Skip to content

A command-line tool (and Lambda) for collecting Buildkite agent metrics


Notifications You must be signed in to change notification settings


Repository files navigation

Buildkite Agent Metrics

A command-line tool for collecting Buildkite agent metrics, focusing on enabling auto-scaling. Currently AWS Cloudwatch, StatsD, Prometheus, Stackdriver and New Relic are supported.

Build status


Either download the latest binary from Github Releases or install with:

go install


Several running modes are supported. All of them require an Agent Registration Token, found on the Buildkite Agents page.

Running as a Daemon

The simplest deployment is to run as a long-running daemon that collects metrics across all queues in an organization.

buildkite-agent-metrics -token abc123 -interval 30s

Restrict it to a single queue with -queue:

buildkite-agent-metrics -token abc123 -interval 30s -queue my-queue

Restrict it to multiple queues by repeating -queue:

buildkite-agent-metrics -token abc123 -interval 30s -queue my-queue1 -queue my-queue2

When using clusters, you can pass a cluster registration token to gather metrics only for that cluster:

buildkite-agent-metrics -token clustertoken ...

You can repeat -token to gather metrics for multiple clusters:

buildkite-agent-metrics -token clusterAtoken -token clusterBtoken ...

Running as an AWS Lambda

An AWS Lambda bundle is created and published as part of the build process. The lambda will require the cloudwatch:PutMetricData IAM permission.

It requires a provided.al2 environment and respects the following env vars:

  • BUILDKITE_BACKEND : The name of the backend to use (e.g. cloudwatch, statsd, newrelic. For the lambda, prometheus and stackdriver are not supported).
  • BUILDKITE_QUEUE : A comma separated list of Buildkite queues to process (e.g. backend-deploy,ui-deploy).
  • BUILDKITE_QUIET : A boolean specifying that only ERROR log lines must be printed. (e.g. 1, true).
  • BUILDKITE_CLOUDWATCH_DIMENSIONS : A comma separated list in the form of Key=Value,Other=Value containing the Cloudwatch dimensions to index metrics under.

Additionally, one of the following groups of environment variables must be set in order to define how the Lambda function should obtain the required Buildkite Agent API token:

Option 1 - Provide the token(s) as plain-text

  • BUILDKITE_AGENT_TOKEN : The Buildkite Agent API token to use. You can supply multiple tokens comma-separated.

Option 2 - Retrieve token from AWS Systems Manager

  • BUILDKITE_AGENT_TOKEN_SSM_KEY : The parameter name which contains the token value in AWS Systems Manager. You can supply multiple names comma-separated.

Note: Parameters stored as String and SecureString are currently supported.

Option 3 - Retrieve token from AWS Secrets Manager

  • BUILDKITE_AGENT_SECRETS_MANAGER_SECRET_ID: The id of the secret which contains the token value in AWS Secrets Manager. You can supply multiple ids comma-separated.
  • (Optional) BUILDKITE_AGENT_SECRETS_MANAGER_JSON_KEY: The JSON key containing the token value in the secret JSON blob.

Note 1: Both SecretBinary and SecretString are supported. In the case of SecretBinary, the secret payload will be automatically decoded and returned as a plain-text string.

Note 2: BUILDKITE_AGENT_SECRETS_MANAGER_JSON_KEY can be used on secrets of type SecretBinary only if their binary payload corresponds to a valid JSON object containing the provided key.

aws lambda create-function \
  --function-name buildkite-agent-metrics \
  --memory 128 \
  --role arn:aws:iam::account-id:role/execution_role \
  --runtime provided.al2 \
  --zip-file fileb:// \
  --handler handler

Running as a Container

You can build a docker image for the buildkite-agent-metrics following:

docker build -t buildkite-agent-metrics .

This will create a local docker image named as buildkite-agent-metrics that you can tag and push to your own registry.

You can use the command-line arguments in a docker execution in the same way as described before:

docker run --rm buildkite-agent-metrics -token abc123 -interval 30s -queue my-queue

