What is the recommended way to present shared storage to a group of KVM hosts?
I have 2 DSS v7 boxes in an Active-Passive cluster. I would like to have 8-10 VMs running on these boxes with auto failover where the VMs can migrate between 2+ KVM hosts.
I have used file based images with NFS in the past to do this, but I don't think NFS failover is supported with DSS v7.
I can't treat the iSCSI target as a single filesystem between KVM hosts without a clustered filesystem. (gfs, clusterfs seem like a fair amount of extra overhead). IE, one large LV on the DSS boxes in A-P partitioned and formated as ext4 wouldn't work, but gfs should. However, gfs might be a little slow for this application.
I don't want to have to create a LV and replication task on both hosts if i make each virtual machine image a separate LV on the DSS hosts. That will be annoying at about 8-10 machines.
Should I create a large LV that is setup for active-passive on the DSS machines and then provision it into virtual machine images as needed with LVM from the KVM hosts? Then have the KVM hosts access the images as lvm partions.
Any other forum or mailing list for this question?
I'd ask in a KVM / or libvirt, ovirt type forum, but the answer would probably be to just use iSCSI targets for each VM image. That may be a good solution if you have a single storage server. But, it gets pretty clunky trying to to setup open-e for failover with 10-15 (or more as we grow) iscsi targets. That would be mean creating the source LV, target LV, with the same size and scsiID, replication task, etc, and adding them to the cluster for each VM.
Is that the way to go? Does my question make sense to anyone? I feel like I'm overlooking something more simple to manage.
Has anyone had luck with open-e active-passive (or active-active) and using gfs or glusterfs?