only deployment is no longer supported, https://github.com/neondatabase/neon/issues/5624");
        };

        let mut timelines_to_resume_deletions = vec![];
@@ -1147,30 +1143,6 @@ impl Tenant {
        .await
    }

-    /// Create a placeholder Tenant object for a broken tenant
-    pub fn create_broken_tenant(
-        conf: &'static PageServerConf,
-        tenant_shard_id: TenantShardId,
-        remote_storage: GenericRemoteStorage,
-        reason: String,
-    ) -> Arc<Tenant> {
-        Arc::new(Tenant::new(
-            TenantState::Broken {
-                reason,
-                backtrace: String::new(),
-            },
-            conf,
-            AttachedTenantConf::try_from(LocationConf::default()).unwrap(),
-            // Shard identity isn't meaningful for a broken tenant: it's just a placeholder
-            // to occupy the slot for this TenantShardId.
-            ShardIdentity::broken(tenant_shard_id.shard_number, tenant_shard_id.shard_count),
-            None,
-            tenant_shard_id,
-            remote_storage,
-            DeletionQueueClient::broken(),
-        ))
-    }
-
    async fn load_timeline_metadata(
        self: &Arc<Tenant>,
        timeline_ids: HashSet<TimelineId>,
@@ -1877,6 +1849,10 @@ impl Tenant {
        tracing::debug!("Waiting for tasks...");
        task_mgr::shutdown_tasks(None, Some(self.tenant_shard_id), None).await;

+        if let Some(walredo_mgr) = self.walredo_mgr.as_ref() {
+            walredo_mgr.shutdown().await;
+        }
+
        // Wait for any in-flight operations to complete
        self.gate.close().await;

@@ -2494,6 +2470,10 @@ impl Tenant {
        remote_storage: GenericRemoteStorage,
        deletion_queue_client: DeletionQueueClient,
    ) -> Tenant {
+        debug_assert!(
+            !attached_conf.location.generation.is_none() || conf.control_plane_api.is_none()
+        );
+
        let (state, mut rx) = watch::channel(state);

        tokio::spawn(async move {
@@ -2584,45 +2564,22 @@ impl Tenant {
        conf: &'static PageServerConf,
        tenant_shard_id: &TenantShardId,
    ) -> anyhow::Result<LocationConf> {
-        let legacy_config_path = conf.tenant_config_path(tenant_shard_id);
        let config_path = conf.tenant_location_config_path(tenant_shard_id);

        if config_path.exists() {
            // New-style config takes precedence
            let deserialized = Self::read_config(&config_path)?;
            Ok(toml_edit::de::from_document::<LocationConf>(deserialized)?)
-        } else if legacy_config_path.exists() {
-            // Upgrade path: found an old-style configuration only
-            let deserialized = Self::read_config(&legacy_config_path)?;
-
-            let mut tenant_conf = TenantConfOpt::default();
-            for (key, item) in deserialized.iter() {
-                match key {
-                    "tenant_config" => {
-                        tenant_conf = TenantConfOpt::try_from(item.to_owned()).context(format!("Failed to parse config from file '{legacy_config_path}' as pageserver config"))?;
-                    }
-                    _ => bail!(
-                        "config file {legacy_config_path} has unrecognized pageserver option '{key}'"
-                    ),
-                }
-            }
-
-            // Legacy configs are implicitly in attached state, and do not support sharding
-            Ok(LocationConf::attached_single(
-                tenant_conf,
-                Generation::none(),
-                &models::ShardParameters::default(),
-            ))
        } else {
-            // FIXME If the config file is not found, assume that we're attaching
-            // a detached tenant and config is passed via attach command.
-            // https://github.com/neondatabase/neon/issues/1555
-            // OR: we're loading after incomplete deletion that managed to remove config.
-            info!(
-                "tenant config not found in {} or {}",
-                config_path, legacy_config_path
-            );
-            Ok(LocationConf::default())
+            // The config should almost always exist for a tenant directory:
+            //  - When attaching a tenant, the config is the first thing we write
+            //  - When detaching a tenant, we atomically move the directory to a tmp location
+            //    before deleting contents.
+            //
+            // The very rare edge case that can result in a missing config is if we crash during attach
+            // between creating directory and writing config.  Callers should handle that as if the
+            // directory didn't exist.
+            anyhow::bail!("tenant config not found in {}", config_path);
        }
    }

@@ -2644,47 +2601,17 @@ impl Tenant {
        tenant_shard_id: &TenantShardId,
        location_conf: &LocationConf,
    ) -> anyhow::Result<()> {
-        let legacy_config_path = conf.tenant_config_path(tenant_shard_id);
        let config_path = conf.tenant_location_config_path(tenant_shard_id);

-        Self::persist_tenant_config_at(
-            tenant_shard_id,
-            &config_path,
-            &legacy_config_path,
-            location_conf,
-        )
-        .await
+        Self::persist_tenant_config_at(tenant_shard_id, &config_path, location_conf).await
    }

    #[tracing::instrument(skip_all, fields(tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug()))]
    pub(super) async fn persist_tenant_config_at(
        tenant_shard_id: &TenantShardId,
        config_path: &Utf8Path,
-        legacy_config_path: &Utf8Path,
        location_conf: &LocationConf,
    ) -> anyhow::Result<()> {
-        if let LocationMode::Attached(attach_conf) = &location_conf.mode {
-            // The modern-style LocationConf config file requires a generation to be set. In case someone
-            // is running a pageserver without the infrastructure to set generations, write out the legacy-style
-            // config file that only contains TenantConf.
-            //
-            // This will eventually be removed in https://github.com/neondatabase/neon/issues/5388
-
-            if attach_conf.generation.is_none() {
-                tracing::info!(
-                    "Running without generations, writing legacy-style tenant config file"
-                );
-                Self::persist_tenant_config_legacy(
-                    tenant_shard_id,
-                    legacy_config_path,
-                    &location_conf.tenant_conf,
-                )
-                .await?;
-
-                return Ok(());
-            }
-        }
-
        debug!("persisting tenantconf to {config_path}");

        let mut conf_content = r#"# This file contains a specific per-tenant's config.
@@ -2711,37 +2638,6 @@ impl Tenant {
        Ok(())
    }

-    #[tracing::instrument(skip_all, fields(tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug()))]
-    async fn persist_tenant_config_legacy(
-        tenant_shard_id: &TenantShardId,
-        target_config_path: &Utf8Path,
-        tenant_conf: &TenantConfOpt,
-    ) -> anyhow::Result<()> {
-        debug!("persisting tenantconf to {target_config_path}");
-
-        let mut conf_content = r#"# This file contains a specific per-tenant's config.
-#  It is read in case of pageserver restart.
-
-[tenant_config]
-"#
-        .to_string();
-
-        // Convert the config to a toml file.
-        conf_content += &toml_edit::ser::to_string(&tenant_conf)?;
-
-        let temp_path = path_with_suffix_extension(target_config_path, TEMP_FILE_SUFFIX);
-
-        let tenant_shard_id = *tenant_shard_id;
-        let target_config_path = target_config_path.to_owned();
-        let conf_content = conf_content.into_bytes();
-        VirtualFile::crashsafe_overwrite(target_config_path.clone(), temp_path, conf_content)
-            .await
-            .with_context(|| {
-                format!("write tenant {tenant_shard_id} config to {target_config_path}")
-            })?;
-        Ok(())
-    }
-
    //
    // How garbage collection works:
    //
@@ -2960,6 +2856,11 @@ impl Tenant {
                let now = SystemTime::now();
                target.leases.retain(|_, lease| !lease.is_expired(&now));

+                timeline
+                    .metrics
+                    .valid_lsn_lease_count_gauge
+                    .set(target.leases.len() as u64);
+
                match gc_cutoffs.remove(&timeline.timeline_id) {
                    Some(cutoffs) => {
                        target.retain_lsns = branchpoints;
@@ -3925,7 +3826,7 @@ pub(crate) mod harness {
            let preload = tenant
                .preload(&self.remote_storage, CancellationToken::new())
                .await?;
-            tenant.attach(Some(preload), SpawnMode::Eager, ctx).await?;
+            tenant.attach(Some(preload), ctx).await?;

            tenant.state.send_replace(TenantState::Active);
            for timeline in tenant.timelines.lock().unwrap().values() {
@@ -3953,7 +3854,7 @@ pub(crate) mod harness {
            base_img: Option<(Lsn, Bytes)>,
            records: Vec<(Lsn, NeonWalRecord)>,
            _pg_version: u32,
-        ) -> anyhow::Result<Bytes> {
+        ) -> Result<Bytes, walredo::Error> {
            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.
@@ -4007,6 +3908,7 @@ mod tests {
    use storage_layer::PersistentLayerKey;
    use tests::storage_layer::ValuesReconstructState;
    use tests::timeline::{GetVectoredError, ShutdownMode};
+    use timeline::GcInfo;
    use utils::bin_ser::BeSer;
    use utils::id::TenantId;

@@ -6362,7 +6264,7 @@ mod tests {

    #[tokio::test]
    async fn test_vectored_missing_metadata_key_reads() -> anyhow::Result<()> {
-        let harness = TenantHarness::create("test_vectored_missing_data_key_reads")?;
+        let harness = TenantHarness::create("test_vectored_missing_metadata_key_reads")?;
        let (tenant, ctx) = harness.load().await;

        let base_key = Key::from_hex("620000000033333333444444445500000000").unwrap();
@@ -6684,49 +6586,48 @@ mod tests {

        // img layer at 0x10
        let img_layer = (0..10)
-            .map(|id| (get_key(id), test_img(&format!("value {id}@0x10"))))
+            .map(|id| (get_key(id), Bytes::from(format!("value {id}@0x10"))))
            .collect_vec();

        let delta1 = vec![
-            // TODO: we should test a real delta record here, which requires us to add a variant of NeonWalRecord for testing purpose.
            (
                get_key(1),
                Lsn(0x20),
-                Value::Image(test_img("value 1@0x20")),
+                Value::Image(Bytes::from("value 1@0x20")),
            ),
            (
                get_key(2),
                Lsn(0x30),
-                Value::Image(test_img("value 2@0x30")),
+                Value::Image(Bytes::from("value 2@0x30")),
            ),
            (
                get_key(3),
                Lsn(0x40),
-                Value::Image(test_img("value 3@0x40")),
+                Value::Image(Bytes::from("value 3@0x40")),
            ),
        ];
        let delta2 = vec![
            (
                get_key(5),
                Lsn(0x20),
-                Value::Image(test_img("value 5@0x20")),
+                Value::Image(Bytes::from("value 5@0x20")),
            ),
            (
                get_key(6),
                Lsn(0x20),
-                Value::Image(test_img("value 6@0x20")),
+                Value::Image(Bytes::from("value 6@0x20")),
            ),
        ];
        let delta3 = vec![
            (
                get_key(8),
                Lsn(0x40),
-                Value::Image(test_img("value 8@0x40")),
+                Value::Image(Bytes::from("value 8@0x40")),
            ),
            (
                get_key(9),
                Lsn(0x40),
-                Value::Image(test_img("value 9@0x40")),
+                Value::Image(Bytes::from("value 9@0x40")),
            ),
        ];

@@ -6748,9 +6649,42 @@ mod tests {
            guard.cutoffs.horizon = Lsn(0x30);
        }

+        let expected_result = [
+            Bytes::from_static(b"value 0@0x10"),
+            Bytes::from_static(b"value 1@0x20"),
+            Bytes::from_static(b"value 2@0x30"),
+            Bytes::from_static(b"value 3@0x40"),
+            Bytes::from_static(b"value 4@0x10"),
+            Bytes::from_static(b"value 5@0x20"),
+            Bytes::from_static(b"value 6@0x20"),
+            Bytes::from_static(b"value 7@0x10"),
+            Bytes::from_static(b"value 8@0x40"),
+            Bytes::from_static(b"value 9@0x40"),
+        ];
+
+        for (idx, expected) in expected_result.iter().enumerate() {
+            assert_eq!(
+                tline
+                    .get(get_key(idx as u32), Lsn(0x50), &ctx)
+                    .await
+                    .unwrap(),
+                expected
+            );
+        }
+
        let cancel = CancellationToken::new();
        tline.compact_with_gc(&cancel, &ctx).await.unwrap();

+        for (idx, expected) in expected_result.iter().enumerate() {
+            assert_eq!(
+                tline
+                    .get(get_key(idx as u32), Lsn(0x50), &ctx)
+                    .await
+                    .unwrap(),
+                expected
+            );
+        }
+
        // Check if the image layer at the GC horizon contains exactly what we want
        let image_at_gc_horizon = tline
            .inspect_image_layers(Lsn(0x30), &ctx)
@@ -6761,14 +6695,22 @@ mod tests {
            .collect::<Vec<_>>();

        assert_eq!(image_at_gc_horizon.len(), 10);
-        let expected_lsn = [0x10, 0x20, 0x30, 0x10, 0x10, 0x20, 0x20, 0x10, 0x10, 0x10];
+        let expected_result = [
+            Bytes::from_static(b"value 0@0x10"),
+            Bytes::from_static(b"value 1@0x20"),
+            Bytes::from_static(b"value 2@0x30"),
+            Bytes::from_static(b"value 3@0x10"),
+            Bytes::from_static(b"value 4@0x10"),
+            Bytes::from_static(b"value 5@0x20"),
+            Bytes::from_static(b"value 6@0x20"),
+            Bytes::from_static(b"value 7@0x10"),
+            Bytes::from_static(b"value 8@0x10"),
+            Bytes::from_static(b"value 9@0x10"),
+        ];
        for idx in 0..10 {
            assert_eq!(
                image_at_gc_horizon[idx],
-                (
-                    get_key(idx as u32),
-                    test_img(&format!("value {idx}@{:#x}", expected_lsn[idx]))
-                )
+                (get_key(idx as u32), expected_result[idx].clone())
            );
        }

@@ -6801,7 +6743,7 @@ mod tests {
                },
                // The delta layer that is cut in the middle
                PersistentLayerKey {
-                    key_range: Key::MIN..get_key(9),
+                    key_range: get_key(3)..get_key(4),
                    lsn_range: Lsn(0x30)..Lsn(0x41),
                    is_delta: true
                },
@@ -6886,6 +6828,9 @@ mod tests {
            tline.get(get_key(2), Lsn(0x50), &ctx).await?,
            Bytes::from_static(b"0x10,0x20,0x30")
        );
+
+        // Need to remove the limit of "Neon WAL redo requires base image".
+
        // assert_eq!(tline.get(get_key(3), Lsn(0x50), &ctx).await?, Bytes::new());
        // assert_eq!(tline.get(get_key(4), Lsn(0x50), &ctx).await?, Bytes::new());

@@ -6980,4 +6925,174 @@ mod tests {

        Ok(())
    }
+
+    #[tokio::test]
+    async fn test_simple_bottom_most_compaction_deltas() -> anyhow::Result<()> {
+        let harness = TenantHarness::create("test_simple_bottom_most_compaction_deltas")?;
+        let (tenant, ctx) = harness.load().await;
+
+        fn get_key(id: u32) -> Key {
+            // using aux key here b/c they are guaranteed to be inside `collect_keyspace`.
+            let mut key = Key::from_hex("620000000033333333444444445500000000").unwrap();
+            key.field6 = id;
+            key
+        }
+
+        // We create one bottom-most image layer, a delta layer D1 crossing the GC horizon, D2 below the horizon, and D3 above the horizon.
+        //
+        //  | D1 |                       | D3 |
+        // -|    |-- gc horizon -----------------
+        //  |    |                | D2 |
+        // --------- img layer ------------------
+        //
+        // What we should expact from this compaction is:
+        //  | Part of D1 |               | D3 |
+        // --------- img layer with D1+D2 at GC horizon------------------
+
+        // img layer at 0x10
+        let img_layer = (0..10)
+            .map(|id| (get_key(id), Bytes::from(format!("value {id}@0x10"))))
+            .collect_vec();
+
+        let delta1 = vec![
+            (
+                get_key(1),
+                Lsn(0x20),
+                Value::WalRecord(NeonWalRecord::wal_append("@0x20")),
+            ),
+            (
+                get_key(2),
+                Lsn(0x30),
+                Value::WalRecord(NeonWalRecord::wal_append("@0x30")),
+            ),
+            (
+                get_key(3),
+                Lsn(0x28),
+                Value::WalRecord(NeonWalRecord::wal_append("@0x28")),
+            ),
+            (
+                get_key(3),
+                Lsn(0x30),
+                Value::WalRecord(NeonWalRecord::wal_append("@0x30")),
+            ),
+            (
+                get_key(3),
+                Lsn(0x40),
+                Value::WalRecord(NeonWalRecord::wal_append("@0x40")),
+            ),
+        ];
+        let delta2 = vec![
+            (
+                get_key(5),
+                Lsn(0x20),
+                Value::WalRecord(NeonWalRecord::wal_append("@0x20")),
+            ),
+            (
+                get_key(6),
+                Lsn(0x20),
+                Value::WalRecord(NeonWalRecord::wal_append("@0x20")),
+            ),
+        ];
+        let delta3 = vec![
+            (
+                get_key(8),
+                Lsn(0x40),
+                Value::WalRecord(NeonWalRecord::wal_append("@0x40")),
+            ),
+            (
+                get_key(9),
+                Lsn(0x40),
+                Value::WalRecord(NeonWalRecord::wal_append("@0x40")),
+            ),
+        ];
+
+        let tline = tenant
+            .create_test_timeline_with_layers(
+                TIMELINE_ID,
+                Lsn(0x10),
+                DEFAULT_PG_VERSION,
+                &ctx,
+                vec![delta1, delta2, delta3], // delta layers
+                vec![(Lsn(0x10), img_layer)], // image layers
+                Lsn(0x50),
+            )
+            .await?;
+        {
+            // Update GC info
+            let mut guard = tline.gc_info.write().unwrap();
+            *guard = GcInfo {
+                retain_lsns: vec![],
+                cutoffs: GcCutoffs {
+                    pitr: Lsn(0x30),
+                    horizon: Lsn(0x30),
+                },
+                leases: Default::default(),
+            };
+        }
+
+        let expected_result = [
+            Bytes::from_static(b"value 0@0x10"),
+            Bytes::from_static(b"value 1@0x10@0x20"),
+            Bytes::from_static(b"value 2@0x10@0x30"),
+            Bytes::from_static(b"value 3@0x10@0x28@0x30@0x40"),
+            Bytes::from_static(b"value 4@0x10"),
+            Bytes::from_static(b"value 5@0x10@0x20"),
+            Bytes::from_static(b"value 6@0x10@0x20"),
+            Bytes::from_static(b"value 7@0x10"),
+            Bytes::from_static(b"value 8@0x10@0x40"),
+            Bytes::from_static(b"value 9@0x10@0x40"),
+        ];
+
+        let expected_result_at_gc_horizon = [
+            Bytes::from_static(b"value 0@0x10"),
+            Bytes::from_static(b"value 1@0x10@0x20"),
+            Bytes::from_static(b"value 2@0x10@0x30"),
+            Bytes::from_static(b"value 3@0x10@0x28@0x30"),
+            Bytes::from_static(b"value 4@0x10"),
+            Bytes::from_static(b"value 5@0x10@0x20"),
+            Bytes::from_static(b"value 6@0x10@0x20"),
+            Bytes::from_static(b"value 7@0x10"),
+            Bytes::from_static(b"value 8@0x10"),
+            Bytes::from_static(b"value 9@0x10"),
+        ];
+
+        for idx in 0..10 {
+            assert_eq!(
+                tline
+                    .get(get_key(idx as u32), Lsn(0x50), &ctx)
+                    .await
+                    .unwrap(),
+                &expected_result[idx]
+            );
+            assert_eq!(
+                tline
+                    .get(get_key(idx as u32), Lsn(0x30), &ctx)
+                    .await
+                    .unwrap(),
+                &expected_result_at_gc_horizon[idx]
+            );
+        }
+
+        let cancel = CancellationToken::new();
+        tline.compact_with_gc(&cancel, &ctx).await.unwrap();
+
+        for idx in 0..10 {
+            assert_eq!(
+                tline
+                    .get(get_key(idx as u32), Lsn(0x50), &ctx)
+                    .await
+                    .unwrap(),
+                &expected_result[idx]
+            );
+            assert_eq!(
+                tline
+                    .get(get_key(idx as u32), Lsn(0x30), &ctx)
+                    .await
+                    .unwrap(),
+                &expected_result_at_gc_horizon[idx]
+            );
+        }
+
+        Ok(())
+    }
 }
@@ -160,6 +160,7 @@ impl<'a> BlockCursor<'a> {
 ///
 /// The file is assumed to be immutable. This doesn't provide any functions
 /// for modifying the file, nor for invalidating the cache if it is modified.
+#[derive(Clone)]
 pub struct FileBlockReader<'a> {
    pub file: &'a VirtualFile,

@@ -281,22 +281,6 @@ impl LocationConf {
    }
 }

-impl Default for LocationConf {
-    // TODO: this should be removed once tenant loading can guarantee that we are never
-    // loading from a directory without a configuration.
-    // => tech debt since https://github.com/neondatabase/neon/issues/1555
-    fn default() -> Self {
-        Self {
-            mode: LocationMode::Attached(AttachedLocationConfig {
-                generation: Generation::none(),
-                attach_mode: AttachmentMode::Single,
-            }),
-            tenant_conf: TenantConfOpt::default(),
-            shard: ShardIdentity::unsharded(),
-        }
-    }
-}
-
 /// A tenant's calcuated configuration, which is the result of merging a
 /// tenant's TenantConfOpt with the global TenantConf from PageServerConf.
 ///
@@ -212,6 +212,7 @@ impl<'a, const L: usize> OnDiskNode<'a, L> {
 ///
 /// Public reader object, to search the tree.
 ///
+#[derive(Clone)]
 pub struct DiskBtreeReader<R, const L: usize>
 where
    R: BlockReader,
@@ -259,27 +260,38 @@ where
        Ok(result)
    }

-    pub fn iter<'a>(
-        &'a self,
-        start_key: &'a [u8; L],
-        ctx: &'a RequestContext,
-    ) -> DiskBtreeIterator<'a> {
+    pub fn iter<'a>(self, start_key: &'a [u8; L], ctx: &'a RequestContext) -> DiskBtreeIterator<'a>
+    where
+        R: 'a,
+    {
        DiskBtreeIterator {
-            stream: Box::pin(self.get_stream_from(start_key, ctx)),
+            stream: Box::pin(self.into_stream(start_key, ctx)),
        }
    }

    /// Return a stream which yields all key, value pairs from the index
    /// starting from the first key greater or equal to `start_key`.
    ///
-    /// Note that this is a copy of [`Self::visit`].
+    /// Note 1: that this is a copy of [`Self::visit`].
    /// TODO: Once the sequential read path is removed this will become
    /// the only index traversal method.
-    pub fn get_stream_from<'a>(
-        &'a self,
+    ///
+    /// Note 2: this function used to take `&self` but it now consumes `self`. This is due to
+    /// the lifetime constraints of the reader and the stream / iterator it creates. Using `&self`
+    /// requires the reader to be present when the stream is used, and this creates a lifetime
+    /// dependency between the reader and the stream. Now if we want to create an iterator that
+    /// holds the stream, someone will need to keep a reference to the reader, which is inconvenient
+    /// to use from the image/delta layer APIs.
+    ///
+    /// Feel free to add the `&self` variant back if it's necessary.
+    pub fn into_stream<'a>(
+        self,
        start_key: &'a [u8; L],
        ctx: &'a RequestContext,
-    ) -> impl Stream<Item = std::result::Result<(Vec<u8>, u64), DiskBtreeError>> + 'a {
+    ) -> impl Stream<Item = std::result::Result<(Vec<u8>, u64), DiskBtreeError>> + 'a
+    where
+        R: 'a,
+    {
        try_stream! {
            let mut stack = Vec::new();
            stack.push((self.root_blk, None));
@@ -495,17 +495,8 @@ pub async fn init_tenant_mgr(
        let mut location_conf = match location_conf {
            Ok(l) => l,
            Err(e) => {
-                warn!(tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(), "Marking tenant broken, failed to {e:#}");
-
-                tenants.insert(
-                    tenant_shard_id,
-                    TenantSlot::Attached(Tenant::create_broken_tenant(
-                        conf,
-                        tenant_shard_id,
-                        resources.remote_storage.clone(),
-                        format!("{}", e),
-                    )),
-                );
+                // This should only happen in the case of a serialization bug or critical local I/O error: we cannot load this tenant
+                error!(tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(), "Failed to load tenant config, failed to {e:#}");
                continue;
            }
        };
@@ -687,8 +678,7 @@ fn tenant_spawn(
        "Cannot load tenant from empty directory {tenant_path:?}"
    );

-    let remote_storage = resources.remote_storage.clone();
-    let tenant = match Tenant::spawn(
+    let tenant = Tenant::spawn(
        conf,
        tenant_shard_id,
        resources,
@@ -697,13 +687,7 @@ fn tenant_spawn(
        init_order,
        mode,
        ctx,
-    ) {
-        Ok(tenant) => tenant,
-        Err(e) => {
-            error!("Failed to spawn tenant {tenant_shard_id}, reason: {e:#}");
-            Tenant::create_broken_tenant(conf, tenant_shard_id, remote_storage, format!("{e:#}"))
-        }
-    };
+    );

    Ok(tenant)
 }
@@ -262,6 +262,7 @@ impl scheduler::RunningJob for RunningDownload {
 struct CompleteDownload {
    secondary_state: Arc<SecondaryTenant>,
    completed_at: Instant,
+    result: Result<(), UpdateError>,
 }

 impl scheduler::Completion for CompleteDownload {
@@ -286,21 +287,33 @@ impl JobGenerator<PendingDownload, RunningDownload, CompleteDownload, DownloadCo
        let CompleteDownload {
            secondary_state,
            completed_at: _completed_at,
+            result,
        } = completion;

        tracing::debug!("Secondary tenant download completed");

        let mut detail = secondary_state.detail.lock().unwrap();

-        let period = detail
-            .last_download
-            .as_ref()
-            .map(|d| d.upload_period)
-            .unwrap_or(DEFAULT_DOWNLOAD_INTERVAL);
+        match result {
+            Err(UpdateError::Restart) => {
+                // Start downloading again as soon as we can.  This will involve waiting for the scheduler's
+                // scheduling interval.  This slightly reduces the peak download speed of tenants that hit their
+                // deadline and keep restarting, but that also helps give other tenants a chance to execute rather
+                // that letting one big tenant dominate for a long time.
+                detail.next_download = Some(Instant::now());
+            }
+            _ => {
+                let period = detail
+                    .last_download
+                    .as_ref()
+                    .map(|d| d.upload_period)
+                    .unwrap_or(DEFAULT_DOWNLOAD_INTERVAL);

-        // We advance next_download irrespective of errors: we don't want error cases to result in
-        // expensive busy-polling.
-        detail.next_download = Some(Instant::now() + period_jitter(period, 5));
+                // We advance next_download irrespective of errors: we don't want error cases to result in
+                // expensive busy-polling.
+                detail.next_download = Some(Instant::now() + period_jitter(period, 5));
+            }
+        }
    }

    async fn schedule(&mut self) -> SchedulingResult<PendingDownload> {
@@ -396,9 +409,10 @@ impl JobGenerator<PendingDownload, RunningDownload, CompleteDownload, DownloadCo
        (RunningDownload { barrier }, Box::pin(async move {
            let _completion = completion;

-            match TenantDownloader::new(conf, &remote_storage, &secondary_state)
+            let result = TenantDownloader::new(conf, &remote_storage, &secondary_state)
                .download(&download_ctx)
-                .await
+                .await;
+            match &result
            {
                Err(UpdateError::NoData) => {
                    tracing::info!("No heatmap found for tenant.  This is fine if it is new.");
@@ -415,6 +429,9 @@ impl JobGenerator<PendingDownload, RunningDownload, CompleteDownload, DownloadCo
                Err(e @ (UpdateError::DownloadError(_) | UpdateError::Other(_))) => {
                    tracing::error!("Error while downloading tenant: {e}");
                },
+                Err(UpdateError::Restart) => {
+                    tracing::info!("Download reached deadline & will restart to update heatmap")
+                }
                Ok(()) => {}
            };

@@ -436,6 +453,7 @@ impl JobGenerator<PendingDownload, RunningDownload, CompleteDownload, DownloadCo
            CompleteDownload {
                secondary_state,
                completed_at: Instant::now(),
+                result
            }
        }.instrument(info_span!(parent: None, "secondary_download", tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug()))))
    }
@@ -452,6 +470,11 @@ struct TenantDownloader<'a> {
 /// Errors that may be encountered while updating a tenant
 #[derive(thiserror::Error, Debug)]
 enum UpdateError {
+    /// This is not a true failure, but it's how a download indicates that it would like to be restarted by
+    /// the scheduler, to pick up the latest heatmap
+    #[error("Reached deadline, restarting downloads")]
+    Restart,
+
    #[error("No remote data found")]
    NoData,
    #[error("Insufficient local storage space")]
@@ -603,6 +626,26 @@ impl<'a> TenantDownloader<'a> {
                self.prepare_timelines(&heatmap, heatmap_mtime).await?;
        }

+        // Calculate a deadline for downloads: if downloading takes longer than this, it is useful to drop out and start again,
+        // so that we are always using reasonably a fresh heatmap.  Otherwise, if we had really huge content to download, we might
+        // spend 10s of minutes downloading layers we don't need.
+        // (see https://github.com/neondatabase/neon/issues/8182)
+        let deadline = {
+            let period = self
+                .secondary_state
+                .detail
+                .lock()
+                .unwrap()
+                .last_download
+                .as_ref()
+                .map(|d| d.upload_period)
+                .unwrap_or(DEFAULT_DOWNLOAD_INTERVAL);
+
+            // Use double the period: we are not promising to complete within the period, this is just a heuristic
+            // to keep using a "reasonably fresh" heatmap.
+            Instant::now() + period * 2
+        };
+
        // Download the layers in the heatmap
        for timeline in heatmap.timelines {
            let timeline_state = timeline_states
@@ -618,7 +661,7 @@ impl<'a> TenantDownloader<'a> {
            }

            let timeline_id = timeline.timeline_id;
-            self.download_timeline(timeline, timeline_state, ctx)
+            self.download_timeline(timeline, timeline_state, deadline, ctx)
                .instrument(tracing::info_span!(
                    "secondary_download_timeline",
                    tenant_id=%tenant_shard_id.tenant_id,
@@ -827,26 +870,28 @@ impl<'a> TenantDownloader<'a> {
        .and_then(|x| x)
    }

-    async fn download_timeline(
+    /// Download heatmap layers that are not present on local disk, or update their
+    /// access time if they are already present.
+    async fn download_timeline_layers(
        &self,
+        tenant_shard_id: &TenantShardId,
        timeline: HeatMapTimeline,
        timeline_state: SecondaryDetailTimeline,
+        deadline: Instant,
        ctx: &RequestContext,
-    ) -> Result<(), UpdateError> {
-        debug_assert_current_span_has_tenant_and_timeline_id();
-        let tenant_shard_id = self.secondary_state.get_tenant_shard_id();
-
+    ) -> (Result<(), UpdateError>, Vec<HeatMapLayer>) {
        // Accumulate updates to the state
        let mut touched = Vec::new();

-        tracing::debug!(timeline_id=%timeline.timeline_id, "Downloading layers, {} in heatmap", timeline.layers.len());
-
-        // Download heatmap layers that are not present on local disk, or update their
-        // access time if they are already present.
        for layer in timeline.layers {
            if self.secondary_state.cancel.is_cancelled() {
                tracing::debug!("Cancelled -- dropping out of layer loop");
-                return Err(UpdateError::Cancelled);
+                return (Err(UpdateError::Cancelled), touched);
+            }
+
+            if Instant::now() > deadline {
+                // We've been running downloads for a while, restart to download latest heatmap.
+                return (Err(UpdateError::Restart), touched);
            }

            // Existing on-disk layers: just update their access time.
@@ -916,20 +961,43 @@ impl<'a> TenantDownloader<'a> {

            match self
                .download_layer(tenant_shard_id, &timeline.timeline_id, layer, ctx)
-                .await?
+                .await
            {
-                Some(layer) => touched.push(layer),
-                None => {
+                Ok(Some(layer)) => touched.push(layer),
+                Ok(None) => {
                    // Not an error but we didn't download it: remote layer is missing.  Don't add it to the list of
                    // things to consider touched.
                }
+                Err(e) => {
+                    return (Err(e), touched);
+                }
            }
        }

-        // Write updates to state to record layers we just downloaded or touched.
+        (Ok(()), touched)
+    }
+
+    async fn download_timeline(
+        &self,
+        timeline: HeatMapTimeline,
+        timeline_state: SecondaryDetailTimeline,
+        deadline: Instant,
+        ctx: &RequestContext,
+    ) -> Result<(), UpdateError> {
+        debug_assert_current_span_has_tenant_and_timeline_id();
+        let tenant_shard_id = self.secondary_state.get_tenant_shard_id();
+        let timeline_id = timeline.timeline_id;
+
+        tracing::debug!(timeline_id=%timeline_id, "Downloading layers, {} in heatmap", timeline.layers.len());
+
+        let (result, touched) = self
+            .download_timeline_layers(tenant_shard_id, timeline, timeline_state, deadline, ctx)
+            .await;
+
+        // Write updates to state to record layers we just downloaded or touched, irrespective of whether the overall result was successful
        {
            let mut detail = self.secondary_state.detail.lock().unwrap();
-            let timeline_detail = detail.timelines.entry(timeline.timeline_id).or_default();
+            let timeline_detail = detail.timelines.entry(timeline_id).or_default();

            tracing::info!("Wrote timeline_detail for {} touched layers", touched.len());

@@ -943,14 +1011,14 @@ impl<'a> TenantDownloader<'a> {
                        let local_path = local_layer_path(
                            self.conf,
                            tenant_shard_id,
-                            &timeline.timeline_id,
+                            &timeline_id,
                            &t.name,
                            &t.metadata.generation,
                        );
                        e.insert(OnDiskState::new(
                            self.conf,
                            tenant_shard_id,
-                            &timeline.timeline_id,
+                            &timeline_id,
                            t.name,
                            t.metadata.clone(),
                            t.access_time,
@@ -961,7 +1029,7 @@ impl<'a> TenantDownloader<'a> {
            }
        }

-        Ok(())
+        result
    }

    /// Call this during timeline download if a layer will _not_ be downloaded, to update progress statistics
@@ -367,10 +367,9 @@ async fn upload_tenant_heatmap(
    debug_assert_current_span_has_tenant_id();

    let generation = tenant.get_generation();
+    debug_assert!(!generation.is_none());
    if generation.is_none() {
-        // We do not expect this: generations were implemented before heatmap uploads.  However,
-        // handle it so that we don't have to make the generation in the heatmap an Option<>
-        // (Generation::none is not serializable)
+        // We do not expect this: None generations should only appear in historic layer metadata, not in running Tenants
        tracing::warn!("Skipping heatmap upload for tenant with generation==None");
        return Ok(UploadHeatmapOutcome::Skipped);
    }
@@ -928,7 +928,6 @@ impl DeltaLayerInner {
    }

    /// Load all key-values in the delta layer, should be replaced by an iterator-based interface in the future.
-    #[cfg(test)]
    pub(super) async fn load_key_values(
        &self,
        ctx: &RequestContext,
@@ -941,7 +940,7 @@ impl DeltaLayerInner {
        );
        let mut result = Vec::new();
        let mut stream =
-            Box::pin(self.stream_index_forwards(&index_reader, &[0; DELTA_KEY_SIZE], ctx));
+            Box::pin(self.stream_index_forwards(index_reader, &[0; DELTA_KEY_SIZE], ctx));
        let block_reader = FileBlockReader::new(&self.file, self.file_id);
        let cursor = block_reader.block_cursor();
        let mut buf = Vec::new();
@@ -976,7 +975,7 @@ impl DeltaLayerInner {
        ctx: &RequestContext,
    ) -> anyhow::Result<Vec<VectoredRead>>
    where
-        Reader: BlockReader,
+        Reader: BlockReader + Clone,
    {
        let ctx = RequestContextBuilder::extend(ctx)
            .page_content_kind(PageContentKind::DeltaLayerBtreeNode)
@@ -986,7 +985,7 @@ impl DeltaLayerInner {
            let mut range_end_handled = false;

            let start_key = DeltaKey::from_key_lsn(&range.start, lsn_range.start);
-            let index_stream = index_reader.get_stream_from(&start_key.0, &ctx);
+            let index_stream = index_reader.clone().into_stream(&start_key.0, &ctx);
            let mut index_stream = std::pin::pin!(index_stream);

            while let Some(index_entry) = index_stream.next().await {
@@ -1241,7 +1240,7 @@ impl DeltaLayerInner {
            block_reader,
        );

-        let stream = self.stream_index_forwards(&tree_reader, &[0u8; DELTA_KEY_SIZE], ctx);
+        let stream = self.stream_index_forwards(tree_reader, &[0u8; DELTA_KEY_SIZE], ctx);
        let stream = stream.map_ok(|(key, lsn, pos)| Item::Actual(key, lsn, pos));
        // put in a sentinel value for getting the end offset for last item, and not having to
        // repeat the whole read part
@@ -1300,7 +1299,7 @@ impl DeltaLayerInner {
                        offsets.start.pos(),
                        offsets.end.pos(),
                        meta,
-                        max_read_size,
+                        Some(max_read_size),
                    ))
                }
            } else {
@@ -1459,17 +1458,17 @@ impl DeltaLayerInner {

    fn stream_index_forwards<'a, R>(
        &'a self,
-        reader: &'a DiskBtreeReader<R, DELTA_KEY_SIZE>,
+        reader: DiskBtreeReader<R, DELTA_KEY_SIZE>,
        start: &'a [u8; DELTA_KEY_SIZE],
        ctx: &'a RequestContext,
    ) -> impl futures::stream::Stream<
        Item = Result<(Key, Lsn, BlobRef), crate::tenant::disk_btree::DiskBtreeError>,
    > + 'a
    where
-        R: BlockReader,
+        R: BlockReader + 'a,
    {
        use futures::stream::TryStreamExt;
-        let stream = reader.get_stream_from(start, ctx);
+        let stream = reader.into_stream(start, ctx);
        stream.map_ok(|(key, value)| {
            let key = DeltaKey::from_slice(&key);
            let (key, lsn) = (key.key(), key.lsn());
@@ -1493,6 +1492,24 @@ impl DeltaLayerInner {
        );
        offset
    }
+
+    #[cfg(test)]
+    pub(crate) fn iter<'a>(&'a self, ctx: &'a RequestContext) -> DeltaLayerIterator<'a> {
+        let block_reader = FileBlockReader::new(&self.file, self.file_id);
+        let tree_reader =
+            DiskBtreeReader::new(self.index_start_blk, self.index_root_blk, block_reader);
+        DeltaLayerIterator {
+            delta_layer: self,
+            ctx,
+            index_iter: tree_reader.iter(&[0; DELTA_KEY_SIZE], ctx),
+            key_values_batch: std::collections::VecDeque::new(),
+            is_end: false,
+            planner: crate::tenant::vectored_blob_io::StreamingVectoredReadPlanner::new(
+                1024 * 8192, // The default value. Unit tests might use a different value. 1024 * 8K = 8MB buffer.
+                1024,        // The default value. Unit tests might use a different value
+            ),
+        }
+    }
 }

