Принцип работы бессерверного выполнения

In serverless, the function is the unit of deployment and scaling. The platform provisions compute only when an event arrives, and scales to zero when idle.

Functions run in response to events — an HTTP request, a queue message, a file upload, a schedule. The event payload lands in your handler.

Invocations are synchronous (caller waits for the result) or asynchronous (event is queued, processed later, and retried on failure).

A handler takes an event and returns a response — the platform manages everything around it. Run this to see it in action.

The first call on fresh capacity hits a cold start: the platform fetches code, boots a runtime, and initializes. Warm calls afterward are much faster.

Tame cold starts by keeping packages small, picking fast-starting runtimes, and using provisioned concurrency on latency-critical paths.

Each concurrent event runs in its own instance. The platform scales out automatically, so 1000 simultaneous requests can mean 1000 parallel executions.

An instance handles one request, then gets reused for the next. Code outside the handler runs once per environment — perfect for setting up shared connections.

Functions are stateless: never count on in-memory data surviving between calls, since instances come and go. Persist state in a database, cache, or object store.

Functions cap out on duration, memory, and payload size. Break long work into steps or another service — and note that more memory also means more CPU.

You pay for invocations × duration × memory. Idle functions cost nothing, which makes serverless economical for spiky or low-volume workloads.

Recap of the execution model: event-driven sync/async calls, cold starts you can mitigate, automatic concurrency, environment reuse, statelessness, and pay-per-use.