common/Thread:pthread_attr_destroy(thread_attr) when done with it
Crash signature (v1):
Crash signature (v2):
#3 Updated by Loïc Dachary over 5 years ago
- Assignee changed from Sage Weil to Loïc Dachary
(11:11:57 AM) ***loicd looking at http://qa-proxy.ceph.com/teuthology/smithfarm-2015-10-12_02:31:17-rados-firefly-backports---basic-multi/1102086/remote/burnupi33/log/valgrind/osd.5.log.gz (11:13:33 AM) loicd: smithfarm: did you check if this job succeeds with the firefly branch as it is ? (11:14:40 AM) smithfarm: loicd: nope (11:14:53 AM) BranchPredictor: smithfarm: it might me related to https://github.com/ceph/ceph/pull/5433 (11:15:09 AM) BranchPredictor: (i mean, this pr might fix that) (11:15:44 AM) loicd: looks like a good candidate indeed (11:16:26 AM) smithfarm: loicd, BranchPredictor: that fix is in firefly-backports https://github.com/ceph/ceph/pull/5989 (11:16:28 AM) loicd: smithfarm: I would dig the past firefly runs to see how this job did in the past two weeks (i.e. after the valgrind check was re-enabled) (11:17:03 AM) loicd: smithfarm: damn, that *was* a good candidate ;-) (11:17:40 AM) smithfarm: loicd: ok, firefly is without that fix, so if the valgrind test succeeds there, PR5989 might be the culprit (11:18:28 AM) smithfarm: BranchPredictor: is it possible that the PR might break it instead of fixing it? ;-) (11:18:36 AM) BranchPredictor: smithfarm: actually, yes. (11:18:38 AM) smithfarm: or maybe there is an issue with the backport itself (11:19:08 AM) BranchPredictor: thread_attr is malloced and then it is passed to pthread_attr_init (11:19:28 AM) BranchPredictor: after that it is passed to pthread_attr_destroy, but it is not free'd. (11:20:07 AM) BranchPredictor: I haven't noticed this before. (11:20:48 AM) BranchPredictor: smithfarm: better yet, get this: https://github.com/ceph/ceph/pull/5337 (11:21:05 AM) BranchPredictor: it gets rid of malloc/free at all. (11:21:30 AM) BranchPredictor: and lacks pthread_attr_destroy...