This repository provides an easy way to deploy etcd as Docker containers or as standalone processes.
Docker deployment with docker-compose
The provided docker-compose.sample.yml file demonstrates how to quickly
deply a 3-node cluster. This sample will run as configured. If you're running
in a Rancher environment, take advantage of host
affinity. Just edit the values in the affinity labels to match your
environment.
To start a 3-node sample cluster:
docker-compose -f docker-compose.sample.yml up -d
Note: This sample is meant for development or demonstration purposes only, as it does not build a robust cluster across multiple Docker hosts. Using affinity labels, you can very easily create a robust cluster.
All data is stored in /data. ETCD_DATA_DIR=/data is set when the image is built.
You'll want to map a persistent volume to /data.
To deploy the first etcd node:
docker run -d --name infra0 -p 2379:2379 -p 2380:2380 -e ETCD_NAME=infra0 -v /data aptplatforms/easy-etcd
Alternatively, you can deploy the node without mapped ports, so then the service can only be accessed from inside the Docker network:
docker run -d --name infra0 -e ETCD_NAME=infra0 -v /data aptplatforms/easy-etcd
Find the internal IP address of the first etcd node.
docker inspect --format='{{range .NetworkSettings.Networks}}{{.IPAddress}}{{end}}' infra0
> 172.17.0.2
Deploy second and subsequent nodes as follows. The only difference is the
presence of the CLUSTER_LEAD variable.
docker run -d --name infra1 -p 2379:2379 -p 2380:2380 -e CLUSTER_LEAD=172.17.0.2 -v /data aptplatforms/easy-etcd
You repeat the above command to add as many more nodes as you want!
Remember that the whole point of having a cluster is to avoid failure, so make sure you use multiple Docker hosts. Having all etcd nodes in the same Docker host is not going to help if the host goes down.
The boot.sh script can be used to deploy etcd as a regular process. The
etcd service binaries must be installed before using this script.
To deploy the first etcd node:
ETCD_NAME=infra0 ./boot.sh &
Assuming the FIRST_ETCD_IP_ADDRESS variable is set to the IP address of the
first etcd node, you can deploy a second node as follows:
CLUSTER_LEAD=$FIRST_ETCD_IP_ADDRESS ETCD_NAME=infra1 ./boot.sh &
The script runs the service on the default 2379 and 2380 ports. If you want to create more than one node in a host, you need to specify different ports for the additional nodes:
CLUSTER_LEAD=$FIRST_ETCD_IP_ADDRESS CLIENT_PORT=2479 PEER_PORT=2480 ./boot.sh &
You repeat the above command to add as many more nodes as you want!
The boot script accepts a number of options through environment
variables. Below is a complete list. If you want to tweak etcd further, see etcd Configuration flags.
| Variable | Value |
|---|---|
ETCD_NAME |
The name to give to the etcd node. You'll thank yourself later for setting a reasonable ETCD_NAME on each node especially if you're running in Docker. Cluster data includes this name and gets very unfriendly if you teardown an instance and rebuild it with the same ETCD_DATADIR. Default: $(hostname). |
ETCD_INITIAL_CLUSTER_TOKEN |
The name of the cluster. Default: etcd-cluster-token. |
CLUSTER_LEAD |
The IP address of an already running node in the cluster. If this variable is set, the new node joins the existing cluster. If this variable is not set, the node starts a new cluster. Default: <empty>. |
LISTEN_IP_ADDRESS |
The IP address where etcd listens for requests. Default: 0.0.0.0. |
CLIENT_PORT |
The port number where etcd should listen to client requests. Default: 2379. |
PEER_PORT |
The port number where etcd should listen to requests from peers. Default: 2380. |
ETCD_DATADIR |
The directory to store persistent data. Default: /data. |
Thanks to Miguel Grinberg for his easy-etcd. This is my modification of his work. Mainly I was trying to learn more about etcd itself by making changes to things that I didn't understand without DoingItMyself while at the same time adding notes about Rancher and docker-compose deployment.