On Aug 24, 2010, at 12:50 PM, Ian Stokes-Rees wrote:
On 8/24/10 12:56 PM, Jaime Frey wrote:
On Aug 24, 2010, at 10:56 AM, Ian Stokes-Rees wrote:
I have my
X509_USER_PROXY pointing to a valid proxy cert in my
NFS-mounted home directory
but condor seems to insist on looking for it in /tmp,
despite what the
docs say here:
I can copy the proxy cert to /tmp, but I need it in the NFS
other services on other hosts, and copying it around is just
problems later, when one expires, and the other is
Is this a bug, or am I doing something wrong? Trace below.
The problem is that by default, the environment variables
that exist when you run condor_submit aren't propagated to the
environment that the condor_dagman job runs under. So when
condor_dagman runs condor_submit to run jobs in the dag,
condor_submit doesn't see X509_USER_PROXY.
You have two options:
1) Do as Zach suggests and set x509userproxy in the dag node
I should have mentioned that I already do this in the classad files
referenced by the DAG:
X509userproxy = /nfs/home/ijstokes/.globus/x509up_u1004
So that doesn't solve the problem, although perhaps if I took the
intermediate DAG classad, modified it to include this, and submitted
that, then it would work.
If your submit files contain the x509userproxy attribute, then that's the file Condor will use, regardless of the default proxy or X509_USER_PROXY environment variable.
2) Use the -import_env argument to condor_submit_dag.
I'll give it a try. Why wouldn't this be the default? The
sub-processes squash the environment they automatically inherit from
Maybe it should be the default. But it won't happen automatically, because condor_dagman is spawned by condor_schedd. There's no direct connection back to your user environment. condor_submit_dag needs to explicitly copy your environment into the job description of the dagman job.
Thanks and regards,
UW-Madison Condor Team