AWS Lambda Concurrency and Throttling

• Lambda Concurrency Basics:
  • Lambda functions can scale automatically and rapidly.
  • Low-scale invocations result in fewer concurrent executions, while high-scale events can lead to up to 1000 concurrent Lambda executions.
• Reserved Concurrency:
  • You can limit the number of concurrent executions by setting a reserved concurrency at the function level.
  • Example: Setting a reserved concurrency of 50 means that the Lambda function can only have up to 50 concurrent executions.
  • Invocations exceeding this limit will be throttled.
• Throttling Behavior:
  • Synchronous invocations: Throttling results in a 429 error.
  • Asynchronous invocations: Automatically retried and may go to a Dead Letter Queue (DLQ) if retries fail.
• Increasing Concurrency Limits:
  • To have more than 1000 concurrent executions, you must open a support ticket with AWS to request higher limits.
• Concurrency Limit Implications:
  • The concurrency limit applies to all functions in your AWS account.
  • If one function exhausts the concurrency limit, other functions may be throttled.
  • Example Scenario: A promotion causes one Lambda function to use all available concurrency, leading to throttling of other functions.
• Asynchronous Invocations and Throttling:
  • If Lambda reaches its concurrency limit, additional asynchronous requests are throttled.
  • Throttled events are retried for up to six hours, with retry intervals increasing exponentially from one second to a maximum of five minutes.

Cold Starts and Provisioned Concurrency

• Cold Starts:
  • Occurs when a new Lambda instance is created, and code initialization takes place.
  • Can result in higher latency for the first request, potentially impacting user experience.
• Provisioned Concurrency:
  • Allocates concurrency in advance to avoid cold starts.
  • Ensures lower latency for all invocations.
  • Can be managed with Application Auto Scaling for scheduling or maintaining target utilization.
• Improvements in VPC:
  • AWS has made improvements to reduce cold starts within a VPC, as detailed in a blog post from October/November 2019.

Additional Resources

• Diagrams and further explanations are available in the provided links for reserved concurrency and provisioned concurrency.
• It is recommended to review these diagrams in your own time for a better understanding.

Hands-On Practice

• The next step in the training involves practical exercises to see how concurrency works in AWS Lambda.