Every request to /work keeps a CPU core busy for 20 ms
(/work?ms=50 for more, at most 200) and answers with the name of the pod that served it.
A HorizontalPodAutoscaler watches the CPU usage: from about one request per second it adds pods,
up to four, and removes them again a minute after the load stops.
Generate load with load-test.sh in the repository folder stefan-autoscale-demo.