Activity keypointing
During a restart, the CICS® log manager scans the log backwards to recover unit of work information.
The log is scanned backwards as follows:
- To begin with, CICS reads all records sequentially as far back as the last complete activity keypoint written before termination. To minimize the time taken for this scan, it is important that you do not specify an activity keypoint frequency zero. For information about setting a suitable activity keypoint frequency for your CICS region, see The activity keypoint frequency (AKPFREQ).
- When CICS reaches
the last-written complete activity keypoint, it extracts all the information
relating to in-flight units of work, including indoubt units of work. With
this information, CICS continues reading backwards, but this time reading
only the records for units of work that are identified in the activity keypoint.
Reading continues until CICS has read all the records for the units of work
identified by the activity keypoint.
This process means that completed units of work, including shunted backout-failed and commit-failed units of work, are ignored in this part of the log scan. This is significant for the retrieval of user-written log records. User-written records cannot be presented at the XRCINPT global user exit program (see Writing user-recovery data) if they are part of units of work that CICS skips in this part of the log scan process.
Figure 1. System log scan
during restart

Here are some steps you can take to ensure that system log stream
sizes, and thus restart times, are kept to a minimum:
- Keep to a minimum the amount of data that has to be read. This means specifying an activity
keypoint frequency that is non-zero, and which is:
- Long enough to avoid excessive keypointing
- Short enough to avoid large volumes of data between keypoints.
For information about calculating system log stream structure sizes, see Structure size for system log usage.
- Except for your own recovery records that you need during emergency restart, do not write your own data to the system log (for example, audit trail data).
- In particular, do not write information to the system log that needs to be kept. (See Avoiding retention periods on the system log).
- If you write a long-running transaction that updates recoverable resources, ensure that it takes a syncpoint at regular intervals.