try: log cancelations within timeline_detail_handler

this in effort to hunt down https://github.com/neondatabase/cloud/issues/4998

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

@@ -0,0 +1,184 @@
+name: 'Create Allure report'
+description: 'Generate Allure report from uploaded by actions/allure-report-store tests results'
+
+outputs:
+  report-url:
+    description: 'Allure report URL'
+    value: ${{ steps.generate-report.outputs.report-url }}
+  report-json-url:
+    description: 'Allure report JSON URL'
+    value: ${{ steps.generate-report.outputs.report-json-url }}
+
+runs:
+  using: "composite"
+
+  steps:
+    # We're using some of env variables quite offen, so let's set them once.
+    #
+    # It would be nice to have them set in common runs.env[0] section, but it doesn't work[1]
+    #
+    # - [0] https://docs.github.com/en/actions/creating-actions/metadata-syntax-for-github-actions#runsenv
+    # - [1] https://github.com/neondatabase/neon/pull/3907#discussion_r1154703456
+    #
+    - name: Set variables
+      shell: bash -euxo pipefail {0}
+      run: |
+        PR_NUMBER=$(jq --raw-output .pull_request.number "$GITHUB_EVENT_PATH" || true)
+        if [ "${PR_NUMBER}" != "null" ]; then
+          BRANCH_OR_PR=pr-${PR_NUMBER}
+        elif [ "${GITHUB_REF_NAME}" = "main" ] || [ "${GITHUB_REF_NAME}" = "release" ]; then
+          # Shortcut for special branches
+          BRANCH_OR_PR=${GITHUB_REF_NAME}
+        else
+          BRANCH_OR_PR=branch-$(printf "${GITHUB_REF_NAME}" | tr -c "[:alnum:]._-" "-")
+        fi
+
+        LOCK_FILE=reports/${BRANCH_OR_PR}/lock.txt
+
+        WORKDIR=/tmp/${BRANCH_OR_PR}-$(date +%s)
+        mkdir -p ${WORKDIR}
+
+        echo "BRANCH_OR_PR=${BRANCH_OR_PR}" >> $GITHUB_ENV
+        echo "LOCK_FILE=${LOCK_FILE}"       >> $GITHUB_ENV
+        echo "WORKDIR=${WORKDIR}"           >> $GITHUB_ENV
+        echo "BUCKET=${BUCKET}"             >> $GITHUB_ENV
+      env:
+        BUCKET: neon-github-public-dev
+
+    # TODO: We can replace with a special docker image with Java and Allure pre-installed
+    - uses: actions/setup-java@v3
+      with:
+        distribution: 'temurin'
+        java-version: '17'
+
+    - name: Install Allure
+      shell: bash -euxo pipefail {0}
+      run: |
+        if ! which allure; then
+          ALLURE_ZIP=allure-${ALLURE_VERSION}.zip
+          wget -q https://github.com/allure-framework/allure2/releases/download/${ALLURE_VERSION}/${ALLURE_ZIP}
+          echo "${ALLURE_ZIP_MD5}  ${ALLURE_ZIP}" | md5sum -c
+          unzip -q ${ALLURE_ZIP}
+          echo "$(pwd)/allure-${ALLURE_VERSION}/bin" >> $GITHUB_PATH
+          rm -f ${ALLURE_ZIP}
+        fi
+      env:
+        ALLURE_VERSION: 2.22.0
+        ALLURE_ZIP_MD5: d5c9f0989b896482536956340a7d5ec9
+
+    # Potentially we could have several running build for the same key (for example, for the main branch), so we use improvised lock for this
+    - name: Acquire lock
+      shell: bash -euxo pipefail {0}
+      run: |
+        LOCK_TIMEOUT=300 # seconds
+
+        LOCK_CONTENT="${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}"
+        echo ${LOCK_CONTENT} > ${WORKDIR}/lock.txt
+
+        # Do it up to 5 times to avoid race condition
+        for _ in $(seq 1 5); do
+          for i in $(seq 1 ${LOCK_TIMEOUT}); do
+            LOCK_ACQUIRED=$(aws s3api head-object --bucket neon-github-public-dev --key ${LOCK_FILE} | jq --raw-output '.LastModified' || true)
+            # `date --date="..."` is supported only by gnu date (i.e. it doesn't work on BSD/macOS)
+            if [ -z "${LOCK_ACQUIRED}" ] || [ "$(( $(date +%s) - $(date --date="${LOCK_ACQUIRED}" +%s) ))" -gt "${LOCK_TIMEOUT}" ]; then
+              break
+            fi
+            sleep 1
+          done
+
+          aws s3 mv --only-show-errors ${WORKDIR}/lock.txt "s3://${BUCKET}/${LOCK_FILE}"
+
+          # Double-check that exactly THIS run has acquired the lock
+          aws s3 cp --only-show-errors "s3://${BUCKET}/${LOCK_FILE}" ./lock.txt
+          if [ "$(cat lock.txt)" = "${LOCK_CONTENT}" ]; then
+            break
+          fi
+        done
+
+    - name: Generate and publish final Allure report
+      id: generate-report
+      shell: bash -euxo pipefail {0}
+      run: |
+        REPORT_PREFIX=reports/${BRANCH_OR_PR}
+        RAW_PREFIX=reports-raw/${BRANCH_OR_PR}/${GITHUB_RUN_ID}
+
+        # Get previously uploaded data for this run
+        ZSTD_NBTHREADS=0
+
+        S3_FILEPATHS=$(aws s3api list-objects-v2 --bucket ${BUCKET} --prefix ${RAW_PREFIX}/ | jq --raw-output  '.Contents[].Key')
+        if [ -z "$S3_FILEPATHS" ]; then
+          # There's no previously uploaded data for this $GITHUB_RUN_ID
+          exit 0
+        fi
+        for S3_FILEPATH in ${S3_FILEPATHS}; do
+          time aws s3 cp --only-show-errors "s3://${BUCKET}/${S3_FILEPATH}" "${WORKDIR}"
+
+          archive=${WORKDIR}/$(basename $S3_FILEPATH)
+          mkdir -p ${archive%.tar.zst}
+          time tar -xf ${archive} -C ${archive%.tar.zst}
+          rm -f ${archive}
+        done
+
+        # Get history trend
+        time aws s3 cp --recursive --only-show-errors "s3://${BUCKET}/${REPORT_PREFIX}/latest/history" "${WORKDIR}/latest/history" || true
+
+        # Generate report
+        time allure generate --clean --output ${WORKDIR}/report ${WORKDIR}/*
+
+        # Replace a logo link with a redirect to the latest version of the report
+        sed -i 's|<a href="." class=|<a href="https://'${BUCKET}'.s3.amazonaws.com/'${REPORT_PREFIX}'/latest/index.html?nocache='"'+Date.now()+'"'" class=|g' ${WORKDIR}/report/app.js
+
+        # Upload a history and the final report (in this particular order to not to have duplicated history in 2 places)
+        time aws s3 mv --recursive --only-show-errors "${WORKDIR}/report/history" "s3://${BUCKET}/${REPORT_PREFIX}/latest/history"
+        time aws s3 mv --recursive --only-show-errors "${WORKDIR}/report" "s3://${BUCKET}/${REPORT_PREFIX}/${GITHUB_RUN_ID}"
+
+        REPORT_URL=https://${BUCKET}.s3.amazonaws.com/${REPORT_PREFIX}/${GITHUB_RUN_ID}/index.html
+
+        # Generate redirect
+        cat <<EOF > ${WORKDIR}/index.html
+          <!DOCTYPE html>
+
+          <meta charset="utf-8">
+          <title>Redirecting to ${REPORT_URL}</title>
+          <meta http-equiv="refresh" content="0; URL=${REPORT_URL}">
+        EOF
+        time aws s3 cp --only-show-errors ${WORKDIR}/index.html "s3://${BUCKET}/${REPORT_PREFIX}/latest/index.html"
+
+        echo "report-url=${REPORT_URL}"                                   >> $GITHUB_OUTPUT
+        echo "report-json-url=${REPORT_URL%/index.html}/data/suites.json" >> $GITHUB_OUTPUT
+
+    - name: Release lock
+      if: always()
+      shell: bash -euxo pipefail {0}
+      run: |
+        aws s3 cp --only-show-errors "s3://${BUCKET}/${LOCK_FILE}" ./lock.txt || exit 0
+
+        if [ "$(cat lock.txt)" = "${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}" ]; then
+          aws s3 rm "s3://${BUCKET}/${LOCK_FILE}"
+        fi
+
+    - name: Cleanup
+      if: always()
+      shell: bash -euxo pipefail {0}
+      run: |
+        if [ -d "${WORKDIR}" ]; then
+          rm -rf ${WORKDIR}
+        fi
+
+    - uses: actions/github-script@v6
+      if: always()
+      env:
+        REPORT_URL: ${{ steps.generate-report.outputs.report-url }}
+        COMMIT_SHA: ${{ github.event.pull_request.head.sha || github.sha }}
+      with:
+        script: |
+          const { REPORT_URL, COMMIT_SHA } = process.env
+
+          await github.rest.repos.createCommitStatus({
+            owner: context.repo.owner,
+            repo: context.repo.repo,
+            sha: `${COMMIT_SHA}`,
+            state: 'success',
+            target_url: `${REPORT_URL}`,
+            context: 'Allure report',
+          })

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

@@ -0,0 +1,72 @@
+name: 'Store Allure results'
+description: 'Upload test results to be used by actions/allure-report-generate'
+
+inputs:
+  report-dir:
+    description: 'directory with test results generated by tests'
+    required: true
+  unique-key:
+    description: 'string to distinguish different results in the same run'
+    required: true
+
+runs:
+  using: "composite"
+
+  steps:
+    - name: Set variables
+      shell: bash -euxo pipefail {0}
+      run: |
+        PR_NUMBER=$(jq --raw-output .pull_request.number "$GITHUB_EVENT_PATH" || true)
+        if [ "${PR_NUMBER}" != "null" ]; then
+          BRANCH_OR_PR=pr-${PR_NUMBER}
+        elif [ "${GITHUB_REF_NAME}" = "main" ] || [ "${GITHUB_REF_NAME}" = "release" ]; then
+          # Shortcut for special branches
+          BRANCH_OR_PR=${GITHUB_REF_NAME}
+        else
+          BRANCH_OR_PR=branch-$(printf "${GITHUB_REF_NAME}" | tr -c "[:alnum:]._-" "-")
+        fi
+
+        echo "BRANCH_OR_PR=${BRANCH_OR_PR}" >> $GITHUB_ENV
+        echo "REPORT_DIR=${REPORT_DIR}"     >> $GITHUB_ENV
+      env:
+        REPORT_DIR: ${{ inputs.report-dir }}
+
+    - name: Upload test results
+      shell: bash -euxo pipefail {0}
+      run: |
+        REPORT_PREFIX=reports/${BRANCH_OR_PR}
+        RAW_PREFIX=reports-raw/${BRANCH_OR_PR}/${GITHUB_RUN_ID}
+
+        # Add metadata
+        cat <<EOF > ${REPORT_DIR}/executor.json
+          {
+            "name": "GitHub Actions",
+            "type": "github",
+            "url": "https://${BUCKET}.s3.amazonaws.com/${REPORT_PREFIX}/latest/index.html",
+            "buildOrder": ${GITHUB_RUN_ID},
+            "buildName": "GitHub Actions Run #${GITHUB_RUN_NUMBER}/${GITHUB_RUN_ATTEMPT}",
+            "buildUrl": "${GITHUB_SERVER_URL}/${GITHUB_REPOSITORY}/actions/runs/${GITHUB_RUN_ID}/attempts/${GITHUB_RUN_ATTEMPT}",
+            "reportUrl": "https://${BUCKET}.s3.amazonaws.com/${REPORT_PREFIX}/${GITHUB_RUN_ID}/index.html",
+            "reportName": "Allure Report"
+          }
+        EOF
+
+        cat <<EOF > ${REPORT_DIR}/environment.properties
+          COMMIT_SHA=${COMMIT_SHA}
+        EOF
+
+        ARCHIVE="${UNIQUE_KEY}-${GITHUB_RUN_ATTEMPT}-$(date +%s).tar.zst"
+        ZSTD_NBTHREADS=0
+
+        time tar -C ${REPORT_DIR} -cf ${ARCHIVE} --zstd .
+        time aws s3 mv --only-show-errors ${ARCHIVE} "s3://${BUCKET}/${RAW_PREFIX}/${ARCHIVE}"
+      env:
+        UNIQUE_KEY: ${{ inputs.unique-key }}
+        COMMIT_SHA: ${{ github.event.pull_request.head.sha || github.sha }}
+        BUCKET: neon-github-public-dev
+
+    - name: Cleanup
+      if: always()
+      shell: bash -euxo pipefail {0}
+      run: |
+        rm -rf ${REPORT_DIR}

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

@@ -1,254 +0,0 @@
-name: 'Create Allure report'
-description: 'Create and publish Allure report'
-
-inputs:
-  action:
-    desctiption: 'generate or store'
-    required: true
-  build_type:
-    description: '`build_type` from run-python-test-set action'
-    required: true
-  test_selection:
-    description: '`test_selector` from run-python-test-set action'
-    required: false
-outputs:
-  report-url:
-    description: 'Allure report URL'
-    value: ${{ steps.generate-report.outputs.report-url }}
-  report-json-url:
-    description: 'Allure report JSON URL'
-    value: ${{ steps.generate-report.outputs.report-json-url }}
-
-runs:
-  using: "composite"
-
-  steps:
-    # We're using some of env variables quite offen, so let's set them once.
-    #
-    # It would be nice to have them set in common runs.env[0] section, but it doesn't work[1]
-    #
-    # - [0] https://docs.github.com/en/actions/creating-actions/metadata-syntax-for-github-actions#runsenv
-    # - [1] https://github.com/neondatabase/neon/pull/3907#discussion_r1154703456
-    #
-    - name: Set common environment variables
-      shell: bash -euxo pipefail {0}
-      run: |
-        echo "BUILD_TYPE=${BUILD_TYPE}"   >> $GITHUB_ENV
-        echo "BUCKET=${BUCKET}"           >> $GITHUB_ENV
-        echo "TEST_OUTPUT=${TEST_OUTPUT}" >> $GITHUB_ENV
-      env:
-        BUILD_TYPE: ${{ inputs.build_type }}
-        BUCKET: neon-github-public-dev
-        TEST_OUTPUT: /tmp/test_output
-
-    - name: Validate input parameters
-      shell: bash -euxo pipefail {0}
-      run: |
-        if [ "${{ inputs.action }}" != "store" ] && [ "${{ inputs.action }}" != "generate" ]; then
-          echo >&2 "Unknown inputs.action type '${{ inputs.action }}'; allowed 'generate' or 'store' only"
-          exit 1
-        fi
-
-        if [ -z "${{ inputs.test_selection }}" ] && [ "${{ inputs.action }}" == "store" ]; then
-          echo >&2 "inputs.test_selection must be set for 'store' action"
-          exit 2
-        fi
-
-    - name: Calculate variables
-      id: calculate-vars
-      shell: bash -euxo pipefail {0}
-      run: |
-        # TODO: for manually triggered workflows (via workflow_dispatch) we need to have a separate key
-
-        pr_number=$(jq --raw-output .pull_request.number "$GITHUB_EVENT_PATH" || true)
-        if [ "${pr_number}" != "null" ]; then
-          key=pr-${pr_number}
-        elif [ "${GITHUB_REF_NAME}" = "main" ]; then
-          # Shortcut for a special branch
-          key=main
-        elif [ "${GITHUB_REF_NAME}" = "release" ]; then
-          # Shortcut for a special branch
-          key=release
-        else
-          key=branch-$(printf "${GITHUB_REF_NAME}" | tr -c "[:alnum:]._-" "-")
-        fi
-        echo "KEY=${key}" >> $GITHUB_OUTPUT
-
-        # Sanitize test selection to remove `/` and any other special characters
-        # Use printf instead of echo to avoid having `\n` at the end of the string
-        test_selection=$(printf "${{ inputs.test_selection }}" | tr -c "[:alnum:]._-" "-" )
-        echo "TEST_SELECTION=${test_selection}" >> $GITHUB_OUTPUT
-
-    - uses: actions/setup-java@v3
-      if: ${{ inputs.action == 'generate' }}
-      with:
-        distribution: 'temurin'
-        java-version: '17'
-
-    - name: Install Allure
-      if: ${{ inputs.action == 'generate' }}
-      shell: bash -euxo pipefail {0}
-      run: |
-        if ! which allure; then
-          ALLURE_ZIP=allure-${ALLURE_VERSION}.zip
-          wget -q https://github.com/allure-framework/allure2/releases/download/${ALLURE_VERSION}/${ALLURE_ZIP}
-          echo "${ALLURE_ZIP_MD5}  ${ALLURE_ZIP}" | md5sum -c
-          unzip -q ${ALLURE_ZIP}
-          echo "$(pwd)/allure-${ALLURE_VERSION}/bin" >> $GITHUB_PATH
-          rm -f ${ALLURE_ZIP}
-        fi
-      env:
-        ALLURE_VERSION: 2.21.0
-        ALLURE_ZIP_MD5: c8db4dd8e2a7882583d569ed2c82879c
-
-    - name: Upload Allure results
-      if: ${{ inputs.action == 'store' }}
-      env:
-        REPORT_PREFIX: reports/${{ steps.calculate-vars.outputs.KEY }}/${{ inputs.build_type }}
-        RAW_PREFIX: reports-raw/${{ steps.calculate-vars.outputs.KEY }}/${{ inputs.build_type }}
-        TEST_SELECTION: ${{ steps.calculate-vars.outputs.TEST_SELECTION }}
-      shell: bash -euxo pipefail {0}
-      run: |
-        # Add metadata
-        cat <<EOF > $TEST_OUTPUT/allure/results/executor.json
-          {
-            "name": "GitHub Actions",
-            "type": "github",
-            "url": "https://${BUCKET}.s3.amazonaws.com/${REPORT_PREFIX}/latest/index.html",
-            "buildOrder": ${GITHUB_RUN_ID},
-            "buildName": "GitHub Actions Run #${{ github.run_number }}/${GITHUB_RUN_ATTEMPT}",
-            "buildUrl": "${GITHUB_SERVER_URL}/${GITHUB_REPOSITORY}/actions/runs/${GITHUB_RUN_ID}/attempts/${GITHUB_RUN_ATTEMPT}",
-            "reportUrl": "https://${BUCKET}.s3.amazonaws.com/${REPORT_PREFIX}/${GITHUB_RUN_ID}/index.html",
-            "reportName": "Allure Report"
-          }
-        EOF
-        cat <<EOF > $TEST_OUTPUT/allure/results/environment.properties
-          TEST_SELECTION=${{ inputs.test_selection }}
-          BUILD_TYPE=${BUILD_TYPE}
-        EOF
-
-        ARCHIVE="${GITHUB_RUN_ID}-${TEST_SELECTION}-${GITHUB_RUN_ATTEMPT}-$(date +%s).tar.zst"
-        ZSTD_NBTHREADS=0
-
-        tar -C ${TEST_OUTPUT}/allure/results -cf ${ARCHIVE} --zstd .
-        aws s3 mv --only-show-errors ${ARCHIVE} "s3://${BUCKET}/${RAW_PREFIX}/${ARCHIVE}"
-
-    # Potentially we could have several running build for the same key (for example for the main branch), so we use improvised lock for this
-    - name: Acquire Allure lock
-      if: ${{ inputs.action == 'generate' }}
-      shell: bash -euxo pipefail {0}
-      env:
-        LOCK_FILE: reports/${{ steps.calculate-vars.outputs.KEY }}/lock.txt
-        TEST_SELECTION: ${{ steps.calculate-vars.outputs.TEST_SELECTION }}
-      run: |
-        LOCK_TIMEOUT=300 # seconds
-
-        for _ in $(seq 1 5); do
-          for i in $(seq 1 ${LOCK_TIMEOUT}); do
-            LOCK_ADDED=$(aws s3api head-object --bucket neon-github-public-dev --key ${LOCK_FILE} | jq --raw-output '.LastModified' || true)
-            # `date --date="..."` is supported only by gnu date (i.e. it doesn't work on BSD/macOS)
-            if [ -z "${LOCK_ADDED}" ] || [ "$(( $(date +%s) - $(date --date="${LOCK_ADDED}" +%s) ))" -gt "${LOCK_TIMEOUT}" ]; then
-              break
-            fi
-            sleep 1
-          done
-          echo "${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}-${TEST_SELECTION}" > lock.txt
-          aws s3 mv --only-show-errors lock.txt "s3://${BUCKET}/${LOCK_FILE}"
-
-          # A double-check that exactly WE have acquired the lock
-          aws s3 cp --only-show-errors "s3://${BUCKET}/${LOCK_FILE}" ./lock.txt
-          if [ "$(cat lock.txt)" = "${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}-${TEST_SELECTION}" ]; then
-            break
-          fi
-        done
-
-    - name: Generate and publish final Allure report
-      if: ${{ inputs.action == 'generate' }}
-      id: generate-report
-      env:
-        REPORT_PREFIX: reports/${{ steps.calculate-vars.outputs.KEY }}/${{ inputs.build_type }}
-        RAW_PREFIX: reports-raw/${{ steps.calculate-vars.outputs.KEY }}/${{ inputs.build_type }}
-      shell: bash -euxo pipefail {0}
-      run: |
-        # Get previously uploaded data for this run
-        ZSTD_NBTHREADS=0
-
-        s3_filepaths=$(aws s3api list-objects-v2 --bucket ${BUCKET} --prefix ${RAW_PREFIX}/${GITHUB_RUN_ID}- | jq --raw-output  '.Contents[].Key')
-        if [ -z "$s3_filepaths" ]; then
-          # There's no previously uploaded data for this run
-          exit 0
-        fi
-        for s3_filepath in ${s3_filepaths}; do
-          aws s3 cp --only-show-errors "s3://${BUCKET}/${s3_filepath}" "${TEST_OUTPUT}/allure/"
-
-          archive=${TEST_OUTPUT}/allure/$(basename $s3_filepath)
-          mkdir -p ${archive%.tar.zst}
-          tar -xf ${archive} -C ${archive%.tar.zst}
-          rm -f ${archive}
-        done
-
-        # Get history trend
-        aws s3 cp --recursive --only-show-errors "s3://${BUCKET}/${REPORT_PREFIX}/latest/history" "${TEST_OUTPUT}/allure/latest/history" || true
-
-        # Generate report
-        allure generate --clean --output $TEST_OUTPUT/allure/report $TEST_OUTPUT/allure/*
-
-        # Replace a logo link with a redirect to the latest version of the report
-        sed -i 's|<a href="." class=|<a href="https://'${BUCKET}'.s3.amazonaws.com/'${REPORT_PREFIX}'/latest/index.html" class=|g' $TEST_OUTPUT/allure/report/app.js
-
-        # Upload a history and the final report (in this particular order to not to have duplicated history in 2 places)
-        aws s3 mv --recursive --only-show-errors "${TEST_OUTPUT}/allure/report/history" "s3://${BUCKET}/${REPORT_PREFIX}/latest/history"
-        aws s3 mv --recursive --only-show-errors "${TEST_OUTPUT}/allure/report" "s3://${BUCKET}/${REPORT_PREFIX}/${GITHUB_RUN_ID}"
-
-        REPORT_URL=https://${BUCKET}.s3.amazonaws.com/${REPORT_PREFIX}/${GITHUB_RUN_ID}/index.html
-
-        # Generate redirect
-        cat <<EOF > ${TEST_OUTPUT}/allure/index.html
-          <!DOCTYPE html>
-
-          <meta charset="utf-8">
-          <title>Redirecting to ${REPORT_URL}</title>
-          <meta http-equiv="refresh" content="0; URL=${REPORT_URL}">
-        EOF
-        aws s3 cp --only-show-errors ${TEST_OUTPUT}/allure/index.html "s3://${BUCKET}/${REPORT_PREFIX}/latest/index.html"
-
-        echo "[Allure Report](${REPORT_URL})" >> ${GITHUB_STEP_SUMMARY}
-        echo "report-url=${REPORT_URL}" >> $GITHUB_OUTPUT
-        echo "report-json-url=${REPORT_URL%/index.html}/data/suites.json" >> $GITHUB_OUTPUT
-
-    - name: Release Allure lock
-      if: ${{ inputs.action == 'generate' && always() }}
-      shell: bash -euxo pipefail {0}
-      env:
-        LOCK_FILE: reports/${{ steps.calculate-vars.outputs.KEY }}/lock.txt
-        TEST_SELECTION: ${{ steps.calculate-vars.outputs.TEST_SELECTION }}
-      run: |
-        aws s3 cp --only-show-errors "s3://${BUCKET}/${LOCK_FILE}" ./lock.txt || exit 0
-
-        if [ "$(cat lock.txt)" = "${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}-${TEST_SELECTION}" ]; then
-          aws s3 rm "s3://${BUCKET}/${LOCK_FILE}"
-        fi
-
-    - name: Cleanup
-      if: always()
-      shell: bash -euxo pipefail {0}
-      run: |
-        rm -rf ${TEST_OUTPUT}/allure
-
-    - uses: actions/github-script@v6
-      if: ${{ inputs.action == 'generate' && always() }}
-      env:
-        REPORT_URL: ${{ steps.generate-report.outputs.report-url }}
-        SHA: ${{ github.event.pull_request.head.sha || github.sha }}
-      with:
-        script: |
-          const { REPORT_URL, BUILD_TYPE, SHA } = process.env
-
-          await github.rest.repos.createCommitStatus({
-            owner: context.repo.owner,
-            repo: context.repo.repo,
-            sha: `${SHA}`,
-            state: 'success',
-            target_url: `${REPORT_URL}`,
-            context: `Allure report / ${BUILD_TYPE}`,
-          })

