Currently the default requests values for CPU and memory are insufficient for large configurations. The single worker process spawned from the core available to the container based on the values in the resources means that it can easily get blocked and therefore bring down the apicast service. Additionally as only the worker process is killed and not the main process this makes it harder for Ops to detect any problems and the gateway can be done for a long period of time without understanding where the problem is.

See the suggested new values in this comment.

Most likely as seen in our own tests setting the following default values would be recommended:

{"limits": {"cpu": "3", "memory": "512Mi"}, "requests": {"cpu": "2", "memory": "256Mi"}}