Bug 227269 – [macOS] Suspend WebProcesses that are in the process cache

• Status: RESOLVED
• Resolution: FIXED
• Component: WebKit2

[Chris Dumez]

Suspend WebProcesses that are in the process cache on macOS to make sure they use no CPU.

[Chris Dumez]

Created attachment 431993 [details]
Patch

[Geoffrey Garen]

Comment on attachment 431993 [details]
Patch

r=me

[EWS]

Committed r279155 (239059@main): <https://commits.webkit.org/239059@main>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 431993 [details].

[Radar WebKit Bug Importer]

<rdar://problem/79639863>

[Chris Dumez]

Reverted r279155 for reason:

Caused a PLT5 regression

Committed r279411 (239276@main): <https://commits.webkit.org/239276@main>

[Chris Dumez]

I can reproduce the issue on PLT5 locally. I am investigating. 2 of the pages hang for an extended period of time of warm loads (but eventually recover). It looks like the cached process ends up exiting in this case and the load eventually happens in a fresh process. I don't see any crash report though.

[Chris Dumez]

(In reply to Chris Dumez from comment #6)
> I can reproduce the issue on PLT5 locally. I am investigating. 2 of the
> pages hang for an extended period of time of warm loads (but eventually
> recover). It looks like the cached process ends up exiting in this case and
> the load eventually happens in a fresh process. I don't see any crash report
> though.

Oh, I think the pages that cause trouble are the ones with service workers. If I look at the logs, I see this logging right after we resume the cached process:
com.apple.WebKit	2021-07-01 14:51:37.071485 -0700	ServiceWorker	error	28186	com.apple.WebKit.WebContent.Development	Failed to terminate service worker with identifier 1535, killing the service worker process

And then the Cached WebProcess dies.

[Chris Dumez]

Created attachment 432739 [details]
Patch

[Geoffrey Garen]

Comment on attachment 432739 [details]
Patch

r=me

[EWS]

Committed r279495 (239347@main): <https://commits.webkit.org/239347@main>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 432739 [details].

[Chris Dumez]

Reverted r279495 for reason:

Seems to have caused a ~2% PLT5 regression

Committed r279583 (239414@main): <https://commits.webkit.org/239414@main>

[Chris Dumez]

Created attachment 432962 [details]
Patch

[Chris Dumez]

(In reply to Chris Dumez from comment #12)
> Created attachment 432962 [details]
> Patch

New iteration, I am currently evaluating its impact on PLT5.

[Chris Dumez]

Created attachment 432966 [details]
Patch

[Chris Dumez]

Comment on attachment 432966 [details]
Patch

A/B bots say no PLT5 regression with this latest iteration.

[Geoffrey Garen]

Comment on attachment 432966 [details]
Patch

r=me

[EWS]

Committed r279644 (239457@main): <https://commits.webkit.org/239457@main>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 432966 [details].