Repository navigation
fix: stop compute log capture deadlocking a command at exit (#1165) * fix: release the compute log lock before rejecting a write to a finished capture A write that reached a capture already marked complete built its rejection while still holding the storage lock. Building it imports Python, which can release the GIL; the main thread then pruning the logs waited for that lock while holding the GIL, so neither could continue and the command never exited. The lock is now released before the rejection is r... #171
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| name: Release Performance | |
| run-name: >- | |
| ${{ github.event.pull_request.title || github.event.head_commit.message || github.workflow }} | |
| # Every push to main compares main with the latest release on the same runner. A regression fails | |
| # the run and opens or updates one issue labelled performance-regression; it does not gate the | |
| # release pull request. Pull requests that change the comparison itself run it too. | |
| on: | |
| push: | |
| branches: | |
| - main | |
| workflow_dispatch: | |
| inputs: | |
| baseline: | |
| description: "Published sqlbuild version to compare against (default: the release before the candidate)" | |
| required: false | |
| type: string | |
| pull_request: | |
| paths: | |
| - .github/workflows/release-performance.yml | |
| - scripts/compare_release_performance.py | |
| - scripts/release_performance/** | |
| - scripts/compile_performance_ratio/_helpers/edit.py | |
| - scripts/compile_performance_ratio/_helpers/measure.py | |
| - scripts/compile_performance_ratio/_helpers/report.py | |
| - scripts/compile_performance_ratio/constants.py | |
| - scripts/compile_performance_ratio/exceptions.py | |
| - scripts/compile_performance_ratio/models.py | |
| - scripts/cold_compile_performance/_helpers/dense_project.py | |
| - scripts/cold_compile_performance/constants.py | |
| permissions: | |
| contents: read | |
| # Every main commit gets its own comparison, so a later push never cancels the run that would | |
| # report an earlier regression. Pull request runs supersede older runs of the same pull request. | |
| concurrency: | |
| group: >- | |
| release-performance-${{ github.event_name }}-${{ | |
| github.event_name == 'push' && github.sha || github.ref }} | |
| cancel-in-progress: ${{ github.event_name == 'pull_request' }} | |
| jobs: | |
| compare: | |
| name: Compare with the previous release | |
| runs-on: ubuntu-24.04 | |
| timeout-minutes: 90 | |
| outputs: | |
| regressed: ${{ steps.compare.outputs.regressed }} | |
| steps: | |
| - uses: actions/checkout@v4 | |
| with: | |
| # Release tags identify the previous release before its wheels reach PyPI. | |
| fetch-depth: 0 | |
| - uses: astral-sh/setup-uv@v5 | |
| with: | |
| version: "0.12.3" | |
| enable-cache: true | |
| cache-dependency-glob: | | |
| uv.lock | |
| Cargo.lock | |
| - uses: dtolnay/rust-toolchain@master | |
| with: | |
| toolchain: stable | |
| - name: Install Python | |
| run: uv python install 3.12 | |
| - name: Install dependencies | |
| run: uv sync --locked --all-groups --all-extras | |
| - name: Build candidate wheel | |
| run: | | |
| uv build --wheel --out-dir "${RUNNER_TEMP}/candidate" | |
| uv build --wheel crates/sqlbuild-cli --out-dir "${RUNNER_TEMP}/candidate" | |
| # A release tag created while the candidate built names the baseline's benchmark generator. | |
| - name: Fetch release tags | |
| run: git fetch --tags --force --quiet origin | |
| # Temporary: one-model edit compiles use MAX_EDIT_COMPILE_TIME_RATIO (1.35x) because an edited | |
| # model recompiles uncached; tighten it back to 1.15x once native rendering is the default. | |
| - name: Compare candidate and previous release on this runner | |
| id: compare | |
| env: | |
| BASELINE: ${{ inputs.baseline }} | |
| EVENT_NAME: ${{ github.event_name }} | |
