Re: [nfsv4] 3530bis Issue 37: More ops need to be able to return LEASE_MOVED

[Trond Myklebust <[e-mail]>]

On Thu, 2010-11-04 at 15:56 -0600, Robert Thurlow wrote:
> Robert Thurlow wrote:
> > Hi folks,
> > 
> > This is issue 37 from 
> > http://github.com/loghyr/3530bis/blob/master/tasklist.txt.
> > 
> > In implementing NFSv4 migration support, we ran into a few more
> > ops which we believe should be able to return NFS4ERR_LEASE_MOVED.
> > The reason we have LEASE_MOVED is to notify the client that some
> > other filesystem it has used has migrated, and that it should react
> > to that to preserve its state.  Operations which include state
> > tokens should help with notification by failing with LEASE_MOVED.
> > 
> > The following OPs use state but do not now return LEASE_MOVED:
> >         OPEN_CONFIRM
> >         OPEN_DOWNGRADE
> >         SETATTR (when the file size is set)
> > 
> > IIRC from the 3530bis call, there were questions about whether
> > it was a good idea for OPEN_CONFIRM to return LEASE_MOVED.  I
> > believe it is, since state is used in the operation.  A possible
> > counter-argument is that delay can be inserted between OPEN and
> > OPEN_CONFIRM for LEASE_MOVED handling, leading to unconfirmed
> > open state being discarded, which is at least inconvenient.  A
> > server could of course choose to extend timeouts for a client
> > that had been sent a LEASE_MOVED notification.
> > 
> > If someone remembers a reason why one of these ops should not
> > return LEASE_MOVED, or you wish to comment on the above, please
> > let us know.
> 
> I don't think this has had a response.  If you disagree,
> now's the time to speak up.

I disagree with any changes here until we've had a chance to finish
discussing the replay issues.

Trond