Proxy release 2025-06-03 06:01 UTC

## Problem

fix https://github.com/neondatabase/neon/issues/12101; this is a quick
hack and we need better API in the future.

In `get_db_size`, we call `get_reldir_size` for every relation. However,
we do the same deserializing the reldir directory thing for every
relation. This creates huge CPU overhead.

## Summary of changes

Get and deserialize the reldir v1 key once and use it across all
get_rel_size requests.

---------

Signed-off-by: Alex Chi Z <chi@neon.tech>

.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"
@@ -4203,6 +4236,7 @@ name = "pagebench"
 version = "0.1.0"
 dependencies = [
  "anyhow",
+ "async-trait",
  "camino",
  "clap",
  "futures",
@@ -4211,12 +4245,15 @@ dependencies = [
  "humantime-serde",
  "pageserver_api",
  "pageserver_client",
+ "pageserver_page_api",
  "rand 0.8.5",
  "reqwest",
  "serde",
  "serde_json",
  "tokio",
+ "tokio-stream",
  "tokio-util",
+ "tonic 0.13.1",
  "tracing",
  "utils",
  "workspace_hack",
@@ -4269,8 +4306,10 @@ dependencies = [
  "enumset",
  "fail",
  "futures",
+ "hashlink",
  "hex",
  "hex-literal",
+ "http 1.1.0",
  "http-utils",
  "humantime",
  "humantime-serde",
@@ -4280,13 +4319,14 @@ 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 +4335,7 @@ dependencies = [
  "postgres_connection",
  "postgres_ffi",
  "postgres_initdb",
+ "posthog_client_lite",
  "pprof",
  "pq_proto",
  "procfs",
@@ -4305,7 +4346,7 @@ dependencies = [
  "reqwest",
  "rpds",
  "rstest",
- "rustls 0.23.18",
+ "rustls 0.23.27",
  "scopeguard",
  "send-future",
  "serde",
@@ -4324,13 +4365,17 @@ 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",
+ "tower 0.5.2",
  "tracing",
  "tracing-utils",
+ "twox-hash",
  "url",
  "utils",
  "uuid",
@@ -4355,7 +4400,7 @@ dependencies = [
  "humantime",
  "humantime-serde",
  "itertools 0.10.5",
- "nix 0.27.1",
+ "nix 0.30.1",
  "once_cell",
  "postgres_backend",
  "postgres_ffi",
@@ -4416,6 +4461,22 @@ 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 +4853,14 @@ 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",
 ]
@@ -4853,11 +4914,16 @@ 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",
 ]
 
@@ -4906,7 +4972,7 @@ dependencies = [
  "inferno 0.12.0",
  "num",
  "paste",
- "prost 0.13.3",
+ "prost 0.13.5",
 ]
 
 [[package]]
@@ -5011,12 +5077,12 @@ 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]]
@@ -5054,7 +5120,7 @@ dependencies = [
  "once_cell",
  "petgraph",
  "prettyplease",
- "prost 0.13.3",
+ "prost 0.13.5",
  "prost-types 0.13.3",
  "regex",
  "syn 2.0.100",
@@ -5076,9 +5142,9 @@ 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",
@@ -5102,7 +5168,7 @@ 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]]
@@ -5150,7 +5216,7 @@ 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",
@@ -5184,7 +5250,7 @@ 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",
@@ -5203,13 +5269,14 @@ 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",
@@ -5426,13 +5493,13 @@ 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",
 ]
@@ -5880,15 +5947,15 @@ 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",
 ]
@@ -5977,6 +6044,17 @@ 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"
@@ -6028,7 +6106,7 @@ dependencies = [
  "regex",
  "remote_storage",
  "reqwest",
- "rustls 0.23.18",
+ "rustls 0.23.27",
  "safekeeper_api",
  "safekeeper_client",
  "scopeguard",
@@ -6045,7 +6123,7 @@ dependencies = [
  "tokio",
  "tokio-io-timeout",
  "tokio-postgres",
- "tokio-rustls 0.26.0",
+ "tokio-rustls 0.26.2",
  "tokio-stream",
  "tokio-tar",
  "tokio-util",
@@ -6217,7 +6295,7 @@ checksum = "255914a8e53822abd946e2ce8baa41d4cded6b8e938913b7f7b9da5b7ab44335"
 dependencies = [
  "httpdate",
  "reqwest",
- "rustls 0.23.18",
+ "rustls 0.23.27",
  "sentry-backtrace",
  "sentry-contexts",
  "sentry-core",
@@ -6646,11 +6724,11 @@ 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",
@@ -6692,7 +6770,7 @@ dependencies = [
  "regex",
  "reqwest",
  "routerify",
- "rustls 0.23.18",
+ "rustls 0.23.27",
  "rustls-native-certs 0.8.0",
  "safekeeper_api",
  "safekeeper_client",
@@ -6707,7 +6785,7 @@ dependencies = [
  "tokio",
  "tokio-postgres",
  "tokio-postgres-rustls",
- "tokio-rustls 0.26.0",
+ "tokio-rustls 0.26.2",
  "tokio-util",
  "tracing",
  "utils",
@@ -6745,7 +6823,7 @@ dependencies = [
  "postgres_ffi",
  "remote_storage",
  "reqwest",
- "rustls 0.23.18",
+ "rustls 0.23.27",
  "rustls-native-certs 0.8.0",
  "serde",
  "serde_json",
@@ -7279,10 +7357,10 @@ 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",
 ]
 
@@ -7326,12 +7404,11 @@ 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",
 ]
 
@@ -7429,7 +7506,7 @@ 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",
@@ -7448,18 +7525,41 @@ 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",
@@ -7467,9 +7567,9 @@ 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",
@@ -7479,6 +7579,19 @@ 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"
@@ -7487,16 +7600,11 @@ 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]]
@@ -7507,9 +7615,12 @@ 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",
@@ -7660,6 +7771,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "e8189decb5ac0fa7bc8b96b7cb9b2701d60d48805aca84a238004d665fcc4008"
 dependencies = [
  "matchers",
+ "nu-ansi-term",
  "once_cell",
  "regex",
  "serde",
@@ -7673,6 +7785,27 @@ 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"
@@ -7815,7 +7948,7 @@ dependencies = [
  "base64 0.22.1",
  "log",
  "once_cell",
- "rustls 0.23.18",
+ "rustls 0.23.27",
  "rustls-pki-types",
  "url",
  "webpki-roots",
@@ -7898,7 +8031,7 @@ dependencies = [
  "humantime",
  "jsonwebtoken",
  "metrics",
- "nix 0.27.1",
+ "nix 0.30.1",
  "once_cell",
  "pem",
  "pin-project-lite",
@@ -8010,7 +8143,7 @@ dependencies = [
  "pageserver_api",
  "postgres_ffi",
  "pprof",
- "prost 0.13.3",
+ "prost 0.13.5",
  "remote_storage",
  "serde",
  "serde_json",
@@ -8430,6 +8563,8 @@ dependencies = [
  "ahash",
  "anstream",
  "anyhow",
+ "axum",
+ "axum-core",
  "base64 0.13.1",
  "base64 0.21.7",
  "base64ct",
@@ -8452,10 +8587,8 @@ dependencies = [
  "fail",
  "form_urlencoded",
  "futures-channel",
- "futures-core",
  "futures-executor",
  "futures-io",
- "futures-task",
  "futures-util",
  "generic-array",
  "getrandom 0.2.11",
@@ -8466,14 +8599,14 @@ 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",
@@ -8485,19 +8618,18 @@ dependencies = [
  "once_cell",
  "p256 0.13.2",
  "parquet",
- "percent-encoding",
  "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",
@@ -8515,15 +8647,15 @@ 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",
@@ -23,6 +24,7 @@ members = [
     "libs/postgres_ffi",
     "libs/safekeeper_api",
     "libs/desim",
+    "libs/neon-shmem",
     "libs/utils",
     "libs/consumption_metrics",
     "libs/postgres_backend",
@@ -127,7 +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"
@@ -147,7 +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"
@@ -197,7 +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"] }
 
@@ -244,6 +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/" }
@@ -251,23 +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/" }
@@ -277,7 +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:${PATH}"
 ARG RUSTFILT_VERSION=0.2.1
@@ -310,13 +310,13 @@ RUN curl -sSO https://static.rust-lang.org/rustup/dist/$(uname -m)-unknown-linux
     . "$HOME/.cargo/env" && \
     cargo --version && rustup --version && \
     rustup component add llvm-tools rustfmt clippy && \
-    cargo install rustfilt            --version ${RUSTFILT_VERSION} && \
-    cargo install cargo-hakari        --version ${CARGO_HAKARI_VERSION} && \
-    cargo install cargo-deny --locked --version ${CARGO_DENY_VERSION} && \
-    cargo install cargo-hack          --version ${CARGO_HACK_VERSION} && \
-    cargo install cargo-nextest       --version ${CARGO_NEXTEST_VERSION} && \
-    cargo install cargo-chef --locked --version ${CARGO_CHEF_VERSION} && \
-    cargo install diesel_cli          --version ${CARGO_DIESEL_CLI_VERSION} \
+    cargo install rustfilt            --version ${RUSTFILT_VERSION} --locked && \
+    cargo install cargo-hakari        --version ${CARGO_HAKARI_VERSION} --locked && \
+    cargo install cargo-deny          --version ${CARGO_DENY_VERSION} --locked && \
+    cargo install cargo-hack          --version ${CARGO_HACK_VERSION} --locked && \
+    cargo install cargo-nextest       --version ${CARGO_NEXTEST_VERSION} --locked && \
+    cargo install cargo-chef          --version ${CARGO_CHEF_VERSION} --locked && \
+    cargo install diesel_cli          --version ${CARGO_DIESEL_CLI_VERSION} --locked \
                                       --features postgres-bundled --no-default-features && \
     rm -rf /home/nonroot/.cargo/registry && \
     rm -rf /home/nonroot/.cargo/git

