[Linux-cluster] clvm running with redundant gnbd servers

[James Garratt]

I'm setting up a GNBD cluster with clvmd on the clients for the purpose of
running a xen cluster. I've been playing with this for a few months now and
I've almost got everything working, However I still have one outstanding
issue that even after extensive searches of documentation and goggle I can't
find an answer to.

My setup:
2 gnbd servers (running rhel5)
5 gnbd clients (running centos5)
GNBD servers are connected to a SAN via redundant paths.
Servers export multiple GNBDs with different names but with matching UIDs
for each device they export.
Clients import all GNBDs from each server.
multipath.conf has been configured on the clients to see the GNBDs
lvm.conf has been configured on the clients to filter everything except the
local disks and /dev/mpath/*

My problem is that if I put the two GNBD servers in the same cluster as the
GNBD clients then I get warnings as the servers can't see the Volume Groups
being used by the clients. If I put the servers in a separate cluster then
fencing can not work properly in the event of a server crash and multipath
locks up until the server is running again. Is there a way to tell clvm to
ignore some of the cluster nodes or is there another solution to this
problem?

Any advice or pointers to relevant documentation would be appreciated.

Regards,

James Garratt
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/linux-cluster/attachments/20090105/24c5f9f5/attachment.htm>