Open sigmundch opened 6 years ago
I can't work backward to the run this is from from the log - I'll have to search. It seems like we could get the process killing, where all firefox processes are killed, into the test.py run on the shard, if this is a shard, because it seems like if we killed all the firefox.exe processes, we wouldn't get this error.
Here is the run from above: https://ci.chromium.org/p/dart/builders/luci.dart.ci.sandbox/dart2js-strong-win-x64-ff/11
I believe every red FF-win bot recently is due to this reason, it often fails on shard_5, but it has failed on other shards as well.
I think we need to move the taskkill step into the test.py process, so that it has a chance to execute on a shard. At least a test.py run that has started chrome or ff browsers should (if it is running on a builder) kill all the browsers on that machine before exiting. I'm not sure if these FF processes were started directly from test.py, or if they could be orphans of other FF processes.
It appears there is some infrastructure flaky behavior in the FF windows bot.
See https://logs.chromium.org/logs/dart/buildbucket/cr-buildbucket.appspot.com/8941183950889847392/+/steps/dart2js_tests_shard_5_on_Windows/0/stdout for an example failure.
/cc @whesse @athomas