COBOL Manual for TNS/E Programs (H06.03+)

Fault-Tolerant Processes
HP COBOL Manual for TNS/E Programs—520347-003
32-14
TMF
•
You can compile your program with STARTBACKUP and CHECKPOINT
statements in place, but disable them for batch testing by running the program with
the run-time parameter NONSTOP OFF. With NONSTOP OFF, the program does
not attempt to execute the STARTBACKUP and CHECKPOINT statements. To
establish the run-time parameter, use the TACL command PARAM (or the SMU
routine SMU_Param_PutText_ if you use a COBOL process to start your batch test
programs).
•
When you are convinced that the program works without being fault tolerant, start
testing the fault tolerance. Execute the program as a process pair, and use a
symbolic debugger to set breakpoints at which you stop the primary process. You
can set different breakpoints in the primary and backup processes to determine
whether the backup process takes over properly.
•
When all controlled testing is complete, obtain the use of a dedicated system and
try full integrated testing in which you run the application as a process pair and
stop one of the two processors on which the application is running.
TMF
TMF simplifies the task of maintaining data consistency for distributed databases that
are being updated by several independent processes.
TMF was developed to help solve two problems faced by designers of database
processing applications:
•
What happens if a transaction starts to change a database and fails to complete all
of its changes?
•
How can a database be recovered after a catastrophic, multicomponent system
failure?
Before TMF was available, the answer to the first question was to write application
code to back out partial transactions and restore the affected parts of the database to
the condition they were in when the failed transaction began. Further, this application
code had to be fault tolerant to overcome possible single points of failure.
The answer to question two was to stop the system periodically and make tape or disk
backups. Then if a catastrophic failure occurred, you could restore the database from
the tape or disk and redo the transactions that had been done since the backup. This
required considerable design and coding effort, particularly if you had to do it for
several applications.
TMF has packaged the solution to these two questions of backout and recovery. TMF
automatically backs out incomplete transactions and provides a mechanism for
periodically taking online dumps and a roll-forward procedure to automate the
repetition of post-backup transactions.