- Integrated Systems
- About Us
- Integrated Systems
- About Us
08-04-2010 10:29 PM
New to clusters but company i am working for has grown and now wants to implment a cluster.
requirements are as follows redhat 5.4
2 r2600 itanium boxes.
the eva will replicate data from /data1 to /data1 area on second eva. Have hp continuous access
the nodes will be attached to both eva.
The cluster nodes when failing will failover to each other.
The caveat is that they want in the case of a total disaster that the node that is active with the active eva must fail over to the other node as well as the replicated data on the second eva or the replicated to eva.
Has anyone done this before and where would i find a how to or document to read to set this type of environment up in.
Thank you in advance
Solved! Go to Solution.
08-05-2010 12:39 AMSolution
So, a storage failover in addition to the node failover. In the HP-UX world, this would be covered by the MetroCluster add-on to the Serviceguard cluster framework. But a RedHat Cluster does not have anything like that as a ready-to-use component, so you'll have to add this feature on your own if you want *automatic* storage failover.
(If a *manual* failover in case of a loss of EVA is acceptable, the problem is easier: just design and test the procedures required to switch the cluster [or what is left of it] to the replicated EVA, and *document those procedures in exhaustive detail*.)
To achieve MetroCluster-like functionality using RedHat Cluster, you'll have to find out the commands you'll need to give to the EVA in the event of a storage failover, create a script that sends those commands automatically and add that script as a resource to the definition of a RedHat cluster service (= package in Serviceguard terminology).
Because this action must be the first when starting up the clustered service, all the other resources of that service must be "children" of that resource.
<script name="switch_eva_replication" file="/usr/local/sbin/switch_eva_replication.sh">
The /usr/local/sbin/switch_eva_replication.sh script should be written sort of like a /etc/init.d startup script: the cluster will invoke it with the "start" parameter when starting up the clustered service, and with the "stop" parameter when stopping it.
The script must assume that the other node may have failed abruptly: it might have burned to a crisp or been flooded by 10000 litres of raw sewage. Therefore, the "start" action of the script should be prepared to do all the necessary work, whatever it takes. The "stop" action can be empty.
You also cannot use a quorum disk. A quorum disk would have to be on a single EVA and would be a weak point in your cluster: if one node *and* the EVA that contains the quorum disk fails, your cluster cannot automatically failover to the remaining node.
If you're new to clusters in general, I'd strongly suggest you to go on a cluster administration course. I can recommend RedHat's RH436, "Clustering and Storage Management". While it does not cover storage failover, it will give you general RedHat cluster knowledge and some experience in its use.
08-05-2010 09:02 PM