Bug 51807 – fast/blockflow/broken-ideograph-[small-caps|font].html sometimes fails on Leopard buildbots (due to ATSServer leak)

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

[Adam Barth]

http://build.webkit.org/results/Leopard%20Intel%20Debug%20(Tests)/r74885%20(25337)/results.html

This test has been failing for many days.

[Adam Barth]

Failure seen as early as http://build.webkit.org/builders/Leopard%20Intel%20Debug%20%28Tests%29/builds/25125

Sat Dec 25 09:57:48 2010

[Adam Barth]

Skipped in http://trac.webkit.org/changeset/74888

[Adam Barth]

As far as I can tell, this test started failing during some kind of buildbot maintenance.  Prior to build 25125, there don't seem to be any failures, but 25125 seems to be the first build with a new buildbot configuration.  The same revision was tested in a prior build, which did not show the error (although it did how other failing tests which magically disappeared).

Thoughts?

[Adam Barth]

Now that I've skipped this test, 
fast/blockflow/broken-ideographic-font.html has started failing instead, which is the next test alphabetically.  That would seem to implicate box-shadow-vertical-rl.html, which is the previous test alphabetically.

[Adam Barth]

I unskipped the test in http://trac.webkit.org/changeset/75683.  The problem seems likely to be caused by an earlier test.

[Adam Roben (:aroben)]

This test started failing again on the Leopard bots today around r81371/r81372. None of the revisions around then seem related. The test seems to be failing in both Debug and Release and on all of the slaves, so it's not configuration-specific.

[Adam Roben (:aroben)]

Here's an example of the current failure: http://build.webkit.org/results/Leopard%20Intel%20Release%20(Tests)/r81376%20(29906)/fast/blockflow/broken-ideographic-font-pretty-diff.html

[Adam Roben (:aroben)]

This seems to have stopped failing again.

[Jessie Berlin]

<rdar://problem/9273894>

[Jessie Berlin]

The test has resumed failing: http://build.webkit.org/results/Leopard%20Intel%20Release%20(Tests)/r84943%20(31499)/results.html

I am not sure when this bout of failing began.

[Andy Estes]

*** Bug 59649 has been marked as a duplicate of this bug. ***

[Andy Estes]

(In reply to comment #10)
> The test has resumed failing: http://build.webkit.org/results/Leopard%20Intel%20Release%20(Tests)/r84943%20(31499)/results.html
> 
> I am not sure when this bout of failing began.

As far back on the bot as I can see, which is currently http://build.webkit.org/builders/Leopard%20Intel%20Release%20%28Tests%29/builds/31371

[Andy Estes]

I'm going to skip fast/blockflow/broken-ideograph-font.html on Leopard. It doesn't seem related to box-shadow-vertical-rl.html since broken-ideograph-small-caps.html isn't skipped and doesn't fail. It's hard to say if this is a failure or a rebaseline situation without seeing pixel results.

[Andy Estes]

Skipped in http://trac.webkit.org/changeset/85130

[Adam Roben (:aroben)]

*** Bug 59780 has been marked as a duplicate of this bug. ***

[Adam Roben (:aroben)]

It looks like these test failures are being caused by an ATSServer leak. ATSServer slowly uses more and more memory and eventually these tests start failing. Cycling ATSServer using launchctl stop/start seems to fix things.

Dan Bernstein suggests that we cycle ATSServer at the start or end of each test run to avoid this issue.

[Adam Roben (:aroben)]

Unskipped the tests in r85306.

[Adam Roben (:aroben)]

Created attachment 91645 [details]
Stop and restart ATSServer before each test run on Leopard

[Adam Roben (:aroben)]

Committed r85307: <http://trac.webkit.org/changeset/85307>

[WebKit Review Bot]

http://trac.webkit.org/changeset/85306 might have broken SnowLeopard Intel Release (Build)

[WebKit Review Bot]

http://trac.webkit.org/changeset/85307 might have broken SnowLeopard Intel Release (Build)