Your service calls a third party API that occasionally becomes slow. How do you configure the client so that does not take you down?

Set explicit connect and read timeouts on the client, because several Java HTTP clients default to waiting indefinitely, and add an overall request timeout. Bound the connection pool for that host, retry only idempotent calls with backoff and jitter, and add a circuit breaker so sustained failures fail fast. Decide the fallback ahead of time: a cached value, a degraded response, or a clear error rather than a hanging request.

I start from the assumption that the defaults are wrong, because several clients will happily wait forever on a read, and that is exactly how a slow dependency turns into every thread in my service being parked. So connect timeout short, read timeout based on their actual latency distribution rather than a round number, and an overall request timeout on top since retries and redirects can stack. Then isolation: that dependency gets its own connection pool with a maximum, so when it degrades it cannot consume all the capacity that other calls need. Retries only on idempotent operations, capped at two attempts with exponential backoff and jitter, and a circuit breaker in front so that once the error rate crosses a threshold I fail immediately for a cooldown rather than piling requests onto something already struggling. The last piece is what the user gets, decided in advance. On a shipping rates integration we served cached rates with a staleness marker rather than failing checkout, and that turned a vendor outage into a minor accuracy problem instead of lost orders. I test it with a proxy that injects delays.