[Linux-cluster] GFS consistency error

[Bob Peterson]

On Fri, 2007-11-09 at 12:04 +0100, Jos Vos wrote:
> Hi,
> 
> One node in our two-node cluster goes down twice a day now (after this
> message it's fenced by the other node) with the following messages:
> 
> Nov  9 11:13:12 node1 kernel: GFS: fsid=my_cluster:vol3.1: fatal: filesystem consistency error
> Nov  9 11:13:12 node1 kernel: GFS: fsid=my_cluster:vol3.1:   inode = 609689846/609689846
> Nov  9 11:13:12 node1 kernel: GFS: fsid=my_cluster:vol3.1:   function = gfs_change_nlink
> Nov  9 11:13:12 node1 kernel: GFS: fsid=my_cluster:vol3.1:   file = /builddir/build/BUILD/gfs-kmod-0.1.16/_kmod_build_/src/gfs/inode.c, line = 844
> Nov  9 11:13:12 node1 kernel: GFS: fsid=my_cluster:vol3.1:   time = 1194603192
> Nov  9 11:13:12 node1 kernel: GFS: fsid=my_cluster:vol3.1: about to withdraw from the cluster
> Nov  9 11:13:12 node1 kernel: GFS: fsid=my_cluster:vol3.1: telling LM to withdraw
> 
> This is on a RHEL5-clone.
> 
> Is this really a GFS fs error and should we do a fsck?  If yes, question is
> of course how this inconsistency could happen.  If not, what can we do?
> 
> Thanks,
Hi Jos,

These problems are usually caused by file system corruption.
I recommend unmounting from all nodes, then running gfs_fsck to
clean up the corruption.  There can be many reasons for the
corruption; I tried to give the most common reasons in the faq.

Regards,

Bob Peterson