fix

Some tests still explicitly specify version 3 of the safekeeper
walproposer protocol. Remove the explicit opt in from the tests as v3 is
the default now since #11518.

We don't touch the places where a test exercises both v2 and v3. Those
we leave for #12021.

Part of https://github.com/neondatabase/neon/issues/10326

.github/actions/neon-project-create/action.yml

@@ -49,10 +49,6 @@ inputs:
     description: 'A JSON object with project settings'
     required: false
     default: '{}'
-  default_endpoint_settings:
-    description: 'A JSON object with the default endpoint settings'
-    required: false
-    default: '{}'
 
 outputs:
   dsn:
@@ -139,21 +135,6 @@ runs:
             -H "Accept: application/json" -H "Content-Type: application/json" -H "Authorization: Bearer ${ADMIN_API_KEY}" \
             -d "{\"scheduling\": \"Essential\"}"
         fi
-        # XXX
-        # This is a workaround for the default endpoint settings, which currently do not allow some settings in the public API.
-        # https://github.com/neondatabase/cloud/issues/27108
-        if [[ -n ${DEFAULT_ENDPOINT_SETTINGS} && ${DEFAULT_ENDPOINT_SETTINGS} != "{}" ]] ; then
-          PROJECT_DATA=$(curl -X GET \
-              "https://${API_HOST}/regions/${REGION_ID}/api/v1/admin/projects/${project_id}" \
-              -H "Accept: application/json" -H "Content-Type: application/json" -H "Authorization: Bearer ${ADMIN_API_KEY}" \
-              -d "{\"scheduling\": \"Essential\"}"
-          )
-          NEW_DEFAULT_ENDPOINT_SETTINGS=$(echo ${PROJECT_DATA} | jq -rc ".project.default_endpoint_settings + ${DEFAULT_ENDPOINT_SETTINGS}")
-          curl -X POST --fail \
-                "https://${API_HOST}/regions/${REGION_ID}/api/v1/admin/projects/${project_id}/default_endpoint_settings" \
-                -H "Accept: application/json" -H "Content-Type: application/json" -H "Authorization: Bearer ${ADMIN_API_KEY}" \
-                --data "${NEW_DEFAULT_ENDPOINT_SETTINGS}"
-        fi
         
 
       env:
@@ -171,4 +152,3 @@ runs:
         PSQL: ${{ inputs.psql_path }}
         LD_LIBRARY_PATH: ${{ inputs.libpq_lib_path }}
         PROJECT_SETTINGS: ${{ inputs.project_settings }}
-        DEFAULT_ENDPOINT_SETTINGS: ${{ inputs.default_endpoint_settings }}

.github/workflows/_build-and-test-locally.yml

@@ -279,18 +279,14 @@ jobs:
           # run all non-pageserver tests
           ${cov_prefix} cargo nextest run $CARGO_FLAGS $CARGO_FEATURES -E '!package(pageserver)'
 
-          # run pageserver tests with different settings
-          for get_vectored_concurrent_io in sequential sidecar-task; do
-            for io_engine in std-fs tokio-epoll-uring ; do
-                for io_mode in buffered direct direct-rw ; do
-                  NEON_PAGESERVER_UNIT_TEST_GET_VECTORED_CONCURRENT_IO=$get_vectored_concurrent_io \
-                  NEON_PAGESERVER_UNIT_TEST_VIRTUAL_FILE_IOENGINE=$io_engine \
-                  NEON_PAGESERVER_UNIT_TEST_VIRTUAL_FILE_IO_MODE=$io_mode \
-                  ${cov_prefix} \
-                  cargo nextest run $CARGO_FLAGS $CARGO_FEATURES  -E 'package(pageserver)'
-              done
-            done
-          done
+          # run pageserver tests
+          # (When developing new pageserver features gated by config fields, we commonly make the rust
+          # unit tests sensitive to an environment variable NEON_PAGESERVER_UNIT_TEST_FEATURENAME.
+          # Then run the nextest invocation below for all relevant combinations. Singling out the
+          # pageserver tests from non-pageserver tests cuts down the time it takes for this CI step.)
+          NEON_PAGESERVER_UNIT_TEST_VIRTUAL_FILE_IOENGINE=tokio-epoll-uring  \
+          ${cov_prefix} \
+          cargo nextest run $CARGO_FLAGS $CARGO_FEATURES  -E 'package(pageserver)'
 
           # Run separate tests for real S3
           export ENABLE_REAL_S3_REMOTE_STORAGE=nonempty
@@ -405,8 +401,6 @@ jobs:
           CHECK_ONDISK_DATA_COMPATIBILITY: nonempty
           BUILD_TAG: ${{ inputs.build-tag }}
           PAGESERVER_VIRTUAL_FILE_IO_ENGINE: tokio-epoll-uring
-          PAGESERVER_GET_VECTORED_CONCURRENT_IO: sidecar-task
-          PAGESERVER_VIRTUAL_FILE_IO_MODE: direct-rw
           USE_LFC: ${{ matrix.lfc_state == 'with-lfc' && 'true' || 'false' }}
 
       # Temporary disable this step until we figure out why it's so flaky

.github/workflows/build_and_test.yml

@@ -314,7 +314,8 @@ jobs:
           test_selection: performance
           run_in_parallel: false
           save_perf_report: ${{ github.ref_name == 'main' }}
-          extra_params: --splits 5 --group ${{ matrix.pytest_split_group }}
+          # test_pageserver_max_throughput_getpage_at_latest_lsn is run in separate workflow periodic_pagebench.yml because it needs snapshots
+          extra_params: --splits 5 --group ${{ matrix.pytest_split_group }} --ignore=test_runner/performance/pageserver/pagebench/test_pageserver_max_throughput_getpage_at_latest_lsn.py
           benchmark_durations: ${{ needs.get-benchmarks-durations.outputs.json }}
           pg_version: v16
           aws-oidc-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
@@ -323,8 +324,6 @@ jobs:
           PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
           TEST_RESULT_CONNSTR: "${{ secrets.REGRESS_TEST_RESULT_CONNSTR_NEW }}"
           PAGESERVER_VIRTUAL_FILE_IO_ENGINE: tokio-epoll-uring
-          PAGESERVER_GET_VECTORED_CONCURRENT_IO: sidecar-task
-          PAGESERVER_VIRTUAL_FILE_IO_MODE: direct-rw
           SYNC_BETWEEN_TESTS: true
       # XXX: no coverage data handling here, since benchmarks are run on release builds,
       # while coverage is currently collected for the debug ones
@@ -965,7 +964,7 @@ jobs:
           fi
 
       - name: Verify docker-compose example and test extensions
-        timeout-minutes: 20
+        timeout-minutes: 60
         env:
           TAG: >-
             ${{

.github/workflows/cloud-extensions.yml

@@ -35,7 +35,7 @@ jobs:
       matrix:
         pg-version: [16, 17]
 
-    runs-on: [ self-hosted, small ]
+    runs-on: us-east-2
     container:
       # We use the neon-test-extensions image here as it contains the source code for the extensions.
       image: ghcr.io/neondatabase/neon-test-extensions-v${{ matrix.pg-version }}:latest
@@ -71,20 +71,7 @@ jobs:
           region_id: ${{ inputs.region_id || 'aws-us-east-2' }}
           postgres_version: ${{ matrix.pg-version }}
           project_settings: ${{ steps.project-settings.outputs.settings }}
-          # We need these settings to get the expected output results.
-          # We cannot use the environment variables e.g. PGTZ due to
-          # https://github.com/neondatabase/neon/issues/1287
-          default_endpoint_settings: >
-            {
-              "pg_settings": {
-                "DateStyle": "Postgres,MDY",
-                "TimeZone": "America/Los_Angeles",
-                "compute_query_id": "off",
-                "neon.allow_unstable_extensions": "on"
-              }
-            }
           api_key: ${{ secrets.NEON_STAGING_API_KEY }}
-          admin_api_key: ${{ secrets.NEON_STAGING_ADMIN_API_KEY }}
 
       - name: Run the regression tests
         run: /run-tests.sh -r /ext-src

.github/workflows/neon_extra_builds.yml

@@ -63,8 +63,10 @@ jobs:
 
       - name: Filter out only v-string for build matrix
         id: postgres_changes
+        env:
+          CHANGES: ${{ steps.files_changed.outputs.changes }}
         run: |
-          v_strings_only_as_json_array=$(echo ${{ steps.files_changed.outputs.chnages }} | jq '.[]|select(test("v\\d+"))' | jq --slurp -c)
+          v_strings_only_as_json_array=$(echo ${CHANGES} | jq '.[]|select(test("v\\d+"))' | jq --slurp -c)
           echo "changes=${v_strings_only_as_json_array}" | tee -a "${GITHUB_OUTPUT}"
 
   check-macos-build:

.github/workflows/periodic_pagebench.yml

@@ -1,4 +1,4 @@
-name: Periodic pagebench performance test on dedicated EC2 machine in eu-central-1 region
+name: Periodic pagebench performance test on unit-perf hetzner runner
 
 on:
   schedule:
@@ -8,7 +8,7 @@ on:
     #        │   │ ┌───────────── day of the month (1 - 31)
     #        │   │ │ ┌───────────── month (1 - 12 or JAN-DEC)
     #        │   │ │ │ ┌───────────── day of the week (0 - 6 or SUN-SAT)
-    - cron: '0 */3 * * *' # Runs every 3 hours
+    - cron: '0 */4 * * *' # Runs every 4 hours
   workflow_dispatch: # Allows manual triggering of the workflow
     inputs:
       commit_hash:
@@ -16,6 +16,11 @@ on:
         description: 'The long neon repo commit hash for the system under test (pageserver) to be tested.'
         required: false
         default: ''
+      recreate_snapshots:
+        type: boolean
+        description: 'Recreate snapshots - !!!WARNING!!! We should only recreate snapshots if the previous ones are no longer compatible. Otherwise benchmarking results are not comparable across runs.'
+        required: false
+        default: false
 
 defaults:
   run:
@@ -29,13 +34,13 @@ permissions:
   contents: read
 
 jobs:
-  trigger_bench_on_ec2_machine_in_eu_central_1:
+  run_periodic_pagebench_test:
     permissions:
       id-token: write # aws-actions/configure-aws-credentials
       statuses: write
       contents: write
       pull-requests: write
-    runs-on: [ self-hosted, small ]
+    runs-on: [ self-hosted, unit-perf ]
     container:
       image: ghcr.io/neondatabase/build-tools:pinned-bookworm
       credentials:
@@ -44,10 +49,13 @@ jobs:
       options: --init
     timeout-minutes: 360  # Set the timeout to 6 hours
     env:
-      API_KEY: ${{ secrets.PERIODIC_PAGEBENCH_EC2_RUNNER_API_KEY }}
       RUN_ID: ${{ github.run_id }}
-      AWS_DEFAULT_REGION : "eu-central-1"
-      AWS_INSTANCE_ID : "i-02a59a3bf86bc7e74"
+      DEFAULT_PG_VERSION: 16
+      BUILD_TYPE: release
+      RUST_BACKTRACE: 1
+      # NEON_ENV_BUILDER_USE_OVERLAYFS_FOR_SNAPSHOTS: 1 - doesn't work without root in container
+      S3_BUCKET: neon-github-public-dev
+      PERF_TEST_RESULT_CONNSTR: "${{ secrets.PERF_TEST_RESULT_CONNSTR }}"
     steps:
     # we don't need the neon source code because we run everything remotely
     # however we still need the local github actions to run the allure step below
@@ -56,99 +64,194 @@ jobs:
       with:
         egress-policy: audit
 
-    - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+    - name: Set up the environment which depends on $RUNNER_TEMP on nvme drive
+      id: set-env
+      shell: bash -euxo pipefail {0}
+      run: |
+        {
+          echo "NEON_DIR=${RUNNER_TEMP}/neon"
+          echo "NEON_BIN=${RUNNER_TEMP}/neon/bin"
+          echo "POSTGRES_DISTRIB_DIR=${RUNNER_TEMP}/neon/pg_install"
+          echo "LD_LIBRARY_PATH=${RUNNER_TEMP}/neon/pg_install/v${DEFAULT_PG_VERSION}/lib"
+          echo "BACKUP_DIR=${RUNNER_TEMP}/instance_store/saved_snapshots"
+          echo "TEST_OUTPUT=${RUNNER_TEMP}/neon/test_output"
+          echo "PERF_REPORT_DIR=${RUNNER_TEMP}/neon/test_output/perf-report-local"
+          echo "ALLURE_DIR=${RUNNER_TEMP}/neon/test_output/allure-results"
+          echo "ALLURE_RESULTS_DIR=${RUNNER_TEMP}/neon/test_output/allure-results/results"
+        } >> "$GITHUB_ENV"
 
-    - name: Show my own (github runner) external IP address - usefull for IP allowlisting
-      run: curl https://ifconfig.me
+        echo "allure_results_dir=${RUNNER_TEMP}/neon/test_output/allure-results/results" >> "$GITHUB_OUTPUT"
 
-    - name: Assume AWS OIDC role that allows to manage (start/stop/describe... EC machine)
-      uses: aws-actions/configure-aws-credentials@e3dd6a429d7300a6a4c196c26e071d42e0343502 # v4.0.2
+    - uses: aws-actions/configure-aws-credentials@e3dd6a429d7300a6a4c196c26e071d42e0343502 # v4.0.2
       with:
         aws-region: eu-central-1
-        role-to-assume: ${{ vars.DEV_AWS_OIDC_ROLE_MANAGE_BENCHMARK_EC2_VMS_ARN }}
-        role-duration-seconds: 3600
-
-    - name: Start EC2 instance and wait for the instance to boot up
-      run: |
-        aws ec2 start-instances --instance-ids $AWS_INSTANCE_ID
-        aws ec2 wait instance-running --instance-ids $AWS_INSTANCE_ID
-        sleep 60 # sleep some time to allow cloudinit and our API server to start up
-
-    - name: Determine public IP of the EC2 instance and set env variable EC2_MACHINE_URL_US
-      run: |
-        public_ip=$(aws ec2 describe-instances --instance-ids $AWS_INSTANCE_ID --query 'Reservations[*].Instances[*].PublicIpAddress' --output text)
-        echo "Public IP of the EC2 instance: $public_ip"
-        echo "EC2_MACHINE_URL_US=https://${public_ip}:8443" >> $GITHUB_ENV
-
+        role-to-assume: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
+        role-duration-seconds: 18000 # max 5 hours (needed in case commit hash is still being built)
     - name: Determine commit hash
+      id: commit_hash
+      shell: bash -euxo pipefail {0}
       env:
         INPUT_COMMIT_HASH: ${{ github.event.inputs.commit_hash }}
       run: |
-        if [ -z "$INPUT_COMMIT_HASH" ]; then
-          echo "COMMIT_HASH=$(curl -s https://api.github.com/repos/neondatabase/neon/commits/main | jq -r '.sha')" >> $GITHUB_ENV
+        if [[ -z "${INPUT_COMMIT_HASH}" ]]; then
+          COMMIT_HASH=$(curl -s https://api.github.com/repos/neondatabase/neon/commits/main | jq -r '.sha')
+          echo "COMMIT_HASH=$COMMIT_HASH" >> $GITHUB_ENV
+          echo "commit_hash=$COMMIT_HASH" >> "$GITHUB_OUTPUT"
           echo "COMMIT_HASH_TYPE=latest" >> $GITHUB_ENV
         else
-          echo "COMMIT_HASH=$INPUT_COMMIT_HASH" >> $GITHUB_ENV
+          COMMIT_HASH="${INPUT_COMMIT_HASH}"
+          echo "COMMIT_HASH=$COMMIT_HASH" >> $GITHUB_ENV
+          echo "commit_hash=$COMMIT_HASH" >> "$GITHUB_OUTPUT"
           echo "COMMIT_HASH_TYPE=manual" >> $GITHUB_ENV
         fi
+    - name: Checkout the neon repository at given commit hash
+      uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+      with:
+        ref: ${{ steps.commit_hash.outputs.commit_hash }}
 
-    - name: Start Bench with run_id
+    # does not reuse ./.github/actions/download because we need to download the artifact for the given commit hash
+    # example artifact
+    # s3://neon-github-public-dev/artifacts/48b870bc078bd2c450eb7b468e743b9c118549bf/15036827400/1/neon-Linux-X64-release-artifact.tar.zst /instance_store/artifacts/neon-Linux-release-artifact.tar.zst
+    - name: Determine artifact S3_KEY for given commit hash and download and extract artifact
+      id: artifact_prefix
+      shell: bash -euxo pipefail {0}
+      env:
+        ARCHIVE: ${{ runner.temp }}/downloads/neon-${{ runner.os }}-${{ runner.arch }}-release-artifact.tar.zst
+        COMMIT_HASH: ${{ env.COMMIT_HASH }}
+        COMMIT_HASH_TYPE: ${{ env.COMMIT_HASH_TYPE }}
       run: |
-        curl -k -X 'POST' \
-        "${EC2_MACHINE_URL_US}/start_test/${GITHUB_RUN_ID}" \
-        -H 'accept: application/json' \
-        -H 'Content-Type: application/json' \
-        -H "Authorization: Bearer $API_KEY" \
-        -d "{\"neonRepoCommitHash\": \"${COMMIT_HASH}\", \"neonRepoCommitHashType\": \"${COMMIT_HASH_TYPE}\"}"
+        attempt=0
+        max_attempts=24 # 5 minutes * 24 = 2 hours
 
-    - name: Poll Test Status
-      id: poll_step
-      run: |
-        status=""
-        while [[ "$status" != "failure" && "$status" != "success" ]]; do
-          response=$(curl -k -X 'GET' \
-          "${EC2_MACHINE_URL_US}/test_status/${GITHUB_RUN_ID}" \
-          -H 'accept: application/json' \
-          -H "Authorization: Bearer $API_KEY")
-          echo "Response: $response"
-          set +x
-          status=$(echo $response | jq -r '.status')
-          echo "Test status: $status"
-          if [[ "$status" == "failure" ]]; then
-            echo "Test failed"
-            exit 1 # Fail the job step if status is failure
-          elif [[ "$status" == "success" || "$status" == "null" ]]; then
+        while [[ $attempt -lt $max_attempts ]]; do
+          # the following command will fail until the artifacts are available ...
+          S3_KEY=$(aws s3api list-objects-v2 --bucket "$S3_BUCKET" --prefix "artifacts/$COMMIT_HASH/" \
+            | jq -r '.Contents[]?.Key' \
+            | grep "neon-${{ runner.os }}-${{ runner.arch }}-release-artifact.tar.zst" \
+            | sort --version-sort \
+            | tail -1) || true # ... thus ignore errors from the command
+          if [[ -n "${S3_KEY}" ]]; then
+            echo "Artifact found: $S3_KEY"
+            echo "S3_KEY=$S3_KEY" >> $GITHUB_ENV
             break
-          elif [[ "$status" == "too_many_runs" ]]; then
-            echo "Too many runs already running"
-            echo "too_many_runs=true" >> "$GITHUB_OUTPUT"
-            exit 1
           fi
-
-          sleep 60 # Poll every 60 seconds
+          
+          # Increment attempt counter and sleep for 5 minutes
+          attempt=$((attempt + 1))
+          echo "Attempt $attempt of $max_attempts to find artifacts in S3 bucket s3://$S3_BUCKET/artifacts/$COMMIT_HASH failed. Retrying in 5 minutes..."
+          sleep 300 # Sleep for 5 minutes
         done
 
-    - name: Retrieve Test Logs
-      if: always() && steps.poll_step.outputs.too_many_runs != 'true'
-      run: |
-        curl -k -X 'GET' \
-        "${EC2_MACHINE_URL_US}/test_log/${GITHUB_RUN_ID}" \
-        -H 'accept: application/gzip' \
-        -H "Authorization: Bearer $API_KEY" \
-        --output "test_log_${GITHUB_RUN_ID}.gz"
+        if [[ -z "${S3_KEY}" ]]; then
+          echo "Error: artifact not found in S3 bucket s3://$S3_BUCKET/artifacts/$COMMIT_HASH" after 2 hours
+        else
+          mkdir -p $(dirname $ARCHIVE)
+          time aws s3 cp --only-show-errors s3://$S3_BUCKET/${S3_KEY} ${ARCHIVE}
+          mkdir -p ${NEON_DIR}
+          time tar -xf ${ARCHIVE} -C ${NEON_DIR}
+          rm -f ${ARCHIVE}
+        fi
 
-    - name: Unzip Test Log and Print it into this job's log
-      if: always() && steps.poll_step.outputs.too_many_runs != 'true'
+    - name: Download snapshots from S3
+      if: ${{ github.event_name != 'workflow_dispatch' || github.event.inputs.recreate_snapshots == 'false' || github.event.inputs.recreate_snapshots == '' }}
+      id: download_snapshots
+      shell: bash -euxo pipefail {0}
       run: |
-        gzip -d "test_log_${GITHUB_RUN_ID}.gz"
-        cat "test_log_${GITHUB_RUN_ID}"
+        # Download the snapshots from S3
+        mkdir -p ${TEST_OUTPUT}
+        mkdir -p $BACKUP_DIR
+        cd $BACKUP_DIR
+        mkdir parts
+        cd parts
+        PART=$(aws s3api list-objects-v2 --bucket $S3_BUCKET --prefix performance/pagebench/ \
+          | jq -r '.Contents[]?.Key' \
+          | grep -E 'shared-snapshots-[0-9]{4}-[0-9]{2}-[0-9]{2}' \
+          | sort \
+          | tail -1)
+        echo "Latest PART: $PART"
+        if [[ -z "$PART" ]]; then
+          echo "ERROR: No matching S3 key found" >&2
+          exit 1
+        fi
+        S3_KEY=$(dirname $PART)
+        time aws s3 cp --only-show-errors --recursive s3://${S3_BUCKET}/$S3_KEY/ .
+        cd $TEST_OUTPUT
+        time cat $BACKUP_DIR/parts/* | zstdcat | tar --extract --preserve-permissions
+        rm -rf ${BACKUP_DIR}
+
+    - name: Cache poetry deps
+      uses: actions/cache@v4
+      with:
+        path: ~/.cache/pypoetry/virtualenvs
+        key: v2-${{ runner.os }}-${{ runner.arch }}-python-deps-bookworm-${{ hashFiles('poetry.lock') }}
+
+    - name: Install Python deps
+      shell: bash -euxo pipefail {0}
+      run: ./scripts/pysync
+
+    # we need high number of open files for pagebench
+    - name: show ulimits
+      shell: bash -euxo pipefail {0}
+      run: |
+        ulimit -a
+
+    - name: Run pagebench testcase
+      shell: bash -euxo pipefail {0}
+      env:
+        CI: false  # need to override this env variable set by github to enforce using snapshots
+      run: |
+        export PLATFORM=hetzner-unit-perf-${COMMIT_HASH_TYPE}
+        # report the commit hash of the neon repository in the revision of the test results
+        export GITHUB_SHA=${COMMIT_HASH}
+        rm -rf ${PERF_REPORT_DIR}
+        rm -rf ${ALLURE_RESULTS_DIR}
+        mkdir -p ${PERF_REPORT_DIR}
+        mkdir -p ${ALLURE_RESULTS_DIR}
+        PARAMS="--alluredir=${ALLURE_RESULTS_DIR} --tb=short --verbose -rA"
+        EXTRA_PARAMS="--out-dir ${PERF_REPORT_DIR} --durations-path $TEST_OUTPUT/benchmark_durations.json"
+        # run only two selected tests
+        # environment set by parent:
+        # RUST_BACKTRACE=1 DEFAULT_PG_VERSION=16 BUILD_TYPE=release
+        ./scripts/pytest ${PARAMS} test_runner/performance/pageserver/pagebench/test_pageserver_max_throughput_getpage_at_latest_lsn.py::test_pageserver_characterize_throughput_with_n_tenants ${EXTRA_PARAMS}
+        ./scripts/pytest ${PARAMS} test_runner/performance/pageserver/pagebench/test_pageserver_max_throughput_getpage_at_latest_lsn.py::test_pageserver_characterize_latencies_with_1_client_and_throughput_with_many_clients_one_tenant ${EXTRA_PARAMS}
+
+    - name: upload the performance metrics to the Neon performance database which is used by grafana dashboards to display the results
+      shell: bash -euxo pipefail {0}
+      run: |
+        export REPORT_FROM="$PERF_REPORT_DIR"
+        export GITHUB_SHA=${COMMIT_HASH}
+        time ./scripts/generate_and_push_perf_report.sh
+
+    - name: Upload test results
+      if: ${{ !cancelled() }}
+      uses: ./.github/actions/allure-report-store
+      with:
+        report-dir:  ${{ steps.set-env.outputs.allure_results_dir }}
+        unique-key: ${{ env.BUILD_TYPE }}-${{ env.DEFAULT_PG_VERSION }}-${{ runner.arch }}
+        aws-oidc-role-arn:  ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
 
     - name: Create Allure report
+      id: create-allure-report
       if: ${{ !cancelled() }}
       uses: ./.github/actions/allure-report-generate
       with:
         aws-oidc-role-arn: ${{ vars.DEV_AWS_OIDC_ROLE_ARN }}
 
+    - name: Upload snapshots
+      if: ${{ github.event_name == 'workflow_dispatch' && github.event.inputs.recreate_snapshots != 'false' && github.event.inputs.recreate_snapshots != '' }}
+      id: upload_snapshots
+      shell: bash -euxo pipefail {0}
+      run: |
+        mkdir -p $BACKUP_DIR
+        cd $TEST_OUTPUT
+        tar --create --preserve-permissions --file - shared-snapshots | zstd -o $BACKUP_DIR/shared_snapshots.tar.zst
+        cd $BACKUP_DIR
+        mkdir parts
+        split -b 1G shared_snapshots.tar.zst ./parts/shared_snapshots.tar.zst.part.
+        SNAPSHOT_DATE=$(date +%F)  # YYYY-MM-DD
+        cd parts
+        time aws s3 cp --recursive . s3://${S3_BUCKET}/performance/pagebench/shared-snapshots-${SNAPSHOT_DATE}/
+
     - name: Post to a Slack channel
       if: ${{ github.event.schedule && failure() }}
       uses: slackapi/slack-github-action@fcfb566f8b0aab22203f066d80ca1d7e4b5d05b3 # v1.27.1
@@ -157,26 +260,22 @@ jobs:
         slack-message: "Periodic pagebench testing on dedicated hardware: ${{ job.status }}\n${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
       env:
         SLACK_BOT_TOKEN: ${{ secrets.SLACK_BOT_TOKEN }}
-
+        
     - name: Cleanup Test Resources
       if: always()
+      shell: bash -euxo pipefail {0}
+      env:
+        ARCHIVE: ${{ runner.temp }}/downloads/neon-${{ runner.os }}-${{ runner.arch }}-release-artifact.tar.zst
       run: |
-        curl -k -X 'POST' \
-        "${EC2_MACHINE_URL_US}/cleanup_test/${GITHUB_RUN_ID}" \
-        -H 'accept: application/json' \
-        -H "Authorization: Bearer $API_KEY" \
-        -d ''
+        # Cleanup the test resources
+        if [[ -d "${BACKUP_DIR}" ]]; then
+          rm -rf ${BACKUP_DIR}
+        fi
+        if [[ -d "${TEST_OUTPUT}" ]]; then
+          rm -rf ${TEST_OUTPUT}
+        fi
+        if [[ -d "${NEON_DIR}" ]]; then
+          rm -rf ${NEON_DIR}
+        fi
+        rm -rf $(dirname $ARCHIVE)
 
-    - name: Assume AWS OIDC role that allows to manage (start/stop/describe... EC machine)
-      if: always() && steps.poll_step.outputs.too_many_runs != 'true'
-      uses: aws-actions/configure-aws-credentials@e3dd6a429d7300a6a4c196c26e071d42e0343502 # v4.0.2
-      with:
-        aws-region: eu-central-1
-        role-to-assume: ${{ vars.DEV_AWS_OIDC_ROLE_MANAGE_BENCHMARK_EC2_VMS_ARN }}
-        role-duration-seconds: 3600
-
-    - name: Stop EC2 instance and wait for the instance to be stopped
-      if: always() && steps.poll_step.outputs.too_many_runs != 'true'
-      run: |
-        aws ec2 stop-instances --instance-ids $AWS_INSTANCE_ID
-        aws ec2 wait instance-stopped --instance-ids $AWS_INSTANCE_ID

Cargo.lock

@@ -1112,6 +1112,12 @@ version = "1.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "baf1de4339761588bc0619e3cbc0120ee582ebb74b53b4efbf79117bd2da40fd"
 
+[[package]]
+name = "cfg_aliases"
+version = "0.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "613afe47fcd5fac7ccf1db93babcb082c5994d996f20b8b159f2ad1658eb5724"
+
 [[package]]
 name = "cgroups-rs"
 version = "0.3.3"
@@ -1270,7 +1276,7 @@ version = "0.1.0"
 dependencies = [
  "anyhow",
  "chrono",
- "indexmap 2.0.1",
+ "indexmap 2.9.0",
  "jsonwebtoken",
  "regex",
  "remote_storage",
@@ -1302,11 +1308,11 @@ dependencies = [
  "flate2",
  "futures",
  "http 1.1.0",
- "indexmap 2.0.1",
+ "indexmap 2.9.0",
  "itertools 0.10.5",
  "jsonwebtoken",
  "metrics",
- "nix 0.27.1",
+ "nix 0.30.1",
  "notify",
  "num_cpus",
  "once_cell",
@@ -1429,7 +1435,7 @@ dependencies = [
  "humantime-serde",
  "hyper 0.14.30",
  "jsonwebtoken",
- "nix 0.27.1",
+ "nix 0.30.1",
  "once_cell",
  "pageserver_api",
  "pageserver_client",
@@ -2591,7 +2597,7 @@ dependencies = [
  "futures-sink",
  "futures-util",
  "http 0.2.9",
- "indexmap 2.0.1",
+ "indexmap 2.9.0",
  "slab",
  "tokio",
  "tokio-util",
@@ -2610,7 +2616,7 @@ dependencies = [
  "futures-sink",
  "futures-util",
  "http 1.1.0",
- "indexmap 2.0.1",
+ "indexmap 2.9.0",
  "slab",
  "tokio",
  "tokio-util",
@@ -2857,14 +2863,14 @@ dependencies = [
  "pprof",
  "regex",
  "routerify",
- "rustls 0.23.18",
+ "rustls 0.23.27",
  "rustls-pemfile 2.1.1",
  "serde",
  "serde_json",
  "serde_path_to_error",
  "thiserror 1.0.69",
  "tokio",
- "tokio-rustls 0.26.0",
+ "tokio-rustls 0.26.2",
  "tokio-stream",
  "tokio-util",
  "tracing",
@@ -3194,12 +3200,12 @@ dependencies = [
 
 [[package]]
 name = "indexmap"
-version = "2.0.1"
+version = "2.9.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ad227c3af19d4914570ad36d30409928b75967c298feb9ea1969db3a610bb14e"
+checksum = "cea70ddb795996207ad57735b50c5982d8844f38ba9ee5f1aedcfb708a2aa11e"
 dependencies = [
  "equivalent",
- "hashbrown 0.14.5",
+ "hashbrown 0.15.2",
  "serde",
 ]
 
@@ -3222,7 +3228,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "232929e1d75fe899576a3d5c7416ad0d88dbfbb3c3d6aa00873a7408a50ddb88"
 dependencies = [
  "ahash",
- "indexmap 2.0.1",
+ "indexmap 2.9.0",
  "is-terminal",
  "itoa",
  "log",
@@ -3245,7 +3251,7 @@ dependencies = [
  "crossbeam-utils",
  "dashmap 6.1.0",
  "env_logger",
- "indexmap 2.0.1",
+ "indexmap 2.9.0",
  "itoa",
  "log",
  "num-format",
@@ -3512,9 +3518,9 @@ dependencies = [
 
 [[package]]
 name = "libc"
-version = "0.2.169"
+version = "0.2.172"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "b5aba8db14291edd000dfcc4d620c7ebfb122c613afb886ca8803fa4e128a20a"
+checksum = "d750af042f7ef4f724306de029d18836c26c1765a54a6a3f094cbd23a7267ffa"
 
 [[package]]
 name = "libloading"
@@ -3788,6 +3794,16 @@ version = "0.8.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "e5ce46fe64a9d73be07dcbe690a38ce1b293be448fd8ce1e6c1b8062c9f72c6a"
 
+[[package]]
+name = "neon-shmem"
+version = "0.1.0"
+dependencies = [
+ "nix 0.30.1",
+ "tempfile",
+ "thiserror 1.0.69",
+ "workspace_hack",
+]
+
 [[package]]
 name = "never-say-never"
 version = "6.6.666"
@@ -3821,12 +3837,13 @@ dependencies = [
 
 [[package]]
 name = "nix"
-version = "0.27.1"
+version = "0.30.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "2eb04e9c688eff1c89d72b407f168cf79bb9e867a9d3323ed6c01519eb9cc053"
+checksum = "74523f3a35e05aba87a1d978330aef40f67b0304ac79c1c00b294c9830543db6"
 dependencies = [
  "bitflags 2.8.0",
  "cfg-if",
+ "cfg_aliases",
  "libc",
  "memoffset 0.9.0",
 ]
@@ -3881,6 +3898,16 @@ dependencies = [
  "winapi",
 ]
 
+[[package]]
+name = "nu-ansi-term"
+version = "0.46.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "77a8165726e8236064dbb45459242600304b42a5ea24ee2948e18e023bf7ba84"
+dependencies = [
+ "overload",
+ "winapi",
+]
+
 [[package]]
 name = "num"
 version = "0.4.1"
@@ -4085,7 +4112,7 @@ dependencies = [
  "opentelemetry-http",
  "opentelemetry-proto",
  "opentelemetry_sdk",
- "prost 0.13.3",
+ "prost 0.13.5",
  "reqwest",
  "thiserror 1.0.69",
 ]
@@ -4098,8 +4125,8 @@ checksum = "a6e05acbfada5ec79023c85368af14abd0b307c015e9064d249b2a950ef459a6"
 dependencies = [
  "opentelemetry",
  "opentelemetry_sdk",
- "prost 0.13.3",
- "tonic",
+ "prost 0.13.5",
+ "tonic 0.12.3",
 ]
 
 [[package]]
@@ -4165,6 +4192,12 @@ version = "0.5.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "4030760ffd992bef45b0ae3f10ce1aba99e33464c90d14dd7c039884963ddc7a"
 
+[[package]]
+name = "overload"
+version = "0.1.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b15813163c1d831bf4a13c3610c05c0d03b39feb07f7e09fa234dac9b15aaf39"
+
 [[package]]
 name = "p256"
 version = "0.11.1"
@@ -4269,6 +4302,7 @@ dependencies = [
  "enumset",
  "fail",
  "futures",
+ "hashlink",
  "hex",
  "hex-literal",
  "http-utils",
@@ -4280,13 +4314,14 @@ dependencies = [
  "jsonwebtoken",
  "md5",
  "metrics",
- "nix 0.27.1",
+ "nix 0.30.1",
  "num-traits",
  "num_cpus",
  "once_cell",
  "pageserver_api",
  "pageserver_client",
  "pageserver_compaction",
+ "pageserver_page_api",
  "pem",
  "pin-project-lite",
  "postgres-protocol",
@@ -4295,6 +4330,7 @@ dependencies = [
  "postgres_connection",
  "postgres_ffi",
  "postgres_initdb",
+ "posthog_client_lite",
  "pprof",
  "pq_proto",
  "procfs",
@@ -4305,7 +4341,7 @@ dependencies = [
  "reqwest",
  "rpds",
  "rstest",
- "rustls 0.23.18",
+ "rustls 0.23.27",
  "scopeguard",
  "send-future",
  "serde",
@@ -4324,13 +4360,16 @@ dependencies = [
  "tokio-epoll-uring",
  "tokio-io-timeout",
  "tokio-postgres",
- "tokio-rustls 0.26.0",
+ "tokio-rustls 0.26.2",
  "tokio-stream",
  "tokio-tar",
  "tokio-util",
  "toml_edit",
+ "tonic 0.13.1",
+ "tonic-reflection",
  "tracing",
  "tracing-utils",
+ "twox-hash",
  "url",
  "utils",
  "uuid",
@@ -4355,7 +4394,7 @@ dependencies = [
  "humantime",
  "humantime-serde",
  "itertools 0.10.5",
- "nix 0.27.1",
+ "nix 0.30.1",
  "once_cell",
  "postgres_backend",
  "postgres_ffi",
@@ -4416,6 +4455,22 @@ dependencies = [
  "workspace_hack",
 ]
 
+[[package]]
+name = "pageserver_page_api"
+version = "0.1.0"
+dependencies = [
+ "bytes",
+ "pageserver_api",
+ "postgres_ffi",
+ "prost 0.13.5",
+ "smallvec",
+ "thiserror 1.0.69",
+ "tonic 0.13.1",
+ "tonic-build",
+ "utils",
+ "workspace_hack",
+]
+
 [[package]]
 name = "papaya"
 version = "0.2.1"
@@ -4792,14 +4847,14 @@ dependencies = [
  "bytes",
  "once_cell",
  "pq_proto",
- "rustls 0.23.18",
+ "rustls 0.23.27",
  "rustls-pemfile 2.1.1",
  "serde",
  "thiserror 1.0.69",
  "tokio",
  "tokio-postgres",
  "tokio-postgres-rustls",
- "tokio-rustls 0.26.0",
+ "tokio-rustls 0.26.2",
  "tokio-util",
  "tracing",
 ]
@@ -4848,6 +4903,24 @@ dependencies = [
  "workspace_hack",
 ]
 
+[[package]]
+name = "posthog_client_lite"
+version = "0.1.0"
+dependencies = [
+ "anyhow",
+ "arc-swap",
+ "reqwest",
+ "serde",
+ "serde_json",
+ "sha2",
+ "thiserror 1.0.69",
+ "tokio",
+ "tokio-util",
+ "tracing",
+ "tracing-utils",
+ "workspace_hack",
+]
+
 [[package]]
 name = "powerfmt"
 version = "0.2.0"
@@ -4893,7 +4966,7 @@ dependencies = [
  "inferno 0.12.0",
  "num",
  "paste",
- "prost 0.13.3",
+ "prost 0.13.5",
 ]
 
 [[package]]
@@ -4998,12 +5071,12 @@ dependencies = [
 
 [[package]]
 name = "prost"
-version = "0.13.3"
+version = "0.13.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "7b0487d90e047de87f984913713b85c601c05609aad5b0df4b4573fbf69aa13f"
+checksum = "2796faa41db3ec313a31f7624d9286acf277b52de526150b7e69f3debf891ee5"
 dependencies = [
  "bytes",
- "prost-derive 0.13.3",
+ "prost-derive 0.13.5",
 ]
 
 [[package]]
@@ -5041,7 +5114,7 @@ dependencies = [
  "once_cell",
  "petgraph",
  "prettyplease",
- "prost 0.13.3",
+ "prost 0.13.5",
  "prost-types 0.13.3",
  "regex",
  "syn 2.0.100",
@@ -5063,9 +5136,9 @@ dependencies = [
 
 [[package]]
 name = "prost-derive"
-version = "0.13.3"
+version = "0.13.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e9552f850d5f0964a4e4d0bf306459ac29323ddfbae05e35a7c0d35cb0803cc5"
+checksum = "8a56d757972c98b346a9b766e3f02746cde6dd1cd1d1d563472929fdd74bec4d"
 dependencies = [
  "anyhow",
  "itertools 0.12.1",
@@ -5089,7 +5162,7 @@ version = "0.13.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "4759aa0d3a6232fb8dbdb97b61de2c20047c68aca932c7ed76da9d788508d670"
 dependencies = [
- "prost 0.13.3",
+ "prost 0.13.5",
 ]
 
 [[package]]
@@ -5137,7 +5210,7 @@ dependencies = [
  "hyper 0.14.30",
  "hyper 1.4.1",
  "hyper-util",
- "indexmap 2.0.1",
+ "indexmap 2.9.0",
  "ipnet",
  "itertools 0.10.5",
  "itoa",
@@ -5171,7 +5244,7 @@ dependencies = [
  "rsa",
  "rstest",
  "rustc-hash 1.1.0",
- "rustls 0.23.18",
+ "rustls 0.23.27",
  "rustls-native-certs 0.8.0",
  "rustls-pemfile 2.1.1",
  "scopeguard",
@@ -5190,13 +5263,14 @@ dependencies = [
  "tokio",
  "tokio-postgres",
  "tokio-postgres2",
- "tokio-rustls 0.26.0",
+ "tokio-rustls 0.26.2",
  "tokio-tungstenite 0.21.0",
  "tokio-util",
  "tracing",
  "tracing-log",
  "tracing-opentelemetry",
  "tracing-subscriber",
+ "tracing-test",
  "tracing-utils",
  "try-lock",
  "typed-json",
@@ -5413,13 +5487,13 @@ dependencies = [
  "num-bigint",
  "percent-encoding",
  "pin-project-lite",
- "rustls 0.23.18",
+ "rustls 0.23.27",
  "rustls-native-certs 0.8.0",
  "ryu",
  "sha1_smol",
  "socket2",
  "tokio",
- "tokio-rustls 0.26.0",
+ "tokio-rustls 0.26.2",
  "tokio-util",
  "url",
 ]
@@ -5867,15 +5941,15 @@ dependencies = [
 
 [[package]]
 name = "rustls"
-version = "0.23.18"
+version = "0.23.27"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "9c9cc1d47e243d655ace55ed38201c19ae02c148ae56412ab8750e8f0166ab7f"
+checksum = "730944ca083c1c233a75c09f199e973ca499344a2b7ba9e755c457e86fb4a321"
 dependencies = [
  "log",
  "once_cell",
  "ring",
  "rustls-pki-types",
- "rustls-webpki 0.102.8",
+ "rustls-webpki 0.103.3",
  "subtle",
  "zeroize",
 ]
@@ -5964,6 +6038,17 @@ dependencies = [
  "untrusted",
 ]
 
+[[package]]
+name = "rustls-webpki"
+version = "0.103.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e4a72fe2bcf7a6ac6fd7d0b9e5cb68aeb7d4c0a0271730218b3e92d43b4eb435"
+dependencies = [
+ "ring",
+ "rustls-pki-types",
+ "untrusted",
+]
+
 [[package]]
 name = "rustversion"
 version = "1.0.12"
@@ -6015,7 +6100,7 @@ dependencies = [
  "regex",
  "remote_storage",
  "reqwest",
- "rustls 0.23.18",
+ "rustls 0.23.27",
  "safekeeper_api",
  "safekeeper_client",
  "scopeguard",
@@ -6032,7 +6117,7 @@ dependencies = [
  "tokio",
  "tokio-io-timeout",
  "tokio-postgres",
- "tokio-rustls 0.26.0",
+ "tokio-rustls 0.26.2",
  "tokio-stream",
  "tokio-tar",
  "tokio-util",
@@ -6204,7 +6289,7 @@ checksum = "255914a8e53822abd946e2ce8baa41d4cded6b8e938913b7f7b9da5b7ab44335"
 dependencies = [
  "httpdate",
  "reqwest",
- "rustls 0.23.18",
+ "rustls 0.23.27",
  "sentry-backtrace",
  "sentry-contexts",
  "sentry-core",
@@ -6633,11 +6718,11 @@ dependencies = [
  "metrics",
  "once_cell",
  "parking_lot 0.12.1",
- "prost 0.13.3",
- "rustls 0.23.18",
+ "prost 0.13.5",
+ "rustls 0.23.27",
  "tokio",
- "tokio-rustls 0.26.0",
- "tonic",
+ "tokio-rustls 0.26.2",
+ "tonic 0.13.1",
  "tonic-build",
  "tracing",
  "utils",
@@ -6679,7 +6764,7 @@ dependencies = [
  "regex",
  "reqwest",
  "routerify",
- "rustls 0.23.18",
+ "rustls 0.23.27",
  "rustls-native-certs 0.8.0",
  "safekeeper_api",
  "safekeeper_client",
@@ -6694,7 +6779,7 @@ dependencies = [
  "tokio",
  "tokio-postgres",
  "tokio-postgres-rustls",
- "tokio-rustls 0.26.0",
+ "tokio-rustls 0.26.2",
  "tokio-util",
  "tracing",
  "utils",
@@ -6732,7 +6817,7 @@ dependencies = [
  "postgres_ffi",
  "remote_storage",
  "reqwest",
- "rustls 0.23.18",
+ "rustls 0.23.27",
  "rustls-native-certs 0.8.0",
  "serde",
  "serde_json",
@@ -7266,10 +7351,10 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "04fb792ccd6bbcd4bba408eb8a292f70fc4a3589e5d793626f45190e6454b6ab"
 dependencies = [
  "ring",
- "rustls 0.23.18",
+ "rustls 0.23.27",
  "tokio",
  "tokio-postgres",
- "tokio-rustls 0.26.0",
+ "tokio-rustls 0.26.2",
  "x509-certificate",
 ]
 
@@ -7313,12 +7398,11 @@ dependencies = [
 
 [[package]]
 name = "tokio-rustls"
-version = "0.26.0"
+version = "0.26.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "0c7bc40d0e5a97695bb96e27995cd3a08538541b0a846f65bba7a359f36700d4"
+checksum = "8e727b36a1a0e8b74c376ac2211e40c2c8af09fb4013c60d910495810f008e9b"
 dependencies = [
- "rustls 0.23.18",
- "rustls-pki-types",
+ "rustls 0.23.27",
  "tokio",
 ]
 
@@ -7416,7 +7500,7 @@ version = "0.22.14"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "f21c7aaf97f1bd9ca9d4f9e73b0a6c74bd5afef56f2bc931943a6e1c37e04e38"
 dependencies = [
- "indexmap 2.0.1",
+ "indexmap 2.9.0",
  "serde",
  "serde_spanned",
  "toml_datetime",
@@ -7435,18 +7519,41 @@ dependencies = [
  "http 1.1.0",
  "http-body 1.0.0",
  "http-body-util",
+ "percent-encoding",
+ "pin-project",
+ "prost 0.13.5",
+ "tokio-stream",
+ "tower-layer",
+ "tower-service",
+ "tracing",
+]
+
+[[package]]
+name = "tonic"
+version = "0.13.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7e581ba15a835f4d9ea06c55ab1bd4dce26fc53752c69a04aac00703bfb49ba9"
+dependencies = [
+ "async-trait",
+ "axum",
+ "base64 0.22.1",
+ "bytes",
+ "h2 0.4.4",
+ "http 1.1.0",
+ "http-body 1.0.0",
+ "http-body-util",
  "hyper 1.4.1",
  "hyper-timeout",
  "hyper-util",
  "percent-encoding",
  "pin-project",
- "prost 0.13.3",
+ "prost 0.13.5",
  "rustls-native-certs 0.8.0",
- "rustls-pemfile 2.1.1",
+ "socket2",
  "tokio",
- "tokio-rustls 0.26.0",
+ "tokio-rustls 0.26.2",
  "tokio-stream",
- "tower 0.4.13",
+ "tower 0.5.2",
  "tower-layer",
  "tower-service",
  "tracing",
@@ -7454,9 +7561,9 @@ dependencies = [
 
 [[package]]
 name = "tonic-build"
-version = "0.12.3"
+version = "0.13.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "9557ce109ea773b399c9b9e5dca39294110b74f1f342cb347a80d1fce8c26a11"
+checksum = "eac6f67be712d12f0b41328db3137e0d0757645d8904b4cb7d51cd9c2279e847"
 dependencies = [
  "prettyplease",
  "proc-macro2",
@@ -7466,6 +7573,19 @@ dependencies = [
  "syn 2.0.100",
 ]
 
+[[package]]
+name = "tonic-reflection"
+version = "0.13.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f9687bd5bfeafebdded2356950f278bba8226f0b32109537c4253406e09aafe1"
+dependencies = [
+ "prost 0.13.5",
+ "prost-types 0.13.3",
+ "tokio",
+ "tokio-stream",
+ "tonic 0.13.1",
+]
+
 [[package]]
 name = "tower"
 version = "0.4.13"
@@ -7474,16 +7594,11 @@ checksum = "b8fa9be0de6cf49e536ce1851f987bd21a43b771b09473c3549a6c853db37c1c"
 dependencies = [
  "futures-core",
  "futures-util",
- "indexmap 1.9.3",
  "pin-project",
  "pin-project-lite",
- "rand 0.8.5",
- "slab",
  "tokio",
- "tokio-util",
  "tower-layer",
  "tower-service",
- "tracing",
 ]
 
 [[package]]
@@ -7494,9 +7609,12 @@ checksum = "d039ad9159c98b70ecfd540b2573b97f7f52c3e8d9f8ad57a24b916a536975f9"
 dependencies = [
  "futures-core",
  "futures-util",
+ "indexmap 2.9.0",
  "pin-project-lite",
+ "slab",
  "sync_wrapper 1.0.1",
  "tokio",
+ "tokio-util",
  "tower-layer",
  "tower-service",
  "tracing",
@@ -7647,6 +7765,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "e8189decb5ac0fa7bc8b96b7cb9b2701d60d48805aca84a238004d665fcc4008"
 dependencies = [
  "matchers",
+ "nu-ansi-term",
  "once_cell",
  "regex",
  "serde",
@@ -7660,6 +7779,27 @@ dependencies = [
  "tracing-serde",
 ]
 
+[[package]]
+name = "tracing-test"
+version = "0.2.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "557b891436fe0d5e0e363427fc7f217abf9ccd510d5136549847bdcbcd011d68"
+dependencies = [
+ "tracing-core",
+ "tracing-subscriber",
+ "tracing-test-macro",
+]
+
+[[package]]
+name = "tracing-test-macro"
+version = "0.2.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "04659ddb06c87d233c566112c1c9c5b9e98256d9af50ec3bc9c8327f873a7568"
+dependencies = [
+ "quote",
+ "syn 2.0.100",
+]
+
 [[package]]
 name = "tracing-utils"
 version = "0.1.0"
@@ -7802,7 +7942,7 @@ dependencies = [
  "base64 0.22.1",
  "log",
  "once_cell",
- "rustls 0.23.18",
+ "rustls 0.23.27",
  "rustls-pki-types",
  "url",
  "webpki-roots",
@@ -7885,7 +8025,7 @@ dependencies = [
  "humantime",
  "jsonwebtoken",
  "metrics",
- "nix 0.27.1",
+ "nix 0.30.1",
  "once_cell",
  "pem",
  "pin-project-lite",
@@ -7997,7 +8137,7 @@ dependencies = [
  "pageserver_api",
  "postgres_ffi",
  "pprof",
- "prost 0.13.3",
+ "prost 0.13.5",
  "remote_storage",
  "serde",
  "serde_json",
@@ -8417,6 +8557,8 @@ dependencies = [
  "ahash",
  "anstream",
  "anyhow",
+ "axum",
+ "axum-core",
  "base64 0.13.1",
  "base64 0.21.7",
  "base64ct",
@@ -8451,14 +8593,14 @@ dependencies = [
  "hyper 0.14.30",
  "hyper 1.4.1",
  "hyper-util",
- "indexmap 1.9.3",
- "indexmap 2.0.1",
+ "indexmap 2.9.0",
  "itertools 0.12.1",
  "lazy_static",
  "libc",
  "log",
  "memchr",
  "nix 0.26.4",
+ "nix 0.30.1",
  "nom",
  "num",
  "num-bigint",
@@ -8472,16 +8614,16 @@ dependencies = [
  "parquet",
  "prettyplease",
  "proc-macro2",
- "prost 0.13.3",
+ "prost 0.13.5",
  "quote",
  "rand 0.8.5",
  "regex",
  "regex-automata 0.4.3",
  "regex-syntax 0.8.2",
  "reqwest",
- "rustls 0.23.18",
+ "rustls 0.23.27",
  "rustls-pki-types",
- "rustls-webpki 0.102.8",
+ "rustls-webpki 0.103.3",
  "scopeguard",
  "sec1 0.7.3",
  "serde",
@@ -8499,15 +8641,15 @@ dependencies = [
  "time",
  "time-macros",
  "tokio",
- "tokio-rustls 0.26.0",
+ "tokio-rustls 0.26.2",
  "tokio-stream",
  "tokio-util",
  "toml_edit",
- "tonic",
- "tower 0.4.13",
+ "tower 0.5.2",
  "tracing",
  "tracing-core",
  "tracing-log",
+ "tracing-subscriber",
  "url",
  "uuid",
  "zeroize",

