Skip to content

Latest commit

 

History

History
19 lines (11 loc) · 2.33 KB

README.md

File metadata and controls

19 lines (11 loc) · 2.33 KB
Raw

Condor utilities

Usage

  1. In run.sh, set all the parameters: the number of events per file, the location of the configuration text file (which describes each set of Pandora jobs), the location of a Pandora setup script, the number of files per job, whether to validate and some details about the validation.

  2. In the configuration text file (default location config.txt), write the description of the jobs to be run. The first row is the header row; each subsequent row represents a set of Pandora jobs to be run. The first four headings are PandoraLocation, SettingsFileLocation, RecoOption and SampleLocation (in that order), which must be filled in for every job.

  3. Further headings in the configuration file can be used to represent changes to make to the provided settings file using sed. To do this, replace the value in the settings file with a dummy name, surrounded by £s, then use the dummy name as the header. E.g. if we want to run a job with slicing on and slicing off, replace <ShouldPerformSlicing>true</ShouldPerformSlicing> in the settings file with <ShouldPerformSlicing>£ShouldPerformSlicing£</ShouldPerformSlicing>, then add a heading ShouldPerformSlicing to the config file, and set it as true or false for each row. Every row must have a value for every heading. If the heading was for one settings file and does not feature in another, then an arbitrary value can be put in the config file and sed will do nothing. This allows for easy parameter sweeps.

  4. From the base directory, run source run.sh and the jobs will be run batch-wise. Concatenated root files are saved in catroots, any stdout from the validation script is saved in results. The results are numbered numerically by batch number, beginning at 0.

Tips/issues

  • Running source run.sh deletes any files in catroots, roots, results, xmls, xml_bases etc., so move the results out of there if they're to be saved.

  • To curate the jobs, it watches condor_q and counts the remaining jobs so it can concatenate them and run validation once they're all done. This means that running other condor jobs will cause this to wait until all the jobs are done, regardless of whether it started them or not. There's probably a way to get the job ID through the Python script and then use that to only look at the relevant jobs.