2014-02-08 00:41:03 +00:00
|
|
|
---
|
|
|
|
layout: "intro"
|
2014-04-14 19:47:55 +00:00
|
|
|
page_title: "Consul Cluster"
|
2014-02-08 00:41:03 +00:00
|
|
|
sidebar_current: "gettingstarted-join"
|
2015-12-27 15:45:59 +00:00
|
|
|
description: >
|
|
|
|
When a Consul agent is started, it begins as an isolated cluster of its own.
|
2016-01-05 17:48:04 +00:00
|
|
|
To learn about other cluster members, the agent must join one or more other
|
2015-12-27 15:45:59 +00:00
|
|
|
nodes using a provided join address. In this step, we will set up a two-node
|
|
|
|
cluster and join the nodes together.
|
2014-02-08 00:41:03 +00:00
|
|
|
---
|
|
|
|
|
2014-04-14 19:47:55 +00:00
|
|
|
# Consul Cluster
|
2014-02-08 00:41:03 +00:00
|
|
|
|
2015-03-14 18:13:41 +00:00
|
|
|
We've started our first agent and registered and queried a service on that
|
|
|
|
agent. This showed how easy it is to use Consul but didn't show how this could
|
|
|
|
be extended to a scalable, production-grade service discovery infrastructure.
|
|
|
|
In this step, we'll create our first real cluster with multiple members.
|
2014-02-08 00:41:03 +00:00
|
|
|
|
2015-03-14 18:13:41 +00:00
|
|
|
When a Consul agent is started, it begins without knowledge of any other node:
|
|
|
|
it is an isolated cluster of one. To learn about other cluster members, the
|
2014-11-26 11:49:39 +00:00
|
|
|
agent must _join_ an existing cluster. To join an existing cluster, it only
|
|
|
|
needs to know about a _single_ existing member. After it joins, the agent will
|
|
|
|
gossip with this member and quickly discover the other members in the cluster.
|
|
|
|
A Consul agent can join any other agent, not just agents in server mode.
|
2014-02-08 00:41:03 +00:00
|
|
|
|
|
|
|
## Starting the Agents
|
|
|
|
|
2015-03-14 18:13:41 +00:00
|
|
|
To simulate a more realistic cluster, we will start a two node cluster via
|
|
|
|
[Vagrant](https://www.vagrantup.com/). The Vagrantfile we will be using can
|
|
|
|
be found in the [demo section of the Consul repo]
|
|
|
|
(https://github.com/hashicorp/consul/tree/master/demo/vagrant-cluster).
|
2014-02-08 00:41:03 +00:00
|
|
|
|
2015-03-14 18:13:41 +00:00
|
|
|
We first boot our two nodes:
|
2014-02-08 00:41:03 +00:00
|
|
|
|
2014-10-19 23:40:10 +00:00
|
|
|
```text
|
2015-03-14 18:13:41 +00:00
|
|
|
$ vagrant up
|
|
|
|
```
|
|
|
|
|
|
|
|
Once the systems are available, we can ssh into them to begin configuration
|
|
|
|
of our cluster. We start by logging in to the first node:
|
|
|
|
|
|
|
|
```text
|
|
|
|
$ vagrant ssh n1
|
|
|
|
```
|
|
|
|
|
2015-12-27 15:45:59 +00:00
|
|
|
In our previous examples, we used the [`-dev`
|
|
|
|
flag](/docs/agent/options.html#_dev) to quickly set up a development server.
|
|
|
|
However, this is not sufficient for use in a clustered environment. We will
|
|
|
|
omit the `-dev` flag from here on, and instead specify our clustering flags as
|
|
|
|
outlined below.
|
|
|
|
|
2015-03-14 18:13:41 +00:00
|
|
|
Each node in a cluster must have a unique name. By default, Consul uses the
|
|
|
|
hostname of the machine, but we'll manually override it using the [`-node`
|
|
|
|
command-line option](/docs/agent/options.html#_node).
|
|
|
|
|
|
|
|
We will also specify a [`bind` address](/docs/agent/options.html#_bind):
|
|
|
|
this is the address that Consul listens on, and it *must* be accessible by
|
|
|
|
all other nodes in the cluster. While a `bind` address is not strictly
|
2016-09-28 16:20:36 +00:00
|
|
|
necessary, it's always best to provide one. Consul will by default attempt to
|
|
|
|
listen on the first private IP on a system, but will fail to start with an
|
|
|
|
error if multiple private IPs are found. Since production servers often
|
|
|
|
have multiple interfaces, specifying a `bind` address assures that you will
|
2015-03-14 18:13:41 +00:00
|
|
|
never bind Consul to the wrong interface.
|
|
|
|
|
|
|
|
The first node will act as our sole server in this cluster, and we indicate
|
2015-12-27 15:45:59 +00:00
|
|
|
this with the [`server` switch](/docs/agent/options.html#_server).
