[Gluster-devel] healing xattr mismatches in snapshots of volumes

[Pranith Kumar Karampuri]

hi,

For snapshot design it was decided that frequent operations like 'cp -a 
<large-dir>' should not pause while taking snapshots. This can lead to 
metadata/xattr mismatches without any afr changelogs because those fops are not 
going to be blocked while taking snapshots on different bricks. Afr at the time 
of lookup detects metadata mismatches in stat structures and initiates a 
self-heal. But if there are any xattr mismatches between two replicas then 
there is no capability at the moment to detect and heal them. This enhancement 
tracks the changes required to achieve this change.

For afr to detect there are mismatches in xattrs, it should be able to query 
i.e. listxattr in lookup itself, to avoid extra over the network listxattr call 
latencies.
This requires changes in posix xlator.

Afr should use this feature in posix to 'listxattr' and see if there are any 
mismatches and trigger self-heals on that file probably at the time of 
fresh-lookup. This behaviour can be exposed in volumes that are exported as 
part of snapshot.

The change is being tracked at 
https://bugzilla.redhat.com/show_bug.cgi?id=1078061
I sent the patch which implements the changes required in posix xlator @ 
http://review.gluster.org/7293
I will be posting changes in afr once afrv2 gets merged upstream.


Pranith