[Gluster-devel] self heal makes client loop indefinately

[Vincent Régnard]

Vincent Régnard a écrit :
> Vincent Régnard a écrit :
>> Vincent Régnard a écrit :
>>> NB: the directory where mismatch are found has a large number of 
>>> files (a few tens of thouthands). When I ls in that directory (on 
>>> glusterfs mount point); it takes ages to return the answer. There are 
>>> also many symlinks in that directory (about 20%).
>>> I saw a stat-prefetch translator that I dont have presently, can this 
>>> help in that situation ? Is it a post 1.3.5 functionality ? (1.3.6?)
>>>
>>
>> Upgrading to 1.3.6 avoid loadaverage increase. But behaviour is the 
>> same, client loops indefinitely the same way, same log messages. And 
>> doing a ls on /mnt/glusterfs mount point hangs also indefinitely, a 
>> kill -9 on ls does nothing, pocess is uninterruptible.
>>
> 
> Probleme seams to be I was mounting the GFS volume on a mount point on a
> file system not having extended attributes. Mounting the volume on a
> mouint point where extended_attrs is available seams to work better.
> Should'nt we encounter at least a warning when doing such a bad operation ?
> 

More precision, the problem is not having the GFS mounted somewhere, but 
accessing the files on this GFS via symlinks path on a FS.

For reasons I dont understand, accessing the files in /home with the 
following symlinks is causing the loop problem. If I remove these links, 
and access the glusterfs home FS via /mnt/gluster, behaviour is normal.

# ls /home/ -l

lrwxrwxrwx    1 root     root           23 Oct 18 09:51 dist -> 
/mnt/gluster/home
lrwxrwxrwx    1 root     root           12 Oct 18 09:42 ftponly -> 
dist/ftponly

This looks like a symlink loop problem.

-- 
Vincent Régnard
vregnard at tbs-internet.com
TBS-internet.com
027 630 5902