»Consul Agent

The Consul agent is the core process of Consul. The agent maintains membership information, registers services, runs checks, responds to queries, and more. The agent must run on every node that is part of a Consul cluster.

Any agent may run in one of two modes: client or server. A server node takes on the additional responsibility of being part of the consensus quorum. These nodes take part in Raft and provide strong consistency and availability in the case of failure. The higher burden on the server nodes means that usually they should be run on dedicated instances -- they are more resource intensive than a client node. Client nodes make up the majority of the cluster, and they are very lightweight as they interface with the server nodes for most operations and maintain very little state of their own.

»Running an Agent

The agent is started with the consul agent command. This command blocks, running forever or until told to quit. You can test a local agent by following the Getting Started tutorials.

The agent command takes a variety of configuration options, but most have sane defaults.

When running consul agent, you should see output similar to this:

$ consul agent -data-dir=/tmp/consul
==> Starting Consul agent...
==> Consul agent running!
       Node name: 'Armons-MacBook-Air'
      Datacenter: 'dc1'
          Server: false (bootstrap: false)
     Client Addr: (HTTP: 8500, DNS: 8600)
    Cluster Addr: (LAN: 8301, WAN: 8302)

==> Log data will now stream in as it occurs:

    [INFO] serf: EventMemberJoin: Armons-MacBook-Air.local
$ consul agent -data-dir=/tmp/consul==> Starting Consul agent...==> Consul agent running!       Node name: 'Armons-MacBook-Air'      Datacenter: 'dc1'          Server: false (bootstrap: false)     Client Addr: (HTTP: 8500, DNS: 8600)    Cluster Addr: (LAN: 8301, WAN: 8302)
==> Log data will now stream in as it occurs:
    [INFO] serf: EventMemberJoin: Armons-MacBook-Air.local

There are several important messages that consul agent outputs:

  • Node name: This is a unique name for the agent. By default, this is the hostname of the machine, but you may customize it using the -node flag.

  • Datacenter: This is the datacenter in which the agent is configured to run. Consul has first-class support for multiple datacenters; however, to work efficiently, each node must be configured to report its datacenter. The -datacenter flag can be used to set the datacenter. For single-DC configurations, the agent will default to "dc1".

  • Server: This indicates whether the agent is running in server or client mode. Server nodes have the extra burden of participating in the consensus quorum, storing cluster state, and handling queries. Additionally, a server may be in "bootstrap" mode. Multiple servers cannot be in bootstrap mode as that would put the cluster in an inconsistent state.

  • Client Addr: This is the address used for client interfaces to the agent. This includes the ports for the HTTP and DNS interfaces. By default, this binds only to localhost. If you change this address or port, you'll have to specify a -http-addr whenever you run commands such as consul members to indicate how to reach the agent. Other applications can also use the HTTP address and port to control Consul.

  • Cluster Addr: This is the address and set of ports used for communication between Consul agents in a cluster. Not all Consul agents in a cluster have to use the same port, but this address MUST be reachable by all other nodes.

When running under systemd on Linux, Consul notifies systemd by sending READY=1 to the $NOTIFY_SOCKET when a LAN join has completed. For this either the join or retry_join option has to be set and the service definition file has to have Type=notify set.

»Stopping an Agent

An agent can be stopped in two ways: gracefully or forcefully. Servers and Clients both behave differently depending on the leave that is performed. There are two potential states a process can be in after a system signal is sent: left and failed.

To gracefully halt an agent, send the process an interrupt signal (usually Ctrl-C from a terminal, or running kill -INT consul_pid ). For more information on different signals sent by the kill command, see here

When a Client is gracefully exited, the agent first notifies the cluster it intends to leave the cluster. This way, other cluster members notify the cluster that the node has left.

When a Server is gracefully exited, the server will not be marked as left. This is to minimally impact the consensus quorum. Instead, the Server will be marked as failed. To remove a server from the cluster, the force-leave command is used. Using force-leave will put the server instance in a left state so long as the Server agent is not alive.

Alternatively, you can forcibly stop an agent by sending it a kill -KILL consul_pid signal. This will stop any agent immediately. The rest of the cluster will eventually (usually within seconds) detect that the node has died and notify the cluster that the node has failed.

For client agents, the difference between a node failing and a node leaving may not be important for your use case. For example, for a web server and load balancer setup, both result in the same outcome: the web node is removed from the load balancer pool.

The skip_leave_on_interrupt and leave_on_terminate configuration options allow you to adjust this behavior.


Every agent in the Consul cluster goes through a lifecycle. Understanding this lifecycle is useful for building a mental model of an agent's interactions with a cluster and how the cluster treats a node.

When an agent is first started, it does not know about any other node in the cluster. To discover its peers, it must join the cluster. This is done with the join command or by providing the proper configuration to auto-join on start. Once a node joins, this information is gossiped to the entire cluster, meaning all nodes will eventually be aware of each other. If the agent is a server, existing servers will begin replicating to the new node.

In the case of a network failure, some nodes may be unreachable by other nodes. In this case, unreachable nodes are marked as failed. It is impossible to distinguish between a network failure and an agent crash, so both cases are handled the same. Once a node is marked as failed, this information is updated in the service catalog.

When a node leaves, it specifies its intent to do so, and the cluster marks that node as having left. Unlike the failed case, all of the services provided by a node are immediately deregistered. If the agent was a server, replication to it will stop.

To prevent an accumulation of dead nodes (nodes in either failed or left states), Consul will automatically remove dead nodes out of the catalog. This process is called reaping. This is currently done on a configurable interval of 72 hours (changing the reap interval is not recommended due to its consequences during outage situations). Reaping is similar to leaving, causing all associated services to be deregistered.