 /// A set of data associated with a delta layer key and its value
@@ -1552,6 +1569,70 @@ impl<'a> pageserver_compaction::interface::CompactionDeltaEntry<'a, Key> for Del
    }
 }

+#[cfg(test)]
+pub struct DeltaLayerIterator<'a> {
+    delta_layer: &'a DeltaLayerInner,
+    ctx: &'a RequestContext,
+    planner: crate::tenant::vectored_blob_io::StreamingVectoredReadPlanner,
+    index_iter: crate::tenant::disk_btree::DiskBtreeIterator<'a>,
+    key_values_batch: std::collections::VecDeque<(Key, Lsn, Value)>,
+    is_end: bool,
+}
+
+#[cfg(test)]
+impl<'a> DeltaLayerIterator<'a> {
+    /// Retrieve a batch of key-value pairs into the iterator buffer.
+    async fn next_batch(&mut self) -> anyhow::Result<()> {
+        assert!(self.key_values_batch.is_empty());
+        assert!(!self.is_end);
+
+        let plan = loop {
+            if let Some(res) = self.index_iter.next().await {
+                let (raw_key, value) = res?;
+                let key = Key::from_slice(&raw_key[..KEY_SIZE]);
+                let lsn = DeltaKey::extract_lsn_from_buf(&raw_key);
+                let blob_ref = BlobRef(value);
+                let offset = blob_ref.pos();
+                if let Some(batch_plan) = self.planner.handle(key, lsn, offset, BlobFlag::None) {
+                    break batch_plan;
+                }
+            } else {
+                self.is_end = true;
+                let data_end_offset = self.delta_layer.index_start_offset();
+                break self.planner.handle_range_end(data_end_offset);
+            }
+        };
+        let vectored_blob_reader = VectoredBlobReader::new(&self.delta_layer.file);
+        let mut next_batch = std::collections::VecDeque::new();
+        let buf_size = plan.size();
+        let buf = BytesMut::with_capacity(buf_size);
+        let blobs_buf = vectored_blob_reader
+            .read_blobs(&plan, buf, self.ctx)
+            .await?;
+        let frozen_buf = blobs_buf.buf.freeze();
+        for meta in blobs_buf.blobs.iter() {
+            let value = Value::des(&frozen_buf[meta.start..meta.end])?;
+            next_batch.push_back((meta.meta.key, meta.meta.lsn, value));
+        }
+        self.key_values_batch = next_batch;
+        Ok(())
+    }
+
+    pub async fn next(&mut self) -> anyhow::Result<Option<(Key, Lsn, Value)>> {
+        if self.key_values_batch.is_empty() {
+            if self.is_end {
+                return Ok(None);
+            }
+            self.next_batch().await?;
+        }
+        Ok(Some(
+            self.key_values_batch
+                .pop_front()
+                .expect("should not be empty"),
+        ))
+    }
+}
+
 #[cfg(test)]
 mod test {
    use std::collections::BTreeMap;
@@ -1561,6 +1642,9 @@ mod test {
    use rand::RngCore;

    use super::*;
+    use crate::tenant::harness::TIMELINE_ID;
+    use crate::tenant::vectored_blob_io::StreamingVectoredReadPlanner;
+    use crate::tenant::Tenant;
    use crate::{
        context::DownloadBehavior,
        task_mgr::TaskKind,
@@ -1857,7 +1941,7 @@ mod test {
            .finish(entries_meta.key_range.end, &timeline, &ctx)
            .await?;

-        let inner = resident.as_delta(&ctx).await?;
+        let inner = resident.get_as_delta(&ctx).await?;

        let file_size = inner.file.metadata().await?.len();
        tracing::info!(
@@ -2044,11 +2128,11 @@ mod test {

            let copied_layer = writer.finish(Key::MAX, &branch, ctx).await.unwrap();

-            copied_layer.as_delta(ctx).await.unwrap();
+            copied_layer.get_as_delta(ctx).await.unwrap();

            assert_keys_and_values_eq(
-                new_layer.as_delta(ctx).await.unwrap(),
-                copied_layer.as_delta(ctx).await.unwrap(),
+                new_layer.get_as_delta(ctx).await.unwrap(),
+                copied_layer.get_as_delta(ctx).await.unwrap(),
                truncate_at,
                ctx,
            )
@@ -2073,7 +2157,7 @@ mod test {
            source.index_root_blk,
            &source_reader,
        );
-        let source_stream = source.stream_index_forwards(&source_tree, &start_key, ctx);
+        let source_stream = source.stream_index_forwards(source_tree, &start_key, ctx);
        let source_stream = source_stream.filter(|res| match res {
            Ok((_, lsn, _)) => ready(lsn < &truncated_at),
            _ => ready(true),
@@ -2086,7 +2170,7 @@ mod test {
            truncated.index_root_blk,
            &truncated_reader,
        );
-        let truncated_stream = truncated.stream_index_forwards(&truncated_tree, &start_key, ctx);
+        let truncated_stream = truncated.stream_index_forwards(truncated_tree, &start_key, ctx);
        let mut truncated_stream = std::pin::pin!(truncated_stream);

        let mut scratch_left = Vec::new();
@@ -2127,4 +2211,116 @@ mod test {
            assert_eq!(utils::Hex(&scratch_left), utils::Hex(&scratch_right));
        }
    }
+
+    async fn produce_delta_layer(
+        tenant: &Tenant,
+        tline: &Arc<Timeline>,
+        mut deltas: Vec<(Key, Lsn, Value)>,
+        ctx: &RequestContext,
+    ) -> anyhow::Result<ResidentLayer> {
+        deltas.sort_by(|(k1, l1, _), (k2, l2, _)| (k1, l1).cmp(&(k2, l2)));
+        let (key_start, _, _) = deltas.first().unwrap();
+        let (key_max, _, _) = deltas.first().unwrap();
+        let lsn_min = deltas.iter().map(|(_, lsn, _)| lsn).min().unwrap();
+        let lsn_max = deltas.iter().map(|(_, lsn, _)| lsn).max().unwrap();
+        let lsn_end = Lsn(lsn_max.0 + 1);
+        let mut writer = DeltaLayerWriter::new(
+            tenant.conf,
+            tline.timeline_id,
+            tenant.tenant_shard_id,
+            *key_start,
+            (*lsn_min)..lsn_end,
+            ctx,
+        )
+        .await?;
+        let key_end = key_max.next();
+
+        for (key, lsn, value) in deltas {
+            writer.put_value(key, lsn, value, ctx).await?;
+        }
+        let delta_layer = writer.finish(key_end, tline, ctx).await?;
+
+        Ok::<_, anyhow::Error>(delta_layer)
+    }
+
+    async fn assert_delta_iter_equal(
+        delta_iter: &mut DeltaLayerIterator<'_>,
+        expect: &[(Key, Lsn, Value)],
+    ) {
+        let mut expect_iter = expect.iter();
+        loop {
+            let o1 = delta_iter.next().await.unwrap();
+            let o2 = expect_iter.next();
+            assert_eq!(o1.is_some(), o2.is_some());
+            if o1.is_none() && o2.is_none() {
+                break;
+            }
+            let (k1, l1, v1) = o1.unwrap();
+            let (k2, l2, v2) = o2.unwrap();
+            assert_eq!(&k1, k2);
+            assert_eq!(l1, *l2);
+            assert_eq!(&v1, v2);
+        }
+    }
+
+    #[tokio::test]
+    async fn delta_layer_iterator() {
+        use crate::repository::Value;
+        use bytes::Bytes;
+
+        let harness = TenantHarness::create("delta_layer_iterator").unwrap();
+        let (tenant, ctx) = harness.load().await;
+
+        let tline = tenant
+            .create_test_timeline(TIMELINE_ID, Lsn(0x10), DEFAULT_PG_VERSION, &ctx)
+            .await
+            .unwrap();
+
+        fn get_key(id: u32) -> Key {
+            let mut key = Key::from_hex("000000000033333333444444445500000000").unwrap();
+            key.field6 = id;
+            key
+        }
+        const N: usize = 1000;
+        let test_deltas = (0..N)
+            .map(|idx| {
+                (
+                    get_key(idx as u32 / 10),
+                    Lsn(0x10 * ((idx as u64) % 10 + 1)),
+                    Value::Image(Bytes::from(format!("img{idx:05}"))),
+                )
+            })
+            .collect_vec();
+        let resident_layer = produce_delta_layer(&tenant, &tline, test_deltas.clone(), &ctx)
+            .await
+            .unwrap();
+        let delta_layer = resident_layer.get_as_delta(&ctx).await.unwrap();
+        for max_read_size in [1, 1024] {
+            for batch_size in [1, 2, 4, 8, 3, 7, 13] {
+                println!("running with batch_size={batch_size} max_read_size={max_read_size}");
+                // Test if the batch size is correctly determined
+                let mut iter = delta_layer.iter(&ctx);
+                iter.planner = StreamingVectoredReadPlanner::new(max_read_size, batch_size);
+                let mut num_items = 0;
+                for _ in 0..3 {
+                    iter.next_batch().await.unwrap();
+                    num_items += iter.key_values_batch.len();
+                    if max_read_size == 1 {
+                        // every key should be a batch b/c the value is larger than max_read_size
+                        assert_eq!(iter.key_values_batch.len(), 1);
+                    } else {
+                        assert_eq!(iter.key_values_batch.len(), batch_size);
+                    }
+                    if num_items >= N {
+                        break;
+                    }
+                    iter.key_values_batch.clear();
+                }
+                // Test if the result is correct
+                let mut iter = delta_layer.iter(&ctx);
+                iter.planner = StreamingVectoredReadPlanner::new(max_read_size, batch_size);
+                assert_delta_iter_equal(&mut iter, &test_deltas).await;
+            }
+        }
+    }
 }
@@ -486,7 +486,6 @@ impl ImageLayerInner {
    }

    /// Load all key-values in the delta layer, should be replaced by an iterator-based interface in the future.
-    #[cfg(test)]
    pub(super) async fn load_key_values(
        &self,
        ctx: &RequestContext,
@@ -495,7 +494,7 @@ impl ImageLayerInner {
        let tree_reader =
            DiskBtreeReader::new(self.index_start_blk, self.index_root_blk, &block_reader);
        let mut result = Vec::new();
-        let mut stream = Box::pin(tree_reader.get_stream_from(&[0; KEY_SIZE], ctx));
+        let mut stream = Box::pin(tree_reader.into_stream(&[0; KEY_SIZE], ctx));
        let block_reader = FileBlockReader::new(&self.file, self.file_id);
        let cursor = block_reader.block_cursor();
        while let Some(item) = stream.next().await {
@@ -544,7 +543,7 @@ impl ImageLayerInner {
            let mut search_key: [u8; KEY_SIZE] = [0u8; KEY_SIZE];
            range.start.write_to_byte_slice(&mut search_key);

-            let index_stream = tree_reader.get_stream_from(&search_key, &ctx);
+            let index_stream = tree_reader.clone().into_stream(&search_key, &ctx);
            let mut index_stream = std::pin::pin!(index_stream);

            while let Some(index_entry) = index_stream.next().await {
@@ -689,6 +688,24 @@ impl ImageLayerInner {
            };
        }
    }
+
+    #[cfg(test)]
+    pub(crate) fn iter<'a>(&'a self, ctx: &'a RequestContext) -> ImageLayerIterator<'a> {
+        let block_reader = FileBlockReader::new(&self.file, self.file_id);
+        let tree_reader =
+            DiskBtreeReader::new(self.index_start_blk, self.index_root_blk, block_reader);
+        ImageLayerIterator {
+            image_layer: self,
+            ctx,
+            index_iter: tree_reader.iter(&[0; KEY_SIZE], ctx),
+            key_values_batch: std::collections::VecDeque::new(),
+            is_end: false,
+            planner: crate::tenant::vectored_blob_io::StreamingVectoredReadPlanner::new(
+                1024 * 8192, // The default value. Unit tests might use a different value. 1024 * 8K = 8MB buffer.
+                1024,        // The default value. Unit tests might use a different value
+            ),
+        }
+    }
 }

 /// A builder object for constructing a new image layer.
@@ -943,11 +960,77 @@ impl Drop for ImageLayerWriter {
    }
 }

+#[cfg(test)]
+pub struct ImageLayerIterator<'a> {
+    image_layer: &'a ImageLayerInner,
+    ctx: &'a RequestContext,
+    planner: crate::tenant::vectored_blob_io::StreamingVectoredReadPlanner,
+    index_iter: crate::tenant::disk_btree::DiskBtreeIterator<'a>,
+    key_values_batch: std::collections::VecDeque<(Key, Lsn, Value)>,
+    is_end: bool,
+}
+
+#[cfg(test)]
+impl<'a> ImageLayerIterator<'a> {
+    /// Retrieve a batch of key-value pairs into the iterator buffer.
+    async fn next_batch(&mut self) -> anyhow::Result<()> {
+        assert!(self.key_values_batch.is_empty());
+        assert!(!self.is_end);
+
+        let plan = loop {
+            if let Some(res) = self.index_iter.next().await {
+                let (raw_key, offset) = res?;
+                if let Some(batch_plan) = self.planner.handle(
+                    Key::from_slice(&raw_key[..KEY_SIZE]),
+                    self.image_layer.lsn,
+                    offset,
+                    BlobFlag::None,
+                ) {
+                    break batch_plan;
+                }
+            } else {
+                self.is_end = true;
+                let payload_end = self.image_layer.index_start_blk as u64 * PAGE_SZ as u64;
+                break self.planner.handle_range_end(payload_end);
+            }
+        };
+        let vectored_blob_reader = VectoredBlobReader::new(&self.image_layer.file);
+        let mut next_batch = std::collections::VecDeque::new();
+        let buf_size = plan.size();
+        let buf = BytesMut::with_capacity(buf_size);
+        let blobs_buf = vectored_blob_reader
+            .read_blobs(&plan, buf, self.ctx)
+            .await?;
+        let frozen_buf: Bytes = blobs_buf.buf.freeze();
+        for meta in blobs_buf.blobs.iter() {
+            let img_buf = frozen_buf.slice(meta.start..meta.end);
+            next_batch.push_back((meta.meta.key, self.image_layer.lsn, Value::Image(img_buf)));
+        }
+        self.key_values_batch = next_batch;
+        Ok(())
+    }
+
+    pub async fn next(&mut self) -> anyhow::Result<Option<(Key, Lsn, Value)>> {
+        if self.key_values_batch.is_empty() {
+            if self.is_end {
+                return Ok(None);
+            }
+            self.next_batch().await?;
+        }
+        Ok(Some(
+            self.key_values_batch
+                .pop_front()
+                .expect("should not be empty"),
+        ))
+    }
+}
+
 #[cfg(test)]
 mod test {
-    use std::time::Duration;
+    use std::{sync::Arc, time::Duration};

    use bytes::Bytes;
+    use itertools::Itertools;
    use pageserver_api::{
        key::Key,
        shard::{ShardCount, ShardIdentity, ShardNumber, ShardStripeSize},
@@ -959,11 +1042,19 @@ mod test {
    };

    use crate::{
-        tenant::{config::TenantConf, harness::TenantHarness},
+        context::RequestContext,
+        repository::Value,
+        tenant::{
+            config::TenantConf,
+            harness::{TenantHarness, TIMELINE_ID},
+            storage_layer::ResidentLayer,
+            vectored_blob_io::StreamingVectoredReadPlanner,
+            Tenant, Timeline,
+        },
        DEFAULT_PG_VERSION,
    };

-    use super::ImageLayerWriter;
+    use super::{ImageLayerIterator, ImageLayerWriter};

    #[tokio::test]
    async fn image_layer_rewrite() {
@@ -1134,4 +1225,111 @@ mod test {
            }
        }
    }
+
+    async fn produce_image_layer(
+        tenant: &Tenant,
+        tline: &Arc<Timeline>,
+        mut images: Vec<(Key, Bytes)>,
+        lsn: Lsn,
+        ctx: &RequestContext,
+    ) -> anyhow::Result<ResidentLayer> {
+        images.sort();
+        let (key_start, _) = images.first().unwrap();
+        let (key_last, _) = images.last().unwrap();
+        let key_end = key_last.next();
+        let key_range = *key_start..key_end;
+        let mut writer = ImageLayerWriter::new(
+            tenant.conf,
+            tline.timeline_id,
+            tenant.tenant_shard_id,
+            &key_range,
+            lsn,
+            ctx,
+        )
+        .await?;
+
+        for (key, img) in images {
+            writer.put_image(key, img, ctx).await?;
+        }
+        let img_layer = writer.finish(tline, ctx).await?;
+
+        Ok::<_, anyhow::Error>(img_layer)
+    }
+
+    async fn assert_img_iter_equal(
+        img_iter: &mut ImageLayerIterator<'_>,
+        expect: &[(Key, Bytes)],
+        expect_lsn: Lsn,
+    ) {
+        let mut expect_iter = expect.iter();
+        loop {
+            let o1 = img_iter.next().await.unwrap();
+            let o2 = expect_iter.next();
+            match (o1, o2) {
+                (None, None) => break,
+                (Some((k1, l1, v1)), Some((k2, i2))) => {
+                    let Value::Image(i1) = v1 else {
+                        panic!("expect Value::Image")
+                    };
+                    assert_eq!(&k1, k2);
+                    assert_eq!(l1, expect_lsn);
+                    assert_eq!(&i1, i2);
+                }
+                (o1, o2) => panic!("iterators length mismatch: {:?}, {:?}", o1, o2),
+            }
+        }
+    }
+
+    #[tokio::test]
+    async fn image_layer_iterator() {
+        let harness = TenantHarness::create("image_layer_iterator").unwrap();
+        let (tenant, ctx) = harness.load().await;
+
+        let tline = tenant
+            .create_test_timeline(TIMELINE_ID, Lsn(0x10), DEFAULT_PG_VERSION, &ctx)
+            .await
+            .unwrap();
+
+        fn get_key(id: u32) -> Key {
+            let mut key = Key::from_hex("000000000033333333444444445500000000").unwrap();
+            key.field6 = id;
+            key
+        }
+        const N: usize = 1000;
+        let test_imgs = (0..N)
+            .map(|idx| (get_key(idx as u32), Bytes::from(format!("img{idx:05}"))))
+            .collect_vec();
+        let resident_layer =
+            produce_image_layer(&tenant, &tline, test_imgs.clone(), Lsn(0x10), &ctx)
+                .await
+                .unwrap();
+        let img_layer = resident_layer.get_as_image(&ctx).await.unwrap();
+        for max_read_size in [1, 1024] {
+            for batch_size in [1, 2, 4, 8, 3, 7, 13] {
+                println!("running with batch_size={batch_size} max_read_size={max_read_size}");
+                // Test if the batch size is correctly determined
+                let mut iter = img_layer.iter(&ctx);
+                iter.planner = StreamingVectoredReadPlanner::new(max_read_size, batch_size);
+                let mut num_items = 0;
+                for _ in 0..3 {
+                    iter.next_batch().await.unwrap();
+                    num_items += iter.key_values_batch.len();
+                    if max_read_size == 1 {
+                        // every key should be a batch b/c the value is larger than max_read_size
+                        assert_eq!(iter.key_values_batch.len(), 1);
+                    } else {
+                        assert_eq!(iter.key_values_batch.len(), batch_size);
+                    }
+                    if num_items >= N {
+                        break;
+                    }
+                    iter.key_values_batch.clear();
+                }
+                // Test if the result is correct
+                let mut iter = img_layer.iter(&ctx);
+                iter.planner = StreamingVectoredReadPlanner::new(max_read_size, batch_size);
+                assert_img_iter_equal(&mut iter, &test_imgs, Lsn(0x10)).await;
+            }
+        }
+    }
 }
@@ -622,18 +622,16 @@ impl InMemoryLayer {

        let end_lsn = *self.end_lsn.get().unwrap();

-        let keys: Vec<_> = if let Some(key_range) = key_range {
+        let key_count = if let Some(key_range) = key_range {
            inner
                .index
                .iter()
                .filter(|(k, _)| key_range.contains(k))
-                .map(|(k, m)| (k.to_i128(), m))
-                .collect()
+                .count()
        } else {
-            inner.index.iter().map(|(k, m)| (k.to_i128(), m)).collect()
+            inner.index.len()
        };
-
-        if keys.is_empty() {
+        if key_count == 0 {
            return Ok(None);
        }

@@ -93,16 +93,12 @@ pub(crate) struct Layer(Arc<LayerInner>);

 impl std::fmt::Display for Layer {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        if matches!(self.0.generation, Generation::Broken) {
-            write!(f, "{}-broken", self.layer_desc().short_id())
-        } else {
-            write!(
-                f,
-                "{}{}",
-                self.layer_desc().short_id(),
-                self.0.generation.get_suffix()
-            )
-        }
+        write!(
+            f,
+            "{}{}",
+            self.layer_desc().short_id(),
+            self.0.generation.get_suffix()
+        )
    }
 }

@@ -389,7 +385,6 @@ impl Layer {
    }

    /// Get all key/values in the layer. Should be replaced with an iterator-based API in the future.
-    #[cfg(test)]
    pub(crate) async fn load_key_values(
        &self,
        ctx: &RequestContext,
@@ -1774,7 +1769,6 @@ impl DownloadedLayer {
        }
    }

-    #[cfg(test)]
    async fn load_key_values(
        &self,
        owner: &Arc<LayerInner>,
@@ -1905,7 +1899,7 @@ impl ResidentLayer {
    }

    #[cfg(test)]
-    pub(crate) async fn as_delta(
+    pub(crate) async fn get_as_delta(
        &self,
        ctx: &RequestContext,
    ) -> anyhow::Result<&delta_layer::DeltaLayerInner> {
@@ -1915,6 +1909,18 @@ impl ResidentLayer {
            Image(_) => Err(anyhow::anyhow!("image layer")),
        }
    }
+
+    #[cfg(test)]
+    pub(crate) async fn get_as_image(
+        &self,
+        ctx: &RequestContext,
+    ) -> anyhow::Result<&image_layer::ImageLayerInner> {
+        use LayerKind::*;
+        match self.downloaded.get(&self.owner.0, ctx).await? {
+            Image(ref d) => Ok(d),
+            Delta(_) => Err(anyhow::anyhow!("delta layer")),
+        }
+    }
 }

 impl AsLayerDesc for ResidentLayer {
@@ -686,6 +686,7 @@ pub enum GetLogicalSizePriority {
 pub(crate) enum CompactFlags {
    ForceRepartition,
    ForceImageLayerCreation,
+    EnhancedGcBottomMostCompaction,
 }

 impl std::fmt::Debug for Timeline {
@@ -1096,7 +1097,6 @@ impl Timeline {
    /// scan iterator interface. We could optimize this interface later to avoid some checks in the vectored
    /// get path to maintain and split the probing and to-be-probe keyspace. We also need to ensure that
    /// the scan operation will not cause OOM in the future.
-    #[allow(dead_code)]
    pub(crate) async fn scan(
        &self,
        keyspace: KeySpace,
@@ -5481,12 +5481,12 @@ impl Timeline {
        }
        images.sort_unstable_by(|(ka, _), (kb, _)| ka.cmp(kb));
        let min_key = *images.first().map(|(k, _)| k).unwrap();
-        let max_key = images.last().map(|(k, _)| k).unwrap().next();
+        let end_key = images.last().map(|(k, _)| k).unwrap().next();
        let mut image_layer_writer = ImageLayerWriter::new(
            self.conf,
            self.timeline_id,
            self.tenant_shard_id,
-            &(min_key..max_key),
+            &(min_key..end_key),
            lsn,
            ctx,
        )
@@ -5518,7 +5518,7 @@ impl Timeline {
        let last_record_lsn = self.get_last_record_lsn();
        deltas.sort_unstable_by(|(ka, la, _), (kb, lb, _)| (ka, la).cmp(&(kb, lb)));
        let min_key = *deltas.first().map(|(k, _, _)| k).unwrap();
-        let max_key = deltas.last().map(|(k, _, _)| k).unwrap().next();
+        let end_key = deltas.last().map(|(k, _, _)| k).unwrap().next();
        let min_lsn = *deltas.iter().map(|(_, lsn, _)| lsn).min().unwrap();
        let max_lsn = *deltas.iter().map(|(_, lsn, _)| lsn).max().unwrap();
        assert!(
@@ -5541,7 +5541,7 @@ impl Timeline {
        for (key, lsn, val) in deltas {
            delta_layer_writer.put_value(key, lsn, val, ctx).await?;
        }
-        let delta_layer = delta_layer_writer.finish(max_key, self, ctx).await?;
+        let delta_layer = delta_layer_writer.finish(end_key, self, ctx).await?;

        {
            let mut guard = self.layers.write().await;
@@ -47,10 +47,14 @@ impl Timeline {
    /// TODO: cancellation
    pub(crate) async fn compact_legacy(
        self: &Arc<Self>,
-        _cancel: &CancellationToken,
+        cancel: &CancellationToken,
        flags: EnumSet<CompactFlags>,
        ctx: &RequestContext,
    ) -> Result<(), CompactionError> {
+        if flags.contains(CompactFlags::EnhancedGcBottomMostCompaction) {
+            return self.compact_with_gc(cancel, ctx).await;
+        }
+
        // High level strategy for compaction / image creation:
        //
        // 1. First, calculate the desired "partitioning" of the
@@ -959,13 +963,20 @@ impl Timeline {
    /// the GC horizon without considering retain_lsns. Then, it does a full compaction over all these delta
    /// layers and image layers, which generates image layers on the gc horizon, drop deltas below gc horizon,
    /// and create delta layers with all deltas >= gc horizon.
-    #[cfg(test)]
    pub(crate) async fn compact_with_gc(
        self: &Arc<Self>,
        _cancel: &CancellationToken,
        ctx: &RequestContext,
    ) -> Result<(), CompactionError> {
        use crate::tenant::storage_layer::ValueReconstructState;
+        use std::collections::BTreeSet;
+
+        info!("running enhanced gc bottom-most compaction");
+
+        scopeguard::defer! {
+            info!("done enhanced gc bottom-most compaction");
+        };
+
        // Step 0: pick all delta layers + image layers below/intersect with the GC horizon.
        // The layer selection has the following properties:
        // 1. If a layer is in the selection, all layers below it are in the selection.
@@ -974,6 +985,11 @@ impl Timeline {
            let guard = self.layers.read().await;
            let layers = guard.layer_map();
            let gc_info = self.gc_info.read().unwrap();
+            if !gc_info.retain_lsns.is_empty() || !gc_info.leases.is_empty() {
+                return Err(CompactionError::Other(anyhow!(
+                    "enhanced legacy compaction currently does not support retain_lsns (branches)"
+                )));
+            }
            let gc_cutoff = Lsn::min(gc_info.cutoffs.horizon, gc_info.cutoffs.pitr);
            let mut selected_layers = Vec::new();
            // TODO: consider retain_lsns
@@ -985,21 +1001,36 @@ impl Timeline {
            }
            (selected_layers, gc_cutoff)
        };
+        info!(
+            "picked {} layers for compaction with gc_cutoff={}",
+            layer_selection.len(),
+            gc_cutoff
+        );
        // Step 1: (In the future) construct a k-merge iterator over all layers. For now, simply collect all keys + LSNs.
+        // Also, collect the layer information to decide when to split the new delta layers.
        let mut all_key_values = Vec::new();
+        let mut delta_split_points = BTreeSet::new();
        for layer in &layer_selection {
            all_key_values.extend(layer.load_key_values(ctx).await?);
+            let desc = layer.layer_desc();
+            if desc.is_delta() {
+                // TODO: is it correct to only record split points for deltas intersecting with the GC horizon? (exclude those below/above the horizon)
+                // so that we can avoid having too many small delta layers.
+                let key_range = desc.get_key_range();
+                delta_split_points.insert(key_range.start);
+                delta_split_points.insert(key_range.end);
+            }
        }
        // Key small to large, LSN low to high, if the same LSN has both image and delta due to the merge of delta layers and
-        // image layers, make image appear later than delta.
+        // image layers, make image appear before than delta.
        struct ValueWrapper<'a>(&'a crate::repository::Value);
        impl Ord for ValueWrapper<'_> {
            fn cmp(&self, other: &Self) -> std::cmp::Ordering {
                use crate::repository::Value;
                use std::cmp::Ordering;
                match (self.0, other.0) {
-                    (Value::Image(_), Value::WalRecord(_)) => Ordering::Greater,
-                    (Value::WalRecord(_), Value::Image(_)) => Ordering::Less,
+                    (Value::Image(_), Value::WalRecord(_)) => Ordering::Less,
+                    (Value::WalRecord(_), Value::Image(_)) => Ordering::Greater,
                    _ => Ordering::Equal,
                }
            }
@@ -1018,13 +1049,6 @@ impl Timeline {
        all_key_values.sort_by(|(k1, l1, v1), (k2, l2, v2)| {
            (k1, l1, ValueWrapper(v1)).cmp(&(k2, l2, ValueWrapper(v2)))
        });
-        let max_lsn = all_key_values
-            .iter()
-            .map(|(_, lsn, _)| lsn)
-            .max()
-            .copied()
-            .unwrap()
-            + 1;
        // Step 2: Produce images+deltas. TODO: ensure newly-produced delta does not overlap with other deltas.
        // Data of the same key.
        let mut accumulated_values = Vec::new();
@@ -1043,14 +1067,24 @@ impl Timeline {
            // We have a list of deltas/images. We want to create image layers while collect garbages.
            for (key, lsn, val) in accumulated_values.iter().rev() {
                if *lsn > horizon {
-                    keys_above_horizon.push((*key, *lsn, val.clone())); // TODO: ensure one LSN corresponds to either delta or image instead of both
+                    if let Some((_, prev_lsn, _)) = keys_above_horizon.last_mut() {
+                        if *prev_lsn == *lsn {
+                            // The case that we have an LSN with both data from the delta layer and the image layer. As
+                            // `ValueWrapper` ensures that an image is ordered before a delta at the same LSN, we simply
+                            // drop this delta and keep the image.
+                            //
+                            // For example, we have delta layer key1@0x10, key1@0x20, and image layer key1@0x10, we will
+                            // keep the image for key1@0x10 and the delta for key1@0x20. key1@0x10 delta will be simply
+                            // dropped.
+                            continue;
+                        }
+                    }
+                    keys_above_horizon.push((*key, *lsn, val.clone()));
                } else if *lsn <= horizon {
                    match val {
                        crate::repository::Value::Image(image) => {
-                            if lsn <= &horizon {
-                                base_image = Some((*lsn, image.clone()));
-                                break;
-                            }
+                            base_image = Some((*lsn, image.clone()));
+                            break;
                        }
                        crate::repository::Value::WalRecord(wal) => {
                            delta_above_base_image.push((*lsn, wal.clone()));
@@ -1058,7 +1092,7 @@ impl Timeline {
                    }
                }
            }
-            delta_above_base_image.reverse();
+            // do not reverse delta_above_base_image, reconstruct state expects reversely-ordered records
            keys_above_horizon.reverse();
            let state = ValueReconstructState {
                img: base_image,
@@ -1068,15 +1102,59 @@ impl Timeline {
            Ok((keys_above_horizon, img))
        }

-        let mut delta_layer_writer = DeltaLayerWriter::new(
-            self.conf,
-            self.timeline_id,
-            self.tenant_shard_id,
-            all_key_values.first().unwrap().0,
-            gc_cutoff..max_lsn, // TODO: off by one?
-            ctx,
-        )
-        .await?;
+        async fn flush_deltas(
+            deltas: &mut Vec<(Key, Lsn, crate::repository::Value)>,
+            last_key: Key,
+            delta_split_points: &[Key],
+            current_delta_split_point: &mut usize,
+            tline: &Arc<Timeline>,
+            gc_cutoff: Lsn,
+            ctx: &RequestContext,
+        ) -> anyhow::Result<Option<ResidentLayer>> {
+            // Check if we need to split the delta layer. We split at the original delta layer boundary to avoid
+            // overlapping layers.
+            //
+            // If we have a structure like this:
+            //
+            // | Delta 1 |         | Delta 4 |
+            // |---------| Delta 2 |---------|
+            // | Delta 3 |         | Delta 5 |
+            //
+            // And we choose to compact delta 2+3+5. We will get an overlapping delta layer with delta 1+4.
+            // A simple solution here is to split the delta layers using the original boundary, while this
+            // might produce a lot of small layers. This should be improved and fixed in the future.
+            let mut need_split = false;
+            while *current_delta_split_point < delta_split_points.len()
+                && last_key >= delta_split_points[*current_delta_split_point]
+            {
+                *current_delta_split_point += 1;
+                need_split = true;
+            }
+            if !need_split {
+                return Ok(None);
+            }
+            let deltas = std::mem::take(deltas);
+            if deltas.is_empty() {
+                return Ok(None);
+            }
+            let end_lsn = deltas.iter().map(|(_, lsn, _)| lsn).max().copied().unwrap() + 1;
+            let mut delta_layer_writer = DeltaLayerWriter::new(
+                tline.conf,
+                tline.timeline_id,
+                tline.tenant_shard_id,
+                deltas.first().unwrap().0,
+                gc_cutoff..end_lsn,
+                ctx,
+            )
+            .await?;
+            let key_end = deltas.last().unwrap().0.next();
+            for (key, lsn, val) in deltas {
+                delta_layer_writer.put_value(key, lsn, val, ctx).await?;
+            }
+            let delta_layer = delta_layer_writer.finish(key_end, tline, ctx).await?;
+            Ok(Some(delta_layer))
+        }
+
        let mut image_layer_writer = ImageLayerWriter::new(
            self.conf,
            self.timeline_id,
@@ -1087,6 +1165,10 @@ impl Timeline {
        )
        .await?;

+        let mut delta_values = Vec::new();
+        let delta_split_points = delta_split_points.into_iter().collect_vec();
+        let mut current_delta_split_point = 0;
+        let mut delta_layers = Vec::new();
        for item @ (key, _, _) in &all_key_values {
            if &last_key == key {
                accumulated_values.push(item);
@@ -1094,34 +1176,63 @@ impl Timeline {
                let (deltas, image) =
                    flush_accumulated_states(self, last_key, &accumulated_values, gc_cutoff)
                        .await?;
+                // Put the image into the image layer. Currently we have a single big layer for the compaction.
                image_layer_writer.put_image(last_key, image, ctx).await?;
-                for (key, lsn, val) in deltas {
-                    delta_layer_writer.put_value(key, lsn, val, ctx).await?;
-                }
+                delta_values.extend(deltas);
+                delta_layers.extend(
+                    flush_deltas(
+                        &mut delta_values,
+                        last_key,
+                        &delta_split_points,
+                        &mut current_delta_split_point,
+                        self,
+                        gc_cutoff,
+                        ctx,
+                    )
+                    .await?,
+                );
                accumulated_values.clear();
                accumulated_values.push(item);
                last_key = *key;
            }
        }
+
+        // TODO: move this part to the loop body
        let (deltas, image) =
            flush_accumulated_states(self, last_key, &accumulated_values, gc_cutoff).await?;
+        // Put the image into the image layer. Currently we have a single big layer for the compaction.
        image_layer_writer.put_image(last_key, image, ctx).await?;
-        for (key, lsn, val) in deltas {
-            delta_layer_writer.put_value(key, lsn, val, ctx).await?;
-        }
-        accumulated_values.clear();
-        // TODO: split layers
-        let delta_layer = delta_layer_writer.finish(last_key, self, ctx).await?;
+        delta_values.extend(deltas);
+        delta_layers.extend(
+            flush_deltas(
+                &mut delta_values,
+                last_key,
+                &delta_split_points,
+                &mut current_delta_split_point,
+                self,
+                gc_cutoff,
+                ctx,
+            )
+            .await?,
+        );
+
        let image_layer = image_layer_writer.finish(self, ctx).await?;
+        info!(
+            "produced {} delta layers and {} image layers",
+            delta_layers.len(),
+            1
+        );
+        let mut compact_to = Vec::new();
+        compact_to.extend(delta_layers);
+        compact_to.push(image_layer);
        // Step 3: Place back to the layer map.
        {
            let mut guard = self.layers.write().await;
-            guard.finish_gc_compaction(
-                &layer_selection,
-                &[delta_layer.clone(), image_layer.clone()],
-                &self.metrics,
-            )
+            guard.finish_gc_compaction(&layer_selection, &compact_to, &self.metrics)
        };
+
+        self.remote_client
+            .schedule_compaction_update(&layer_selection, &compact_to)?;
        Ok(())
    }
 }
@@ -227,7 +227,6 @@ impl LayerManager {
    }

    /// Called when a GC-compaction is completed.
-    #[cfg(test)]
    pub(crate) fn finish_gc_compaction(
        &mut self,
        compact_from: &[Layer],
@@ -20,6 +20,7 @@ use std::num::NonZeroUsize;

 use bytes::BytesMut;
 use pageserver_api::key::Key;
+use tokio_epoll_uring::BoundedBuf;
 use utils::lsn::Lsn;
 use utils::vec_map::VecMap;