Cargo.toml

@@ -9,6 +9,7 @@ members = [
     "pageserver/ctl",
     "pageserver/client",
     "pageserver/pagebench",
+    "pageserver/page_api",
     "proxy",
     "safekeeper",
     "safekeeper/client",
@@ -23,9 +24,11 @@ members = [
     "libs/postgres_ffi",
     "libs/safekeeper_api",
     "libs/desim",
+    "libs/neon-shmem",
     "libs/utils",
     "libs/consumption_metrics",
     "libs/postgres_backend",
+    "libs/posthog_client_lite",
     "libs/pq_proto",
     "libs/tenant_size_model",
     "libs/metrics",
@@ -126,7 +129,7 @@ md5 = "0.7.0"
 measured = { version = "0.0.22", features=["lasso"] }
 measured-process = { version = "0.0.22" }
 memoffset = "0.9"
-nix = { version = "0.27", features = ["dir", "fs", "process", "socket", "signal", "poll"] }
+nix = { version = "0.30.1", features = ["dir", "fs", "mman", "process", "socket", "signal", "poll"] }
 # Do not update to >= 7.0.0, at least. The update will have a significant impact
 # on compute startup metrics (start_postgres_ms), >= 25% degradation.
 notify = "6.0.0"
@@ -146,7 +149,7 @@ pin-project-lite = "0.2"
 pprof = { version = "0.14", features = ["criterion", "flamegraph", "frame-pointer", "prost-codec"] }
 procfs = "0.16"
 prometheus = {version = "0.13", default-features=false, features = ["process"]} # removes protobuf dependency
-prost = "0.13"
+prost = "0.13.5"
 rand = "0.8"
 redis = { version = "0.29.2", features = ["tokio-rustls-comp", "keep-alive"] }
 regex = "1.10.2"
@@ -196,7 +199,8 @@ tokio-tar = "0.3"
 tokio-util = { version = "0.7.10", features = ["io", "rt"] }
 toml = "0.8"
 toml_edit = "0.22"
-tonic = {version = "0.12.3", default-features = false, features = ["channel", "tls", "tls-roots"]}
+tonic = { version = "0.13.1", default-features = false, features = ["channel", "codegen", "prost", "router", "server", "tls-ring", "tls-native-roots"] }
+tonic-reflection = { version = "0.13.1", features = ["server"] }
 tower = { version = "0.5.2", default-features = false }
 tower-http = { version = "0.6.2", features = ["auth", "request-id", "trace"] }
 
@@ -243,6 +247,7 @@ azure_storage_blobs = { git = "https://github.com/neondatabase/azure-sdk-for-rus
 ## Local libraries
 compute_api = { version = "0.1", path = "./libs/compute_api/" }
 consumption_metrics = { version = "0.1", path = "./libs/consumption_metrics/" }
+desim = { version = "0.1", path = "./libs/desim" }
 endpoint_storage = { version = "0.0.1", path = "./endpoint_storage/" }
 http-utils = { version = "0.1", path = "./libs/http-utils/" }
 metrics = { version = "0.1", path = "./libs/metrics/" }
@@ -250,23 +255,24 @@ pageserver = { path = "./pageserver" }
 pageserver_api = { version = "0.1", path = "./libs/pageserver_api/" }
 pageserver_client = { path = "./pageserver/client" }
 pageserver_compaction = { version = "0.1", path = "./pageserver/compaction/" }
+pageserver_page_api = { path = "./pageserver/page_api" }
 postgres_backend = { version = "0.1", path = "./libs/postgres_backend/" }
 postgres_connection = { version = "0.1", path = "./libs/postgres_connection/" }
 postgres_ffi = { version = "0.1", path = "./libs/postgres_ffi/" }
 postgres_initdb = { path = "./libs/postgres_initdb" }
+posthog_client_lite = { version = "0.1", path = "./libs/posthog_client_lite" }
 pq_proto = { version = "0.1", path = "./libs/pq_proto/" }
 remote_storage = { version = "0.1", path = "./libs/remote_storage/" }
 safekeeper_api = { version = "0.1", path = "./libs/safekeeper_api" }
 safekeeper_client = { path = "./safekeeper/client" }
-desim = { version = "0.1", path = "./libs/desim" }
 storage_broker = { version = "0.1", path = "./storage_broker/" } # Note: main broker code is inside the binary crate, so linking with the library shouldn't be heavy.
 storage_controller_client = { path = "./storage_controller/client" }
 tenant_size_model = { version = "0.1", path = "./libs/tenant_size_model/" }
 tracing-utils = { version = "0.1", path = "./libs/tracing-utils/" }
 utils = { version = "0.1", path = "./libs/utils/" }
 vm_monitor = { version = "0.1", path = "./libs/vm_monitor/" }
-walproposer = { version = "0.1", path = "./libs/walproposer/" }
 wal_decoder = { version = "0.1", path = "./libs/wal_decoder" }
+walproposer = { version = "0.1", path = "./libs/walproposer/" }
 
 ## Common library dependency
 workspace_hack = { version = "0.1", path = "./workspace_hack/" }
@@ -276,7 +282,7 @@ criterion = "0.5.1"
 rcgen = "0.13"
 rstest = "0.18"
 camino-tempfile = "1.0.2"
-tonic-build = "0.12"
+tonic-build = "0.13.1"
 
 [patch.crates-io]
 

build-tools.Dockerfile

@@ -155,7 +155,7 @@ RUN set -e \
 
 # Keep the version the same as in compute/compute-node.Dockerfile and
 # test_runner/regress/test_compute_metrics.py.
-ENV SQL_EXPORTER_VERSION=0.17.0
+ENV SQL_EXPORTER_VERSION=0.17.3
 RUN curl -fsSL \
     "https://github.com/burningalchemist/sql_exporter/releases/download/${SQL_EXPORTER_VERSION}/sql_exporter-${SQL_EXPORTER_VERSION}.linux-$(case "$(uname -m)" in x86_64) echo amd64;; aarch64) echo arm64;; esac).tar.gz" \
     --output sql_exporter.tar.gz \
@@ -292,7 +292,7 @@ WORKDIR /home/nonroot
 
 # Rust
 # Please keep the version of llvm (installed above) in sync with rust llvm (`rustc --version --verbose | grep LLVM`)
-ENV RUSTC_VERSION=1.86.0
+ENV RUSTC_VERSION=1.87.0
 ENV RUSTUP_HOME="/home/nonroot/.rustup"
 ENV PATH="/home/nonroot/.cargo/bin:${PATH}"
 ARG RUSTFILT_VERSION=0.2.1

compute/compute-node.Dockerfile

@@ -582,6 +582,38 @@ RUN make -j $(getconf _NPROCESSORS_ONLN) && \
     make -j $(getconf _NPROCESSORS_ONLN) install && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/hypopg.control
 
+#########################################################################################
+#
+# Layer "online_advisor-build"
+# compile online_advisor extension
+#
+#########################################################################################
+FROM build-deps AS online_advisor-src
+ARG PG_VERSION
+
+# online_advisor supports all Postgres version starting from PG14, but prior to PG17 has to be included in preload_shared_libraries
+# last release 1.0 - May 15, 2025
+WORKDIR /ext-src
+RUN case "${PG_VERSION:?}" in \
+    "v17") \
+        ;; \
+    *) \
+        echo "skipping the version of online_advistor for $PG_VERSION" && exit 0 \
+        ;; \
+    esac && \
+	wget https://github.com/knizhnik/online_advisor/archive/refs/tags/1.0.tar.gz -O online_advisor.tar.gz && \
+    echo "059b7d9e5a90013a58bdd22e9505b88406ce05790675eb2d8434e5b215652d54 online_advisor.tar.gz" | sha256sum --check && \
+    mkdir online_advisor-src && cd online_advisor-src && tar xzf ../online_advisor.tar.gz --strip-components=1 -C .
+
+FROM pg-build AS online_advisor-build
+COPY --from=online_advisor-src /ext-src/ /ext-src/
+WORKDIR /ext-src/
+RUN if [ -d online_advisor-src ]; then \
+	    cd online_advisor-src && \
+        make -j install && \
+        echo 'trusted = true' >> /usr/local/pgsql/share/extension/online_advisor.control; \
+    fi
+
 #########################################################################################
 #
 # Layer "pg_hashids-build"
@@ -1117,8 +1149,8 @@ RUN wget https://github.com/microsoft/onnxruntime/archive/refs/tags/v1.18.1.tar.
     mkdir onnxruntime-src && cd onnxruntime-src && tar xzf ../onnxruntime.tar.gz --strip-components=1 -C . && \
     echo "#nothing to test here" > neon-test.sh
 
-RUN wget https://github.com/neondatabase-labs/pgrag/archive/refs/tags/v0.1.1.tar.gz -O pgrag.tar.gz &&  \
-    echo "087b2ecd11ba307dc968042ef2e9e43dc04d9ba60e8306e882c407bbe1350a50 pgrag.tar.gz" | sha256sum --check && \
+RUN wget https://github.com/neondatabase-labs/pgrag/archive/refs/tags/v0.1.2.tar.gz -O pgrag.tar.gz &&  \
+    echo "7361654ea24f08cbb9db13c2ee1c0fe008f6114076401bb871619690dafc5225 pgrag.tar.gz" | sha256sum --check && \
     mkdir pgrag-src && cd pgrag-src && tar xzf ../pgrag.tar.gz --strip-components=1 -C .
 
 FROM rust-extensions-build-pgrx14 AS pgrag-build
@@ -1648,6 +1680,7 @@ COPY --from=pg_jsonschema-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg_graphql-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg_tiktoken-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=hypopg-build /usr/local/pgsql/ /usr/local/pgsql/
+COPY --from=online_advisor-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg_hashids-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=rum-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pgtap-build /usr/local/pgsql/ /usr/local/pgsql/
@@ -1751,17 +1784,17 @@ ARG TARGETARCH
 RUN if [ "$TARGETARCH" = "amd64" ]; then\
         postgres_exporter_sha256='59aa4a7bb0f7d361f5e05732f5ed8c03cc08f78449cef5856eadec33a627694b';\
         pgbouncer_exporter_sha256='c9f7cf8dcff44f0472057e9bf52613d93f3ffbc381ad7547a959daa63c5e84ac';\
-        sql_exporter_sha256='38e439732bbf6e28ca4a94d7bc3686d3fa1abdb0050773d5617a9efdb9e64d08';\
+        sql_exporter_sha256='9a41127a493e8bfebfe692bf78c7ed2872a58a3f961ee534d1b0da9ae584aaab';\
     else\
         postgres_exporter_sha256='d1dedea97f56c6d965837bfd1fbb3e35a3b4a4556f8cccee8bd513d8ee086124';\
         pgbouncer_exporter_sha256='217c4afd7e6492ae904055bc14fe603552cf9bac458c063407e991d68c519da3';\
-        sql_exporter_sha256='11918b00be6e2c3a67564adfdb2414fdcbb15a5db76ea17d1d1a944237a893c6';\
+        sql_exporter_sha256='530e6afc77c043497ed965532c4c9dfa873bc2a4f0b3047fad367715c0081d6a';\
     fi\
     && curl -sL https://github.com/prometheus-community/postgres_exporter/releases/download/v0.17.1/postgres_exporter-0.17.1.linux-${TARGETARCH}.tar.gz\
      | tar xzf - --strip-components=1 -C.\
     && curl -sL https://github.com/prometheus-community/pgbouncer_exporter/releases/download/v0.10.2/pgbouncer_exporter-0.10.2.linux-${TARGETARCH}.tar.gz\
      | tar xzf - --strip-components=1 -C.\
-    && curl -sL https://github.com/burningalchemist/sql_exporter/releases/download/0.17.0/sql_exporter-0.17.0.linux-${TARGETARCH}.tar.gz\
+    && curl -sL https://github.com/burningalchemist/sql_exporter/releases/download/0.17.3/sql_exporter-0.17.3.linux-${TARGETARCH}.tar.gz\
      | tar xzf - --strip-components=1 -C.\
     && echo "${postgres_exporter_sha256} postgres_exporter" | sha256sum -c -\
     && echo "${pgbouncer_exporter_sha256} pgbouncer_exporter" | sha256sum -c -\
@@ -1814,7 +1847,7 @@ COPY docker-compose/ext-src/ /ext-src/
 COPY --from=pg-build /postgres /postgres
 #COPY --from=postgis-src /ext-src/ /ext-src/
 COPY --from=plv8-src /ext-src/ /ext-src/
-#COPY --from=h3-pg-src /ext-src/ /ext-src/
+COPY --from=h3-pg-src /ext-src/h3-pg-src /ext-src/h3-pg-src
 COPY --from=postgresql-unit-src /ext-src/ /ext-src/
 COPY --from=pgvector-src /ext-src/ /ext-src/
 COPY --from=pgjwt-src /ext-src/ /ext-src/
@@ -1823,6 +1856,7 @@ COPY --from=pgjwt-src /ext-src/ /ext-src/
 COPY --from=pg_graphql-src /ext-src/ /ext-src/
 #COPY --from=pg_tiktoken-src /ext-src/ /ext-src/
 COPY --from=hypopg-src /ext-src/ /ext-src/
+COPY --from=online_advisor-src /ext-src/ /ext-src/
 COPY --from=pg_hashids-src /ext-src/ /ext-src/
 COPY --from=rum-src /ext-src/ /ext-src/
 COPY --from=pgtap-src /ext-src/ /ext-src/
@@ -1971,7 +2005,8 @@ COPY --from=sql_exporter_preprocessor --chmod=0644 /home/nonroot/compute/etc/sql
 COPY --from=sql_exporter_preprocessor --chmod=0644 /home/nonroot/compute/etc/neon_collector_autoscaling.yml /etc/neon_collector_autoscaling.yml
 
 # Make the libraries we built available
-RUN echo '/usr/local/lib' >> /etc/ld.so.conf && /sbin/ldconfig
+COPY --chmod=0666 compute/etc/ld.so.conf.d/00-neon.conf /etc/ld.so.conf.d/00-neon.conf
+RUN /sbin/ldconfig
 
 # rsyslog config permissions
 # directory for rsyslogd pid file

compute/etc/ld.so.conf.d/00-neon.conf

@@ -0,0 +1 @@
+/usr/local/lib

compute/patches/rum.patch

@@ -7,7 +7,7 @@ index 255e616..1c6edb7 100644
  			 RelationGetRelationName(index));
  
 +#ifdef NEON_SMGR
-+	smgr_start_unlogged_build(index->rd_smgr);
++	smgr_start_unlogged_build(RelationGetSmgr(index));
 +#endif
 +
  	initRumState(&buildstate.rumstate, index);
@@ -18,7 +18,7 @@ index 255e616..1c6edb7 100644
  	rumUpdateStats(index, &buildstate.buildStats, buildstate.rumstate.isBuild);
  
 +#ifdef NEON_SMGR
-+	smgr_finish_unlogged_build_phase_1(index->rd_smgr);
++	smgr_finish_unlogged_build_phase_1(RelationGetSmgr(index));
 +#endif
 +
  	/*
@@ -29,7 +29,7 @@ index 255e616..1c6edb7 100644
  	}
  
 +#ifdef NEON_SMGR
-+	smgr_end_unlogged_build(index->rd_smgr);
++	smgr_end_unlogged_build(RelationGetSmgr(index));
 +#endif
 +
  	/*

compute_tools/src/compute.rs

@@ -695,25 +695,18 @@ impl ComputeNode {
                 let log_directory_path = Path::new(&self.params.pgdata).join("log");
                 let log_directory_path = log_directory_path.to_string_lossy().to_string();
 
-                // Add project_id,endpoint_id tag to identify the logs.
+                // Add project_id,endpoint_id to identify the logs.
                 //
                 // These ids are passed from cplane,
-                // for backwards compatibility (old computes that don't have them),
-                // we set them to None.
-                // TODO: Clean up this code when all computes have them.
-                let tag: Option<String> = match (
-                    pspec.spec.project_id.as_deref(),
-                    pspec.spec.endpoint_id.as_deref(),
-                ) {
-                    (Some(project_id), Some(endpoint_id)) => {
-                        Some(format!("{project_id}/{endpoint_id}"))
-                    }
-                    (Some(project_id), None) => Some(format!("{project_id}/None")),
-                    (None, Some(endpoint_id)) => Some(format!("None,{endpoint_id}")),
-                    (None, None) => None,
-                };
+                let endpoint_id = pspec.spec.endpoint_id.as_deref().unwrap_or("");
+                let project_id = pspec.spec.project_id.as_deref().unwrap_or("");
 
-                configure_audit_rsyslog(log_directory_path.clone(), tag, &remote_endpoint)?;
+                configure_audit_rsyslog(
+                    log_directory_path.clone(),
+                    endpoint_id,
+                    project_id,
+                    &remote_endpoint,
+                )?;
 
                 // Launch a background task to clean up the audit logs
                 launch_pgaudit_gc(log_directory_path);

compute_tools/src/config_template/compute_audit_rsyslog_template.conf

@@ -2,10 +2,24 @@
 module(load="imfile")
 
 # Input configuration for log files in the specified directory
-# Replace {log_directory} with the directory containing the log files
-input(type="imfile" File="{log_directory}/*.log" Tag="{tag}" Severity="info" Facility="local0")
+# The messages can be multiline. The start of the message is a timestamp
+# in "%Y-%m-%d %H:%M:%S.%3N GMT" (so timezone hardcoded).
+# Replace log_directory with the directory containing the log files
+input(type="imfile" File="{log_directory}/*.log"
+  Tag="pgaudit_log" Severity="info" Facility="local5"
+  startmsg.regex="^[[:digit:]]{{4}}-[[:digit:]]{{2}}-[[:digit:]]{{2}} [[:digit:]]{{2}}:[[:digit:]]{{2}}:[[:digit:]]{{2}}.[[:digit:]]{{3}} GMT,")
+
 # the directory to store rsyslog state files
 global(workDirectory="/var/log/rsyslog")
 
-# Forward logs to remote syslog server
-*.* @@{remote_endpoint}
+# Construct json, endpoint_id and project_id as additional metadata
+set $.json_log!endpoint_id = "{endpoint_id}";
+set $.json_log!project_id = "{project_id}";
+set $.json_log!msg = $msg;
+
+# Template suitable for rfc5424 syslog format
+template(name="PgAuditLog" type="string"
+    string="<%PRI%>1 %TIMESTAMP:::date-rfc3339% %HOSTNAME% - - - - %$.json_log%")
+
+# Forward to remote syslog receiver (@@<hostname>:<port>;format
+local5.info @@{remote_endpoint};PgAuditLog

compute_tools/src/pg_helpers.rs

@@ -213,8 +213,10 @@ impl Escaping for PgIdent {
 
         // Find the first suitable tag that is not present in the string.
         // Postgres' max role/DB name length is 63 bytes, so even in the
-        // worst case it won't take long.
-        while self.contains(&format!("${tag}$")) || self.contains(&format!("${outer_tag}$")) {
+        // worst case it won't take long. Outer tag is always `tag + "x"`,
+        // so if `tag` is not present in the string, `outer_tag` is not
+        // present in the string either.
+        while self.contains(&tag.to_string()) {
             tag += "x";
             outer_tag = tag.clone() + "x";
         }

compute_tools/src/rsyslog.rs

@@ -27,6 +27,40 @@ fn get_rsyslog_pid() -> Option<String> {
     }
 }
 
+fn wait_for_rsyslog_pid() -> Result<String, anyhow::Error> {
+    const MAX_WAIT: Duration = Duration::from_secs(5);
+    const INITIAL_SLEEP: Duration = Duration::from_millis(2);
+
+    let mut sleep_duration = INITIAL_SLEEP;
+    let start = std::time::Instant::now();
+    let mut attempts = 1;
+
+    for attempt in 1.. {
+        attempts = attempt;
+        match get_rsyslog_pid() {
+            Some(pid) => return Ok(pid),
+            None => {
+                if start.elapsed() >= MAX_WAIT {
+                    break;
+                }
+                info!(
+                    "rsyslogd is not running, attempt {}. Sleeping for {} ms",
+                    attempt,
+                    sleep_duration.as_millis()
+                );
+                std::thread::sleep(sleep_duration);
+                sleep_duration *= 2;
+            }
+        }
+    }
+
+    Err(anyhow::anyhow!(
+        "rsyslogd is not running after waiting for {} seconds and {} attempts",
+        attempts,
+        start.elapsed().as_secs()
+    ))
+}
+
 // Restart rsyslogd to apply the new configuration.
 // This is necessary, because there is no other way to reload the rsyslog configuration.
 //
@@ -36,27 +70,29 @@ fn get_rsyslog_pid() -> Option<String> {
 // TODO: test it properly
 //
 fn restart_rsyslog() -> Result<()> {
-    let old_pid = get_rsyslog_pid().context("rsyslogd is not running")?;
-    info!("rsyslogd is running with pid: {}, restart it", old_pid);
-
     // kill it to restart
     let _ = Command::new("pkill")
         .arg("rsyslogd")
         .output()
-        .context("Failed to stop rsyslogd")?;
+        .context("Failed to restart rsyslogd")?;
+
+    // ensure rsyslogd is running
+    wait_for_rsyslog_pid()?;
 
     Ok(())
 }
 
 pub fn configure_audit_rsyslog(
     log_directory: String,
-    tag: Option<String>,
+    endpoint_id: &str,
+    project_id: &str,
     remote_endpoint: &str,
 ) -> Result<()> {
     let config_content: String = format!(
         include_str!("config_template/compute_audit_rsyslog_template.conf"),
         log_directory = log_directory,
-        tag = tag.unwrap_or("".to_string()),
+        endpoint_id = endpoint_id,
+        project_id = project_id,
         remote_endpoint = remote_endpoint
     );
 
@@ -131,15 +167,11 @@ pub fn configure_postgres_logs_export(conf: PostgresLogsRsyslogConfig) -> Result
         return Ok(());
     }
 
-    // When new config is empty we can simply remove the configuration file.
+    // Nothing to configure
     if new_config.is_empty() {
-        info!("removing rsyslog config file: {}", POSTGRES_LOGS_CONF_PATH);
-        match std::fs::remove_file(POSTGRES_LOGS_CONF_PATH) {
-            Ok(_) => {}
-            Err(err) if err.kind() == ErrorKind::NotFound => {}
-            Err(err) => return Err(err.into()),
-        }
-        restart_rsyslog()?;
+        // When the configuration is removed, PostgreSQL will stop sending data
+        // to the files watched by rsyslog, so restarting rsyslog is more effort
+        // than just ignoring this change.
         return Ok(());
     }
 

compute_tools/tests/pg_helpers_tests.rs

@@ -71,6 +71,14 @@ test.escaping = 'here''s a backslash \\ and a quote '' and a double-quote " hoor
             ("name$$$", ("$x$name$$$$x$", "xx")),
             ("name$$$$", ("$x$name$$$$$x$", "xx")),
             ("name$x$", ("$xx$name$x$$xx$", "xxx")),
+            ("x", ("$xx$x$xx$", "xxx")),
+            ("xx", ("$xxx$xx$xxx$", "xxxx")),
+            ("$x", ("$xx$$x$xx$", "xxx")),
+            ("x$", ("$xx$x$$xx$", "xxx")),
+            ("$x$", ("$xx$$x$$xx$", "xxx")),
+            ("xx$", ("$xxx$xx$$xxx$", "xxxx")),
+            ("$xx", ("$xxx$$xx$xxx$", "xxxx")),
+            ("$xx$", ("$xxx$$xx$$xxx$", "xxxx")),
         ];
 
         for (input, expected) in test_cases {

control_plane/safekeepers.conf

@@ -2,8 +2,10 @@
 [pageserver]
 listen_pg_addr = '127.0.0.1:64000'
 listen_http_addr = '127.0.0.1:9898'
+listen_grpc_addr = '127.0.0.1:51051'
 pg_auth_type = 'Trust'
 http_auth_type = 'Trust'
+grpc_auth_type = 'Trust'
 
 [[safekeepers]]
 id = 1

control_plane/simple.conf

@@ -4,8 +4,10 @@
 id=1
 listen_pg_addr = '127.0.0.1:64000'
 listen_http_addr = '127.0.0.1:9898'
+listen_grpc_addr = '127.0.0.1:51051'
 pg_auth_type = 'Trust'
 http_auth_type = 'Trust'
+grpc_auth_type = 'Trust'
 
 [[safekeepers]]
 id = 1

control_plane/src/background_process.rs

@@ -14,7 +14,7 @@
 
 use std::ffi::OsStr;
 use std::io::Write;
-use std::os::unix::prelude::AsRawFd;
+use std::os::fd::AsFd;
 use std::os::unix::process::CommandExt;
 use std::path::Path;
 use std::process::Command;
@@ -356,7 +356,7 @@ where
             let file = pid_file::claim_for_current_process(&path).expect("claim pid file");
             // Remove the FD_CLOEXEC flag on the pidfile descriptor so that the pidfile
             // remains locked after exec.
-            nix::fcntl::fcntl(file.as_raw_fd(), FcntlArg::F_SETFD(FdFlag::empty()))
+            nix::fcntl::fcntl(file.as_fd(), FcntlArg::F_SETFD(FdFlag::empty()))
                 .expect("remove FD_CLOEXEC");
             // Don't run drop(file), it would close the file before we actually exec.
             std::mem::forget(file);

control_plane/src/bin/neon_local.rs

@@ -8,7 +8,6 @@
 use std::borrow::Cow;
 use std::collections::{BTreeSet, HashMap};
 use std::fs::File;
-use std::os::fd::AsRawFd;
 use std::path::PathBuf;
 use std::process::exit;
 use std::str::FromStr;
@@ -31,8 +30,9 @@ use control_plane::safekeeper::SafekeeperNode;
 use control_plane::storage_controller::{
     NeonStorageControllerStartArgs, NeonStorageControllerStopArgs, StorageController,
 };
-use nix::fcntl::{FlockArg, flock};
+use nix::fcntl::{Flock, FlockArg};
 use pageserver_api::config::{
+    DEFAULT_GRPC_LISTEN_PORT as DEFAULT_PAGESERVER_GRPC_PORT,
     DEFAULT_HTTP_LISTEN_PORT as DEFAULT_PAGESERVER_HTTP_PORT,
     DEFAULT_PG_LISTEN_PORT as DEFAULT_PAGESERVER_PG_PORT,
 };
@@ -749,16 +749,16 @@ struct TimelineTreeEl {
 
 /// A flock-based guard over the neon_local repository directory
 struct RepoLock {
-    _file: File,
+    _file: Flock<File>,
 }
 
 impl RepoLock {
     fn new() -> Result<Self> {
         let repo_dir = File::open(local_env::base_path())?;
-        let repo_dir_fd = repo_dir.as_raw_fd();
-        flock(repo_dir_fd, FlockArg::LockExclusive)?;
-
-        Ok(Self { _file: repo_dir })
+        match Flock::lock(repo_dir, FlockArg::LockExclusive) {
+            Ok(f) => Ok(Self { _file: f }),
+            Err((_, e)) => Err(e).context("flock error"),
+        }
     }
 }
 
@@ -1008,13 +1008,16 @@ fn handle_init(args: &InitCmdArgs) -> anyhow::Result<LocalEnv> {
                     let pageserver_id = NodeId(DEFAULT_PAGESERVER_ID.0 + i as u64);
                     let pg_port = DEFAULT_PAGESERVER_PG_PORT + i;
                     let http_port = DEFAULT_PAGESERVER_HTTP_PORT + i;
+                    let grpc_port = DEFAULT_PAGESERVER_GRPC_PORT + i;
                     NeonLocalInitPageserverConf {
                         id: pageserver_id,
                         listen_pg_addr: format!("127.0.0.1:{pg_port}"),
                         listen_http_addr: format!("127.0.0.1:{http_port}"),
                         listen_https_addr: None,
+                        listen_grpc_addr: Some(format!("127.0.0.1:{grpc_port}")),
                         pg_auth_type: AuthType::Trust,
                         http_auth_type: AuthType::Trust,
+                        grpc_auth_type: AuthType::Trust,
                         other: Default::default(),
                         // Typical developer machines use disks with slow fsync, and we don't care
                         // about data integrity: disable disk syncs.
@@ -1276,6 +1279,7 @@ async fn handle_timeline(cmd: &TimelineCmd, env: &mut local_env::LocalEnv) -> Re
                 mode: pageserver_api::models::TimelineCreateRequestMode::Branch {
                     ancestor_timeline_id,
                     ancestor_start_lsn: start_lsn,
+                    read_only: false,
                     pg_version: None,
                 },
             };

control_plane/src/local_env.rs

@@ -278,8 +278,10 @@ pub struct PageServerConf {
     pub listen_pg_addr: String,
     pub listen_http_addr: String,
     pub listen_https_addr: Option<String>,
+    pub listen_grpc_addr: Option<String>,
     pub pg_auth_type: AuthType,
     pub http_auth_type: AuthType,
+    pub grpc_auth_type: AuthType,
     pub no_sync: bool,
 }
 
@@ -290,8 +292,10 @@ impl Default for PageServerConf {
             listen_pg_addr: String::new(),
             listen_http_addr: String::new(),
             listen_https_addr: None,
+            listen_grpc_addr: None,
             pg_auth_type: AuthType::Trust,
             http_auth_type: AuthType::Trust,
+            grpc_auth_type: AuthType::Trust,
             no_sync: false,
         }
     }
@@ -306,8 +310,10 @@ pub struct NeonLocalInitPageserverConf {
     pub listen_pg_addr: String,
     pub listen_http_addr: String,
     pub listen_https_addr: Option<String>,
+    pub listen_grpc_addr: Option<String>,
     pub pg_auth_type: AuthType,
     pub http_auth_type: AuthType,
+    pub grpc_auth_type: AuthType,
     #[serde(default, skip_serializing_if = "std::ops::Not::not")]
     pub no_sync: bool,
     #[serde(flatten)]
@@ -321,8 +327,10 @@ impl From<&NeonLocalInitPageserverConf> for PageServerConf {
             listen_pg_addr,
             listen_http_addr,
             listen_https_addr,
+            listen_grpc_addr,
             pg_auth_type,
             http_auth_type,
+            grpc_auth_type,
             no_sync,
             other: _,
         } = conf;
@@ -331,7 +339,9 @@ impl From<&NeonLocalInitPageserverConf> for PageServerConf {
             listen_pg_addr: listen_pg_addr.clone(),
             listen_http_addr: listen_http_addr.clone(),
             listen_https_addr: listen_https_addr.clone(),
+            listen_grpc_addr: listen_grpc_addr.clone(),
             pg_auth_type: *pg_auth_type,
+            grpc_auth_type: *grpc_auth_type,
             http_auth_type: *http_auth_type,
             no_sync: *no_sync,
         }
@@ -707,8 +717,10 @@ impl LocalEnv {
                     listen_pg_addr: String,
                     listen_http_addr: String,
                     listen_https_addr: Option<String>,
+                    listen_grpc_addr: Option<String>,
                     pg_auth_type: AuthType,
                     http_auth_type: AuthType,
+                    grpc_auth_type: AuthType,
                     #[serde(default)]
                     no_sync: bool,
                 }
@@ -732,8 +744,10 @@ impl LocalEnv {
                     listen_pg_addr,
                     listen_http_addr,
                     listen_https_addr,
+                    listen_grpc_addr,
                     pg_auth_type,
                     http_auth_type,
+                    grpc_auth_type,
                     no_sync,
                 } = config_toml;
                 let IdentityTomlSubset {
@@ -750,8 +764,10 @@ impl LocalEnv {
                     listen_pg_addr,
                     listen_http_addr,
                     listen_https_addr,
+                    listen_grpc_addr,
                     pg_auth_type,
                     http_auth_type,
+                    grpc_auth_type,
                     no_sync,
                 };
                 pageservers.push(conf);

control_plane/src/pageserver.rs

@@ -129,7 +129,9 @@ impl PageServerNode {
             ));
         }
 
-        if conf.http_auth_type != AuthType::Trust || conf.pg_auth_type != AuthType::Trust {
+        if [conf.http_auth_type, conf.pg_auth_type, conf.grpc_auth_type]
+            .contains(&AuthType::NeonJWT)
+        {
             // Keys are generated in the toplevel repo dir, pageservers' workdirs
             // are one level below that, so refer to keys with ../
             overrides.push("auth_validation_public_key_path='../auth_public_key.pem'".to_owned());
@@ -546,6 +548,16 @@ impl PageServerNode {
                 .map(serde_json::from_str)
                 .transpose()
                 .context("Falied to parse 'sampling_ratio'")?,
+            relsize_snapshot_cache_capacity: settings
+                .remove("relsize snapshot cache capacity")
+                .map(|x| x.parse::<usize>())
+                .transpose()
+                .context("Falied to parse 'relsize_snapshot_cache_capacity' as integer")?,
+            basebackup_cache_enabled: settings
+                .remove("basebackup_cache_enabled")
+                .map(|x| x.parse::<bool>())
+                .transpose()
+                .context("Failed to parse 'basebackup_cache_enabled' as bool")?,
         };
         if !settings.is_empty() {
             bail!("Unrecognized tenant settings: {settings:?}")

docker-compose/compute_wrapper/shell/compute.sh

@@ -14,6 +14,14 @@ PG_VERSION=${PG_VERSION:-14}
 CONFIG_FILE_ORG=/var/db/postgres/configs/config.json
 CONFIG_FILE=/tmp/config.json
 
+# Test that the first library path that the dynamic loader looks in is the path
+# that we use for custom compiled software
+first_path="$(ldconfig --verbose 2>/dev/null \
+    | grep --invert-match ^$'\t' \
+    | cut --delimiter=: --fields=1 \
+    | head --lines=1)"
+test "$first_path" == '/usr/local/lib' || true # Remove the || true in a follow-up PR. Needed for backwards compat.
+
 echo "Waiting pageserver become ready."
 while ! nc -z pageserver 6400; do
      sleep 1;

docker-compose/ext-src/alter_db.sh

@@ -0,0 +1,8 @@
+#!/bin/bash
+# We need these settings to get the expected output results.
+# We cannot use the environment variables e.g. PGTZ due to
+# https://github.com/neondatabase/neon/issues/1287
+export DATABASE=${1:-contrib_regression}
+psql -c "ALTER DATABASE ${DATABASE} SET neon.allow_unstable_extensions='on'" \
+     -c "ALTER DATABASE ${DATABASE} SET DateStyle='Postgres,MDY'" \
+     -c "ALTER DATABASE ${DATABASE} SET TimeZone='America/Los_Angeles'" \

docker-compose/ext-src/h3-pg-src/neon-test.sh

@@ -0,0 +1,16 @@
+#!/usr/bin/env bash
+set -ex
+cd "$(dirname "${0}")"
+PG_REGRESS=$(dirname "$(pg_config --pgxs)")/../test/regress/pg_regress
+dropdb --if-exists contrib_regression
+createdb contrib_regression
+cd h3_postgis/test
+psql -d contrib_regression -c "CREATE EXTENSION postgis" -c "CREATE EXTENSION postgis_raster" -c "CREATE EXTENSION h3" -c "CREATE EXTENSION h3_postgis"
+TESTS=$(echo sql/* | sed 's|sql/||g; s|\.sql||g')
+${PG_REGRESS} --use-existing --dbname contrib_regression ${TESTS}
+cd ../../h3/test
+TESTS=$(echo sql/* | sed 's|sql/||g; s|\.sql||g')
+dropdb --if-exists contrib_regression
+createdb contrib_regression
+psql -d contrib_regression -c "CREATE EXTENSION h3"
+${PG_REGRESS} --use-existing --dbname contrib_regression ${TESTS}

docker-compose/ext-src/h3-pg-src/test-upgrade.sh

@@ -0,0 +1,7 @@
+#!/bin/sh
+set -ex
+cd "$(dirname ${0})"
+PG_REGRESS=$(dirname "$(pg_config --pgxs)")/../test/regress/pg_regress
+cd h3/test
+TESTS=$(echo sql/* | sed 's|sql/||g; s|\.sql||g')
+${PG_REGRESS} --use-existing --inputdir=./ --bindir='/usr/local/pgsql/bin'  --dbname=contrib_regression  ${TESTS}

docker-compose/ext-src/online_advisor-src/neon-test.sh

@@ -0,0 +1,6 @@
+#!/bin/sh
+set -ex
+cd "$(dirname "${0}")"
+if [ -f Makefile ]; then
+  make installcheck
+fi

docker-compose/ext-src/online_advisor-src/regular-test.sh

@@ -0,0 +1,9 @@
+#!/bin/sh
+set -ex
+cd "$(dirname ${0})"
+[ -f Makefile ] || exit 0
+dropdb --if-exist contrib_regression
+createdb contrib_regression
+PG_REGRESS=$(dirname "$(pg_config --pgxs)")/../test/regress/pg_regress
+TESTS=$(echo sql/* | sed 's|sql/||g; s|\.sql||g')
+${PG_REGRESS} --use-existing --inputdir=./ --bindir='/usr/local/pgsql/bin' --dbname=contrib_regression ${TESTS}

docker-compose/ext-src/pg_graphql-src/regular-test.sh

@@ -18,6 +18,7 @@ TESTS=${TESTS/row_level_security/}
 TESTS=${TESTS/sqli_connection/}
 dropdb --if-exist contrib_regression
 createdb contrib_regression
+. ../alter_db.sh
 psql -v ON_ERROR_STOP=1 -f test/fixtures.sql -d contrib_regression
 ${REGRESS} --use-existing --dbname=contrib_regression --inputdir=${TESTDIR} ${TESTS}
 

docker-compose/ext-src/pgrag-src/regular-test.sh

@@ -3,6 +3,7 @@ set -ex
 cd "$(dirname "${0}")"
 dropdb --if-exist contrib_regression
 createdb contrib_regression
+. ../alter_db.sh
 psql -d contrib_regression -c "CREATE EXTENSION vector" -c "CREATE EXTENSION rag"
 PG_REGRESS=$(dirname "$(pg_config --pgxs)")/../test/regress/pg_regress
 ${PG_REGRESS} --inputdir=./ --bindir='/usr/local/pgsql/bin'    --use-existing --load-extension=vector --load-extension=rag --dbname=contrib_regression basic_functions text_processing api_keys chunking_functions document_processing embedding_api_functions voyageai_functions

docker-compose/ext-src/pgx_ulid-src/Makefile

@@ -20,5 +20,6 @@ installcheck: regression-test
 regression-test:
 	dropdb --if-exists contrib_regression
 	createdb contrib_regression
+	../alter_db.sh
 	psql -d contrib_regression -c "CREATE EXTENSION $(EXTNAME)"
 	$(PG_REGRESS) --inputdir=. --outputdir=. --use-existing --dbname=contrib_regression $(REGRESS)

docker-compose/ext-src/plv8-src/regular-test.sh

@@ -3,6 +3,7 @@ set -ex
 cd "$(dirname ${0})"
 dropdb --if-exist contrib_regression
 createdb contrib_regression
+. ../alter_db.sh
 PG_REGRESS=$(dirname "$(pg_config --pgxs)")/../test/regress/pg_regress
 REGRESS="$(make -n installcheck | awk '{print substr($0,index($0,"init-extension"));}')"
 REGRESS="${REGRESS/startup_perms/}"

docker-compose/ext-src/rag_bge_small_en_v15-src/Makefile

@@ -11,5 +11,6 @@ PG_REGRESS := $(dir $(PGXS))../../src/test/regress/pg_regress
 installcheck:
 	dropdb --if-exists contrib_regression
 	createdb contrib_regression
+	../alter_db.sh
 	psql -d contrib_regression -c "CREATE EXTENSION vector" -c "CREATE EXTENSION rag_bge_small_en_v15"
 	$(PG_REGRESS) --use-existing --dbname=contrib_regression $(REGRESS)

docker-compose/ext-src/rag_jina_reranker_v1_tiny_en-src/Makefile

@@ -11,5 +11,6 @@ PG_REGRESS := $(dir $(PGXS))../../src/test/regress/pg_regress
 installcheck:
 	dropdb --if-exists contrib_regression
 	createdb contrib_regression
+	../alter_db.sh
 	psql -d contrib_regression -c "CREATE EXTENSION vector" -c "CREATE EXTENSION rag_jina_reranker_v1_tiny_en"
 	$(PG_REGRESS) --use-existing --dbname=contrib_regression $(REGRESS)

docker-compose/ext-src/rum-src/regular-test.sh

@@ -3,5 +3,6 @@ set -ex
 cd "$(dirname ${0})"
 dropdb --if-exist contrib_regression
 createdb contrib_regression
+. ../alter_db.sh
 PG_REGRESS=$(dirname "$(pg_config --pgxs)")/../test/regress/pg_regress
 ${PG_REGRESS} --inputdir=./ --bindir='/usr/local/pgsql/bin' --use-existing --dbname=contrib_regression rum rum_hash ruminv timestamp orderby orderby_hash altorder altorder_hash limits int2 int4 int8 float4 float8 money oid time timetz date interval macaddr inet cidr text varchar char bytea bit varbit numeric rum_weight expr array

docker-compose/test_extensions_upgrade.sh

@@ -82,7 +82,8 @@ EXTENSIONS='[
 {"extname": "pg_ivm", "extdir": "pg_ivm-src"},
 {"extname": "pgjwt", "extdir": "pgjwt-src"},
 {"extname": "pgtap", "extdir": "pgtap-src"},
-{"extname": "pg_repack", "extdir": "pg_repack-src"}
+{"extname": "pg_repack", "extdir": "pg_repack-src"},
+{"extname": "h3", "extdir": "h3-pg-src"}
 ]'
 EXTNAMES=$(echo ${EXTENSIONS} | jq -r '.[].extname' | paste -sd ' ' -)
 COMPUTE_TAG=${NEW_COMPUTE_TAG} docker compose --profile test-extensions up --quiet-pull --build -d

docs/rfcs/043-bottom-most-gc-compaction.md

@@ -0,0 +1,194 @@
+# Bottommost Garbage-Collection Compaction
+
+## Summary
+
+The goal of this doc is to propose a way to reliably collect garbages below the GC horizon. This process is called bottom-most garbage-collect-compaction, and is part of the broader legacy-enhanced compaction that we plan to implement in the future.
+
+## Motivation
+
+The current GC algorithm will wait until the covering via image layers before collecting the garbages of a key region. Relying on image layer generation to generate covering images is not reliable. There are prior arts to generate feedbacks from the GC algorithm to the image generation process to accelerate garbage collection, but it slows down the system and creates write amplification.
+
+# Basic Idea
+
+![](images/036-bottom-most-gc-compaction/01-basic-idea.svg)
+
+The idea of bottom-most compaction is simple: we rewrite all layers that are below or intersect with the GC horizon to produce a flat level of image layers at the GC horizon and deltas above the GC horizon. In this process,
+
+- All images and deltas ≤ GC horizon LSN will be dropped. This process collects garbages.
+- We produce images for all keys involved in the compaction process at the GC horizon.
+
+Therefore, it can precisely collect all garbages below the horizon, and reduce the space amplification, i.e., in the staircase pattern (test_gc_feedback).
+
+![The staircase pattern in test_gc_feedback in the original compaction algorithm. The goal is to collect garbage below the red horizontal line.](images/036-bottom-most-gc-compaction/12-staircase-test-gc-feedback.png)
+
+The staircase pattern in test_gc_feedback in the original compaction algorithm. The goal is to collect garbage below the red horizontal line.
+
+# Branches
+
+With branches, the bottom-most compaction should retain a snapshot of the keyspace at the `retain_lsn` so that the child branch can access data at the branch point. This requires some modifications to the basic bottom-most compaction algorithm that we sketched above. 
+
+![](images/036-bottom-most-gc-compaction/03-retain-lsn.svg)
+
+## Single Timeline w/ Snapshots: handle `retain_lsn`
+
+First let’s look into the case where we create branches over the main branch but don’t write any data to them (aka “snapshots”).
+
+The bottom-most compaction algorithm collects all deltas and images of a key and can make decisions on what data to retain. Given that we have a single key’s history as below:
+
+```
+LSN 0x10 -> A
+LSN 0x20 -> append B
+retain_lsn: 0x20
+LSN 0x30 -> append C
+LSN 0x40 -> append D
+retain_lsn: 0x40
+LSN 0x50 -> append E
+GC horizon: 0x50
+LSN 0x60 -> append F
+```
+
+The algorithm will produce:
+
+```
+LSN 0x20 -> AB
+(drop all history below the earliest retain_lsn)
+LSN 0x40 -> ABCD
+(assume the cost of replaying 2 deltas is higher than storing the full image, we generate an image here)
+LSN 0x50 -> append E
+(replay one delta is cheap)
+LSN 0x60 -> append F
+(keep everything as-is above the GC horizon)
+```
+
+![](images/036-bottom-most-gc-compaction/05-btmgc-parent.svg)
+
+What happens is that we balance the space taken by each retain_lsn and the cost of replaying deltas during the bottom-most compaction process. This is controlled by a threshold. If `count(deltas) < $threshold`, the deltas will be retained. Otherwise, an image will be generated and the deltas will be dropped.
+
+In the example above, the `$threshold` is 2.
+
+## Child Branches with data: pull + partial images
+
+In the previous section we have shown how bottom-most compaction respects `retain_lsn` so that all data that was readable at branch creation remains readable. But branches can have data on their own, and that data can fall out of the branch’s PITR window. So, this section explains how we deal with that.
+
+We will run the same bottom-most compaction for these branches, to ensure the space amplification on the child branch is reasonable. 
+
+```
+branch_lsn: 0x20
+LSN 0x30 -> append P
+LSN 0x40 -> append Q
+LSN 0x50 -> append R
+GC horizon: 0x50
+LSN 0x60 -> append S
+```
+
+Note that bottom-most compaction happens on a per-timeline basis. When it processes this key, it only reads the history from LSN 0x30 without a base image. Therefore, on child branches, the bottom-most compaction process will make image creation decisions based on the same `count(deltas) < $threshold` criteria, and if it decides to create an image, the base image will be retrieved from the ancestor branch.
+
+```
+branch_lsn: 0x20
+LSN 0x50 -> ABPQR
+(we pull the image at LSN 0x20 from the ancestor branch to get AB, and then apply append PQ to the page; we replace the record at 0x40 with an image and drop the delta)
+GC horizon: 0x50
+LSN 0x60 -> append S
+```
+
+![](images/036-bottom-most-gc-compaction/06-btmgc-child.svg)
+
+Note that for child branches, we do not create image layers for the images when bottom-most compaction runs. Instead, we drop the 0x30/0x40/0x50 delta records and directly place the image ABPQR@0x50 into the delta layer, which serves as a sparse image layer. For child branches, if we create image layers, we will need to put all keys in the range into the image layer. This causes space bloat and slow compactions. In this proposal, the compaction process will only compact and process keys modified inside the child branch.
+
+# Result
+
+Bottom-most compaction ensures all garbage under the GC horizon gets collected right away (compared with “eventually” in the current algorithm). Meanwhile, it generates images at each of the retain_lsn to ensure branch reads are fast. As we make per-key decisions on whether to generate an image or not, the theoretical lower bound of the storage space we need to retain for a branch is lower than before.
+
+Before: min(sum(logs for each key), sum(image for each key)), for each partition — we always generate image layers on a key range
+
+After: sum(min(logs for each key, image for each key))
+
+# Compaction Trigger
+
+The bottom-most compaction can be automatically triggered. The goal of the trigger is that it should ensure a constant factor for write amplification. Say that the user write 1GB of WAL into the system, we should write 1GB x C data to S3. The legacy compaction algorithm does not have such a constant factor C. The data we write to S3 is quadratic to the logical size of the database (see [A Theoretical View of Neon Storage](https://www.notion.so/A-Theoretical-View-of-Neon-Storage-8d7ad7555b0c41b2a3597fa780911194?pvs=21)).
+
+We propose the following compaction trigger that generates a constant write amplification factor. Write amplification >= total writes to S3 / total user writes. We only analyze the write amplification caused by the bottom-most GC-compaction process, ignoring the legacy create image layers amplification.
+
+Given that we have ***X*** bytes of the delta layers above the GC horizon, ***A*** bytes of the delta layers intersecting with the GC horizon, ***B*** bytes of the delta layers below the GC horizon, and ***C*** bytes of the image layers below the GC horizon.
+
+The legacy GC + compaction loop will always keep ***A*** unchanged, reduce ***B and C*** when there are image layers covering the key range. This yields 0 write amplification (only file deletions) and extra ***B*** bytes of space.
+
+![](images/036-bottom-most-gc-compaction/09-btmgc-analysis-2.svg)
+
+The bottom-most compaction proposed here will split ***A*** into deltas above the GC horizon and below the GC horizon. Everything below the GC horizon will be image layers after the compaction (not considering branches). Therefore, this yields ***A+C*** extra write traffic each iteration, plus 0 extra space.
+
+![](images/036-bottom-most-gc-compaction/07-btmgc-analysis-1.svg)
+
+Also considering read amplification (below the GC horizon). When a read request reaches the GC horizon, the read amplification will be (A+B+C)/C=1+(A+B)/C. Reducing ***A*** and ***B*** can help reduce the read amplification below the GC horizon.
+
+The metrics-based trigger will wait until a point that space amplification is not that large and write amplification is not that large before the compaction gets triggered. The trigger is defined as **(A+B)/C ≥ 1 (or some other ratio)**.
+
+To reason about this trigger, consider the two cases:
+
+**Data Ingestion**
+
+User keeps ingesting data into the database, which indicates that WAL size roughly equals to the database logical size. The compaction gets triggered only when the newly-written WAL roughly equals to the current bottom-most image size (=X). Therefore, it’s triggered when the database size gets doubled. This is a reasonable amount of work. Write amplification is 2X/X=1 for the X amount of data written.
+
+![](images/036-bottom-most-gc-compaction/10-btmgc-analysis-3.svg)
+
+**Updates/Deletion**
+
+In this case, WAL size will be larger than the database logical size ***D***. The compaction gets triggered for every ***D*** bytes of WAL written. Therefore, for every ***D*** bytes of WAL, we rewrite the bottom-most layer, which produces an extra ***D*** bytes of write amplification. This incurs exactly 2x write amplification (by the write of D), 1.5x write amplification (if we count from the start of the process) and no space amplification. 
+
+![](images/036-bottom-most-gc-compaction/11-btmgc-analysis-4.svg)
+
+Note that here I try to reason that write amplification is a constant (i.e., the data we write to S3 is proportional to the data the user write). The main problem with the current legacy compaction algorithm is that write amplification is proportional to the database size.
+
+The next step is to optimize the write amplification above the GC horizon (i.e., change the image creation criteria, top-most compaction, or introduce tiered compaction), to ensure the write amplification of the whole system is a constant factor.
+
+20GB layers → +20GB layers → delete 20GB, need 40GB temporary space
+
+# Sub-Compactions
+
+The gc-compaction algorithm may take a long time and we need to split the job into multiple sub-compaction jobs.
+
+![](images/036-bottom-most-gc-compaction/13-job-split.svg)
+
+As in the figure, the auto-trigger schedules a compaction job covering the full keyspace below a specific LSN. In such case that we cannot finish compacting it in one run in a reasonable amount of time, the algorithm will vertically split it into multiple jobs (in this case, 5).
+
+Each gc-compaction job will create one level of delta layers and one flat level of image layers for each LSN. Those layers will be automatically split based on size, which means that if the sub-compaction job produces 1GB of deltas, it will produce 4 * 256MB delta layers. For those layers that is not fully contained within the sub-compaction job rectangles, it will be rewritten to only contain the keys outside of the key range.
+
+# Implementation
+
+The main implementation of gc-compaction is in `compaction.rs`.
+
+* `compact_with_gc`: The main loop of gc-compaction. It takes a rectangle range of the layer map and compact that specific range. It selects layers intersecting with the rectangle, downloads the layers, creates the k-merge iterator to read those layers in the key-lsn order, and decide which keys to keep or insert a reconstructed page. The process is the basic unit of a gc-compaction and is not interruptable. If the process gets preempted by L0 compaction, it has to be restarted from scratch. For layers overlaps with the rectangle but not fully inside, the main loop will also rewrite them so that the new layer (or two layers if both left and right ends are outside of the rectangle) has the same LSN range as the original one but only contain the keys outside of the compaction range.
+* `gc_compaction_split_jobs`: Splits a big gc-compaction job into sub-compactions based on heuristics in the layer map. The function looks at the layer map and splits the compaction job based on the size of the layers so that each compaction job only pulls ~4GB of layer files.
+* `generate_key_retention` and `KeyHistoryRetention`: Implements the algorithm described in the "basic idea" and "branch" chapter of this RFC. It takes a vector of history of a key (key-lsn-value) and decides which LSNs of the key to retain. If there are too many deltas between two retain_lsns, it will reconstruct the page and insert an image into the compaction result. Also, we implement `KeyHistoryRetention::verify` to ensure the generated result is not corrupted -- all retain_lsns and all LSNs above the gc-horizon should be accessible.
+* `GcCompactionQueue`: the automatic trigger implementation for gc-compaction. `GcCompactionQueue::iteration` is called at the end of the tenant compaction loop. It will then call `trigger_auto_compaction` to decide whether to trigger a gc-compaction job for this tenant. If yes, the compaction-job will be added to the compaction queue, and the queue will be slowly drained once there are no other compaction jobs running. gc-compaction has the lowest priority. If a sub-compaction job is not successful or gets preempted by L0 compaction (see limitations for reasons why a compaction job would fail), it will _not_ be retried.
+* Changes to `index_part.json`: we added a `last_completed_lsn` field to the index part for the auto-trigger to decide when to trigger a compaction.
+* Changes to the read path: when gc-compaction updates the layer map, all reads need to wait. See `gc_compaction_layer_update_lock` and comments in the code path for more information.
+
+Gc-compaction can also be scheduled over the HTTP API. Example:
+
+```
+curl 'localhost:9898/v1/tenant/:tenant_id/timeline/:timeline_id/compact?enhanced_gc_bottom_most_compaction=true&dry_run=true' -X PUT -H "Content-Type: application/json" -d '{"scheduled": true, "compact_key_range": { "start": "000000067F0000A0000002A1CF0100000000", "end": "000000067F0000A0000002A1D70100000000" } }'
+```
+
+The `dry_run` mode can be specified in the query string so that the compaction will go through all layers to estimate how much space can be saved without writing the compaction result into the layer map.
+
+The auto-trigger is controlled by tenant-level flag `gc_compaction_enabled`. If this is set to false, no gc-compaction will be automatically scheduled on this tenant (but manual trigger still works).
+
+# Next Steps
+
+There are still some limitations of gc-compaction itself that needs to be resolved and tested,
+
+- gc-compaction is currently only automatically triggered on root branches. We have not tested gc-compaction on child branches in staging.
+- gc-compaction will skip aux key regions because of the possible conflict with the assumption of aux file tombstones.
+- gc-compaction does not consider keyspaces at retain_lsns and only look at keys in the layers. This also causes us giving up some sub-compaction jobs because a key might have part of its history available due to traditional GC removing part of the history.
+- We limit gc-compaction to run over shards <= 150GB to avoid gc-compaction taking too much time blocking other compaction jobs. The sub-compaction split algorithm needs to be improved to be able to split vertically and horizontally. Also, we need to move the download layer process out of the compaction loop so that we don't block other compaction jobs for too long.
+- The compaction trigger always schedules gc-compaction from the lowest LSN to the gc-horizon. Currently we do not schedule compaction jobs that only selects layers in the middle. Allowing this could potentially reduce the number of layers read/write throughout the process.
+- gc-compaction will give up if there are too many layers to rewrite or if there are not enough disk space for the compaction.
+- gc-compaction sometimes fails with "no key produced during compaction", which means that all existing keys within the compaction range can be collected; but we don't have a way to write this information back to the layer map -- we cannot generate an empty image layer.
+- We limit the maximum size of deltas for a single key to 512MB. If above this size, gc-compaction will give up. This can be resolved by changing `generate_key_retention` to be a stream instead of requiring to collect all the key history.
+
+In the future,
+
+- Top-most compaction: ensure we always have an image coverage for the latest data (or near the latest data), so that reads will be fast at the latest LSN.
+- Tiered compaction on deltas: ensure read from any LSN is fast.
+- Per-timeline compaction → tenant-wide compaction?