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 $PG_VERSION" && exit 0 \
+        ;; \
+    esac && \
+	wget https://github.com/knizhnik/online_advisor/archive/refs/tags/1.0.tar.gz -O online_advisor.tar.gz && \
+    echo "059b7d9e5a90013a58bdd22e9505b88406ce05790675eb2d8434e5b215652d54 online_advisor.tar.gz" | sha256sum --check && \
+    mkdir online_advisor-src && cd online_advisor-src && tar xzf ../online_advisor.tar.gz --strip-components=1 -C .
+
+FROM pg-build AS online_advisor-build
+COPY --from=online_advisor-src /ext-src/ /ext-src/
+WORKDIR /ext-src/
+RUN if [ -d online_advisor-src ]; then \
+	    cd online_advisor-src && \
+        make -j install && \
+        echo 'trusted = true' >> /usr/local/pgsql/share/extension/online_advisor.control; \
+    fi
+
 #########################################################################################
 #
 # Layer "pg_hashids-build"
@@ -1648,6 +1680,7 @@ COPY --from=pg_jsonschema-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg_graphql-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg_tiktoken-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=hypopg-build /usr/local/pgsql/ /usr/local/pgsql/
+COPY --from=online_advisor-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pg_hashids-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=rum-build /usr/local/pgsql/ /usr/local/pgsql/
 COPY --from=pgtap-build /usr/local/pgsql/ /usr/local/pgsql/
@@ -1751,17 +1784,17 @@ ARG TARGETARCH
 RUN if [ "$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/

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/bin/compute_ctl.rs

@@ -57,21 +57,6 @@ use tracing::{error, info};
 use url::Url;
 use utils::failpoint_support;
 
-// Compatibility hack: if the control plane specified any remote-ext-config
-// use the default value for extension storage proxy gateway.
-// Remove this once the control plane is updated to pass the gateway URL
-fn parse_remote_ext_base_url(arg: &str) -> Result<String> {
-    const FALLBACK_PG_EXT_GATEWAY_BASE_URL: &str =
-        "http://pg-ext-s3-gateway.pg-ext-s3-gateway.svc.cluster.local";
-
-    Ok(if arg.starts_with("http") {
-        arg
-    } else {
-        FALLBACK_PG_EXT_GATEWAY_BASE_URL
-    }
-    .to_owned())
-}
-
 #[derive(Parser)]
 #[command(rename_all = "kebab-case")]
 struct Cli {
@@ -79,9 +64,8 @@ struct Cli {
     pub pgbin: String,
 
     /// The base URL for the remote extension storage proxy gateway.
-    /// Should be in the form of `http(s)://<gateway-hostname>[:<port>]`.
-    #[arg(short = 'r', long, value_parser = parse_remote_ext_base_url, alias = "remote-ext-config")]
-    pub remote_ext_base_url: Option<String>,
+    #[arg(short = 'r', long)]
+    pub remote_ext_base_url: Option<Url>,
 
     /// The port to bind the external listening HTTP server to. Clients running
     /// outside the compute will talk to the compute through this port. Keep
@@ -136,6 +120,10 @@ struct Cli {
         requires = "compute-id"
     )]
     pub control_plane_uri: Option<String>,
+
+    /// Interval in seconds for collecting installed extensions statistics
+    #[arg(long, default_value = "3600")]
+    pub installed_extensions_collection_interval: u64,
 }
 
 fn main() -> Result<()> {
@@ -179,6 +167,7 @@ fn main() -> Result<()> {
             cgroup: cli.cgroup,
             #[cfg(target_os = "linux")]
             vm_monitor_addr: cli.vm_monitor_addr,
+            installed_extensions_collection_interval: cli.installed_extensions_collection_interval,
         },
         config,
     )?;
@@ -271,18 +260,4 @@ mod test {
     fn verify_cli() {
         Cli::command().debug_assert()
     }
-
-    #[test]
-    fn parse_pg_ext_gateway_base_url() {
-        let arg = "http://pg-ext-s3-gateway2";
-        let result = super::parse_remote_ext_base_url(arg).unwrap();
-        assert_eq!(result, arg);
-
-        let arg = "pg-ext-s3-gateway";
-        let result = super::parse_remote_ext_base_url(arg).unwrap();
-        assert_eq!(
-            result,
-            "http://pg-ext-s3-gateway.pg-ext-s3-gateway.svc.cluster.local"
-        );
-    }
 }

compute_tools/src/bin/fast_import.rs

@@ -339,6 +339,8 @@ async fn run_dump_restore(
     destination_connstring: String,
 ) -> Result<(), anyhow::Error> {
     let dumpdir = workdir.join("dumpdir");
+    let num_jobs = num_cpus::get().to_string();
+    info!("using {num_jobs} jobs for dump/restore");
 
     let common_args = [
         // schema mapping (prob suffices to specify them on one side)
@@ -354,7 +356,7 @@ async fn run_dump_restore(
         "directory".to_string(),
         // concurrency
         "--jobs".to_string(),
-        num_cpus::get().to_string(),
+        num_jobs,
         // progress updates
         "--verbose".to_string(),
     ];

compute_tools/src/compute.rs

@@ -31,6 +31,7 @@ use std::time::{Duration, Instant};
 use std::{env, fs};
 use tokio::spawn;
 use tracing::{Instrument, debug, error, info, instrument, warn};
+use url::Url;
 use utils::id::{TenantId, TimelineId};
 use utils::lsn::Lsn;
 use utils::measured_stream::MeasuredReader;
@@ -96,7 +97,10 @@ pub struct ComputeNodeParams {
     pub internal_http_port: u16,
 
     /// the address of extension storage proxy gateway
-    pub remote_ext_base_url: Option<String>,
+    pub remote_ext_base_url: Option<Url>,
+
+    /// Interval for installed extensions collection
+    pub installed_extensions_collection_interval: u64,
 }
 
 /// Compute node info shared across several `compute_ctl` threads.
@@ -695,25 +699,18 @@ 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);
@@ -749,17 +746,7 @@ impl ComputeNode {
 
             let conf = self.get_tokio_conn_conf(None);
             tokio::task::spawn(async {
-                let res = get_installed_extensions(conf).await;
-                match res {
-                    Ok(extensions) => {
-                        info!(
-                            "[NEON_EXT_STAT] {}",
-                            serde_json::to_string(&extensions)
-                                .expect("failed to serialize extensions list")
-                        );
-                    }
-                    Err(err) => error!("could not get installed extensions: {err:?}"),
-                }
+                let _ = installed_extensions(conf).await;
             });
         }
 
@@ -789,6 +776,9 @@ impl ComputeNode {
         // Log metrics so that we can search for slow operations in logs
         info!(?metrics, postmaster_pid = %postmaster_pid, "compute start finished");
 
+        // Spawn the extension stats background task
+        self.spawn_extension_stats_task();
+
         if pspec.spec.prewarm_lfc_on_startup {
             self.prewarm_lfc();
         }
@@ -2199,6 +2189,41 @@ LIMIT 100",
             info!("Pageserver config changed");
         }
     }
+
+    pub fn spawn_extension_stats_task(&self) {
+        let conf = self.tokio_conn_conf.clone();
+        let installed_extensions_collection_interval =
+            self.params.installed_extensions_collection_interval;
+        tokio::spawn(async move {
+            // An initial sleep is added to ensure that two collections don't happen at the same time.
+            // The first collection happens during compute startup.
+            tokio::time::sleep(tokio::time::Duration::from_secs(
+                installed_extensions_collection_interval,
+            ))
+            .await;
+            let mut interval = tokio::time::interval(tokio::time::Duration::from_secs(
+                installed_extensions_collection_interval,
+            ));
+            loop {
+                interval.tick().await;
+                let _ = installed_extensions(conf.clone()).await;
+            }
+        });
+    }
+}
+
+pub async fn installed_extensions(conf: tokio_postgres::Config) -> Result<()> {
+    let res = get_installed_extensions(conf).await;
+    match res {
+        Ok(extensions) => {
+            info!(
+                "[NEON_EXT_STAT] {}",
+                serde_json::to_string(&extensions).expect("failed to serialize extensions list")
+            );
+        }
+        Err(err) => error!("could not get installed extensions: {err:?}"),
+    }
+    Ok(())
 }
 
 pub fn forward_termination_signal() {

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/extension_server.rs

@@ -83,6 +83,7 @@ use reqwest::StatusCode;
 use tar::Archive;
 use tracing::info;
 use tracing::log::warn;
+use url::Url;
 use zstd::stream::read::Decoder;
 
 use crate::metrics::{REMOTE_EXT_REQUESTS_TOTAL, UNKNOWN_HTTP_STATUS};
@@ -158,14 +159,14 @@ fn parse_pg_version(human_version: &str) -> PostgresMajorVersion {
 pub async fn download_extension(
     ext_name: &str,
     ext_path: &RemotePath,
-    remote_ext_base_url: &str,
+    remote_ext_base_url: &Url,
     pgbin: &str,
 ) -> Result<u64> {
     info!("Download extension {:?} from {:?}", ext_name, ext_path);
 
     // TODO add retry logic
     let download_buffer =
-        match download_extension_tar(remote_ext_base_url, &ext_path.to_string()).await {
+        match download_extension_tar(remote_ext_base_url.as_str(), &ext_path.to_string()).await {
             Ok(buffer) => buffer,
             Err(error_message) => {
                 return Err(anyhow::anyhow!(

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

@@ -20,7 +20,7 @@ first_path="$(ldconfig --verbose 2>/dev/null \
     | grep --invert-match ^$'\t' \
     | cut --delimiter=: --fields=1 \
     | head --lines=1)"
-test "$first_path" == '/usr/local/lib' || true # Remove the || true in a follow-up PR. Needed for backwards compat.
+test "$first_path" == '/usr/local/lib'
 
 echo "Waiting pageserver become ready."
 while ! nc -z pageserver 6400; do

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

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/metrics/src/hll.rs

@@ -107,7 +107,7 @@ impl<const N: usize> MetricType for HyperLogLogState<N> {
 }
 
 impl<const N: usize> HyperLogLogState<N> {
-    pub fn measure(&self, item: &impl Hash) {
+    pub fn measure(&self, item: &(impl Hash + ?Sized)) {
         // changing the hasher will break compatibility with previous measurements.
         self.record(BuildHasherDefault::<xxh3::Hash64>::default().hash_one(item));
     }

libs/metrics/src/lib.rs

@@ -27,6 +27,7 @@ pub use prometheus::{
 
 pub mod launch_timestamp;
 mod wrappers;
+pub use prometheus;
 pub use wrappers::{CountedReader, CountedWriter};
 mod hll;
 pub use hll::{HyperLogLog, HyperLogLogState, HyperLogLogVec};