@@ -77,7 +78,7 @@ pub(crate) struct VectoredReadBuilder {
    start: u64,
    end: u64,
    blobs_at: VecMap<u64, BlobMeta>,
-    max_read_size: usize,
+    max_read_size: Option<usize>,
 }

 impl VectoredReadBuilder {
@@ -90,7 +91,7 @@ impl VectoredReadBuilder {
        start_offset: u64,
        end_offset: u64,
        meta: BlobMeta,
-        max_read_size: usize,
+        max_read_size: Option<usize>,
    ) -> Self {
        let mut blobs_at = VecMap::default();
        blobs_at
@@ -111,7 +112,13 @@ impl VectoredReadBuilder {
    pub(crate) fn extend(&mut self, start: u64, end: u64, meta: BlobMeta) -> VectoredReadExtended {
        tracing::trace!(start, end, "trying to extend");
        let size = (end - start) as usize;
-        if self.end == start && self.size() + size <= self.max_read_size {
+        if self.end == start && {
+            if let Some(max_read_size) = self.max_read_size {
+                self.size() + size <= max_read_size
+            } else {
+                true
+            }
+        } {
            self.end = end;
            self.blobs_at
                .append(start, meta)
@@ -157,7 +164,7 @@ pub struct VectoredReadPlanner {
    // Arguments for previous blob passed into [`VectoredReadPlanner::handle`]
    prev: Option<(Key, Lsn, u64, BlobFlag)>,

-    max_read_size: usize,
+    max_read_size: Option<usize>,
 }

 impl VectoredReadPlanner {
@@ -165,7 +172,20 @@ impl VectoredReadPlanner {
        Self {
            blobs: BTreeMap::new(),
            prev: None,
-            max_read_size,
+            max_read_size: Some(max_read_size),
+        }
+    }
+
+    /// This function should *only* be used if the caller has a way to control the limit. e.g., in [`StreamingVectoredReadPlanner`],
+    /// it uses the vectored read planner to avoid duplicated logic on handling blob start/end, while expecting the vectored
+    /// read planner to give a single read to a continuous range of bytes in the image layer. Therefore, it does not need the
+    /// code path to split reads into chunks of `max_read_size`, and controls the read size itself.
+    #[cfg(test)]
+    pub(crate) fn new_caller_controlled_max_limit() -> Self {
+        Self {
+            blobs: BTreeMap::new(),
+            prev: None,
+            max_read_size: None,
        }
    }

@@ -297,8 +317,9 @@ impl<'a> VectoredBlobReader<'a> {
        );
        let buf = self
            .file
-            .read_exact_at_n(buf, read.start, read.size(), ctx)
-            .await?;
+            .read_exact_at(buf.slice(0..read.size()), read.start, ctx)
+            .await?
+            .into_inner();

        let blobs_at = read.blobs_at.as_slice();
        let start_offset = blobs_at.first().expect("VectoredRead is never empty").0;
@@ -354,6 +375,87 @@ impl<'a> VectoredBlobReader<'a> {
    }
 }

+/// Read planner used in [`crate::tenant::storage_layer::image_layer::ImageLayerIterator`]. It provides a streaming API for
+/// getting read blobs. It returns a batch when `handle` gets called and when the current key would exceed the read_size and
+/// max_cnt constraints. Underlying it uses [`VectoredReadPlanner`].
+#[cfg(test)]
+pub struct StreamingVectoredReadPlanner {
+    planner: VectoredReadPlanner,
+    /// Max read size per batch
+    max_read_size: u64,
+    /// Max item count per batch
+    max_cnt: usize,
+    /// The first offset of this batch
+    this_batch_first_offset: Option<u64>,
+    /// Size of the current batch
+    cnt: usize,
+}
+
+#[cfg(test)]
+impl StreamingVectoredReadPlanner {
+    pub fn new(max_read_size: u64, max_cnt: usize) -> Self {
+        assert!(max_cnt > 0);
+        assert!(max_read_size > 0);
+        Self {
+            // We want to have exactly one read syscall (plus several others for index lookup) for each `next_batch` call.
+            // Therefore, we enforce `self.max_read_size` by ourselves instead of using the VectoredReadPlanner's capability,
+            // to avoid splitting into two I/Os.
+            planner: VectoredReadPlanner::new_caller_controlled_max_limit(),
+            max_cnt,
+            max_read_size,
+            this_batch_first_offset: None,
+            cnt: 0,
+        }
+    }
+
+    fn emit(&mut self, this_batch_first_offset: u64) -> VectoredRead {
+        let planner = std::mem::replace(
+            &mut self.planner,
+            VectoredReadPlanner::new_caller_controlled_max_limit(),
+        );
+        self.this_batch_first_offset = Some(this_batch_first_offset);
+        self.cnt = 1;
+        let mut batch = planner.finish();
+        assert_eq!(batch.len(), 1, "should have exactly one read batch");
+        batch.pop().unwrap()
+    }
+
+    pub fn handle(
+        &mut self,
+        key: Key,
+        lsn: Lsn,
+        offset: u64,
+        flag: BlobFlag,
+    ) -> Option<VectoredRead> {
+        if let Some(begin_offset) = self.this_batch_first_offset {
+            // Each batch will have at least one item b/c `self.this_batch_first_offset` is set
+            // after one item gets processed
+            if offset - begin_offset > self.max_read_size {
+                self.planner.handle_range_end(offset); // End the current batch with the offset
+                let batch = self.emit(offset); // Produce a batch
+                self.planner.handle(key, lsn, offset, flag); // Add this key to the next batch
+                return Some(batch);
+            }
+        } else {
+            self.this_batch_first_offset = Some(offset)
+        }
+        if self.cnt >= self.max_cnt {
+            self.planner.handle_range_end(offset); // End the current batch with the offset
+            let batch = self.emit(offset); // Produce a batch
+            self.planner.handle(key, lsn, offset, flag); // Add this key to the next batch
+            return Some(batch);
+        }
+        self.planner.handle(key, lsn, offset, flag); // Add this key to the current batch
+        self.cnt += 1;
+        None
+    }
+
+    pub fn handle_range_end(&mut self, offset: u64) -> VectoredRead {
+        self.planner.handle_range_end(offset);
+        self.emit(offset)
+    }
+}
+
 #[cfg(test)]
 mod tests {
    use super::*;
@@ -13,7 +13,7 @@
 use crate::context::RequestContext;
 use crate::metrics::{StorageIoOperation, STORAGE_IO_SIZE, STORAGE_IO_TIME_METRIC};

-use crate::page_cache::PageWriteGuard;
+use crate::page_cache::{PageWriteGuard, PAGE_SZ};
 use crate::tenant::TENANTS_SEGMENT_NAME;
 use camino::{Utf8Path, Utf8PathBuf};
 use once_cell::sync::OnceCell;
@@ -48,6 +48,7 @@ pub(crate) mod owned_buffers_io {
    //! but for the time being we're proving out the primitives in the neon.git repo
    //! for faster iteration.

+    pub(crate) mod slice;
    pub(crate) mod write;
    pub(crate) mod util {
        pub(crate) mod size_tracking_writer;
@@ -143,16 +144,17 @@ struct SlotInner {
 /// Impl of [`tokio_epoll_uring::IoBuf`] and [`tokio_epoll_uring::IoBufMut`] for [`PageWriteGuard`].
 struct PageWriteGuardBuf {
    page: PageWriteGuard<'static>,
-    init_up_to: usize,
 }
 // Safety: the [`PageWriteGuard`] gives us exclusive ownership of the page cache slot,
 // and the location remains stable even if [`Self`] or the [`PageWriteGuard`] is moved.
+// Page cache pages are zero-initialized, so, wrt uninitialized memory we're good.
+// (Page cache tracks separately whether the contents are valid, see `PageWriteGuard::mark_valid`.)
 unsafe impl tokio_epoll_uring::IoBuf for PageWriteGuardBuf {
    fn stable_ptr(&self) -> *const u8 {
        self.page.as_ptr()
    }
    fn bytes_init(&self) -> usize {
-        self.init_up_to
+        self.page.len()
    }
    fn bytes_total(&self) -> usize {
        self.page.len()
@@ -166,8 +168,8 @@ unsafe impl tokio_epoll_uring::IoBufMut for PageWriteGuardBuf {
    }

    unsafe fn set_init(&mut self, pos: usize) {
+        // There shouldn't really be any reason to call this API since bytes_init() == bytes_total().
        assert!(pos <= self.page.len());
-        self.init_up_to = pos;
    }
 }

@@ -585,37 +587,37 @@ impl VirtualFile {
        Ok(self.pos)
    }

-    pub async fn read_exact_at<B>(
+    /// Read the file contents in range `offset..(offset + slice.bytes_total())` into `slice[0..slice.bytes_total()]`.
+    ///
+    /// The returned `Slice<Buf>` is equivalent to the input `slice`, i.e., it's the same view into the same buffer.
+    pub async fn read_exact_at<Buf>(
        &self,
-        buf: B,
+        slice: Slice<Buf>,
        offset: u64,
        ctx: &RequestContext,
-    ) -> Result<B, Error>
+    ) -> Result<Slice<Buf>, Error>
    where
-        B: IoBufMut + Send,
+        Buf: IoBufMut + Send,
    {
-        let (buf, res) = read_exact_at_impl(buf, offset, None, |buf, offset| {
-            self.read_at(buf, offset, ctx)
-        })
-        .await;
-        res.map(|()| buf)
-    }
+        let assert_we_return_original_bounds = if cfg!(debug_assertions) {
+            Some((slice.stable_ptr() as usize, slice.bytes_total()))
+        } else {
+            None
+        };

-    pub async fn read_exact_at_n<B>(
-        &self,
-        buf: B,
-        offset: u64,
-        count: usize,
-        ctx: &RequestContext,
-    ) -> Result<B, Error>
-    where
-        B: IoBufMut + Send,
-    {
-        let (buf, res) = read_exact_at_impl(buf, offset, Some(count), |buf, offset| {
-            self.read_at(buf, offset, ctx)
-        })
-        .await;
-        res.map(|()| buf)
+        let original_bounds = slice.bounds();
+        let (buf, res) =
+            read_exact_at_impl(slice, offset, |buf, offset| self.read_at(buf, offset, ctx)).await;
+        let res = res.map(|_| buf.slice(original_bounds));
+
+        if let Some(original_bounds) = assert_we_return_original_bounds {
+            if let Ok(slice) = &res {
+                let returned_bounds = (slice.stable_ptr() as usize, slice.bytes_total());
+                assert_eq!(original_bounds, returned_bounds);
+            }
+        }
+
+        res
    }

    /// Like [`Self::read_exact_at`] but for [`PageWriteGuard`].
@@ -625,13 +627,11 @@ impl VirtualFile {
        offset: u64,
        ctx: &RequestContext,
    ) -> Result<PageWriteGuard<'static>, Error> {
-        let buf = PageWriteGuardBuf {
-            page,
-            init_up_to: 0,
-        };
-        let res = self.read_exact_at(buf, offset, ctx).await;
-        res.map(|PageWriteGuardBuf { page, .. }| page)
-            .map_err(|e| Error::new(ErrorKind::Other, e))
+        let buf = PageWriteGuardBuf { page }.slice_full();
+        debug_assert_eq!(buf.bytes_total(), PAGE_SZ);
+        self.read_exact_at(buf, offset, ctx)
+            .await
+            .map(|slice| slice.into_inner().page)
    }

    // Copied from https://doc.rust-lang.org/1.72.0/src/std/os/unix/fs.rs.html#219-235
@@ -722,14 +722,14 @@ impl VirtualFile {
        (buf, Ok(n))
    }

-    pub(crate) async fn read_at<B>(
+    pub(crate) async fn read_at<Buf>(
        &self,
-        buf: B,
+        buf: tokio_epoll_uring::Slice<Buf>,
        offset: u64,
        _ctx: &RequestContext, /* TODO: use for metrics: https://github.com/neondatabase/neon/issues/6107 */
-    ) -> (B, Result<usize, Error>)
+    ) -> (tokio_epoll_uring::Slice<Buf>, Result<usize, Error>)
    where
-        B: tokio_epoll_uring::BoundedBufMut + Send,
+        Buf: tokio_epoll_uring::IoBufMut + Send,
    {
        let file_guard = match self.lock_file().await {
            Ok(file_guard) => file_guard,
@@ -781,26 +781,16 @@ impl VirtualFile {
 }

 // Adapted from https://doc.rust-lang.org/1.72.0/src/std/os/unix/fs.rs.html#117-135
-pub async fn read_exact_at_impl<B, F, Fut>(
-    buf: B,
+pub async fn read_exact_at_impl<Buf, F, Fut>(
+    mut buf: tokio_epoll_uring::Slice<Buf>,
    mut offset: u64,
-    count: Option<usize>,
    mut read_at: F,
-) -> (B, std::io::Result<()>)
+) -> (Buf, std::io::Result<()>)
 where
-    B: IoBufMut + Send,
-    F: FnMut(tokio_epoll_uring::Slice<B>, u64) -> Fut,
-    Fut: std::future::Future<Output = (tokio_epoll_uring::Slice<B>, std::io::Result<usize>)>,
+    Buf: IoBufMut + Send,
+    F: FnMut(tokio_epoll_uring::Slice<Buf>, u64) -> Fut,
+    Fut: std::future::Future<Output = (tokio_epoll_uring::Slice<Buf>, std::io::Result<usize>)>,
 {
-    let mut buf: tokio_epoll_uring::Slice<B> = match count {
-        Some(count) => {
-            assert!(count <= buf.bytes_total());
-            assert!(count > 0);
-            buf.slice(..count) // may include uninitialized memory
-        }
-        None => buf.slice_full(), // includes all the uninitialized memory
-    };
-
    while buf.bytes_total() != 0 {
        let res;
        (buf, res) = read_at(buf, offset).await;
@@ -882,7 +872,7 @@ mod test_read_exact_at_impl {

    #[tokio::test]
    async fn test_basic() {
-        let buf = Vec::with_capacity(5);
+        let buf = Vec::with_capacity(5).slice_full();
        let mock_read_at = Arc::new(tokio::sync::Mutex::new(MockReadAt {
            expectations: VecDeque::from(vec![Expectation {
                offset: 0,
@@ -890,7 +880,7 @@ mod test_read_exact_at_impl {
                result: Ok(vec![b'a', b'b', b'c', b'd', b'e']),
            }]),
        }));
-        let (buf, res) = read_exact_at_impl(buf, 0, None, |buf, offset| {
+        let (buf, res) = read_exact_at_impl(buf, 0, |buf, offset| {
            let mock_read_at = Arc::clone(&mock_read_at);
            async move { mock_read_at.lock().await.read_at(buf, offset).await }
        })
@@ -899,33 +889,13 @@ mod test_read_exact_at_impl {
        assert_eq!(buf, vec![b'a', b'b', b'c', b'd', b'e']);
    }

-    #[tokio::test]
-    async fn test_with_count() {
-        let buf = Vec::with_capacity(5);
-        let mock_read_at = Arc::new(tokio::sync::Mutex::new(MockReadAt {
-            expectations: VecDeque::from(vec![Expectation {
-                offset: 0,
-                bytes_total: 3,
-                result: Ok(vec![b'a', b'b', b'c']),
-            }]),
-        }));
-
-        let (buf, res) = read_exact_at_impl(buf, 0, Some(3), |buf, offset| {
-            let mock_read_at = Arc::clone(&mock_read_at);
-            async move { mock_read_at.lock().await.read_at(buf, offset).await }
-        })
-        .await;
-        assert!(res.is_ok());
-        assert_eq!(buf, vec![b'a', b'b', b'c']);
-    }
-
    #[tokio::test]
    async fn test_empty_buf_issues_no_syscall() {
-        let buf = Vec::new();
+        let buf = Vec::new().slice_full();
        let mock_read_at = Arc::new(tokio::sync::Mutex::new(MockReadAt {
            expectations: VecDeque::new(),
        }));
-        let (_buf, res) = read_exact_at_impl(buf, 0, None, |buf, offset| {
+        let (_buf, res) = read_exact_at_impl(buf, 0, |buf, offset| {
            let mock_read_at = Arc::clone(&mock_read_at);
            async move { mock_read_at.lock().await.read_at(buf, offset).await }
        })
@@ -935,7 +905,7 @@ mod test_read_exact_at_impl {

    #[tokio::test]
    async fn test_two_read_at_calls_needed_until_buf_filled() {
-        let buf = Vec::with_capacity(4);
+        let buf = Vec::with_capacity(4).slice_full();
        let mock_read_at = Arc::new(tokio::sync::Mutex::new(MockReadAt {
            expectations: VecDeque::from(vec![
                Expectation {
@@ -950,7 +920,7 @@ mod test_read_exact_at_impl {
                },
            ]),
        }));
-        let (buf, res) = read_exact_at_impl(buf, 0, None, |buf, offset| {
+        let (buf, res) = read_exact_at_impl(buf, 0, |buf, offset| {
            let mock_read_at = Arc::clone(&mock_read_at);
            async move { mock_read_at.lock().await.read_at(buf, offset).await }
        })
@@ -961,7 +931,7 @@ mod test_read_exact_at_impl {

    #[tokio::test]
    async fn test_eof_before_buffer_full() {
-        let buf = Vec::with_capacity(3);
+        let buf = Vec::with_capacity(3).slice_full();
        let mock_read_at = Arc::new(tokio::sync::Mutex::new(MockReadAt {
            expectations: VecDeque::from(vec![
                Expectation {
@@ -981,7 +951,7 @@ mod test_read_exact_at_impl {
                },
            ]),
        }));
-        let (_buf, res) = read_exact_at_impl(buf, 0, None, |buf, offset| {
+        let (_buf, res) = read_exact_at_impl(buf, 0, |buf, offset| {
            let mock_read_at = Arc::clone(&mock_read_at);
            async move { mock_read_at.lock().await.read_at(buf, offset).await }
        })
@@ -1051,27 +1021,29 @@ impl VirtualFile {
        ctx: &RequestContext,
    ) -> Result<crate::tenant::block_io::BlockLease<'_>, std::io::Error> {
        use crate::page_cache::PAGE_SZ;
-        let buf = vec![0; PAGE_SZ];
-        let buf = self
-            .read_exact_at(buf, blknum as u64 * (PAGE_SZ as u64), ctx)
+        let slice = Vec::with_capacity(PAGE_SZ).slice_full();
+        assert_eq!(slice.bytes_total(), PAGE_SZ);
+        let slice = self
+            .read_exact_at(slice, blknum as u64 * (PAGE_SZ as u64), ctx)
            .await?;
-        Ok(crate::tenant::block_io::BlockLease::Vec(buf))
+        Ok(crate::tenant::block_io::BlockLease::Vec(slice.into_inner()))
    }

    async fn read_to_end(&mut self, buf: &mut Vec<u8>, ctx: &RequestContext) -> Result<(), Error> {
        let mut tmp = vec![0; 128];
        loop {
-            let res;
-            (tmp, res) = self.read_at(tmp, self.pos, ctx).await;
+            let slice = tmp.slice(..128);
+            let (slice, res) = self.read_at(slice, self.pos, ctx).await;
            match res {
                Ok(0) => return Ok(()),
                Ok(n) => {
                    self.pos += n as u64;
-                    buf.extend_from_slice(&tmp[..n]);
+                    buf.extend_from_slice(&slice[..n]);
                }
                Err(ref e) if e.kind() == std::io::ErrorKind::Interrupted => {}
                Err(e) => return Err(e),
            }
+            tmp = slice.into_inner();
        }
    }
 }
@@ -1185,6 +1157,7 @@ mod tests {
    use crate::task_mgr::TaskKind;

    use super::*;
+    use owned_buffers_io::slice::SliceExt;
    use rand::seq::SliceRandom;
    use rand::thread_rng;
    use rand::Rng;
@@ -1206,13 +1179,16 @@ mod tests {
    impl MaybeVirtualFile {
        async fn read_exact_at(
            &self,
-            mut buf: Vec<u8>,
+            mut slice: tokio_epoll_uring::Slice<Vec<u8>>,
            offset: u64,
            ctx: &RequestContext,
-        ) -> Result<Vec<u8>, Error> {
+        ) -> Result<tokio_epoll_uring::Slice<Vec<u8>>, Error> {
            match self {
-                MaybeVirtualFile::VirtualFile(file) => file.read_exact_at(buf, offset, ctx).await,
-                MaybeVirtualFile::File(file) => file.read_exact_at(&mut buf, offset).map(|()| buf),
+                MaybeVirtualFile::VirtualFile(file) => file.read_exact_at(slice, offset, ctx).await,
+                MaybeVirtualFile::File(file) => {
+                    let rust_slice: &mut [u8] = slice.as_mut_rust_slice_full_zeroed();
+                    file.read_exact_at(rust_slice, offset).map(|()| slice)
+                }
            }
        }
        async fn write_all_at<B: BoundedBuf<Buf = Buf>, Buf: IoBuf + Send>(
@@ -1286,9 +1262,12 @@ mod tests {
            len: usize,
            ctx: &RequestContext,
        ) -> Result<String, Error> {
-            let buf = vec![0; len];
-            let buf = self.read_exact_at(buf, pos, ctx).await?;
-            Ok(String::from_utf8(buf).unwrap())
+            let slice = Vec::with_capacity(len).slice_full();
+            assert_eq!(slice.bytes_total(), len);
+            let slice = self.read_exact_at(slice, pos, ctx).await?;
+            let vec = slice.into_inner();
+            assert_eq!(vec.len(), len);
+            Ok(String::from_utf8(vec).unwrap())
        }
    }

@@ -1507,7 +1486,11 @@ mod tests {
                let mut rng = rand::rngs::OsRng;
                for _ in 1..1000 {
                    let f = &files[rng.gen_range(0..files.len())];
-                    buf = f.read_exact_at(buf, 0, &ctx).await.unwrap();
+                    buf = f
+                        .read_exact_at(buf.slice_full(), 0, &ctx)
+                        .await
+                        .unwrap()
+                        .into_inner();
                    assert!(buf == SAMPLE);
                }
            });
@@ -107,7 +107,7 @@ use std::{
    sync::atomic::{AtomicU8, Ordering},
 };

-use super::{FileGuard, Metadata};
+use super::{owned_buffers_io::slice::SliceExt, FileGuard, Metadata};

 #[cfg(target_os = "linux")]
 fn epoll_uring_error_to_std(e: tokio_epoll_uring::Error<std::io::Error>) -> std::io::Error {
@@ -120,38 +120,29 @@ fn epoll_uring_error_to_std(e: tokio_epoll_uring::Error<std::io::Error>) -> std:
 }

 impl IoEngine {
-    pub(super) async fn read_at<B>(
+    pub(super) async fn read_at<Buf>(
        &self,
        file_guard: FileGuard,
        offset: u64,
-        mut buf: B,
-    ) -> ((FileGuard, B), std::io::Result<usize>)
+        mut slice: tokio_epoll_uring::Slice<Buf>,
+    ) -> (
+        (FileGuard, tokio_epoll_uring::Slice<Buf>),
+        std::io::Result<usize>,
+    )
    where
-        B: tokio_epoll_uring::BoundedBufMut + Send,
+        Buf: tokio_epoll_uring::IoBufMut + Send,
    {
        match self {
            IoEngine::NotSet => panic!("not initialized"),
            IoEngine::StdFs => {
-                // SAFETY: `dst` only lives at most as long as this match arm, during which buf remains valid memory.
-                let dst = unsafe {
-                    std::slice::from_raw_parts_mut(buf.stable_mut_ptr(), buf.bytes_total())
-                };
-                let res = file_guard.with_std_file(|std_file| std_file.read_at(dst, offset));
-                if let Ok(nbytes) = &res {
-                    assert!(*nbytes <= buf.bytes_total());
-                    // SAFETY: see above assertion
-                    unsafe {
-                        buf.set_init(*nbytes);
-                    }
-                }
-                #[allow(dropping_references)]
-                drop(dst);
-                ((file_guard, buf), res)
+                let rust_slice = slice.as_mut_rust_slice_full_zeroed();
+                let res = file_guard.with_std_file(|std_file| std_file.read_at(rust_slice, offset));
+                ((file_guard, slice), res)
            }
            #[cfg(target_os = "linux")]
            IoEngine::TokioEpollUring => {
                let system = tokio_epoll_uring_ext::thread_local_system().await;
-                let (resources, res) = system.read(file_guard, offset, buf).await;
+                let (resources, res) = system.read(file_guard, offset, slice).await;
                (resources, res.map_err(epoll_uring_error_to_std))
            }
        }
@@ -0,0 +1,121 @@
+use tokio_epoll_uring::BoundedBuf;
+use tokio_epoll_uring::BoundedBufMut;
+use tokio_epoll_uring::IoBufMut;
+use tokio_epoll_uring::Slice;
+
+pub(crate) trait SliceExt {
+    /// Get a `&mut[0..self.bytes_total()`] slice, for when you need to do borrow-based IO.
+    ///
+    /// See the test case `test_slice_full_zeroed` for the difference to just doing `&slice[..]`
+    fn as_mut_rust_slice_full_zeroed(&mut self) -> &mut [u8];
+}
+
+impl<B> SliceExt for Slice<B>
+where
+    B: IoBufMut,
+{
+    #[inline(always)]
+    fn as_mut_rust_slice_full_zeroed(&mut self) -> &mut [u8] {
+        // zero-initialize the uninitialized parts of the buffer so we can create a Rust slice
+        //
+        // SAFETY: we own `slice`, don't write outside the bounds
+        unsafe {
+            let to_init = self.bytes_total() - self.bytes_init();
+            self.stable_mut_ptr()
+                .add(self.bytes_init())
+                .write_bytes(0, to_init);
+            self.set_init(self.bytes_total());
+        };
+        let bytes_total = self.bytes_total();
+        &mut self[0..bytes_total]
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use std::io::Read;
+
+    use super::*;
+    use bytes::Buf;
+    use tokio_epoll_uring::Slice;
+
+    #[test]
+    fn test_slice_full_zeroed() {
+        let make_fake_file = || bytes::BytesMut::from(&b"12345"[..]).reader();
+
+        // before we start the test, let's make sure we have a shared understanding of what slice_full does
+        {
+            let buf = Vec::with_capacity(3);
+            let slice: Slice<_> = buf.slice_full();
+            assert_eq!(slice.bytes_init(), 0);
+            assert_eq!(slice.bytes_total(), 3);
+            let rust_slice = &slice[..];
+            assert_eq!(
+                rust_slice.len(),
+                0,
+                "Slice only derefs to a &[u8] of the initialized part"
+            );
+        }
+
+        // and also let's establish a shared understanding of .slice()
+        {
+            let buf = Vec::with_capacity(3);
+            let slice: Slice<_> = buf.slice(0..2);
+            assert_eq!(slice.bytes_init(), 0);
+            assert_eq!(slice.bytes_total(), 2);
+            let rust_slice = &slice[..];
+            assert_eq!(
+                rust_slice.len(),
+                0,
+                "Slice only derefs to a &[u8] of the initialized part"
+            );
+        }
+
+        // the above leads to the easy mistake of using slice[..] for borrow-based IO like so:
+        {
+            let buf = Vec::with_capacity(3);
+            let mut slice: Slice<_> = buf.slice_full();
+            assert_eq!(slice[..].len(), 0);
+            let mut file = make_fake_file();
+            file.read_exact(&mut slice[..]).unwrap(); // one might think this reads 3 bytes but it reads 0
+            assert_eq!(&slice[..] as &[u8], &[][..] as &[u8]);
+        }
+
+        // With owned buffers IO like with VirtualFilem, you could totally
+        // pass in a `Slice` with bytes_init()=0 but bytes_total()=5
+        // and it will read 5 bytes into the slice, and return a slice that has bytes_init()=5.
+        {
+            // TODO: demo
+        }
+
+        //
+        // Ok, now that we have a shared understanding let's demo how to use the extension trait.
+        //
+
+        // slice_full()
+        {
+            let buf = Vec::with_capacity(3);
+            let mut slice: Slice<_> = buf.slice_full();
+            let rust_slice = slice.as_mut_rust_slice_full_zeroed();
+            assert_eq!(rust_slice.len(), 3);
+            assert_eq!(rust_slice, &[0, 0, 0]);
+            let mut file = make_fake_file();
+            file.read_exact(rust_slice).unwrap();
+            assert_eq!(rust_slice, b"123");
+            assert_eq!(&slice[..], b"123");
+        }
+
+        // .slice(..)
+        {
+            let buf = Vec::with_capacity(3);
+            let mut slice: Slice<_> = buf.slice(0..2);
+            let rust_slice = slice.as_mut_rust_slice_full_zeroed();
+            assert_eq!(rust_slice.len(), 2);
+            assert_eq!(rust_slice, &[0, 0]);
+            let mut file = make_fake_file();
+            file.read_exact(rust_slice).unwrap();
+            assert_eq!(rust_slice, b"12");
+            assert_eq!(&slice[..], b"12");
+        }
+    }
+}
@@ -343,7 +343,33 @@ impl WalIngest {
                        xlog_checkpoint.oldestActiveXid,
                        self.checkpoint.oldestActiveXid
                    );
-                    self.checkpoint.oldestActiveXid = xlog_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
@@ -375,6 +401,7 @@ impl WalIngest {
                if info == pg_constants::XLOG_RUNNING_XACTS {
                    let xlrec = crate::walrecord::XlRunningXacts::decode(&mut buf);
                    self.checkpoint.oldestActiveXid = xlrec.oldest_running_xid;
+                    self.checkpoint_modified = true;
                }
            }
            pg_constants::RM_REPLORIGIN_ID => {
@@ -1277,13 +1304,10 @@ impl WalIngest {
            xlrec.pageno, xlrec.oldest_xid, xlrec.oldest_xid_db
        );

-        // Here we treat oldestXid and oldestXidDB
-        // differently from postgres redo routines.
-        // In postgres checkpoint.oldestXid lags behind xlrec.oldest_xid
-        // until checkpoint happens and updates the value.
-        // Here we can use the most recent value.
-        // It's just an optimization, though and can be deleted.
-        // TODO Figure out if there will be any issues with replica.
+        // In Postgres, oldestXid and oldestXidDB are updated in memory when the CLOG is
+        // 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;
        self.checkpoint_modified = true;
@@ -1384,14 +1408,31 @@ impl WalIngest {
            // Note: The multixact members can wrap around, even within one WAL record.
            offset = offset.wrapping_add(n_this_page as u32);
        }
-        if xlrec.mid >= self.checkpoint.nextMulti {
-            self.checkpoint.nextMulti = xlrec.mid + 1;
-            self.checkpoint_modified = true;
-        }
-        if xlrec.moff + xlrec.nmembers > self.checkpoint.nextMultiOffset {
-            self.checkpoint.nextMultiOffset = xlrec.moff + xlrec.nmembers;
+        let next_offset = offset;
+        assert!(xlrec.moff.wrapping_add(xlrec.nmembers) == next_offset);
+
+        // Update next-multi-xid and next-offset
+        //
+        // NB: In PostgreSQL, the next-multi-xid stored in the control file is allowed to
+        // go to 0, and it's fixed up by skipping to FirstMultiXactId in functions that
+        // read it, like GetNewMultiXactId(). This is different from how nextXid is
+        // incremented! nextXid skips over < FirstNormalTransactionId when the the value
+        // is stored, so it's never 0 in a checkpoint.
+        //
+        // I don't know why it's done that way, it seems less error-prone to skip over 0
+        // when the value is stored rather than when it's read. But let's do it the same
+        // way here.
+        let next_multi_xid = xlrec.mid.wrapping_add(1);
+
+        if self
+            .checkpoint
+            .update_next_multixid(next_multi_xid, next_offset)
+        {
            self.checkpoint_modified = true;
        }
+
+        // Also update the next-xid with the highest member. According to the comments in
+        // multixact_redo(), this shouldn't be necessary, but let's do the same here.
        let max_mbr_xid = xlrec.members.iter().fold(None, |acc, mbr| {
            if let Some(max_xid) = acc {
                if mbr.xid.wrapping_sub(max_xid) as i32 > 0 {
@@ -40,6 +40,7 @@ use std::time::Duration;
 use std::time::Instant;
 use tracing::*;
 use utils::lsn::Lsn;
+use utils::sync::gate::GateError;
 use utils::sync::heavier_once_cell;

 ///
@@ -53,10 +54,18 @@ pub struct PostgresRedoManager {
    tenant_shard_id: TenantShardId,
    conf: &'static PageServerConf,
    last_redo_at: std::sync::Mutex<Option<Instant>>,
-    /// The current [`process::WalRedoProcess`] that is used by new redo requests.
-    /// We use [`heavier_once_cell`] for coalescing the spawning, but the redo
-    /// requests don't use the [`heavier_once_cell::Guard`] to keep ahold of the
+    /// We use [`heavier_once_cell`] for
+    ///
+    /// 1. coalescing the lazy spawning of walredo processes ([`ProcessOnceCell::Spawned`])
+    /// 2. prevent new processes from being spawned on [`Self::shutdown`] (=> [`ProcessOnceCell::ManagerShutDown`]).
+    ///
+    /// # Spawning
+    ///
+    /// Redo requests use the once cell to coalesce onto one call to [`process::WalRedoProcess::launch`].
+    ///
+    /// Notably, requests don't use the [`heavier_once_cell::Guard`] to keep ahold of the
    /// their process object; we use [`Arc::clone`] for that.
+    ///
    /// This is primarily because earlier implementations that didn't  use [`heavier_once_cell`]
    /// had that behavior; it's probably unnecessary.
    /// The only merit of it is that if one walredo process encounters an error,
@@ -65,7 +74,63 @@ pub struct PostgresRedoManager {
    /// still be using the old redo process. But, those other tasks will most likely
    /// encounter an error as well, and errors are an unexpected condition anyway.
    /// So, probably we could get rid of the `Arc` in the future.
-    redo_process: heavier_once_cell::OnceCell<Arc<process::WalRedoProcess>>,
+    ///
+    /// # Shutdown
+    ///
+    /// See [`Self::launched_processes`].
+    redo_process: heavier_once_cell::OnceCell<ProcessOnceCell>,
+
+    /// Gate that is entered when launching a walredo process and held open
+    /// until the process has been `kill()`ed and `wait()`ed upon.
+    ///
+    /// Manager shutdown waits for this gate to close after setting the
+    /// [`ProcessOnceCell::ManagerShutDown`] state in [`Self::redo_process`].
+    ///
+    /// This type of usage is a bit unusual because gates usually keep track of
+    /// concurrent operations, e.g., every [`Self::request_redo`] that is inflight.
+    /// But we use it here to keep track of the _processes_ that we have launched,
+    /// which may outlive any individual redo request because
+    /// - we keep walredo process around until its quiesced to amortize spawn cost and
+    /// - the Arc may be held by multiple concurrent redo requests, so, just because
+    ///   you replace the [`Self::redo_process`] cell's content doesn't mean the
+    ///   process gets killed immediately.
+    ///
+    /// We could simplify this by getting rid of the [`Arc`].
+    /// See the comment on [`Self::redo_process`] for more details.
+    launched_processes: utils::sync::gate::Gate,
+}
+
+/// See [`PostgresRedoManager::redo_process`].
+enum ProcessOnceCell {
+    Spawned(Arc<Process>),
+    ManagerShutDown,
+}
+
+struct Process {
+    _launched_processes_guard: utils::sync::gate::GateGuard,
+    process: process::WalRedoProcess,
+}
+
+impl std::ops::Deref for Process {
+    type Target = process::WalRedoProcess;
+
+    fn deref(&self) -> &Self::Target {
+        &self.process
+    }
+}
+
+#[derive(Debug, thiserror::Error)]
+pub enum Error {
+    #[error("cancelled")]
+    Cancelled,
+    #[error(transparent)]
+    Other(#[from] anyhow::Error),
+}
+
+macro_rules! bail {
+    ($($arg:tt)*) => {
+        return Err($crate::walredo::Error::Other(::anyhow::anyhow!($($arg)*)));
+    }
 }

 ///
@@ -88,9 +153,9 @@ impl PostgresRedoManager {
        base_img: Option<(Lsn, Bytes)>,
        records: Vec<(Lsn, NeonWalRecord)>,
        pg_version: u32,
-    ) -> anyhow::Result<Bytes> {
+    ) -> Result<Bytes, Error> {
        if records.is_empty() {
-            anyhow::bail!("invalid WAL redo request with no records");
+            bail!("invalid WAL redo request with no records");
        }

        let base_img_lsn = base_img.as_ref().map(|p| p.0).unwrap_or(Lsn::INVALID);
@@ -148,10 +213,10 @@ impl PostgresRedoManager {
                    chrono::Utc::now().checked_sub_signed(chrono::Duration::from_std(age).ok()?)
                })
            },
-            process: self
-                .redo_process
-                .get()
-                .map(|p| WalRedoManagerProcessStatus { pid: p.id() }),
+            process: self.redo_process.get().and_then(|p| match &*p {
+                ProcessOnceCell::Spawned(p) => Some(WalRedoManagerProcessStatus { pid: p.id() }),
+                ProcessOnceCell::ManagerShutDown => None,
+            }),
        }
    }
 }
@@ -170,9 +235,39 @@ impl PostgresRedoManager {
            conf,
            last_redo_at: std::sync::Mutex::default(),
            redo_process: heavier_once_cell::OnceCell::default(),
+            launched_processes: utils::sync::gate::Gate::default(),
        }
    }

