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

Re: [HTCondor-users] thoughts on HTCondor python bindings submit improvements



> I think it would be extremely useful to have a way to easily take a submit file on disk and to submit it from python.  That should be a standalone interface

 

Where would you hang this?  on the transaction() object? on the schedd object?

 

> with schedd.transaction() as txn:

>   for foo in iterdata:

>      sub.queue(txn, foo)

 

This unfortunately would mean that we would have to defer talking to the schedd until we tear down the txn object, we couldnât do anything in the queue() method except save the data for later processing.   Late materialization (which we hope will become the default someday) means that what we send to the schedd is a cluster ad, submit digest, and a complete set of iterdata.

 

So the c++ code needs to get all of that in a single call. hence the design that gives us

 

with schedd.transaction() as txn : sub.queue(txn, iterdata)

 

instead of

 

with schedd.transaction() as txn : for foo in iterdata: sub.queue(txn, foo)

 

Of course, we could do both, but that would be non-pythonic (right?)

 

-tj

 

 

From: HTCondor-users [mailto:htcondor-users-bounces@xxxxxxxxxxx] On Behalf Of Brian Bockelman
Sent: Tuesday, April 24, 2018 8:45 PM
To: HTCondor-Users Mail List <htcondor-users@xxxxxxxxxxx>
Subject: Re: [HTCondor-users] thoughts on HTCondor python bindings submit improvements

 

Hi TJ,

 

I think it would be extremely useful to have a way to easily take a submit file on disk and to submit it from python.  That should be a standalone interface -- not 100% sure we should try to wedge it into the Submit object.  Perhaps something like this:

 

with schedd.transaction() as txn:

    htcondor.submit_file(fp)

 

?

 

Anyhow, I've always struggled with ways to combine condor submit files and python approaches.  One idea may be to do something like this:

 

with schedd.transaction() as txn:

   for foo in iterdata:

      sub.queue(txn, foo)

 

where foo is a dictionary-like object that expands any missing macros in the submit object.

 

Brian



On Apr 24, 2018, at 4:42 PM, John M Knoeller <johnkn@xxxxxxxxxxx> wrote:

 

I have been thinking about how to make HTCondor python bindings submit as capable as condor_submit,

 

The main thing missing right now is that the python bindings supports only the simplest possible queue statement.

I think that the bindings needs to support the full range of queue options, but in a pythonic way, so Iâm thinking this

 

  1. change the htcondor.Submit class so it can be constructed using the text of a valid submit file. like this

 

sub = hcondor.Submit(âââ

                executable = /bin/echo

                queue 2 args in ( one, two, three )

                âââ)

 

  1. add a method to the htcondor.Submit class that lets you get an iterator from the queue itemdata, like this

 

for item in sub.itemdata() :

print item

 

which (given the submit file above) would print:

{âargsâ : âoneâ}

{âargsâ : âtwoâ}

{âargsâ : âthreeâ}

 

  1. add a method to the htcondor.Submit class that would queue jobs to the schedd using an iterator for the queue itemdata.  like this

 

with schedd.transaction() as txn : 

sub.queue_with_iter(txn, 2, iter(itemdata))

 

                which would submit 2 jobs for each item returned by the iterator.

The iterator could be sub.itemdata(), or any iterator that returned an equivalent type of item â either a dict, or a string.

 

The reason this would need to be a new queue method on the submit object is that the current queue method already

allows a third argument that is a list that is used to return the submitted job classads.   This would conflict with adding

an optional third argument to be the queue itemdata iterator.   Having a new method also allows the return type to be

change to be more pythonic â return a tuple of <clusterid,classad> for instance.

 

  1. (optionally) change the current queue method on the htcondor.Submit object to honor any QUEUE statement supplied in the constructor of the Submit object.  It would be a runtime error to pass more than just the txn argument to the queue method in this case. (since the count argument would conflict)

 

Thoughts?

Is anyone currently using the third argument to the queue() method of the current htcondor.Submit class?

Would anyone be harmed if I just removed that argument and replaced it with an optional iterator?

 

-tj

 

 

 

 

_______________________________________________
HTCondor-users mailing list
To unsubscribe, send a message to 
htcondor-users-request@xxxxxxxxxxx with a
subject: Unsubscribe
You can also unsubscribe by visiting
https://lists.cs.wisc.edu/mailman/listinfo/htcondor-users

The archives can be found at:
https://lists.cs.wisc.edu/archive/htcondor-users/