[Gluster-devel] Gluster volume start <volname> force + EC nfs mount

[Pranith Kumar Karampuri]

hi Xavi,
        All gluster commands which can restart nfs process can lead to 
inconsistent versions on the file/directory if the gluster-nfs process 
dies just at the time of updating versions. I don't see any way to fix 
this problem as the NFS process is killed with SIGKILL.

Directory and metadata heals can recover from versions not being same. I 
think we need to add logic in data self-heal code where even when the 
versions don't match it should go ahead and check if the data matches on 
the bricks. i.e. Read from 'k' (n=k+m) number of bricks and see if the 
data matches on the rest of the redundancy bricks('m'). If it all 
matches then it should just set the versions same.

Any other ideas?

Also added gluster-devel to check if anyother component had to deal with 
similar problems.

Pranith