.github/actions/run-python-test-set/action.yml

@@ -197,14 +197,13 @@ runs:
       uses: ./.github/actions/upload
       with:
         name: compatibility-snapshot-${{ inputs.build_type }}-pg14-${{ github.run_id }}
-        # The path includes a test name (test_create_snapshot) and directory that the test creates (compatibility_snapshot_pg14), keep the path in sync with the test
-        path: /tmp/test_output/test_create_snapshot/compatibility_snapshot_pg14/
+        # Directory is created by test_compatibility.py::test_create_snapshot, keep the path in sync with the test
+        path: /tmp/test_output/compatibility_snapshot_pg14/
         prefix: latest
 
-    - name: Create Allure report
+    - name: Upload test results
       if: ${{ !cancelled() }}
-      uses: ./.github/actions/allure-report
+      uses: ./.github/actions/allure-report-store
       with:
-        action: store
-        build_type: ${{ inputs.build_type }}
-        test_selection: ${{ inputs.test_selection }}
+        report-dir: /tmp/test_output/allure/results
+        unique-key: ${{ inputs.build_type }}

.github/workflows/benchmarking.yml

@@ -93,10 +93,7 @@ jobs:
 
     - name: Create Allure report
       if: ${{ !cancelled() }}
-      uses: ./.github/actions/allure-report
-      with:
-        action: generate
-        build_type: ${{ env.BUILD_TYPE }}
+      uses: ./.github/actions/allure-report-generate
 
     - name: Post to a Slack channel
       if: ${{ github.event.schedule && failure() }}
@@ -283,10 +280,7 @@ jobs:
 
     - name: Create Allure report
       if: ${{ !cancelled() }}
-      uses: ./.github/actions/allure-report
-      with:
-        action: generate
-        build_type: ${{ env.BUILD_TYPE }}
+      uses: ./.github/actions/allure-report-generate
 
     - name: Post to a Slack channel
       if: ${{ github.event.schedule && failure() }}
@@ -380,10 +374,7 @@ jobs:
 
     - name: Create Allure report
       if: ${{ !cancelled() }}
-      uses: ./.github/actions/allure-report
-      with:
-        action: generate
-        build_type: ${{ env.BUILD_TYPE }}
+      uses: ./.github/actions/allure-report-generate
 
     - name: Post to a Slack channel
       if: ${{ github.event.schedule && failure() }}
@@ -476,10 +467,7 @@ jobs:
 
     - name: Create Allure report
       if: ${{ !cancelled() }}
-      uses: ./.github/actions/allure-report
-      with:
-        action: generate
-        build_type: ${{ env.BUILD_TYPE }}
+      uses: ./.github/actions/allure-report-generate
 
     - name: Post to a Slack channel
       if: ${{ github.event.schedule && failure() }}
@@ -566,16 +554,13 @@ jobs:
 
     - name: Create Allure report
       if: ${{ !cancelled() }}
-      uses: ./.github/actions/allure-report
-      with:
-        action: generate
-        build_type: ${{ env.BUILD_TYPE }}
+      uses: ./.github/actions/allure-report-generate
 
     - name: Post to a Slack channel
       if: ${{ github.event.schedule && failure() }}
       uses: slackapi/slack-github-action@v1
       with:
         channel-id: "C033QLM5P7D" # dev-staging-stream
-        slack-message: "Periodic TPC-H perf testing ${{ matrix.platform }}: ${{ job.status }}\n${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
+        slack-message: "Periodic User example perf testing ${{ matrix.platform }}: ${{ job.status }}\n${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
       env:
         SLACK_BOT_TOKEN: ${{ secrets.SLACK_BOT_TOKEN }}

.github/workflows/build_and_test.yml

@@ -330,6 +330,7 @@ jobs:
       fail-fast: false
       matrix:
         build_type: [ debug, release ]
+        pg_version: [ v14, v15 ]
     steps:
       - name: Checkout
         uses: actions/checkout@v3
@@ -350,11 +351,12 @@ jobs:
           real_s3_secret_access_key: "${{ secrets.AWS_SECRET_ACCESS_KEY_CI_TESTS_S3 }}"
           rerun_flaky: true
         env:
+          DEFAULT_PG_VERSION: ${{ matrix.pg_version }}
           TEST_RESULT_CONNSTR: ${{ secrets.REGRESS_TEST_RESULT_CONNSTR }}
           CHECK_ONDISK_DATA_COMPATIBILITY: nonempty
 
       - name: Merge and upload coverage data
-        if: matrix.build_type == 'debug'
+        if: matrix.build_type == 'debug' && matrix.pg_version == 'v14'
         uses: ./.github/actions/save-coverage-data
 
   benchmarks:
@@ -399,21 +401,10 @@ jobs:
     steps:
       - uses: actions/checkout@v3
 
-      - name: Create Allure report (debug)
+      - name: Create Allure report
         if: ${{ !cancelled() }}
-        id: create-allure-report-debug
-        uses: ./.github/actions/allure-report
-        with:
-          action: generate
-          build_type: debug
-
-      - name: Create Allure report (release)
-        if: ${{ !cancelled() }}
-        id: create-allure-report-release
-        uses: ./.github/actions/allure-report
-        with:
-          action: generate
-          build_type: release
+        id: create-allure-report
+        uses: ./.github/actions/allure-report-generate
 
       - uses: actions/github-script@v6
         if: >
@@ -423,52 +414,37 @@ jobs:
           # Retry script for 5XX server errors: https://github.com/actions/github-script#retries
           retries: 5
           script: |
-            const reports = [{
-              buildType: "debug",
-              reportUrl: "${{ steps.create-allure-report-debug.outputs.report-url }}",
-              jsonUrl:   "${{ steps.create-allure-report-debug.outputs.report-json-url }}",
-            }, {
-              buildType: "release",
-              reportUrl: "${{ steps.create-allure-report-release.outputs.report-url }}",
-              jsonUrl:   "${{ steps.create-allure-report-release.outputs.report-json-url }}",
-            }]
+            const report = {
+              reportUrl:     "${{ steps.create-allure-report.outputs.report-url }}",
+              reportJsonUrl: "${{ steps.create-allure-report.outputs.report-json-url }}",
+            }
 
             const script = require("./scripts/pr-comment-test-report.js")
             await script({
               github,
               context,
               fetch,
-              reports,
+              report,
             })
 
       - name: Store Allure test stat in the DB
-        if: >
-          !cancelled() && (
-            steps.create-allure-report-debug.outputs.report-url ||
-            steps.create-allure-report-release.outputs.report-url
-          )
+        if: ${{ !cancelled() && steps.create-allure-report.outputs.report-json-url }}
         env:
-          SHA: ${{ github.event.pull_request.head.sha || github.sha }}
-          REPORT_JSON_URL_DEBUG: ${{ steps.create-allure-report-debug.outputs.report-json-url }}
-          REPORT_JSON_URL_RELEASE: ${{ steps.create-allure-report-release.outputs.report-json-url }}
+          COMMIT_SHA: ${{ github.event.pull_request.head.sha || github.sha }}
+          REPORT_JSON_URL: ${{ steps.create-allure-report.outputs.report-json-url }}
           TEST_RESULT_CONNSTR: ${{ secrets.REGRESS_TEST_RESULT_CONNSTR }}
         run: |
           ./scripts/pysync
 
-          for report_url in $REPORT_JSON_URL_DEBUG $REPORT_JSON_URL_RELEASE; do
-            if [ -z "$report_url" ]; then
-              continue
-            fi
+          curl --fail --output suites.json "${REPORT_JSON_URL}"
+          export BUILD_TYPE=unified
+          export DATABASE_URL="$TEST_RESULT_CONNSTR"
 
-            if [[ "$report_url" == "$REPORT_JSON_URL_DEBUG" ]]; then
-              BUILD_TYPE=debug
-            else
-              BUILD_TYPE=release
-            fi
-
-            curl --fail --output suites.json "${report_url}"
-            DATABASE_URL="$TEST_RESULT_CONNSTR" poetry run python3 scripts/ingest_regress_test_result.py --revision ${SHA} --reference ${GITHUB_REF} --build-type ${BUILD_TYPE} --ingest suites.json
-          done
+          poetry run python3 scripts/ingest_regress_test_result.py \
+            --revision ${COMMIT_SHA} \
+            --reference ${GITHUB_REF} \
+            --build-type ${BUILD_TYPE} \
+            --ingest suites.json
 
   coverage-report:
     runs-on: [ self-hosted, gen3, small ]

Cargo.lock

@@ -2859,22 +2859,22 @@ dependencies = [
 
 [[package]]
 name = "pin-project"
-version = "1.0.12"
+version = "1.1.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ad29a609b6bcd67fee905812e544992d216af9d755757c05ed2d0e15a74c6ecc"
+checksum = "c95a7476719eab1e366eaf73d0260af3021184f18177925b07f54b30089ceead"
 dependencies = [
  "pin-project-internal",
 ]
 
 [[package]]
 name = "pin-project-internal"
-version = "1.0.12"
+version = "1.1.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "069bdb1e05adc7a8990dce9cc75370895fbe4e3d58b9b73bf1aee56359344a55"
+checksum = "39407670928234ebc5e6e580247dd567ad73a3578460c5990f9503df207e8f07"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn 1.0.109",
+ "syn 2.0.15",
 ]
 
 [[package]]
@@ -4979,6 +4979,7 @@ dependencies = [
  "metrics",
  "nix",
  "once_cell",
+ "pin-project",
  "pin-project-lite",
  "pq_proto",
  "rand",

libs/pageserver_api/src/models.rs

@@ -48,13 +48,33 @@ pub enum TenantState {
 }
 
 impl TenantState {
-    pub fn has_in_progress_downloads(&self) -> bool {
+    pub fn attachment_status(&self) -> TenantAttachmentStatus {
+        use TenantAttachmentStatus::*;
         match self {
-            Self::Loading => true,
-            Self::Attaching => true,
-            Self::Active => false,
-            Self::Stopping => false,
-            Self::Broken { .. } => false,
+            // The attach procedure writes the marker file before adding the Attaching tenant to the tenants map.
+            // So, technically, we can return Attached here.
+            // However, as soon as Console observes Attached, it will proceed with the Postgres-level health check.
+            // But, our attach task might still be fetching the remote timelines, etc.
+            // So, return `Maybe` while Attaching, making Console wait for the attach task to finish.
+            Self::Attaching => Maybe,
+            // tenant mgr startup distinguishes attaching from loading via marker file.
+            // If it's loading, there is no attach marker file, i.e., attach had finished in the past.
+            Self::Loading => Attached,
+            // We only reach Active after successful load / attach.
+            // So, call atttachment status Attached.
+            Self::Active => Attached,
+            // If the (initial or resumed) attach procedure fails, the tenant becomes Broken.
+            // However, it also becomes Broken if the regular load fails.
+            // We would need a separate TenantState variant to distinguish these cases.
+            // However, there's no practical difference from Console's perspective.
+            // It will run a Postgres-level health check as soon as it observes Attached.
+            // That will fail on Broken tenants.
+            // Console can then rollback the attach, or, wait for operator to fix the Broken tenant.
+            Self::Broken { .. } => Attached,
+            // Why is Stopping a Maybe case? Because, during pageserver shutdown,
+            // we set the Stopping state irrespective of whether the tenant
+            // has finished attaching or not.
+            Self::Stopping => Maybe,
         }
     }
 
@@ -209,16 +229,25 @@ impl TenantConfigRequest {
     }
 }
 
+/// See [`TenantState::attachment_status`] and the OpenAPI docs for context.
+#[derive(Serialize, Deserialize, Clone)]
+#[serde(rename_all = "snake_case")]
+pub enum TenantAttachmentStatus {
+    Maybe,
+    Attached,
+}
+
 #[serde_as]
 #[derive(Serialize, Deserialize, Clone)]
 pub struct TenantInfo {
     #[serde_as(as = "DisplayFromStr")]
     pub id: TenantId,
+    // NB: intentionally not part of OpenAPI, we don't want to commit to a specific set of TenantState's
     pub state: TenantState,
     /// Sum of the size of all layer files.
     /// If a layer is present in both local FS and S3, it counts only once.
     pub current_physical_size: Option<u64>, // physical size is only included in `tenant_status` endpoint
-    pub has_in_progress_downloads: Option<bool>,
+    pub attachment_status: TenantAttachmentStatus,
 }
 
 /// This represents the output of the "timeline_detail" and "timeline_list" API calls.
@@ -691,7 +720,7 @@ mod tests {
             id: TenantId::generate(),
             state: TenantState::Active,
             current_physical_size: Some(42),
-            has_in_progress_downloads: Some(false),
+            attachment_status: TenantAttachmentStatus::Attached,
         };
         let expected_active = json!({
             "id": original_active.id.to_string(),
@@ -699,7 +728,7 @@ mod tests {
                 "slug": "Active",
             },
             "current_physical_size": 42,
-            "has_in_progress_downloads": false,
+            "attachment_status": "attached",
         });
 
         let original_broken = TenantInfo {
@@ -709,7 +738,7 @@ mod tests {
                 backtrace: "backtrace info".into(),
             },
             current_physical_size: Some(42),
-            has_in_progress_downloads: Some(false),
+            attachment_status: TenantAttachmentStatus::Attached,
         };
         let expected_broken = json!({
             "id": original_broken.id.to_string(),
@@ -721,7 +750,7 @@ mod tests {
                 }
             },
             "current_physical_size": 42,
-            "has_in_progress_downloads": false,
+            "attachment_status": "attached",
         });
 
         assert_eq!(

libs/postgres_ffi/src/pg_constants.rs

@@ -146,6 +146,10 @@ pub const XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED: u8 = (1 << 0) as u8;
 pub const XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED: u8 = (1 << 1) as u8;
 pub const XLH_DELETE_ALL_VISIBLE_CLEARED: u8 = (1 << 0) as u8;
 
+// From replication/message.h
+pub const XLOG_LOGICAL_MESSAGE: u8 = 0x00;
+
+// From rmgrlist.h
 pub const RM_XLOG_ID: u8 = 0;
 pub const RM_XACT_ID: u8 = 1;
 pub const RM_SMGR_ID: u8 = 2;
@@ -157,6 +161,7 @@ pub const RM_RELMAP_ID: u8 = 7;
 pub const RM_STANDBY_ID: u8 = 8;
 pub const RM_HEAP2_ID: u8 = 9;
 pub const RM_HEAP_ID: u8 = 10;
+pub const RM_LOGICALMSG_ID: u8 = 21;
 
 // from xlogreader.h
 pub const XLR_INFO_MASK: u8 = 0x0F;

libs/utils/Cargo.toml

@@ -41,6 +41,9 @@ pq_proto.workspace = true
 metrics.workspace = true
 workspace_hack.workspace = true
 
