Re: [Condor-users] condor_run vs. condor_submit and non-nfs directories

["Erik Paulson" <epaulson@xxxxxxxxxxxx>]

On 5/30/07, Christoph Spielmann <cspielma@xxxxxxxxxx> wrote:
>  Erik Paulson wrote:
>  On 5/30/07, Christoph Spielmann <cspielma@xxxxxxxxxx> wrote:
>
>
>  hi everybody!
>
> We use condor on one of our linux-clusters here. The installation seems
> to be okey, but when i try to submit a job to condor from a non-nfs
> directory it failes with the famous condor_shadow (condor_SHADOW)
> EXITING WITH STATUS 112 error message. The detailled error message is:
>
> 5/30 12:15:37 (2203.0) (29451): Job 2203.0 going into Hold state (code
> 6,2): Error from starter on vm2@xxxxxxxxxxxxxxxxxxxxxxxx: Failed to
> execute '/tmp/.condor_run.29439': No such file or directory
> 5/30 12:15:37 (2203.0) (29451): ZKM: setting default map to (null)
> 5/30 12:15:37 (2203.0) (29451): **** condor_shadow (condor_SHADOW)
> EXITING WITH STATUS 112
>
> I searched the mailing-list archives and found quite alot of ppl with
> the same problems but none of the proposed solutions worked for us. We
> tried to work with version 6.8.5 and 6.9.2 both dynamically linked. The
> problem shows up on both versions. Sometimes it does work but in 99 % of
> the trial runs it doesn't.
>
> The funny thing is that it doesn't work when i use condor_run in
> combination with a shell-command like /bin/hostname or /bin/date but
> when i write a simple hello-world c-program, a submit description file
> for that c-program and submit the description file with condor_submit it
> works as expected. Even on non-nfs directories!
>
>
>  condor_run does not use file transfer. You must have a shared
> filesystem to use condor_run, or at least have the executable in the
> same place in every machine. (That is why /bin/hostname works).
>
> I'd bet the reason it works on a few occasions is that every now and
> then your job runs on the submit machine, and can find the executable.
>
> -Erik
> _______________________________________________
> Condor-users mailing list
> To unsubscribe, send a message to
> condor-users-request@xxxxxxxxxxx with a
> subject: Unsubscribe
> You can also unsubscribe by visiting
> https://lists.cs.wisc.edu/mailman/listinfo/condor-users
>
> The archives can be found at:
> https://lists.cs.wisc.edu/archive/condor-users/
>
>
>  Hello Erik!
>
>  Well i just checked and both hostname AND date are on all machines in the
> same place (/bin) so that's not the problem. Actually the filesystem-root of
> the nodes is mounted via nfs just machine-specific things like /tmp, /etc...
> are mounted seperately on each machine. But are mounted all on the same
> place of course...

But that's still your problem - if you use condor_run from /tmp,
Condor assumes that /tmp is shared between all machines. However, /tmp
is private to each machine, which is why Condor on the remote machine
complains:

Error from starter on vm2@xxxxxxxxxxxxxxxxxxxxxxxx: Failed to execute
'/tmp/.condor_run.29439': No such file or directory

You need to use condor_run in a directory that's shared between all
hosts (like /home or /users or however you might have you machines
setup.)

Or, just don't use condor_run, and just use condor_submit with file transfer.

-Erik

> _______________________________________________
> Condor-users mailing list
> To unsubscribe, send a message to
> condor-users-request@xxxxxxxxxxx with a
> subject: Unsubscribe
> You can also unsubscribe by visiting
> https://lists.cs.wisc.edu/mailman/listinfo/condor-users
>
> The archives can be found at:
> https://lists.cs.wisc.edu/archive/condor-users/