Long-running transactions
Do not design long-running transactions in such a way that they write frequently to the system log stream, in a single unit of work, across multiple activity keypoints.
If a long-running transaction does not take regular syncpoints while regularly initiating writes to the system log, CICS® is prevented from purging units of work that completed after the long-running task started.
In Figure 1, the long-running
transaction is updating recoverable resources, causing writes to the system
log. All the updates are in the same unit of work because it does not take
explicit syncpoints. The oldest deletion point is earlier than activity keypoint
number 5, preventing CICS from deleting any completed units of work that
lie between activity keypoints 5 and 8. In this example, the long-running
transaction will eventually cause the system log stream to spill onto secondary
storage.
Figure 1. The effect of a ‘bad’ long-running transaction
