Limit the amount of bot time for reference build benchmark that just fails completely |
|||||||
Issue descriptionWith issue 680138 fixed, we will no longer need to disable failing reference build benchmark. However, this comes with the annoyance of wasting bot time for those builds if they just don't produce any useful data. In https://build.chromium.org/p/chromium.perf/builders/Win%2010%20High-DPI%20Perf/builds/449, memory.desktop.reference is taking 13 mins, 27 secs to fail (the non reference build pass after 26 minutes) Proposal: we fail the benchmark completely if the first 3 stories in the page fail in a row. Thoughts?
,
Mar 31 2017
,
Apr 3 2017
If the benchmark fails completely and consistently, we should either replace the reference build, or just stop running the reference build at all on that bot. On system health I'm only aware of two bots where this is an issue: go/catabug/3399
,
Apr 13 2017
We decided not to do anything about this. The main reason is that we don't really care about the cycle time of reference build as it is not a lot more than TOT build. The strategy proposed in #1 can save some times, but won't be as much when we shard benchmark by stories, so not much win with more code complexity. Long term: we will carefully track the cycle time of all benchmarks with perf dashboard & will look carefully when a ref benchmarks' cycle time blow up.
,
Apr 13 2017
,
Aug 10 2017
This is actually on the table again since with Stephen's sharding algorithm, most benchmark will not be sharded (only ones that are cut in the middle).
,
Aug 10 2017
Also we found out in code review (https://codereview.chromium.org/2998773002/) today that we can make this happens using 1 line CL to add "--max-failures=5" to all the ref build benchmark.
,
Aug 14 2017
This makes so much sense to me.
,
Aug 14 2017
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/10450252daa708cd59929774aba0c27fff671fc0 commit 10450252daa708cd59929774aba0c27fff671fc0 Author: Ned Nguyen <nednguyen@google.com> Date: Mon Aug 14 20:37:26 2017 Set maximum number of failures for reference build benchmarks to be 5 Bug: 707236 Change-Id: Id129869ac38eed55eb0ce0f5b9daf34e04f5210c Reviewed-on: https://chromium-review.googlesource.com/611314 Reviewed-by: Stephen Martinis <martiniss@chromium.org> Reviewed-by: Juan Antonio Navarro Pérez <perezju@chromium.org> Commit-Queue: Ned Nguyen <nednguyen@google.com> Cr-Commit-Position: refs/heads/master@{#494170} [modify] https://crrev.com/10450252daa708cd59929774aba0c27fff671fc0/testing/buildbot/chromium.perf.fyi.json [modify] https://crrev.com/10450252daa708cd59929774aba0c27fff671fc0/testing/buildbot/chromium.perf.json [modify] https://crrev.com/10450252daa708cd59929774aba0c27fff671fc0/tools/perf/core/perf_data_generator.py [modify] https://crrev.com/10450252daa708cd59929774aba0c27fff671fc0/tools/perf/core/perf_data_generator_unittest.py
,
Aug 15
This issue has been Available for over a year. If it's no longer important or seems unlikely to be fixed, please consider closing it out. If it is important, please re-triage the issue. Sorry for the inconvenience if the bug really should have been left as Available. For more details visit https://www.chromium.org/issue-tracking/autotriage - Your friendly Sheriffbot
,
Aug 15
With #9, I think this is fixed |
|||||||
►
Sign in to add a comment |
|||||||
Comment 1 by nedngu...@google.com
, Mar 31 2017