Uploaded image for project: 'Network Observability'
  1. Network Observability
  2. NETOBSERV-1766

Increase loki.WriteBatchSize to 10MB default

    • Icon: Bug Bug
    • Resolution: Done
    • Icon: Major Major
    • netobserv-1.7
    • None
    • FLP
    • False
    • None
    • False
    • Hide
      Previously, there was an inconsistency with the default Loki WriteBatchSize configuration: it was set to 100KB in the CRD default, and 10MB in the OLM sample/default configuration. Both are now aligned to 10MB, which generally provides better performances and less resource footprint.
      Show
      Previously, there was an inconsistency with the default Loki WriteBatchSize configuration: it was set to 100KB in the CRD default, and 10MB in the OLM sample/default configuration. Both are now aligned to 10MB, which generally provides better performances and less resource footprint.
    • NetObserv - Sprint 258, NetObserv - Sprint 259
    • Important

      Description of problem:

      Currently, loki.WriteBatchSize defaults to 100KB, however recent perf tests finds lower default value could cause higher FLP CPU utilization. Comparison Sheet https://docs.google.com/spreadsheets/d/1l9oYXu80HqR4ZsqMd4gIwo1xK59buK8SYptsdXbeVQk/edit?gid=31203062#gid=31203062

      Steps to Reproduce:

      1. Run NDH test with loki.writeBatchSize of 10MB vs default 100KB
      2.
      3.
      

      Actual results:

      Comparison Sheet https://docs.google.com/spreadsheets/d/1l9oYXu80HqR4ZsqMd4gIwo1xK59buK8SYptsdXbeVQk/edit?gid=31203062#gid=31203062
       

      Expected results:

      FLP CPU util can be lower when loki.writeBatchSize is increased to 10MB

      Slack convo: https://redhat-internal.slack.com/archives/C03SHCD5C9M/p1721650562374129 

              jpinsonn@redhat.com Julien Pinsonneau
              rhn-support-memodi Mehul Modi
              Votes:
              0 Vote for this issue
              Watchers:
              4 Start watching this issue

                Created:
                Updated:
                Resolved: