RedHat clustering question

[JL Billman]

I am new to RedHat clustering but very familiar with ServiceGuard. My question is how do I make sure that applications do not failback to the node they were originally started on after an initial failover. I know that this is the default behavior on ServiceGuard. Any pointed on what I need to do in the cluster.conf file?

[Ivan Ferreira]

Please see this link:

http://kbase.redhat.com/faq/docs/DOC-5922;jsessionid=BE77A1DF335B1467E56717D7BE2380F7.066ef7ba

And please, keep asigning points.

Cheers.

[Steven E. Protter]

Shalom,

The principles of SG are relavent here.

Much is the same. Red Hat Cluster will operate (badly) without benefit of a lock disk.

A recent feature, imitating lock disk functionality called qdisk or quorum disk was recently added.

To do configuration of a RHCS cluster, you need GUI access to the system.

It must be running at run level 5.

From console or another Linux system:

ssh -X hostname system-config-cluster

You then get a nice GUI that lets you do such things as set up the cluster.

This is nothing like SG, but the principles are the same.

Fencing performs the RHCS (Red Hat Cluster Suite) equivalent of TOC in SG. IF your cluster has a proper fencing device like a supported UPS, fiber switch or HP iLO card, it can be configured to reboot cluster nodes if there is a danger of data corruption.

Resources are plentiful and you can configure floating ip addrsses, everything you are familiar with in SG.

A key aspect of keeping your package running where you want it is a failover domain.

If you want a package to run only on one node, create a failover domain including only that node.

If you want a package to fail over properly between nodes create a domain that includes both nodes.

It is possible with priority to dictate a package favor one node over the other.

Package creation is fun though complex.

I will run through an example.

I want to run an httpd server.

I choose to create a resource in the cluster that will call the httpd start script and include a floating IP address. Both of these are resources.

I then create a service that includes these two resources and uses a failover domain that includes both nodes.

RHCS for RHEL 4 requires 4 daemons
ccsd
cman
fenced
rgmangager

RHCS for RHEL 5 requires two daemons
cman
rgmanager

cman includes the first three daemons in one daemon.

Perhaps I've given too much information, but I tried to tailor this reply to SG issues to be of assistance.

This is the right place to deal with specific questions as you go along.

SEP

[Stephen Doud]

If you use the default settings in the package configuration file, FAILBACK_POLICY is set to MANUAL.
This means that the package will NOT fail back to the primary host upon its return to the cluster. The legacy package configuration file states:
# The MANUAL policy means no attempt will be made to move the package
# back to its primary node when it is running on an adoptive node.
#