An API backed by Lambda returns high p95 latency after deployment. Which two telemetry sources are most useful first?