docs/rfcs/images/036-bottom-most-gc-compaction/01-basic-idea.svg

@@ -0,0 +1,135 @@
+<?xml version="1.0" encoding="UTF-8" standalone="no"?>
+<!DOCTYPE svg PUBLIC "-//W3C//DTD SVG 1.1//EN" "http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd">
+<svg xmlns="http://www.w3.org/2000/svg" xmlns:xl="http://www.w3.org/1999/xlink" xmlns:dc="http://purl.org/dc/elements/1.1/" version="1.1" viewBox="82 284 863 375" width="863" height="375">
+  <defs/>
+  <g id="01-basic-idea" stroke-opacity="1" stroke-dasharray="none" stroke="none" fill="none" fill-opacity="1">
+    <title>01-basic-idea</title>
+    <rect fill="white" x="82" y="284" width="863" height="375"/>
+    <g id="01-basic-idea_Layer_1">
+      <title>Layer 1</title>
+      <g id="Graphic_2">
+        <rect x="234" y="379.5" width="203.5" height="17.5" fill="white"/>
+        <rect x="234" y="379.5" width="203.5" height="17.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_3">
+        <rect x="453.5" y="379.5" width="203.5" height="17.5" fill="white"/>
+        <rect x="453.5" y="379.5" width="203.5" height="17.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_4">
+        <rect x="672.5" y="379.5" width="203.5" height="17.5" fill="white"/>
+        <rect x="672.5" y="379.5" width="203.5" height="17.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_5">
+        <rect x="234" y="288.5" width="127" height="77.5" fill="white"/>
+        <rect x="234" y="288.5" width="127" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_6">
+        <rect x="375" y="288.5" width="127" height="77.5" fill="white"/>
+        <rect x="375" y="288.5" width="127" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_7">
+        <rect x="516" y="288.5" width="127" height="77.5" fill="white"/>
+        <rect x="516" y="288.5" width="127" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_8">
+        <rect x="657" y="288.5" width="127" height="77.5" fill="white"/>
+        <rect x="657" y="288.5" width="127" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_9">
+        <rect x="798" y="288.5" width="78" height="77.5" fill="white"/>
+        <rect x="798" y="288.5" width="78" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Line_11">
+        <line x1="185.5" y1="326.75" x2="943.7734" y2="326.75" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_12">
+        <text transform="translate(87 318.026)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="0" y="15" xml:space="preserve">GC Horizon</tspan>
+        </text>
+      </g>
+      <g id="Graphic_13">
+        <text transform="translate(106.41 372.886)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="8.39" y="15" xml:space="preserve">Images </tspan>
+          <tspan font-family="Helvetica Neue" font-size="10" fill="black" x="29132252e-19" y="28.447998" xml:space="preserve">at earlier LSN</tspan>
+        </text>
+      </g>
+      <g id="Graphic_14">
+        <text transform="translate(121.92 289.578)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="8739676e-19" y="15" xml:space="preserve">Deltas</tspan>
+        </text>
+      </g>
+      <g id="Graphic_15">
+        <path d="M 517.125 423.5 L 553.375 423.5 L 553.375 482 L 571.5 482 L 535.25 512 L 499 482 L 517.125 482 Z" fill="white"/>
+        <path d="M 517.125 423.5 L 553.375 423.5 L 553.375 482 L 571.5 482 L 535.25 512 L 499 482 L 517.125 482 Z" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_26">
+        <rect x="234" y="599.474" width="203.5" height="17.5" fill="white"/>
+        <rect x="234" y="599.474" width="203.5" height="17.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_25">
+        <rect x="453.5" y="599.474" width="203.5" height="17.5" fill="white"/>
+        <rect x="453.5" y="599.474" width="203.5" height="17.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_24">
+        <rect x="672.5" y="599.474" width="203.5" height="17.5" fill="white"/>
+        <rect x="672.5" y="599.474" width="203.5" height="17.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_23">
+        <rect x="234" y="533" width="127" height="52.974" fill="white"/>
+        <rect x="234" y="533" width="127" height="52.974" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_22">
+        <rect x="375" y="533" width="310.5" height="52.974" fill="white"/>
+        <rect x="375" y="533" width="310.5" height="52.974" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_21">
+        <rect x="702.5" y="533" width="173.5" height="52.974" fill="white"/>
+        <rect x="702.5" y="533" width="173.5" height="52.974" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Line_18">
+        <line x1="185.5" y1="607.724" x2="943.7734" y2="607.724" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_16">
+        <text transform="translate(121.92 538)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="8739676e-19" y="15" xml:space="preserve">Deltas</tspan>
+        </text>
+      </g>
+      <g id="Graphic_27">
+        <text transform="translate(114.8 592.86)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="3488765e-18" y="15" xml:space="preserve">Images </tspan>
+          <tspan font-family="Helvetica Neue" font-size="10" fill="black" x="4.01" y="28.447998" xml:space="preserve">at GC LSN</tspan>
+        </text>
+      </g>
+      <g id="Graphic_28">
+        <rect x="243.06836" y="300" width="624.3633" height="17.5" fill="#c0ffc0"/>
+        <text transform="translate(248.06836 301.068)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="13" fill="black" x="233.52364" y="12" xml:space="preserve">Deltas above GC Horizon</tspan>
+        </text>
+      </g>
+      <g id="Graphic_30">
+        <rect x="243.06836" y="335.5" width="624.3633" height="17.5" fill="#c0ffff"/>
+        <text transform="translate(248.06836 336.568)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="13" fill="black" x="233.89414" y="12" xml:space="preserve">Deltas below GC Horizon</tspan>
+        </text>
+      </g>
+      <g id="Graphic_32">
+        <rect x="243.06836" y="550.737" width="624.3633" height="17.5" fill="#c0ffc0"/>
+        <text transform="translate(248.06836 551.805)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="13" fill="black" x="233.52364" y="12" xml:space="preserve">Deltas above GC Horizon</tspan>
+        </text>
+      </g>
+      <g id="Graphic_33">
+        <rect x="304" y="630.474" width="485.5" height="28.447998" fill="#c0ffff"/>
+        <text transform="translate(309 637.016)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="13" fill="black" x="63.095" y="12" xml:space="preserve">Deltas and image below GC Horizon gets garbage-collected</tspan>
+        </text>
+      </g>
+      <g id="Graphic_34">
+        <text transform="translate(576.5 444.0325)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="12" fill="black" x="0" y="11" xml:space="preserve">WAL replay of deltas+image below GC Horizon</tspan>
+          <tspan font-family="Helvetica Neue" font-size="12" fill="black" x="0" y="25.336" xml:space="preserve">Reshuffle deltas</tspan>
+        </text>
+      </g>
+    </g>
+  </g>
+</svg>

docs/rfcs/images/036-bottom-most-gc-compaction/03-retain-lsn.svg

@@ -0,0 +1,141 @@
+<?xml version="1.0" encoding="UTF-8" standalone="no"?>
+<!DOCTYPE svg PUBLIC "-//W3C//DTD SVG 1.1//EN" "http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd">
+<svg xmlns="http://www.w3.org/2000/svg" xmlns:xl="http://www.w3.org/1999/xlink" xmlns:dc="http://purl.org/dc/elements/1.1/" version="1.1" viewBox="-104 215 863 335" width="863" height="335">
+  <defs>
+    <marker orient="auto" overflow="visible" markerUnits="strokeWidth" id="FilledArrow_Marker" stroke-linejoin="miter" stroke-miterlimit="10" viewBox="-1 -4 10 8" markerWidth="10" markerHeight="8" color="#7f8080">
+      <g>
+        <path d="M 8 0 L 0 -3 L 0 3 Z" fill="currentColor" stroke="currentColor" stroke-width="1"/>
+      </g>
+    </marker>
+  </defs>
+  <g id="03-retain-lsn" stroke-opacity="1" stroke-dasharray="none" stroke="none" fill="none" fill-opacity="1">
+    <title>03-retain-lsn</title>
+    <rect fill="white" x="-104" y="215" width="863" height="335"/>
+    <g id="03-retain-lsn_Layer_1">
+      <title>Layer 1</title>
+      <g id="Graphic_28">
+        <rect x="48" y="477" width="203.5" height="9.990005" fill="white"/>
+        <rect x="48" y="477" width="203.5" height="9.990005" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_27">
+        <rect x="267.5" y="477" width="203.5" height="9.990005" fill="white"/>
+        <rect x="267.5" y="477" width="203.5" height="9.990005" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_26">
+        <rect x="486.5" y="477" width="203.5" height="9.990005" fill="white"/>
+        <rect x="486.5" y="477" width="203.5" height="9.990005" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Line_20">
+        <line x1="-.5" y1="387.172" x2="757.7734" y2="387.172" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_19">
+        <text transform="translate(-99 378.448)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="0" y="15" xml:space="preserve">GC Horizon</tspan>
+        </text>
+      </g>
+      <g id="Graphic_31">
+        <rect x="48.25" y="410" width="203.5" height="9.990005" fill="white"/>
+        <rect x="48.25" y="410" width="203.5" height="9.990005" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_30">
+        <rect x="267.75" y="410" width="203.5" height="9.990005" fill="white"/>
+        <rect x="267.75" y="410" width="203.5" height="9.990005" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_29">
+        <rect x="486.75" y="410" width="203.5" height="9.990005" fill="white"/>
+        <rect x="486.75" y="410" width="203.5" height="9.990005" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_34">
+        <rect x="48.25" y="431.495" width="113.75" height="34" fill="white"/>
+        <rect x="48.25" y="431.495" width="113.75" height="34" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_33">
+        <rect x="172.5" y="431.495" width="203.5" height="34" fill="white"/>
+        <rect x="172.5" y="431.495" width="203.5" height="34" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_32">
+        <rect x="386.5" y="431.495" width="303.5" height="34" fill="white"/>
+        <rect x="386.5" y="431.495" width="303.5" height="34" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_37">
+        <rect x="48" y="498.495" width="203.5" height="9.990005" fill="white"/>
+        <rect x="48" y="498.495" width="203.5" height="9.990005" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_36">
+        <rect x="267.5" y="498.495" width="203.5" height="9.990005" fill="white"/>
+        <rect x="267.5" y="498.495" width="203.5" height="9.990005" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_35">
+        <rect x="486.5" y="498.495" width="203.5" height="9.990005" fill="white"/>
+        <rect x="486.5" y="498.495" width="203.5" height="9.990005" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Line_38">
+        <line x1="-10.48" y1="535.5395" x2="39.318294" y2="508.24794" marker-end="url(#FilledArrow_Marker)" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_39">
+        <text transform="translate(-96.984 526.3155)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="40500936e-20" y="15" xml:space="preserve">retain_lsn 1</tspan>
+        </text>
+      </g>
+      <g id="Line_41">
+        <line x1="-10.48" y1="507.0915" x2="38.90236" y2="485.8992" marker-end="url(#FilledArrow_Marker)" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_40">
+        <text transform="translate(-96.984 497.8675)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="40500936e-20" y="15" xml:space="preserve">retain_lsn 2</tspan>
+        </text>
+      </g>
+      <g id="Line_43">
+        <line x1="-10.48" y1="478.6435" x2="39.44267" y2="453.01616" marker-end="url(#FilledArrow_Marker)" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_42">
+        <text transform="translate(-96.984 469.4195)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="40500936e-20" y="15" xml:space="preserve">retain_lsn 3</tspan>
+        </text>
+      </g>
+      <g id="Line_45">
+        <line x1="-10.48" y1="448.495" x2="39.65061" y2="419.90015" marker-end="url(#FilledArrow_Marker)" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_44">
+        <text transform="translate(-96.984 439.271)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="40500936e-20" y="15" xml:space="preserve">retain_lsn 4</tspan>
+        </text>
+      </g>
+      <g id="Graphic_46">
+        <rect x="335.46477" y="215.5" width="353.4299" height="125.495" fill="white"/>
+        <rect x="335.46477" y="215.5" width="353.4299" height="125.495" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_48">
+        <text transform="translate(549.3766 317.547)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="6536993e-19" y="15" xml:space="preserve">Dependent Branch</tspan>
+        </text>
+      </g>
+      <g id="Graphic_50">
+        <text transform="translate(340.43824 317.547)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="40500936e-20" y="15" xml:space="preserve">retain_lsn 3</tspan>
+        </text>
+      </g>
+      <g id="Line_57">
+        <line x1="323.90685" y1="248.8045" x2="714.9232" y2="248.8045" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_56">
+        <text transform="translate(165.91346 240.0805)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="35811354e-19" y="15" xml:space="preserve">Branch GC Horizon</tspan>
+        </text>
+      </g>
+      <g id="Graphic_58">
+        <rect x="493.9232" y="301.6405" width="107.45294" height="9.990005" fill="white"/>
+        <rect x="493.9232" y="301.6405" width="107.45294" height="9.990005" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_59">
+        <text transform="translate(358.9232 277.276)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="0" y="15" xml:space="preserve">Partial Image Coverage</tspan>
+        </text>
+      </g>
+      <g id="Graphic_60">
+        <rect x="354.1732" y="301.6405" width="107.45294" height="9.990005" fill="white"/>
+        <rect x="354.1732" y="301.6405" width="107.45294" height="9.990005" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+    </g>
+  </g>
+</svg>

docs/rfcs/images/036-bottom-most-gc-compaction/05-btmgc-parent.svg

@@ -0,0 +1,187 @@
+<?xml version="1.0" encoding="UTF-8" standalone="no"?>
+<!DOCTYPE svg PUBLIC "-//W3C//DTD SVG 1.1//EN" "http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd">
+<svg xmlns="http://www.w3.org/2000/svg" xmlns:xl="http://www.w3.org/1999/xlink" xmlns:dc="http://purl.org/dc/elements/1.1/" version="1.1" viewBox="-235 426 864 366" width="864" height="366">
+  <defs/>
+  <g id="05-btmgc-parent" stroke-opacity="1" stroke-dasharray="none" stroke="none" fill="none" fill-opacity="1">
+    <title>05-btmgc-parent</title>
+    <rect fill="white" x="-235" y="426" width="864" height="366"/>
+    <g id="05-btmgc-parent_Layer_1">
+      <title>Layer 1</title>
+      <g id="Graphic_23">
+        <rect x="-83" y="510.15" width="203.5" height="26.391998" fill="white"/>
+        <rect x="-83" y="510.15" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-78 516.178)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="12" fill="black" x="51.714" y="11" xml:space="preserve">Append C@0x30</tspan>
+        </text>
+      </g>
+      <g id="Graphic_22">
+        <rect x="136.5" y="510.15" width="203.5" height="26.391998" fill="white"/>
+        <rect x="136.5" y="510.15" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_21">
+        <rect x="355.5" y="510.15" width="203.5" height="26.391998" fill="white"/>
+        <rect x="355.5" y="510.15" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Line_20">
+        <line x1="-100.448" y1="459.224" x2="626.77344" y2="459.224" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_19">
+        <text transform="translate(-230 450.5)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="0" y="15" xml:space="preserve">GC Horizon</tspan>
+        </text>
+      </g>
+      <g id="Graphic_18">
+        <rect x="-82.75" y="426.748" width="203.5" height="26.391998" fill="white"/>
+        <rect x="-82.75" y="426.748" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-77.75 432.776)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="12" fill="black" x="52.602" y="11" xml:space="preserve">Append F@0x60</tspan>
+        </text>
+      </g>
+      <g id="Graphic_17">
+        <rect x="136.75" y="426.748" width="203.5" height="26.391998" fill="white"/>
+        <rect x="136.75" y="426.748" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_16">
+        <rect x="355.75" y="426.748" width="203.5" height="26.391998" fill="white"/>
+        <rect x="355.75" y="426.748" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_15">
+        <rect x="-82.75" y="464.645" width="113.75" height="34" fill="white"/>
+        <rect x="-82.75" y="464.645" width="113.75" height="34" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-77.75 467.309)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="12" fill="black" x="7.505" y="11" xml:space="preserve">Append E@0x50</tspan>
+          <tspan font-family="Helvetica Neue" font-size="12" fill="black" x="6.947" y="25.336" xml:space="preserve">Append D@0x40</tspan>
+        </text>
+      </g>
+      <g id="Graphic_14">
+        <rect x="41.5" y="464.645" width="203.5" height="34" fill="white"/>
+        <rect x="41.5" y="464.645" width="203.5" height="34" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_13">
+        <rect x="255.5" y="464.645" width="303.5" height="34" fill="white"/>
+        <rect x="255.5" y="464.645" width="303.5" height="34" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_12">
+        <rect x="-83" y="548.047" width="203.5" height="26.391998" fill="white"/>
+        <rect x="-83" y="548.047" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-78 554.075)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="12" fill="black" x="26.796" y="11" xml:space="preserve">A@0x10, Append B@0x20</tspan>
+        </text>
+      </g>
+      <g id="Graphic_11">
+        <rect x="136.5" y="548.047" width="203.5" height="26.391998" fill="white"/>
+        <rect x="136.5" y="548.047" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_10">
+        <rect x="355.5" y="548.047" width="203.5" height="26.391998" fill="white"/>
+        <rect x="355.5" y="548.047" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Line_24">
+        <line x1="-104" y1="542" x2="610.5" y2="542" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-dasharray="4.0,4.0" stroke-width="1"/>
+      </g>
+      <g id="Graphic_25">
+        <text transform="translate(-139.604 534.5)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="14" fill="black" x="0" y="13" xml:space="preserve">0x20</tspan>
+        </text>
+      </g>
+      <g id="Graphic_28">
+        <text transform="translate(-139.604 452.556)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="14" fill="black" x="0" y="13" xml:space="preserve">0x50</tspan>
+        </text>
+      </g>
+      <g id="Line_30">
+        <line x1="-100.448" y1="481.145" x2="614.052" y2="481.145" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-dasharray="4.0,4.0" stroke-width="1"/>
+      </g>
+      <g id="Graphic_29">
+        <text transform="translate(-139.604 473.449)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="14" fill="black" x="0" y="13" xml:space="preserve">0x40</tspan>
+        </text>
+      </g>
+      <g id="Line_48">
+        <line x1="-99.448" y1="701.513" x2="627.77344" y2="701.513" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_47">
+        <text transform="translate(-229 692.789)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="0" y="15" xml:space="preserve">GC Horizon</tspan>
+        </text>
+      </g>
+      <g id="Graphic_46">
+        <rect x="-81.75" y="670.496" width="113.75" height="26.391998" fill="white"/>
+        <rect x="-81.75" y="670.496" width="113.75" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-76.75 676.524)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="12" fill="black" x="7.727" y="11" xml:space="preserve">Append F@0x60</tspan>
+        </text>
+      </g>
+      <g id="Graphic_43">
+        <rect x="-81.75" y="708.393" width="113.75" height="34" fill="white"/>
+        <rect x="-81.75" y="708.393" width="113.75" height="34" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-76.75 718.225)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="12" fill="black" x="7.505" y="11" xml:space="preserve">Append E@0x50</tspan>
+        </text>
+      </g>
+      <g id="Line_37">
+        <line x1="-101" y1="777.2665" x2="613.5" y2="777.2665" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-dasharray="4.0,4.0" stroke-width="1"/>
+      </g>
+      <g id="Graphic_36">
+        <text transform="translate(-138.604 769.7665)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="14" fill="black" x="0" y="13" xml:space="preserve">0x20</tspan>
+        </text>
+      </g>
+      <g id="Graphic_33">
+        <text transform="translate(-138.604 694.845)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="14" fill="black" x="0" y="13" xml:space="preserve">0x50</tspan>
+        </text>
+      </g>
+      <g id="Line_32">
+        <line x1="-99.448" y1="755.089" x2="615.052" y2="755.089" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-dasharray="4.0,4.0" stroke-width="1"/>
+      </g>
+      <g id="Graphic_31">
+        <text transform="translate(-138.604 747.393)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="14" fill="black" x="0" y="13" xml:space="preserve">0x40</tspan>
+        </text>
+      </g>
+      <g id="Graphic_40">
+        <rect x="-82" y="770.909" width="203.5" height="14.107002" fill="white"/>
+        <rect x="-82" y="770.909" width="203.5" height="14.107002" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-77 770.7945)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="12" fill="black" x="70.836" y="11" xml:space="preserve">AB@0x20</tspan>
+        </text>
+      </g>
+      <g id="Graphic_39">
+        <rect x="137.5" y="770.909" width="203.5" height="14.107002" fill="white"/>
+        <rect x="137.5" y="770.909" width="203.5" height="14.107002" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_38">
+        <rect x="356.5" y="770.909" width="203.5" height="14.107002" fill="white"/>
+        <rect x="356.5" y="770.909" width="203.5" height="14.107002" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_54">
+        <rect x="-81.75" y="748.5355" width="203.5" height="14.107002" fill="white"/>
+        <rect x="-81.75" y="748.5355" width="203.5" height="14.107002" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-76.75 748.421)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="12" fill="black" x="62.28" y="11" xml:space="preserve">ABCD@0x40</tspan>
+        </text>
+      </g>
+      <g id="Graphic_53">
+        <rect x="137.75" y="748.5355" width="203.5" height="14.107002" fill="white"/>
+        <rect x="137.75" y="748.5355" width="203.5" height="14.107002" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_52">
+        <rect x="356.75" y="748.5355" width="203.5" height="14.107002" fill="white"/>
+        <rect x="356.75" y="748.5355" width="203.5" height="14.107002" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_57">
+        <path d="M 211.32422 585 L 265.17578 585 L 265.17578 611.332 L 287.84375 611.332 L 238.25 633.117 L 188.65625 611.332 L 211.32422 611.332 Z" fill="white"/>
+        <path d="M 211.32422 585 L 265.17578 585 L 265.17578 611.332 L 287.84375 611.332 L 238.25 633.117 L 188.65625 611.332 L 211.32422 611.332 Z" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_60">
+        <rect x="359" y="692.858" width="203.5" height="14.107002" fill="white"/>
+        <rect x="359" y="692.858" width="203.5" height="14.107002" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_59">
+        <rect x="41.5" y="693.858" width="303" height="14.107002" fill="white"/>
+        <rect x="41.5" y="693.858" width="303" height="14.107002" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+    </g>
+  </g>
+</svg>

docs/rfcs/images/036-bottom-most-gc-compaction/06-btmgc-child.svg

@@ -0,0 +1,184 @@
+<?xml version="1.0" encoding="UTF-8" standalone="no"?>
+<!DOCTYPE svg PUBLIC "-//W3C//DTD SVG 1.1//EN" "http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd">
+<svg xmlns="http://www.w3.org/2000/svg" xmlns:xl="http://www.w3.org/1999/xlink" xmlns:dc="http://purl.org/dc/elements/1.1/" version="1.1" viewBox="-413 471 931 354" width="931" height="354">
+  <defs/>
+  <g id="06-btmgc-child" stroke-opacity="1" stroke-dasharray="none" stroke="none" fill="none" fill-opacity="1">
+    <title>06-btmgc-child</title>
+    <rect fill="white" x="-413" y="471" width="931" height="354"/>
+    <g id="06-btmgc-child_Layer_1">
+      <title>Layer 1</title>
+      <g id="Graphic_47">
+        <rect x="-412" y="594.402" width="928" height="28.447998" fill="white"/>
+        <rect x="-412" y="594.402" width="928" height="28.447998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_46">
+        <rect x="-205" y="555.552" width="203.5" height="26.391998" fill="white"/>
+        <rect x="-205" y="555.552" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-200 561.58)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="12" fill="black" x="52.158" y="11" xml:space="preserve">Append P@0x30</tspan>
+        </text>
+      </g>
+      <g id="Graphic_45">
+        <rect x="14.5" y="555.552" width="203.5" height="26.391998" fill="white"/>
+        <rect x="14.5" y="555.552" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_44">
+        <rect x="233.5" y="555.552" width="203.5" height="26.391998" fill="white"/>
+        <rect x="233.5" y="555.552" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Line_43">
+        <line x1="-222.448" y1="504.724" x2="504.77344" y2="504.724" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_42">
+        <text transform="translate(-352 496)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="0" y="15" xml:space="preserve">GC Horizon</tspan>
+        </text>
+      </g>
+      <g id="Graphic_41">
+        <rect x="-204.75" y="472.15" width="203.5" height="26.391998" fill="white"/>
+        <rect x="-204.75" y="472.15" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-199.75 478.178)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="12" fill="black" x="52.158" y="11" xml:space="preserve">Append S@0x60</tspan>
+        </text>
+      </g>
+      <g id="Graphic_40">
+        <rect x="14.75" y="472.15" width="203.5" height="26.391998" fill="white"/>
+        <rect x="14.75" y="472.15" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_39">
+        <rect x="233.75" y="472.15" width="203.5" height="26.391998" fill="white"/>
+        <rect x="233.75" y="472.15" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_38">
+        <rect x="-204.75" y="510.047" width="113.75" height="34" fill="white"/>
+        <rect x="-204.75" y="510.047" width="113.75" height="34" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-199.75 512.711)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="12" fill="black" x="7.061" y="11" xml:space="preserve">Append R@0x50</tspan>
+          <tspan font-family="Helvetica Neue" font-size="12" fill="black" x="6.611" y="25.336" xml:space="preserve">Append Q@0x40</tspan>
+        </text>
+      </g>
+      <g id="Graphic_37">
+        <rect x="-80.5" y="510.047" width="203.5" height="34" fill="white"/>
+        <rect x="-80.5" y="510.047" width="203.5" height="34" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_36">
+        <rect x="133.5" y="510.047" width="303.5" height="34" fill="white"/>
+        <rect x="133.5" y="510.047" width="303.5" height="34" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_33">
+        <text transform="translate(-261.604 498.056)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="14" fill="black" x="0" y="13" xml:space="preserve">0x50</tspan>
+        </text>
+      </g>
+      <g id="Line_30">
+        <line x1="-224" y1="607.9115" x2="490.5" y2="607.9115" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-dasharray="4.0,4.0" stroke-width="1"/>
+      </g>
+      <g id="Graphic_29">
+        <text transform="translate(-261.604 600.4115)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="14" fill="black" x="0" y="13" xml:space="preserve">0x20</tspan>
+        </text>
+      </g>
+      <g id="Graphic_28">
+        <rect x="-205" y="601.554" width="203.5" height="14.107002" fill="white"/>
+        <rect x="-205" y="601.554" width="203.5" height="14.107002" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-200 601.4395)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="12" fill="black" x="70.836" y="11" xml:space="preserve">AB@0x20</tspan>
+        </text>
+      </g>
+      <g id="Graphic_27">
+        <rect x="14.5" y="601.554" width="203.5" height="14.107002" fill="white"/>
+        <rect x="14.5" y="601.554" width="203.5" height="14.107002" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_26">
+        <rect x="233.5" y="601.554" width="203.5" height="14.107002" fill="white"/>
+        <rect x="233.5" y="601.554" width="203.5" height="14.107002" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_25">
+        <text transform="translate(-407 599.1875)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="0" y="15" xml:space="preserve">Ancestor Branch</tspan>
+        </text>
+      </g>
+      <g id="Graphic_24">
+        <rect x="-411" y="795.46" width="928" height="28.447998" fill="white"/>
+        <rect x="-411" y="795.46" width="928" height="28.447998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Line_20">
+        <line x1="-221.448" y1="755.528" x2="505.77344" y2="755.528" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_19">
+        <text transform="translate(-351 746.804)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="0" y="15" xml:space="preserve">GC Horizon</tspan>
+        </text>
+      </g>
+      <g id="Graphic_18">
+        <rect x="-203.75" y="723.579" width="203.25" height="26.391998" fill="white"/>
+        <rect x="-203.75" y="723.579" width="203.25" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-198.75 729.607)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="12" fill="black" x="52.033" y="11" xml:space="preserve">Append S@0x60</tspan>
+        </text>
+      </g>
+      <g id="Graphic_10">
+        <text transform="translate(-260.604 748.86)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="14" fill="black" x="0" y="13" xml:space="preserve">0x50</tspan>
+        </text>
+      </g>
+      <g id="Line_7">
+        <line x1="-223" y1="808.9695" x2="491.5" y2="808.9695" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-dasharray="4.0,4.0" stroke-width="1"/>
+      </g>
+      <g id="Graphic_6">
+        <text transform="translate(-260.604 801.4695)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="14" fill="black" x="0" y="13" xml:space="preserve">0x20</tspan>
+        </text>
+      </g>
+      <g id="Graphic_5">
+        <rect x="-204" y="802.612" width="203.5" height="14.107002" fill="white"/>
+        <rect x="-204" y="802.612" width="203.5" height="14.107002" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-199 802.4975)" fill="#b1001c">
+          <tspan font-family="Helvetica Neue" font-size="12" fill="#b1001c" x="70.836" y="11" xml:space="preserve">AB</tspan>
+          <tspan font-family="Helvetica Neue" font-size="12" fill="black" y="11" xml:space="preserve">@0x20</tspan>
+        </text>
+      </g>
+      <g id="Graphic_4">
+        <rect x="15.5" y="802.612" width="203.5" height="14.107002" fill="white"/>
+        <rect x="15.5" y="802.612" width="203.5" height="14.107002" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_3">
+        <rect x="234.5" y="802.612" width="203.5" height="14.107002" fill="white"/>
+        <rect x="234.5" y="802.612" width="203.5" height="14.107002" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_2">
+        <text transform="translate(-406 800.2455)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="0" y="15" xml:space="preserve">Ancestor Branch</tspan>
+        </text>
+      </g>
+      <g id="Graphic_48">
+        <path d="M 89.32422 639.081 L 143.17578 639.081 L 143.17578 665.413 L 165.84375 665.413 L 116.25 687.198 L 66.65625 665.413 L 89.32422 665.413 Z" fill="white"/>
+        <path d="M 89.32422 639.081 L 143.17578 639.081 L 143.17578 665.413 L 165.84375 665.413 L 116.25 687.198 L 66.65625 665.413 L 89.32422 665.413 Z" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_49">
+        <rect x="-204" y="762.428" width="203.5" height="26.391998" fill="white"/>
+        <rect x="-204" y="762.428" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-199 768.456)" fill="#b1001c">
+          <tspan font-family="Helvetica Neue" font-size="12" fill="#b1001c" x="58.278" y="11" xml:space="preserve">AB</tspan>
+          <tspan font-family="Helvetica Neue" font-size="12" fill="black" y="11" xml:space="preserve">PQR@0x50</tspan>
+        </text>
+      </g>
+      <g id="Graphic_59">
+        <rect x="14.5" y="723.579" width="203.5" height="26.391998" fill="white"/>
+        <rect x="14.5" y="723.579" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_58">
+        <rect x="233.5" y="723.579" width="203.5" height="26.391998" fill="white"/>
+        <rect x="233.5" y="723.579" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_63">
+        <rect x="9" y="762.085" width="203.5" height="26.391998" fill="white"/>
+        <rect x="9" y="762.085" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_62">
+        <rect x="225" y="762.085" width="213" height="26.391998" fill="white"/>
+        <rect x="225" y="762.085" width="213" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+    </g>
+  </g>
+</svg>

docs/rfcs/images/036-bottom-most-gc-compaction/07-btmgc-analysis-1.svg

@@ -0,0 +1,180 @@
+<?xml version="1.0" encoding="UTF-8" standalone="no"?>
+<!DOCTYPE svg PUBLIC "-//W3C//DTD SVG 1.1//EN" "http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd">
+<svg xmlns="http://www.w3.org/2000/svg" xmlns:xl="http://www.w3.org/1999/xlink" xmlns:dc="http://purl.org/dc/elements/1.1/" version="1.1" viewBox="-556 476 923 411" width="923" height="411">
+  <defs/>
+  <g id="07-btmgc-analysis-1" stroke-opacity="1" stroke-dasharray="none" stroke="none" fill="none" fill-opacity="1">
+    <title>07-btmgc-analysis-1</title>
+    <rect fill="white" x="-556" y="476" width="923" height="411"/>
+    <g id="07-btmgc-analysis-1_Layer_1">
+      <title>Layer 1</title>
+      <g id="Graphic_85">
+        <rect x="-404" y="609.062" width="203.5" height="17.5" fill="white"/>
+        <rect x="-404" y="609.062" width="203.5" height="17.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_84">
+        <rect x="-184.5" y="609.062" width="203.5" height="17.5" fill="white"/>
+        <rect x="-184.5" y="609.062" width="203.5" height="17.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_83">
+        <rect x="34.5" y="609.062" width="203.5" height="17.5" fill="white"/>
+        <rect x="34.5" y="609.062" width="203.5" height="17.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_82">
+        <rect x="-404" y="479.922" width="127" height="77.5" fill="white"/>
+        <rect x="-404" y="479.922" width="127" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_81">
+        <rect x="-263" y="479.922" width="127" height="77.5" fill="white"/>
+        <rect x="-263" y="479.922" width="127" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_80">
+        <rect x="-122" y="479.922" width="127" height="77.5" fill="white"/>
+        <rect x="-122" y="479.922" width="127" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_79">
+        <rect x="19" y="479.922" width="127" height="77.5" fill="white"/>
+        <rect x="19" y="479.922" width="127" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_78">
+        <rect x="160" y="479.922" width="78" height="77.5" fill="white"/>
+        <rect x="160" y="479.922" width="78" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Line_77">
+        <line x1="-452.5" y1="518.172" x2="251" y2="518.172" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_76">
+        <text transform="translate(-551 509.448)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="0" y="15" xml:space="preserve">GC Horizon</tspan>
+        </text>
+      </g>
+      <g id="Graphic_75">
+        <text transform="translate(-531.59 602.448)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="8.39" y="15" xml:space="preserve">Images </tspan>
+          <tspan font-family="Helvetica Neue" font-size="10" fill="black" x="29132252e-19" y="28.447998" xml:space="preserve">at earlier LSN</tspan>
+        </text>
+      </g>
+      <g id="Graphic_74">
+        <text transform="translate(-516.08 481)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="8739676e-19" y="15" xml:space="preserve">Deltas</tspan>
+        </text>
+      </g>
+      <g id="Graphic_73">
+        <path d="M -120.675 651.5 L -84.425 651.5 L -84.425 710 L -66.3 710 L -102.55 740 L -138.8 710 L -120.675 710 Z" fill="white"/>
+        <path d="M -120.675 651.5 L -84.425 651.5 L -84.425 710 L -66.3 710 L -102.55 740 L -138.8 710 L -120.675 710 Z" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_72">
+        <rect x="-403.8" y="827.474" width="203.5" height="17.5" fill="white"/>
+        <rect x="-403.8" y="827.474" width="203.5" height="17.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_71">
+        <rect x="-184.3" y="827.474" width="203.5" height="17.5" fill="white"/>
+        <rect x="-184.3" y="827.474" width="203.5" height="17.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_70">
+        <rect x="34.7" y="827.474" width="203.5" height="17.5" fill="white"/>
+        <rect x="34.7" y="827.474" width="203.5" height="17.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_69">
+        <rect x="-403.8" y="761" width="127" height="52.974" fill="white"/>
+        <rect x="-403.8" y="761" width="127" height="52.974" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_68">
+        <rect x="-262.8" y="761" width="310.5" height="52.974" fill="white"/>
+        <rect x="-262.8" y="761" width="310.5" height="52.974" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_67">
+        <rect x="64.7" y="761" width="173.5" height="52.974" fill="white"/>
+        <rect x="64.7" y="761" width="173.5" height="52.974" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Line_66">
+        <line x1="-452.3" y1="835.724" x2="251.2" y2="835.724" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_65">
+        <text transform="translate(-515.88 766)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="8739676e-19" y="15" xml:space="preserve">Deltas</tspan>
+        </text>
+      </g>
+      <g id="Graphic_64">
+        <text transform="translate(-523 820.86)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="3488765e-18" y="15" xml:space="preserve">Images </tspan>
+          <tspan font-family="Helvetica Neue" font-size="10" fill="black" x="4.01" y="28.447998" xml:space="preserve">at GC LSN</tspan>
+        </text>
+      </g>
+      <g id="Graphic_63">
+        <rect x="-394.93164" y="491.422" width="624.3633" height="17.5" fill="#c0ffc0"/>
+        <text transform="translate(-389.93164 492.49)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="13" fill="black" x="233.52364" y="12" xml:space="preserve">Deltas above GC Horizon</tspan>
+        </text>
+      </g>
+      <g id="Graphic_62">
+        <rect x="-394.93164" y="526.922" width="624.3633" height="17.5" fill="#c0ffff"/>
+        <text transform="translate(-389.93164 527.99)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="13" fill="black" x="233.89414" y="12" xml:space="preserve">Deltas below GC Horizon</tspan>
+        </text>
+      </g>
+      <g id="Graphic_61">
+        <rect x="-394.73164" y="778.737" width="624.3633" height="17.5" fill="#c0ffc0"/>
+        <text transform="translate(-389.73164 779.805)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="13" fill="black" x="233.52364" y="12" xml:space="preserve">Deltas above GC Horizon</tspan>
+        </text>
+      </g>
+      <g id="Graphic_60">
+        <rect x="-333.8" y="858.474" width="485.5" height="28.447998" fill="#c0ffff"/>
+        <text transform="translate(-328.8 865.016)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="13" fill="black" x="63.095" y="12" xml:space="preserve">Deltas and image below GC Horizon gets garbage-collected</tspan>
+        </text>
+      </g>
+      <g id="Graphic_86">
+        <text transform="translate(263 499.724)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="32" fill="black" x="0" y="30" xml:space="preserve">size=A</tspan>
+        </text>
+      </g>
+      <g id="Line_87">
+        <line x1="260.87012" y1="479.068" x2="360.71387" y2="479.068" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Line_88">
+        <line x1="260.87012" y1="561" x2="360.71387" y2="561" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_89">
+        <rect x="-403.8" y="569" width="161.8" height="28.447998" fill="white"/>
+        <rect x="-403.8" y="569" width="161.8" height="28.447998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_90">
+        <rect x="-229.5" y="569.018" width="277.2" height="28.447998" fill="white"/>
+        <rect x="-229.5" y="569.018" width="277.2" height="28.447998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_91">
+        <rect x="64.7" y="569.018" width="173.5" height="28.447998" fill="white"/>
+        <rect x="64.7" y="569.018" width="173.5" height="28.447998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Line_92">
+        <line x1="262" y1="602" x2="361.84375" y2="602" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Line_93">
+        <line x1="263" y1="625.562" x2="362.84375" y2="625.562" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_94">
+        <text transform="translate(264.53787 562.276)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="32" fill="black" x="14210855e-21" y="30" xml:space="preserve">size=B</tspan>
+        </text>
+      </g>
+      <g id="Graphic_95">
+        <text transform="translate(285.12 599.5)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="24" fill="black" x="0" y="23" xml:space="preserve">size=C</tspan>
+        </text>
+      </g>
+      <g id="Graphic_98">
+        <text transform="translate(264.53787 773.772)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="26" fill="black" x="8881784e-19" y="25" xml:space="preserve">A</tspan>
+          <tspan font-family="Lucida Grande" font-size="26" fill="black" y="25" xml:space="preserve">↓</tspan>
+        </text>
+      </g>
+      <g id="Graphic_97">
+        <text transform="translate(265.87013 815.5)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="26" fill="black" x="6536993e-19" y="25" xml:space="preserve">B</tspan>
+          <tspan font-family="Lucida Grande" font-size="26" fill="black" y="25" xml:space="preserve">↓</tspan>
+        </text>
+      </g>
+    </g>
+  </g>
+</svg>

docs/rfcs/images/036-bottom-most-gc-compaction/08-optimization.svg

@@ -0,0 +1,158 @@
+<?xml version="1.0" encoding="UTF-8" standalone="no"?>
+<!DOCTYPE svg PUBLIC "-//W3C//DTD SVG 1.1//EN" "http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd">
+<svg xmlns="http://www.w3.org/2000/svg" xmlns:xl="http://www.w3.org/1999/xlink" xmlns:dc="http://purl.org/dc/elements/1.1/" version="1.1" viewBox="-235 406 586 424" width="586" height="424">
+  <defs/>
+  <g id="08-optimization" stroke-opacity="1" stroke-dasharray="none" stroke="none" fill="none" fill-opacity="1">
+    <title>08-optimization</title>
+    <rect fill="white" x="-235" y="406" width="586" height="424"/>
+    <g id="08-optimization_Layer_1">
+      <title>Layer 1</title>
+      <g id="Graphic_22">
+        <rect x="-100.448" y="509.902" width="203.5" height="26.391998" fill="white"/>
+        <rect x="-100.448" y="509.902" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_21">
+        <rect x="118.552" y="509.902" width="203.5" height="26.391998" fill="white"/>
+        <rect x="118.552" y="509.902" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Line_20">
+        <line x1="-101.79572" y1="420.322" x2="349.5" y2="420.322" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_19">
+        <text transform="translate(-230 411.598)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="0" y="15" xml:space="preserve">GC Horizon</tspan>
+        </text>
+      </g>
+      <g id="Graphic_17">
+        <rect x="-100.198" y="426.5" width="203.5" height="26.391998" fill="white"/>
+        <rect x="-100.198" y="426.5" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_16">
+        <rect x="118.802" y="426.5" width="203.5" height="26.391998" fill="white"/>
+        <rect x="118.802" y="426.5" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_14">
+        <rect x="-100.198" y="464.397" width="108.25" height="34" fill="white"/>
+        <rect x="-100.198" y="464.397" width="108.25" height="34" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_13">
+        <rect x="18.552" y="464.397" width="303.5" height="34" fill="white"/>
+        <rect x="18.552" y="464.397" width="303.5" height="34" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_11">
+        <rect x="-100.448" y="547.799" width="203.5" height="26.391998" fill="white"/>
+        <rect x="-100.448" y="547.799" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_10">
+        <rect x="118.552" y="547.799" width="203.5" height="26.391998" fill="white"/>
+        <rect x="118.552" y="547.799" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Line_24">
+        <line x1="-104" y1="542" x2="339.4011" y2="542" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-dasharray="4.0,4.0" stroke-width="1"/>
+      </g>
+      <g id="Graphic_25">
+        <text transform="translate(-139.604 534.5)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="14" fill="black" x="0" y="13" xml:space="preserve">0x20</tspan>
+        </text>
+      </g>
+      <g id="Line_27">
+        <line x1="-101.79572" y1="459.098" x2="341.6054" y2="459.098" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-dasharray="4.0,4.0" stroke-width="1"/>
+      </g>
+      <g id="Graphic_26">
+        <text transform="translate(-139.604 451.402)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="14" fill="black" x="0" y="13" xml:space="preserve">0x50</tspan>
+        </text>
+      </g>
+      <g id="Graphic_28">
+        <text transform="translate(-139.604 413.654)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="14" fill="black" x="0" y="13" xml:space="preserve">0x60</tspan>
+        </text>
+      </g>
+      <g id="Line_30">
+        <line x1="-101.79572" y1="481.145" x2="341.6054" y2="481.145" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-dasharray="4.0,4.0" stroke-width="1"/>
+      </g>
+      <g id="Graphic_29">
+        <text transform="translate(-139.604 473.449)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="14" fill="black" x="0" y="13" xml:space="preserve">0x40</tspan>
+        </text>
+      </g>
+      <g id="Graphic_77">
+        <rect x="-100.448" y="765.19595" width="203.5" height="26.391998" fill="white"/>
+        <rect x="-100.448" y="765.19595" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_76">
+        <rect x="118.552" y="765.19595" width="203.5" height="26.391998" fill="white"/>
+        <rect x="118.552" y="765.19595" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Line_75">
+        <line x1="-101.79572" y1="637.317" x2="349.5" y2="637.317" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_74">
+        <text transform="translate(-230 628.593)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="0" y="15" xml:space="preserve">GC Horizon</tspan>
+        </text>
+      </g>
+      <g id="Graphic_73">
+        <rect x="-100.198" y="681.794" width="203.5" height="26.391998" fill="white"/>
+        <rect x="-100.198" y="681.794" width="203.5" height="26.391998" stroke="#b1001c" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_72">
+        <rect x="118.802" y="681.794" width="203.5" height="26.391998" fill="white"/>
+        <rect x="118.802" y="681.794" width="203.5" height="26.391998" stroke="#b1001c" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_71">
+        <rect x="-100.198" y="719.69096" width="108.25" height="34" fill="white"/>
+        <rect x="-100.198" y="719.69096" width="108.25" height="34" stroke="#b1001c" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_70">
+        <rect x="18.552" y="719.69096" width="303.5" height="34" fill="white"/>
+        <rect x="18.552" y="719.69096" width="303.5" height="34" stroke="#b1001c" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_69">
+        <rect x="-100.448" y="803.09295" width="203.5" height="26.391998" fill="white"/>
+        <rect x="-100.448" y="803.09295" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_68">
+        <rect x="118.552" y="803.09295" width="203.5" height="26.391998" fill="white"/>
+        <rect x="118.552" y="803.09295" width="203.5" height="26.391998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Line_67">
+        <line x1="-104" y1="797.294" x2="339.4011" y2="797.294" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-dasharray="4.0,4.0" stroke-width="1"/>
+      </g>
+      <g id="Graphic_66">
+        <text transform="translate(-139.604 789.794)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="14" fill="black" x="0" y="13" xml:space="preserve">0x20</tspan>
+        </text>
+      </g>
+      <g id="Graphic_63">
+        <text transform="translate(-139.604 630.649)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="14" fill="black" x="0" y="13" xml:space="preserve">0x70</tspan>
+        </text>
+      </g>
+      <g id="Line_62">
+        <line x1="-101.79572" y1="736.439" x2="341.6054" y2="736.439" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-dasharray="4.0,4.0" stroke-width="1"/>
+      </g>
+      <g id="Graphic_61">
+        <text transform="translate(-139.604 728.743)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="14" fill="black" x="0" y="13" xml:space="preserve">0x40</tspan>
+        </text>
+      </g>
+      <g id="Graphic_79">
+        <rect x="-100.198" y="644.393" width="168.198" height="26.391998" fill="white"/>
+        <rect x="-100.198" y="644.393" width="168.198" height="26.391998" stroke="#b1001c" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_78">
+        <rect x="80" y="644.393" width="242.302" height="26.391998" fill="white"/>
+        <rect x="80" y="644.393" width="242.302" height="26.391998" stroke="#b1001c" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Line_81">
+        <line x1="-101.79572" y1="714.139" x2="341.6054" y2="714.139" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-dasharray="1.0,4.0" stroke-width="1"/>
+      </g>
+      <g id="Graphic_80">
+        <text transform="translate(-139.604 706.443)" fill="#a5a5a5">
+          <tspan font-family="Helvetica Neue" font-size="14" fill="#a5a5a5" x="0" y="13" xml:space="preserve">0x50</tspan>
+        </text>
+      </g>
+    </g>
+  </g>
+</svg>

