Ramana Raja

**Handling old clients that don't get MDS auth caps info from the MDS** > W.r.t. old clients, the MDS does not error out an fsync() call -- that needs to...

@phlogistonjohn I'm half way through reviewing this PR. Sorry about the delay. @adk3798 I see that you're running this through testing. Can you post the test result here? I'd like...

> > @phlogistonjohn I'm half way through reviewing this PR. Sorry about the delay. > > @adk3798 I see that you're running this through testing. Can you post the test...

> Please clarify [#46209 (comment)](https://github.com/ceph/ceph/pull/46209#discussion_r949515079) Instead of using NonFatalError, I made the following change to make the`nfs cluster create` command idempotent when a cluster already exists, ``` diff --git a/src/pybind/mgr/nfs/cluster.py...

> > Please clarify [#46209 (comment)](https://github.com/ceph/ceph/pull/46209#discussion_r949515079) > > Instead of using NonFatalError, I made the following change to make the`nfs cluster create` command idempotent when a cluster already exists, >...

> > > Please clarify [#46209 (comment)](https://github.com/ceph/ceph/pull/46209#discussion_r949515079) > > > > > > Instead of using NonFatalError, I made the following change to make the`nfs cluster create` command idempotent when...

> > One question though. In NonFatalError definition, was it intentional to output self.msg in the stdout position instead of stderr position of the return tuple. I guess it doesn't...

The code looks good. @dparmar18 How are you testing this fix? I remember seeing a FS tracker ticket to test new clients with old MDS server. I can't find that...

> ... > > > > This will make it more complicated than the fix in this PR currently and hard to maintain in MDS side ? > > >...

@lxbsz , also what's your opinion on Greg's suggestion of switching to the conventional behavior of root_squash enabled, which is to map uid 0 to anonymous uid. Please see, https://tracker.ceph.com/issues/56067#note-11...