Discussion:
[Pacemaker] Multiple live cib in one pacemaker
Adam Błaszczykowski
2015-02-19 14:26:03 UTC
Permalink
Hello,
I am using Pacemaker 1.1.12 together with Corosync 2.4.3 in my cluster
environment. I have two nodes in cluster that are in different LAN
locations. It may be situation that nodes will not be able to connect with
each other because of network failure. In that situation split brain can
occur if we add new configuration on both nodes but that configuration will
be different on each node.

1. After repairing network failure, both nodes will rejoin to cluster, but
there are two different configurations. Which cib configuration will be
used ? Is it possible that configuration on first node will be overwriten
by second node ?
2. Is there any possibility to use multiple live cib configurations to
prevent from losing configuration in split brain situation ?

My split brain example:

1. Both nodes are up and cib configuration is synced
2. node1 has vip1 as resource, node2 has vip2 as resource
3. network failure, nodes cannot sync configuration
4. admin add vip2 on node1
5. second admin add vip3 on node2 (now we have split brain in configuration)
6. network failure resolved, nodes are rejoining the cluster
7. cib configuration is synced and probably we lost configuration from
node1 or node2

How to solve this problem ?

Thank you in advance !
Best Regards.
Adam Blaszczykowski
Mark Gardner
2015-02-19 14:48:37 UTC
Permalink
Fixed by using a third node.

On Thu, Feb 19, 2015 at 8:26 AM, Adam Błaszczykowski <
Post by Adam Błaszczykowski
Hello,
I am using Pacemaker 1.1.12 together with Corosync 2.4.3 in my cluster
environment. I have two nodes in cluster that are in different LAN
locations. It may be situation that nodes will not be able to connect with
each other because of network failure. In that situation split brain can
occur if we add new configuration on both nodes but that configuration will
be different on each node.
1. After repairing network failure, both nodes will rejoin to cluster, but
there are two different configurations. Which cib configuration will be
used ? Is it possible that configuration on first node will be overwriten
by second node ?
2. Is there any possibility to use multiple live cib configurations to
prevent from losing configuration in split brain situation ?
1. Both nodes are up and cib configuration is synced
2. node1 has vip1 as resource, node2 has vip2 as resource
3. network failure, nodes cannot sync configuration
4. admin add vip2 on node1
5. second admin add vip3 on node2 (now we have split brain in
configuration)
6. network failure resolved, nodes are rejoining the cluster
7. cib configuration is synced and probably we lost configuration from
node1 or node2
How to solve this problem ?
Thank you in advance !
Best Regards.
Adam Blaszczykowski
_______________________________________________
http://oss.clusterlabs.org/mailman/listinfo/pacemaker
Project Home: http://www.clusterlabs.org
Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
Bugs: http://bugs.clusterlabs.org
--
~ Mark
Gardner ~

***@gmail.com
***
Kristoffer Grönlund
2015-02-19 14:52:33 UTC
Permalink
Post by Adam Błaszczykowski
Hello,
I am using Pacemaker 1.1.12 together with Corosync 2.4.3 in my cluster
environment. I have two nodes in cluster that are in different LAN
locations. It may be situation that nodes will not be able to connect with
each other because of network failure. In that situation split brain can
occur if we add new configuration on both nodes but that configuration will
be different on each node.
Running a single cluster across different LANs with an unreliable
network connection doesn't seem like a situation supported by
Pacemaker. You probably need to look at something like booth [1] to get
this to work reliably.

Booth allows resources to migrate between different clusters, using
tickets and an arbiter in a third location.

[1]: https://github.com/ClusterLabs/booth
--
// Kristoffer Grönlund
// ***@suse.com

_______________________________________________
Pacemaker mailing list: ***@oss.clusterlabs.org
http://oss.clusterlabs.org/mailman/listinfo/pacemaker

Project Home: http://www.clusterlabs.org
Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
Bugs: http://bugs.clu
Andrew Beekhof
2015-02-19 19:17:43 UTC
Permalink
Post by Kristoffer Grönlund
Post by Adam Błaszczykowski
Hello,
I am using Pacemaker 1.1.12 together with Corosync 2.4.3 in my cluster
environment. I have two nodes in cluster that are in different LAN
locations. It may be situation that nodes will not be able to connect with
each other because of network failure. In that situation split brain can
occur if we add new configuration on both nodes but that configuration will
be different on each node.
Running a single cluster across different LANs with an unreliable
network connection doesn't seem like a situation supported by
Pacemaker. You probably need to look at something like booth [1] to get
this to work reliably.
Booth allows resources to migrate between different clusters, using
tickets and an arbiter in a third location.
And probably sbd
Post by Kristoffer Grönlund
[1]: https://github.com/ClusterLabs/booth
--
// Kristoffer Grönlund
_______________________________________________
http://oss.clusterlabs.org/mailman/listinfo/pacemaker
Project Home: http://www.clusterlabs.org
Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
Bugs: http://bugs.clusterlabs.org
_______________________________________________
Pacemaker mailing list: ***@oss.clusterlabs.org
http://oss.clusterlabs.org/mailman/listinfo/pacemaker

Project Home: http://www.clusterlabs.org
Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
Bugs: http:

Loading...