Hi Joseph,
> We have production-level servers running the RH2.4.18+XFS kernel. We have
> found
> under heavy load (NFS and samba) the filesystem will lock up completely, the
> last thing being logged is:
>
> Aug 16 08:07:03 phobos smbd[8842]: request_oplock_break: no response received
> to oplock break request to pid 23785 on port 39505 for dev = 802, inode =
> 173896
> 772, file_id = 126
> Aug 16 08:07:48 phobos kernel: lease timed out
What version of Samba are you using? I also experienced these messages -
but not the complete lockups you describe - they resulted in "busy inodes"
being reported at umount time, but no corruption of any kind occurred as a
result.
Versions of samba prior to 2.2.5 have numerous bugs that cause oplock
break requests to be missed - I reported these to the samba guys and a few
patches went in for 2.2.5 to help out. I think a more general (but
less-often hit) problem was found just after the 2.2.5 release as well.
Moving to the SAMBA_2_2 CVS code from 4 July 2002 completely eliminated
the problem of Samba letting oplock leases time out for me - and the
problems that arose as a result of this went away. I too am hoping that
the patches that went into 2.4.19 will help out with this too. Let me
know if you do try the 2.4.19 CVS tree with the old version of samba and
the hangs that you experience go away.
Regards,
Chris
|