Deadline not propagated upstream

A user request flows through gateway to service-A to service-B to service-C (all gRPC). At 21:00, service-C develops a latency tail (p99.9 ~4s). Soon the whole chain is melting down: the gateway times out, but downstream you observe service-B and service-C still busily processing requests whose callers have already given up. Dashboards: huge 'context deadline exceeded' counts; service-B/C CPU is high doing 'wasted' work; goroutine/thread counts climbing across the chain. Recent context: timeouts are configured per-hop as fixed values (each hop has its own 5s timeout) with no deadline propagation. How do you triage and mitigate?