+# needed to have pin-projected with PinnedDrop
+pin-project = "1.1"
+
 [dev-dependencies]
 byteorder.workspace = true
 bytes.workspace = true

libs/utils/scripts/restore_from_wal.sh

@@ -1,21 +1,21 @@
 #!/bin/bash
+
+set -euxo pipefail
+
 PG_BIN=$1
 WAL_PATH=$2
 DATA_DIR=$3
 PORT=$4
-SYSID=`od -A n -j 24 -N 8 -t d8 $WAL_PATH/000000010000000000000002* | cut -c 3-`
-rm -fr $DATA_DIR
-env -i LD_LIBRARY_PATH=$PG_BIN/../lib $PG_BIN/initdb -E utf8 -U cloud_admin -D $DATA_DIR --sysid=$SYSID
-echo port=$PORT >> $DATA_DIR/postgresql.conf
-REDO_POS=0x`$PG_BIN/pg_controldata -D $DATA_DIR | fgrep "REDO location"| cut -c 42-`
+SYSID=$(od -A n -j 24 -N 8 -t d8 "$WAL_PATH"/000000010000000000000002* | cut -c 3-)
+rm -fr "$DATA_DIR"
+env -i LD_LIBRARY_PATH="$PG_BIN"/../lib "$PG_BIN"/initdb -E utf8 -U cloud_admin -D "$DATA_DIR" --sysid="$SYSID"
+echo port="$PORT" >> "$DATA_DIR"/postgresql.conf
+REDO_POS=0x$("$PG_BIN"/pg_controldata -D "$DATA_DIR" | grep -F "REDO location"| cut -c 42-)
 declare -i WAL_SIZE=$REDO_POS+114
-$PG_BIN/pg_ctl -D $DATA_DIR -l logfile start
-$PG_BIN/pg_ctl -D $DATA_DIR -l logfile stop -m immediate
-cp $DATA_DIR/pg_wal/000000010000000000000001 .
-cp $WAL_PATH/* $DATA_DIR/pg_wal/
-if [ -f $DATA_DIR/pg_wal/*.partial ]
-then
-	(cd $DATA_DIR/pg_wal ; for partial in \*.partial ; do mv $partial `basename $partial .partial` ; done)
-fi
-dd if=000000010000000000000001 of=$DATA_DIR/pg_wal/000000010000000000000001 bs=$WAL_SIZE count=1 conv=notrunc
+"$PG_BIN"/pg_ctl -D "$DATA_DIR" -l logfile start
+"$PG_BIN"/pg_ctl -D "$DATA_DIR" -l logfile stop -m immediate
+cp "$DATA_DIR"/pg_wal/000000010000000000000001 .
+cp "$WAL_PATH"/* "$DATA_DIR"/pg_wal/
+for partial in "$DATA_DIR"/pg_wal/*.partial ; do mv "$partial" "${partial%.partial}" ; done
+dd if=000000010000000000000001 of="$DATA_DIR"/pg_wal/000000010000000000000001 bs=$WAL_SIZE count=1 conv=notrunc
 rm -f 000000010000000000000001

libs/utils/scripts/restore_from_wal_archive.sh

@@ -1,20 +0,0 @@
-PG_BIN=$1
-WAL_PATH=$2
-DATA_DIR=$3
-PORT=$4
-SYSID=`od -A n -j 24 -N 8 -t d8 $WAL_PATH/000000010000000000000002* | cut -c 3-`
-rm -fr $DATA_DIR /tmp/pg_wals
-mkdir /tmp/pg_wals
-env -i LD_LIBRARY_PATH=$PG_BIN/../lib $PG_BIN/initdb -E utf8 -U cloud_admin -D $DATA_DIR --sysid=$SYSID
-echo port=$PORT >> $DATA_DIR/postgresql.conf
-REDO_POS=0x`$PG_BIN/pg_controldata -D $DATA_DIR | fgrep "REDO location"| cut -c 42-`
-declare -i WAL_SIZE=$REDO_POS+114
-cp $WAL_PATH/* /tmp/pg_wals
-if [ -f $DATA_DIR/pg_wal/*.partial ]
-then
-	(cd /tmp/pg_wals ; for partial in \*.partial ; do  mv $partial `basename $partial .partial` ; done)
-fi
-dd if=$DATA_DIR/pg_wal/000000010000000000000001 of=/tmp/pg_wals/000000010000000000000001 bs=$WAL_SIZE count=1 conv=notrunc
-echo > $DATA_DIR/recovery.signal
-rm -f $DATA_DIR/pg_wal/*
-echo "restore_command = 'cp /tmp/pg_wals/%f %p'" >> $DATA_DIR/postgresql.conf

libs/utils/src/lib.rs

@@ -60,24 +60,43 @@ pub mod tracing_span_assert;
 
 pub mod rate_limit;
 
-/// use with fail::cfg("$name", "return(2000)")
-#[macro_export]
-macro_rules! failpoint_sleep_millis_async {
-    ($name:literal) => {{
-        let should_sleep: Option<std::time::Duration> = (|| {
-            fail::fail_point!($name, |v: Option<_>| {
-                let millis = v.unwrap().parse::<u64>().unwrap();
-                Some(Duration::from_millis(millis))
-            });
-            None
-        })();
-        if let Some(d) = should_sleep {
-            tracing::info!("failpoint {:?}: sleeping for {:?}", $name, d);
-            tokio::time::sleep(d).await;
-            tracing::info!("failpoint {:?}: sleep done", $name);
-        }
-    }};
+mod failpoint_macro_helpers {
+
+    /// use with fail::cfg("$name", "return(2000)")
+    ///
+    /// The effect is similar to a "sleep(2000)" action, i.e. we sleep for the
+    /// specified time (in milliseconds). The main difference is that we use async
+    /// tokio sleep function. Another difference is that we print lines to the log,
+    /// which can be useful in tests to check that the failpoint was hit.
+    #[macro_export]
+    macro_rules! failpoint_sleep_millis_async {
+        ($name:literal) => {{
+            // If the failpoint is used with a "return" action, set should_sleep to the
+            // returned value (as string). Otherwise it's set to None.
+            let should_sleep = (|| {
+                ::fail::fail_point!($name, |x| x);
+                ::std::option::Option::None
+            })();
+
+            // Sleep if the action was a returned value
+            if let ::std::option::Option::Some(duration_str) = should_sleep {
+                $crate::failpoint_sleep_helper($name, duration_str).await
+            }
+        }};
+    }
+
+    // Helper function used by the macro. (A function has nicer scoping so we
+    // don't need to decorate everything with "::")
+    pub async fn failpoint_sleep_helper(name: &'static str, duration_str: String) {
+        let millis = duration_str.parse::<u64>().unwrap();
+        let d = std::time::Duration::from_millis(millis);
+
+        tracing::info!("failpoint {:?}: sleeping for {:?}", name, d);
+        tokio::time::sleep(d).await;
+        tracing::info!("failpoint {:?}: sleep done", name);
+    }
 }
+pub use failpoint_macro_helpers::failpoint_sleep_helper;
 
 /// This is a shortcut to embed git sha into binaries and avoid copying the same build script to all packages
 ///
@@ -128,3 +147,56 @@ macro_rules! const_assert {
         const _: () = assert!($($args)*);
     };
 }
+
+pub mod cancel_log {
+    pub trait CancelationLoggingExt {
+        fn log_being_canceled(self, step: &'static str) -> LogCancelation<Self>
+        where
+            Self: Sized;
+    }
+
+    impl<Fut: std::future::Future> CancelationLoggingExt for Fut {
+        fn log_being_canceled(self, step: &'static str) -> LogCancelation<Self>
+        where
+            Self: Sized,
+        {
+            LogCancelation {
+                inner: self,
+                name: Some(step),
+                span: tracing::Span::current(),
+            }
+        }
+    }
+
+    #[pin_project::pin_project(PinnedDrop)]
+    pub struct LogCancelation<Fut> {
+        #[pin]
+        inner: Fut,
+        name: Option<&'static str>,
+        span: tracing::Span,
+    }
+
+    impl<Fut: std::future::Future> std::future::Future for LogCancelation<Fut> {
+        type Output = Fut::Output;
+
+        fn poll(
+            self: std::pin::Pin<&mut Self>,
+            cx: &mut std::task::Context<'_>,
+        ) -> std::task::Poll<Self::Output> {
+            let this = self.project();
+            let ready = std::task::ready!(this.inner.poll(cx));
+            this.name.take();
+            std::task::Poll::Ready(ready)
+        }
+    }
+
+    #[pin_project::pinned_drop]
+    impl<Fut> PinnedDrop for LogCancelation<Fut> {
+        fn drop(self: std::pin::Pin<&mut Self>) {
+            if let Some(name) = self.name {
+                let _g = self.span.enter();
+                tracing::info!("canceled while waiting for {name}");
+            }
+        }
+    }
+}

pageserver/src/consumption_metrics.rs

@@ -5,7 +5,7 @@
 //!
 use crate::context::{DownloadBehavior, RequestContext};
 use crate::task_mgr::{self, TaskKind, BACKGROUND_RUNTIME};
-use crate::tenant::mgr;
+use crate::tenant::{mgr, LogicalSizeCalculationCause};
 use anyhow;
 use chrono::Utc;
 use consumption_metrics::{idempotency_key, Event, EventChunk, EventType, CHUNK_SIZE};
@@ -164,7 +164,8 @@ pub async fn collect_metrics_iteration(
                     timeline_written_size,
                 ));
 
-                match timeline.get_current_logical_size(ctx) {
+                let span = info_span!("collect_metrics_iteration", tenant_id = %timeline.tenant_id, timeline_id = %timeline.timeline_id);
+                match span.in_scope(|| timeline.get_current_logical_size(ctx)) {
                     // Only send timeline logical size when it is fully calculated.
                     Ok((size, is_exact)) if is_exact => {
                         current_metrics.push((
@@ -334,7 +335,9 @@ pub async fn calculate_synthetic_size_worker(
 
                     if let Ok(tenant) = mgr::get_tenant(tenant_id, true).await
                     {
-                        if let Err(e) = tenant.calculate_synthetic_size(ctx).await {
+                        if let Err(e) = tenant.calculate_synthetic_size(
+                            LogicalSizeCalculationCause::ConsumptionMetricsSyntheticSize,
+                            ctx).await {
                             error!("failed to calculate synthetic size for tenant {}: {}", tenant_id, e);
                         }
                     }

pageserver/src/http/openapi_spec.yml

@@ -346,23 +346,23 @@ paths:
         starts writing to the tenant's S3 state unless it receives one of the
         distinguished errors below that state otherwise.
 
-        The method to identify whether a request has arrived at the pageserver, and
-        whether it has succeeded, is to poll for the tenant status to reach "Active"
-        or "Broken" state. These values are currently not explicitly documented in
-        the API spec.
-        Polling for `has_in_progress_downloads == false` is INCORRECT because that
-        value can turn `false` during shutdown while the Attach operation is still
-        unfinished.
+        If a client receives a not-distinguished response, e.g., a network timeout,
+        it MUST retry the /attach request and poll again for the tenant's
+        attachment status.
+
+        After the client has received a 202, it MUST poll the tenant's
+        attachment status (field `attachment_status`) to reach state `attached`.
+        If the `attachment_status` is missing, the client MUST retry the `/attach`
+        request (goto previous paragraph). This is a robustness measure in case the tenant
+        status endpoint is buggy, but the attach operation is ongoing.
 
         There is no way to cancel an in-flight request.
 
-        If a client receives a not-distinguished response, e.g., a network timeout,
-        it MUST retry the /attach request and poll again for tenant status.
-
-        In any case, it must
-        * NOT ASSUME that the /attach request has been lost in the network,
-        * NOT ASSUME that the request has been lost based on a subsequent
-          tenant status request returning 404. (The request may still be in flight!)
+        In any case, the client
+        * MUST NOT ASSUME that the /attach request has been lost in the network,
+        * MUST NOT ASSUME that the request has been lost, based on the observation
+          that a subsequent tenant status request returns 404. The request may
+          still be in flight. It must be retried.
       responses:
         "202":
           description: Tenant attaching scheduled
@@ -888,13 +888,27 @@ components:
       type: object
       required:
         - id
+        - attachment_status
       properties:
         id:
           type: string
         current_physical_size:
           type: integer
-        has_in_progress_downloads:
-          type: boolean
+        attachment_status:
+          description: |
+            Status of this tenant's attachment to this pageserver.
+
+            - `maybe` means almost nothing, don't read anything into it
+              except for the fact that the pageserver _might_ be already
+              writing to the tenant's S3 state, so, DO NOT ATTACH the
+              tenant to any other pageserver, or we risk split-brain.
+            - `attached` means that the attach operation has completed,
+              maybe successfully, maybe not. Perform a health check at
+              the Postgres level to determine healthiness of the tenant.
+
+            See the tenant `/attach` endpoint for more information.
+          type: string
+          enum: [ "maybe", "attached" ]
     TenantCreateInfo:
       type: object
       properties:

pageserver/src/http/routes.rs

@@ -25,10 +25,11 @@ use crate::tenant::config::TenantConfOpt;
 use crate::tenant::mgr::{TenantMapInsertError, TenantStateError};
 use crate::tenant::size::ModelInputs;
 use crate::tenant::storage_layer::LayerAccessStatsReset;
-use crate::tenant::{PageReconstructError, Timeline};
+use crate::tenant::{LogicalSizeCalculationCause, PageReconstructError, Timeline};
 use crate::{config::PageServerConf, tenant::mgr};
 use utils::{
     auth::JwtAuth,
+    cancel_log::CancelationLoggingExt,
     http::{
         endpoint::{self, attach_openapi_ui, auth_middleware, check_permission_with},
         error::{ApiError, HttpErrorBody},
@@ -105,6 +106,9 @@ impl From<PageReconstructError> for ApiError {
             PageReconstructError::Cancelled => {
                 ApiError::InternalServerError(anyhow::anyhow!("request was cancelled"))
             }
+            PageReconstructError::AncestorStopping(_) => {
+                ApiError::InternalServerError(anyhow::Error::new(pre))
+            }
             PageReconstructError::WalRedo(pre) => {
                 ApiError::InternalServerError(anyhow::Error::new(pre))
             }
@@ -169,6 +173,8 @@ async fn build_timeline_info(
     include_non_incremental_logical_size: bool,
     ctx: &RequestContext,
 ) -> anyhow::Result<TimelineInfo> {
+    crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
+
     let mut info = build_timeline_info_common(timeline, ctx)?;
     if include_non_incremental_logical_size {
         // XXX we should be using spawn_ondemand_logical_size_calculation here.
@@ -181,6 +187,7 @@ async fn build_timeline_info(
                     CancellationToken::new(),
                     ctx,
                 )
+                .log_being_canceled("get_current_logical_size_non_incremental")
                 .await?,
         );
     }
@@ -191,6 +198,7 @@ fn build_timeline_info_common(
     timeline: &Arc<Timeline>,
     ctx: &RequestContext,
 ) -> anyhow::Result<TimelineInfo> {
+    crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
     let last_record_lsn = timeline.get_last_record_lsn();
     let (wal_source_connstr, last_received_msg_lsn, last_received_msg_ts) = {
         let guard = timeline.last_received_wal.lock().unwrap();
@@ -263,25 +271,28 @@ async fn timeline_create_handler(mut request: Request<Body>) -> Result<Response<
 
     let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Error);
 
-    let tenant = mgr::get_tenant(tenant_id, true).await?;
-    match tenant.create_timeline(
-        new_timeline_id,
-        request_data.ancestor_timeline_id.map(TimelineId::from),
-        request_data.ancestor_start_lsn,
-        request_data.pg_version.unwrap_or(crate::DEFAULT_PG_VERSION),
-        &ctx,
-    )
-    .instrument(info_span!("timeline_create", tenant = %tenant_id, new_timeline = ?request_data.new_timeline_id, timeline_id = %new_timeline_id, lsn=?request_data.ancestor_start_lsn, pg_version=?request_data.pg_version))
-    .await {
-        Ok(Some(new_timeline)) => {
-            // Created. Construct a TimelineInfo for it.
-            let timeline_info = build_timeline_info_common(&new_timeline, &ctx)
-                .map_err(ApiError::InternalServerError)?;
-            json_response(StatusCode::CREATED, timeline_info)
+    async {
+        let tenant = mgr::get_tenant(tenant_id, true).await?;
+        match tenant.create_timeline(
+            new_timeline_id,
+            request_data.ancestor_timeline_id.map(TimelineId::from),
+            request_data.ancestor_start_lsn,
+            request_data.pg_version.unwrap_or(crate::DEFAULT_PG_VERSION),
+            &ctx,
+        )
+        .await {
+            Ok(Some(new_timeline)) => {
+                // Created. Construct a TimelineInfo for it.
+                let timeline_info = build_timeline_info_common(&new_timeline, &ctx)
+                    .map_err(ApiError::InternalServerError)?;
+                json_response(StatusCode::CREATED, timeline_info)
+            }
+            Ok(None) => json_response(StatusCode::CONFLICT, ()), // timeline already exists
+            Err(err) => Err(ApiError::InternalServerError(err)),
         }
-        Ok(None) => json_response(StatusCode::CONFLICT, ()), // timeline already exists
-        Err(err) => Err(ApiError::InternalServerError(err)),
     }
+    .instrument(info_span!("timeline_create", tenant = %tenant_id, new_timeline = ?request_data.new_timeline_id, timeline_id = %new_timeline_id, lsn=?request_data.ancestor_start_lsn, pg_version=?request_data.pg_version))
+    .await
 }
 
 async fn timeline_list_handler(request: Request<Body>) -> Result<Response<Body>, ApiError> {
@@ -303,6 +314,7 @@ async fn timeline_list_handler(request: Request<Body>) -> Result<Response<Body>,
                 include_non_incremental_logical_size.unwrap_or(false),
                 &ctx,
             )
+            .instrument(info_span!("build_timeline_info", timeline_id = %timeline.timeline_id))
             .await
             .context("Failed to convert tenant timeline {timeline_id} into the local one: {e:?}")
             .map_err(ApiError::InternalServerError)?;
@@ -328,7 +340,9 @@ async fn timeline_detail_handler(request: Request<Body>) -> Result<Response<Body
     let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
 
     let timeline_info = async {
-        let tenant = mgr::get_tenant(tenant_id, true).await?;
+        let tenant = mgr::get_tenant(tenant_id, true)
+            .log_being_canceled("get_tenant")
+            .await?;
 
         let timeline = tenant
             .get_timeline(timeline_id, false)
@@ -339,6 +353,7 @@ async fn timeline_detail_handler(request: Request<Body>) -> Result<Response<Body
             include_non_incremental_logical_size.unwrap_or(false),
             &ctx,
         )
+        .log_being_canceled("build_timeline_info")
         .await
         .context("get local timeline info")
         .map_err(ApiError::InternalServerError)?;
@@ -346,6 +361,7 @@ async fn timeline_detail_handler(request: Request<Body>) -> Result<Response<Body
         Ok::<_, ApiError>(timeline_info)
     }
     .instrument(info_span!("timeline_detail", tenant = %tenant_id, timeline = %timeline_id))
+    .log_being_canceled("whole response")
     .await?;
 
     json_response(StatusCode::OK, timeline_info)
@@ -467,7 +483,7 @@ async fn tenant_list_handler(request: Request<Body>) -> Result<Response<Body>, A
             id: *id,
             state: state.clone(),
             current_physical_size: None,
-            has_in_progress_downloads: Some(state.has_in_progress_downloads()),
+            attachment_status: state.attachment_status(),
         })
         .collect::<Vec<TenantInfo>>();
 
@@ -492,7 +508,7 @@ async fn tenant_status(request: Request<Body>) -> Result<Response<Body>, ApiErro
             id: tenant_id,
             state: state.clone(),
             current_physical_size: Some(current_physical_size),
-            has_in_progress_downloads: Some(state.has_in_progress_downloads()),
+            attachment_status: state.attachment_status(),
         })
     }
     .instrument(info_span!("tenant_status_handler", tenant = %tenant_id))
@@ -527,7 +543,11 @@ async fn tenant_size_handler(request: Request<Body>) -> Result<Response<Body>, A
 
     // this can be long operation
     let inputs = tenant
-        .gather_size_inputs(retention_period, &ctx)
+        .gather_size_inputs(
+            retention_period,
+            LogicalSizeCalculationCause::TenantSizeHandler,
+            &ctx,
+        )
         .await
         .map_err(ApiError::InternalServerError)?;
 

pageserver/src/metrics.rs

@@ -30,6 +30,7 @@ const STORAGE_TIME_OPERATIONS: &[&str] = &[
     "create images",
     "init logical size",
     "logical size",
+    "imitate logical size",
     "load layer map",
     "gc",
 ];
@@ -186,6 +187,16 @@ static PERSISTENT_BYTES_WRITTEN: Lazy<IntCounterVec> = Lazy::new(|| {
     .expect("failed to define a metric")
 });
 
+pub(crate) static EVICTION_ITERATION_DURATION: Lazy<HistogramVec> = Lazy::new(|| {
+    register_histogram_vec!(
+        "pageserver_eviction_iteration_duration_seconds_global",
+        "Time spent on a single eviction iteration",
+        &["period_secs", "threshold_secs"],
+        STORAGE_OP_BUCKETS.into(),
+    )
+    .expect("failed to define a metric")
+});
+
 static EVICTIONS: Lazy<IntCounterVec> = Lazy::new(|| {
     register_int_counter_vec!(
         "pageserver_evictions",
@@ -478,6 +489,15 @@ pub static TENANT_TASK_EVENTS: Lazy<IntCounterVec> = Lazy::new(|| {
     .expect("Failed to register tenant_task_events metric")
 });
 
