Some test failures are caused by the large amount parallelism that NRWT employs which can stress some parts of whatever platform the tests are running on (e.g. see bug 82233). At the conclusion of the parallel stage, we should run the failures serially rather than in parallel so that we can eliminate these errors as a factor in test failures.
Created attachment 133891 [details]
Comment on attachment 133891 [details]
I'm a bit concerned about what side effects modifying the value in the options might have ... have you run this through test-webkitpy?
I was actually more envisioning that you'd pass in the # of workers to use to _run_tests(), which I think (should?) be a safer change, but I suppose that might have unexpected ramifications as well.
Created attachment 133893 [details]
Comment on attachment 133893 [details]
View in context: https://bugs.webkit.org/attachment.cgi?id=133893&action=review
> + def _run_tests(self, file_list, result_summary, num_workers=0):
Sorry for all the nitpicking, but I usually only like to use optional parameters as a last choice. You should be able to just pass in num_workers to both places it's called (line 897 and line 911), and then change line 751 to min(num_workers, len(all_shards).
Created attachment 133897 [details]
> Sorry for all the nitpicking...
Not a problem :-) When in Rome...
Comment on attachment 133897 [details]
Clearing flags on attachment: 133897
Committed r112181: <http://trac.webkit.org/changeset/112181>
All reviewed patches have been landed. Closing bug.