Hi All,
As we know in glusterfs we have one known issue specific to preserve metadata consistency for directory in
pure distributed environment.
Problem:
After brick stop if application do any metadata operation(like permission,uid/gid change, xattr change) on directory and restart
the brick, metadata is not consistent on stopped brick because dht is not able to heal it properly.We have tried to fix
this problem from the patch https://review.gluster.org/#/c/15468/ and we would like to highlight the change in behavior after
merge this patch.
Solution:
This patch will change the current behavior of metadata operation on directory.
As of now we don't restrict application to run any metadata operation on directory if brick is stopped even brick is belong to the hash
sub-volume for that directory but after apply this patch we would not allow metadata operation(attribute,xattr change) on directory
if hashed subvolume (MDS) is down for that directory.
After merge this patch we will consider hash sub-volume as a MDS for that directory and before wind any fop call at dht level to next xlator
we do check the status of MDS, if stopped brick is the MDS of that directory we will not allow to run the operation on directory otherwise
dht wind a call to next xlator.
Regards
Mohit Agrawal
_______________________________________________ Gluster-devel mailing list Gluster-devel@xxxxxxxxxxx http://lists.gluster.org/mailman/listinfo/gluster-devel
