2013/5/10 Liu Yuan <namei.unix at gmail.com>: >This problem can be solved by giving the sheeps same zone ID (with -z >option). sheeps which share same zone ID are not used for storing >dentical objects for redundancy. Now I understand better what zones are. I didn't dig into it yet :-) Anyway, better go for md use. > Valerio's suggestions looks good to me that we only offer an option to > force a cluster recovery (re-weighting all the nodes based on the space) > and plug/unplug just does a local re-balance as default. Is your new > design considering this? Yes. For me it's ok like that: plug/unplug of a device, no triggering of the whole cluster. cluster recovery done "manual". I.e. The day I expanded the cluster adding disks, my backups were going to start after 1h. Backups write on a guest running on the cluster. If cluster recover was triggered, I don't know if backup were able to write while cluster recovering. Sure slower. Just to avoid confusion: what is 'collie cluster recover' doing? I used it only to start the cluster with a missing node. Does it have other scope of use? May it be while cluster is running? |