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
@@ -667,9 +713,12 @@ impl Default for ConfigToml {
             enable_tls_page_service_api: false,
             dev_mode: false,
             timeline_import_config: TimelineImportConfig {
-                import_job_concurrency: NonZeroUsize::new(128).unwrap(),
-                import_job_soft_size_limit: NonZeroUsize::new(1024 * 1024 * 1024).unwrap(),
+                import_job_concurrency: NonZeroUsize::new(32).unwrap(),
+                import_job_soft_size_limit: NonZeroUsize::new(256 * 1024 * 1024).unwrap(),
+                import_job_checkpoint_threshold: NonZeroUsize::new(32).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 +336,33 @@ 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,
+    /// Soft limit for the job size
+    /// This needs to remain constant throughout the import
+    pub job_soft_size_limit: usize,
+}
+
 impl ShardImportStatus {
     pub fn is_terminal(&self) -> bool {
         match self {
-            ShardImportStatus::InProgress => false,
+            ShardImportStatus::InProgress(_) => false,
             ShardImportStatus::Done | ShardImportStatus::Error(_) => true,
         }
     }
@@ -386,6 +405,8 @@ 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 +635,10 @@ 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 +768,12 @@ 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 +819,8 @@ 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 +922,12 @@ 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 +967,8 @@ impl TenantConfig {
             gc_compaction_initial_threshold_kb,
             gc_compaction_ratio_percent,
             sampling_ratio,
+            relsize_snapshot_cache_capacity,
+            basebackup_cache_enabled,
         })
     }
 
@@ -1036,6 +1077,12 @@ 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 +1850,6 @@ pub struct TopTenantShardsResponse {
 }
 
 pub mod virtual_file {
-    use std::sync::LazyLock;
 
     #[derive(
         Copy,
@@ -1851,15 +1897,7 @@ 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
         }
     }
 
@@ -1896,7 +1934,7 @@ pub enum PagestreamFeMessage {
 }
 
 // Wrapped in libpq CopyData
-#[derive(strum_macros::EnumProperty)]
+#[derive(Debug, strum_macros::EnumProperty)]
 pub enum PagestreamBeMessage {
     Exists(PagestreamExistsResponse),
     Nblocks(PagestreamNblocksResponse),
@@ -2007,7 +2045,7 @@ pub enum PagestreamProtocolVersion {
 
 pub type RequestId = u64;
 
-#[derive(Debug, PartialEq, Eq, Clone, Copy)]
+#[derive(Debug, Default, PartialEq, Eq, Clone, Copy)]
 pub struct PagestreamRequest {
     pub reqid: RequestId,
     pub request_lsn: Lsn,
@@ -2026,7 +2064,7 @@ pub struct PagestreamNblocksRequest {
     pub rel: RelTag,
 }
 
-#[derive(Debug, PartialEq, Eq, Clone, Copy)]
+#[derive(Debug, Default, PartialEq, Eq, Clone, Copy)]
 pub struct PagestreamGetPageRequest {
     pub hdr: PagestreamRequest,
     pub rel: RelTag,

libs/pageserver_api/src/reltag.rs

@@ -24,7 +24,7 @@ use serde::{Deserialize, Serialize};
 // FIXME: should move 'forknum' as last field to keep this consistent with Postgres.
 // Then we could replace the custom Ord and PartialOrd implementations below with
 // deriving them. This will require changes in walredoproc.c.
-#[derive(Debug, PartialEq, Eq, Hash, Clone, Copy, Serialize, Deserialize)]
+#[derive(Debug, Default, PartialEq, Eq, Hash, Clone, Copy, Serialize, Deserialize)]
 pub struct RelTag {
     pub forknum: u8,
     pub spcnode: Oid,
@@ -184,12 +184,12 @@ pub enum SlruKind {
     MultiXactOffsets,
 }
 
-impl SlruKind {
-    pub fn to_str(&self) -> &'static str {
+impl fmt::Display for SlruKind {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
         match self {
-            Self::Clog => "pg_xact",
-            Self::MultiXactMembers => "pg_multixact/members",
-            Self::MultiXactOffsets => "pg_multixact/offsets",
+            Self::Clog => write!(f, "pg_xact"),
+            Self::MultiXactMembers => write!(f, "pg_multixact/members"),
+            Self::MultiXactOffsets => write!(f, "pg_multixact/offsets"),
         }
     }
 }

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/posthog_client_lite/Cargo.toml

@@ -6,9 +6,14 @@ license.workspace = true
 
 [dependencies]
 anyhow.workspace = true
+arc-swap.workspace = true
 reqwest.workspace = true
-serde.workspace = true
 serde_json.workspace = true
+serde.workspace = true
 sha2.workspace = true
-workspace_hack.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 +1,64 @@
+//! 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 tracing::{Instrument, info_span};
+
+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!("Feature flag updated");
+                }
+                tracing::info!("PostHog feature resolver stopped");
+            }
+            .instrument(info_span!("posthog_feature_resolver")),
+        );
+    }
+
+    pub fn feature_store(&self) -> Arc<FeatureStore> {
+        self.feature_store.load_full()
+    }
+}

libs/posthog_client_lite/src/lib.rs

@@ -1,5 +1,9 @@
 //! 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};
@@ -20,8 +24,7 @@ pub enum PostHogEvaluationError {
 
 #[derive(Deserialize)]
 pub struct LocalEvaluationResponse {
-    #[allow(dead_code)]
-    flags: Vec<LocalEvaluationFlag>,
+    pub flags: Vec<LocalEvaluationFlag>,
 }
 
 #[derive(Deserialize)]
@@ -34,7 +37,7 @@ pub struct LocalEvaluationFlag {
 #[derive(Deserialize)]
 pub struct LocalEvaluationFlagFilters {
     groups: Vec<LocalEvaluationFlagFilterGroup>,
-    multivariate: LocalEvaluationFlagMultivariate,
+    multivariate: Option<LocalEvaluationFlagMultivariate>,
 }
 
 #[derive(Deserialize)]
@@ -94,6 +97,12 @@ impl FeatureStore {
         }
     }
 
+    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 {
@@ -245,7 +254,7 @@ impl FeatureStore {
         }
     }
 
-    /// Evaluate a multivariate feature flag. Returns `None` if the flag is not available or if there are errors
+    /// Evaluate a multivariate feature flag. Returns an error if the flag is not available or if there are errors
     /// during the evaluation.
     ///
     /// The parsing logic is as follows:
@@ -263,10 +272,15 @@ impl FeatureStore {
     /// 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.
+    ///
+    /// Error handling: the caller should inspect the error and decide the behavior when a feature flag
+    /// cannot be evaluated (i.e., default to false if it cannot be resolved). The error should *not* be
+    /// propagated beyond where the feature flag gets resolved.
     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");
@@ -276,10 +290,39 @@ impl FeatureStore {
             flag_key,
             hash_on_global_rollout_percentage,
             hash_on_group_rollout_percentage,
-            &HashMap::new(),
+            properties,
         )
     }
 
+    /// Evaluate a boolean feature flag. Returns  an error if the flag is not available or if there are errors
+    /// during the evaluation.
+    ///
+    /// The parsing logic is as follows:
+    ///
+    /// * Generate a consistent hash for the tenant-feature.
+    /// * Match each filter group.
+    ///   - If a group is matched, it will first determine whether the user is in the range of the rollout
+    ///     percentage.
+    ///   - If the hash falls within the group's rollout percentage, return true.
+    /// * 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.
+    ///
+    /// Returns `Ok(())` if the feature flag evaluates to true. In the future, it will return a payload.
+    ///
+    /// Error handling: the caller should inspect the error and decide the behavior when a feature flag
+    /// cannot be evaluated (i.e., default to false if it cannot be resolved). The error should *not* be
+    /// propagated beyond where the feature flag gets resolved.
+    pub fn evaluate_boolean(
+        &self,
+        flag_key: &str,
+        user_id: &str,
+        properties: &HashMap<String, PostHogFlagFilterPropertyValue>,
+    ) -> Result<(), PostHogEvaluationError> {
+        let hash_on_global_rollout_percentage = Self::consistent_hash(user_id, flag_key, "boolean");
+        self.evaluate_boolean_inner(flag_key, hash_on_global_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.
@@ -306,6 +349,11 @@ impl FeatureStore {
                     flag_key
                 )));
             }
+            let Some(ref multivariate) = flag_config.filters.multivariate else {
+                return Err(PostHogEvaluationError::Internal(format!(
+                    "No multivariate available, should use evaluate_boolean?: {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.
@@ -314,7 +362,7 @@ impl FeatureStore {
                     GroupEvaluationResult::MatchedAndOverride(variant) => return Ok(variant),
                     GroupEvaluationResult::MatchedAndEvaluate => {
                         let mut percentage = 0;
-                        for variant in &flag_config.filters.multivariate.variants {
+                        for variant in &multivariate.variants {
                             percentage += variant.rollout_percentage;
                             if self
                                 .evaluate_percentage(hash_on_global_rollout_percentage, percentage)
@@ -342,6 +390,89 @@ impl FeatureStore {
             )))
         }
     }
+
+    /// 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_boolean_inner(
+        &self,
+        flag_key: &str,
+        hash_on_global_rollout_percentage: f64,
+        properties: &HashMap<String, PostHogFlagFilterPropertyValue>,
+    ) -> Result<(), 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
+                )));
+            }
+            if flag_config.filters.multivariate.is_some() {
+                return Err(PostHogEvaluationError::Internal(format!(
+                    "This looks like a multivariate flag, should use evaluate_multivariate?: {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_global_rollout_percentage, properties)? {
+                    GroupEvaluationResult::MatchedAndOverride(_) => {
+                        return Err(PostHogEvaluationError::Internal(format!(
+                            "Boolean flag cannot have overrides: {}",
+                            flag_key
+                        )));
+                    }
+                    GroupEvaluationResult::MatchedAndEvaluate => {
+                        return Ok(());
+                    }
+                    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
+            )))
+        }
+    }
+
+    /// Infer whether a feature flag is a boolean flag by checking if it has a multivariate filter.
+    pub fn is_feature_flag_boolean(&self, flag_key: &str) -> Result<bool, PostHogEvaluationError> {
+        if let Some(flag_config) = self.flags.get(flag_key) {
+            Ok(flag_config.filters.multivariate.is_none())
+        } else {
+            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.
@@ -360,37 +491,16 @@ impl FeatureStore {
 /// 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 server API key.
-    server_api_key: String,
-    /// The client API key.
-    client_api_key: String,
-    /// The project ID.
-    project_id: String,
-    /// The private API URL.
-    private_api_url: String,
-    /// The public API URL.
-    public_api_url: String,
+    /// The config.
+    config: PostHogClientConfig,
     /// The HTTP client.
     client: reqwest::Client,
 }
 
 impl PostHogClient {
-    pub fn new(
-        server_api_key: String,
-        client_api_key: String,
-        project_id: String,
-        private_api_url: String,
-        public_api_url: String,
-    ) -> Self {
+    pub fn new(config: PostHogClientConfig) -> Self {
         let client = reqwest::Client::new();
-        Self {
-            server_api_key,
-            client_api_key,
-            project_id,
-            private_api_url,
-            public_api_url,
-            client,
-        }
+        Self { config, client }
     }
 
     pub fn new_with_us_region(
@@ -398,13 +508,13 @@ impl PostHogClient {
         client_api_key: String,
         project_id: String,
     ) -> Self {
-        Self::new(
+        Self::new(PostHogClientConfig {
             server_api_key,
             client_api_key,
             project_id,
-            "https://us.posthog.com".to_string(),
-            "https://us.i.posthog.com".to_string(),
-        )
+            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.
@@ -422,15 +532,23 @@ impl PostHogClient {
         // with bearer token of self.server_api_key
         let url = format!(
             "{}/api/projects/{}/feature_flags/local_evaluation",
-            self.private_api_url, self.project_id
+            self.config.private_api_url, self.config.project_id
         );
         let response = self
             .client
             .get(url)
-            .bearer_auth(&self.server_api_key)
+            .bearer_auth(&self.config.server_api_key)
             .send()
             .await?;
+        let status = response.status();
         let body = response.text().await?;
+        if !status.is_success() {
+            return Err(anyhow::anyhow!(
+                "Failed to get feature flags: {}, {}",
+                status,
+                body
+            ));
+        }
         Ok(serde_json::from_str(&body)?)
     }
 