docs/rfcs/images/036-bottom-most-gc-compaction/09-btmgc-analysis-2.svg

@@ -0,0 +1,184 @@
+<?xml version="1.0" encoding="UTF-8" standalone="no"?>
+<!DOCTYPE svg PUBLIC "-//W3C//DTD SVG 1.1//EN" "http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd">
+<svg xmlns="http://www.w3.org/2000/svg" xmlns:xl="http://www.w3.org/1999/xlink" xmlns:dc="http://purl.org/dc/elements/1.1/" version="1.1" viewBox="-562 479 876 429" width="876" height="429">
+  <defs/>
+  <g id="09-btmgc-analysis-2" stroke-opacity="1" stroke-dasharray="none" stroke="none" fill="none" fill-opacity="1">
+    <title>09-btmgc-analysis-2</title>
+    <rect fill="white" x="-562" y="479" width="876" height="429"/>
+    <g id="09-btmgc-analysis-2_Layer_1">
+      <title>Layer 1</title>
+      <g id="Graphic_85">
+        <rect x="-404" y="622.386" width="203.5" height="17.5" fill="white"/>
+        <rect x="-404" y="622.386" width="203.5" height="17.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-399 621.912)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="90.974" y="15" xml:space="preserve">C</tspan>
+        </text>
+      </g>
+      <g id="Graphic_84">
+        <rect x="-184.5" y="622.386" width="203.5" height="17.5" fill="white"/>
+        <rect x="-184.5" y="622.386" width="203.5" height="17.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-179.5 621.912)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="90.974" y="15" xml:space="preserve">C</tspan>
+        </text>
+      </g>
+      <g id="Graphic_83">
+        <rect x="34.5" y="622.386" width="203.5" height="17.5" fill="white"/>
+        <rect x="34.5" y="622.386" width="203.5" height="17.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(39.5 621.912)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="90.974" y="15" xml:space="preserve">C</tspan>
+        </text>
+      </g>
+      <g id="Graphic_82">
+        <rect x="-404" y="479.922" width="127" height="77.5" fill="white"/>
+        <rect x="-404" y="479.922" width="127" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-399 509.448)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="53.316" y="15" xml:space="preserve">A</tspan>
+        </text>
+      </g>
+      <g id="Graphic_81">
+        <rect x="-263" y="479.922" width="127" height="77.5" fill="white"/>
+        <rect x="-263" y="479.922" width="127" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-258 509.448)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="53.316" y="15" xml:space="preserve">A</tspan>
+        </text>
+      </g>
+      <g id="Graphic_80">
+        <rect x="-122" y="479.922" width="127" height="77.5" fill="white"/>
+        <rect x="-122" y="479.922" width="127" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-117 509.448)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="53.316" y="15" xml:space="preserve">A</tspan>
+        </text>
+      </g>
+      <g id="Graphic_79">
+        <rect x="19" y="479.922" width="127" height="77.5" fill="white"/>
+        <rect x="19" y="479.922" width="127" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(24 509.448)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="53.316" y="15" xml:space="preserve">A</tspan>
+        </text>
+      </g>
+      <g id="Graphic_78">
+        <rect x="160" y="479.922" width="78" height="77.5" fill="white"/>
+        <rect x="160" y="479.922" width="78" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(165 509.448)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="28.816" y="15" xml:space="preserve">A</tspan>
+        </text>
+      </g>
+      <g id="Line_77">
+        <line x1="-452.5" y1="518.172" x2="251" y2="518.172" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_76">
+        <text transform="translate(-551 509.448)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="0" y="15" xml:space="preserve">GC Horizon</tspan>
+        </text>
+      </g>
+      <g id="Graphic_73">
+        <path d="M -120.675 651.5 L -84.425 651.5 L -84.425 710 L -66.3 710 L -102.55 740 L -138.8 710 L -120.675 710 Z" fill="white"/>
+        <path d="M -120.675 651.5 L -84.425 651.5 L -84.425 710 L -66.3 710 L -102.55 740 L -138.8 710 L -120.675 710 Z" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_89">
+        <rect x="-403.8" y="582.324" width="161.8" height="28.447998" fill="white"/>
+        <rect x="-403.8" y="582.324" width="161.8" height="28.447998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-398.8 587.324)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="70.42" y="15" xml:space="preserve">B</tspan>
+        </text>
+      </g>
+      <g id="Graphic_90">
+        <rect x="-229.5" y="582.342" width="277.2" height="28.447998" fill="white"/>
+        <rect x="-229.5" y="582.342" width="277.2" height="28.447998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-224.5 587.342)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="128.12" y="15" xml:space="preserve">B</tspan>
+        </text>
+      </g>
+      <g id="Graphic_91">
+        <rect x="64.7" y="582.342" width="173.5" height="28.447998" fill="white"/>
+        <rect x="64.7" y="582.342" width="173.5" height="28.447998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(69.7 587.342)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="76.27" y="15" xml:space="preserve">B</tspan>
+        </text>
+      </g>
+      <g id="Graphic_97">
+        <rect x="-403.8" y="564.842" width="490.8" height="12.157997" fill="white"/>
+        <rect x="-403.8" y="564.842" width="490.8" height="12.157997" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-398.8 561.697)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="234.624" y="15" xml:space="preserve">C</tspan>
+        </text>
+      </g>
+      <g id="Graphic_109">
+        <rect x="28.6" y="889.964" width="203.5" height="17.5" fill="white"/>
+        <rect x="28.6" y="889.964" width="203.5" height="17.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(33.6 889.49)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="90.974" y="15" xml:space="preserve">C</tspan>
+        </text>
+      </g>
+      <g id="Graphic_108">
+        <rect x="-409.9" y="747.5" width="127" height="77.5" fill="white"/>
+        <rect x="-409.9" y="747.5" width="127" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-404.9 777.026)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="53.316" y="15" xml:space="preserve">A</tspan>
+        </text>
+      </g>
+      <g id="Graphic_107">
+        <rect x="-268.9" y="747.5" width="127" height="77.5" fill="white"/>
+        <rect x="-268.9" y="747.5" width="127" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-263.9 777.026)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="53.316" y="15" xml:space="preserve">A</tspan>
+        </text>
+      </g>
+      <g id="Graphic_106">
+        <rect x="-127.9" y="747.5" width="127" height="77.5" fill="white"/>
+        <rect x="-127.9" y="747.5" width="127" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-122.9 777.026)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="53.316" y="15" xml:space="preserve">A</tspan>
+        </text>
+      </g>
+      <g id="Graphic_105">
+        <rect x="13.1" y="747.5" width="127" height="77.5" fill="white"/>
+        <rect x="13.1" y="747.5" width="127" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(18.1 777.026)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="53.316" y="15" xml:space="preserve">A</tspan>
+        </text>
+      </g>
+      <g id="Graphic_104">
+        <rect x="154.1" y="747.5" width="78" height="77.5" fill="white"/>
+        <rect x="154.1" y="747.5" width="78" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(159.1 777.026)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="28.816" y="15" xml:space="preserve">A</tspan>
+        </text>
+      </g>
+      <g id="Line_103">
+        <line x1="-458.4" y1="785.75" x2="245.1" y2="785.75" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_102">
+        <text transform="translate(-556.9 777.026)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="0" y="15" xml:space="preserve">GC Horizon</tspan>
+        </text>
+      </g>
+      <g id="Graphic_99">
+        <rect x="58.8" y="849.92" width="173.5" height="28.447998" fill="white"/>
+        <rect x="58.8" y="849.92" width="173.5" height="28.447998" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(63.8 854.92)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="76.27" y="15" xml:space="preserve">B</tspan>
+        </text>
+      </g>
+      <g id="Graphic_98">
+        <rect x="-409.7" y="832.42" width="490.8" height="12.157997" fill="white"/>
+        <rect x="-409.7" y="832.42" width="490.8" height="12.157997" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(-404.7 829.275)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="234.624" y="15" xml:space="preserve">C</tspan>
+        </text>
+      </g>
+      <g id="Graphic_112">
+        <text transform="translate(273 797.5)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="26" fill="black" x="6536993e-19" y="25" xml:space="preserve">B</tspan>
+          <tspan font-family="Lucida Grande" font-size="26" fill="black" y="25" xml:space="preserve">↓</tspan>
+        </text>
+      </g>
+      <g id="Graphic_113">
+        <text transform="translate(273 833.974)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="26" fill="black" x="42277293e-20" y="25" xml:space="preserve">C</tspan>
+          <tspan font-family="Lucida Grande" font-size="26" fill="black" y="25" xml:space="preserve">↓</tspan>
+        </text>
+      </g>
+    </g>
+  </g>
+</svg>

docs/rfcs/images/036-bottom-most-gc-compaction/10-btmgc-analysis-3.svg

@@ -0,0 +1,81 @@
+<?xml version="1.0" encoding="UTF-8" standalone="no"?>
+<!DOCTYPE svg PUBLIC "-//W3C//DTD SVG 1.1//EN" "http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd">
+<svg xmlns="http://www.w3.org/2000/svg" xmlns:xl="http://www.w3.org/1999/xlink" xmlns:dc="http://purl.org/dc/elements/1.1/" version="1.1" viewBox="-12 920 809 269" width="809" height="269">
+  <defs/>
+  <g id="10-btmgc-analysis-3" stroke-opacity="1" stroke-dasharray="none" stroke="none" fill="none" fill-opacity="1">
+    <title>10-btmgc-analysis-3</title>
+    <rect fill="white" x="-12" y="920" width="809" height="269"/>
+    <g id="10-btmgc-analysis-3_Layer_1">
+      <title>Layer 1</title>
+      <g id="Graphic_13">
+        <rect x="433.7" y="949" width="63.559346" height="77.5" fill="white"/>
+        <rect x="433.7" y="949" width="63.559346" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(438.7 978.526)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="8.107673" y="15" xml:space="preserve">1/5 X</tspan>
+        </text>
+      </g>
+      <g id="Graphic_12">
+        <rect x="503.7654" y="949" width="63.559346" height="77.5" fill="white"/>
+        <rect x="503.7654" y="949" width="63.559346" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(508.7654 978.526)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="8.107673" y="15" xml:space="preserve">1/5 X</tspan>
+        </text>
+      </g>
+      <g id="Graphic_11">
+        <rect x="574.8318" y="949" width="63.559346" height="77.5" fill="white"/>
+        <rect x="574.8318" y="949" width="63.559346" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(579.8318 978.526)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="8.107673" y="15" xml:space="preserve">1/5 X</tspan>
+        </text>
+      </g>
+      <g id="Graphic_10">
+        <rect x="645.3977" y="949" width="63.559346" height="77.5" fill="white"/>
+        <rect x="645.3977" y="949" width="63.559346" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(650.3977 978.526)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="8.107673" y="15" xml:space="preserve">1/5 X</tspan>
+        </text>
+      </g>
+      <g id="Line_8">
+        <line x1="92" y1="934.276" x2="795.5" y2="934.276" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_7">
+        <text transform="translate(-6.500003 925.552)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="0" y="15" xml:space="preserve">GC Horizon</tspan>
+        </text>
+      </g>
+      <g id="Graphic_2">
+        <rect x="113.2" y="1033.92" width="321.3" height="12.157997" fill="white"/>
+        <rect x="113.2" y="1033.92" width="321.3" height="12.157997" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(118.2 1030.775)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="150.762" y="15" xml:space="preserve">X</tspan>
+        </text>
+      </g>
+      <g id="Graphic_17">
+        <path d="M 420.125 1062 L 456.375 1062 L 456.375 1120.5 L 474.5 1120.5 L 438.25 1150.5 L 402 1120.5 L 420.125 1120.5 Z" fill="white"/>
+        <path d="M 420.125 1062 L 456.375 1062 L 456.375 1120.5 L 474.5 1120.5 L 438.25 1150.5 L 402 1120.5 L 420.125 1120.5 Z" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Line_27">
+        <line x1="93" y1="1164.224" x2="796.5" y2="1164.224" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_26">
+        <text transform="translate(-5.5000034 1155.5)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="0" y="15" xml:space="preserve">GC Horizon</tspan>
+        </text>
+      </g>
+      <g id="Graphic_25">
+        <rect x="114" y="1173.5" width="641.8" height="12.157997" fill="white"/>
+        <rect x="114" y="1173.5" width="641.8" height="12.157997" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(119 1170.355)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="306.564" y="15" xml:space="preserve">2X</tspan>
+        </text>
+      </g>
+      <g id="Graphic_33">
+        <rect x="715.96355" y="949" width="63.559346" height="77.5" fill="white"/>
+        <rect x="715.96355" y="949" width="63.559346" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(720.96355 978.526)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="8.107673" y="15" xml:space="preserve">1/5 X</tspan>
+        </text>
+      </g>
+    </g>
+  </g>
+</svg>

docs/rfcs/images/036-bottom-most-gc-compaction/11-btmgc-analysis-4.svg

@@ -0,0 +1,81 @@
+<?xml version="1.0" encoding="UTF-8" standalone="no"?>
+<!DOCTYPE svg PUBLIC "-//W3C//DTD SVG 1.1//EN" "http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd">
+<svg xmlns="http://www.w3.org/2000/svg" xmlns:xl="http://www.w3.org/1999/xlink" xmlns:dc="http://purl.org/dc/elements/1.1/" version="1.1" viewBox="-12 920 809 269" width="809" height="269">
+  <defs/>
+  <g id="11-btmgc-analysis-4" stroke-opacity="1" stroke-dasharray="none" stroke="none" fill="none" fill-opacity="1">
+    <title>11-btmgc-analysis-4</title>
+    <rect fill="white" x="-12" y="920" width="809" height="269"/>
+    <g id="11-btmgc-analysis-4_Layer_1">
+      <title>Layer 1</title>
+      <g id="Graphic_13">
+        <rect x="113" y="949" width="127" height="77.5" fill="white"/>
+        <rect x="113" y="949" width="127" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(118 978.526)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="39.084" y="15" xml:space="preserve">1/5 D</tspan>
+        </text>
+      </g>
+      <g id="Graphic_12">
+        <rect x="253" y="949" width="127" height="77.5" fill="white"/>
+        <rect x="253" y="949" width="127" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(258 978.526)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="39.084" y="15" xml:space="preserve">1/5 D</tspan>
+        </text>
+      </g>
+      <g id="Graphic_11">
+        <rect x="395" y="949" width="127" height="77.5" fill="white"/>
+        <rect x="395" y="949" width="127" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(400 978.526)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="39.084" y="15" xml:space="preserve">1/5 D</tspan>
+        </text>
+      </g>
+      <g id="Graphic_10">
+        <rect x="536" y="949" width="127" height="77.5" fill="white"/>
+        <rect x="536" y="949" width="127" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(541 978.526)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="39.084" y="15" xml:space="preserve">1/5 D</tspan>
+        </text>
+      </g>
+      <g id="Graphic_9">
+        <rect x="677" y="949" width="78" height="77.5" fill="white"/>
+        <rect x="677" y="949" width="78" height="77.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(682 978.526)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="14.584" y="15" xml:space="preserve">1/5 D</tspan>
+        </text>
+      </g>
+      <g id="Line_8">
+        <line x1="92" y1="934.276" x2="795.5" y2="934.276" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_7">
+        <text transform="translate(-6.500003 925.552)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="0" y="15" xml:space="preserve">GC Horizon</tspan>
+        </text>
+      </g>
+      <g id="Graphic_2">
+        <rect x="113.2" y="1033.92" width="641.8" height="12.157997" fill="white"/>
+        <rect x="113.2" y="1033.92" width="641.8" height="12.157997" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(118.2 1030.775)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="310.268" y="15" xml:space="preserve">D</tspan>
+        </text>
+      </g>
+      <g id="Graphic_17">
+        <path d="M 420.125 1062 L 456.375 1062 L 456.375 1120.5 L 474.5 1120.5 L 438.25 1150.5 L 402 1120.5 L 420.125 1120.5 Z" fill="white"/>
+        <path d="M 420.125 1062 L 456.375 1062 L 456.375 1120.5 L 474.5 1120.5 L 438.25 1150.5 L 402 1120.5 L 420.125 1120.5 Z" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Line_27">
+        <line x1="93" y1="1164.224" x2="796.5" y2="1164.224" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_26">
+        <text transform="translate(-5.5000034 1155.5)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="0" y="15" xml:space="preserve">GC Horizon</tspan>
+        </text>
+      </g>
+      <g id="Graphic_25">
+        <rect x="114" y="1173.5" width="641.8" height="12.157997" fill="white"/>
+        <rect x="114" y="1173.5" width="641.8" height="12.157997" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(119 1170.355)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="310.268" y="15" xml:space="preserve">D</tspan>
+        </text>
+      </g>
+    </g>
+  </g>
+</svg>

docs/rfcs/images/036-bottom-most-gc-compaction/13-job-split.svg

@@ -0,0 +1,176 @@
+<?xml version="1.0" encoding="UTF-8" standalone="no"?>
+<!DOCTYPE svg PUBLIC "-//W3C//DTD SVG 1.1//EN" "http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd">
+<svg version="1.1" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns="http://www.w3.org/2000/svg" xmlns:xl="http://www.w3.org/1999/xlink" viewBox="210 271 870 514" width="870" height="514">
+  <defs/>
+  <g id="gc-compaction-split" stroke-dasharray="none" fill-opacity="1" stroke="none" fill="none" stroke-opacity="1">
+    <title>gc-compaction-split</title>
+    <rect fill="white" x="210" y="271" width="870" height="514"/>
+    <g id="gc-compaction-split_Layer_1">
+      <title>Layer 1</title>
+      <g id="Graphic_12">
+        <rect x="241" y="272" width="213" height="50.5" fill="white"/>
+        <rect x="241" y="272" width="213" height="50.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_11">
+        <rect x="468.72266" y="272" width="213" height="50.5" fill="white"/>
+        <rect x="468.72266" y="272" width="213" height="50.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_10">
+        <rect x="695.72266" y="272" width="213" height="50.5" fill="white"/>
+        <rect x="695.72266" y="272" width="213" height="50.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_9">
+        <rect x="241" y="337.3711" width="303.5" height="50.5" fill="white"/>
+        <rect x="241" y="337.3711" width="303.5" height="50.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_8">
+        <rect x="556.2617" y="337.3711" width="352.46094" height="50.5" fill="white"/>
+        <rect x="556.2617" y="337.3711" width="352.46094" height="50.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_7">
+        <rect x="241" y="402.7422" width="667.72266" height="50.5" fill="white"/>
+        <rect x="241" y="402.7422" width="667.72266" height="50.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Line_6">
+        <line x1="211" y1="355.5" x2="947.4961" y2="355.5" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_5">
+        <text transform="translate(952.4961 346.776)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="0" y="15" xml:space="preserve">branch point</tspan>
+        </text>
+      </g>
+      <g id="Line_4">
+        <line x1="212" y1="438.5182" x2="948.4961" y2="438.5182" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_3">
+        <text transform="translate(953.4961 429.7942)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="0" y="15" xml:space="preserve">last branch point</tspan>
+        </text>
+      </g>
+      <g id="Graphic_13">
+        <rect x="241" y="272" width="127.99101" height="181.24219" fill="#3a8eed" fill-opacity=".5"/>
+        <text transform="translate(246 353.3971)" fill="white">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="white" x="38.835502" y="15" xml:space="preserve">Job 1</tspan>
+        </text>
+      </g>
+      <g id="Graphic_57">
+        <rect x="359" y="647.96484" width="551.72266" height="50.5" fill="white"/>
+        <rect x="359" y="647.96484" width="551.72266" height="50.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_54">
+        <rect x="359" y="517.22266" width="96" height="50.5" fill="white"/>
+        <rect x="359" y="517.22266" width="96" height="50.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_53">
+        <rect x="469.72266" y="517.22266" width="213" height="50.5" fill="white"/>
+        <rect x="469.72266" y="517.22266" width="213" height="50.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_52">
+        <rect x="696.72266" y="517.22266" width="213" height="50.5" fill="white"/>
+        <rect x="696.72266" y="517.22266" width="213" height="50.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_51">
+        <rect x="359" y="582.59375" width="186.5" height="50.5" fill="white"/>
+        <rect x="359" y="582.59375" width="186.5" height="50.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_50">
+        <rect x="557.2617" y="582.59375" width="352.46094" height="50.5" fill="white"/>
+        <rect x="557.2617" y="582.59375" width="352.46094" height="50.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Line_49">
+        <line x1="212" y1="600.72266" x2="948.4961" y2="600.72266" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_48">
+        <text transform="translate(953.4961 591.99866)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="0" y="15" xml:space="preserve">branch point</tspan>
+        </text>
+      </g>
+      <g id="Line_47">
+        <line x1="213" y1="683.74084" x2="949.4961" y2="683.74084" stroke="#7f8080" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_46">
+        <text transform="translate(954.4961 675.01685)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="0" y="15" xml:space="preserve">last branch point</tspan>
+        </text>
+      </g>
+      <g id="Graphic_63">
+        <rect x="376.72525" y="272" width="127.99101" height="181.24219" fill="#3a8eed" fill-opacity=".5"/>
+        <text transform="translate(381.72525 353.3971)" fill="white">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="white" x="38.835502" y="15" xml:space="preserve">Job 2</tspan>
+        </text>
+      </g>
+      <g id="Graphic_64">
+        <rect x="511.39405" y="272" width="127.99101" height="181.24219" fill="#3a8eed" fill-opacity=".5"/>
+        <text transform="translate(516.39405 353.3971)" fill="white">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="white" x="38.835502" y="15" xml:space="preserve">Job 3</tspan>
+        </text>
+      </g>
+      <g id="Graphic_65">
+        <rect x="646.06285" y="272" width="127.99101" height="181.24219" fill="#3a8eed" fill-opacity=".5"/>
+        <text transform="translate(651.06285 353.3971)" fill="white">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="white" x="38.835502" y="15" xml:space="preserve">Job 4</tspan>
+        </text>
+      </g>
+      <g id="Graphic_66">
+        <rect x="780.73165" y="272" width="127.99101" height="181.24219" fill="#3a8eed" fill-opacity=".5"/>
+        <text transform="translate(785.73165 353.3971)" fill="white">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="white" x="38.835502" y="15" xml:space="preserve">Job 5</tspan>
+        </text>
+      </g>
+      <g id="Graphic_56">
+        <rect x="243.5" y="517.22266" width="125.49101" height="181.24219" fill="#ccc"/>
+        <rect x="243.5" y="517.22266" width="125.49101" height="181.24219" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_55">
+        <rect x="243.5" y="673.46484" width="125.49101" height="17.5" fill="#6b7ca5"/>
+        <rect x="243.5" y="673.46484" width="125.49101" height="17.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_68">
+        <rect x="379.22525" y="517.22266" width="125.49101" height="181.24219" fill="#ccc"/>
+        <rect x="379.22525" y="517.22266" width="125.49101" height="181.24219" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_67">
+        <rect x="379.22525" y="673.46484" width="125.49101" height="17.5" fill="#6b7ca5"/>
+        <rect x="379.22525" y="673.46484" width="125.49101" height="17.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_70">
+        <rect x="514.22525" y="517.22266" width="125.49101" height="181.24219" fill="#ccc"/>
+        <rect x="514.22525" y="517.22266" width="125.49101" height="181.24219" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_69">
+        <rect x="514.22525" y="673.46484" width="125.49101" height="17.5" fill="#6b7ca5"/>
+        <rect x="514.22525" y="673.46484" width="125.49101" height="17.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_72">
+        <rect x="649.22525" y="517.22266" width="125.49101" height="181.24219" fill="#ccc"/>
+        <rect x="649.22525" y="517.22266" width="125.49101" height="181.24219" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_71">
+        <rect x="649.22525" y="673.46484" width="125.49101" height="17.5" fill="#6b7ca5"/>
+        <rect x="649.22525" y="673.46484" width="125.49101" height="17.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_74">
+        <rect x="785.23165" y="517.22266" width="125.49101" height="181.24219" fill="#ccc"/>
+        <rect x="785.23165" y="517.22266" width="125.49101" height="181.24219" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_73">
+        <rect x="785.23165" y="673.46484" width="125.49101" height="17.5" fill="#6b7ca5"/>
+        <rect x="785.23165" y="673.46484" width="125.49101" height="17.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+      </g>
+      <g id="Graphic_78">
+        <rect x="241" y="731.3359" width="125.49101" height="27.26953" fill="#ccc"/>
+        <rect x="241" y="731.3359" width="125.49101" height="27.26953" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(246 735.7467)" fill="black">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="black" x="17.297502" y="15" xml:space="preserve">Delta Layer</tspan>
+        </text>
+      </g>
+      <g id="Graphic_79">
+        <rect x="241" y="766.759" width="125.49101" height="17.5" fill="#6b7ca5"/>
+        <rect x="241" y="766.759" width="125.49101" height="17.5" stroke="black" stroke-linecap="round" stroke-linejoin="round" stroke-width="1"/>
+        <text transform="translate(246 766.285)" fill="white">
+          <tspan font-family="Helvetica Neue" font-size="16" fill="white" x="13.737502" y="15" xml:space="preserve">Image Layer</tspan>
+        </text>
+      </g>
+    </g>
+  </g>
+</svg>

endpoint_storage/src/app.rs

@@ -462,6 +462,8 @@ MC4CAQAwBQYDK2VwBCIEID/Drmc1AA6U/znNRWpF3zEGegOATQxfkdWxitcOMsIH
         if var(REAL_S3_ENV).is_ok() {
             assert!(body.contains("remote_storage_s3_deleted_objects_total"));
         }
+
+        #[cfg(target_os = "linux")]
         assert!(body.contains("process_threads"));
     }
 

libs/neon-shmem/Cargo.toml

@@ -0,0 +1,13 @@
+[package]
+name = "neon-shmem"
+version = "0.1.0"
+edition.workspace = true
+license.workspace = true
+
+[dependencies]
+thiserror.workspace = true
+nix.workspace=true
+workspace_hack = { version = "0.1", path = "../../workspace_hack" }
+
+[target.'cfg(target_os = "macos")'.dependencies]
+tempfile = "3.14.0"

libs/neon-shmem/src/lib.rs

@@ -0,0 +1,418 @@
+//! Shared memory utilities for neon communicator
+
+use std::num::NonZeroUsize;
+use std::os::fd::{AsFd, BorrowedFd, OwnedFd};
+use std::ptr::NonNull;
+use std::sync::atomic::{AtomicUsize, Ordering};
+
+use nix::errno::Errno;
+use nix::sys::mman::MapFlags;
+use nix::sys::mman::ProtFlags;
+use nix::sys::mman::mmap as nix_mmap;
+use nix::sys::mman::munmap as nix_munmap;
+use nix::unistd::ftruncate as nix_ftruncate;
+
+/// ShmemHandle represents a shared memory area that can be shared by processes over fork().
+/// Unlike shared memory allocated by Postgres, this area is resizable, up to 'max_size' that's
+/// specified at creation.
+///
+/// The area is backed by an anonymous file created with memfd_create(). The full address space for
+/// 'max_size' is reserved up-front with mmap(), but whenever you call [`ShmemHandle::set_size`],
+/// the underlying file is resized. Do not access the area beyond the current size. Currently, that
+/// will cause the file to be expanded, but we might use mprotect() etc. to enforce that in the
+/// future.
+pub struct ShmemHandle {
+    /// memfd file descriptor
+    fd: OwnedFd,
+
+    max_size: usize,
+
+    // Pointer to the beginning of the shared memory area. The header is stored there.
+    shared_ptr: NonNull<SharedStruct>,
+
+    // Pointer to the beginning of the user data
+    pub data_ptr: NonNull<u8>,
+}
+
+/// This is stored at the beginning in the shared memory area.
+struct SharedStruct {
+    max_size: usize,
+
+    /// Current size of the backing file. The high-order bit is used for the RESIZE_IN_PROGRESS flag
+    current_size: AtomicUsize,
+}
+
+const RESIZE_IN_PROGRESS: usize = 1 << 63;
+
+const HEADER_SIZE: usize = std::mem::size_of::<SharedStruct>();
+
+/// Error type returned by the ShmemHandle functions.
+#[derive(thiserror::Error, Debug)]
+#[error("{msg}: {errno}")]
+pub struct Error {
+    pub msg: String,
+    pub errno: Errno,
+}
+
+impl Error {
+    fn new(msg: &str, errno: Errno) -> Error {
+        Error {
+            msg: msg.to_string(),
+            errno,
+        }
+    }
+}
+
+impl ShmemHandle {
+    /// Create a new shared memory area. To communicate between processes, the processes need to be
+    /// fork()'d after calling this, so that the ShmemHandle is inherited by all processes.
+    ///
+    /// If the ShmemHandle is dropped, the memory is unmapped from the current process. Other
+    /// processes can continue using it, however.
+    pub fn new(name: &str, initial_size: usize, max_size: usize) -> Result<ShmemHandle, Error> {
+        // create the backing anonymous file.
+        let fd = create_backing_file(name)?;
+
+        Self::new_with_fd(fd, initial_size, max_size)
+    }
+
+    fn new_with_fd(
+        fd: OwnedFd,
+        initial_size: usize,
+        max_size: usize,
+    ) -> Result<ShmemHandle, Error> {
+        // We reserve the high-order bit for the RESIZE_IN_PROGRESS flag, and the actual size
+        // is a little larger than this because of the SharedStruct header. Make the upper limit
+        // somewhat smaller than that, because with anything close to that, you'll run out of
+        // memory anyway.
+        if max_size >= 1 << 48 {
+            panic!("max size {} too large", max_size);
+        }
+        if initial_size > max_size {
+            panic!("initial size {initial_size} larger than max size {max_size}");
+        }
+
+        // The actual initial / max size is the one given by the caller, plus the size of
+        // 'SharedStruct'.
+        let initial_size = HEADER_SIZE + initial_size;
+        let max_size = NonZeroUsize::new(HEADER_SIZE + max_size).unwrap();
+
+        // Reserve address space for it with mmap
+        //
+        // TODO: Use MAP_HUGETLB if possible
+        let start_ptr = unsafe {
+            nix_mmap(
+                None,
+                max_size,
+                ProtFlags::PROT_READ | ProtFlags::PROT_WRITE,
+                MapFlags::MAP_SHARED,
+                &fd,
+                0,
+            )
+        }
+        .map_err(|e| Error::new("mmap failed: {e}", e))?;
+
+        // Reserve space for the initial size
+        enlarge_file(fd.as_fd(), initial_size as u64)?;
+
+        // Initialize the header
+        let shared: NonNull<SharedStruct> = start_ptr.cast();
+        unsafe {
+            shared.write(SharedStruct {
+                max_size: max_size.into(),
+                current_size: AtomicUsize::new(initial_size),
+            })
+        };
+
+        // The user data begins after the header
+        let data_ptr = unsafe { start_ptr.cast().add(HEADER_SIZE) };
+
+        Ok(ShmemHandle {
+            fd,
+            max_size: max_size.into(),
+            shared_ptr: shared,
+            data_ptr,
+        })
+    }
+
+    // return reference to the header
+    fn shared(&self) -> &SharedStruct {
+        unsafe { self.shared_ptr.as_ref() }
+    }
+
+    /// Resize the shared memory area. 'new_size' must not be larger than the 'max_size' specified
+    /// when creating the area.
+    ///
+    /// This may only be called from one process/thread concurrently. We detect that case
+    /// and return an Error.
+    pub fn set_size(&self, new_size: usize) -> Result<(), Error> {
+        let new_size = new_size + HEADER_SIZE;
+        let shared = self.shared();
+
+        if new_size > self.max_size {
+            panic!(
+                "new size ({} is greater than max size ({})",
+                new_size, self.max_size
+            );
+        }
+        assert_eq!(self.max_size, shared.max_size);
+
+        // Lock the area by setting the bit in 'current_size'
+        //
+        // Ordering::Relaxed would probably be sufficient here, as we don't access any other memory
+        // and the posix_fallocate/ftruncate call is surely a synchronization point anyway. But
+        // since this is not performance-critical, better safe than sorry .
+        let mut old_size = shared.current_size.load(Ordering::Acquire);
+        loop {
+            if (old_size & RESIZE_IN_PROGRESS) != 0 {
+                return Err(Error::new(
+                    "concurrent resize detected",
+                    Errno::UnknownErrno,
+                ));
+            }
+            match shared.current_size.compare_exchange(
+                old_size,
+                new_size,
+                Ordering::Acquire,
+                Ordering::Relaxed,
+            ) {
+                Ok(_) => break,
+                Err(x) => old_size = x,
+            }
+        }
+
+        // Ok, we got the lock.
+        //
+        // NB: If anything goes wrong, we *must* clear the bit!
+        let result = {
+            use std::cmp::Ordering::{Equal, Greater, Less};
+            match new_size.cmp(&old_size) {
+                Less => nix_ftruncate(&self.fd, new_size as i64).map_err(|e| {
+                    Error::new("could not shrink shmem segment, ftruncate failed: {e}", e)
+                }),
+                Equal => Ok(()),
+                Greater => enlarge_file(self.fd.as_fd(), new_size as u64),
+            }
+        };
+
+        // Unlock
+        shared.current_size.store(
+            if result.is_ok() { new_size } else { old_size },
+            Ordering::Release,
+        );
+
+        result
+    }
+
+    /// Returns the current user-visible size of the shared memory segment.
+    ///
+    /// NOTE: a concurrent set_size() call can change the size at any time. It is the caller's
+    /// responsibility not to access the area beyond the current size.
+    pub fn current_size(&self) -> usize {
+        let total_current_size =
+            self.shared().current_size.load(Ordering::Relaxed) & !RESIZE_IN_PROGRESS;
+        total_current_size - HEADER_SIZE
+    }
+}
+
+impl Drop for ShmemHandle {
+    fn drop(&mut self) {
+        // SAFETY: The pointer was obtained from mmap() with the given size.
+        // We unmap the entire region.
+        let _ = unsafe { nix_munmap(self.shared_ptr.cast(), self.max_size) };
+        // The fd is dropped automatically by OwnedFd.
+    }
+}
+
+/// Create a "backing file" for the shared memory area. On Linux, use memfd_create(), to create an
+/// anonymous in-memory file. One macos, fall back to a regular file. That's good enough for
+/// development and testing, but in production we want the file to stay in memory.
+///
+/// disable 'unused_variables' warnings, because in the macos path, 'name' is unused.
+#[allow(unused_variables)]
+fn create_backing_file(name: &str) -> Result<OwnedFd, Error> {
+    #[cfg(not(target_os = "macos"))]
+    {
+        nix::sys::memfd::memfd_create(name, nix::sys::memfd::MFdFlags::empty())
+            .map_err(|e| Error::new("memfd_create failed: {e}", e))
+    }
+    #[cfg(target_os = "macos")]
+    {
+        let file = tempfile::tempfile().map_err(|e| {
+            Error::new(
+                "could not create temporary file to back shmem area: {e}",
+                nix::errno::Errno::from_raw(e.raw_os_error().unwrap_or(0)),
+            )
+        })?;
+        Ok(OwnedFd::from(file))
+    }
+}
+
+fn enlarge_file(fd: BorrowedFd, size: u64) -> Result<(), Error> {
+    // Use posix_fallocate() to enlarge the file. It reserves the space correctly, so that
+    // we don't get a segfault later when trying to actually use it.
+    #[cfg(not(target_os = "macos"))]
+    {
+        nix::fcntl::posix_fallocate(fd, 0, size as i64).map_err(|e| {
+            Error::new(
+                "could not grow shmem segment, posix_fallocate failed: {e}",
+                e,
+            )
+        })
+    }
+    // As a fallback on macos, which doesn't have posix_fallocate, use plain 'fallocate'
+    #[cfg(target_os = "macos")]
+    {
+        nix::unistd::ftruncate(fd, size as i64)
+            .map_err(|e| Error::new("could not grow shmem segment, ftruncate failed: {e}", e))
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    use nix::unistd::ForkResult;
+    use std::ops::Range;
+
+    /// check that all bytes in given range have the expected value.
+    fn assert_range(ptr: *const u8, expected: u8, range: Range<usize>) {
+        for i in range {
+            let b = unsafe { *(ptr.add(i)) };
+            assert_eq!(expected, b, "unexpected byte at offset {}", i);
+        }
+    }
+
+    /// Write 'b' to all bytes in the given range
+    fn write_range(ptr: *mut u8, b: u8, range: Range<usize>) {
+        unsafe { std::ptr::write_bytes(ptr.add(range.start), b, range.end - range.start) };
+    }
+
+    // simple single-process test of growing and shrinking
+    #[test]
+    fn test_shmem_resize() -> Result<(), Error> {
+        let max_size = 1024 * 1024;
+        let init_struct = ShmemHandle::new("test_shmem_resize", 0, max_size)?;
+
+        assert_eq!(init_struct.current_size(), 0);
+
+        // Initial grow
+        let size1 = 10000;
+        init_struct.set_size(size1).unwrap();
+        assert_eq!(init_struct.current_size(), size1);
+
+        // Write some data
+        let data_ptr = init_struct.data_ptr.as_ptr();
+        write_range(data_ptr, 0xAA, 0..size1);
+        assert_range(data_ptr, 0xAA, 0..size1);
+
+        // Shrink
+        let size2 = 5000;
+        init_struct.set_size(size2).unwrap();
+        assert_eq!(init_struct.current_size(), size2);
+
+        // Grow again
+        let size3 = 20000;
+        init_struct.set_size(size3).unwrap();
+        assert_eq!(init_struct.current_size(), size3);
+
+        // Try to read it. The area that was shrunk and grown again should read as all zeros now
+        assert_range(data_ptr, 0xAA, 0..5000);
+        assert_range(data_ptr, 0, 5000..size1);
+
+        // Try to grow beyond max_size
+        //let size4 = max_size + 1;
+        //assert!(init_struct.set_size(size4).is_err());
+
+        // Dropping init_struct should unmap the memory
+        drop(init_struct);
+
+        Ok(())
+    }
+
+    /// This is used in tests to coordinate between test processes. It's like std::sync::Barrier,
+    /// but is stored in the shared memory area and works across processes. It's implemented by
+    /// polling, because e.g. standard rust mutexes are not guaranteed to work across processes.
+    struct SimpleBarrier {
+        num_procs: usize,
+        count: AtomicUsize,
+    }
+
+    impl SimpleBarrier {
+        unsafe fn init(ptr: *mut SimpleBarrier, num_procs: usize) {
+            unsafe {
+                *ptr = SimpleBarrier {
+                    num_procs,
+                    count: AtomicUsize::new(0),
+                }
+            }
+        }
+
+        pub fn wait(&self) {
+            let old = self.count.fetch_add(1, Ordering::Relaxed);
+
+            let generation = old / self.num_procs;
+
+            let mut current = old + 1;
+            while current < (generation + 1) * self.num_procs {
+                std::thread::sleep(std::time::Duration::from_millis(10));
+                current = self.count.load(Ordering::Relaxed);
+            }
+        }
+    }
+
+    #[test]
+    fn test_multi_process() {
+        // Initialize
+        let max_size = 1_000_000_000_000;
+        let init_struct = ShmemHandle::new("test_multi_process", 0, max_size).unwrap();
+        let ptr = init_struct.data_ptr.as_ptr();
+
+        // Store the SimpleBarrier in the first 1k of the area.
+        init_struct.set_size(10000).unwrap();
+        let barrier_ptr: *mut SimpleBarrier = unsafe {
+            ptr.add(ptr.align_offset(std::mem::align_of::<SimpleBarrier>()))
+                .cast()
+        };
+        unsafe { SimpleBarrier::init(barrier_ptr, 2) };
+        let barrier = unsafe { barrier_ptr.as_ref().unwrap() };
+
+        // Fork another test process. The code after this runs in both processes concurrently.
+        let fork_result = unsafe { nix::unistd::fork().unwrap() };
+
+        // In the parent, fill bytes between 1000..2000. In the child, between 2000..3000
+        if fork_result.is_parent() {
+            write_range(ptr, 0xAA, 1000..2000);
+        } else {
+            write_range(ptr, 0xBB, 2000..3000);
+        }
+        barrier.wait();
+        // Verify the contents. (in both processes)
+        assert_range(ptr, 0xAA, 1000..2000);
+        assert_range(ptr, 0xBB, 2000..3000);
+
+        // Grow, from the child this time
+        let size = 10_000_000;
+        if !fork_result.is_parent() {
+            init_struct.set_size(size).unwrap();
+        }
+        barrier.wait();
+
+        // make some writes at the end
+        if fork_result.is_parent() {
+            write_range(ptr, 0xAA, (size - 10)..size);
+        } else {
+            write_range(ptr, 0xBB, (size - 20)..(size - 10));
+        }
+        barrier.wait();
+
+        // Verify the contents. (This runs in both processes)
+        assert_range(ptr, 0, (size - 1000)..(size - 20));
+        assert_range(ptr, 0xBB, (size - 20)..(size - 10));
+        assert_range(ptr, 0xAA, (size - 10)..size);
+
+        if let ForkResult::Parent { child } = fork_result {
+            nix::sys::wait::waitpid(child, None).unwrap();
+        }
+    }
+}

libs/pageserver_api/src/config.rs

@@ -8,6 +8,8 @@ pub const DEFAULT_PG_LISTEN_PORT: u16 = 64000;
 pub const DEFAULT_PG_LISTEN_ADDR: &str = formatcp!("127.0.0.1:{DEFAULT_PG_LISTEN_PORT}");
 pub const DEFAULT_HTTP_LISTEN_PORT: u16 = 9898;
 pub const DEFAULT_HTTP_LISTEN_ADDR: &str = formatcp!("127.0.0.1:{DEFAULT_HTTP_LISTEN_PORT}");
+// TODO: gRPC is disabled by default for now, but the port is used in neon_local.
+pub const DEFAULT_GRPC_LISTEN_PORT: u16 = 51051; // storage-broker already uses 50051
 
 use std::collections::HashMap;
 use std::num::{NonZeroU64, NonZeroUsize};
@@ -43,6 +45,21 @@ pub struct NodeMetadata {
     pub other: HashMap<String, serde_json::Value>,
 }
 
+/// PostHog integration config.
+#[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
+pub struct PostHogConfig {
+    /// PostHog project ID
+    pub project_id: String,
+    /// Server-side (private) API key
+    pub server_api_key: String,
+    /// Client-side (public) API key
+    pub client_api_key: String,
+    /// Private API URL
+    pub private_api_url: String,
+    /// Public API URL
+    pub public_api_url: String,
+}
+
 /// `pageserver.toml`
 ///
 /// We use serde derive with `#[serde(default)]` to generate a deserializer
@@ -104,6 +121,7 @@ pub struct ConfigToml {
     pub listen_pg_addr: String,
     pub listen_http_addr: String,
     pub listen_https_addr: Option<String>,
+    pub listen_grpc_addr: Option<String>,
     pub ssl_key_file: Utf8PathBuf,
     pub ssl_cert_file: Utf8PathBuf,
     #[serde(with = "humantime_serde")]
@@ -123,6 +141,7 @@ pub struct ConfigToml {
     pub http_auth_type: AuthType,
     #[serde_as(as = "serde_with::DisplayFromStr")]
     pub pg_auth_type: AuthType,
+    pub grpc_auth_type: AuthType,
     pub auth_validation_public_key_path: Option<Utf8PathBuf>,
     pub remote_storage: Option<RemoteStorageConfig>,
     pub tenant_config: TenantConfigToml,
@@ -182,7 +201,11 @@ pub struct ConfigToml {
     pub tracing: Option<Tracing>,
     pub enable_tls_page_service_api: bool,
     pub dev_mode: bool,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub posthog_config: Option<PostHogConfig>,
     pub timeline_import_config: TimelineImportConfig,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub basebackup_cache_config: Option<BasebackupCacheConfig>,
 }
 
 #[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
@@ -235,7 +258,7 @@ pub enum PageServiceProtocolPipelinedBatchingStrategy {
     ScatteredLsn,
 }
 
-#[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
+#[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
 #[serde(tag = "mode", rename_all = "kebab-case")]
 pub enum GetVectoredConcurrentIo {
     /// The read path is fully sequential: layers are visited
@@ -305,6 +328,27 @@ impl From<OtelExporterProtocol> for tracing_utils::Protocol {
 pub struct TimelineImportConfig {
     pub import_job_concurrency: NonZeroUsize,
     pub import_job_soft_size_limit: NonZeroUsize,
+    pub import_job_checkpoint_threshold: NonZeroUsize,
+}
+
+#[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
+#[serde(default)]
+pub struct BasebackupCacheConfig {
+    #[serde(with = "humantime_serde")]
+    pub cleanup_period: Duration,
+    // FIXME: Support max_size_bytes.
+    // pub max_size_bytes: usize,
+    pub max_size_entries: i64,
+}
+
+impl Default for BasebackupCacheConfig {
+    fn default() -> Self {
+        Self {
+            cleanup_period: Duration::from_secs(60),
+            // max_size_bytes: 1024 * 1024 * 1024, // 1 GiB
+            max_size_entries: 1000,
+        }
+    }
 }
 
 pub mod statvfs {
@@ -490,6 +534,14 @@ pub struct TenantConfigToml {
     /// Tenant level performance sampling ratio override. Controls the ratio of get page requests
     /// that will get perf sampling for the tenant.
     pub sampling_ratio: Option<Ratio>,
+
+    /// Capacity of relsize snapshot cache (used by replicas).
+    pub relsize_snapshot_cache_capacity: usize,
+
+    /// Enable preparing basebackup on XLOG_CHECKPOINT_SHUTDOWN and using it in basebackup requests.
+    // FIXME: Remove skip_serializing_if when the feature is stable.
+    #[serde(skip_serializing_if = "std::ops::Not::not")]
+    pub basebackup_cache_enabled: bool,
 }
 
 pub mod defaults {
@@ -557,6 +609,7 @@ impl Default for ConfigToml {
             listen_pg_addr: (DEFAULT_PG_LISTEN_ADDR.to_string()),
             listen_http_addr: (DEFAULT_HTTP_LISTEN_ADDR.to_string()),
             listen_https_addr: (None),
+            listen_grpc_addr: None, // TODO: default to 127.0.0.1:51051
             ssl_key_file: Utf8PathBuf::from(DEFAULT_SSL_KEY_FILE),
             ssl_cert_file: Utf8PathBuf::from(DEFAULT_SSL_CERT_FILE),
             ssl_cert_reload_period: Duration::from_secs(60),
@@ -573,6 +626,7 @@ impl Default for ConfigToml {
             pg_distrib_dir: None, // Utf8PathBuf::from("./pg_install"), // TODO: formely, this was std::env::current_dir()
             http_auth_type: (AuthType::Trust),
             pg_auth_type: (AuthType::Trust),
+            grpc_auth_type: (AuthType::Trust),
             auth_validation_public_key_path: (None),
             remote_storage: None,
             broker_endpoint: (storage_broker::DEFAULT_ENDPOINT
@@ -639,23 +693,15 @@ impl Default for ConfigToml {
             tenant_config: TenantConfigToml::default(),
             no_sync: None,
             wal_receiver_protocol: DEFAULT_WAL_RECEIVER_PROTOCOL,
-            page_service_pipelining: if !cfg!(test) {
-                PageServicePipeliningConfig::Serial
-            } else {
-                // Do not turn this into the default until scattered reads have been
-                // validated and rolled-out fully.
-                PageServicePipeliningConfig::Pipelined(PageServicePipeliningConfigPipelined {
+            page_service_pipelining: PageServicePipeliningConfig::Pipelined(
+                PageServicePipeliningConfigPipelined {
                     max_batch_size: NonZeroUsize::new(32).unwrap(),
                     execution: PageServiceProtocolPipelinedExecutionStrategy::ConcurrentFutures,
                     batching: PageServiceProtocolPipelinedBatchingStrategy::ScatteredLsn,
-                })
-            },
-            get_vectored_concurrent_io: if !cfg!(test) {
-                GetVectoredConcurrentIo::Sequential
-            } else {
-                GetVectoredConcurrentIo::SidecarTask
-            },
-            enable_read_path_debugging: if cfg!(test) || cfg!(feature = "testing") {
+                },
+            ),
+            get_vectored_concurrent_io: GetVectoredConcurrentIo::SidecarTask,
+            enable_read_path_debugging: if cfg!(feature = "testing") {
                 Some(true)
             } else {
                 None
@@ -669,7 +715,10 @@ impl Default for ConfigToml {
             timeline_import_config: TimelineImportConfig {
                 import_job_concurrency: NonZeroUsize::new(128).unwrap(),
                 import_job_soft_size_limit: NonZeroUsize::new(1024 * 1024 * 1024).unwrap(),
+                import_job_checkpoint_threshold: NonZeroUsize::new(128).unwrap(),
             },
+            basebackup_cache_config: None,
+            posthog_config: None,
         }
     }
 }
@@ -736,6 +785,7 @@ pub mod tenant_conf_defaults {
     pub const DEFAULT_GC_COMPACTION_VERIFICATION: bool = true;
     pub const DEFAULT_GC_COMPACTION_INITIAL_THRESHOLD_KB: u64 = 5 * 1024 * 1024; // 5GB
     pub const DEFAULT_GC_COMPACTION_RATIO_PERCENT: u64 = 100;
+    pub const DEFAULT_RELSIZE_SNAPSHOT_CACHE_CAPACITY: usize = 1000;
 }
 
 impl Default for TenantConfigToml {
@@ -793,6 +843,8 @@ impl Default for TenantConfigToml {
             gc_compaction_initial_threshold_kb: DEFAULT_GC_COMPACTION_INITIAL_THRESHOLD_KB,
             gc_compaction_ratio_percent: DEFAULT_GC_COMPACTION_RATIO_PERCENT,
             sampling_ratio: None,
+            relsize_snapshot_cache_capacity: DEFAULT_RELSIZE_SNAPSHOT_CACHE_CAPACITY,
+            basebackup_cache_enabled: false,
         }
     }
 }

libs/pageserver_api/src/key.rs

@@ -910,6 +910,11 @@ impl Key {
         self.field1 == 0x00 && self.field4 != 0 && self.field6 != 0xffffffff
     }
 
+    #[inline(always)]
+    pub fn is_rel_block_of_rel(&self, rel: Oid) -> bool {
+        self.is_rel_block_key() && self.field4 == rel
+    }
+
     #[inline(always)]
     pub fn is_rel_dir_key(&self) -> bool {
         self.field1 == 0x00

libs/pageserver_api/src/models.rs

@@ -336,14 +336,30 @@ impl TimelineCreateRequest {
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq)]
 pub enum ShardImportStatus {
-    InProgress,
+    InProgress(Option<ShardImportProgress>),
     Done,
     Error(String),
 }
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq)]
+pub enum ShardImportProgress {
+    V1(ShardImportProgressV1),
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, Eq)]
+pub struct ShardImportProgressV1 {
+    /// Total number of jobs in the import plan
+    pub jobs: usize,
+    /// Number of jobs completed
+    pub completed: usize,
+    /// Hash of the plan
+    pub import_plan_hash: u64,
+}
+
 impl ShardImportStatus {
     pub fn is_terminal(&self) -> bool {
         match self {
-            ShardImportStatus::InProgress => false,
+            ShardImportStatus::InProgress(_) => false,
             ShardImportStatus::Done | ShardImportStatus::Error(_) => true,
         }
     }
@@ -386,6 +402,8 @@ pub enum TimelineCreateRequestMode {
         // using a flattened enum, so, it was an accepted field, and
         // we continue to accept it by having it here.
         pg_version: Option<u32>,
+        #[serde(default, skip_serializing_if = "std::ops::Not::not")]
+        read_only: bool,
     },
     ImportPgdata {
         import_pgdata: TimelineCreateRequestModeImportPgdata,
@@ -614,6 +632,10 @@ pub struct TenantConfigPatch {
     pub gc_compaction_ratio_percent: FieldPatch<u64>,
     #[serde(skip_serializing_if = "FieldPatch::is_noop")]
     pub sampling_ratio: FieldPatch<Option<Ratio>>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub relsize_snapshot_cache_capacity: FieldPatch<usize>,
+    #[serde(skip_serializing_if = "FieldPatch::is_noop")]
+    pub basebackup_cache_enabled: FieldPatch<bool>,
 }
 
 /// Like [`crate::config::TenantConfigToml`], but preserves the information
@@ -743,6 +765,12 @@ pub struct TenantConfig {
 
     #[serde(skip_serializing_if = "Option::is_none")]
     pub sampling_ratio: Option<Option<Ratio>>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub relsize_snapshot_cache_capacity: Option<usize>,
+
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub basebackup_cache_enabled: Option<bool>,
 }
 
 impl TenantConfig {
@@ -788,6 +816,8 @@ impl TenantConfig {
             mut gc_compaction_initial_threshold_kb,
             mut gc_compaction_ratio_percent,
             mut sampling_ratio,
+            mut relsize_snapshot_cache_capacity,
+            mut basebackup_cache_enabled,
         } = self;
 
         patch.checkpoint_distance.apply(&mut checkpoint_distance);
@@ -889,6 +919,12 @@ impl TenantConfig {
             .gc_compaction_ratio_percent
             .apply(&mut gc_compaction_ratio_percent);
         patch.sampling_ratio.apply(&mut sampling_ratio);
+        patch
+            .relsize_snapshot_cache_capacity
+            .apply(&mut relsize_snapshot_cache_capacity);
+        patch
+            .basebackup_cache_enabled
+            .apply(&mut basebackup_cache_enabled);
 
         Ok(Self {
             checkpoint_distance,
@@ -928,6 +964,8 @@ impl TenantConfig {
             gc_compaction_initial_threshold_kb,
             gc_compaction_ratio_percent,
             sampling_ratio,
+            relsize_snapshot_cache_capacity,
+            basebackup_cache_enabled,
         })
     }
 
@@ -1036,6 +1074,12 @@ impl TenantConfig {
                 .gc_compaction_ratio_percent
                 .unwrap_or(global_conf.gc_compaction_ratio_percent),
             sampling_ratio: self.sampling_ratio.unwrap_or(global_conf.sampling_ratio),
+            relsize_snapshot_cache_capacity: self
+                .relsize_snapshot_cache_capacity
+                .unwrap_or(global_conf.relsize_snapshot_cache_capacity),
+            basebackup_cache_enabled: self
+                .basebackup_cache_enabled
+                .unwrap_or(global_conf.basebackup_cache_enabled),
         }
     }
 }
@@ -1803,7 +1847,6 @@ pub struct TopTenantShardsResponse {
 }
 
 pub mod virtual_file {
-    use std::sync::LazyLock;
 
     #[derive(
         Copy,
@@ -1851,15 +1894,7 @@ pub mod virtual_file {
 
     impl IoMode {
         pub fn preferred() -> Self {
-            // The default behavior when running Rust unit tests without any further
-            // flags is to use the newest behavior (DirectRw).
-            // The CI uses the environment variable to unit tests for all different modes.
-            // NB: the Python regression & perf tests have their own defaults management
-            // that writes pageserver.toml; they do not use this variable.
-            static ENV_OVERRIDE: LazyLock<Option<IoMode>> = LazyLock::new(|| {
-                utils::env::var_serde_json_string("NEON_PAGESERVER_UNIT_TEST_VIRTUAL_FILE_IO_MODE")
-            });
-            ENV_OVERRIDE.unwrap_or(IoMode::DirectRw)
+            IoMode::DirectRw
         }
     }
 

libs/pageserver_api/src/upcall_api.rs

@@ -4,6 +4,7 @@
 //! See docs/rfcs/025-generation-numbers.md
 
 use serde::{Deserialize, Serialize};
+use utils::generation::Generation;
 use utils::id::{NodeId, TimelineId};
 
 use crate::controller_api::NodeRegisterRequest;
@@ -63,9 +64,17 @@ pub struct ValidateResponseTenant {
     pub valid: bool,
 }
 
