Your write fix works! I can copy a local /usr to the NFS reexported
GlusterFS, and it copies correctly (verified with an rsync).
However, I still get flaky reads. Input/output errors occur while
copying from the reexport back to it, resulting in a partial copy. Also,
"du" on a complex directory within the reexport will come up with
substantially different values everytime it's run. Reads seem to have a
very lossy view of the filesystem...
So, there's still a bug in there somewhere.
My GlusterFS is a unified set of AFRs, with readahead and writebehind. I
tried without readahead, but the problem still occurs. For convenience,
my NFS client is also my NFS server.
Any ideas?
Thanks,
Brent
On Wed, 4 Jul 2007, Anand Avati wrote:
Brent,
Fixed this, i did find an issue in copying 'TO' glusterfs via nfs-rexport,
the NFS compat layer in the fuse kernel module sends a 'setattr' with a 0
valid flag. Data was not getting copied to the backend. now the copy-to goes
through, though I could not see any errors while copying 'OUT', probably the
same fix fixed that too.
We still dont support NFS re-export officially, though we are more than
happy to fix issues found by users (as long as they are reasonable and dont
mean a huge change in the glusterfs codebase, which is unlikely)
thanks,
avati
2007/7/3, Brent A Nelson <brent@xxxxxxxxxxxx>:
I should point out that copies FROM the NFS reexported glusterfs also
have issues. While they mostly work, there are a number of input/output
errors that crop up during the copy.
So, read nearly works but with numerous glitches, and write works except
for actually writing data.
Thanks,
Brent
On Sun, 1 Jul 2007, Brent A Nelson wrote:
> When attempting NFS reexport from the latest TLA, I have trouble with
writes,
> at least using cp. From a Solaris NFS client, I get "write: Not owner"
> messages for each file; from a Linux NFS client, I get "Operation not
> permitted" messages. In both cases, the copy appeared to create
directories
> and symlinks correctly, but the files copied were all empty.
>
> Perhaps this is related to the "ACL issue", perhaps not...
>
> Thanks,
>
> Brent
>
_______________________________________________
Gluster-devel mailing list
Gluster-devel@xxxxxxxxxx
http://lists.nongnu.org/mailman/listinfo/gluster-devel
--
Anand V. Avati
