Repository navigation
ci: unstick star chart, fix weekly benchmarks, pin all actions to Node 24 SHAs - #626
Merged
Merged
Conversation
The repo has held 40 stars since 2026-09-30, so the daily job found no movement and the chart's line stopped at that date, reading as a frozen chart. Carry the last count forward to today when rendering; the stored series still records only days the count moved.
…nchmark workflows - Pin actions/*, astral-sh/setup-uv, msys2/setup-msys2 and softprops/action-gh-release to the commit of their latest release, with the tag in a comment (the policy the benchmark workflows already enforced). Clears the Node 20 deprecation on every job. - benchmark-memory: move off benchmark-latency's identical Sunday cron; both share benchmark-stats-production, where GitHub keeps one pending run, so latency cancelled memory before it got a runner. Raise the 60 min job timeout that the 5-allocator suite outgrew to 90 (repo variable BENCHMARK_MEMORY_TIMEOUT_MINUTES overrides). - benchmark-pprof-tax: upload the validation report only when validation ran, so a matrix failure is not buried under a missing-file error.
#414 added both to ci/build_pprof_tax_configurations.py's equivalence and fork-only key lists but not to the Rust copy, so benchmark-pprof-tax-run rejected every real run before measuring (every weekly run since 09-27).
The overhead check (control p50 <= 5% of measured p50, measured p99 > 2x control p99) was fatal for every allocator, and tcmalloc's tiny-fixed-64/1 cell sits at timer resolution on hosted runners, so benchmark-latency has failed every week. Keep it a hard gate for mimalloc-pprof; record overhead_valid for references (still checked for consistency). Name the thresholds in Rust and in the Python site validator, and write the raw run before validating so a failed run uploads its evidence.
… in tests The memory checker enforces 60 minutes as a hard budget, so raising it is an owner decision (escalated on the issue), not part of this fix. Two workflow tests matched checkout by its old tag; match the action, not the ref.
Local-Gate: v1 tree=3572d92417ec8f94e84bd8a9b58b9b561ea78b3a secs=67 lanes=lint:run
Ci-Attestation: {"at":1792967571,"gate":"python/all/lint","host":"linux-x86_64","key":"44d2f62f96209c34eace739e0010857fa8449ed308c3a56e5069da2f57267d14","lane":"lint","parents":["32b48501de26e6b9dd4ac40006ea464f50e1a593"],"secs":67,"stamp":"5c7273c53be9fa9e7dceb7aca69c286c","tree":"3572d92417ec8f94e84bd8a9b58b9b561ea78b3a","v":1,"via":"run"}
zackees
added
the
ci-full
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes the stuck README star chart, the three scheduled benchmark workflows that have failed every week, and moves every action off deprecated Node 20 onto SHA-pinned releases.
Star chart (
star-history.yml"succeeded" daily but the chart stopped at 09-30)The repo has held 40 stars since 2026-09-30, so the job correctly found no movement and never pushed; the line just ended at that date.
ci/star_history.pynow carries the last count forward to today when rendering (extend_to);star-history.jsonstill records only days the count moved (verified unchanged against the data branch).Actions: pinned to full SHAs, all on Node 24
249
uses:refs foractions/*,astral-sh/setup-uv,msys2/setup-msys2,softprops/action-gh-release→owner/name@<sha> # vX.Y.Zat each latest release (the SHA-pin policy the benchmark workflows already enforced, now everywhere). Breaking notes reviewed per major: download-artifact v5 by-ID path change (no by-ID downloads here), v8 hash-mismatch now errors (wanted); setup-uv v7server-url/v8 manifest removal (unused), v10 cache disabled on untrusted events (fine); setup-python v7 removedpip-install(unused).zackees/*actions left as the owner's own refs (setup-soldr@v0,docker-mac-x64@main). Labelledci-fullso the full matrix, macOS native and release-path jobs exercise the new pins.benchmark-pprof-tax (failing every week since 09-27)
#414 added
MI_DIAGNOSTICS/MI_MEMEVTtoci/build_pprof_tax_configurations.py's key lists but not torust/benchmark-suite/src/pprof_tax.rs, so the run binary rejected every real run in 0.05 s:upstream-baseline cmake_cache does not carry exactly the equivalence cache keys(run 40163077276). Rust lists + synthetic fixture updated; the check is restored, not relaxed. The upload step now runs only when validation ran, instead of adding a misleading missing-file error.benchmark-latency (failing every week)
overhead_is_valid(control p50 ≤ 5% measured p50, measured p99 > 2× control p99) was fatal for every allocator; tcmalloctiny-fixed-64/1sits at timer resolution on hosted runners, so it never passed. It stays a hard gate formimalloc-pprof; references recordoverhead_valid(still cross-checked for consistency) — mirrored inci/benchmark_report.py's site validator. Thresholds are now named constants (rule 11). The raw run is written before validation so a failed run uploads evidence.benchmark-memory (cancelled/failed since August)
Same Sunday 08:41 cron as benchmark-latency and the same
benchmark-stats-productiongroup: GitHub keeps one pending run per group, so latency cancelled memory before it got a runner (run 40178014027,jobs: []). Moved to Saturday. Not changed: the 60-minute job timeout, which the 5-allocator suite now exceeds (suite step 43 → 49 → >53 min + ~7 min builds).ci/check_benchmark_memory_workflow.pyenforces 60 as a hard budget, so that is escalated as an owner decision rather than raised here.Tests: benchmark-suite 177 passed; local gate 45/45.