|
|
|
|
|
|
|
|
The [`-bootstrap-expect` flag](/docs/agent/options.html#_bootstrap_expect)
|
|
|
|
hints to the Consul server the number of additional server nodes we are
|
|
|
|
expecting to join. The purpose of this flag is to delay the bootstrapping of
|
|
|
|
the replicated log until the expected number of servers has successfully joined.
|
|
|
|
You can read more about this in the [bootstrapping
|
|
|
|
guide](/docs/guides/bootstrapping.html).
|
|
|
|
|
|
|
|
Finally, we add the [`config-dir` flag](/docs/agent/options.html#_config_dir),
|
|
|
|
marking where service and check definitions can be found.
|
2015-03-14 18:13:41 +00:00
|
|
|
|
|
|
|
All together, these settings yield a
|
|
|
|
[`consul agent`](/docs/commands/agent.html) command like this:
|
|
|
|
|
|
|
|
```text
|
2016-09-27 02:28:39 +00:00
|
|
|
vagrant@n1:~$ consul agent -server -bootstrap-expect=1 \
|
|
|
|
-data-dir=/tmp/consul -node=agent-one -bind=172.20.20.10 \
|
|
|
|
-config-dir=/etc/consul.d
|
2014-02-08 00:41:03 +00:00
|
|
|
...
|
|
|
|
```
|
|
|
|
|
2015-03-14 18:13:41 +00:00
|
|
|
Now, in another terminal, we will connect to the second node:
|
|
|
|
|
|
|
|
```text
|
|
|
|
$ vagrant ssh n2
|
|
|
|
```
|
|
|
|
|
|
|
|
This time, we set the [`bind` address](/docs/agent/options.html#_bind)
|
|
|
|
address to match the IP of the second node as specified in the Vagrantfile
|
|
|
|
and the [`node` name](/docs/agent/options.html#_node) to be `agent-two`.
|
|
|
|
Since this node will not be a Consul server, we don't provide a
|
|
|
|
[`server` switch](/docs/agent/options.html#_server).
|
|
|
|
|
|
|
|
All together, these settings yield a
|
|
|
|
[`consul agent`](/docs/commands/agent.html) command like this:
|
2014-02-08 00:41:03 +00:00
|
|
|
|
2014-10-19 23:40:10 +00:00
|
|
|
```text
|
2016-09-27 02:28:39 +00:00
|
|
|
vagrant@n2:~$ consul agent -data-dir=/tmp/consul -node=agent-two \
|
|
|
|
-bind=172.20.20.11 -config-dir=/etc/consul.d
|
2014-02-08 00:41:03 +00:00
|
|
|
...
|
|
|
|
```
|
|
|
|
|
2015-03-14 18:13:41 +00:00
|
|
|
At this point, you have two Consul agents running: one server and one client.
|
|
|
|
The two Consul agents still don't know anything about each other and are each
|
|
|
|
part of their own single-node clusters. You can verify this by running
|
|
|
|
[`consul members`](/docs/commands/members.html) against each agent and noting
|
|
|
|
that only one member is visible to each agent.
|
2014-02-08 00:41:03 +00:00
|
|
|
|
|
|
|
## Joining a Cluster
|
|
|
|
|
2015-03-14 18:13:41 +00:00
|
|
|
Now, we'll tell the first agent to join the second agent by running
|
|
|
|
the following commands in a new terminal:
|
2014-02-08 00:41:03 +00:00
|
|
|
|
2014-10-19 23:40:10 +00:00
|
|
|
```text
|
2015-03-14 18:13:41 +00:00
|
|
|
$ vagrant ssh n1
|
|
|
|
...
|
|
|
|
vagrant@n1:~$ consul join 172.20.20.11
|
2014-02-08 00:41:03 +00:00
|
|
|
Successfully joined cluster by contacting 1 nodes.
|
|
|
|
```
|
|
|
|
|
|
|
|
You should see some log output in each of the agent logs. If you read
|
|
|
|
carefully, you'll see that they received join information. If you
|
2015-03-14 18:21:54 +00:00
|
|
|
run [`consul members`](/docs/commands/members.html) against each agent,
|
|
|
|
you'll see that both agents now know about each other:
|
2014-02-08 00:41:03 +00:00
|
|
|
|
2014-10-19 23:40:10 +00:00
|
|
|
```text
|
2015-03-14 18:13:41 +00:00
|
|
|
vagrant@n2:~$ consul members
|
|
|
|
Node Address Status Type Build Protocol
|
|
|
|
agent-two 172.20.20.11:8301 alive client 0.5.0 2
|
|
|
|
agent-one 172.20.20.10:8301 alive server 0.5.0 2
|
2014-02-08 00:41:03 +00:00
|
|
|
```
|
|
|
|
|
2014-11-26 11:49:39 +00:00
|
|
|
-> **Remember:** To join a cluster, a Consul agent only needs to
|
2014-02-08 00:41:03 +00:00
|
|
|
learn about <em>one existing member</em>. After joining the cluster, the
|
|
|
|
agents gossip with each other to propagate full membership information.