+    /// Shut down the WAL redo manager.
+    ///
+    /// After this future completes
+    /// - no redo process is running
+    /// - no new redo process will be spawned
+    /// - redo requests that need walredo process will fail with [`Error::Cancelled`]
+    /// - [`apply_neon`]-only redo requests may still work, but this may change in the future
+    ///
+    /// # Cancel-Safety
+    ///
+    /// This method is cancellation-safe.
+    pub async fn shutdown(&self) {
+        // prevent new processes from being spawned
+        let permit = match self.redo_process.get_or_init_detached().await {
+            Ok(guard) => {
+                let (proc, permit) = guard.take_and_deinit();
+                drop(proc); // this just drops the Arc, its refcount may not be zero yet
+                permit
+            }
+            Err(permit) => permit,
+        };
+        self.redo_process
+            .set(ProcessOnceCell::ManagerShutDown, permit);
+        // wait for ongoing requests to drain and the refcounts of all Arc<WalRedoProcess> that
+        // we ever launched to drop to zero, which when it happens synchronously kill()s & wait()s
+        // for the underlying process.
+        self.launched_processes.close().await;
+    }
+
    /// This type doesn't have its own background task to check for idleness: we
    /// rely on our owner calling this function periodically in its own housekeeping
    /// loops.
@@ -203,38 +298,48 @@ impl PostgresRedoManager {
        records: &[(Lsn, NeonWalRecord)],
        wal_redo_timeout: Duration,
        pg_version: u32,
-    ) -> anyhow::Result<Bytes> {
+    ) -> Result<Bytes, Error> {
        *(self.last_redo_at.lock().unwrap()) = Some(Instant::now());

        let (rel, blknum) = key.to_rel_block().context("invalid record")?;
        const MAX_RETRY_ATTEMPTS: u32 = 1;
        let mut n_attempts = 0u32;
        loop {
-            let proc: Arc<process::WalRedoProcess> =
-                match self.redo_process.get_or_init_detached().await {
-                    Ok(guard) => Arc::clone(&guard),
-                    Err(permit) => {
-                        // don't hold poison_guard, the launch code can bail
-                        let start = Instant::now();
-                        let proc = Arc::new(
-                            process::WalRedoProcess::launch(
+            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();
+                    let proc = Arc::new(Process {
+                            _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"
+                                ),
+                            },
+                            process: process::WalRedoProcess::launch(
                                self.conf,
                                self.tenant_shard_id,
                                pg_version,
                            )
                            .context("launch walredo process")?,
-                        );
-                        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(Arc::clone(&proc), permit);
-                        proc
-                    }
-                };
+                        });
+                    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 started_at = std::time::Instant::now();

@@ -299,12 +404,17 @@ impl PostgresRedoManager {
                match self.redo_process.get() {
                    None => (),
                    Some(guard) => {
-                        if Arc::ptr_eq(&proc, &*guard) {
-                            // 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.
+                        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.
+                                }
+                            }
                        }
                    }
                }
@@ -315,7 +425,7 @@ impl PostgresRedoManager {
            }
            n_attempts += 1;
            if n_attempts > MAX_RETRY_ATTEMPTS || result.is_ok() {
-                return result;
+                return result.map_err(Error::Other);
            }
        }
    }
@@ -329,7 +439,7 @@ impl PostgresRedoManager {
        lsn: Lsn,
        base_img: Option<Bytes>,
        records: &[(Lsn, NeonWalRecord)],
-    ) -> anyhow::Result<Bytes> {
+    ) -> Result<Bytes, Error> {
        let start_time = Instant::now();

        let mut page = BytesMut::new();
@@ -338,7 +448,7 @@ impl PostgresRedoManager {
            page.extend_from_slice(&fpi[..]);
        } else {
            // All the current WAL record types that we can handle require a base image.
-            anyhow::bail!("invalid neon WAL redo request with no base image");
+            bail!("invalid neon WAL redo request with no base image");
        }

        // Apply all the WAL records in the batch
@@ -12,6 +12,8 @@
 #include "fmgr.h"

 #include "miscadmin.h"
+#include "access/subtrans.h"
+#include "access/twophase.h"
 #include "access/xact.h"
 #include "access/xlog.h"
 #include "storage/buf_internals.h"
@@ -22,10 +24,12 @@
 #include "replication/logical.h"
 #include "replication/slot.h"
 #include "replication/walsender.h"
+#include "storage/proc.h"
 #include "storage/procsignal.h"
 #include "tcop/tcopprot.h"
 #include "funcapi.h"
 #include "access/htup_details.h"
+#include "utils/builtins.h"
 #include "utils/pg_lsn.h"
 #include "utils/guc.h"
 #include "utils/wait_event.h"
@@ -41,7 +45,6 @@ PG_MODULE_MAGIC;
 void		_PG_init(void);

 static int	logical_replication_max_snap_files = 300;
-bool primary_is_running = false;

 static void
 InitLogicalReplicationMonitor(void)
@@ -267,6 +270,293 @@ LogicalSlotsMonitorMain(Datum main_arg)
 	}
 }

+/*
+ * XXX: These private to procarray.c, but we need them here.
+ */
+#define PROCARRAY_MAXPROCS	(MaxBackends + max_prepared_xacts)
+#define TOTAL_MAX_CACHED_SUBXIDS \
+	((PGPROC_MAX_CACHED_SUBXIDS + 1) * PROCARRAY_MAXPROCS)
+
+/*
+ * Restore running-xact information by scanning the CLOG at startup.
+ *
+ * In PostgreSQL, a standby always has to wait for a running-xacts WAL record
+ * to arrive before it can start accepting queries. Furthermore, if there are
+ * transactions with too many subxids (> 64) open to fit in the in-memory
+ * subxids cache, the running-xacts record will be marked as "suboverflowed",
+ * and the standby will need to also wait for the currently in-progress
+ * transactions to finish.
+ *
+ * That's not great in PostgreSQL, because a hot standby does not necessary
+ * open up for queries immediately as you might expect. But it's worse in
+ * Neon: A standby in Neon doesn't need to start WAL replay from a checkpoint
+ * record; it can start at any LSN. Postgres arranges things so that there is
+ * a running-xacts record soon after every checkpoint record, but when you
+ * start from an arbitrary LSN, that doesn't help. If the primary is idle, or
+ * not running at all, it might never write a new running-xacts record,
+ * leaving the replica in a limbo where it can never start accepting queries.
+ *
+ * To mitigate that, we have an additional mechanism to find the running-xacts
+ * information: we scan the CLOG, making note of any XIDs not marked as
+ * committed or aborted. They are added to the Postgres known-assigned XIDs
+ * array by calling ProcArrayApplyRecoveryInfo() in the caller of this
+ * function.
+ *
+ * There is one big limitation with that mechanism: The size of the
+ * known-assigned XIDs is limited, so if there are a lot of in-progress XIDs,
+ * we have to give up. Furthermore, we don't know how many of the in-progress
+ * XIDs are subtransactions, and if we use up all the space in the
+ * known-assigned XIDs array for subtransactions, we might run out of space in
+ * the array later during WAL replay, causing the replica to shut down with
+ * "ERROR: too many KnownAssignedXids". The safe # of XIDs that we can add to
+ * the known-assigned array without risking that error later is very low,
+ * merely PGPROC_MAX_CACHED_SUBXIDS == 64, so we take our chances and use up
+ * to half of the known-assigned XIDs array for the subtransactions, even
+ * though that risks getting the error later.
+ *
+ * Note: It's OK if the recovered list of XIDs includes some transactions that
+ * have crashed in the primary, and hence will never commit. They will be seen
+ * as in-progress, until we see a new next running-acts record with an
+ * oldestActiveXid that invalidates them. That's how the known-assigned XIDs
+ * array always works.
+ *
+ * If scraping the CLOG doesn't succeed for some reason, like the subxid
+ * overflow, Postgres will fall back to waiting for a running-xacts record
+ * like usual.
+ *
+ * Returns true if a complete list of in-progress XIDs was scraped.
+ */
+static bool
+RestoreRunningXactsFromClog(CheckPoint *checkpoint, TransactionId **xids, int *nxids)
+{
+	TransactionId from;
+	TransactionId till;
+	int			max_xcnt;
+	TransactionId *prepared_xids = NULL;
+	int			n_prepared_xids;
+	TransactionId *restored_xids = NULL;
+	int			n_restored_xids;
+	int			next_prepared_idx;
+
+	Assert(*xids == NULL);
+
+	/*
+	 * If the checkpoint doesn't have a valid oldestActiveXid, bail out. We
+	 * don't know where to start the scan.
+	 *
+	 * This shouldn't happen, because the pageserver always maintains a valid
+	 * oldestActiveXid nowadays. Except when starting at an old point in time
+	 * that was ingested before the pageserver was taught to do that.
+	 */
+	if (!TransactionIdIsValid(checkpoint->oldestActiveXid))
+	{
+		elog(LOG, "cannot restore running-xacts from CLOG because oldestActiveXid is not set");
+		goto fail;
+	}
+
+	/*
+	 * We will scan the CLOG starting from the oldest active XID.
+	 *
+	 * In some corner cases, the oldestActiveXid from the last checkpoint
+	 * might already have been truncated from the CLOG. That is,
+	 * oldestActiveXid might be older than oldestXid. That's possible because
+	 * oldestActiveXid is only updated at checkpoints. After the last
+	 * checkpoint, the oldest transaction might have committed, and the CLOG
+	 * might also have been already truncated. So if oldestActiveXid is older
+	 * than oldestXid, start at oldestXid instead. (Otherwise we'd try to
+	 * access CLOG segments that have already been truncated away.)
+	 */
+	from = TransactionIdPrecedes(checkpoint->oldestXid, checkpoint->oldestActiveXid)
+		? checkpoint->oldestActiveXid : checkpoint->oldestXid;
+	till = XidFromFullTransactionId(checkpoint->nextXid);
+
+	/*
+	 * To avoid "too many KnownAssignedXids" error later during replay, we
+	 * limit number of collected transactions. This is a tradeoff: if we are
+	 * willing to consume more of the KnownAssignedXids space for the XIDs
+	 * now, that allows us to start up, but we might run out of space later.
+	 *
+	 * The size of the KnownAssignedXids array is TOTAL_MAX_CACHED_SUBXIDS,
+	 * which is (PGPROC_MAX_CACHED_SUBXIDS + 1) * PROCARRAY_MAXPROCS). In
+	 * PostgreSQL, that's always enough because the primary will always write
+	 * an XLOG_XACT_ASSIGNMENT record if a transaction has more than
+	 * PGPROC_MAX_CACHED_SUBXIDS subtransactions. Seeing that record allows
+	 * the standby to mark the XIDs in pg_subtrans and removing them from the
+	 * KnowingAssignedXids array.
+	 *
+	 * Here, we don't know which XIDs belong to subtransactions that have
+	 * already been WAL-logged with an XLOG_XACT_ASSIGNMENT record. If we
+	 * wanted to be totally safe and avoid the possibility of getting a "too
+	 * many KnownAssignedXids" error later, we would have to limit ourselves
+	 * to PGPROC_MAX_CACHED_SUBXIDS, which is not much. And that includes top
+	 * transaction IDs too, because we cannot distinguish between top
+	 * transaction IDs and subtransactions here.
+	 *
+	 * Somewhat arbitrarily, we use up to half of KnownAssignedXids. That
+	 * strikes a sensible balance between being useful, and risking a "too
+	 * many KnownAssignedXids" error later.
+	 */
+	max_xcnt = TOTAL_MAX_CACHED_SUBXIDS / 2;
+
+	/*
+	 * Collect XIDs of prepared transactions in an array. This includes only
+	 * their top-level XIDs. We assume that StandbyRecoverPreparedTransactions
+	 * has already been called, so we can find all the sub-transactions in
+	 * pg_subtrans.
+	 */
+	PrescanPreparedTransactions(&prepared_xids, &n_prepared_xids);
+	qsort(prepared_xids, n_prepared_xids, sizeof(TransactionId), xidLogicalComparator);
+
+	/*
+	 * Scan the CLOG, collecting in-progress XIDs into 'restored_xids'.
+	 */
+	elog(DEBUG1, "scanning CLOG between %u and %u for in-progress XIDs", from, till);
+	restored_xids = (TransactionId *) palloc(max_xcnt * sizeof(TransactionId));
+	n_restored_xids = 0;
+	next_prepared_idx = 0;
+	for (TransactionId xid = from; xid != till;)
+	{
+		XLogRecPtr	xidlsn;
+		XidStatus	xidstatus;
+
+		xidstatus = TransactionIdGetStatus(xid, &xidlsn);
+
+		/*
+		 * "Merge" the prepared transactions into the restored_xids array as
+		 * we go.  The prepared transactions array is sorted. This is mostly
+		 * a sanity check to ensure that all the prpeared transactions are
+		 * seen as in-progress. (There is a check after the loop that we didn't
+		 * miss any.)
+		 */
+		if (next_prepared_idx < n_prepared_xids && xid == prepared_xids[next_prepared_idx])
+		{
+			/*
+			 * This is a top-level transaction ID of a prepared transaction.
+			 * Include it in the array.
+			 */
+
+			/* sanity check */
+			if (xidstatus != TRANSACTION_STATUS_IN_PROGRESS)
+			{
+				elog(LOG, "prepared transaction %u has unexpected status %X, cannot restore running-xacts from CLOG",
+					 xid, xidstatus);
+				Assert(false);
+				goto fail;
+			}
+
+			elog(DEBUG1, "XID %u: was next prepared xact (%d / %d)", xid, next_prepared_idx, n_prepared_xids);
+			next_prepared_idx++;
+		}
+		else if (xidstatus == TRANSACTION_STATUS_COMMITTED)
+		{
+			elog(DEBUG1, "XID %u: was committed", xid);
+			goto skip;
+		}
+		else if (xidstatus == TRANSACTION_STATUS_ABORTED)
+		{
+			elog(DEBUG1, "XID %u: was aborted", xid);
+			goto skip;
+		}
+		else if (xidstatus == TRANSACTION_STATUS_IN_PROGRESS)
+		{
+			/*
+			 * In-progress transactions are included in the array.
+			 *
+			 * Except subtransactions of the prepared transactions. They are
+			 * already set in pg_subtrans, and hence don't need to be tracked
+			 * in the known-assigned XIDs array.
+			 */
+			if (n_prepared_xids > 0)
+			{
+				TransactionId parent = SubTransGetParent(xid);
+
+				if (TransactionIdIsValid(parent))
+				{
+					/*
+					 * This is a subtransaction belonging to a prepared
+					 * transaction.
+					 *
+					 * Sanity check that it is in the prepared XIDs array. It
+					 * should be, because StandbyRecoverPreparedTransactions
+					 * populated pg_subtrans, and no other XID should be set
+					 * in it yet. (This also relies on the fact that
+					 * StandbyRecoverPreparedTransactions sets the parent of
+					 * each subxid to point directly to the top-level XID,
+					 * rather than restoring the original subtransaction
+					 * hierarchy.)
+					 */
+					if (bsearch(&parent, prepared_xids, next_prepared_idx,
+								sizeof(TransactionId), xidLogicalComparator) == NULL)
+					{
+						elog(LOG, "sub-XID %u has unexpected parent %u, cannot restore running-xacts from CLOG",
+							 xid, parent);
+						Assert(false);
+						goto fail;
+					}
+					elog(DEBUG1, "XID %u: was a subtransaction of prepared xid %u", xid, parent);
+					goto skip;
+				}
+			}
+
+			/* include it in the array */
+			elog(DEBUG1, "XID %u: is in progress", xid);
+		}
+		else
+		{
+			/*
+			 * SUB_COMMITTED is a transient state used at commit. We don't
+			 * expect to see that here.
+			 */
+			elog(LOG, "XID %u has unexpected status %X in pg_xact, cannot restore running-xacts from CLOG",
+				 xid, xidstatus);
+			Assert(false);
+			goto fail;
+		}
+
+		if (n_restored_xids >= max_xcnt)
+		{
+			/*
+			 * Overflowed. We won't be able to install the RunningTransactions
+			 * snapshot.
+			 */
+			elog(LOG, "too many running xacts to restore from the CLOG; oldestXid=%u oldestActiveXid=%u nextXid %u",
+				 checkpoint->oldestXid, checkpoint->oldestActiveXid,
+				 XidFromFullTransactionId(checkpoint->nextXid));
+			goto fail;
+		}
+
+		restored_xids[n_restored_xids++] = xid;
+
+	skip:
+		TransactionIdAdvance(xid);
+		continue;
+	}
+
+	/* sanity check */
+	if (next_prepared_idx != n_prepared_xids)
+	{
+		elog(LOG, "prepared transaction ID %u was not visited in the CLOG scan, cannot restore running-xacts from CLOG",
+			 prepared_xids[next_prepared_idx]);
+		Assert(false);
+		goto fail;
+	}
+
+	elog(LOG, "restored %d running xacts by scanning the CLOG; oldestXid=%u oldestActiveXid=%u nextXid %u",
+		 n_restored_xids, checkpoint->oldestXid, checkpoint->oldestActiveXid, XidFromFullTransactionId(checkpoint->nextXid));
+	*nxids = n_restored_xids;
+	*xids = restored_xids;
+	return true;
+
+ fail:
+	*nxids = 0;
+	*xids = NULL;
+	if (restored_xids)
+		pfree(restored_xids);
+	if (prepared_xids)
+		pfree(prepared_xids);
+	return false;
+}
+
 void
 _PG_init(void)
 {
@@ -289,15 +579,8 @@ _PG_init(void)

 	pg_init_extension_server();

-	DefineCustomBoolVariable(
-		"neon.primary_is_running",
-		"true if the primary was running at replica startup. false otherwise",
-		NULL,
-		&primary_is_running,
-		false,
-		PGC_POSTMASTER,
-		0,
-		NULL, NULL, NULL);
+	restore_running_xacts_callback = RestoreRunningXactsFromClog;
+
 	/*
 	 * Important: This must happen after other parts of the extension are
 	 * loaded, otherwise any settings to GUCs that were set before the
@@ -1447,7 +1447,7 @@ RecvAppendResponses(Safekeeper *sk)
 			 * core as this is kinda expected scenario.
 			 */
 			disable_core_dump();
-			wp_log(PANIC, "WAL acceptor %s:%s with term " INT64_FORMAT " rejected our request, our term " INT64_FORMAT "",
+			wp_log(PANIC, "WAL acceptor %s:%s with term " INT64_FORMAT " rejected our request, our term " INT64_FORMAT ", meaning another compute is running at the same time, and it conflicts with us",
 				   sk->host, sk->port,
 				   sk->appendResponse.term, wp->propTerm);
 		}
@@ -63,6 +63,8 @@ char	   *wal_acceptors_list = "";
 int			wal_acceptor_reconnect_timeout = 1000;
 int			wal_acceptor_connection_timeout = 10000;

+/* Set to true in the walproposer bgw. */
+static bool am_walproposer;
 static WalproposerShmemState *walprop_shared;
 static WalProposerConfig walprop_config;
 static XLogRecPtr sentPtr = InvalidXLogRecPtr;
@@ -76,6 +78,7 @@ static HotStandbyFeedback agg_hs_feedback;

 static void nwp_shmem_startup_hook(void);
 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 bool backpressure_throttling_impl(void);
@@ -111,7 +114,8 @@ init_walprop_config(bool syncSafekeepers)
 {
 	walprop_config.neon_tenant = neon_tenant;
 	walprop_config.neon_timeline = neon_timeline;
-	walprop_config.safekeepers_list = wal_acceptors_list;
+	/* WalProposerCreate scribbles directly on it, so pstrdup */
+	walprop_config.safekeepers_list = pstrdup(wal_acceptors_list);
 	walprop_config.safekeeper_reconnect_timeout = wal_acceptor_reconnect_timeout;
 	walprop_config.safekeeper_connection_timeout = wal_acceptor_connection_timeout;
 	walprop_config.wal_segment_size = wal_segment_size;
@@ -151,6 +155,7 @@ WalProposerMain(Datum main_arg)

 	init_walprop_config(false);
 	walprop_pg_init_bgworker();
+	am_walproposer = true;
 	walprop_pg_load_libpqwalreceiver();

 	wp = WalProposerCreate(&walprop_config, walprop_pg);
@@ -189,10 +194,10 @@ nwp_register_gucs(void)
 							   NULL,	/* long_desc */
 							   &wal_acceptors_list, /* valueAddr */
 							   "",	/* bootValue */
-							   PGC_POSTMASTER,
+							   PGC_SIGHUP,
 							   GUC_LIST_INPUT,	/* extensions can't use*
 												 * GUC_LIST_QUOTE */
-							   NULL, NULL, NULL);
+							   NULL, assign_neon_safekeepers, NULL);

 	DefineCustomIntVariable(
 							"neon.safekeeper_reconnect_timeout",
@@ -215,6 +220,33 @@ nwp_register_gucs(void)
 							NULL, NULL, NULL);
 }

+/*
+ * GUC assign_hook for neon.safekeepers. Restarts walproposer through FATAL if
+ * the list changed.
+ */
+static void
+assign_neon_safekeepers(const char *newval, void *extra)
+{
+	if (!am_walproposer)
+		return;
+
+	if (!newval) {
+		/* should never happen */
+		wpg_log(FATAL, "neon.safekeepers is empty");
+	}
+
+	/* 
+	 * TODO: restarting through FATAL is stupid and introduces 1s delay before
+	 * next bgw start. We should refactor walproposer to allow graceful exit and
+	 * thus remove this delay.
+	 */
+	if (strcmp(wal_acceptors_list, newval) != 0)
+	{
+		wpg_log(FATAL, "restarting walproposer to change safekeeper list from %s to %s",
+				wal_acceptors_list, newval);
+	}
+}
+
 /*  Check if we need to suspend inserts because of lagging replication. */
 static uint64
 backpressure_lag_impl(void)
@@ -363,7 +395,7 @@ walprop_register_bgworker(void)
 	snprintf(bgw.bgw_function_name, BGW_MAXLEN, "WalProposerMain");
 	snprintf(bgw.bgw_name, BGW_MAXLEN, "WAL proposer");
 	snprintf(bgw.bgw_type, BGW_MAXLEN, "WAL proposer");
-	bgw.bgw_restart_time = 5;
+	bgw.bgw_restart_time = 1;
 	bgw.bgw_notify_pid = 0;
 	bgw.bgw_main_arg = (Datum) 0;

@@ -1639,6 +1671,18 @@ walprop_pg_wait_event_set(WalProposer *wp, long timeout, Safekeeper **sk, uint32
 		late_cv_trigger = ConditionVariableCancelSleep();
 #endif

+	/*
+	 * Process config if requested. This restarts walproposer if safekeepers
+	 * list changed. Don't do that for sync-safekeepers because quite probably
+	 * it (re-reading config) won't work without some effort, and
+	 * sync-safekeepers should be quick to finish anyway.
+	 */
+	if (!wp->config->syncSafekeepers && ConfigReloadPending)
+	{
+		ConfigReloadPending = false;
+		ProcessConfigFile(PGC_SIGHUP);
+	}
+
 	/*
 	 * If wait is terminated by latch set (walsenders' latch is set on each
 	 * wal flush). (no need for pm death check due to WL_EXIT_ON_PM_DEATH)
@@ -7,7 +7,7 @@ OBJS = \
 	neontest.o

 EXTENSION = neon_test_utils
-DATA = neon_test_utils--1.1.sql
+DATA = neon_test_utils--1.2.sql
 PGFILEDESC = "neon_test_utils - helpers for neon testing and debugging"

 PG_CONFIG = pg_config
@@ -41,7 +41,7 @@ RETURNS bytea
 AS 'MODULE_PATHNAME', 'get_raw_page_at_lsn_ex'
 LANGUAGE C PARALLEL UNSAFE;

-CREATE FUNCTION neon_xlogflush(lsn pg_lsn)
+CREATE FUNCTION neon_xlogflush(lsn pg_lsn DEFAULT NULL)
 RETURNS VOID
 AS 'MODULE_PATHNAME', 'neon_xlogflush'
 LANGUAGE C PARALLEL UNSAFE;
@@ -1,6 +1,6 @@
 # neon_test_utils extension
 comment = 'helpers for neon testing and debugging'
-default_version = '1.1'
+default_version = '1.2'
 module_pathname = '$libdir/neon_test_utils'
 relocatable = true
 trusted = true
@@ -15,6 +15,7 @@
 #include "access/relation.h"
 #include "access/xact.h"
 #include "access/xlog.h"
+#include "access/xlog_internal.h"
 #include "catalog/namespace.h"
 #include "fmgr.h"
 #include "funcapi.h"
@@ -444,11 +445,46 @@ get_raw_page_at_lsn_ex(PG_FUNCTION_ARGS)

 /*
 * Directly calls XLogFlush(lsn) to flush WAL buffers.
+ *
+ * If 'lsn' is not specified (is NULL), flush all generated WAL.
 */
 Datum
 neon_xlogflush(PG_FUNCTION_ARGS)
 {
-	XLogRecPtr	lsn = PG_GETARG_LSN(0);
+	XLogRecPtr	lsn;
+
+	if (RecoveryInProgress())
+		ereport(ERROR,
+				(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+				 errmsg("recovery is in progress"),
+				 errhint("cannot flush WAL during recovery.")));
+
+	if (!PG_ARGISNULL(0))
+		lsn = PG_GETARG_LSN(0);
+	else
+	{
+		lsn = GetXLogInsertRecPtr();
+
+		/*---
+		 * The LSN returned by GetXLogInsertRecPtr() is the position where the
+		 * next inserted record would begin. If the last record ended just at
+		 * the page boundary, the next record will begin after the page header
+		 * on the next page, and that's what GetXLogInsertRecPtr().returns,
+		 * but the page header has not been written yet. If we tried to flush
+		 * it, XLogFlush() would throw an error:
+		 *
+		 * ERROR : xlog flush request %X/%X is not satisfied --- flushed only to %X/%X
+		 *
+		 * To avoid that, if the insert position points to just after the page
+		 * header, back off to page boundary.
+		 */
+		if (lsn % XLOG_BLCKSZ == SizeOfXLogShortPHD &&
+			XLogSegmentOffset(lsn, wal_segment_size) > XLOG_BLCKSZ)
+			lsn -= SizeOfXLogShortPHD;
+		else if (lsn % XLOG_BLCKSZ == SizeOfXLogLongPHD &&
+				 XLogSegmentOffset(lsn, wal_segment_size) < XLOG_BLCKSZ)
+			lsn -= SizeOfXLogLongPHD;
+	}

 	XLogFlush(lsn);
 	PG_RETURN_VOID();
@@ -168,16 +168,15 @@ close_range_syscall(unsigned int start_fd, unsigned int count, unsigned int flag
 static void
 enter_seccomp_mode(void)
 {
-
 	/*
 	 * The pageserver process relies on us to close all the file descriptors
 	 * it potentially leaked to us, _before_ we start processing potentially dangerous
 	 * wal records. See the comment in the Rust code that launches this process.
 	 */
-	int err;
-	if (err = close_range_syscall(3, ~0U, 0)) {
-		ereport(FATAL, (errcode(ERRCODE_SYSTEM_ERROR), errmsg("seccomp: could not close files >= fd 3")));
-	}
+	if (close_range_syscall(3, ~0U, 0) != 0)
+		ereport(FATAL,
+				(errcode(ERRCODE_SYSTEM_ERROR),
+				 errmsg("seccomp: could not close files >= fd 3")));

 	PgSeccompRule syscalls[] =
 	{
@@ -10,7 +10,7 @@ use itertools::Itertools;
 use proxy::config::TlsServerEndPoint;
 use proxy::context::RequestMonitoring;
 use proxy::metrics::{Metrics, ThreadPoolMetrics};
-use proxy::proxy::{copy_bidirectional_client_compute, run_until_cancelled};
+use proxy::proxy::{copy_bidirectional_client_compute, run_until_cancelled, ErrorSource};
 use rustls::pki_types::PrivateKeyDer;
 use tokio::net::TcpListener;

@@ -286,7 +286,10 @@ async fn handle_client(

    // Starting from here we only proxy the client's traffic.
    info!("performing the proxy pass...");
-    let _ = copy_bidirectional_client_compute(&mut tls_stream, &mut client).await?;

-    Ok(())
+    match copy_bidirectional_client_compute(&mut tls_stream, &mut client).await {
+        Ok(_) => Ok(()),
+        Err(ErrorSource::Client(err)) => Err(err).context("client"),
+        Err(ErrorSource::Compute(err)) => Err(err).context("compute"),
+    }
 }
@@ -5,7 +5,7 @@ use std::fmt::{self, Display};
 use crate::auth::IpPattern;

 use crate::intern::{BranchIdInt, EndpointIdInt, ProjectIdInt};
-use crate::proxy::retry::ShouldRetry;
+use crate::proxy::retry::CouldRetry;

 /// Generic error response with human-readable description.
 /// Note that we can't always present it to user as is.
@@ -64,45 +64,47 @@ impl Display for ConsoleError {
    }
 }

-impl ShouldRetry for ConsoleError {
+impl CouldRetry for ConsoleError {
    fn could_retry(&self) -> bool {
-        if self.status.is_none() || self.status.as_ref().unwrap().details.retry_info.is_none() {
-            // retry some temporary failures because the compute was in a bad state
-            // (bad request can be returned when the endpoint was in transition)
-            return match &self {
-                ConsoleError {
-                    http_status_code: http::StatusCode::BAD_REQUEST,
-                    ..
-                } => true,
-                // don't retry when quotas are exceeded
-                ConsoleError {
-                    http_status_code: http::StatusCode::UNPROCESSABLE_ENTITY,
-                    ref error,
-                    ..
-                } => !error.contains("compute time quota of non-primary branches is exceeded"),
-                // locked can be returned when the endpoint was in transition
-                // or when quotas are exceeded. don't retry when quotas are exceeded
-                ConsoleError {
-                    http_status_code: http::StatusCode::LOCKED,
-                    ref error,
-                    ..
-                } => {
-                    !error.contains("quota exceeded")
-                        && !error.contains("the limit for current plan reached")
-                }
-                _ => false,
-            };
+        // If the error message does not have a status,
+        // the error is unknown and probably should not retry automatically
+        let Some(status) = &self.status else {
+            return false;
+        };
+
+        // retry if the retry info is set.
+        if status.details.retry_info.is_some() {
+            return true;
        }

-        // retry if the response has a retry delay
-        if let Some(retry_info) = self
-            .status
-            .as_ref()
-            .and_then(|s| s.details.retry_info.as_ref())
-        {
-            retry_info.retry_delay_ms > 0
-        } else {
-            false
+        // if no retry info set, attempt to use the error code to guess the retry state.
+        let reason = status
+            .details
+            .error_info
+            .map_or(Reason::Unknown, |e| e.reason);
+        match reason {
+            // not a transitive error
+            Reason::RoleProtected => false,
+            // on retry, it will still not be found
+            Reason::ResourceNotFound
+            | Reason::ProjectNotFound
+            | Reason::EndpointNotFound
+            | Reason::BranchNotFound => false,
+            // we were asked to go away
+            Reason::RateLimitExceeded
+            | Reason::NonDefaultBranchComputeTimeExceeded
+            | Reason::ActiveTimeQuotaExceeded
+            | Reason::ComputeTimeQuotaExceeded
+            | Reason::WrittenDataQuotaExceeded
+            | Reason::DataTransferQuotaExceeded
+            | Reason::LogicalSizeQuotaExceeded => false,
+            // transitive error. control plane is currently busy
+            // but might be ready soon
+            Reason::RunningOperations => true,
+            Reason::ConcurrencyLimitReached => true,
+            Reason::LockAlreadyTaken => true,
+            // unknown error. better not retry it.
+            Reason::Unknown => false,
        }
    }
 }
@@ -121,7 +123,7 @@ pub struct Details {
    pub user_facing_message: Option<UserFacingMessage>,
 }

-#[derive(Debug, Deserialize)]
+#[derive(Copy, Clone, Debug, Deserialize)]
 pub struct ErrorInfo {
    pub reason: Reason,
    // Schema could also have `metadata` field, but it's not structured. Skip it for now.
@@ -129,30 +131,59 @@ pub struct ErrorInfo {

 #[derive(Clone, Copy, Debug, Deserialize, Default)]
 pub enum Reason {
+    /// RoleProtected indicates that the role is protected and the attempted operation is not permitted on protected roles.
    #[serde(rename = "ROLE_PROTECTED")]
    RoleProtected,
+    /// ResourceNotFound indicates that a resource (project, endpoint, branch, etc.) wasn't found,
+    /// usually due to the provided ID not being correct or because the subject doesn't have enough permissions to
+    /// access the requested resource.
+    /// Prefer a more specific reason if possible, e.g., ProjectNotFound, EndpointNotFound, etc.
    #[serde(rename = "RESOURCE_NOT_FOUND")]
    ResourceNotFound,
+    /// ProjectNotFound indicates that the project wasn't found, usually due to the provided ID not being correct,
+    /// or that the subject doesn't have enough permissions to access the requested project.
    #[serde(rename = "PROJECT_NOT_FOUND")]
    ProjectNotFound,
+    /// EndpointNotFound indicates that the endpoint wasn't found, usually due to the provided ID not being correct,
+    /// or that the subject doesn't have enough permissions to access the requested endpoint.
    #[serde(rename = "ENDPOINT_NOT_FOUND")]
    EndpointNotFound,
+    /// BranchNotFound indicates that the branch wasn't found, usually due to the provided ID not being correct,
+    /// or that the subject doesn't have enough permissions to access the requested branch.
    #[serde(rename = "BRANCH_NOT_FOUND")]
    BranchNotFound,
+    /// RateLimitExceeded indicates that the rate limit for the operation has been exceeded.
    #[serde(rename = "RATE_LIMIT_EXCEEDED")]
    RateLimitExceeded,
+    /// NonDefaultBranchComputeTimeExceeded indicates that the compute time quota of non-default branches has been
+    /// exceeded.
    #[serde(rename = "NON_PRIMARY_BRANCH_COMPUTE_TIME_EXCEEDED")]
-    NonPrimaryBranchComputeTimeExceeded,
+    NonDefaultBranchComputeTimeExceeded,
+    /// ActiveTimeQuotaExceeded indicates that the active time quota was exceeded.
    #[serde(rename = "ACTIVE_TIME_QUOTA_EXCEEDED")]
    ActiveTimeQuotaExceeded,
+    /// ComputeTimeQuotaExceeded indicates that the compute time quota was exceeded.
    #[serde(rename = "COMPUTE_TIME_QUOTA_EXCEEDED")]
    ComputeTimeQuotaExceeded,
+    /// WrittenDataQuotaExceeded indicates that the written data quota was exceeded.
    #[serde(rename = "WRITTEN_DATA_QUOTA_EXCEEDED")]
    WrittenDataQuotaExceeded,
+    /// DataTransferQuotaExceeded indicates that the data transfer quota was exceeded.
    #[serde(rename = "DATA_TRANSFER_QUOTA_EXCEEDED")]
    DataTransferQuotaExceeded,
+    /// LogicalSizeQuotaExceeded indicates that the logical size quota was exceeded.
    #[serde(rename = "LOGICAL_SIZE_QUOTA_EXCEEDED")]
    LogicalSizeQuotaExceeded,
+    /// RunningOperations indicates that the project already has some running operations
+    /// and scheduling of new ones is prohibited.
+    #[serde(rename = "RUNNING_OPERATIONS")]
+    RunningOperations,
+    /// ConcurrencyLimitReached indicates that the concurrency limit for an action was reached.
+    #[serde(rename = "CONCURRENCY_LIMIT_REACHED")]
+    ConcurrencyLimitReached,
+    /// LockAlreadyTaken indicates that the we attempted to take a lock that was already taken.
+    #[serde(rename = "LOCK_ALREADY_TAKEN")]
+    LockAlreadyTaken,
    #[default]
    #[serde(other)]
    Unknown,
@@ -170,7 +201,7 @@ impl Reason {
    }
 }

-#[derive(Debug, Deserialize)]
+#[derive(Copy, Clone, Debug, Deserialize)]
 pub struct RetryInfo {
    pub retry_delay_ms: u64,
 }
@@ -25,9 +25,9 @@ use tracing::info;

 pub mod errors {
    use crate::{
-        console::messages::{self, ConsoleError},
+        console::messages::{self, ConsoleError, Reason},
        error::{io_error, ReportableError, UserFacingError},
-        proxy::retry::ShouldRetry,
+        proxy::retry::CouldRetry,
    };
    use thiserror::Error;

@@ -76,21 +76,22 @@ pub mod errors {
                ApiError::Console(e) => {
                    use crate::error::ErrorKind::*;
                    match e.get_reason() {
-                        crate::console::messages::Reason::RoleProtected => User,
-                        crate::console::messages::Reason::ResourceNotFound => User,
-                        crate::console::messages::Reason::ProjectNotFound => User,
-                        crate::console::messages::Reason::EndpointNotFound => User,
-                        crate::console::messages::Reason::BranchNotFound => User,
-                        crate::console::messages::Reason::RateLimitExceeded => ServiceRateLimit,
-                        crate::console::messages::Reason::NonPrimaryBranchComputeTimeExceeded => {
-                            User
-                        }
-                        crate::console::messages::Reason::ActiveTimeQuotaExceeded => User,
-                        crate::console::messages::Reason::ComputeTimeQuotaExceeded => User,
-                        crate::console::messages::Reason::WrittenDataQuotaExceeded => User,
-                        crate::console::messages::Reason::DataTransferQuotaExceeded => User,
-                        crate::console::messages::Reason::LogicalSizeQuotaExceeded => User,
-                        crate::console::messages::Reason::Unknown => match &e {
+                        Reason::RoleProtected => User,
+                        Reason::ResourceNotFound => User,
+                        Reason::ProjectNotFound => User,
+                        Reason::EndpointNotFound => User,
+                        Reason::BranchNotFound => User,
+                        Reason::RateLimitExceeded => ServiceRateLimit,
+                        Reason::NonDefaultBranchComputeTimeExceeded => User,
+                        Reason::ActiveTimeQuotaExceeded => User,
+                        Reason::ComputeTimeQuotaExceeded => User,
+                        Reason::WrittenDataQuotaExceeded => User,
+                        Reason::DataTransferQuotaExceeded => User,
+                        Reason::LogicalSizeQuotaExceeded => User,
+                        Reason::ConcurrencyLimitReached => ControlPlane,
+                        Reason::LockAlreadyTaken => ControlPlane,
+                        Reason::RunningOperations => ControlPlane,
+                        Reason::Unknown => match &e {
                            ConsoleError {
                                http_status_code:
                                    http::StatusCode::NOT_FOUND | http::StatusCode::NOT_ACCEPTABLE,
@@ -128,7 +129,7 @@ pub mod errors {
        }
    }

