we were able to resolve it by using ip for the kube server instead of the hostname. This is caused by load balancers not supporting HTTP/2. Instead of using the https://hostname URL to Kube, use the IP address, and disable SSL verification

kubectl config set-cluster $NAME --user=$USER --server=$KUBE_URL --insecure-skip-tls-verify=true

Ran into this today when trying to use Garden.io for a cluster running in Jelastic.

Found the solution in this Github comment:

First acquire a local binary for Tiller (server-version of Helm), either by compiling or by downloading it from the release page.

Then run:

$ export HELM_HOST=":44134"
$ tiller -listen ${HELM_HOST} -alsologtostderr >/dev/null 2>&1 &

This will run a local version of the Kubernetes Helm Server. Now try your original command again, kubectl, that will delegate to this local Helm instead and manage to connect.

I have seen this before and there were a couple of issues to address:

1. The Nginx that was proxying requests did not contain the following config (required for HTTP2 support):

   proxy_set_header Upgrade 'websocket';
   proxy_set_header Connection 'Upgrade';

2. And we also had to switch our Classic Load Balancer on AWS (ELB) with an Application Load Balancer (ALB).

TBH, the question is a bit vague. Where are you seeing this exactly? It needs more context.