Bug 246472

Summary:	[GTK] Flaky timeouts in some WTF API tests
Product:	WebKit	Reporter:	Lauro Moura <lmoura>
Component:	WebKitGTK	Assignee:	Nobody <webkit-unassigned>
Status:	NEW ---
Severity:	Normal	CC:	abstractmachines, bugs-noreply, clopez
Priority:	P2
Version:	WebKit Nightly Build
Hardware:	Unspecified
OS:	Unspecified

Description Lauro Moura 2022-10-13 14:04:55 PDT

The following tests are frequent flaky timeouts in the GTK-X11-Release bots:

WTF_Condition.OneProducerTenConsumersHundredSlotsNotifyAll
WTF_Condition.OneProducerTenConsumersHundredSlotsNotifyOne
WTF_RunLoop.ManyTimes

In the post-commit bot, it started around https://commits.webkit.org 254468@main

Comment 1 Lauro Moura 2022-10-13 16:29:01 PDT

Gardened in https://commits.webkit.org/255511@main

Comment 2 Amanda Falke 2023-08-10 11:26:31 PDT

A few WTF_Condition* tests are frequent flaky timeouts in GTK release bots;
gardened in https://commits.webkit.org/266748@main.

Comment 3 Carlos Alberto Lopez Perez 2023-11-09 11:57:43 PST

I have updated a bunch of expectations here from "gtk" to "all" as several of this failures were happening also on the EWS for WPE API tests and also added two new expected timeouts on 270468@main


This API tests (TestWTF) are kind of core, so I would expect that any issue that happens on GTK will happen also on WPE and viceversa. maybe we should use "all" for every expectation on this sub-tests. Just a thought.

A bit more info: I suspect some of the timeouts are simply slow tests.

For example the test WTF_Lock.ManyContendedLongSections takes 5 seconds to run on a Ryzen 9 5950X 16-Core with Kernel 6.0 but it runs on just 1 second on a i7-3770 4-core with Kernel 5.10

Looking at the strace you can see that the time is all spent on doing futex syscall. I suspect the difference on performance maybe has to be with this syscalls that for a reason that I don't know are faster on one machine than the other. It can be also related to the kernel version.