Probes and Self-Healing

Part of the Kubernetes course

A process that has crashed is easy to deal with. It is gone, the operating system noticed, and Kubernetes starts a new one. The hard case is the process that is still running and no longer working: a web server whose database connection pool has deadlocked, a worker stuck in an infinite loop, an application that leaked memory until every request times out. From the outside the process looks alive.

Nothing in the operating system can tell you the difference, because "working" is a question only your application can answer. So Kubernetes asks it. A probe is a small, repeated question the cluster puts to your container, and the container's answer decides what happens next.