multicluster-runtime provider that watches kubeconfig secrets in a namespace and dynamically engages controllers 🚀 .

AI creates PR (on my machine). I review. Can I approve? What does the Github/lab netiquette say?.

After a little operational hickup where the app got deleted and GitOps eventually recreated everything, OH: "I only wish that the db would have been also deleted to test backups".

This costed me a day and a half:. When watching via Golang types in controller-runtime, but use unstructured when reading from the cache, you are in trouble. Of course, 2 informers are run and they are not consistent aka events arrive and you don't see the changes in the cache.

Linter.

In programming, I like to call that "working blindly" until that moment when you suddenly realize that you connected A to B. After so many years, I am still amazed about these moments when everything comes together. And how I did not see it coming in this full focus mind set.

Try to build a bridge from A to B over a deep valley. No car will be able to pass before a certain point. Not even a bicycle or human can. The measure of progress is not the thing that can pass. The measure is not continuous. There is a point when everything comes together.

Project managers hate them. They don't fit into the mantra of routine work where you want to produce value continuously like a machine that keeps outputting goods. These things are different.

There are tasks that are not iterative. Tasks that take days or weeks or more of super focused time until you see success. I call them Heavy Liftings. Making them iterative with sensible intermediate steps makes them infeasible or way slower.

The real planning tool is gdocs. Jira is an asynchronous replica. Mostly out of date. Never consistent.

Jira must be disrupted. Please.

PRDs are conversation starters not specification documents. Amen 🙏 😍.

And I didn‘t mean it should go away. It‘s the best for a collaborative culture. Neither without opinions nor without ability to listen and revise a team can really work together.

Strong opinions, loosely held. Still a thing.

„directly share“ in @LinkedIn irritates me every single time. It‘s a dishonest term. It really means „shared with network“ and NOT directly with you. Pure clickbait.

Talking about critical path. MVP is the spiritual successor in a less plannable environment. Visibility of necessary focus is what they have in common.

RT @fabiankramm: @the_sttts Yes exactly, we have a runtime that starts the actual pods inside another container (the vNode) that uses linux….

RT @fabiankramm: @the_sttts @the_sttts so the idea of vNode is not to run a separate kubelet (although you could), instead it wraps and sta….

Where does it differ from previous kube-in-docker approaches in the last 10 years, most prominently implemented in kind today.

@fabiankramm please correct where I am wrong. If I was building that, not sure I would run a full kubelet, or in general kubelet expecting full node ownership would be my first concern. Am curious which route you took.