[torqueusers] output/error file size limit?
gs at hep.ucl.ac.uk
Fri Oct 28 10:08:17 MDT 2005
> > We are looking at a way to limit the size of the .oJOBNUM and .eJOBNUM
> > files generated by Torque without limiting the "real output" eg root
> > files. In the odd case these files could grow so much they disrupt
> > the operation of the compute node they run on.
> > I looked at the man page for pbs_resources_linux and it has an option
> > listed as "file The largest size of any single file that may be created
> > by the job. Units: size.
> > If I interpret this correctly, it that looks like it limits ALL output
> > which is not very useful for our purpose. Is my interpretation correct, if
> > yes, is there any other option to achieve the wished limitation?
> I don't think that limits the aggregate amount of output, just the
> maximum size of any one file. The code uses setrlimit(RLIMIT_FSIZE,...)
> to accomplish this. However, I'm not sure if this limit would apply to
> the spooled stdout and stderr files.
> You could try setting quotas on $PBS_HOME/spool on each of the compute
> nodes, if the OS and filesystem you're using support them.
Setting user quotas on the spool directory is feasible. How is PBS going
to treat a running job that happens to kill the quota?
More information about the torqueusers