In this lab you will deploy an application on Docker Enterprise Edition that takes advantage of some of the latest features of Docker Universal Control Plane (UCP). UCP is an included component of Docker EE Standard and Docker EE Advanced. The tutorial will lead you through building a compose file that can deploy a full application on UCP in one click. Capabilities that you will use in this application deployment include:
- Docker services
- Application scaling and failure mitigation
- Layer 7 load balancing
- Overlay networking
- Application secrets
- Application health checks
Difficulty: Intermediate
Time: Approximately 45 - 60 minutes
Tasks:
When you encounter a phrase in between < and > you are meant to substitute in a different value.
This lab is best done on three separate nodes, though it can be done with a single one. The requirements are as follows:
- 3 Linux hosts on a version and distribution that support Docker
- Docker Engine 17.03 EE (or higher) installed on each node
- Open network access between the nodes
- Ports
443and30000-30005open on the lab nodes externally to the lab user
Docker EE needs to be installed on each node. Docker EE installation instructions can be found here. The nodes can be in a cloud provider environment or on locally hosted VMs.
For the remainder of this lab the three nodes will be referred to as node0, node1, and node2.
Your nodes may be using private IP addresses along with mapped public IP addresses or the lab environment may be entirely private. We will be referring to these IPs as node0-private-ip, node0-public-ip, or node0-public-dns. If you are using only private IPs then you will replace the public IP/DNS with the private equivalent.
The following task will guide you through how to create a UCP cluster on your hosts.
- Log in to
node0of the three nodes you have been given for this lab. You may be prompted whether you want to continue. Answeryesand then enter the password.
$ ssh [email protected]
The authenticity of host 'node0-smwqii1akqh.southcentralus.cloudapp.azure.com (13.65.212.221)' can't be established.
ECDSA key fingerprint is SHA256:BKHHGwzrRx/zIuO7zwvyq5boa/5o2cZD9OTlOlOWJvY.
Are you sure you want to continue connecting (yes/no)? yes
Warning: Permanently added 'node0-smwqii1akqh.southcentralus.cloudapp.azure.com,13.65.212.221' (ECDSA) to the list of known hosts.
[email protected]'s password:
Welcome to Ubuntu 16.04.2 LTS (GNU/Linux 4.4.0-72-generic x86_64)
- Check to make sure you are running the correct Docker version. At a minimum you should be running
17.03 EE
$ docker version
Client:
Version: 17.03.0-ee-1
API version: 1.26
Go version: go1.7.5
Git commit: 9094a76
Built: Wed Mar 1 01:20:54 2017
OS/Arch: linux/amd64
Server:
Version: 17.03.0-ee-1
API version: 1.26 (minimum version 1.12)
Go version: go1.7.5
Git commit: 9094a76
Built: Wed Mar 1 01:20:54 2017
OS/Arch: linux/amd64
Experimental: false
- Run the UCP installer to install the UCP Controller node.
You will have to supply the following values to the install command:
--ucp-password- This can be a password of your choosing--san- This should be the public IP ofnode0
You may want to open up a text editor to enter in your ucp password and --san value.
docker run --rm -it --name ucp \
-v /var/run/docker.sock:/var/run/docker.sock \
docker/ucp:2.1.3 install \
--debug \
--admin-username admin \
--admin-password <your-password> \
--san <node0-public-dns> \
--host-address <node0-private-ip>
This is an example of what your final install command might look like ...
docker run --rm -it --name ucp \
-v /var/run/docker.sock:/var/run/docker.sock \
docker/ucp:2.1.3 install \
--admin-username admin \
--admin-password <your-password> \
--san <node0-public-dns> \
--host-address <node0-private-ip>
It will take up to 30 seconds to install.
- Log in to UCP by going to
https://<node0-public-dns>in your browswer
Depending on what browser you are using, you will receive a warning about the connection. Proceed through to the UCP URL. The warning is occurring because we UCP uses privately signed certificates by default. In a production installation we would add our own certificates that would be trusted by our browser.
Log in as the user admin with the password that you supplied in step 3. You will be asked to upload a license. Skip this step. You will continue the lab without the license.
You now have a UCP cluster with a single node. Next you are going to add two nodes to the cluster. These nodes are known as Worker nodes and are the nodes that host application containers.
This is not a requirement for the lab but is a best practice for any production clusters.
-
In the UCP GUI, click through to Resources / Nodes. Click "+ Add Node" and check the box "Add node as a manager"
-
On the same screen click "Copy to Clipboard." This will copy the UCP join command.
-
Log in to the other nodes that you would like to configure as managers and paste this join command on the CLI.
- In the UCP GUI, click through to Resources / Nodes. Click "+ Add Node" and then click "Copy to Clipboard."