-    impl ShouldRetry for ApiError {
+    impl CouldRetry for ApiError {
        fn could_retry(&self) -> bool {
            match self {
                // retry some transport errors
@@ -239,6 +240,17 @@ pub mod errors {
            }
        }
    }
+
+    impl CouldRetry for WakeComputeError {
+        fn could_retry(&self) -> bool {
+            match self {
+                WakeComputeError::BadComputeAddress(_) => false,
+                WakeComputeError::ApiError(e) => e.could_retry(),
+                WakeComputeError::TooManyConnections => false,
+                WakeComputeError::TooManyConnectionAttempts(_) => false,
+            }
+        }
+    }
 }

 /// Auth secret which is managed by the cloud.
@@ -8,6 +8,7 @@ pub mod passthrough;
 pub mod retry;
 pub mod wake_compute;
 pub use copy_bidirectional::copy_bidirectional_client_compute;
+pub use copy_bidirectional::ErrorSource;

 use crate::{
    auth,
@@ -148,8 +149,11 @@ pub async fn task_main(
                    ctx.log_connect();
                    match p.proxy_pass().instrument(span.clone()).await {
                        Ok(()) => {}
-                        Err(e) => {
-                            error!(parent: &span, "per-client task finished with an error: {e:#}");
+                        Err(ErrorSource::Client(e)) => {
+                            error!(parent: &span, "per-client task finished with an IO error from the client: {e:#}");
+                        }
+                        Err(ErrorSource::Compute(e)) => {
+                            error!(parent: &span, "per-client task finished with an IO error from the compute: {e:#}");
                        }
                    }
                }
@@ -7,7 +7,7 @@ use crate::{
    error::ReportableError,
    metrics::{ConnectOutcome, ConnectionFailureKind, Metrics, RetriesMetricGroup, RetryType},
    proxy::{
-        retry::{retry_after, ShouldRetry},
+        retry::{retry_after, should_retry, CouldRetry},
        wake_compute::wake_compute,
    },
    Host,
@@ -17,6 +17,8 @@ use pq_proto::StartupMessageParams;
 use tokio::time;
 use tracing::{error, info, warn};

+use super::retry::ShouldRetryWakeCompute;
+
 const CONNECT_TIMEOUT: time::Duration = time::Duration::from_secs(2);

 /// If we couldn't connect, a cached connection info might be to blame
@@ -104,7 +106,7 @@ pub async fn connect_to_compute<M: ConnectMechanism, B: ComputeConnectBackend>(
    connect_to_compute_retry_config: RetryConfig,
 ) -> Result<M::Connection, M::Error>
 where
-    M::ConnectError: ShouldRetry + std::fmt::Debug,
+    M::ConnectError: CouldRetry + ShouldRetryWakeCompute + std::fmt::Debug,
    M::Error: From<WakeComputeError>,
 {
    let mut num_retries = 0;
@@ -139,10 +141,10 @@ where

    error!(error = ?err, "could not connect to compute node");

-    let node_info = if !node_info.cached() || !err.should_retry_database_address() {
+    let node_info = if !node_info.cached() || !err.should_retry_wake_compute() {
        // If we just recieved this from cplane and dodn't get it from cache, we shouldn't retry.
        // Do not need to retrieve a new node_info, just return the old one.
-        if !err.should_retry(num_retries, connect_to_compute_retry_config) {
+        if should_retry(&err, num_retries, connect_to_compute_retry_config) {
            Metrics::get().proxy.retries_metric.observe(
                RetriesMetricGroup {
                    outcome: ConnectOutcome::Failed,
@@ -188,9 +190,8 @@ where
                return Ok(res);
            }
            Err(e) => {
-                let retriable = e.should_retry(num_retries, connect_to_compute_retry_config);
-                if !retriable {
-                    error!(error = ?e, num_retries, retriable, "couldn't connect to compute node");
+                if !should_retry(&e, num_retries, connect_to_compute_retry_config) {
+                    error!(error = ?e, num_retries, retriable = false, "couldn't connect to compute node");
                    Metrics::get().proxy.retries_metric.observe(
                        RetriesMetricGroup {
                            outcome: ConnectOutcome::Failed,
@@ -200,9 +201,10 @@ where
                    );
                    return Err(e.into());
                }
-                warn!(error = ?e, num_retries, retriable, "couldn't connect to compute node");
+
+                warn!(error = ?e, num_retries, retriable = true, "couldn't connect to compute node");
            }
-        }
+        };

        let wait_duration = retry_after(num_retries, connect_to_compute_retry_config);
        num_retries += 1;
@@ -13,12 +13,39 @@ enum TransferState {
    Done(u64),
 }

+#[derive(Debug)]
+pub enum ErrorDirection {
+    Read(io::Error),
+    Write(io::Error),
+}
+
+impl ErrorSource {
+    fn from_client(err: ErrorDirection) -> ErrorSource {
+        match err {
+            ErrorDirection::Read(client) => Self::Client(client),
+            ErrorDirection::Write(compute) => Self::Compute(compute),
+        }
+    }
+    fn from_compute(err: ErrorDirection) -> ErrorSource {
+        match err {
+            ErrorDirection::Write(client) => Self::Client(client),
+            ErrorDirection::Read(compute) => Self::Compute(compute),
+        }
+    }
+}
+
+#[derive(Debug)]
+pub enum ErrorSource {
+    Client(io::Error),
+    Compute(io::Error),
+}
+
 fn transfer_one_direction<A, B>(
    cx: &mut Context<'_>,
    state: &mut TransferState,
    r: &mut A,
    w: &mut B,
-) -> Poll<io::Result<u64>>
+) -> Poll<Result<u64, ErrorDirection>>
 where
    A: AsyncRead + AsyncWrite + Unpin + ?Sized,
    B: AsyncRead + AsyncWrite + Unpin + ?Sized,
@@ -32,7 +59,7 @@ where
                *state = TransferState::ShuttingDown(count);
            }
            TransferState::ShuttingDown(count) => {
-                ready!(w.as_mut().poll_shutdown(cx))?;
+                ready!(w.as_mut().poll_shutdown(cx)).map_err(ErrorDirection::Write)?;
                *state = TransferState::Done(*count);
            }
            TransferState::Done(count) => return Poll::Ready(Ok(*count)),
@@ -44,7 +71,7 @@ where
 pub async fn copy_bidirectional_client_compute<Client, Compute>(
    client: &mut Client,
    compute: &mut Compute,
-) -> Result<(u64, u64), std::io::Error>
+) -> Result<(u64, u64), ErrorSource>
 where
    Client: AsyncRead + AsyncWrite + Unpin + ?Sized,
    Compute: AsyncRead + AsyncWrite + Unpin + ?Sized,
@@ -54,9 +81,11 @@ where

    poll_fn(|cx| {
        let mut client_to_compute_result =
-            transfer_one_direction(cx, &mut client_to_compute, client, compute)?;
+            transfer_one_direction(cx, &mut client_to_compute, client, compute)
+                .map_err(ErrorSource::from_client)?;
        let mut compute_to_client_result =
-            transfer_one_direction(cx, &mut compute_to_client, compute, client)?;
+            transfer_one_direction(cx, &mut compute_to_client, compute, client)
+                .map_err(ErrorSource::from_compute)?;

        // Early termination checks from compute to client.
        if let TransferState::Done(_) = compute_to_client {
@@ -65,18 +94,20 @@ where
                // Initiate shutdown
                client_to_compute = TransferState::ShuttingDown(buf.amt);
                client_to_compute_result =
-                    transfer_one_direction(cx, &mut client_to_compute, client, compute)?;
+                    transfer_one_direction(cx, &mut client_to_compute, client, compute)
+                        .map_err(ErrorSource::from_client)?;
            }
        }

-        // Early termination checks from compute to client.
+        // Early termination checks from client to compute.
        if let TransferState::Done(_) = client_to_compute {
            if let TransferState::Running(buf) = &compute_to_client {
                info!("Client is done, terminate compute");
                // Initiate shutdown
                compute_to_client = TransferState::ShuttingDown(buf.amt);
                compute_to_client_result =
-                    transfer_one_direction(cx, &mut compute_to_client, client, compute)?;
+                    transfer_one_direction(cx, &mut compute_to_client, compute, client)
+                        .map_err(ErrorSource::from_compute)?;
            }
        }

@@ -138,7 +169,7 @@ impl CopyBuffer {
        cx: &mut Context<'_>,
        mut reader: Pin<&mut R>,
        mut writer: Pin<&mut W>,
-    ) -> Poll<io::Result<usize>>
+    ) -> Poll<Result<usize, ErrorDirection>>
    where
        R: AsyncRead + ?Sized,
        W: AsyncWrite + ?Sized,
@@ -149,11 +180,11 @@ impl CopyBuffer {
                // Top up the buffer towards full if we can read a bit more
                // data - this should improve the chances of a large write
                if !me.read_done && me.cap < me.buf.len() {
-                    ready!(me.poll_fill_buf(cx, reader.as_mut()))?;
+                    ready!(me.poll_fill_buf(cx, reader.as_mut())).map_err(ErrorDirection::Read)?;
                }
                Poll::Pending
            }
-            res => res,
+            res => res.map_err(ErrorDirection::Write),
        }
    }

@@ -162,7 +193,7 @@ impl CopyBuffer {
        cx: &mut Context<'_>,
        mut reader: Pin<&mut R>,
        mut writer: Pin<&mut W>,
-    ) -> Poll<io::Result<u64>>
+    ) -> Poll<Result<u64, ErrorDirection>>
    where
        R: AsyncRead + ?Sized,
        W: AsyncWrite + ?Sized,
@@ -176,12 +207,13 @@ impl CopyBuffer {

                match self.poll_fill_buf(cx, reader.as_mut()) {
                    Poll::Ready(Ok(())) => (),
-                    Poll::Ready(Err(err)) => return Poll::Ready(Err(err)),
+                    Poll::Ready(Err(err)) => return Poll::Ready(Err(ErrorDirection::Read(err))),
                    Poll::Pending => {
                        // Try flushing when the reader has no progress to avoid deadlock
                        // when the reader depends on buffered writer.
                        if self.need_flush {
-                            ready!(writer.as_mut().poll_flush(cx))?;
+                            ready!(writer.as_mut().poll_flush(cx))
+                                .map_err(ErrorDirection::Write)?;
                            self.need_flush = false;
                        }

@@ -194,10 +226,10 @@ impl CopyBuffer {
            while self.pos < self.cap {
                let i = ready!(self.poll_write_buf(cx, reader.as_mut(), writer.as_mut()))?;
                if i == 0 {
-                    return Poll::Ready(Err(io::Error::new(
+                    return Poll::Ready(Err(ErrorDirection::Write(io::Error::new(
                        io::ErrorKind::WriteZero,
                        "write zero byte into writer",
-                    )));
+                    ))));
                } else {
                    self.pos += i;
                    self.amt += i as u64;
@@ -216,7 +248,7 @@ impl CopyBuffer {
            // If we've written all the data and we've seen EOF, flush out the
            // data and finish the transfer.
            if self.pos == self.cap && self.read_done {
-                ready!(writer.as_mut().poll_flush(cx))?;
+                ready!(writer.as_mut().poll_flush(cx)).map_err(ErrorDirection::Write)?;
                return Poll::Ready(Ok(self.amt));
            }
        }
@@ -10,13 +10,15 @@ use tokio::io::{AsyncRead, AsyncWrite};
 use tracing::info;
 use utils::measured_stream::MeasuredStream;

+use super::copy_bidirectional::ErrorSource;
+
 /// Forward bytes in both directions (client <-> compute).
 #[tracing::instrument(skip_all)]
 pub async fn proxy_pass(
    client: impl AsyncRead + AsyncWrite + Unpin,
    compute: impl AsyncRead + AsyncWrite + Unpin,
    aux: MetricsAuxInfo,
-) -> anyhow::Result<()> {
+) -> Result<(), ErrorSource> {
    let usage = USAGE_METRICS.register(Ids {
        endpoint_id: aux.endpoint_id,
        branch_id: aux.branch_id,
@@ -66,9 +68,11 @@ pub struct ProxyPassthrough<P, S> {
 }

 impl<P, S: AsyncRead + AsyncWrite + Unpin> ProxyPassthrough<P, S> {
-    pub async fn proxy_pass(self) -> anyhow::Result<()> {
+    pub async fn proxy_pass(self) -> Result<(), ErrorSource> {
        let res = proxy_pass(self.client, self.compute.stream, self.aux).await;
-        self.compute.cancel_closure.try_cancel_query().await?;
+        if let Err(err) = self.compute.cancel_closure.try_cancel_query().await {
+            tracing::error!(?err, "could not cancel the query in the database");
+        }
        res
    }
 }
@@ -2,20 +2,22 @@ use crate::{compute, config::RetryConfig};
 use std::{error::Error, io};
 use tokio::time;

-pub trait ShouldRetry {
+pub trait CouldRetry {
+    /// Returns true if the error could be retried
    fn could_retry(&self) -> bool;
-    fn should_retry(&self, num_retries: u32, config: RetryConfig) -> bool {
-        match self {
-            _ if num_retries >= config.max_retries => false,
-            err => err.could_retry(),
-        }
-    }
-    fn should_retry_database_address(&self) -> bool {
-        true
-    }
 }

-impl ShouldRetry for io::Error {
+pub trait ShouldRetryWakeCompute {
+    /// Returns true if we need to invalidate the cache for this node.
+    /// If false, we can continue retrying with the current node cache.
+    fn should_retry_wake_compute(&self) -> bool;
+}
+
+pub fn should_retry(err: &impl CouldRetry, num_retries: u32, config: RetryConfig) -> bool {
+    num_retries < config.max_retries && err.could_retry()
+}
+
+impl CouldRetry for io::Error {
    fn could_retry(&self) -> bool {
        use std::io::ErrorKind;
        matches!(
@@ -25,7 +27,7 @@ impl ShouldRetry for io::Error {
    }
 }

-impl ShouldRetry for tokio_postgres::error::DbError {
+impl CouldRetry for tokio_postgres::error::DbError {
    fn could_retry(&self) -> bool {
        use tokio_postgres::error::SqlState;
        matches!(
@@ -36,7 +38,9 @@ impl ShouldRetry for tokio_postgres::error::DbError {
                | &SqlState::SQLCLIENT_UNABLE_TO_ESTABLISH_SQLCONNECTION,
        )
    }
-    fn should_retry_database_address(&self) -> bool {
+}
+impl ShouldRetryWakeCompute for tokio_postgres::error::DbError {
+    fn should_retry_wake_compute(&self) -> bool {
        use tokio_postgres::error::SqlState;
        // Here are errors that happens after the user successfully authenticated to the database.
        // TODO: there are pgbouncer errors that should be retried, but they are not listed here.
@@ -53,7 +57,7 @@ impl ShouldRetry for tokio_postgres::error::DbError {
    }
 }

-impl ShouldRetry for tokio_postgres::Error {
+impl CouldRetry for tokio_postgres::Error {
    fn could_retry(&self) -> bool {
        if let Some(io_err) = self.source().and_then(|x| x.downcast_ref()) {
            io::Error::could_retry(io_err)
@@ -63,29 +67,33 @@ impl ShouldRetry for tokio_postgres::Error {
            false
        }
    }
-    fn should_retry_database_address(&self) -> bool {
-        if let Some(io_err) = self.source().and_then(|x| x.downcast_ref()) {
-            io::Error::should_retry_database_address(io_err)
-        } else if let Some(db_err) = self.source().and_then(|x| x.downcast_ref()) {
-            tokio_postgres::error::DbError::should_retry_database_address(db_err)
+}
+impl ShouldRetryWakeCompute for tokio_postgres::Error {
+    fn should_retry_wake_compute(&self) -> bool {
+        if let Some(db_err) = self.source().and_then(|x| x.downcast_ref()) {
+            tokio_postgres::error::DbError::should_retry_wake_compute(db_err)
        } else {
+            // likely an IO error. Possible the compute has shutdown and the
+            // cache is stale.
            true
        }
    }
 }

-impl ShouldRetry for compute::ConnectionError {
+impl CouldRetry for compute::ConnectionError {
    fn could_retry(&self) -> bool {
        match self {
            compute::ConnectionError::Postgres(err) => err.could_retry(),
            compute::ConnectionError::CouldNotConnect(err) => err.could_retry(),
+            compute::ConnectionError::WakeComputeError(err) => err.could_retry(),
            _ => false,
        }
    }
-    fn should_retry_database_address(&self) -> bool {
+}
+impl ShouldRetryWakeCompute for compute::ConnectionError {
+    fn should_retry_wake_compute(&self) -> bool {
        match self {
-            compute::ConnectionError::Postgres(err) => err.should_retry_database_address(),
-            compute::ConnectionError::CouldNotConnect(err) => err.should_retry_database_address(),
+            compute::ConnectionError::Postgres(err) => err.should_retry_wake_compute(),
            // the cache entry was not checked for validity
            compute::ConnectionError::TooManyConnectionAttempts(_) => false,
            _ => true,
@@ -5,21 +5,21 @@ mod mitm;
 use std::time::Duration;

 use super::connect_compute::ConnectMechanism;
-use super::retry::ShouldRetry;
+use super::retry::CouldRetry;
 use super::*;
 use crate::auth::backend::{
    ComputeCredentialKeys, ComputeCredentials, ComputeUserInfo, MaybeOwned, TestBackend,
 };
 use crate::config::{CertResolver, RetryConfig};
 use crate::console::caches::NodeInfoCache;
-use crate::console::messages::{ConsoleError, MetricsAuxInfo};
+use crate::console::messages::{ConsoleError, Details, MetricsAuxInfo, Status};
 use crate::console::provider::{CachedAllowedIps, CachedRoleSecret, ConsoleBackend};
 use crate::console::{self, CachedNodeInfo, NodeInfo};
 use crate::error::ErrorKind;
-use crate::proxy::retry::retry_after;
 use crate::{http, sasl, scram, BranchId, EndpointId, ProjectId};
 use anyhow::{bail, Context};
 use async_trait::async_trait;
+use retry::{retry_after, ShouldRetryWakeCompute};
 use rstest::rstest;
 use rustls::pki_types;
 use tokio_postgres::config::SslMode;
@@ -438,11 +438,16 @@ impl std::fmt::Display for TestConnectError {

 impl std::error::Error for TestConnectError {}

-impl ShouldRetry for TestConnectError {
+impl CouldRetry for TestConnectError {
    fn could_retry(&self) -> bool {
        self.retryable
    }
 }
+impl ShouldRetryWakeCompute for TestConnectError {
+    fn should_retry_wake_compute(&self) -> bool {
+        true
+    }
+}

 #[async_trait]
 impl ConnectMechanism for TestConnectMechanism {
@@ -485,7 +490,7 @@ impl TestBackend for TestConnectMechanism {
            ConnectAction::Wake => Ok(helper_create_cached_node_info(self.cache)),
            ConnectAction::WakeFail => {
                let err = console::errors::ApiError::Console(ConsoleError {
-                    http_status_code: http::StatusCode::FORBIDDEN,
+                    http_status_code: http::StatusCode::BAD_REQUEST,
                    error: "TEST".into(),
                    status: None,
                });
@@ -496,7 +501,15 @@ impl TestBackend for TestConnectMechanism {
                let err = console::errors::ApiError::Console(ConsoleError {
                    http_status_code: http::StatusCode::BAD_REQUEST,
                    error: "TEST".into(),
-                    status: None,
+                    status: Some(Status {
+                        code: "error".into(),
+                        message: "error".into(),
+                        details: Details {
+                            error_info: None,
+                            retry_info: Some(console::messages::RetryInfo { retry_delay_ms: 1 }),
+                            user_facing_message: None,
+                        },
+                    }),
                });
                assert!(err.could_retry());
                Err(console::errors::WakeComputeError::ApiError(err))
@@ -1,18 +1,16 @@
 use crate::config::RetryConfig;
-use crate::console::messages::ConsoleError;
+use crate::console::messages::{ConsoleError, Reason};
 use crate::console::{errors::WakeComputeError, provider::CachedNodeInfo};
 use crate::context::RequestMonitoring;
 use crate::metrics::{
    ConnectOutcome, ConnectionFailuresBreakdownGroup, Metrics, RetriesMetricGroup, RetryType,
    WakeupFailureKind,
 };
-use crate::proxy::retry::retry_after;
+use crate::proxy::retry::{retry_after, should_retry};
 use hyper1::StatusCode;
-use std::ops::ControlFlow;
 use tracing::{error, info, warn};

 use super::connect_compute::ComputeConnectBackend;
-use super::retry::ShouldRetry;

 pub async fn wake_compute<B: ComputeConnectBackend>(
    num_retries: &mut u32,
@@ -22,9 +20,8 @@ pub async fn wake_compute<B: ComputeConnectBackend>(
 ) -> Result<CachedNodeInfo, WakeComputeError> {
    let retry_type = RetryType::WakeCompute;
    loop {
-        let wake_res = api.wake_compute(ctx).await;
-        match handle_try_wake(wake_res, *num_retries, config) {
-            Err(e) => {
+        match api.wake_compute(ctx).await {
+            Err(e) if !should_retry(&e, *num_retries, config) => {
                error!(error = ?e, num_retries, retriable = false, "couldn't wake compute node");
                report_error(&e, false);
                Metrics::get().proxy.retries_metric.observe(
@@ -36,11 +33,11 @@ pub async fn wake_compute<B: ComputeConnectBackend>(
                );
                return Err(e);
            }
-            Ok(ControlFlow::Continue(e)) => {
+            Err(e) => {
                warn!(error = ?e, num_retries, retriable = true, "couldn't wake compute node");
                report_error(&e, true);
            }
-            Ok(ControlFlow::Break(n)) => {
+            Ok(n) => {
                Metrics::get().proxy.retries_metric.observe(
                    RetriesMetricGroup {
                        outcome: ConnectOutcome::Success,
@@ -63,70 +60,28 @@ pub async fn wake_compute<B: ComputeConnectBackend>(
    }
 }

-/// Attempts to wake up the compute node.
-/// * Returns Ok(Continue(e)) if there was an error waking but retries are acceptable
-/// * Returns Ok(Break(node)) if the wakeup succeeded
-/// * Returns Err(e) if there was an error
-pub fn handle_try_wake(
-    result: Result<CachedNodeInfo, WakeComputeError>,
-    num_retries: u32,
-    config: RetryConfig,
-) -> Result<ControlFlow<CachedNodeInfo, WakeComputeError>, WakeComputeError> {
-    match result {
-        Err(err) => match &err {
-            WakeComputeError::ApiError(api) if api.should_retry(num_retries, config) => {
-                Ok(ControlFlow::Continue(err))
-            }
-            _ => Err(err),
-        },
-        // Ready to try again.
-        Ok(new) => Ok(ControlFlow::Break(new)),
-    }
-}
-
 fn report_error(e: &WakeComputeError, retry: bool) {
    use crate::console::errors::ApiError;
    let kind = match e {
        WakeComputeError::BadComputeAddress(_) => WakeupFailureKind::BadComputeAddress,
        WakeComputeError::ApiError(ApiError::Transport(_)) => WakeupFailureKind::ApiTransportError,
        WakeComputeError::ApiError(ApiError::Console(e)) => match e.get_reason() {
-            crate::console::messages::Reason::RoleProtected => {
-                WakeupFailureKind::ApiConsoleBadRequest
-            }
-            crate::console::messages::Reason::ResourceNotFound => {
-                WakeupFailureKind::ApiConsoleBadRequest
-            }
-            crate::console::messages::Reason::ProjectNotFound => {
-                WakeupFailureKind::ApiConsoleBadRequest
-            }
-            crate::console::messages::Reason::EndpointNotFound => {
-                WakeupFailureKind::ApiConsoleBadRequest
-            }
-            crate::console::messages::Reason::BranchNotFound => {
-                WakeupFailureKind::ApiConsoleBadRequest
-            }
-            crate::console::messages::Reason::RateLimitExceeded => {
-                WakeupFailureKind::ApiConsoleLocked
-            }
-            crate::console::messages::Reason::NonPrimaryBranchComputeTimeExceeded => {
-                WakeupFailureKind::QuotaExceeded
-            }
-            crate::console::messages::Reason::ActiveTimeQuotaExceeded => {
-                WakeupFailureKind::QuotaExceeded
-            }
-            crate::console::messages::Reason::ComputeTimeQuotaExceeded => {
-                WakeupFailureKind::QuotaExceeded
-            }
-            crate::console::messages::Reason::WrittenDataQuotaExceeded => {
-                WakeupFailureKind::QuotaExceeded
-            }
-            crate::console::messages::Reason::DataTransferQuotaExceeded => {
-                WakeupFailureKind::QuotaExceeded
-            }
-            crate::console::messages::Reason::LogicalSizeQuotaExceeded => {
-                WakeupFailureKind::QuotaExceeded
-            }
-            crate::console::messages::Reason::Unknown => match e {
+            Reason::RoleProtected => WakeupFailureKind::ApiConsoleBadRequest,
+            Reason::ResourceNotFound => WakeupFailureKind::ApiConsoleBadRequest,
+            Reason::ProjectNotFound => WakeupFailureKind::ApiConsoleBadRequest,
+            Reason::EndpointNotFound => WakeupFailureKind::ApiConsoleBadRequest,
+            Reason::BranchNotFound => WakeupFailureKind::ApiConsoleBadRequest,
+            Reason::RateLimitExceeded => WakeupFailureKind::ApiConsoleLocked,
+            Reason::NonDefaultBranchComputeTimeExceeded => WakeupFailureKind::QuotaExceeded,
+            Reason::ActiveTimeQuotaExceeded => WakeupFailureKind::QuotaExceeded,
+            Reason::ComputeTimeQuotaExceeded => WakeupFailureKind::QuotaExceeded,
+            Reason::WrittenDataQuotaExceeded => WakeupFailureKind::QuotaExceeded,
+            Reason::DataTransferQuotaExceeded => WakeupFailureKind::QuotaExceeded,
+            Reason::LogicalSizeQuotaExceeded => WakeupFailureKind::QuotaExceeded,
+            Reason::ConcurrencyLimitReached => WakeupFailureKind::ApiConsoleLocked,
+            Reason::LockAlreadyTaken => WakeupFailureKind::ApiConsoleLocked,
+            Reason::RunningOperations => WakeupFailureKind::ApiConsoleLocked,
+            Reason::Unknown => match e {
                ConsoleError {
                    http_status_code: StatusCode::LOCKED,
                    ref error,
@@ -16,7 +16,10 @@ use crate::{
    context::RequestMonitoring,
    error::{ErrorKind, ReportableError, UserFacingError},
    intern::EndpointIdInt,
-    proxy::{connect_compute::ConnectMechanism, retry::ShouldRetry},
+    proxy::{
+        connect_compute::ConnectMechanism,
+        retry::{CouldRetry, ShouldRetryWakeCompute},
+    },
    rate_limiter::EndpointRateLimiter,
    Host,
 };
@@ -179,7 +182,7 @@ impl UserFacingError for HttpConnError {
    }
 }

-impl ShouldRetry for HttpConnError {
+impl CouldRetry for HttpConnError {
    fn could_retry(&self) -> bool {
        match self {
            HttpConnError::ConnectionError(e) => e.could_retry(),
@@ -190,9 +193,11 @@ impl ShouldRetry for HttpConnError {
            HttpConnError::TooManyConnectionAttempts(_) => false,
        }
    }
-    fn should_retry_database_address(&self) -> bool {
+}
+impl ShouldRetryWakeCompute for HttpConnError {
+    fn should_retry_wake_compute(&self) -> bool {
        match self {
-            HttpConnError::ConnectionError(e) => e.should_retry_database_address(),
+            HttpConnError::ConnectionError(e) => e.should_retry_wake_compute(),
            // we never checked cache validity
            HttpConnError::TooManyConnectionAttempts(_) => false,
            _ => true,
@@ -1,3 +1,4 @@
+use crate::proxy::ErrorSource;
 use crate::{
    cancellation::CancellationHandlerMain,
    config::ProxyConfig,
@@ -7,6 +8,7 @@ use crate::{
    proxy::{handle_client, ClientMode},
    rate_limiter::EndpointRateLimiter,
 };
+use anyhow::Context as _;
 use bytes::{Buf, BufMut, Bytes, BytesMut};
 use framed_websockets::{Frame, OpCode, WebSocketServer};
 use futures::{Sink, Stream};
@@ -165,7 +167,11 @@ pub async fn serve_websocket(
        Ok(Some(p)) => {
            ctx.set_success();
            ctx.log_connect();
-            p.proxy_pass().await
+            match p.proxy_pass().await {
+                Ok(()) => Ok(()),
+                Err(ErrorSource::Client(err)) => Err(err).context("client"),
+                Err(ErrorSource::Compute(err)) => Err(err).context("compute"),
+            }
        }
    }
 }
@@ -28,7 +28,8 @@ use utils::pid_file;

 use metrics::set_build_info_metric;
 use safekeeper::defaults::{
-    DEFAULT_HEARTBEAT_TIMEOUT, DEFAULT_HTTP_LISTEN_ADDR, DEFAULT_MAX_OFFLOADER_LAG_BYTES,
+    DEFAULT_CONTROL_FILE_SAVE_INTERVAL, DEFAULT_HEARTBEAT_TIMEOUT, DEFAULT_HTTP_LISTEN_ADDR,
+    DEFAULT_MAX_OFFLOADER_LAG_BYTES, DEFAULT_PARTIAL_BACKUP_CONCURRENCY,
    DEFAULT_PARTIAL_BACKUP_TIMEOUT, DEFAULT_PG_LISTEN_ADDR,
 };
 use safekeeper::http;
@@ -172,6 +173,7 @@ struct Args {
    walsenders_keep_horizon: bool,
    /// Enable partial backup. If disabled, safekeeper will not upload partial
    /// segments to remote storage.
+    /// TODO: now partial backup is always enabled, remove this flag.
    #[arg(long)]
    partial_backup_enabled: bool,
    /// Controls how long backup will wait until uploading the partial segment.
@@ -181,6 +183,18 @@ struct Args {
    /// be used in tests.
    #[arg(long)]
    disable_periodic_broker_push: bool,
+    /// Enable automatic switching to offloaded state.
+    #[arg(long)]
+    enable_offload: bool,
+    /// Delete local WAL files after offloading. When disabled, they will be left on disk.
+    #[arg(long)]
+    delete_offloaded_wal: bool,
+    /// Pending updates to control file will be automatically saved after this interval.
+    #[arg(long, value_parser = humantime::parse_duration, default_value = DEFAULT_CONTROL_FILE_SAVE_INTERVAL)]
+    control_file_save_interval: Duration,
+    /// Number of allowed concurrent uploads of partial segments to remote storage.
+    #[arg(long, default_value = DEFAULT_PARTIAL_BACKUP_CONCURRENCY)]
+    partial_backup_concurrency: usize,
 }

 // Like PathBufValueParser, but allows empty string.
@@ -328,9 +342,13 @@ async fn main() -> anyhow::Result<()> {
        sk_auth_token,
        current_thread_runtime: args.current_thread_runtime,
        walsenders_keep_horizon: args.walsenders_keep_horizon,
-        partial_backup_enabled: args.partial_backup_enabled,
+        partial_backup_enabled: true,
        partial_backup_timeout: args.partial_backup_timeout,
        disable_periodic_broker_push: args.disable_periodic_broker_push,
+        enable_offload: args.enable_offload,
+        delete_offloaded_wal: args.delete_offloaded_wal,
+        control_file_save_interval: args.control_file_save_interval,
+        partial_backup_concurrency: args.partial_backup_concurrency,
    };

    // initialize sentry if SENTRY_DSN is provided
@@ -72,6 +72,9 @@ impl FileStorage {
        conf: &SafeKeeperConf,
        state: TimelinePersistentState,
    ) -> Result<FileStorage> {
+        // we don't support creating new timelines in offloaded state
+        assert!(matches!(state.eviction_state, EvictionState::Present));
+
        let store = FileStorage {
            timeline_dir,
            no_sync: conf.no_sync,
@@ -103,7 +106,7 @@ impl FileStorage {
    }

    /// Load control file from given directory.
-    pub fn load_control_file_from_dir(timeline_dir: &Utf8Path) -> Result<TimelinePersistentState> {
+    fn load_control_file_from_dir(timeline_dir: &Utf8Path) -> Result<TimelinePersistentState> {
        let path = timeline_dir.join(CONTROL_FILE_NAME);
        Self::load_control_file(path)
    }
@@ -15,7 +15,7 @@ use crate::{
    control_file::{FileStorage, Storage},
    pull_timeline::{create_temp_timeline_dir, load_temp_timeline, validate_temp_timeline},
    state::TimelinePersistentState,
-    timeline::{FullAccessTimeline, Timeline, TimelineError},
+    timeline::{Timeline, TimelineError, WalResidentTimeline},
    wal_backup::copy_s3_segments,
    wal_storage::{wal_file_paths, WalReader},
    GlobalTimelines,
@@ -46,7 +46,7 @@ pub async fn handle_request(request: Request) -> Result<()> {
        }
    }

-    let source_tli = request.source.full_access_guard().await?;
+    let source_tli = request.source.wal_residence_guard().await?;

    let conf = &GlobalTimelines::get_global_config();
    let ttid = request.destination_ttid;
@@ -159,7 +159,7 @@ pub async fn handle_request(request: Request) -> Result<()> {
 }

 async fn copy_disk_segments(
-    tli: &FullAccessTimeline,
+    tli: &WalResidentTimeline,
    wal_seg_size: usize,
    start_lsn: Lsn,
    end_lsn: Lsn,
@@ -183,7 +183,7 @@ async fn copy_disk_segments(
        let copy_end = copy_end - segment_start;

        let wal_file_path = {
-            let (normal, partial) = wal_file_paths(tli_dir_path, segment, wal_seg_size)?;
+            let (normal, partial) = wal_file_paths(tli_dir_path, segment, wal_seg_size);

            if segment == last_segment {
                partial
@@ -28,7 +28,8 @@ use crate::send_wal::WalSenderState;
 use crate::state::TimelineMemState;
 use crate::state::TimelinePersistentState;
 use crate::timeline::get_timeline_dir;
-use crate::timeline::FullAccessTimeline;
+use crate::timeline::WalResidentTimeline;
+use crate::timeline_manager;
 use crate::GlobalTimelines;
 use crate::SafeKeeperConf;

@@ -168,6 +169,7 @@ pub struct Memory {
    pub last_removed_segno: XLogSegNo,
    pub epoch_start_lsn: Lsn,
    pub mem_state: TimelineMemState,
+    pub mgr_status: timeline_manager::Status,

    // PhysicalStorage state.
    pub write_lsn: Lsn,
@@ -326,7 +328,7 @@ pub struct TimelineDigest {
 }

 pub async fn calculate_digest(
-    tli: &FullAccessTimeline,
+    tli: &WalResidentTimeline,
    request: TimelineDigestRequest,
 ) -> Result<TimelineDigest> {
    if request.from_lsn > request.until_lsn {
@@ -214,10 +214,10 @@ async fn timeline_snapshot_handler(request: Request<Body>) -> Result<Response<Bo
    let tli = GlobalTimelines::get(ttid).map_err(ApiError::from)?;
    // Note: with evicted timelines it should work better then de-evict them and
    // stream; probably start_snapshot would copy partial s3 file to dest path
-    // and stream control file, or return FullAccessTimeline if timeline is not
+    // and stream control file, or return WalResidentTimeline if timeline is not
    // evicted.
    let tli = tli
-        .full_access_guard()
+        .wal_residence_guard()
        .await
        .map_err(ApiError::InternalServerError)?;

@@ -283,7 +283,7 @@ async fn timeline_digest_handler(request: Request<Body>) -> Result<Response<Body

    let tli = GlobalTimelines::get(ttid).map_err(ApiError::from)?;
    let tli = tli
-        .full_access_guard()
+        .wal_residence_guard()
        .await
        .map_err(ApiError::InternalServerError)?;

@@ -306,7 +306,7 @@ async fn timeline_checkpoint_handler(request: Request<Body>) -> Result<Response<
    tli.write_shared_state()
        .await
        .sk
-        .state
+        .state_mut()
        .flush()
        .await
        .map_err(ApiError::InternalServerError)?;
@@ -21,7 +21,7 @@ use crate::safekeeper::{
 };
 use crate::safekeeper::{Term, TermHistory, TermLsn};
 use crate::state::TimelinePersistentState;
-use crate::timeline::FullAccessTimeline;
+use crate::timeline::WalResidentTimeline;
 use crate::GlobalTimelines;
 use postgres_backend::PostgresBackend;
 use postgres_ffi::encode_logical_message;
@@ -102,7 +102,7 @@ pub async fn handle_json_ctrl<IO: AsyncRead + AsyncWrite + Unpin>(
 async fn prepare_safekeeper(
    ttid: TenantTimelineId,
    pg_version: u32,
-) -> anyhow::Result<FullAccessTimeline> {
+) -> anyhow::Result<WalResidentTimeline> {
    let tli = GlobalTimelines::create(
        ttid,
        ServerInfo {
@@ -115,11 +115,11 @@ async fn prepare_safekeeper(
    )
    .await?;

-    tli.full_access_guard().await
+    tli.wal_residence_guard().await
 }

 async fn send_proposer_elected(
-    tli: &FullAccessTimeline,
+    tli: &WalResidentTimeline,
    term: Term,
    lsn: Lsn,
 ) -> anyhow::Result<()> {
@@ -151,7 +151,7 @@ pub struct InsertedWAL {
 /// Extend local WAL with new LogicalMessage record. To do that,
 /// create AppendRequest with new WAL and pass it to safekeeper.
 pub async fn append_logical_message(
-    tli: &FullAccessTimeline,
+    tli: &WalResidentTimeline,
    msg: &AppendLogicalMessage,
 ) -> anyhow::Result<InsertedWAL> {
    let wal_data = encode_logical_message(&msg.lm_prefix, &msg.lm_message);
@@ -28,6 +28,8 @@ pub mod safekeeper;
 pub mod send_wal;
 pub mod state;
 pub mod timeline;
+pub mod timeline_eviction;
+pub mod timeline_guard;
 pub mod timeline_manager;
 pub mod timelines_set;
 pub mod wal_backup;
@@ -49,6 +51,8 @@ pub mod defaults {
    pub const DEFAULT_HEARTBEAT_TIMEOUT: &str = "5000ms";
    pub const DEFAULT_MAX_OFFLOADER_LAG_BYTES: u64 = 128 * (1 << 20);
    pub const DEFAULT_PARTIAL_BACKUP_TIMEOUT: &str = "15m";
+    pub const DEFAULT_CONTROL_FILE_SAVE_INTERVAL: &str = "300s";
+    pub const DEFAULT_PARTIAL_BACKUP_CONCURRENCY: &str = "5";
 }

