Locking mechanisms and database integrity in DB/DC and DCCTL

IMS offers a choice of locking; you can use program isolation (PI) locking or the services of the Internal Resource Lock Manager (IRLM). The IRLM component is used as an integral part of data sharing.

With program isolation, all activity (modifying the database and creating messages) of an application program that is active in the DB⁄DC environment is isolated from any other application programs that are active in the system. The isolation persists until that application program confirms, by reaching a synchronization point, that the data it has modified or created is valid.

Note: The PI lock manager supports a maximum of 63 waiting application programs in an IMS system. If an application program that must wait for a lock exceeds the maximum that the PI lock manager supports, the application program terminates with an abend 2478, is backed out by IMS, and its locks are released. If the requestor is a message-driven program (MPP, JMP, IFP, or BMP) the message is returned to the queue and is reprocessed.

The locking mechanisms are also used to:

Remove the effects of an abnormally terminated application program
Perform the processing required for a ROLL, ROLB, or ROLS call
Resolve deadlock situations

For all the above processing, the removal of database updates and held output messages is done from the previous synchronization point up to the current status. A synchronization point is defined as the point at which an application program can be restarted. The first such point for an application program is its initial scheduling. The most common synchronization point is when a GU to the message queue occurs. By issuing a call for the next message, a program in single message mode is indicating the start of a cycle of processing and the completion of any previous work. At this time, any output messages that are queued to a temporary destination are sent to their final destination, and database updates are committed.

An application program can also issue a CHKP call, which forces a synchronization point. For application programs executing in multiple message mode, or BMPs that are not transaction driven, the synchronization point is the time of either the initial scheduling or the last CHKP call.

Another aspect of program isolation is the control of database updates at the segment occurrence level. During the scheduling process, IMS analyzes the intent of an application program toward the database it uses. If a conflict exists with the database usage of a currently scheduled transaction and a candidate for scheduling because an application program needs exclusive use of the database, the scheduling process must select another transaction code and try again. If exclusive intent is not a factor (this is usually the case), application programs are scheduled concurrently. IMS controls the interleaved ownership of database segments with a locking mechanism. As application programs execute, they enqueue on the database records and release those resources, either after update or when the application program reaches a synchronization point.

Possible deadlock situations are resolved in a manner transparent to application programs and terminal operators. When IMS detects a deadlock situation, one of the application programs involved in the deadlock is abnormally terminated with a special abnormal termination code. The abnormal termination causes the activity of the terminated program to be dynamically backed out to a previous synchronization point. Its held resources are released. This allows other application programs to complete their processing. The special code causes the transaction that was being processed to be saved. The application program is rescheduled.

In DBCTL, if a deadlock situation forces an abnormal termination of a CCTL thread, that thread is not saved or retried by DBCTL. The CCTL, upon receiving certain deadlock termination codes, retries its transaction.

If a BMP is selected to be dynamically backed out, it cannot be rescheduled and terminates at its latest synchronization point. If the BMP did not access the message queues for input or issue CHKP calls, the BMP terminates during scheduling and all the BMP database update activity is nullified.

If an ODBA thread is active when IMS DB shuts down or terminates abnormally, the ODBA application thread is terminated. The ODBA application program is not terminated, but is no longer able to make calls on the thread.