You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The scheduled nightly 43125899270, tested SHA 0bdc3a7b2a582cbb65a44e07a879775a9fc71fb1, produced valid drag measurements, but daily-benchmark could not find them and failed with SUITE ui sample missing; INCOMPLETE ui.drag.node.frameP95,ui.drag.pan.frameP95.
Evidence and verified cause
perf-outputs job/logs succeeded. The upload log explicitly says the least common ancestor of its paths is /.
I downloaded and inspected artifact 13903062139. Its ZIP contains home/runner/work/OpenDirector/OpenDirector/nightly-perf.json alongside tmp/opendirector-drag-nightly-*.cpuprofile.
The JSON matches the run/SHA and contains node frameP95 49.9 ms and pan frameP95 66.7 ms. Render verification passed (H.264/AAC, 640×360, 48 frames).
daily-benchmark job/logs successfully downloads that artifact into nightly-perf/, but only passes --perf when nightly-perf/nightly-perf.json exists.
Workflow producer mixes a workspace-relative JSON and absolute /tmp profiles in one upload, so the consumer's assumed root layout is wrong.
Classification: confirmed CI artifact-layout defect, not missing browser execution or a demonstrated application performance defect.
Reproduction
Download the linked nightly-perf ZIP and list its members. Extract into a directory named nightly-perf; the consumer's test -f nightly-perf/nightly-perf.json fails although the nested JSON exists. This archive inspection was performed; I did not rerun a workflow or browser.
Expected: the downstream trend report consumes this run's finite node/pan measurements and records matching source run/SHA. Actual: both are discarded and reported missing.
Stage the report and profiles under a single explicit artifact directory (or split artifacts), align the consumer path, and add an upload-layout/consumer regression test. Preserve real missing-sample failure behavior.
The same run also reports storage.delete.trash 172.4 ms vs 107.2 ms previously (+61%). That is a separate single-run threshold signal, with root cause unverified; fixing artifact wiring must not hide it or promise the whole lane will pass.
Current applicability and dedup
Current main af428f9c8ddbbd616106f1bdd4d189370175395c retains the identical workflow. All 297 issue/PR records across 3 pages were inspected, plus relevant #249/#232/#273/#153 discussion and a final updates read. #249 fixed browser/perf execution but not this artifact handoff; #232 correctly fails missing measurements. No existing report of this exact layout defect found. Refs #200, #213, #273.
Activity
[bugfix] ci: keep the nightly perf report at its artifact root (#299) (…