Hi everyone,
I’m a university student who recently got interested in Kubernetes, so please bear with me if my questions reflect gaps in my knowledge. I’m still learning.
I know that companies in industry and research institutions doing scientific computing on HPC systems use Kubernetes for container orchestration. It’s also fairly well known that Kubernetes is notorious for its large configuration surface area; the cluster autoscaler alone exposes around 30 parameters, and that’s just one component.
My question is: how do the people who set up and maintain Kubernetes clusters actually arrive at a good configuration?
Is it primarily a manual, iterative process — tweak things progressively, respond to problems as they surface, and stop once things are “good enough”? And if so, does that mean most production clusters are never truly optimized, just satisficed?
In a perfect world, there would be a tool that takes your workload characteristics and system constraints as input and outputs a near-optimal Kubernetes configuration. I’m curious whether something like that exists, whether anyone is actively working on it, and — perhaps most interestingly — whether practitioners would even want it. Is the configuration complexity seen as a problem worth solving systematically, or is the current empirical approach considered acceptable?
Would love to hear from anyone with hands-on experience.
Thank you.