What is gRPC monitoring?
gRPC services don't serve web pages, so a website check can't see them, and a TCP check only proves the port is open. A gRPC monitor calls the standard health service, grpc.health.v1.Health/Check, on a schedule. If the server answers SERVING within 10 seconds, it's up. Anything else, from NOT_SERVING to a connection that never opens, is down and alerts your team.
It's the same health service that Kubernetes probes, Envoy and cloud load balancers can use, so if your servers already register it there's nothing to add.
Checking one service or the whole server
Leave the service name empty to ask about the server as a whole. If one server hosts several services and reports each on its own, enter the full name, like shop.v1.Cart, and the monitor only goes down when that service does. Names are case-sensitive and include the package.
For servers behind authentication, add metadata such as authorization: Bearer … and it's sent with every call. Public endpoints use TLS; tick plaintext for a server that speaks HTTP/2 without it.
Errors that point at the fix
When a check fails, the alert names the reason in plain words: the server doesn't have the health service (UNIMPLEMENTED), it doesn't know the service you named (NOT_FOUND), it says NOT_SERVING, it wants credentials, or a proxy in front answered with HTTP 502 or a web page instead of passing gRPC through. When the TLS setting doesn't match the server, the alert says which way to change it.
A failed check is retried and confirmed from other regions before it counts, so a dropped connection doesn't page anyone. The certificate's expiry is read from each TLS check, with warnings before it runs out.