+#[derive(Serialize, Deserialize)]
+pub struct TimelineImportStatusRequest {
+    pub tenant_shard_id: TenantShardId,
+    pub timeline_id: TimelineId,
+    pub generation: Generation,
+}
+
 #[derive(Serialize, Deserialize)]
 pub struct PutTimelineImportStatusRequest {
     pub tenant_shard_id: TenantShardId,
     pub timeline_id: TimelineId,
     pub status: ShardImportStatus,
+    pub generation: Generation,
 }

libs/pageserver_api/src/value.rs

@@ -36,6 +36,24 @@ impl Value {
             Value::WalRecord(rec) => rec.will_init(),
         }
     }
+
+    #[inline(always)]
+    pub fn estimated_size(&self) -> usize {
+        match self {
+            Value::Image(image) => image.len(),
+            Value::WalRecord(NeonWalRecord::AuxFile {
+                content: Some(content),
+                ..
+            }) => content.len(),
+            Value::WalRecord(NeonWalRecord::Postgres { rec, .. }) => rec.len(),
+            Value::WalRecord(NeonWalRecord::ClogSetAborted { xids }) => xids.len() * 4,
+            Value::WalRecord(NeonWalRecord::ClogSetCommitted { xids, .. }) => xids.len() * 4,
+            Value::WalRecord(NeonWalRecord::MultixactMembersCreate { members, .. }) => {
+                members.len() * 8
+            }
+            _ => 8192, /* use image size as the estimation */
+        }
+    }
 }
 
 #[derive(Debug, PartialEq)]

libs/posthog_client_lite/Cargo.toml

@@ -0,0 +1,19 @@
+[package]
+name = "posthog_client_lite"
+version = "0.1.0"
+edition = "2024"
+license.workspace = true
+
+[dependencies]
+anyhow.workspace = true
+arc-swap.workspace = true
+reqwest.workspace = true
+serde_json.workspace = true
+serde.workspace = true
+sha2.workspace = true
+thiserror.workspace = true
+tokio = { workspace = true, features = ["process", "sync", "fs", "rt", "io-util", "time"] }
+tokio-util.workspace = true
+tracing-utils.workspace = true
+tracing.workspace = true
+workspace_hack.workspace = true

libs/posthog_client_lite/src/background_loop.rs

@@ -0,0 +1,59 @@
+//! A background loop that fetches feature flags from PostHog and updates the feature store.
+
+use std::{sync::Arc, time::Duration};
+
+use arc_swap::ArcSwap;
+use tokio_util::sync::CancellationToken;
+
+use crate::{FeatureStore, PostHogClient, PostHogClientConfig};
+
+/// A background loop that fetches feature flags from PostHog and updates the feature store.
+pub struct FeatureResolverBackgroundLoop {
+    posthog_client: PostHogClient,
+    feature_store: ArcSwap<FeatureStore>,
+    cancel: CancellationToken,
+}
+
+impl FeatureResolverBackgroundLoop {
+    pub fn new(config: PostHogClientConfig, shutdown_pageserver: CancellationToken) -> Self {
+        Self {
+            posthog_client: PostHogClient::new(config),
+            feature_store: ArcSwap::new(Arc::new(FeatureStore::new())),
+            cancel: shutdown_pageserver,
+        }
+    }
+
+    pub fn spawn(self: Arc<Self>, handle: &tokio::runtime::Handle, refresh_period: Duration) {
+        let this = self.clone();
+        let cancel = self.cancel.clone();
+        handle.spawn(async move {
+            tracing::info!("Starting PostHog feature resolver");
+            let mut ticker = tokio::time::interval(refresh_period);
+            ticker.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Skip);
+            loop {
+                tokio::select! {
+                    _ = ticker.tick() => {}
+                    _ = cancel.cancelled() => break
+                }
+                let resp = match this
+                    .posthog_client
+                    .get_feature_flags_local_evaluation()
+                    .await
+                {
+                    Ok(resp) => resp,
+                    Err(e) => {
+                        tracing::warn!("Cannot get feature flags: {}", e);
+                        continue;
+                    }
+                };
+                let feature_store = FeatureStore::new_with_flags(resp.flags);
+                this.feature_store.store(Arc::new(feature_store));
+            }
+            tracing::info!("PostHog feature resolver stopped");
+        });
+    }
+
+    pub fn feature_store(&self) -> Arc<FeatureStore> {
+        self.feature_store.load_full()
+    }
+}

libs/posthog_client_lite/src/lib.rs

@@ -0,0 +1,636 @@
+//! A lite version of the PostHog client that only supports local evaluation of feature flags.
+
+mod background_loop;
+
+pub use background_loop::FeatureResolverBackgroundLoop;
+
+use std::collections::HashMap;
+
+use serde::{Deserialize, Serialize};
+use serde_json::json;
+use sha2::Digest;
+
+#[derive(Debug, thiserror::Error)]
+pub enum PostHogEvaluationError {
+    /// The feature flag is not available, for example, because the local evaluation data is not populated yet.
+    #[error("Feature flag not available: {0}")]
+    NotAvailable(String),
+    #[error("No condition group is matched")]
+    NoConditionGroupMatched,
+    /// Real errors, e.g., the rollout percentage does not add up to 100.
+    #[error("Failed to evaluate feature flag: {0}")]
+    Internal(String),
+}
+
+#[derive(Deserialize)]
+pub struct LocalEvaluationResponse {
+    pub flags: Vec<LocalEvaluationFlag>,
+}
+
+#[derive(Deserialize)]
+pub struct LocalEvaluationFlag {
+    key: String,
+    filters: LocalEvaluationFlagFilters,
+    active: bool,
+}
+
+#[derive(Deserialize)]
+pub struct LocalEvaluationFlagFilters {
+    groups: Vec<LocalEvaluationFlagFilterGroup>,
+    multivariate: LocalEvaluationFlagMultivariate,
+}
+
+#[derive(Deserialize)]
+pub struct LocalEvaluationFlagFilterGroup {
+    variant: Option<String>,
+    properties: Option<Vec<LocalEvaluationFlagFilterProperty>>,
+    rollout_percentage: i64,
+}
+
+#[derive(Deserialize)]
+pub struct LocalEvaluationFlagFilterProperty {
+    key: String,
+    value: PostHogFlagFilterPropertyValue,
+    operator: String,
+}
+
+#[derive(Debug, Serialize, Deserialize)]
+#[serde(untagged)]
+pub enum PostHogFlagFilterPropertyValue {
+    String(String),
+    Number(f64),
+    Boolean(bool),
+    List(Vec<String>),
+}
+
+#[derive(Deserialize)]
+pub struct LocalEvaluationFlagMultivariate {
+    variants: Vec<LocalEvaluationFlagMultivariateVariant>,
+}
+
+#[derive(Deserialize)]
+pub struct LocalEvaluationFlagMultivariateVariant {
+    key: String,
+    rollout_percentage: i64,
+}
+
+pub struct FeatureStore {
+    flags: HashMap<String, LocalEvaluationFlag>,
+}
+
+impl Default for FeatureStore {
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
+enum GroupEvaluationResult {
+    MatchedAndOverride(String),
+    MatchedAndEvaluate,
+    Unmatched,
+}
+
+impl FeatureStore {
+    pub fn new() -> Self {
+        Self {
+            flags: HashMap::new(),
+        }
+    }
+
+    pub fn new_with_flags(flags: Vec<LocalEvaluationFlag>) -> Self {
+        let mut store = Self::new();
+        store.set_flags(flags);
+        store
+    }
+
+    pub fn set_flags(&mut self, flags: Vec<LocalEvaluationFlag>) {
+        self.flags.clear();
+        for flag in flags {
+            self.flags.insert(flag.key.clone(), flag);
+        }
+    }
+
+    /// Generate a consistent hash for a user ID (e.g., tenant ID).
+    ///
+    /// The implementation is different from PostHog SDK. In PostHog SDK, it is sha1 of `user_id.distinct_id.salt`.
+    /// However, as we do not upload all of our tenant IDs to PostHog, we do not have the PostHog distinct_id for a
+    /// tenant. Therefore, the way we compute it is sha256 of `user_id.feature_id.salt`.
+    fn consistent_hash(user_id: &str, flag_key: &str, salt: &str) -> f64 {
+        let mut hasher = sha2::Sha256::new();
+        hasher.update(user_id);
+        hasher.update(".");
+        hasher.update(flag_key);
+        hasher.update(".");
+        hasher.update(salt);
+        let hash = hasher.finalize();
+        let hash_int = u64::from_le_bytes(hash[..8].try_into().unwrap());
+        hash_int as f64 / u64::MAX as f64
+    }
+
+    /// Evaluate a condition. Returns an error if the condition cannot be evaluated due to parsing error or missing
+    /// property.
+    fn evaluate_condition(
+        &self,
+        operator: &str,
+        provided: &PostHogFlagFilterPropertyValue,
+        requested: &PostHogFlagFilterPropertyValue,
+    ) -> Result<bool, PostHogEvaluationError> {
+        match operator {
+            "exact" => {
+                let PostHogFlagFilterPropertyValue::String(provided) = provided else {
+                    // Left should be a string
+                    return Err(PostHogEvaluationError::Internal(format!(
+                        "The left side of the condition is not a string: {:?}",
+                        provided
+                    )));
+                };
+                let PostHogFlagFilterPropertyValue::List(requested) = requested else {
+                    // Right should be a list of string
+                    return Err(PostHogEvaluationError::Internal(format!(
+                        "The right side of the condition is not a list: {:?}",
+                        requested
+                    )));
+                };
+                Ok(requested.contains(provided))
+            }
+            "lt" | "gt" => {
+                let PostHogFlagFilterPropertyValue::String(requested) = requested else {
+                    // Right should be a string
+                    return Err(PostHogEvaluationError::Internal(format!(
+                        "The right side of the condition is not a string: {:?}",
+                        requested
+                    )));
+                };
+                let Ok(requested) = requested.parse::<f64>() else {
+                    return Err(PostHogEvaluationError::Internal(format!(
+                        "Can not parse the right side of the condition as a number: {:?}",
+                        requested
+                    )));
+                };
+                // Left can either be a number or a string
+                let provided = match provided {
+                    PostHogFlagFilterPropertyValue::Number(provided) => *provided,
+                    PostHogFlagFilterPropertyValue::String(provided) => {
+                        let Ok(provided) = provided.parse::<f64>() else {
+                            return Err(PostHogEvaluationError::Internal(format!(
+                                "Can not parse the left side of the condition as a number: {:?}",
+                                provided
+                            )));
+                        };
+                        provided
+                    }
+                    _ => {
+                        return Err(PostHogEvaluationError::Internal(format!(
+                            "The left side of the condition is not a number or a string: {:?}",
+                            provided
+                        )));
+                    }
+                };
+                match operator {
+                    "lt" => Ok(provided < requested),
+                    "gt" => Ok(provided > requested),
+                    op => Err(PostHogEvaluationError::Internal(format!(
+                        "Unsupported operator: {}",
+                        op
+                    ))),
+                }
+            }
+            _ => Err(PostHogEvaluationError::Internal(format!(
+                "Unsupported operator: {}",
+                operator
+            ))),
+        }
+    }
+
+    /// Evaluate a percentage.
+    fn evaluate_percentage(&self, mapped_user_id: f64, percentage: i64) -> bool {
+        mapped_user_id <= percentage as f64 / 100.0
+    }
+
+    /// Evaluate a filter group for a feature flag. Returns an error if there are errors during the evaluation.
+    ///
+    /// Return values:
+    /// Ok(GroupEvaluationResult::MatchedAndOverride(variant)): matched and evaluated to this value
+    /// Ok(GroupEvaluationResult::MatchedAndEvaluate): condition matched but no variant override, use the global rollout percentage
+    /// Ok(GroupEvaluationResult::Unmatched): condition unmatched
+    fn evaluate_group(
+        &self,
+        group: &LocalEvaluationFlagFilterGroup,
+        hash_on_group_rollout_percentage: f64,
+        provided_properties: &HashMap<String, PostHogFlagFilterPropertyValue>,
+    ) -> Result<GroupEvaluationResult, PostHogEvaluationError> {
+        if let Some(ref properties) = group.properties {
+            for property in properties {
+                if let Some(value) = provided_properties.get(&property.key) {
+                    // The user provided the property value
+                    if !self.evaluate_condition(
+                        property.operator.as_ref(),
+                        value,
+                        &property.value,
+                    )? {
+                        return Ok(GroupEvaluationResult::Unmatched);
+                    }
+                } else {
+                    // We cannot evaluate, the property is not available
+                    return Err(PostHogEvaluationError::NotAvailable(format!(
+                        "The required property in the condition is not available: {}",
+                        property.key
+                    )));
+                }
+            }
+        }
+
+        // The group has no condition matchers or we matched the properties
+        if self.evaluate_percentage(hash_on_group_rollout_percentage, group.rollout_percentage) {
+            if let Some(ref variant_override) = group.variant {
+                Ok(GroupEvaluationResult::MatchedAndOverride(
+                    variant_override.clone(),
+                ))
+            } else {
+                Ok(GroupEvaluationResult::MatchedAndEvaluate)
+            }
+        } else {
+            Ok(GroupEvaluationResult::Unmatched)
+        }
+    }
+
+    /// Evaluate a multivariate feature flag. Returns `None` if the flag is not available or if there are errors
+    /// during the evaluation.
+    ///
+    /// The parsing logic is as follows:
+    ///
+    /// * Match each filter group.
+    ///   - If a group is matched, it will first determine whether the user is in the range of the group's rollout
+    ///     percentage. We will generate a consistent hash for the user ID on the group rollout percentage. This hash
+    ///     is shared across all groups.
+    ///   - If the hash falls within the group's rollout percentage, return the variant if it's overridden, or
+    ///   - Evaluate the variant using the global config and the global rollout percentage.
+    /// * Otherwise, continue with the next group until all groups are evaluated and no group is within the
+    ///   rollout percentage.
+    /// * If there are no matching groups, return an error.
+    ///
+    /// Example: we have a multivariate flag with 3 groups of the configured global rollout percentage: A (10%), B (20%), C (70%).
+    /// There is a single group with a condition that has a rollout percentage of 10% and it does not have a variant override.
+    /// Then, we will have 1% of the users evaluated to A, 2% to B, and 7% to C.
+    pub fn evaluate_multivariate(
+        &self,
+        flag_key: &str,
+        user_id: &str,
+        properties: &HashMap<String, PostHogFlagFilterPropertyValue>,
+    ) -> Result<String, PostHogEvaluationError> {
+        let hash_on_global_rollout_percentage =
+            Self::consistent_hash(user_id, flag_key, "multivariate");
+        let hash_on_group_rollout_percentage =
+            Self::consistent_hash(user_id, flag_key, "within_group");
+        self.evaluate_multivariate_inner(
+            flag_key,
+            hash_on_global_rollout_percentage,
+            hash_on_group_rollout_percentage,
+            properties,
+        )
+    }
+
+    /// Evaluate a multivariate feature flag. Note that we directly take the mapped user ID
+    /// (a consistent hash ranging from 0 to 1) so that it is easier to use it in the tests
+    /// and avoid duplicate computations.
+    ///
+    /// Use a different consistent hash for evaluating the group rollout percentage.
+    /// The behavior: if the condition is set to rolling out to 10% of the users, and
+    /// we set the variant A to 20% in the global config, then 2% of the total users will
+    /// be evaluated to variant A.
+    ///
+    /// Note that the hash to determine group rollout percentage is shared across all groups. So if we have two
+    /// exactly-the-same conditions with 10% and 20% rollout percentage respectively, a total of 20% of the users
+    /// will be evaluated (versus 30% if group evaluation is done independently).
+    pub(crate) fn evaluate_multivariate_inner(
+        &self,
+        flag_key: &str,
+        hash_on_global_rollout_percentage: f64,
+        hash_on_group_rollout_percentage: f64,
+        properties: &HashMap<String, PostHogFlagFilterPropertyValue>,
+    ) -> Result<String, PostHogEvaluationError> {
+        if let Some(flag_config) = self.flags.get(flag_key) {
+            if !flag_config.active {
+                return Err(PostHogEvaluationError::NotAvailable(format!(
+                    "The feature flag is not active: {}",
+                    flag_key
+                )));
+            }
+            // TODO: sort the groups so that variant overrides always get evaluated first and it follows the PostHog
+            // Python SDK behavior; for now we do not configure conditions without variant overrides in Neon so it
+            // does not matter.
+            for group in &flag_config.filters.groups {
+                match self.evaluate_group(group, hash_on_group_rollout_percentage, properties)? {
+                    GroupEvaluationResult::MatchedAndOverride(variant) => return Ok(variant),
+                    GroupEvaluationResult::MatchedAndEvaluate => {
+                        let mut percentage = 0;
+                        for variant in &flag_config.filters.multivariate.variants {
+                            percentage += variant.rollout_percentage;
+                            if self
+                                .evaluate_percentage(hash_on_global_rollout_percentage, percentage)
+                            {
+                                return Ok(variant.key.clone());
+                            }
+                        }
+                        // This should not happen because the rollout percentage always adds up to 100, but just in case that PostHog
+                        // returned invalid spec, we return an error.
+                        return Err(PostHogEvaluationError::Internal(format!(
+                            "Rollout percentage does not add up to 100: {}",
+                            flag_key
+                        )));
+                    }
+                    GroupEvaluationResult::Unmatched => continue,
+                }
+            }
+            // If no group is matched, the feature is not available, and up to the caller to decide what to do.
+            Err(PostHogEvaluationError::NoConditionGroupMatched)
+        } else {
+            // The feature flag is not available yet
+            Err(PostHogEvaluationError::NotAvailable(format!(
+                "Not found in the local evaluation spec: {}",
+                flag_key
+            )))
+        }
+    }
+}
+
+pub struct PostHogClientConfig {
+    /// The server API key.
+    pub server_api_key: String,
+    /// The client API key.
+    pub client_api_key: String,
+    /// The project ID.
+    pub project_id: String,
+    /// The private API URL.
+    pub private_api_url: String,
+    /// The public API URL.
+    pub public_api_url: String,
+}
+
+/// A lite PostHog client.
+///
+/// At the point of writing this code, PostHog does not have a functional Rust client with feature flag support.
+/// This is a lite version that only supports local evaluation of feature flags and only supports those JSON specs
+/// that will be used within Neon.
+///
+/// PostHog is designed as a browser-server system: the browser (client) side uses the client key and is exposed
+/// to the end users; the server side uses a server key and is not exposed to the end users. The client and the
+/// server has different API keys and provide a different set of APIs. In Neon, we only have the server (that is
+/// pageserver), and it will use both the client API and the server API. So we need to store two API keys within
+/// our PostHog client.
+///
+/// The server API is used to fetch the feature flag specs. The client API is used to capture events in case we
+/// want to report the feature flag usage back to PostHog. The current plan is to use PostHog only as an UI to
+/// configure feature flags so it is very likely that the client API will not be used.
+pub struct PostHogClient {
+    /// The config.
+    config: PostHogClientConfig,
+    /// The HTTP client.
+    client: reqwest::Client,
+}
+
+impl PostHogClient {
+    pub fn new(config: PostHogClientConfig) -> Self {
+        let client = reqwest::Client::new();
+        Self { config, client }
+    }
+
+    pub fn new_with_us_region(
+        server_api_key: String,
+        client_api_key: String,
+        project_id: String,
+    ) -> Self {
+        Self::new(PostHogClientConfig {
+            server_api_key,
+            client_api_key,
+            project_id,
+            private_api_url: "https://us.posthog.com".to_string(),
+            public_api_url: "https://us.i.posthog.com".to_string(),
+        })
+    }
+
+    /// Fetch the feature flag specs from the server.
+    ///
+    /// This is unfortunately an undocumented API at:
+    /// - <https://posthog.com/docs/api/feature-flags#get-api-projects-project_id-feature_flags-local_evaluation>
+    /// - <https://posthog.com/docs/feature-flags/local-evaluation>
+    ///
+    /// The handling logic in [`FeatureStore`] mostly follows the Python API implementation.
+    /// See `_compute_flag_locally` in <https://github.com/PostHog/posthog-python/blob/master/posthog/client.py>
+    pub async fn get_feature_flags_local_evaluation(
+        &self,
+    ) -> anyhow::Result<LocalEvaluationResponse> {
+        // BASE_URL/api/projects/:project_id/feature_flags/local_evaluation
+        // with bearer token of self.server_api_key
+        let url = format!(
+            "{}/api/projects/{}/feature_flags/local_evaluation",
+            self.config.private_api_url, self.config.project_id
+        );
+        let response = self
+            .client
+            .get(url)
+            .bearer_auth(&self.config.server_api_key)
+            .send()
+            .await?;
+        let body = response.text().await?;
+        Ok(serde_json::from_str(&body)?)
+    }
+
+    /// Capture an event. This will only be used to report the feature flag usage back to PostHog, though
+    /// it also support a lot of other functionalities.
+    ///
+    /// <https://posthog.com/docs/api/capture>
+    pub async fn capture_event(
+        &self,
+        event: &str,
+        distinct_id: &str,
+        properties: &HashMap<String, PostHogFlagFilterPropertyValue>,
+    ) -> anyhow::Result<()> {
+        // PUBLIC_URL/capture/
+        // with bearer token of self.client_api_key
+        let url = format!("{}/capture/", self.config.public_api_url);
+        self.client
+            .post(url)
+            .body(serde_json::to_string(&json!({
+                "api_key": self.config.client_api_key,
+                "distinct_id": distinct_id,
+                "event": event,
+                "properties": properties,
+            }))?)
+            .send()
+            .await?;
+        Ok(())
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn data() -> &'static str {
+        r#"{
+            "flags": [
+                {
+                    "id": 132794,
+                    "team_id": 152860,
+                    "name": "",
+                    "key": "gc-compaction",
+                    "filters": {
+                        "groups": [
+                            {
+                                "variant": "enabled-stage-2",
+                                "properties": [
+                                    {
+                                        "key": "plan_type",
+                                        "type": "person",
+                                        "value": [
+                                            "free"
+                                        ],
+                                        "operator": "exact"
+                                    },
+                                    {
+                                        "key": "pageserver_remote_size",
+                                        "type": "person",
+                                        "value": "10000000",
+                                        "operator": "lt"
+                                    }
+                                ],
+                                "rollout_percentage": 50
+                            },
+                            {
+                                "properties": [
+                                    {
+                                        "key": "plan_type",
+                                        "type": "person",
+                                        "value": [
+                                            "free"
+                                        ],
+                                        "operator": "exact"
+                                    },
+                                    {
+                                        "key": "pageserver_remote_size",
+                                        "type": "person",
+                                        "value": "10000000",
+                                        "operator": "lt"
+                                    }
+                                ],
+                                "rollout_percentage": 80
+                            }
+                        ],
+                        "payloads": {},
+                        "multivariate": {
+                            "variants": [
+                                {
+                                    "key": "disabled",
+                                    "name": "",
+                                    "rollout_percentage": 90
+                                },
+                                {
+                                    "key": "enabled-stage-1",
+                                    "name": "",
+                                    "rollout_percentage": 10
+                                },
+                                {
+                                    "key": "enabled-stage-2",
+                                    "name": "",
+                                    "rollout_percentage": 0
+                                },
+                                {
+                                    "key": "enabled-stage-3",
+                                    "name": "",
+                                    "rollout_percentage": 0
+                                },
+                                {
+                                    "key": "enabled",
+                                    "name": "",
+                                    "rollout_percentage": 0
+                                }
+                            ]
+                        }
+                    },
+                    "deleted": false,
+                    "active": true,
+                    "ensure_experience_continuity": false,
+                    "has_encrypted_payloads": false,
+                    "version": 6
+                }
+            ],
+            "group_type_mapping": {},
+            "cohorts": {}
+        }"#
+    }
+
+    #[test]
+    fn parse_local_evaluation() {
+        let data = data();
+        let _: LocalEvaluationResponse = serde_json::from_str(data).unwrap();
+    }
+
+    #[test]
+    fn evaluate_multivariate() {
+        let mut store = FeatureStore::new();
+        let response: LocalEvaluationResponse = serde_json::from_str(data()).unwrap();
+        store.set_flags(response.flags);
+
+        // This lacks the required properties and cannot be evaluated.
+        let variant =
+            store.evaluate_multivariate_inner("gc-compaction", 1.00, 0.40, &HashMap::new());
+        assert!(matches!(
+            variant,
+            Err(PostHogEvaluationError::NotAvailable(_))
+        ),);
+
+        let properties_unmatched = HashMap::from([
+            (
+                "plan_type".to_string(),
+                PostHogFlagFilterPropertyValue::String("paid".to_string()),
+            ),
+            (
+                "pageserver_remote_size".to_string(),
+                PostHogFlagFilterPropertyValue::Number(1000.0),
+            ),
+        ]);
+
+        // This does not match any group so there will be an error.
+        let variant =
+            store.evaluate_multivariate_inner("gc-compaction", 1.00, 0.40, &properties_unmatched);
+        assert!(matches!(
+            variant,
+            Err(PostHogEvaluationError::NoConditionGroupMatched)
+        ),);
+        let variant =
+            store.evaluate_multivariate_inner("gc-compaction", 0.80, 0.80, &properties_unmatched);
+        assert!(matches!(
+            variant,
+            Err(PostHogEvaluationError::NoConditionGroupMatched)
+        ),);
+
+        let properties = HashMap::from([
+            (
+                "plan_type".to_string(),
+                PostHogFlagFilterPropertyValue::String("free".to_string()),
+            ),
+            (
+                "pageserver_remote_size".to_string(),
+                PostHogFlagFilterPropertyValue::Number(1000.0),
+            ),
+        ]);
+
+        // It matches the first group as 0.10 <= 0.50 and the properties are matched. Then it gets evaluated to the variant override.
+        let variant = store.evaluate_multivariate_inner("gc-compaction", 0.10, 0.10, &properties);
+        assert_eq!(variant.unwrap(), "enabled-stage-2".to_string());
+
+        // It matches the second group as 0.50 <= 0.60 <= 0.80 and the properties are matched. Then it gets evaluated using the global percentage.
+        let variant = store.evaluate_multivariate_inner("gc-compaction", 0.99, 0.60, &properties);
+        assert_eq!(variant.unwrap(), "enabled-stage-1".to_string());
+        let variant = store.evaluate_multivariate_inner("gc-compaction", 0.80, 0.60, &properties);
+        assert_eq!(variant.unwrap(), "disabled".to_string());
+
+        // It matches the group conditions but not the group rollout percentage.
+        let variant = store.evaluate_multivariate_inner("gc-compaction", 1.00, 0.90, &properties);
+        assert!(matches!(
+            variant,
+            Err(PostHogEvaluationError::NoConditionGroupMatched)
+        ),);
+    }
+}

libs/proxy/postgres-protocol2/src/message/frontend.rs

@@ -25,6 +25,7 @@ where
     Ok(())
 }
 
+#[derive(Debug)]
 pub enum BindError {
     Conversion(Box<dyn Error + marker::Sync + Send>),
     Serialization(io::Error),
@@ -288,6 +289,12 @@ pub fn sync(buf: &mut BytesMut) {
     write_body(buf, |_| Ok::<(), io::Error>(())).unwrap();
 }
 
+#[inline]
+pub fn flush(buf: &mut BytesMut) {
+    buf.put_u8(b'H');
+    write_body(buf, |_| Ok::<(), io::Error>(())).unwrap();
+}
+
 #[inline]
 pub fn terminate(buf: &mut BytesMut) {
     buf.put_u8(b'X');

libs/proxy/postgres-types2/src/lib.rs

@@ -9,7 +9,6 @@ use std::error::Error;
 use std::fmt;
 use std::sync::Arc;
 
-use bytes::BytesMut;
 use fallible_iterator::FallibleIterator;
 #[doc(inline)]
 pub use postgres_protocol2::Oid;
@@ -27,41 +26,6 @@ macro_rules! accepts {
     )
 }
 
-/// Generates an implementation of `ToSql::to_sql_checked`.
-///
-/// All `ToSql` implementations should use this macro.
-macro_rules! to_sql_checked {
-    () => {
-        fn to_sql_checked(
-            &self,
-            ty: &$crate::Type,
-            out: &mut $crate::private::BytesMut,
-        ) -> ::std::result::Result<
-            $crate::IsNull,
-            Box<dyn ::std::error::Error + ::std::marker::Sync + ::std::marker::Send>,
-        > {
-            $crate::__to_sql_checked(self, ty, out)
-        }
-    };
-}
-
-// WARNING: this function is not considered part of this crate's public API.
-// It is subject to change at any time.
-#[doc(hidden)]
-pub fn __to_sql_checked<T>(
-    v: &T,
-    ty: &Type,
-    out: &mut BytesMut,
-) -> Result<IsNull, Box<dyn Error + Sync + Send>>
-where
-    T: ToSql,
-{
-    if !T::accepts(ty) {
-        return Err(Box::new(WrongType::new::<T>(ty.clone())));
-    }
-    v.to_sql(ty, out)
-}
-
 // mod pg_lsn;
 #[doc(hidden)]
 pub mod private;
@@ -142,7 +106,7 @@ pub enum Kind {
     /// An array type along with the type of its elements.
     Array(Type),
     /// A range type along with the type of its elements.
-    Range(Type),
+    Range(Oid),
     /// A multirange type along with the type of its elements.
     Multirange(Type),
     /// A domain type along with its underlying type.
@@ -377,43 +341,6 @@ pub enum IsNull {
     No,
 }
 
-/// A trait for types that can be converted into Postgres values.
-pub trait ToSql: fmt::Debug {
-    /// Converts the value of `self` into the binary format of the specified
-    /// Postgres `Type`, appending it to `out`.
-    ///
-    /// The caller of this method is responsible for ensuring that this type
-    /// is compatible with the Postgres `Type`.
-    ///
-    /// The return value indicates if this value should be represented as
-    /// `NULL`. If this is the case, implementations **must not** write
-    /// anything to `out`.
-    fn to_sql(&self, ty: &Type, out: &mut BytesMut) -> Result<IsNull, Box<dyn Error + Sync + Send>>
-    where
-        Self: Sized;
-
-    /// Determines if a value of this type can be converted to the specified
-    /// Postgres `Type`.
-    fn accepts(ty: &Type) -> bool
-    where
-        Self: Sized;
-
-    /// An adaptor method used internally by Rust-Postgres.
-    ///
-    /// *All* implementations of this method should be generated by the
-    /// `to_sql_checked!()` macro.
-    fn to_sql_checked(
-        &self,
-        ty: &Type,
-        out: &mut BytesMut,
-    ) -> Result<IsNull, Box<dyn Error + Sync + Send>>;
-
-    /// Specify the encode format
-    fn encode_format(&self, _ty: &Type) -> Format {
-        Format::Binary
-    }
-}
-
 /// Supported Postgres message format types
 ///
 /// Using Text format in a message assumes a Postgres `SERVER_ENCODING` of `UTF8`
@@ -424,52 +351,3 @@ pub enum Format {
     /// Compact, typed binary format
     Binary,
 }
-
-impl ToSql for &str {
-    fn to_sql(&self, ty: &Type, w: &mut BytesMut) -> Result<IsNull, Box<dyn Error + Sync + Send>> {
-        match *ty {
-            ref ty if ty.name() == "ltree" => types::ltree_to_sql(self, w),
-            ref ty if ty.name() == "lquery" => types::lquery_to_sql(self, w),
-            ref ty if ty.name() == "ltxtquery" => types::ltxtquery_to_sql(self, w),
-            _ => types::text_to_sql(self, w),
-        }
-        Ok(IsNull::No)
-    }
-
-    fn accepts(ty: &Type) -> bool {
-        match *ty {
-            Type::VARCHAR | Type::TEXT | Type::BPCHAR | Type::NAME | Type::UNKNOWN => true,
-            ref ty
-                if (ty.name() == "citext"
-                    || ty.name() == "ltree"
-                    || ty.name() == "lquery"
-                    || ty.name() == "ltxtquery") =>
-            {
-                true
-            }
-            _ => false,
-        }
-    }
-
-    to_sql_checked!();
-}
-
-macro_rules! simple_to {
-    ($t:ty, $f:ident, $($expected:ident),+) => {
-        impl ToSql for $t {
-            fn to_sql(&self,
-                      _: &Type,
-                      w: &mut BytesMut)
-                      -> Result<IsNull, Box<dyn Error + Sync + Send>> {
-                types::$f(*self, w);
-                Ok(IsNull::No)
-            }
-
-            accepts!($($expected),+);
-
-            to_sql_checked!();
-        }
-    }
-}
-
-simple_to!(u32, oid_to_sql, OID);

libs/proxy/postgres-types2/src/type_gen.rs

@@ -393,7 +393,7 @@ impl Inner {
         }
     }
 
-    pub fn oid(&self) -> Oid {
+    pub const fn const_oid(&self) -> Oid {
         match *self {
             Inner::Bool => 16,
             Inner::Bytea => 17,
@@ -580,7 +580,14 @@ impl Inner {
             Inner::TstzmultiRangeArray => 6153,
             Inner::DatemultiRangeArray => 6155,
             Inner::Int8multiRangeArray => 6157,
+            Inner::Other(_) => u32::MAX,
+        }
+    }
+
+    pub fn oid(&self) -> Oid {
+        match *self {
             Inner::Other(ref u) => u.oid,
+            _ => self.const_oid(),
         }
     }
 
@@ -727,17 +734,17 @@ impl Inner {
             Inner::JsonbArray => &Kind::Array(Type(Inner::Jsonb)),
             Inner::AnyRange => &Kind::Pseudo,
             Inner::EventTrigger => &Kind::Pseudo,
-            Inner::Int4Range => &Kind::Range(Type(Inner::Int4)),
+            Inner::Int4Range => &const { Kind::Range(Inner::Int4.const_oid()) },
             Inner::Int4RangeArray => &Kind::Array(Type(Inner::Int4Range)),
-            Inner::NumRange => &Kind::Range(Type(Inner::Numeric)),
+            Inner::NumRange => &const { Kind::Range(Inner::Numeric.const_oid()) },
             Inner::NumRangeArray => &Kind::Array(Type(Inner::NumRange)),
-            Inner::TsRange => &Kind::Range(Type(Inner::Timestamp)),
+            Inner::TsRange => &const { Kind::Range(Inner::Timestamp.const_oid()) },
             Inner::TsRangeArray => &Kind::Array(Type(Inner::TsRange)),
-            Inner::TstzRange => &Kind::Range(Type(Inner::Timestamptz)),
+            Inner::TstzRange => &const { Kind::Range(Inner::Timestamptz.const_oid()) },
             Inner::TstzRangeArray => &Kind::Array(Type(Inner::TstzRange)),
-            Inner::DateRange => &Kind::Range(Type(Inner::Date)),
+            Inner::DateRange => &const { Kind::Range(Inner::Date.const_oid()) },
             Inner::DateRangeArray => &Kind::Array(Type(Inner::DateRange)),
-            Inner::Int8Range => &Kind::Range(Type(Inner::Int8)),
+            Inner::Int8Range => &const { Kind::Range(Inner::Int8.const_oid()) },
             Inner::Int8RangeArray => &Kind::Array(Type(Inner::Int8Range)),
             Inner::Jsonpath => &Kind::Simple,
             Inner::JsonpathArray => &Kind::Array(Type(Inner::Jsonpath)),

libs/proxy/tokio-postgres2/src/client.rs

@@ -1,14 +1,12 @@
 use std::collections::HashMap;
 use std::fmt;
 use std::net::IpAddr;
-use std::sync::Arc;
 use std::task::{Context, Poll};
 use std::time::Duration;
 
 use bytes::BytesMut;
 use fallible_iterator::FallibleIterator;
 use futures_util::{TryStreamExt, future, ready};
-use parking_lot::Mutex;
 use postgres_protocol2::message::backend::Message;
 use postgres_protocol2::message::frontend;
 use serde::{Deserialize, Serialize};
@@ -16,29 +14,52 @@ use tokio::sync::mpsc;
 
 use crate::codec::{BackendMessages, FrontendMessage};
 use crate::config::{Host, SslMode};
-use crate::connection::{Request, RequestMessages};
 use crate::query::RowStream;
 use crate::simple_query::SimpleQueryStream;
 use crate::types::{Oid, Type};
 use crate::{
-    CancelToken, Error, ReadyForQueryStatus, SimpleQueryMessage, Statement, Transaction,
-    TransactionBuilder, query, simple_query,
+    CancelToken, Error, ReadyForQueryStatus, SimpleQueryMessage, Transaction, TransactionBuilder,
+    query, simple_query,
 };
 
 pub struct Responses {
+    /// new messages from conn
     receiver: mpsc::Receiver<BackendMessages>,
+    /// current batch of messages
     cur: BackendMessages,
+    /// number of total queries sent.
+    waiting: usize,
+    /// number of ReadyForQuery messages received.
+    received: usize,
 }
 
 impl Responses {
     pub fn poll_next(&mut self, cx: &mut Context<'_>) -> Poll<Result<Message, Error>> {
         loop {
-            match self.cur.next().map_err(Error::parse)? {
-                Some(Message::ErrorResponse(body)) => return Poll::Ready(Err(Error::db(body))),
-                Some(message) => return Poll::Ready(Ok(message)),
-                None => {}
+            // get the next saved message
+            if let Some(message) = self.cur.next().map_err(Error::parse)? {
+                let received = self.received;
+
+                // increase the query head if this is the last message.
+                if let Message::ReadyForQuery(_) = message {
+                    self.received += 1;
+                }
+
+                // check if the client has skipped this query.
+                if received + 1 < self.waiting {
+                    // grab the next message.
+                    continue;
+                }
+
+                // convenience: turn the error messaage into a proper error.
+                let res = match message {
+                    Message::ErrorResponse(body) => Err(Error::db(body)),
+                    message => Ok(message),
+                };
+                return Poll::Ready(res);
             }
 
+            // get the next batch of messages.
             match ready!(self.receiver.poll_recv(cx)) {
                 Some(messages) => self.cur = messages,
                 None => return Poll::Ready(Err(Error::closed())),
@@ -55,44 +76,87 @@ impl Responses {
 /// (corresponding to the queries in the [crate::prepare] module).
 #[derive(Default)]
 pub(crate) struct CachedTypeInfo {
-    /// A statement for basic information for a type from its
-    /// OID. Corresponds to [TYPEINFO_QUERY](crate::prepare::TYPEINFO_QUERY) (or its
-    /// fallback).
-    pub(crate) typeinfo: Option<Statement>,
-
     /// Cache of types already looked up.
     pub(crate) types: HashMap<Oid, Type>,
 }
 
 pub struct InnerClient {
-    sender: mpsc::UnboundedSender<Request>,
+    sender: mpsc::UnboundedSender<FrontendMessage>,
+    responses: Responses,
 
     /// A buffer to use when writing out postgres commands.
-    buffer: Mutex<BytesMut>,
+    buffer: BytesMut,
 }
 
 impl InnerClient {
-    pub fn send(&self, messages: RequestMessages) -> Result<Responses, Error> {
-        let (sender, receiver) = mpsc::channel(1);
-        let request = Request { messages, sender };
-        self.sender.send(request).map_err(|_| Error::closed())?;
-
-        Ok(Responses {
-            receiver,
-            cur: BackendMessages::empty(),
-        })
+    pub fn start(&mut self) -> Result<PartialQuery, Error> {
+        self.responses.waiting += 1;
+        Ok(PartialQuery(Some(self)))
     }
 
-    /// Call the given function with a buffer to be used when writing out
-    /// postgres commands.
-    pub fn with_buf<F, R>(&self, f: F) -> R
+    // pub fn send_with_sync<F>(&mut self, f: F) -> Result<&mut Responses, Error>
+    // where
+    //     F: FnOnce(&mut BytesMut) -> Result<(), Error>,
+    // {
+    //     self.start()?.send_with_sync(f)
+    // }
+
+    pub fn send_simple_query(&mut self, query: &str) -> Result<&mut Responses, Error> {
+        self.responses.waiting += 1;
+
+        self.buffer.clear();
+        // simple queries do not need sync.
+        frontend::query(query, &mut self.buffer).map_err(Error::encode)?;
+        let buf = self.buffer.split().freeze();
+        self.send_message(FrontendMessage::Raw(buf))
+    }
+
+    fn send_message(&mut self, messages: FrontendMessage) -> Result<&mut Responses, Error> {
+        self.sender.send(messages).map_err(|_| Error::closed())?;
+        Ok(&mut self.responses)
+    }
+}
+
+pub struct PartialQuery<'a>(Option<&'a mut InnerClient>);
+
+impl Drop for PartialQuery<'_> {
+    fn drop(&mut self) {
+        if let Some(client) = self.0.take() {
+            client.buffer.clear();
+            frontend::sync(&mut client.buffer);
+            let buf = client.buffer.split().freeze();
+            let _ = client.send_message(FrontendMessage::Raw(buf));
+        }
+    }
+}
+
+impl<'a> PartialQuery<'a> {
+    pub fn send_with_flush<F>(&mut self, f: F) -> Result<&mut Responses, Error>
     where
-        F: FnOnce(&mut BytesMut) -> R,
+        F: FnOnce(&mut BytesMut) -> Result<(), Error>,
     {
-        let mut buffer = self.buffer.lock();
-        let r = f(&mut buffer);
-        buffer.clear();
-        r
+        let client = self.0.as_deref_mut().unwrap();
+
+        client.buffer.clear();
+        f(&mut client.buffer)?;
+        frontend::flush(&mut client.buffer);
+        let buf = client.buffer.split().freeze();
+        client.send_message(FrontendMessage::Raw(buf))
+    }
+
+    pub fn send_with_sync<F>(mut self, f: F) -> Result<&'a mut Responses, Error>
+    where
+        F: FnOnce(&mut BytesMut) -> Result<(), Error>,
+    {
+        let client = self.0.as_deref_mut().unwrap();
+
+        client.buffer.clear();
+        f(&mut client.buffer)?;
+        frontend::sync(&mut client.buffer);
+        let buf = client.buffer.split().freeze();
+        let _ = client.send_message(FrontendMessage::Raw(buf));
+
+        Ok(&mut self.0.take().unwrap().responses)
     }
 }
 
@@ -109,7 +173,7 @@ pub struct SocketConfig {
 /// The client is one half of what is returned when a connection is established. Users interact with the database
 /// through this client object.
 pub struct Client {
-    inner: Arc<InnerClient>,
+    inner: InnerClient,
     cached_typeinfo: CachedTypeInfo,
 
     socket_config: SocketConfig,
@@ -120,17 +184,24 @@ pub struct Client {
 
 impl Client {
     pub(crate) fn new(
-        sender: mpsc::UnboundedSender<Request>,
+        sender: mpsc::UnboundedSender<FrontendMessage>,
+        receiver: mpsc::Receiver<BackendMessages>,
         socket_config: SocketConfig,
         ssl_mode: SslMode,
         process_id: i32,
         secret_key: i32,
     ) -> Client {
         Client {
-            inner: Arc::new(InnerClient {
+            inner: InnerClient {
                 sender,
+                responses: Responses {
+                    receiver,
+                    cur: BackendMessages::empty(),
+                    waiting: 0,
+                    received: 0,
+                },
                 buffer: Default::default(),
-            }),
+            },
             cached_typeinfo: Default::default(),
 
             socket_config,
@@ -145,19 +216,29 @@ impl Client {
         self.process_id
     }
 
-    pub(crate) fn inner(&self) -> &Arc<InnerClient> {
-        &self.inner
+    pub(crate) fn inner_mut(&mut self) -> &mut InnerClient {
+        &mut self.inner
     }
 
     /// Pass text directly to the Postgres backend to allow it to sort out typing itself and
     /// to save a roundtrip
-    pub async fn query_raw_txt<S, I>(&self, statement: &str, params: I) -> Result<RowStream, Error>
+    pub async fn query_raw_txt<S, I>(
+        &mut self,
+        statement: &str,
+        params: I,
+    ) -> Result<RowStream, Error>
     where
         S: AsRef<str>,
         I: IntoIterator<Item = Option<S>>,
         I::IntoIter: ExactSizeIterator,
     {
-        query::query_txt(&self.inner, statement, params).await
+        query::query_txt(
+            &mut self.inner,
+            &mut self.cached_typeinfo,
+            statement,
+            params,
+        )
+        .await
     }
 
     /// Executes a sequence of SQL statements using the simple query protocol, returning the resulting rows.
@@ -173,12 +254,15 @@ impl Client {
     /// Prepared statements should be use for any query which contains user-specified data, as they provided the
     /// functionality to safely embed that data in the request. Do not form statements via string concatenation and pass
     /// them to this method!
-    pub async fn simple_query(&self, query: &str) -> Result<Vec<SimpleQueryMessage>, Error> {
+    pub async fn simple_query(&mut self, query: &str) -> Result<Vec<SimpleQueryMessage>, Error> {
         self.simple_query_raw(query).await?.try_collect().await
     }
 
-    pub(crate) async fn simple_query_raw(&self, query: &str) -> Result<SimpleQueryStream, Error> {
-        simple_query::simple_query(self.inner(), query).await
+    pub(crate) async fn simple_query_raw(
+        &mut self,
+        query: &str,
+    ) -> Result<SimpleQueryStream, Error> {
+        simple_query::simple_query(self.inner_mut(), query).await
     }
 
     /// Executes a sequence of SQL statements using the simple query protocol.
@@ -191,15 +275,11 @@ impl Client {
     /// Prepared statements should be use for any query which contains user-specified data, as they provided the
     /// functionality to safely embed that data in the request. Do not form statements via string concatenation and pass
     /// them to this method!
-    pub async fn batch_execute(&self, query: &str) -> Result<ReadyForQueryStatus, Error> {
-        simple_query::batch_execute(self.inner(), query).await
+    pub async fn batch_execute(&mut self, query: &str) -> Result<ReadyForQueryStatus, Error> {
+        simple_query::batch_execute(self.inner_mut(), query).await
     }
 
     pub async fn discard_all(&mut self) -> Result<ReadyForQueryStatus, Error> {
-        // clear the prepared statements that are about to be nuked from the postgres session
-
-        self.cached_typeinfo.typeinfo = None;
-
         self.batch_execute("discard all").await
     }
 
@@ -208,7 +288,7 @@ impl Client {
     /// The transaction will roll back by default - use the `commit` method to commit it.
     pub async fn transaction(&mut self) -> Result<Transaction<'_>, Error> {
         struct RollbackIfNotDone<'me> {
-            client: &'me Client,
+            client: &'me mut Client,
             done: bool,
         }
 
@@ -218,14 +298,7 @@ impl Client {
                     return;
                 }
 
-                let buf = self.client.inner().with_buf(|buf| {
-                    frontend::query("ROLLBACK", buf).unwrap();
-                    buf.split().freeze()
-                });
-                let _ = self
-                    .client
-                    .inner()
-                    .send(RequestMessages::Single(FrontendMessage::Raw(buf)));
+                let _ = self.client.inner.send_simple_query("ROLLBACK");
             }
         }
 
@@ -239,7 +312,7 @@ impl Client {
                 client: self,
                 done: false,
             };
-            self.batch_execute("BEGIN").await?;
+            cleaner.client.batch_execute("BEGIN").await?;
             cleaner.done = true;
         }
 
@@ -265,11 +338,6 @@ impl Client {
         }
     }
 
-    /// Query for type information
-    pub(crate) async fn get_type_inner(&mut self, oid: Oid) -> Result<Type, Error> {
-        crate::prepare::get_type(&self.inner, &mut self.cached_typeinfo, oid).await
-    }
-
     /// Determines if the connection to the server has already closed.
     ///
     /// In that case, all future queries will fail.

libs/proxy/tokio-postgres2/src/codec.rs

@@ -1,21 +1,16 @@
 use std::io;
 
-use bytes::{Buf, Bytes, BytesMut};
+use bytes::{Bytes, BytesMut};
 use fallible_iterator::FallibleIterator;
 use postgres_protocol2::message::backend;
-use postgres_protocol2::message::frontend::CopyData;
 use tokio_util::codec::{Decoder, Encoder};
 
 pub enum FrontendMessage {
     Raw(Bytes),
-    CopyData(CopyData<Box<dyn Buf + Send>>),
 }
 
 pub enum BackendMessage {
-    Normal {
-        messages: BackendMessages,
-        request_complete: bool,
-    },
+    Normal { messages: BackendMessages },
     Async(backend::Message),
 }
 
@@ -44,7 +39,6 @@ impl Encoder<FrontendMessage> for PostgresCodec {
     fn encode(&mut self, item: FrontendMessage, dst: &mut BytesMut) -> io::Result<()> {
         match item {
             FrontendMessage::Raw(buf) => dst.extend_from_slice(&buf),
-            FrontendMessage::CopyData(data) => data.write(dst),
         }
 
         Ok(())
@@ -57,7 +51,6 @@ impl Decoder for PostgresCodec {
 
     fn decode(&mut self, src: &mut BytesMut) -> Result<Option<BackendMessage>, io::Error> {
         let mut idx = 0;
-        let mut request_complete = false;
 
         while let Some(header) = backend::Header::parse(&src[idx..])? {
             let len = header.len() as usize + 1;
@@ -82,7 +75,6 @@ impl Decoder for PostgresCodec {
             idx += len;
 
             if header.tag() == backend::READY_FOR_QUERY_TAG {
-                request_complete = true;
                 break;
             }
         }
@@ -92,7 +84,6 @@ impl Decoder for PostgresCodec {
         } else {
             Ok(Some(BackendMessage::Normal {
                 messages: BackendMessages(src.split_to(idx)),
-                request_complete,
             }))
         }
     }

libs/proxy/tokio-postgres2/src/connect.rs

@@ -59,9 +59,11 @@ where
         connect_timeout: config.connect_timeout,
     };
 
-    let (sender, receiver) = mpsc::unbounded_channel();
+    let (client_tx, conn_rx) = mpsc::unbounded_channel();
+    let (conn_tx, client_rx) = mpsc::channel(4);
     let client = Client::new(
-        sender,
+        client_tx,
+        client_rx,
         socket_config,
         config.ssl_mode,
         process_id,
@@ -74,7 +76,7 @@ where
         .map(|m| BackendMessage::Async(Message::NoticeResponse(m)))
         .collect();
 
-    let connection = Connection::new(stream, delayed, parameters, receiver);
+    let connection = Connection::new(stream, delayed, parameters, conn_tx, conn_rx);
 
     Ok((client, connection))
 }

libs/proxy/tokio-postgres2/src/connection.rs

@@ -4,7 +4,6 @@ use std::pin::Pin;
 use std::task::{Context, Poll};
 
 use bytes::BytesMut;
-use fallible_iterator::FallibleIterator;
 use futures_util::{Sink, Stream, ready};
 use postgres_protocol2::message::backend::Message;
 use postgres_protocol2::message::frontend;
@@ -19,30 +18,12 @@ use crate::error::DbError;
 use crate::maybe_tls_stream::MaybeTlsStream;
 use crate::{AsyncMessage, Error, Notification};
 
-pub enum RequestMessages {
-    Single(FrontendMessage),
-}
-
-pub struct Request {
-    pub messages: RequestMessages,
-    pub sender: mpsc::Sender<BackendMessages>,
-}
-
-pub struct Response {
-    sender: PollSender<BackendMessages>,
-}
-
 #[derive(PartialEq, Debug)]
 enum State {
     Active,
     Closing,
 }
 
-enum WriteReady {
-    Terminating,
-    WaitingOnRead,
-}
-
 /// A connection to a PostgreSQL database.
 ///
 /// This is one half of what is returned when a new connection is established. It performs the actual IO with the
@@ -56,9 +37,11 @@ pub struct Connection<S, T> {
     pub stream: Framed<MaybeTlsStream<S, T>, PostgresCodec>,
     /// HACK: we need this in the Neon Proxy to forward params.
     pub parameters: HashMap<String, String>,
-    receiver: mpsc::UnboundedReceiver<Request>,
+
+    sender: PollSender<BackendMessages>,
+    receiver: mpsc::UnboundedReceiver<FrontendMessage>,
+
     pending_responses: VecDeque<BackendMessage>,
-    responses: VecDeque<Response>,
     state: State,
 }
 
@@ -71,14 +54,15 @@ where
         stream: Framed<MaybeTlsStream<S, T>, PostgresCodec>,
         pending_responses: VecDeque<BackendMessage>,
         parameters: HashMap<String, String>,
-        receiver: mpsc::UnboundedReceiver<Request>,
+        sender: mpsc::Sender<BackendMessages>,
+        receiver: mpsc::UnboundedReceiver<FrontendMessage>,
     ) -> Connection<S, T> {
         Connection {
             stream,
             parameters,
+            sender: PollSender::new(sender),
             receiver,
             pending_responses,
-            responses: VecDeque::new(),
             state: State::Active,
         }
     }
@@ -110,7 +94,7 @@ where
                 }
             };
 
