[DRMAA-WG] testsuite -- control and job_ps

Peter Tröger peter at troeger.eu
Tue Nov 4 07:48:33 CST 2008


> I noticed at least in 2 places (suspending and resuming a job) that
> the testsuite calls drmaa_control and expects drmaa_job_ps to return a
> new state right after that. As far as I remember, the spec says that
> it is only guaranteed that drmaa_control requests the state change
> from DRM and that it doesn't have to block until the change takes
> place actually.

Your are right, citing GFD.130 Section 9.8:

"This method SHALL return once the action has been acknowledged by the  
DRM system, but
MAY return before the action has been completed."

I think you talk about the cases where check_job_state() is  
immediately called after drmaa_control() .  In the SVN trunk of  
test_drmaa.c, I would blame everything around line 1887, 2076, and  
2091. Do you have other locations in mind ?

I would propose to relax 'check_job_state()' with some timeout  
mechanism, in order to give all DRM's enough time to perform the state  
change in a visible way.

Regards,
Peter.



More information about the drmaa-wg mailing list