In this post, I’ll be discussing the use of monitoring as a tool to optimize the cost and performance of AWS Lambda. I’ve worked on a number of teams, and almost without exception, the need to put monitoring in place has featured prominently in early plans. Tickets are usually created and discussed, and then placed in the backlog, where they seem to enter a cycle of being important—but never quite enough to be placed ahead of the next phase of work.
In reality, especially in a continuous development environment, monitoring should be a top priority, and with the tools available in AWS and organizations like Sumo Logic, setting up basic monitoring shouldn’t take more than a couple of hours, or a day at most.
What exactly is AWS Lambda?
AWS Lambda from Amazon Web Services (AWS) allows an organization to introduce functionality into the AWS ecosystem without the need to provision and maintain servers. A Lambda function can be uploaded and configured, and then can be executed as frequently as needed without further intervention.
Contrary to a typical server environment, you don’t have any control, nor do you require insight into the data elements like CPU usage or available memory, and you don’t have to worry about scaling your functionality to meet increased demand. Once a Lambda has been deployed, the cost is based on the number of requests and a few key elements we’ll discuss shortly.
How to setup AWS Lambda monitoring, and what to track
Before we get into how to set up monitoring and what data elements should be tracked, it is vital that you put an effective monitoring system in place. And with that decision made, AWS helps you right from the start. Monitoring is handled automatically by AWS Lambda, which means less time configuring and more time analyzing results. Logs are automatically sent to Amazon Cloudwatch where a user can view basic metrics, or harness the power of an external reporting system, and gain key insights into how Lambda is performing. The Sumo Logic App for AWS Lambda uses the Lambda logs via CloudWatch and visualizes operational and performance trends about all the Lambda functions in your account, providing insight into executions such as memory and duration usage, broken down by function versions or aliases.
The pricing model for functionality deployed using AWS Lambda is calculated by the number of requests received, and the time and resources needed to process the request. Therefore, the key metrics that need to be considered are:
- Number of requests and associated error counts.
- Resources required for execution.
- Time required for execution or latency.
- Request and error counts in Lambda
The cost per request is the simplest of the three factors. In the typical business environment, the goal is to drive traffic to the business’ offerings; thus, increasing the number of requests is key. Monitoring of these metrics should compare the number of actual visitors with the number of requests being made to the function. Depending on the function and how often a user is expected to interact with it, you can quickly determine what an acceptable ratio might be—1:1 or 1:3. Variances from this should be investigated to determine what the cause might be.
Lambda Resources usage and processing time
When Lambda is first configured, you can specify the expected memory usage of the function. The actual runtime usage may differ, and is reported on a per-request basis. Amazon then factors the cost of the request based on how much memory is used, and for how long. The latency is based on 100ms segments.
If, for example, you find that your function typically completes in a time between 290ms and 310ms, there would be a definite cost savings if the function could be optimized to perform consistently in less than 300ms. These optimizations, however, would need to be analyzed to determine whether they increase resource usage for that same time, and if that increase in performance is worth an increase in cost.
For more information on how Amazon calculates costs relative to resource usage and execution time, you can visit the AWS Lambda pricing page.
AWS Lambda: The big picture
Finally, when implementing and considering metrics, it is important to consider the big picture. One of my very first attempts with a Lambda function yielded exceptional numbers with respect to performance and utilization of resources. The process was blazingly fast, and barely used any resources at all. It wasn’t until I looked at the request and error metrics that I realized that over 90% of my requests were being rejected immediately.
While monitoring can’t make your business decisions for you, having a solid monitoring system in place will give an objective view of how your functions are performing, and data to support the decisions you need to make for your business.
Optimizing AWS Lambda Cost and Performance API Workflows is published by the Sumo Logic DevOps Community. If you’d like to learn more or contribute, visit devops.sumologic.com. Also, be sure to check out Sumo Logic Developers for free tools and code that will enable you to monitor and troubleshoot applications from code to production.
About the Author
Mike Mackrory is a Global citizen who has settled down in the Pacific Northwest – for now. By day he works as a Senior Engineer on a Quality Engineering team and by night he writes, consults on several web based projects and runs a marginally successful eBay sticker business. When he’s not tapping on the keys, he can be found hiking, fishing and exploring both the urban and the rural landscape with his kids. Always happy to help out another developer, he has a definite preference for helping those who bring gifts of gourmet donuts, craft beer and/or Single-malt Scotch.
