Skip to content

Cloud pricing term

What is Cold start?

A cold start is the extra startup work and delay required when a platform must create a fresh runtime instance before serving work.

Also called: startup latency, container startup, function cold start, scale-from-zero delay.

The expensive mistake

Counting only the application's useful execution time. Image pulls, runtime initialization, dependency loading, and connection setup can add latency and, on some platforms, billable duration. Eliminating idle compute can therefore trade money for slower first responses.

Where it appears on your bill

Cold-start latency is not normally an invoice line. Its billable portion is folded into task, instance, CPU, or memory duration.

Worked example

AWS Fargate bills Linux tasks from image download and applies a one-minute minimum. At the captured 1 vCPU and 2 GB rates, a task that performs 10 seconds of useful work but stops within a minute is billed at least 60 x ($0.000011244 + 2 x $0.000001235) = $0.00082284. Repeating 10,000 such starts reaches $8.2284 before transfer.

Example prices checked:

Primary sources