On 06.08.2013 10:31, Pranith Kumar Karampuri wrote: > Nux, > I suggest you don't do this. As soon as the bricks are connected > to gluster self-heal daemon, it already does this. And every 10 > minutes it checks if there is anything to be healed and heals. Understood, but that might not be soon enough for some scenarios. From what you told me, my current setup might cause a split-brain. My setup is replicated volume exported via NFS from 2 servers, HA is achieved by keepalived. If the keepalive master goes down, the IP is switched to the other server and all is well, but when the master comes back and it claims the IP, then NFS operations will happen on it. If NFS operations happen on files that have changed in the meanwhile and heal is not yet triggered, then we have a split brain. My solution to this was to let glusterd come up, initiate heal and _after_ that to start keepalived. -- Sent from the Delta quadrant using Borg technology! Nux! www.nux.ro
