Container networking on multiple Docker hosts

Prerequisite

The multihost networking is available out the box with libnetwork since Docker 1.9

This required to setup a key-value store first

several are supported: etcd / consul / zookeeper
keeps all the information regarding (networks / subnetworks, IP addresses of Docker hosts / containers, …)

Creation of a key-value store

Several steps are needed to run the key value store

Create dedicated Docker host with Machine docker-machine create -d virtualbox consul
Switch to the context of the newly created host eval "$(docker-machine env consul)"
Run container based on progirum/consul image docker run -d -p "8500:8500" -h "consul" progrium/consul -server -bootstrap

Creation of Docker hosts that will run application containers

As for consul, we use Docker Machine to create 2 test Docker hosts

Host 1

$ docker-machine create \
-d virtualbox \
--engine-opt="cluster-store=consul://$(docker-machine ip consul):8500" \
--engine-opt="cluster-advertise=eth1:2376" \
host1

$ docker $(docker-machine config host1) network ls
NETWORK ID          NAME                DRIVER
14753b15c63e        bridge              bridge
2cc7d35a48e3        none                null
ad05eeca763a        host                host

Host 2

$ docker-machine create \
-d virtualbox \
--engine-opt="cluster-store=consul://$(docker-machine ip consul):8500" \
--engine-opt="cluster-advertise=eth1:2376" \
host2

$ docker $(docker-machine config host2) network ls
NETWORK ID          NAME                DRIVER
b7765c98adbf        bridge              bridge
48244d2fca3b        none                null
36a3858b68c8        host                host

As we’ve seen in a previous chapter, 3 default networks are available on each host: bridge / none / host. We will create an overlay user defined network and benefit from the embedded DNS name server that enables container communication across nodes.

Creation of an overlay network

As seen befire, a user defined network can easily be created. Let’s create an overlay network, named appnet, from host1.

docker $(docker-machine config host1) network create -d overlay appnet

This network is also visible from host2 as we can see below.

$ docker $(docker-machine config host1) network ls
NETWORK ID          NAME                DRIVER
acd47b4c062d        appnet              overlay
14753b15c63e        bridge              bridge
2cc7d35a48e3        none                null
ad05eeca763a        host                host

$ docker $(docker-machine config host2) network ls
NETWORK ID          NAME                DRIVER
acd47b4c062d        appnet              overlay
b7765c98adbf        bridge              bridge
48244d2fca3b        none                null
36a3858b68c8        host                host

Check cross host communication

Run the mongo container, based on mongo 3.2 offical image, on appnet network from host1

docker $(docker-machine config host1) run -d --name mongo --net=appnet mongo:3.2

Run the box container, based on busybox) on appnet network from host2

docker $(docker-machine config host2) run -ti --name box --net=appnet busybox sh

Even if box and mongo do not run on the same host, box can communicate with mongo container using its name through the DNS name server embedded in Docker 1.10+

/ # ping mongo
PING mongo (10.0.0.2): 56 data bytes
64 bytes from 10.0.0.2: seq=0 ttl=64 time=0.553 ms
…
/ # ping mongo.appnet
PING mongo.appnet (10.0.0.2): 56 data bytes
64 bytes from 10.0.0.2: seq=0 ttl=64 time=0.474 ms
…