[Gluster-users] Broken AFR - DU / DF - part 2

Stas Oskin stas.oskin at gmail.com
Thu Apr 2 08:38:07 UTC 2009


I wanted to update with results from my tests.

So, up to full disk it worked fine, no memory leaks and equal size was
reported on all the directories - export on both servers, and the mounted
one on client.
I didn't stop/start the cluster nodes - just let it run without any stop.

After it reached the full size of the cluster, things seem to break

DF reports one size on client and on one of the servers, but another size on
the another server:

glusterfs              31G   29G  1.3G  96% /mnt/media

server 1:
/dev/hda4              31G   29G  1.3G  96% /media

server 2:
/dev/hda4              31G   18G   13G  60% /media

DU over mounted and exported directories reports same sizes on all the

So, because one of the servers reports that almost all space is taken while
it's not true, the cluster has lost almost half of the available space.
AFAIK GlusterFS always reports the free space provided by the first server
indeed, but it completely incorrect here.

What is more interesting, that it always happens with the first server
defined in client vol file. I switched the servers order this time just to
verify it, and it happened on the server that become the first one.

There are no 0 size files this time though.

So, what should be my steps to help the devs to find this issue? I'm
currently leaving all running for some time to see how it develops further.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20090402/54fe83da/attachment.html>

More information about the Gluster-users mailing list