When a lot of updates on the same services are received, the memory of the speaker pods grow and slowly comes down only after the service stops updating.

4.10 (all the frr based versions)

Always.

Create a bgp - enabled service, and continuosly delete it's ip via a script like:

while :do 
kubectl patch svc nginx --subresource=status -p '{"status":{"loadBalancer":{"ingress":[]}}}' 
sleep 0.5 
done

This will cause the controller to restore the ip, and another update to be processed by the speaker

The memory of the speakers grow

The memory of the speaker stay stable

It's a corner case related to environmental issues (something that keeps changing the service)