I tried to get a 3 node bare metal cluster running, and i came to the conclusion that its not worth it unless you have over 50+ services running on over 5+ servers and really really need instant automatic failover. Otherwise docker compose is so so so much easier.
I found very poor documentation on how to recover from failure scenarios, i dont like how the control plane has to be separate from the workers, i happen to choose k3s which was abandoned, i never understood why etcd NEEDS more than 2, i hated all the "beta" configuration tags/domains whatever they are called, there was no good way to know when they changed or what they changed to, I thought the whole point of helm was that you could update things easily but its the opposite: it breaks things. I couldn't believe there was no built in support for uninterruptible power supplies? maybe just a k3s thing.
I am sure it works for huge scale entire/multiple data center stuff, but anything in one location i bet is a mistake
I found very poor documentation on how to recover from failure scenarios, i dont like how the control plane has to be separate from the workers, i happen to choose k3s which was abandoned, i never understood why etcd NEEDS more than 2, i hated all the "beta" configuration tags/domains whatever they are called, there was no good way to know when they changed or what they changed to, I thought the whole point of helm was that you could update things easily but its the opposite: it breaks things. I couldn't believe there was no built in support for uninterruptible power supplies? maybe just a k3s thing.
I am sure it works for huge scale entire/multiple data center stuff, but anything in one location i bet is a mistake