@@ -446,11 +564,11 @@ impl PostHogClient {
     ) -> anyhow::Result<()> {
         // PUBLIC_URL/capture/
         // with bearer token of self.client_api_key
-        let url = format!("{}/capture/", self.public_api_url);
+        let url = format!("{}/capture/", self.config.public_api_url);
         self.client
             .post(url)
             .body(serde_json::to_string(&json!({
-                "api_key": self.client_api_key,
+                "api_key": self.config.client_api_key,
                 "distinct_id": distinct_id,
                 "event": event,
                 "properties": properties,
@@ -467,95 +585,162 @@ mod tests {
 
     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
-                }
+  "flags": [
+    {
+      "id": 141807,
+      "team_id": 152860,
+      "name": "",
+      "key": "image-compaction-boundary",
+      "filters": {
+        "groups": [
+          {
+            "variant": null,
+            "properties": [
+              {
+                "key": "plan_type",
+                "type": "person",
+                "value": [
+                  "free"
+                ],
+                "operator": "exact"
+              }
             ],
-            "group_type_mapping": {},
-            "cohorts": {}
-        }"#
+            "rollout_percentage": 40
+          },
+          {
+            "variant": null,
+            "properties": [],
+            "rollout_percentage": 10
+          }
+        ],
+        "payloads": {},
+        "multivariate": null
+      },
+      "deleted": false,
+      "active": true,
+      "ensure_experience_continuity": false,
+      "has_encrypted_payloads": false,
+      "version": 1
+    },
+    {
+      "id": 135586,
+      "team_id": 152860,
+      "name": "",
+      "key": "boolean-flag",
+      "filters": {
+        "groups": [
+          {
+            "variant": null,
+            "properties": [
+              {
+                "key": "plan_type",
+                "type": "person",
+                "value": [
+                  "free"
+                ],
+                "operator": "exact"
+              }
+            ],
+            "rollout_percentage": 47
+          }
+        ],
+        "payloads": {},
+        "multivariate": null
+      },
+      "deleted": false,
+      "active": true,
+      "ensure_experience_continuity": false,
+      "has_encrypted_payloads": false,
+      "version": 1
+    },
+    {
+      "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": 7
+    }
+  ],
+  "group_type_mapping": {},
+  "cohorts": {}
+}"#
     }
 
     #[test]
@@ -631,4 +816,125 @@ mod tests {
             Err(PostHogEvaluationError::NoConditionGroupMatched)
         ),);
     }
+
+    #[test]
+    fn evaluate_boolean_1() {
+        // The `boolean-flag` feature flag only has one group that matches on the free user.
+
+        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_boolean_inner("boolean-flag", 1.00, &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_boolean_inner("boolean-flag", 1.00, &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_boolean_inner("boolean-flag", 0.10, &properties);
+        assert!(variant.is_ok());
+
+        // It matches the group conditions but not the group rollout percentage.
+        let variant = store.evaluate_boolean_inner("boolean-flag", 1.00, &properties);
+        assert!(matches!(
+            variant,
+            Err(PostHogEvaluationError::NoConditionGroupMatched)
+        ),);
+    }
+
+    #[test]
+    fn evaluate_boolean_2() {
+        // The `image-compaction-boundary` feature flag has one group that matches on the free user and a group that matches on all users.
+
+        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_boolean_inner("image-compaction-boundary", 1.00, &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 the filtered group but the all user group.
+        let variant =
+            store.evaluate_boolean_inner("image-compaction-boundary", 1.00, &properties_unmatched);
+        assert!(matches!(
+            variant,
+            Err(PostHogEvaluationError::NoConditionGroupMatched)
+        ),);
+        let variant =
+            store.evaluate_boolean_inner("image-compaction-boundary", 0.05, &properties_unmatched);
+        assert!(variant.is_ok());
+
+        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 0.30 <= 0.40 and the properties are matched. Then it gets evaluated to the variant override.
+        let variant = store.evaluate_boolean_inner("image-compaction-boundary", 0.30, &properties);
+        assert!(variant.is_ok());
+
+        // It matches the group conditions but not the group rollout percentage.
+        let variant = store.evaluate_boolean_inner("image-compaction-boundary", 1.00, &properties);
+        assert!(matches!(
+            variant,
+            Err(PostHogEvaluationError::NoConditionGroupMatched)
+        ),);
+
+        // It matches the second "all" group conditions.
+        let variant = store.evaluate_boolean_inner("image-compaction-boundary", 0.09, &properties);
+        assert!(variant.is_ok());
+    }
 }

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/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/leaky_bucket.rs

@@ -28,6 +28,7 @@ use std::time::Duration;
 use tokio::sync::Notify;
 use tokio::time::Instant;
 
+#[derive(Clone, Copy)]
 pub struct LeakyBucketConfig {
     /// This is the "time cost" of a single request unit.
     /// Should loosely represent how long it takes to handle a request unit in active resource time.

libs/utils/src/lib.rs

@@ -73,6 +73,7 @@ pub mod error;
 /// async timeout helper
 pub mod timeout;
 
+pub mod span;
 pub mod sync;
 
 pub mod failpoint_support;

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/span.rs

@@ -0,0 +1,19 @@
+//! Tracing span helpers.
+
+/// Records the given fields in the current span, as a single call. The fields must already have
+/// been declared for the span (typically with empty values).
+#[macro_export]
+macro_rules! span_record {
+    ($($tokens:tt)*) => {$crate::span_record_in!(::tracing::Span::current(), $($tokens)*)};
+}
+
+/// Records the given fields in the given span, as a single call. The fields must already have been
+/// declared for the span (typically with empty values).
+#[macro_export]
+macro_rules! span_record_in {
+    ($span:expr, $($tokens:tt)*) => {
+        if let Some(meta) = $span.metadata() {
+            $span.record_all(&tracing::valueset!(meta.fields(), $($tokens)*));
+        }
+    };
+}

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 +17,70 @@ 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.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 +89,20 @@ 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
+tower.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/benches/bench_metrics.rs

@@ -264,10 +264,56 @@ mod propagation_of_cached_label_value {
     }
 }
 
+criterion_group!(histograms, histograms::bench_bucket_scalability);
+mod histograms {
+    use std::time::Instant;
+
+    use criterion::{BenchmarkId, Criterion};
+    use metrics::core::Collector;
+
+    pub fn bench_bucket_scalability(c: &mut Criterion) {
+        let mut g = c.benchmark_group("bucket_scalability");
+
+        for n in [1, 4, 8, 16, 32, 64, 128, 256] {
+            g.bench_with_input(BenchmarkId::new("nbuckets", n), &n, |b, n| {
+                b.iter_custom(|iters| {
+                    let buckets: Vec<f64> = (0..*n).map(|i| i as f64 * 100.0).collect();
+                    let histo = metrics::Histogram::with_opts(
+                        metrics::prometheus::HistogramOpts::new("name", "help")
+                            .buckets(buckets.clone()),
+                    )
+                    .unwrap();
+                    let start = Instant::now();
+                    for i in 0..usize::try_from(iters).unwrap() {
+                        histo.observe(buckets[i % buckets.len()]);
+                    }
+                    let elapsed = start.elapsed();
+                    // self-test
+                    let mfs = histo.collect();
+                    assert_eq!(mfs.len(), 1);
+                    let metrics = mfs[0].get_metric();
+                    assert_eq!(metrics.len(), 1);
+                    let histo = metrics[0].get_histogram();
+                    let buckets = histo.get_bucket();
+                    assert!(
+                        buckets
+                            .iter()
+                            .enumerate()
+                            .all(|(i, b)| b.get_cumulative_count()
+                                >= i as u64 * (iters / buckets.len() as u64))
+                    );
+                    elapsed
+                })
+            });
+        }
+    }
+}
+
 criterion_main!(
     label_values,
     single_metric_multicore_scalability,
-    propagation_of_cached_label_value
+    propagation_of_cached_label_value,
+    histograms,
 );
 
 /*
@@ -290,6 +336,14 @@ propagation_of_cached_label_value__naive/nthreads/8 time:   [211.50 ns 214.44 ns
 propagation_of_cached_label_value__long_lived_reference_per_thread/nthreads/1 time:   [14.135 ns 14.147 ns 14.160 ns]
 propagation_of_cached_label_value__long_lived_reference_per_thread/nthreads/4 time:   [14.243 ns 14.255 ns 14.268 ns]
 propagation_of_cached_label_value__long_lived_reference_per_thread/nthreads/8 time:   [14.470 ns 14.682 ns 14.895 ns]
+bucket_scalability/nbuckets/1     time:   [30.352 ns 30.353 ns 30.354 ns]
+bucket_scalability/nbuckets/4     time:   [30.464 ns 30.465 ns 30.467 ns]
+bucket_scalability/nbuckets/8     time:   [30.569 ns 30.575 ns 30.584 ns]
+bucket_scalability/nbuckets/16      time:   [30.961 ns 30.965 ns 30.969 ns]
+bucket_scalability/nbuckets/32      time:   [35.691 ns 35.707 ns 35.722 ns]
+bucket_scalability/nbuckets/64      time:   [47.829 ns 47.898 ns 47.974 ns]
+bucket_scalability/nbuckets/128     time:   [73.479 ns 73.512 ns 73.545 ns]
+bucket_scalability/nbuckets/256     time:   [127.92 ns 127.94 ns 127.96 ns]
 
