docker-swarm-tutorial

In Docker 1.12, a built-in swarm mode was introduced. This is a tutorial to walk through the basic features on a local cluster of hosts.

Prerequisites

• Latest Docker Toolbox that supports Docker >1.12
  • Note: If you have the new native Docker applications for windows or mac, check out this link to learn how to get the native apps to play nice with Toolbox: https://docs.docker.com/docker-for-mac/docker-toolbox/#/docker-toolbox-and-docker-for-mac-coexistence. I assume instructions are similar for Windows... but no guarantees. If in doubt, uninstall the native apps for this tutorial.
• Internet access

Setup machines

We are going to be following the Docker swarm "Getting Started" tutorial. https://docs.docker.com/engine/swarm/swarm-tutorial/. The prerequisite for the tutorial is a group of 3 machines that can communicate on ports 2377, 7946, and 4789. Unfortunately, they don't give any guidance on setting those up... so let's do that first.

In my opinion, the easiest way to setup these machines is with docker-machine, which was installed when you installed Docker Toolbox.

The three machines we want to create:

• manager1
• worker1
• worker2

To create our 3 machines...

On mac/linux

docker-machine create --driver virtualbox manager1
docker-machine create --driver virtualbox worker1
docker-machine create --driver virtualbox worker2

On windows

For windows machines, if you have Hyper-V installed, you'll need to create a virtual switch and use the hyperv driver. See https://github.com/SamuelDebruyn/chipsncookies-site/blob/master/content/post/run-docker-on-hyper-v-with-docker-machine.md

Once you have Hyper-V configured and your new switch configured (I named mine docker-switch), the commands you'll run should look like this:

docker-machine create -d hyperv --hyperv-virtual-switch docker-switch manager1
docker-machine create -d hyperv --hyperv-virtual-switch docker-switch worker1
docker-machine create -d hyperv --hyperv-virtual-switch docker-switch worker2

After you have created your machines, you'll want to set your Docker client to connect to manager1. To find out how to do that run:

docker-machine env manager1

At the bottom of the output, there should be instructions on what to do next. On Mac/Linux it should say to do the following:

eval $(docker-machine env manager1)

Continue Docker Tutorial

Now you can continue the tutorial from here: https://docs.docker.com/engine/swarm/swarm-tutorial/create-swarm, but with one key difference. Whenever the tutorial asks you to ssh to a machine, use the following command to ssh to the machine:

docker-machine ssh <machine-name>

When you need the ip address of your manager machine, run the following:

docker-machine ip manager1

Tutorial with Rancher

Setup/Prerequisites

We are going to use Rancher to visualize what's happening on our machines. http://rancher.com

• Start rancher on manager1

eval $(docker-machine env manager1)
docker run -d --restart=always -p 8080:8080 rancher/server

• Wait a few mins and then navigate to http://:8080.
• Add all 3 machines as hosts in rancher following these directions: http://docs.rancher.com/rancher/latest/en/hosts/custom/

You are now ready to try out some various components of Docker 1.12 and the changes will be visualized in Rancher

• Create the swarm
  • Create swarm manager

docker swarm init --advertise-addr $(docker-machine ip manager1)

• Add worker nodes to swarm. After running the above command, the output will have commands that will need to be run on each worker node.

Creating a network

Not new in Docker 1.12, but will need this for some future steps.

• Create an overlay network. From Docker docs: "You can create an overlay network on a manager node running in swarm mode without an external key-value store. The swarm makes the overlay network available only to nodes in the swarm that require it for a service. When you create a service that uses the overlay network, the manager node automatically extends the overlay network to nodes that run service tasks." See https://docs.docker.com/engine/userguide/networking/#/an-overlay-network-with-docker-engine-swarm-mode for more details.

docker network create --driver overlay my_network

Create, scale, update, remove a service

• Create

docker service create --name my_web --replicas 1 --publish 8181:80 nginx:1.10

Verify you can hit the service from any of the machines in your cluster at port 8181.

• Scale

docker service scale my_web=5

• Update to latest nginx

docker service update --image nginx:1.11 my_web

• Revert back to 1.10 nginx, 2 servers at a time and with a 10s delay

docker service update --image nginx:1.10 --update-delay 10s --update-parallelism 2 my_web

• Drain a node to take it out of service

docker node update --availability drain worker1

• Make the node leave the swarm

docker-machine ssh worker1
docker swarm leave

• Drain another node to take it out of service

docker node update --availability drain worker2

• Change your mind

docker node update --availability active worker2

• Remove the service

docker service rm my_web

Try out overlay networks and routing mesh

• Create overlay network if you didn't above

docker network create --driver overlay my_network

• Create service that will be pinged

docker service create --network my_network --name my_web nginx

• Try to ping from service that is not on some network

docker service create --name pinger alpine ping my_web

What happens?

• Let's make that give up at some point

docker service create --name pinger --restart-max-attempts 5 alpine ping my_web

• Try to ping from service that is on the network

docker service create --name pinger --network my_network alpine ping my_web

• Let's confirm this with wget Create an nginx service

docker service create --replicas 3 --name my_web --network my_network nginx

Run a box we can wget from

docker service create --name my-busybox --network my_network busybox sleep 30000

Find where busy box started:

docker service ps my-busybox

Open interactive shell in busybox TODO Run nslookup on my_web

nslookup my-web

Run wget

wget -O- my-web

• Run service in DNSRR mode instead of VIP mode

docker service rm my_web

docker service create --replicas 3 --name my_web --network my_network --endpoint-mode dnsrr nginx

Now run nslookup from busybox again.

Global services

• Run something "globally" meaning one on every machine

docker service create --name logging --mode global alpine ping microsoft.com

Labels

Other Docker 1.12 goodness

Health Check instruction

https://github.com/tomwillfixit/healthcheck