Update to HTCondor Job submission

1. Support for -spool Option: Users can now submit HTCondor jobs from EOS avoiding "ERROR: EOS Submission is not currently supported by the HTCondor Service." with the use of -spool option of condor_submit. HTCondor will asynchronously transfers output files (.err, .out, .log) from its source to the schedd and then to the execution node instead of being immediately transferred back to the submission node.

  • Need to add to ~/.config/bamboorc

    [htcondor]
    spool= True
  • You can include the transfer_output_files attribute in ~/.config/bamboorc  to specify which files should be copied back to the schedd.

  • Or Manually retrieve output files once they are completed

    condor_transfer_data <JobID>

2. Custom Output File Paths: You can also use transfer_output_remaps or output_destination attributes to change the destination path for output files. This means output files no longer need to be transferred back to the job’s initial working directory, which makes PR !252 no longer necessary.

This part need more testing

  • Missing root:// Protocol: HTCondor needs the root:// protocol to interact with EOS for file transfers. Without it, HTCondor cannot locate or write to the specified directory.

3. Fix some HTcondor attributes where previously by passed through Bamboo:

  • request_memory
  • request_cpus
  • request_disk
  • accounting_group attribute is now mandatory for USTC users.

4. Fix for output files naming: Previously, output files (.err, .out, .log) did not include the clusterID in their names, which cause resubmitted jobs to overwrite existing output files for the same job ID.

Edited by Khawla Jaffel

Merge request reports

Loading