On 06/08/2015 02:38 AM, Юрий Полторацкий wrote:

Hi,

I have made a lab with a config listed below and have got unexpected result. Someone, tell me, please, where did I go wrong?

I am testing oVirt. Data Center has two clusters: the first as a computing with three nodes (node1, node2, node3); the second as a storage (node5, node6) based on glusterfs (replica 2).

I want the storage to be HA. I have read here next:
For a replicated volume with two nodes and one brick on each machine, if the server-side quorum is enabled and one of the nodes goes offline, the other node will also be taken offline because of the quorum configuration. As a result, the high availability provided by the replication is ineffective. To prevent this situation, a dummy node can be added to the trusted storage pool which does not contain any bricks. This ensures that even if one of the nodes which contains data goes offline, the other node will remain online. Note that if the dummy node and one of the data nodes goes offline, the brick on other node will be also be taken offline, and will result in data unavailability.

So, I have added my "Engine" (not self-hosted) as a dummy node without a brick and have configured quorum as listed below:
cluster.quorum-type: fixed
cluster.quorum-count: 1
cluster.server-quorum-type: server
cluster.server-quorum-ratio: 51%


Then, I've run a VM and have dropped the network link from node6, after one a hour have switched back the link and after a while have got a split-brain. But why? No one could write to the brick on node6: the VM was running on node3 and node1 was SPM.

It could have happened that after node6 came up, the client(s) saw a temporary disconnect of node 5 and a write happened at that time. When the node 5 is connected again, we have AFR xattrs on both nodes blaming each other, causing split-brain. For a replica 2 setup. it is best to set the client-quorum to auto instead of fixed. What this means is that the first node of the replica must always be up for writes to be permitted. If the first node goes down, the volume becomes read-only. 

For better availability , it would be better to use a replica 3 volume with (again with client-quorum set to auto). If you are using glusterfs 3.7, you can also consider using the arbiter configuration [1] for replica 3.

[1] https://github.com/gluster/glusterfs/blob/master/doc/features/afr-arbiter-volumes.md

Thanks,
Ravi

Gluster's log from node6:
Июн 07 15:35:06 node6.virt.local etc-glusterfs-glusterd.vol[28491]: [2015-06-07 12:35:06.106270] C [MSGID: 106002] [glusterd-server-quorum.c:356:glusterd_do_volume_quorum_action] 0-management: Server quorum lost for volume vol3. Stopping local bricks.
Июн 07 16:30:06 node6.virt.local etc-glusterfs-glusterd.vol[28491]: [2015-06-07 13:30:06.261505] C [MSGID: 106003] [glusterd-server-quorum.c:351:glusterd_do_volume_quorum_action] 0-management: Server quorum regained for volume vol3. Starting local bricks.


gluster> volume heal vol3 info
Brick node5.virt.local:/storage/brick12/
/5d0bb2f3-f903-4349-b6a5-25b549affe5f/dom_md/ids - Is in split-brain

Number of entries: 1

Brick node6.virt.local:/storage/brick13/
/5d0bb2f3-f903-4349-b6a5-25b549affe5f/dom_md/ids - Is in split-brain

Number of entries: 1


gluster> volume info vol3
 
Volume Name: vol3
Type: Replicate
Volume ID: 69ba8c68-6593-41ca-b1d9-40b3be50ac80
Status: Started
Number of Bricks: 1 x 2 = 2
Transport-type: tcp
Bricks:
Brick1: node5.virt.local:/storage/brick12
Brick2: node6.virt.local:/storage/brick13
Options Reconfigured:
storage.owner-gid: 36
storage.owner-uid: 36
cluster.server-quorum-type: server
cluster.quorum-type: fixed
network.remote-dio: enable
cluster.eager-lock: enable
performance.stat-prefetch: off
performance.io-cache: off
performance.read-ahead: off
performance.quick-read: off
auth.allow: *
user.cifs: disable
nfs.disable: on
performance.readdir-ahead: on
cluster.quorum-count: 1
cluster.server-quorum-ratio: 51%

06.06.2015 12:09, Юрий Полторацкий пишет:

Hi,

I want to build a HA storage based on two servers. I want that if one goes down, my storage will be available in RW mode.

If I will use replica 2, then split-brain can occur. To avoid this I would use a quorum. As I understand correctly, I can use quorum on a client side, on a server side, or on both. I want to add a dummy node without a brick and make such config:

cluster.quorum-type: fixed
cluster.quorum-count: 1
cluster.server-quorum-type: server
cluster.server-quorum-ratio: 51%

I expect that client will have access in RW mode until one brick alive. On the other side if server's quorum will not meet, then bricks will be RO.

Say, HOST1 with a brick BRICK1, HOST2 with a brick BRICK2, and HOST3 without a brick.

Once HOST1 lose a network connection, than on this node server quorum will not meet and the brick BRICK1 will not be able for writing. But on HOST2 there is no problem with server quorum (HOST2 + HOST3 > 51%) and that's why BRICK2 still accessible for writing. With client's quorum there is no problem also - one brick is alive, so client can write on it.

I have made a lab using KVM on my desktop and it seems to be worked well and as expected.

The main question is:

Can I use such a storage for production?

Thanks.

_______________________________________________
Gluster-users mailing list
Gluster-users@gluster.org
http://www.gluster.org/mailman/listinfo/gluster-users