+pub static BACKGROUND_LOOP_PERIOD_OVERRUN_COUNT: Lazy<IntCounterVec> = Lazy::new(|| {
+    register_int_counter_vec!(
+        "pageserver_background_loop_period_overrun_count",
+        "Incremented whenever warn_when_period_overrun() logs a warning.",
+        &["task", "period"],
+    )
+    .expect("failed to define a metric")
+});
+
 // walreceiver metrics
 
 pub static WALRECEIVER_STARTED_CONNECTIONS: Lazy<IntCounter> = Lazy::new(|| {
@@ -688,6 +708,7 @@ pub struct TimelineMetrics {
     pub compact_time_histo: StorageTimeMetrics,
     pub create_images_time_histo: StorageTimeMetrics,
     pub logical_size_histo: StorageTimeMetrics,
+    pub imitate_logical_size_histo: StorageTimeMetrics,
     pub load_layer_map_histo: StorageTimeMetrics,
     pub garbage_collect_histo: StorageTimeMetrics,
     pub last_record_gauge: IntGauge,
@@ -720,6 +741,8 @@ impl TimelineMetrics {
         let create_images_time_histo =
             StorageTimeMetrics::new("create images", &tenant_id, &timeline_id);
         let logical_size_histo = StorageTimeMetrics::new("logical size", &tenant_id, &timeline_id);
+        let imitate_logical_size_histo =
+            StorageTimeMetrics::new("imitate logical size", &tenant_id, &timeline_id);
         let load_layer_map_histo =
             StorageTimeMetrics::new("load layer map", &tenant_id, &timeline_id);
         let garbage_collect_histo = StorageTimeMetrics::new("gc", &tenant_id, &timeline_id);
@@ -756,6 +779,7 @@ impl TimelineMetrics {
             compact_time_histo,
             create_images_time_histo,
             logical_size_histo,
+            imitate_logical_size_histo,
             garbage_collect_histo,
             load_layer_map_histo,
             last_record_gauge,
@@ -1216,4 +1240,7 @@ pub fn preinitialize_metrics() {
     // Initialize it eagerly, so that our alert rule can distinguish absence of the metric from metric value 0.
     assert_eq!(UNEXPECTED_ONDEMAND_DOWNLOADS.get(), 0);
     UNEXPECTED_ONDEMAND_DOWNLOADS.reset();
+
+    // Same as above for this metric, but, it's a Vec-type metric for which we don't know all the labels.
+    BACKGROUND_LOOP_PERIOD_OVERRUN_COUNT.reset();
 }

pageserver/src/page_service.rs

@@ -256,7 +256,10 @@ async fn page_service_conn_main(
     //
     // no write timeout is used, because the kernel is assumed to error writes after some time.
     let mut socket = tokio_io_timeout::TimeoutReader::new(socket);
-    socket.set_timeout(Some(std::time::Duration::from_secs(60 * 10)));
+
+    // timeout should be lower, but trying out multiple days for
+    // <https://github.com/neondatabase/neon/issues/4205>
+    socket.set_timeout(Some(std::time::Duration::from_secs(60 * 60 * 24 * 3)));
     let socket = std::pin::pin!(socket);
 
     // XXX: pgbackend.run() should take the connection_ctx,

pageserver/src/pgdatadir_mapping.rs

@@ -500,6 +500,8 @@ impl Timeline {
         cancel: CancellationToken,
         ctx: &RequestContext,
     ) -> Result<u64, CalculateLogicalSizeError> {
+        crate::tenant::debug_assert_current_span_has_tenant_and_timeline_id();
+
         // Fetch list of database dirs and iterate them
         let buf = self.get(DBDIR_KEY, lsn, ctx).await.context("read dbdir")?;
         let dbdir = DbDirectory::des(&buf).context("deserialize db directory")?;

pageserver/src/tenant.rs

@@ -97,7 +97,10 @@ mod timeline;
 
 pub mod size;
 
-pub use timeline::{LocalLayerInfoForDiskUsageEviction, PageReconstructError, Timeline};
+pub(crate) use timeline::debug_assert_current_span_has_tenant_and_timeline_id;
+pub use timeline::{
+    LocalLayerInfoForDiskUsageEviction, LogicalSizeCalculationCause, PageReconstructError, Timeline,
+};
 
 // re-export this function so that page_cache.rs can use it.
 pub use crate::tenant::ephemeral_file::writeback as writeback_ephemeral_file;
@@ -593,16 +596,19 @@ impl Tenant {
     /// finishes. You can use wait_until_active() to wait for the task to
     /// complete.
     ///
-    pub fn spawn_attach(
+    pub(crate) fn spawn_attach(
         conf: &'static PageServerConf,
         tenant_id: TenantId,
         remote_storage: GenericRemoteStorage,
         ctx: &RequestContext,
-    ) -> Arc<Tenant> {
+    ) -> anyhow::Result<Arc<Tenant>> {
         // XXX: Attach should provide the config, especially during tenant migration.
         //      See https://github.com/neondatabase/neon/issues/1555
         let tenant_conf = TenantConfOpt::default();
 
+        Self::attach_idempotent_create_marker_file(conf, tenant_id)
+            .context("create attach marker file")?;
+
         let wal_redo_manager = Arc::new(PostgresRedoManager::new(conf, tenant_id));
         let tenant = Arc::new(Tenant::new(
             TenantState::Attaching,
@@ -635,7 +641,46 @@ impl Tenant {
                 Ok(())
             },
         );
-        tenant
+        Ok(tenant)
+    }
+
+    fn attach_idempotent_create_marker_file(
+        conf: &'static PageServerConf,
+        tenant_id: TenantId,
+    ) -> anyhow::Result<()> {
+        // Create directory with marker file to indicate attaching state.
+        // The load_local_tenants() function in tenant::mgr relies on the marker file
+        // to determine whether a tenant has finished attaching.
+        let tenant_dir = conf.tenant_path(&tenant_id);
+        let marker_file = conf.tenant_attaching_mark_file_path(&tenant_id);
+        debug_assert_eq!(marker_file.parent().unwrap(), tenant_dir);
+        // TODO: should use tokio::fs here, but
+        // 1. caller is not async, for good reason (it holds tenants map lock)
+        // 2. we'd need to think about cancel safety. Turns out dropping a tokio::fs future
+        //    doesn't wait for the activity in the fs thread pool.
+        crashsafe::create_dir_all(&tenant_dir).context("create tenant directory")?;
+        match fs::OpenOptions::new()
+            .write(true)
+            .create_new(true)
+            .open(&marker_file)
+        {
+            Ok(_) => {}
+            Err(e) if e.kind() == std::io::ErrorKind::AlreadyExists => {
+                // Either this is a retry of attach or there is a concurrent task also doing attach for this tenant.
+                // We cannot distinguish this here.
+                // The caller is responsible for ensuring there's no concurrent attach for a tenant.
+                {} // fsync again, we don't know if that already happened
+            }
+            err => {
+                err.context("create tenant attaching marker file")?;
+                unreachable!("we covered the Ok() case above");
+            }
+        }
+        crashsafe::fsync_file_and_parent(&marker_file)
+            .context("fsync tenant attaching marker file and parent")?;
+        debug_assert!(tenant_dir.is_dir());
+        debug_assert!(marker_file.is_file());
+        Ok(())
     }
 
     ///
@@ -643,26 +688,15 @@ impl Tenant {
     ///
     #[instrument(skip_all, fields(tenant_id=%self.tenant_id))]
     async fn attach(self: &Arc<Tenant>, ctx: RequestContext) -> anyhow::Result<()> {
-        // Create directory with marker file to indicate attaching state.
-        // The load_local_tenants() function in tenant::mgr relies on the marker file
-        // to determine whether a tenant has finished attaching.
-        let tenant_dir = self.conf.tenant_path(&self.tenant_id);
         let marker_file = self.conf.tenant_attaching_mark_file_path(&self.tenant_id);
-        debug_assert_eq!(marker_file.parent().unwrap(), tenant_dir);
-        if tenant_dir.exists() {
-            if !marker_file.is_file() {
-                anyhow::bail!(
-                    "calling Tenant::attach with a tenant directory that doesn't have the attaching marker file:\ntenant_dir: {}\nmarker_file: {}",
-                    tenant_dir.display(), marker_file.display());
-            }
-        } else {
-            crashsafe::create_dir_all(&tenant_dir).context("create tenant directory")?;
-            fs::File::create(&marker_file).context("create tenant attaching marker file")?;
-            crashsafe::fsync_file_and_parent(&marker_file)
-                .context("fsync tenant attaching marker file and parent")?;
+        if !tokio::fs::try_exists(&marker_file)
+            .await
+            .context("check for existence of marker file")?
+        {
+            anyhow::bail!(
+                "implementation error: marker file should exist at beginning of this function"
+            );
         }
-        debug_assert!(tenant_dir.is_dir());
-        debug_assert!(marker_file.is_file());
 
         // Get list of remote timelines
         // download index files for every tenant timeline
@@ -839,11 +873,15 @@ impl Tenant {
     }
 
     /// Create a placeholder Tenant object for a broken tenant
-    pub fn create_broken_tenant(conf: &'static PageServerConf, tenant_id: TenantId) -> Arc<Tenant> {
+    pub fn create_broken_tenant(
+        conf: &'static PageServerConf,
+        tenant_id: TenantId,
+        reason: String,
+    ) -> Arc<Tenant> {
         let wal_redo_manager = Arc::new(PostgresRedoManager::new(conf, tenant_id));
         Arc::new(Tenant::new(
             TenantState::Broken {
-                reason: "create_broken_tenant".into(),
+                reason,
                 backtrace: String::new(),
             },
             conf,
@@ -876,7 +914,7 @@ impl Tenant {
             Ok(conf) => conf,
             Err(e) => {
                 error!("load tenant config failed: {:?}", e);
-                return Tenant::create_broken_tenant(conf, tenant_id);
+                return Tenant::create_broken_tenant(conf, tenant_id, format!("{e:#}"));
             }
         };
 
@@ -1226,8 +1264,24 @@ impl Tenant {
             "Cannot create timelines on inactive tenant"
         );
 
-        if self.get_timeline(new_timeline_id, false).is_ok() {
+        if let Ok(existing) = self.get_timeline(new_timeline_id, false) {
             debug!("timeline {new_timeline_id} already exists");
+
+            if let Some(remote_client) = existing.remote_client.as_ref() {
+                // Wait for uploads to complete, so that when we return Ok, the timeline
+                // is known to be durable on remote storage. Just like we do at the end of
+                // this function, after we have created the timeline ourselves.
+                //
+                // We only really care that the initial version of `index_part.json` has
+                // been uploaded. That's enough to remember that the timeline
+                // exists. However, there is no function to wait specifically for that so
+                // we just wait for all in-progress uploads to finish.
+                remote_client
+                    .wait_completion()
+                    .await
+                    .context("wait for timeline uploads to complete")?;
+            }
+
             return Ok(None);
         }
 
@@ -1269,6 +1323,17 @@ impl Tenant {
             }
         };
 
+        if let Some(remote_client) = loaded_timeline.remote_client.as_ref() {
+            // Wait for the upload of the 'index_part.json` file to finish, so that when we return
+            // Ok, the timeline is durable in remote storage.
+            let kind = ancestor_timeline_id
+                .map(|_| "branched")
+                .unwrap_or("bootstrapped");
+            remote_client.wait_completion().await.with_context(|| {
+                format!("wait for {} timeline initial uploads to complete", kind)
+            })?;
+        }
+
         Ok(Some(loaded_timeline))
     }
 
@@ -2338,17 +2403,18 @@ impl Tenant {
             src_timeline.initdb_lsn,
             src_timeline.pg_version,
         );
-        let mut timelines = self.timelines.lock().unwrap();
-        let new_timeline = self
-            .prepare_timeline(
+
+        let new_timeline = {
+            let mut timelines = self.timelines.lock().unwrap();
+            self.prepare_timeline(
                 dst_id,
                 &metadata,
                 timeline_uninit_mark,
                 false,
                 Some(Arc::clone(src_timeline)),
             )?
-            .initialize_with_lock(ctx, &mut timelines, true, true)?;
-        drop(timelines);
+            .initialize_with_lock(ctx, &mut timelines, true, true)?
+        };
 
         // Root timeline gets its layers during creation and uploads them along with the metadata.
         // A branch timeline though, when created, can get no writes for some time, hence won't get any layers created.
@@ -2606,6 +2672,7 @@ impl Tenant {
         // `max_retention_period` overrides the cutoff that is used to calculate the size
         // (only if it is shorter than the real cutoff).
         max_retention_period: Option<u64>,
+        cause: LogicalSizeCalculationCause,
         ctx: &RequestContext,
     ) -> anyhow::Result<size::ModelInputs> {
         let logical_sizes_at_once = self
@@ -2627,6 +2694,7 @@ impl Tenant {
             logical_sizes_at_once,
             max_retention_period,
             &mut shared_cache,
+            cause,
             ctx,
         )
         .await
@@ -2636,8 +2704,12 @@ impl Tenant {
     /// This is periodically called by background worker.
     /// result is cached in tenant struct
     #[instrument(skip_all, fields(tenant_id=%self.tenant_id))]
-    pub async fn calculate_synthetic_size(&self, ctx: &RequestContext) -> anyhow::Result<u64> {
-        let inputs = self.gather_size_inputs(None, ctx).await?;
+    pub async fn calculate_synthetic_size(
+        &self,
+        cause: LogicalSizeCalculationCause,
+        ctx: &RequestContext,
+    ) -> anyhow::Result<u64> {
+        let inputs = self.gather_size_inputs(None, cause, ctx).await?;
 
         let size = inputs.calculate()?;
 
@@ -3397,14 +3469,26 @@ mod tests {
             .gc_iteration(Some(TIMELINE_ID), 0x10, Duration::ZERO, &ctx)
             .await?;
 
+        // The branchpoints should contain all timelines, even ones marked
+        // as Broken.
+        {
+            let branchpoints = &tline.gc_info.read().unwrap().retain_lsns;
+            assert_eq!(branchpoints.len(), 1);
+            assert_eq!(branchpoints[0], Lsn(0x40));
+        }
+
+        // You can read the key from the child branch even though the parent is
+        // Broken, as long as you don't need to access data from the parent.
         assert_eq!(
-            newtline.get(*TEST_KEY, Lsn(0x50), &ctx).await?,
-            TEST_IMG(&format!("foo at {}", Lsn(0x40)))
+            newtline.get(*TEST_KEY, Lsn(0x70), &ctx).await?,
+            TEST_IMG(&format!("foo at {}", Lsn(0x70)))
         );
 
-        let branchpoints = &tline.gc_info.read().unwrap().retain_lsns;
-        assert_eq!(branchpoints.len(), 1);
-        assert_eq!(branchpoints[0], Lsn(0x40));
+        // This needs to traverse to the parent, and fails.
+        let err = newtline.get(*TEST_KEY, Lsn(0x50), &ctx).await.unwrap_err();
+        assert!(err
+            .to_string()
+            .contains("will not become active. Current state: Broken"));
 
         Ok(())
     }

pageserver/src/tenant/config.rs

@@ -292,77 +292,93 @@ fn bad_duration<'a>(field_name: &'static str, value: &'a str) -> impl 'a + Fn()
     move || format!("Cannot parse `{field_name}` duration {value:?}")
 }
 
-impl TryFrom<&'_ TenantCreateRequest> for TenantConfOpt {
-    type Error = anyhow::Error;
-
-    fn try_from(request_data: &TenantCreateRequest) -> Result<Self, Self::Error> {
+impl TenantConfOpt {
+    #[allow(clippy::too_many_arguments)]
+    fn from_request(
+        checkpoint_distance: Option<u64>,
+        checkpoint_timeout: &Option<String>,
+        compaction_target_size: Option<u64>,
+        compaction_period: &Option<String>,
+        compaction_threshold: Option<usize>,
+        gc_horizon: Option<u64>,
+        gc_period: &Option<String>,
+        image_creation_threshold: Option<usize>,
+        pitr_interval: &Option<String>,
+        walreceiver_connect_timeout: &Option<String>,
+        lagging_wal_timeout: &Option<String>,
+        max_lsn_wal_lag: Option<NonZeroU64>,
+        trace_read_requests: Option<bool>,
+        eviction_policy: &Option<serde_json::Value>,
+        min_resident_size_override: Option<u64>,
+        evictions_low_residence_duration_metric_threshold: &Option<String>,
+    ) -> Result<Self, anyhow::Error> {
         let mut tenant_conf = TenantConfOpt::default();
 
-        if let Some(gc_period) = &request_data.gc_period {
+        if let Some(gc_period) = &gc_period {
             tenant_conf.gc_period = Some(
                 humantime::parse_duration(gc_period)
                     .with_context(bad_duration("gc_period", gc_period))?,
             );
         }
-        tenant_conf.gc_horizon = request_data.gc_horizon;
-        tenant_conf.image_creation_threshold = request_data.image_creation_threshold;
+        tenant_conf.gc_horizon = gc_horizon;
+        tenant_conf.image_creation_threshold = image_creation_threshold;
 
-        if let Some(pitr_interval) = &request_data.pitr_interval {
+        if let Some(pitr_interval) = &pitr_interval {
             tenant_conf.pitr_interval = Some(
                 humantime::parse_duration(pitr_interval)
                     .with_context(bad_duration("pitr_interval", pitr_interval))?,
             );
         }
 
-        if let Some(walreceiver_connect_timeout) = &request_data.walreceiver_connect_timeout {
+        if let Some(walreceiver_connect_timeout) = &walreceiver_connect_timeout {
             tenant_conf.walreceiver_connect_timeout = Some(
                 humantime::parse_duration(walreceiver_connect_timeout).with_context(
                     bad_duration("walreceiver_connect_timeout", walreceiver_connect_timeout),
                 )?,
             );
         }
-        if let Some(lagging_wal_timeout) = &request_data.lagging_wal_timeout {
+        if let Some(lagging_wal_timeout) = &lagging_wal_timeout {
             tenant_conf.lagging_wal_timeout = Some(
                 humantime::parse_duration(lagging_wal_timeout)
                     .with_context(bad_duration("lagging_wal_timeout", lagging_wal_timeout))?,
             );
         }
-        if let Some(max_lsn_wal_lag) = request_data.max_lsn_wal_lag {
+        if let Some(max_lsn_wal_lag) = max_lsn_wal_lag {
             tenant_conf.max_lsn_wal_lag = Some(max_lsn_wal_lag);
         }
-        if let Some(trace_read_requests) = request_data.trace_read_requests {
+        if let Some(trace_read_requests) = trace_read_requests {
             tenant_conf.trace_read_requests = Some(trace_read_requests);
         }
 
-        tenant_conf.checkpoint_distance = request_data.checkpoint_distance;
-        if let Some(checkpoint_timeout) = &request_data.checkpoint_timeout {
+        tenant_conf.checkpoint_distance = checkpoint_distance;
+        if let Some(checkpoint_timeout) = &checkpoint_timeout {
             tenant_conf.checkpoint_timeout = Some(
                 humantime::parse_duration(checkpoint_timeout)
                     .with_context(bad_duration("checkpoint_timeout", checkpoint_timeout))?,
             );
         }
 
-        tenant_conf.compaction_target_size = request_data.compaction_target_size;
-        tenant_conf.compaction_threshold = request_data.compaction_threshold;
+        tenant_conf.compaction_target_size = compaction_target_size;
+        tenant_conf.compaction_threshold = compaction_threshold;
 
-        if let Some(compaction_period) = &request_data.compaction_period {
+        if let Some(compaction_period) = &compaction_period {
             tenant_conf.compaction_period = Some(
                 humantime::parse_duration(compaction_period)
                     .with_context(bad_duration("compaction_period", compaction_period))?,
             );
         }
 
-        if let Some(eviction_policy) = &request_data.eviction_policy {
+        if let Some(eviction_policy) = &eviction_policy {
             tenant_conf.eviction_policy = Some(
                 serde::Deserialize::deserialize(eviction_policy)
                     .context("parse field `eviction_policy`")?,
             );
         }
 
-        tenant_conf.min_resident_size_override = request_data.min_resident_size_override;
+        tenant_conf.min_resident_size_override = min_resident_size_override;
 
         if let Some(evictions_low_residence_duration_metric_threshold) =
-            &request_data.evictions_low_residence_duration_metric_threshold
+            &evictions_low_residence_duration_metric_threshold
         {
             tenant_conf.evictions_low_residence_duration_metric_threshold = Some(
                 humantime::parse_duration(evictions_low_residence_duration_metric_threshold)
@@ -377,81 +393,53 @@ impl TryFrom<&'_ TenantCreateRequest> for TenantConfOpt {
     }
 }
 
