What’s the Best Way to Monitor AWS Lambda and API Gateway Without Going Crazy?

Honestly, if you're frustrated with Lambda, consider using KEDA on Kubernetes. It can lighten the load and streamline your monitoring process.

I created a Lambda function that activates when CloudWatch alarms go off. It utilizes Contributor Insights for automatic error tracking across Lambda functions. It also collects X-Ray traces to pinpoint exactly where the failures are happening, along with recent error logs and user impact. So instead of just seeing "Lambda errors increased," I get specific insights like which function is failing, the specific issue, and which users it affects.

Honestly, what you're describing is pretty standard for APM tools like Datadog and NewRelic. They already provide auto-discovery of APIs, uptime tracking, latency monitoring, and error reporting without much customization. If you're on a budget, consider designing a Lambda to list your API Gateways and build your own alerting mechanism. For tracking errors, CloudWatch will still need to play a part, but you could automate alerts with another Lambda.

CloudWatch provides reliable metrics right out of the box, like error rates and latencies. You can easily set up alarms and dashboards. If you think it’s complex, wrapping it in a CDK construct or Terraform module could standardize everything for new Lambdas and APIs. Also, for AI summaries, you could link a Lambda to alarms to analyze logs and generate summaries.

You can easily set up Canaries along with a dashboard and alerts. That helps keep an eye on your Lambda functions effectively.

CloudWatch Logs Insights can be super helpful here. Just spend a few minutes writing some queries and save them. I have a simple dashboard with my top queries pinned, like checking latency percentiles and error rates. For alerts, I use SNS topics that trigger a Lambda to format the messages before they go to Slack. Once your queries are set up, you'll barely have to mess with them again as they take care of much of the monitoring for you.

You can configure your Lambda to only log errors, which cuts down your storage costs and makes it easier to identify issues. While this method isn't perfect for every application, it reduces noise significantly.