How does AWS Lambda's scaling work?

Example Answer

Lambda adds execution environments as demand rises, within account, function, request-rate, and scaling limits. I would estimate concurrency from arrival rate times average duration, then leave headroom and check the actual quotas. For 200 requests per second taking 0.25 seconds each, average concurrency is about 50; bursts can need more. Reserved concurrency can limit pressure on a database, but it does not warm environments. When capacity is exhausted, a synchronous caller can receive throttling, while asynchronous invocation and event sources have their own buffering and retry behavior. I would design backpressure and retries around the actual trigger.

Capacity check

Track concurrent executions, throttles, duration, downstream connections, and queue age where applicable. A function can saturate a dependency before it reaches its Lambda quota. Low average duration does not remove separate requests-per-second limits.

Load-test a sudden burst as well as steady traffic. Use bounded retries with jitter and idempotent operations; unbounded client retries can amplify an overload. Check the current Region/account quotas instead of assuming a default limit applies to every account.

Practice follow-up

Would you increase concurrency if database connections are already saturated?

Sources

Lambda concurrency and scaling; Lambda asynchronous retry behavior.