+impl TryFrom<&'_ TenantCreateRequest> for TenantConfOpt {
+    type Error = anyhow::Error;
+
+    fn try_from(request_data: &TenantCreateRequest) -> Result<Self, Self::Error> {
+        Self::from_request(
+            request_data.checkpoint_distance,
+            &request_data.checkpoint_timeout,
+            request_data.compaction_target_size,
+            &request_data.compaction_period,
+            request_data.compaction_threshold,
+            request_data.gc_horizon,
+            &request_data.gc_period,
+            request_data.image_creation_threshold,
+            &request_data.pitr_interval,
+            &request_data.walreceiver_connect_timeout,
+            &request_data.lagging_wal_timeout,
+            request_data.max_lsn_wal_lag,
+            request_data.trace_read_requests,
+            &request_data.eviction_policy,
+            request_data.min_resident_size_override,
+            &request_data.evictions_low_residence_duration_metric_threshold,
+        )
+    }
+}
+
 impl TryFrom<&'_ TenantConfigRequest> for TenantConfOpt {
     type Error = anyhow::Error;
 
     fn try_from(request_data: &TenantConfigRequest) -> Result<Self, Self::Error> {
-        let mut tenant_conf = TenantConfOpt::default();
-        if let Some(gc_period) = &request_data.gc_period {
-            tenant_conf.gc_period = Some(
-                humantime::parse_duration(gc_period)
-                    .with_context(bad_duration("gc_period", gc_period))?,
-            );
-        }
-        tenant_conf.gc_horizon = request_data.gc_horizon;
-        tenant_conf.image_creation_threshold = request_data.image_creation_threshold;
-
-        if let Some(pitr_interval) = &request_data.pitr_interval {
-            tenant_conf.pitr_interval = Some(
-                humantime::parse_duration(pitr_interval)
-                    .with_context(bad_duration("pitr_interval", pitr_interval))?,
-            );
-        }
-        if let Some(walreceiver_connect_timeout) = &request_data.walreceiver_connect_timeout {
-            tenant_conf.walreceiver_connect_timeout = Some(
-                humantime::parse_duration(walreceiver_connect_timeout).with_context(
-                    bad_duration("walreceiver_connect_timeout", walreceiver_connect_timeout),
-                )?,
-            );
-        }
-        if let Some(lagging_wal_timeout) = &request_data.lagging_wal_timeout {
-            tenant_conf.lagging_wal_timeout = Some(
-                humantime::parse_duration(lagging_wal_timeout)
-                    .with_context(bad_duration("lagging_wal_timeout", lagging_wal_timeout))?,
-            );
-        }
-        tenant_conf.max_lsn_wal_lag = request_data.max_lsn_wal_lag;
-        tenant_conf.trace_read_requests = request_data.trace_read_requests;
-
-        tenant_conf.checkpoint_distance = request_data.checkpoint_distance;
-        if let Some(checkpoint_timeout) = &request_data.checkpoint_timeout {
-            tenant_conf.checkpoint_timeout = Some(
-                humantime::parse_duration(checkpoint_timeout)
-                    .with_context(bad_duration("checkpoint_timeout", checkpoint_timeout))?,
-            );
-        }
-        tenant_conf.compaction_target_size = request_data.compaction_target_size;
-        tenant_conf.compaction_threshold = request_data.compaction_threshold;
-
-        if let Some(compaction_period) = &request_data.compaction_period {
-            tenant_conf.compaction_period = Some(
-                humantime::parse_duration(compaction_period)
-                    .with_context(bad_duration("compaction_period", compaction_period))?,
-            );
-        }
-
-        if let Some(eviction_policy) = &request_data.eviction_policy {
-            tenant_conf.eviction_policy = Some(
-                serde::Deserialize::deserialize(eviction_policy)
-                    .context("parse field `eviction_policy`")?,
-            );
-        }
-
-        tenant_conf.min_resident_size_override = request_data.min_resident_size_override;
-
-        if let Some(evictions_low_residence_duration_metric_threshold) =
-            &request_data.evictions_low_residence_duration_metric_threshold
-        {
-            tenant_conf.evictions_low_residence_duration_metric_threshold = Some(
-                humantime::parse_duration(evictions_low_residence_duration_metric_threshold)
-                    .with_context(bad_duration(
-                        "evictions_low_residence_duration_metric_threshold",
-                        evictions_low_residence_duration_metric_threshold,
-                    ))?,
-            );
-        }
-
-        Ok(tenant_conf)
+        Self::from_request(
+            request_data.checkpoint_distance,
+            &request_data.checkpoint_timeout,
+            request_data.compaction_target_size,
+            &request_data.compaction_period,
+            request_data.compaction_threshold,
+            request_data.gc_horizon,
+            &request_data.gc_period,
+            request_data.image_creation_threshold,
+            &request_data.pitr_interval,
+            &request_data.walreceiver_connect_timeout,
+            &request_data.lagging_wal_timeout,
+            request_data.max_lsn_wal_lag,
+            request_data.trace_read_requests,
+            &request_data.eviction_policy,
+            request_data.min_resident_size_override,
+            &request_data.evictions_low_residence_duration_metric_threshold,
+        )
     }
 }
 

pageserver/src/tenant/mgr.rs

@@ -186,10 +186,20 @@ pub fn schedule_local_tenant_processing(
     let tenant = if conf.tenant_attaching_mark_file_path(&tenant_id).exists() {
         info!("tenant {tenant_id} has attaching mark file, resuming its attach operation");
         if let Some(remote_storage) = remote_storage {
-            Tenant::spawn_attach(conf, tenant_id, remote_storage, ctx)
+            match Tenant::spawn_attach(conf, tenant_id, remote_storage, ctx) {
+                Ok(tenant) => tenant,
+                Err(e) => {
+                    error!("Failed to spawn_attach tenant {tenant_id}, reason: {e:#}");
+                    Tenant::create_broken_tenant(conf, tenant_id, format!("{e:#}"))
+                }
+            }
         } else {
             warn!("tenant {tenant_id} has attaching mark file, but pageserver has no remote storage configured");
-            Tenant::create_broken_tenant(conf, tenant_id)
+            Tenant::create_broken_tenant(
+                conf,
+                tenant_id,
+                "attaching mark file present but no remote storage configured".to_string(),
+            )
         }
     } else {
         info!("tenant {tenant_id} is assumed to be loadable, starting load operation");
@@ -466,7 +476,8 @@ pub async fn attach_tenant(
             "Cannot attach tenant {tenant_id}, local tenant directory already exists"
         );
 
-        let tenant = Tenant::spawn_attach(conf, tenant_id, remote_storage, ctx);
+        let tenant =
+            Tenant::spawn_attach(conf, tenant_id, remote_storage, ctx).context("spawn_attach")?;
         vacant_entry.insert(tenant);
         Ok(())
     })

pageserver/src/tenant/size.rs

@@ -11,7 +11,7 @@ use tokio_util::sync::CancellationToken;
 use crate::context::RequestContext;
 use crate::pgdatadir_mapping::CalculateLogicalSizeError;
 
-use super::Tenant;
+use super::{LogicalSizeCalculationCause, Tenant};
 use crate::tenant::Timeline;
 use utils::id::TimelineId;
 use utils::lsn::Lsn;
@@ -126,6 +126,7 @@ pub(super) async fn gather_inputs(
     limit: &Arc<Semaphore>,
     max_retention_period: Option<u64>,
     logical_size_cache: &mut HashMap<(TimelineId, Lsn), u64>,
+    cause: LogicalSizeCalculationCause,
     ctx: &RequestContext,
 ) -> anyhow::Result<ModelInputs> {
     // refresh is needed to update gc related pitr_cutoff and horizon_cutoff
@@ -318,7 +319,15 @@ pub(super) async fn gather_inputs(
 
     // We left the 'size' field empty in all of the Segments so far.
     // Now find logical sizes for all of the points that might need or benefit from them.
-    fill_logical_sizes(&timelines, &mut segments, limit, logical_size_cache, ctx).await?;
+    fill_logical_sizes(
+        &timelines,
+        &mut segments,
+        limit,
+        logical_size_cache,
+        cause,
+        ctx,
+    )
+    .await?;
 
     Ok(ModelInputs {
         segments,
@@ -336,6 +345,7 @@ async fn fill_logical_sizes(
     segments: &mut [SegmentMeta],
     limit: &Arc<Semaphore>,
     logical_size_cache: &mut HashMap<(TimelineId, Lsn), u64>,
+    cause: LogicalSizeCalculationCause,
     ctx: &RequestContext,
 ) -> anyhow::Result<()> {
     let timeline_hash: HashMap<TimelineId, Arc<Timeline>> = HashMap::from_iter(
@@ -373,13 +383,17 @@ async fn fill_logical_sizes(
                 let timeline = Arc::clone(timeline_hash.get(&timeline_id).unwrap());
                 let parallel_size_calcs = Arc::clone(limit);
                 let ctx = ctx.attached_child();
-                joinset.spawn(calculate_logical_size(
-                    parallel_size_calcs,
-                    timeline,
-                    lsn,
-                    ctx,
-                    cancel.child_token(),
-                ));
+                joinset.spawn(
+                    calculate_logical_size(
+                        parallel_size_calcs,
+                        timeline,
+                        lsn,
+                        cause,
+                        ctx,
+                        cancel.child_token(),
+                    )
+                    .in_current_span(),
+                );
             }
             e.insert(cached_size);
         }
@@ -482,6 +496,7 @@ async fn calculate_logical_size(
     limit: Arc<tokio::sync::Semaphore>,
     timeline: Arc<crate::tenant::Timeline>,
     lsn: utils::lsn::Lsn,
+    cause: LogicalSizeCalculationCause,
     ctx: RequestContext,
     cancel: CancellationToken,
 ) -> Result<TimelineAtLsnSizeResult, RecvError> {
@@ -490,7 +505,7 @@ async fn calculate_logical_size(
         .expect("global semaphore should not had been closed");
 
     let size_res = timeline
-        .spawn_ondemand_logical_size_calculation(lsn, ctx, cancel)
+        .spawn_ondemand_logical_size_calculation(lsn, cause, ctx, cancel)
         .instrument(info_span!("spawn_ondemand_logical_size_calculation"))
         .await?;
     Ok(TimelineAtLsnSizeResult(timeline, lsn, size_res))

pageserver/src/tenant/tasks.rs

@@ -259,6 +259,7 @@ pub(crate) async fn random_init_delay(
     }
 }
 
+/// Attention: the `task` and `period` beocme labels of a pageserver-wide prometheus metric.
 pub(crate) fn warn_when_period_overrun(elapsed: Duration, period: Duration, task: &str) {
     // Duration::ZERO will happen because it's the "disable [bgtask]" value.
     if elapsed >= period && period != Duration::ZERO {
@@ -271,5 +272,8 @@ pub(crate) fn warn_when_period_overrun(elapsed: Duration, period: Duration, task
             task,
             "task iteration took longer than the configured period"
         );
+        crate::metrics::BACKGROUND_LOOP_PERIOD_OVERRUN_COUNT
+            .with_label_values(&[task, &format!("{}", period.as_secs())])
+            .inc();
     }
 }

pageserver/src/tenant/timeline.rs

@@ -396,6 +396,9 @@ pub enum PageReconstructError {
     /// The operation was cancelled
     Cancelled,
 
+    /// The ancestor of this is being stopped
+    AncestorStopping(TimelineId),
+
     /// An error happened replaying WAL records
     #[error(transparent)]
     WalRedo(#[from] crate::walredo::WalRedoError),
@@ -414,6 +417,9 @@ impl std::fmt::Debug for PageReconstructError {
                 )
             }
             Self::Cancelled => write!(f, "cancelled"),
+            Self::AncestorStopping(timeline_id) => {
+                write!(f, "ancestor timeline {timeline_id} is being stopped")
+            }
             Self::WalRedo(err) => err.fmt(f),
         }
     }
@@ -432,11 +438,22 @@ impl std::fmt::Display for PageReconstructError {
                 )
             }
             Self::Cancelled => write!(f, "cancelled"),
+            Self::AncestorStopping(timeline_id) => {
+                write!(f, "ancestor timeline {timeline_id} is being stopped")
+            }
             Self::WalRedo(err) => err.fmt(f),
         }
     }
 }
 
+#[derive(Clone, Copy)]
+pub enum LogicalSizeCalculationCause {
+    Initial,
+    ConsumptionMetricsSyntheticSize,
+    EvictionTaskImitation,
+    TenantSizeHandler,
+}
+
 /// Public interface functions
 impl Timeline {
     /// Get the LSN where this branch was created
@@ -926,6 +943,31 @@ impl Timeline {
         self.state.subscribe()
     }
 
+    pub async fn wait_to_become_active(
+        &self,
+        _ctx: &RequestContext, /* Prepare for use by cancellation */
+    ) -> Result<(), TimelineState> {
+        let mut receiver = self.state.subscribe();
+        loop {
+            let current_state = *receiver.borrow_and_update();
+            match current_state {
+                TimelineState::Loading => {
+                    receiver
+                        .changed()
+                        .await
+                        .expect("holding a reference to self");
+                }
+                TimelineState::Active { .. } => {
+                    return Ok(());
+                }
+                TimelineState::Broken { .. } | TimelineState::Stopping => {
+                    // There's no chance the timeline can transition back into ::Active
+                    return Err(current_state);
+                }
+            }
+        }
+    }
+
     pub fn layer_map_info(&self, reset: LayerAccessStatsReset) -> LayerMapInfo {
         let layer_map = self.layers.read().unwrap();
         let mut in_memory_layers = Vec::with_capacity(layer_map.frozen_layers.len() + 1);
@@ -1839,18 +1881,31 @@ impl Timeline {
                 // to spawn_ondemand_logical_size_calculation.
                 let cancel = CancellationToken::new();
                 let calculated_size = match self_clone
-                    .logical_size_calculation_task(lsn, &background_ctx, cancel)
+                    .logical_size_calculation_task(lsn, LogicalSizeCalculationCause::Initial, &background_ctx, cancel)
                     .await
                 {
                     Ok(s) => s,
                     Err(CalculateLogicalSizeError::Cancelled) => {
                         // Don't make noise, this is a common task.
-                        // In the unlikely case that there ihs another call to this function, we'll retry
+                        // In the unlikely case that there is another call to this function, we'll retry
                         // because initial_logical_size is still None.
                         info!("initial size calculation cancelled, likely timeline delete / tenant detach");
                         return Ok(());
                     }
-                    x @ Err(_) => x.context("Failed to calculate logical size")?,
+                    Err(CalculateLogicalSizeError::Other(err)) => {
+                        if let Some(e @ PageReconstructError::AncestorStopping(_)) =
+                            err.root_cause().downcast_ref()
+                        {
+                            // This can happen if the timeline parent timeline switches to
+                            // Stopping state while we're still calculating the initial
+                            // timeline size for the child, for example if the tenant is
+                            // being detached or the pageserver is shut down. Like with
+                            // CalculateLogicalSizeError::Cancelled, don't make noise.
+                            info!("initial size calculation failed because the timeline or its ancestor is Stopping, likely because the tenant is being detached: {e:#}");
+                            return Ok(());
+                        }
+                        return Err(err.context("Failed to calculate logical size"));
+                    }
                 };
 
                 // we cannot query current_logical_size.current_size() to know the current
@@ -1886,13 +1941,14 @@ impl Timeline {
                 // so that we prevent future callers from spawning this task
                 permit.forget();
                 Ok(())
-            },
+            }.in_current_span(),
         );
     }
 
     pub fn spawn_ondemand_logical_size_calculation(
         self: &Arc<Self>,
         lsn: Lsn,
+        cause: LogicalSizeCalculationCause,
         ctx: RequestContext,
         cancel: CancellationToken,
     ) -> oneshot::Receiver<Result<u64, CalculateLogicalSizeError>> {
@@ -1915,22 +1971,26 @@ impl Timeline {
             false,
             async move {
                 let res = self_clone
-                    .logical_size_calculation_task(lsn, &ctx, cancel)
+                    .logical_size_calculation_task(lsn, cause, &ctx, cancel)
                     .await;
                 let _ = sender.send(res).ok();
                 Ok(()) // Receiver is responsible for handling errors
-            },
+            }
+            .in_current_span(),
         );
         receiver
     }
 
-    #[instrument(skip_all, fields(tenant = %self.tenant_id, timeline = %self.timeline_id))]
+    #[instrument(skip_all)]
     async fn logical_size_calculation_task(
         self: &Arc<Self>,
         lsn: Lsn,
+        cause: LogicalSizeCalculationCause,
         ctx: &RequestContext,
         cancel: CancellationToken,
     ) -> Result<u64, CalculateLogicalSizeError> {
+        debug_assert_current_span_has_tenant_and_timeline_id();
+
         let mut timeline_state_updates = self.subscribe_for_state_updates();
         let self_calculation = Arc::clone(self);
 
@@ -1938,7 +1998,7 @@ impl Timeline {
             let cancel = cancel.child_token();
             let ctx = ctx.attached_child();
             self_calculation
-                .calculate_logical_size(lsn, cancel, &ctx)
+                .calculate_logical_size(lsn, cause, cancel, &ctx)
                 .await
         });
         let timeline_state_cancellation = async {
@@ -1993,6 +2053,7 @@ impl Timeline {
     pub async fn calculate_logical_size(
         &self,
         up_to_lsn: Lsn,
+        cause: LogicalSizeCalculationCause,
         cancel: CancellationToken,
         ctx: &RequestContext,
     ) -> Result<u64, CalculateLogicalSizeError> {
@@ -2026,7 +2087,15 @@ impl Timeline {
         if let Some(size) = self.current_logical_size.initialized_size(up_to_lsn) {
             return Ok(size);
         }
-        let timer = self.metrics.logical_size_histo.start_timer();
+        let storage_time_metrics = match cause {
+            LogicalSizeCalculationCause::Initial
+            | LogicalSizeCalculationCause::ConsumptionMetricsSyntheticSize
+            | LogicalSizeCalculationCause::TenantSizeHandler => &self.metrics.logical_size_histo,
+            LogicalSizeCalculationCause::EvictionTaskImitation => {
+                &self.metrics.imitate_logical_size_histo
+            }
+        };
+        let timer = storage_time_metrics.start_timer();
         let logical_size = self
             .get_current_logical_size_non_incremental(up_to_lsn, cancel, ctx)
             .await?;
@@ -2218,6 +2287,46 @@ impl Timeline {
                     Ok(timeline) => timeline,
                     Err(e) => return Err(PageReconstructError::from(e)),
                 };
+
+                // It's possible that the ancestor timeline isn't active yet, or
+                // is active but hasn't yet caught up to the branch point. Wait
+                // for it.
+                //
+                // This cannot happen while the pageserver is running normally,
+                // because you cannot create a branch from a point that isn't
+                // present in the pageserver yet. However, we don't wait for the
+                // branch point to be uploaded to cloud storage before creating
+                // a branch. I.e., the branch LSN need not be remote consistent
+                // for the branching operation to succeed.
+                //
+                // Hence, if we try to load a tenant in such a state where
+                // 1. the existence of the branch was persisted (in IndexPart and/or locally)
+                // 2. but the ancestor state is behind branch_lsn because it was not yet persisted
+                // then we will need to wait for the ancestor timeline to
+                // re-stream WAL up to branch_lsn before we access it.
+                //
+                // How can a tenant get in such a state?
+                // - ungraceful pageserver process exit
+                // - detach+attach => this is a bug, https://github.com/neondatabase/neon/issues/4219
+                //
+                // NB: this could be avoided by requiring
+                //   branch_lsn >= remote_consistent_lsn
+                // during branch creation.
+                match ancestor.wait_to_become_active(ctx).await {
+                    Ok(()) => {}
+                    Err(state) if state == TimelineState::Stopping => {
+                        return Err(PageReconstructError::AncestorStopping(ancestor.timeline_id));
+                    }
+                    Err(state) => {
+                        return Err(PageReconstructError::Other(anyhow::anyhow!(
+                            "Timeline {} will not become active. Current state: {:?}",
+                            ancestor.timeline_id,
+                            &state,
+                        )));
+                    }
+                }
+                ancestor.wait_lsn(timeline.ancestor_lsn, ctx).await?;
+
                 timeline_owned = ancestor;
                 timeline = &*timeline_owned;
                 prev_lsn = Lsn(u64::MAX);

pageserver/src/tenant/timeline/eviction_task.rs

@@ -30,7 +30,7 @@ use crate::{
     tenant::{
         config::{EvictionPolicy, EvictionPolicyLayerAccessThreshold},
         storage_layer::PersistentLayer,
-        Tenant,
+        LogicalSizeCalculationCause, Tenant,
     },
 };
 
@@ -120,6 +120,13 @@ impl Timeline {
                 }
                 let elapsed = start.elapsed();
                 crate::tenant::tasks::warn_when_period_overrun(elapsed, p.period, "eviction");
+                crate::metrics::EVICTION_ITERATION_DURATION
+                    .get_metric_with_label_values(&[
+                        &format!("{}", p.period.as_secs()),
+                        &format!("{}", p.threshold.as_secs()),
+                    ])
+                    .unwrap()
+                    .observe(elapsed.as_secs_f64());
                 ControlFlow::Continue(start + p.period)
             }
         }
@@ -335,7 +342,12 @@ impl Timeline {
 
