Question

The auth service has a brief 4-second hiccup at 09:30 (a leader election in its datastore). Instead of recovering, it gets dramatically worse: request volume to it suddenly triples and stays there, latency stays high, and it keeps falling over every time it tries to come back. Dashboards: inbound RPS to auth is 3x baseline though product traffic is flat; the gateway shows a surge of retried requests; each auth recovery attempt is immediately re-buried. The clients (mobile + 6 internal services) all retry failed auth calls 3x with a fixed 200ms delay. Explain the dynamics and how you break the cycle.