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.

Open source

Migrate from Grafana Agent Operator to Grafana Agent Flow

With the release of Grafana Agent Flow, Grafana Agent Operator is no longer the recommended way to deploy Grafana Agent in Kubernetes. Some of the Operator functionality has moved into Grafana Agent Flow itself, and the Helm Chart has replaced the remaining functionality.

  • The Monitor types (PodMonitor, ServiceMonitor, Probe, and PodLogs) are all supported natively by Grafana Agent Flow. You are no longer required to use the Operator to consume those CRDs for dynamic monitoring in your cluster.
  • The parts of the Operator that deploy the Grafana Agent itself (GrafanaAgent, MetricsInstance, and LogsInstance CRDs) are deprecated. Operator users should use the Grafana Agent [Helm Chart][] to deploy Grafana Agent directly to your clusters.

This guide provides some steps to get started with Grafana Agent Flow for users coming from Grafana Agent Operator.

Deploy Grafana Agent Flow with Helm

  1. Create a values.yaml file, which contains options for deploying your Grafana Agent. You can start with the default values and customize as you see fit, or start with this snippet, which should be a good starting point for what the Operator does.

    yaml
    agent:
      mode: 'flow'
      configMap:
        create: true
      clustering:
        enabled: true
    controller:
      type: 'statefulset'
      replicas: 2
    crds:
      create: false

    This configuration deploys Grafana Agent Flow as a StatefulSet using the built-in clustering functionality to allow distributing scrapes across all Grafana Agent Pods.

    This is one of many deployment possible modes. For example, you may want to use a DaemonSet to collect host-level logs or metrics. See the Grafana Agent Flow deployment guide for more details about different topologies.

  2. Create a Grafana Agent configuration file, agent.river.

    In the next step, you add to this configuration as you convert MetricsInstances. You can add any additional configuration to this file as you need.

  3. Install the Grafana Helm repository:

    helm repo add grafana https://grafana.github.io/helm-charts
    helm repo update
  4. Create a Helm release. You can name the release anything you like. The following command installs a release called grafana-agent-metrics in the monitoring namespace.

    shell
    helm upgrade grafana-agent-metrics grafana/grafana-agent -i -n monitoring -f values.yaml --set-file agent.configMap.content=agent.river

    This command uses the --set-file flag to pass the configuration file as a Helm value so that you can continue to edit it as a regular River file.

Convert MetricsIntances to Grafana Agent Flow components

A MetricsInstance resource primarily defines:

  • The remote endpoints Grafana Agent Flow should send metrics to.
  • The PodMonitor, ServiceMonitor, and Probe resources this Grafana Agent should discover.

You can use these functions in Grafana Agent Flow with the prometheus.remote_write, prometheus.operator.podmonitors, prometheus.operator.servicemonitors, and prometheus.operator.probes components respectively.

The following River sample is equivalent to the MetricsInstance from the operator guide.

river

// read the credentials secret for remote_write authorization
remote.kubernetes.secret "credentials" {
  namespace = "monitoring"
  name = "primary-credentials-metrics"
}

prometheus.remote_write "primary" {
    endpoint {
        url = "https://<PROMETHEUS_URL>/api/v1/push"
        basic_auth {
            username = nonsensitive(remote.kubernetes.secret.credentials.data["username"])
            password = remote.kubernetes.secret.credentials.data["password"]
        }
    }
}

prometheus.operator.podmonitors "primary" {
    forward_to = [prometheus.remote_write.primary.receiver]
    // leave out selector to find all podmonitors in the entire cluster
    selector {
        match_labels = {instance = "primary"}
    }
}

prometheus.operator.servicemonitors "primary" {
    forward_to = [prometheus.remote_write.primary.receiver]
    // leave out selector to find all servicemonitors in the entire cluster
    selector {
        match_labels = {instance = "primary"}
    }
}

Replace the following:

  • <PROMETHEUS_URL>: The endpoint you want to send metrics to.

This configuration discovers all PodMonitor, ServiceMonitor, and Probe resources in your cluster that match the label selector instance=primary. It then scrapes metrics from the targets and forward them to your remote write endpoint.

You may need to customize this configuration further if you use additional features in your MetricsInstance resources. Refer to the documentation for the relevant components for additional information:

Collecting Logs

Our current recommendation is to create an additional DaemonSet deployment of Grafana Agents to scrape logs.

