Bug 63795 - ReportCrash destabilizes new-run-webkit-tests
Summary: ReportCrash destabilizes new-run-webkit-tests
Status: RESOLVED FIXED
Alias: None
Product: WebKit
Classification: Unclassified
Component: New Bugs (show other bugs)
Version: 528+ (Nightly build)
Hardware: Unspecified Unspecified
: P2 Normal
Assignee: Adam Barth
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2011-07-01 02:09 PDT by Adam Barth
Modified: 2011-07-01 13:21 PDT (History)
4 users (show)

See Also:


Attachments
Patch (5.62 KB, patch)
2011-07-01 02:12 PDT, Adam Barth
no flags Details | Formatted Diff | Diff

Note You need to log in before you can comment on or make changes to this bug.
Description Adam Barth 2011-07-01 02:09:54 PDT
ReportCrash destabilizes new-run-webkit-tests
Comment 1 Adam Barth 2011-07-01 02:12:51 PDT
Created attachment 99445 [details]
Patch
Comment 2 Adam Barth 2011-07-01 11:14:41 PDT
Comment on attachment 99445 [details]
Patch

Clearing flags on attachment: 99445

Committed r90246: <http://trac.webkit.org/changeset/90246>
Comment 3 Adam Barth 2011-07-01 11:14:44 PDT
All reviewed patches have been landed.  Closing bug.
Comment 4 Eric Seidel (no email) 2011-07-01 11:21:06 PDT
Comment on attachment 99445 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=99445&action=review

> Tools/Scripts/webkitpy/layout_tests/layout_package/manager.py:671
> +                if self._port.executive().running_pids(self._port.is_crash_reporter):

This call is expensive.  How often do we run this?
Comment 5 Eric Seidel (no email) 2011-07-01 11:21:23 PDT
CCing ossy in case this slows down the Qt bots again.
Comment 6 Adam Barth 2011-07-01 11:30:47 PDT
> This call is expensive.  How often do we run this?

Once a second.  If it turns out to be a problem, I have an idea how to optimize it.
Comment 7 Eric Seidel (no email) 2011-07-01 11:37:24 PDT
It's not entirely clear to me why ReportCrash would destablize the run, since DRT should theoretically be taking up 100% of the CPU before it crashes, so the fact that ReportCrash is taking up 100% of that core in DRT's stead should be OK.

Do we run more DRTs than we have cores?  (Actually I think we run 2x iirc?)
Comment 8 Adam Barth 2011-07-01 11:41:51 PDT
Maybe it's something specific to my system?  ReportCrash make the UI in all my programs sluggish, maybe because causing the kernel to do a lot of work somehow?

Anecdotally, this patch has been a big stability win on my machine.  Crashes in some tests used to trigger other tests to timeout, and that doesn't happen anymore.
Comment 9 Dirk Pranke 2011-07-01 12:50:10 PDT
Comment on attachment 99445 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=99445&action=review

> Tools/Scripts/webkitpy/layout_tests/layout_package/single_test_runner.py:132
> +        self._port.executive().wait_newest(self._port.is_crash_reporter)

You might want to wait for the crash reporter in worker._run_test instead, since you can do in between tests (when the test timeout does not apply).
Comment 10 Adam Barth 2011-07-01 13:21:34 PDT
> You might want to wait for the crash reporter in worker._run_test instead, since you can do in between tests (when the test timeout does not apply).

https://bugs.webkit.org/show_bug.cgi?id=63837