[Gluster-users] Anyone have that find command handy

Dan Bretherton d.a.bretherton at reading.ac.uk
Fri Jul 22 23:40:10 UTC 2011

> *Luis Cerezo* lec at luiscerezo.org 
> <mailto:gluster-users%40gluster.org?Subject=Re%3A%20%5BGluster-users%5D%20Anyone%20have%20that%20find%20command%20handy&In-Reply-To=%3C3F7AF906-46B9-4166-8F6D-5400FB0B7EE9%40luiscerezo.org%3E>
> /Thu Jul 21 17:19:45 PDT 2011/
> ------------------------------------------------------------------------
> i would also add a -size 0.
> -luis
>> On Jul 21, 2011, at 9:30 AM, Joe Landman wrote:
>> >/  On 07/21/2011 11:21 AM, Joe Landman wrote:
>> />>/  On 07/21/2011 11:17 AM, Burnash, James wrote:
>> />>>/  Hi Joe.
>> />>>/
>> />/
>> />/  Found it:
>> />/
>> />/    find $local -type f -perm +01000 -exec rm -v '{}' \;
>> />/
>> />/  for $local
>> />/
>> />/  /
*I too would strongly suggest this, because I have been seeing real, 
non-zero size files with "---------T" permissions since upgrading to 
3.2.0.  These files are visible at the mount point and usually (although 
not always) have the correct ownership (i.e. non root).  In each case 
there is always a different version of the file on another pair of 
bricks, which is different in size and almost always older than the 'T' 
file (as I call them), but I did see one case where the alternative 
version was newer than the 'T' file.  I have only seen a few dozen of 
these so far, but what worries me is that GlusterFS could be deleting 
them as if they are "stale link files", a message I often see in the 
client logs.

I have no idea which of the two alternative versions of these files are 
correct - the 'T' files or their alternatives - possibly both.  They 
might both be valid output from two successive model runs for example.  
Only the owners of these files can say which versions are correct; one 
user has agreed to take a look at some examples that I saved but hasn't 
had time to do it yet.  Deleting a 'T' file (or moving it to another 
location as I always do) results in the alternative version appearing at 
the mount point. The same thing happens after removing the trusted.gfid 
xattr from the 'T' file, as is done to files with GFID errors by Vikas 
Gorur's new GFID mismatch fixing tools I believe.  Therefore there is a 
real danger of 'T' files being deleted, either by "find ... -execrm{} 
\;" or by GlusterFS itself, resulting in the loss of important data.  
What certainly does happen is that some of the 'T' files become hidden 
from the mount point (possibly as a result of GFID matcher fixer 
scripts), and even if they are not actually deleted it can be just as 
disruptive for the files' owners.

I sincerely hope that the GFID bug fixes in 3.2.2 will stop this from 
occurring, but I don't know if the 'T' file problem is related to GFID 
mismatches and it is too early to tell whether or not the problem has 
been solved by upgrading to 3.2.2.  I am going to monitor the situation 
by doing "find .  -perm 1000  -size +0b -exec /bin/ls -l {} \;" in the 
mount points of all the volumes, and on the back-end filesystems, as 
often as I can.



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20110723/4c9ada70/attachment.html>

More information about the Gluster-users mailing list