[GE users] qresub as other user

Sebastian Stark stark at tuebingen.mpg.de
Wed Nov 16 16:18:12 GMT 2005


    [ The following text is in the "iso-8859-1" character set. ]
    [ Your display is set for the "ISO-8859-10" character set.  ]
    [ Some special characters may be displayed incorrectly. ]

On Wednesday 16 November 2005 16:55, Andreas Haas wrote:
> > > There is a job error and a queue error for error conditions. Job
> > > error is for cases when the end user made a failure that must be
> > > fixed before rerun is possible. The queue error is for problem with
> > > set-up problems which can be fixed by admins only.
> >
> > 99% of the errors in our cluster are like this:
> >
> >   failed changing into working directory:11/15/2005 17:48:30
> > [1792:20744]: error: can't chdir to /agbs/cluster/chrisd: No such file or
> > direct
> >
> > (of course people notice right *after* submitting thousands of jobs...)
> >
> > In case of a not accessible NFS mount a user error occures that can not
> > be fixed by the user.
>
> I understand. The problem here is that one can not tell easily whose
> fault it was:
>
> (1) it is clearly the users fault if he used -cwd option to run a job on
>     a machine where the qsub current working directory doesn't exist and
>     should not exist

Agreed.

> (2) it is the admins fault if the directory should exist though but it
>     wasn't mounted.

> Question is simply how should Grid Engine know whether that directory
> should have been available? Not even use of -cwd option can be used as
> reasonable indication.
>
> Though Grid Engine could treat this always as queue error, but then
> any ordinary user could activate queue error state for the entire
> cluster simply by misusing -cwd option ...

I clearly see this as a user error. A user should always check if a directory 
is writable before he's actually trying to put stuff into it. Especially if 
it is a program that will be run at an unpredictable time in the future.

So I think Grid Engine is doing exactly the right thing here. (besides mailing 
me hundreds of error messages...)

-- 
Sebastian Stark -- http://www.kyb.tuebingen.mpg.de/~stark
Max Planck Institute for Biological Cybernetics

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list