Bug 216035 – [Win] The WPT server is not always starting when running layout tests

• Status: RESOLVED
• Resolution: FIXED
• Component: Tools / Tests

[Per Arne Vollan]

The Web platform test server is not always starting when running layout tests using Cygwin.

[Per Arne Vollan]

Created attachment 407679 [details]
Patch

[Per Arne Vollan]

Created attachment 407684 [details]
Patch

[Alexey Proskuryakov]

Comment on attachment 407684 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=407684&action=review

> Tools/ChangeLog:10
> +        issue by waiting up to 60 seconds with polling every 10th second. A unit test had to be modified, since it was overriding the

Why is it desirable to poll rarely? This will delay starting tests by 5 seconds on average.

[Per Arne Vollan]

(In reply to Alexey Proskuryakov from comment #3)
> Comment on attachment 407684 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=407684&action=review
> 
> > Tools/ChangeLog:10
> > +        issue by waiting up to 60 seconds with polling every 10th second. A unit test had to be modified, since it was overriding the
> 
> Why is it desirable to poll rarely? This will delay starting tests by 5
> seconds on average.

Polling frequently seems to prevent the WPT server on Cygwin to start in a timely manner.

Perhaps it would be better to make this change Cygwin-only?

[Alexey Proskuryakov]

Interesting. Or perhaps whatever prevents it from starting can be fixed. I don't know much about how the server works.

[youenn fablet]

Comment on attachment 407684 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=407684&action=review

>>> Tools/ChangeLog:10
>>> +        issue by waiting up to 60 seconds with polling every 10th second. A unit test had to be modified, since it was overriding the
>> 
>> Why is it desirable to poll rarely? This will delay starting tests by 5 seconds on average.
> 
> Polling frequently seems to prevent the WPT server on Cygwin to start in a timely manner.
> 
> Perhaps it would be better to make this change Cygwin-only?

I think it is ok to increase the timer for all platforms.
Increasing the polling time for Cygwin-only is probably fine as a stop-gap but we should try to understand why that happens.
@gsnedders might have an idea of how other CI are using WPT with Cygwin.

[Sam Sneddon [:gsnedders]]

> @gsnedders might have an idea of how other CI are using WPT with Cygwin.

They aren't. Hence I don't really have any idea here!

[Per Arne Vollan]

(In reply to Alexey Proskuryakov from comment #5)
> Interesting. Or perhaps whatever prevents it from starting can be fixed. I
> don't know much about how the server works.

I am not too familiar with the WPT server either :) I can look into why it appears to take longer time to start up on Windows. While that investigation is ongoing, I can upload a patch to extend the wait time for Cygwin only, so the affected bot(s) can resume running layout tests.

Thanks for reviewing!

[Per Arne Vollan]

Created attachment 407763 [details]
Patch

[Per Arne Vollan]

(In reply to youenn fablet from comment #6)
> Comment on attachment 407684 [details]
> Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=407684&action=review
> 
> >>> Tools/ChangeLog:10
> >>> +        issue by waiting up to 60 seconds with polling every 10th second. A unit test had to be modified, since it was overriding the
> >> 
> >> Why is it desirable to poll rarely? This will delay starting tests by 5 seconds on average.
> > 
> > Polling frequently seems to prevent the WPT server on Cygwin to start in a timely manner.
> > 
> > Perhaps it would be better to make this change Cygwin-only?
> 
> I think it is ok to increase the timer for all platforms.
> Increasing the polling time for Cygwin-only is probably fine as a stop-gap
> but we should try to understand why that happens.
> @gsnedders might have an idea of how other CI are using WPT with Cygwin.

As a stop-gap, I have uploaded a patch which only changes the behavior for Cygwin. I can investigate why the server startup time seems to be longer on Windows. I also seem to remember that WPT tests are quite slow on Windows. I am not sure if we reached a conclusion as to why, the last time we looked into this.

Thanks for reviewing!

[Per Arne Vollan]

Comment on attachment 407763 [details]
Patch

I believe the GTK failures are unrelated to this patch.

[EWS]

Committed r266460: <https://trac.webkit.org/changeset/266460>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 407763 [details].

[Radar WebKit Bug Importer]

<rdar://problem/68211737>