Repository navigation
ci(bench): compare the Dart build time over three interleaved runs per side (#1132) - #1254
Merged
Merged
Conversation
…r side (#1132) The grid benchmark compared one run per side on FrameTiming.totalSpan, which includes vsync wait and software rasterising on a shared VM, so the base itself moved by 40% between runs and PRs that did not touch the grid got regression comments. - Three runs per side, interleaved (base, PR, ...); medians are compared. - Only build p50 / p90 flag a regression: worse by more than 10% and 0.2 ms, and slower in every pair. Raster, total and stutters stay in the table as information. - Stutters count gaps over 25 ms (a missed vsync at 60 Hz under xvfb). - Unit tests for the comparator run in CI.
github-actions
Bot
added
performance
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Closes #1132
Changes
benchmark.yml: three runs per side, interleaved (base, PR, base, PR, base, PR), so a slow stretch of the shared runner lands on both sides. The base runs the PR's copy ofbenchmark/, so both sides print the same lines.bench_compare.py: takes--baseand--headlogs and compares medians.buildp50 and p90 can flag a regression: worse by more than 10% and 0.2 ms, and slower in every pair.buildp99,raster,totaland stutters are in the table, marked as information.grid_perf_bench.dart: also printsstutters(>25ms), a missed vsync at 60 Hz (xvfb). The 120 Hz count stays for local runs.scripts/ci/test_bench_compare.py(13 cases, including the feat(workspace): Explain and Cancel in the SQL toolbar #1131 numbers that used to comment) runs in the CI analyze job. It passes locally.Still informational; the job never blocks a merge. It costs about three more benchmark runs per PR that touches the grid paths.
This PR changes the workflow itself, so the Grid performance job on it runs the new comparison.