Gluster users - May 2013 - Can a distributed-only volume be HA for writes?

I'm testing out using gluster for storing backup images. I don't have
any data redundancy requirements beyond RAID, I just want the volume to
still be writable when one (or more?) nodes are down.

I tried it, but I'm getting a "transport endpoint is not
connected"
error when I try to write to a volume where not all the servers are
reachable. Some writes fail, some writes succeed. I'm assuming this is
because the path hash points to the missing server. Is there some way to
get the client to try the write on another server? Currently I'm testing
3.4.0-0.3.alpha3 on CentOS 6.4 (i686! my vintage test servers don't do
long mode unfortunately)

If I missed some documentation that talks about this, please point me to it!

Here are some specifics of what I tried:

[root at 192.168.254.50 ~]# gluster volume info
 
Volume Name: testvol
Type: Distribute
Volume ID: 45db51e0-18ed-4180-882e-f208ffa01452
Status: Started
Number of Bricks: 4
Transport-type: tcp
Bricks:
Brick1: 192.168.254.50:/mnt/brick1
Brick2: 192.168.254.51:/mnt/brick1
Brick3: 192.168.254.50:/mnt/brick2
Brick4: 192.168.254.51:/mnt/brick2

My client has testvol mounted:
192.168.254.50:/testvol on /mnt/gluster-test type fuse.glusterfs
(rw,default_permissions,allow_other,max_read=131072)

I crudely killed all gluster services on 192.168.254.51 with "pkill
-KILL gluster".
From the other node,
[root at 192.168.254.50 ~]# gluster peer status
Number of Peers: 1

Hostname: 192.168.254.51
Uuid: a9793617-3813-4721-a827-475790685f2c
State: Peer in Cluster (Disconnected)

Try to write to the volume from a client:

[root at client testvol]# dd if=/dev/zero of=foo bs=1M count=100
100+0 records in
100+0 records out
104857600 bytes (105 MB) copied, 1.30653 s, 80.3 MB/s

[root at client testvol]# dd if=/dev/zero of=bar bs=1M count=100
dd: opening `bar': Transport endpoint is not connected

My client logs show entries like this:
[2013-05-07 16:47:14.776711] W [common-utils.c:2330:gf_ports_reserved]
0-glusterfs-socket:  is not a valid port identifier
[2013-05-07 16:47:14.776948] W [socket.c:514:__socket_rwv]
0-testvol-client-1: readv failed (No data available)
[2013-05-07 16:47:14.779218] W [common-utils.c:2330:gf_ports_reserved]
0-glusterfs-socket:  is not a valid port identifier
[2013-05-07 16:47:14.779380] W [socket.c:514:__socket_rwv]
0-testvol-client-3: readv failed (No data available)
[2013-05-07 16:47:16.732909] W
[client-rpc-fops.c:2624:client3_3_lookup_cbk] 0-testvol-client-1: remote
operation failed: Transport endpoint is not connected. Path: /
(00000000-0000-0000-0000-000000000001)



- Andrew

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 901 bytes
Desc: OpenPGP digital signature
URL:
<http://supercolony.gluster.org/pipermail/gluster-users/attachments/20130507/749fa9ca/attachment.sig>

No, it must be replicated or distributed-replicated for that.

On 7 May 2013 18:04, "Andrew Denton" <andrewd at sterling.net>
wrote:
>
> I'm testing out using gluster for storing backup images. I don't
have
> any data redundancy requirements beyond RAID, I just want the volume to
> still be writable when one (or more?) nodes are down.
>
> I tried it, but I'm getting a "transport endpoint is not
connected"
> error when I try to write to a volume where not all the servers are
> reachable. Some writes fail, some writes succeed. I'm assuming this is
> because the path hash points to the missing server. Is there some way to
> get the client to try the write on another server? Currently I'm
testing
> 3.4.0-0.3.alpha3 on CentOS 6.4 (i686! my vintage test servers don't do
> long mode unfortunately)
>
> If I missed some documentation that talks about this, please point me to
it!
>
> Here are some specifics of what I tried:
>
> [root at 192.168.254.50 ~]# gluster volume info
>
> Volume Name: testvol
> Type: Distribute
> Volume ID: 45db51e0-18ed-4180-882e-f208ffa01452
> Status: Started
> Number of Bricks: 4
> Transport-type: tcp
> Bricks:
> Brick1: 192.168.254.50:/mnt/brick1
> Brick2: 192.168.254.51:/mnt/brick1
> Brick3: 192.168.254.50:/mnt/brick2
> Brick4: 192.168.254.51:/mnt/brick2
>
> My client has testvol mounted:
> 192.168.254.50:/testvol on /mnt/gluster-test type fuse.glusterfs
> (rw,default_permissions,allow_other,max_read=131072)
>
> I crudely killed all gluster services on 192.168.254.51 with "pkill
> -KILL gluster".
> From the other node,
> [root at 192.168.254.50 ~]# gluster peer status
> Number of Peers: 1
>
> Hostname: 192.168.254.51
> Uuid: a9793617-3813-4721-a827-475790685f2c
> State: Peer in Cluster (Disconnected)
>
> Try to write to the volume from a client:
>
> [root at client testvol]# dd if=/dev/zero of=foo bs=1M count=100
> 100+0 records in
> 100+0 records out
> 104857600 bytes (105 MB) copied, 1.30653 s, 80.3 MB/s
>
> [root at client testvol]# dd if=/dev/zero of=bar bs=1M count=100
> dd: opening `bar': Transport endpoint is not connected
>
> My client logs show entries like this:
> [2013-05-07 16:47:14.776711] W [common-utils.c:2330:gf_ports_reserved]
> 0-glusterfs-socket:  is not a valid port identifier
> [2013-05-07 16:47:14.776948] W [socket.c:514:__socket_rwv]
> 0-testvol-client-1: readv failed (No data available)
> [2013-05-07 16:47:14.779218] W [common-utils.c:2330:gf_ports_reserved]
> 0-glusterfs-socket:  is not a valid port identifier
> [2013-05-07 16:47:14.779380] W [socket.c:514:__socket_rwv]
> 0-testvol-client-3: readv failed (No data available)
> [2013-05-07 16:47:16.732909] W
> [client-rpc-fops.c:2624:client3_3_lookup_cbk] 0-testvol-client-1: remote
> operation failed: Transport endpoint is not connected. Path: /
> (00000000-0000-0000-0000-000000000001)
>
>
>
> - Andrew
>
>
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://supercolony.gluster.org/mailman/listinfo/gluster-users
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://supercolony.gluster.org/pipermail/gluster-users/attachments/20130508/d4a9790b/attachment.html>