The string you copied will look something like the following:
docker swarm join --token SWMTKN-1-5mql67at3mftfxdhoelmufv0f50id358xyyeps4gk9odgxfoym-4nqy2vbs5gzi1yhydhn20nh33 172.31.24.143:2377
This is a Swarm join token. It includes the IP address of the UCP/Swarm manager so do not change it in the command. It is a secret token used by nodes so that they can securely join the rest of the UCP cluster.
- Log in to
node1.
$ ssh ubuntu@<node1-public-ip>
- On the command line run the Swarm join token command you copied from UCP. You will get a status message indicating that this node has joined the cluster.
$ docker swarm join \
> --token SWMTKN-1-1dg967kx56j9s0l0o8t8oytwutacsspzjt6f2h2i31s3cevmcm-7tihlxtl2e2uztmxjhtgs5orz \
> 172.31.30.254:2377
This node joined a swarm as a worker.
This indicates that this node is now joining your UCP cluster.
-
Repeat steps 1 & 2 for
node2 -
Go to the UCP GUI and click on Resources / Nodes. You should now see that all of your nodes listed with their respective role as Manager or Worker.
Congratulations! You have successfully installed and deployed a full UCP cluster. Wasn't that easy? You are now ready to move on to the rest of the lab.
Compose is a specification for defining and running multi-container Docker applications. With compose, you use a compose file to configure your application’s services. Then, using a single command, you create and start all the services from your configuration. A single compose file can define all aspects of your application deployment including networking, health checks, secrets, and much more. The full specification for compose is defined here.
To deploy an application on UCP or Swarm, you create a service. Frequently a service will be the image for a microservice within the context of some larger application. Examples of services might include an HTTP server, a database, or any other type of executable program that you wish to run in a distributed environment. UCP schedules services across a cluster of UCP worker nodes. Services are managed by UCP throughout their lifecycle and can get rescheduled if they die, scaled up and down, gracefully terminated and more.
A Docker Stack is an instantiation of a compose file. When a compose file is deployed on UCP it is deployed as a stack. This stack is a group of related services (applications) that can be defined and deployed together in a single compose file.
In this section we will deploy the Docker Pets application using a compose file. In the following sections we will add features to our compose file and make our application progressively more complex and feature-full. Docker Pets is a simple web app that records votes for different animals and uses a persistent backend to record the votes. It's comprised of two images:
chrch/docker-petsis a front-end Python Flask container that serves up random images of housepets, depending on the given configurationconsul(which will be used in a later compose file) is a back-end KV store that stores the number of visits that the web services receive. It's configured to bootstrap itself with 3 replicas so that we have fault tolerant persistence.
This is the first iteration of our compose file for the Docker Pets application:
version: '3.1'
services:
web:
image: chrch/docker-pets:1.0
ports:
- 5000
deploy:
replicas: 2
healthcheck:
interval: 10s
timeout: 2s
retries: 3
version: '3.1'is the version of the compose format we are using.web:is the name that we are giving this service.image: chrch/docker-pets:1.0defines the image and version that we are deploying in this service.ports:configures the ports that we expose for our application. Our application listens on port5000so we are exposing port5000internally and mapping it to a random ephemeral port externally. UCP will take care of the port mapping and application load balancing for us.healthcheck:defines the health check for our application. We are setting theintervalfor how often the check runs and how manytimeoutswe allow before we consider the container to be unhealthy.
- Log in to your UCP GUI and go to Resources / Stacks & Applications. Click Deploy.
Paste the above compose file text into the box under Application Definition. In the Application Name box write pets. Click the Create button.
You should see a success message:
Creating network pets_default
Creating service pets_web
Your pets application stack is now deployed and live! It's running as two stateless containers serving up a web page.
- Go to Resources and click on the stack that you just deployed. You will see that we deployed a service called
pets_weband a network calledpets_default. Click on thepets_webservice and you will see all of the configured options for this service. Some of these are taken from our compose file and some are default options.
- On the bottom of the
pets_webpage, UCP will show what ports it is exposing the application on. Click on this link. It will take you to the IP:port of where your application is being served.
In your browser you should now see the deployed Docker Pets app. It serves up an image of different pets. Click on "Serve Another Pet" and it will reload the picture.
So far we have deployed a service as a single container. Our application will need some level of redundancy in case there is a crash or node failure, so we are going to scale the web service so that it's made of multiple containers running on different hosts.
- Go to Resources / Services /
pets_web/ Scheduling. Edit the Scale parameter and change it from2to3. Click the checkmark and then Save Changes. After a few moments on the Services page we can see that the Status will change to3/3as the new container is scheduled and deployed in the cluster. Click onpets_web/ Tasks. It shows the nodes where ourwebcontainers were deployed on.
