Re: NFSv4 console messages (locks lost etc.)

Lars Eggert wrote:
> Hi,
> 
> I should have mentioned that the server is FreeBSD -STABLE running
> newnfs, and the network isn't partitioned (because I access the box
> over SSH at the same time I see these messages.)
> 
> They only appear under heavy NFS load (portmaster build of math/R in
> this case.)
> 
Oh, one more thing. If the server hits NFSRV_V4STATELIMIT for a total
count of opens and locks (defined in sys/fs/nfs/nfsport.h) it will stop
doing opens and pretty well come to a grinding halt.

The default of 500000 seems pretty generous, but since it was chosen
to avoid resource exhaustion on a single core PIII with 256Mbytes of
RAM, increasing it is probably safe for more modern server hardware.
(You'll need to edit sys/fs/nfs/nfsport.h and rebuild a kernel. It isn't
 a sysctl.)

rick

> Lars
> 
> On Jun 29, 2013, at 2:32, Rick Macklem <rmacklem_at_uoguelph.ca> wrote:
> 
> > Lars Eggert wrote:
> >> Hi,
> >> 
> >> on a -CURRENT client, I get quite a number of console messages
> >> under
> >> heavy NFSv4 load, such as:
> >> 
> >> nfsv4 expired locks lost
> > Means the lease expired on the NFSv4 server somehow. Lease
> > expiry is "bad news" and there is no way to recover locks
> > lost because of it.
> >> nfscl: never fnd open
> > Usually, opens can be recovered after a lease expiry, but it
> > might be broken. Since lease expiry should never happen during
> > normal operation (see below), it doesn't get a lot of testing.
> > 
> >> nfscl: never fnd open
> >> nfscl: never fnd open
> >> nfsv4 expired locks lost
> >> nfscl: never fnd open
> >> nfscl: never fnd open
> >> nfsv4 expired locks lost
> >> nfsv4 expired locks lost
> >> nfsv4 expired locks lost
> >> nfsv4 expired locks lost
> >> nfsv4 expired locks lost
> >> nfscl: never fnd open
> >> 
> >> Can I ignore them? Can I turn them off?
> >> 
> > Well, these should never happen during normal, correct operation.
> > The
> > "nfsv4 expired locks lost" implies lease expiry. This should only
> > happen
> > when the client is network partitioned from the server for more
> > than
> > a lease duration (chosen by the server, but typically about
> > 1minute).
> > The client does a Renew Op every 1/2 lease durations to avoid this.
> > Also,
> > any state related operation (open/lock/locku/close/etc) is supposed
> > to
> > renew the lease implicitly.
> > 
> > If you are getting network partitions happening, then you really
> > need
> > to fix the network.
> > 
> > If not, then if you watch network traffic with something like
> > wireshark
> > and see Renew Ops happening at regular intervals, then I can only
> > suggest
> > that the server is somehow broken for NFSv4. You should also look
> > for
> > NFS4ERR_EXPIRED error replies to operations related to state
> > (open/lock/locku/close).
> > That is the server reply which indicates the lease expiry. If the
> > server is
> > never returning this, I have no idea how the client would generate
> > the above
> > messages, but it does indicate a client NFSv4 bug if that is the
> > case.
> > 
> > Switching all mounts to NFSv3 will get rid of the above, although
> > it is
> > not exactly a fix;-)
> > 
> > rick
> > 
> >> Thanks,
> >> Lars
> >> _______________________________________________
> >> freebsd-current_at_freebsd.org mailing list
> >> http://lists.freebsd.org/mailman/listinfo/freebsd-current
> >> To unsubscribe, send any mail to
> >> "freebsd-current-unsubscribe_at_freebsd.org"
> >> 
> 
> _______________________________________________
> freebsd-current_at_freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-current
> To unsubscribe, send any mail to
> "freebsd-current-unsubscribe_at_freebsd.org"
>