[eaceaser]

The queues are local to each host so you'd lose those writes, but since writes are written to a journal on disk, you can apply some redundancy strategy to that as necessary. Depends on the SLA really.