Re: [HTCondor-users] Scratch Directory problem

[Brian Candler <b.candler@xxxxxxxxx>]

On 27/05/2013 15:01, Michael Fienen wrote:
> I'm running into a problem regarding temporary run directories. I am submitting jobs in Vanilla universe to a cluster of Linux Machines. For some reason, on one of them, rather than the jobs running in, for example, /var/lib/condor/execute/dir_<PID>/.…. where <PID> is a unique number tied to the process ID, in some cases, they end up running in roughly the same location they were submitted from (like /home/username/launch_dir)
Have you set FILESYSTEM_DOMAIN to the same value on every machine? Then 
this is going to happen.

You can fix this by putting
FILESYSTEM_DOMAIN = $(FULL_HOSTNAME)
in the config. Or, in every submit file, you can set 
"should_transfer_files = yes"
> Relatedly, I hoped that I could just exclude a machine in requirements like (Target.Machine != "xxx.xxx.xxx.xxx") but that didn't work either.
I think you need to use FQDN for that.
Requirements = (Machine != "host.your.domain")
But normally a better approach would be to classify your machines:

HasWibble = True
STARTD_EXPRS = HasWibble, $(STARTD_EXPRS)

and then your job can say

Requirements = HasWibble