The idea that something of the monstrous complexity of k8s is easy is pretty funny to me. I think if you have less than than 2 full time experts on k8s at hand, you're basically nuts if you use it for some non-toy project. In my experience, you can and will experience interesting failure scenarios.

If you don't have state, why not just either use something serverless/fully-managed (beanstalk, lambda, cloudflare workers whatever) if you really need to scale up and down (or have very limited devops/sysadmin capacity) or deploy like 2 or 3 bare metal machines or VMs?

Either sounds like a lot less work to manage and troubleshoot than some freaking k8s cluster.

Bare metal I'd think is the first choice for a large rdbms where you have skilled dedicated personnel that can manage it.

If not rather use a specialist service like RDS for anything with serious uptime/throughput requirements.

k8s doesn't really make sense to me unless it's for spinning up lots of instances, like for test or dev envs or like in the article where they host DBs for people.