 Results on an i3en.3xlarge instance
 
@@ -344,6 +398,14 @@ propagation_of_cached_label_value__naive/nthreads/8     time:   [434.87 ns 456.4
 propagation_of_cached_label_value__long_lived_reference_per_thread/nthreads/1     time:   [3.3767 ns 3.3974 ns 3.4220 ns]
 propagation_of_cached_label_value__long_lived_reference_per_thread/nthreads/4     time:   [3.6105 ns 4.2355 ns 5.1463 ns]
 propagation_of_cached_label_value__long_lived_reference_per_thread/nthreads/8     time:   [4.0889 ns 4.9714 ns 6.0779 ns]
+bucket_scalability/nbuckets/1     time:   [4.8455 ns 4.8542 ns 4.8646 ns]
+bucket_scalability/nbuckets/4     time:   [4.5663 ns 4.5722 ns 4.5787 ns]
+bucket_scalability/nbuckets/8     time:   [4.5531 ns 4.5670 ns 4.5842 ns]
+bucket_scalability/nbuckets/16      time:   [4.6392 ns 4.6524 ns 4.6685 ns]
+bucket_scalability/nbuckets/32      time:   [6.0302 ns 6.0439 ns 6.0589 ns]
+bucket_scalability/nbuckets/64      time:   [10.608 ns 10.644 ns 10.691 ns]
+bucket_scalability/nbuckets/128     time:   [22.178 ns 22.316 ns 22.483 ns]
+bucket_scalability/nbuckets/256     time:   [42.190 ns 42.328 ns 42.492 ns]
 
 Results on a Hetzner AX102 AMD Ryzen 9 7950X3D 16-Core Processor
 
@@ -362,5 +424,13 @@ propagation_of_cached_label_value__naive/nthreads/8     time:   [164.24 ns 170.1
 propagation_of_cached_label_value__long_lived_reference_per_thread/nthreads/1     time:   [2.2915 ns 2.2960 ns 2.3012 ns]
 propagation_of_cached_label_value__long_lived_reference_per_thread/nthreads/4     time:   [2.5726 ns 2.6158 ns 2.6624 ns]
 propagation_of_cached_label_value__long_lived_reference_per_thread/nthreads/8     time:   [2.7068 ns 2.8243 ns 2.9824 ns]
+bucket_scalability/nbuckets/1     time:   [6.3998 ns 6.4288 ns 6.4684 ns]
+bucket_scalability/nbuckets/4     time:   [6.3603 ns 6.3620 ns 6.3637 ns]
+bucket_scalability/nbuckets/8     time:   [6.1646 ns 6.1654 ns 6.1667 ns]
+bucket_scalability/nbuckets/16      time:   [6.1341 ns 6.1391 ns 6.1454 ns]
+bucket_scalability/nbuckets/32      time:   [8.2206 ns 8.2254 ns 8.2301 ns]
+bucket_scalability/nbuckets/64      time:   [13.988 ns 13.994 ns 14.000 ns]
+bucket_scalability/nbuckets/128     time:   [28.180 ns 28.216 ns 28.251 ns]
+bucket_scalability/nbuckets/256     time:   [54.914 ns 54.931 ns 54.951 ns]
 
 */

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 +1,239 @@
+// 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 gRPC status response (e.g. errors) will terminate the stream. The
+  // stream may be shared by multiple Postgres backends, so we avoid this by
+  // sending them as GetPageResponse.status_code 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 backwards compatibility: used when an older client version sends a class
+  // that a newer server version has removed.
+  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.
+  GetPageStatusCode status_code = 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_code != OK.
+  repeated bytes page_image = 4;
+}
+
+// A GetPageResponse status code.
+//
+// These are effectively equivalent to gRPC statuses. However, we use a bidirectional stream
+// (potentially shared by many backends), and a gRPC status response would terminate the stream so
+// we send GetPageResponse messages with these codes instead.
+enum GetPageStatusCode {
+  // Unknown status. For forwards compatibility: used when an older client version receives a new
+  // status code from a newer server version.
+  GET_PAGE_STATUS_CODE_UNKNOWN = 0;
+  // The request was successful.
+  GET_PAGE_STATUS_CODE_OK = 1;
+  // The page did not exist. The tenant/timeline/shard has already been
+  // validated during stream setup.
+  GET_PAGE_STATUS_CODE_NOT_FOUND = 2;
+  // The request was invalid.
+  GET_PAGE_STATUS_CODE_INVALID_REQUEST = 3;
+  // The request failed due to an internal server error.
+  GET_PAGE_STATUS_CODE_INTERNAL_ERROR = 4;
+  // The tenant is rate limited. Slow down and retry later.
+  GET_PAGE_STATUS_CODE_SLOW_DOWN = 5;
+  // NB: shutdown errors are emitted as a gRPC Unavailable status.
+  //
+  // TODO: consider adding a GET_PAGE_STATUS_CODE_LAYER_DOWNLOAD in the case of a layer download.
+  // This could free up the server task to process other requests while the 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 +1,610 @@
+//! 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 std::fmt::Display;
+
+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:?}"))
+    }
+}
+
+impl From<ProtocolError> for tonic::Status {
+    fn from(err: ProtocolError) -> Self {
+        tonic::Status::invalid_argument(format!("{err}"))
+    }
+}
+
+/// 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. If not given, the request will read at the
+    /// request_lsn and wait for it to arrive if necessary. 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 Display for ReadLsn {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        let req_lsn = self.request_lsn;
+        if let Some(mod_lsn) = self.not_modified_since_lsn {
+            write!(f, "{req_lsn}>={mod_lsn}")
+        } else {
+            req_lsn.fmt(f)
+        }
+    }
+}
+
+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 class. 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_code: GetPageStatusCode,
+    /// 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_code: pb.status_code.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_code: response.status_code.into(),
+            reason: response.reason.unwrap_or_default(),
+            page_image: response.page_images.into_vec(),
+        }
+    }
+}
+
+/// A GetPage response status code.
+///
+/// These are effectively equivalent to gRPC statuses. However, we use a bidirectional stream
+/// (potentially shared by many backends), and a gRPC status response would terminate the stream so
+/// we send GetPageResponse messages with these codes instead.
+#[derive(Clone, Copy, Debug)]
+pub enum GetPageStatusCode {
+    /// 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.
+    InvalidRequest,
+    /// The request failed due to an internal server error.
+    InternalError,
+    /// The tenant is rate limited. Slow down and retry later.
+    SlowDown,
+}
+
+impl From<proto::GetPageStatusCode> for GetPageStatusCode {
+    fn from(pb: proto::GetPageStatusCode) -> Self {
+        match pb {
+            proto::GetPageStatusCode::Unknown => Self::Unknown,
+            proto::GetPageStatusCode::Ok => Self::Ok,
+            proto::GetPageStatusCode::NotFound => Self::NotFound,
+            proto::GetPageStatusCode::InvalidRequest => Self::InvalidRequest,
+            proto::GetPageStatusCode::InternalError => Self::InternalError,
+            proto::GetPageStatusCode::SlowDown => Self::SlowDown,
+        }
+    }
+}
+
+impl From<i32> for GetPageStatusCode {
+    fn from(status_code: i32) -> Self {
+        proto::GetPageStatusCode::try_from(status_code)
+            .unwrap_or(proto::GetPageStatusCode::Unknown)
+            .into()
+    }
+}
+
+impl From<GetPageStatusCode> for proto::GetPageStatusCode {
+    fn from(status_code: GetPageStatusCode) -> Self {
+        match status_code {
+            GetPageStatusCode::Unknown => Self::Unknown,
+            GetPageStatusCode::Ok => Self::Ok,
+            GetPageStatusCode::NotFound => Self::NotFound,
+            GetPageStatusCode::InvalidRequest => Self::InvalidRequest,
+            GetPageStatusCode::InternalError => Self::InternalError,
+            GetPageStatusCode::SlowDown => Self::SlowDown,
+        }
+    }
+}
+
+impl From<GetPageStatusCode> for i32 {
+    fn from(status_code: GetPageStatusCode) -> Self {
+        proto::GetPageStatusCode::from(status_code).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> {
+        // TODO: can a segment legitimately be empty?
+        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/Cargo.toml

@@ -8,6 +8,7 @@ license.workspace = true
 
 [dependencies]
 anyhow.workspace = true
+async-trait.workspace = true
 camino.workspace = true
 clap.workspace = true
 futures.workspace = true
@@ -15,14 +16,17 @@ hdrhistogram.workspace = true
 humantime.workspace = true
 humantime-serde.workspace = true
 rand.workspace = true
-reqwest.workspace=true
+reqwest.workspace = true
 serde.workspace = true
 serde_json.workspace = true
 tracing.workspace = true
 tokio.workspace = true
+tokio-stream.workspace = true
 tokio-util.workspace = true
+tonic.workspace = true
 
 pageserver_client.workspace = true
 pageserver_api.workspace = true
+pageserver_page_api.workspace = true
 utils = { path = "../../libs/utils/" }
 workspace_hack = { version = "0.1", path = "../../workspace_hack" }

pageserver/pagebench/src/cmd/getpage_latest_lsn.rs

@@ -7,11 +7,15 @@ use std::sync::{Arc, Mutex};
 use std::time::{Duration, Instant};
 
 use anyhow::Context;
+use async_trait::async_trait;
 use camino::Utf8PathBuf;
 use pageserver_api::key::Key;
 use pageserver_api::keyspace::KeySpaceAccum;
-use pageserver_api::models::{PagestreamGetPageRequest, PagestreamRequest};
+use pageserver_api::models::{
+    PagestreamGetPageRequest, PagestreamGetPageResponse, PagestreamRequest,
+};
 use pageserver_api::shard::TenantShardId;
+use pageserver_page_api::proto;
 use rand::prelude::*;
 use tokio::task::JoinSet;
 use tokio_util::sync::CancellationToken;
@@ -22,6 +26,12 @@ use utils::lsn::Lsn;
 use crate::util::tokio_thread_local_stats::AllThreadLocalStats;
 use crate::util::{request_stats, tokio_thread_local_stats};
 
+#[derive(clap::ValueEnum, Clone, Debug)]
+enum Protocol {
+    Libpq,
+    Grpc,
+}
+
 /// GetPage@LatestLSN, uniformly distributed across the compute-accessible keyspace.
 #[derive(clap::Parser)]
 pub(crate) struct Args {
@@ -35,6 +45,8 @@ pub(crate) struct Args {
     num_clients: NonZeroUsize,
     #[clap(long)]
     runtime: Option<humantime::Duration>,
+    #[clap(long, value_enum, default_value = "libpq")]
+    protocol: Protocol,
     /// Each client sends requests at the given rate.
     ///
     /// If a request takes too long and we should be issuing a new request already,
@@ -65,6 +77,9 @@ pub(crate) struct Args {
     #[clap(long, default_value = "1")]
     queue_depth: NonZeroUsize,
 
