[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [HTCondor-users] API

On Mon, Dec 18, 2017 at 4:44 PM, Todd L Miller <tlmiller@xxxxxxxxxxx> wrote:
>> I have a python script and it kicks off a bunch of threads.  Each of
>> those threads kicks off another bunch of threads. Each of those
>> threads runs a python script using Popen. Currently this is run from
>> cron on one host. I would like to move this to condor to take
>> advantage of multiple machines and achieve parallelism.
> ...
>> Does using the API for this seem like the best way?
>         Generally, the right API depends on how much communication is
> necessary between the different threads (and which direction it is), and how
> long each thread lasts.  I tend to think that HTCondor works best when
> scheduling jobs that don't need to communicate with each other and take
> about an hour to run.  If that describes the last step of your application
> (the Python script called by popen()), then you might benefit from replacing
> that popen() with an HTCondor job.
>         It it takes a while to get to that point, you could consider
> replacing the second-level threads with HTCondor jobs as well.  If the
> second level takes a while, but the scripts it popen()s do not, maybe you
> want to start HTCondor jobs instead of second-level threads but not replace
> the popen()s.

Thanks for the advise. I will start with replacing the popen() with a
condor job.
>> Can anyone point me at an example for using the API?
>         Brian Bockleman has been working on improving the documentation for
> the Python API.  The documentation is here:
> https://htcondor-python.readthedocs.io/en/latest/
> His tutorial from earlier this year has links to earlier tutorials and
> other resources:
> http://research.cs.wisc.edu/htcondor/HTCondorWeek2017/presentations/TueBockelman_Python.pdf

Thanks for this as well. It has more info then the docs I was looking at.