Scaling in Kubernetes possible? loadtesting with multiple kea-dhcp4 pods
I am attempting to load-test a Kea deployment in Kubernetes at 10,000 DORA transactions per second. I am having problems running multiple instances of Kea DHCPv4 2.0.1 in a large Kubernetes cluster.
For simplicity, the setup is:
- perfdhcp for testing
- only DHCPv4, serving on UDP 67
- memfile lease DB
- 2.0.1 with default config
- subnet configured with relay to match IP of client pod(s) running perfdhcp
In all cases, only a SINGLE pod processes DORA requests, regardless of how many pods are created in the replicaset. I have checked LB configuration suits other UDP load-balanced services. If I kill the pod that is actually processing requests, then a new pod picks up the requests, and occasionally I can get multiple pods processing in this way - but it seems very unreliable.
Is there any advice on configuration for configuring 10 - 15 instances of Kea in cluster? Eventually we want to use a persisted DB backend (Mysql/Postgresql) but for now, just memfile performance would be fine for testing purposes.