Bug 249444
| Summary: | WK1 tests are taking a long time to run 'extract-test-results' | ||
|---|---|---|---|
| Product: | WebKit | Reporter: | Matt Woodrow <mattwoodrow> |
| Component: | Tools / Tests | Assignee: | Nobody <webkit-unassigned> |
| Status: | NEW | ||
| Severity: | Normal | CC: | achristensen, ap, jbedard, ryanhaddad, webkit-bug-importer |
| Priority: | P2 | Keywords: | InRadar |
| Version: | WebKit Nightly Build | ||
| Hardware: | Unspecified | ||
| OS: | Unspecified | ||
| Bug Depends on: | 249614, 249658, 249724 | ||
| Bug Blocks: | |||
Matt Woodrow
It looks like extracting test results is frequently taking 10+ minutes, adding up to a lot when we often run tests twice.
My build: https://ews-build.webkit.org/#/builders/73/builds/22639
Others picked at random:
https://ews-build.webkit.org/#/builders/73/builds/22631 (17 minutes!)
https://ews-build.webkit.org/#/builders/73/builds/22619
| Attachments | ||
|---|---|---|
| Add attachment proposed patch, testcase, etc. |
Alexey Proskuryakov
Looks like we have 35K files in the archive, most of them stderr output of this kind:
2022-12-15 18:20:33.500 DumpRenderTree[27762:191348898] nil host used in call to allowsSpecificHTTPSCertificateForHost
2022-12-15 18:20:33.500 DumpRenderTree[27762:191348898] nil host used in call to allowsAnyHTTPSCertificateForHost:
Not sure if something changed recently, but we should definitely fix this.
Alexey Proskuryakov
What's also weird is that these are not in results.html.
Are we just failing to clean the results directory between runs?
Alexey Proskuryakov
No - checking the dates of -stderr.txt files on a bot, that's not what is happening, they are all recent.
Jonathan Bedard
I also was able to reproduce this problem very easily on a local build. I'm going to see if I can figure out the cause of the problem with code inspection, but we might consider attempting to bisect.
Jonathan Bedard
From testing locally, looks like there are 2 problems:
1) It's somehow possible for us to get empty allowed hosts or local host aliases. I still haven't figured out _how_ that's possible, though, because according to the code, it shouldn't be possible.
2) Clearly out our database caches is racy because these caches are shared between all workers. This race seems so obvious, though, that I'm unclear what I'm missing.
Jonathan Bedard
Not convinced https://bugs.webkit.org/show_bug.cgi?id=249614 is problem, but it does seem to be a problem.
Alexey Proskuryakov
Turns out that these were a known bug fixed in macOS Ventura, so we should just filter out output on older OSes:
> 2022-12-15 18:20:33.500 DumpRenderTree[27762:191348898] nil host used in call to allowsSpecificHTTPSCertificateForHost
> 2022-12-15 18:20:33.500 DumpRenderTree[27762:191348898] nil host used in call to allowsAnyHTTPSCertificateForHost:
I didn't check if this resolves enough of the spurious logs.
Radar WebKit Bug Importer
<rdar://problem/103657108>