1617272 : Probleme mit HTCondor Jobs

Created: 2026-07-15T08:21:23Z - current status: new

Summary

A user reports that their NAF batch jobs, which previously ran successfully, are failing with an error related to Git configuration. The stderr indicates:

error: could not lock config file /afs/desy.de/user/[REDACTED]/.gitconfig: File exists

The stdout shows that the script attempts to execute:

git config --global --add safe.directory /cvmfs/belle.cern.ch/tools

The issue persists even after a reboot. Other jobs using the same environment script run fine.

Anonymized Issue Description

Jobs fail during environment setup due to a conflict when modifying the global .gitconfig file in AFS (/afs/desy.de/user/[USERNAME]/). The error suggests a race condition or permission issue when multiple processes attempt to modify the same file simultaneously. The problematic command (git config --global) is part of the script, though its necessity is unclear, especially since other jobs sharing the same script succeed.

---

Possible Solution / Next Steps

  1. Check for Concurrent Access:
  2. Multiple jobs may be trying to modify /afs/desy.de/user/[USERNAME]/.gitconfig simultaneously. Since AFS handles file locking differently than local filesystems, this can cause conflicts.

  3. Workaround: Modify the script to use --local instead of --global for Git configurations (if project-specific settings suffice), or ensure the .gitconfig file is pre-configured before job submission.

  4. Verify File Permissions:

  5. Ensure the user has write permissions for .gitconfig in AFS. Run ls -la /afs/desy.de/user/[USERNAME] to confirm ownership/access rights.

  6. Prevent Redundant Configurations:

  7. Add a check in the script to skip git config commands if the setting already exists: bash git config --global --get-all safe.directory | grep -q "/cvmfs/belle.cern.ch/tools" || \ git config --global --add safe.directory /cvmfs/belle.cern.ch/tools

  8. Test Interactively:

  9. Use condor_submit -i to debug interactively. Verify whether the issue reproduces in a single session or only under concurrent job execution.

  10. Contact Experiment-Specific Support:

  11. If unresolved, reach out to naf-belle-support@desy.de (as this involves Belle-related CVMFS paths).

References