Some LRMs don't have a specific way to mark jobs that are canceled (e.g., PBS). Currently, PBS looks for an exit code of 265 (SIGKILL), but this does not seem to be reliable.
A possible solution would be to mark a job as CANCELED whenever the job is detected as ended if cancel() was previously called. This may hide some corner cases when failures unrelated to cancel() being called happen after the call and which may be relevant in some cases. However, in most normal usage situations, it should properly mark canceled jobs as canceled. We can mitigate the problem with the corner cases by logging the actual exit status from the job.
hategan
commented
4 months ago
Some LRMs don't have a specific way to mark jobs that are canceled (e.g., PBS). Currently, PBS looks for an exit code of 265 (SIGKILL), but this does not seem to be reliable.
A possible solution would be to mark a job as CANCELED whenever the job is detected as ended if
cancel()was previously called. This may hide some corner cases when failures unrelated tocancel()being called happen after the call and which may be relevant in some cases. However, in most normal usage situations, it should properly mark canceled jobs as canceled. We can mitigate the problem with the corner cases by logging the actual exit status from the job.