Removing StateTransition checks from kill/delete client functions?

[sfayer]

Hi,

The stable DIRAC/src/DIRAC/Interfaces/API/Dirac.py killJob and deleteJob functions do a transition check to filter the list of jobs for ones that can be killed/deleted respectively. There are a couple of side-effects of doing this:

• You can't use killJob on a job that's in the waiting state: The server side does a check and will delete the job instead, but the filter in the client stops this from happening.
• If all jobs are removed from the list by the filter, the client will post an empty list, resulting in the confusing error Invalid job specification: []
• The documentation for deleteJob claims "If running, these jobs will be first killed", but this is no-longer true: The JobManager will do this but the filter removes running jobs from the list so it doesn't happen.

I propose removing the filter from these two functions: There is enough filtering done on the server side to handle all of the possible cases in a sensible way, the extra client-side filtering doesn't seem to add very much (from my perspective), what do people think?

Regards,
Simon

[fstagni]

This check was introduced in https://github.com/DIRACGrid/DIRAC/pull/5752/files. IIRC (I might be wrong), the reason for introducing this check is that calling commands like dirac-wms-job-kill on a non-kill-able job would have returned an S_OK result (while the status would have not changed).

[sfayer]

Understood... Perhaps we could do one of the following?

Either extend the filter in both cases so that it checks for both possible states that may be applied?
e.g. something like:

-            res = JobStatus.checkJobStateTransition(jobID, JobStatus.KILLED)
-            if res["OK"]:
-                jobIDsToKill.append(jobID)
+            can_kill = JobStatus.checkJobStateTransition(jobID, JobStatus.KILLED)["OK"]
+            can_del = JobStatus.checkJobStateTransition(jobID, JobStatus.DELETED)["OK"]
+            if can_kill or can_del:
+                jobIDsToKill.append(jobID)

Or remove the client filter and modify dirac-wms-job-kill and ...job-delete so that they process the result["InvalidJobIDs"] field to tell the user if there were jobs that didn't get updated?

Regards,
Simon

[fstagni]

extend the filter in both cases so that it checks for both possible states that may be applied? e.g. something like:

-            res = JobStatus.checkJobStateTransition(jobID, JobStatus.KILLED)
-            if res["OK"]:
-                jobIDsToKill.append(jobID)
+            can_kill = JobStatus.checkJobStateTransition(jobID, JobStatus.KILLED)["OK"]
+            can_del = JobStatus.checkJobStateTransition(jobID, JobStatus.DELETED)["OK"]
+            if can_kill or can_del:
+                jobIDsToKill.append(jobID)
```> 

This, I would say.

[fstagni]

Would you take care of the PR?

[sfayer]

Sure, #6665 .