How do I implement exponential backoff for API retries?

Calculate the delay as baseDelay × 2^attempt, capped at a maximum (e.g., 60 seconds). Add random jitter to prevent all clients retrying simultaneously: delay = random(0, min(maxDelay, baseDelay × 2^attempt)). Always check the Retry-After response header first — if present, use it instead of the calculated backoff. Typical values: baseDelay=1s, maxDelay=60s, maxRetries=5.

Why is jitter important in retry logic?

Without jitter, all clients that hit a rate limit at the same time retry at exactly the same intervals, recreating the same traffic spike repeatedly (thundering herd problem). Jitter adds randomness to spread retries across the backoff window. Full jitter (delay = random(0, calculated_delay)) provides the best distribution. AWS research shows full jitter reduces total completion time by up to 40% compared to no jitter.

What is a retry budget and when should I use it?

A retry budget limits the percentage of total requests that can be retries (typically 10%). Without it, if many concurrent requests all start retrying, the retry traffic can exceed the original request volume, overwhelming an already-struggling server. Implement it by tracking the ratio of retries to total requests in a sliding time window and only allowing a retry if the ratio is below the threshold.

What HTTP status codes should trigger automatic retries?

Retry on: 429 (Too Many Requests), 500 (Internal Server Error), 502 (Bad Gateway), 503 (Service Unavailable), and 504 (Gateway Timeout). Do NOT retry on: 400 (Bad Request), 401 (Unauthorized — refresh token first), 403 (Forbidden), 404 (Not Found), or 422 (Validation Error). These client errors won't resolve by retrying the same request. The exception is 401: refresh the token, then retry once.

How does a circuit breaker complement retry logic?

A circuit breaker stops sending requests entirely when failures exceed a threshold (e.g., 5 consecutive failures), preventing your client from hammering a down server. After a cooldown period, it allows one test request through (half-open state). If it succeeds, the circuit closes and normal traffic resumes. If it fails, the circuit stays open. This protects both the server (less load) and the client (faster failure instead of waiting through retries).

Retry Logic for Rate-Limited APIs