Avoiding long running transactions

Avoid designing applications in which TMF transactions remain active for more than a few tens of seconds. Long running transactions can cause the circular audit trail of TMF to fill up. Because TMF is a critical system-wide resource, TMF protects itself by backing out application transactions that are active for too long.

Suppose that the processing within an application is driven by getting messages from a queue, and that the application gets a message from the queue and processes the message within a unit of work. Typically, an application calls MQGET with the wait option and within syncpoint control to get a message from the queue.

If the application is using a global unit of work instead, the specified wait interval on the MQGET call must be short to avoid a long running transaction. This means that the application might need to issue the MQGET call more than once before it retrieves a message.