Menu

Caution

Grafana Alloy is the new name for our distribution of the OTel collector. Grafana Agent has been deprecated and is in Long-Term Support (LTS) through October 31, 2025. Grafana Agent will reach an End-of-Life (EOL) on November 1, 2025. Read more about why we recommend migrating to Grafana Alloy.

Important: This documentation is about an older version. It's relevant only to the release noted, many of the features and functions have been updated or replaced. Please view the current version.

DocumentationGrafana AgentFlow modeGet startedCollect Prometheus metrics
Open source

Collect Prometheus metrics

Grafana Agent Flow can be configured to collect Prometheus metrics and forward them to any Prometheus-compatible database.

This topic describes how to:

  • Configure metrics delivery
  • Collect metrics from Kubernetes Pods

Components used in this topic

Before you begin

  • Ensure that you have basic familiarity with instrumenting applications with Prometheus.
  • Have a set of Prometheus exports or applications exposing Prometheus metrics that you want to collect metrics from.
  • Identify where you will write collected metrics. Metrics may be written to Prometheus or Prometheus-compatible endpoints such as Grafana Mimir, Grafana Cloud, or Grafana Enterprise Metrics.
  • Be familiar with the concept of Components in Grafana Agent Flow.

Configure metrics delivery

Before components can collect Prometheus metrics, you must have a component responsible for writing those metrics somewhere.

The prometheus.remote_write component is responsible for delivering Prometheus metrics to one or Prometheus-compatible endpoints. Once a prometheus.remote_write component is defined, other Grafana Agent Flow components can be used to forward metrics to it.

To configure a prometheus.remote_write component for metrics delivery, complete the following steps:

  1. Add the following prometheus.remote_write component to your configuration file:

    river 
    prometheus.remote_write "LABEL" {
  endpoint {
    url = "PROMETHEUS_URL"
  }
}

    1. Replace LABEL with a label to use for the component, such as default. The label chosen must be unique across all prometheus.remote_write components in the same configuration file.

    2. Replace PROMETHEUS_URL with the full URL of the Prometheus-compatible endpoint where metrics will be sent, such as https://prometheus-us-central1.grafana.net/api/prom/push.

  2. If your endpoint requires basic authentication, paste the following inside of the endpoint block:

    river 
    basic_auth {
  username = "USERNAME"
  password = "PASSWORD"
}

    1. Replace USERNAME with the basic authentication username to use.

    2. Replace PASSWORD with the basic authentication password or API key to use.

  3. If you have more than one endpoint to write metrics to, repeat the endpoint block for additional endpoints.

The following example demonstrates configuring prometheus.remote_write with multiple endpoints and mixed usage of basic authentication, and a prometheus.scrape component which forwards metrics to it:

river 
prometheus.remote_write "default" {
  endpoint {
    url = "http://localhost:9090/api/prom/push"
  }

  endpoint {
    url = "https://prometheus-us-central1.grafana.net/api/prom/push"

    // Get basic authentication based on environment variables.
    basic_auth {
      username = env("REMOTE_WRITE_USERNAME")
      password = env("REMOTE_WRITE_PASSWORD")
    }
  }
}

prometheus.scrape "example" {
  // Collect metrics from Grafana Agent's default listen address.
  targets = [{
    __address__ = "127.0.0.1:12345",
  }]

  forward_to = [prometheus.remote_write.default.receiver]
}

For more information on configuring metrics delivery, refer to prometheus.remote_write.

Collect metrics from Kubernetes Pods

Grafana Agent Flow can be configured to collect metrics from Kubernetes Pods by:

  1. Discovering Kubernetes Pods to collect metrics from.
  2. Collecting metrics from those discovered Pods.

To collect metrics from Kubernetes Pods, complete the following steps:

  1. Follow Configure metrics delivery to ensure collected metrics can be written somewhere.

  2. Discover Kubernetes Pods:

    1. Add the following discovery.kubernetes component to your configuration file to discover every Pod in the cluster across all Namespaces:

      river 
      discovery.kubernetes "DISCOVERY_LABEL" {
  role = "pod"
}

      This will generate one Prometheus target for every exposed port on every discovered Pod.

      1. Replace DISCOVERY_LABEL with a label to use for the component, such as pods. The label chosen must be unique across all discovery.kubernetes components in the same configuration file.