+    #[clap(long)]
+    only_relnode: Option<u32>,
+
     targets: Option<Vec<TenantTimelineId>>,
 }
 
@@ -206,7 +221,12 @@ 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();
@@ -295,7 +315,20 @@ async fn main_impl(
                 .unwrap();
 
         Box::pin(async move {
-            client_libpq(args, worker_id, ss, cancel, rps_period, ranges, weights).await
+            let client: Box<dyn Client> = match args.protocol {
+                Protocol::Libpq => Box::new(
+                    LibpqClient::new(args.page_service_connstring.clone(), worker_id.timeline)
+                        .await
+                        .unwrap(),
+                ),
+
+                Protocol::Grpc => Box::new(
+                    GrpcClient::new(args.page_service_connstring.clone(), worker_id.timeline)
+                        .await
+                        .unwrap(),
+                ),
+            };
+            run_worker(args, client, ss, cancel, rps_period, ranges, weights).await
         })
     };
 
@@ -347,23 +380,15 @@ async fn main_impl(
     anyhow::Ok(())
 }
 
-async fn client_libpq(
+async fn run_worker(
     args: &Args,
-    worker_id: WorkerId,
+    mut client: Box<dyn Client>,
     shared_state: Arc<SharedState>,
     cancel: CancellationToken,
     rps_period: Option<Duration>,
     ranges: Vec<KeyRange>,
     weights: rand::distributions::weighted::WeightedIndex<i128>,
 ) {
-    let client = pageserver_client::page_service::Client::new(args.page_service_connstring.clone())
-        .await
-        .unwrap();
-    let mut client = client
-        .pagestream(worker_id.timeline.tenant_id, worker_id.timeline.timeline_id)
-        .await
-        .unwrap();
-
     shared_state.start_work_barrier.wait().await;
     let client_start = Instant::now();
     let mut ticks_processed = 0;
@@ -407,12 +432,12 @@ async fn client_libpq(
                     blkno: block_no,
                 }
             };
-            client.getpage_send(req).await.unwrap();
+            client.send_get_page(req).await.unwrap();
             inflight.push_back(start);
         }
 
         let start = inflight.pop_front().unwrap();
-        client.getpage_recv().await.unwrap();
+        client.recv_get_page().await.unwrap();
         let end = Instant::now();
         shared_state.live_stats.request_done();
         ticks_processed += 1;
@@ -434,3 +459,104 @@ async fn client_libpq(
         }
     }
 }
+
+/// A benchmark client, to allow switching out the transport protocol.
+///
+/// For simplicity, this just uses separate asynchronous send/recv methods. The send method could
+/// return a future that resolves when the response is received, but we don't really need it.
+#[async_trait]
+trait Client: Send {
+    /// Sends an asynchronous GetPage request to the pageserver.
+    async fn send_get_page(&mut self, req: PagestreamGetPageRequest) -> anyhow::Result<()>;
+
+    /// Receives the next GetPage response from the pageserver.
+    async fn recv_get_page(&mut self) -> anyhow::Result<PagestreamGetPageResponse>;
+}
+
+/// A libpq-based Pageserver client.
+struct LibpqClient {
+    inner: pageserver_client::page_service::PagestreamClient,
+}
+
+impl LibpqClient {
+    async fn new(connstring: String, ttid: TenantTimelineId) -> anyhow::Result<Self> {
+        let inner = pageserver_client::page_service::Client::new(connstring)
+            .await?
+            .pagestream(ttid.tenant_id, ttid.timeline_id)
+            .await?;
+        Ok(Self { inner })
+    }
+}
+
+#[async_trait]
+impl Client for LibpqClient {
+    async fn send_get_page(&mut self, req: PagestreamGetPageRequest) -> anyhow::Result<()> {
+        self.inner.getpage_send(req).await
+    }
+
+    async fn recv_get_page(&mut self) -> anyhow::Result<PagestreamGetPageResponse> {
+        self.inner.getpage_recv().await
+    }
+}
+
+/// A gRPC client using the raw, no-frills gRPC client.
+struct GrpcClient {
+    req_tx: tokio::sync::mpsc::Sender<proto::GetPageRequest>,
+    resp_rx: tonic::Streaming<proto::GetPageResponse>,
+}
+
+impl GrpcClient {
+    async fn new(connstring: String, ttid: TenantTimelineId) -> anyhow::Result<Self> {
+        let mut client = pageserver_page_api::proto::PageServiceClient::connect(connstring).await?;
+
+        // The channel has a buffer size of 1, since 0 is not allowed. It does not matter, since the
+        // benchmark will control the queue depth (i.e. in-flight requests) anyway, and requests are
+        // buffered by Tonic and the OS too.
+        let (req_tx, req_rx) = tokio::sync::mpsc::channel(1);
+        let req_stream = tokio_stream::wrappers::ReceiverStream::new(req_rx);
+        let mut req = tonic::Request::new(req_stream);
+        let metadata = req.metadata_mut();
+        metadata.insert("neon-tenant-id", ttid.tenant_id.to_string().try_into()?);
+        metadata.insert("neon-timeline-id", ttid.timeline_id.to_string().try_into()?);
+        metadata.insert("neon-shard-id", "0000".try_into()?);
+
+        let resp = client.get_pages(req).await?;
+        let resp_stream = resp.into_inner();
+
+        Ok(Self {
+            req_tx,
+            resp_rx: resp_stream,
+        })
+    }
+}
+
+#[async_trait]
+impl Client for GrpcClient {
+    async fn send_get_page(&mut self, req: PagestreamGetPageRequest) -> anyhow::Result<()> {
+        let req = proto::GetPageRequest {
+            request_id: 0,
+            request_class: proto::GetPageClass::Normal as i32,
+            read_lsn: Some(proto::ReadLsn {
+                request_lsn: req.hdr.request_lsn.0,
+                not_modified_since_lsn: req.hdr.not_modified_since.0,
+            }),
+            rel: Some(req.rel.into()),
+            block_number: vec![req.blkno],
+        };
+        self.req_tx.send(req).await?;
+        Ok(())
+    }
+
+    async fn recv_get_page(&mut self) -> anyhow::Result<PagestreamGetPageResponse> {
+        let resp = self.resp_rx.message().await?.unwrap();
+        anyhow::ensure!(
+            resp.status_code == proto::GetPageStatusCode::Ok as i32,
+            "unexpected status code: {}",
+            resp.status_code
+        );
+        Ok(PagestreamGetPageResponse {
+            page: resp.page_image[0].clone(),
+            req: PagestreamGetPageRequest::default(), // dummy
+        })
+    }
+}

pageserver/src/basebackup.rs

@@ -65,6 +65,30 @@ impl From<GetVectoredError> for BasebackupError {
     }
 }
 
+impl From<BasebackupError> for postgres_backend::QueryError {
+    fn from(err: BasebackupError) -> Self {
+        use postgres_backend::QueryError;
+        use pq_proto::framed::ConnectionError;
+        match err {
+            BasebackupError::Client(err, _) => QueryError::Disconnected(ConnectionError::Io(err)),
+            BasebackupError::Server(err) => QueryError::Other(err),
+            BasebackupError::Shutdown => QueryError::Shutdown,
+        }
+    }
+}
+
+impl From<BasebackupError> for tonic::Status {
+    fn from(err: BasebackupError) -> Self {
+        use tonic::Code;
+        let code = match &err {
+            BasebackupError::Client(_, _) => Code::Cancelled,
+            BasebackupError::Server(_) => Code::Internal,
+            BasebackupError::Shutdown => Code::Unavailable,
+        };
+        tonic::Status::new(code, err.to_string())
+    }
+}
+
 /// Create basebackup with non-rel data in it.
 /// Only include relational data if 'full_backup' is true.
 ///
@@ -144,7 +168,7 @@ where
         replica,
         ctx,
         io_concurrency: IoConcurrency::spawn_from_conf(
-            timeline.conf,
+            timeline.conf.get_vectored_concurrent_io,
             timeline
                 .gate
                 .enter()
@@ -248,7 +272,7 @@ where
     async fn flush(&mut self) -> Result<(), BasebackupError> {
         let nblocks = self.buf.len() / BLCKSZ as usize;
         let (kind, segno) = self.current_segment.take().unwrap();
-        let segname = format!("{}/{:>04X}", kind.to_str(), segno);
+        let segname = format!("{kind}/{segno:>04X}");
         let header = new_tar_header(&segname, self.buf.len() as u64)?;
         self.ar
             .append(&header, self.buf.as_slice())
@@ -343,7 +367,7 @@ 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 +402,7 @@ 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 +541,7 @@ 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 +601,7 @@ 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 +655,7 @@ 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 +804,25 @@ fn start_pageserver(
         } else {
             None
         },
+        basebackup_cache,
     );
 
+    // 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(
+            tenant_manager.clone(),
+            grpc_auth,
+            otel_guard.as_ref().map(|g| g.dispatch.clone()),
+            grpc_listener,
+        )?);
+    }
+
     // All started up! Now just sit and wait for shutdown signal.
     BACKGROUND_RUNTIME.block_on(async move {
         let signal_token = CancellationToken::new();
@@ -783,6 +841,7 @@ 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 +855,14 @@ 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;
@@ -51,13 +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,
-    ) -> impl Future<Output = Result<Option<ShardImportStatus>, RetryForeverError>> + Send;
+        generation: Generation,
+    ) -> impl Future<Output = Result<ShardImportStatus, RetryForeverError>> + Send;
 }
 
 impl StorageControllerUpcallClient {
@@ -102,6 +104,7 @@ impl StorageControllerUpcallClient {
         &self,
         url: &url::Url,
         request: R,
+        method: reqwest::Method,
     ) -> Result<T, RetryForeverError>
     where
         R: Serialize,
@@ -111,7 +114,7 @@ impl StorageControllerUpcallClient {
             || async {
                 let response = self
                     .http_client
-                    .post(url.clone())
+                    .request(method.clone(), url.clone())
                     .json(&request)
                     .send()
                     .await?;
@@ -220,7 +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(),
@@ -273,7 +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);
             }
@@ -292,6 +299,7 @@ impl StorageControllerUpcallApi for StorageControllerUpcallClient {
         &self,
         tenant_shard_id: TenantShardId,
         timeline_id: TimelineId,
+        generation: Generation,
         status: ShardImportStatus,
     ) -> Result<(), RetryForeverError> {
         let url = self
@@ -302,10 +310,12 @@ 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
@@ -313,33 +323,22 @@ impl StorageControllerUpcallApi for StorageControllerUpcallClient {
         &self,
         tenant_shard_id: TenantShardId,
         timeline_id: TimelineId,
-    ) -> Result<Option<ShardImportStatus>, RetryForeverError> {
+        generation: Generation,
+    ) -> Result<ShardImportStatus, RetryForeverError> {
         let url = self
             .base_url
-            .join(format!("timeline_import_status/{}/{}", tenant_shard_id, timeline_id).as_str())
+            .join("timeline_import_status")
             .expect("Failed to build path");
 
