Check peak API Gateway traffic against the account RPS quota and burst allowance, and see requests in flight and spare headroom.
API Gateway throttles with a token bucket: the steady quota refills the bucket every second and the burst allowance is its depth, so traffic above the quota is only served until the bucket drains. Requests in flight follow Little's law and matter because slow integrations hold connections open, and the hard 29-second integration timeout caps how slow a backend may be. Quotas differ by region and can be raised, so check your Service Quotas console rather than assuming the defaults. The RPS quota is per region and shared by every API in the account, so one noisy API can throttle an unrelated production endpoint — knowing your utilization is what lets you ask for the increase before the incident rather than during it.
AWS API Gateway Throughput
quota utilization = peak req/s ÷ effective limit; requests in flight = peak req/s × average latency in seconds; burst absorbed = burst allowance ÷ (peak − limit).
quota utilization = peak req/s ÷ effective limit; requests in flight = peak req/s × average latency in seconds; burst absorbed = burst allowance ÷ (peak − limit). API Gateway throttles with a token bucket: the steady quota refills the bucket every second and the burst allowance is its depth, so traffic above the quota is only served until the bucket drains. Requests in flight follow Little's law and matter because slow integrations hold connections open, and the hard 29-second integration timeout caps how slow a backend may be. Quotas differ by region and can be raised, so check your Service Quotas console rather than assuming the defaults.
The RPS quota is per region and shared by every API in the account, so one noisy API can throttle an unrelated production endpoint — knowing your utilization is what lets you ask for the increase before the incident rather than during it.
This calculator takes 6 inputs: Peak request rate, Average end-to-end latency, Account steady-state quota, Burst allowance, Stage or method throttle, Integration timeout. The pre-filled defaults are a realistic starting point — replace them with figures from your own environment for a result you can act on.
A 429 Too Many Requests with a Retry-After hint on REST APIs. It is returned by the gateway before your integration runs, so the request never reaches Lambda and never appears in your function's error metrics — you have to watch the gateway's 4XXError and Count metrics to see it.