A Service has healthy pods and endpoints but requests still fail

A new node group joins an EKS cluster and only pods placed there start failing for Service traffic despite being fully Ready.