To deploy our MongoDB instances, we choose to use the official Docker images: MongoDB 3.2
By inspecting into the image, we can easily find that the entry point is: /entrypoint.sh.
We need a data volume for better data persistency:
docker run --name mongo1-data -v /data/db busybox
To start a simple MongoDB instance, which doesn't support replica set and has no authentication:
docker run -it --rm --volume-from mongo1-data -p 27017:27017 mongo:latest
MongoDB uses key files for inner communication within shards and replica set clusters. Therefore, we need to generate a key file for our replica set:
openssl rand -base64 258 > /root/mongo_key
Open another terminal, connect to the Mongo Shell, and then add a root user to the admin database.
Stop the previous temporary MongoDB container, and start the formal one with full authentication and replica set:
docker run -d --name mongo1 -p 27017:27017 -v /root/mongo_key:/mongo_key mongo --repliSet default --auth --keyFile /mongo_key
Connect to the database, use rs.initiate() to initialize the node to be primary. However, there is one more step. By default, a Docker container will get a random string at creation, which will be used as its host name. This random host name will later be used as the host in rs configuration, but other containers cannot use such host name for access. So we must reconfigure it to real host names:
> var c = rs.config()
> c.members[0].host = "172.17.0.1:27017"
> rs.reconfig(c)
After setting up the primary node, we can start more MongoDB nodes and add them one by one into the replica set:
docker run -d --name mongo2 -p 27018:27017 -v /root/mongo_key:/mongo_key mongo --repliSet default --auth --keyFile /mongo_key
> rs.add("172.17.0.1:27018")