On Tue, Jul 14, 2009 at 01:13:03PM -0500, Charles Duffy wrote:
Scott Baker wrote:
You're supposed to have /var/lib/libvirt/images mounted via shared storage (nfs/cifs/etc) on both machines.
I've heard of folks having trouble doing live migration over NFS -- something with stronger concurrency guarantees (GFS, a shared iSCSI or FC mount [possibly with cLVM to partition it up], etc) is likely to be a safer bet.
(Indeed, looking at the documented preferred environment for VMware's VMotion is probably a good guideline for anything that's going to be trusted with real production usage -- a separate bonded pair of NICs on each host, connected to a dedicated network for communication with the SAN hosting the backing store).
A method I'm going to be trying out very shortly is using DRBD in dual-primary mode. DRBD will keep LVM volumes (block devices) for each VM in sync between two nodes. Upon migration, writes will cease on node A and when ready, begin on node B. You won't have concurrent writes, so it will be safe to use DRBD directly without a clustering FS on top (GFS, OCFS2, etc...). DRBD has scripts for Xen that take care of the details for you. Haven't seen the same for KVM/QEMU yet, but I imagine it would be pretty similar. -- Garry Dolley ARP Networks, Inc. | http://www.arpnetworks.com | (818) 206-0181 Data center, VPS, and IP Transit solutions Member Los Angeles County REACT, Unit 336 | WQGK336 Blog http://scie.nti.st