Q: What does a third-party request timeout mean and where can it happen? A: To know how to fix third-party request timeout, start by understanding that timeouts can occur at the client, in the network, or on the server. A connect timeout happens before the TCP handshake while a read timeout occurs after connect when waiting for bytes, and HTTP 408/504 indicate client-side or gateway timeouts respectively. Q: How should I triage a stalled API call to a third-party provider? A: Do initial checks: look at the provider status page and test a light health or ping endpoint with curl from your app host and another region. Use curl timers and separate connect vs read (–connect-timeout and –max-time), check DNS and TLS with dig and openssl, and validate auth, rate limits, and IP allowlists as needed. Q: What client timeout and retry settings are recommended to prevent third-party timeouts? A: Set a short connect timeout (2–5s) and a realistic read timeout (10–30s) based on the API’s p95, and match or stay below proxy or platform limits. Retry only idempotent calls using exponential backoff and jitter (for example 0.5s, 1s, 2s, 4s), use idempotency keys for write operations, add a circuit breaker, and keep connections alive with HTTP keep-alive or HTTP/2. Q: How can I reduce request payloads or work to lower the chance of timeouts? A: Request fewer fields using sparse fieldsets, paginate large lists, and apply server-side filters or compression (gzip/Brotli) to reduce payload size and transfer time. Use ETag or If-Modified-Since to skip full payloads, chunk large uploads, and batch small calls when supported to avoid chattiness. Q: Which proxy, load balancer, or platform time limits should I align with? A: Map every hop and align your timeouts with reverse proxy and platform caps such as Nginx proxy_read_timeout, HAProxy timeouts, Envoy idle_timeout, AWS ALB idle timeout (default 60s), AWS API Gateway synchronous limit (~29s), Cloudflare typical 100s, and Heroku’s 30s stall timeout. Your lowest timeout in the chain decides the outcome, so set per-hop budgets accordingly. Q: What observability and testing practices catch third-party timeouts early? A: Add correlation IDs and log timing stages (DNS, connect, TLS, TTFB, total) and track latency percentiles, error rates, and retry counts to spot regressions. Run regional synthetic probes and load tests, use distributed tracing to identify slow spans, and alert on rising p95 latency and timeouts, not just 5xx spikes. Q: What safer fallback patterns should I use when third-party requests time out? A: Serve cached or partial results, use stale-while-revalidate, queue slow work for background processing, and prefer webhooks or polling for long-running jobs so requests return quickly. Also implement bulkheads, rate-limit retries, and a graceful UI that shows spinners with time limits and retry options. Q: What common pitfalls should I avoid when trying to fix third-party request timeouts? A: Avoid setting timeouts too high, creating retry storms, retrying non-idempotent POSTs without idempotency keys, and ignoring platform caps like API Gateway’s ~29s limit. Also avoid not using keep-alive, unbounded GraphQL queries, big uncompressed payloads, and hardcoded DNS that breaks provider failover.