    2. To limit the Namespaces that Pods are discovered in, add the following block inside of the discovery.kubernetes component:

      river 
      namespaces {
  own_namespace = true
  names         = [NAMESPACE_NAMES]
}

      1. If you do not want to search for Pods in the Namespace Grafana Agent is running in, set own_namespace to false.

      2. Replace NAMESPACE_NAMES with a comma-delimited list of strings representing Namespaces to search. Each string must be wrapped in double quotes. For example, "default","kube-system".

    3. To use a field selector to limit the number of discovered Pods, add the following block inside of the discovery.kubernetes component:

      river 
      selectors {
  role  = "pod"
  field = "FIELD_SELECTOR"
}

      1. Replace FIELD_SELECTOR with the Kubernetes field selector to use, such as metadata.name=my-service. For more information on field selectors, refer to the Kubernetes documentation on Field Selectors.

      2. Create additional selectors blocks for each field selector you want to apply.

    4. To use a label selector to limit the number of discovered Pods, add the following block inside of the discovery.kubernetes component:

      river 
      selectors {
  role  = "pod"
  label = "LABEL_SELECTOR"
}

      1. Replace LABEL_SELECTOR with the Kubernetes label selector to use, such as environment in (production, qa). For more information on label selectors, refer to the Kubernetes documentation on Labels and Selectors.

      2. Create additional selectors blocks for each label selector you want to apply.

  3. Collect metrics from discovered Pods:

    1. Add the following prometheus.scrape component to your configuration file:

      river 
      prometheus.scrape "SCRAPE_LABEL" {
  targets    = discovery.kubernetes.DISCOVERY_LABEL.targets
  forward_to = [prometheus.remote_write.REMOTE_WRITE_LABEL.receiver]
}

      1. Replace SCRAPE_LABEL with a label to use for the component, such as pods. The label chosen must be unique across all prometeus.scrape components in the same configuration file.

      2. Replace DISCOVERY_LABEL with the label chosen for the discovery.kubernetes component in step 2.1.1.

      3. Replace REMOTE_WRITE_LABEL with the label chosen for your existing prometheus.remote_write component.

The following example demonstrates configuring Grafana Agent to collect metrics from running production Kubernetes Pods in the default Namespace:

river 
discovery.kubernetes "pods" {
  role = "pod"

  namespaces {
    own_namespace = false

    names = ["default"]
  }

  selectors {
    role  = "pod"
    label = "environment in (production)"
  }
}

prometheus.scrape "pods" {
  targets    = discovery.kubernetes.pods.targets
  forward_to = [prometheus.remote_write.default.receiver]
}

prometheus.remote_write "default" {
  endpoint {
    url = "http://localhost:9090/api/prom/push"
  }
}

For more information on configuring Kubernetes service delivery and collecting metrics, refer to discovery.kubernetes and prometheus.scrape.

Collect metrics from Kubernetes Services

Grafana Agent Flow can be configured to collect metrics from Kubernetes Services by:

  1. Discovering Kubernetes Services to collect metrics from.
  2. Collecting metrics from those discovered Services.

To collect metrics from Kubernetes Services, complete the following steps:

  1. Follow Configure metrics delivery to ensure collected metrics can be written somewhere.

  2. Discover Kubernetes Services:

    1. Add the following discovery.kubernetes component to your configuration file to discover every Services in the cluster across all Namespaces:

      river 
      discovery.kubernetes "DISCOVERY_LABEL" {
  role = "service"
}

      This will generate one Prometheus target for every exposed port on every discovered Service.

      1. Replace DISCOVERY_LABEL with a label to use for the component, such as services. The label chosen must be unique across all discovery.kubernetes components in the same configuration file.

    2. To limit the Namespaces that Services are discovered in, add the following block inside of the discovery.kubernetes component:

      river 
      namespaces {
  own_namespace = true
  names         = [NAMESPACE_NAMES]
}

      1. If you do not want to search for Services in the Namespace Grafana Agent is running in, set own_namespace to false.

      2. Replace NAMESPACE_NAMES with a comma-delimited list of strings representing Namespaces to search. Each string must be wrapped in double quotes. For example, "default","kube-system".

    3. To use a field selector to limit the number of discovered Services, add the following block inside of the discovery.kubernetes component:

      river 
      selectors {
  role  = "service"
  field = "FIELD_SELECTOR"
}

      1. Replace FIELD_SELECTOR with the Kubernetes field selector to use, such as metadata.name=my-service. For more information on field selectors, refer to the Kubernetes documentation on Field Selectors.

