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

The run command

The run command runs Grafana Agent Flow in the foreground until an interrupt is received.

Usage

Usage:

  • AGENT_MODE=flow grafana-agent run [FLAG ...] PATH_NAME

  • grafana-agent-flow run [FLAG ...] PATH_NAME

    Replace the following:

    • FLAG: One or more flags that define the input and output of the command.
    • PATH_NAME: Required. The Grafana Agent configuration file/directory path.

If the PATH_NAME argument is not provided, or if the configuration path can’t be loaded or contains errors during the initial load, the run command will immediately exit and show an error message.

If you give the PATH_NAME argument a directory path, the agent will find *.river files (ignoring nested directories) and load them as a single configuration source. However, component names must be unique across all River files, and configuration blocks must not be repeated.

Grafana Agent Flow will continue to run if subsequent reloads of the configuration file fail, potentially marking components as unhealthy depending on the nature of the failure. When this happens, Grafana Agent Flow will continue functioning in the last valid state.

run launches an HTTP server that exposes metrics about itself and its components. The HTTP server is also exposes a UI at / for debugging running components.

The following flags are supported:

  • --server.http.enable-pprof: Enable /debug/pprof profiling endpoints. (default true)
  • --server.http.memory-addr: Address to listen for in-memory HTTP traffic on (default agent.internal:12345).
  • --server.http.listen-addr: Address to listen for HTTP traffic on (default 127.0.0.1:12345).
  • --server.http.ui-path-prefix: Base path where the UI is exposed (default /).
  • --storage.path: Base directory where components can store data (default data-agent/).
  • --disable-reporting: Disable data collection (default false).
  • --cluster.enabled: Start the Agent in clustered mode (default false).
  • --cluster.node-name: The name to use for this node (defaults to the environment’s hostname).
  • --cluster.join-addresses: Comma-separated list of addresses to join the cluster at (default ""). Mutually exclusive with --cluster.discover-peers.
  • --cluster.discover-peers: List of key-value tuples for discovering peers (default ""). Mutually exclusive with --cluster.join-addresses.
  • --cluster.rejoin-interval: How often to rejoin the list of peers (default "60s").
  • --cluster.advertise-address: Address to advertise to other cluster nodes (default "").
  • --cluster.advertise-interfaces: List of interfaces used to infer an address to advertise. Set to all to use all available network interfaces on the system. (default "eth0,en0").
  • --cluster.max-join-peers: Number of peers to join from the discovered set (default 5).
  • --cluster.name: Name to prevent nodes without this identifier from joining the cluster (default "").
  • --config.format: The format of the source file. Supported formats: flow, prometheus, promtail, static (default "flow").
  • --config.bypass-conversion-errors: Enable bypassing errors when converting (default false).

Update the configuration file

The configuration file can be reloaded from disk by either:

  • Sending an HTTP POST request to the /-/reload endpoint.
  • Sending a SIGHUP signal to the Grafana Agent process.

When this happens, the component controller synchronizes the set of running components with the latest set of components specified in the configuration file. Components that are no longer defined in the configuration file after reloading are shut down, and components that have been added to the configuration file since the previous reload are created.

All components managed by the component controller are reevaluated after reloading.

Clustering (beta)

The --cluster.enabled command-line argument starts Grafana Agent in clustering mode. The rest of the --cluster.* command-line flags can be used to configure how nodes discover and connect to one another.

Each cluster member’s name must be unique within the cluster. Nodes which try to join with a conflicting name are rejected and will fall back to bootstrapping a new cluster of their own.

Peers communicate over HTTP/2 on the agent’s built-in HTTP server. Each node must be configured to accept connections on --server.http.listen-addr and the address defined or inferred in --cluster.advertise-address.

If the --cluster.advertise-address flag is not explicitly set, the agent tries to infer a suitable one from --cluster.advertise-interfaces. If --cluster.advertise-interfaces is not explicitly set, the agent will infer one from the eth0 and en0 local network interfaces. The agent will fail to start if it can’t determine the advertised address. Since Windows does not use the interface names eth0 or en0, Windows users must explicitly pass at least one valid network interface for --cluster.advertise-interfaces or a value for --cluster.advertise-address.

The comma-separated list of addresses provided in --cluster.join-addresses can either be IP addresses with an optional port, or DNS records to lookup. The ports on the list of addresses default to the port used for the HTTP listener if not explicitly provided. We recommend that you align the port numbers on as many nodes as possible to simplify the deployment process.

The --cluster.discover-peers command-line flag expects a list of tuples in the form of provider=XXX key=val key=val .... Clustering uses the go-discover package to discover peers and fetch their IP addresses, based on the chosen provider and the filtering key-values it supports. Clustering supports the default set of providers available in go-discover and registers the k8s provider on top.

If either the key or the value in a tuple pair contains a space, a backslash, or double quotes, then it must be quoted with double quotes. Within this quoted string, the backslash can be used to escape double quotes or the backslash itself.

The --cluster.rejoin-interval flag defines how often each node should rediscover peers based on the contents of the --cluster.join-addresses and --cluster.discover-peers flags and try to rejoin them. This operation is useful for addressing split-brain issues if the initial bootstrap is unsuccessful and for making clustering easier to manage in dynamic environments. To disable this behavior, set the --cluster.rejoin-interval flag to "0s".

Discovering peers using the --cluster.join-addresses and --cluster.discover-peers flags only happens on startup; after that, cluster nodes depend on gossiping messages with each other to converge on the cluster’s state.

The first node that is used to bootstrap a new cluster (also known as the “seed node”) can either omit the flags that specify peers to join or can try to connect to itself.

To join or rejoin a cluster, the agent will try to connect to a certain number of peers limited by the --cluster.max-join-peers flag. This flag can be useful for clusters of significant sizes because connecting to a high number of peers can be an expensive operation. To disable this behavior, set the --cluster.max-join-peers flag to 0. If the value of --cluster.max-join-peers is higher than the number of peers discovered, the agent will connect to all of them.

The --cluster.name flag can be used to prevent clusters from accidentally merging. When --cluster.name is provided, nodes will only join peers who share the same cluster name value. By default, the cluster name is empty, and any node that doesn’t set the flag can join. Attempting to join a cluster with a wrong --cluster.name will result in a “failed to join memberlist” error.

Clustering states

Clustered agents are in one of three states:

  • Viewer: The agent has a read-only view of the cluster and is not participating in workload distribution.

  • Participant: The agent is participating in workload distribution for components that have clustering enabled.

  • Terminating: The agent is shutting down and will no longer assign new work to itself.

Agents initially join the cluster in the viewer state and then transition to the participant state after the process startup completes. Agents then transition to the terminating state when shutting down.

The current state of a clustered agent is shown on the clustering page in the UI.

Configuration conversion (beta)

When you use the --config.format command-line argument with a value other than flow, Grafana Agent converts the configuration file from the source format to River and immediately starts running with the new configuration. This conversion uses the converter API described in the grafana-agent-flow convert docs.

If you also use the --config.bypass-conversion-errors command-line argument, Grafana Agent will ignore any errors from the converter. Use this argument with caution because the resulting conversion may not be equivalent to the original configuration.