[Gluster-users] Fwd: files not syncing up with glusterfs 3.1.2

[paul simpson]

hi shehjar,

many thanks for your reply.

that is 4 servers - serving ~30 clients - 95% linux, 5% mac.  all NFS.
>>
> Mac OS as a nfs client remains untested against Gluster NFS. Do you see
> these errors on Mac or Linux clients?


only tested locking on linux so far.


 - sometimes root can stat/read the file in question while the user cannot!
>>  i can remount the same NFS share to another mount point - and i can then
>> see that with the same user.
>>
> I think that may be occurring because NFS+LDAP requires a slightly
> different authentication scheme as compared to a NFS only setup. Please try
> the same test without LDAP in the middle.


will do tomorrow - been crazy busy fire fighting today.  (btw, root always
seemed to work - which is obviously not an ldap authenticated user - but
will try another one...)


> - sample output of g1 nfs.log file:
>>
>> [2011-02-18 15:27:07.201433] I [io-stats.c:338:io_stats_dump_fd]
>> glustervol1:       Filename :
>> glustervol1:       Filename :
>> /production/conan/hda/published/shot/backup/.svn/tmp/entries
>>
>> ...so, the files not working don't have lifetime, read/written lines after
>> their log entry.
>>
>> I'll need the log for the NFS server in TRACE log level when you run a
> command that results in any of the errors above. i.e. stale file handle,
> remote IO error and input/output error.
>

will do.

> Locking is part of the core GlusterFS protocol but the NFS server module
does not have NLM support yet(NLM is the locking protocol associated with
NFSv3).
> On linux, the workaround is generally to mount with the -o nolock option
although I dont see why excluding this option results in stale file handle
and other
> errors. let me go through the complete thread, I'll reply elsewhere.

oh, thats a biggie - and not mentioned in any docs i've read so far.  i
think that a features table comparing NFS to gluster would be most useful.
 until that table appears - are there any other nfs v3 features that gluster
NFS doesn't support?  would allow us to manage our own expectations and use
gluster in an appropriate fashion...

> Today, if locking among multiple client machines is a must-have, you'll
have to use FUSE.

i have tried svn with multiple clients using FUSE - and have experienced the
same input/output errors.  should NLM locking in the fuse client work as per
NFS v3?

again thanks, logs will follow asap.

-paul