Uploaded image for project: 'Network Observability'
  1. Network Observability
  2. NETOBSERV-926

Connection tracking + Kafka might generate incorrect counters

    • False
    • None
    • False
    • Previously, the when sending to kafka, the kafka message key was ignored causing error with connection tracking. Now the key is used for partitioning so each flow from the same connection are sent to the same processor.
    • NetObserv - Sprint 234, NetObserv - Sprint 235, NetObserv - Sprint 236, NetObserv - Sprint 237, NetObserv - Sprint 238, NetObserv - Sprint 239, NetObserv - Sprint 240, NetObserv - Sprint 241, NetObserv - Sprint 242

      Connection tracking is processed per node (FLP) and maintains counter states. As traffic comes from EBPF agents that are also bound to a node, there is an automatic consistency in the n-tuple ids that are processed.

      When Kafka is used, FLP receives "random" flow data - loosing that n-tuple ids consistency. So the counter that it produces will be flawed, ie. they might not reflect the actual total of bytes/packets received.

            ocazade@redhat.com Olivier Cazade
            jtakvori Joel Takvorian
            Amogh Rameshappa Devapura Amogh Rameshappa Devapura
            0 Vote for this issue
            7 Start watching this issue