On Thu, Sep 06, 2007 at 09:30:55AM -0400, Christopher Barry wrote: > On Thu, 2007-09-06 at 08:22 -0400, Lon Hohberger wrote: > > On Wed, Sep 05, 2007 at 06:16:32PM -0400, Christopher Barry wrote: > > > > > > The cluster nodes will run GFS, the director will not. Only one director > > > will be active with a VIP, load will balance across all 6 VMs. The > > > crossover will actually have VLANs on it that will allow a separate > > > heartbeat net, but it was getting a bit tricky with ASCII art ;) > > > > > > Can anyone see any issues that may arise where quorum could create a > > > split brain scenario? What would be the best way to approach votes, etc. > > > here? > > > > So, two physical boxes hosting LVS to virtual machines as the real > > servers (how ironic, actually...). Said real server cluster is using > > GFS to share the data? > > > > (I want to make sure I understand the question here) > > > > > Hi Lon, > > It is a bit ironic, isn't it ;) Yes, you are correct; the vm > real-servers are sharing a gfs volume. No real issues, but your qdiskd heuristics should be based on "can I talk to a physical node in the cluster" or something like that. Basically, you need to implement a solution which will allow all-but-one node to fail in the "virtual machine" cluster. This way, if you lose half of the VMs, you can still maintain a quorum. -- Lon Hohberger - Software Engineer - Red Hat, Inc. -- Linux-cluster mailing list Linux-cluster@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/linux-cluster
