Question

After enrolling a service into the service mesh (Envoy sidecar injected), end-to-end p99 rose ~40ms and a small but steady fraction of requests now fail with 503 'upstream connect error' even though the application itself logs success and is healthy. The 503s and the added latency correlate with high request concurrency. App CPU is fine, but the sidecar container is hitting its CPU limit and you see Envoy circuit-breaker / pending-request-overflow stats incrementing. No application or downstream change — only the mesh enrollment. How do you triage and mitigate?