The message killing connection/stream because serving request timed out and response had been started means that we see timeouts of requests (after 60s) and the response has partly be written. This is not the typical timeout due to slow etcd as then the response is usually not started yet. This looks more like outgoing network issues, i.e. the apiserver cannot send back the data to the client in time. I see something about Istio is known to reconfigure networking quite a bit. The chance is high that the issue is on that side. Note: we are working improving the error output "apiserver panic'd ...." as this is technically a panic, but a gracefully handled one. It is expected in this situation and is not critical for the apiserver, compare https://github.com/kubernetes/kubernetes/pull/95002.