Readiness Probes

A new instance of your .NET service takes several seconds to warm its in-memory cache before it can answer correctly.

Azure Load Balancer starts sending it live traffic the instant the process starts running, not once it's actually ready. So some requests during a scale-up hit a half-warmed instance and get back a wrong, or empty, answer.

The instance existing and the instance being ready are two different things, and only one of them was actually being checked. That's exactly what readiness probes fix.

A readiness probe makes the load balancer actually wait: have the instance expose a health check endpoint that only reports healthy once its own startup work has actually finished. Configure Azure Load Balancer, or Azure Application Gateway, to hold traffic from any instance until that endpoint reports healthy, so traffic never arrives before the instance is genuinely ready, no matter how long startup takes.

Use: Microsoft.Extensions.Diagnostics.HealthChecks

Don't guess how long startup takes. Ask the instance directly, and believe its answer.

A store unlocks its doors the moment the staff arrive, before the shelves are stocked or the registers are counted. Customers walk in to a store that isn't actually ready to serve them.

Keeping the doors locked until the staff are genuinely ready avoids that, even if it means opening a few minutes later than the posted time.

A door being unlocked isn't the same as a store being ready. Neither is a process running the same as a service being ready.