We have a deployment of a simple Node.js application with replica=2. The first pod comes up and is perfectly running, the second pod is in state CrashLoopBackOff, because its readiness probe failed:
Readiness probe failed: Get http://100.107.65.32:8000/: dial tcp 100.107.65.32:8000: getsockopt: connection refused
Unfortunately, the logs are empty, no matter whether you specify --previous or not.
It can't be a problem with the container or the underlying image, since pod 1 works without any hassle. I have found https://github.com/kubernetes/kubernetes/issues/62594, but that is open and the solution proposed there is rather a workaround than a fix, especially since it does not explain anything on why this happens.
Any ideas of how to proceed with this?
The problem could be solved
Actually, the pod was just takings its time to start, and the probes started too early, when the pod was not yet ready. Increasing the initial probe delay helped.