Uploaded image for project: 'OCP Technical Release Team'
  1. OCP Technical Release Team
  2. TRT-813

Audit disruption fallbacks for safety/accuracy



    • Story
    • Resolution: Done
    • Normal
    • None
    • None
    • None
    • False
    • None
    • False


      In next_best_guess.go, we have logic that will fall back to nurp combos if we don't have one for the current job.

      Some important things to remember:

      • right now data doesn't go into the query_results.json file in origin if we do not have the required 100 runs to do a semi-accurate P99.
      • data also doesn't go into the query_results.json file in origin for older releases, we only have 4.13 data in master of origin now.

      We recently hit this with gcp ovn micro upgrades, which we don't really have data for, falling back to minor upgrades, and showing big differences for LB backend.

      How often are we falling back across CI? How often are we failing if we fall back?
      search.ci might help us see this.

      Does the list of fallbacks look safe, should some be removed?

      Investigate what happens when we move origin master from 4.13 to 4.14. This would be handled in the ci-tools repo when we generate the PR to update this file, and Egli did add some logic for this case. I cannot immediately remember what it does though, fallbacks could be important during this time.




            rh-ee-dmistry Deep Mistry
            rhn-engineering-dgoodwin Devan Goodwin
            0 Vote for this issue
            2 Start watching this issue