Bug 36380 – REGESSION: websocket/tests/frame-lengths.html times out on Tiger bot

• Component: Tools / Tests
• Status: RESOLVED FIXED

[Eric Seidel (no email)]

REGESSION: websocket/tests/frame-lengths.html times out on Tiger bot

This has been happening for days, if not weeks and has rendered the bot useless.

We need to find resolution to this soon, even if that means skipping this test on tiger.

Two recent failures:
http://build.webkit.org/results/Tiger%20Intel%20Release/r56247%20(9828)/websocket/tests/frame-lengths-diffs.txt
http://build.webkit.org/results/Tiger%20Intel%20Release/r56243%20(9825)/websocket/tests/frame-lengths-diffs.txt

This fails every Tiger build.

[Eric Seidel (no email)]

This has been failing since at least 2/22/09:
https://bugs.webkit.org/show_bug.cgi?id=35041#c2

I think we should just skip this test on Tiger since it's been ignored for nearly a month already.

[Eric Seidel (no email)]

This test was added by Alexey 3 months ago:
http://trac.webkit.org/browser/trunk/LayoutTests/websocket/tests/frame-lengths.html

[Alexey Proskuryakov]

This test has always timed out on slow machines. I think that we need to increase the timeout.

[Eric Seidel (no email)]

How do we do that?

[Alexey Proskuryakov]

Just change these lines in run-webkit-tests, and make matching changes to DRT, I guess.

# DumpRenderTree has an internal timeout of 15 seconds, so this must be > 15.
my $timeoutSeconds = 20;

[Eric Seidel (no email)]

That seems wrong.  Why would we want to make all tests possibly take longer when timing out, instead of modifying this single test?

new-run-webkit-tests has a way of marking a single test as slow, but we don't have that ability for run-webkit-tests yet.

Seems we should either cut this test down to size or skip it on slow machines instead.  No?

[Alexey Proskuryakov]

I'd agree if only one test was timing out, but in fact, many tests are timing out on slower machines. Didn't we discuss this on webkit-dev a few days ago?

How much of a problem would increasing the timeout be in practice? Normally, no tests are timing out, so such a change wouldn't affect anything.

[Eric Seidel (no email)]

I must have missed the webkit-dev discussion.

I reverse my position.  r=me to increase the timeout.  Once we have the new-webkit-tests technology of individual "slow tests", we can push the timeout back down.

[Alexey Proskuryakov]

Created attachment 51183 [details]
proposed patch

[Alexey Proskuryakov]

*** Bug 35041 has been marked as a duplicate of this bug. ***

[Eric Seidel (no email)]

A little trivia:
new-webkit-tests's timeout limit is 6 seconds for release and 12 for debug. :)  You can mark individual tests as SLOW and then they get 5X the base timeout.

[Alexey Proskuryakov]

Committed revision 56262.

[Alexey Proskuryakov]

> A little trivia:

Is there a global override? Running the tests on PPC/Debug, or on a mobile platform would be a pain without one.

[Eric Seidel (no email)]

The architecture is such that one could have a per-port timeout.  Although that's not currently wired up.

You can also (already) pass a timeout over the command line.  See:
new-run-webkit-tests --help