- Now go back to the application
<node1-public-ip>:<published-port>in your browser. Click Server Another Pet a few times to see the page get reloaded. You should see the Container ID changing between three different values. UCP is automatically load balancing your requests between the three containers in thepets_webservice.
Now we are going to deploy a second service along with our Docker Pets application. It's called the Visualizer and it visually shows how containers are scheduled across a UCP cluster.
Now we are going to update the pets stack with the following compose file. We have added a couple things to this compose file. You are going to copy the entire compose file below and place it in to the application deployment field.
version: '3.1'
services:
web:
image: chrch/docker-pets:1.0
deploy:
replicas: 3
ports:
- 5000
healthcheck:
interval: 10s
timeout: 2s
retries: 3
visualizer:
image: manomarks/visualizer
ports:
- 8080
deploy:
placement:
constraints: [node.role == manager]
volumes:
- /var/run/docker.sock:/var/run/docker.sock
replicas: 3defines how many identical copies of a service container we want UCP to schedule.visualizer:is a second service we are going deploy as a part of this stack.constraints: [node.role == manager]is a scheduling requirement we are applying so that thevisualizeris only scheduled on the manager node.- /var/run/docker.sock:/var/run/docker.sockis a host-mounted volume we are mounting inside our container. This volume is being used so that thevisualizercan communicate directly with the local docker engine.
- Go to Resources / Stacks & Applications / Deploy. Paste the above compose file text into the box under Application Definition with the title
pets.
You should see the following output:
Updating service pets_web (id: vyp6gx092d1o6z7t2wy996i7u)
Creating service pets_visualizer
The pets_visualizer service is created and our existing pets_web service is updated. Because no pets_web parameters were changed in this compose file, there are no actions done to the web containers.
- Go to the
<node1-public-ip>:<published-portthat is listed on the bottom of the page of thepets_visualizerservice in your browser window. This could actually be any of your three nodes because all traffic is being load balanced. This shows you the nodes of your UCP cluster and where containers are scheduled. You should see that thepets_webcontainer is evenly distributed across your nodes.
Now that we have a redundant application that we can view with the Visualizer, we are going to simulate a failure to test UCP's ability to heal applications. We will shut off one of the Docker engines. This will simulate a node failure.
-
Bring the Visualizer app up in your browser. Make note of how the
petscontainers are scheduled across your hosts. They should be distributed equally across all hosts. -
Log in to the commandline of one of your worker nodes (be sure not to do this to the manager node). Shut the Docker engine off with the following command.
$ sudo service docker stop
docker stop/waiting
This will turn off the Docker engine and bring down all of the containers on this host.
- Now watch the visualizer app in your browser. You will see one of the nodes go red, indicating that UCP has detected a node failure. Any containers on this node will now get rescheduled on to other nodes. Since we defined it in our compose file with
replicas: 3, UCP will ensure that we always have3copies of thewebcontainer running in our cluster.
- Now return
ucp-worker-1to a healthy state by turning the Docker engine on again.
$ sudo service docker start
docker start/running, process 22882
- Finally, decommission this
petsstack in the UCP GUI by clicking onpets/ Actions / Remove Stack. Confirm the removal. You have just removed all of the containers and also the networks that were created when the stack was first deployed.
In this task we are going to add another service to the stack. Up to this point the Docker Pets application was a set of stateless web servers. Now we are going to add a persistent backend that will enhance the functionality of the application. We will use consul as a redundant backend which will store persistent data for our app, distributed across a set of nodes.
The resulting application will have 3 web frontend containers and 3 db consul containers. A backend network will be deployed for secure communication between web and db. The app is exposing HTTP endpoints for different services on ports 5000 and 7000. UCP will publish these ports on each node in the UCP cluster. Application traffic to any of the external ports will get load balanced to healthy containers.
In this step we will deploy a new compose file that adds functionality on top of the previous compose files.
version: '3.1'
services:
web:
image: chrch/docker-pets:1.0
deploy:
replicas: 3
ports:
- 5000
- 7000
healthcheck:
interval: 10s
timeout: 2s
retries: 3
environment:
DB: 'db'
networks:
- backend
db:
image: consul:0.7.2
command: agent -server -ui -client=0.0.0.0 -bootstrap-expect=3 -retry-join=db -retry-join=db -retry-join=db -retry-interval 5s
deploy:
replicas: 3
ports:
- 8500
environment:
CONSUL_BIND_INTERFACE: 'eth2'
CONSUL_LOCAL_CONFIG: '{"skip_leave_on_interrupt": true}'
networks:
- backend
visualizer:
image: manomarks/visualizer
ports:
- 8080
deploy:
placement:
constraints: [node.role == manager]
volumes:
- /var/run/docker.sock:/var/run/docker.sock
networks:
backend:
- 7000is exposing a second port on our application. This port will serve traffic to administrate the app.environment:defines environment variables that are set inside the container. In this compose file we are settingDB=db. Our backend service is nameddbso we are passing in the service name to the front endwebservice. During operations, built-in Docker DNS will resolve the service name to the IPs of the service's containers.image: consul:0.7.2is the image and version for our backend data persistence store. We are deploying 3 replicas for a highly available backend.command:is passing a specific command line argument to the consul image.networks: backend:defines an overlay network that bothwebanddbwill connect to to communicate.
