[pnfs] NFSv4.1 (DEVELOPER ONLY)

[Steve Dickson]

On 08/17/2009 12:37 PM, J. Bruce Fields wrote:
> On Mon, Aug 17, 2009 at 12:19:27PM -0400, William A. (Andy) Adamson wrote:
>> I think NFSv4.0 should be moved from EXPERIMENTAL before NFSv4.1 gets
>> moved from DEVELOPER -> EXPERIMENTAL!
>>
>> What is left to do in NFSv4.0 to allow this to happen?
> 
> Todo's I know of that should probably be fixed first:
> 
> 	- on the client side (we probably want to move client and server
> 	  out of experimental at roughly the same time, if possible?)
> 	  there's the mount negotiation to finish.  Steve intends his
> 	  mount configuration stuff to be a first step.
Which was committed...

> 	- server-side exports coordinated with v2/v3: I need to take a
> 	  day or two to think about Steve's patches.  Will do that soon!
This has been in Fedora's rawhide kernel since early July... with
now issues.. yet! ;-)

> 	- server-side reboot recovery code: the other kernel people
> 	  really hate the current v4 recovery code, and it has indeed
> 	  been a generator of weird bugs.  I tried to pick this up early
> 	  this week and write down a basic design, then got distracted.
> 	  Would like to get this fixed sooner rather than later just
> 	  because it will likely be a backwards-incompatible change,
> 	  hence easier to do while there are fewer users.
> 	- proper delegation enforcement: currently e.g. if you edit a
> 	  delegated file with a text editor, delegation-holding clients
> 	  don't get an update (because the server is not correctly
> 	  breaking leases on rename/unlink).  I have a patch series that
> 	  needs a few more bug fixes and some review.
Are these in a bug report somewhere? And it sound like that latter is
a bit more important than the former..

> 
> Some other todo's I can think of that I'm not might also block
> EXPERIMENTAL removal (but I'm not as sure):
> 
> 	- Turn on 4.0 reply cache.  Theoretical problem, not sure if
> 	  we've seen it in practice.
Won't enabling nfs41 help with this?

> 	- Make sure deferrals don't replay non-idempotent operations.
> 	  (Neil is close to fixing this.)  Theoretical problem, not sure
> 	  if we've seen it in practice.
So probably not a show stopper... 

steved.