I had nfs-kernel mostly working with earlier GlusterFS versions,
Linux-to-Linux (except that CWD in a shell cded into the mount would
eventually disappear on an idle shell); Linux-to-Solaris had very
unreliable rm -rf behavior. It took a lot of effort on the part of
GlusterFS developers to get to that point.
I would suggest that, if you don't get acceptable behavior from
nfs-kernel, give the latest release of UNFS3 a try. In testing, at least,
I had it working without issue, reexporting a GlusterFS via NFS to Linux
and Solaris. Performance was quite reasonable, and I very much get the
impression that there's less to go wrong with user-mode vs. kernel-mode
NFS, which is very hard to get right (and is probably very easy to break
with a small change to GlusterFS code).
Thanks,
Brent
PS I assume you know to set fsid= on your NFS exports, both for nfs-kernel
and UNFS3 (otherwise, you probably wouldn't have even been able to mount,
and you need to turn off direct-io on GlusterFS with
direct-io-mode=DISABLE on your GlusterFS mount options (or writes will
fail horribly, at best giving you zero-length files). I don't recall for
certain, but direct-io may not be an issue with UNFS3.
On Fri, 3 Oct 2008, Sam Wouters wrote:
Hi,
do you have any update about this?
I'm trying to reexport a glusterfs mount with nfs using glusterfs 1.3.12,
using nfs-kernel.
I can list and read files perfectly through the nfs share, however when trying
to write I get a "nfsd: write err=-1" in my nfs debug. Tracing the process, I
see the file being written but after fh_verify ?by nfs, the file gets emptied.
This happens with a basic storage brick as well as when using afr.
It doesn't seem to be a fuse problem, because exporting an sshfs fuse mount
works perfectly.
Any ideas?
thanks,
Sam
_______________________________________________
Gluster-devel mailing list
Gluster-devel@xxxxxxxxxx
http://lists.nongnu.org/mailman/listinfo/gluster-devel
