When workloads are deployed on different nodes which communicates with each other, flows are captured by ebpf-agents running on each node. This is a problem for features like Network Policies Correlation (OVN Observability) where in UI the flows which are captured by one of the agents where the client pod is running and   network policies didn't have impact (because the server workload that had an Ingress policy setup to deny  traffic had its flow captured by agent running on its node) shows "n/a" causing confusion for user. Only the flow captured by agent running on same node as server workload had Network Event populated for the flow.

Attached raw flow json files for both flows captured.

1.  Set up client and server pods on different nodes.
2. Set up Ingress policy to deny all traffic for server pod NS.
3. On Network Traffic page, see flows between client and server pods.

Flows are duplicated across agents when they're across nodes.

Improved user experience to identify network events better regardless of how pods are deployed