Cold Start

What it is

A cold start is the cost of starting an execution environment before it can handle work.

The first request or job pays that startup cost.

The next ones often do not, at least for a while.

When it matters

It matters most when traffic is bursty, infrequent, or latency-sensitive.

If the system keeps instances warm, the effect is smaller.

If it spins up on demand, the delay becomes easier to notice.

Common mistake

The classic mistake is treating cold start as only a serverless problem.

It shows up anywhere startup cost is meaningful:

• workers
• containers
• functions
• just-in-time initialization

Short example

A background job runs once every 20 minutes.

Every run needs to boot the runtime, load dependencies, and connect to services before doing useful work.

That startup time is the cold start.

Why it helps

Understanding cold start helps you separate:

• execution time
• startup time

That changes how you reason about latency, autoscaling, and workload shape.

Cold start is not the work itself. It is the tax you pay before the work can begin.