That's your error: you are evaluating k8s on a use-case that is not representative.
Once you scale up the number of nodes, the number of applications, and the years of uptime the complexity grows exponentially.
Now your team needs to be able to debug each component in the OS stack as well as k8s itself.
In the long run you always pay the price for unnecessary complexity - and with interests.
That's your error: you are evaluating k8s on a use-case that is not representative.
Once you scale up the number of nodes, the number of applications, and the years of uptime the complexity grows exponentially.
Now your team needs to be able to debug each component in the OS stack as well as k8s itself.
In the long run you always pay the price for unnecessary complexity - and with interests.