Comment 36 for bug 1640547

OK, I understand what your requirements are now. With that in mind, I guess the best thing to do is:

1. run stress-ng with the -k flag, this keeps all the process names as "stress-ng" rather than "stress-ng-${stressor-name} - this way we can nuke them using killall -9 stress-ng later on.
2. Initially send a SIGALRM to the stress-ng parent
3. Give it say 60s or so to try to terminate cleanly, this still may not occur if we have lots of pending I/O still waiting on a write()/read() or close().
3. Kill all stress-ng stressors thereafter using killall -9 stress-ng