[gridengine users] Is it bad practice to have a submitted/running job submit and wait for other jobs?

Jesse Becker beckerjes at mail.nih.gov
Mon May 9 22:06:39 UTC 2011


On Mon, May 09, 2011 at 05:32:37PM -0400, William Deegan wrote:
>Greetings,
>
>I'm trying to take some existing regression logic from a customer and
>have it utilize the grid.
>They have a script which launches N qsubs, ideally that job could be
>fired off with -hold_jid on the job which builds the software.
>
>so
>
>build <--hold_jid<---master script<=== N qsub'd jobs
>
>Is this OK? or considered  bad practice, and if bad, why?

While not exactly the same, we do simlar stuff like this all the time.
(That doens't mean it's "correct" though, just that you're not alone in
being "wrong" <grin>)

We actually have a few cases in our processing pipeline where we want
a driving script to block waiting on jobs, then start up again.  To do this, we
submit a bunch of jobs with known names ("Foo_1", "Foo_2", etc), then
the driving script submits a job along the lines of:

	 qrsh -now n -hold_jid 'Foo_*' /bin/true

This eventually runs, and control passes back to the driving script.


-- 
Jesse Becker
NHGRI Linux support (Digicon Contractor)



More information about the users mailing list