A cold start is the delay before a serverless function handles a request after scaling down to zero. The cloud must find hardware, start the runtime environment, and load the function code; this boot-up can take a second or two. Scaling to zero saves money when idle, but the next request pays the startup penalty. section 9.4.4
To reduce the delay, keep the function’s deployment package lightweight and minimize external dependencies. That helps the environment start faster, but serverless still brings constraints such as execution time limits and stateless code. section 9.5.2 section 9.12
When a pause is acceptable
A cold start may frustrate someone waiting on an API response, but a one-second delay can go unnoticed while a function processes a queue message, sends email, or generates a nightly report. section 9.4.4






