[Gluster-users] frequent split-brain detected, aborting selfheal; background meta-data self-heal failed

Jeff Darcy jdarcy at redhat.com
Tue Jan 8 21:09:18 UTC 2013


On 1/8/13 3:56 PM, Tomasz Chmielewski wrote:
> Also, what would be the best way to "heal it" manually?
> 
> Is something like this acceptable, when issued on both servers (basically, removing that attribute)?
> 
> setfattr -x trusted.afr.shared-client-0 /data/gluster/lfd/techstudiolfc/pub
> setfattr -x trusted.afr.shared-client-1 /data/gluster/lfd/techstudiolfc/pub


We generally try to discourage this kind of thing, because it has so often
ended up making a bad problem worse, but it's your data and only you can make
the call about risks vs. benefits.  Those commands are on the right track.  If
you're *sure* that the two copies of the directory are identical, then you can
remove the trusted.afr xattrs.  Just make sure you don't remove the gfid as
well.  You could also zero them out ("-v 0x000000000000000000000000" instead of
"-x") and/or only do that for one copy to get a slight extra margin of safety.



More information about the Gluster-users mailing list