-        Ok(backoff::retry(
-            || async {
-                let response = self.http_client.get(url.clone()).send().await?;
+        let request = TimelineImportStatusRequest {
+            tenant_shard_id,
+            timeline_id,
+            generation,
+        };
 
-                if let Err(err) = response.error_for_status_ref() {
-                    if matches!(err.status(), Some(reqwest::StatusCode::NOT_FOUND)) {
-                        return Ok(None);
-                    } else {
-                        return Err(err);
-                    }
-                }
-                response.json::<ShardImportStatus>().await.map(Some)
-            },
-            |_| false,
-            3,
-            u32::MAX,
-            "storage controller upcall",
-            &self.cancel,
-        )
-        .await
-        .ok_or(RetryForeverError::ShuttingDown)?
-        .expect("We retry forever, this should never be reached"))
+        let response: ShardImportStatus = self
+            .retry_http_forever(&url, request, reqwest::Method::GET)
+            .await?;
+        Ok(response)
     }
 }

pageserver/src/deletion_queue.rs

@@ -793,6 +793,7 @@ mod test {
             &self,
             _tenant_shard_id: TenantShardId,
             _timeline_id: TimelineId,
+            _generation: Generation,
             _status: pageserver_api::models::ShardImportStatus,
         ) -> Result<(), RetryForeverError> {
             unimplemented!()
@@ -802,7 +803,8 @@ mod test {
             &self,
             _tenant_shard_id: TenantShardId,
             _timeline_id: TimelineId,
-        ) -> Result<Option<ShardImportStatus>, RetryForeverError> {
+            _generation: Generation,
+        ) -> Result<ShardImportStatus, RetryForeverError> {
             unimplemented!()
         }
     }

pageserver/src/disk_usage_eviction_task.rs

@@ -837,7 +837,30 @@ async fn collect_eviction_candidates(
                 continue;
             }
             let info = tl.get_local_layers_for_disk_usage_eviction().await;
-            debug!(tenant_id=%tl.tenant_shard_id.tenant_id, shard_id=%tl.tenant_shard_id.shard_slug(), timeline_id=%tl.timeline_id, "timeline resident layers count: {}", info.resident_layers.len());
+            debug!(
+                tenant_id=%tl.tenant_shard_id.tenant_id,
+                shard_id=%tl.tenant_shard_id.shard_slug(),
+                timeline_id=%tl.timeline_id,
+                "timeline resident layers count: {}", info.resident_layers.len()
+            );
+
+            tenant_candidates.extend(info.resident_layers.into_iter());
+            max_layer_size = max_layer_size.max(info.max_layer_size.unwrap_or(0));
+
+            if cancel.is_cancelled() {
+                return Ok(EvictionCandidates::Cancelled);
+            }
+        }
+
+        // Also consider layers of timelines being imported for eviction
+        for tl in tenant.list_importing_timelines() {
+            let info = tl.timeline.get_local_layers_for_disk_usage_eviction().await;
+            debug!(
+                tenant_id=%tl.timeline.tenant_shard_id.tenant_id,
+                shard_id=%tl.timeline.tenant_shard_id.shard_slug(),
+                timeline_id=%tl.timeline.timeline_id,
+                "timeline resident layers count: {}", info.resident_layers.len()
+            );
 
             tenant_candidates.extend(info.resident_layers.into_iter());
             max_layer_size = max_layer_size.max(info.max_layer_size.unwrap_or(0));

pageserver/src/feature_resolver.rs

@@ -0,0 +1,104 @@
+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.
+    ///
+    /// Error handling: the caller should inspect the error and decide the behavior when a feature flag
+    /// cannot be evaluated (i.e., default to false if it cannot be resolved). The error should *not* be
+    /// propagated beyond where the feature flag gets resolved.
+    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(),
+            ))
+        }
+    }
+
+    /// Evaluate a boolean feature flag. Currently, we do not support any properties.
+    ///
+    /// Returns `Ok(())` if the flag is evaluated to true, otherwise returns an error.
+    ///
+    /// Error handling: the caller should inspect the error and decide the behavior when a feature flag
+    /// cannot be evaluated (i.e., default to false if it cannot be resolved). The error should *not* be
+    /// propagated beyond where the feature flag gets resolved.
+    pub fn evaluate_boolean(
+        &self,
+        flag_key: &str,
+        tenant_id: TenantId,
+    ) -> Result<(), PostHogEvaluationError> {
+        if let Some(inner) = &self.inner {
+            inner.feature_store().evaluate_boolean(
+                flag_key,
+                &tenant_id.to_string(),
+                &HashMap::new(),
+            )
+        } else {
+            Err(PostHogEvaluationError::NotAvailable(
+                "PostHog integration is not enabled".to_string(),
+            ))
+        }
+    }
+
+    pub fn is_feature_flag_boolean(&self, flag_key: &str) -> Result<bool, PostHogEvaluationError> {
+        if let Some(inner) = &self.inner {
+            inner.feature_store().is_feature_flag_boolean(flag_key)
+        } else {
+            Err(PostHogEvaluationError::NotAvailable(
+                "PostHog integration is not enabled".to_string(),
+            ))
+        }
+    }
+}

pageserver/src/http/openapi_spec.yml

@@ -353,6 +353,33 @@ paths:
         "200":
           description: OK
 
+  /v1/tenant/{tenant_shard_id}/timeline/{timeline_id}/mark_invisible:
+    parameters:
+      - name: tenant_shard_id
+        in: path
+        required: true
+        schema:
+          type: string
+      - name: timeline_id
+        in: path
+        required: true
+        schema:
+          type: string
+          format: hex
+    put:
+      requestBody:
+        content:
+          application/json:
+            schema:
+              type: object
+              properties:
+                is_visible:
+                  type: boolean
+                  default: false
+      responses:
+        "200":
+          description: OK
+
   /v1/tenant/{tenant_shard_id}/location_config:
     parameters:
       - name: tenant_shard_id
@@ -626,6 +653,8 @@ paths:
                   format: hex
                 pg_version:
                   type: integer
+                read_only:
+                  type: boolean
                 existing_initdb_timeline_id:
                   type: string
                   format: hex

pageserver/src/http/routes.rs

@@ -370,6 +370,18 @@ impl From<crate::tenant::secondary::SecondaryTenantError> for ApiError {
     }
 }
 
+impl From<crate::tenant::FinalizeTimelineImportError> for ApiError {
+    fn from(err: crate::tenant::FinalizeTimelineImportError) -> ApiError {
+        use crate::tenant::FinalizeTimelineImportError::*;
+        match err {
+            ImportTaskStillRunning => {
+                ApiError::ResourceUnavailable("Import task still running".into())
+            }
+            ShuttingDown => ApiError::ShuttingDown,
+        }
+    }
+}
+
 // Helper function to construct a TimelineInfo struct for a timeline
 async fn build_timeline_info(
     timeline: &Arc<Timeline>,
@@ -449,7 +461,7 @@ async fn build_timeline_info_common(
     // Internally we distinguish between the planned GC cutoff (PITR point) and the "applied" GC cutoff (where we
     // actually trimmed data to), which can pass each other when PITR is changed.
     let min_readable_lsn = std::cmp::max(
-        timeline.get_gc_cutoff_lsn(),
+        timeline.get_gc_cutoff_lsn().unwrap_or_default(),
         *timeline.get_applied_gc_cutoff_lsn(),
     );
 
@@ -572,6 +584,7 @@ async fn timeline_create_handler(
         TimelineCreateRequestMode::Branch {
             ancestor_timeline_id,
             ancestor_start_lsn,
+            read_only: _,
             pg_version: _,
         } => tenant::CreateTimelineParams::Branch(tenant::CreateTimelineParamsBranch {
             new_timeline_id,
@@ -3199,7 +3212,7 @@ async fn list_aux_files(
             .await?;
 
     let io_concurrency = IoConcurrency::spawn_from_conf(
-        state.conf,
+        state.conf.get_vectored_concurrent_io,
         timeline.gate.enter().map_err(|_| ApiError::Cancelled)?,
     );
 
@@ -3532,10 +3545,7 @@ async fn activate_post_import_handler(
 
         tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
 
-        tenant
-            .finalize_importing_timeline(timeline_id)
-            .await
-            .map_err(ApiError::InternalServerError)?;
+        tenant.finalize_importing_timeline(timeline_id).await?;
 
         match tenant.get_timeline(timeline_id, false) {
             Ok(_timeline) => {
@@ -3653,6 +3663,46 @@ async fn read_tar_eof(mut reader: (impl tokio::io::AsyncRead + Unpin)) -> anyhow
     Ok(())
 }
 
+async fn tenant_evaluate_feature_flag(
+    request: Request<Body>,
+    _cancel: CancellationToken,
+) -> Result<Response<Body>, ApiError> {
+    let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
+    check_permission(&request, Some(tenant_shard_id.tenant_id))?;
+
+    let flag: String = must_parse_query_param(&request, "flag")?;
+    let as_type: String = must_parse_query_param(&request, "as")?;
+
+    let state = get_state(&request);
+
+    async {
+        let tenant = state
+            .tenant_manager
+            .get_attached_tenant_shard(tenant_shard_id)?;
+        if as_type == "boolean" {
+            let result = tenant.feature_resolver.evaluate_boolean(&flag, tenant_shard_id.tenant_id);
+            let result = result.map(|_| true).map_err(|e| e.to_string());
+            json_response(StatusCode::OK, result)
+        } else if as_type == "multivariate" {
+            let result = tenant.feature_resolver.evaluate_multivariate(&flag, tenant_shard_id.tenant_id).map_err(|e| e.to_string());
+            json_response(StatusCode::OK, result)
+        } else {
+            // Auto infer the type of the feature flag.
+            let is_boolean = tenant.feature_resolver.is_feature_flag_boolean(&flag).map_err(|e| ApiError::InternalServerError(anyhow::anyhow!("{e}")))?;
+            if is_boolean {
+                let result = tenant.feature_resolver.evaluate_boolean(&flag, tenant_shard_id.tenant_id);
+                let result = result.map(|_| true).map_err(|e| e.to_string());
+                json_response(StatusCode::OK, result)
+            } else {
+                let result = tenant.feature_resolver.evaluate_multivariate(&flag, tenant_shard_id.tenant_id).map_err(|e| e.to_string());
+                json_response(StatusCode::OK, result)
+            }
+        }
+    }
+    .instrument(info_span!("tenant_evaluate_feature_flag", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug()))
+    .await
+}
+
 /// Common functionality of all the HTTP API handlers.
 ///
 /// - Adds a tracing span to each request (by `request_span`)
@@ -4029,5 +4079,8 @@ pub fn make_router(
             "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/activate_post_import",
             |r| api_handler(r, activate_post_import_handler),
         )
+        .get("/v1/tenant/:tenant_shard_id/feature_flag", |r| {
+            api_handler(r, tenant_evaluate_feature_flag)
+        })
         .any(handler_404))
 }

