FORTRAN Reference Manual

Fault-Tolerant Programming
FORTRAN Reference Manual—528615-001
16-12
Starting a New Backup Process
Starting a New Backup Process
The following list describes the possible actions of the new primary process—formerly
the backup process—after a takeover from the former primary process as a result of
either a failure or a call to a Guardian routine to stop the process.
•
If the former primary process called STOP or PROCESS_STOP_ and the START
BACKUP statement did not set bit 13 in its OPTION specifier, the backup process
also stops immediately.
•
If there have been more than ten takeovers by the backup process, FORTRAN
does not start another backup process, and returns from the CHECKPOINT
statement with BACKUPSTATUS = 5000.
•
If the START BACKUP statement had bit 11 set (recreate a backup process
immediately after a takeover) in its OPTION specifier, FORTRAN attempts to
create a new backup process in the former primary’s processor. If the takeover was
not caused by a processor failure and FORTRAN cannot start a new backup
process in the former primary processor, FORTRAN terminates your process.
Otherwise, it returns from the CHECKPOINT statement with BACKUPSTATUS =
100 or BACKUPSTATUS = 101.
•
If the START BACKUP statement did not set bit 11 in its OPTION specifier,
FORTRAN allows the new primary process to run without a backup for a while, and
arranges for the next CHECKPOINT statement to attempt to create a new backup
process in the former primary’s processor. It then returns from the present
CHECKPOINT statement with BACKUPSTATUS = 100 or 101.
If the former primary’s processor failed, FORTRAN does not attempt to create a
new backup process, but only returns from the CHECKPOINT statement with
BACKUPSTATUS = 102.
When the FORTRAN run-time system cannot start a new backup process because the
former primary’s processor is down, the application program must implement one of
the following strategies:
•
Run without a backup for the remainder of the program’s execution.
•
Periodically execute a START BACKUP statement on the failed processor. This
could be done every time a CHECKPOINT statement returns BACKUPSTATUS =
1000 (backup CPU down).
•
Execute a START BACKUP statement specifying a different processor.