When a service goes down

Security is having a bad afternoon. Switch its service off and ask both sides again. Federation needs every owner to answer right now; the projection answers from its last copy.

Switch services off (only for you):
Ready

GraphQL Federation

The router still answers, but only with what it could fetch. Missing parts come back null, with an error naming the subgraph.

Run it.

Answer
Errors

    

Event-driven projection

The query service never calls HR, Employment or Security, so their outage does not reach it. The cost: the answer is only as fresh as the last event.

Run it.

Answer
Console

    
Things to try: In the Apollo version this lesson meant kubectl scale deployment security-subgraph --replicas=0. Here a switch on your request tells that subgraph to fail, for you only.