Bug 1431081

Description of problem:
operations as done by rsnapshot easily trigger faulty geo-replication state

Version-Release number of selected component (if applicable):
3.10, but 3.8 affected the same way

How reproducible:
100%

Steps to Reproduce:
0) create volume on master and slave and setup geo-replication between them
1) mount master volume and cd to it (nfs or fuse doesn't make a difference)
2) simulate a rsnapshot run that updates a symlink like this:

mkdir /tmp/symlinkbug
ln -f -s /does/not/exist /tmp/symlinkbug/a_symlink
rsync -a /tmp/symlinkbug ./
cp -al symlinkbug symlinkbug.0
ln -f -s /does/not/exist2 /tmp/symlinkbug/a_symlink
rsync -a /tmp/symlinkbug ./
cp -al symlinkbug symlinkbug.1

(rsnapshot uses hardlinks between rotations, that's why it is using cp -al)

Actual results:
geo-replication goes to faulty state, and symlinkbug/a_symlink still points to the old location

Expected results:
geo-replication should update the link destination, and not choke/go into faulty state on the second cp -al

Additional info:
you can pause in between the steps, add checkpoints and verify those to have all steps synced separately - doesn't make a difference.