The AKS control plane is provided cost free. And you can run stateless applications in k8s for years without them needing any maintenance.
I agree that almost nothing in the article is realistic, even if I have reached a similar conclusion in different ways.
- Kubernetes is easy if you run a big number of stateless applications.There are easier ways to do that in the cloud thought. I like the pattern of running the database in whatever way the DBA is used to (probably in a VM in which they ssh/RDP into).
- Kubernetes is great if you deploy your product for clients in different clouds. You tell them we need a k8s cluster of API version XYZ and their staff provision one for you. That way you can support Azure, GCP, Oracle and AWS without having to learn many of their APIs. That abstraction is leaky, hopefully your clients can setup the ingress controllers and storage correctly.
- If you are running your app for your own org in the cloud and you don't intend to move, you probably don't need the extra layers of complexity. Running managed k8s is a cloud API over a cloud API.
- If you don't run anything stateful in kubernetes its much easier - there is plenty of SaaS you can use, for example: RDS for DB, Datadog for monitoring, email providers, managed redis. Use kubernetes only for code you've written and understand it's behaviour. Don't be tempted to many things with helm, just the bare essentials.
- Kubernetes is great if you are a really large organization, because if you standardise on it you standardise infrastructurele, and you can move engineers between projects. And can have an ops silo again! /s
- If you are a big org with a DevOps team, you can create a custom Paas over kubernetes. I.e. when a dev runs make deploy staging1 that actually wraps kubectl and magically builds,tests and deploys the current code.