 #[derive(Debug, Clone)]
@@ -85,6 +89,10 @@ pub struct SafeKeeperConf {
    pub partial_backup_enabled: bool,
    pub partial_backup_timeout: Duration,
    pub disable_periodic_broker_push: bool,
+    pub enable_offload: bool,
+    pub delete_offloaded_wal: bool,
+    pub control_file_save_interval: Duration,
+    pub partial_backup_concurrency: usize,
 }

 impl SafeKeeperConf {
@@ -124,6 +132,10 @@ impl SafeKeeperConf {
            partial_backup_enabled: false,
            partial_backup_timeout: Duration::from_secs(0),
            disable_periodic_broker_push: false,
+            enable_offload: false,
+            delete_offloaded_wal: false,
+            control_file_save_interval: Duration::from_secs(1),
+            partial_backup_concurrency: 1,
        }
    }
 }
@@ -5,15 +5,15 @@ use std::{
    time::{Instant, SystemTime},
 };

-use ::metrics::{register_histogram, GaugeVec, Histogram, IntGauge, DISK_WRITE_SECONDS_BUCKETS};
+use ::metrics::{register_histogram, GaugeVec, Histogram, IntGauge, DISK_FSYNC_SECONDS_BUCKETS};
 use anyhow::Result;
 use futures::Future;
 use metrics::{
    core::{AtomicU64, Collector, Desc, GenericCounter, GenericGaugeVec, Opts},
    proto::MetricFamily,
-    register_int_counter, register_int_counter_pair, register_int_counter_pair_vec,
-    register_int_counter_vec, Gauge, IntCounter, IntCounterPair, IntCounterPairVec, IntCounterVec,
-    IntGaugeVec,
+    register_histogram_vec, register_int_counter, register_int_counter_pair,
+    register_int_counter_pair_vec, register_int_counter_vec, Gauge, HistogramVec, IntCounter,
+    IntCounterPair, IntCounterPairVec, IntCounterVec, IntGaugeVec,
 };
 use once_cell::sync::Lazy;

@@ -48,7 +48,7 @@ pub static WRITE_WAL_SECONDS: Lazy<Histogram> = Lazy::new(|| {
    register_histogram!(
        "safekeeper_write_wal_seconds",
        "Seconds spent writing and syncing WAL to a disk in a single request",
-        DISK_WRITE_SECONDS_BUCKETS.to_vec()
+        DISK_FSYNC_SECONDS_BUCKETS.to_vec()
    )
    .expect("Failed to register safekeeper_write_wal_seconds histogram")
 });
@@ -56,7 +56,7 @@ pub static FLUSH_WAL_SECONDS: Lazy<Histogram> = Lazy::new(|| {
    register_histogram!(
        "safekeeper_flush_wal_seconds",
        "Seconds spent syncing WAL to a disk",
-        DISK_WRITE_SECONDS_BUCKETS.to_vec()
+        DISK_FSYNC_SECONDS_BUCKETS.to_vec()
    )
    .expect("Failed to register safekeeper_flush_wal_seconds histogram")
 });
@@ -64,10 +64,28 @@ pub static PERSIST_CONTROL_FILE_SECONDS: Lazy<Histogram> = Lazy::new(|| {
    register_histogram!(
        "safekeeper_persist_control_file_seconds",
        "Seconds to persist and sync control file",
-        DISK_WRITE_SECONDS_BUCKETS.to_vec()
+        DISK_FSYNC_SECONDS_BUCKETS.to_vec()
    )
    .expect("Failed to register safekeeper_persist_control_file_seconds histogram vec")
 });
+pub static WAL_STORAGE_OPERATION_SECONDS: Lazy<HistogramVec> = Lazy::new(|| {
+    register_histogram_vec!(
+        "safekeeper_wal_storage_operation_seconds",
+        "Seconds spent on WAL storage operations",
+        &["operation"],
+        DISK_FSYNC_SECONDS_BUCKETS.to_vec()
+    )
+    .expect("Failed to register safekeeper_wal_storage_operation_seconds histogram vec")
+});
+pub static MISC_OPERATION_SECONDS: Lazy<HistogramVec> = Lazy::new(|| {
+    register_histogram_vec!(
+        "safekeeper_misc_operation_seconds",
+        "Seconds spent on miscellaneous operations",
+        &["operation"],
+        DISK_FSYNC_SECONDS_BUCKETS.to_vec()
+    )
+    .expect("Failed to register safekeeper_misc_operation_seconds histogram vec")
+});
 pub static PG_IO_BYTES: Lazy<IntCounterVec> = Lazy::new(|| {
    register_int_counter_vec!(
        "safekeeper_pg_io_bytes_total",
@@ -126,7 +144,7 @@ pub static BROKER_PUSH_ALL_UPDATES_SECONDS: Lazy<Histogram> = Lazy::new(|| {
    register_histogram!(
        "safekeeper_broker_push_update_seconds",
        "Seconds to push all timeline updates to the broker",
-        DISK_WRITE_SECONDS_BUCKETS.to_vec()
+        DISK_FSYNC_SECONDS_BUCKETS.to_vec()
    )
    .expect("Failed to register safekeeper_broker_push_update_seconds histogram vec")
 });
@@ -32,7 +32,7 @@ use crate::{
        routes::TimelineStatus,
    },
    safekeeper::Term,
-    timeline::{get_tenant_dir, get_timeline_dir, FullAccessTimeline, Timeline, TimelineError},
+    timeline::{get_tenant_dir, get_timeline_dir, Timeline, TimelineError, WalResidentTimeline},
    wal_storage::{self, open_wal_file, Storage},
    GlobalTimelines, SafeKeeperConf,
 };
@@ -46,7 +46,7 @@ use utils::{

 /// Stream tar archive of timeline to tx.
 #[instrument(name = "snapshot", skip_all, fields(ttid = %tli.ttid))]
-pub async fn stream_snapshot(tli: FullAccessTimeline, tx: mpsc::Sender<Result<Bytes>>) {
+pub async fn stream_snapshot(tli: WalResidentTimeline, tx: mpsc::Sender<Result<Bytes>>) {
    if let Err(e) = stream_snapshot_guts(tli, tx.clone()).await {
        // Error type/contents don't matter as they won't can't reach the client
        // (hyper likely doesn't do anything with it), but http stream will be
@@ -66,7 +66,7 @@ pub struct SnapshotContext {
    pub flush_lsn: Lsn,
    pub wal_seg_size: usize,
    // used to remove WAL hold off in Drop.
-    pub tli: FullAccessTimeline,
+    pub tli: WalResidentTimeline,
 }

 impl Drop for SnapshotContext {
@@ -80,7 +80,7 @@ impl Drop for SnapshotContext {
 }

 pub async fn stream_snapshot_guts(
-    tli: FullAccessTimeline,
+    tli: WalResidentTimeline,
    tx: mpsc::Sender<Result<Bytes>>,
 ) -> Result<()> {
    // tokio-tar wants Write implementor, but we have mpsc tx <Result<Bytes>>;
@@ -135,7 +135,7 @@ pub async fn stream_snapshot_guts(
    Ok(())
 }

-impl FullAccessTimeline {
+impl WalResidentTimeline {
    /// Start streaming tar archive with timeline:
    /// 1) stream control file under lock;
    /// 2) hold off WAL removal;
@@ -160,6 +160,7 @@ impl FullAccessTimeline {
        ar: &mut tokio_tar::Builder<W>,
    ) -> Result<SnapshotContext> {
        let mut shared_state = self.write_shared_state().await;
+        let wal_seg_size = shared_state.get_wal_seg_size();

        let cf_path = self.get_timeline_dir().join(CONTROL_FILE_NAME);
        let mut cf = File::open(cf_path).await?;
@@ -173,19 +174,19 @@ impl FullAccessTimeline {
        // lock and setting `wal_removal_on_hold` later, it guarantees that WAL
        // won't be removed until we're done.
        let from_lsn = min(
-            shared_state.sk.state.remote_consistent_lsn,
-            shared_state.sk.state.backup_lsn,
+            shared_state.sk.state().remote_consistent_lsn,
+            shared_state.sk.state().backup_lsn,
        );
        if from_lsn == Lsn::INVALID {
            // this is possible if snapshot is called before handling first
            // elected message
            bail!("snapshot is called on uninitialized timeline");
        }
-        let from_segno = from_lsn.segment_number(shared_state.get_wal_seg_size());
-        let term = shared_state.sk.get_term();
-        let last_log_term = shared_state.sk.get_last_log_term();
+        let from_segno = from_lsn.segment_number(wal_seg_size);
+        let term = shared_state.sk.state().acceptor_state.term;
+        let last_log_term = shared_state.sk.last_log_term();
        let flush_lsn = shared_state.sk.flush_lsn();
-        let upto_segno = flush_lsn.segment_number(shared_state.get_wal_seg_size());
+        let upto_segno = flush_lsn.segment_number(wal_seg_size);
        // have some limit on max number of segments as a sanity check
        const MAX_ALLOWED_SEGS: u64 = 1000;
        let num_segs = upto_segno - from_segno + 1;
@@ -206,14 +207,18 @@ impl FullAccessTimeline {
        }
        shared_state.wal_removal_on_hold = true;

+        // Drop shared_state to release the lock, before calling wal_residence_guard().
+        drop(shared_state);
+
+        let tli_copy = self.wal_residence_guard().await?;
        let bctx = SnapshotContext {
            from_segno,
            upto_segno,
            term,
            last_log_term,
            flush_lsn,
-            wal_seg_size: shared_state.get_wal_seg_size(),
-            tli: self.clone(),
+            wal_seg_size,
+            tli: tli_copy,
        };

        Ok(bctx)
@@ -225,8 +230,8 @@ impl FullAccessTimeline {
    /// forget this if snapshotting fails mid the way.
    pub async fn finish_snapshot(&self, bctx: &SnapshotContext) -> Result<()> {
        let shared_state = self.read_shared_state().await;
-        let term = shared_state.sk.get_term();
-        let last_log_term = shared_state.sk.get_last_log_term();
+        let term = shared_state.sk.state().acceptor_state.term;
+        let last_log_term = shared_state.sk.last_log_term();
        // There are some cases to relax this check (e.g. last_log_term might
        // change, but as long as older history is strictly part of new that's
        // fine), but there is no need to do it.
@@ -6,7 +6,7 @@ use crate::handler::SafekeeperPostgresHandler;
 use crate::safekeeper::AcceptorProposerMessage;
 use crate::safekeeper::ProposerAcceptorMessage;
 use crate::safekeeper::ServerInfo;
-use crate::timeline::FullAccessTimeline;
+use crate::timeline::WalResidentTimeline;
 use crate::wal_service::ConnectionId;
 use crate::GlobalTimelines;
 use anyhow::{anyhow, Context};
@@ -213,7 +213,7 @@ impl SafekeeperPostgresHandler {
        &mut self,
        pgb: &mut PostgresBackend<IO>,
    ) -> Result<(), QueryError> {
-        let mut tli: Option<FullAccessTimeline> = None;
+        let mut tli: Option<WalResidentTimeline> = None;
        if let Err(end) = self.handle_start_wal_push_guts(pgb, &mut tli).await {
            // Log the result and probably send it to the client, closing the stream.
            let handle_end_fut = pgb.handle_copy_stream_end(end);
@@ -233,7 +233,7 @@ impl SafekeeperPostgresHandler {
    pub async fn handle_start_wal_push_guts<IO: AsyncRead + AsyncWrite + Unpin>(
        &mut self,
        pgb: &mut PostgresBackend<IO>,
-        tli: &mut Option<FullAccessTimeline>,
+        tli: &mut Option<WalResidentTimeline>,
    ) -> Result<(), CopyStreamHandlerEnd> {
        // Notify the libpq client that it's allowed to send `CopyData` messages
        pgb.write_message(&BeMessage::CopyBothResponse).await?;
@@ -269,11 +269,11 @@ impl SafekeeperPostgresHandler {
                    .get_walreceivers()
                    .pageserver_feedback_tx
                    .subscribe();
-            *tli = Some(timeline.clone());
+            *tli = Some(timeline.wal_residence_guard().await?);

            tokio::select! {
                // todo: add read|write .context to these errors
-                r = network_reader.run(msg_tx, msg_rx, reply_tx, timeline.clone(), next_msg) => r,
+                r = network_reader.run(msg_tx, msg_rx, reply_tx, timeline, next_msg) => r,
                r = network_write(pgb, reply_rx, pageserver_feedback_rx) => r,
            }
        } else {
@@ -323,7 +323,7 @@ struct NetworkReader<'a, IO> {
 impl<'a, IO: AsyncRead + AsyncWrite + Unpin> NetworkReader<'a, IO> {
    async fn read_first_message(
        &mut self,
-    ) -> Result<(FullAccessTimeline, ProposerAcceptorMessage), CopyStreamHandlerEnd> {
+    ) -> Result<(WalResidentTimeline, ProposerAcceptorMessage), CopyStreamHandlerEnd> {
        // Receive information about server to create timeline, if not yet.
        let next_msg = read_message(self.pgb_reader).await?;
        let tli = match next_msg {
@@ -340,7 +340,7 @@ impl<'a, IO: AsyncRead + AsyncWrite + Unpin> NetworkReader<'a, IO> {
                let tli =
                    GlobalTimelines::create(self.ttid, server_info, Lsn::INVALID, Lsn::INVALID)
                        .await?;
-                tli.full_access_guard().await?
+                tli.wal_residence_guard().await?
            }
            _ => {
                return Err(CopyStreamHandlerEnd::Other(anyhow::anyhow!(
@@ -356,7 +356,7 @@ impl<'a, IO: AsyncRead + AsyncWrite + Unpin> NetworkReader<'a, IO> {
        msg_tx: Sender<ProposerAcceptorMessage>,
        msg_rx: Receiver<ProposerAcceptorMessage>,
        reply_tx: Sender<AcceptorProposerMessage>,
-        tli: FullAccessTimeline,
+        tli: WalResidentTimeline,
        next_msg: ProposerAcceptorMessage,
    ) -> Result<(), CopyStreamHandlerEnd> {
        *self.acceptor_handle = Some(WalAcceptor::spawn(
@@ -451,7 +451,7 @@ const KEEPALIVE_INTERVAL: Duration = Duration::from_secs(1);
 /// replies to reply_tx; reading from socket and writing to disk in parallel is
 /// beneficial for performance, this struct provides writing to disk part.
 pub struct WalAcceptor {
-    tli: FullAccessTimeline,
+    tli: WalResidentTimeline,
    msg_rx: Receiver<ProposerAcceptorMessage>,
    reply_tx: Sender<AcceptorProposerMessage>,
    conn_id: Option<ConnectionId>,
@@ -464,7 +464,7 @@ impl WalAcceptor {
    ///
    /// conn_id None means WalAcceptor is used by recovery initiated at this safekeeper.
    pub fn spawn(
-        tli: FullAccessTimeline,
+        tli: WalResidentTimeline,
        msg_rx: Receiver<ProposerAcceptorMessage>,
        reply_tx: Sender<AcceptorProposerMessage>,
        conn_id: Option<ConnectionId>,
@@ -21,7 +21,7 @@ use utils::{id::NodeId, lsn::Lsn, postgres_client::wal_stream_connection_config}

 use crate::receive_wal::{WalAcceptor, REPLY_QUEUE_SIZE};
 use crate::safekeeper::{AppendRequest, AppendRequestHeader};
-use crate::timeline::FullAccessTimeline;
+use crate::timeline::WalResidentTimeline;
 use crate::{
    http::routes::TimelineStatus,
    receive_wal::MSG_QUEUE_SIZE,
@@ -36,7 +36,7 @@ use crate::{
 /// Entrypoint for per timeline task which always runs, checking whether
 /// recovery for this safekeeper is needed and starting it if so.
 #[instrument(name = "recovery task", skip_all, fields(ttid = %tli.ttid))]
-pub async fn recovery_main(tli: FullAccessTimeline, conf: SafeKeeperConf) {
+pub async fn recovery_main(tli: WalResidentTimeline, conf: SafeKeeperConf) {
    info!("started");

    let cancel = tli.cancel.clone();
@@ -66,12 +66,12 @@ pub async fn recovery_main(tli: FullAccessTimeline, conf: SafeKeeperConf) {
 /// depending on assembled quorum (e.g. classic picture 8 from Raft paper).
 /// Thus we don't try to predict it here.
 async fn recovery_needed(
-    tli: &FullAccessTimeline,
+    tli: &WalResidentTimeline,
    heartbeat_timeout: Duration,
 ) -> RecoveryNeededInfo {
    let ss = tli.read_shared_state().await;
-    let term = ss.sk.state.acceptor_state.term;
-    let last_log_term = ss.sk.get_last_log_term();
+    let term = ss.sk.state().acceptor_state.term;
+    let last_log_term = ss.sk.last_log_term();
    let flush_lsn = ss.sk.flush_lsn();
    // note that peers contain myself, but that's ok -- we are interested only in peers which are strictly ahead of us.
    let mut peers = ss.get_peers(heartbeat_timeout);
@@ -195,7 +195,7 @@ impl From<&PeerInfo> for Donor {
 const CHECK_INTERVAL_MS: u64 = 2000;

 /// Check regularly whether we need to start recovery.
-async fn recovery_main_loop(tli: FullAccessTimeline, conf: SafeKeeperConf) {
+async fn recovery_main_loop(tli: WalResidentTimeline, conf: SafeKeeperConf) {
    let check_duration = Duration::from_millis(CHECK_INTERVAL_MS);
    loop {
        let recovery_needed_info = recovery_needed(&tli, conf.heartbeat_timeout).await;
@@ -205,7 +205,12 @@ async fn recovery_main_loop(tli: FullAccessTimeline, conf: SafeKeeperConf) {
                    "starting recovery from donor {}: {}",
                    donor.sk_id, recovery_needed_info
                );
-                match recover(tli.clone(), donor, &conf).await {
+                let res = tli.wal_residence_guard().await;
+                if let Err(e) = res {
+                    warn!("failed to obtain guard: {}", e);
+                    continue;
+                }
+                match recover(res.unwrap(), donor, &conf).await {
                    // Note: 'write_wal rewrites WAL written before' error is
                    // expected here and might happen if compute and recovery
                    // concurrently write the same data. Eventually compute
@@ -228,7 +233,7 @@ async fn recovery_main_loop(tli: FullAccessTimeline, conf: SafeKeeperConf) {
 /// Recover from the specified donor. Returns message explaining normal finish
 /// reason or error.
 async fn recover(
-    tli: FullAccessTimeline,
+    tli: WalResidentTimeline,
    donor: &Donor,
    conf: &SafeKeeperConf,
 ) -> anyhow::Result<String> {
@@ -314,7 +319,7 @@ async fn recover(

 // Pull WAL from donor, assuming handshake is already done.
 async fn recovery_stream(
-    tli: FullAccessTimeline,
+    tli: WalResidentTimeline,
    donor: &Donor,
    start_streaming_at: Lsn,
    conf: &SafeKeeperConf,
@@ -364,10 +369,10 @@ async fn recovery_stream(
    // As in normal walreceiver, do networking and writing to disk in parallel.
    let (msg_tx, msg_rx) = channel(MSG_QUEUE_SIZE);
    let (reply_tx, reply_rx) = channel(REPLY_QUEUE_SIZE);
-    let wa = WalAcceptor::spawn(tli.clone(), msg_rx, reply_tx, None);
+    let wa = WalAcceptor::spawn(tli.wal_residence_guard().await?, msg_rx, reply_tx, None);

    let res = tokio::select! {
-        r = network_io(physical_stream, msg_tx, donor.clone(), tli.clone(), conf.clone()) => r,
+        r = network_io(physical_stream, msg_tx, donor.clone(), tli, conf.clone()) => r,
        r = read_replies(reply_rx, donor.term) => r.map(|()| None),
    };

@@ -398,7 +403,7 @@ async fn network_io(
    physical_stream: ReplicationStream,
    msg_tx: Sender<ProposerAcceptorMessage>,
    donor: Donor,
-    tli: FullAccessTimeline,
+    tli: WalResidentTimeline,
    conf: SafeKeeperConf,
 ) -> anyhow::Result<Option<String>> {
    let mut physical_stream = pin!(physical_stream);
@@ -8,7 +8,7 @@ use crate::timeline_manager::StateSnapshot;
 /// While it is safe to use inmem values for determining horizon,
 /// we use persistent to make possible normal states less surprising.
 /// All segments covering LSNs before horizon_lsn can be removed.
-pub fn calc_horizon_lsn(state: &StateSnapshot, extra_horizon_lsn: Option<Lsn>) -> Lsn {
+pub(crate) fn calc_horizon_lsn(state: &StateSnapshot, extra_horizon_lsn: Option<Lsn>) -> Lsn {
    use std::cmp::min;

    let mut horizon_lsn = min(
@@ -15,6 +15,7 @@ use storage_broker::proto::SafekeeperTimelineInfo;
 use tracing::*;

 use crate::control_file;
+use crate::metrics::MISC_OPERATION_SECONDS;
 use crate::send_wal::HotStandbyFeedback;

 use crate::state::TimelineState;
@@ -499,7 +500,11 @@ where
    /// Accepts a control file storage containing the safekeeper state.
    /// State must be initialized, i.e. contain filled `tenant_id`, `timeline_id`
    /// and `server` (`wal_seg_size` inside it) fields.
-    pub fn new(state: CTRL, wal_store: WAL, node_id: NodeId) -> Result<SafeKeeper<CTRL, WAL>> {
+    pub fn new(
+        state: TimelineState<CTRL>,
+        wal_store: WAL,
+        node_id: NodeId,
+    ) -> Result<SafeKeeper<CTRL, WAL>> {
        if state.tenant_id == TenantId::from([0u8; 16])
            || state.timeline_id == TimelineId::from([0u8; 16])
        {
@@ -512,7 +517,7 @@ where

        Ok(SafeKeeper {
            term_start_lsn: Lsn(0),
-            state: TimelineState::new(state),
+            state,
            wal_store,
            node_id,
        })
@@ -526,11 +531,6 @@ where
            .up_to(self.flush_lsn())
    }

-    /// Get current term.
-    pub fn get_term(&self) -> Term {
-        self.state.acceptor_state.term
-    }
-
    pub fn get_last_log_term(&self) -> Term {
        self.state
            .acceptor_state
@@ -697,6 +697,10 @@ where
        &mut self,
        msg: &ProposerElected,
    ) -> Result<Option<AcceptorProposerMessage>> {
+        let _timer = MISC_OPERATION_SECONDS
+            .with_label_values(&["handle_elected"])
+            .start_timer();
+
        info!("received ProposerElected {:?}", msg);
        if self.state.acceptor_state.term < msg.term {
            let mut state = self.state.start_change();
@@ -912,10 +916,8 @@ where
        )))
    }

-    /// Update timeline state with peer safekeeper data.
+    /// Update commit_lsn from peer safekeeper data.
    pub async fn record_safekeeper_info(&mut self, sk_info: &SafekeeperTimelineInfo) -> Result<()> {
-        let mut sync_control_file = false;
-
        if (Lsn(sk_info.commit_lsn) != Lsn::INVALID) && (sk_info.last_log_term != INVALID_TERM) {
            // Note: the check is too restrictive, generally we can update local
            // commit_lsn if our history matches (is part of) history of advanced
@@ -924,29 +926,6 @@ where
                self.update_commit_lsn(Lsn(sk_info.commit_lsn)).await?;
            }
        }
-
-        self.state.inmem.backup_lsn = max(Lsn(sk_info.backup_lsn), self.state.inmem.backup_lsn);
-        sync_control_file |= self.state.backup_lsn + (self.state.server.wal_seg_size as u64)
-            < self.state.inmem.backup_lsn;
-
-        self.state.inmem.remote_consistent_lsn = max(
-            Lsn(sk_info.remote_consistent_lsn),
-            self.state.inmem.remote_consistent_lsn,
-        );
-        sync_control_file |= self.state.remote_consistent_lsn
-            + (self.state.server.wal_seg_size as u64)
-            < self.state.inmem.remote_consistent_lsn;
-
-        self.state.inmem.peer_horizon_lsn = max(
-            Lsn(sk_info.peer_horizon_lsn),
-            self.state.inmem.peer_horizon_lsn,
-        );
-        sync_control_file |= self.state.peer_horizon_lsn + (self.state.server.wal_seg_size as u64)
-            < self.state.inmem.peer_horizon_lsn;
-
-        if sync_control_file {
-            self.state.flush().await?;
-        }
        Ok(())
    }
 }
@@ -1039,7 +1018,7 @@ mod tests {
            persisted_state: test_sk_state(),
        };
        let wal_store = DummyWalStore { lsn: Lsn(0) };
-        let mut sk = SafeKeeper::new(storage, wal_store, NodeId(0)).unwrap();
+        let mut sk = SafeKeeper::new(TimelineState::new(storage), wal_store, NodeId(0)).unwrap();

        // check voting for 1 is ok
        let vote_request = ProposerAcceptorMessage::VoteRequest(VoteRequest { term: 1 });
@@ -1055,7 +1034,7 @@ mod tests {
            persisted_state: state,
        };

-        sk = SafeKeeper::new(storage, sk.wal_store, NodeId(0)).unwrap();
+        sk = SafeKeeper::new(TimelineState::new(storage), sk.wal_store, NodeId(0)).unwrap();

        // and ensure voting second time for 1 is not ok
        vote_resp = sk.process_msg(&vote_request).await;
@@ -1072,7 +1051,7 @@ mod tests {
        };
        let wal_store = DummyWalStore { lsn: Lsn(0) };

-        let mut sk = SafeKeeper::new(storage, wal_store, NodeId(0)).unwrap();
+        let mut sk = SafeKeeper::new(TimelineState::new(storage), wal_store, NodeId(0)).unwrap();

        let mut ar_hdr = AppendRequestHeader {
            term: 1,
@@ -5,7 +5,7 @@ use crate::handler::SafekeeperPostgresHandler;
 use crate::metrics::RECEIVED_PS_FEEDBACKS;
 use crate::receive_wal::WalReceivers;
 use crate::safekeeper::{Term, TermLsn};
-use crate::timeline::FullAccessTimeline;
+use crate::timeline::WalResidentTimeline;
 use crate::wal_service::ConnectionId;
 use crate::wal_storage::WalReader;
 use crate::GlobalTimelines;
@@ -387,10 +387,10 @@ impl SafekeeperPostgresHandler {
        term: Option<Term>,
    ) -> Result<(), QueryError> {
        let tli = GlobalTimelines::get(self.ttid).map_err(|e| QueryError::Other(e.into()))?;
-        let full_access = tli.full_access_guard().await?;
+        let residence_guard = tli.wal_residence_guard().await?;

        if let Err(end) = self
-            .handle_start_replication_guts(pgb, start_pos, term, full_access)
+            .handle_start_replication_guts(pgb, start_pos, term, residence_guard)
            .await
        {
            let info = tli.get_safekeeper_info(&self.conf).await;
@@ -407,7 +407,7 @@ impl SafekeeperPostgresHandler {
        pgb: &mut PostgresBackend<IO>,
        start_pos: Lsn,
        term: Option<Term>,
-        tli: FullAccessTimeline,
+        tli: WalResidentTimeline,
    ) -> Result<(), CopyStreamHandlerEnd> {
        let appname = self.appname.clone();

@@ -458,7 +458,8 @@ impl SafekeeperPostgresHandler {

        let mut sender = WalSender {
            pgb,
-            tli: tli.clone(),
+            // should succeed since we're already holding another guard
+            tli: tli.wal_residence_guard().await?,
            appname,
            start_pos,
            end_pos,
@@ -527,7 +528,7 @@ impl EndWatch {
 /// A half driving sending WAL.
 struct WalSender<'a, IO> {
    pgb: &'a mut PostgresBackend<IO>,
-    tli: FullAccessTimeline,
+    tli: WalResidentTimeline,
    appname: Option<String>,
    // Position since which we are sending next chunk.
    start_pos: Lsn,
@@ -736,7 +737,7 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> WalSender<'_, IO> {
 struct ReplyReader<IO> {
    reader: PostgresBackendReader<IO>,
    ws_guard: Arc<WalSenderGuard>,
-    tli: FullAccessTimeline,
+    tli: WalResidentTimeline,
 }

 impl<IO: AsyncRead + AsyncWrite + Unpin> ReplyReader<IO> {
@@ -189,7 +189,12 @@ where

    /// Persist given state. c.f. start_change.
    pub async fn finish_change(&mut self, s: &TimelinePersistentState) -> Result<()> {
-        self.pers.persist(s).await?;
+        if s.eq(&*self.pers) {
+            // nothing to do if state didn't change
+        } else {
+            self.pers.persist(s).await?;
+        }
+
        // keep in memory values up to date
        self.inmem.commit_lsn = s.commit_lsn;
        self.inmem.backup_lsn = s.backup_lsn;
@@ -31,12 +31,15 @@ use crate::safekeeper::{
    INVALID_TERM,
 };
 use crate::send_wal::WalSenders;
-use crate::state::{TimelineMemState, TimelinePersistentState};
+use crate::state::{EvictionState, TimelineMemState, TimelinePersistentState, TimelineState};
+use crate::timeline_guard::ResidenceGuard;
+use crate::timeline_manager::{AtomicStatus, ManagerCtl};
 use crate::timelines_set::TimelinesSet;
 use crate::wal_backup::{self};
+use crate::wal_backup_partial::{PartialRemoteSegment, RateLimiter};
 use crate::{control_file, safekeeper::UNKNOWN_SERVER_VERSION};

-use crate::metrics::FullTimelineInfo;
+use crate::metrics::{FullTimelineInfo, WalStorageMetrics, MISC_OPERATION_SECONDS};
 use crate::wal_storage::{Storage as wal_storage_iface, WalReader};
 use crate::{debug_dump, timeline_manager, wal_storage};
 use crate::{GlobalTimelines, SafeKeeperConf};
@@ -132,8 +135,9 @@ impl<'a> DerefMut for WriteGuardSharedState<'a> {

 impl<'a> Drop for WriteGuardSharedState<'a> {
    fn drop(&mut self) {
-        let term_flush_lsn = TermLsn::from((self.guard.sk.get_term(), self.guard.sk.flush_lsn()));
-        let commit_lsn = self.guard.sk.state.inmem.commit_lsn;
+        let term_flush_lsn =
+            TermLsn::from((self.guard.sk.last_log_term(), self.guard.sk.flush_lsn()));
+        let commit_lsn = self.guard.sk.state().inmem.commit_lsn;

        let _ = self.tli.term_flush_lsn_watch_tx.send_if_modified(|old| {
            if *old != term_flush_lsn {
@@ -162,10 +166,150 @@ impl<'a> Drop for WriteGuardSharedState<'a> {
    }
 }

+/// This structure is stored in shared state and represents the state of the timeline.
+/// Usually it holds SafeKeeper, but it also supports offloaded timeline state. In this
+/// case, SafeKeeper is not available (because WAL is not present on disk) and all
+/// operations can be done only with control file.
+pub enum StateSK {
+    Loaded(SafeKeeper<control_file::FileStorage, wal_storage::PhysicalStorage>),
+    Offloaded(Box<TimelineState<control_file::FileStorage>>),
+    // Not used, required for moving between states.
+    Empty,
+}
+
+impl StateSK {
+    pub fn flush_lsn(&self) -> Lsn {
+        match self {
+            StateSK::Loaded(sk) => sk.wal_store.flush_lsn(),
+            StateSK::Offloaded(state) => match state.eviction_state {
+                EvictionState::Offloaded(flush_lsn) => flush_lsn,
+                _ => panic!("StateSK::Offloaded mismatches with eviction_state from control_file"),
+            },
+            StateSK::Empty => unreachable!(),
+        }
+    }
+
+    /// Get a reference to the control file's timeline state.
+    pub fn state(&self) -> &TimelineState<control_file::FileStorage> {
+        match self {
+            StateSK::Loaded(sk) => &sk.state,
+            StateSK::Offloaded(ref s) => s,
+            StateSK::Empty => unreachable!(),
+        }
+    }
+
+    pub fn state_mut(&mut self) -> &mut TimelineState<control_file::FileStorage> {
+        match self {
+            StateSK::Loaded(sk) => &mut sk.state,
+            StateSK::Offloaded(ref mut s) => s,
+            StateSK::Empty => unreachable!(),
+        }
+    }
+
+    pub fn last_log_term(&self) -> Term {
+        self.state()
+            .acceptor_state
+            .get_last_log_term(self.flush_lsn())
+    }
+
+    /// Close open WAL files to release FDs.
+    fn close_wal_store(&mut self) {
+        if let StateSK::Loaded(sk) = self {
+            sk.wal_store.close();
+        }
+    }
+
+    /// Update timeline state with peer safekeeper data.
+    pub async fn record_safekeeper_info(&mut self, sk_info: &SafekeeperTimelineInfo) -> Result<()> {
+        // update commit_lsn if safekeeper is loaded
+        match self {
+            StateSK::Loaded(sk) => sk.record_safekeeper_info(sk_info).await?,
+            StateSK::Offloaded(_) => {}
+            StateSK::Empty => unreachable!(),
+        }
+
+        // update everything else, including remote_consistent_lsn and backup_lsn
+        let mut sync_control_file = false;
+        let state = self.state_mut();
+        let wal_seg_size = state.server.wal_seg_size as u64;
+
+        state.inmem.backup_lsn = max(Lsn(sk_info.backup_lsn), state.inmem.backup_lsn);
+        sync_control_file |= state.backup_lsn + wal_seg_size < state.inmem.backup_lsn;
+
+        state.inmem.remote_consistent_lsn = max(
+            Lsn(sk_info.remote_consistent_lsn),
+            state.inmem.remote_consistent_lsn,
+        );
+        sync_control_file |=
+            state.remote_consistent_lsn + wal_seg_size < state.inmem.remote_consistent_lsn;
+
+        state.inmem.peer_horizon_lsn =
+            max(Lsn(sk_info.peer_horizon_lsn), state.inmem.peer_horizon_lsn);
+        sync_control_file |= state.peer_horizon_lsn + wal_seg_size < state.inmem.peer_horizon_lsn;
+
+        if sync_control_file {
+            state.flush().await?;
+        }
+        Ok(())
+    }
+
+    /// Previously known as epoch_start_lsn. Needed only for reference in some APIs.
+    pub fn term_start_lsn(&self) -> Lsn {
+        match self {
+            StateSK::Loaded(sk) => sk.term_start_lsn,
+            StateSK::Offloaded(_) => Lsn(0),
+            StateSK::Empty => unreachable!(),
+        }
+    }
+
+    /// Used for metrics only.
+    pub fn wal_storage_metrics(&self) -> WalStorageMetrics {
+        match self {
+            StateSK::Loaded(sk) => sk.wal_store.get_metrics(),
+            StateSK::Offloaded(_) => WalStorageMetrics::default(),
+            StateSK::Empty => unreachable!(),
+        }
+    }
+
+    /// Returns WAL storage internal LSNs for debug dump.
+    pub fn wal_storage_internal_state(&self) -> (Lsn, Lsn, Lsn, bool) {
+        match self {
+            StateSK::Loaded(sk) => sk.wal_store.internal_state(),
+            StateSK::Offloaded(_) => {
+                let flush_lsn = self.flush_lsn();
+                (flush_lsn, flush_lsn, flush_lsn, false)
+            }
+            StateSK::Empty => unreachable!(),
+        }
+    }
+
+    /// Access to SafeKeeper object. Panics if offloaded, should be good to use from WalResidentTimeline.
+    pub fn safekeeper(
+        &mut self,
+    ) -> &mut SafeKeeper<control_file::FileStorage, wal_storage::PhysicalStorage> {
+        match self {
+            StateSK::Loaded(sk) => sk,
+            StateSK::Offloaded(_) => {
+                panic!("safekeeper is offloaded, cannot be used")
+            }
+            StateSK::Empty => unreachable!(),
+        }
+    }
+
+    /// Moves control file's state structure out of the enum. Used to switch states.
+    fn take_state(self) -> TimelineState<control_file::FileStorage> {
+        match self {
+            StateSK::Loaded(sk) => sk.state,
+            StateSK::Offloaded(state) => *state,
+            StateSK::Empty => unreachable!(),
+        }
+    }
+}
+
 /// Shared state associated with database instance
 pub struct SharedState {
    /// Safekeeper object
-    pub(crate) sk: SafeKeeper<control_file::FileStorage, wal_storage::PhysicalStorage>,
+    pub(crate) sk: StateSK,
    /// In memory list containing state of peers sent in latest messages from them.
    pub(crate) peers_info: PeersInfo,
    // True value hinders old WAL removal; this is used by snapshotting. We
@@ -203,10 +347,10 @@ impl SharedState {
            control_file::FileStorage::create_new(timeline_dir.clone(), conf, state)?;
        let wal_store =
            wal_storage::PhysicalStorage::new(ttid, timeline_dir, conf, &control_store)?;
-        let sk = SafeKeeper::new(control_store, wal_store, conf.my_id)?;
+        let sk = SafeKeeper::new(TimelineState::new(control_store), wal_store, conf.my_id)?;

        Ok(Self {
-            sk,
+            sk: StateSK::Loaded(sk),
            peers_info: PeersInfo(vec![]),
            wal_removal_on_hold: false,
        })
@@ -220,18 +364,30 @@ impl SharedState {
            bail!(TimelineError::UninitializedWalSegSize(*ttid));
        }

-        let wal_store =
-            wal_storage::PhysicalStorage::new(ttid, timeline_dir, conf, &control_store)?;
+        let sk = match control_store.eviction_state {
+            EvictionState::Present => {
+                let wal_store =
+                    wal_storage::PhysicalStorage::new(ttid, timeline_dir, conf, &control_store)?;
+                StateSK::Loaded(SafeKeeper::new(
+                    TimelineState::new(control_store),
+                    wal_store,
+                    conf.my_id,
+                )?)
+            }
+            EvictionState::Offloaded(_) => {
+                StateSK::Offloaded(Box::new(TimelineState::new(control_store)))
+            }
+        };

