The Vagrantfile can be used to spin few nodes with Docker Swarm and Docker API configured. The boxes are running Ubuntu 16.04(Xenial) amd64.
Vagrant and VirtualBox need to be installed on your workstation.
By default vagrant up spins up 3 machines: manager, worker1, worker2. You can adjust how many workers you want in the Vagrantfile, by setting the numworkers variable. Manager, by default, has address "192.168.10.2", workers have consecutive ips.
numworkers = 2If your provisioner is Virtualbox, you can modify the vm allocations for memory and cpu by changing these variables:
vmmemory = 512numcpu = 1/etc/hosts on every machine is populated with an IP address and a name of every other machine, so that names are resolved within the cluster. This mechanism is not idempotent, reprovisioning will append the hosts again.
By default, vagrant will create a docker swarm cluster. You can run
AUTO_START_SWARM=false vagrant up to provision individual docker hosts
Go to the master node and start docker swarm:
(host)# vagrant ssh manager
(manager)# docker swarm init --advertise-addr 192.168.10.2
docker swarm join \
--token SWMTKN-1-59h28hcbb8gzs2xs24oyh7hjvc7fp8skjzvnpw9cksmp96m4y2-35er9ai3u1f1ae5esb7x8l1hx \
192.168.10.2:2377After starting swarm, you can use my testing Docker image to play with. It is called darek/goweb and is a super simple Web app, displaying the hostname, and a version. They can be used to play with swarm rolling update feature. The container exposes port 8080.
Now join the swarm on the nodes with the command from the manager, do it on both nodes. You can verify that nodes are in the cluster by doing docker nodes ls on the manager.
vagrant@manager:~$ docker node ls
ID HOSTNAME STATUS AVAILABILITY MANAGER STATUS
0rtuyz07e0wazmxvoed1llmx3 * manager Ready Active Leader
1iab1w9m5znyzmh7hpzyn7rrw worker2 Ready Active
4mmca5rxrxxhb0tb7s5du5hpe worker1 Ready ActiveNow create a web service, with 1 replica, on any of the nodes:
docker service create --name web --replicas 3 --publish 8080:8080 darek/goweb:1.0You can see the status od the service with docker service ps web.
vagrant@manager:~$ docker service ps web
ID NAME IMAGE NODE DESIRED STATE CURRENT STATE ERROR
7m0s3ikyjyvps4xgpk655uv9k web.1 darek/goweb:1.0 manager Running Running 16 seconds ago
You can scale up the service if you want:
docker service scale web=6Now, let us verify it works, docker swarm will load balance requests to all of the service instances:
vagrant ssh worker1
sudo apt-get install httpie -y
http localhost:8080Do the http command serveral times, you will notice different hostnames every time. This is because swarm handler load balancing for us.
Let's do a rolling update of the service, first let's scale it to 10 - you already know how. First we will update the defintion of the service to say that updates will have a delay of 5s.
docker service update --update-delay 5s webLet's us update the service now:
docker service update --image darek/goweb:2.0 webNow docker will update service one by one with a 5s delay between the updates. If you want to introduce parallelism to updates you can specify if with the --update-parallelism flag.
docker service update --update-parallelism 2 webLet's see how it goes:
vagrant@manager:~$ docker service ps web
ID NAME IMAGE NODE DESIRED STATE CURRENT STATE ERROR
7m0s3ikyjyvps4xgpk655uv9k web.1 darek/goweb:1.0 manager Running Running 8 minutes ago
f49k99apn8yzve0uh8zsyigw6 web.2 darek/goweb:1.0 worker1 Running Running 7 minutes ago
8t9x9qqqxztzd8msuccbx43mg web.3 darek/goweb:2.0 manager Running Running 11 seconds ago
38pl9czfhaztifjsb605wi2m9 \_ web.3 darek/goweb:1.0 worker1 Shutdown Shutdown 16 seconds ago
5y4iaxmv1jhdw5b7qroessmr9 web.4 darek/goweb:1.0 worker2 Running Running 7 minutes ago
2zhfna5m15cd482hbt06kc139 web.5 darek/goweb:2.0 worker1 Running Preparing 1 seconds ago
3ydym0s1iq0psko60zi66ojvs \_ web.5 darek/goweb:1.0 manager Shutdown Shutdown less than a second ago
9feao0z26oc75qgoktuc16pro web.6 darek/goweb:1.0 manager Running Running about a minute ago
14tsult13blurmrh2klf7sm7i web.7 darek/goweb:2.0 worker2 Running Running 43 seconds ago
1qtcsaohat94puoym9i6zsadf \_ web.7 darek/goweb:1.0 manager Shutdown Shutdown 48 seconds ago
8d2qnv19urq2yzlmqzud41ffa web.8 darek/goweb:1.0 worker2 Running Running about a minute ago
9shytfqyotuuyaj5hiddnymd4 web.9 darek/goweb:1.0 worker2 Running Running about a minute ago
9kr6ehtv093cnmdxo55sy4pl9 web.10 darek/goweb:2.0 worker1 Running Running 27 seconds ago
1yvwgb25d8fbarrabyeqha843 \_ web.10 darek/goweb:1.0 worker1 Shutdown Shutdown 31 seconds agoSome services are already updated. Nice.
You can also inspect the service by docker service inspect --pretty web.
vagrant@manager:~$ docker service inspect --pretty web
ID: 0iqjzgky9p7ilky60xff4ovgt
Name: web
Mode: Replicated
Replicas: 10
Update status:
State: completed
Started: 3 minutes ago
Completed: 48 seconds ago
Message: update completed
Placement:
UpdateConfig:
Parallelism: 1
Delay: 10s
On failure: pause
ContainerSpec:
Image: darek/goweb:2.0
Resources:
Ports:
Protocol = tcp
TargetPort = 8080
PublishedPort = 8080After the update, you can verify that the version is really 2.0 with the httpie command. Let's now delete the service.
docker service rm webList Docker containers
curl http://192.168.10.2:4243/containers/json | jq .Docker Info
curl http://192.168.10.2:4243/info | jq .Destroy the vm's with: vagrant destroy --force.
Inspired by: https://github.com/tdi/vagrant-docker-swarm