[Pacemaker] streamed writes fail with migration for NFS v3 over TCP

[Bob Haxo]

Hi Karl,

I have not encountered stale file handles with NFSv3 migration with
streamed write failures. And I'm pretty certain that at least some of
the time I wait more than 90 sec for the migration to happen before
declaring failure and migrating back to the original server.

I would first like to determine where the problem is.  Since streamed
reads and writes work across migrations for NFSv3 over UDP, I think the
problem is in the tcp layer.  But, since streamed reads work across
migrations for NFSv3 over TCP, I'm left wondering what is the difference
between how the NFSv3 reads and writes are handled.  

Someone with more NFS internals experience maybe could point out where
the problem is occurring and propose a workaround or fix.  Anyone have
any suggestions?

Cheers,
Bob Haxo
SGI

On Tue, 2009-05-19 at 21:04 -0500, Karl Katzke wrote:

> Bob - 
> 
> 
> No, I don't, but I'm very interested in your progress. I also have it
> (sometimes) working over NFSv4 but ... also sometimes not working. The
> other issue is that after a HA migration, we sometimes see the file
> handles on the client machines hit a 90 second timeout and go stale;
> restarting the NFS client on the machines will clear this problem and
> bring back the file handles. Have you hit that problem? 
> 
> 
> 
> UDP is undesirable for us because we're on a large university network
> that sometimes drops packets, and I'm not aware of a way to checksum
> tranfers automatically to avoid corruption. 
> 
> 
> 
> I'm a bit new to NFS (especially v4), so I'm sorry that I can't help
> any more than that... and to make matters worse, I've had to shelf
> testing on HAE this week due to some other urgent configurations and
> moves that need to happen. 
> 
> 
> 
> Please keep me posted on your progress and let me know what
> configurations you've tried (besides UDP)... 
> 
> 
> Thanks,
> -K
> 
> ---
> Karl Katzke
> Systems Analyst II
> TAMU - DRGS
> 
> 
> 
> 
> >>> Bob Haxo <bhaxo at sgi.com> 05/19/09 5:16 PM >>>
> Greetings,
> 
> I find that streamed writes fail with migration for NFS v3 over TCP.
> Not every time, but almost every time.
> 
> Streamed writes continue nicely across many migrations for NFS v3 over
> UDP.
> 
> With TCP, writes continue with migration back to the initial server.
> 
> Does anyone have HA NFS migrations working for NFS over TCP?
> 
> Suggestions?
> 
> Cheers,
> Bob haxo
> SGI
> 
> 
> _______________________________________________
> Pacemaker mailing list
> Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.clusterlabs.org/pipermail/pacemaker/attachments/20090520/67ccea5f/attachment-0001.htm>