1---
2layout: "docs"
3page_title: "Agent"
4sidebar_current: "docs-agent-running"
5description: |-
6  The Serf agent is the core process of Serf. The agent maintains membership information, propagates events, invokes event handlers, detects failures, and more. The agent must run on every node that is part of a Serf cluster.
7---
8
9# Serf Agent
10
11The Serf agent is the core process of Serf. The agent maintains membership
12information, propagates events, invokes event handlers, detects failures,
13and more. The agent must run on every node that is part of a Serf cluster.
14
15## Running an Agent
16
17The agent is started with the `serf agent` command. This command blocks,
18running forever or until told to quit. The agent command takes a variety
19of configuration options but the defaults are usually good enough. When
20running `serf agent`, you should see output similar to that below:
21
22```
23$ serf agent
24==> Starting Serf agent...
25==> Serf agent running!
26    Node name: 'mitchellh.local'
27    Bind addr: '0.0.0.0:7946'
28     RPC addr: '127.0.0.1:7373'
29    Encrypted: false
30     Snapshot: false
31      Profile: lan
32
33==> Log data will now stream in as it occurs:
34
352013/10/22 10:35:33 [INFO] Serf agent starting
362013/10/22 10:35:33 [INFO] serf: EventMemberJoin: mitchellh.local 127.0.0.1
372013/10/22 10:35:33 [INFO] Serf agent started
382013/10/22 10:35:33 [INFO] agent: Received event: member-join
39...
40```
41
42There are six important components that `serf agent` outputs:
43
44* **Node name**: This is a unique name for the agent. By default this
45  is the hostname of the machine, but you may customize it to whatever
46  you'd like using the `-node` flag.
47
48* **Bind addr**: This is the address and port used for communication between
49  Serf agents in a cluster. Every Serf agent in a cluster does not have to
50  use the same port.
51
52* **RPC addr**: This is the address and port used for RPC communications
53  for other `serf` commands. Other Serf commands such as `serf members`
54  connect to a running agent and use RPC to query and control the agent.
55  By default, this binds only to localhost on the default port. If you
56  change this address, you'll have to specify an `-rpc-addr` to commands
57  such as `serf members` so they know how to talk to the agent. This is also
58  the address other applications can use over [RPC to control Serf](/docs/agent/rpc.html).
59
60* **Encrypted**: This shows if Serf is encrypting all traffic that it
61  sends and expects to receive. It is a good sanity check to avoid sending
62  non-encrypted traffic over any public networks. You can read more about
63  [encryption here](/docs/agent/encryption.html).
64
65* **Snapshot**: This shows if Serf snapshotting is enabled. The snapshot
66  file enables Serf to automatically re-join a cluster after failure and
67  prevents replay of events that have already been seen. It requires storing
68  state on disk, and [must be configured](/docs/agent/options.html)
69  using a CLI flag or in the configuration directory. If it is not provided,
70  other nodes will still attempt to reconnect on recovery, however the node
71  will take longer to join the cluster and will replay old events.
72
73* **Profile**: The profile controls various timing values which should
74  be appropriate to the environment Serf is running in. It defaults to
75  optimizing for a LAN environment, but can also be set for WAN or
76  local-only communication. The profile can be set in
77  the [configuration](/docs/agent/options.html).
78
79## Stopping an Agent
80
81An agent can be stopped in two ways: gracefully or forcefully. To gracefully
82halt an agent, send the process an interrupt signal, which is usually
83`Ctrl-C` from a terminal. When gracefully exiting, the agent first notifies
84the cluster it intends to leave the cluster. This way, other cluster members
85notify the cluster that the node has _left_.
86
87Alternatively, you can force kill the agent by sending it a kill signal.
88When force killed, the agent ends immediately. The rest of the cluster will
89eventually (usually within seconds) detect that the node has died and will
90notify the cluster that the node has _failed_.
91
92The difference between a node _failing_ and a node _leaving_ may not be
93important for your use case. For example, for a web server and load
94balancer setup, both result in the same action: remove the web node
95from the load balancer pool. But for other situations, you may handle
96each scenario differently.
97