1893807 - Recent intermittent unexpected passes on Windows in WPT IndexedDB/structured-clone.any.worker.html?101-last

Status: RESOLVED FIXED

(Core :: Storage: IndexedDB, task)

Description

[Thomas Wisniewski [:twisniewski]]

Recently this WPT began to unexpectedly pass its Not serializable: symbol: Symbol(desc) case on Windows (which Jari says is due to recent nsThreadPool changes).

Our current test expectations for Windows there are [FAIL, NOTRUN]. After a chat with :janv and :jjalkanen, we decided we should adjust them to [PASS, FAIL, NOTRUN].

Comment 1

[Thomas Wisniewski [:twisniewski]]

Attachment: Bug 1893807 - update expectations for structured-clone.any.worker.html?101-last; r?jari

Comment 2

[Jens Stutte [:jstutte]]

I assume this is related to:

bug 1878149 - NO and YES: there used to be a defect which caused this test to hang at the end (a bug, might still be relevant, root cause should be removed by nsThreadPool or D206588, we used to add idle timers when closing threads with the IDB custom thread pool, this test case has > 100 instances and it caused a shutdown hang), otherwise it fails because filelist and symbol cloning is incomplete (enhancement)

from bug 1878131 comment 2.

Comment 3

[Jan Varga [:janv]]

Isn't that comment about TEST-UNEXPECTED-TIMEOUT ?
I tried to run structured-clone.any.worker.html locally on mac and I got a timeout.

Comment 4

[Thomas Wisniewski [:twisniewski]]

Would it be worth updating the expectation at the same time to also include TIMEOUT for Mac (and/or other platforms), as well as PASS for Windows?

Comment 5

[Jens Stutte [:jstutte]]

I was just noting in bug 1893912 that quite some IndexedDB WPT tests seem to run close to our timeout limits, and on our MAC test machines (which are slow) this might be even more a problem. Can we just raise the timeout value for WPT tests to see if that solves some issues?

Comment 6

[Jens Stutte [:jstutte]]

(In reply to Jan Varga [:janv] from comment #3)

I tried to run structured-clone.any.worker.html locally on mac and I got a timeout.

Oh, locally on mac means probably on a fast machine.

Comment 7

[Andrew Sutherland [:asuth] (he/him)]

(In reply to Jens Stutte [:jstutte] from comment #5)

I was just noting in bug 1893912 that quite some IndexedDB WPT tests seem to run close to our timeout limits, and on our MAC test machines (which are slow) this might be even more a problem. Can we just raise the timeout value for WPT tests to see if that solves some issues?

It's really bizarre that these tests should ever time out. Maybe there's horrible I/O happening on the test runners? One notable thing is that structured-clone.any.js is using {durability: 'relaxed'} so when bug 1883045 lands maybe the tests will all get a lot happier?

Comment 8

[Jens Stutte [:jstutte]]

IIUC the test runners are shared resources that host several VMs on the same hardware, another VM causing heavy IO load could definitely interfere, I think. But the captured profiles do not show any anomaly on the IO track, so I am quite unsure if it is more probable we see an over-provisioning of virtual CPU cores in relation to the physical ones that slows down the entire execution when all VMs are on high CPU load, not just IO?

Comment 9

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/5c04f92b37ae
update expectations for structured-clone.any.worker.html?101-last; r=dom-storage-reviewers,asuth

Comment 10

[Serban Stanca [:SerbanS]]

https://hg.mozilla.org/mozilla-central/rev/5c04f92b37ae