|
|
|
|
|
2015-03-05 00:12:14 +00:00
|
|
|
## Auto-joining a Cluster on Start
|
2015-03-14 18:13:41 +00:00
|
|
|
Ideally, whenever a new node is brought up in your datacenter, it should
|
|
|
|
automatically join the Consul cluster without human intervention. To
|
|
|
|
accomplish this, you can use
|
|
|
|
[Atlas by HashiCorp](https://atlas.hashicorp.com?utm_source=oss&utm_medium=getting-started-join&utm_campaign=consul)
|
|
|
|
and the [`-atlas-join` flag](/docs/agent/options.html#_atlas_join).
|
|
|
|
An example configuration is shown below:
|
2015-03-05 00:12:14 +00:00
|
|
|
|
|
|
|
```text
|
|
|
|
$ consul agent -atlas-join \
|
|
|
|
-atlas=ATLAS_USERNAME/infrastructure \
|
|
|
|
-atlas-token="YOUR_ATLAS_TOKEN"
|
|
|
|
```
|
2015-03-14 18:13:41 +00:00
|
|
|
To get an Atlas username and token,
|
|
|
|
[create an account](https://atlas.hashicorp.com/account/new?utm_source=oss&utm_medium=getting-started-join&utm_campaign=consul)
|
|
|
|
and replace the respective values in your Consul configuration with your
|
|
|
|
credentials. Now, whenever a new node comes up with a Consul agent, it
|
|
|
|
will automatically join your Consul cluster without any hardcoded
|
|
|
|
configuration.
|
|
|
|
|
2015-03-14 18:21:54 +00:00
|
|
|
Alternatively, you can join a cluster at startup using the
|
2016-01-13 22:44:01 +00:00
|
|
|
[`-join` flag](https://www.consul.io/docs/agent/options.html#_join) or
|
|
|
|
[`start_join` setting](https://www.consul.io/docs/agent/options.html#start_join)
|
2015-03-14 18:21:54 +00:00
|
|
|
with hardcoded addresses of other known Consul agents.
|
2014-02-08 00:41:03 +00:00
|
|
|
|
2014-04-14 19:47:55 +00:00
|
|
|
## Querying Nodes
|
|
|
|
|
|
|
|
Just like querying services, Consul has an API for querying the
|
|
|
|
nodes themselves. You can do this via the DNS or HTTP API.
|
|
|
|
|
|
|
|
For the DNS API, the structure of the names is `NAME.node.consul` or
|
2015-03-05 23:04:40 +00:00
|
|
|
`NAME.node.DATACENTER.consul`. If the datacenter is omitted, Consul
|
2014-04-14 19:47:55 +00:00
|
|
|
will only search the local datacenter.
|
|
|
|
|
2015-03-14 18:13:41 +00:00
|
|
|
For example, from "agent-one", we can query for the address of the
|
|
|
|
node "agent-two":
|
2014-04-14 19:47:55 +00:00
|
|
|
|
|
|
|
```
|
2015-03-14 18:13:41 +00:00
|
|
|
vagrant@n1:~$ dig @127.0.0.1 -p 8600 agent-two.node.consul
|
2014-04-14 19:47:55 +00:00
|
|
|
...
|
|
|
|
|
|
|
|
;; QUESTION SECTION:
|
|
|
|
;agent-two.node.consul. IN A
|
|
|
|
|
|
|
|
;; ANSWER SECTION:
|
|
|
|
agent-two.node.consul. 0 IN A 172.20.20.11
|
|
|
|
```
|
|
|
|
|
|
|
|
The ability to look up nodes in addition to services is incredibly
|
|
|
|
useful for system administration tasks. For example, knowing the address
|
2015-03-14 18:13:41 +00:00
|
|
|
of the node to SSH into is as easy as making the node a part of the
|
|
|
|
Consul cluster and querying it.
|
2014-04-14 19:47:55 +00:00
|
|
|
|
2014-02-08 00:41:03 +00:00
|
|
|
## Leaving a Cluster
|
|
|
|
|
|
|
|
To leave the cluster, you can either gracefully quit an agent (using
|
|
|
|
`Ctrl-C`) or force kill one of the agents. Gracefully leaving allows
|
2015-03-14 18:13:41 +00:00
|
|
|
the node to transition into the _left_ state; otherwise, other nodes
|
2014-02-08 00:41:03 +00:00
|
|
|
will detect it as having _failed_. The difference is covered
|
2015-03-14 18:13:41 +00:00
|
|
|
in more detail [here](/intro/getting-started/agent.html#stopping).
|
|
|
|
|
|
|
|
## Next Steps
|
|
|
|
|
|
|
|
We now have a multi-node Consul cluster up and running. Let's make
|
|
|
|
our services more robust by giving them [health checks](checks.html)!
|