The hard part is waking up again!

When working with Kubernetes, it is trivial to find out when your workload is not getting any HTTP requests and is therefore redundant. It is easy to scale it down. Putting something in place that can recognize that an application is needed again, that's where the fun starts! When we were faced with this issue we didn't want to have proxies or service meshes to do that. Rather we dug deep how Kubernetes routing works and planted a temporary handler, which upon reception of request would wake up the application.

In this talk I'll demonstrate how all that maze of disconnected but cooperating controllers works, how to use the Endpoint API and how to achieve scaling to zero and back up with no networking overhead and just plain Java. And also how to test all of that without breaking anybody's cluster.