Fermilab/HTCondor Minutes February 13, 2015

Tony, Parag, Neha, Zack, Steve, Todd

Agenda and Notes:

Google Compute Engine:

  • 8.2.7 works again with GCE

Nova Gahp:

  • not ready yet
  • TTU fully consumed with NSF cloud project
  • Todd asked them to put source on github or send code to HTCondor team

Gangliad: We have 2 Collectors running in HA mode. The manual suggests only running one condor_gangliad. How should condor_gangliad be configured? Can it be managed by HTCondor HA?

  • if collector and ganglia can handle it then there is no reason why we cannot run two gangliad's (one per collector)

Changing ownership of HTCondor daemons - How to change ownership of condor daemons to start as root instead of non privileged user without affecting existing jobs

  • Daemons can have permissions changed, but you have to worry about all the infrastrure surrounding them. e.g. change ownership/permissions on $(SPOOL) directory and files, all the log files, etc.
  • Alternatively, you can tell HTCondor to run as specific user
  • config file must be owned by root
  • ProcLog file owned by root

Generic vs. License Classads - it seems like license ads are inserted as special type (LicenseAd) while the generic ads are inserted with whatever is set in MyType (glideresource in this case) of the classad

  • See email for deatils
  • HTCondor team will look at this offline and respond via email

Any pitfalls with an incremental upgrade from 8.0.7 -> 8.2.7. Incremental in this case means upgrade collectors from 8.0.7 to 8.2.7, then upgrade submitters, then workers.

GLOW VO - Has a problem with OSG where sites are not enforcing job policies and admins "freak out" about when GLOW jobs run multihtreaded jobs and ban the VO. The problem is the even *nix utilities like sort are now multithreaded.

  • ideally, sites would use something like cgroups to enforce limits
  • discussed CPU_AFFINITY
  • discussed how to pin jobs to cores dynamically
  • glideinWMS Frontend allow custom scripts to be run that can modify the pilot