Re: [classad-users] Dependencies and repetitions in DAGMan


Date: Fri, 5 Jun 2009 14:57:44 -0400
From: Marc Tardif <marc@xxxxxxxxxxxxx>
Subject: Re: [classad-users] Dependencies and repetitions in DAGMan
* Alain Roy <roy@xxxxxxxxxxx> [2009-06-05 13:03 -0500]:
> On Jun 5, 2009, at 12:52 PM, Marc Tardif wrote:
>> I'm not sure I completely understand but I suspect this might provide
>> exactly the kind of flexibility needed to address my remaining use
>> cases for using condor. To make sure I understand, lets take part of
>> the above example:
>>
>>    # Filename: B.dag
>>    Script PRE B modify-script
>>    JOB B B.condor
>>
>> So, when you say "edit B", does that mean that the modify-script would
>> modify B.condor on the fly?
>
> Yes. But I think that B.condor has to exist when you submit the DAG  
> file, and it might read the "Log = " line at the time and you can't  
> change it. I'm not sure about that.
>
>> For example, it could perhaps create the
>> file to contain multiple jobs:
>
> No. DAGMan has a restriction of one job per submit file. However, you  
> could submit a DAG (which is a single job) that turns around and submits 
> lots of jobs.

Thanks for the explanation, adding an extra level of indirection with
a DAG now makes sense. This actually addresses several of the pending
use cases I've been trying to solve.

There is one last use case which I'm not quite sure how to solve:

  How could I express a job which should be run once on two different
  machines?

This is seems similar to the RETRY keyword, except that the job must
be retried on different machines. Might there be a trick to solve this
problem?

Thanks again,
Marc

[← Prev in Thread] Current Thread [Next in Thread→]