Dear All,
I've made much progress on the mounting side of things - glusterfs is now clustering perfectly my Debian sub-servers in my "head" Leopard xServe - but I was still unable to write to the mounted filesystem without errors/refusals, and I only recently understood why: my xServe is unable to read/write to an xfs filesystem that is mounted into the local userspace through macfuse.
I know that without macFuse on a Leopard server, glusterfs has "server-only" capabilities. I also read that it is possible to re-export a glusterfs cluster through nfs, which leopard is able to understand - but I read that (a non-kernel) Fuse is necessary to do this, yet if the filesystem is first mounted into the local userspace, Leopard won't be able to read it! Catch-22...
For the time being, the only solution I see is to redo my network architecture to make one of the Debian sub-servers the nfs exporter (instead of having the Debian servers clustered and mounted in the Leopard head server) - yet is it possible that a unique Debian server be both a client and server? This way I could have a Debian server cluster and NFS-export two other debian servers, at the same time as being part of another cluster mounted on another Debian server. Does this sound feasable?
Thanks, best,
Josef.Dear Amar,
Hello - did you manage to look into the directory related problems since?
Thank you,
Josef. On Jan 3, 2009, at 09:53 , Amar Tumballi (bulde) wrote: hi 'At Work',
I got similar report in another user of glusterfs in macfuse mailing list too. I will look into this mac 'directory' related issues on monday. Will get back to you after I investigate it.
Regards,
Amar
2009/1/3 At Work <admin@xxxxxxxxxxx> What's more, I see that the proper permissions and UID are being forwarded to the remote filesystem - as the user of the service creating the files exists only on the "head" server, is it possible that it is the remote server is refusing to do a mkdir and chown directories? This would be odd, as it would seem logical that it would be the mount-point server that would decide who gets to read or write.
What of the "glusterfs-fuse" error I get every two seconds? Is this in your domain, or should I be asking this of the FUSE developers?
Thanks, best. That's it exactly. As it stands I have glusterfs server (or its server.vol file) on the sub-servers setting up (and exporting?) the bricks, and the OS X uses only the client.vol file to import and assemble the remote bricks into a cluster. Also, yes, the problems are as you say: I can read/write files, but I cannot create/upload/rename directories.
Here is a copy of the server.vol files from two servers:
matserve01:
volume posix01a
type storage/posix
option directory /raid01a/clients
end-volume
volume raid01a
type features/locks
subvolumes posix01a
end-volume
volume posix01b
type storage/posix
option directory /raid01b/clients
end-volume
volume raid01b
type features/locks
subvolumes posix01b
end-volume
### Add network serving capability to above exports.
volume server
type protocol/server
option transport-type tcp
subvolumes raid01a raid01b
option auth.addr.raid01a.allow 192.168.1.* # Allow access to "raid01a" volume
option auth.addr.raid01b.allow 192.168.1.* # Allow access to "raid01b" volume
end-volume
matserve02:
volume posix02a
type storage/posix
option directory /raid02a/clients
end-volume
volume raid02a
type features/locks
subvolumes posix02a
end-volume
volume posix02b
type storage/posix
option directory /raid02b/clients
end-volume
volume raid02b
type features/locks
subvolumes posix02b
end-volume
### Add network serving capability to above exports.
volume server
type protocol/server
option transport-type tcp
subvolumes raid02a raid02b
option auth.addr.raid02a.allow 192.168.1.* # Allow access to "raid02a" volume
option auth.addr.raid02b.allow 192.168.1.* # Allow access to "raid02b" volume
end-volume ...and the client.vol file from the OS X server.
### Add client feature and attach to remote subvolume of server1
# import RAID a's on matserve01 & matserve02
volume rRaid01a
type protocol/client
option transport-type tcp/client
option remote-host 192.168.1.6 # IP address of the remote brick
option remote-subvolume raid01a # name of the remote volume
end-volume
volume rRaid02a
type protocol/client
option transport-type tcp/client
option remote-host 192.168.1.7 # IP address of the remote brick
option remote-subvolume raid02a # name of the remote volume
end-volume
## add c, d, e, etc sections as bays expand for each server
###################
### Add client feature and attach to remote subvolume of server2
# combine raid a's
volume cluster0102a
type cluster/afr
subvolumes rRaid01a rRaid02a
end-volume
## add c, d, e, etc sections as bays expand for each server
###################
...you may notice that I am for the time being assembling but one cluster (a) - for testing purposes.
Does all this seem correct to you?
On Jan 2, 2009, at 14:17 , Krishna Srinivas wrote: Schomburg, You have 4 servers and one client. Each server has to export 2 directories /raid01a and /raid01b (FUSE do not play any role on the servers). On the client machine the glusterfs mounts using the client vol file combining all the exported directories. This would be a typical setup in your case. How is your setup? Can you mail the client vol file? According to your mail creation of directory fails. But creation/read/write of files are fine. Right? Krishna On Fri, Jan 2, 2009 at 5:01 PM, Jake Maul < jakemaul@xxxxxxxxx> wrote: On Fri, Jan 2, 2009 at 3:55 AM, At Work <admin@xxxxxxxxxxx> wrote:
Thank you for your rapid reply. Just one question: by "leave your fstab
mount alone" do you mean leave it mount the xfs disk on startup?
Yes. Mount your XFS partition via fstab as you normally would.
As for the rest.... dunno what to tell ya. Maybe one of the glusterfs
devs can chime in with some ideas.
Good luck,
Jake
This problem is odd to say the least - when I do a 'mount' after activating
the glusterfs client and cluster on Leopard, I get the following:
glusterfs on /Volumes/raid0102a (fusefs, local, synchronous)
...and on the Debian host server I get:
fusectl on /sys/fs/fuse/connections type fusectl (rw) # seems to be a
fuse connection - should fuse-accessible mounts go here?
