Integration of JURECA
Technical Issues
-
Create StartD IDTOKEN to join the c4p-htcondor-dev
pool and send it to Vadim -
Add HTCondor StartD configuration to https://gitlab-p4n.aip.de/compute4punch/c4p-htcondor-configs -
Add GitLab deployment token to access this configuration and send it to Vadim -
Run HTCondor worker in Jülich that connects to the c4p-htcondor-dev
pool
Politcal & Policy Issues
-
FZJ requires that every user needs an account on JURECA and runs only workflows belonging to her/his account. This is in principle against the idea of Compute4PUNCH to share integrated resources as much as possible by utilising a proxy user to integrate resources. -
The nodes at FZJ HPCs do not have outgoing network connectivity. JUWELS nodes even not an IP network. On JURECA we could use a tunnel via the login node to access at least the HTCondor and CVMFS services as a workaround. This will certainly not scale for any data transfers to Storage4PUNCH. -
Disabled user namespaces avoid utilisation of multiple container layers in order to isolate HTCondor WN from the actual job environment provided a different container. Workaround could be to use a local HTCondor installation and just run the job environment in a dedicated container.
Edited by Manuel Giffels