When ncp is not running and Kubernetes endpoints if modified, the Loadbalancer pool members will not be updated to reflect change in endpoints in NSX-T

Products

• VMware NSX

Issue/Introduction

Symptoms:
Consider kubernetes service 'x' and port 'y'. NCP is currently not running.
Consider the case, pod 'z' which is added as an endpoint for service 'x' and port 'y' when NCP is down. On NCP start, container in pod 'z' will not be added to Loadbalancer pool members.
Consider the case, pod 'z' which is as an endpoint for service 'x' and port 'y', is removed when NCP is down. On NCP start, container in pod 'z' will not be removed to Loadbalancer pool members.

Environment

VMware NSX-T

Resolution

This is a known issue affecting VMware NSX-T 2.1.x.

Currently, there is no resolution.

Workaround:
After NCP is up and running, trigger the kubernetes endpoint modify event. An endpoint modify event can be triggered, by adding/deleting a pod which matches the service 'x' labelselectors or adding/removing a container (having the same containerPort as the targetPort in service 'x') from the pod (which has labels same as the labelselectors in the kubernetes service). Once such an endpoint modify event is triggered, the Loadbalancer pool members will reflect the current state of service endpoints.