VM not live migrated when storage not online

[Terry Xu]

Hello,
I found there is not vm live-migration when storage is not online, whose disk is on this offline storage, even if vm is in HA group.
Is that ok? How I can deal with the storage error by HA? Storage failure caused by node network error. Other nodes' networks are good.
So storages are online on other nodes.
Thanks!

 

[t.lamprecht]

Hi,

I found there is not vm live-migration when storage is not online, whose disk is on this offline storage, even if vm is in HA group.
Is that ok?

I mean, if the VM runs and the storage is offline even just from POV of this node, there's no way to ensure a clean disk state in the VM? There may be outstanding writes not yet been synced to the storage box or locking state that since the connection broke got out of sync, so this is not really possible to do in a safe way.

Recovery would still work, e.g., if the whole node gets fenced - there's some feature request to allow checking for such things and self fence the node in that case, but really only useful if just a single NIC or switch port fails, not if the whole network or storage hosts fail, as then it would cause more interruption (full cluster reboot).

IME it may often be easier and cover more practical failures by reducing coupling to a single point of failure, like the storage box, and the (network) paths to it in your case.. This can be done by either having bonding (fallback) interface to the storage box or use a hyper-converged solution like Ceph on PVE directly.