Ask HN: How do you design an efficient transacitonal/durable task system?

As someone who wants to design a cloud resource scheduling system similar to EC2, I always wonder how they works in depth. I know I'm going to need a task queue, but which? Kafka? NATS (JetStream)? RabbitMQ? Or if they really just design your own persistent durable task queue with some black magic which is not open sourced.

I still don't know to this day. Plus I have a focus on running that as some kind of Kubernetes operator as well.