[Bugs] [Bug 1432542] New: Glusterd crashes when restarted with many volumes

[bugzilla at redhat.com]

https://bugzilla.redhat.com/show_bug.cgi?id=1432542

            Bug ID: 1432542
           Summary: Glusterd crashes when restarted with many volumes
           Product: GlusterFS
           Version: mainline
         Component: glusterd
          Assignee: bugs at gluster.org
          Reporter: jdarcy at redhat.com
                CC: bugs at gluster.org



This was actually found in a test for bug 1430860, which was about *glusterfsd*
crashing under the same load.  That test never reproduced the original bug, but
it could reliably reproduce this one.  Examples include:

  https://build.gluster.org/job/centos6-regression/3607/consoleFull
  two crashes in attach_brick/send_attach_req  

  https://build.gluster.org/job/centos6-regression/3608/consoleFull
  first crash in attach_brick/send_attach_req
  second crash in __synclock_unlock/list_del_init

  https://build.gluster.org/job/centos6-regression/3609/consoleFull
  one crash in __synclock_unlock/list_del_init

Investigation showed that the problem was with the lock manipulation done
around retries in attach_brick.  Turns out that's not safe after all,
especially when it allows a new operation to overlap with one already in
progress - as the test does.  A fix, which moves retries into a separate thread
and tracks connection states more carefully, is forthcoming shortly.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are the assignee for the bug.