AWS Lambda Under Load: Where Events Wait, Why Requests Throttle, and What You Pay
Your Lambda handler takes 200 ms. So why are requests throttling, and why is billed duration higher?
The answer starts before your code runs.
• S3 events enter Lambda’s asynchronous queue. SQS and Kinesis are polled and processed in batches.
• Reserved concurrency protects capacity; provisioned concurrency prepares environments.
• Throttling can happen because of concurrency, request-rate or scaling limits.
• Compute usage is configured memory × billed duration. INIT time matters too.
2 GB × 350 ms = 0.70 GB-seconds.
For architects, the key question is: when capacity runs out, where does the event wait, who retries it, and how do we prevent duplicate business actions?
Save this flow for your next serverless design review.
Top comments (0)