         // imitiate on-restart initial logical size
         let size = self
-            .calculate_logical_size(lsn, cancel.clone(), ctx)
+            .calculate_logical_size(
+                lsn,
+                LogicalSizeCalculationCause::EvictionTaskImitation,
+                cancel.clone(),
+                ctx,
+            )
             .instrument(info_span!("calculate_logical_size"))
             .await;
 
@@ -407,9 +419,15 @@ impl Timeline {
             .inner();
 
         let mut throwaway_cache = HashMap::new();
-        let gather =
-            crate::tenant::size::gather_inputs(tenant, limit, None, &mut throwaway_cache, ctx)
-                .instrument(info_span!("gather_inputs"));
+        let gather = crate::tenant::size::gather_inputs(
+            tenant,
+            limit,
+            None,
+            &mut throwaway_cache,
+            LogicalSizeCalculationCause::EvictionTaskImitation,
+            ctx,
+        )
+        .instrument(info_span!("gather_inputs"));
 
         tokio::select! {
             _ = cancel.cancelled() => {}

pageserver/src/walingest.rs

@@ -305,6 +305,15 @@ impl<'a> WalIngest<'a> {
                     self.checkpoint_modified = true;
                 }
             }
+        } else if decoded.xl_rmid == pg_constants::RM_LOGICALMSG_ID {
+            let info = decoded.xl_info & pg_constants::XLR_RMGR_INFO_MASK;
+            if info == pg_constants::XLOG_LOGICAL_MESSAGE {
+                // This is a convenient way to make the WAL ingestion pause at
+                // particular point in the WAL. For more fine-grained control,
+                // we could peek into the message and only pause if it contains
+                // a particular string, for example, but this is enough for now.
+                utils::failpoint_sleep_millis_async!("wal-ingest-logical-message-sleep");
+            }
         }
 
         // Iterate through all the blocks that the record modifies, and

pgxn/neon/libpagestore.c

@@ -192,8 +192,9 @@ retry:
 		{
 			if (!PQconsumeInput(pageserver_conn))
 			{
-				neon_log(LOG, "could not get response from pageserver: %s",
-						 PQerrorMessage(pageserver_conn));
+				char	   *msg = pchomp(PQerrorMessage(pageserver_conn));
+				neon_log(LOG, "could not get response from pageserver: %s", msg);
+				pfree(msg);
 				return -1;
 			}
 		}
@@ -343,7 +344,7 @@ pageserver_receive(void)
 			resp = NULL;
 		}
 		else if (rc == -2)
-			neon_log(ERROR, "could not read COPY data: %s", PQerrorMessage(pageserver_conn));
+			neon_log(ERROR, "could not read COPY data: %s", pchomp(PQerrorMessage(pageserver_conn)));
 		else
 			neon_log(ERROR, "unexpected PQgetCopyData return value: %d", rc);
 	}
@@ -367,7 +368,7 @@ pageserver_flush(void)
 	}
 	else if (PQflush(pageserver_conn))
 	{
-		char	   *msg = PQerrorMessage(pageserver_conn);
+		char	   *msg = pchomp(PQerrorMessage(pageserver_conn));
 
 		pageserver_disconnect();
 		neon_log(ERROR, "failed to flush page requests: %s", msg);

pgxn/neon/pagestore_client.h

@@ -52,7 +52,7 @@ typedef struct
 #define NEON_TAG "[NEON_SMGR] "
 #define neon_log(tag, fmt, ...) ereport(tag,                                  \
 										(errmsg(NEON_TAG fmt, ##__VA_ARGS__), \
-										 errhidestmt(true), errhidecontext(true), internalerrposition(0)))
+										 errhidestmt(true), errhidecontext(true), errposition(0), internalerrposition(0)))
 
 /*
  * supertype of all the Neon*Request structs below

proxy/src/auth.rs

@@ -7,6 +7,7 @@ mod credentials;
 pub use credentials::ClientCredentials;
 
 mod password_hack;
+pub use password_hack::parse_endpoint_param;
 use password_hack::PasswordHackPayload;
 
 mod flow;
@@ -44,10 +45,10 @@ pub enum AuthErrorImpl {
     #[error(
         "Endpoint ID is not specified. \
         Either please upgrade the postgres client library (libpq) for SNI support \
-        or pass the endpoint ID (first part of the domain name) as a parameter: '?options=project%3D<endpoint-id>'. \
+        or pass the endpoint ID (first part of the domain name) as a parameter: '?options=endpoint%3D<endpoint-id>'. \
         See more at https://neon.tech/sni"
     )]
-    MissingProjectName,
+    MissingEndpointName,
 
     #[error("password authentication failed for user '{0}'")]
     AuthFailed(Box<str>),
@@ -88,7 +89,7 @@ impl UserFacingError for AuthError {
             AuthFailed(_) => self.to_string(),
             BadAuthMethod(_) => self.to_string(),
             MalformedPassword(_) => self.to_string(),
-            MissingProjectName => self.to_string(),
+            MissingEndpointName => self.to_string(),
             Io(_) => "Internal error".to_string(),
         }
     }

proxy/src/auth/backend/hacks.rs

@@ -52,8 +52,8 @@ pub async fn password_hack(
         .authenticate()
         .await?;
 
-    info!(project = &payload.project, "received missing parameter");
-    creds.project = Some(payload.project);
+    info!(project = &payload.endpoint, "received missing parameter");
+    creds.project = Some(payload.endpoint);
 
     let mut node = api.wake_compute(extra, creds).await?;
     node.config.password(payload.password);

proxy/src/auth/credentials.rs

@@ -1,6 +1,7 @@
 //! User credentials used in authentication.
 
-use crate::error::UserFacingError;
+use crate::{auth::password_hack::parse_endpoint_param, error::UserFacingError};
+use itertools::Itertools;
 use pq_proto::StartupMessageParams;
 use std::collections::HashSet;
 use thiserror::Error;
@@ -61,7 +62,15 @@ impl<'a> ClientCredentials<'a> {
         // Project name might be passed via PG's command-line options.
         let project_option = params
             .options_raw()
-            .and_then(|mut options| options.find_map(|opt| opt.strip_prefix("project=")))
+            .and_then(|options| {
+                // We support both `project` (deprecated) and `endpoint` options for backward compatibility.
+                // However, if both are present, we don't exactly know which one to use.
+                // Therefore we require that only one of them is present.
+                options
+                    .filter_map(parse_endpoint_param)
+                    .at_most_one()
+                    .ok()?
+            })
             .map(|name| name.to_string());
 
         let project_from_domain = if let Some(sni_str) = sni {
@@ -177,6 +186,51 @@ mod tests {
         Ok(())
     }
 
+    #[test]
+    fn parse_endpoint_from_options() -> anyhow::Result<()> {
+        let options = StartupMessageParams::new([
+            ("user", "john_doe"),
+            ("options", "-ckey=1 endpoint=bar -c geqo=off"),
+        ]);
+
+        let creds = ClientCredentials::parse(&options, None, None)?;
+        assert_eq!(creds.user, "john_doe");
+        assert_eq!(creds.project.as_deref(), Some("bar"));
+
+        Ok(())
+    }
+
+    #[test]
+    fn parse_three_endpoints_from_options() -> anyhow::Result<()> {
+        let options = StartupMessageParams::new([
+            ("user", "john_doe"),
+            (
+                "options",
+                "-ckey=1 endpoint=one endpoint=two endpoint=three -c geqo=off",
+            ),
+        ]);
+
+        let creds = ClientCredentials::parse(&options, None, None)?;
+        assert_eq!(creds.user, "john_doe");
+        assert!(creds.project.is_none());
+
+        Ok(())
+    }
+
+    #[test]
+    fn parse_when_endpoint_and_project_are_in_options() -> anyhow::Result<()> {
+        let options = StartupMessageParams::new([
+            ("user", "john_doe"),
+            ("options", "-ckey=1 endpoint=bar project=foo -c geqo=off"),
+        ]);
+
+        let creds = ClientCredentials::parse(&options, None, None)?;
+        assert_eq!(creds.user, "john_doe");
+        assert!(creds.project.is_none());
+
+        Ok(())
+    }
+
     #[test]
     fn parse_projects_identical() -> anyhow::Result<()> {
         let options = StartupMessageParams::new([("user", "john_doe"), ("options", "project=baz")]);

proxy/src/auth/flow.rs

@@ -91,7 +91,7 @@ impl<S: AsyncRead + AsyncWrite + Unpin> AuthFlow<'_, S, PasswordHack> {
             // the user neither enabled SNI nor resorted to any other method
             // for passing the project name we rely on. We should show them
             // the most helpful error message and point to the documentation.
-            .ok_or(AuthErrorImpl::MissingProjectName)?;
+            .ok_or(AuthErrorImpl::MissingEndpointName)?;
 
         Ok(payload)
     }

proxy/src/auth/password_hack.rs

@@ -6,27 +6,55 @@
 use bstr::ByteSlice;
 
 pub struct PasswordHackPayload {
-    pub project: String,
+    pub endpoint: String,
     pub password: Vec<u8>,
 }
 
 impl PasswordHackPayload {
     pub fn parse(bytes: &[u8]) -> Option<Self> {
         // The format is `project=<utf-8>;<password-bytes>`.
-        let mut iter = bytes.strip_prefix(b"project=")?.splitn_str(2, ";");
-        let project = iter.next()?.to_str().ok()?.to_owned();
+        let mut iter = bytes.splitn_str(2, ";");
+        let endpoint = iter.next()?.to_str().ok()?;
+        let endpoint = parse_endpoint_param(endpoint)?.to_owned();
         let password = iter.next()?.to_owned();
 
-        Some(Self { project, password })
+        Some(Self { endpoint, password })
     }
 }
 
+pub fn parse_endpoint_param(bytes: &str) -> Option<&str> {
+    bytes
+        .strip_prefix("project=")
+        .or_else(|| bytes.strip_prefix("endpoint="))
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
 
     #[test]
-    fn parse_password_hack_payload() {
+    fn parse_endpoint_param_fn() {
+        let input = "";
+        assert!(parse_endpoint_param(input).is_none());
+
+        let input = "project=";
+        assert_eq!(parse_endpoint_param(input), Some(""));
+
+        let input = "project=foobar";
+        assert_eq!(parse_endpoint_param(input), Some("foobar"));
+
+        let input = "endpoint=";
+        assert_eq!(parse_endpoint_param(input), Some(""));
+
+        let input = "endpoint=foobar";
+        assert_eq!(parse_endpoint_param(input), Some("foobar"));
+
+        let input = "other_option=foobar";
+        assert!(parse_endpoint_param(input).is_none());
+    }
+
+    #[test]
+    fn parse_password_hack_payload_project() {
         let bytes = b"";
         assert!(PasswordHackPayload::parse(bytes).is_none());
 
@@ -34,13 +62,33 @@ mod tests {
         assert!(PasswordHackPayload::parse(bytes).is_none());
 
         let bytes = b"project=;";
-        let payload = PasswordHackPayload::parse(bytes).expect("parsing failed");
-        assert_eq!(payload.project, "");
+        let payload: PasswordHackPayload =
+            PasswordHackPayload::parse(bytes).expect("parsing failed");
+        assert_eq!(payload.endpoint, "");
         assert_eq!(payload.password, b"");
 
         let bytes = b"project=foobar;pass;word";
         let payload = PasswordHackPayload::parse(bytes).expect("parsing failed");
-        assert_eq!(payload.project, "foobar");
+        assert_eq!(payload.endpoint, "foobar");
+        assert_eq!(payload.password, b"pass;word");
+    }
+
+    #[test]
+    fn parse_password_hack_payload_endpoint() {
+        let bytes = b"";
+        assert!(PasswordHackPayload::parse(bytes).is_none());
+
+        let bytes = b"endpoint=";
+        assert!(PasswordHackPayload::parse(bytes).is_none());
+
+        let bytes = b"endpoint=;";
+        let payload = PasswordHackPayload::parse(bytes).expect("parsing failed");
+        assert_eq!(payload.endpoint, "");
+        assert_eq!(payload.password, b"");
+
+        let bytes = b"endpoint=foobar;pass;word";
+        let payload = PasswordHackPayload::parse(bytes).expect("parsing failed");
+        assert_eq!(payload.endpoint, "foobar");
         assert_eq!(payload.password, b"pass;word");
     }
 }

proxy/src/compute.rs

@@ -1,4 +1,4 @@
-use crate::{cancellation::CancelClosure, error::UserFacingError};
+use crate::{auth::parse_endpoint_param, cancellation::CancelClosure, error::UserFacingError};
 use futures::{FutureExt, TryFutureExt};
 use itertools::Itertools;
 use pq_proto::StartupMessageParams;
@@ -279,7 +279,7 @@ fn filtered_options(params: &StartupMessageParams) -> Option<String> {
     #[allow(unstable_name_collisions)]
     let options: String = params
         .options_raw()?
-        .filter(|opt| !opt.starts_with("project="))
+        .filter(|opt| parse_endpoint_param(opt).is_none())
         .intersperse(" ") // TODO: use impl from std once it's stabilized
         .collect();
 

scripts/flaky_tests.py

@@ -21,6 +21,7 @@ FLAKY_TESTS_QUERY = """
                 jsonb_array_elements(jsonb_array_elements(data -> 'children') -> 'children') -> 'name' as suite,
                 jsonb_array_elements(jsonb_array_elements(jsonb_array_elements(data -> 'children') -> 'children') -> 'children') -> 'name' as test,
                 jsonb_array_elements(jsonb_array_elements(jsonb_array_elements(data -> 'children') -> 'children') -> 'children') -> 'status' as status,
+                jsonb_array_elements(jsonb_array_elements(jsonb_array_elements(data -> 'children') -> 'children') -> 'children') -> 'retriesStatusChange' as retries_status_change,
                 to_timestamp((jsonb_array_elements(jsonb_array_elements(jsonb_array_elements(data -> 'children') -> 'children') -> 'children') -> 'time' -> 'start')::bigint / 1000)::date as timestamp
             FROM
                 regress_test_results
@@ -29,7 +30,7 @@ FLAKY_TESTS_QUERY = """
         ) data
     WHERE
         timestamp > CURRENT_DATE - INTERVAL '%s' day
-        AND status::text IN ('"failed"', '"broken"')
+        AND (status::text IN ('"failed"', '"broken"') OR retries_status_change::boolean)
     ;
 """
 

scripts/pr-comment-test-report.js

@@ -1,6 +1,5 @@
 //
 // The script parses Allure reports and posts a comment with a summary of the test results to the PR.
-// It accepts an array of items and creates a comment with a summary for each one (for "release" and "debug", together or separately if any of them failed to be generated).
 //
 // The comment is updated on each run with the latest results.
 //
@@ -13,19 +12,37 @@
 //         github,
 //         context,
 //         fetch,
-//         reports: [{...}, ...], // each report is expected to have "buildType", "reportUrl", and "jsonUrl" properties
+//         report: {
+//           reportUrl: "...",
+//           reportJsonUrl: "...",
+//         },
 //       })
 //
 
-module.exports = async ({ github, context, fetch, reports }) => {
+// Analog of Python's defaultdict.
+//
+// const dm = new DefaultMap(() => new DefaultMap(() => []))
+// dm["firstKey"]["secondKey"].push("value")
+//
+class DefaultMap extends Map {
+    constructor(getDefaultValue) {
+        return new Proxy({}, {
+            get: (target, name) => name in target ? target[name] : (target[name] = getDefaultValue(name))
+        })
+    }
+}
+
+module.exports = async ({ github, context, fetch, report }) => {
     // Marker to find the comment in the subsequent runs
     const startMarker = `<!--AUTOMATIC COMMENT START #${context.payload.number}-->`
+    // Let users know that the comment is updated automatically
+    const autoupdateNotice = `<div align="right"><sub>The comment gets automatically updated with the latest test results :recycle:</sub></div>`
     // GitHub bot id taken from (https://api.github.com/users/github-actions[bot])
     const githubActionsBotId = 41898282
     // The latest commit in the PR URL
     const commitUrl = `${context.serverUrl}/${context.repo.owner}/${context.repo.repo}/pull/${context.payload.number}/commits/${context.payload.pull_request.head.sha}`
     // Commend body itself
-    let commentBody = `${startMarker}\n### Test results for ${commitUrl}:\n___\n`
+    let commentBody = `${startMarker}\n`
 
     // Common parameters for GitHub API requests
     const ownerRepoParams = {
@@ -33,76 +50,119 @@ module.exports = async ({ github, context, fetch, reports }) => {
         repo: context.repo.repo,
     }
 
-    for (const report of reports) {
-        const {buildType, reportUrl, jsonUrl} = report
+    const {reportUrl, reportJsonUrl} = report
 
-        if (!reportUrl || !jsonUrl) {
-            commentBody += `#### ${buildType} build: no tests were run or test report is not available\n`
-            continue
-        }
+    if (!reportUrl || !reportJsonUrl) {
+        commentBody += `#### No tests were run or test report is not available\n`
+        commentBody += autoupdateNotice
+        return
+    }
 
-        const suites = await (await fetch(jsonUrl)).json()
+    const suites = await (await fetch(reportJsonUrl)).json()
 
-        // Allure distinguishes "failed" (with an assertion error) and "broken" (with any other error) tests.
-        // For this report it's ok to treat them in the same way (as failed).
-        failedTests = []
-        passedTests = []
-        skippedTests = []
+    // Allure distinguishes "failed" (with an assertion error) and "broken" (with any other error) tests.
+    // For this report it's ok to treat them in the same way (as failed).
+    const failedTests = new DefaultMap(() => new DefaultMap(() => []))
+    const passedTests = new DefaultMap(() => new DefaultMap(() => []))
+    const skippedTests = new DefaultMap(() => new DefaultMap(() => []))
+    const retriedTests = new DefaultMap(() => new DefaultMap(() => []))
+    const flakyTests = new DefaultMap(() => new DefaultMap(() => []))
 
-        retriedTests = []
-        retriedStatusChangedTests = []
+    let failedTestsCount = 0
+    let passedTestsCount = 0
+    let skippedTestsCount = 0
+    let flakyTestsCount = 0
 
-        for (const parentSuite of suites.children) {
-            for (const suite of parentSuite.children) {
-                for (const test of suite.children) {
-                    pytestName = `${parentSuite.name.replace(".", "/")}/${suite.name}.py::${test.name}`
-                    test.pytestName = pytestName
+    const pgVersions = new Set()
+    const buildTypes = new Set()
 
-                    if (test.status === "passed") {
-                        passedTests.push(test);
-                    } else if (test.status === "failed" || test.status === "broken") {
-                        failedTests.push(test);
-                    } else if (test.status === "skipped") {
-                        skippedTests.push(test);
-                    }
+    for (const parentSuite of suites.children) {
+        for (const suite of parentSuite.children) {
+            for (const test of suite.children) {
+                let buildType, pgVersion
+                const match = test.name.match(/[\[-](?<buildType>debug|release)-pg(?<pgVersion>\d+)[-\]]/)?.groups
+                if (match) {
+                    ({buildType, pgVersion} = match)
+                } else {
+                    // It's ok, we embed BUILD_TYPE and Postgres Version into the test name only for regress suite and do not for other suites (like performance).
+                    console.info(`Cannot get BUILD_TYPE and Postgres Version from test name: "${test.name}", defaulting to "release" and "14"`)
 
-                    if (test.retriesCount > 0) {
-                        retriedTests.push(test);
+                    buildType = "release"
+                    pgVersion = "14"
+                }
 
-                        if (test.retriedStatusChangedTests) {
-                            retriedStatusChangedTests.push(test);
-                        }
+                pgVersions.add(pgVersion)
+                buildTypes.add(buildType)
+
+                // Removing build type and PostgreSQL version from the test name to make it shorter
+                const testName = test.name.replace(new RegExp(`${buildType}-pg${pgVersion}-?`), "").replace("[]", "")
+                test.pytestName = `${parentSuite.name.replace(".", "/")}/${suite.name}.py::${testName}`
+
+                if (test.status === "passed") {
+                    passedTests[pgVersion][buildType].push(test)
+                    passedTestsCount += 1
+                } else if (test.status === "failed" || test.status === "broken") {
+                    failedTests[pgVersion][buildType].push(test)
+                    failedTestsCount += 1
+                } else if (test.status === "skipped") {
+                    skippedTests[pgVersion][buildType].push(test)
+                    skippedTestsCount += 1
+                }
+
+                if (test.retriesCount > 0) {
+                    retriedTests[pgVersion][buildType].push(test)
+
+                    if (test.retriesStatusChange) {
+                        flakyTests[pgVersion][buildType].push(test)
+                        flakyTestsCount += 1
                     }
                 }
             }
         }
+    }
 
