Restarting highly available cluster applications
To restart an application, the application needs to know its state at the time of the failover or switchover.
State information is application specific; therefore, the application must determine what information is needed. Without any state information, the application can be restarted on your PC. However, you must reestablish your position within the application.
Several methods are available to save application state information for the backup system. Each application needs to determine which method works best for it.
- The application can transfer all state information to the requesting client system. When a switchover or failover occurs, the application uses the stored state on the client to reestablish the state in the new server. This can be accomplished by using the Distribute Information API or Clustered Hash Table APIs.
- The application can replicate state information (such as job information and other control structures that are associated with the application) on a real-time basis. For every change in the structures, the application sends the change over to the backup system.
- The application can store pertinent state information that is associated with it in the exit program data portion of the cluster resource group for that application. This method assumes that a small amount of state information is required. You can use the Change Cluster Resource Group (QcstChangeClusterResourceGroup) API to do this.
- The application can store state information in a data object that is being replicated to the backup systems along with the application's data.
- The application can store state information in a data object contained in the switchable IASP that also contains the application's data.
- The application can store the state information about the client.
- No state information is saved, and you need to perform the recovery.
Note: The amount of information that is required to be saved is lessened if
the application uses some form of checkpoint-restart processing. State information
is only saved at predetermined application checkpoints. A restart takes you
back to the last known checkpoint which is similar to how database's commitment
control processing works.