[ClusterLabs] Antw: principal questions to a two-node cluster
kgaillot at redhat.com
Fri Apr 24 10:05:47 EDT 2015
----- Original Message -----
> >>> "Lentes, Bernd" <bernd.lentes at helmholtz-muenchen.de> schrieb am
> >>> 24.04.2015
> 13:07 in Nachricht
> <15785B7E063D464C86DD482FCAE4EBA501CAC3CA529F at XCH11.scidom.de>:
> > Hi,
> > I get the impression that the basic setup is not easy, but very important,
> > because when the cluster is running it is difficult to change something.
> > This is my idea: SAN offers one or more volumes to the nodes ==> disks
> > appears in the nodes from the SAN are used as physical volumes ==> physical
> > volume/s are combined to a volume group ==> logical volumes are created on
> > of this vg, one lv for each virtual machine ==> formatted with a file
> > system
> > like ext3 (because each virtual machine does not run concurrently on both
> > nodes this should work) ==> each virtual machine in a raw file, one vm for
> > each lv. In this setup I use cLVM.
> > I made a draft (sorry, i'm not an illustrator):
> > https://hmgubox.helmholtz-muenchen.de:8001/f/d4e18bc12a/
> > The green circle means that I'd like to move a vm to another host, maybe
> > live migration.
> > What I like to know:
> > - could it work ?
> > - is live migration possible with it ?
> > - what is about adding a third node later ?
> From some performance tests I had made I got the impression that LVM (or
> device mapper) introduces a layer of buffering on top of the multipath device
> (or SCSI disks). I'm not sure whether live migration is safe this way. The
> experts will tell you, I guess...
In the past, I've used a very similar setup with live migration. We used DRBD instead of a SAN, and Xen for the virtualization. We had cLVM set up as you described, exporting the LVs to the VMs formatted ext4 (for the most part). The cluster will never mount the fs on more than one node, so you're fine as long as your admins don't accidentally mount the fs when it's active somewhere else. :-) You could use a cluster file system if you want more safety there, but that does add more complexity. Proper fencing is essential unless you're willing to corrupt all your data in a split-brain.
cLVM requires DLM; the cluster can manage that too.
> Users mailing list: Users at clusterlabs.org
> Project Home: http://www.clusterlabs.org
> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
> Bugs: http://bugs.clusterlabs.org
Ken Gaillot <kgaillot at redhat.com>
More information about the Users