-        const totalTestsCount = failedTests.length + passedTests.length + skippedTests.length
-        commentBody += `#### ${buildType} build: ${totalTestsCount} tests run: ${passedTests.length} passed, ${failedTests.length} failed, ${skippedTests.length} skipped ([full report](${reportUrl}))\n`
-        if (failedTests.length > 0) {
-            commentBody += `Failed tests:\n`
-            for (const test of failedTests) {
-                const allureLink = `${reportUrl}#suites/${test.parentUid}/${test.uid}`
+    const totalTestsCount = failedTestsCount + passedTestsCount + skippedTestsCount
+    commentBody += `### ${totalTestsCount} tests run: ${passedTestsCount} passed, ${failedTestsCount} failed, ${skippedTestsCount} skipped ([full report](${reportUrl}) for ${commitUrl})\n___\n`
 
-                commentBody += `- [\`${test.pytestName}\`](${allureLink})`
-                if (test.retriesCount > 0) {
-                    commentBody += ` (ran [${test.retriesCount + 1} times](${allureLink}/retries))`
+    // Print test resuls from the newest to the oldest PostgreSQL version for release and debug builds.
+    for (const pgVersion of Array.from(pgVersions).sort().reverse()) {
+        for (const buildType of Array.from(buildTypes).sort().reverse()) {
+            if (failedTests[pgVersion][buildType].length > 0) {
+                commentBody += `#### PostgreSQL ${pgVersion} (${buildType} build)\n\n`
+                commentBody += `Failed tests:\n`
+                for (const test of failedTests[pgVersion][buildType]) {
+                    const allureLink = `${reportUrl}#suites/${test.parentUid}/${test.uid}`
+
+                    commentBody += `- [\`${test.pytestName}\`](${allureLink})`
+                    if (test.retriesCount > 0) {
+                        commentBody += ` (ran [${test.retriesCount + 1} times](${allureLink}/retries))`
+                    }
+                    commentBody += "\n"
                 }
                 commentBody += "\n"
             }
-            commentBody += "\n"
         }
-        if (retriedStatusChangedTests > 0) {
-            commentBody += `Flaky tests:\n`
-            for (const test of retriedStatusChangedTests) {
-                const status = test.status === "passed" ? ":white_check_mark:" : ":x:"
-                commentBody += `- ${status} [\`${test.pytestName}\`](${reportUrl}#suites/${test.parentUid}/${test.uid}/retries)\n`
-            }
-            commentBody += "\n"
-        }
-        commentBody += "___\n"
     }
 
+    if (flakyTestsCount > 0) {
+        commentBody += "<details>\n<summary>Flaky tests</summary>\n\n"
+        for (const pgVersion of Array.from(pgVersions).sort().reverse()) {
+            for (const buildType of Array.from(buildTypes).sort().reverse()) {
+                if (flakyTests[pgVersion][buildType].length > 0) {
+                    commentBody += `#### PostgreSQL ${pgVersion} (${buildType} build)\n\n`
+                    for (const test of flakyTests[pgVersion][buildType]) {
+                        const status = test.status === "passed" ? ":white_check_mark:" : ":x:"
+                        commentBody += `- ${status} [\`${test.pytestName}\`](${reportUrl}#suites/${test.parentUid}/${test.uid}/retries)\n`
+                    }
+                    commentBody += "\n"
+                }
+            }
+        }
+        commentBody += "\n</details>\n"
+    }
+
+    commentBody += autoupdateNotice
+
     const { data: comments } = await github.rest.issues.listComments({
         issue_number: context.payload.number,
         ...ownerRepoParams,

test_runner/conftest.py

@@ -1,5 +1,6 @@
 pytest_plugins = (
     "fixtures.pg_version",
+    "fixtures.allure",
     "fixtures.neon_fixtures",
     "fixtures.benchmark_fixture",
     "fixtures.pg_stats",

test_runner/fixtures/allure.py

@@ -0,0 +1,25 @@
+import os
+
+import pytest
+
+from fixtures.pg_version import DEFAULT_VERSION, PgVersion
+
+"""
+Set of utilities to make Allure report more informative.
+
+- It adds BUILD_TYPE and DEFAULT_PG_VERSION to the test names (only in test_runner/regress)
+to make tests distinguishable in Allure report.
+"""
+
+
+@pytest.fixture(scope="function", autouse=True)
+def allure_noop():
+    pass
+
+
+def pytest_generate_tests(metafunc):
+    if "test_runner/regress" in metafunc.definition._nodeid:
+        build_type = os.environ.get("BUILD_TYPE", "DEBUG").lower()
+        pg_version = PgVersion(os.environ.get("DEFAULT_PG_VERSION", DEFAULT_VERSION))
+
+        metafunc.parametrize("allure_noop", [f"{build_type}-pg{pg_version}"])

test_runner/fixtures/benchmark_fixture.py

@@ -451,13 +451,17 @@ def pytest_terminal_summary(
     revision = os.getenv("GITHUB_SHA", "local")
     platform = os.getenv("PLATFORM", "local")
 
-    terminalreporter.section("Benchmark results", "-")
+    is_header_printed = False
 
     result = []
     for test_report in terminalreporter.stats.get("passed", []):
         result_entry = []
 
         for _, recorded_property in test_report.user_properties:
+            if not is_header_printed:
+                terminalreporter.section("Benchmark results", "-")
+                is_header_printed = True
+
             terminalreporter.write(
                 "{}.{}: ".format(test_report.head_line, recorded_property["name"])
             )
@@ -485,7 +489,6 @@ def pytest_terminal_summary(
 
     out_dir = config.getoption("out_dir")
     if out_dir is None:
-        warnings.warn("no out dir provided to store performance test results")
         return
 
     if not result:

test_runner/fixtures/pageserver/http.py

@@ -272,6 +272,7 @@ class PageserverHttpClient(requests.Session):
         new_timeline_id: Optional[TimelineId] = None,
         ancestor_timeline_id: Optional[TimelineId] = None,
         ancestor_start_lsn: Optional[Lsn] = None,
+        **kwargs,
     ) -> Dict[Any, Any]:
         body: Dict[str, Any] = {
             "new_timeline_id": str(new_timeline_id) if new_timeline_id else None,
@@ -281,7 +282,9 @@ class PageserverHttpClient(requests.Session):
         if pg_version != PgVersion.NOT_SET:
             body["pg_version"] = int(pg_version)
 
-        res = self.post(f"http://localhost:{self.port}/v1/tenant/{tenant_id}/timeline", json=body)
+        res = self.post(
+            f"http://localhost:{self.port}/v1/tenant/{tenant_id}/timeline", json=body, **kwargs
+        )
         self.verbose_error(res)
         if res.status_code == 409:
             raise Exception(f"could not create timeline: already exists for id {new_timeline_id}")

test_runner/fixtures/pg_version.py

@@ -46,6 +46,20 @@ class PgVersion(str, enum.Enum):
 DEFAULT_VERSION: PgVersion = PgVersion.V14
 
 
+def skip_on_postgres(version: PgVersion, reason: str):
+    return pytest.mark.skipif(
+        PgVersion(os.environ.get("DEFAULT_PG_VERSION", DEFAULT_VERSION)) is version,
+        reason=reason,
+    )
+
+
+def xfail_on_postgres(version: PgVersion, reason: str):
+    return pytest.mark.xfail(
+        PgVersion(os.environ.get("DEFAULT_PG_VERSION", DEFAULT_VERSION)) is version,
+        reason=reason,
+    )
+
+
 def pytest_addoption(parser: Parser):
     parser.addoption(
         "--pg-version",

test_runner/regress/test_compatibility.py

@@ -16,7 +16,7 @@ from fixtures.neon_fixtures import (
 )
 from fixtures.pageserver.http import PageserverHttpClient
 from fixtures.pageserver.utils import wait_for_last_record_lsn, wait_for_upload
-from fixtures.pg_version import PgVersion
+from fixtures.pg_version import PgVersion, skip_on_postgres
 from fixtures.types import Lsn
 from pytest import FixtureRequest
 
@@ -41,12 +41,15 @@ check_ondisk_data_compatibility_if_enabled = pytest.mark.skipif(
 )
 
 
-# Note: if renaming this test, don't forget to update a reference to it in a workflow file:
-# "Upload compatibility snapshot" step in .github/actions/run-python-test-set/action.yml
-@check_ondisk_data_compatibility_if_enabled
+@skip_on_postgres(PgVersion.V15, "Compatibility tests doesn't support Postgres 15 yet")
 @pytest.mark.xdist_group("compatibility")
 @pytest.mark.order(before="test_forward_compatibility")
-def test_create_snapshot(neon_env_builder: NeonEnvBuilder, pg_bin: PgBin, test_output_dir: Path):
+def test_create_snapshot(
+    neon_env_builder: NeonEnvBuilder,
+    pg_bin: PgBin,
+    top_output_dir: Path,
+    test_output_dir: Path,
+):
     # The test doesn't really test anything
     # it creates a new snapshot for releases after we tested the current version against the previous snapshot in `test_backward_compatibility`.
     #
@@ -86,10 +89,14 @@ def test_create_snapshot(neon_env_builder: NeonEnvBuilder, pg_bin: PgBin, test_o
         sk.stop()
     env.pageserver.stop()
 
-    shutil.copytree(test_output_dir, test_output_dir / "compatibility_snapshot_pg14")
-    # Directory `test_output_dir / "compatibility_snapshot_pg14"` is uploaded to S3 in a workflow, keep the name in sync with it
+    # Directory `compatibility_snapshot_dir` is uploaded to S3 in a workflow, keep the name in sync with it
+    compatibility_snapshot_dir = top_output_dir / "compatibility_snapshot_pg14"
+    if compatibility_snapshot_dir.exists():
+        shutil.rmtree(compatibility_snapshot_dir)
+    shutil.copytree(test_output_dir, compatibility_snapshot_dir)
 
 
+@skip_on_postgres(PgVersion.V15, "Compatibility tests doesn't support Postgres 15 yet")
 @check_ondisk_data_compatibility_if_enabled
 @pytest.mark.xdist_group("compatibility")
 @pytest.mark.order(after="test_create_snapshot")
@@ -148,11 +155,13 @@ def test_backward_compatibility(
     ), "Breaking changes are allowed by ALLOW_BACKWARD_COMPATIBILITY_BREAKAGE, but the test has passed without any breakage"
 
 
+@skip_on_postgres(PgVersion.V15, "Compatibility tests doesn't support Postgres 15 yet")
 @check_ondisk_data_compatibility_if_enabled
 @pytest.mark.xdist_group("compatibility")
 @pytest.mark.order(after="test_create_snapshot")
 def test_forward_compatibility(
     test_output_dir: Path,
+    top_output_dir: Path,
     port_distributor: PortDistributor,
     pg_version: PgVersion,
     request: FixtureRequest,
@@ -174,9 +183,7 @@ def test_forward_compatibility(
     ), "COMPATIBILITY_POSTGRES_DISTRIB_DIR is not set. It should be set to a pg_install directrory (ideally generated by the previous version of Neon)"
     compatibility_postgres_distrib_dir = Path(compatibility_postgres_distrib_dir_env).resolve()
 
-    compatibility_snapshot_dir = (
-        test_output_dir.parent / "test_create_snapshot" / "compatibility_snapshot_pg14"
-    )
+    compatibility_snapshot_dir = top_output_dir / "compatibility_snapshot_pg14"
 
     breaking_changes_allowed = (
         os.environ.get("ALLOW_FORWARD_COMPATIBILITY_BREAKAGE", "false").lower() == "true"

test_runner/regress/test_disk_usage_eviction.py

@@ -136,9 +136,7 @@ def eviction_env(request, neon_env_builder: NeonEnvBuilder, pg_bin: PgBin) -> Ev
     env.pageserver.allowed_errors.append(r".* running disk usage based eviction due to pressure.*")
 
     # remove the initial tenant
-    ## why wait for upload queue? => https://github.com/neondatabase/neon/issues/3865
     assert env.initial_timeline
-    wait_for_upload_queue_empty(pageserver_http, env.initial_tenant, env.initial_timeline)
     pageserver_http.tenant_detach(env.initial_tenant)
     assert isinstance(env.remote_storage, LocalFsStorage)
     tenant_remote_storage = env.remote_storage.root / "tenants" / str(env.initial_tenant)

test_runner/regress/test_hot_standby.py

@@ -1,7 +1,9 @@
 import pytest
 from fixtures.neon_fixtures import NeonEnv
+from fixtures.pg_version import PgVersion, xfail_on_postgres
 
 
+@xfail_on_postgres(PgVersion.V15, reason="https://github.com/neondatabase/neon/pull/4182")
 @pytest.mark.timeout(1800)
 def test_hot_standby(neon_simple_env: NeonEnv):
     env = neon_simple_env

test_runner/regress/test_pg_regress.py

@@ -5,6 +5,7 @@ from pathlib import Path
 
 import pytest
 from fixtures.neon_fixtures import NeonEnv, check_restored_datadir_content
+from fixtures.pg_version import PgVersion, xfail_on_postgres
 
 
 # Run the main PostgreSQL regression tests, in src/test/regress.
@@ -71,6 +72,7 @@ def test_pg_regress(
 #
 # This runs for a long time, especially in debug mode, so use a larger-than-default
 # timeout.
+@xfail_on_postgres(PgVersion.V15, reason="https://github.com/neondatabase/neon/pull/4213")
 @pytest.mark.timeout(1800)
 def test_isolation(
     neon_simple_env: NeonEnv,

test_runner/regress/test_proxy.py

@@ -5,16 +5,18 @@ import pytest
 from fixtures.neon_fixtures import PSQL, NeonProxy, VanillaPostgres
 
 
-def test_proxy_select_1(static_proxy: NeonProxy):
+@pytest.mark.parametrize("option_name", ["project", "endpoint"])
+def test_proxy_select_1(static_proxy: NeonProxy, option_name: str):
     """
     A simplest smoke test: check proxy against a local postgres instance.
     """
 
-    out = static_proxy.safe_psql("select 1", options="project=generic-project-name")
+    out = static_proxy.safe_psql("select 1", options=f"{option_name}=generic-project-name")
     assert out[0][0] == 1
 
 
-def test_password_hack(static_proxy: NeonProxy):
+@pytest.mark.parametrize("option_name", ["project", "endpoint"])
+def test_password_hack(static_proxy: NeonProxy, option_name: str):
     """
     Check the PasswordHack auth flow: an alternative to SCRAM auth for
     clients which can't provide the project/endpoint name via SNI or `options`.
@@ -23,11 +25,12 @@ def test_password_hack(static_proxy: NeonProxy):
     user = "borat"
     password = "password"
     static_proxy.safe_psql(
-        f"create role {user} with login password '{password}'", options="project=irrelevant"
+        f"create role {user} with login password '{password}'",
+        options=f"{option_name}=irrelevant",
     )
 
     # Note the format of `magic`!
-    magic = f"project=irrelevant;{password}"
+    magic = f"{option_name}=irrelevant;{password}"
     static_proxy.safe_psql("select 1", sslsni=0, user=user, password=magic)
 
     # Must also check that invalid magic won't be accepted.
@@ -56,55 +59,62 @@ async def test_link_auth(vanilla_pg: VanillaPostgres, link_proxy: NeonProxy):
     assert out == "42"
 
 
-def test_proxy_options(static_proxy: NeonProxy):
+@pytest.mark.parametrize("option_name", ["project", "endpoint"])
+def test_proxy_options(static_proxy: NeonProxy, option_name: str):
     """
     Check that we pass extra `options` to the PostgreSQL server:
-    * `project=...` shouldn't be passed at all (otherwise postgres will raise an error).
+    * `project=...` and `endpoint=...` shouldn't be passed at all
+    * (otherwise postgres will raise an error).
     * everything else should be passed as-is.
     """
 
-    options = "project=irrelevant -cproxytest.option=value"
+    options = f"{option_name}=irrelevant -cproxytest.option=value"
     out = static_proxy.safe_psql("show proxytest.option", options=options)
     assert out[0][0] == "value"
 
-    options = "-c proxytest.foo=\\ str project=irrelevant"
+    options = f"-c proxytest.foo=\\ str {option_name}=irrelevant"
     out = static_proxy.safe_psql("show proxytest.foo", options=options)
     assert out[0][0] == " str"
 
 
-def test_auth_errors(static_proxy: NeonProxy):
+@pytest.mark.parametrize("option_name", ["project", "endpoint"])
+def test_auth_errors(static_proxy: NeonProxy, option_name: str):
     """
     Check that we throw very specific errors in some unsuccessful auth scenarios.
     """
 
     # User does not exist
     with pytest.raises(psycopg2.Error) as exprinfo:
-        static_proxy.connect(user="pinocchio", options="project=irrelevant")
+        static_proxy.connect(user="pinocchio", options=f"{option_name}=irrelevant")
     text = str(exprinfo.value).strip()
     assert text.endswith("password authentication failed for user 'pinocchio'")
 
     static_proxy.safe_psql(
-        "create role pinocchio with login password 'magic'", options="project=irrelevant"
+        "create role pinocchio with login password 'magic'",
+        options=f"{option_name}=irrelevant",
     )
 
     # User exists, but password is missing
     with pytest.raises(psycopg2.Error) as exprinfo:
-        static_proxy.connect(user="pinocchio", password=None, options="project=irrelevant")
+        static_proxy.connect(user="pinocchio", password=None, options=f"{option_name}=irrelevant")
     text = str(exprinfo.value).strip()
     assert text.endswith("password authentication failed for user 'pinocchio'")
 
     # User exists, but password is wrong
     with pytest.raises(psycopg2.Error) as exprinfo:
-        static_proxy.connect(user="pinocchio", password="bad", options="project=irrelevant")
+        static_proxy.connect(user="pinocchio", password="bad", options=f"{option_name}=irrelevant")
     text = str(exprinfo.value).strip()
     assert text.endswith("password authentication failed for user 'pinocchio'")
 
     # Finally, check that the user can connect
-    with static_proxy.connect(user="pinocchio", password="magic", options="project=irrelevant"):
+    with static_proxy.connect(
+        user="pinocchio", password="magic", options=f"{option_name}=irrelevant"
+    ):
         pass
 
 
-def test_forward_params_to_client(static_proxy: NeonProxy):
+@pytest.mark.parametrize("option_name", ["project", "endpoint"])
+def test_forward_params_to_client(static_proxy: NeonProxy, option_name: str):
     """
     Check that we forward all necessary PostgreSQL server params to client.
     """
@@ -130,7 +140,7 @@ def test_forward_params_to_client(static_proxy: NeonProxy):
         where name = any(%s)
     """
 
-    with static_proxy.connect(options="project=irrelevant") as conn:
+    with static_proxy.connect(options=f"{option_name}=irrelevant") as conn:
         with conn.cursor() as cur:
             cur.execute(query, (reported_params_subset,))
             for name, value in cur.fetchall():
@@ -138,17 +148,18 @@ def test_forward_params_to_client(static_proxy: NeonProxy):
                 assert conn.get_parameter_status(name) == value
 
 
+@pytest.mark.parametrize("option_name", ["project", "endpoint"])
 @pytest.mark.timeout(5)
-def test_close_on_connections_exit(static_proxy: NeonProxy):
+def test_close_on_connections_exit(static_proxy: NeonProxy, option_name: str):
     # Open two connections, send SIGTERM, then ensure that proxy doesn't exit
     # until after connections close.
-    with static_proxy.connect(options="project=irrelevant"), static_proxy.connect(
-        options="project=irrelevant"
+    with static_proxy.connect(options=f"{option_name}=irrelevant"), static_proxy.connect(
+        options=f"{option_name}=irrelevant"
     ):
         static_proxy.terminate()
         with pytest.raises(subprocess.TimeoutExpired):
             static_proxy.wait_for_exit(timeout=2)
         # Ensure we don't accept any more connections
         with pytest.raises(psycopg2.OperationalError):
-            static_proxy.connect(options="project=irrelevant")
+            static_proxy.connect(options=f"{option_name}=irrelevant")
     static_proxy.wait_for_exit()

test_runner/regress/test_remote_storage.py

@@ -2,11 +2,12 @@
 # env NEON_PAGESERVER_OVERRIDES="remote_storage={local_path='/tmp/neon_zzz/'}" poetry ......
 
 import os
+import queue
 import shutil
 import threading
 import time
 from pathlib import Path
-from typing import Dict, List, Tuple
+from typing import Dict, List, Optional, Tuple
 
 import pytest
 from fixtures.log_helper import log
@@ -26,6 +27,7 @@ from fixtures.pageserver.utils import (
 )
 from fixtures.types import Lsn, TenantId, TimelineId
 from fixtures.utils import print_gc_result, query_scalar, wait_until
+from requests import ReadTimeout
 
