Frank Filz

Not really, at least if you are doing NFSv4.,x since an NFSv4.x lock is dependent on an open stateid and both are dependent on a clientid (which for NFSv4.1 also...

Yes, all nodes in a cluster must synchronize grace period. You can't do grace period per inode because you have no way of knowing for which inodes recovery is complete...

You would need to send sm-notify from the load balancer node. You might be able to send it from the Ganesha node and make it look like it came from...

> @ffilz Thanks the reply, I have another question, sorry for that. How can I make client reclaim lock when create session, even without ganesha restart ? We need this...

> > Possibly you could even persist the union of the access modes and deny modes to allow non-conflicting opens to occur. > > @ffilz As the NFSv4 proto describes,...

> > If possible, I suggest you try and join our community calls at 7:00 AM Pacific Time on Tuesdays. I think your needs are in much alignment with Googles....

Sorry I forgot to mention the new NFSV4 { } parameter Server_Scope which is in V4.0 that allows configuring all the nodes with the same server scope. That may be...

From RFC 7530 Section 9.6.2. Server Failure and Recovery: The period of special handling of locking and READs and WRITEs, equal in duration to the lease period, is referred to...

So certainly we COULD get more smart about when we can reliably avoid returning NFS4ERR_GRACE. More persistent information would help. Knowing we CAN'T hand out layouts and delegations would help.

> > If possible, I suggest you try and join our community calls at 7:00 AM Pacific Time on Tuesdays. > > @ffilz Hi, what's the link of community calls...