sartg wrote:
Hi, sorry for reviving an old post.
Does this mean that as long disk failures are not simultaneously, theres enough time between disk failures to finish the rebuild and i have spare chunklets free, i can have multiple disk failures without losing the volumes?
Yes.
And when you've used up all your spare chunklets the system will use free chunklets if available. So as long as you have free space in the system, you can lose a huge amount of drives (as long as every rebuild is allowed to complete before the next failure).
Edit: as for simultaneous failures you are down to RAID level and failure domains (node pair unless filtered in CPG). So a 8 node system with RAID6 (and no HA cage), you could lose 8 drives (2 behind each node pair) without losing data as long as it's the "right" drives. With HA cage you could lose all drives in one cage per node pair....
On the other side, without HA cage and only RAID5, you also could get data loss on 2 simultaneous failures if it was the "wrong" drives (ie behind same node pair).... and even worse you could get a minor data loss with a single disk failure if you encounter an URE during rebuild (read 3PAR RAID6 whitepaper).