Bug 227455

Summary:

[webkitcorepy] Fix race condition in TaskPool unittests

Product:

WebKit

Reporter:

Jonathan Bedard <jbedard>

Component:

Tools / Tests

Assignee:

Jonathan Bedard <jbedard>

Status:

RESOLVED FIXED

Severity:

Normal

CC:

aakash_jain, darin, dewei_zhu, ryanhaddad, slewis, webkit-bug-importer

Priority:

Keywords:

InRadar

Version:

WebKit Nightly Build

Hardware:

Unspecified

OS:

Unspecified

Attachments:

Description	Flags
Patch	none
Patch	ews-feeder: commit-queue-

Jonathan Bedard

Reported 2021-06-28 11:16:38 PDT

webkitcorepy.tests.task_pool_unittest.TaskPoolUnittest.test_invalid_shutdown is racey on EWS, because it relies on the parent process terminating it's children before the children have a chance to finish their computation.

Attachments
Patch (1.44 KB, patch) 2021-06-28 11:21 PDT, Jonathan Bedard	no flags	Details Formatted Diff Diff
Patch (1.52 KB, patch) 2021-07-28 08:39 PDT, Jonathan Bedard	ews-feeder: commit-queue-	Details Formatted Diff Diff
Show Obsolete (1) View All Add attachment proposed patch, testcase, etc.

Radar WebKit Bug Importer

Comment 1 2021-06-28 11:16:52 PDT

<rdar://problem/79873003>

Jonathan Bedard

Comment 2 2021-06-28 11:21:42 PDT

Created attachment 432410 [details] Patch

Jonathan Bedard

Comment 3 2021-06-28 11:22:05 PDT

Failures like this one https://ews-build.webkit.org/#/builders/5/builds/55039/steps/11/logs/stdio are the ones I'm trying to address.

Jonathan Bedard

Comment 4 2021-06-28 11:23:55 PDT

History of this test: https://results.webkit.org/?suite=webkitpy-tests&test=webkitcorepy.tests.task_pool_unittest.TaskPoolUnittest.test_invalid_shutdown

dewei_zhu

Comment 5 2021-06-28 12:02:32 PDT

Comment on attachment 432410 [details] Patch r=me, this seems to be a mitigation, but theoretically doesn't fix the race itself.

Jonathan Bedard

Comment 6 2021-06-28 12:06:14 PDT

(In reply to dewei_zhu from comment #5) > Comment on attachment 432410 [details] > Patch > > r=me, this seems to be a mitigation, but theoretically doesn't fix the race > itself. It doesn't fix the race, but it's hard to see how we could (and still retain the test). The test is basically verifying that the parent can and does terminate a child process if it's taking too long to complete. We clearly want some kind of limit on how long the child runs, in case the parent actually isn't terminating the child

EWS

Comment 7 2021-06-28 12:11:39 PDT

Committed r279345 (239213@main): <https://commits.webkit.org/239213@main> All reviewed patches have been landed. Closing bug and clearing flags on attachment 432410 [details].

Jonathan Bedard

Comment 8 2021-07-28 08:39:48 PDT

Reopening to attach new patch.

Jonathan Bedard

Comment 9 2021-07-28 08:39:49 PDT

Created attachment 434429 [details] Patch

Jonathan Bedard

Comment 10 2021-07-28 08:40:59 PDT

The original change didn't fix our flakiness. This test isn't super valuable in the first place, I think we should just delete it.

Darin Adler

Comment 11 2021-07-28 10:16:53 PDT

Comment on attachment 434429 [details] Patch How does deleting a function fix a race condition?

Darin Adler

Comment 12 2021-07-28 10:17:26 PDT

Comment on attachment 434429 [details] Patch Oh, I get it, deleting a test (not just a function).

dewei_zhu

Comment 13 2021-07-28 17:57:48 PDT

Comment on attachment 434429 [details] Patch r=me

EWS

Comment 14 2021-07-28 18:48:53 PDT

Committed r280412 (240052@main): <https://commits.webkit.org/240052@main> All reviewed patches have been landed. Closing bug and clearing flags on attachment 434429 [details].

Note You need to log in before you can comment on or make changes to this bug.