Prometheus Amazon ECS discovery

Prometheus has native Amazon EC2 discovery capabilities, but it does not have the capacity to discover ECS instances that can be scraped by Prometheus. This program is a Prometheus File Service Discovery (file_sd_config) integration that bridges said gap.

Help

Run prometheus-ecs-discovery --help to get information.

The command line parameters that can be used are:

• -config.cluster (string): the name of a cluster to scrape (defaults to scraping all clusters)
• -config.scrape-interval (duration): interval at which to scrape the AWS API for ECS service discovery information (default 1m0s)
• -config.scrape-times (int): how many times to scrape before exiting (0 = infinite)
• -config.write-to (string): path of file to write ECS service discovery information to (default "ecs_file_sd.yml")
• -config.role-arn (string): ARN of the role to assume when scraping the AWS API (optional)
• -config.server-name-label (string): Docker label to define the server name (default "PROMETHEUS_EXPORTER_SERVER_NAME")
• -config.job-name-label (string): Docker label to define the job name (default "PROMETHEUS_EXPORTER_JOB_NAME")
• -config.path-label (string): Docker label to define the scrape path of the application (default "PROMETHEUS_EXPORTER_PATH")
• -config.filter-label (string): docker label (and optional value) to filter on "NAME_OF_LABEL[=VALUE]".
• -config.port-label (string): Docker label to define the scrape port of the application (if missing an application won't be scraped) (default "PROMETHEUS_EXPORTER_PORT")

Usage

First, build this program using the usual go get mechanism.

Then, run it as follows:

• Ensure the program can write to a directory readable by your Prometheus master instance(s).
• Export the usual AWS_REGION, AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY into the environment of the program, making sure that the keys have access to the EC2 / ECS APIs (IAM policies should include ECS:ListClusters, ECS:ListTasks, ECS:DescribeTask, EC2:DescribeInstances, ECS:DescribeContainerInstances, ECS:DescribeTasks, ECS:DescribeTaskDefinition, ECS:DescribeClusters). If the program needs to assume a different role to obtain access, this role's ARN may be passed in via the --config.role-arn option. This option also allows for cross-account access, depending on which account the role is defined in.
• Start the program, using the command line option -config.write-to to point the program to the specific folder that your Prometheus master can read from.
• Add a file_sd_config to your Prometheus master:

scrape_configs:
- job_name: ecs
  file_sd_configs:
    - files:
      - /path/to/ecs_file_sd.yml
      refresh_interval: 10m
  # Drop unwanted labels using the labeldrop action
  metric_relabel_configs:
    - regex: task_arn
      action: labeldrop

To scrape the containers add following docker labels to them:

• PROMETHEUS_EXPORTER_PORT specify the container port where prometheus scrapes (mandatory)
• PROMETHEUS_EXPORTER_SERVER_NAME specify the hostname here, per default ip is used (optional)
• PROMETHEUS_EXPORTER_JOB_NAME specify job name here (optional)
• PROMETHEUS_EXPORTER_PATH specify alternative scrape path here (optional)
• PROMETHEUS_EXPORTER_SCHEME specify an alternative scheme here, default is http (optional)

By docker labels one means dockerLabels map in ECS task definition JSONs like that:

{
  ...
  "containerDefinitions": [
    {
      ...
      "dockerLabels": {
        "PROMETHEUS_EXPORTER_PORT": "5000"
      }
    }
  ]
  ...
}

That's it. You should begin seeing the program scraping the AWS APIs and writing the discovery file (by default it does that every minute, and by default Prometheus will reload the file the minute it is written). After reloading your Prometheus master configuration, this program will begin informing via the discovery file of new targets that Prometheus must scrape.