Q: What is the difference between client timeouts and server timeouts? A: Client timeouts are limits you set in your code or gateway and cap how long your app will wait, while server timeouts are the provider’s limits on how long they will spend on a request. You control only the client side, so design client timeouts with the provider’s server limits in mind. Q: How should I allocate connect, read, and total time budgets? A: Set all three: keep connect short, give read most of the budget, and set a total deadline that covers DNS, connect, TLS, and reads while avoiding a single infinite read timeout. Use allocations like connect 10–20% of the total, TLS/first byte 10–20%, read 60–80%, and reserve 20–40% of the budget for retries. For example, user-facing calls might use connect 200–500 ms, read 1–2 s, total 2–3 s, while background jobs can use connect ~500 ms, read 5–15 s, total 10–30 s. Q: How can I increase timeout for third-party requests without breaking apps? A: To increase timeout for third-party requests reliably, set clear client, proxy, and job-level deadlines, add retries with jittered backoff and circuit breakers, and cap concurrency and connection pools. Back the change with data from logs and SLAs, provide fallbacks like caching or partial views, and load or chaos test before you ship. Q: When should I retry requests and how many retries are safe? A: Retry only idempotent calls or safe network errors (timeouts, resets) and do not retry on server 4xx responses. Use exponential backoff with jitter (for example, start around 200 ms and double with jitter), cap retries to about 2–3 attempts, and keep per-try timeouts short so you stay within the deadline budget. Q: What metrics and logs should I monitor when tuning timeouts? A: Track latency percentiles (p50, p90, p95, p99) per endpoint, region, and provider alongside timeout rate, retry rate, and final success rate. Also monitor connection errors versus read timeouts, in-flight request count, queue wait time, and thread/CPU usage during load spikes to guide tuning. Q: How do circuit breakers and concurrency limits protect my app when I increase timeout for third-party requests? A: When you increase timeout for third-party requests, circuit breakers stop sending traffic to a failing provider after many recent errors or timeouts and use short half-open windows to probe recovery; deadlines and request budgets enforce that every request finishes within a set time. Limiting concurrency and pool sizes prevents long waits from creating excessive in-flight requests, and backpressure rejects or queues calls with a short queue timeout. Q: What fallback and graceful degradation strategies should I use if a partner is slow? A: Cache recent responses for hot endpoints and serve cached data on timeout to protect user experience. Return partial views (for example, hide slow components like ratings), use bulkheads to isolate a slow partner, and define clear fallbacks for critical UI paths. Q: How should I test and tune timeout changes before rolling them out? A: Replay real traffic with higher latency and inject timeouts or packet loss to confirm retries and breakers act as planned. Load test by doubling or tripling latency, watch error rates and user-times, and document values to review them quarterly or after big traffic shifts.