[Gluster-users] inconsistent file content across clients

[Jeff Darcy]

On 11/18/2010 06:05 PM, Lana Deere wrote:
> I'm using glusterfs 3.1.0, 4 distributed nodes, rdma transport, server
> nodes running CentOS 5.5, client nodes running CentOS 5.4.  Access via
> the native/fuse client, fstab entry:
>   storage0:/RaidData /mnt/RaidData glusterfs defaults,_netdev,transport=rdma 0 0
> 
> I'm seeing some file consistency issues.  Here's an example, where a second
> host becomes out of sync with the first host once "vi" changes a file.  Both
> hosts h1 and h2 are in the same working directory.
> 
> ...
> 
> If I now go to a third client which was uninvolved with the above, it
> has the current version of the file (same as on h1).  But even a long
> time later (say, 30 minutes), h2 still sometimes shows the file as
> missing and other times shows the original content, but never
> the correct new content.  However, if I now use h2 to delete the
> file, it immediately vanishes from h1 as well.
> 
> Is this expected behavior, and is there a way to force the files to
> synchronize more closely?

This looks a lot like
http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=2027 (fixed in
3.1.1) to me.  In the meantime, the bug report contains instructions for
a workaround.