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

[Jeff Darcy]

On 1/8/13 3:27 PM, Tomasz Chmielewski wrote:
> On 01/08/2013 09:02 PM, Jeff Darcy wrote:
> 
>> * Why are there so many metadata operations, when those are usually rare?
> 
> It could be our environment make the above more likely, because all servers
> accessing glusterfs run a cronjob at the same time:
> 
> - look up data from a database - create any "missing" directories, i.e.
> mkdir /gluster/mount/...
> 
> 
> It is "a bit" reproducible with the following script run at the same time:
> 
> for i in $(seq 1 1000); do mkdir -p /gluster/mount/dir/$i
> /gluster/mount/dir/$i/_logs /gluster/mount/dir/$i/pub done
> 
> 
> When running this on 2 servers, I can see (successful) self heal kicking in
> on one of the servers. I didn't test it very intensively - but on the other
> hand, I *only* see these failing self-heals on directories created
> automatically, as described above.
> 
> Does it make any sense?

Not yet, but it looks like an interesting test case.  I'll try to run it on
some of my own machines and see if I can reproduce it.  Thanks!