the three things that come to mind are: 1. we've had reports that grpc-node will occasionally hang/fail in a pretty opaque way. bumping your client version may help as it brings a different grpc-node version with it. 1. if you have an L7 load balancer in between the client and the cluster, such as an ALB, they occasionally cause problems. we use an L4 (? might be L3) load balancer for this reason (a Network Load Balancer in the AWS case) 1. what's the call that you're making? if you're trying to return the world with a lookupresources call it can fail in a way similar to what you're seeing.