Supported command line flags

$ buildkite-agent-metrics --help
Usage of buildkite-agent-metrics:
  -backend string
    	Specify the backend to use: cloudwatch, statsd, prometheus, stackdriver (default "cloudwatch")
  -cloudwatch-dimensions string
    	Cloudwatch dimensions to index metrics under, in the form of Key=Value, Other=Value
  -cloudwatch-region string
    	AWS Region to connect to, defaults to $AWS_REGION or us-east-1
    	Show debug output
    	Show full http traces
    	Whether to only print metrics
  -endpoint string
    	A custom Buildkite Agent API endpoint (default "")
  -interval duration
    	Update metrics every interval, rather than once
  -newrelic-app-name string
    	New Relic application name for metric events
  -newrelic-license-key string
    	New Relic license key for publishing events
  -prometheus-addr string
    	Prometheus metrics transport bind address (default ":8080")
  -prometheus-path string
    	Prometheus metrics transport path (default "/metrics")
  -queue value
    	Specific queues to process
    	Only print errors
  -stackdriver-projectid string
    	Specify Stackdriver Project ID
  -statsd-host string
    	Specify the StatsD server (default "")
    	Whether your StatsD server supports tagging like Datadog
  -token string
    	A Buildkite Agent Registration Token
    	Show the version


By default metrics will be submitted to CloudWatch but the backend can be switched to StatsD or Prometheus using the command-line argument -backend statsd or -backend prometheus respectively.


The Cloudwatch backend supports the following arguments:

  • -cloudwatch-dimensions: A optional custom dimension in the form of Key=Value, Key=Value

StatsD (Datadog)

The StatsD backend supports the following arguments:

  • -statsd-host HOST: The StatsD host and port (defaults to
  • -statsd-tags: Some StatsD servers like the agent provided by Datadog support tags. If specified, metrics will be tagged by queue otherwise metrics will include the queue name in the metric. Only enable this option if you know your StatsD server supports tags.


The Prometheus backend supports the following arguments:

  • -prometheus-addr: The local address to listen on (defaults to :8080).
  • -prometheus-path: The path under prometheus-addr to expose metrics on (defaults to /metrics).


The Stackdriver backend supports the following arguments:

  • -stackdriver-projectid: The Google Cloud Platform project to report metrics for.

The New Relic backend supports the following arguments:

  • -newrelic-app-name: String for the New Relic app name
  • -newrelic-license-key: The New Relic license key. Must be of type INGEST

Upgrading from v2 to v3

  1. The -org argument is no longer needed
  2. The -token argument is now an Agent Registration Token — the same used in the Buildkite Agent configuration file, and found on the Buildkite Agents page.
  3. Build and pipeline metrics have been removed, focusing on agents and jobs by queue for auto–scaling. If you have a compelling reason to gather build or pipeline metrics please continue to use the previous version or open an issue with details.


This tool is built with Go 1.20+ and assumes Go Modules by default.

You can build and run the binary tool locally with Go installed:

go run *.go -token [buildkite agent registration token]

Currently this will publish metrics to Cloudwatch under the custom metric prefix of Buildkite, using AWS credentials from your environment. The machine will require the cloudwatch:PutMetricData IAM permission.

The token package

It is an abstraction layer enabling the retrieval of a Buildkite Agent API token from different sources.

The current supported sources are:

  • AWS Systems Manager (a.k.a parameter store).
  • AWS Secrets Manager.
  • OS environment variable.


All the tests for AWS dependant resources require their corresponding auto-generated mocks. Thus, before running them, you need to generate such mocks by executing:

go generate token/secretsmanager_test.go
go generate token/ssm_test.go


The following metrics are gathered when no specific queue is supplied:

Buildkite > (Org) > RunningJobsCount
Buildkite > (Org) > ScheduledJobsCount
Buildkite > (Org) > UnfinishedJobsCount
Buildkite > (Org) > WaitingJobsCount
Buildkite > (Org) > IdleAgentsCount
Buildkite > (Org) > BusyAgentsCount
Buildkite > (Org) > BusyAgentPercentage
Buildkite > (Org) > TotalAgentsCount

Buildkite > (Org, Queue) > RunningJobsCount
Buildkite > (Org, Queue) > ScheduledJobsCount
Buildkite > (Org, Queue) > UnfinishedJobsCount
Buildkite > (Org, Queue) > WaitingJobsCount
Buildkite > (Org, Queue) > IdleAgentsCount
Buildkite > (Org, Queue) > BusyAgentsCount
Buildkite > (Org, Queue) > BusyAgentPercentage
Buildkite > (Org, Queue) > TotalAgentsCount

When a queue is specified, only that queue's metrics are published.

We send metrics for Jobs in the following states:

  • Scheduled: the job hasn't been assigned to an agent yet. If you have agent capacity, this value should be close to 0.
  • Waiting: the job is known to exist but isn't schedulable yet due to dependencies, wait statements, etc. This information is mostly useful to an autoscaler, since it represents work that will start soon.
  • Running: an agent is actively executing this job.


See (MIT)