[Rivet] Running Rivet on the grid

[Andy Buckley]

Frank Siegert wrote:
> Andy Buckley, Monday 13 October 2008:
>> I don't know if the API is public, but the glite commands certainly are
>> Python scripts making use of some API:
> 
> Not all of them. In fact, the job-status seems to be the only one.

Ah, damn. I just chose to look at the one that gave me an error
(actually because of bad interactions between Python 2.5 and 2.3 in my
environment)!

There are more Python ones, as you can see by running "file
/opt/glite/bin/*", but the binary programs dominate.

>> Then there are modules called UIutils, UIchecks, Job, ... and the
>> catchy glite_wmsui_*Wrapper which do the Griddy stuff:
> 
> Yes, thanks. I am still trying to dig through these, but so far can't 
> really tell, how it is going to be useful. If you find anything vaguely 
> similar to documentation related to this "API", please let me know.

I had a look on the gLite site and found it hard to identify anything
useful through all the acronyms. The most relevant-looking thing was
this, which looks like a silly SQL wrapper, but may be useful for
getting job information:
https://edms.cern.ch/document/503614/1.5.1

We should ask people like Phil Roffe and David Grellscheid if they know
any more about middleware APIs... I certainly know nothing!

> If this is all that exists in terms of API/documentation, I am inclined to 
> calling the glite-wms-* commands through os.system, as they seem to be 
> the better documented and maybe even more stable choice. What do you 
> think?

Maybe, depending on how parsable their output is. Although the
"commands" and "subprocess" modules provide better control than os.system.

Andy

-- 
Dr Andy Buckley
Institute for Particle Physics Phenomenology
Durham University
0191 3343798 | 0191 3732613 | www.insectnation.org