-            let (mut messages, request_complete) = match message {
+            let messages = match message {
                 BackendMessage::Async(Message::NoticeResponse(body)) => {
                     let error = DbError::parse(&mut body.fields()).map_err(Error::parse)?;
                     return Poll::Ready(Ok(AsyncMessage::Notice(error)));
@@ -131,41 +115,19 @@ where
                     continue;
                 }
                 BackendMessage::Async(_) => unreachable!(),
-                BackendMessage::Normal {
-                    messages,
-                    request_complete,
-                } => (messages, request_complete),
+                BackendMessage::Normal { messages } => messages,
             };
 
-            let mut response = match self.responses.pop_front() {
-                Some(response) => response,
-                None => match messages.next().map_err(Error::parse)? {
-                    Some(Message::ErrorResponse(error)) => {
-                        return Poll::Ready(Err(Error::db(error)));
-                    }
-                    _ => return Poll::Ready(Err(Error::unexpected_message())),
-                },
-            };
-
-            match response.sender.poll_reserve(cx) {
+            match self.sender.poll_reserve(cx) {
                 Poll::Ready(Ok(())) => {
-                    let _ = response.sender.send_item(messages);
-                    if !request_complete {
-                        self.responses.push_front(response);
-                    }
+                    let _ = self.sender.send_item(messages);
                 }
                 Poll::Ready(Err(_)) => {
-                    // we need to keep paging through the rest of the messages even if the receiver's hung up
-                    if !request_complete {
-                        self.responses.push_front(response);
-                    }
+                    return Poll::Ready(Err(Error::closed()));
                 }
                 Poll::Pending => {
-                    self.responses.push_front(response);
-                    self.pending_responses.push_back(BackendMessage::Normal {
-                        messages,
-                        request_complete,
-                    });
+                    self.pending_responses
+                        .push_back(BackendMessage::Normal { messages });
                     trace!("poll_read: waiting on sender");
                     return Poll::Pending;
                 }
@@ -174,7 +136,7 @@ where
     }
 
     /// Fetch the next client request and enqueue the response sender.
-    fn poll_request(&mut self, cx: &mut Context<'_>) -> Poll<Option<RequestMessages>> {
+    fn poll_request(&mut self, cx: &mut Context<'_>) -> Poll<Option<FrontendMessage>> {
         if self.receiver.is_closed() {
             return Poll::Ready(None);
         }
@@ -182,10 +144,7 @@ where
         match self.receiver.poll_recv(cx) {
             Poll::Ready(Some(request)) => {
                 trace!("polled new request");
-                self.responses.push_back(Response {
-                    sender: PollSender::new(request.sender),
-                });
-                Poll::Ready(Some(request.messages))
+                Poll::Ready(Some(request))
             }
             Poll::Ready(None) => Poll::Ready(None),
             Poll::Pending => Poll::Pending,
@@ -194,7 +153,7 @@ where
 
     /// Process client requests and write them to the postgres connection, flushing if necessary.
     /// client -> postgres
-    fn poll_write(&mut self, cx: &mut Context<'_>) -> Poll<Result<WriteReady, Error>> {
+    fn poll_write(&mut self, cx: &mut Context<'_>) -> Poll<Result<(), Error>> {
         loop {
             if Pin::new(&mut self.stream)
                 .poll_ready(cx)
@@ -209,14 +168,14 @@ where
 
             match self.poll_request(cx) {
                 // send the message to postgres
-                Poll::Ready(Some(RequestMessages::Single(request))) => {
+                Poll::Ready(Some(request)) => {
                     Pin::new(&mut self.stream)
                         .start_send(request)
                         .map_err(Error::io)?;
                 }
                 // No more messages from the client, and no more responses to wait for.
                 // Send a terminate message to postgres
-                Poll::Ready(None) if self.responses.is_empty() => {
+                Poll::Ready(None) => {
                     trace!("poll_write: at eof, terminating");
                     let mut request = BytesMut::new();
                     frontend::terminate(&mut request);
@@ -228,16 +187,7 @@ where
 
                     trace!("poll_write: sent eof, closing");
                     trace!("poll_write: done");
-                    return Poll::Ready(Ok(WriteReady::Terminating));
-                }
-                // No more messages from the client, but there are still some responses to wait for.
-                Poll::Ready(None) => {
-                    trace!(
-                        "poll_write: at eof, pending responses {}",
-                        self.responses.len()
-                    );
-                    ready!(self.poll_flush(cx))?;
-                    return Poll::Ready(Ok(WriteReady::WaitingOnRead));
+                    return Poll::Ready(Ok(()));
                 }
                 // Still waiting for a message from the client.
                 Poll::Pending => {
@@ -298,7 +248,7 @@ where
             // if the state is still active, try read from and write to postgres.
             let message = self.poll_read(cx)?;
             let closing = self.poll_write(cx)?;
-            if let Poll::Ready(WriteReady::Terminating) = closing {
+            if let Poll::Ready(()) = closing {
                 self.state = State::Closing;
             }
 

libs/proxy/tokio-postgres2/src/error/mod.rs

@@ -86,6 +86,27 @@ pub struct DbError {
 }
 
 impl DbError {
+    pub fn new_test_error(code: SqlState, message: String) -> Self {
+        DbError {
+            severity: "ERROR".to_string(),
+            parsed_severity: Some(Severity::Error),
+            code,
+            message,
+            detail: None,
+            hint: None,
+            position: None,
+            where_: None,
+            schema: None,
+            table: None,
+            column: None,
+            datatype: None,
+            constraint: None,
+            file: None,
+            line: None,
+            routine: None,
+        }
+    }
+
     pub(crate) fn parse(fields: &mut ErrorFields<'_>) -> io::Result<DbError> {
         let mut severity = None;
         let mut parsed_severity = None;

libs/proxy/tokio-postgres2/src/generic_client.rs

@@ -1,9 +1,6 @@
 #![allow(async_fn_in_trait)]
 
-use postgres_protocol2::Oid;
-
 use crate::query::RowStream;
-use crate::types::Type;
 use crate::{Client, Error, Transaction};
 
 mod private {
@@ -15,20 +12,17 @@ mod private {
 /// This trait is "sealed", and cannot be implemented outside of this crate.
 pub trait GenericClient: private::Sealed {
     /// Like `Client::query_raw_txt`.
-    async fn query_raw_txt<S, I>(&self, statement: &str, params: I) -> Result<RowStream, Error>
+    async fn query_raw_txt<S, I>(&mut self, statement: &str, params: I) -> Result<RowStream, Error>
     where
         S: AsRef<str> + Sync + Send,
         I: IntoIterator<Item = Option<S>> + Sync + Send,
         I::IntoIter: ExactSizeIterator + Sync + Send;
-
-    /// Query for type information
-    async fn get_type(&mut self, oid: Oid) -> Result<Type, Error>;
 }
 
 impl private::Sealed for Client {}
 
 impl GenericClient for Client {
-    async fn query_raw_txt<S, I>(&self, statement: &str, params: I) -> Result<RowStream, Error>
+    async fn query_raw_txt<S, I>(&mut self, statement: &str, params: I) -> Result<RowStream, Error>
     where
         S: AsRef<str> + Sync + Send,
         I: IntoIterator<Item = Option<S>> + Sync + Send,
@@ -36,17 +30,12 @@ impl GenericClient for Client {
     {
         self.query_raw_txt(statement, params).await
     }
-
-    /// Query for type information
-    async fn get_type(&mut self, oid: Oid) -> Result<Type, Error> {
-        self.get_type_inner(oid).await
-    }
 }
 
 impl private::Sealed for Transaction<'_> {}
 
 impl GenericClient for Transaction<'_> {
-    async fn query_raw_txt<S, I>(&self, statement: &str, params: I) -> Result<RowStream, Error>
+    async fn query_raw_txt<S, I>(&mut self, statement: &str, params: I) -> Result<RowStream, Error>
     where
         S: AsRef<str> + Sync + Send,
         I: IntoIterator<Item = Option<S>> + Sync + Send,
@@ -54,9 +43,4 @@ impl GenericClient for Transaction<'_> {
     {
         self.query_raw_txt(statement, params).await
     }
-
-    /// Query for type information
-    async fn get_type(&mut self, oid: Oid) -> Result<Type, Error> {
-        self.client_mut().get_type(oid).await
-    }
 }

libs/proxy/tokio-postgres2/src/lib.rs

@@ -18,7 +18,6 @@ pub use crate::statement::{Column, Statement};
 pub use crate::tls::NoTls;
 pub use crate::transaction::Transaction;
 pub use crate::transaction_builder::{IsolationLevel, TransactionBuilder};
-use crate::types::ToSql;
 
 /// After executing a query, the connection will be in one of these states
 #[derive(Clone, Copy, Debug, PartialEq)]
@@ -120,9 +119,3 @@ pub enum SimpleQueryMessage {
     /// The number of rows modified or selected is returned.
     CommandComplete(u64),
 }
-
-fn slice_iter<'a>(
-    s: &'a [&'a (dyn ToSql + Sync)],
-) -> impl ExactSizeIterator<Item = &'a (dyn ToSql + Sync)> + 'a {
-    s.iter().map(|s| *s as _)
-}

libs/proxy/tokio-postgres2/src/prepare.rs

@@ -1,19 +1,14 @@
-use std::future::Future;
-use std::pin::Pin;
-use std::sync::Arc;
-
-use bytes::Bytes;
+use bytes::BytesMut;
 use fallible_iterator::FallibleIterator;
-use futures_util::{TryStreamExt, pin_mut};
-use postgres_protocol2::message::backend::Message;
+use postgres_protocol2::IsNull;
+use postgres_protocol2::message::backend::{Message, RowDescriptionBody};
 use postgres_protocol2::message::frontend;
-use tracing::debug;
+use postgres_protocol2::types::oid_to_sql;
+use postgres_types2::Format;
 
-use crate::client::{CachedTypeInfo, InnerClient};
-use crate::codec::FrontendMessage;
-use crate::connection::RequestMessages;
+use crate::client::{CachedTypeInfo, PartialQuery, Responses};
 use crate::types::{Kind, Oid, Type};
-use crate::{Column, Error, Statement, query, slice_iter};
+use crate::{Column, Error, Row, Statement};
 
 pub(crate) const TYPEINFO_QUERY: &str = "\
 SELECT t.typname, t.typtype, t.typelem, r.rngsubtype, t.typbasetype, n.nspname, t.typrelid
@@ -23,22 +18,51 @@ INNER JOIN pg_catalog.pg_namespace n ON t.typnamespace = n.oid
 WHERE t.oid = $1
 ";
 
+/// we need to make sure we close this prepared statement.
+struct CloseStmt<'a, 'b> {
+    client: Option<&'a mut PartialQuery<'b>>,
+    name: &'static str,
+}
+
+impl<'a> CloseStmt<'a, '_> {
+    fn close(mut self) -> Result<&'a mut Responses, Error> {
+        let client = self.client.take().unwrap();
+        client.send_with_flush(|buf| {
+            frontend::close(b'S', self.name, buf).map_err(Error::encode)?;
+            Ok(())
+        })
+    }
+}
+
+impl Drop for CloseStmt<'_, '_> {
+    fn drop(&mut self) {
+        if let Some(client) = self.client.take() {
+            let _ = client.send_with_flush(|buf| {
+                frontend::close(b'S', self.name, buf).map_err(Error::encode)?;
+                Ok(())
+            });
+        }
+    }
+}
+
 async fn prepare_typecheck(
-    client: &Arc<InnerClient>,
+    client: &mut PartialQuery<'_>,
     name: &'static str,
     query: &str,
-    types: &[Type],
 ) -> Result<Statement, Error> {
-    let buf = encode(client, name, query, types)?;
-    let mut responses = client.send(RequestMessages::Single(FrontendMessage::Raw(buf)))?;
+    let responses = client.send_with_flush(|buf| {
+        frontend::parse(name, query, [], buf).map_err(Error::encode)?;
+        frontend::describe(b'S', name, buf).map_err(Error::encode)?;
+        Ok(())
+    })?;
 
     match responses.next().await? {
         Message::ParseComplete => {}
         _ => return Err(Error::unexpected_message()),
     }
 
-    let parameter_description = match responses.next().await? {
-        Message::ParameterDescription(body) => body,
+    match responses.next().await? {
+        Message::ParameterDescription(_) => {}
         _ => return Err(Error::unexpected_message()),
     };
 
@@ -48,13 +72,6 @@ async fn prepare_typecheck(
         _ => return Err(Error::unexpected_message()),
     };
 
-    let mut parameters = vec![];
-    let mut it = parameter_description.parameters();
-    while let Some(oid) = it.next().map_err(Error::parse)? {
-        let type_ = Type::from_oid(oid).ok_or_else(Error::unexpected_message)?;
-        parameters.push(type_);
-    }
-
     let mut columns = vec![];
     if let Some(row_description) = row_description {
         let mut it = row_description.fields();
@@ -65,98 +82,168 @@ async fn prepare_typecheck(
         }
     }
 
-    Ok(Statement::new(client, name, parameters, columns))
+    Ok(Statement::new(name, columns))
 }
 
-fn encode(client: &InnerClient, name: &str, query: &str, types: &[Type]) -> Result<Bytes, Error> {
-    if types.is_empty() {
-        debug!("preparing query {}: {}", name, query);
-    } else {
-        debug!("preparing query {} with types {:?}: {}", name, types, query);
-    }
-
-    client.with_buf(|buf| {
-        frontend::parse(name, query, types.iter().map(Type::oid), buf).map_err(Error::encode)?;
-        frontend::describe(b'S', name, buf).map_err(Error::encode)?;
-        frontend::sync(buf);
-        Ok(buf.split().freeze())
-    })
-}
-
-pub async fn get_type(
-    client: &Arc<InnerClient>,
-    typecache: &mut CachedTypeInfo,
-    oid: Oid,
-) -> Result<Type, Error> {
+fn try_from_cache(typecache: &CachedTypeInfo, oid: Oid) -> Option<Type> {
     if let Some(type_) = Type::from_oid(oid) {
-        return Ok(type_);
+        return Some(type_);
     }
 
     if let Some(type_) = typecache.types.get(&oid) {
-        return Ok(type_.clone());
+        return Some(type_.clone());
     };
 
-    let stmt = typeinfo_statement(client, typecache).await?;
+    None
+}
 
-    let rows = query::query(client, stmt, slice_iter(&[&oid])).await?;
-    pin_mut!(rows);
+pub async fn parse_row_description(
+    client: &mut PartialQuery<'_>,
+    typecache: &mut CachedTypeInfo,
+    row_description: Option<RowDescriptionBody>,
+) -> Result<Vec<Column>, Error> {
+    let mut columns = vec![];
 
-    let row = match rows.try_next().await? {
-        Some(row) => row,
-        None => return Err(Error::unexpected_message()),
+    if let Some(row_description) = row_description {
+        let mut it = row_description.fields();
+        while let Some(field) = it.next().map_err(Error::parse)? {
+            let type_ = try_from_cache(typecache, field.type_oid()).unwrap_or(Type::UNKNOWN);
+            let column = Column::new(field.name().to_string(), type_, field);
+            columns.push(column);
+        }
+    }
+
+    let all_known = columns.iter().all(|c| c.type_ != Type::UNKNOWN);
+    if all_known {
+        // all known, return early.
+        return Ok(columns);
+    }
+
+    let typeinfo = "neon_proxy_typeinfo";
+
+    // make sure to close the typeinfo statement before exiting.
+    let mut guard = CloseStmt {
+        name: typeinfo,
+        client: None,
+    };
+    let client = guard.client.insert(client);
+
+    // get the typeinfo statement.
+    let stmt = prepare_typecheck(client, typeinfo, TYPEINFO_QUERY).await?;
+
+    for column in &mut columns {
+        column.type_ = get_type(client, typecache, &stmt, column.type_oid()).await?;
+    }
+
+    // cancel the close guard.
+    let responses = guard.close()?;
+
+    match responses.next().await? {
+        Message::CloseComplete => {}
+        _ => return Err(Error::unexpected_message()),
+    }
+
+    Ok(columns)
+}
+
+async fn get_type(
+    client: &mut PartialQuery<'_>,
+    typecache: &mut CachedTypeInfo,
+    stmt: &Statement,
+    mut oid: Oid,
+) -> Result<Type, Error> {
+    let mut stack = vec![];
+    let mut type_ = loop {
+        if let Some(type_) = try_from_cache(typecache, oid) {
+            break type_;
+        }
+
+        let row = exec(client, stmt, oid).await?;
+        if stack.len() > 8 {
+            return Err(Error::unexpected_message());
+        }
+
+        let name: String = row.try_get(0)?;
+        let type_: i8 = row.try_get(1)?;
+        let elem_oid: Oid = row.try_get(2)?;
+        let rngsubtype: Option<Oid> = row.try_get(3)?;
+        let basetype: Oid = row.try_get(4)?;
+        let schema: String = row.try_get(5)?;
+        let relid: Oid = row.try_get(6)?;
+
+        let kind = if type_ == b'e' as i8 {
+            Kind::Enum
+        } else if type_ == b'p' as i8 {
+            Kind::Pseudo
+        } else if basetype != 0 {
+            Kind::Domain(basetype)
+        } else if elem_oid != 0 {
+            stack.push((name, oid, schema));
+            oid = elem_oid;
+            continue;
+        } else if relid != 0 {
+            Kind::Composite(relid)
+        } else if let Some(rngsubtype) = rngsubtype {
+            Kind::Range(rngsubtype)
+        } else {
+            Kind::Simple
+        };
+
+        let type_ = Type::new(name, oid, kind, schema);
+        typecache.types.insert(oid, type_.clone());
+        break type_;
     };
 
-    let name: String = row.try_get(0)?;
-    let type_: i8 = row.try_get(1)?;
-    let elem_oid: Oid = row.try_get(2)?;
-    let rngsubtype: Option<Oid> = row.try_get(3)?;
-    let basetype: Oid = row.try_get(4)?;
-    let schema: String = row.try_get(5)?;
-    let relid: Oid = row.try_get(6)?;
-
-    let kind = if type_ == b'e' as i8 {
-        Kind::Enum
-    } else if type_ == b'p' as i8 {
-        Kind::Pseudo
-    } else if basetype != 0 {
-        Kind::Domain(basetype)
-    } else if elem_oid != 0 {
-        let type_ = get_type_rec(client, typecache, elem_oid).await?;
-        Kind::Array(type_)
-    } else if relid != 0 {
-        Kind::Composite(relid)
-    } else if let Some(rngsubtype) = rngsubtype {
-        let type_ = get_type_rec(client, typecache, rngsubtype).await?;
-        Kind::Range(type_)
-    } else {
-        Kind::Simple
-    };
-
-    let type_ = Type::new(name, oid, kind, schema);
-    typecache.types.insert(oid, type_.clone());
+    while let Some((name, oid, schema)) = stack.pop() {
+        type_ = Type::new(name, oid, Kind::Array(type_), schema);
+        typecache.types.insert(oid, type_.clone());
+    }
 
     Ok(type_)
 }
 
-fn get_type_rec<'a>(
-    client: &'a Arc<InnerClient>,
-    typecache: &'a mut CachedTypeInfo,
-    oid: Oid,
-) -> Pin<Box<dyn Future<Output = Result<Type, Error>> + Send + 'a>> {
-    Box::pin(get_type(client, typecache, oid))
-}
+/// exec the typeinfo statement returning one row.
+async fn exec(
+    client: &mut PartialQuery<'_>,
+    statement: &Statement,
+    param: Oid,
+) -> Result<Row, Error> {
+    let responses = client.send_with_flush(|buf| {
+        encode_bind(statement, param, "", buf);
+        frontend::execute("", 0, buf).map_err(Error::encode)?;
+        Ok(())
+    })?;
 
-async fn typeinfo_statement(
-    client: &Arc<InnerClient>,
-    typecache: &mut CachedTypeInfo,
-) -> Result<Statement, Error> {
-    if let Some(stmt) = &typecache.typeinfo {
-        return Ok(stmt.clone());
+    match responses.next().await? {
+        Message::BindComplete => {}
+        _ => return Err(Error::unexpected_message()),
     }
 
-    let typeinfo = "neon_proxy_typeinfo";
-    let stmt = prepare_typecheck(client, typeinfo, TYPEINFO_QUERY, &[]).await?;
+    let row = match responses.next().await? {
+        Message::DataRow(body) => Row::new(statement.clone(), body, Format::Binary)?,
+        _ => return Err(Error::unexpected_message()),
+    };
 
-    typecache.typeinfo = Some(stmt.clone());
-    Ok(stmt)
+    match responses.next().await? {
+        Message::CommandComplete(_) => {}
+        _ => return Err(Error::unexpected_message()),
+    };
+
+    Ok(row)
+}
+
+fn encode_bind(statement: &Statement, param: Oid, portal: &str, buf: &mut BytesMut) {
+    frontend::bind(
+        portal,
+        statement.name(),
+        [Format::Binary as i16],
+        [param],
+        |param, buf| {
+            oid_to_sql(param, buf);
+            Ok(IsNull::No)
+        },
+        [Format::Binary as i16],
+        buf,
+    )
+    .unwrap();
 }

libs/proxy/tokio-postgres2/src/query.rs

@@ -1,76 +1,43 @@
-use std::fmt;
-use std::marker::PhantomPinned;
 use std::pin::Pin;
-use std::sync::Arc;
 use std::task::{Context, Poll};
 
-use bytes::{BufMut, Bytes, BytesMut};
-use fallible_iterator::FallibleIterator;
+use bytes::BufMut;
 use futures_util::{Stream, ready};
-use pin_project_lite::pin_project;
 use postgres_protocol2::message::backend::Message;
 use postgres_protocol2::message::frontend;
-use postgres_types2::{Format, ToSql, Type};
-use tracing::debug;
+use postgres_types2::Format;
 
-use crate::client::{InnerClient, Responses};
-use crate::codec::FrontendMessage;
-use crate::connection::RequestMessages;
-use crate::types::IsNull;
-use crate::{Column, Error, ReadyForQueryStatus, Row, Statement};
+use crate::client::{CachedTypeInfo, InnerClient, Responses};
+use crate::{Error, ReadyForQueryStatus, Row, Statement};
 
-struct BorrowToSqlParamsDebug<'a>(&'a [&'a (dyn ToSql + Sync)]);
-
-impl fmt::Debug for BorrowToSqlParamsDebug<'_> {
-    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
-        f.debug_list().entries(self.0.iter()).finish()
-    }
-}
-
-pub async fn query<'a, I>(
-    client: &InnerClient,
-    statement: Statement,
-    params: I,
-) -> Result<RowStream, Error>
-where
-    I: IntoIterator<Item = &'a (dyn ToSql + Sync)>,
-    I::IntoIter: ExactSizeIterator,
-{
-    let buf = if tracing::enabled!(tracing::Level::DEBUG) {
-        let params = params.into_iter().collect::<Vec<_>>();
-        debug!(
-            "executing statement {} with parameters: {:?}",
-            statement.name(),
-            BorrowToSqlParamsDebug(params.as_slice()),
-        );
-        encode(client, &statement, params)?
-    } else {
-        encode(client, &statement, params)?
-    };
-    let responses = start(client, buf).await?;
-    Ok(RowStream {
-        statement,
-        responses,
-        command_tag: None,
-        status: ReadyForQueryStatus::Unknown,
-        output_format: Format::Binary,
-        _p: PhantomPinned,
-    })
-}
-
-pub async fn query_txt<S, I>(
-    client: &Arc<InnerClient>,
+pub async fn query_txt<'a, S, I>(
+    client: &'a mut InnerClient,
+    typecache: &mut CachedTypeInfo,
     query: &str,
     params: I,
-) -> Result<RowStream, Error>
+) -> Result<RowStream<'a>, Error>
 where
     S: AsRef<str>,
     I: IntoIterator<Item = Option<S>>,
     I::IntoIter: ExactSizeIterator,
 {
     let params = params.into_iter();
+    let mut client = client.start()?;
 
-    let buf = client.with_buf(|buf| {
+    // Flow:
+    // 1. Parse the query
+    // 2. Inspect the row description for OIDs
+    // 3. If there's any OIDs we don't already know about, perform the typeinfo routine
+    // 4. Execute the query
+    // 5. Sync.
+    //
+    // The typeinfo routine:
+    // 1. Parse the typeinfo query
+    // 2. Execute the query on each OID
+    // 3. If the result does not match an OID we know, repeat 2.
+
+    // parse the query and get type info
+    let responses = client.send_with_flush(|buf| {
         frontend::parse(
             "",                 // unnamed prepared statement
             query,              // query to parse
@@ -79,7 +46,30 @@ where
         )
         .map_err(Error::encode)?;
         frontend::describe(b'S', "", buf).map_err(Error::encode)?;
-        // Bind, pass params as text, retrieve as binary
+        Ok(())
+    })?;
+
+    match responses.next().await? {
+        Message::ParseComplete => {}
+        _ => return Err(Error::unexpected_message()),
+    }
+
+    match responses.next().await? {
+        Message::ParameterDescription(_) => {}
+        _ => return Err(Error::unexpected_message()),
+    };
+
+    let row_description = match responses.next().await? {
+        Message::RowDescription(body) => Some(body),
+        Message::NoData => None,
+        _ => return Err(Error::unexpected_message()),
+    };
+
+    let columns =
+        crate::prepare::parse_row_description(&mut client, typecache, row_description).await?;
+
+    let responses = client.send_with_sync(|buf| {
+        // Bind, pass params as text, retrieve as text
         match frontend::bind(
             "",                 // empty string selects the unnamed portal
             "",                 // unnamed prepared statement
@@ -102,173 +92,55 @@ where
 
         // Execute
         frontend::execute("", 0, buf).map_err(Error::encode)?;
-        // Sync
-        frontend::sync(buf);
 
-        Ok(buf.split().freeze())
+        Ok(())
     })?;
 
-    // now read the responses
-    let mut responses = client.send(RequestMessages::Single(FrontendMessage::Raw(buf)))?;
-
-    match responses.next().await? {
-        Message::ParseComplete => {}
-        _ => return Err(Error::unexpected_message()),
-    }
-
-    let parameter_description = match responses.next().await? {
-        Message::ParameterDescription(body) => body,
-        _ => return Err(Error::unexpected_message()),
-    };
-
-    let row_description = match responses.next().await? {
-        Message::RowDescription(body) => Some(body),
-        Message::NoData => None,
-        _ => return Err(Error::unexpected_message()),
-    };
-
     match responses.next().await? {
         Message::BindComplete => {}
         _ => return Err(Error::unexpected_message()),
     }
 
-    let mut parameters = vec![];
-    let mut it = parameter_description.parameters();
-    while let Some(oid) = it.next().map_err(Error::parse)? {
-        let type_ = Type::from_oid(oid).unwrap_or(Type::UNKNOWN);
-        parameters.push(type_);
-    }
-
-    let mut columns = vec![];
-    if let Some(row_description) = row_description {
-        let mut it = row_description.fields();
-        while let Some(field) = it.next().map_err(Error::parse)? {
-            let type_ = Type::from_oid(field.type_oid()).unwrap_or(Type::UNKNOWN);
-            let column = Column::new(field.name().to_string(), type_, field);
-            columns.push(column);
-        }
-    }
-
     Ok(RowStream {
-        statement: Statement::new_anonymous(parameters, columns),
         responses,
+        statement: Statement::new("", columns),
         command_tag: None,
         status: ReadyForQueryStatus::Unknown,
         output_format: Format::Text,
-        _p: PhantomPinned,
     })
 }
 
-async fn start(client: &InnerClient, buf: Bytes) -> Result<Responses, Error> {
-    let mut responses = client.send(RequestMessages::Single(FrontendMessage::Raw(buf)))?;
-
-    match responses.next().await? {
-        Message::BindComplete => {}
-        _ => return Err(Error::unexpected_message()),
-    }
-
-    Ok(responses)
+/// A stream of table rows.
+pub struct RowStream<'a> {
+    responses: &'a mut Responses,
+    output_format: Format,
+    pub statement: Statement,
+    pub command_tag: Option<String>,
+    pub status: ReadyForQueryStatus,
 }
 
-pub fn encode<'a, I>(client: &InnerClient, statement: &Statement, params: I) -> Result<Bytes, Error>
-where
-    I: IntoIterator<Item = &'a (dyn ToSql + Sync)>,
-    I::IntoIter: ExactSizeIterator,
-{
-    client.with_buf(|buf| {
-        encode_bind(statement, params, "", buf)?;
-        frontend::execute("", 0, buf).map_err(Error::encode)?;
-        frontend::sync(buf);
-        Ok(buf.split().freeze())
-    })
-}
-
-pub fn encode_bind<'a, I>(
-    statement: &Statement,
-    params: I,
-    portal: &str,
-    buf: &mut BytesMut,
-) -> Result<(), Error>
-where
-    I: IntoIterator<Item = &'a (dyn ToSql + Sync)>,
-    I::IntoIter: ExactSizeIterator,
-{
-    let param_types = statement.params();
-    let params = params.into_iter();
-
-    assert!(
-        param_types.len() == params.len(),
-        "expected {} parameters but got {}",
-        param_types.len(),
-        params.len()
-    );
-
-    let (param_formats, params): (Vec<_>, Vec<_>) = params
-        .zip(param_types.iter())
-        .map(|(p, ty)| (p.encode_format(ty) as i16, p))
-        .unzip();
-
-    let params = params.into_iter();
-
-    let mut error_idx = 0;
-    let r = frontend::bind(
-        portal,
-        statement.name(),
-        param_formats,
-        params.zip(param_types).enumerate(),
-        |(idx, (param, ty)), buf| match param.to_sql_checked(ty, buf) {
-            Ok(IsNull::No) => Ok(postgres_protocol2::IsNull::No),
-            Ok(IsNull::Yes) => Ok(postgres_protocol2::IsNull::Yes),
-            Err(e) => {
-                error_idx = idx;
-                Err(e)
-            }
-        },
-        Some(1),
-        buf,
-    );
-    match r {
-        Ok(()) => Ok(()),
-        Err(frontend::BindError::Conversion(e)) => Err(Error::to_sql(e, error_idx)),
-        Err(frontend::BindError::Serialization(e)) => Err(Error::encode(e)),
-    }
-}
-
-pin_project! {
-    /// A stream of table rows.
-    pub struct RowStream {
-        statement: Statement,
-        responses: Responses,
-        command_tag: Option<String>,
-        output_format: Format,
-        status: ReadyForQueryStatus,
-        #[pin]
-        _p: PhantomPinned,
-    }
-}
-
-impl Stream for RowStream {
+impl Stream for RowStream<'_> {
     type Item = Result<Row, Error>;
 
     fn poll_next(self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Option<Self::Item>> {
-        let this = self.project();
+        let this = self.get_mut();
         loop {
             match ready!(this.responses.poll_next(cx)?) {
                 Message::DataRow(body) => {
                     return Poll::Ready(Some(Ok(Row::new(
                         this.statement.clone(),
                         body,
-                        *this.output_format,
+                        this.output_format,
                     )?)));
                 }
                 Message::EmptyQueryResponse | Message::PortalSuspended => {}
                 Message::CommandComplete(body) => {
                     if let Ok(tag) = body.tag() {
-                        *this.command_tag = Some(tag.to_string());
+                        this.command_tag = Some(tag.to_string());
                     }
                 }
                 Message::ReadyForQuery(status) => {
-                    *this.status = status.into();
+                    this.status = status.into();
                     return Poll::Ready(None);
                 }
                 _ => return Poll::Ready(Some(Err(Error::unexpected_message()))),
@@ -276,24 +148,3 @@ impl Stream for RowStream {
         }
     }
 }
-
-impl RowStream {
-    /// Returns information about the columns of data in the row.
-    pub fn columns(&self) -> &[Column] {
-        self.statement.columns()
-    }
-
-    /// Returns the command tag of this query.
-    ///
-    /// This is only available after the stream has been exhausted.
-    pub fn command_tag(&self) -> Option<String> {
-        self.command_tag.clone()
-    }
-
-    /// Returns if the connection is ready for querying, with the status of the connection.
-    ///
-    /// This might be available only after the stream has been exhausted.
-    pub fn ready_status(&self) -> ReadyForQueryStatus {
-        self.status
-    }
-}

libs/proxy/tokio-postgres2/src/simple_query.rs

@@ -1,19 +1,14 @@
-use std::marker::PhantomPinned;
 use std::pin::Pin;
 use std::sync::Arc;
 use std::task::{Context, Poll};
 
-use bytes::Bytes;
 use fallible_iterator::FallibleIterator;
 use futures_util::{Stream, ready};
 use pin_project_lite::pin_project;
 use postgres_protocol2::message::backend::Message;
-use postgres_protocol2::message::frontend;
 use tracing::debug;
 
 use crate::client::{InnerClient, Responses};
-use crate::codec::FrontendMessage;
-use crate::connection::RequestMessages;
 use crate::{Error, ReadyForQueryStatus, SimpleQueryMessage, SimpleQueryRow};
 
 /// Information about a column of a single query row.
@@ -33,28 +28,28 @@ impl SimpleColumn {
     }
 }
 
-pub async fn simple_query(client: &InnerClient, query: &str) -> Result<SimpleQueryStream, Error> {
+pub async fn simple_query<'a>(
+    client: &'a mut InnerClient,
+    query: &str,
+) -> Result<SimpleQueryStream<'a>, Error> {
     debug!("executing simple query: {}", query);
 
-    let buf = encode(client, query)?;
-    let responses = client.send(RequestMessages::Single(FrontendMessage::Raw(buf)))?;
+    let responses = client.send_simple_query(query)?;
 
     Ok(SimpleQueryStream {
         responses,
         columns: None,
         status: ReadyForQueryStatus::Unknown,
-        _p: PhantomPinned,
     })
 }
 
 pub async fn batch_execute(
-    client: &InnerClient,
+    client: &mut InnerClient,
     query: &str,
 ) -> Result<ReadyForQueryStatus, Error> {
     debug!("executing statement batch: {}", query);
 
-    let buf = encode(client, query)?;
-    let mut responses = client.send(RequestMessages::Single(FrontendMessage::Raw(buf)))?;
+    let responses = client.send_simple_query(query)?;
 
     loop {
         match responses.next().await? {
@@ -68,25 +63,16 @@ pub async fn batch_execute(
     }
 }
 
-pub(crate) fn encode(client: &InnerClient, query: &str) -> Result<Bytes, Error> {
-    client.with_buf(|buf| {
-        frontend::query(query, buf).map_err(Error::encode)?;
-        Ok(buf.split().freeze())
-    })
-}
-
 pin_project! {
     /// A stream of simple query results.
-    pub struct SimpleQueryStream {
-        responses: Responses,
+    pub struct SimpleQueryStream<'a> {
+        responses: &'a mut Responses,
         columns: Option<Arc<[SimpleColumn]>>,
         status: ReadyForQueryStatus,
-        #[pin]
-        _p: PhantomPinned,
     }
 }
 
-impl SimpleQueryStream {
+impl SimpleQueryStream<'_> {
     /// Returns if the connection is ready for querying, with the status of the connection.
     ///
     /// This might be available only after the stream has been exhausted.
@@ -95,7 +81,7 @@ impl SimpleQueryStream {
     }
 }
 
-impl Stream for SimpleQueryStream {
+impl Stream for SimpleQueryStream<'_> {
     type Item = Result<SimpleQueryMessage, Error>;
 
     fn poll_next(self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Option<Self::Item>> {

libs/proxy/tokio-postgres2/src/statement.rs

@@ -1,35 +1,15 @@
 use std::fmt;
-use std::sync::{Arc, Weak};
+use std::sync::Arc;
 
+use crate::types::Type;
 use postgres_protocol2::Oid;
 use postgres_protocol2::message::backend::Field;
-use postgres_protocol2::message::frontend;
-
-use crate::client::InnerClient;
-use crate::codec::FrontendMessage;
-use crate::connection::RequestMessages;
-use crate::types::Type;
 
 struct StatementInner {
-    client: Weak<InnerClient>,
     name: &'static str,
-    params: Vec<Type>,
     columns: Vec<Column>,
 }
 
-impl Drop for StatementInner {
-    fn drop(&mut self) {
-        if let Some(client) = self.client.upgrade() {
-            let buf = client.with_buf(|buf| {
-                frontend::close(b'S', self.name, buf).unwrap();
-                frontend::sync(buf);
-                buf.split().freeze()
-            });
-            let _ = client.send(RequestMessages::Single(FrontendMessage::Raw(buf)));
-        }
-    }
-}
-
 /// A prepared statement.
 ///
 /// Prepared statements can only be used with the connection that created them.
@@ -37,38 +17,14 @@ impl Drop for StatementInner {
 pub struct Statement(Arc<StatementInner>);
 
 impl Statement {
-    pub(crate) fn new(
-        inner: &Arc<InnerClient>,
-        name: &'static str,
-        params: Vec<Type>,
-        columns: Vec<Column>,
-    ) -> Statement {
-        Statement(Arc::new(StatementInner {
-            client: Arc::downgrade(inner),
-            name,
-            params,
-            columns,
-        }))
-    }
-
-    pub(crate) fn new_anonymous(params: Vec<Type>, columns: Vec<Column>) -> Statement {
-        Statement(Arc::new(StatementInner {
-            client: Weak::new(),
-            name: "<anonymous>",
-            params,
-            columns,
-        }))
+    pub(crate) fn new(name: &'static str, columns: Vec<Column>) -> Statement {
+        Statement(Arc::new(StatementInner { name, columns }))
     }
 
     pub(crate) fn name(&self) -> &str {
         self.0.name
     }
 
-    /// Returns the expected types of the statement's parameters.
-    pub fn params(&self) -> &[Type] {
-        &self.0.params
-    }
-
     /// Returns information about the columns returned when the statement is queried.
     pub fn columns(&self) -> &[Column] {
         &self.0.columns
@@ -78,7 +34,7 @@ impl Statement {
 /// Information about a column of a query.
 pub struct Column {
     name: String,
-    type_: Type,
+    pub(crate) type_: Type,
 
     // raw fields from RowDescription
     table_oid: Oid,

libs/proxy/tokio-postgres2/src/transaction.rs

@@ -1,7 +1,3 @@
-use postgres_protocol2::message::frontend;
-
-use crate::codec::FrontendMessage;
-use crate::connection::RequestMessages;
 use crate::query::RowStream;
 use crate::{CancelToken, Client, Error, ReadyForQueryStatus};
 
@@ -20,14 +16,7 @@ impl Drop for Transaction<'_> {
             return;
         }
 
-        let buf = self.client.inner().with_buf(|buf| {
-            frontend::query("ROLLBACK", buf).unwrap();
-            buf.split().freeze()
-        });
-        let _ = self
-            .client
-            .inner()
-            .send(RequestMessages::Single(FrontendMessage::Raw(buf)));
+        let _ = self.client.inner_mut().send_simple_query("ROLLBACK");
     }
 }
 
@@ -54,7 +43,11 @@ impl<'a> Transaction<'a> {
     }
 
     /// Like `Client::query_raw_txt`.
-    pub async fn query_raw_txt<S, I>(&self, statement: &str, params: I) -> Result<RowStream, Error>
+    pub async fn query_raw_txt<S, I>(
+        &mut self,
+        statement: &str,
+        params: I,
+    ) -> Result<RowStream, Error>
     where
         S: AsRef<str>,
         I: IntoIterator<Item = Option<S>>,

libs/remote_storage/src/azure_blob.rs

@@ -330,11 +330,18 @@ impl AzureBlobStorage {
                 if let Err(DownloadError::Timeout) = &next_item {
                     timeout_try_cnt += 1;
                     if timeout_try_cnt <= 5 {
-                        continue;
+                        continue 'outer;
                     }
                 }
 
-                let next_item = next_item?;
+                let next_item = match next_item {
+                    Ok(next_item) => next_item,
+                    Err(e) => {
+                        // The error is potentially retryable, so we must rewind the loop after yielding.
+                        yield Err(e);
+                        continue 'outer;
+                    },
+                };
 
                 // Log a warning if we saw two timeouts in a row before a successful request
                 if timeout_try_cnt > 2 {

libs/remote_storage/src/s3_bucket.rs

@@ -657,7 +657,14 @@ impl RemoteStorage for S3Bucket {
                     res = request => Ok(res),
                     _ = tokio::time::sleep(self.timeout) => Err(DownloadError::Timeout),
                     _ = cancel.cancelled() => Err(DownloadError::Cancelled),
-                }?;
+                };
+
+                if let Err(DownloadError::Timeout) = &response {
+                    yield Err(DownloadError::Timeout);
+                    continue 'outer;
+                }
+
+                let response = response?; // always yield cancellation errors and stop the stream
 
                 let response = response
                     .context("Failed to list S3 prefixes")

libs/utils/src/crashsafe.rs

@@ -1,7 +1,7 @@
 use std::borrow::Cow;
 use std::fs::{self, File};
 use std::io::{self, Write};
-use std::os::fd::AsRawFd;
+use std::os::fd::AsFd;
 
 use camino::{Utf8Path, Utf8PathBuf};
 
@@ -210,13 +210,13 @@ pub fn overwrite(
 
 /// Syncs the filesystem for the given file descriptor.
 #[cfg_attr(target_os = "macos", allow(unused_variables))]
-pub fn syncfs(fd: impl AsRawFd) -> anyhow::Result<()> {
+pub fn syncfs(fd: impl AsFd) -> anyhow::Result<()> {
     // Linux guarantees durability for syncfs.
     // POSIX doesn't have syncfs, and further does not actually guarantee durability of sync().
     #[cfg(target_os = "linux")]
     {
         use anyhow::Context;
-        nix::unistd::syncfs(fd.as_raw_fd()).context("syncfs")?;
+        nix::unistd::syncfs(fd).context("syncfs")?;
     }
     #[cfg(target_os = "macos")]
     {

libs/utils/src/fs_ext/rename_noreplace.rs

@@ -11,9 +11,9 @@ pub fn rename_noreplace<P1: ?Sized + NixPath, P2: ?Sized + NixPath>(
         #[cfg(all(target_os = "linux", target_env = "gnu"))]
         {
             nix::fcntl::renameat2(
-                None,
+                nix::fcntl::AT_FDCWD,
                 src,
-                None,
+                nix::fcntl::AT_FDCWD,
                 dst,
                 nix::fcntl::RenameFlags::RENAME_NOREPLACE,
             )

libs/utils/src/lock_file.rs

@@ -1,6 +1,6 @@
 //! A module to create and read lock files.
 //!
-//! File locking is done using [`fcntl::flock`] exclusive locks.
+//! File locking is done using [`nix::fcntl::Flock`] exclusive locks.
 //! The only consumer of this module is currently
 //! [`pid_file`](crate::pid_file). See the module-level comment
 //! there for potential pitfalls with lock files that are used
@@ -9,26 +9,25 @@
 use std::fs;
 use std::io::{Read, Write};
 use std::ops::Deref;
-use std::os::unix::prelude::AsRawFd;
 
 use anyhow::Context;
 use camino::{Utf8Path, Utf8PathBuf};
 use nix::errno::Errno::EAGAIN;
-use nix::fcntl;
+use nix::fcntl::{Flock, FlockArg};
 
 use crate::crashsafe;
 
-/// A handle to an open and unlocked, but not-yet-written lock file.
+/// A handle to an open and flocked, but not-yet-written lock file.
 /// Returned by [`create_exclusive`].
 #[must_use]
 pub struct UnwrittenLockFile {
     path: Utf8PathBuf,
-    file: fs::File,
+    file: Flock<fs::File>,
 }
 
 /// Returned by [`UnwrittenLockFile::write_content`].
 #[must_use]
-pub struct LockFileGuard(fs::File);
+pub struct LockFileGuard(Flock<fs::File>);
 
 impl Deref for LockFileGuard {
     type Target = fs::File;
@@ -67,17 +66,14 @@ pub fn create_exclusive(lock_file_path: &Utf8Path) -> anyhow::Result<UnwrittenLo
         .open(lock_file_path)
         .context("open lock file")?;
 
-    let res = fcntl::flock(
-        lock_file.as_raw_fd(),
-        fcntl::FlockArg::LockExclusiveNonblock,
-    );
+    let res = Flock::lock(lock_file, FlockArg::LockExclusiveNonblock);
     match res {
-        Ok(()) => Ok(UnwrittenLockFile {
+        Ok(lock_file) => Ok(UnwrittenLockFile {
             path: lock_file_path.to_owned(),
             file: lock_file,
         }),
-        Err(EAGAIN) => anyhow::bail!("file is already locked"),
-        Err(e) => Err(e).context("flock error"),
+        Err((_, EAGAIN)) => anyhow::bail!("file is already locked"),
+        Err((_, e)) => Err(e).context("flock error"),
     }
 }
 
@@ -105,32 +101,37 @@ pub enum LockFileRead {
 /// Check the [`LockFileRead`] variants for details.
 pub fn read_and_hold_lock_file(path: &Utf8Path) -> anyhow::Result<LockFileRead> {
     let res = fs::OpenOptions::new().read(true).open(path);
-    let mut lock_file = match res {
+    let lock_file = match res {
         Ok(f) => f,
         Err(e) => match e.kind() {
             std::io::ErrorKind::NotFound => return Ok(LockFileRead::NotExist),
             _ => return Err(e).context("open lock file"),
         },
     };
-    let res = fcntl::flock(
-        lock_file.as_raw_fd(),
-        fcntl::FlockArg::LockExclusiveNonblock,
-    );
+    let res = Flock::lock(lock_file, FlockArg::LockExclusiveNonblock);
     // We need the content regardless of lock success / failure.
     // But, read it after flock so that, if it succeeded, the content is consistent.
-    let mut content = String::new();
-    lock_file
-        .read_to_string(&mut content)
-        .context("read lock file")?;
     match res {
-        Ok(()) => Ok(LockFileRead::NotHeldByAnyProcess(
-            LockFileGuard(lock_file),
-            content,
-        )),
-        Err(EAGAIN) => Ok(LockFileRead::LockedByOtherProcess {
-            not_locked_file: lock_file,
-            content,
-        }),
-        Err(e) => Err(e).context("flock error"),
+        Ok(mut locked_file) => {
+            let mut content = String::new();
+            locked_file
+                .read_to_string(&mut content)
+                .context("read lock file")?;
+            Ok(LockFileRead::NotHeldByAnyProcess(
+                LockFileGuard(locked_file),
+                content,
+            ))
+        }
+        Err((mut not_locked_file, EAGAIN)) => {
+            let mut content = String::new();
+            not_locked_file
+                .read_to_string(&mut content)
+                .context("read lock file")?;
+            Ok(LockFileRead::LockedByOtherProcess {
+                not_locked_file,
+                content,
+            })
+        }
+        Err((_, e)) => Err(e).context("flock error"),
     }
 }

libs/utils/src/tracing_span_assert.rs

@@ -127,12 +127,12 @@ macro_rules! __check_fields_present {
 
             match check_fields_present0($extractors) {
                 Ok(FoundEverything) => Ok(()),
-                Ok(Unconfigured) if cfg!(test) => {
+                Ok(Unconfigured) if cfg!(feature = "testing") => {
                     // allow unconfigured in tests
                     Ok(())
                 },
                 Ok(Unconfigured) => {
-                    panic!("utils::tracing_span_assert: outside of #[cfg(test)] expected tracing to be configured with tracing_error::ErrorLayer")
+                    panic!(r#"utils::tracing_span_assert: outside of #[cfg(feature = "testing")] expected tracing to be configured with tracing_error::ErrorLayer"#)
                 },
                 Err(missing) => Err(missing)
             }

libs/walproposer/src/walproposer.rs

@@ -1,6 +1,7 @@
 #![allow(clippy::todo)]
 
 use std::ffi::CString;
+use std::str::FromStr;
 
 use postgres_ffi::WAL_SEGMENT_SIZE;
 use utils::id::TenantTimelineId;
@@ -173,6 +174,8 @@ pub struct Config {
     pub ttid: TenantTimelineId,
     /// List of safekeepers in format `host:port`
     pub safekeepers_list: Vec<String>,
+    /// libpq connection info options
+    pub safekeeper_conninfo_options: String,
     /// Safekeeper reconnect timeout in milliseconds
     pub safekeeper_reconnect_timeout: i32,
     /// Safekeeper connection timeout in milliseconds
@@ -202,6 +205,9 @@ impl Wrapper {
             .into_bytes_with_nul();
         assert!(safekeepers_list_vec.len() == safekeepers_list_vec.capacity());
         let safekeepers_list = safekeepers_list_vec.as_mut_ptr() as *mut std::ffi::c_char;
+        let safekeeper_conninfo_options = CString::from_str(&config.safekeeper_conninfo_options)
+            .unwrap()
+            .into_raw();
 
         let callback_data = Box::into_raw(Box::new(api)) as *mut ::std::os::raw::c_void;
 
@@ -209,6 +215,7 @@ impl Wrapper {
             neon_tenant,
             neon_timeline,
             safekeepers_list,
+            safekeeper_conninfo_options,
             safekeeper_reconnect_timeout: config.safekeeper_reconnect_timeout,
             safekeeper_connection_timeout: config.safekeeper_connection_timeout,
             wal_segment_size: WAL_SEGMENT_SIZE as i32, // default 16MB
@@ -576,6 +583,7 @@ mod tests {
         let config = crate::walproposer::Config {
             ttid,
             safekeepers_list: vec!["localhost:5000".to_string()],
+            safekeeper_conninfo_options: String::new(),
             safekeeper_reconnect_timeout: 1000,
             safekeeper_connection_timeout: 10000,
             sync_safekeepers: true,

pageserver/Cargo.toml

@@ -17,49 +17,69 @@ anyhow.workspace = true
 arc-swap.workspace = true
 async-compression.workspace = true
 async-stream.workspace = true
-bit_field.workspace = true
 bincode.workspace = true
+bit_field.workspace = true
 byteorder.workspace = true
 bytes.workspace = true
-camino.workspace = true
 camino-tempfile.workspace = true
+camino.workspace = true
 chrono = { workspace = true, features = ["serde"] }
 clap = { workspace = true, features = ["string"] }
 consumption_metrics.workspace = true
 crc32c.workspace = true
 either.workspace = true
+enum-map.workspace = true
+enumset = { workspace = true, features = ["serde"]}
 fail.workspace = true
 futures.workspace = true
+hashlink.workspace = true
 hex.workspace = true
-humantime.workspace = true
+http-utils.workspace = true
 humantime-serde.workspace = true
+humantime.workspace = true
 hyper0.workspace = true
 itertools.workspace = true
 jsonwebtoken.workspace = true
 md5.workspace = true
+metrics.workspace = true
 nix.workspace = true
-# hack to get the number of worker threads tokio uses
-num_cpus.workspace = true
+num_cpus.workspace = true # hack to get the number of worker threads tokio uses
 num-traits.workspace = true
 once_cell.workspace = true
+pageserver_api.workspace = true
+pageserver_client.workspace = true # for ResponseErrorMessageExt TOOD refactor that
+pageserver_compaction.workspace = true
+pageserver_page_api.workspace = true
+pem.workspace = true
 pin-project-lite.workspace = true
 postgres_backend.workspace = true
+postgres_connection.workspace = true
+postgres_ffi.workspace = true
+postgres_initdb.workspace = true
 postgres-protocol.workspace = true
 postgres-types.workspace = true
-postgres_initdb.workspace = true
+posthog_client_lite.workspace = true
 pprof.workspace = true
+pq_proto.workspace = true
 rand.workspace = true
 range-set-blaze = { version = "0.1.16", features = ["alloc"] }
 regex.workspace = true
+remote_storage.workspace = true
+reqwest.workspace = true
+rpds.workspace = true
 rustls.workspace = true
 scopeguard.workspace = true
 send-future.workspace = true
-serde.workspace = true
 serde_json = { workspace = true, features = ["raw_value"] }
 serde_path_to_error.workspace = true
 serde_with.workspace = true
+serde.workspace = true
+smallvec.workspace = true
+storage_broker.workspace = true
+strum_macros.workspace = true
+strum.workspace = true
 sysinfo.workspace = true
-tokio-tar.workspace = true
+tenant_size_model.workspace = true
 thiserror.workspace = true
 tikv-jemallocator.workspace = true
 tokio = { workspace = true, features = ["process", "sync", "fs", "rt", "io-util", "time"] }
@@ -68,34 +88,19 @@ tokio-io-timeout.workspace = true
 tokio-postgres.workspace = true
 tokio-rustls.workspace = true
 tokio-stream.workspace = true
+tokio-tar.workspace = true
 tokio-util.workspace = true
 toml_edit = { workspace = true, features = [ "serde" ] }
+tonic.workspace = true
+tonic-reflection.workspace = true
 tracing.workspace = true
 tracing-utils.workspace = true
 url.workspace = true
-walkdir.workspace = true
-metrics.workspace = true
-pageserver_api.workspace = true
-pageserver_client.workspace = true # for ResponseErrorMessageExt TOOD refactor that
-pageserver_compaction.workspace = true
-pem.workspace = true
-postgres_connection.workspace = true
-postgres_ffi.workspace = true
-pq_proto.workspace = true
-remote_storage.workspace = true
-storage_broker.workspace = true
-tenant_size_model.workspace = true
-http-utils.workspace = true
 utils.workspace = true
-workspace_hack.workspace = true
-reqwest.workspace = true
-rpds.workspace = true
-enum-map.workspace = true
-enumset = { workspace = true, features = ["serde"]}
-strum.workspace = true
-strum_macros.workspace = true
 wal_decoder.workspace = true
-smallvec.workspace = true
+walkdir.workspace = true
+workspace_hack.workspace = true
+twox-hash.workspace = true
 
