[sheepdog-users] Crash due to assert failure in inode_coherence_update

[Vedvyas shanbhogue]

I have sheepdog running with TGT multipath. I noticed this crash when I
took one of the nodes offline and started bringing it back online again.
Looks like the VDI lock state assertion failed. I see the zookeeper
signaling the other node going away and that should have decremented the
participant count but since this node (the one that crashed) was still
alive I did not expect the count to go to 0. The only place I could see the
lock_state will go to UNLOCKED is if the participant count goes to 0.
Anyone else seen this crash? Any ideas?

May 08 07:40:16  EMERG [main] inode_coherence_update(817) Asserting
`entry->lock_state == LOCK_STATE_SHARED' failed.
May 08 07:40:16  EMERG [main] crash_handler(277) sheep exits unexpectedly
(Aborted).
May 08 07:40:16  EMERG [main] sd_backtrace(833) sheep.c:279: crash_handler
May 08 07:40:16  EMERG [main] sd_backtrace(847)
/lib64/libpthread.so.0(+0xf0ff) [0x7f7732bcc0ff]
May 08 07:40:16  EMERG [main] sd_backtrace(847)
/lib64/libc.so.6(gsignal+0x36) [0x7f7731d035f6]
May 08 07:40:16  EMERG [main] sd_backtrace(847)
/lib64/libc.so.6(abort+0x147) [0x7f7731d04ce7]
May 08 07:40:16  EMERG [main] sd_backtrace(833) vdi.c:817 (discriminator
1): inode_coherence_update
May 08 07:40:16  EMERG [main] sd_backtrace(833) group.c:978:
sd_notify_handler
May 08 07:40:16  EMERG [main] sd_backtrace(833) zookeeper.c:1252:
zk_event_handler
May 08 07:40:16  EMERG [main] sd_backtrace(833) event.c:210: do_event_loop
May 08 07:40:16  EMERG [main] sd_backtrace(833) sheep.c:1012: main
May 08 07:40:16  EMERG [main] sd_backtrace(847)
/lib64/libc.so.6(__libc_start_main+0xf4) [0x7f7731cefb14]
May 08 07:40:16  EMERG [main] sd_backtrace(847) sheep() [0x4060b8]

vs
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.wpkg.org/pipermail/sheepdog-users/attachments/20160508/97ae3e5d/attachment.html>