Am 19.09.2012 um 18:38 schrieb Dave Love:

> William Hay <[email protected]> writes:
> 
>>>> As far as I remember, you can distinguish qlogin and command-less qrsh
>>>> with JSV if necessary.
>>> 
>>> For the client side JSV yes, as the name of the command will be
>>> output for CLIENT. But not for the ones started by the qmaster
>>> (i.e. server side JSV) as the CMDNAME will be NONE in both cases and
>>> CONTEXT is "server"/CLIENT is "qmaster". You can't even decide
>>> between `qrsh` with command and `qsub` AFAICS.
> 
> It sounds as if a bug report is called for.  Could someone who knows
> what seems to be missing there put it in the tracker?

Do you think something is missing? For serverside JSV it's just always:

CLIENT=qmaster
CONTEXT=server
CMDNAME=NONE

and that is fixed. You mean something like ISSUING_CLIENT should be there?

-- Reuti


>> We currently distinguish between qrsh/qlogin and batch jobs in the
>> server side JSV by the presence of QRSH_PORT in the requested
>> environment.  I believe the presence of this environment variable is
>> an undocumented feature and IIRC at least one of the forks was
>> proposing to get rid of it.  Also if you are worried about abuse
>> someone could probably set QRSH_PORT by hand when submitting a qsub
>> command
> 
> Isn't the port info crucial to making the comms work?
> 
> I assume it would be straightforward at least to have the client stuff
> QRLOGIN in the environment if it's not there at that stage, but that
> doesn't sound like the right thing, especially in view of the strictures
> about environment size.
> 
> Why do people need to make the interactive/batch distinction?  I wonder
> what really needs testing, especially as there's also potential DRMAA.
> 
> -- 
> Community Grid Engine:  http://arc.liv.ac.uk/SGE/

_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users

Reply via email to