การสร้างระบบไร้เซิร์ฟเวอร์ที่ทนทาน

Welcome! In this lesson, we'll dive into designing highly resilient and fault-tolerant serverless applications. Even though AWS manages much of the infrastructure, your functions still need to handle failures gracefully.

We'll explore key architectural patterns to ensure your applications remain stable and performant, even when things go wrong.

In a serverless world, your functions often interact with many other services: databases, APIs, message queues. These interactions happen over a network, and networks can be unreliable.

• Transient Failures: Brief network glitches or service slowdowns.
• Downstream Service Issues: A service your Lambda calls might be temporarily unavailable.
• Unexpected Data: Malformed input can cause your function to crash.

Designing for these "failures" is crucial for a stable system.

For certain invocation types, AWS Lambda automatically retries your function if it fails. This is a powerful built-in resilience mechanism for asynchronous invocations.

For example, if an SQS queue triggers your Lambda and your function errors, Lambda (or SQS) will retry the invocation a few times. This helps overcome transient issues without any code changes.

However, retries aren't a silver bullet; they can lead to duplicate processing if not handled carefully.

When retries happen, your function might execute the same operation multiple times. This is where idempotency comes in.

An idempotent operation is one that can be applied multiple times without changing the result beyond the initial application.

• Example: Setting a value (x = 5) is idempotent.
• Non-Example: Incrementing a value (x++) is NOT idempotent, as each retry would change the value.

For resilient systems, many operations should strive to be idempotent.

To make your Lambda functions idempotent, you often need to track the state of a request. This typically involves:

1. Generating a unique Idempotency Key for each request (e.g., from request ID, event source ID).
2. Checking if this key has already been processed before performing the core logic.
3. Storing the result or status of the operation associated with the key.

This ensures that even if a function is retried, the core side-effect only occurs once.

The Circuit Breaker pattern is a powerful way to prevent a failing service from causing cascading failures throughout your application.

Imagine a call to an external API that starts failing. Continuously retrying it will just waste resources and slow down your function. A circuit breaker detects this and "opens" the circuit, stopping calls to the failing service temporarily.

This gives the failing service time to recover and prevents your application from getting bogged down.

A circuit breaker typically has three states:

• Closed: Operations proceed as normal. If failures exceed a threshold, it transitions to Open.
• Open: All calls to the protected service immediately fail (or return a fallback). After a timeout, it transitions to Half-Open.
• Half-Open: A limited number of test calls are allowed through. If these succeed, it transitions back to Closed. If they fail, it returns to Open.

This intelligent behavior allows for self-healing.

Implementing a full circuit breaker involves managing state (failures, success counts, last failure time). For serverless, this state might be stored in a shared cache (like ElastiCache) or a database.

While complex to implement from scratch in a simple Lambda, understanding the logic is key. Libraries exist for various languages to help, or you can leverage AWS services like Step Functions to orchestrate retry logic with delays.

Another crucial resilience pattern is using timeouts for external calls. If your Lambda function calls another service (e.g., a database, an HTTP API), that call could hang indefinitely if the service is unresponsive.

Configuring a timeout ensures your function doesn't wait forever, freeing up resources and allowing for retry logic to kick in faster. AWS Lambda itself has a configurable timeout, but you should also set timeouts within your code for specific external requests.

Consider a Lambda function that processes incoming orders. If the function fails after successfully deducting payment but before updating the order status in a database, and then retries, what problem could arise if the payment deduction is NOT idempotent?

We've covered essential patterns for building resilient serverless applications:

• Retries: Lambda's built-in mechanism for transient errors.
• Idempotency: Ensuring operations can be safely retried without unintended side-effects (e.g., duplicate charges).
• Circuit Breakers: Preventing cascading failures by intelligently stopping calls to failing services.
• Timeouts: Protecting against unresponsive external services.

By applying these principles, you can create serverless systems that gracefully handle inevitable failures.