      2. Create additional selectors blocks for each field selector you want to apply.

    4. To use a label selector to limit the number of discovered Services, add the following block inside of the discovery.kubernetes component:

      river 
      selectors {
  role  = "service"
  label = "LABEL_SELECTOR"
}

      1. Replace LABEL_SELECTOR with the Kubernetes label selector to use, such as environment in (production, qa). For more information on label selectors, refer to the Kubernetes documentation on Labels and Selectors.

      2. Create additional selectors blocks for each label selector you want to apply.

  3. Collect metrics from discovered Services:

    1. Add the following prometheus.scrape component to your configuration file:

      river 
      prometheus.scrape "SCRAPE_LABEL" {
  targets    = discovery.kubernetes.DISCOVERY_LABEL.targets
  forward_to = [prometheus.remote_write.REMOTE_WRITE_LABEL.receiver]
}

      1. Replace SCRAPE_LABEL with a label to use for the component, such as services. The label chosen must be unique across all prometeus.scrape components in the same configuration file.

      2. Replace DISCOVERY_LABEL with the label chosen for the discovery.kubernetes component in step 2.1.1.

      3. Replace REMOTE_WRITE_LABEL with the label chosen for your existing prometheus.remote_write component.

The following example demonstrates configuring Grafana Agent to collect metrics from running production Kubernetes Services in the default Namespace:

river 
discovery.kubernetes "services" {
  role = "service"

  namespaces {
    own_namespace = false

    names = ["default"]
  }

  selectors {
    role  = "service"
    label = "environment in (production)"
  }
}

prometheus.scrape "services" {
  targets    = discovery.kubernetes.services.targets
  forward_to = [prometheus.remote_write.default.receiver]
}

prometheus.remote_write "default" {
  endpoint {
    url = "http://localhost:9090/api/prom/push"
  }
}

For more information on configuring Kubernetes service delivery and collecting metrics, refer to discovery.kubernetes and prometheus.scrape.

Collect metrics from custom targets

Grafana Agent Flow can be configured to collect metrics from a custom set of targets without the need for service discovery.

To collect metrics from a custom set of targets, complete the following steps:

  1. Follow Configure metrics delivery to ensure collected metrics can be written somewhere.

  2. Add the following prometheus.scrape component to your configuration file:

    river 
    prometheus.scrape "SCRAPE_LABEL" {
  targets    = [TARGET_LIST]
  forward_to = [prometheus.remote_write.REMOTE_WRITE_LABEL.receiver]
}

    1. Replace SCRAPE_LABEL with a label to use for the component, such as custom_targets. The label chosen must be unique across all prometheus.scrape components in the same configuration file.

    2. Replace TARGET_LIST with a comma-delimited list of Objects denoting the Prometheus target. Each object must conform to the following rules:

      • There must be an __address__ key denoting the HOST:PORT of the target to collect metrics from.

      • To explicitly specify which protocol to use, set the __scheme__ key to "http" or "https". If the __scheme__ key is not provided, the protocol to use is inherited by the settings of the prometheus.scrape component (default "http").

      • To explicitly specify which HTTP path to collect metrics from, set the __metrics_path__ key to the HTTP path to use. If the __metrics_path__ key is not provided, the protocol to use is inherited by the settings of the prometheus.scrape component (default "/metrics").

      • Add additional keys as desired to inject extra labels to collected metrics. Any label starting with two underscores (__) will be dropped prior to scraping.

    3. Replace REMOTE_WRITE_LABEL with the label chosen for your existing prometheus.remote_write component.

The following example demonstrates configuring prometheus.scrape to collect metrics from a custom set of endpoints:

river 
prometheus.scrape "custom_targets" {
  targets = [
    {
      __address__ = "prometheus:9090",
    },
    {
      __address__ = "mimir:8080",
      __scheme__  = "https",
    },
    {
      __address__      = "custom-application:80",
      __metrics_path__ = "/custom-metrics–path",
    },
    {
      __address__ = "grafana-agent:12345",
      application = "grafana-agent",
      environment = "production",
    },
  ]

  forward_to = [prometheus.remote_write.default.receiver]
}

prometheus.remote_write "default" {
  endpoint {
    url = "http://localhost:9090/api/prom/push"
  }
}