/dev/sdb1 on /raid01a type xfs (rw) # raid block a
/dev/sdc1 on /raid01b type xfs (rw) # raid block b
...and in the glusterfs log I get:
2009-01-02 11:06:42 E [fuse-bridge.c:279:fuse_loc_fill] fuse-bridge: failed
to search parent for 576 ((null))
2009-01-02 11:06:42 E [fuse-bridge.c:703:do_chmod] glusterfs-fuse: 2: CHMOD
576 ((null)) (fuse_loc_fill() failed)
2009-01-02 11:06:42 E [fuse-bridge.c:279:fuse_loc_fill] fuse-bridge: failed
to search parent for 576 ((null))
2009-01-02 11:06:42 E [fuse-bridge.c:581:fuse_getattr] glusterfs-fuse: 1:
GETATTR 576 (fuse_loc_fill() failed)
2009-01-02 11:08:16 E [fuse-bridge.c:279:fuse_loc_fill] fuse-bridge: failed
to search parent for 578 ((null))
2009-01-02 11:08:16 E [fuse-bridge.c:2193:fuse_getxattr] glusterfs-fuse: 2:
GETXATTR (null)/578 (com.apple.FinderInfo) (fuse_loc_fill() failed)
2009-01-02 11:08:16 E [fuse-bridge.c:279:fuse_loc_fill] fuse-bridge: failed
to search parent for 578 ((null))
2009-01-02 11:08:16 E [fuse-bridge.c:2193:fuse_getxattr] glusterfs-fuse: 2:
GETXATTR (null)/578 (com.apple.FinderInfo) (fuse_loc_fill() failed)
2009-01-02 11:08:17 E [fuse-bridge.c:279:fuse_loc_fill] fuse-bridge: failed
to search parent for 578 ((null))
2009-01-02 11:08:17 E [fuse-bridge.c:2193:fuse_getxattr] glusterfs-fuse: 0:
GETXATTR (null)/578 (com.apple.FinderInfo) (fuse_loc_fill() failed)
2009-01-02 11:09:58 E [fuse-bridge.c:279:fuse_loc_fill] fuse-bridge: failed
to search parent for 578 ((null))
2009-01-02 11:09:58 E [fuse-bridge.c:581:fuse_getattr] glusterfs-fuse: 1:
GETATTR 578 (fuse_loc_fill() failed)
...and the last two lines are repeated every few minutes.
Am I correct in understanding that I have no need for FUSE on the Debian
servers? There seems to be a bridge-failure of some sort going on here.
On Jan 2, 2009, at 08:34 , Jake Maul wrote:
On the brick server (the content server... the one with the
XFS-formatted volume), FUSE is actually not used or even needed as far
as I can tell. Leave your fstab mount alone, and treat GlusterFS as a
pure replacement for NFS's /etc/exports.
FUSE only comes into play on the client side, where it's no longer
relevant what the underlying filesystem is. If I'm reading you right,
your XServe is the client in this scenario. Perhaps Mac OSX's FUSE
implementation is strange somehow, I'm not familiar with it.
Otherwise, it sounds to me like you're doing it right. Sounds like
either a permissions problem or a bug somewhere (first guesses would
be Mac OSX's FUSE, or GlusterFS client on OSX).
On Thu, Jan 1, 2009 at 11:55 PM, admin@xxxxxxxxxxx <admin@xxxxxxxxxxx>
wrote:
Dear All,
I'm afraid I'm a bit new to this. I hope I'm not missing the obvious, but
in
all the documentation I can't seem to find a clear answer to my problem.
I have a head server (Leopard X serve) that will be used as a mount point
for four sub-servers (Debian Etch) that each have two SATA RAID 5 blocks
running an XFS filesystem.
Before I switched to glusterfs, I would do an NFS export (/etc/exports)
of
the XFS filesystem mounted in /etc/fstab. I have since cancelled
(commented
out) the NFS export, but I am not quite sure what to do about the fstab:
Should I mount the drives using this file, then export the filesystem
using
glusterfs? Or should it be glusterfs doing the mounting? What role does
FUSE
have in the mount operation?
The RAID drives are at /dev/sdb and /dev/sdc, and their filesystems are
accessible at /dev/sdb1 and /dev/sdc1 - should I be mounting these with
glusterfs (instead of mounting them to a folder in the server root as I
am
doing presently)?
With my present configuration, all works correctly if I mount the raid
drives individually, yet when I mirror two drives across two servers
using
AFS things get wonky - I can upload files to a folder (and see that they
have indeed been replicated to both drives), yet I am unable to create a
new
folder (it becomes an inaccessible icon).
Thank you for any advice.
Best,
J.M. Schomburg.
_______________________________________________
Gluster-devel mailing list
Gluster-devel@xxxxxxxxxx
http://lists.nongnu.org/mailman/listinfo/gluster-devel
_______________________________________________
Gluster-devel mailing list
Gluster-devel@xxxxxxxxxx
http://lists.nongnu.org/mailman/listinfo/gluster-devel
_______________________________________________
Gluster-devel mailing list
Gluster-devel@xxxxxxxxxx
http://lists.nongnu.org/mailman/listinfo/gluster-devel
_______________________________________________ Gluster-devel mailing list
_______________________________________________
Gluster-devel mailing list
Gluster-devel@xxxxxxxxxx
http://lists.nongnu.org/mailman/listinfo/gluster-devel
--
Amar Tumballi
Gluster/GlusterFS Hacker
[bulde on #gluster/irc.gnu.org]
http://www.zresearch.com - Commoditizing Super Storage!
|
