[bionsystem]

I give a counter example to a general statement, in logic, it is enough to debunk the original statement. My point is, SOME things wouldn't run predictably inside the cluster (in that case without pinning to nodes, which the article says isn't necessary, and which in general defeats the purpose), so you'd need to run some things outside of it.

As per standardized knowledge, I can't see how somebody even proficient with kube, could jump into any app and troubleshoot bad behavior. Apps each have their quirks and subtleties, specific components that behave a certain way. The layers still exists, the kube cluster itself (which again has many component options at every layer of the stack ; hard to know them all), and the app (which will require at least some specialty knowledge).

If it's just about pushing helm charts we wouldn't need SRE anymore, just a CI.

[bakies]

The same way you're describing sysadmins knowing how to look at a linux system and understand what's going on. I feel exactly I can jump into any app and troubleshoot on k8s. And I draw on a lot on top of my previous sysadmin exp.

[bionsystem]

So you agree that it is not kube specific, then ?

[bakies]

The debugging? Not really. It comes from sysadmining. But k8s is a standardized way to deploy software amongst a fleet of machines. Which makes it easy to debug complex systems because I know how to access it all.