Todd Tannenbaum | 8 Feb 18:32 2012
Picon

drmaa and persistent job handles


Hi DRMAA experts -

IIRC, last time I looked into DRMAA it used ephemeral job ids - in other
words, when submitting a job via DRMAA you got back a handle to the job,
but this handle was meaningless across a restart of the application
(similar to a file descriptor). So if application X submits a job and
then application X is restarted (e.g. the submit machine is rebooted),
there was no guarantee that X could continue to monitor/manage the job.

Is the above still true?  If so, is there a well-known work-around
and/or are there plans to address the issue in a future DRMAA spec?

Thanks much,
Todd

--

-- 
Todd Tannenbaum                       University of Wisconsin-Madison
Center for High Throughput Computing  Department of Computer Sciences
Peter Tröger | 9 Feb 15:00 2012
Picon

DRMAA is now GFD.194

Dear all,

with some minor corrections (thanks to Daniel Katz), DRMAAv2 is now an officially approved OGF specification:

http://www.ogf.org/gf/docs/?final

Congratulations and thanks to everyone who helped with this. 

How do we proceed now ?

I rejected a request for DRMAA sessions at OGF 34, since we need to finalize the crucial C binding first. 
Please, state your ideas on how to proceed with this in the most efficient way. The latest header file is
always here:

http://bit.ly/xH2rfb

One idea would be to sharpen the C binding concepts by a process-based reference implementation of a fake
job submission system. Anybody interested to join me ?

Best regards,
Peter.

Peter Tröger | 9 Feb 16:18 2012
Picon

Re: drmaa and persistent job handles

Hi Todd,

you might have seen the announcement about DRMAAv2 becoming an official OGF standard. 
'Old DRMAA' had the problem you described. DRMAAv2 has a concept of session persistency, which leads to the
demanded reusability of job identifiers.

The workaround for 'old DRMAA' was to assume that job identifiers remain valid between sessions. For all
implementations I know, this is a valid assumption, since the DRM system identifiers are used as DRMAA job identifiers.

Best regards,
Peter.

P.S.: We have a strong interested in DRMAAv2 adoption for Condor. Please contact me if you see a change to
make that happen.

Am 08.02.2012 um 18:32 schrieb Todd Tannenbaum:

> 
> Hi DRMAA experts -
> 
> IIRC, last time I looked into DRMAA it used ephemeral job ids - in other
> words, when submitting a job via DRMAA you got back a handle to the job,
> but this handle was meaningless across a restart of the application
> (similar to a file descriptor). So if application X submits a job and
> then application X is restarted (e.g. the submit machine is rebooted),
> there was no guarantee that X could continue to monitor/manage the job.
> 
> Is the above still true?  If so, is there a well-known work-around
> and/or are there plans to address the issue in a future DRMAA spec?
> 
(Continue reading)

Daniel Templeton | 9 Feb 18:06 2012

Re: DRMAA is now GFD.194

And there was much rejoicing!  Congrats everyone!

Daniel

On 2/9/12 6:00 AM, Peter Tröger wrote:
> Dear all,
>
> with some minor corrections (thanks to Daniel Katz), DRMAAv2 is now an officially approved OGF specification:
>
> http://www.ogf.org/gf/docs/?final
>
> Congratulations and thanks to everyone who helped with this.
>
> How do we proceed now ?
>
> I rejected a request for DRMAA sessions at OGF 34, since we need to finalize the crucial C binding first.
> Please, state your ideas on how to proceed with this in the most efficient way. The latest header file is
always here:
>
> http://bit.ly/xH2rfb
>
> One idea would be to sharpen the C binding concepts by a process-based reference implementation of a fake
job submission system. Anybody interested to join me ?
>
> Best regards,
> Peter.
>
>
>
>
(Continue reading)

Peter Tröger | 29 Feb 09:07 2012
Picon

DRMAA at OGF 34

Dear all,

even though we planned to have the next big DRMAA meeting at OGF35 (June), I was asked by the OCCI people to
join them already at OGF34. There is some interest in getting the OCCI<->DRMAA binding done.

If any of you wants to join us - I will be in Oxford from Tuesday to Wednesday.

Best regards,
Peter.

Attachment (smime.p7s): application/pkcs7-signature, 5869 bytes
Dear all,

even though we planned to have the next big DRMAA meeting at OGF35 (June), I was asked by the OCCI people to
join them already at OGF34. There is some interest in getting the OCCI<->DRMAA binding done.

If any of you wants to join us - I will be in Oxford from Tuesday to Wednesday.

Best regards,
Peter.


Gmane