Lustre discuss - Oct 2010 - Client lockup: Problems with bug 22935?

Hi all,
we are running Lustre 1.8.3 on Redhat 2.6.18-164.11.1.el5 kernels and on
one client the Lustre module ''crashed'' twice in 24 hours:

The first error message looks like we are running into bug 22935. For
the second error I could not find anything in bugzilla.
There are no errors on the Lustre servers.

Could those two problems be related?

Does anybody know a workaround for this? Should we disable statahead for
now?

Happy to provide more information.

Thanks for your help,
Karen


--- First error----
Oct 31 18:01:44 server kernel: LustreError:
4887:0:(statahead.c:1164:ll_statahead_exit())
ASSERTION(lli->lli_opendir_pid == cfs_curproc_pid()) failed
Oct 31 18:01:44 server kernel: LustreError:
4887:0:(statahead.c:1164:ll_statahead_exit()) LBUG
Oct 31 18:01:44 server kernel:  [<ffffffff887c0a69>]
ll_statahead_exit+0xd9/0x510 [lustre]
Oct 31 18:01:44 server kernel:  [<ffffffff887b63bd>]
ll_i2gids+0x7d/0x150 [lustre]
Oct 31 18:01:44 server kernel:  [<ffffffff887b6517>]
ll_prepare_mdc_op_data+0x87/0x110 [lustre]
Oct 31 18:01:44 server kernel:  [<ffffffff8876f9df>]
ll_revalidate_it+0x86f/0x1050 [lustre]
Oct 31 18:01:44 server kernel:  [<ffffffff887b9280>]
ll_mdc_blocking_ast+0x0/0x520 [lustre]
Oct 31 18:01:44 server kernel:  [<ffffffff887b5911>]
ll_convert_intent+0xb1/0x170 [lustre]
Oct 31 18:01:44 server kernel:  [<ffffffff88770d64>]
ll_revalidate_nd+0x194/0x390 [lustre]


---Second error----
Nov  1 03:17:23 server kernel: LustreError:
5341:0:(file.c:435:ll_file_open()) ASSERTION(lli->lli_sai == NULL) failed
Nov  1 03:17:23 server kernel: LustreError:
5341:0:(file.c:435:ll_file_open()) LBUG
Nov  1 03:17:23 server kernel:  [<ffffffff887860c6>]
ll_file_open+0x226/0xd10 [lustre]
Nov  1 03:17:23 server kernel:  [<ffffffff88785ea0>]
ll_file_open+0x0/0xd10 [lustre]
Nov  1 03:17:23 server kernel:  [<ffffffff8876fdb2>]
ll_revalidate_nd+0x1e2/0x390 [lustre]
Nov  1 03:17:23 server kernel: LustreError: dumping log to
/tmp/lustre-log.1288534643.5341

On 2010-10-31, at 16:30, Karen Liesenfeld wrote:
> The first error message looks like we are running into bug 22935. For
> the second error I could not find anything in bugzilla.
> There are no errors on the Lustre servers.
> 
> Could those two problems be related?
They both look like statahead problems.
> Does anybody know a workaround for this? Should we disable statahead for
> now?
I believe 1.8.4 has fixes for the statahead code, and 22935 shows that a fix was
landed for the upcoming 1.8.5 release as well.  In the meantime, you could
disable stathead on the clients.

Cheers, Andreas
--
Andreas Dilger
Lustre Technical Lead
Oracle Corporation Canada Inc.

On 01/11/10 19:22, Andreas Dilger wrote:
> On 2010-10-31, at 16:30, Karen Liesenfeld wrote:
>> The first error message looks like we are running into bug 22935. For
>> the second error I could not find anything in bugzilla.
>> There are no errors on the Lustre servers.
>>
>> Could those two problems be related?
> 
> They both look like statahead problems.
> 
>> Does anybody know a workaround for this? Should we disable statahead
for
>> now?
> 
> I believe 1.8.4 has fixes for the statahead code, and 22935 shows that a
fix was landed for the upcoming 1.8.5 release as well.  In the meantime, you
could disable stathead on the clients.
> 
Thanks for your help.
I''ll test disabling statahead until we can upgrade.

Thanks,
Karen