Question

A company runs a fleet of custom UDP-based real-time telemetry servers behind a Network Load Balancer (NLB) in a single VPC. Operations reports that unhealthy backend instances continue to receive UDP traffic, causing packet loss for clients. The team confirms the target group protocol is set to UDP and instances are registered correctly. What is the MOST likely cause and appropriate fix?

Accepted Answer

UDP target groups cannot use UDP for health checks, so a separate TCP or HTTP health check port must be configured on the targets to detect and drain unhealthy instances.

Answer

NLB does not support UDP; migrate the listener and target group to an Application Load Balancer with a UDP listener.

Answer

Cross-zone load balancing must be disabled so the NLB only routes to healthy targets in the same Availability Zone.

Answer

Connection draining (deregistration delay) must be set to 0 seconds so unhealthy targets are removed immediately.