To help the users of this exercise to learn how to use docker and where to get help when they are stuck in the real world, I have tried to create exercises that forces the user to look in documentation and experiment.
This means, that the exercise set is not a simple copy & paste exercise
- About this exercise set
- Table of content
- Setup a Docker Swarm Cluster
- Deploy your first service
- Create a complete Docker Swarm stack
- Scaling the frontend service
- Network security
- Playing with node failure
- Specify placement for the services
- Setting limits for the application containers
- Log aggregation
- References
In this part we use Play With Docker as a free provider of Alpine Linux Virtual MachineS in the cloud. These can be used to build and run Docker containers and to play around with clusters in Swarm Mode.
This can be found at: http://play-with-docker.com/
If you are not at at all able to use vim, which is the only editor available on play with docker, you can install another.
$ apk update
$ apk add nano
Or install the docker-machine driver for Play With Docker, to be able to edit files locally on your own machine.
Start by creating three machines using the "+Add New Instance" button.
Chose a machine to be the manager and initiate the Docker Swarm cluster on this machine.
$ docker swarm init --advertise-addr eth0
In the output of this command, we get the command that can be used to join workers to the cluster
$ docker swarm join --token SWMTKN-1-2ypam... {ip}:2377
Run this command on the two other machines. Now we can validate that we have a cluster by going to the manager note and run the following command.
$ docker node ls
ID HOSTNAME STATUS AVAILABILITY MANAGER STATUS
3kkk6vvth1gu5hi3v93q2har3 * node1 Ready Active Leader
207qifbk21o6h05afa4ypabk3 node2 Ready Active
dne037y5lvndl37kccj2028dk node3 Ready Active
To be able to see whats running on our cluster in a graphical interface we use the docker services called Visualizer.
Create a service stack with the visualizer container.
docker service create --name=viz \
-p=8081:8080 \
--constraint=node.role==manager \
--mount=type=bind,src=/var/run/docker.sock,dst=/var/run/docker.sock \
dockersamples/visualizer:stable
Check the new endpoint on the manager node (port 8081) to see our cluster and the visualizer container.
Delete the service we created above.
$ docker service ls
ID NAME MODE REPLICAS IMAGE PORTS
rjvz7l1tzoal viz replicated 1/1 dockersamples/visualizer:stable *:8081->8080/tcp
$ docker service rm viz
Change the command we just ran to a docker-compose file with the name docker-visualize.yml. Looking something like this: (Note that the volume mapping does not have quite the same format in the command line as it does in the docker-compose file)
version: "3"
services:
visualizer:
image: {that image we want to use}
ports:
- {port stuff}
volumes:
- {volume mapping}
deploy:
placement:
constraints:
- {constraints}
Redeploy the visualizer with the yaml file, and validate that it works!
$ docker stack deploy --compose-file docker-visualize.yml viz
[Note] This command is used BOTH for deploying a stack first time and for updating the desired state of the stack.
We will now look at an example application called VotingApp from one the Docker Labs.
- A Python webapp which lets you vote between two options
- A Redis queue which collects new votes
- A .NET worker which consumes votes and stores them in...
- A Postgres database backed by a Docker volume
- A Node.js webapp which shows the results of the voting in real time
We can deploy the stack with this description.
version: "3"
services:
redis:
image: redis:alpine
ports:
- "6379"
db:
image: postgres:9.4
volumes:
- db-data:/var/lib/postgresql/data
vote:
image: dockersamples/examplevotingapp_vote:before
ports:
- 5000:80
depends_on:
- redis
result:
image: dockersamples/examplevotingapp_result:before
ports:
- 5001:80
depends_on:
- db
worker:
image: dockersamples/examplevotingapp_worker
depends_on:
- db
- redis
volumes:
db-data:
Start by creating a yml file with the data above and deploy it to our cluster as in the last exercise. Check that you can see all five containers in the visualizer. You should also check that you can see the two frontends at port 5000 and 5001. Try voting.
News has spread like wildfire about our new awesome voting app and we need to have more services available for voting.
Call the docker service command line with a --help param to figure out how to scale the vote service to 5 instances.
$ docker service --help
Once you have scaled the service:
- Check that you can see all five containers in the visualizer.
- Check that the vote page is now handled by different containers.
- The container handling a request is shown on the vote page.
Update the yml file with the new deployment info, stating that we want to have 5 replicas of the vote service. Docker Compose deploy documentation
The next step in making our voting application more production ready is to make sure that the services only has access to what they need.
More specifically, we want the services in the voting application to be on different networks. So that the frontend application vote only has access to the redis cache and not to the postgres database or the result application.
We also need the worker service to have access to both networks, because it is the one responsible for moving votes from the redis cache to the postgres database.
| Frontend | Backend |
|---|---|
| vote | worker |
| redis | db |
| worker | result |
Create the two networks in the yml file and specify the networks on the services. Note that this is something that docker has changed a lot over time. Newest documentation is here: Docker compose endpoint mode
To simulate a failing node (bloody hardware failing all the time) try deleting one of the worker nodes in the Play With Docker interface. When doing this keep an eye on the visualizer and see how all the containers are redeployed on the remaining node.
Try creating two new nodes and joining them to the cluster. By default docker swarm will not move running containers to the new nodes. Try scaling up the vote or worker services and see how the containers are placed.
Next task on the production list is to make sure that our result database are only running on machines that have ssd. Add a label called disk to all our worker nodes, with disk=hdd for two of them and disk=ssd on the last worker See documentation for adding labels.
Update the yml file with conditions on the db service that specify that it can only be deployed on disk=ssd. Validate by looking in the visualizer.
You can also change the other services to make sure that only the database runs on machines with ssd.
Try setting the manager as drained so no containers are scheduled on the manager.
The next important thing to make our new awesome voting application more production ready is to limit the cpu of all the services. This will help prevent a situation where on single service goes crazy and consumes all the resources on the host machine. An example of setting the limits from command line could look like this:
docker service deploy \
--limit-cpu 0.25 \
--reserve-cpu 0.1 \
webapp
The reserve-cpu parameter tells docker how much it can pack the container on a host, so that each container will always have at least the reserved cpu amount available.
Update the yml file with conditions on all services that specify a cpu limit and a reserved cpu amount.
You can also look at limiting the services on memory.
When we run many services on may machines, it becomes very important to have centralized logging. You should try to setup central logging of your Docker Swarm and send the logs from the different services to that service.
There are plenty applications for central logging
- Splunk
- Humio
- Logstash, ElasticSearch and Kibana
- PaperTrail
- Graylog2
- ...
For all containers that logs to standard out docker can collect the logs for us.
Create a free account at PaperTrails and setup a logsprout container on each Docker Swarm node to send the data to PaperTrails
Remember that Docker Swarm has the notion of mode for a service that can either be replicated or global.
https://help.papertrailapp.com/kb/configuration/configuring-centralized-logging-from-docker/
Because the logsprout container gets all the logs from the Docker socket, and not by actually talking to any of the other containers, it does not need to know anything about them, or be able to send them requests.
Create a new .yml stack file with the logsprout container in and deploy the stack to the cluster.