Can you confirm that it's impossible to avoid downtime, even with HA, because of forced updates?

Surely that's what HA is? no downtime as you update each node one at a time?

If it's impossible then it's a dealbreaker.