 
 #
@@ -626,10 +628,7 @@ def test_empty_branch_remote_storage_upload(
 
     new_branch_name = "new_branch"
     new_branch_timeline_id = env.neon_cli.create_branch(new_branch_name, "main", env.initial_tenant)
-
-    with env.endpoints.create_start(new_branch_name, tenant_id=env.initial_tenant) as endpoint:
-        wait_for_last_flush_lsn(env, endpoint, env.initial_tenant, new_branch_timeline_id)
-    wait_upload_queue_empty(client, env.initial_tenant, new_branch_timeline_id)
+    assert_nothing_to_upload(client, env.initial_tenant, new_branch_timeline_id)
 
     timelines_before_detach = set(
         map(
@@ -658,13 +657,19 @@ def test_empty_branch_remote_storage_upload(
     ), f"Expected to have same timelines after reattach, but got {timelines_after_detach}"
 
 
-# Branches off a root branch, but does not write anything to the new branch, so it has a metadata file only.
-# Ensures the branch is not on the remote storage and restarts the pageserver — the branch should be uploaded after the restart.
 @pytest.mark.parametrize("remote_storage_kind", [RemoteStorageKind.LOCAL_FS])
 def test_empty_branch_remote_storage_upload_on_restart(
     neon_env_builder: NeonEnvBuilder,
     remote_storage_kind: RemoteStorageKind,
 ):
+    """
+    Branches off a root branch, but does not write anything to the new branch, so
+    it has a metadata file only.
+
+    Ensures the branch is not on the remote storage and restarts the pageserver
+    — the upload should be scheduled by load, and create_timeline should await
+    for it even though it gets 409 Conflict.
+    """
     neon_env_builder.enable_remote_storage(
         remote_storage_kind=remote_storage_kind,
         test_name="test_empty_branch_remote_storage_upload_on_restart",
@@ -673,35 +678,87 @@ def test_empty_branch_remote_storage_upload_on_restart(
     env = neon_env_builder.init_start()
     client = env.pageserver.http_client()
 
-    new_branch_name = "new_branch"
-    new_branch_timeline_id = env.neon_cli.create_branch(new_branch_name, "main", env.initial_tenant)
+    client.configure_failpoints(("before-upload-index", "return"))
 
-    with env.endpoints.create_start(new_branch_name, tenant_id=env.initial_tenant) as endpoint:
-        wait_for_last_flush_lsn(env, endpoint, env.initial_tenant, new_branch_timeline_id)
-    wait_upload_queue_empty(client, env.initial_tenant, new_branch_timeline_id)
+    new_branch_timeline_id = TimelineId.generate()
 
+    with pytest.raises(ReadTimeout):
+        client.timeline_create(
+            tenant_id=env.initial_tenant,
+            ancestor_timeline_id=env.initial_timeline,
+            new_timeline_id=new_branch_timeline_id,
+            pg_version=env.pg_version,
+            timeout=4,
+        )
+
+    env.pageserver.allowed_errors.append(
+        f".*POST.* path=/v1/tenant/{env.initial_tenant}/timeline.* request was dropped before completing"
+    )
+
+    # index upload is now hitting the failpoint, should not block the shutdown
     env.pageserver.stop()
 
-    # Remove new branch from the remote storage
-    assert isinstance(env.remote_storage, LocalFsStorage)
-    new_branch_on_remote_storage = (
-        env.remote_storage.root
-        / "tenants"
-        / str(env.initial_tenant)
-        / "timelines"
-        / str(new_branch_timeline_id)
+    timeline_path = (
+        Path("tenants") / str(env.initial_tenant) / "timelines" / str(new_branch_timeline_id)
     )
-    assert (
-        new_branch_on_remote_storage.is_dir()
-    ), f"'{new_branch_on_remote_storage}' path does not exist on the remote storage"
-    shutil.rmtree(new_branch_on_remote_storage)
 
-    env.pageserver.start()
+    local_metadata = env.repo_dir / timeline_path / "metadata"
+    assert local_metadata.is_file(), "timeout cancelled timeline branching, not the upload"
 
-    wait_upload_queue_empty(client, env.initial_tenant, new_branch_timeline_id)
+    assert isinstance(env.remote_storage, LocalFsStorage)
+    new_branch_on_remote_storage = env.remote_storage.root / timeline_path
     assert (
-        new_branch_on_remote_storage.is_dir()
-    ), f"New branch should have been reuploaded on pageserver restart to the remote storage path '{new_branch_on_remote_storage}'"
+        not new_branch_on_remote_storage.exists()
+    ), "failpoint should had prohibited index_part.json upload"
+
+    # during reconciliation we should had scheduled the uploads and on the
+    # retried create_timeline, we will await for those to complete on next
+    # client.timeline_create
+    env.pageserver.start(extra_env_vars={"FAILPOINTS": "before-upload-index=return"})
+
+    # sleep a bit to force the upload task go into exponential backoff
+    time.sleep(1)
+
+    q: queue.Queue[Optional[PageserverApiException]] = queue.Queue()
+    barrier = threading.Barrier(2)
+
+    def create_in_background():
+        barrier.wait()
+        try:
+            client.timeline_create(
+                tenant_id=env.initial_tenant,
+                ancestor_timeline_id=env.initial_timeline,
+                new_timeline_id=new_branch_timeline_id,
+                pg_version=env.pg_version,
+            )
+            q.put(None)
+        except PageserverApiException as e:
+            q.put(e)
+
+    create_thread = threading.Thread(target=create_in_background)
+    create_thread.start()
+
+    try:
+        # maximize chances of actually waiting for the uploads by create_timeline
+        barrier.wait()
+
+        assert not new_branch_on_remote_storage.exists(), "failpoint should had stopped uploading"
+
+        client.configure_failpoints(("before-upload-index", "off"))
+        conflict = q.get()
+
+        assert conflict, "create_timeline should not have succeeded"
+        assert (
+            conflict.status_code == 409
+        ), "timeline was created before restart, and uploads scheduled during initial load, so we expect 409 conflict"
+
+        assert_nothing_to_upload(client, env.initial_tenant, new_branch_timeline_id)
+
+        assert (
+            new_branch_on_remote_storage / "index_part.json"
+        ).is_file(), "uploads scheduled during initial load should had been awaited for"
+    finally:
+        create_thread.join()
 
 
 def wait_upload_queue_empty(
@@ -752,4 +809,17 @@ def get_queued_count(
     return int(val)
 
 
+def assert_nothing_to_upload(
+    client: PageserverHttpClient,
+    tenant_id: TenantId,
+    timeline_id: TimelineId,
+):
+    """
+    Check last_record_lsn == remote_consistent_lsn. Assert works only for empty timelines, which
+    do not have anything to compact or gc.
+    """
+    detail = client.timeline_detail(tenant_id, timeline_id)
+    assert Lsn(detail["last_record_lsn"]) == Lsn(detail["remote_consistent_lsn"])
+
+
 # TODO Test that we correctly handle GC of files that are stuck in upload queue.

test_runner/regress/test_tenant_relocation.py

@@ -1,5 +1,7 @@
 import os
+import shutil
 import threading
+import time
 from contextlib import closing, contextmanager
 from pathlib import Path
 from typing import Any, Dict, Optional, Tuple
@@ -12,6 +14,8 @@ from fixtures.neon_fixtures import (
     NeonEnv,
     NeonEnvBuilder,
     PortDistributor,
+    RemoteStorageKind,
+    available_remote_storages,
 )
 from fixtures.pageserver.http import PageserverHttpClient
 from fixtures.pageserver.utils import (
@@ -512,3 +516,225 @@ def test_tenant_relocation(
             if line.startswith("listen_pg_addr"):
                 lines[i] = f"listen_pg_addr = 'localhost:{env.pageserver.service_port.pg}'"
         (env.repo_dir / "config").write_text("\n".join(lines))
+
+
+# Simulate hard crash of pageserver and re-attach a tenant with a branch
+#
+# This exercises a race condition after tenant attach, where the
+# branch point on the ancestor timeline is greater than the ancestor's
+# last-record LSN. We had a bug where GetPage incorrectly followed the
+# timeline to the ancestor without waiting for the missing WAL to
+# arrive.
+@pytest.mark.parametrize("remote_storage_kind", available_remote_storages())
+def test_emergency_relocate_with_branches_slow_replay(
+    neon_env_builder: NeonEnvBuilder,
+    remote_storage_kind: RemoteStorageKind,
+):
+    neon_env_builder.enable_remote_storage(
+        remote_storage_kind=remote_storage_kind,
+        test_name="test_emergency_relocate_with_branches_slow_replay",
+    )
+
+    env = neon_env_builder.init_start()
+    env.pageserver.is_testing_enabled_or_skip()
+    pageserver_http = env.pageserver.http_client()
+
+    # Prepare for the test:
+    #
+    # - Main branch, with a table and two inserts to it.
+    # - A logical replication message between the inserts, so that we can conveniently
+    #   pause the WAL ingestion between the two inserts.
+    # - Child branch, created after the inserts
+    tenant_id, _ = env.neon_cli.create_tenant()
+
+    main_endpoint = env.endpoints.create_start("main", tenant_id=tenant_id)
+    with main_endpoint.cursor() as cur:
+        cur.execute("CREATE TABLE test_reattach (t text)")
+        cur.execute("INSERT INTO test_reattach VALUES ('before pause')")
+
+        cur.execute("SELECT pg_logical_emit_message(false, 'neon-test', 'between inserts')")
+
+        cur.execute("INSERT INTO test_reattach VALUES ('after pause')")
+        current_lsn = Lsn(query_scalar(cur, "SELECT pg_current_wal_flush_lsn()"))
+
+    main_endpoint.stop()
+    env.neon_cli.create_branch("child", tenant_id=tenant_id, ancestor_start_lsn=current_lsn)
+
+    # Now kill the pageserver, remove the tenant directory, and restart. This simulates
+    # the scenario that a pageserver dies unexpectedly and cannot be recovered, so we relocate
+    # the tenant to a different pageserver. We reuse the same pageserver because it's
+    # simpler than initializing a new one from scratch, but the effect on the single tenant
+    # is the same.
+    env.pageserver.stop(immediate=True)
+    shutil.rmtree(Path(env.repo_dir) / "tenants" / str(tenant_id))
+    env.pageserver.start()
+
+    # This fail point will pause the WAL ingestion on the main branch, after the
+    # the first insert
+    pageserver_http.configure_failpoints([("wal-ingest-logical-message-sleep", "return(5000)")])
+
+    # Attach and wait a few seconds to give it time to load the tenants, attach to the
+    # safekeepers, and to stream and ingest the WAL up to the pause-point.
+    before_attach_time = time.time()
+    pageserver_http.tenant_attach(tenant_id)
+    time.sleep(3)
+
+    # The wal ingestion on the main timeline should now be paused at the fail point.
+    # Run a query on the child branch. The GetPage requests for this should recurse to the
+    # parent timeline, and wait for the WAL to be ingested there. Otherwise it won't see
+    # the second insert.
+    child_endpoint = env.endpoints.create_start("child", tenant_id=tenant_id)
+    with child_endpoint.cursor() as cur:
+        cur.execute("SELECT * FROM test_reattach")
+        assert cur.fetchall() == [("before pause",), ("after pause",)]
+
+    # Sanity check that the failpoint was reached
+    assert env.pageserver.log_contains('failpoint "wal-ingest-logical-message-sleep": sleep done')
+    assert time.time() - before_attach_time > 5
+
+    # Clean up
+    pageserver_http.configure_failpoints(("wal-ingest-logical-message-sleep", "off"))
+
+
+# Simulate hard crash of pageserver and re-attach a tenant with a branch
+#
+# This exercises the same race condition after as
+# 'test_emergency_relocate_with_branches_slow_replay', but this test case
+# is closer to the original scenario where we originally found the
+# issue.
+#
+# In this scenario, the incorrect result to get-request leads to
+# *permanent damage* in the child timeline, because ingesting the WAL
+# on the child timeline depended on incorrect view of the parent. This
+# test reproduced one such case; the symptom was an error on the child, when
+# trying to connect to the child endpoint after re-attaching the tenant:
+#
+# FATAL: database "neondb" does not exist
+#
+# In the original case where we bumped into this, the error was slightly
+# different:
+#
+# FATAL:  "base/16385" is not a valid data directory
+# DETAIL:  File "base/16385/PG_VERSION" is missing.
+#
+### Detailed explanation of the original bug and why it lead to that error:
+#
+# The WAL on the main and the child branches look like this:
+#
+#    Main                                  Child
+# 1. CREATE DATABASE
+#           <child branch is created>
+# 2. CREATE TABLE AS SELECT ...            3. CREATE TABLE AS SELECT ...
+#
+# None of these WAL records have been flushed to disk or uploaded to remote
+# storage in the pageserver yet, when the tenant is detached.
+#
+# After detach and re-attach, a walreceiver is spawned on both timelines.
+# They will connect to the safekeepers and start ingesting the WAL
+# from their respective IndexParts' `disk_consistent_lsn` onward.
+#
+# The bug occurs if the child branch's walreceiver runs before the
+# main's.  It receives the SMGR_CREATE WAL record emitted by the
+# CREATE TABLE statement (3.), and applies it, without seeing the
+# effect of the CREATE DATABASE statement.
+#
+# To understand why that leads to a 'File "base/16385/PG_VERSION" is
+# missing' error, let's look at what the handlers for the WAL records
+# do:
+#
+# CREATE DATABASE WAL record is handled by ingest_xlog_dbase_create:
+#
+#    ingest_xlog_dbase_create:
+#     put_relmap_file()
+#       // NOTE 'true': It means that there is a relmapper and PG_VERSION file
+# 1:    let r = dbdir.dbdirs.insert((spcnode, dbnode), true);
+#
+#
+# CREATE TABLE emits an SMGR_CREATE record, which is handled by:
+#
+#    ingest_xlog_smgr_create:
+#      put_rel_creation:
+#        ...
+#        let mut rel_dir = if dbdir.dbdirs.get(&(rel.spcnode, rel.dbnode)).is_none() {
+# 2:         // Didn't exist. Update dbdir
+#            dbdir.dbdirs.insert((rel.spcnode, rel.dbnode), false);
+#            let buf = DbDirectory::ser(&dbdir)?;
+#            self.put(DBDIR_KEY, Value::Image(buf.into()));
+#
+#            // and create the RelDirectory
+#            RelDirectory::default()
+#        } else {
+# 3:         // reldir already exists, fetch it
+#            RelDirectory::des(&self.get(rel_dir_key, ctx).await?)?
+#        };
+#
+#
+# In the correct ordering, the SMGR_CREATE record is applied after the
+# CREATE DATABASE record. The CREATE DATABASE creates the entry in the
+# 'dbdir', with the 'true' flag that indicates that PG_VERSION exists
+# (1). The SMGR_CREATE handler calls put_rel_creation, which finds the
+# dbdir entry that the CREATE DATABASE record created, and takes the
+# "reldir already exists, fetch it" else-branch at the if statement (3).
+#
+# In the incorrect ordering, the child walreceiver applies the
+# SMGR_CREATE record without seeing the effects of the CREATE
+# DATABASE. In that case, put_rel_creation takes the "Didn't
+# exist. Update dbir" path (2), and inserts an entry in the
+# DbDirectory with 'false' to indicate there is no PG_VERSION file.
+#
+@pytest.mark.parametrize("remote_storage_kind", available_remote_storages())
+def test_emergency_relocate_with_branches_createdb(
+    neon_env_builder: NeonEnvBuilder,
+    remote_storage_kind: RemoteStorageKind,
+):
+    neon_env_builder.enable_remote_storage(
+        remote_storage_kind=remote_storage_kind,
+        test_name="test_emergency_relocate_with_branches_createdb",
+    )
+
+    env = neon_env_builder.init_start()
+    pageserver_http = env.pageserver.http_client()
+
+    # create new nenant
+    tenant_id, _ = env.neon_cli.create_tenant()
+
+    main_endpoint = env.endpoints.create_start("main", tenant_id=tenant_id)
+    with main_endpoint.cursor() as cur:
+        cur.execute("SELECT pg_logical_emit_message(false, 'neon-test', 'between inserts')")
+
+        cur.execute("CREATE DATABASE neondb")
+        current_lsn = Lsn(query_scalar(cur, "SELECT pg_current_wal_flush_lsn()"))
+    env.neon_cli.create_branch("child", tenant_id=tenant_id, ancestor_start_lsn=current_lsn)
+
+    with main_endpoint.cursor(dbname="neondb") as cur:
+        cur.execute("CREATE TABLE test_migrate_one AS SELECT generate_series(1,100)")
+    main_endpoint.stop()
+
+    child_endpoint = env.endpoints.create_start("child", tenant_id=tenant_id)
+    with child_endpoint.cursor(dbname="neondb") as cur:
+        cur.execute("CREATE TABLE test_migrate_one AS SELECT generate_series(1,200)")
+    child_endpoint.stop()
+
+    # Kill the pageserver, remove the tenant directory, and restart
+    env.pageserver.stop(immediate=True)
+    shutil.rmtree(Path(env.repo_dir) / "tenants" / str(tenant_id))
+    env.pageserver.start()
+
+    # Wait before ingesting the WAL for CREATE DATABASE on the main branch. The original
+    # bug reproduced easily even without this, as there is always some delay between
+    # loading the timeline and establishing the connection to the safekeeper to stream and
+    # ingest the WAL, but let's make this less dependent on accidental timing.
+    pageserver_http.configure_failpoints([("wal-ingest-logical-message-sleep", "return(5000)")])
+    before_attach_time = time.time()
+    pageserver_http.tenant_attach(tenant_id)
+
+    child_endpoint.start()
+    with child_endpoint.cursor(dbname="neondb") as cur:
+        assert query_scalar(cur, "SELECT count(*) FROM test_migrate_one") == 200
+
+    # Sanity check that the failpoint was reached
+    assert env.pageserver.log_contains('failpoint "wal-ingest-logical-message-sleep": sleep done')
+    assert time.time() - before_attach_time > 5
+
+    # Clean up
+    pageserver_http.configure_failpoints(("wal-ingest-logical-message-sleep", "off"))

test_runner/regress/test_tenant_size.py

@@ -11,6 +11,7 @@ from fixtures.neon_fixtures import (
     wait_for_wal_insert_lsn,
 )
 from fixtures.pageserver.http import PageserverHttpClient
+from fixtures.pg_version import PgVersion, xfail_on_postgres
 from fixtures.types import Lsn, TenantId, TimelineId
 
 
@@ -512,6 +513,7 @@ def test_single_branch_get_tenant_size_grows(
     assert size_after == prev, "size after restarting pageserver should not have changed"
 
 
+@xfail_on_postgres(PgVersion.V15, reason="Test significantly more flaky on Postgres 15")
 def test_get_tenant_size_with_multiple_branches(
     neon_env_builder: NeonEnvBuilder, test_output_dir: Path
 ):

test_runner/regress/test_wal_restore.py

@@ -1,14 +1,20 @@
+import sys
 from pathlib import Path
 
+import pytest
 from fixtures.neon_fixtures import (
     NeonEnvBuilder,
     PgBin,
     PortDistributor,
     VanillaPostgres,
 )
-from fixtures.types import TenantId
+from fixtures.types import TenantId, TimelineId
 
 
+@pytest.mark.skipif(
+    sys.platform != "linux",
+    reason="restore_from_wal.sh supports only Linux",
+)
 def test_wal_restore(
     neon_env_builder: NeonEnvBuilder,
     pg_bin: PgBin,
@@ -22,6 +28,7 @@ def test_wal_restore(
     endpoint = env.endpoints.create_start("test_wal_restore")
     endpoint.safe_psql("create table t as select generate_series(1,300000)")
     tenant_id = TenantId(endpoint.safe_psql("show neon.tenant_id")[0][0])
+    timeline_id = TimelineId(endpoint.safe_psql("show neon.timeline_id")[0][0])
     env.neon_cli.pageserver_stop()
     port = port_distributor.get_port()
     data_dir = test_output_dir / "pgsql.restored"
@@ -30,9 +37,16 @@ def test_wal_restore(
     ) as restored:
         pg_bin.run_capture(
             [
-                str(base_dir / "libs/utils/scripts/restore_from_wal.sh"),
+                str(base_dir / "libs" / "utils" / "scripts" / "restore_from_wal.sh"),
                 str(pg_distrib_dir / f"v{env.pg_version}/bin"),
-                str(test_output_dir / "repo" / "safekeepers" / "sk1" / str(tenant_id) / "*"),
+                str(
+                    test_output_dir
+                    / "repo"
+                    / "safekeepers"
+                    / "sk1"
+                    / str(tenant_id)
+                    / str(timeline_id)
+                ),
                 str(data_dir),
                 str(port),
             ]