 [target.'cfg(target_os = "linux")'.dependencies]
 procfs.workspace = true

pageserver/client/src/mgmt_api.rs

@@ -1,5 +1,6 @@
 use std::collections::HashMap;
 use std::error::Error as _;
+use std::time::Duration;
 
 use bytes::Bytes;
 use detach_ancestor::AncestorDetached;
@@ -819,4 +820,25 @@ impl Client {
             .await
             .map(|resp| resp.status())
     }
+
+    pub async fn activate_post_import(
+        &self,
+        tenant_shard_id: TenantShardId,
+        timeline_id: TimelineId,
+        activate_timeline_timeout: Duration,
+    ) -> Result<TimelineInfo> {
+        let uri = format!(
+            "{}/v1/tenant/{}/timeline/{}/activate_post_import?timeline_activate_timeout_ms={}",
+            self.mgmt_api_endpoint,
+            tenant_shard_id,
+            timeline_id,
+            activate_timeline_timeout.as_millis()
+        );
+
+        self.request(Method::PUT, uri, ())
+            .await?
+            .json()
+            .await
+            .map_err(Error::ReceiveBody)
+    }
 }

pageserver/page_api/Cargo.toml

@@ -0,0 +1,19 @@
+[package]
+name = "pageserver_page_api"
+version = "0.1.0"
+edition.workspace = true
+license.workspace = true
+
+[dependencies]
+bytes.workspace = true
+pageserver_api.workspace = true
+postgres_ffi.workspace = true
+prost.workspace = true
+smallvec.workspace = true
+thiserror.workspace = true
+tonic.workspace = true
+utils.workspace = true
+workspace_hack.workspace = true
+
+[build-dependencies]
+tonic-build.workspace = true

pageserver/page_api/build.rs

@@ -0,0 +1,13 @@
+use std::env;
+use std::path::PathBuf;
+
+/// Generates Rust code from .proto Protobuf schemas, along with a binary file
+/// descriptor set for Protobuf schema reflection.
+fn main() -> Result<(), Box<dyn std::error::Error>> {
+    let out_dir = PathBuf::from(env::var("OUT_DIR")?);
+    tonic_build::configure()
+        .bytes(["."])
+        .file_descriptor_set_path(out_dir.join("page_api_descriptor.bin"))
+        .compile_protos(&["proto/page_service.proto"], &["proto"])
+        .map_err(|err| err.into())
+}

pageserver/page_api/proto/page_service.proto

@@ -0,0 +1,233 @@
+// Page service, presented by pageservers for computes.
+//
+// This is the compute read path. It primarily serves page versions at given
+// LSNs, but also base backups, SLRU segments, and relation metadata.
+//
+// EXPERIMENTAL: this is still under development and subject to change.
+//
+// Request metadata headers:
+// - authorization: JWT token ("Bearer <token>"), if auth is enabled
+// - neon-tenant-id: tenant ID ("7c4a1f9e3bd6470c8f3e21a65bd2e980")
+// - neon-shard-id: shard ID, as <number><count> in hex ("0b10" = shard 11 of 16, 0-based)
+// - neon-timeline-id: timeline ID ("f08c4e9a2d5f76b1e3a7c2d8910f4b3e")
+//
+// The service can be accessed via e.g. grpcurl:
+//
+//    ```
+//    grpcurl \
+//      -plaintext \
+//      -H "neon-tenant-id: 7c4a1f9e3bd6470c8f3e21a65bd2e980" \
+//      -H "neon-shard-id: 0b10" \
+//      -H "neon-timeline-id: f08c4e9a2d5f76b1e3a7c2d8910f4b3e" \
+//      -H "authorization: Bearer $JWT" \
+//      -d '{"read_lsn": {"request_lsn": 1234567890}, "rel": {"spc_oid": 1663, "db_oid": 1234, "rel_number": 5678, "fork_number": 0}}'
+//      localhost:51051 page_api.PageService/CheckRelExists
+//    ```
+//
+// TODO: consider adding neon-compute-mode ("primary", "static", "replica").
+// However, this will require reconnecting when changing modes.
+//
+// TODO: write implementation guidance on
+// - Health checks
+// - Tracing, OpenTelemetry
+// - Compression
+
+syntax = "proto3";
+package page_api;
+
+service PageService {
+  // Returns whether a relation exists.
+  rpc CheckRelExists(CheckRelExistsRequest) returns (CheckRelExistsResponse);
+
+  // Fetches a base backup.
+  rpc GetBaseBackup (GetBaseBackupRequest) returns (stream GetBaseBackupResponseChunk);
+
+  // Returns the total size of a database, as # of bytes.
+  rpc GetDbSize (GetDbSizeRequest) returns (GetDbSizeResponse);
+
+  // Fetches pages.
+  //
+  // This is implemented as a bidirectional streaming RPC for performance. Unary
+  // requests incur costs for e.g. HTTP/2 stream setup, header parsing,
+  // authentication, and so on -- with streaming, we only pay these costs during
+  // the initial stream setup. This ~doubles throughput in benchmarks. Other
+  // RPCs use regular unary requests, since they are not as frequent and
+  // performance-critical, and this simplifies implementation.
+  //
+  // NB: a status response (e.g. errors) will terminate the stream. The stream
+  // may be shared by e.g. multiple Postgres backends, so we should avoid this.
+  // Most errors are therefore sent as GetPageResponse.status instead.
+  rpc GetPages (stream GetPageRequest) returns (stream GetPageResponse);
+
+  // Returns the size of a relation, as # of blocks.
+  rpc GetRelSize (GetRelSizeRequest) returns (GetRelSizeResponse);
+
+  // Fetches an SLRU segment.
+  rpc GetSlruSegment (GetSlruSegmentRequest) returns (GetSlruSegmentResponse);
+}
+
+// The LSN a request should read at.
+message ReadLsn {
+  // The request's read LSN. Required.
+  uint64 request_lsn = 1;
+  // If given, the caller guarantees that the page has not been modified since
+  // this LSN. Must be smaller than or equal to request_lsn. This allows the
+  // Pageserver to serve an old page without waiting for the request LSN to
+  // arrive. Valid for all request types.
+  //
+  // It is undefined behaviour to make a request such that the page was, in
+  // fact, modified between request_lsn and not_modified_since_lsn. The
+  // Pageserver might detect it and return an error, or it might return the old
+  // page version or the new page version. Setting not_modified_since_lsn equal
+  // to request_lsn is always safe, but can lead to unnecessary waiting.
+  uint64 not_modified_since_lsn = 2;
+}
+
+// A relation identifier.
+message RelTag {
+    uint32 spc_oid = 1;
+    uint32 db_oid = 2;
+    uint32 rel_number = 3;
+    uint32 fork_number = 4;
+}
+
+// Checks whether a relation exists, at the given LSN. Only valid on shard 0,
+// other shards will error.
+message CheckRelExistsRequest {
+  ReadLsn read_lsn = 1;
+  RelTag rel = 2;
+}
+
+message CheckRelExistsResponse {
+  bool exists = 1;
+}
+
+// Requests a base backup at a given LSN.
+message GetBaseBackupRequest {
+  // The LSN to fetch a base backup at.
+  ReadLsn read_lsn = 1;
+  // If true, logical replication slots will not be created.
+  bool replica = 2;
+}
+
+// Base backup response chunk, returned as an ordered stream.
+message GetBaseBackupResponseChunk {
+  // A basebackup data chunk. The size is undefined, but bounded by the 4 MB
+  // gRPC message size limit.
+  bytes chunk = 1;
+}
+
+// Requests the size of a database, as # of bytes. Only valid on shard 0, other
+// shards will error.
+message GetDbSizeRequest {
+  ReadLsn read_lsn = 1;
+  uint32 db_oid = 2;
+}
+
+message GetDbSizeResponse {
+  uint64 num_bytes = 1;
+}
+
+// Requests one or more pages.
+message GetPageRequest {
+  // A request ID. Will be included in the response. Should be unique for
+  // in-flight requests on the stream.
+  uint64 request_id = 1;
+  // The request class.
+  GetPageClass request_class = 2;
+  // The LSN to read at.
+  ReadLsn read_lsn = 3;
+  // The relation to read from.
+  RelTag rel = 4;
+  // Page numbers to read. Must belong to the remote shard.
+  //
+  // Multiple pages will be executed as a single batch by the Pageserver,
+  // amortizing layer access costs and parallelizing them. This may increase the
+  // latency of any individual request, but improves the overall latency and
+  // throughput of the batch as a whole.
+  //
+  // TODO: this causes an allocation in the common single-block case. The sender
+  // can use a SmallVec to stack-allocate it, but Prost will always deserialize
+  // into a heap-allocated Vec. Consider optimizing this.
+  //
+  // TODO: we might be able to avoid a sort or something if we mandate that these
+  // are always in order. But we can't currenly rely on this on the server, because
+  // of compatibility with the libpq protocol handler.
+  repeated uint32 block_number = 5;
+}
+
+// A GetPageRequest class. Primarily intended for observability, but may also be
+// used for prioritization in the future.
+enum GetPageClass {
+  // Unknown class. For forwards compatibility: used when the client sends a
+  // class that the server doesn't know about.
+  GET_PAGE_CLASS_UNKNOWN = 0;
+  // A normal request. This is the default.
+  GET_PAGE_CLASS_NORMAL = 1;
+  // A prefetch request. NB: can only be classified on pg < 18.
+  GET_PAGE_CLASS_PREFETCH = 2;
+  // A background request (e.g. vacuum).
+  GET_PAGE_CLASS_BACKGROUND = 3;
+}
+
+// A GetPage response.
+//
+// A batch response will contain all of the requested pages. We could eagerly
+// emit individual pages as soon as they are ready, but on a readv() Postgres
+// holds buffer pool locks on all pages in the batch and we'll only return once
+// the entire batch is ready, so no one can make use of the individual pages.
+message GetPageResponse {
+  // The original request's ID.
+  uint64 request_id = 1;
+  // The response status code.
+  GetPageStatus status = 2;
+  // A string describing the status, if any.
+  string reason = 3;
+  // The 8KB page images, in the same order as the request. Empty if status != OK.
+  repeated bytes page_image = 4;
+}
+
+// A GetPageResponse status code. Since we use a bidirectional stream, we don't
+// want to send errors as gRPC statuses, since this would terminate the stream.
+enum GetPageStatus {
+  // Unknown status. For forwards compatibility: used when the server sends a
+  // status code that the client doesn't know about.
+  GET_PAGE_STATUS_UNKNOWN = 0;
+  // The request was successful.
+  GET_PAGE_STATUS_OK = 1;
+  // The page did not exist. The tenant/timeline/shard has already been
+  // validated during stream setup.
+  GET_PAGE_STATUS_NOT_FOUND = 2;
+  // The request was invalid.
+  GET_PAGE_STATUS_INVALID = 3;
+  // The tenant is rate limited. Slow down and retry later.
+  GET_PAGE_STATUS_SLOW_DOWN = 4;
+  // TODO: consider adding a GET_PAGE_STATUS_LAYER_DOWNLOAD in the case of a
+  // layer download. This could free up the server task to process other
+  // requests while the layer download is in progress.
+}
+
+// Fetches the size of a relation at a given LSN, as # of blocks. Only valid on
+// shard 0, other shards will error.
+message GetRelSizeRequest {
+  ReadLsn read_lsn = 1;
+  RelTag rel = 2;
+}
+
+message GetRelSizeResponse {
+  uint32 num_blocks = 1;
+}
+
+// Requests an SLRU segment. Only valid on shard 0, other shards will error.
+message GetSlruSegmentRequest {
+  ReadLsn read_lsn = 1;
+  uint32 kind = 2;
+  uint32 segno = 3;
+}
+
+// Returns an SLRU segment.
+//
+// These are up 32 pages (256 KB), so we can send them as a single response.
+message GetSlruSegmentResponse {
+  bytes segment = 1;
+}

pageserver/page_api/src/lib.rs

@@ -0,0 +1,23 @@
+//! This crate provides the Pageserver's page API. It contains:
+//!
+//! * proto/page_service.proto: the Protobuf schema for the page API.
+//! * proto: auto-generated Protobuf types for gRPC.
+//!
+//! This crate is used by both the client and the server. Try to keep it slim.
+
+// Code generated by protobuf.
+pub mod proto {
+    tonic::include_proto!("page_api");
+
+    /// File descriptor set for Protobuf schema reflection. This allows using
+    /// e.g. grpcurl with the API.
+    pub const FILE_DESCRIPTOR_SET: &[u8] =
+        tonic::include_file_descriptor_set!("page_api_descriptor");
+
+    pub use page_service_client::PageServiceClient;
+    pub use page_service_server::{PageService, PageServiceServer};
+}
+
+mod model;
+
+pub use model::*;

pageserver/page_api/src/model.rs

@@ -0,0 +1,581 @@
+//! Structs representing the canonical page service API.
+//!
+//! These mirror the autogenerated Protobuf types. The differences are:
+//!
+//! - Types that are in fact required by the API are not Options. The protobuf "required"
+//!   attribute is deprecated and 'prost' marks a lot of members as optional because of that.
+//!   (See <https://github.com/tokio-rs/prost/issues/800> for a gripe on this)
+//!
+//! - Use more precise datatypes, e.g. Lsn and uints shorter than 32 bits.
+//!
+//! - Validate protocol invariants, via try_from() and try_into().
+
+use bytes::Bytes;
+use postgres_ffi::Oid;
+use smallvec::SmallVec;
+// TODO: split out Lsn, RelTag, SlruKind, Oid and other basic types to a separate crate, to avoid
+// pulling in all of their other crate dependencies when building the client.
+use utils::lsn::Lsn;
+
+use crate::proto;
+
+/// A protocol error. Typically returned via try_from() or try_into().
+#[derive(thiserror::Error, Debug)]
+pub enum ProtocolError {
+    #[error("field '{0}' has invalid value '{1}'")]
+    Invalid(&'static str, String),
+    #[error("required field '{0}' is missing")]
+    Missing(&'static str),
+}
+
+impl ProtocolError {
+    /// Helper to generate a new ProtocolError::Invalid for the given field and value.
+    pub fn invalid(field: &'static str, value: impl std::fmt::Debug) -> Self {
+        Self::Invalid(field, format!("{value:?}"))
+    }
+}
+
+/// The LSN a request should read at.
+#[derive(Clone, Copy, Debug)]
+pub struct ReadLsn {
+    /// The request's read LSN.
+    pub request_lsn: Lsn,
+    /// If given, the caller guarantees that the page has not been modified since this LSN. Must be
+    /// smaller than or equal to request_lsn. This allows the Pageserver to serve an old page
+    /// without waiting for the request LSN to arrive. Valid for all request types.
+    ///
+    /// It is undefined behaviour to make a request such that the page was, in fact, modified
+    /// between request_lsn and not_modified_since_lsn. The Pageserver might detect it and return an
+    /// error, or it might return the old page version or the new page version. Setting
+    /// not_modified_since_lsn equal to request_lsn is always safe, but can lead to unnecessary
+    /// waiting.
+    pub not_modified_since_lsn: Option<Lsn>,
+}
+
+impl ReadLsn {
+    /// Validates the ReadLsn.
+    pub fn validate(&self) -> Result<(), ProtocolError> {
+        if self.request_lsn == Lsn::INVALID {
+            return Err(ProtocolError::invalid("request_lsn", self.request_lsn));
+        }
+        if self.not_modified_since_lsn > Some(self.request_lsn) {
+            return Err(ProtocolError::invalid(
+                "not_modified_since_lsn",
+                self.not_modified_since_lsn,
+            ));
+        }
+        Ok(())
+    }
+}
+
+impl TryFrom<proto::ReadLsn> for ReadLsn {
+    type Error = ProtocolError;
+
+    fn try_from(pb: proto::ReadLsn) -> Result<Self, Self::Error> {
+        let read_lsn = Self {
+            request_lsn: Lsn(pb.request_lsn),
+            not_modified_since_lsn: match pb.not_modified_since_lsn {
+                0 => None,
+                lsn => Some(Lsn(lsn)),
+            },
+        };
+        read_lsn.validate()?;
+        Ok(read_lsn)
+    }
+}
+
+impl TryFrom<ReadLsn> for proto::ReadLsn {
+    type Error = ProtocolError;
+
+    fn try_from(read_lsn: ReadLsn) -> Result<Self, Self::Error> {
+        read_lsn.validate()?;
+        Ok(Self {
+            request_lsn: read_lsn.request_lsn.0,
+            not_modified_since_lsn: read_lsn.not_modified_since_lsn.unwrap_or_default().0,
+        })
+    }
+}
+
+// RelTag is defined in pageserver_api::reltag.
+pub type RelTag = pageserver_api::reltag::RelTag;
+
+impl TryFrom<proto::RelTag> for RelTag {
+    type Error = ProtocolError;
+
+    fn try_from(pb: proto::RelTag) -> Result<Self, Self::Error> {
+        Ok(Self {
+            spcnode: pb.spc_oid,
+            dbnode: pb.db_oid,
+            relnode: pb.rel_number,
+            forknum: pb
+                .fork_number
+                .try_into()
+                .map_err(|_| ProtocolError::invalid("fork_number", pb.fork_number))?,
+        })
+    }
+}
+
+impl From<RelTag> for proto::RelTag {
+    fn from(rel_tag: RelTag) -> Self {
+        Self {
+            spc_oid: rel_tag.spcnode,
+            db_oid: rel_tag.dbnode,
+            rel_number: rel_tag.relnode,
+            fork_number: rel_tag.forknum as u32,
+        }
+    }
+}
+
+/// Checks whether a relation exists, at the given LSN. Only valid on shard 0, other shards error.
+#[derive(Clone, Copy, Debug)]
+pub struct CheckRelExistsRequest {
+    pub read_lsn: ReadLsn,
+    pub rel: RelTag,
+}
+
+impl TryFrom<proto::CheckRelExistsRequest> for CheckRelExistsRequest {
+    type Error = ProtocolError;
+
+    fn try_from(pb: proto::CheckRelExistsRequest) -> Result<Self, Self::Error> {
+        Ok(Self {
+            read_lsn: pb
+                .read_lsn
+                .ok_or(ProtocolError::Missing("read_lsn"))?
+                .try_into()?,
+            rel: pb.rel.ok_or(ProtocolError::Missing("rel"))?.try_into()?,
+        })
+    }
+}
+
+pub type CheckRelExistsResponse = bool;
+
+impl From<proto::CheckRelExistsResponse> for CheckRelExistsResponse {
+    fn from(pb: proto::CheckRelExistsResponse) -> Self {
+        pb.exists
+    }
+}
+
+impl From<CheckRelExistsResponse> for proto::CheckRelExistsResponse {
+    fn from(exists: CheckRelExistsResponse) -> Self {
+        Self { exists }
+    }
+}
+
+/// Requests a base backup at a given LSN.
+#[derive(Clone, Copy, Debug)]
+pub struct GetBaseBackupRequest {
+    /// The LSN to fetch a base backup at.
+    pub read_lsn: ReadLsn,
+    /// If true, logical replication slots will not be created.
+    pub replica: bool,
+}
+
+impl TryFrom<proto::GetBaseBackupRequest> for GetBaseBackupRequest {
+    type Error = ProtocolError;
+
+    fn try_from(pb: proto::GetBaseBackupRequest) -> Result<Self, Self::Error> {
+        Ok(Self {
+            read_lsn: pb
+                .read_lsn
+                .ok_or(ProtocolError::Missing("read_lsn"))?
+                .try_into()?,
+            replica: pb.replica,
+        })
+    }
+}
+
+impl TryFrom<GetBaseBackupRequest> for proto::GetBaseBackupRequest {
+    type Error = ProtocolError;
+
+    fn try_from(request: GetBaseBackupRequest) -> Result<Self, Self::Error> {
+        Ok(Self {
+            read_lsn: Some(request.read_lsn.try_into()?),
+            replica: request.replica,
+        })
+    }
+}
+
+pub type GetBaseBackupResponseChunk = Bytes;
+
+impl TryFrom<proto::GetBaseBackupResponseChunk> for GetBaseBackupResponseChunk {
+    type Error = ProtocolError;
+
+    fn try_from(pb: proto::GetBaseBackupResponseChunk) -> Result<Self, Self::Error> {
+        if pb.chunk.is_empty() {
+            return Err(ProtocolError::Missing("chunk"));
+        }
+        Ok(pb.chunk)
+    }
+}
+
+impl TryFrom<GetBaseBackupResponseChunk> for proto::GetBaseBackupResponseChunk {
+    type Error = ProtocolError;
+
+    fn try_from(chunk: GetBaseBackupResponseChunk) -> Result<Self, Self::Error> {
+        if chunk.is_empty() {
+            return Err(ProtocolError::Missing("chunk"));
+        }
+        Ok(Self { chunk })
+    }
+}
+
+/// Requests the size of a database, as # of bytes. Only valid on shard 0, other shards will error.
+#[derive(Clone, Copy, Debug)]
+pub struct GetDbSizeRequest {
+    pub read_lsn: ReadLsn,
+    pub db_oid: Oid,
+}
+
+impl TryFrom<proto::GetDbSizeRequest> for GetDbSizeRequest {
+    type Error = ProtocolError;
+
+    fn try_from(pb: proto::GetDbSizeRequest) -> Result<Self, Self::Error> {
+        Ok(Self {
+            read_lsn: pb
+                .read_lsn
+                .ok_or(ProtocolError::Missing("read_lsn"))?
+                .try_into()?,
+            db_oid: pb.db_oid,
+        })
+    }
+}
+
+impl TryFrom<GetDbSizeRequest> for proto::GetDbSizeRequest {
+    type Error = ProtocolError;
+
+    fn try_from(request: GetDbSizeRequest) -> Result<Self, Self::Error> {
+        Ok(Self {
+            read_lsn: Some(request.read_lsn.try_into()?),
+            db_oid: request.db_oid,
+        })
+    }
+}
+
+pub type GetDbSizeResponse = u64;
+
+impl From<proto::GetDbSizeResponse> for GetDbSizeResponse {
+    fn from(pb: proto::GetDbSizeResponse) -> Self {
+        pb.num_bytes
+    }
+}
+
+impl From<GetDbSizeResponse> for proto::GetDbSizeResponse {
+    fn from(num_bytes: GetDbSizeResponse) -> Self {
+        Self { num_bytes }
+    }
+}
+
+/// Requests one or more pages.
+#[derive(Clone, Debug)]
+pub struct GetPageRequest {
+    /// A request ID. Will be included in the response. Should be unique for in-flight requests on
+    /// the stream.
+    pub request_id: RequestID,
+    /// The request class.
+    pub request_class: GetPageClass,
+    /// The LSN to read at.
+    pub read_lsn: ReadLsn,
+    /// The relation to read from.
+    pub rel: RelTag,
+    /// Page numbers to read. Must belong to the remote shard.
+    ///
+    /// Multiple pages will be executed as a single batch by the Pageserver, amortizing layer access
+    /// costs and parallelizing them. This may increase the latency of any individual request, but
+    /// improves the overall latency and throughput of the batch as a whole.
+    pub block_numbers: SmallVec<[u32; 1]>,
+}
+
+impl TryFrom<proto::GetPageRequest> for GetPageRequest {
+    type Error = ProtocolError;
+
+    fn try_from(pb: proto::GetPageRequest) -> Result<Self, Self::Error> {
+        if pb.block_number.is_empty() {
+            return Err(ProtocolError::Missing("block_number"));
+        }
+        Ok(Self {
+            request_id: pb.request_id,
+            request_class: pb.request_class.into(),
+            read_lsn: pb
+                .read_lsn
+                .ok_or(ProtocolError::Missing("read_lsn"))?
+                .try_into()?,
+            rel: pb.rel.ok_or(ProtocolError::Missing("rel"))?.try_into()?,
+            block_numbers: pb.block_number.into(),
+        })
+    }
+}
+
+impl TryFrom<GetPageRequest> for proto::GetPageRequest {
+    type Error = ProtocolError;
+
+    fn try_from(request: GetPageRequest) -> Result<Self, Self::Error> {
+        if request.block_numbers.is_empty() {
+            return Err(ProtocolError::Missing("block_number"));
+        }
+        Ok(Self {
+            request_id: request.request_id,
+            request_class: request.request_class.into(),
+            read_lsn: Some(request.read_lsn.try_into()?),
+            rel: Some(request.rel.into()),
+            block_number: request.block_numbers.into_vec(),
+        })
+    }
+}
+
+/// A GetPage request ID.
+pub type RequestID = u64;
+
+/// A GetPage request class.
+#[derive(Clone, Copy, Debug)]
+pub enum GetPageClass {
+    /// Unknown status. For backwards compatibility: used when an older client version sends a class
+    /// that a newer server version has removed.
+    Unknown,
+    /// A normal request. This is the default.
+    Normal,
+    /// A prefetch request. NB: can only be classified on pg < 18.
+    Prefetch,
+    /// A background request (e.g. vacuum).
+    Background,
+}
+
+impl From<proto::GetPageClass> for GetPageClass {
+    fn from(pb: proto::GetPageClass) -> Self {
+        match pb {
+            proto::GetPageClass::Unknown => Self::Unknown,
+            proto::GetPageClass::Normal => Self::Normal,
+            proto::GetPageClass::Prefetch => Self::Prefetch,
+            proto::GetPageClass::Background => Self::Background,
+        }
+    }
+}
+
+impl From<i32> for GetPageClass {
+    fn from(class: i32) -> Self {
+        proto::GetPageClass::try_from(class)
+            .unwrap_or(proto::GetPageClass::Unknown)
+            .into()
+    }
+}
+
+impl From<GetPageClass> for proto::GetPageClass {
+    fn from(class: GetPageClass) -> Self {
+        match class {
+            GetPageClass::Unknown => Self::Unknown,
+            GetPageClass::Normal => Self::Normal,
+            GetPageClass::Prefetch => Self::Prefetch,
+            GetPageClass::Background => Self::Background,
+        }
+    }
+}
+
+impl From<GetPageClass> for i32 {
+    fn from(class: GetPageClass) -> Self {
+        proto::GetPageClass::from(class).into()
+    }
+}
+
+/// A GetPage response.
+///
+/// A batch response will contain all of the requested pages. We could eagerly emit individual pages
+/// as soon as they are ready, but on a readv() Postgres holds buffer pool locks on all pages in the
+/// batch and we'll only return once the entire batch is ready, so no one can make use of the
+/// individual pages.
+#[derive(Clone, Debug)]
+pub struct GetPageResponse {
+    /// The original request's ID.
+    pub request_id: RequestID,
+    /// The response status code.
+    pub status: GetPageStatus,
+    /// A string describing the status, if any.
+    pub reason: Option<String>,
+    /// The 8KB page images, in the same order as the request. Empty if status != OK.
+    pub page_images: SmallVec<[Bytes; 1]>,
+}
+
+impl From<proto::GetPageResponse> for GetPageResponse {
+    fn from(pb: proto::GetPageResponse) -> Self {
+        Self {
+            request_id: pb.request_id,
+            status: pb.status.into(),
+            reason: Some(pb.reason).filter(|r| !r.is_empty()),
+            page_images: pb.page_image.into(),
+        }
+    }
+}
+
+impl From<GetPageResponse> for proto::GetPageResponse {
+    fn from(response: GetPageResponse) -> Self {
+        Self {
+            request_id: response.request_id,
+            status: response.status.into(),
+            reason: response.reason.unwrap_or_default(),
+            page_image: response.page_images.into_vec(),
+        }
+    }
+}
+
+/// A GetPage response status.
+#[derive(Clone, Copy, Debug)]
+pub enum GetPageStatus {
+    /// Unknown status. For forwards compatibility: used when an older client version receives a new
+    /// status code from a newer server version.
+    Unknown,
+    /// The request was successful.
+    Ok,
+    /// The page did not exist. The tenant/timeline/shard has already been validated during stream
+    /// setup.
+    NotFound,
+    /// The request was invalid.
+    Invalid,
+    /// The tenant is rate limited. Slow down and retry later.
+    SlowDown,
+}
+
+impl From<proto::GetPageStatus> for GetPageStatus {
+    fn from(pb: proto::GetPageStatus) -> Self {
+        match pb {
+            proto::GetPageStatus::Unknown => Self::Unknown,
+            proto::GetPageStatus::Ok => Self::Ok,
+            proto::GetPageStatus::NotFound => Self::NotFound,
+            proto::GetPageStatus::Invalid => Self::Invalid,
+            proto::GetPageStatus::SlowDown => Self::SlowDown,
+        }
+    }
+}
+
+impl From<i32> for GetPageStatus {
+    fn from(status: i32) -> Self {
+        proto::GetPageStatus::try_from(status)
+            .unwrap_or(proto::GetPageStatus::Unknown)
+            .into()
+    }
+}
+
+impl From<GetPageStatus> for proto::GetPageStatus {
+    fn from(status: GetPageStatus) -> Self {
+        match status {
+            GetPageStatus::Unknown => Self::Unknown,
+            GetPageStatus::Ok => Self::Ok,
+            GetPageStatus::NotFound => Self::NotFound,
+            GetPageStatus::Invalid => Self::Invalid,
+            GetPageStatus::SlowDown => Self::SlowDown,
+        }
+    }
+}
+
+impl From<GetPageStatus> for i32 {
+    fn from(status: GetPageStatus) -> Self {
+        proto::GetPageStatus::from(status).into()
+    }
+}
+
+// Fetches the size of a relation at a given LSN, as # of blocks. Only valid on shard 0, other
+// shards will error.
+pub struct GetRelSizeRequest {
+    pub read_lsn: ReadLsn,
+    pub rel: RelTag,
+}
+
+impl TryFrom<proto::GetRelSizeRequest> for GetRelSizeRequest {
+    type Error = ProtocolError;
+
+    fn try_from(proto: proto::GetRelSizeRequest) -> Result<Self, Self::Error> {
+        Ok(Self {
+            read_lsn: proto
+                .read_lsn
+                .ok_or(ProtocolError::Missing("read_lsn"))?
+                .try_into()?,
+            rel: proto.rel.ok_or(ProtocolError::Missing("rel"))?.try_into()?,
+        })
+    }
+}
+
+impl TryFrom<GetRelSizeRequest> for proto::GetRelSizeRequest {
+    type Error = ProtocolError;
+
+    fn try_from(request: GetRelSizeRequest) -> Result<Self, Self::Error> {
+        Ok(Self {
+            read_lsn: Some(request.read_lsn.try_into()?),
+            rel: Some(request.rel.into()),
+        })
+    }
+}
+
+pub type GetRelSizeResponse = u32;
+
+impl From<proto::GetRelSizeResponse> for GetRelSizeResponse {
+    fn from(proto: proto::GetRelSizeResponse) -> Self {
+        proto.num_blocks
+    }
+}
+
+impl From<GetRelSizeResponse> for proto::GetRelSizeResponse {
+    fn from(num_blocks: GetRelSizeResponse) -> Self {
+        Self { num_blocks }
+    }
+}
+
+/// Requests an SLRU segment. Only valid on shard 0, other shards will error.
+pub struct GetSlruSegmentRequest {
+    pub read_lsn: ReadLsn,
+    pub kind: SlruKind,
+    pub segno: u32,
+}
+
+impl TryFrom<proto::GetSlruSegmentRequest> for GetSlruSegmentRequest {
+    type Error = ProtocolError;
+
+    fn try_from(pb: proto::GetSlruSegmentRequest) -> Result<Self, Self::Error> {
+        Ok(Self {
+            read_lsn: pb
+                .read_lsn
+                .ok_or(ProtocolError::Missing("read_lsn"))?
+                .try_into()?,
+            kind: u8::try_from(pb.kind)
+                .ok()
+                .and_then(SlruKind::from_repr)
+                .ok_or_else(|| ProtocolError::invalid("slru_kind", pb.kind))?,
+            segno: pb.segno,
+        })
+    }
+}
+
+impl TryFrom<GetSlruSegmentRequest> for proto::GetSlruSegmentRequest {
+    type Error = ProtocolError;
+
+    fn try_from(request: GetSlruSegmentRequest) -> Result<Self, Self::Error> {
+        Ok(Self {
+            read_lsn: Some(request.read_lsn.try_into()?),
+            kind: request.kind as u32,
+            segno: request.segno,
+        })
+    }
+}
+
+pub type GetSlruSegmentResponse = Bytes;
+
+impl TryFrom<proto::GetSlruSegmentResponse> for GetSlruSegmentResponse {
+    type Error = ProtocolError;
+
+    fn try_from(pb: proto::GetSlruSegmentResponse) -> Result<Self, Self::Error> {
+        if pb.segment.is_empty() {
+            return Err(ProtocolError::Missing("segment"));
+        }
+        Ok(pb.segment)
+    }
+}
+
+impl TryFrom<GetSlruSegmentResponse> for proto::GetSlruSegmentResponse {
+    type Error = ProtocolError;
+
+    fn try_from(segment: GetSlruSegmentResponse) -> Result<Self, Self::Error> {
+        if segment.is_empty() {
+            return Err(ProtocolError::Missing("segment"));
+        }
+        Ok(Self { segment })
+    }
+}
+
+// SlruKind is defined in pageserver_api::reltag.
+pub type SlruKind = pageserver_api::reltag::SlruKind;

pageserver/pagebench/src/cmd/getpage_latest_lsn.rs

@@ -65,6 +65,9 @@ pub(crate) struct Args {
     #[clap(long, default_value = "1")]
     queue_depth: NonZeroUsize,
 
+    #[clap(long)]
+    only_relnode: Option<u32>,
+
     targets: Option<Vec<TenantTimelineId>>,
 }
 
@@ -206,7 +209,12 @@ async fn main_impl(
                     for r in partitioning.keys.ranges.iter() {
                         let mut i = r.start;
                         while i != r.end {
-                            if i.is_rel_block_key() {
+                            let mut include = true;
+                            include &= i.is_rel_block_key();
+                            if let Some(only_relnode) = args.only_relnode {
+                                include &= i.is_rel_block_of_rel(only_relnode);
+                            }
+                            if include {
                                 filtered.add_key(i);
                             }
                             i = i.next();

pageserver/src/basebackup.rs

@@ -144,7 +144,7 @@ where
         replica,
         ctx,
         io_concurrency: IoConcurrency::spawn_from_conf(
-            timeline.conf,
+            timeline.conf.get_vectored_concurrent_io,
             timeline
                 .gate
                 .enter()
@@ -343,7 +343,7 @@ where
             // Gather non-relational files from object storage pages.
             let slru_partitions = self
                 .timeline
-                .get_slru_keyspace(Version::Lsn(self.lsn), self.ctx)
+                .get_slru_keyspace(Version::at(self.lsn), self.ctx)
                 .await?
                 .partition(
                     self.timeline.get_shard_identity(),
@@ -378,7 +378,7 @@ where
             // Otherwise only include init forks of unlogged relations.
             let rels = self
                 .timeline
-                .list_rels(spcnode, dbnode, Version::Lsn(self.lsn), self.ctx)
+                .list_rels(spcnode, dbnode, Version::at(self.lsn), self.ctx)
                 .await?;
             for &rel in rels.iter() {
                 // Send init fork as main fork to provide well formed empty
@@ -517,7 +517,7 @@ where
     async fn add_rel(&mut self, src: RelTag, dst: RelTag) -> Result<(), BasebackupError> {
         let nblocks = self
             .timeline
-            .get_rel_size(src, Version::Lsn(self.lsn), self.ctx)
+            .get_rel_size(src, Version::at(self.lsn), self.ctx)
             .await?;
 
         // If the relation is empty, create an empty file
@@ -577,7 +577,7 @@ where
         let relmap_img = if has_relmap_file {
             let img = self
                 .timeline
-                .get_relmap_file(spcnode, dbnode, Version::Lsn(self.lsn), self.ctx)
+                .get_relmap_file(spcnode, dbnode, Version::at(self.lsn), self.ctx)
                 .await?;
 
             if img.len()
@@ -631,7 +631,7 @@ where
             if !has_relmap_file
                 && self
                     .timeline
-                    .list_rels(spcnode, dbnode, Version::Lsn(self.lsn), self.ctx)
+                    .list_rels(spcnode, dbnode, Version::at(self.lsn), self.ctx)
                     .await?
                     .is_empty()
             {

pageserver/src/basebackup_cache.rs

@@ -0,0 +1,518 @@
+use std::{collections::HashMap, sync::Arc};
+
+use async_compression::tokio::write::GzipEncoder;
+use camino::{Utf8Path, Utf8PathBuf};
+use metrics::core::{AtomicU64, GenericCounter};
+use pageserver_api::{config::BasebackupCacheConfig, models::TenantState};
+use tokio::{
+    io::{AsyncWriteExt, BufWriter},
+    sync::mpsc::{UnboundedReceiver, UnboundedSender},
+};
+use tokio_util::sync::CancellationToken;
+use utils::{
+    id::{TenantId, TenantTimelineId, TimelineId},
+    lsn::Lsn,
+    shard::TenantShardId,
+};
+
+use crate::{
+    basebackup::send_basebackup_tarball,
+    context::{DownloadBehavior, RequestContext},
+    metrics::{BASEBACKUP_CACHE_ENTRIES, BASEBACKUP_CACHE_PREPARE, BASEBACKUP_CACHE_READ},
+    task_mgr::TaskKind,
+    tenant::{
+        Timeline,
+        mgr::{TenantManager, TenantSlot},
+    },
+};
+
+pub struct BasebackupPrepareRequest {
+    pub tenant_shard_id: TenantShardId,
+    pub timeline_id: TimelineId,
+    pub lsn: Lsn,
+}
+
+pub type BasebackupPrepareSender = UnboundedSender<BasebackupPrepareRequest>;
+pub type BasebackupPrepareReceiver = UnboundedReceiver<BasebackupPrepareRequest>;
+
+type BasebackupRemoveEntrySender = UnboundedSender<Utf8PathBuf>;
+type BasebackupRemoveEntryReceiver = UnboundedReceiver<Utf8PathBuf>;
+
+/// BasebackupCache stores cached basebackup archives for timelines on local disk.
+///
+/// The main purpose of this cache is to speed up the startup process of compute nodes
+/// after scaling to zero.
+/// Thus, the basebackup is stored only for the latest LSN of the timeline and with
+/// fixed set of parameters (gzip=true, full_backup=false, replica=false, prev_lsn=none).
+///
+/// The cache receives prepare requests through the `BasebackupPrepareSender` channel,
+/// generates a basebackup from the timeline in the background, and stores it on disk.
+///
+/// Basebackup requests are pretty rare. We expect ~thousands of entries in the cache
+/// and ~1 RPS for get requests.
+pub struct BasebackupCache {
+    data_dir: Utf8PathBuf,
+    config: BasebackupCacheConfig,
+    tenant_manager: Arc<TenantManager>,
+    remove_entry_sender: BasebackupRemoveEntrySender,
+
+    entries: std::sync::Mutex<HashMap<TenantTimelineId, Lsn>>,
+
+    cancel: CancellationToken,
+
+    read_hit_count: GenericCounter<AtomicU64>,
+    read_miss_count: GenericCounter<AtomicU64>,
+    read_err_count: GenericCounter<AtomicU64>,
+
+    prepare_ok_count: GenericCounter<AtomicU64>,
+    prepare_skip_count: GenericCounter<AtomicU64>,
+    prepare_err_count: GenericCounter<AtomicU64>,
+}
+
+impl BasebackupCache {
+    /// Creates a BasebackupCache and spawns the background task.
+    /// The initialization of the cache is performed in the background and does not
+    /// block the caller. The cache will return `None` for any get requests until
+    /// initialization is complete.
+    pub fn spawn(
+        runtime_handle: &tokio::runtime::Handle,
+        data_dir: Utf8PathBuf,
+        config: Option<BasebackupCacheConfig>,
+        prepare_receiver: BasebackupPrepareReceiver,
+        tenant_manager: Arc<TenantManager>,
+        cancel: CancellationToken,
+    ) -> Arc<Self> {
+        let (remove_entry_sender, remove_entry_receiver) = tokio::sync::mpsc::unbounded_channel();
+
+        let enabled = config.is_some();
+
+        let cache = Arc::new(BasebackupCache {
+            data_dir,
+            config: config.unwrap_or_default(),
+            tenant_manager,
+            remove_entry_sender,
+
+            entries: std::sync::Mutex::new(HashMap::new()),
+
+            cancel,
+
+            read_hit_count: BASEBACKUP_CACHE_READ.with_label_values(&["hit"]),
+            read_miss_count: BASEBACKUP_CACHE_READ.with_label_values(&["miss"]),
+            read_err_count: BASEBACKUP_CACHE_READ.with_label_values(&["error"]),
+
+            prepare_ok_count: BASEBACKUP_CACHE_PREPARE.with_label_values(&["ok"]),
+            prepare_skip_count: BASEBACKUP_CACHE_PREPARE.with_label_values(&["skip"]),
+            prepare_err_count: BASEBACKUP_CACHE_PREPARE.with_label_values(&["error"]),
+        });
+
+        if enabled {
+            runtime_handle.spawn(
+                cache
+                    .clone()
+                    .background(prepare_receiver, remove_entry_receiver),
+            );
+        }
+
+        cache
+    }
+
+    /// Gets a basebackup entry from the cache.
+    /// If the entry is found, opens a file with the basebackup archive and returns it.
+    /// The open file descriptor will prevent the file system from deleting the file
+    /// even if the entry is removed from the cache in the background.
+    pub async fn get(
+        &self,
+        tenant_id: TenantId,
+        timeline_id: TimelineId,
+        lsn: Lsn,
+    ) -> Option<tokio::fs::File> {
+        // Fast path. Check if the entry exists using the in-memory state.
+        let tti = TenantTimelineId::new(tenant_id, timeline_id);
+        if self.entries.lock().unwrap().get(&tti) != Some(&lsn) {
+            self.read_miss_count.inc();
+            return None;
+        }
+
+        let path = self.entry_path(tenant_id, timeline_id, lsn);
+
+        match tokio::fs::File::open(path).await {
+            Ok(file) => {
+                self.read_hit_count.inc();
+                Some(file)
+            }
+            Err(e) => {
+                if e.kind() == std::io::ErrorKind::NotFound {
+                    // We may end up here if the basebackup was concurrently removed by the cleanup task.
+                    self.read_miss_count.inc();
+                } else {
+                    self.read_err_count.inc();
+                    tracing::warn!("Unexpected error opening basebackup cache file: {:?}", e);
+                }
+                None
+            }
+        }
+    }
+
+    // Private methods.
+
+    fn entry_filename(tenant_id: TenantId, timeline_id: TimelineId, lsn: Lsn) -> String {
+        // The default format for LSN is 0/ABCDEF.
+        // The backslash is not filename friendly, so serialize it as plain hex.
+        let lsn = lsn.0;
+        format!("basebackup_{tenant_id}_{timeline_id}_{lsn:016X}.tar.gz")
+    }
+
+    fn entry_path(&self, tenant_id: TenantId, timeline_id: TimelineId, lsn: Lsn) -> Utf8PathBuf {
+        self.data_dir
+            .join(Self::entry_filename(tenant_id, timeline_id, lsn))
+    }
+
+    fn entry_tmp_path(
+        &self,
+        tenant_id: TenantId,
+        timeline_id: TimelineId,
+        lsn: Lsn,
+    ) -> Utf8PathBuf {
+        self.data_dir
+            .join("tmp")
+            .join(Self::entry_filename(tenant_id, timeline_id, lsn))
+    }
+
+    fn parse_entry_filename(filename: &str) -> Option<(TenantId, TimelineId, Lsn)> {
+        let parts: Vec<&str> = filename
+            .strip_prefix("basebackup_")?
+            .strip_suffix(".tar.gz")?
+            .split('_')
+            .collect();
+        if parts.len() != 3 {
+            return None;
+        }
+        let tenant_id = parts[0].parse::<TenantId>().ok()?;
+        let timeline_id = parts[1].parse::<TimelineId>().ok()?;
+        let lsn = Lsn(u64::from_str_radix(parts[2], 16).ok()?);
+
+        Some((tenant_id, timeline_id, lsn))
+    }
+
+    async fn cleanup(&self) -> anyhow::Result<()> {
+        // Cleanup tmp directory.
+        let tmp_dir = self.data_dir.join("tmp");
+        let mut tmp_dir = tokio::fs::read_dir(&tmp_dir).await?;
+        while let Some(dir_entry) = tmp_dir.next_entry().await? {
+            if let Err(e) = tokio::fs::remove_file(dir_entry.path()).await {
+                tracing::warn!("Failed to remove basebackup cache tmp file: {:#}", e);
+            }
+        }
+
+        // Remove outdated entries.
+        let entries_old = self.entries.lock().unwrap().clone();
+        let mut entries_new = HashMap::new();
+        for (tenant_shard_id, tenant_slot) in self.tenant_manager.list() {
+            if !tenant_shard_id.is_shard_zero() {
+                continue;
+            }
+            let TenantSlot::Attached(tenant) = tenant_slot else {
+                continue;
+            };
+            let tenant_id = tenant_shard_id.tenant_id;
+
+            for timeline in tenant.list_timelines() {
+                let tti = TenantTimelineId::new(tenant_id, timeline.timeline_id);
+                if let Some(&entry_lsn) = entries_old.get(&tti) {
+                    if timeline.get_last_record_lsn() <= entry_lsn {
+                        entries_new.insert(tti, entry_lsn);
+                    }
+                }
+            }
+        }
+
+        for (&tti, &lsn) in entries_old.iter() {
+            if !entries_new.contains_key(&tti) {
+                self.remove_entry_sender
+                    .send(self.entry_path(tti.tenant_id, tti.timeline_id, lsn))
+                    .unwrap();
+            }
+        }
+
+        BASEBACKUP_CACHE_ENTRIES.set(entries_new.len() as i64);
+        *self.entries.lock().unwrap() = entries_new;
+
+        Ok(())
+    }
+
+    async fn on_startup(&self) -> anyhow::Result<()> {
+        // Create data_dir and tmp directory if they do not exist.
+        tokio::fs::create_dir_all(&self.data_dir.join("tmp"))
+            .await
+            .map_err(|e| {
+                anyhow::anyhow!(
+                    "Failed to create basebackup cache data_dir {:?}: {:?}",
+                    self.data_dir,
+                    e
+                )
+            })?;
+
+        // Read existing entries from the data_dir and add them to in-memory state.
+        let mut entries = HashMap::new();
+        let mut dir = tokio::fs::read_dir(&self.data_dir).await?;
+        while let Some(dir_entry) = dir.next_entry().await? {
+            let filename = dir_entry.file_name();
+
+            if filename == "tmp" {
+                // Skip the tmp directory.
+                continue;
+            }
+
+            let parsed = Self::parse_entry_filename(filename.to_string_lossy().as_ref());
+            let Some((tenant_id, timeline_id, lsn)) = parsed else {
+                tracing::warn!("Invalid basebackup cache file name: {:?}", filename);
+                continue;
+            };
+
+            let tti = TenantTimelineId::new(tenant_id, timeline_id);
+
+            use std::collections::hash_map::Entry::*;
+
+            match entries.entry(tti) {
+                Occupied(mut entry) => {
+                    let entry_lsn = *entry.get();
+                    // Leave only the latest entry, remove the old one.
+                    if lsn < entry_lsn {
+                        self.remove_entry_sender.send(self.entry_path(
+                            tenant_id,
+                            timeline_id,
+                            lsn,
+                        ))?;
+                    } else if lsn > entry_lsn {
+                        self.remove_entry_sender.send(self.entry_path(
+                            tenant_id,
+                            timeline_id,
+                            entry_lsn,
+                        ))?;
+                        entry.insert(lsn);
+                    } else {
+                        // Two different filenames parsed to the same timline_id and LSN.
+                        // Should never happen.
+                        return Err(anyhow::anyhow!(
+                            "Duplicate basebackup cache entry with the same LSN: {:?}",
+                            filename
+                        ));
+                    }
+                }
+                Vacant(entry) => {
+                    entry.insert(lsn);
+                }
+            }
+        }
+
+        BASEBACKUP_CACHE_ENTRIES.set(entries.len() as i64);
+        *self.entries.lock().unwrap() = entries;
+
+        Ok(())
+    }
+
+    async fn background(
+        self: Arc<Self>,
+        mut prepare_receiver: BasebackupPrepareReceiver,
+        mut remove_entry_receiver: BasebackupRemoveEntryReceiver,
+    ) {
+        // Panic in the background is a safe fallback.
+        // It will drop receivers and the cache will be effectively disabled.
+        self.on_startup()
+            .await
+            .expect("Failed to initialize basebackup cache");
+
+        let mut cleanup_ticker = tokio::time::interval(self.config.cleanup_period);
+        cleanup_ticker.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Skip);
+
+        loop {
+            tokio::select! {
+                Some(req) = prepare_receiver.recv() => {
+                    if let Err(err) = self.prepare_basebackup(
+                        req.tenant_shard_id,
+                        req.timeline_id,
+                        req.lsn,
+                    ).await {
+                        tracing::info!("Failed to prepare basebackup: {:#}", err);
+                        self.prepare_err_count.inc();
+                        continue;
+                    }
+                }
+                Some(req) = remove_entry_receiver.recv() => {
+                    if let Err(e) = tokio::fs::remove_file(req).await {
+                        tracing::warn!("Failed to remove basebackup cache file: {:#}", e);
+                    }
+                }
+                _ = cleanup_ticker.tick() => {
+                    self.cleanup().await.unwrap_or_else(|e| {
+                        tracing::warn!("Failed to clean up basebackup cache: {:#}", e);
+                    });
+                }
+                _ = self.cancel.cancelled() => {
+                    tracing::info!("BasebackupCache background task cancelled");
+                    break;
+                }
+            }
+        }
+    }
+
+    /// Prepare a basebackup for the given timeline.
+    ///
+    /// If the basebackup already exists with a higher LSN or the timeline already
+    /// has a higher last_record_lsn, skip the preparation.
+    ///
+    /// The basebackup is prepared in a temporary directory and then moved to the final
+    /// location to make the operation atomic.
+    async fn prepare_basebackup(
+        &self,
+        tenant_shard_id: TenantShardId,
+        timeline_id: TimelineId,
+        req_lsn: Lsn,
+    ) -> anyhow::Result<()> {
+        tracing::info!(
+            tenant_id = %tenant_shard_id.tenant_id,
+            %timeline_id,
+            %req_lsn,
+            "Preparing basebackup for timeline",
+        );
+
+        let tti = TenantTimelineId::new(tenant_shard_id.tenant_id, timeline_id);
+
+        {
+            let entries = self.entries.lock().unwrap();
+            if let Some(&entry_lsn) = entries.get(&tti) {
+                if entry_lsn >= req_lsn {
+                    tracing::info!(
+                        %timeline_id,
+                        %req_lsn,
+                        %entry_lsn,
+                        "Basebackup entry already exists for timeline with higher LSN, skipping basebackup",
+                    );
+                    self.prepare_skip_count.inc();
+                    return Ok(());
+                }
+            }
+
+            if entries.len() as i64 >= self.config.max_size_entries {
+                tracing::info!(
+                    %timeline_id,
+                    %req_lsn,
+                    "Basebackup cache is full, skipping basebackup",
+                );
+                self.prepare_skip_count.inc();
+                return Ok(());
+            }
+        }
+
+        let tenant = self
+            .tenant_manager
+            .get_attached_tenant_shard(tenant_shard_id)?;
+
+        let tenant_state = tenant.current_state();
+        if tenant_state != TenantState::Active {
+            anyhow::bail!(
+                "Tenant {} is not active, current state: {:?}",
+                tenant_shard_id.tenant_id,
+                tenant_state
+            )
+        }
+
+        let timeline = tenant.get_timeline(timeline_id, true)?;
+
+        let last_record_lsn = timeline.get_last_record_lsn();
+        if last_record_lsn > req_lsn {
+            tracing::info!(
+                %timeline_id,
+                %req_lsn,
+                %last_record_lsn,
+                "Timeline has a higher LSN than the requested one, skipping basebackup",
+            );
+            self.prepare_skip_count.inc();
+            return Ok(());
+        }
+
+        let entry_tmp_path = self.entry_tmp_path(tenant_shard_id.tenant_id, timeline_id, req_lsn);
+
+        let res = self
+            .prepare_basebackup_tmp(&entry_tmp_path, &timeline, req_lsn)
+            .await;
+
+        if let Err(err) = res {
+            tracing::info!("Failed to prepare basebackup tmp file: {:#}", err);
+            // Try to clean up tmp file. If we fail, the background clean up task will take care of it.
+            match tokio::fs::remove_file(&entry_tmp_path).await {
+                Ok(_) => {}
+                Err(e) if e.kind() == std::io::ErrorKind::NotFound => {}
+                Err(e) => {
+                    tracing::info!("Failed to remove basebackup tmp file: {:?}", e);
+                }
+            }
+            return Err(err);
+        }
+
+        // Move the tmp file to the final location atomically.
+        let entry_path = self.entry_path(tenant_shard_id.tenant_id, timeline_id, req_lsn);
+        tokio::fs::rename(&entry_tmp_path, &entry_path).await?;
+
+        let mut entries = self.entries.lock().unwrap();
+        if let Some(old_lsn) = entries.insert(tti, req_lsn) {
+            // Remove the old entry if it exists.
+            self.remove_entry_sender
+                .send(self.entry_path(tenant_shard_id.tenant_id, timeline_id, old_lsn))
+                .unwrap();
+        }
+        BASEBACKUP_CACHE_ENTRIES.set(entries.len() as i64);
+
+        self.prepare_ok_count.inc();
+        Ok(())
+    }
+
+    /// Prepares a basebackup in a temporary file.
+    async fn prepare_basebackup_tmp(
+        &self,
+        emptry_tmp_path: &Utf8Path,
+        timeline: &Arc<Timeline>,
+        req_lsn: Lsn,
+    ) -> anyhow::Result<()> {
+        let ctx = RequestContext::new(TaskKind::BasebackupCache, DownloadBehavior::Download);
+        let ctx = ctx.with_scope_timeline(timeline);
+
+        let file = tokio::fs::File::create(emptry_tmp_path).await?;
+        let mut writer = BufWriter::new(file);
+
+        let mut encoder = GzipEncoder::with_quality(
+            &mut writer,
+            // Level::Best because compression is not on the hot path of basebackup requests.
+            // The decompression is almost not affected by the compression level.
+            async_compression::Level::Best,
+        );
+
+        // We may receive a request before the WAL record is applied to the timeline.
+        // Wait for the requested LSN to be applied.
+        timeline
+            .wait_lsn(
+                req_lsn,
+                crate::tenant::timeline::WaitLsnWaiter::BaseBackupCache,
+                crate::tenant::timeline::WaitLsnTimeout::Default,
+                &ctx,
+            )
+            .await?;
+
+        send_basebackup_tarball(
+            &mut encoder,
+            timeline,
+            Some(req_lsn),
+            None,
+            false,
+            false,
+            &ctx,
+        )
+        .await?;
+
+        encoder.shutdown().await?;
+        writer.flush().await?;
+        writer.into_inner().sync_all().await?;
+
+        Ok(())
+    }
+}

pageserver/src/bin/pageserver.rs

@@ -16,10 +16,12 @@ use http_utils::tls_certs::ReloadingCertificateResolver;
 use metrics::launch_timestamp::{LaunchTimestamp, set_launch_timestamp_metric};
 use metrics::set_build_info_metric;
 use nix::sys::socket::{setsockopt, sockopt};
+use pageserver::basebackup_cache::BasebackupCache;
 use pageserver::config::{PageServerConf, PageserverIdentity, ignored_fields};
 use pageserver::controller_upcall_client::StorageControllerUpcallClient;
 use pageserver::deletion_queue::DeletionQueue;
 use pageserver::disk_usage_eviction_task::{self, launch_disk_usage_global_eviction_task};
+use pageserver::feature_resolver::FeatureResolver;
 use pageserver::metrics::{STARTUP_DURATION, STARTUP_IS_LOADING};
 use pageserver::task_mgr::{
     BACKGROUND_RUNTIME, COMPUTE_REQUEST_RUNTIME, MGMT_REQUEST_RUNTIME, WALRECEIVER_RUNTIME,
@@ -387,23 +389,30 @@ fn start_pageserver(
     // We need to release the lock file only when the process exits.
     std::mem::forget(lock_file);
 
