Re: [ogsa-wg] [DRMAA-WG] Grid Command Line interfaces

4 Jun 2008


      DRMAA is narrowly focused on offering an API, but as was already pointed 
out, as soon as you have a Perl binding (which we do), the API itself 
also becomes a set of commands.

Daniel

Andre Merzky wrote:
...
Well, DRMAA is an API specification as well, so does not
really define command line tools in the spec.  There are a
number of tools available, but I am not qualified to say if
the DRMAA group intended to have a complete API mapping, or
plan to standardize these tools, etc.
Cc to the DRMAA group, maybe they have input on the topic...
Quoting [Omer F. Rana] (Jun 04 2008):
...
Date: Wed, 04 Jun 2008 16:35:13 +0100
From: "Omer F. Rana" <o.f.rana@cs.cardiff.ac.uk>
To: Andre Merzky <andre@merzky.net>
Subject: Re: [ogsa-wg] Grid Command Line interfaces
Hi Andre,
Thanks. It might be useful to understand the relationship between these 
and commands provided
in DRMAA perhaps? Just a thought.
best wishes
Omer
Andre Merzky wrote:
...
Hi Omer
Quoting [Omer F. Rana] (Jun 04 2008):
...
Hi,
Doesn't DRMAA already do this?
Andre: I thought the aim of SAGA was to provide a programmatic
API rather than a command line tools interface?
Of course, you are perfectly right.  However, several people
(including our won group) use SAGA for the purpose to write
such command line tools (which are then middleware
independent), and thus we considered mappings from the SAGA
API to command line tools since quite a while...
Best, Andre.
...
regards
Omer
Steven Newhouse wrote:
...
It may make sense to define common tools for job:
Submit
Status
Terminate
I'm not sure what broader interest we would have to do generic SAGA 
commands.
Steven
...
-----Original Message-----
From: Andre Merzky [mailto:andre@merzky.net]
Sent: Wednesday, June 04, 2008 2:48 PM
To: Steven Newhouse
Cc: Andre Merzky; ogsa-wg@ogf.org; ogsa-hpcp-wg@ogf.org
Subject: Re: [ogsa-wg] Grid Command Line interfaces
Quoting [Steven Newhouse] (Jun 04 2008):
>> is a SAGA command line binding something you would
>> conider worth pursuing?  We actually started to do
>> something like that, in a pet project...
>>       
>>            
>>                 
> Do you mean the ability to implement any defined command
> line interface using the SAGA APIs? (i.e. internal to the
> command) Or To define a set of command line tools to cover
> elements of the SAGA API?
>     
>          
>               
The latter.  For example, for the SAGA call
class saga::filesystem::file
{
  void copy (saga::url src, saga::url tgt, it flags);
}
define the command line tool
saga_file_copy [flags] <src> <tgt>
flags:
      session related flags
        -s|--session <s>   run command in session s
        -c|--context <c>   use context c
operational flags
        -a|--async=<Sync|Async|Task>
                            use async mode Sync, Async or Task
                            default is Sync
      call specific flags
        -r|--recursive  copy recursively
        -o|--overwrite  overwrite target if exists
        ...
So, the command line tools would basically reflect what we
define in the SAGA API spec, with a set of flags which are
consistent for all command line tools such defined.
A session could look like:
# saga_create_context --name=my_context --type=UserPass --user=anon
<prompts for password>
# saga_create_session --name my_session --add_context=my_context
# /bin/date | saga_file_cat --session=my_session --write
gsiftp://localhost/tmp/in.dat
# saga_file_copy --session=my_session gsiftp://localhost/tmp/in.dat
gsiftp://remotehost/tmp/out.dat
# saga_file_cat  --session=my_session gsiftp://remotehost/tmp/out.dat
Wed Jun  4 14:43:27 CEST 2008
or, with some default assumptions of course (default session
and context):
# /bin/date | saga_file_cat --write gsiftp://localhost/tmp/in.dat
# saga_file_copy gsiftp://localhost/tmp/in.dat
gsiftp://remotehost/tmp/out.dat
# saga_file_cat  gsiftp://remotehost/tmp/out.dat
Wed Jun  4 14:43:27 CEST 2008
Best, Andre.
PS.: As for option (a) of yours: yes, that would be trivial to
   implement in SAGA :-)  Well, at least it would be easy (one
   needs to add some magick for state management, to keep track
   of async ops and security credentials between separate calls
   to different tools.
> Steven
>     
>          
>               
--
Nothing is ever easy.
--
ogsa-wg mailing list
ogsa-wg@ogf.org
http://www.ogf.org/mailman/listinfo/ogsa-wg

Daniel Templeton

tags

participants (1)