+++ This bug was initially created as a clone of Bug #743594 +++ We were hit by a rash of OSX 10.7 mochitest failures across different slaves where the common thread is mass bustage and the Bluetooth Setup Assistant showing a crashed window on screen (*NOT* the setup wizard itself). Can we take another look at bug 743594 as I think this shows that the wizard popping up isn't always innocuous. https://tbpl.mozilla.org/php/getParsedLog.php?id=19933482&tree=Mozilla-Inbound https://tbpl.mozilla.org/php/getParsedLog.php?id=19940899&tree=Mozilla-Inbound
Does that "restore running apps after a reboot" misfeature, the one that explains why iCal is running even during suites that aren't the one with the test that opens it, also restore the crash reporter, so we might have a set of slaves which always have that crash report rather than always having the setup assistant?
This is a pretty frequent failure. Can we please get someone on the relenge side to look at this? https://tbpl.mozilla.org/php/getParsedLog.php?id=19952759&tree=Mozilla-Inbound
I tried all the steps in bug 743594 to disable the bluetooth setup assistant on 10.7 and they didn't work. I also tried many more steps that I found on various Apple forums to no avail. Still looking...
Still haven't found a solution to fix this. I looked at tbpl and this doesn't seem to be happening recently, but as mentioned, it's an intermittent problem.
The command to disable it is defaults write "/Library/Preferences/com.apple.Bluetooth" BluetoothAutoSeekPointingDevice -bool NO defaults write "/Library/Preferences/com.apple.Bluetooth" BluetoothAutoSeekKeyboard -bool NO I've tested it on a slave and it does indeed disable the Bluetooth setup assistant for keyboard and mouse. I'll write a patch for puppet
Created attachment 717222 [details] [diff] [review] patch I've tested in staging
Actually the previous patch just disabled the bluetooth setup assistant. The resume state should also be disabled, so the crashed applications don't start up again. I'll attach another patch.
Comment on attachment 717222 [details] [diff] [review] patch Ooops, wanted to remove r?
Created attachment 717268 [details] [diff] [review] patch Okay this patch does two things 1) disables the find a keyboard or mouse via bluetooth setup assistant 2) disables the restoring apps upon relogin I've tested this in staging on talos-r4-lion-001 and it works after a sudo reboot.
Created attachment 717277 [details] [diff] [review] patch to make bool value consistent as false compared to previous patch
Comment on attachment 717277 [details] [diff] [review] patch and deployed to production. I'm watching some machines that have puppetized since then and are running mochitests.
If the error in comment 27 is still occurring, then it's not related to 1) apps being restored after a restart or 2) the bluetooth setup assistant randomly starting. I took talos-r4-lion-008, the machine from comment 27, disabled it in slavealloc, rebooted it, and then started a bunch of applications. Then I rebooted it via sudo reboot, and it didn't restart any applications upon reboot, just like the slave I used in staging. One thing I did notice is that there are 150+ icons on the desktop for text files with different timestamps on them. Perhaps they are causing problems when the tests try to find focus? Just a guess.
It's always hard to tell whether something intermittent has gone away on a weekend, but given the relative frequency Thursday and Friday and the number of jobs we ran Saturday, I would have expected around five Saturday if it was still happening, and we saw zero. I just figured comment 27 was the last one that managed to sneak in before you told it to stop. The text files are a fun, but separate problem: we think that the test which was leaving them no longer leaves them, but we also think that it's possible that the desktop background (the combination of the actual background and the hundreds of text file icons) affects perf test numbers. You would think that would make us want to remove them all at once, leaving a clear non-code mark that the numbers changed from that, but in fact we're reimaging slaves without them (well, I presume we are, unless we reimage from a refimage that has a desktop-full?) and presuming that they aren't coming back on those reimaged slaves, letting the hypothetical perf improvement roll out as noise. For bonus fun, I've seen screenshots with a pile of .part files, so we probably have other tests which clean up fine when they pass, but leave a .part when they fail, so even post-reimage we should expect the desktop to slowly fill up again, with parts instead of text.
Okay thanks for the clarification regarding the text and .part files. I guess this bug can be closed now :-)
talos-r4-lion-001 has a note saying that is being used for this bug. Is the machine ready to go back to the pool? Does it need re-imaging? Thanks in advance!
Armen: I'm using it for bug 786679 now, I've updated slavealloc.
(In reply to Phil Ringnalda (:philor) from comment #30) > The text files are a fun, but separate problem: we think that the test which > was leaving them no longer leaves them, but we also think that it's possible > that the desktop background (the combination of the actual background and > the hundreds of text file icons) affects perf test numbers... Filed bug 851123 for this.