Load balancing
Health check
A regular probe a load balancer sends each backend, to stop using those that fail.
A good check tests what the application needs, not just that the server answers: a server can run and still be
broken. A few failures take a backend out, a few successes put it back. That is failover: checks every second
with fall 2 notice a crashed server within 2 to 3 seconds, redispatch retries the requests caught in between, and
rise 2 lets it rejoin once it answers again. A wrong check path turns away servers that work.