콜드 스타트 및 프로비저닝된 동시성

When you invoke an AWS Lambda function for the first time, or after a period of inactivity, you might notice a slight delay. This delay is known as a cold start.

During a cold start, AWS needs to prepare the execution environment for your function before your code can run. It's like starting a computer from scratch.

Lambda functions are designed to be stateless and ephemeral. To save resources, AWS 'unloads' execution environments when they are not actively processing requests.

When a cold start occurs, the Lambda service performs several steps:

• Downloads your code package.
• Starts the runtime (e.g., Python, Node.js, Java).
• Initializes your function's dependencies and any global code outside the main handler.

These steps contribute to the initial latency.

Cold starts can significantly impact the user experience, especially for latency-sensitive applications like:

• API backends: Users might experience slower response times.
• Interactive web services: Initial page loads or actions could feel sluggish.
• Real-time data processing: Delays in processing can cascade.

For infrequent background tasks, cold starts might be less noticeable, but for interactive services, they are a critical concern.

Here's a basic Python Lambda function. While this code runs quickly, it's the type of function that experiences cold starts.

The 'cold start' overhead happens before the lambda_handler itself runs, as AWS prepares the environment.

The duration of a cold start can vary based on several factors:

• Runtime: Languages like Java and .NET often have longer cold starts due to larger runtimes and JVM/CLR startup times, compared to Node.js or Python.
• Memory: Functions allocated more memory generally have faster CPU performance and can initialize quicker.
• Package Size: Larger deployment packages take longer for AWS to download and extract.
• VPC Configuration: Functions configured to run within a Virtual Private Cloud (VPC) might incur additional latency for network interface initialization.

To address the latency introduced by cold starts, AWS offers Provisioned Concurrency (PC). This feature keeps a specified number of execution environments for your Lambda function pre-initialized and ready to respond instantly.

Think of it like having a car engine already warmed up and running, rather than starting it from cold.

When you enable Provisioned Concurrency for a Lambda function, AWS actively maintains the requested number of execution environments in an initialized state. These environments are kept 'warm' indefinitely.

When an invocation arrives for a function with PC enabled:

• It's routed directly to one of these pre-initialized environments.
• The cold start phase is completely bypassed.
• Your function code executes immediately with minimal latency.

This ensures consistent, low-latency performance.

You can configure Provisioned Concurrency for a specific version or alias of your Lambda function.

This can be done through:

• The AWS Management Console (Lambda service settings).
• The AWS CLI (Command Line Interface).
• Infrastructure as Code (IaC) tools like AWS Serverless Application Model (SAM) or the Serverless Framework.

You simply specify the number of concurrent instances you want to provision.

Provisioned Concurrency is a powerful tool for optimizing latency, but it's important to understand its implications:

• Cost: Unlike standard Lambda where you only pay for execution time, you pay for Provisioned Concurrency even when your function is idle. This cost is for keeping the environments warm.
• Best Use Cases: It's ideal for critical, user-facing applications requiring consistent low latency, such as interactive APIs or chatbots.
• When Not to Use: For infrequent, non-latency-sensitive background tasks, the extra cost of PC might not be justified.

Test your understanding of cold starts and Provisioned Concurrency.

In this lesson, we explored the concept of cold starts in AWS Lambda – the initial delay when an execution environment needs to be prepared. We learned how factors like runtime, memory, and package size can influence their duration and impact user experience.

To combat cold starts, we introduced Provisioned Concurrency (PC), a powerful feature that keeps a specified number of function instances warm and ready, ensuring consistent, low-latency performance for critical applications. Remember to consider the cost implications when deciding to use PC.