On Tue, Dec 10, 2013 at 5:05 PM, Karl Tomlinson <mozn...@karlt.net> wrote: > > Yes, sounds like all processes in the test are being killed. Most tests > have a shorter time out to kill with signals that generate stacks, > before this higher level time out kicks in.
Here's an excerpt from the log, with tons of setup/build/valgrind details omitted: ========= Started 'mock_mozilla -r ...' failed (results: 2, elapsed: 1 hrs, 5 mins, 9 secs) (at 2013-12-10 16:48:20.585382) ========= mock_mozilla -r mozilla-centos6-x86_64 --cwd /builds/slave/m-cen-l64-valgrind-00000000000/. --unpriv --shell '/usr/bin/env ... ... ... --1757:1:mallocfr newSuperblock at 0x42E8B3000 (pszB 1048544) owner VALGRIND/exectxt --1757:1:mallocfr newSuperblock at 0x42E9BF000 (pszB 41942 command timed out: 1200 seconds without output, attempting to kill program finished with exit code 247 elapsedTime=3908.864089 ========= Finished 'mock_mozilla -r ...' failed (results: 2, elapsed: 1 hrs, 5 mins, 9 secs) (at 2013-12-10 17:53:30.356878) ========= Two things to note here. - The last Valgrind line (starting with "--1757:1:mallocfr") is truncated -- compare it to the previous line. - The "Started" line states how long the whole step (which timed out) took. So all this output clearly wasn't written to this file until afterwards -- i.e. it wasn't streaming the output on the fly. I've looked at buildbot's code and I can see it is responsible for this, and that it has 64KB buffers that it uses for this output. It's a real shame the output is buffered; I'd dearly love to see what is the last output before the hang. Nick _______________________________________________ dev-platform mailing list dev-platform@lists.mozilla.org https://lists.mozilla.org/listinfo/dev-platform
