z/OS MVS Setting Up a Sysplex
Previous topic | Next topic | Contents | Contact z/OS | Library | PDF


Persistent Scenarios in Coupling Facility Failure Situations

z/OS MVS Setting Up a Sysplex
SA23-1399-00

When a CF fails, the scenarios involving persistent connectors to structures in the CF are a little more complicated and perhaps less intuitive. Generally, the following occurs:
  • When the CF fails, a loss of coupling facility connectivity is recognized by attached z/OS® systems. If a structure is duplexed, duplexing will be stopped to allow the structure instance in the other CF to be used without further recovery actions. If a structure is not duplexed, the loss of CF connectivity is reported to the connectors to the structure.
  • If the CF structure connectors are able to rebuild the structure (into a different CF), the connections remain active and switch over to using the new rebuilt structure instance.
  • If rebuild is not supported or fails, the connectors disconnect abnormally, indicating a recovery bind to the structure for which connectivity was lost. The persistent connections enter the disconnecting/failing state, and when the failures have been seen and confirmed by other connectors to the structure, the persistent connections enter the failed-persistent state.
  • When an application associated with a failed-persistent connection attempts to reconnect to the structure, because there is still a recovery bind to the old instance of the structure (which, as far as z/OS knows, is still allocated in the CF to which connectivity was lost), z/OS will attempt to reconnect to that old structure instance. There are several cases here:
    • If the CF has been restored and the structure is still allocated in that coupling facility (that is, there was only a loss of connectivity to the CF, as opposed to a loss of the coupling facility and its data), then the reconnect will be successful and the connector will do recovery processing using the data in the structure. Generally speaking, recovery from logs or other manual recovery procedures will not be necessary in this case.
    • If the CF has been restored and the structure is no longer allocated in that CF (that is, the CF failed and the structure and all of its data was lost), z/OS realizes that the old instance of the structure is no longer allocated, allocates a new instance of the structure, and reports to the connector that the connection request succeeded but resulted in the allocation of a new structure instance. The application will need to accomplish its recovery through alternative means, because the data in the structure is not present in this case. This might mean recovery from logs or other manual recovery procedures that must be performed.
    • If the CF is still unavailable to the system attempting to reconnect but available to some system in the Parallel Sysplex®, the attempt to connect to the structure will fail for that reason. If the CF remains unavailable to the system indefinitely, then attempts to connect to the structure will continue to fail indefinitely.

      Connectivity should be restored between the system and the CF. If this cannot be done, it might be possible to restart the application on a system that has connectivity to the CF, or to manually rebuild the structure into a CF that has the required connectivity. As the last resort, when no active connectors to the structure remain, deallocation of the structure can be manually forced - allowing an attempt to connect to the structure to allocate a new structure instance as in the previous bullet.

    • If the CF is still unavailable to all systems in the sysplex, CFRM will force deallocation of the inaccessible structure instance. Since no system has connectivity to the CF, the structure instance will become pending deletion. The attempt to reconnect will cause a new instance of the structure to be allocated in a different CF. The application will need to accomplish its recovery through alternative means, because the data in the structure is not present in this case. This might mean recovery from logs or other manual recovery procedures must be performed.

For more information on structure and connection persistence, and on forcing persistent structures and connections, see z/OS MVS Programming: Sysplex Services Guide.

Go to the previous page Go to the next page




Copyright IBM Corporation 1990, 2014