Chapter 13. Clustering

Dynamic schema generation

The introduction of dynamic schema generation provides a lot of flexibility for end users to plug into Red Hat Enterprise Linux High Availability Add-on custom resource and fence agents, and still retain the possibility to validate their /etc/cluster.conf configuration file against those agents. It is a strict requirement that custom agents provide correct metadata output and that the agents must be installed on all cluster nodes.

Clustered Samba on GFS2

Support for Samba in a clustered environment is now fully supported in Red Hat Enterprise Linux 6.2. Samba clustering relies on a clustered file system being available and shared on all nodes. In the Red Hat Enterprise Linux context, Samba clustering has been configured to work with GFS2, the native shared storage file system.

Clustered Samba (more specifically CTDB) provides the ability for the metadata to span multiple physical hosts in a cluster. CTDB will automatically recover and repair node-specific databases in case of node failures. It also provides high availability features like node monitoring and failover.
Support for redundant ring for standalone Corosync

Red Hat Enterprise Linux 6.2 introduces support for redundant ring with autorecovery feature as a Technology Preview. Refer to the Technical Notes for a list of known issues associated with this Technology Preview.

corosync-cpgtool

The corosync-cpgtool now specifies both interfaces in a dual ring configuration. This feature is a Technology Preview.

Disabling rgmanager in /etc/cluster.conf

As a consequence of converting the /etc/cluster.conf configuration file to be used by pacemaker, rgmanager must be disabled. The risk of not doing this is high; after a successful conversion, it would be possible to start rgmanager and pacemaker on the same host, managing the same resources.

Consequently, Red Hat Enterprise Linux 6.2 includes a feature (as a Technology Preview) that forces the following requirements:
  • rgmanager must refuse to start if it sees the <rm disabled="1"> flag in /etc/cluster.conf.
  • rgmanager must stop any resources and exit if the <rm disabled="1"> flag appears in /etc/cluster.conf during a reconfiguration.