[torqueusers] Re: Unexpected 'qsub: Access from host not allowed, or unknown host' error

[Kevin Murphy]

I think this is a buffer overrun issue caused by the fact that the 
PBS_DEPEND_LEN constant (2040 -- very inadequate for my purposes) is 
used for allocation purposes but is never used to limit copying into the 
buffer.  I increased PBS_DEPEND_LEN to 32768 and recompiled, and my 
pipeline submits fine now.

However, the buffer overrun bug in parse_depend.c needs to be fixed.

-Kevin

Kevin Murphy wrote:
> I'm not certain that this problem is caused by some sort of numeric or 
> size limit, but here's what I'm experiencing (with Torque 2.3.1 on 
> CentOS 5.1):
>
> For "no good reason", a qsub is failing with this error:
>
> qsub: Access from host not allowed, or unknown host
>
> This job would have been the 2,348th job in a pipeline of dependent 
> jobs, all placed in a held state initially, with the very first job in 
> a manual hold state.
>
> The "#PBS -W" line in the qsub script is 13,984 bytes long (thanks to 
> a hefty 'depend' clause).
>
> This pipeline works fine with a few hundred files but breaks with a 
> few thousand.  My submission code is in Perl, and as far as I know, 
> it's not subject to any size limits that would cause peculiar 
> malfunction.
>
> Thanks,
> Kevin Murphy
>
>