An orchestrator that runs containers across a cluster — placing them, restarting what dies, scaling with load, and routing traffic to healthy instances.
It is the standard for products that must scale or survive node failure. It is also a system with real operational cost, and plenty of products are better off without it.
We recommend it when the load, the team, or the compliance requirement justifies it, and a managed platform otherwise. Running a cluster you do not need is a tax paid monthly, in attention.