/cloudwatch_exporter

A CloudWatch exporter for Prometheus coded in Go, with multi-region support

Primary LanguageGoApache License 2.0Apache-2.0

CloudWatch Exporter

An AWS CloudWatch exporter for Prometheus coded in Go, with multi-region and dynamic target support.

How to configure

The configuration is in YAML and tries to stay in the same spirit as the official exporter.

tasks:
 - name: billing
   default_region: us-east-1
   metrics:
    - aws_namespace: "AWS/Billing"
      aws_dimensions: [Currency]
      aws_dimensions_select:
        Currency: [USD]
      aws_metric_name: EstimatedCharges
      aws_statistics: [Maximum]
      range_seconds: 86400

 - name: ec2_cloudwatch
   metrics:
    - aws_namespace: "AWS/EC2"
      aws_dimensions: [InstanceId]
      aws_dimensions_select:
        InstanceId: [$_target]
      aws_metric_name: CPUUtilization
      aws_statistics: [Average]

    - aws_namespace: "AWS/EC2"
      aws_dimensions: [InstanceId]
      aws_dimensions_select:
        InstanceId: [$_target]
      aws_metric_name: NetworkOut
      aws_statistics: [Average]

  - name: vpn_mon
    metrics:
     - aws_namespace: "AWS/VPN"
       aws_dimensions: [VpnId]
       aws_dimensions_select:
         VpnId: [$_target]
       aws_metric_name: TunnelState
       aws_statistics: [Average]
       range_seconds: 3600

What are "Tasks" and "$_target"?

Tasks are used to describe a CloudWatch scrape that can be reused on a whole set of instances and even cross-region.

The $_target token in the dimensions select is used to pass a parameter given by Prometheus (for example a __meta tag with service discovery).

For example a scrape URL looks like this:

http://localhost:9042/scrape?task=ec2_cloudwatch&target=i-0123456789&region=eu-west-1

With the example configuration above, this means that the CloudWatch exporter will scrape the CPUUtilization and the NetworkOut metrics when the dimension InstanceId will be equal to i-0123456789 in the eu-west-1 region according to the configuration of the task: ec2_cloudwatch.

Hot reload of the configuration

Let's say you can't afford to kill the process and restart it for any reason and you need to modify the configuration on the fly. It's possible! Just call the /reload endpoint.

How to configure Prometheus

  - job_name: 'aws_billing'
    metrics_path: '/scrape'
    params:
      task: [billing]
    static_configs:
      - targets: ['localhost:9042']

  - job_name: 'ec2_cloudwatch'
    metrics_path: '/scrape'
    ec2_sd_configs:
      - region: eu-west-1
    params:
      region: [eu-west-1]
    relabel_configs:
      - source_labels: [__meta_ec2_tag_role]
        regex: webapp
        action: keep
      - source_labels: [job]
        target_label: __param_task
      - source_labels: [__meta_ec2_instance_id]
        target_label: __param_target
      - target_label: __address__
        replacement: 'localhost:9042'

  - job_name: 'vpn_mon'
    metrics_path: '/scrape'
    params:
      task: [vpn_mon]
    static_configs:
      - targets: ['vpn-aabbccdd']
    relabel_configs:
      - source_labels: [__address__]
        target_label: __param_target
      - target_label: __address__
        replacement: 'localhost:9042'

Thanks to Prometheus relabelling feature, in the second job configuration, we tell it to use the job_name as the task parameter and to use the __meta_ec2_instance_id as the target parameter. The region is specified in the params section.

The Billing example is there to demonstrate the multi-region capability of this exporter, the default_region parameter is specified in the exporter's configuration.

Note: It would also work if no default_region was specified but a params block with the region parameter was set in the Prometheus configuration.

Endpoints

Endpoint Description
/metrics Gathers metrics from the CloudWatch exporter itself such as the total number of requests made to the AWS CloudWatch API.
/scrape Gathers metrics from the CloudWatch API depending on the task and (optionally) the target passed as parameters.
/reload Does a live reload of the configuration without restarting the process

Why this remake?

We felt left out when we wanted to monitor hundreds of machines on AWS thanks to CloudWatch when using the original exporter. We wanted to be able to use the service EC2 discovery functionnality provided by Prometheus to dynamically monitor our fleet.

Regarding our requirements, installing Java runtime was also a bit of an issue, so we decided to make it "compliant" with the rest of the Prometheus project by using Golang.

TODO

This exporter is still in its early stages! It still lacks the dimensions_select_regex parameter and the DynamoDB special use-cases. Any help and/or criticism is welcome!

_dimensions_select_regex has been added here

End Note

This exporter is largely inspired by the official CloudWatch Exporter and we'd like to thank all the contributors who participated to the original project.

This project is licensed under the Apache 2.0 license.