UWS POST PHASE=RUN

Matthew Graham mjg at cacr.caltech.edu
Wed Nov 5 20:34:39 CET 2014


Hi,

On Nov 5, 2014, at 11:22 AM, Walter Landry wrote:

>> On Nov 5, 2014, at 8:33 AM, Matthew Graham wrote:
>>> On Nov 5, 2014, at 7:53 AM, Walter Landry wrote:
>>>> Just to be clear, I am proposing getting rid of PENDING.  To handle
>>>> the Patrick's use-case, you can ask the service for estimates, but you
>>>> can not modify anything that is persistent.  You can only cancel them.
>>> 
>>> So a job is submitted and automatically entered into a QUEUED state
>>> which is then persistent?
> 
> If you only ask for an estimate, nothing is queued.  The service
> promptly forgets about your request after it returns an estimate.

But PENDING can also be about a scheduled job (see below).

> 
>>> The QUEUED state can be useful in third-party data transfers where
>>> you are negotiating with two VOSpaces and do not want the job
>>> running until both client and server are agreed on the
>>> transaction. This is not an estimate but negotiation.
> 
> Can you give an example of this negotiation?  What kind of information
> is being exchanged?

I want to transfer data from VOSpace1 (V1) to VOSpace2 (V2). I contact V2 to negotiate the transfer details and schedule the transfer - estimation would be useless here since as you say all details are promptly forgotten. I then contact V1, again the appropriate negotiation and when I am happy that this looks good, initiate the transfer with PHASE = RUN.

> 
>>> More generally, though, the problem with getting rid of PENDING is
>>> that it will not only impact UWS but also VOSpace and possibly TAP
>>> and the code bases for those already out there.
> 
> Getting rid of PENDING would certainly not be a minor modification to
> UWS.  It would, however, make UWS easier to implement, since it
> integrates better with existing queuing software (e.g. SLURM, LSF,
> PBS).


But the current usage integrates with other IVOA specs and I think the cost of this change is prohibitive and not essential.

	Cheers,

	Matthew



More information about the grid mailing list