838134 - [10.6] Massive log spam on Mozmill OS X test machines: "plugin-container ... unknown error code: invalid pixel format"

Status: RESOLVED INCOMPLETE

(External Software Affecting Firefox Graveyard :: Flash (Adobe), defect)

Description

[Henrik Skupin [:whimboo][⌚️UTC+1]]

In the last couple of days I had to delete a lot of our old logs given that we get log sizes per testrun of about 100MB on OS X instead of the usual 16KB ones. This completely fills up our Mozmill CI master.

As far as I can remember it hasn't been started lately. It should be around for a while but looks like it got even more worse.

Here some of the output:

2013-02-02 05:17:26.098 firefox-bin[21354:903] invalid pixel format             
2013-02-02 05:17:26.100 firefox-bin[21354:903] invalid context
[..]
Sat Feb  2 05:23:58 release2-osx-106.qa.mtv1.mozilla.com plugin-container[21410]
 <Error>: unknown error code: invalid pixel format


$ glxinfo | grep 'direct rendering'

Tue Feb  5 04:44:55 release2-osx-106.qa.mtv1.mozilla.com glxinfo[30591] <Error>: unknown error code: invalid pixel format
Tue Feb  5 04:44:55 release2-osx-106.qa.mtv1.mozilla.com glxinfo[30591] <Error>: kCGErrorFailure: Set a breakpoint @ CGErrorBreakpoint() to catch errors as they are logged.
X Error of failed request:  GLXBadContext
  Major opcode of failed request:  0 ()
  Serial number of failed request:  16
  Current serial number in output stream:  16

Comment 1

[Steven Michaud [:smichaud] (Retired)]

I've seen both the "invalid pixel format" and "invalid context" messages before -- sometimes with non-Mozilla apps.  But I've never seen them in large numbers.

The glxinfo results seem to indicate something wrong with the machine (hardware and/or software).

> hasn't been started lately

You mean restarted?  Maybe it's best just to try that :-)

Comment 2

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

I am confused -- why is glxinfo output relevant on OSX?  Just to run something that will exercise various parts of the graphics subsystem (even going through X11 and then to OSX)?

Comment 3

[Henrik Skupin [:whimboo][⌚️UTC+1]]

I have seen bug 728822 and this information has been requested there too. So I thought it would have been good to give this extra information. If that doesn't help you I would appreciate feedback how to further debug or fix it.

Comment 4

[Henrik Skupin [:whimboo][⌚️UTC+1]]

I have checked several of those logs and the only machines affected are running OS X 10.6. The other machines running OS X 10.7 and 10.8 are not showing this behavior.

Comment 5

[Steven Michaud [:smichaud] (Retired)]

It'd be very useful to have a regression range here.  Without one there's probably nothing we can do.

If you can't find a regression range for Firefox, I'd start looking for one for Mozmill itself, or for how your 10.6 machines are set up.

Comment 6

[Henrik Skupin [:whimboo][⌚️UTC+1]]

This problem occurs whenever we use embed or object to make use of a plugin. It's only happening for our endurance tests because no other test is using flash at the moment.

One test case which can be used to reproduce it manually is:
http://mozqa.com/data/firefox/plugins/flash/test_swf_object.html

It doesn't matter if the appropriate plugin is enabled or not. The log output will happen all the time. I will check older builds if that's a regression or not.

Comment 7

[Henrik Skupin [:whimboo][⌚️UTC+1]]

So I was wrong. This is indeed the Flash plugin (v 11.5) which is installed on those machines. Whenever it is enabled we get tons of those lines. Given that we haven't seen this in the past one of the lastest major upgrades of Flash should have introduced this problem.

Comment 8

[Henrik Skupin [:whimboo][⌚️UTC+1]]

I went back to the earliest version of Flash supported on OS X 10.6 which is 10_3r181_14 and it can still be seen. So meanwhile I think that this is a bad interaction between Flash and the system, and that most likely an OS X update has triggered that behavior. I can see for log entries even outside of Flash when I start Firefox.

Comment 9

[Henrik Skupin [:whimboo][⌚️UTC+1]]

Given that this will most likely never be fixed on OS X 10.6 we agreed on to disable Flash for our tests. We will not be able to see real endurance results for Flash anymore but that shouldn't matter for 10.6.

I will leave it open for a final statement from Adobe if there is something which can be done or not.

Comment 10

[BMO Automation]

Version and milestone values are being reset to defaults as part of product refactoring.