Should I consider NFS v4 production/stable now?

[J. Bruce Fields]

On Fri, Dec 15, 2006 at 01:56:41AM -0800, mike wrote:
> According to this page:
> http://wiki.linux-nfs.org/index.php/Server_state
> 
> It looks like everything is ready (and has been for a couple kernel
> versions) for Linux NFS v4...

All the basic features are there, yes.

> Last time I tried it though, it seemed like it wasn't necessarily
> ready for production.

Do you know whether any bugs that you hit are currently still
unresolved?

> It also looks like replication and migration might be coming soon...
> would this be a true possibility that in the NFS v4 protocol itself it
> would allow for a failover machine (and no longer require DRBD or
> other mirroring mechanisms and manual failover/migration?)

It will still require shared storage for the exported filesystem--DRBD,
or a cluster filesystem, or something.

One thing it should allow is moving clients without having to play with
the servers' ip addresses.

> I guess it's a two part query... also when I was switching from v4
> back to v3, I had to change which services were running, since
> rpcidmapd is used in v4, but not v3 (or whatever...) and other
> services are used in v3, but not v4.

Yes.  There's not much harm in just leaving rpc.idmapd running, though.
Note that it doesn't listen on any network sockets.  The security
exposure should be minimal.

> Just trying to get some thoughts - thanks.

I'm sitting here doing my work on a machine with a home directory on
NFSv4.  So it works for me.  I also know of some outstanding bugs.  But
it's a little hard for me to decide at what point to declare it ready
for a given use.

The NFSv4 server implementation still depends on the EXPERIMENTAL option
in the kernel config, and one thing I may have some personal influence
over is the decision about when to remove that.  There are a few
problems I'd like to see solved first--I'll see if I can post a brief
summary later this afternoon.

--b.