-    // Bind the HTTP and libpq ports early, so that if they are in use by some other
-    // process, we error out early.
-    let http_addr = &conf.listen_http_addr;
-    info!("Starting pageserver http handler on {http_addr}");
-    let http_listener = tcp_listener::bind(http_addr)?;
+    // Bind the HTTP, libpq, and gRPC ports early, to error out if they are
+    // already in use.
+    info!(
+        "Starting pageserver http handler on {} with auth {:#?}",
+        conf.listen_http_addr, conf.http_auth_type
+    );
+    let http_listener = tcp_listener::bind(&conf.listen_http_addr)?;
 
     let https_listener = match conf.listen_https_addr.as_ref() {
         Some(https_addr) => {
-            info!("Starting pageserver https handler on {https_addr}");
+            info!(
+                "Starting pageserver https handler on {https_addr} with auth {:#?}",
+                conf.http_auth_type
+            );
             Some(tcp_listener::bind(https_addr)?)
         }
         None => None,
     };
 
-    let pg_addr = &conf.listen_pg_addr;
-    info!("Starting pageserver pg protocol handler on {pg_addr}");
-    let pageserver_listener = tcp_listener::bind(pg_addr)?;
+    info!(
+        "Starting pageserver pg protocol handler on {} with auth {:#?}",
+        conf.listen_pg_addr, conf.pg_auth_type,
+    );
+    let pageserver_listener = tcp_listener::bind(&conf.listen_pg_addr)?;
 
     // Enable SO_KEEPALIVE on the socket, to detect dead connections faster.
     // These are configured via net.ipv4.tcp_keepalive_* sysctls.
@@ -412,6 +421,15 @@ fn start_pageserver(
     // support enabling keepalives while using the default OS sysctls.
     setsockopt(&pageserver_listener, sockopt::KeepAlive, &true)?;
 
+    let mut grpc_listener = None;
+    if let Some(grpc_addr) = &conf.listen_grpc_addr {
+        info!(
+            "Starting pageserver gRPC handler on {grpc_addr} with auth {:#?}",
+            conf.grpc_auth_type
+        );
+        grpc_listener = Some(tcp_listener::bind(grpc_addr).map_err(|e| anyhow!("{e}"))?);
+    }
+
     // Launch broker client
     // The storage_broker::connect call needs to happen inside a tokio runtime thread.
     let broker_client = WALRECEIVER_RUNTIME
@@ -439,7 +457,8 @@ fn start_pageserver(
     // Initialize authentication for incoming connections
     let http_auth;
     let pg_auth;
-    if conf.http_auth_type == AuthType::NeonJWT || conf.pg_auth_type == AuthType::NeonJWT {
+    let grpc_auth;
+    if [conf.http_auth_type, conf.pg_auth_type, conf.grpc_auth_type].contains(&AuthType::NeonJWT) {
         // unwrap is ok because check is performed when creating config, so path is set and exists
         let key_path = conf.auth_validation_public_key_path.as_ref().unwrap();
         info!("Loading public key(s) for verifying JWT tokens from {key_path:?}");
@@ -447,20 +466,23 @@ fn start_pageserver(
         let jwt_auth = JwtAuth::from_key_path(key_path)?;
         let auth: Arc<SwappableJwtAuth> = Arc::new(SwappableJwtAuth::new(jwt_auth));
 
-        http_auth = match &conf.http_auth_type {
+        http_auth = match conf.http_auth_type {
             AuthType::Trust => None,
             AuthType::NeonJWT => Some(auth.clone()),
         };
-        pg_auth = match &conf.pg_auth_type {
+        pg_auth = match conf.pg_auth_type {
+            AuthType::Trust => None,
+            AuthType::NeonJWT => Some(auth.clone()),
+        };
+        grpc_auth = match conf.grpc_auth_type {
             AuthType::Trust => None,
             AuthType::NeonJWT => Some(auth),
         };
     } else {
         http_auth = None;
         pg_auth = None;
+        grpc_auth = None;
     }
-    info!("Using auth for http API: {:#?}", conf.http_auth_type);
-    info!("Using auth for pg connections: {:#?}", conf.pg_auth_type);
 
     let tls_server_config = if conf.listen_https_addr.is_some() || conf.enable_tls_page_service_api
     {
@@ -501,6 +523,12 @@ fn start_pageserver(
     // Set up remote storage client
     let remote_storage = BACKGROUND_RUNTIME.block_on(create_remote_storage_client(conf))?;
 
+    let feature_resolver = create_feature_resolver(
+        conf,
+        shutdown_pageserver.clone(),
+        BACKGROUND_RUNTIME.handle(),
+    )?;
+
     // Set up deletion queue
     let (deletion_queue, deletion_workers) = DeletionQueue::new(
         remote_storage.clone(),
@@ -541,6 +569,8 @@ fn start_pageserver(
         pageserver::l0_flush::L0FlushGlobalState::new(conf.l0_flush.clone());
 
     // Scan the local 'tenants/' directory and start loading the tenants
+    let (basebackup_prepare_sender, basebackup_prepare_receiver) =
+        tokio::sync::mpsc::unbounded_channel();
     let deletion_queue_client = deletion_queue.new_client();
     let background_purges = mgr::BackgroundPurges::default();
     let tenant_manager = BACKGROUND_RUNTIME.block_on(mgr::init_tenant_mgr(
@@ -551,12 +581,23 @@ fn start_pageserver(
             remote_storage: remote_storage.clone(),
             deletion_queue_client,
             l0_flush_global_state,
+            basebackup_prepare_sender,
+            feature_resolver,
         },
         order,
         shutdown_pageserver.clone(),
     ))?;
     let tenant_manager = Arc::new(tenant_manager);
 
+    let basebackup_cache = BasebackupCache::spawn(
+        BACKGROUND_RUNTIME.handle(),
+        conf.basebackup_cache_dir(),
+        conf.basebackup_cache_config.clone(),
+        basebackup_prepare_receiver,
+        Arc::clone(&tenant_manager),
+        shutdown_pageserver.child_token(),
+    );
+
     BACKGROUND_RUNTIME.spawn({
         let shutdown_pageserver = shutdown_pageserver.clone();
         let drive_init = async move {
@@ -763,8 +804,27 @@ fn start_pageserver(
         } else {
             None
         },
+        basebackup_cache.clone(),
     );
 
+    // Spawn a Pageserver gRPC server task. It will spawn separate tasks for
+    // each stream/request.
+    //
+    // TODO: this uses a separate Tokio runtime for the page service. If we want
+    // other gRPC services, they will need their own port and runtime. Is this
+    // necessary?
+    let mut page_service_grpc = None;
+    if let Some(grpc_listener) = grpc_listener {
+        page_service_grpc = Some(page_service::spawn_grpc(
+            conf,
+            tenant_manager.clone(),
+            grpc_auth,
+            otel_guard.as_ref().map(|g| g.dispatch.clone()),
+            grpc_listener,
+            basebackup_cache,
+        )?);
+    }
+
     // All started up! Now just sit and wait for shutdown signal.
     BACKGROUND_RUNTIME.block_on(async move {
         let signal_token = CancellationToken::new();
@@ -783,6 +843,7 @@ fn start_pageserver(
             http_endpoint_listener,
             https_endpoint_listener,
             page_service,
+            page_service_grpc,
             consumption_metrics_tasks,
             disk_usage_eviction_task,
             &tenant_manager,
@@ -796,6 +857,14 @@ fn start_pageserver(
     })
 }
 
+fn create_feature_resolver(
+    conf: &'static PageServerConf,
+    shutdown_pageserver: CancellationToken,
+    handle: &tokio::runtime::Handle,
+) -> anyhow::Result<FeatureResolver> {
+    FeatureResolver::spawn(conf, shutdown_pageserver, handle)
+}
+
 async fn create_remote_storage_client(
     conf: &'static PageServerConf,
 ) -> anyhow::Result<GenericRemoteStorage> {

pageserver/src/config.rs

@@ -14,7 +14,7 @@ use std::time::Duration;
 use anyhow::{Context, bail, ensure};
 use camino::{Utf8Path, Utf8PathBuf};
 use once_cell::sync::OnceCell;
-use pageserver_api::config::{DiskUsageEvictionTaskConfig, MaxVectoredReadBytes};
+use pageserver_api::config::{DiskUsageEvictionTaskConfig, MaxVectoredReadBytes, PostHogConfig};
 use pageserver_api::models::ImageCompressionAlgorithm;
 use pageserver_api::shard::TenantShardId;
 use pem::Pem;
@@ -58,11 +58,16 @@ pub struct PageServerConf {
     pub listen_http_addr: String,
     /// Example: 127.0.0.1:9899
     pub listen_https_addr: Option<String>,
+    /// If set, expose a gRPC API on this address.
+    /// Example: 127.0.0.1:51051
+    ///
+    /// EXPERIMENTAL: this protocol is unstable and under active development.
+    pub listen_grpc_addr: Option<String>,
 
-    /// Path to a file with certificate's private key for https API.
+    /// Path to a file with certificate's private key for https and gRPC API.
     /// Default: server.key
     pub ssl_key_file: Utf8PathBuf,
-    /// Path to a file with a X509 certificate for https API.
+    /// Path to a file with a X509 certificate for https and gRPC API.
     /// Default: server.crt
     pub ssl_cert_file: Utf8PathBuf,
     /// Period to reload certificate and private key from files.
@@ -100,6 +105,8 @@ pub struct PageServerConf {
     pub http_auth_type: AuthType,
     /// authentication method for libpq connections from compute
     pub pg_auth_type: AuthType,
+    /// authentication method for gRPC connections from compute
+    pub grpc_auth_type: AuthType,
     /// Path to a file or directory containing public key(s) for verifying JWT tokens.
     /// Used for both mgmt and compute auth, if enabled.
     pub auth_validation_public_key_path: Option<Utf8PathBuf>,
@@ -231,7 +238,12 @@ pub struct PageServerConf {
     /// This is insecure and should only be used in development environments.
     pub dev_mode: bool,
 
+    /// PostHog integration config.
+    pub posthog_config: Option<PostHogConfig>,
+
     pub timeline_import_config: pageserver_api::config::TimelineImportConfig,
+
+    pub basebackup_cache_config: Option<pageserver_api::config::BasebackupCacheConfig>,
 }
 
 /// Token for authentication to safekeepers
@@ -261,6 +273,10 @@ impl PageServerConf {
         self.workdir.join("metadata.json")
     }
 
+    pub fn basebackup_cache_dir(&self) -> Utf8PathBuf {
+        self.workdir.join("basebackup_cache")
+    }
+
     pub fn deletion_list_path(&self, sequence: u64) -> Utf8PathBuf {
         // Encode a version in the filename, so that if we ever switch away from JSON we can
         // increment this.
@@ -349,6 +365,7 @@ impl PageServerConf {
             listen_pg_addr,
             listen_http_addr,
             listen_https_addr,
+            listen_grpc_addr,
             ssl_key_file,
             ssl_cert_file,
             ssl_cert_reload_period,
@@ -363,6 +380,7 @@ impl PageServerConf {
             pg_distrib_dir,
             http_auth_type,
             pg_auth_type,
+            grpc_auth_type,
             auth_validation_public_key_path,
             remote_storage,
             broker_endpoint,
@@ -406,7 +424,9 @@ impl PageServerConf {
             tracing,
             enable_tls_page_service_api,
             dev_mode,
+            posthog_config,
             timeline_import_config,
+            basebackup_cache_config,
         } = config_toml;
 
         let mut conf = PageServerConf {
@@ -416,6 +436,7 @@ impl PageServerConf {
             listen_pg_addr,
             listen_http_addr,
             listen_https_addr,
+            listen_grpc_addr,
             ssl_key_file,
             ssl_cert_file,
             ssl_cert_reload_period,
@@ -428,6 +449,7 @@ impl PageServerConf {
             max_file_descriptors,
             http_auth_type,
             pg_auth_type,
+            grpc_auth_type,
             auth_validation_public_key_path,
             remote_storage_config: remote_storage,
             broker_endpoint,
@@ -461,6 +483,7 @@ impl PageServerConf {
             enable_tls_page_service_api,
             dev_mode,
             timeline_import_config,
+            basebackup_cache_config,
 
             // ------------------------------------------------------------
             // fields that require additional validation or custom handling
@@ -517,13 +540,16 @@ impl PageServerConf {
                 }
                 None => Vec::new(),
             },
+            posthog_config,
         };
 
         // ------------------------------------------------------------
         // custom validation code that covers more than one field in isolation
         // ------------------------------------------------------------
 
-        if conf.http_auth_type == AuthType::NeonJWT || conf.pg_auth_type == AuthType::NeonJWT {
+        if [conf.http_auth_type, conf.pg_auth_type, conf.grpc_auth_type]
+            .contains(&AuthType::NeonJWT)
+        {
             let auth_validation_public_key_path = conf
                 .auth_validation_public_key_path
                 .get_or_insert_with(|| workdir.join("auth_public_key.pem"));
@@ -544,6 +570,23 @@ impl PageServerConf {
                     ratio.numerator, ratio.denominator
                 )
             );
+
+            let url = Url::parse(&tracing_config.export_config.endpoint)
+                .map_err(anyhow::Error::msg)
+                .with_context(|| {
+                    format!(
+                        "tracing endpoint URL is invalid : {}",
+                        tracing_config.export_config.endpoint
+                    )
+                })?;
+
+            ensure!(
+                url.scheme() == "http" || url.scheme() == "https",
+                format!(
+                    "tracing endpoint URL must start with http:// or https://: {}",
+                    tracing_config.export_config.endpoint
+                )
+            );
         }
 
         IndexEntry::validate_checkpoint_distance(conf.default_tenant_conf.checkpoint_distance)
@@ -660,4 +703,25 @@ mod tests {
         PageServerConf::parse_and_validate(NodeId(0), config_toml, &workdir)
             .expect("parse_and_validate");
     }
+
+    #[test]
+    fn test_config_tracing_endpoint_is_invalid() {
+        let input = r#"
+            control_plane_api = "http://localhost:6666"
+
+            [tracing]
+
+            sampling_ratio = { numerator = 1, denominator = 0 }
+
+            [tracing.export_config]
+            endpoint = "localhost:4317"
+            protocol = "http-binary"
+            timeout = "1ms"
+        "#;
+        let config_toml = toml_edit::de::from_str::<pageserver_api::config::ConfigToml>(input)
+            .expect("config has valid fields");
+        let workdir = Utf8PathBuf::from("/nonexistent");
+        PageServerConf::parse_and_validate(NodeId(0), config_toml, &workdir)
+            .expect_err("parse_and_validate should fail for endpoint without scheme");
+    }
 }

pageserver/src/consumption_metrics/metrics.rs

@@ -18,12 +18,25 @@ use crate::tenant::timeline::logical_size::CurrentLogicalSize;
 // management.
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Hash, serde::Serialize, serde::Deserialize)]
 pub(super) enum Name {
-    /// Timeline last_record_lsn, absolute
+    /// Timeline last_record_lsn, absolute.
     #[serde(rename = "written_size")]
     WrittenSize,
     /// Timeline last_record_lsn, incremental
     #[serde(rename = "written_data_bytes_delta")]
     WrittenSizeDelta,
+    /// Written bytes only on this timeline (not including ancestors):
+    /// written_size - ancestor_lsn
+    ///
+    /// On the root branch, this is equivalent to `written_size`.
+    #[serde(rename = "written_size_since_parent")]
+    WrittenSizeSinceParent,
+    /// PITR history size only on this timeline (not including ancestors):
+    /// last_record_lsn - max(pitr_cutoff, ancestor_lsn).
+    ///
+    /// On the root branch, this is its entire PITR history size. Not emitted if GC hasn't computed
+    /// the PITR cutoff yet. 0 if PITR is disabled.
+    #[serde(rename = "pitr_history_size_since_parent")]
+    PitrHistorySizeSinceParent,
     /// Timeline logical size
     #[serde(rename = "timeline_logical_size")]
     LogicalSize,
@@ -157,6 +170,32 @@ impl MetricsKey {
         .incremental_values()
     }
 
+    /// `written_size` - `ancestor_lsn`.
+    const fn written_size_since_parent(
+        tenant_id: TenantId,
+        timeline_id: TimelineId,
+    ) -> AbsoluteValueFactory {
+        MetricsKey {
+            tenant_id,
+            timeline_id: Some(timeline_id),
+            metric: Name::WrittenSizeSinceParent,
+        }
+        .absolute_values()
+    }
+
+    /// `written_size` - max(`pitr_cutoff`, `ancestor_lsn`).
+    const fn pitr_history_size_since_parent(
+        tenant_id: TenantId,
+        timeline_id: TimelineId,
+    ) -> AbsoluteValueFactory {
+        MetricsKey {
+            tenant_id,
+            timeline_id: Some(timeline_id),
+            metric: Name::PitrHistorySizeSinceParent,
+        }
+        .absolute_values()
+    }
+
     /// Exact [`Timeline::get_current_logical_size`].
     ///
     /// [`Timeline::get_current_logical_size`]: crate::tenant::Timeline::get_current_logical_size
@@ -334,7 +373,13 @@ impl TenantSnapshot {
 struct TimelineSnapshot {
     loaded_at: (Lsn, SystemTime),
     last_record_lsn: Lsn,
+    ancestor_lsn: Lsn,
     current_exact_logical_size: Option<u64>,
+    /// Whether PITR is enabled (pitr_interval > 0).
+    pitr_enabled: bool,
+    /// The PITR cutoff LSN. None if not yet initialized. If PITR is disabled, this is approximately
+    /// Some(last_record_lsn), but may lag behind it since it's computed periodically.
+    pitr_cutoff: Option<Lsn>,
 }
 
 impl TimelineSnapshot {
@@ -354,6 +399,9 @@ impl TimelineSnapshot {
         } else {
             let loaded_at = t.loaded_at;
             let last_record_lsn = t.get_last_record_lsn();
+            let ancestor_lsn = t.get_ancestor_lsn();
+            let pitr_enabled = !t.get_pitr_interval().is_zero();
+            let pitr_cutoff = t.gc_info.read().unwrap().cutoffs.time;
 
             let current_exact_logical_size = {
                 let span = tracing::info_span!("collect_metrics_iteration", tenant_id = %t.tenant_shard_id.tenant_id, timeline_id = %t.timeline_id);
@@ -373,7 +421,10 @@ impl TimelineSnapshot {
             Ok(Some(TimelineSnapshot {
                 loaded_at,
                 last_record_lsn,
+                ancestor_lsn,
                 current_exact_logical_size,
+                pitr_enabled,
+                pitr_cutoff,
             }))
         }
     }
@@ -424,6 +475,8 @@ impl TimelineSnapshot {
 
         let up_to = now;
 
+        let written_size_last = written_size_now.value.max(prev.1); // don't regress
+
         if let Some(delta) = written_size_now.value.checked_sub(prev.1) {
             let key_value = written_size_delta_key.from_until(prev.0, up_to, delta);
             // written_size_delta
@@ -441,6 +494,27 @@ impl TimelineSnapshot {
             });
         }
 
+        // Compute the branch-local written size.
+        let written_size_since_parent_key =
+            MetricsKey::written_size_since_parent(tenant_id, timeline_id);
+        metrics.push(
+            written_size_since_parent_key
+                .at(now, written_size_last.saturating_sub(self.ancestor_lsn.0)),
+        );
+
+        // Compute the branch-local PITR history size. Not emitted if GC hasn't yet computed the
+        // PITR cutoff. 0 if PITR is disabled.
+        let pitr_history_size_since_parent_key =
+            MetricsKey::pitr_history_size_since_parent(tenant_id, timeline_id);
+        if !self.pitr_enabled {
+            metrics.push(pitr_history_size_since_parent_key.at(now, 0));
+        } else if let Some(pitr_cutoff) = self.pitr_cutoff {
+            metrics.push(pitr_history_size_since_parent_key.at(
+                now,
+                written_size_last.saturating_sub(pitr_cutoff.max(self.ancestor_lsn).0),
+            ));
+        }
+
         {
             let factory = MetricsKey::timeline_logical_size(tenant_id, timeline_id);
             let current_or_previous = self

pageserver/src/consumption_metrics/metrics/tests.rs

@@ -12,12 +12,17 @@ fn startup_collected_timeline_metrics_before_advancing() {
     let cache = HashMap::new();
 
     let initdb_lsn = Lsn(0x10000);
+    let pitr_cutoff = Lsn(0x11000);
     let disk_consistent_lsn = Lsn(initdb_lsn.0 * 2);
+    let logical_size = 0x42000;
 
     let snap = TimelineSnapshot {
         loaded_at: (disk_consistent_lsn, SystemTime::now()),
         last_record_lsn: disk_consistent_lsn,
-        current_exact_logical_size: Some(0x42000),
+        ancestor_lsn: Lsn(0),
+        current_exact_logical_size: Some(logical_size),
+        pitr_enabled: true,
+        pitr_cutoff: Some(pitr_cutoff),
     };
 
     let now = DateTime::<Utc>::from(SystemTime::now());
@@ -33,7 +38,11 @@ fn startup_collected_timeline_metrics_before_advancing() {
                 0
             ),
             MetricsKey::written_size(tenant_id, timeline_id).at(now, disk_consistent_lsn.0),
-            MetricsKey::timeline_logical_size(tenant_id, timeline_id).at(now, 0x42000)
+            MetricsKey::written_size_since_parent(tenant_id, timeline_id)
+                .at(now, disk_consistent_lsn.0),
+            MetricsKey::pitr_history_size_since_parent(tenant_id, timeline_id)
+                .at(now, disk_consistent_lsn.0 - pitr_cutoff.0),
+            MetricsKey::timeline_logical_size(tenant_id, timeline_id).at(now, logical_size)
         ]
     );
 }
@@ -49,7 +58,9 @@ fn startup_collected_timeline_metrics_second_round() {
     let before = DateTime::<Utc>::from(before);
 
     let initdb_lsn = Lsn(0x10000);
+    let pitr_cutoff = Lsn(0x11000);
     let disk_consistent_lsn = Lsn(initdb_lsn.0 * 2);
+    let logical_size = 0x42000;
 
     let mut metrics = Vec::new();
     let cache = HashMap::from([MetricsKey::written_size(tenant_id, timeline_id)
@@ -59,7 +70,10 @@ fn startup_collected_timeline_metrics_second_round() {
     let snap = TimelineSnapshot {
         loaded_at: (disk_consistent_lsn, init),
         last_record_lsn: disk_consistent_lsn,
-        current_exact_logical_size: Some(0x42000),
+        ancestor_lsn: Lsn(0),
+        current_exact_logical_size: Some(logical_size),
+        pitr_enabled: true,
+        pitr_cutoff: Some(pitr_cutoff),
     };
 
     snap.to_metrics(tenant_id, timeline_id, now, &mut metrics, &cache);
@@ -69,7 +83,11 @@ fn startup_collected_timeline_metrics_second_round() {
         &[
             MetricsKey::written_size_delta(tenant_id, timeline_id).from_until(before, now, 0),
             MetricsKey::written_size(tenant_id, timeline_id).at(now, disk_consistent_lsn.0),
-            MetricsKey::timeline_logical_size(tenant_id, timeline_id).at(now, 0x42000)
+            MetricsKey::written_size_since_parent(tenant_id, timeline_id)
+                .at(now, disk_consistent_lsn.0),
+            MetricsKey::pitr_history_size_since_parent(tenant_id, timeline_id)
+                .at(now, disk_consistent_lsn.0 - pitr_cutoff.0),
+            MetricsKey::timeline_logical_size(tenant_id, timeline_id).at(now, logical_size)
         ]
     );
 }
@@ -86,7 +104,9 @@ fn startup_collected_timeline_metrics_nth_round_at_same_lsn() {
     let before = DateTime::<Utc>::from(before);
 
     let initdb_lsn = Lsn(0x10000);
+    let pitr_cutoff = Lsn(0x11000);
     let disk_consistent_lsn = Lsn(initdb_lsn.0 * 2);
+    let logical_size = 0x42000;
 
     let mut metrics = Vec::new();
     let cache = HashMap::from([
@@ -103,7 +123,10 @@ fn startup_collected_timeline_metrics_nth_round_at_same_lsn() {
     let snap = TimelineSnapshot {
         loaded_at: (disk_consistent_lsn, init),
         last_record_lsn: disk_consistent_lsn,
-        current_exact_logical_size: Some(0x42000),
+        ancestor_lsn: Lsn(0),
+        current_exact_logical_size: Some(logical_size),
+        pitr_enabled: true,
+        pitr_cutoff: Some(pitr_cutoff),
     };
 
     snap.to_metrics(tenant_id, timeline_id, now, &mut metrics, &cache);
@@ -113,16 +136,18 @@ fn startup_collected_timeline_metrics_nth_round_at_same_lsn() {
         &[
             MetricsKey::written_size_delta(tenant_id, timeline_id).from_until(just_before, now, 0),
             MetricsKey::written_size(tenant_id, timeline_id).at(now, disk_consistent_lsn.0),
-            MetricsKey::timeline_logical_size(tenant_id, timeline_id).at(now, 0x42000)
+            MetricsKey::written_size_since_parent(tenant_id, timeline_id)
+                .at(now, disk_consistent_lsn.0),
+            MetricsKey::pitr_history_size_since_parent(tenant_id, timeline_id)
+                .at(now, disk_consistent_lsn.0 - pitr_cutoff.0),
+            MetricsKey::timeline_logical_size(tenant_id, timeline_id).at(now, logical_size)
         ]
     );
 }
 
+/// Tests that written sizes do not regress across restarts.
 #[test]
 fn post_restart_written_sizes_with_rolled_back_last_record_lsn() {
-    // it can happen that we lose the inmemorylayer but have previously sent metrics and we
-    // should never go backwards
-
     let tenant_id = TenantId::generate();
     let timeline_id = TimelineId::generate();
 
@@ -140,7 +165,10 @@ fn post_restart_written_sizes_with_rolled_back_last_record_lsn() {
     let snap = TimelineSnapshot {
         loaded_at: (Lsn(50), at_restart),
         last_record_lsn: Lsn(50),
+        ancestor_lsn: Lsn(0),
         current_exact_logical_size: None,
+        pitr_enabled: true,
+        pitr_cutoff: Some(Lsn(20)),
     };
 
     let mut cache = HashMap::from([
@@ -169,6 +197,8 @@ fn post_restart_written_sizes_with_rolled_back_last_record_lsn() {
                 0
             ),
             MetricsKey::written_size(tenant_id, timeline_id).at(now, 100),
+            MetricsKey::written_size_since_parent(tenant_id, timeline_id).at(now, 100),
+            MetricsKey::pitr_history_size_since_parent(tenant_id, timeline_id).at(now, 80),
         ]
     );
 
@@ -183,6 +213,157 @@ fn post_restart_written_sizes_with_rolled_back_last_record_lsn() {
         &[
             MetricsKey::written_size_delta(tenant_id, timeline_id).from_until(now, later, 0),
             MetricsKey::written_size(tenant_id, timeline_id).at(later, 100),
+            MetricsKey::written_size_since_parent(tenant_id, timeline_id).at(later, 100),
+            MetricsKey::pitr_history_size_since_parent(tenant_id, timeline_id).at(later, 80),
+        ]
+    );
+}
+
+/// Tests that written sizes do not regress across restarts, even on child branches.
+#[test]
+fn post_restart_written_sizes_with_rolled_back_last_record_lsn_and_ancestor_lsn() {
+    let tenant_id = TenantId::generate();
+    let timeline_id = TimelineId::generate();
+
+    let [later, now, at_restart] = time_backwards();
+
+    // FIXME: tests would be so much easier if we did not need to juggle back and forth
+    // SystemTime and DateTime::<Utc> ... Could do the conversion only at upload time?
+    let now = DateTime::<Utc>::from(now);
+    let later = DateTime::<Utc>::from(later);
+    let before_restart = at_restart - std::time::Duration::from_secs(5 * 60);
+    let way_before = before_restart - std::time::Duration::from_secs(10 * 60);
+    let before_restart = DateTime::<Utc>::from(before_restart);
+    let way_before = DateTime::<Utc>::from(way_before);
+
+    let snap = TimelineSnapshot {
+        loaded_at: (Lsn(50), at_restart),
+        last_record_lsn: Lsn(50),
+        ancestor_lsn: Lsn(40),
+        current_exact_logical_size: None,
+        pitr_enabled: true,
+        pitr_cutoff: Some(Lsn(20)),
+    };
+
+    let mut cache = HashMap::from([
+        MetricsKey::written_size(tenant_id, timeline_id)
+            .at(before_restart, 100)
+            .to_kv_pair(),
+        MetricsKey::written_size_delta(tenant_id, timeline_id)
+            .from_until(
+                way_before,
+                before_restart,
+                // not taken into account, but the timestamps are important
+                999_999_999,
+            )
+            .to_kv_pair(),
+    ]);
+
+    let mut metrics = Vec::new();
+    snap.to_metrics(tenant_id, timeline_id, now, &mut metrics, &cache);
+
+    assert_eq!(
+        metrics,
+        &[
+            MetricsKey::written_size_delta(tenant_id, timeline_id).from_until(
+                before_restart,
+                now,
+                0
+            ),
+            MetricsKey::written_size(tenant_id, timeline_id).at(now, 100),
+            MetricsKey::written_size_since_parent(tenant_id, timeline_id).at(now, 60),
+            MetricsKey::pitr_history_size_since_parent(tenant_id, timeline_id).at(now, 60),
+        ]
+    );
+
+    // now if we cache these metrics, and re-run while "still in recovery"
+    cache.extend(metrics.drain(..).map(|x| x.to_kv_pair()));
+
+    // "still in recovery", because our snapshot did not change
+    snap.to_metrics(tenant_id, timeline_id, later, &mut metrics, &cache);
+
+    assert_eq!(
+        metrics,
+        &[
+            MetricsKey::written_size_delta(tenant_id, timeline_id).from_until(now, later, 0),
+            MetricsKey::written_size(tenant_id, timeline_id).at(later, 100),
+            MetricsKey::written_size_since_parent(tenant_id, timeline_id).at(later, 60),
+            MetricsKey::pitr_history_size_since_parent(tenant_id, timeline_id).at(later, 60),
+        ]
+    );
+}
+
+/// Tests that written sizes do not regress across restarts, even on child branches and
+/// with a PITR cutoff after the branch point.
+#[test]
+fn post_restart_written_sizes_with_rolled_back_last_record_lsn_and_ancestor_lsn_and_pitr_cutoff() {
+    let tenant_id = TenantId::generate();
+    let timeline_id = TimelineId::generate();
+
+    let [later, now, at_restart] = time_backwards();
+
+    // FIXME: tests would be so much easier if we did not need to juggle back and forth
+    // SystemTime and DateTime::<Utc> ... Could do the conversion only at upload time?
+    let now = DateTime::<Utc>::from(now);
+    let later = DateTime::<Utc>::from(later);
+    let before_restart = at_restart - std::time::Duration::from_secs(5 * 60);
+    let way_before = before_restart - std::time::Duration::from_secs(10 * 60);
+    let before_restart = DateTime::<Utc>::from(before_restart);
+    let way_before = DateTime::<Utc>::from(way_before);
+
+    let snap = TimelineSnapshot {
+        loaded_at: (Lsn(50), at_restart),
+        last_record_lsn: Lsn(50),
+        ancestor_lsn: Lsn(30),
+        current_exact_logical_size: None,
+        pitr_enabled: true,
+        pitr_cutoff: Some(Lsn(40)),
+    };
+
+    let mut cache = HashMap::from([
+        MetricsKey::written_size(tenant_id, timeline_id)
+            .at(before_restart, 100)
+            .to_kv_pair(),
+        MetricsKey::written_size_delta(tenant_id, timeline_id)
+            .from_until(
+                way_before,
+                before_restart,
+                // not taken into account, but the timestamps are important
+                999_999_999,
+            )
+            .to_kv_pair(),
+    ]);
+
+    let mut metrics = Vec::new();
+    snap.to_metrics(tenant_id, timeline_id, now, &mut metrics, &cache);
+
+    assert_eq!(
+        metrics,
+        &[
+            MetricsKey::written_size_delta(tenant_id, timeline_id).from_until(
+                before_restart,
+                now,
+                0
+            ),
+            MetricsKey::written_size(tenant_id, timeline_id).at(now, 100),
+            MetricsKey::written_size_since_parent(tenant_id, timeline_id).at(now, 70),
+            MetricsKey::pitr_history_size_since_parent(tenant_id, timeline_id).at(now, 60),
+        ]
+    );
+
+    // now if we cache these metrics, and re-run while "still in recovery"
+    cache.extend(metrics.drain(..).map(|x| x.to_kv_pair()));
+
+    // "still in recovery", because our snapshot did not change
+    snap.to_metrics(tenant_id, timeline_id, later, &mut metrics, &cache);
+
+    assert_eq!(
+        metrics,
+        &[
+            MetricsKey::written_size_delta(tenant_id, timeline_id).from_until(now, later, 0),
+            MetricsKey::written_size(tenant_id, timeline_id).at(later, 100),
+            MetricsKey::written_size_since_parent(tenant_id, timeline_id).at(later, 70),
+            MetricsKey::pitr_history_size_since_parent(tenant_id, timeline_id).at(later, 60),
         ]
     );
 }
@@ -201,7 +382,10 @@ fn post_restart_current_exact_logical_size_uses_cached() {
     let snap = TimelineSnapshot {
         loaded_at: (Lsn(50), at_restart),
         last_record_lsn: Lsn(50),
+        ancestor_lsn: Lsn(0),
         current_exact_logical_size: None,
+        pitr_enabled: true,
+        pitr_cutoff: None,
     };
 
     let cache = HashMap::from([MetricsKey::timeline_logical_size(tenant_id, timeline_id)
@@ -286,16 +470,101 @@ fn time_backwards<const N: usize>() -> [std::time::SystemTime; N] {
     times
 }
 
+/// Tests that disabled PITR history does not yield any history size, even when the PITR cutoff
+/// indicates otherwise.
+#[test]
+fn pitr_disabled_yields_no_history_size() {
+    let tenant_id = TenantId::generate();
+    let timeline_id = TimelineId::generate();
+
+    let mut metrics = Vec::new();
+    let cache = HashMap::new();
+
+    let initdb_lsn = Lsn(0x10000);
+    let pitr_cutoff = Lsn(0x11000);
+    let disk_consistent_lsn = Lsn(initdb_lsn.0 * 2);
+
+    let snap = TimelineSnapshot {
+        loaded_at: (disk_consistent_lsn, SystemTime::now()),
+        last_record_lsn: disk_consistent_lsn,
+        ancestor_lsn: Lsn(0),
+        current_exact_logical_size: None,
+        pitr_enabled: false,
+        pitr_cutoff: Some(pitr_cutoff),
+    };
+
+    let now = DateTime::<Utc>::from(SystemTime::now());
+
+    snap.to_metrics(tenant_id, timeline_id, now, &mut metrics, &cache);
+
+    assert_eq!(
+        metrics,
+        &[
+            MetricsKey::written_size_delta(tenant_id, timeline_id).from_until(
+                snap.loaded_at.1.into(),
+                now,
+                0
+            ),
+            MetricsKey::written_size(tenant_id, timeline_id).at(now, disk_consistent_lsn.0),
+            MetricsKey::written_size_since_parent(tenant_id, timeline_id)
+                .at(now, disk_consistent_lsn.0),
+            MetricsKey::pitr_history_size_since_parent(tenant_id, timeline_id).at(now, 0),
+        ]
+    );
+}
+
+/// Tests that uninitialized PITR cutoff does not emit any history size metric at all.
+#[test]
+fn pitr_uninitialized_does_not_emit_history_size() {
+    let tenant_id = TenantId::generate();
+    let timeline_id = TimelineId::generate();
+
+    let mut metrics = Vec::new();
+    let cache = HashMap::new();
+
+    let initdb_lsn = Lsn(0x10000);
+    let disk_consistent_lsn = Lsn(initdb_lsn.0 * 2);
+
+    let snap = TimelineSnapshot {
+        loaded_at: (disk_consistent_lsn, SystemTime::now()),
+        last_record_lsn: disk_consistent_lsn,
+        ancestor_lsn: Lsn(0),
+        current_exact_logical_size: None,
+        pitr_enabled: true,
+        pitr_cutoff: None,
+    };
+
+    let now = DateTime::<Utc>::from(SystemTime::now());
+
+    snap.to_metrics(tenant_id, timeline_id, now, &mut metrics, &cache);
+
+    assert_eq!(
+        metrics,
+        &[
+            MetricsKey::written_size_delta(tenant_id, timeline_id).from_until(
+                snap.loaded_at.1.into(),
+                now,
+                0
+            ),
+            MetricsKey::written_size(tenant_id, timeline_id).at(now, disk_consistent_lsn.0),
+            MetricsKey::written_size_since_parent(tenant_id, timeline_id)
+                .at(now, disk_consistent_lsn.0),
+        ]
+    );
+}
+
 pub(crate) const fn metric_examples_old(
     tenant_id: TenantId,
     timeline_id: TimelineId,
     now: DateTime<Utc>,
     before: DateTime<Utc>,
-) -> [RawMetric; 5] {
+) -> [RawMetric; 7] {
     [
         MetricsKey::written_size(tenant_id, timeline_id).at_old_format(now, 0),
         MetricsKey::written_size_delta(tenant_id, timeline_id)
             .from_until_old_format(before, now, 0),
+        MetricsKey::written_size_since_parent(tenant_id, timeline_id).at_old_format(now, 0),
+        MetricsKey::pitr_history_size_since_parent(tenant_id, timeline_id).at_old_format(now, 0),
         MetricsKey::timeline_logical_size(tenant_id, timeline_id).at_old_format(now, 0),
         MetricsKey::remote_storage_size(tenant_id).at_old_format(now, 0),
         MetricsKey::synthetic_size(tenant_id).at_old_format(now, 1),
@@ -307,10 +576,12 @@ pub(crate) const fn metric_examples(
     timeline_id: TimelineId,
     now: DateTime<Utc>,
     before: DateTime<Utc>,
-) -> [NewRawMetric; 5] {
+) -> [NewRawMetric; 7] {
     [
         MetricsKey::written_size(tenant_id, timeline_id).at(now, 0),
         MetricsKey::written_size_delta(tenant_id, timeline_id).from_until(before, now, 0),
+        MetricsKey::written_size_since_parent(tenant_id, timeline_id).at(now, 0),
+        MetricsKey::pitr_history_size_since_parent(tenant_id, timeline_id).at(now, 0),
         MetricsKey::timeline_logical_size(tenant_id, timeline_id).at(now, 0),
         MetricsKey::remote_storage_size(tenant_id).at(now, 0),
         MetricsKey::synthetic_size(tenant_id).at(now, 1),

pageserver/src/consumption_metrics/upload.rs

@@ -513,6 +513,14 @@ mod tests {
                 line!(),
                 r#"{"type":"incremental","start_time":"2023-09-14T00:00:00.123456789Z","stop_time":"2023-09-15T00:00:00.123456789Z","metric":"written_data_bytes_delta","idempotency_key":"2023-09-15 00:00:00.123456789 UTC-1-0000","value":0,"tenant_id":"00000000000000000000000000000000","timeline_id":"ffffffffffffffffffffffffffffffff"}"#,
             ),
+            (
+                line!(),
+                r#"{"type":"absolute","time":"2023-09-15T00:00:00.123456789Z","metric":"written_size_since_parent","idempotency_key":"2023-09-15 00:00:00.123456789 UTC-1-0000","value":0,"tenant_id":"00000000000000000000000000000000","timeline_id":"ffffffffffffffffffffffffffffffff"}"#,
+            ),
+            (
+                line!(),
+                r#"{"type":"absolute","time":"2023-09-15T00:00:00.123456789Z","metric":"pitr_history_size_since_parent","idempotency_key":"2023-09-15 00:00:00.123456789 UTC-1-0000","value":0,"tenant_id":"00000000000000000000000000000000","timeline_id":"ffffffffffffffffffffffffffffffff"}"#,
+            ),
             (
                 line!(),
                 r#"{"type":"absolute","time":"2023-09-15T00:00:00.123456789Z","metric":"timeline_logical_size","idempotency_key":"2023-09-15 00:00:00.123456789 UTC-1-0000","value":0,"tenant_id":"00000000000000000000000000000000","timeline_id":"ffffffffffffffffffffffffffffffff"}"#,
@@ -560,7 +568,7 @@ mod tests {
         assert_eq!(upgraded_samples, new_samples);
     }
 
-    fn metric_samples_old() -> [RawMetric; 5] {
+    fn metric_samples_old() -> [RawMetric; 7] {
         let tenant_id = TenantId::from_array([0; 16]);
         let timeline_id = TimelineId::from_array([0xff; 16]);
 
@@ -572,7 +580,7 @@ mod tests {
         super::super::metrics::metric_examples_old(tenant_id, timeline_id, now, before)
     }
 
-    fn metric_samples() -> [NewRawMetric; 5] {
+    fn metric_samples() -> [NewRawMetric; 7] {
         let tenant_id = TenantId::from_array([0; 16]);
         let timeline_id = TimelineId::from_array([0xff; 16]);
 

pageserver/src/controller_upcall_client.rs

@@ -7,7 +7,7 @@ use pageserver_api::models::ShardImportStatus;
 use pageserver_api::shard::TenantShardId;
 use pageserver_api::upcall_api::{
     PutTimelineImportStatusRequest, ReAttachRequest, ReAttachResponse, ReAttachResponseTenant,
-    ValidateRequest, ValidateRequestTenant, ValidateResponse,
+    TimelineImportStatusRequest, ValidateRequest, ValidateRequestTenant, ValidateResponse,
 };
 use reqwest::Certificate;
 use serde::Serialize;
@@ -51,8 +51,15 @@ pub trait StorageControllerUpcallApi {
         &self,
         tenant_shard_id: TenantShardId,
         timeline_id: TimelineId,
+        generation: Generation,
         status: ShardImportStatus,
     ) -> impl Future<Output = Result<(), RetryForeverError>> + Send;
+    fn get_timeline_import_status(
+        &self,
+        tenant_shard_id: TenantShardId,
+        timeline_id: TimelineId,
+        generation: Generation,
+    ) -> impl Future<Output = Result<ShardImportStatus, RetryForeverError>> + Send;
 }
 
 impl StorageControllerUpcallClient {
@@ -97,6 +104,7 @@ impl StorageControllerUpcallClient {
         &self,
         url: &url::Url,
         request: R,
+        method: reqwest::Method,
     ) -> Result<T, RetryForeverError>
     where
         R: Serialize,
@@ -106,7 +114,7 @@ impl StorageControllerUpcallClient {
             || async {
                 let response = self
                     .http_client
-                    .post(url.clone())
+                    .request(method.clone(), url.clone())
                     .json(&request)
                     .send()
                     .await?;
@@ -215,7 +223,9 @@ impl StorageControllerUpcallApi for StorageControllerUpcallClient {
             register: register.clone(),
         };
 
-        let response: ReAttachResponse = self.retry_http_forever(&url, request).await?;
+        let response: ReAttachResponse = self
+            .retry_http_forever(&url, request, reqwest::Method::POST)
+            .await?;
         tracing::info!(
             "Received re-attach response with {} tenants (node {}, register: {:?})",
             response.tenants.len(),
@@ -268,7 +278,9 @@ impl StorageControllerUpcallApi for StorageControllerUpcallClient {
                 return Err(RetryForeverError::ShuttingDown);
             }
 
-            let response: ValidateResponse = self.retry_http_forever(&url, request).await?;
+            let response: ValidateResponse = self
+                .retry_http_forever(&url, request, reqwest::Method::POST)
+                .await?;
             for rt in response.tenants {
                 result.insert(rt.id, rt.valid);
             }
@@ -287,6 +299,7 @@ impl StorageControllerUpcallApi for StorageControllerUpcallClient {
         &self,
         tenant_shard_id: TenantShardId,
         timeline_id: TimelineId,
+        generation: Generation,
         status: ShardImportStatus,
     ) -> Result<(), RetryForeverError> {
         let url = self
@@ -297,9 +310,35 @@ impl StorageControllerUpcallApi for StorageControllerUpcallClient {
         let request = PutTimelineImportStatusRequest {
             tenant_shard_id,
             timeline_id,
+            generation,
             status,
         };
 
-        self.retry_http_forever(&url, request).await
+        self.retry_http_forever(&url, request, reqwest::Method::POST)
+            .await
+    }
+
+    #[tracing::instrument(skip_all)] // so that warning logs from retry_http_forever have context
+    async fn get_timeline_import_status(
+        &self,
+        tenant_shard_id: TenantShardId,
+        timeline_id: TimelineId,
+        generation: Generation,
+    ) -> Result<ShardImportStatus, RetryForeverError> {
+        let url = self
+            .base_url
+            .join("timeline_import_status")
+            .expect("Failed to build path");
+
+        let request = TimelineImportStatusRequest {
+            tenant_shard_id,
+            timeline_id,
+            generation,
+        };
+
+        let response: ShardImportStatus = self
+            .retry_http_forever(&url, request, reqwest::Method::GET)
+            .await?;
+        Ok(response)
     }
 }

pageserver/src/deletion_queue.rs

@@ -663,6 +663,7 @@ mod test {
     use camino::Utf8Path;
     use hex_literal::hex;
     use pageserver_api::key::Key;
+    use pageserver_api::models::ShardImportStatus;
     use pageserver_api::shard::ShardIndex;
     use pageserver_api::upcall_api::ReAttachResponseTenant;
     use remote_storage::{RemoteStorageConfig, RemoteStorageKind};
@@ -792,10 +793,20 @@ mod test {
             &self,
             _tenant_shard_id: TenantShardId,
             _timeline_id: TimelineId,
+            _generation: Generation,
             _status: pageserver_api::models::ShardImportStatus,
         ) -> Result<(), RetryForeverError> {
             unimplemented!()
         }
+
+        async fn get_timeline_import_status(
+            &self,
+            _tenant_shard_id: TenantShardId,
+            _timeline_id: TimelineId,
+            _generation: Generation,
+        ) -> Result<ShardImportStatus, RetryForeverError> {
+            unimplemented!()
+        }
     }
 
     async fn setup(test_name: &str) -> anyhow::Result<TestSetup> {

pageserver/src/feature_resolver.rs

@@ -0,0 +1,65 @@
+use std::{collections::HashMap, sync::Arc, time::Duration};
+
+use posthog_client_lite::{
+    FeatureResolverBackgroundLoop, PostHogClientConfig, PostHogEvaluationError,
+};
+use tokio_util::sync::CancellationToken;
+use utils::id::TenantId;
+
+use crate::config::PageServerConf;
+
+#[derive(Clone)]
+pub struct FeatureResolver {
+    inner: Option<Arc<FeatureResolverBackgroundLoop>>,
+}
+
+impl FeatureResolver {
+    pub fn new_disabled() -> Self {
+        Self { inner: None }
+    }
+
+    pub fn spawn(
+        conf: &PageServerConf,
+        shutdown_pageserver: CancellationToken,
+        handle: &tokio::runtime::Handle,
+    ) -> anyhow::Result<Self> {
+        // DO NOT block in this function: make it return as fast as possible to avoid startup delays.
+        if let Some(posthog_config) = &conf.posthog_config {
+            let inner = FeatureResolverBackgroundLoop::new(
+                PostHogClientConfig {
+                    server_api_key: posthog_config.server_api_key.clone(),
+                    client_api_key: posthog_config.client_api_key.clone(),
+                    project_id: posthog_config.project_id.clone(),
+                    private_api_url: posthog_config.private_api_url.clone(),
+                    public_api_url: posthog_config.public_api_url.clone(),
+                },
+                shutdown_pageserver,
+            );
+            let inner = Arc::new(inner);
+            // TODO: make this configurable
+            inner.clone().spawn(handle, Duration::from_secs(60));
+            Ok(FeatureResolver { inner: Some(inner) })
+        } else {
+            Ok(FeatureResolver { inner: None })
+        }
+    }
+
+    /// Evaluate a multivariate feature flag. Currently, we do not support any properties.
+    pub fn evaluate_multivariate(
+        &self,
+        flag_key: &str,
+        tenant_id: TenantId,
+    ) -> Result<String, PostHogEvaluationError> {
+        if let Some(inner) = &self.inner {
+            inner.feature_store().evaluate_multivariate(
+                flag_key,
+                &tenant_id.to_string(),
+                &HashMap::new(),
+            )
+        } else {
+            Err(PostHogEvaluationError::NotAvailable(
+                "PostHog integration is not enabled".to_string(),
+            ))
+        }
+    }
+}