NFS Locking Problem with Fully Qualified Hostnames

NFS Locking Problem with Fully Qualified Hostnames

Post by charlottejanehamm.. » Wed, 21 Dec 2005 05:11:16



Dear All,

I'm not a unix administrator (he's off sick!) and I'm out of my depth
so I'd appreciate any help anyone could offer on the following problem.

We have an NFS set-up :- both servers and clients are running Solaris
9.  "Casual" use of the NFS mount looks ok - we can create/edit/delete
files etc.  However we have a 3rd party application which will hang
when writing to the NFS mount.  I've traced the system call and it
looks like this:

fcntl(259, F_SETLK, 0xFFFFFFFF7FFF8FC8) (sleeping...)

It sleeps forever (or until it is killed).

Oddly, I've noticed that if I add a new entry in /etc/hosts on the
server which included the UNqualified hostnames of the clients, and
then re-export the filesystems, this seems to work ok.  However, the
normal situation - fully qualified names resolved by DNS - doesn't work
for this particular system call.  We can probably live with this
workaround but it seems a bit "hacky" and was wondering if anyone could
explain what was going on.

Many thanks
C*te

 
 
 

NFS Locking Problem with Fully Qualified Hostnames

Post by Michael Tosc » Wed, 21 Dec 2005 05:25:58



> Dear All,

> I'm not a unix administrator (he's off sick!) and I'm out of my depth
> so I'd appreciate any help anyone could offer on the following problem.

> We have an NFS set-up :- both servers and clients are running Solaris
> 9.  "Casual" use of the NFS mount looks ok - we can create/edit/delete
> files etc.  However we have a 3rd party application which will hang
> when writing to the NFS mount.  I've traced the system call and it
> looks like this:

> fcntl(259, F_SETLK, 0xFFFFFFFF7FFF8FC8) (sleeping...)

> It sleeps forever (or until it is killed).

> Oddly, I've noticed that if I add a new entry in /etc/hosts on the
> server which included the UNqualified hostnames of the clients, and
> then re-export the filesystems, this seems to work ok.  However, the
> normal situation - fully qualified names resolved by DNS - doesn't work
> for this particular system call.  We can probably live with this
> workaround but it seems a bit "hacky" and was wondering if anyone could
> explain what was going on.

> Many thanks
> C*te

Check that you get identical response on the server and the clients:

getent hosts <IP>
getent hosts <host>

Otherwise correct the hosts entry in /etc/nsswitch.conf,
or the related name service.

--


 
 
 

NFS Locking Problem with Fully Qualified Hostnames

Post by Andrew Gabri » Wed, 21 Dec 2005 09:28:36




Quote:

> Check that you get identical response on the server and the clients:

> getent hosts <IP>
> getent hosts <host>

Check ipnodes before hosts (which is what the software does),
unless you commented ipnodes out in /etc/nsswitch.conf.

Quote:> Otherwise correct the hosts entry in /etc/nsswitch.conf,
> or the related name service.

--
Andrew Gabriel
 
 
 

NFS Locking Problem with Fully Qualified Hostnames

Post by Charlotte Hammon » Wed, 21 Dec 2005 19:51:04


Thank you very much Michael and Andrew!

The problem appears to be the /etc/hosts file on the clients which
containted the _unqualified_ hostname, which contrasted with the server
which used DNS only to resolved the _qualified_ hostname.    I don't
understand why this problem only manifests for this single application
while everything else appeared to work ok on the NFS mount.  However,
removing the local /etc/hosts entry so that the client also used DNS
seems to have cleared the problem, so I'm happy.

Many thanks again
C*te




> > Check that you get identical response on the server and the clients:

> > getent hosts <IP>
> > getent hosts <host>

> Check ipnodes before hosts (which is what the software does),
> unless you commented ipnodes out in /etc/nsswitch.conf.

> > Otherwise correct the hosts entry in /etc/nsswitch.conf,
> > or the related name service.

> --
> Andrew Gabriel

 
 
 

NFS Locking Problem with Fully Qualified Hostnames

Post by Michael Tosc » Thu, 22 Dec 2005 06:03:50



> Thank you very much Michael and Andrew!

> The problem appears to be the /etc/hosts file on the clients which
> containted the _unqualified_ hostname, which contrasted with the server
> which used DNS only to resolved the _qualified_ hostname.    I don't
> understand why this problem only manifests for this single application
> while everything else appeared to work ok on the NFS mount.  However,
> removing the local /etc/hosts entry so that the client also used DNS
> seems to have cleared the problem, so I'm happy.

Only locking seems affected. Few applications do locking.

I suspect a bug somewhere in the statd/lockd code,
maybe since a change in the system's name resolution.

--

 
 
 

1. hostname: fully qualify?

Here's a basic question, folks: Under Solaris 2.3, should the hostname be
fully qualified or not?  I'm pretty sure that I've read that the hostname
shouldn't be fully qualified, and that the domain should be specified only
in domainname, but I'm starting to doubt myself.

Thanks,
Bruce

2. connection problem

3. share now requires fully-qualified hostnames?

4. Cyrix 6x86 PR166+ and g++ (drowing no floats)

5. Getting Fully Qualified Hostname

6. linux route

7. hostname fully qualified or not?

8. Fortran 90 Compiler

9. Getting Fully Qualified Hostname

10. Fully qualified hostname?

11. Is There A Common Command To Get A FULLY-QUALIFIED Hostname?

12. Fully qualified hostname

13. Is There A Common Command To Get A FULLY-QUALIFIED Hostname?