587378 - mozilla-central talos Tp4 failures not predicted by try server

Status: RESOLVED INVALID

(Release Engineering :: General, defect)

Description

[Josh Aas]

Attachment: failure log

When I landed "fix v2.6" for bug 578868 on mozilla-central I got a failure that I did not get on try server.

I'm not sure if the problem is my patch or the machine, but in any case I don't think I should be getting a failure on m-c that I can't catch on try.

Comment 1

[Josh Aas]

Is it possible that the machine this failed on does not have "libXt.so" and "libXext.so" installed properly? These are required for loading Flash on Linux.

Comment 2

[[e-mail] (:bhearsum)]

The tests are run on the exact same pool, so I would suspect code or merge issues over machine issues, but I'm happy to help debug this. I see these files on the machines:
[cltbld@talos-r3-fed64-032 ~]$ ls -l /usr/lib64/libXext.so.6
lrwxrwxrwx 1 root root 16 2010-01-19 22:04 /usr/lib64/libXext.so.6 -> libXext.so.6.4.0
[cltbld@talos-r3-fed64-032 ~]$ ls -l /usr/lib64/libXt.so.6
lrwxrwxrwx 1 root root 14 2010-01-19 22:04 /usr/lib64/libXt.so.6 -> libXt.so.6.0.0

Comment 3

[[e-mail] (:bhearsum)]

I'm helping Josh look into this.

Comment 4

[[e-mail] (:bhearsum)]

Possibly related to bug 587815

Comment 5

[Josh Aas]

I figured out what is going on. There is an error in the patch causing the browser to crash. The crash only happens when you try to load a plugin library of the wrong architecture. When talos Tp4 runs happen on mozilla-central 64-bit Linux boxes there is a 32-bit Flash plugin installed to the profile which triggers that condition.

This bug is really about the fact that I couldn't see this happen on the try server and thus I got caught by surprise during the mozilla-central run. Seems like if it can fail on mozilla-central then we should be able to see the failure on try server. Maybe this means try server should run Talos Tp4 and whatever else mozilla-central runs that try doesn't.

Comment 6

[[e-mail] (:bhearsum)]

I'm going to call this INVALID, then. We purposely turned off Talos by default on try pushes in bug 579573. This means that there *will* be occasional things such as this where a Talos run doesn't seem necessary on try, but a freak issue pops on when the push hits mozilla-central. I think that's OK given the improvement in wait times and end-to-end times for a try push.

At some point in the future we may run Talos or some subset of it on Try again, but not at this time.