Hi,
unfortunately I’m a bit light on details, but willing to provide better diagnostics as this moves along.
We’ve had two instances of NFS clients getting stuck with a kernel thread spinning around `nfs4_schedule_state_manager` AFAICT:
The first instance of this was last September on a Qemu VM running a 6.1.45 guest:
root 315344 44.5 0.0 0 0 ? D Sep05 781:38 \_ [172.22.56.83-manager]
It happened a second time in last December on another VM that was likely running 5.15.139. (We downgraded our fleet due to other stability issues from 6.1 to 5.15 in between those two incidents.)
My colleagues told me that no issues were visible in the logs at that time, the systems were generally usable but interacting with anything on NFS was (obviously) stuck. So apparently no (soft-) lock ups and no stalls were recorded, but I’ll ask my colleagues to either alert me or provide all of the logging they can get the next time this happens.
Cheers,
Christian
PS: I’ve also recorded this in https://bugzilla.kernel.org/show_bug.cgi?id=217877 but switching to mail-based workflow now.
--
Christian Theune · [email protected] · +49 345 219401 0
Flying Circus Internet Operations GmbH · https://flyingcircus.io
Leipziger Str. 70/71 · 06108 Halle (Saale) · Deutschland
HR Stendal HRB 21169 · Geschäftsführer: Christian Theune, Christian Zagrodnick
On Fri, Feb 23, 2024 at 02:59:49PM +0100, Christian Theune wrote:
> Hi,
>
> unfortunately I’m a bit light on details, but willing to provide better diagnostics as this moves along.
>
> We’ve had two instances of NFS clients getting stuck with a kernel thread spinning around `nfs4_schedule_state_manager` AFAICT:
>
> The first instance of this was last September on a Qemu VM running a 6.1.45 guest:
>
> root 315344 44.5 0.0 0 0 ? D Sep05 781:38 \_ [172.22.56.83-manager]
>
> It happened a second time in last December on another VM that was likely running 5.15.139. (We downgraded our fleet due to other stability issues from 6.1 to 5.15 in between those two incidents.)
>
> My colleagues told me that no issues were visible in the logs at that time, the systems were generally usable but interacting with anything on NFS was (obviously) stuck. So apparently no (soft-) lock ups and no stalls were recorded, but I’ll ask my colleagues to either alert me or provide all of the logging they can get the next time this happens.
>
> Cheers,
> Christian
>
> PS: I’ve also recorded this in https://bugzilla.kernel.org/show_bug.cgi?id=217877 but switching to mail-based workflow now.
nfs4_schedule_state_manager() is part of the Linux NFS client, so
Cc'ing the client maintainers.
--
Chuck Lever
Hi,
not sure whether I may have missed a response that didn’t make it back to me or any of the lists.
Just in case, because the CC didn’t include the original addendum I made to my report:
Addendum:
I’ve checked kernel changelogs since then but didn’t find anything that I could relate to this aside from *maybe* dfda2a5eb66a685aa6d0b81c0cef1cf8bfe0b3c4 (rename(): fix the locking of subdirectories) which mentions NFS but doesn’t describe the potential impact.
We’re running 5.15.148 now and as it’s been another 2 months there might be the chance of another lockup in the near future ;)
If anyone has ideas on how to debug/approach a reproducer I’d be more than happy to help and try to provide more data.
Cheers,
Christian
--
Christian Theune · [email protected] · +49 345 219401 0
Flying Circus Internet Operations GmbH · https://flyingcircus.io
Leipziger Str. 70/71 · 06108 Halle (Saale) · Deutschland
HR Stendal HRB 21169 · Geschäftsführer: Christian Theune, Christian Zagrodnick
On 5 Mar 2024, at 1:09, Christian Theune wrote:
> Hi,
>
> not sure whether I may have missed a response that didn’t make it back to me or any of the lists.
>
> Just in case, because the CC didn’t include the original addendum I made to my report:
>
> Addendum:
>
> I’ve checked kernel changelogs since then but didn’t find anything that I could relate to this aside from *maybe* dfda2a5eb66a685aa6d0b81c0cef1cf8bfe0b3c4 (rename(): fix the locking of subdirectories) which mentions NFS but doesn’t describe the potential impact.
>
> We’re running 5.15.148 now and as it’s been another 2 months there might be the chance of another lockup in the near future ;)
>
> If anyone has ideas on how to debug/approach a reproducer I’d be more than happy to help and try to provide more data.
>
> Cheers,
> Christian
When the problem occurs, use the
nfs4:nfs4_state_mgr
nfs4:nfs4_state_mgr_failed
tracepoints to see what the state manager might be doing.
Also a network capture might show what the state manager thread is up to if
it is sending operations.
Ben