pageserver/src/lib.rs

@@ -3,12 +3,14 @@
 
 mod auth;
 pub mod basebackup;
+pub mod basebackup_cache;
 pub mod config;
 pub mod consumption_metrics;
 pub mod context;
 pub mod controller_upcall_client;
 pub mod deletion_queue;
 pub mod disk_usage_eviction_task;
+pub mod feature_resolver;
 pub mod http;
 pub mod import_datadir;
 pub mod l0_flush;
@@ -83,6 +85,7 @@ pub async fn shutdown_pageserver(
     http_listener: HttpEndpointListener,
     https_listener: Option<HttpsEndpointListener>,
     page_service: page_service::Listener,
+    grpc_task: Option<CancellableTask>,
     consumption_metrics_worker: ConsumptionMetricsTasks,
     disk_usage_eviction_task: Option<DiskUsageEvictionTask>,
     tenant_manager: &TenantManager,
@@ -176,6 +179,16 @@ pub async fn shutdown_pageserver(
     )
     .await;
 
+    // Shut down the gRPC server task, including request handlers.
+    if let Some(grpc_task) = grpc_task {
+        timed(
+            grpc_task.shutdown(),
+            "shutdown gRPC PageRequestHandler",
+            Duration::from_secs(3),
+        )
+        .await;
+    }
+
     // Shut down all the tenants. This flushes everything to disk and kills
     // the checkpoint and GC tasks.
     timed(

pageserver/src/metrics.rs

@@ -843,23 +843,50 @@ pub(crate) static COMPRESSION_IMAGE_OUTPUT_BYTES: Lazy<IntCounter> = Lazy::new(|
     .expect("failed to define a metric")
 });
 
-pub(crate) static RELSIZE_CACHE_ENTRIES: Lazy<UIntGauge> = Lazy::new(|| {
+pub(crate) static RELSIZE_LATEST_CACHE_ENTRIES: Lazy<UIntGauge> = Lazy::new(|| {
     register_uint_gauge!(
-        "pageserver_relsize_cache_entries",
-        "Number of entries in the relation size cache",
+        "pageserver_relsize_latest_cache_entries",
+        "Number of entries in the latest relation size cache",
     )
     .expect("failed to define a metric")
 });
 
-pub(crate) static RELSIZE_CACHE_HITS: Lazy<IntCounter> = Lazy::new(|| {
-    register_int_counter!("pageserver_relsize_cache_hits", "Relation size cache hits",)
-        .expect("failed to define a metric")
+pub(crate) static RELSIZE_LATEST_CACHE_HITS: Lazy<IntCounter> = Lazy::new(|| {
+    register_int_counter!(
+        "pageserver_relsize_latest_cache_hits",
+        "Latest relation size cache hits",
+    )
+    .expect("failed to define a metric")
 });
 
-pub(crate) static RELSIZE_CACHE_MISSES: Lazy<IntCounter> = Lazy::new(|| {
+pub(crate) static RELSIZE_LATEST_CACHE_MISSES: Lazy<IntCounter> = Lazy::new(|| {
     register_int_counter!(
-        "pageserver_relsize_cache_misses",
-        "Relation size cache misses",
+        "pageserver_relsize_latest_cache_misses",
+        "Relation size latest cache misses",
+    )
+    .expect("failed to define a metric")
+});
+
+pub(crate) static RELSIZE_SNAPSHOT_CACHE_ENTRIES: Lazy<UIntGauge> = Lazy::new(|| {
+    register_uint_gauge!(
+        "pageserver_relsize_snapshot_cache_entries",
+        "Number of entries in the pitr relation size cache",
+    )
+    .expect("failed to define a metric")
+});
+
+pub(crate) static RELSIZE_SNAPSHOT_CACHE_HITS: Lazy<IntCounter> = Lazy::new(|| {
+    register_int_counter!(
+        "pageserver_relsize_snapshot_cache_hits",
+        "Pitr relation size cache hits",
+    )
+    .expect("failed to define a metric")
+});
+
+pub(crate) static RELSIZE_SNAPSHOT_CACHE_MISSES: Lazy<IntCounter> = Lazy::new(|| {
+    register_int_counter!(
+        "pageserver_relsize_snapshot_cache_misses",
+        "Relation size snapshot cache misses",
     )
     .expect("failed to define a metric")
 });
@@ -1039,6 +1066,15 @@ pub(crate) static TENANT_SYNTHETIC_SIZE_METRIC: Lazy<UIntGaugeVec> = Lazy::new(|
     .expect("Failed to register pageserver_tenant_synthetic_cached_size_bytes metric")
 });
 
+pub(crate) static TENANT_OFFLOADED_TIMELINES: Lazy<UIntGaugeVec> = Lazy::new(|| {
+    register_uint_gauge_vec!(
+        "pageserver_tenant_offloaded_timelines",
+        "Number of offloaded timelines of a tenant",
+        &["tenant_id", "shard_id"]
+    )
+    .expect("Failed to register pageserver_tenant_offloaded_timelines metric")
+});
+
 pub(crate) static EVICTION_ITERATION_DURATION: Lazy<HistogramVec> = Lazy::new(|| {
     register_histogram_vec!(
         "pageserver_eviction_iteration_duration_seconds_global",
@@ -1276,11 +1312,44 @@ impl EvictionsWithLowResidenceDuration {
 //
 // Roughly logarithmic scale.
 const STORAGE_IO_TIME_BUCKETS: &[f64] = &[
-    0.000030, // 30 usec
-    0.001000, // 1000 usec
-    0.030,    // 30 ms
-    1.000,    // 1000 ms
-    30.000,   // 30000 ms
+    0.00005,  // 50us
+    0.00006,  // 60us
+    0.00007,  // 70us
+    0.00008,  // 80us
+    0.00009,  // 90us
+    0.0001,   // 100us
+    0.000110, // 110us
+    0.000120, // 120us
+    0.000130, // 130us
+    0.000140, // 140us
+    0.000150, // 150us
+    0.000160, // 160us
+    0.000170, // 170us
+    0.000180, // 180us
+    0.000190, // 190us
+    0.000200, // 200us
+    0.000210, // 210us
+    0.000220, // 220us
+    0.000230, // 230us
+    0.000240, // 240us
+    0.000250, // 250us
+    0.000300, // 300us
+    0.000350, // 350us
+    0.000400, // 400us
+    0.000450, // 450us
+    0.000500, // 500us
+    0.000600, // 600us
+    0.000700, // 700us
+    0.000800, // 800us
+    0.000900, // 900us
+    0.001000, // 1ms
+    0.002000, // 2ms
+    0.003000, // 3ms
+    0.004000, // 4ms
+    0.005000, // 5ms
+    0.01000,  // 10ms
+    0.02000,  // 20ms
+    0.05000,  // 50ms
 ];
 
 /// VirtualFile fs operation variants.
@@ -2198,8 +2267,10 @@ impl BasebackupQueryTimeOngoingRecording<'_> {
         // If you want to change categorize of a specific error, also change it in `log_query_error`.
         let metric = match res {
             Ok(_) => &self.parent.ok,
-            Err(QueryError::Shutdown) => {
-                // Do not observe ok/err for shutdown
+            Err(QueryError::Shutdown) | Err(QueryError::Reconnect) => {
+                // Do not observe ok/err for shutdown/reconnect.
+                // Reconnect error might be raised when the operation is waiting for LSN and the tenant shutdown interrupts
+                // the operation. A reconnect error will be issued and the client will retry.
                 return;
             }
             Err(QueryError::Disconnected(ConnectionError::Io(io_error)))
@@ -3524,11 +3595,14 @@ impl TimelineMetrics {
 }
 
 pub(crate) fn remove_tenant_metrics(tenant_shard_id: &TenantShardId) {
+    let tid = tenant_shard_id.tenant_id.to_string();
+    let shard_id = tenant_shard_id.shard_slug().to_string();
+
     // Only shard zero deals in synthetic sizes
     if tenant_shard_id.is_shard_zero() {
-        let tid = tenant_shard_id.tenant_id.to_string();
         let _ = TENANT_SYNTHETIC_SIZE_METRIC.remove_label_values(&[&tid]);
     }
+    let _ = TENANT_OFFLOADED_TIMELINES.remove_label_values(&[&tid, &shard_id]);
 
     tenant_throttling::remove_tenant_metrics(tenant_shard_id);
 
@@ -4320,6 +4394,42 @@ pub(crate) fn set_tokio_runtime_setup(setup: &str, num_threads: NonZeroUsize) {
         .set(u64::try_from(num_threads.get()).unwrap());
 }
 
+pub(crate) static BASEBACKUP_CACHE_READ: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
+        "pageserver_basebackup_cache_read_total",
+        "Number of read accesses to the basebackup cache grouped by hit/miss/error",
+        &["result"]
+    )
+    .expect("failed to define a metric")
+});
+
+pub(crate) static BASEBACKUP_CACHE_PREPARE: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
+        "pageserver_basebackup_cache_prepare_total",
+        "Number of prepare requests processed by the basebackup cache grouped by ok/skip/error",
+        &["result"]
+    )
+    .expect("failed to define a metric")
+});
+
+pub(crate) static BASEBACKUP_CACHE_ENTRIES: Lazy<IntGauge> = Lazy::new(|| {
+    register_int_gauge!(
+        "pageserver_basebackup_cache_entries_total",
+        "Number of entries in the basebackup cache"
+    )
+    .expect("failed to define a metric")
+});
+
+// FIXME: Support basebackup cache size metrics.
+#[allow(dead_code)]
+pub(crate) static BASEBACKUP_CACHE_SIZE: Lazy<IntGauge> = Lazy::new(|| {
+    register_int_gauge!(
+        "pageserver_basebackup_cache_size_bytes",
+        "Total size of all basebackup cache entries on disk in bytes"
+    )
+    .expect("failed to define a metric")
+});
+
 static PAGESERVER_CONFIG_IGNORED_ITEMS: Lazy<UIntGaugeVec> = Lazy::new(|| {
     register_uint_gauge_vec!(
         "pageserver_config_ignored_items",