Backing out uncommitted updates after a failure
The meaning of the term dynamic backout differs slightly between CICS® and IMS™.
In CICS, dynamic backout means backout as a result of a transaction (or application program) failure. The term transaction backout is used for backout done during CICS emergency restart.
In IMS, dynamic backout means backout as a result of a program failure. In a DBCTL environment, program failures include CICS transaction abends and BMP failures. The IMS /ERESTART command also performs emergency restart backout. IMS provides a batch backout utility, DFSBBO00, which you can use if dynamic backout or emergency restart fails. See IMS: Operations and automation for guidance on when to run this utility, and IMS reference: Database utilities for guidance on how to run it.
Because IMS does the backing out of database updates in a DBCTL environment, we concentrate on IMS backout in this section.
For IMS full function databases, database changes are placed in the log buffers and the database buffers as they are made. Depending on system activity, they may be written before they are committed and so, after a program failure or an IMS system failure, databases may require backout. The IMS log data sets (OLDS) are used for dynamic backout. (See IMS online log data set (OLDS) for more information.) Additionally, if dynamic backout or /ERESTART backout fails, for a database, that database is stopped. The backout is automatically reattempted when the database is restarted.
For DEDBs, no changes are placed in the log buffers until syncpoint processing begins, and no changes are written to the database until a commit has been received. This means that they do not need backout if there is a failure during phase 1 of the syncpoint process. The system can undo the changes by releasing the database buffers that have been modified but not yet written.