Calling a cluster resource group exit program

The cluster resource group exit program is called during different phases of a cluster environment.

This program establishes the environment necessary resiliency for resources within a cluster. The exit program is optional for a resilient device CRG but is required for the other CRG types. When a cluster resource group exit program is used, it is called on the occurrence of cluster-wide events, including the following:

• A node leaves the cluster unexpectedly
• A node leaves the cluster as a result of calling the End Cluster Node (QcstEndClusterNode) API or Remove Cluster Node Entry (QcstRemoveClusterNodeEntry) API
• The cluster is deleted as a result of calling the Delete Cluster (QcstDeleteCluster) API
• A node is activated by calling the Start Cluster Node (QcstStartClusterNode) API
• Communication with a partitioned node is re-established

The exit program completes the following processes:

• Runs in a named activation group or the caller's activation group (*CALLER).
• Ignores the restart parameter if the exit program has an unhandled exception or is canceled.
• Provides a cancel handler.

When a cluster resource group API is run, the exit program is called from a separate job with the user profile specified on the Create Cluster Resource Group (QcstCreateClusterResourceGroup) API. The separate job is automatically created by the API when the exit program is called. If the exit program for a data CRG is unsuccessful or ends abnormally, the cluster resource group exit program is called on all active nodes in the recovery domain by using an action code of Undo. This action code allows any unfinished activity to be backed out and the original state of the cluster resource group to be recovered.

Suppose an unsuccessful switchover occurs for a device CRG. After switching back all the devices, if all of the devices were varied-on successfully on the original primary node, clustering calls the exit program on the original primary node by using an action code of Start.

If the exit program for an application CRG is unsuccessful or ends abnormally, cluster resource services attempt to restart the application if the status of the CRG is active. The cluster resource group exit program is called by using an action code of Restart. If the application cannot be restarted in the specified maximum number of attempts, the cluster resource group exit program is called by using an action code of Failover. The restart count is reset only when the exit program is called by using an action code of start, which can be the result of a start CRG, a failover, or a switchover.

When the cluster resource group is started, the application CRG exit program called on the primary node is not to return control to cluster resource services until the application itself ends or an error occurs. After an application CRG is active, if cluster resource services must notify the application CRG exit program of some event, another instance of the exit program is started in a different job. Any action code other than Start or Restart is expected to be returned.

When a cluster resource group exit program is called, it is passed a set of parameters that identify the cluster event being processed, the current state of the cluster resources, and the expected state of the cluster resources.

For complete information about cluster resource group exit programs, including what information is passed to the exit program for each action code, see Cluster Resource Group Exit Program in the Cluster API documentation. Sample source code has been provided in the QUSRTOOL library which can be used as a basis for writing an exit program. See the TCSTAPPEXT member in the QATTSYSC file.