[SAGA-RG] SAGA Resource Package Comments

[Andre Merzky]

Hi Steve,

On Wed, Jan 16, 2013 at 12:06 PM, Steve Fisher <dr.s.m.fisher at gmail.com> wrote:
>
> On 16 January 2013 09:14, Andre Merzky <andre at merzky.net> wrote:
>>
>> Hi Steve,
>>
>> Either way, thanks for the comments.  This will likely mean that we'll
>> trim down the resource API to simply perform
>>
>>   - describe resource
>>   - acquire resource
>>   - release resurce
>>   - get job/data service for resource
>>
>> Best, Andre.
>>
>
> Andre,
>
> I am happy with the first three operations - the last is less nice as not
> all resources will be able to support, for example,  getJobService(). Better
> perhaps would be getService() where the Service is the same Service class
> that SD provides. This would support services of any type.
>
> Steve

Well, we do have different resource types (*), and so far I don't see
a reason nor argument to change that.  So, for a compute resource, I
would expect to always be able to retrieve a job service endpoint, for
a StorageResource, I would expect to get a filesystem or replica
endpoint.  Also please consider that the main use case for the
resource package is basically:

  - create a vm
  - run a couple of jobs on the VM


Originally, that was supposed to look like:

  vm = resource_manager.aquire_resource (resource_description)
  vm.create_job (job_description)


As we now want to decouple the resource and job parts, this will
likely look like:

  vm = resource_manager.aquire_resource (resource_description)
  js = vm.get_job_service ()
  js.create_job (job_description)


Yes, adding a generic service inspection / retrieval to the resource
makes the package more generic, obviously -- but it will make the main
use cases more
complex, too (error checks left out):

  vm  = resource_manager.aquire_resource (resource_description)
  sds = vm.get_service_descriptions ()
  js  = None
  for sd in sds :
    if sd.type == 'saga.job.service' :
      js = saga.job.service (sd.url)
      break
  js.create_job (job_description)


or, if you allow the filters on the sd retrieval:

  vm  = resource_manager.aquire_resource (resource_description)
  sds = vm.get_service_descriptions
(service_filter='Capabilities="org.ogf.saga.service.job"')
  js  = saga.job.service (sds[0].url)
  js.create_job (job_description)


I rather see the role for SD/ISN to be used for inspection purposes,
to understand / inform the application about the properties of the
resources (number or CPUs, load, lifetime, ...) -- we do have dynamic
scheduling use cases which would need that kind of information about
the resources, and I'd rather expose those data via the existing
SD/ISN packages that getting them out-of-band...

My $0.02,

  Andre.



(*) although in the reduced resource package scope, we don't have any
use for network resources anymore.


-- 
There are only two hard things in Computer Science: cache invalidation
and naming things.

-- Phil Karlton