Schedule Typha for scaling to well-known nodes


If you are on Kubernetes, use the Kubernetes API datastore, and have a cluster of more than 50 nodes, you have probably deployed Typha. Typha is a fan-out proxy that improves performance in larger clusters. Typha agents must accept connections from other agents on a fixed port.

As part of the Calico bootstrap infrastructure, Typha must be available before pod networking begins and uses host networking instead. It opens a port on the node it is scheduled on. By default, it can get scheduled to any node and opens TCP 5473.

To reduce the number of nodes with the port open to a subset of the total, consider scheduling Typha to well-known nodes.