[Gluster-devel] Re: NFS reexport still a little glitchy

[Brent A Nelson]

Taking AFR out of the picture, things do look much better for NFS 
reexport.  If it weren't for the suspected hardlink glitch mentioned in my 
other email, it may have been perfect (copies of /usr via NFS came out 
the same size, although rsync did not, with complaints about the 
troublesome hardlink files), and no errors were logged in glusterfs.log.

So, if you need NFS reexport but not AFR, give it a go...

Thanks,

Brent

On Wed, 25 Jul 2007, Brent A Nelson wrote:

> I've been testing NFS reexport with cp -pr (from a Sun) of a /usr directory 
> from GlusterFS to GlusterFS.  So far, it's been close, but not quite right. 
> The copies never turn out the same, there's always at least some missing 
> files.  The cp reports "cannot access" for a lot of items. GlusterFS logs a 
> ton of op_ret=-1 op_errno=61 errors such as the following:
>
> 2007-07-25 14:36:43 E [afr.c:1234:afr_selfheal_getxattr_cbk] mirror2: 
> (path=/nfs2/share/zoneinfo/right/Pacific/Johnston child=share2-1) op_ret=-1 
> op_errno=61
> 2007-07-25 14:36:43 E [afr.c:1234:afr_selfheal_getxattr_cbk] ns0: 
> (path=/nfs2/share/zoneinfo/right/Pacific/Johnston child=ns0-0) op_ret=-1 
> op_errno=61
> 2007-07-25 14:36:43 E [afr.c:1234:afr_selfheal_getxattr_cbk] ns0: 
> (path=/nfs2/share/zoneinfo/right/Pacific/Johnston child=ns0-1) op_ret=-1 
> op_errno=61
>
> In my tests from today's TLA repository, nfsd is eventually even hanging 
> (this wasn't occurring in earlier patch releases).  Following that, I did an 
> ls from the GlusterFS reexport machine of one of the areas that the NFS 
> client recently complained about and got a similar error, but in a different 
> function:
>
> 2007-07-25 15:51:10 E [afr.c:563:afr_getxattr_cbk] mirror4: 
> (path=/usr/share/dict child=share4-0) op_ret=-1 op_errno=61
> 2007-07-25 15:51:10 E [afr.c:563:afr_getxattr_cbk] mirror4: (path=/usr/share 
> child=share4-0) op_ret=-1 op_errno=61
>
> I'm hoping the errors provide some clue to the NFS glitches (and presumably 
> pins the issue down to AFR), but perhaps they're harmless. Any ideas?
>
> Thanks,
>
> Brent
>
> PS An NFS reexport from an extremely simple (no protocol/*, no unify, no AFR; 
> only storage/posix) GlusterFS volume last week seemed fast and trouble-free. 
> If anyone needs NFS reexport and doesn't need AFR, I haven't tested it (yet), 
> but it's definitely worth a shot!
>