Liveness Checks Made Easy
When working in a Kubernetes environment you should be defining meaningful
readinessProbes and livenessProbes.
However, most teams implement a simple /healthz endpoint which only confirms
that the server is ready to serve requests. This is not a great indicator of
"liveness" as there are often dependencies which need to be alive and reachable
as well such as a database. If you already include a DB connection attempt and
cache ping in your /healthz endpoint then this tool isn't for you. But if
you're looking for a quick way to add meaningful liveness to your kubernetes
pods then read on.
liveness is a simple process which you can run either inside your Docker
container as a background process or as a sidecar container. It takes a list of
resources via command line flags which it will monitor for liveness. It runs an
HTTP server that whenever it receives a request to any URL it will check all the
provided resources to ensure they are still accepting connections. If any
resource has become unavailable it will respond with a 503 Server Unavailable
response otherwise, it will send a 200.
In this way you don't need to reinvent the wheel if you're just looking for simple liveness. It also is very useful for long running services which aren't HTTP server likes background workers (such as those provided by Celery) since it can add that functionality to check all the relevant dependencies are still alive.