[ClusterLabs] Setting n+1 cluster steps

[Ken Gaillot]

----- Original Message -----
> Hi ,
> 
> Thank you very much  for the information ..
> 
> In each of my nodes the data base copy will be different  (used by some
> application)  . In case of failover of one node I want it to failover to
> other node with same data base copy  from failed node .
> Is it possible with pace maker to do achieve this ?

Yes, but every node has to keep a synchronized copy of every database.
Two common alternatives:

1. Set up master/slave resources for each database, where the slaves
use the database software's native replication to keep the data in sync.
If the master fails, Pacemaker will promote one of the slaves to be the
new master.

2. Set up shared storage, and run each database server on only one node.
This can be done via DRBD, a SAN or NAS, clustered filesystems, etc.
A variation of this would have the database server in a container or
virtual server running on the shared storage, and then have Pacemaker
manage the container or virtual server rather than the database server
directly.

Fencing must be configured in any case, to avoid the possibility of
data corruption in a split-brain situation.

> Thanks
> 
> 
> -----Original Message-----
> From: Ken Gaillot [mailto:kgaillot at redhat.com]
> Sent: Wednesday, June 17, 2015 7:41 PM
> To: Cluster Labs - All topics related to open-source clustering welcomed
> Subject: Re: [ClusterLabs] Setting n+1 cluster steps
> 
> ----- Original Message -----
> > Hi  ,
> > 
> > I am referring to cluster from scratch document by pacemaker which
> > only talks about configuration to 1+1 cluster .
> > I need to set n+1 and n+m clusters .Can someone please send me the
> > information where it describes the steps to set these configurations .
> 
> Every cluster is unique, so Clusters From Scratch just gives you an example
> to show the basic workings. You're probably using a high-level configuration
> tool such as crm or pcs; refer to its documentation for how to add nodes.
> 
> Usually it's as simple as
> 
>    pcs cluster node add $NODENAME --start
> 
> to add a node to an existing cluster. Every situation is different however,
> so you'll have to research and experiment. If you configured two_node in
> corosync.conf, or no-quorum-policy=ignore in Pacemaker, you'll want to
> remove that when going above 2 nodes.
> 
> For N+1/N+M you just add all your nodes to the cluster, and Pacemaker will
> host your resources on whichever nodes are available, automatically moving
> them if one goes down or comes up. There's no need to specify one or more
> nodes as "backup", but if you really want to (perhaps because one node is
> slower than the rest), you can use location constraints to prefer any node
> more or less than the others.
> 
> Again, refer to your configuration tool's documentation for how to configure
> constraints etc., but if you want a lower-level view of how it all works,
> see Pacemaker Explained:
> 
> http://clusterlabs.org/doc/en-US/Pacemaker/1.1-pcs/html-single/Pacemaker_Explained/index.html