961075 - Add the build step or else process name to buildbot's generic command timed out failure strings

Status: RESOLVED FIXED

(Release Engineering :: General, defect)

Description

[Ed Morley [:emorley]]

Bug 778688 comment 38 covers a number of intermittent failures where we have the generic log output:
"command timed out: N seconds without output, attempting to kill"

Whilst we've added this to the TBPL regexes, so we can use TBPL's bug suggestion feature - the messages are generic, so many suggestions are shown, eg:

https://bugzilla.mozilla.org/buglist.cgi?quicksearch=command%20timed%20out%20keywords%3Aintermittent

Whilst I'd prefer the worst of these failure modes to be handled by the mozharness/test harness/... itself, we're always going to have edge cases where timeouts occur and it's not worth adding TBPL-compatible failure messages to that script.

As such, I was thinking we should prefix the timeout messages with the build step name or else the process name (former preferred) here:

http://hg.mozilla.org/build/buildbot/file/6d8bc44b5874/slave/buildslave/runprocess.py#l657
   657     def doTimeout(self):
   658         self.timer = None
   659         msg = "command timed out: %d seconds without output" % self.timeout
   660         self.kill(msg)
   661 
   662     def doMaxTimeout(self):
   663         self.maxTimer = None
   664         msg = "command timed out: %d seconds elapsed" % self.maxTime
   665         self.kill(msg)

Now I know buildbot patches are generally a bit more awkward - so don't know if you think we would need to upstream first - or even whether they'd take it the change?

Dustin, what do you think? :-)

Comment 1

[Dustin J. Mitchell [:dustin] (he/him)]

I'd like to see that upstream, sure.

Shipping a change to non-Windows systems is pretty easy - it's done with Puppet.  Windows is still hard.

Comment 2

[Ed Morley [:emorley]]

Upstream PR:
https://github.com/buildbot/buildbot/pull/1130

Comment 4

[Ed Morley [:emorley]]

Attachment: For timeouts include the command being run in the failure string

Backport of upstream commit:
https://github.com/buildbot/buildbot/commit/b66cc92ee5a7ec0e923c7ab2055a38d07ac3515c

I've checked that we won't break any of the current regex:
https://mxr.mozilla.org/build/search?string=command%2Btimed%2Bout

Comment 5

[Dustin J. Mitchell [:dustin] (he/him)]

Comment on attachment 8408096 [details] [diff] [review]
For timeouts include the command being run in the failure string

Assuming you're confident that fake_command works the same way in 0.8.2, this looks just like the patch I merged :)

Comment 6

[Ed Morley [:emorley]]

Landed on default & transplanted to production-0.8, since there are buildbot master changes that require a restart that did not want to be merged across just yet.

https://hg.mozilla.org/build/buildbot/rev/6ac947fa5721
https://hg.mozilla.org/build/buildbot/rev/8a9e33843c3f

Comment 7

[Ed Morley [:emorley]]

This is still waiting for bug 1009584 to actually be deployed, but closing this so it still appearing in bugzilla-todos.

Comment 8

[Ed Morley [:emorley]]

This is deployed on !Windows; bug 1042597 will take care of Windows.

@Sheriffs: Note this bug changes "command timed out: 2400 seconds without output" to "command timed out: 2400 seconds without output running <cmd...>"