| run: | | |
| # A main commit that is not itself a release still carries the latest release's version, | |
| # so it is compared with that release rather than the one before it. A release commit | |
| # bumps the version and is compared with the release before it. | |
| if [[ "${EVENT_NAME}" == "push" && -z "${BASELINE}" ]]; then | |
| project_version() { | |
| python3 -c 'import sys, tomllib; print(tomllib.loads(sys.stdin.read())["project"]["version"])' | |
| } | |
| version="$(project_version < pyproject.toml)" | |
| previous="$(git show HEAD^:pyproject.toml | project_version)" || previous="" | |
| if [[ "${version}" == "${previous}" ]]; then | |
| BASELINE="${version}" | |
| fi | |
| fi | |
| wheels=("${RUNNER_TEMP}"/candidate/sqlbuild-*.whl) | |
| evidence="${RUNNER_TEMP}/release-performance" | |
| mkdir -p "${evidence}" | |
| args=( | |
| --candidate "${wheels[0]}" | |
| --work-dir "${RUNNER_TEMP}/release-performance-work" | |
| --output "${evidence}/release-performance.json" | |
| ) | |
| if [[ -n "${BASELINE}" ]]; then | |
| args+=(--baseline "${BASELINE}") | |
| fi | |
| # The comparison's own exit status, not tee's, decides the outcome. | |
| set +e | |
| uv run python -m scripts.compare_release_performance "${args[@]}" \ | |
| | tee "${evidence}/release-performance.md" | |
| status="${PIPESTATUS[0]}" | |
| set -e | |
| regressed=false | |
| if [[ "${status}" != "0" && -f "${evidence}/release-performance.json" ]] && \ | |
| jq -e 'any(.verdicts[]; .regressed)' "${evidence}/release-performance.json" > /dev/null; then | |
| regressed=true | |
| fi | |
| echo "status=${status}" >> "${GITHUB_OUTPUT}" | |
| echo "regressed=${regressed}" >> "${GITHUB_OUTPUT}" | |
| - name: Upload comparison evidence | |
| if: always() | |
| uses: actions/upload-artifact@v4 | |
| with: | |
| name: release-performance-${{ github.run_id }}-${{ github.run_attempt }} | |
| path: ${{ runner.temp }}/release-performance | |
| if-no-files-found: warn | |
| retention-days: 90 | |
| - name: Fail on a regression or comparison error | |
| if: steps.compare.outputs.status != '0' | |
| env: | |
| STATUS: ${{ steps.compare.outputs.status }} | |
| run: | | |
| echo "The release performance comparison failed with exit status ${STATUS:-unknown}" >&2 | |
| exit 1 | |
| report: | |
| name: Report regression | |
| needs: compare | |
| if: ${{ always() && github.event_name == 'push' && needs.compare.outputs.regressed == 'true' }} | |
| runs-on: ubuntu-24.04 | |
| # Overlapping main runs report one at a time so they update a single regression issue. | |
| concurrency: | |
| group: release-performance-report | |
| cancel-in-progress: false | |
| permissions: | |
| issues: write | |
| steps: | |
| - uses: actions/download-artifact@v4 | |
| with: | |
| name: release-performance-${{ github.run_id }}-${{ github.run_attempt }} | |
| path: evidence | |
| - name: Open or update the performance regression issue | |
| env: | |
| GH_TOKEN: ${{ github.token }} | |
| GH_REPO: ${{ github.repository }} | |
| LABEL: performance-regression | |
| RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }} | |
| run: | | |
| { | |
| echo "Release performance regressed at ${GITHUB_SHA} on main." | |
| echo | |
| echo "Comparison run: ${RUN_URL}" | |
| echo | |
| cat evidence/release-performance.md | |
| } > report.md | |
| gh label create "${LABEL}" --color B60205 \ | |
| --description "Main is slower or larger than the latest release" 2>/dev/null || true | |
| issue="$(gh issue list --label "${LABEL}" --state open --limit 1 \ | |
| --json number --jq '.[0].number // empty')" | |
| if [[ -n "${issue}" ]]; then | |
| gh issue comment "${issue}" --body-file report.md | |
| echo "Updated issue #${issue}" >> "${GITHUB_STEP_SUMMARY}" | |
| else | |
| gh issue create --title "Release performance regression on main" \ | |
| --label "${LABEL}" --body-file report.md | |
| fi |