amaltaro
commented
1 month ago Hi @hassan11196 , I compiled a list of potential datasets/workflows with duplicate lumis and placed them in this text file: https://amaltaro.web.cern.ch/forWMCore/Issue_12092/potential_dups.txt
Please let me know if you prefer it in a different format.
This list is based on:
- searching for the specific error pattern in the JobAccountant component logs in all agents
- extracting the lfn name
- converting the lfn to dataset
- finding a workflow that produces the dataset above
Impact of the bug WMAgent
Describe the bug As described in these 2 comments: https://github.com/dmwm/WMCore/issues/11956#issuecomment-2329503741 and https://github.com/dmwm/WMCore/issues/11956#issuecomment-2329515605
we can conclude that the way JobAccountant deals with jobs that do not have any location for one or more of the output files is not sufficient. The component is properly failing those jobs, but somehow the output files from such jobs are still considered for upcoming jobs (e.g. merge jobs), which is a terrible mistake!
How to reproduce it It was never understood how output files are reported without any location, so the reproducibility of this issue is hard to say.
Expected behavior First, a job should never report an output file without any location. But this is somehow out of the scope of this issue.
Second, if JobAccountant identifies such jobs, in addition to marking those jobs as failed, it should as well disregard any output files generated by those jobs. Those output files cannot be considered for subsequent jobs down in the task dependency chain.
Additional context and error message None