We have components that can scrape pod logs directly from the Kubernetes API without needing a DaemonSet deployment. These are still considered experimental, but if you would like to try them, see the documentation for loki.source.kubernetes and loki.source.podlogs.

These values are close to what the Operator currently deploys for logs:

yaml
agent:
  mode: 'flow'
  configMap:
    create: true
  clustering:
    enabled: false
  controller:
    type: 'daemonset'
  mounts:
    # -- Mount /var/log from the host into the container for log collection.
    varlog: true

This command will install a release named grafana-agent-logs in the monitoring namespace:

helm upgrade grafana-agent-logs grafana/grafana-agent -i -n monitoring -f values-logs.yaml --set-file agent.configMap.content=agent-logs.river

This simple configuration will scrape logs for every pod on each node:

river
// read the credentials secret for remote_write authorization
remote.kubernetes.secret "credentials" {
  namespace = "monitoring"
  name      = "primary-credentials-logs"
}

discovery.kubernetes "pods" {
  role = "pod"
  // limit to pods on this node to reduce the amount you need to filter
  selectors {
    role  = "pod"
    field = "spec.nodeName=" + env("<HOSTNAME>")
  }
}

discovery.relabel "pod_logs" {
  targets = discovery.kubernetes.pods.targets
  rule {
    source_labels = ["__meta_kubernetes_namespace"]
    target_label  = "namespace"
  }
  rule {
    source_labels = ["__meta_kubernetes_pod_name"]
    target_label  = "pod"
  }
  rule {
    source_labels = ["__meta_kubernetes_pod_container_name"]
    target_label  = "container"
  }
  rule {
    source_labels = ["__meta_kubernetes_namespace", "__meta_kubernetes_pod_name"]
    separator     = "/"
    target_label  = "job"
  }
  rule {
    source_labels = ["__meta_kubernetes_pod_uid", "__meta_kubernetes_pod_container_name"]
    separator     = "/"
    action        = "replace"
    replacement   = "/var/log/pods/*$1/*.log"
    target_label  = "__path__"
  }
  rule {
    action = "replace"
    source_labels = ["__meta_kubernetes_pod_container_id"]
    regex = "^(\\w+):\\/\\/.+$"
    replacement = "$1"
    target_label = "tmp_container_runtime"
  }
}

local.file_match "pod_logs" {
  path_targets = discovery.relabel.pod_logs.output
}

loki.source.file "pod_logs" {
  targets    = local.file_match.pod_logs.targets
  forward_to = [loki.process.pod_logs.receiver]
}

// basic processing to parse the container format. You can add additional processing stages
// to match your application logs.
loki.process "pod_logs" {
  stage.match {
    selector = "{tmp_container_runtime=\"containerd\"}"
    // the cri processing stage extracts the following k/v pairs: log, stream, time, flags
    stage.cri {}
    // Set the extract flags and stream values as labels
    stage.labels {
      values = {
        flags   = "",
        stream  = "",
      }
    }
  }

  // if the label tmp_container_runtime from above is docker parse using docker
  stage.match {
    selector = "{tmp_container_runtime=\"docker\"}"
    // the docker processing stage extracts the following k/v pairs: log, stream, time
    stage.docker {}

    // Set the extract stream value as a label
    stage.labels {
      values = {
        stream  = "",
      }
    }
  }

  // drop the temporary container runtime label as it is no longer needed
  stage.label_drop {
    values = ["tmp_container_runtime"]
  }

  forward_to = [loki.write.loki.receiver]
}

loki.write "loki" {
  endpoint {
    url = "https://<LOKI_URL>/loki/api/v1/push"
    basic_auth {
      username = nonsensitive(remote.kubernetes.secret.credentials.data["username"])
      password = remote.kubernetes.secret.credentials.data["password"]
    }
}
}

Replace the following:

  • <LOKI_URL>: The endpoint of your Loki instance.

The logging subsystem is very powerful and has many options for processing logs. For further details, see the component documentation.

Integrations

The Integration CRD isn’t supported with Grafana Agent Flow. However, all static mode integrations have an equivalent component in the prometheus.exporter namespace. The [reference documentation][component documentation] should help convert those integrations to their Grafana Agent Flow equivalent.