        Ok(Self {
-            sk: SafeKeeper::new(control_store, wal_store, conf.my_id)?,
+            sk,
            peers_info: PeersInfo(vec![]),
            wal_removal_on_hold: false,
        })
    }

    pub(crate) fn get_wal_seg_size(&self) -> usize {
-        self.sk.state.server.wal_seg_size as usize
+        self.sk.state().server.wal_seg_size as usize
    }

    fn get_safekeeper_info(
@@ -246,20 +402,20 @@ impl SharedState {
                tenant_id: ttid.tenant_id.as_ref().to_owned(),
                timeline_id: ttid.timeline_id.as_ref().to_owned(),
            }),
-            term: self.sk.state.acceptor_state.term,
-            last_log_term: self.sk.get_last_log_term(),
+            term: self.sk.state().acceptor_state.term,
+            last_log_term: self.sk.last_log_term(),
            flush_lsn: self.sk.flush_lsn().0,
            // note: this value is not flushed to control file yet and can be lost
-            commit_lsn: self.sk.state.inmem.commit_lsn.0,
-            remote_consistent_lsn: self.sk.state.inmem.remote_consistent_lsn.0,
-            peer_horizon_lsn: self.sk.state.inmem.peer_horizon_lsn.0,
+            commit_lsn: self.sk.state().inmem.commit_lsn.0,
+            remote_consistent_lsn: self.sk.state().inmem.remote_consistent_lsn.0,
+            peer_horizon_lsn: self.sk.state().inmem.peer_horizon_lsn.0,
            safekeeper_connstr: conf
                .advertise_pg_addr
                .to_owned()
                .unwrap_or(conf.listen_pg_addr.clone()),
            http_connstr: conf.listen_http_addr.to_owned(),
-            backup_lsn: self.sk.state.inmem.backup_lsn.0,
-            local_start_lsn: self.sk.state.local_start_lsn.0,
+            backup_lsn: self.sk.state().inmem.backup_lsn.0,
+            local_start_lsn: self.sk.state().local_start_lsn.0,
            availability_zone: conf.availability_zone.clone(),
            standby_horizon: standby_apply_lsn.0,
        }
@@ -335,6 +491,7 @@ pub struct Timeline {
    walsenders: Arc<WalSenders>,
    walreceivers: Arc<WalReceivers>,
    timeline_dir: Utf8PathBuf,
+    manager_ctl: ManagerCtl,

    /// Delete/cancel will trigger this, background tasks should drop out as soon as it fires
    pub(crate) cancel: CancellationToken,
@@ -343,6 +500,7 @@ pub struct Timeline {
    pub(crate) broker_active: AtomicBool,
    pub(crate) wal_backup_active: AtomicBool,
    pub(crate) last_removed_segno: AtomicU64,
+    pub(crate) mgr_status: AtomicStatus,
 }

 impl Timeline {
@@ -352,9 +510,9 @@ impl Timeline {

        let shared_state = SharedState::restore(conf, &ttid)?;
        let (commit_lsn_watch_tx, commit_lsn_watch_rx) =
-            watch::channel(shared_state.sk.state.commit_lsn);
+            watch::channel(shared_state.sk.state().commit_lsn);
        let (term_flush_lsn_watch_tx, term_flush_lsn_watch_rx) = watch::channel(TermLsn::from((
-            shared_state.sk.get_term(),
+            shared_state.sk.last_log_term(),
            shared_state.sk.flush_lsn(),
        )));
        let (shared_state_version_tx, shared_state_version_rx) = watch::channel(0);
@@ -373,9 +531,11 @@ impl Timeline {
            walreceivers,
            cancel: CancellationToken::default(),
            timeline_dir: get_timeline_dir(conf, &ttid),
+            manager_ctl: ManagerCtl::new(),
            broker_active: AtomicBool::new(false),
            wal_backup_active: AtomicBool::new(false),
            last_removed_segno: AtomicU64::new(0),
+            mgr_status: AtomicStatus::new(),
        })
    }

@@ -409,9 +569,11 @@ impl Timeline {
            walreceivers,
            cancel: CancellationToken::default(),
            timeline_dir: get_timeline_dir(conf, &ttid),
+            manager_ctl: ManagerCtl::new(),
            broker_active: AtomicBool::new(false),
            wal_backup_active: AtomicBool::new(false),
            last_removed_segno: AtomicU64::new(0),
+            mgr_status: AtomicStatus::new(),
        })
    }

@@ -425,6 +587,7 @@ impl Timeline {
        shared_state: &mut WriteGuardSharedState<'_>,
        conf: &SafeKeeperConf,
        broker_active_set: Arc<TimelinesSet>,
+        partial_backup_rate_limiter: RateLimiter,
    ) -> Result<()> {
        match fs::metadata(&self.timeline_dir).await {
            Ok(_) => {
@@ -442,7 +605,7 @@ impl Timeline {
        fs::create_dir_all(&self.timeline_dir).await?;

        // Write timeline to disk and start background tasks.
-        if let Err(e) = shared_state.sk.state.flush().await {
+        if let Err(e) = shared_state.sk.state_mut().flush().await {
            // Bootstrap failed, cancel timeline and remove timeline directory.
            self.cancel(shared_state);

@@ -455,7 +618,7 @@ impl Timeline {

            return Err(e);
        }
-        self.bootstrap(conf, broker_active_set);
+        self.bootstrap(conf, broker_active_set, partial_backup_rate_limiter);
        Ok(())
    }

@@ -464,13 +627,19 @@ impl Timeline {
        self: &Arc<Timeline>,
        conf: &SafeKeeperConf,
        broker_active_set: Arc<TimelinesSet>,
+        partial_backup_rate_limiter: RateLimiter,
    ) {
+        let (tx, rx) = self.manager_ctl.bootstrap_manager();
+
        // Start manager task which will monitor timeline state and update
        // background tasks.
        tokio::spawn(timeline_manager::main_task(
-            self.clone(),
+            ManagerTimeline { tli: self.clone() },
            conf.clone(),
            broker_active_set,
+            tx,
+            rx,
+            partial_backup_rate_limiter,
        ));
    }

@@ -507,7 +676,7 @@ impl Timeline {
        self.cancel.cancel();
        // Close associated FDs. Nobody will be able to touch timeline data once
        // it is cancelled, so WAL storage won't be opened again.
-        shared_state.sk.wal_store.close();
+        shared_state.sk.close_wal_store();
    }

    /// Returns if timeline is cancelled.
@@ -547,12 +716,15 @@ impl Timeline {
    /// Returns state of the timeline.
    pub async fn get_state(&self) -> (TimelineMemState, TimelinePersistentState) {
        let state = self.read_shared_state().await;
-        (state.sk.state.inmem.clone(), state.sk.state.clone())
+        (
+            state.sk.state().inmem.clone(),
+            TimelinePersistentState::clone(state.sk.state()),
+        )
    }

    /// Returns latest backup_lsn.
    pub async fn get_wal_backup_lsn(&self) -> Lsn {
-        self.read_shared_state().await.sk.state.inmem.backup_lsn
+        self.read_shared_state().await.sk.state().inmem.backup_lsn
    }

    /// Sets backup_lsn to the given value.
@@ -562,7 +734,7 @@ impl Timeline {
        }

        let mut state = self.write_shared_state().await;
-        state.sk.state.inmem.backup_lsn = max(state.sk.state.inmem.backup_lsn, backup_lsn);
+        state.sk.state_mut().inmem.backup_lsn = max(state.sk.state().inmem.backup_lsn, backup_lsn);
        // we should check whether to shut down offloader, but this will be done
        // soon by peer communication anyway.
        Ok(())
@@ -604,7 +776,7 @@ impl Timeline {

    /// Returns flush_lsn.
    pub async fn get_flush_lsn(&self) -> Lsn {
-        self.read_shared_state().await.sk.wal_store.flush_lsn()
+        self.read_shared_state().await.sk.flush_lsn()
    }

    /// Gather timeline data for metrics.
@@ -623,11 +795,11 @@ impl Timeline {
            timeline_is_active: self.broker_active.load(Ordering::Relaxed),
            num_computes: self.walreceivers.get_num() as u32,
            last_removed_segno: self.last_removed_segno.load(Ordering::Relaxed),
-            epoch_start_lsn: state.sk.term_start_lsn,
-            mem_state: state.sk.state.inmem.clone(),
-            persisted_state: state.sk.state.clone(),
-            flush_lsn: state.sk.wal_store.flush_lsn(),
-            wal_storage: state.sk.wal_store.get_metrics(),
+            epoch_start_lsn: state.sk.term_start_lsn(),
+            mem_state: state.sk.state().inmem.clone(),
+            persisted_state: TimelinePersistentState::clone(state.sk.state()),
+            flush_lsn: state.sk.flush_lsn(),
+            wal_storage: state.sk.wal_storage_metrics(),
        })
    }

@@ -636,7 +808,7 @@ impl Timeline {
        let state = self.read_shared_state().await;

        let (write_lsn, write_record_lsn, flush_lsn, file_open) =
-            state.sk.wal_store.internal_state();
+            state.sk.wal_storage_internal_state();

        debug_dump::Memory {
            is_cancelled: self.is_cancelled(),
@@ -646,8 +818,9 @@ impl Timeline {
            active: self.broker_active.load(Ordering::Relaxed),
            num_computes: self.walreceivers.get_num() as u32,
            last_removed_segno: self.last_removed_segno.load(Ordering::Relaxed),
-            epoch_start_lsn: state.sk.term_start_lsn,
-            mem_state: state.sk.state.inmem.clone(),
+            epoch_start_lsn: state.sk.term_start_lsn(),
+            mem_state: state.sk.state().inmem.clone(),
+            mgr_status: self.mgr_status.get(),
            write_lsn,
            write_record_lsn,
            flush_lsn,
@@ -661,34 +834,89 @@ impl Timeline {
        f: impl FnOnce(&mut TimelinePersistentState) -> Result<T>,
    ) -> Result<T> {
        let mut state = self.write_shared_state().await;
-        let mut persistent_state = state.sk.state.start_change();
+        let mut persistent_state = state.sk.state_mut().start_change();
        // If f returns error, we abort the change and don't persist anything.
        let res = f(&mut persistent_state)?;
        // If persisting fails, we abort the change and return error.
-        state.sk.state.finish_change(&persistent_state).await?;
+        state
+            .sk
+            .state_mut()
+            .finish_change(&persistent_state)
+            .await?;
        Ok(res)
    }

    /// Get the timeline guard for reading/writing WAL files.
-    /// TODO: if WAL files are not present on disk (evicted), they will be
-    /// downloaded from S3. Also there will logic for preventing eviction
-    /// while someone is holding FullAccessTimeline guard.
-    pub async fn full_access_guard(self: &Arc<Self>) -> Result<FullAccessTimeline> {
+    /// If WAL files are not present on disk (evicted), they will be automatically
+    /// downloaded from remote storage. This is done in the manager task, which is
+    /// responsible for issuing all guards.
+    ///
+    /// NB: don't use this function from timeline_manager, it will deadlock.
+    /// NB: don't use this function while holding shared_state lock.
+    pub async fn wal_residence_guard(self: &Arc<Self>) -> Result<WalResidentTimeline> {
        if self.is_cancelled() {
            bail!(TimelineError::Cancelled(self.ttid));
        }
-        Ok(FullAccessTimeline { tli: self.clone() })
+
+        debug!("requesting WalResidentTimeline guard");
+        let started_at = Instant::now();
+        let status_before = self.mgr_status.get();
+
+        // Wait 30 seconds for the guard to be acquired. It can time out if someone is
+        // holding the lock (e.g. during `SafeKeeper::process_msg()`) or manager task
+        // is stuck.
+        let res = tokio::time::timeout_at(
+            started_at + Duration::from_secs(30),
+            self.manager_ctl.wal_residence_guard(),
+        )
+        .await;
+
+        let guard = match res {
+            Ok(Ok(guard)) => {
+                let finished_at = Instant::now();
+                let elapsed = finished_at - started_at;
+                MISC_OPERATION_SECONDS
+                    .with_label_values(&["wal_residence_guard"])
+                    .observe(elapsed.as_secs_f64());
+
+                guard
+            }
+            Ok(Err(e)) => {
+                warn!(
+                    "error while acquiring WalResidentTimeline guard, statuses {:?} => {:?}",
+                    status_before,
+                    self.mgr_status.get()
+                );
+                return Err(e);
+            }
+            Err(_) => {
+                warn!(
+                    "timeout while acquiring WalResidentTimeline guard, statuses {:?} => {:?}",
+                    status_before,
+                    self.mgr_status.get()
+                );
+                anyhow::bail!("timeout while acquiring WalResidentTimeline guard");
+            }
+        };
+
+        Ok(WalResidentTimeline::new(self.clone(), guard))
    }
 }

 /// This is a guard that allows to read/write disk timeline state.
-/// All tasks that are using the disk should use this guard.
-#[derive(Clone)]
-pub struct FullAccessTimeline {
+/// All tasks that are trying to read/write WAL from disk should use this guard.
+pub struct WalResidentTimeline {
    pub tli: Arc<Timeline>,
+    _guard: ResidenceGuard,
 }

-impl Deref for FullAccessTimeline {
+impl WalResidentTimeline {
+    pub fn new(tli: Arc<Timeline>, _guard: ResidenceGuard) -> Self {
+        WalResidentTimeline { tli, _guard }
+    }
+}
+
+impl Deref for WalResidentTimeline {
    type Target = Arc<Timeline>;

    fn deref(&self) -> &Self::Target {
@@ -696,7 +924,7 @@ impl Deref for FullAccessTimeline {
    }
 }

-impl FullAccessTimeline {
+impl WalResidentTimeline {
    /// Returns true if walsender should stop sending WAL to pageserver. We
    /// terminate it if remote_consistent_lsn reached commit_lsn and there is no
    /// computes. While there might be nothing to stream already, we learn about
@@ -708,8 +936,8 @@ impl FullAccessTimeline {
        }
        let shared_state = self.read_shared_state().await;
        if self.walreceivers.get_num() == 0 {
-            return shared_state.sk.state.inmem.commit_lsn == Lsn(0) || // no data at all yet
-            reported_remote_consistent_lsn >= shared_state.sk.state.inmem.commit_lsn;
+            return shared_state.sk.state().inmem.commit_lsn == Lsn(0) || // no data at all yet
+            reported_remote_consistent_lsn >= shared_state.sk.state().inmem.commit_lsn;
        }
        false
    }
@@ -717,11 +945,11 @@ impl FullAccessTimeline {
    /// Ensure that current term is t, erroring otherwise, and lock the state.
    pub async fn acquire_term(&self, t: Term) -> Result<ReadGuardSharedState> {
        let ss = self.read_shared_state().await;
-        if ss.sk.state.acceptor_state.term != t {
+        if ss.sk.state().acceptor_state.term != t {
            bail!(
                "failed to acquire term {}, current term {}",
                t,
-                ss.sk.state.acceptor_state.term
+                ss.sk.state().acceptor_state.term
            );
        }
        Ok(ss)
@@ -739,7 +967,7 @@ impl FullAccessTimeline {
        let mut rmsg: Option<AcceptorProposerMessage>;
        {
            let mut shared_state = self.write_shared_state().await;
-            rmsg = shared_state.sk.process_msg(msg).await?;
+            rmsg = shared_state.sk.safekeeper().process_msg(msg).await?;

            // if this is AppendResponse, fill in proper hot standby feedback.
            if let Some(AcceptorProposerMessage::AppendResponse(ref mut resp)) = rmsg {
@@ -769,8 +997,141 @@ impl FullAccessTimeline {
    /// Update in memory remote consistent lsn.
    pub async fn update_remote_consistent_lsn(&self, candidate: Lsn) {
        let mut shared_state = self.write_shared_state().await;
-        shared_state.sk.state.inmem.remote_consistent_lsn =
-            max(shared_state.sk.state.inmem.remote_consistent_lsn, candidate);
+        shared_state.sk.state_mut().inmem.remote_consistent_lsn = max(
+            shared_state.sk.state().inmem.remote_consistent_lsn,
+            candidate,
+        );
+    }
+}
+
+/// This struct contains methods that are used by timeline manager task.
+pub(crate) struct ManagerTimeline {
+    pub(crate) tli: Arc<Timeline>,
+}
+
+impl Deref for ManagerTimeline {
+    type Target = Arc<Timeline>;
+
+    fn deref(&self) -> &Self::Target {
+        &self.tli
+    }
+}
+
+impl ManagerTimeline {
+    pub(crate) fn timeline_dir(&self) -> &Utf8PathBuf {
+        &self.tli.timeline_dir
+    }
+
+    /// Manager requests this state on startup.
+    pub(crate) async fn bootstrap_mgr(&self) -> (bool, Option<PartialRemoteSegment>) {
+        let shared_state = self.read_shared_state().await;
+        let is_offloaded = matches!(
+            shared_state.sk.state().eviction_state,
+            EvictionState::Offloaded(_)
+        );
+        let partial_backup_uploaded = shared_state.sk.state().partial_backup.uploaded_segment();
+
+        (is_offloaded, partial_backup_uploaded)
+    }
+
+    /// Try to switch state Present->Offloaded.
+    pub(crate) async fn switch_to_offloaded(
+        &self,
+        partial: &PartialRemoteSegment,
+    ) -> anyhow::Result<()> {
+        let mut shared = self.write_shared_state().await;
+
+        // updating control file
+        let mut pstate = shared.sk.state_mut().start_change();
+
+        if !matches!(pstate.eviction_state, EvictionState::Present) {
+            bail!(
+                "cannot switch to offloaded state, current state is {:?}",
+                pstate.eviction_state
+            );
+        }
+
+        if partial.flush_lsn != shared.sk.flush_lsn() {
+            bail!(
+                "flush_lsn mismatch in partial backup, expected {}, got {}",
+                shared.sk.flush_lsn(),
+                partial.flush_lsn
+            );
+        }
+
+        if partial.commit_lsn != pstate.commit_lsn {
+            bail!(
+                "commit_lsn mismatch in partial backup, expected {}, got {}",
+                pstate.commit_lsn,
+                partial.commit_lsn
+            );
+        }
+
+        if partial.term != shared.sk.last_log_term() {
+            bail!(
+                "term mismatch in partial backup, expected {}, got {}",
+                shared.sk.last_log_term(),
+                partial.term
+            );
+        }
+
+        pstate.eviction_state = EvictionState::Offloaded(shared.sk.flush_lsn());
+        shared.sk.state_mut().finish_change(&pstate).await?;
+        // control file is now switched to Offloaded state
+
+        // now we can switch shared.sk to Offloaded, shouldn't fail
+        let prev_sk = std::mem::replace(&mut shared.sk, StateSK::Empty);
+        let cfile_state = prev_sk.take_state();
+        shared.sk = StateSK::Offloaded(Box::new(cfile_state));
+
+        Ok(())
+    }
+
+    /// Try to switch state Offloaded->Present.
+    pub(crate) async fn switch_to_present(&self) -> anyhow::Result<()> {
+        let conf = GlobalTimelines::get_global_config();
+        let mut shared = self.write_shared_state().await;
+
+        // trying to restore WAL storage
+        let wal_store = wal_storage::PhysicalStorage::new(
+            &self.ttid,
+            self.timeline_dir.clone(),
+            &conf,
+            shared.sk.state(),
+        )?;
+
+        // updating control file
+        let mut pstate = shared.sk.state_mut().start_change();
+
+        if !matches!(pstate.eviction_state, EvictionState::Offloaded(_)) {
+            bail!(
+                "cannot switch to present state, current state is {:?}",
+                pstate.eviction_state
+            );
+        }
+
+        if wal_store.flush_lsn() != shared.sk.flush_lsn() {
+            bail!(
+                "flush_lsn mismatch in restored WAL, expected {}, got {}",
+                shared.sk.flush_lsn(),
+                wal_store.flush_lsn()
+            );
+        }
+
+        pstate.eviction_state = EvictionState::Present;
+        shared.sk.state_mut().finish_change(&pstate).await?;
+
+        // now we can switch shared.sk to Present, shouldn't fail
+        let prev_sk = std::mem::replace(&mut shared.sk, StateSK::Empty);
+        let cfile_state = prev_sk.take_state();
+        shared.sk = StateSK::Loaded(SafeKeeper::new(cfile_state, wal_store, conf.my_id)?);
+
+        Ok(())
+    }
+
+    /// Update current manager state, useful for debugging manager deadlocks.
+    pub(crate) fn set_status(&self, status: timeline_manager::Status) {
+        self.mgr_status.store(status, Ordering::Relaxed);
    }
 }

@@ -784,13 +1145,13 @@ async fn delete_dir(path: &Utf8PathBuf) -> Result<bool> {
 }

 /// Get a path to the tenant directory. If you just need to get a timeline directory,
-/// use FullAccessTimeline::get_timeline_dir instead.
+/// use WalResidentTimeline::get_timeline_dir instead.
 pub(crate) fn get_tenant_dir(conf: &SafeKeeperConf, tenant_id: &TenantId) -> Utf8PathBuf {
    conf.workdir.join(tenant_id.to_string())
 }

 /// Get a path to the timeline directory. If you need to read WAL files from disk,
-/// use FullAccessTimeline::get_timeline_dir instead. This function does not check
+/// use WalResidentTimeline::get_timeline_dir instead. This function does not check
 /// timeline eviction status and WAL files might not be present on disk.
 pub(crate) fn get_timeline_dir(conf: &SafeKeeperConf, ttid: &TenantTimelineId) -> Utf8PathBuf {
    get_tenant_dir(conf, &ttid.tenant_id).join(ttid.timeline_id.to_string())
@@ -0,0 +1,366 @@
+//! Code related to evicting WAL files to remote storage. The actual upload is done by the
+//! partial WAL backup code. This file has code to delete and re-download WAL files,
+//! cross-validate with partial WAL backup if local file is still present.
+
+use anyhow::Context;
+use camino::Utf8PathBuf;
+use remote_storage::RemotePath;
+use tokio::{
+    fs::File,
+    io::{AsyncRead, AsyncWriteExt},
+};
+use tracing::{debug, info, instrument, warn};
+use utils::crashsafe::durable_rename;
+
+use crate::{
+    timeline_manager::{Manager, StateSnapshot},
+    wal_backup,
+    wal_backup_partial::{self, PartialRemoteSegment},
+    wal_storage::wal_file_paths,
+};
+
+impl Manager {
+    /// Returns true if the timeline is ready for eviction.
+    /// Current criteria:
+    /// - no active tasks
+    /// - control file is flushed (no next event scheduled)
+    /// - no WAL residence guards
+    /// - no pushes to the broker
+    /// - partial WAL backup is uploaded
+    pub(crate) fn ready_for_eviction(
+        &self,
+        next_event: &Option<tokio::time::Instant>,
+        state: &StateSnapshot,
+    ) -> bool {
+        self.backup_task.is_none()
+            && self.recovery_task.is_none()
+            && self.wal_removal_task.is_none()
+            && self.partial_backup_task.is_none()
+            && self.partial_backup_uploaded.is_some()
+            && next_event.is_none()
+            && self.access_service.is_empty()
+            && !self.tli_broker_active.get()
+            && !wal_backup_partial::needs_uploading(state, &self.partial_backup_uploaded)
+            && self
+                .partial_backup_uploaded
+                .as_ref()
+                .unwrap()
+                .flush_lsn
+                .segment_number(self.wal_seg_size)
+                == self.last_removed_segno + 1
+    }
+
+    /// Evict the timeline to remote storage.
+    #[instrument(name = "evict_timeline", skip_all)]
+    pub(crate) async fn evict_timeline(&mut self) {
+        assert!(!self.is_offloaded);
+        let partial_backup_uploaded = match &self.partial_backup_uploaded {
+            Some(p) => p.clone(),
+            None => {
+                warn!("no partial backup uploaded, skipping eviction");
+                return;
+            }
+        };
+
+        info!("starting eviction, using {:?}", partial_backup_uploaded);
+
+        if let Err(e) = do_eviction(self, &partial_backup_uploaded).await {
+            warn!("failed to evict timeline: {:?}", e);
+            return;
+        }
+
+        info!("successfully evicted timeline");
+    }
+
+    /// Restore evicted timeline from remote storage.
+    #[instrument(name = "unevict_timeline", skip_all)]
+    pub(crate) async fn unevict_timeline(&mut self) {
+        assert!(self.is_offloaded);
+        let partial_backup_uploaded = match &self.partial_backup_uploaded {
+            Some(p) => p.clone(),
+            None => {
+                warn!("no partial backup uploaded, cannot unevict");
+                return;
+            }
+        };
+
+        info!("starting uneviction, using {:?}", partial_backup_uploaded);
+
+        if let Err(e) = do_uneviction(self, &partial_backup_uploaded).await {
+            warn!("failed to unevict timeline: {:?}", e);
+            return;
+        }
+
+        info!("successfully restored evicted timeline");
+    }
+}
+
+/// Ensure that content matches the remote partial backup, if local segment exists.
+/// Then change state in control file and in-memory. If `delete_offloaded_wal` is set,
+/// delete the local segment.
+async fn do_eviction(mgr: &mut Manager, partial: &PartialRemoteSegment) -> anyhow::Result<()> {
+    compare_local_segment_with_remote(mgr, partial).await?;
+
+    mgr.tli.switch_to_offloaded(partial).await?;
+    // switch manager state as soon as possible
+    mgr.is_offloaded = true;
+
+    if mgr.conf.delete_offloaded_wal {
+        delete_local_segment(mgr, partial).await?;
+    }
+
+    Ok(())
+}
+
+/// Ensure that content matches the remote partial backup, if local segment exists.
+/// Then download segment to local disk and change state in control file and in-memory.
+async fn do_uneviction(mgr: &mut Manager, partial: &PartialRemoteSegment) -> anyhow::Result<()> {
+    // if the local segment is present, validate it
+    compare_local_segment_with_remote(mgr, partial).await?;
+
+    // atomically download the partial segment
+    redownload_partial_segment(mgr, partial).await?;
+
+    mgr.tli.switch_to_present().await?;
+    // switch manager state as soon as possible
+    mgr.is_offloaded = false;
+
+    Ok(())
+}
+
+/// Delete local WAL segment.
+async fn delete_local_segment(mgr: &Manager, partial: &PartialRemoteSegment) -> anyhow::Result<()> {
+    let local_path = local_segment_path(mgr, partial);
+
+    info!("deleting WAL file to evict: {}", local_path);
+    tokio::fs::remove_file(&local_path).await?;
+    Ok(())
+}
+
+/// Redownload partial segment from remote storage.
+/// The segment is downloaded to a temporary file and then renamed to the final path.
+async fn redownload_partial_segment(
+    mgr: &Manager,
+    partial: &PartialRemoteSegment,
+) -> anyhow::Result<()> {
+    let tmp_file = mgr.tli.timeline_dir().join("remote_partial.tmp");
+    let remote_segfile = remote_segment_path(mgr, partial)?;
+
+    debug!(
+        "redownloading partial segment: {} -> {}",
+        remote_segfile, tmp_file
+    );
+
+    let mut reader = wal_backup::read_object(&remote_segfile, 0).await?;
+    let mut file = File::create(&tmp_file).await?;
+
+    let actual_len = tokio::io::copy(&mut reader, &mut file).await?;
+    let expected_len = partial.flush_lsn.segment_offset(mgr.wal_seg_size);
+
+    if actual_len != expected_len as u64 {
+        anyhow::bail!(
+            "partial downloaded {} bytes, expected {}",
+            actual_len,
+            expected_len
+        );
+    }
+
+    if actual_len > mgr.wal_seg_size as u64 {
+        anyhow::bail!(
+            "remote segment is too long: {} bytes, expected {}",
+            actual_len,
+            mgr.wal_seg_size
+        );
+    }
+    file.set_len(mgr.wal_seg_size as u64).await?;
+    file.flush().await?;
+
+    let final_path = local_segment_path(mgr, partial);
+    info!(
+        "downloaded {} bytes, renaming to {}",
+        final_path, final_path,
+    );
+    if let Err(e) = durable_rename(&tmp_file, &final_path, !mgr.conf.no_sync).await {
+        // Probably rename succeeded, but fsync of it failed. Remove
+        // the file then to avoid using it.
+        tokio::fs::remove_file(tmp_file)
+            .await
+            .or_else(utils::fs_ext::ignore_not_found)?;
+        return Err(e.into());
+    }
+
+    Ok(())
+}
+
+/// Compare local WAL segment with partial WAL backup in remote storage.
+/// If the local segment is not present, the function does nothing.
+/// If the local segment is present, it compares the local segment with the remote one.
+async fn compare_local_segment_with_remote(
+    mgr: &Manager,
+    partial: &PartialRemoteSegment,
+) -> anyhow::Result<()> {
+    let local_path = local_segment_path(mgr, partial);
+
+    match File::open(&local_path).await {
+        Ok(mut local_file) => do_validation(mgr, &mut local_file, mgr.wal_seg_size, partial)
+            .await
+            .context("validation failed"),
+        Err(_) => {
+            info!(
+                "local WAL file {} is not present, skipping validation",
+                local_path
+            );
+            Ok(())
+        }
+    }
+}
+
+/// Compare opened local WAL segment with partial WAL backup in remote storage.
+/// Validate full content of both files.
+async fn do_validation(
+    mgr: &Manager,
+    file: &mut File,
+    wal_seg_size: usize,
+    partial: &PartialRemoteSegment,
+) -> anyhow::Result<()> {
+    let local_size = file.metadata().await?.len() as usize;
+    if local_size != wal_seg_size {
+        anyhow::bail!(
+            "local segment size is invalid: found {}, expected {}",
+            local_size,
+            wal_seg_size
+        );
+    }
+
+    let remote_segfile = remote_segment_path(mgr, partial)?;
+    let mut remote_reader: std::pin::Pin<Box<dyn AsyncRead + Send + Sync>> =
+        wal_backup::read_object(&remote_segfile, 0).await?;
+
+    // remote segment should have bytes excatly up to `flush_lsn`
+    let expected_remote_size = partial.flush_lsn.segment_offset(mgr.wal_seg_size);
+    // let's compare the first `expected_remote_size` bytes
+    compare_n_bytes(&mut remote_reader, file, expected_remote_size).await?;
+    // and check that the remote segment ends here
+    check_end(&mut remote_reader).await?;
+
+    // if local segment is longer, the rest should be zeroes
+    read_n_zeroes(file, mgr.wal_seg_size - expected_remote_size).await?;
+    // and check that the local segment ends here
+    check_end(file).await?;
+
+    Ok(())
+}
+
+fn local_segment_path(mgr: &Manager, partial: &PartialRemoteSegment) -> Utf8PathBuf {
+    let flush_lsn = partial.flush_lsn;
+    let segno = flush_lsn.segment_number(mgr.wal_seg_size);
+    let (_, local_partial_segfile) =
+        wal_file_paths(mgr.tli.timeline_dir(), segno, mgr.wal_seg_size);
+    local_partial_segfile
+}
+
+fn remote_segment_path(
+    mgr: &Manager,
+    partial: &PartialRemoteSegment,
+) -> anyhow::Result<RemotePath> {
+    let remote_timeline_path = wal_backup::remote_timeline_path(&mgr.tli.ttid)?;
+    Ok(partial.remote_path(&remote_timeline_path))
+}
+
+/// Compare first `n` bytes of two readers. If the bytes differ, return an error.
+/// If the readers are shorter than `n`, return an error.
+async fn compare_n_bytes<R1, R2>(reader1: &mut R1, reader2: &mut R2, n: usize) -> anyhow::Result<()>
+where
+    R1: AsyncRead + Unpin,
+    R2: AsyncRead + Unpin,
+{
+    use tokio::io::AsyncReadExt;
+
+    const BUF_SIZE: usize = 32 * 1024;
+
+    let mut buffer1 = vec![0u8; BUF_SIZE];
+    let mut buffer2 = vec![0u8; BUF_SIZE];
+
+    let mut offset = 0;
+
+    while offset < n {
+        let bytes_to_read = std::cmp::min(BUF_SIZE, n - offset);
+
+        let bytes_read1 = reader1
+            .read(&mut buffer1[..bytes_to_read])
+            .await
+            .with_context(|| format!("failed to read from reader1 at offset {}", offset))?;
+        if bytes_read1 == 0 {
+            anyhow::bail!("unexpected EOF from reader1 at offset {}", offset);
+        }
+
+        let bytes_read2 = reader2
+            .read_exact(&mut buffer2[..bytes_read1])
+            .await
+            .with_context(|| {
+                format!(
+                    "failed to read {} bytes from reader2 at offset {}",
+                    bytes_read1, offset
+                )
+            })?;
+        assert!(bytes_read2 == bytes_read1);
+
+        if buffer1[..bytes_read1] != buffer2[..bytes_read2] {
+            let diff_offset = buffer1[..bytes_read1]
+                .iter()
+                .zip(buffer2[..bytes_read2].iter())
+                .position(|(a, b)| a != b)
+                .expect("mismatched buffers, but no difference found");
+            anyhow::bail!("mismatch at offset {}", offset + diff_offset);
+        }
+
+        offset += bytes_read1;
+    }
+
+    Ok(())
+}
+
+async fn check_end<R>(mut reader: R) -> anyhow::Result<()>
+where
+    R: AsyncRead + Unpin,
+{
+    use tokio::io::AsyncReadExt;
+
+    let mut buffer = [0u8; 1];
+    let bytes_read = reader.read(&mut buffer).await?;
+    if bytes_read != 0 {
+        anyhow::bail!("expected EOF, found bytes");
+    }
+    Ok(())
+}
+
+async fn read_n_zeroes<R>(reader: &mut R, n: usize) -> anyhow::Result<()>
+where
+    R: AsyncRead + Unpin,
+{
+    use tokio::io::AsyncReadExt;
+
+    const BUF_SIZE: usize = 32 * 1024;
+    let mut buffer = vec![0u8; BUF_SIZE];
+    let mut offset = 0;
+
+    while offset < n {
+        let bytes_to_read = std::cmp::min(BUF_SIZE, n - offset);
+
+        let bytes_read = reader
+            .read(&mut buffer[..bytes_to_read])
+            .await
+            .context("expected zeroes, got read error")?;
+        if bytes_read == 0 {
+            anyhow::bail!("expected zeroes, got EOF");
+        }
+
+        if buffer[..bytes_read].iter().all(|&b| b == 0) {
+            offset += bytes_read;
+        } else {
+            anyhow::bail!("non-zero byte found");
+        }
+    }
+
+    Ok(())
+}
@@ -0,0 +1,71 @@
+//! Timeline residence guard is needed to ensure that WAL segments are present on disk,
+//! as long as the code is holding the guard. This file implements guard logic, to issue
+//! and drop guards, and to notify the manager when the guard is dropped.
+
+use std::collections::HashSet;
+
+use tracing::{debug, warn};
+
+use crate::timeline_manager::ManagerCtlMessage;
+
+#[derive(Debug, Clone, Copy)]
+pub struct GuardId(u64);
+
+pub struct ResidenceGuard {
+    manager_tx: tokio::sync::mpsc::UnboundedSender<ManagerCtlMessage>,
+    guard_id: GuardId,
+}
+
+impl Drop for ResidenceGuard {
+    fn drop(&mut self) {
+        // notify the manager that the guard is dropped
+        let res = self
+            .manager_tx
+            .send(ManagerCtlMessage::GuardDrop(self.guard_id));
+        if let Err(e) = res {
+            warn!("failed to send GuardDrop message: {:?}", e);
+        }
+    }
+}
+
+/// AccessService is responsible for issuing and dropping residence guards.
+/// All guards are stored in the `guards` set.
+/// TODO: it's possible to add `String` name to each guard, for better observability.
+pub(crate) struct AccessService {
+    next_guard_id: u64,
+    guards: HashSet<u64>,
+    manager_tx: tokio::sync::mpsc::UnboundedSender<ManagerCtlMessage>,
+}
+
+impl AccessService {
+    pub(crate) fn new(manager_tx: tokio::sync::mpsc::UnboundedSender<ManagerCtlMessage>) -> Self {
+        Self {
+            next_guard_id: 0,
+            guards: HashSet::new(),
+            manager_tx,
+        }
+    }
+
+    pub(crate) fn is_empty(&self) -> bool {
+        self.guards.is_empty()
+    }
+
+    pub(crate) fn create_guard(&mut self) -> ResidenceGuard {
+        let guard_id = self.next_guard_id;
+        self.next_guard_id += 1;
+        self.guards.insert(guard_id);
+
+        let guard_id = GuardId(guard_id);
+        debug!("issued a new guard {:?}", guard_id);
+
+        ResidenceGuard {
+            manager_tx: self.manager_tx.clone(),
+            guard_id,
+        }
+    }
+
+    pub(crate) fn drop_guard(&mut self, guard_id: GuardId) {
+        debug!("dropping guard {:?}", guard_id);
+        assert!(self.guards.remove(&guard_id.0));
+    }
+}
@@ -2,66 +2,83 @@
 //! It is spawned alongside each timeline and exits when the timeline is deleted.
 //! It watches for changes in the timeline state and decides when to spawn or kill background tasks.
 //! It also can manage some reactive state, like should the timeline be active for broker pushes or not.
+//!
+//! Be aware that you need to be extra careful with manager code, because it is not respawned on panic.
+//! Also, if it will stuck in some branch, it will prevent any further progress in the timeline.

