Slide 4 of 12

AWS metrics: metric streams

How it works

AWS Metric Streams architecture: AWS pushes metrics to Grafana Cloud via Amazon Data Firehose

Complexity: Moderate | Infrastructure: Firehose + IAM | Latency: Seconds

You create a CloudWatch Metric Stream that selects which metrics to export. The stream delivers to an Amazon Data Firehose stream, which batches the metrics and forwards them to Grafana Cloud’s ingestion endpoint. You deploy the Metric Stream, the Firehose delivery stream, and the IAM roles in your AWS account, and you can template all of it with Terraform or CloudFormation.

You also create a resource metadata scrape job. It periodically fetches your AWS resource tags and metadata, which Grafana Cloud combines with the streamed metrics to enrich them with context like resource ARNs and tags.

Considerations

  • Near real-time delivery, within seconds
  • Scaling without API rate limits
  • Predictable cost at scale
  • No scrape schedules to manage
  • AWS infrastructure to deploy: the Metric Stream, an Amazon Data Firehose delivery stream, and IAM roles
  • A resource metadata scrape job to enrich metrics with resource ARNs and tags
  • Upfront setup, which you can template with Terraform or CloudFormation

Note

If you automate the setup with Terraform, you also need permission to create an access policy token, which typically requires the Admin role.

When to use

  • Production monitoring where seconds-level alerting latency matters
  • Environments that scale across multiple AWS accounts
  • Setups where you want no scrape schedules to manage

Documentation

View the AWS metric streams documentation.

Script

AWS Metric Streams suit ongoing production monitoring. Instead of Grafana Cloud pulling metrics from CloudWatch, AWS pushes them to you.

Here’s the architecture: you create a CloudWatch Metric Stream that selects which metrics to export. That stream delivers to an Amazon Data Firehose stream, which batches and forwards to Grafana Cloud’s ingestion endpoint.

You also create a resource metadata scrape job. It fetches your AWS resource tags and metadata, so Grafana Cloud can enrich the streamed metrics with context like resource ARNs and tags.

The result? Metrics arrive within seconds, which is critical for alerting. Because AWS pushes the data, metrics scale without API rate limits and costs stay predictable.

Long-term maintenance is easier too. There’re no scheduled scrape jobs to manage.

There’s some upfront setup. You’re deploying infrastructure in your AWS account: the Metric Stream itself, a Firehose delivery stream, IAM roles to authorize everything.

You can template this with Terraform or CloudFormation and deploy consistently across all your accounts. The investment pays off quickly in reliability, cost efficiency, and low alerting latency.

/
Previous Next