[Linux-cluster] GFS locks wnen a node fails
Balagopal Pillai
pillai at mathstat.dal.ca
Fri Jul 20 11:01:25 UTC 2007
On Fri, 20 Jul 2007, Maciej Bogucki wrote:
Hi,
I saw this problem when i was looking at gfs as an option for an
hpc cluster. Did you mount with oopses_ok? With that option, if one node
has a problem, gfs mounts on other nodes too tend to hang needing a total
cluster restart. Ideally, the nodes should panic when they have gfs
problems and the other nodes would stay unaffected.
Regards
Balagopal
> Hal napisa?(a):
> > Hallo everybody,
> > I have a test cluster of 4 machines. node0 - gnbd server and gnbd fence server
> > and 3 nodes to mount gfs. The problem is that when I unplug one of the nodes,
> > gfs locks and no one can access it until the node is reconnected.
> >
> > How can this lock be avoided if one node fails?
> > How can I tell that gnbd-fencing is working at all?
> >
> > "gnbd_import -c node0" says nothing even if I do "fence_node node2" I assume
> > fencing is not working am I right?
>
> Hello,
>
> It looks like You don't hava fencing properly configured.
> You have to check Your logs, to see what is going on.
> If Your fence agent failed, GFS filesystem will be freezed(no ro/rw
> operations permited) until You perform manual fencing.
>
> Best Regards
> Maciej Bogucki
>
> --
> Linux-cluster mailing list
> Linux-cluster at redhat.com
> https://www.redhat.com/mailman/listinfo/linux-cluster
>
More information about the Linux-cluster
mailing list