we
try to run vanilla jobs in singularity on an HPC center. The jobs start and the x509userproxy environment is set for the job. However, the variable is e.g. /tmp/condor/execute/dir_1/proxy. We bind mount the executable directory of the job with SINGULARITY_TARGET_DIR
= /srv
Inside
singularity, the x509userproxy file is under /srv/proxy. The environment variable is still /tmp/condor/execute/dir_1/proxy inside the container. So the x509userproxy environment variable is wrong inside the container.
We
use HTCondor version 8.6.2. Do we have to configure something in HTCondor that the x509userproxy variable is set correct inside the singularity container?
This appears to be a bug in HTCondor. When using the SINGULARITY_TARGET_DIR parameter, the condor_starter updates the environment variables it sets for the job that are related to the execute directory to use the location inside the container.
X509_USER_PROXY was left out by mistake.
We will fix this in a future release. Iâve made a ticket in our bug tracking system:
Thanks and regards,
Jaime Frey
UW-Madison HTCondor Project
|