NFS consistency problem?

[Jonathan Schmitt]

Hi Trond,


Am Dienstag, 8. Mai 2007 02:10 schrieb Trond Myklebust:
> If you could reproduce it together with a binary tcpdump (use 'tcpdump
> -s 9000 -w /tmp/dump.dmp' from the server).

Done. You can find an archive with the dumps, the resulting files and the 
configuration files of client and server in the archive available at

http://www.informatik.uni-augsburg.de/~schmijon/nfs_consistency.tar.bz2

I executed the script test_nfs.sh and placed the "sleep n" commands with 3, 4 
and 10 seconds in it. Authentication for ssh was done by Kerberos. As you can 
see, neither of the clients mounted with nocto.

> > So, if you all allow, I would like to summarise here. From your comments
> > I gather, this is not a misconfiguration of server or client by myself,
> > is this correct?
>
> As long as you are not using the 'nocto' option on the client mount
> options, and you are being careful not to make changes within < 1 second
> of each other from different clients then this should all just work out
> of the box.

Ok, so pending your evaluation of the dumps I have found a real bug?

> >  It seems, that maybe the writing may be related to a timing issue
> > with ext3 time resolution, however this seems unlikely because I left up
> > to 3 seconds between each A and B write access and still found only As.
> > Finally, I'm the first one encountering this problem and reporting to
> > this list?
>
> If you have 3 seconds between the A and B writes, then there is a
> problem. The above mentioned tcpdump would really help...

As mentioned above, up to 10 seconds and still not working. Even worse, the 
file contains blanks which shouldn't be there, so it looks like the updates 
are not just ignored...

Kind regards and thanks for your continued help,
Jonathan