AWS is a vast and sprawling set of services. It can be hard to find the hidden gems like this one so I wanted to point this one out.

Structured metrics are very helpful to monitoring the health and function of an software system.

• Do you want to know how long a particular transaction typically takes?
• How fast your database queries are?
• How long external APIs take to respond?
• Fire an alert when a particular function on the site happens too many times? Or too few times?

...plus a million other things specific to whatever system you're working on.

There are a lot of great tools for doing this and one that you might not know about is AWS CloudWatch Metric Filters. If you're already on AWS then you should consider these because it requires only that your application logs to CloudWatch.

If you're on ECS then the awslogs log driver for Docker gets you that nearly for free. By "free" I mean that your application itself can have zero dependencies on AWS services and not require any AWS credentials or libraries to start pumping out metrics that you can visualize, alert on and record over time.

The AWS docs themselves offer the canonical reference for configuring these so I won't go into detail here.

However, the gist is that for a log filter you define the following properties

• A filter pattern for extracting a discrete metric value out of a log entry
• A metric name to store the value in
• An optional dimension for sub-classifying the value
• And finally a log group to extract the metric values from

After that you just run the application and as the logs roll in the metric values get pumped out. Then you can define alarms for alerting on them, graph them, define autoscaling rules from them and more.

To conclude - AWS is big and hairy. While there are benefits to staying platform agnostic, some AWS services don't require much or any coupling of your application code to take advantage of. Cloudwatch Metrics is one of those services and you can get a lot of value out of it with not much effort.