 use std::{
-    sync::Arc,
-    time::{Duration, Instant},
+    sync::{atomic::AtomicUsize, Arc},
+    time::Duration,
 };

 use postgres_ffi::XLogSegNo;
-use tokio::task::{JoinError, JoinHandle};
-use tracing::{info, info_span, instrument, warn, Instrument};
+use serde::{Deserialize, Serialize};
+use tokio::{
+    task::{JoinError, JoinHandle},
+    time::Instant,
+};
+use tracing::{debug, info, info_span, instrument, warn, Instrument};
 use utils::lsn::Lsn;

 use crate::{
-    control_file::Storage,
-    metrics::{MANAGER_ACTIVE_CHANGES, MANAGER_ITERATIONS_TOTAL},
+    control_file::{FileStorage, Storage},
+    metrics::{MANAGER_ACTIVE_CHANGES, MANAGER_ITERATIONS_TOTAL, MISC_OPERATION_SECONDS},
    recovery::recovery_main,
    remove_wal::calc_horizon_lsn,
+    safekeeper::Term,
    send_wal::WalSenders,
-    timeline::{PeerInfo, ReadGuardSharedState, Timeline},
+    state::TimelineState,
+    timeline::{ManagerTimeline, PeerInfo, ReadGuardSharedState, StateSK, WalResidentTimeline},
+    timeline_guard::{AccessService, GuardId, ResidenceGuard},
    timelines_set::{TimelineSetGuard, TimelinesSet},
    wal_backup::{self, WalBackupTaskHandle},
-    wal_backup_partial, SafeKeeperConf,
+    wal_backup_partial::{self, PartialRemoteSegment, RateLimiter},
+    SafeKeeperConf,
 };

-pub struct StateSnapshot {
+pub(crate) struct StateSnapshot {
    // inmem values
-    pub commit_lsn: Lsn,
-    pub backup_lsn: Lsn,
-    pub remote_consistent_lsn: Lsn,
+    pub(crate) commit_lsn: Lsn,
+    pub(crate) backup_lsn: Lsn,
+    pub(crate) remote_consistent_lsn: Lsn,

    // persistent control file values
-    pub cfile_peer_horizon_lsn: Lsn,
-    pub cfile_remote_consistent_lsn: Lsn,
-    pub cfile_backup_lsn: Lsn,
+    pub(crate) cfile_peer_horizon_lsn: Lsn,
+    pub(crate) cfile_remote_consistent_lsn: Lsn,
+    pub(crate) cfile_backup_lsn: Lsn,
+
+    // latest state
+    pub(crate) flush_lsn: Lsn,
+    pub(crate) last_log_term: Term,

    // misc
-    pub cfile_last_persist_at: Instant,
-    pub inmem_flush_pending: bool,
-    pub wal_removal_on_hold: bool,
-    pub peers: Vec<PeerInfo>,
+    pub(crate) cfile_last_persist_at: std::time::Instant,
+    pub(crate) inmem_flush_pending: bool,
+    pub(crate) wal_removal_on_hold: bool,
+    pub(crate) peers: Vec<PeerInfo>,
 }

 impl StateSnapshot {
    /// Create a new snapshot of the timeline state.
    fn new(read_guard: ReadGuardSharedState, heartbeat_timeout: Duration) -> Self {
+        let state = read_guard.sk.state();
        Self {
-            commit_lsn: read_guard.sk.state.inmem.commit_lsn,
-            backup_lsn: read_guard.sk.state.inmem.backup_lsn,
-            remote_consistent_lsn: read_guard.sk.state.inmem.remote_consistent_lsn,
-            cfile_peer_horizon_lsn: read_guard.sk.state.peer_horizon_lsn,
-            cfile_remote_consistent_lsn: read_guard.sk.state.remote_consistent_lsn,
-            cfile_backup_lsn: read_guard.sk.state.backup_lsn,
-            cfile_last_persist_at: read_guard.sk.state.pers.last_persist_at(),
-            inmem_flush_pending: Self::has_unflushed_inmem_state(&read_guard),
+            commit_lsn: state.inmem.commit_lsn,
+            backup_lsn: state.inmem.backup_lsn,
+            remote_consistent_lsn: state.inmem.remote_consistent_lsn,
+            cfile_peer_horizon_lsn: state.peer_horizon_lsn,
+            cfile_remote_consistent_lsn: state.remote_consistent_lsn,
+            cfile_backup_lsn: state.backup_lsn,
+            flush_lsn: read_guard.sk.flush_lsn(),
+            last_log_term: read_guard.sk.last_log_term(),
+            cfile_last_persist_at: state.pers.last_persist_at(),
+            inmem_flush_pending: Self::has_unflushed_inmem_state(state),
            wal_removal_on_hold: read_guard.wal_removal_on_hold,
            peers: read_guard.get_peers(heartbeat_timeout),
        }
    }

-    fn has_unflushed_inmem_state(read_guard: &ReadGuardSharedState) -> bool {
-        let state = &read_guard.sk.state;
+    fn has_unflushed_inmem_state(state: &TimelineState<FileStorage>) -> bool {
        state.inmem.commit_lsn > state.commit_lsn
            || state.inmem.backup_lsn > state.backup_lsn
            || state.inmem.peer_horizon_lsn > state.peer_horizon_lsn
@@ -73,314 +90,576 @@ impl StateSnapshot {
 /// There is no need to check for updates more often than this.
 const REFRESH_INTERVAL: Duration = Duration::from_millis(300);

-/// How often to save the control file if the is no other activity.
-const CF_SAVE_INTERVAL: Duration = Duration::from_secs(300);
+pub enum ManagerCtlMessage {
+    /// Request to get a guard for WalResidentTimeline, with WAL files available locally.
+    GuardRequest(tokio::sync::oneshot::Sender<anyhow::Result<ResidenceGuard>>),
+    /// Request to drop the guard.
+    GuardDrop(GuardId),
+}
+
+impl std::fmt::Debug for ManagerCtlMessage {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        match self {
+            ManagerCtlMessage::GuardRequest(_) => write!(f, "GuardRequest"),
+            ManagerCtlMessage::GuardDrop(id) => write!(f, "GuardDrop({:?})", id),
+        }
+    }
+}
+
+pub struct ManagerCtl {
+    manager_tx: tokio::sync::mpsc::UnboundedSender<ManagerCtlMessage>,
+
+    // this is used to initialize manager, it will be moved out in bootstrap().
+    init_manager_rx:
+        std::sync::Mutex<Option<tokio::sync::mpsc::UnboundedReceiver<ManagerCtlMessage>>>,
+}
+
+impl Default for ManagerCtl {
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
+impl ManagerCtl {
+    pub fn new() -> Self {
+        let (tx, rx) = tokio::sync::mpsc::unbounded_channel();
+        Self {
+            manager_tx: tx,
+            init_manager_rx: std::sync::Mutex::new(Some(rx)),
+        }
+    }
+
+    /// Issue a new guard and wait for manager to prepare the timeline.
+    /// Sends a message to the manager and waits for the response.
+    /// Can be blocked indefinitely if the manager is stuck.
+    pub async fn wal_residence_guard(&self) -> anyhow::Result<ResidenceGuard> {
+        let (tx, rx) = tokio::sync::oneshot::channel();
+        self.manager_tx.send(ManagerCtlMessage::GuardRequest(tx))?;
+
+        // wait for the manager to respond with the guard
+        rx.await
+            .map_err(|e| anyhow::anyhow!("response read fail: {:?}", e))
+            .and_then(std::convert::identity)
+    }
+
+    /// Must be called exactly once to bootstrap the manager.
+    pub fn bootstrap_manager(
+        &self,
+    ) -> (
+        tokio::sync::mpsc::UnboundedSender<ManagerCtlMessage>,
+        tokio::sync::mpsc::UnboundedReceiver<ManagerCtlMessage>,
+    ) {
+        let rx = self
+            .init_manager_rx
+            .lock()
+            .expect("mutex init_manager_rx poisoned")
+            .take()
+            .expect("manager already bootstrapped");
+
+        (self.manager_tx.clone(), rx)
+    }
+}
+
+pub(crate) struct Manager {
+    // configuration & dependencies
+    pub(crate) tli: ManagerTimeline,
+    pub(crate) conf: SafeKeeperConf,
+    pub(crate) wal_seg_size: usize,
+    pub(crate) walsenders: Arc<WalSenders>,
+
+    // current state
+    pub(crate) state_version_rx: tokio::sync::watch::Receiver<usize>,
+    pub(crate) num_computes_rx: tokio::sync::watch::Receiver<usize>,
+    pub(crate) tli_broker_active: TimelineSetGuard,
+    pub(crate) last_removed_segno: XLogSegNo,
+    pub(crate) is_offloaded: bool,
+
+    // background tasks
+    pub(crate) backup_task: Option<WalBackupTaskHandle>,
+    pub(crate) recovery_task: Option<JoinHandle<()>>,
+    pub(crate) wal_removal_task: Option<JoinHandle<anyhow::Result<u64>>>,
+
+    // partial backup
+    pub(crate) partial_backup_task: Option<JoinHandle<Option<PartialRemoteSegment>>>,
+    pub(crate) partial_backup_uploaded: Option<PartialRemoteSegment>,
+
+    // misc
+    pub(crate) access_service: AccessService,
+    pub(crate) partial_backup_rate_limiter: RateLimiter,
+}

 /// This task gets spawned alongside each timeline and is responsible for managing the timeline's
 /// background tasks.
 /// Be careful, this task is not respawned on panic, so it should not panic.
 #[instrument(name = "manager", skip_all, fields(ttid = %tli.ttid))]
 pub async fn main_task(
-    tli: Arc<Timeline>,
+    tli: ManagerTimeline,
    conf: SafeKeeperConf,
    broker_active_set: Arc<TimelinesSet>,
+    manager_tx: tokio::sync::mpsc::UnboundedSender<ManagerCtlMessage>,
+    mut manager_rx: tokio::sync::mpsc::UnboundedReceiver<ManagerCtlMessage>,
+    partial_backup_rate_limiter: RateLimiter,
 ) {
+    tli.set_status(Status::Started);
+
+    let defer_tli = tli.tli.clone();
    scopeguard::defer! {
-        if tli.is_cancelled() {
+        if defer_tli.is_cancelled() {
            info!("manager task finished");
        } else {
            warn!("manager task finished prematurely");
        }
    };

-    // configuration & dependencies
-    let wal_seg_size = tli.get_wal_seg_size().await;
-    let heartbeat_timeout = conf.heartbeat_timeout;
-    let walsenders = tli.get_walsenders();
-    let walreceivers = tli.get_walreceivers();
-
-    // current state
-    let mut state_version_rx = tli.get_state_version_rx();
-    let mut num_computes_rx = walreceivers.get_num_rx();
-    let mut tli_broker_active = broker_active_set.guard(tli.clone());
-    let mut last_removed_segno = 0 as XLogSegNo;
-
-    // list of background tasks
-    let mut backup_task: Option<WalBackupTaskHandle> = None;
-    let mut recovery_task: Option<JoinHandle<()>> = None;
-    let mut partial_backup_task: Option<JoinHandle<()>> = None;
-    let mut wal_removal_task: Option<JoinHandle<anyhow::Result<u64>>> = None;
+    let mut mgr = Manager::new(
+        tli,
+        conf,
+        broker_active_set,
+        manager_tx,
+        partial_backup_rate_limiter,
+    )
+    .await;

    // Start recovery task which always runs on the timeline.
-    if conf.peer_recovery_enabled {
-        match tli.full_access_guard().await {
-            Ok(tli) => {
-                recovery_task = Some(tokio::spawn(recovery_main(tli, conf.clone())));
-            }
-            Err(e) => {
-                warn!("failed to start recovery task: {:?}", e);
-            }
-        }
-    }
-
-    // Start partial backup task which always runs on the timeline.
-    if conf.is_wal_backup_enabled() && conf.partial_backup_enabled {
-        match tli.full_access_guard().await {
-            Ok(tli) => {
-                partial_backup_task = Some(tokio::spawn(wal_backup_partial::main_task(
-                    tli,
-                    conf.clone(),
-                )));
-            }
-            Err(e) => {
-                warn!("failed to start partial backup task: {:?}", e);
-            }
-        }
+    if !mgr.is_offloaded && mgr.conf.peer_recovery_enabled {
+        let tli = mgr.wal_resident_timeline();
+        mgr.recovery_task = Some(tokio::spawn(recovery_main(tli, mgr.conf.clone())));
    }

    let last_state = 'outer: loop {
        MANAGER_ITERATIONS_TOTAL.inc();

-        let state_snapshot = StateSnapshot::new(tli.read_shared_state().await, heartbeat_timeout);
-        let num_computes = *num_computes_rx.borrow();
+        mgr.set_status(Status::StateSnapshot);
+        let state_snapshot = mgr.state_snapshot().await;

-        let is_wal_backup_required = update_backup(
-            &conf,
-            &tli,
-            wal_seg_size,
-            num_computes,
-            &state_snapshot,
-            &mut backup_task,
-        )
-        .await;
+        let mut next_event: Option<Instant> = None;
+        if !mgr.is_offloaded {
+            let num_computes = *mgr.num_computes_rx.borrow();

-        let _is_active = update_is_active(
-            is_wal_backup_required,
-            num_computes,
-            &state_snapshot,
-            &mut tli_broker_active,
-            &tli,
-        );
+            mgr.set_status(Status::UpdateBackup);
+            let is_wal_backup_required = mgr.update_backup(num_computes, &state_snapshot).await;
+            mgr.update_is_active(is_wal_backup_required, num_computes, &state_snapshot);

-        let next_cfile_save = update_control_file_save(&state_snapshot, &tli).await;
+            mgr.set_status(Status::UpdateControlFile);
+            mgr.update_control_file_save(&state_snapshot, &mut next_event)
+                .await;

-        update_wal_removal(
-            &conf,
-            walsenders,
-            &tli,
-            wal_seg_size,
-            &state_snapshot,
-            last_removed_segno,
-            &mut wal_removal_task,
-        )
-        .await;
+            mgr.set_status(Status::UpdateWalRemoval);
+            mgr.update_wal_removal(&state_snapshot).await;

+            mgr.set_status(Status::UpdatePartialBackup);
+            mgr.update_partial_backup(&state_snapshot).await;
+
+            if mgr.conf.enable_offload && mgr.ready_for_eviction(&next_event, &state_snapshot) {
+                mgr.set_status(Status::EvictTimeline);
+                mgr.evict_timeline().await;
+            }
+        }
+
+        mgr.set_status(Status::Wait);
        // wait until something changes. tx channels are stored under Arc, so they will not be
        // dropped until the manager task is finished.
        tokio::select! {
-            _ = tli.cancel.cancelled() => {
+            _ = mgr.tli.cancel.cancelled() => {
                // timeline was deleted
                break 'outer state_snapshot;
            }
            _ = async {
                // don't wake up on every state change, but at most every REFRESH_INTERVAL
                tokio::time::sleep(REFRESH_INTERVAL).await;
-                let _ = state_version_rx.changed().await;
+                let _ = mgr.state_version_rx.changed().await;
            } => {
                // state was updated
            }
-            _ = num_computes_rx.changed() => {
+            _ = mgr.num_computes_rx.changed() => {
                // number of connected computes was updated
            }
-            _ = async {
-                if let Some(timeout) = next_cfile_save {
-                    tokio::time::sleep_until(timeout).await
-                } else {
-                    futures::future::pending().await
-                }
-            } => {
-                // it's time to save the control file
+            _ = sleep_until(&next_event) => {
+                // we were waiting for some event (e.g. cfile save)
            }
-            res = async {
-                if let Some(task) = &mut wal_removal_task {
-                    task.await
-                } else {
-                    futures::future::pending().await
-                }
-            } => {
+            res = await_task_finish(&mut mgr.wal_removal_task) => {
                // WAL removal task finished
-                wal_removal_task = None;
-                update_wal_removal_end(res, &tli, &mut last_removed_segno);
+                mgr.wal_removal_task = None;
+                mgr.update_wal_removal_end(res);
+            }
+            res = await_task_finish(&mut mgr.partial_backup_task) => {
+                // partial backup task finished
+                mgr.partial_backup_task = None;
+                mgr.update_partial_backup_end(res);
+            }
+
+            msg = manager_rx.recv() => {
+                mgr.set_status(Status::HandleMessage);
+                mgr.handle_message(msg).await;
            }
        }
    };
+    mgr.set_status(Status::Exiting);

    // remove timeline from the broker active set sooner, before waiting for background tasks
-    tli_broker_active.set(false);
+    mgr.tli_broker_active.set(false);

    // shutdown background tasks
-    if conf.is_wal_backup_enabled() {
-        wal_backup::update_task(&conf, &tli, false, &last_state, &mut backup_task).await;
+    if mgr.conf.is_wal_backup_enabled() {
+        wal_backup::update_task(&mut mgr, false, &last_state).await;
    }

-    if let Some(recovery_task) = recovery_task {
+    if let Some(recovery_task) = &mut mgr.recovery_task {
        if let Err(e) = recovery_task.await {
            warn!("recovery task failed: {:?}", e);
        }
    }

-    if let Some(partial_backup_task) = partial_backup_task {
+    if let Some(partial_backup_task) = &mut mgr.partial_backup_task {
        if let Err(e) = partial_backup_task.await {
            warn!("partial backup task failed: {:?}", e);
        }
    }

-    if let Some(wal_removal_task) = wal_removal_task {
+    if let Some(wal_removal_task) = &mut mgr.wal_removal_task {
        let res = wal_removal_task.await;
-        update_wal_removal_end(res, &tli, &mut last_removed_segno);
+        mgr.update_wal_removal_end(res);
    }
+
+    mgr.set_status(Status::Finished);
 }

-/// Spawns/kills backup task and returns true if backup is required.
-async fn update_backup(
-    conf: &SafeKeeperConf,
-    tli: &Arc<Timeline>,
-    wal_seg_size: usize,
-    num_computes: usize,
-    state: &StateSnapshot,
-    backup_task: &mut Option<WalBackupTaskHandle>,
-) -> bool {
-    let is_wal_backup_required =
-        wal_backup::is_wal_backup_required(wal_seg_size, num_computes, state);
-
-    if conf.is_wal_backup_enabled() {
-        wal_backup::update_task(conf, tli, is_wal_backup_required, state, backup_task).await;
+impl Manager {
+    async fn new(
+        tli: ManagerTimeline,
+        conf: SafeKeeperConf,
+        broker_active_set: Arc<TimelinesSet>,
+        manager_tx: tokio::sync::mpsc::UnboundedSender<ManagerCtlMessage>,
+        partial_backup_rate_limiter: RateLimiter,
+    ) -> Manager {
+        let (is_offloaded, partial_backup_uploaded) = tli.bootstrap_mgr().await;
+        Manager {
+            conf,
+            wal_seg_size: tli.get_wal_seg_size().await,
+            walsenders: tli.get_walsenders().clone(),
+            state_version_rx: tli.get_state_version_rx(),
+            num_computes_rx: tli.get_walreceivers().get_num_rx(),
+            tli_broker_active: broker_active_set.guard(tli.clone()),
+            last_removed_segno: 0,
+            is_offloaded,
+            backup_task: None,
+            recovery_task: None,
+            wal_removal_task: None,
+            partial_backup_task: None,
+            partial_backup_uploaded,
+            access_service: AccessService::new(manager_tx),
+            tli,
+            partial_backup_rate_limiter,
+        }
    }

-    // update the state in Arc<Timeline>
-    tli.wal_backup_active
-        .store(backup_task.is_some(), std::sync::atomic::Ordering::Relaxed);
-    is_wal_backup_required
-}
-
-/// Update is_active flag and returns its value.
-fn update_is_active(
-    is_wal_backup_required: bool,
-    num_computes: usize,
-    state: &StateSnapshot,
-    tli_broker_active: &mut TimelineSetGuard,
-    tli: &Arc<Timeline>,
-) -> bool {
-    let is_active = is_wal_backup_required
-        || num_computes > 0
-        || state.remote_consistent_lsn < state.commit_lsn;
-
-    // update the broker timeline set
-    if tli_broker_active.set(is_active) {
-        // write log if state has changed
-        info!(
-            "timeline active={} now, remote_consistent_lsn={}, commit_lsn={}",
-            is_active, state.remote_consistent_lsn, state.commit_lsn,
-        );
-
-        MANAGER_ACTIVE_CHANGES.inc();
+    fn set_status(&self, status: Status) {
+        self.tli.set_status(status);
    }

-    // update the state in Arc<Timeline>
-    tli.broker_active
-        .store(is_active, std::sync::atomic::Ordering::Relaxed);
-    is_active
-}
-
-/// Save control file if needed. Returns Instant if we should persist the control file in the future.
-async fn update_control_file_save(
-    state: &StateSnapshot,
-    tli: &Arc<Timeline>,
-) -> Option<tokio::time::Instant> {
-    if !state.inmem_flush_pending {
-        return None;
+    /// Get a WalResidentTimeline.
+    /// Manager code must use this function instead of one from `Timeline`
+    /// directly, because it will deadlock.
+    pub(crate) fn wal_resident_timeline(&mut self) -> WalResidentTimeline {
+        assert!(!self.is_offloaded);
+        let guard = self.access_service.create_guard();
+        WalResidentTimeline::new(self.tli.clone(), guard)
    }

-    if state.cfile_last_persist_at.elapsed() > CF_SAVE_INTERVAL {
-        let mut write_guard = tli.write_shared_state().await;
-        // this can be done in the background because it blocks manager task, but flush() should
-        // be fast enough not to be a problem now
-        if let Err(e) = write_guard.sk.state.flush().await {
-            warn!("failed to save control file: {:?}", e);
+    /// Get a snapshot of the timeline state.
+    async fn state_snapshot(&self) -> StateSnapshot {
+        let _timer = MISC_OPERATION_SECONDS
+            .with_label_values(&["state_snapshot"])
+            .start_timer();
+
+        StateSnapshot::new(
+            self.tli.read_shared_state().await,
+            self.conf.heartbeat_timeout,
+        )
+    }
+
+    /// Spawns/kills backup task and returns true if backup is required.
+    async fn update_backup(&mut self, num_computes: usize, state: &StateSnapshot) -> bool {
+        let is_wal_backup_required =
+            wal_backup::is_wal_backup_required(self.wal_seg_size, num_computes, state);
+
+        if self.conf.is_wal_backup_enabled() {
+            wal_backup::update_task(self, is_wal_backup_required, state).await;
        }

-        None
-    } else {
-        // we should wait until next CF_SAVE_INTERVAL
-        Some((state.cfile_last_persist_at + CF_SAVE_INTERVAL).into())
-    }
-}
-
-/// Spawns WAL removal task if needed.
-async fn update_wal_removal(
-    conf: &SafeKeeperConf,
-    walsenders: &Arc<WalSenders>,
-    tli: &Arc<Timeline>,
-    wal_seg_size: usize,
-    state: &StateSnapshot,
-    last_removed_segno: u64,
-    wal_removal_task: &mut Option<JoinHandle<anyhow::Result<u64>>>,
-) {
-    if wal_removal_task.is_some() || state.wal_removal_on_hold {
-        // WAL removal is already in progress or hold off
-        return;
-    }
-
-    // If enabled, we use LSN of the most lagging walsender as a WAL removal horizon.
-    // This allows to get better read speed for pageservers that are lagging behind,
-    // at the cost of keeping more WAL on disk.
-    let replication_horizon_lsn = if conf.walsenders_keep_horizon {
-        walsenders.laggard_lsn()
-    } else {
-        None
-    };
-
-    let removal_horizon_lsn = calc_horizon_lsn(state, replication_horizon_lsn);
-    let removal_horizon_segno = removal_horizon_lsn
-        .segment_number(wal_seg_size)
-        .saturating_sub(1);
-
-    if removal_horizon_segno > last_removed_segno {
-        // we need to remove WAL
-        let remover = crate::wal_storage::Storage::remove_up_to(
-            &tli.read_shared_state().await.sk.wal_store,
-            removal_horizon_segno,
+        // update the state in Arc<Timeline>
+        self.tli.wal_backup_active.store(
+            self.backup_task.is_some(),
+            std::sync::atomic::Ordering::Relaxed,
        );
-        *wal_removal_task = Some(tokio::spawn(
-            async move {
-                remover.await?;
-                Ok(removal_horizon_segno)
+        is_wal_backup_required
+    }
+
+    /// Update is_active flag and returns its value.
+    fn update_is_active(
+        &mut self,
+        is_wal_backup_required: bool,
+        num_computes: usize,
+        state: &StateSnapshot,
+    ) {
+        let is_active = is_wal_backup_required
+            || num_computes > 0
+            || state.remote_consistent_lsn < state.commit_lsn;
+
+        // update the broker timeline set
+        if self.tli_broker_active.set(is_active) {
+            // write log if state has changed
+            info!(
+                "timeline active={} now, remote_consistent_lsn={}, commit_lsn={}",
+                is_active, state.remote_consistent_lsn, state.commit_lsn,
+            );
+
+            MANAGER_ACTIVE_CHANGES.inc();
+        }
+
+        // update the state in Arc<Timeline>
+        self.tli
+            .broker_active
+            .store(is_active, std::sync::atomic::Ordering::Relaxed);
+    }
+
+    /// Save control file if needed. Returns Instant if we should persist the control file in the future.
+    async fn update_control_file_save(
+        &self,
+        state: &StateSnapshot,
+        next_event: &mut Option<Instant>,
+    ) {
+        if !state.inmem_flush_pending {
+            return;
+        }
+
+        if state.cfile_last_persist_at.elapsed() > self.conf.control_file_save_interval {
+            let mut write_guard = self.tli.write_shared_state().await;
+            // it should be done in the background because it blocks manager task, but flush() should
+            // be fast enough not to be a problem now
+            if let Err(e) = write_guard.sk.state_mut().flush().await {
+                warn!("failed to save control file: {:?}", e);
            }
-            .instrument(info_span!("WAL removal", ttid=%tli.ttid)),
-        ));
+        } else {
+            // we should wait until some time passed until the next save
+            update_next_event(
+                next_event,
+                (state.cfile_last_persist_at + self.conf.control_file_save_interval).into(),
+            );
+        }
+    }
+
+    /// Spawns WAL removal task if needed.
+    async fn update_wal_removal(&mut self, state: &StateSnapshot) {
+        if self.wal_removal_task.is_some() || state.wal_removal_on_hold {
+            // WAL removal is already in progress or hold off
+            return;
+        }
+
+        // If enabled, we use LSN of the most lagging walsender as a WAL removal horizon.
+        // This allows to get better read speed for pageservers that are lagging behind,
+        // at the cost of keeping more WAL on disk.
+        let replication_horizon_lsn = if self.conf.walsenders_keep_horizon {
+            self.walsenders.laggard_lsn()
+        } else {
+            None
+        };
+
+        let removal_horizon_lsn = calc_horizon_lsn(state, replication_horizon_lsn);
+        let removal_horizon_segno = removal_horizon_lsn
+            .segment_number(self.wal_seg_size)
+            .saturating_sub(1);
+
+        if removal_horizon_segno > self.last_removed_segno {
+            // we need to remove WAL
+            let remover = match self.tli.read_shared_state().await.sk {
+                StateSK::Loaded(ref sk) => {
+                    crate::wal_storage::Storage::remove_up_to(&sk.wal_store, removal_horizon_segno)
+                }
+                StateSK::Offloaded(_) => {
+                    // we can't remove WAL if it's not loaded
+                    warn!("unexpectedly trying to run WAL removal on offloaded timeline");
+                    return;
+                }
+                StateSK::Empty => unreachable!(),
+            };
+
+            self.wal_removal_task = Some(tokio::spawn(
+                async move {
+                    remover.await?;
+                    Ok(removal_horizon_segno)
+                }
+                .instrument(info_span!("WAL removal", ttid=%self.tli.ttid)),
+            ));
+        }
+    }
+
+    /// Update the state after WAL removal task finished.
+    fn update_wal_removal_end(&mut self, res: Result<anyhow::Result<u64>, JoinError>) {
+        let new_last_removed_segno = match res {
+            Ok(Ok(segno)) => segno,
+            Err(e) => {
+                warn!("WAL removal task failed: {:?}", e);
+                return;
+            }
+            Ok(Err(e)) => {
+                warn!("WAL removal task failed: {:?}", e);
+                return;
+            }
+        };
+
+        self.last_removed_segno = new_last_removed_segno;
+        // update the state in Arc<Timeline>
+        self.tli
+            .last_removed_segno
+            .store(new_last_removed_segno, std::sync::atomic::Ordering::Relaxed);
+    }
+
+    /// Spawns partial WAL backup task if needed.
+    async fn update_partial_backup(&mut self, state: &StateSnapshot) {
+        // check if partial backup is enabled and should be started
+        if !self.conf.is_wal_backup_enabled() || !self.conf.partial_backup_enabled {
+            return;
+        }
+
+        if self.partial_backup_task.is_some() {
+            // partial backup is already running
+            return;
+        }
+
+        if !wal_backup_partial::needs_uploading(state, &self.partial_backup_uploaded) {
+            // nothing to upload
+            return;
+        }
+
+        // Get WalResidentTimeline and start partial backup task.
+        self.partial_backup_task = Some(tokio::spawn(wal_backup_partial::main_task(
+            self.wal_resident_timeline(),
+            self.conf.clone(),
+            self.partial_backup_rate_limiter.clone(),
+        )));
+    }
+
+    /// Update the state after partial WAL backup task finished.
+    fn update_partial_backup_end(&mut self, res: Result<Option<PartialRemoteSegment>, JoinError>) {
+        match res {
+            Ok(new_upload_state) => {
+                self.partial_backup_uploaded = new_upload_state;
+            }
+            Err(e) => {
+                warn!("partial backup task panicked: {:?}", e);
+            }
+        }
+    }
+
+    /// Handle message arrived from ManagerCtl.
+    async fn handle_message(&mut self, msg: Option<ManagerCtlMessage>) {
+        debug!("received manager message: {:?}", msg);
+        match msg {
+            Some(ManagerCtlMessage::GuardRequest(tx)) => {
+                if self.is_offloaded {
+                    // trying to unevict timeline, but without gurarantee that it will be successful
+                    self.unevict_timeline().await;
+                }
+
+                let guard = if self.is_offloaded {
+                    Err(anyhow::anyhow!("timeline is offloaded, can't get a guard"))
+                } else {
+                    Ok(self.access_service.create_guard())
+                };
+
+                if tx.send(guard).is_err() {
+                    warn!("failed to reply with a guard, receiver dropped");
+                }
+            }
+            Some(ManagerCtlMessage::GuardDrop(guard_id)) => {
+                self.access_service.drop_guard(guard_id);
+            }
+            None => {
+                // can't happen, we're holding the sender
+                unreachable!();
+            }
+        }
    }
 }

-/// Update the state after WAL removal task finished.
-fn update_wal_removal_end(
-    res: Result<anyhow::Result<u64>, JoinError>,
-    tli: &Arc<Timeline>,
-    last_removed_segno: &mut u64,
-) {
-    let new_last_removed_segno = match res {
-        Ok(Ok(segno)) => segno,
-        Err(e) => {
-            warn!("WAL removal task failed: {:?}", e);
-            return;
-        }
-        Ok(Err(e)) => {
-            warn!("WAL removal task failed: {:?}", e);
-            return;
-        }
-    };
-
-    *last_removed_segno = new_last_removed_segno;
-    // update the state in Arc<Timeline>
-    tli.last_removed_segno
-        .store(new_last_removed_segno, std::sync::atomic::Ordering::Relaxed);
+// utility functions
+async fn sleep_until(option: &Option<tokio::time::Instant>) {
+    if let Some(timeout) = option {
+        tokio::time::sleep_until(*timeout).await;
+    } else {
+        futures::future::pending::<()>().await;
+    }
+}
+
+async fn await_task_finish<T>(option: &mut Option<JoinHandle<T>>) -> Result<T, JoinError> {
+    if let Some(task) = option {
+        task.await
+    } else {
+        futures::future::pending().await
+    }
+}
+
+/// Update next_event if candidate is earlier.
+fn update_next_event(next_event: &mut Option<Instant>, candidate: Instant) {
+    if let Some(next) = next_event {
+        if candidate < *next {
+            *next = candidate;
+        }
+    } else {
+        *next_event = Some(candidate);
+    }
+}
+
+#[repr(usize)]
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
+pub enum Status {
+    NotStarted,
+    Started,
+    StateSnapshot,
+    UpdateBackup,
+    UpdateControlFile,
+    UpdateWalRemoval,
+    UpdatePartialBackup,
+    EvictTimeline,
+    Wait,
+    HandleMessage,
+    Exiting,
+    Finished,
+}
+
+/// AtomicStatus is a wrapper around AtomicUsize adapted for the Status enum.
+pub struct AtomicStatus {
+    inner: AtomicUsize,
+}
+
+impl Default for AtomicStatus {
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
+impl AtomicStatus {
+    pub fn new() -> Self {
+        AtomicStatus {
+            inner: AtomicUsize::new(Status::NotStarted as usize),
+        }
+    }
+
+    pub fn load(&self, order: std::sync::atomic::Ordering) -> Status {
+        // Safety: This line of code uses `std::mem::transmute` to reinterpret the loaded value as `Status`.
+        // It is safe to use `transmute` in this context because `Status` is a repr(usize) enum,
+        // which means it has the same memory layout as usize.
+        // However, it is important to ensure that the loaded value is a valid variant of `Status`,
+        // otherwise, the behavior will be undefined.
+        unsafe { std::mem::transmute(self.inner.load(order)) }
+    }
+
+    pub fn get(&self) -> Status {
+        self.load(std::sync::atomic::Ordering::Relaxed)
+    }
+
+    pub fn store(&self, val: Status, order: std::sync::atomic::Ordering) {
+        self.inner.store(val as usize, order);
+    }
 }
@@ -5,6 +5,7 @@
 use crate::safekeeper::ServerInfo;
 use crate::timeline::{get_tenant_dir, get_timeline_dir, Timeline, TimelineError};
 use crate::timelines_set::TimelinesSet;
+use crate::wal_backup_partial::RateLimiter;
 use crate::SafeKeeperConf;
 use anyhow::{bail, Context, Result};
 use camino::Utf8PathBuf;
@@ -23,6 +24,7 @@ struct GlobalTimelinesState {
    conf: Option<SafeKeeperConf>,
    broker_active_set: Arc<TimelinesSet>,
    load_lock: Arc<tokio::sync::Mutex<TimelineLoadLock>>,
+    partial_backup_rate_limiter: RateLimiter,
 }

 // Used to prevent concurrent timeline loading.
@@ -37,8 +39,12 @@ impl GlobalTimelinesState {
    }

    /// Get dependencies for a timeline constructor.
-    fn get_dependencies(&self) -> (SafeKeeperConf, Arc<TimelinesSet>) {
-        (self.get_conf().clone(), self.broker_active_set.clone())
+    fn get_dependencies(&self) -> (SafeKeeperConf, Arc<TimelinesSet>, RateLimiter) {
+        (
+            self.get_conf().clone(),
+            self.broker_active_set.clone(),
+            self.partial_backup_rate_limiter.clone(),
+        )
    }

    /// Insert timeline into the map. Returns error if timeline with the same id already exists.
@@ -66,6 +72,7 @@ static TIMELINES_STATE: Lazy<Mutex<GlobalTimelinesState>> = Lazy::new(|| {
        conf: None,
        broker_active_set: Arc::new(TimelinesSet::default()),
        load_lock: Arc::new(tokio::sync::Mutex::new(TimelineLoadLock)),
+        partial_backup_rate_limiter: RateLimiter::new(1),
    })
 });

@@ -79,6 +86,7 @@ impl GlobalTimelines {
        // lock, so use explicit block
        let tenants_dir = {
            let mut state = TIMELINES_STATE.lock().unwrap();
+            state.partial_backup_rate_limiter = RateLimiter::new(conf.partial_backup_concurrency);
            state.conf = Some(conf);

            // Iterate through all directories and load tenants for all directories
@@ -122,7 +130,7 @@ impl GlobalTimelines {
    /// this function is called during init when nothing else is running, so
    /// this is fine.
    async fn load_tenant_timelines(tenant_id: TenantId) -> Result<()> {
-        let (conf, broker_active_set) = {
+        let (conf, broker_active_set, partial_backup_rate_limiter) = {
            let state = TIMELINES_STATE.lock().unwrap();
            state.get_dependencies()
        };
@@ -145,7 +153,11 @@ impl GlobalTimelines {
                                    .unwrap()
                                    .timelines
                                    .insert(ttid, tli.clone());
-                                tli.bootstrap(&conf, broker_active_set.clone());
+                                tli.bootstrap(
+                                    &conf,
+                                    broker_active_set.clone(),
+                                    partial_backup_rate_limiter.clone(),
+                                );
                            }
                            // If we can't load a timeline, it's most likely because of a corrupted
                            // directory. We will log an error and won't allow to delete/recreate
@@ -178,7 +190,8 @@ impl GlobalTimelines {
        _guard: &tokio::sync::MutexGuard<'a, TimelineLoadLock>,
        ttid: TenantTimelineId,
    ) -> Result<Arc<Timeline>> {
-        let (conf, broker_active_set) = TIMELINES_STATE.lock().unwrap().get_dependencies();
+        let (conf, broker_active_set, partial_backup_rate_limiter) =
+            TIMELINES_STATE.lock().unwrap().get_dependencies();

        match Timeline::load_timeline(&conf, ttid) {
            Ok(timeline) => {
@@ -191,7 +204,7 @@ impl GlobalTimelines {
                    .timelines
                    .insert(ttid, tli.clone());

-                tli.bootstrap(&conf, broker_active_set);
+                tli.bootstrap(&conf, broker_active_set, partial_backup_rate_limiter);

                Ok(tli)
            }
@@ -222,7 +235,7 @@ impl GlobalTimelines {
        commit_lsn: Lsn,
        local_start_lsn: Lsn,
    ) -> Result<Arc<Timeline>> {
-        let (conf, broker_active_set) = {
+        let (conf, broker_active_set, partial_backup_rate_limiter) = {
            let state = TIMELINES_STATE.lock().unwrap();
            if let Ok(timeline) = state.get(&ttid) {
                // Timeline already exists, return it.
@@ -257,7 +270,12 @@ impl GlobalTimelines {
            // Bootstrap is transactional, so if it fails, the timeline will be deleted,
            // and the state on disk should remain unchanged.
            if let Err(e) = timeline
-                .init_new(&mut shared_state, &conf, broker_active_set)
+                .init_new(
+                    &mut shared_state,
+                    &conf,
+                    broker_active_set,
+                    partial_backup_rate_limiter,
+                )
                .await
            {
                // Note: the most likely reason for init failure is that the timeline
--- a/Show More
+++ b/Show More