Hello, just after a bit of implementation advice for planning out a cluster.
Scenario - VMs running on ceph as a high availability cluster, but attached to each VM is an additional drive which is just for storing ephemeral data as a high-speed cache. This data does not need to survive a migration or be synced, but the VM itself does need to be able to be migrated (so the vm.conf must be able to reference disks that exist).
Solutions that I can think of:
Scenario - VMs running on ceph as a high availability cluster, but attached to each VM is an additional drive which is just for storing ephemeral data as a high-speed cache. This data does not need to survive a migration or be synced, but the VM itself does need to be able to be migrated (so the vm.conf must be able to reference disks that exist).
Solutions that I can think of:
- Local zfs drive that is synced once to all nodes so it is present and then find some strategy for "ignoring" sync checks during migration
- NFS mount to a local VM that is not synced - this would need some customisation to allow it to reconnect to the new "local" VM following a migration. Also not sure of NFS overhead will hinder the speed benefits
- Some sort of ceph trickery that achieves the goal?