In Antidote, Lesson Endpoints are spun up as containers within a Kubernetes cluster. As a result, they all have a single eth0 network interface by default. However, as NRE Labs is designed to teach complex infrastructure topics, often a single network interface is not enough. This document will explain how we use a concept called "Connections" to provide multiple interconnections directly between Lesson endpoints.
Imagine you have a lesson definition with three endpoints: vqfx1, vqfx2, and vqfx3. You can think of these like nodes in a graph topology, as all networks are. So, if Endpoints are nodes, then the edges, or the connections between these nodes, are connections:
connections:
- a: vqfx1
b: vqfx2
- a: vqfx2
b: vqfx3
- a: vqfx3
b: vqfx1
This is a simple list of connections from a to b. The first connection is from vqfx1 to vqfx2 and so on. Antidote will create virtual networks for each Connection, and then attach Endpoints to them.
The way these networks will be made available will depend on how the image is built. If a native container, net1, net2, and so forth. If a VM-in-container, it could be anything. In the future we may go with a solution like Kata containers by default, which could also add a wrinkle.
At the moment, the best place to know which network interfaces will be available to you is to consult the the image metadata files, which includes a list of network interfaces that the image makes available to be used. This is an ordered list, and interfaces are consumed in order of the Connections that reference that endpoint.