[sheepdog-users] re-start a cluster using md
namei.unix at gmail.com
Fri May 3 16:23:34 CEST 2013
On 05/03/2013 10:06 PM, Valerio Pachera wrote:
> If I make a mistake, like inverting the order
> sheep /mnt/sdb1,/mnt/sheepdog/obj,/mnt/sheepdog
> Will it fail with an error, or mess up?
meta-store will be stored in /mnt/sdb1, and data store are
/mnt/sheepdog/obj,/mnt/sheepdog, a big mess up
> Also, what if I "forget" a disk
> i.e. (having /mnt/sheepdog for sda, /mnt/sdb1, /mnt/sdc1, /mnt/sdd1)
> sheep /mnt/sheepdog,/mnt/sheepdog/obj,/mnt/sdb1,/mnt/sdc1
> I missed /mnt/sdd1...
> What's going to happen?
Umm, nothing will be reported because we don't store information about
disks for sheep. You need to plug it back to get all data objects
belongs to this node.
So if some disks are broken, this means some objects are lost for this
node, you 1) either need a maunal force recovery by collie 2) or plug
some disk(s) in to trigger an automatic recovery to recover the lost
objects from other sheep.
More information about the sheepdog-users