If one of them fails, the client falls back to the other one. That’s the main reason why you can configure two. I think you’re right that some OSes / devices send half the queries to each server (that’s what I’m seeing too) but the fallback logic is still there too.