Same thing, the engine in this case is "self-hosted", as in, it runs in a VM hosted on the cluster that it is managing. I am a beginner here, but from my understanding, each node is always checking on the health of the engine VM. If the engine is missing (ie, the host running it has gone down), then another available, healthy host will spawn up the engine and you will regain access.
In my experience this has worked very reliably. I have 2 hosts, both are "able" to run the engine VM. If I take one host down, I am not able to load the engine GUI. But if I wait a few minutes, then I regain access, and see that the engine is now running on the remaining healthy host.