การจัดการข้อผิดพลาดและการลองใหม่

In serverless applications, things can go wrong. Your function might fail, a dependency might be unavailable, or an event might be malformed.

Robust error handling is crucial for building resilient systems that can recover gracefully and notify you when issues occur. This lesson explores how AWS Lambda helps you achieve this.

When working with Lambda, it's helpful to understand different types of errors:

• Invocation Errors: Problems occurring before your code runs, like incorrect IAM permissions for Lambda to read from an event source.
• Function Errors: Exceptions thrown by your function code, timeouts, or out-of-memory errors. These are the errors you primarily handle within your code.
• Service Errors: Rare issues with the underlying AWS Lambda service itself.

How Lambda handles errors depends on the invocation type:

• Synchronous: (e.g., via API Gateway, ALB) Lambda returns the error directly to the caller. The caller is responsible for retries.
• Asynchronous: (e.g., via S3, SNS, SQS) Lambda places the event in an internal queue and automatically retries the function if it fails. You don't get immediate feedback.

We'll focus mostly on asynchronous error handling and retries in this lesson.

For asynchronous invocations, Lambda automatically retries your function if it fails due to an unhandled exception or times out.

• By default, Lambda retries twice (total of 3 attempts).
• These retries are automatic and happen with a delay, often with exponential backoff.
• This built-in mechanism helps ensure temporary issues don't lead to lost events.

You can customize the asynchronous invocation settings for your Lambda function:

• Maximum retry attempts: Set this from 0 to 2. Setting it to 0 means no retries.
• Maximum event age: Define how long Lambda should keep an event in its queue for retries, from 60 seconds to 6 hours.

These settings give you control over how long and how often Lambda attempts to process a failed event.

Even with retries, some events might still fail consistently (e.g., due to malformed data). These are often called 'poison pill' messages.

A Dead Letter Queue (DLQ) is a designated destination (an SQS queue or SNS topic) where Lambda sends events that have exhausted all retry attempts.

DLQs are vital for debugging, preventing data loss, and analyzing why certain events couldn't be processed.

To use a DLQ:

1. Create an Amazon SQS queue or SNS topic.
2. Configure your Lambda function's asynchronous invocation settings to point to this SQS queue or SNS topic as its DLQ.
3. Ensure your Lambda function's execution role has permissions to send messages to the chosen DLQ (e.g., sqs:SendMessage or sns:Publish).

This ensures that no event is truly 'lost' even after multiple failures.

While Lambda handles retries, you should still implement error handling within your function code using try-catch blocks. This allows you to:

• Gracefully manage expected errors.
• Log detailed context for debugging.
• Perform cleanup or partial rollbacks before re-throwing an exception to trigger Lambda's retry mechanism.

Try running this simple Java example:

Sometimes, Lambda can't even start your function. These are invocation errors.

• Examples: Incorrect IAM permissions for Lambda to access an S3 bucket trigger, a misconfigured VPC, or exceeding service quotas before execution.
• Detection: These errors often appear in CloudWatch Logs for your function or as `Invocation errors` metrics in the Lambda console. They are not typically handled by your function's try-catch blocks.

Let's test your understanding of error handling and retries in AWS Lambda.

You've learned how to make your serverless applications more robust:

• Differentiated between various types of Lambda errors.
• Understood how synchronous vs. asynchronous invocations handle errors differently.
• Explored Lambda's automatic retry mechanism for asynchronous calls.
• Discovered the importance of Dead Letter Queues (DLQs) for failed events.
• Saw how to implement basic error handling within your code using try-catch.

By applying these techniques, you can build more resilient and reliable serverless systems.