By default, when using the Round Robin (RR) load balancing algorithm in Avi Load Balancer, traffic distribution is done on a per-core basis within each Service Engine (SE). This can result in multiple cores from a single SE selecting the same backend server, which is expected behavior in the default configuration.
You may observe non-uniform distribution across backend servers when:
A single SE with many cores is handling most of the traffic.
You're expecting per-SE server selection (i.e., each SE should pick a unique server, regardless of how many cores it has
To modify the default behavior and ensure that each SE selects only one server, regardless of the number of cores, use the lb_algo_rr_per_se
knob in the pool configuration.
SSH into the Avi Controller and enter shell:
Enter configuration mode for the target pool:
Enable per-SE RR algorithm: