We have 3 hosts with 3 guests, one guest (VM) on each.
The guests provide service redundancy/load-balancing for an application and require shared disk (GFS2). They run in different hosts in case one of the hosts go down. They are active/active so they should have the shared-mount active in all guests at the same time.
From my understanding, the VM fencing (libvirt) will multicast a message to all hosts and make sure the fenced VM gets shutdown/rebooted.
But what happens if the host has a problem and the network is down on the host? It will not receive the message to shutdown the VM and the VM could still be using the shared scsi device.
Because they are VMs, they cannot use scsi IO fencing (no PGR available).
Our previous setup used OCFS2 which used a heartbeat on the drive itself and used autofencing if things went wrong.
I am confident this setup is common out there and should have a reliable solution for that.
Thanks for any ideas