-
Deploy the
petsstack again with the above compose file. -
Once all the service tasks are up go to the
webservice externally published<node1-public-ip>:<port>that maps to the internal port5000. The Docker Pets app is written to take advantage of the stateful backend. Now it gives you the capability to cast a vote for your favorite pet. The vote will be stored by thedbservice along with the number of visits to the application. -
Submit your name and vote to the app.
After you cast your vote you will get redirected back to the pets landing page.
-
Refresh the page a few times with Server Another Pet. You will see the page views climb while you get served across all three
webcontainers. -
Now go to the
webservice externally published<node1-public-ip>:<port>that maps to the internal port7000. This page totals the number of votes that are held in thedbbackend.
- If you go to the
visualizerservice in your browser you will now see that a redundant Consul KV store is deployed on the Swarm. It's storing the votes for the application.
Secrets are any data that an application uses that is sensitive in nature. Secrets can be PKI certificates, passwords, or even config files. UCP handles secrets as a special class of data. Docker secrets are encrypted them at rest, sent to containers through TLS, and are mounted inside containers in a memory-only file that is never stored on disk.
Before we can configure our compose file to use a secret, we have to create the secret so it can be stored in the encrypted UCP key-value store.
- In the UCP GUI go to Resources / Secrets / +Create Secret. Name your secret
admin_password. Enter the secret password of your choice and click Create.
This secret will now be stored encrypted in the UCP data store. When applications request access to it, the secret will be sent encrypted to the container and mounted in a memory-only file on the host.
- Update the
petsstack with the following compose file.
version: '3.1'
services:
web:
image: chrch/docker-pets:1.0
deploy:
replicas: 3
ports:
- 5000
- 7000
healthcheck:
interval: 10s
timeout: 2s
retries: 3
environment:
DB: 'db'
ADMIN_PASSWORD_FILE: '/run/secrets/admin_password'
networks:
- backend
secrets:
- admin_password
db:
image: consul:0.7.2
command: agent -server -ui -client=0.0.0.0 -bootstrap-expect=3 -retry-join=db -retry-join=db -retry-join=db -retry-interval 5s
deploy:
replicas: 3
ports:
- 8500
environment:
CONSUL_BIND_INTERFACE: 'eth2'
CONSUL_LOCAL_CONFIG: '{"skip_leave_on_interrupt": true}'
networks:
- backend
visualizer:
image: manomarks/visualizer
ports:
- 8080
deploy:
placement:
constraints: [node.role == manager]
volumes:
- /var/run/docker.sock:/var/run/docker.sock
networks:
backend:
secrets:
admin_password:
external: true
We have made two additions to this compose file:
ADMIN_PASSWORD_FILE:is an environment variable that tells thewebservice that the secret will be stored at the location/run/secrets/admin_passwordsecrets: admin_passwordreferences the secret that we created in UCP. UCP will send the secret to thewebcontainers wherever they are scheduled.
- Now go to the
<node1-public-ip>:<port>on thepets_webservice that is mapped to the internal port7000. This is the administrator consul that displays the votes. It should be protected by a password now. Use the secret you entered as theadmin_password.
The Docker Pets application is built with a /health endpoint to advertise it's own health on port 5000. Docker uses this endpoint to manage the lifecycle of the application.
- View the application health by going to
<node1-public-ip>:<port>/healthin your browser. You can use theipof any of your UCP nodes. Theportmust be the external port that publishes the internal port5000.
You should receive an OK message indicating that this particular container is healthy.
- Now use your browser and go to the
<node1-public-ip>:<port>/killURL. This will toggle the health to unhealthy for one of thewebcontainers.
You should receive a message similar to:
You have toggled web instance 87660acc389c to unhealthy
- Go to the Visualizer in your browser. You will see the healthcheck on one of the containers go to red. After a few more moments UCP will kill the container and reschedule another one. A default grace period is applied to allow existing application traffic to drain from this container. This simulates an application failure that is recovered by UCP.
A rolling update is a deployment method to slowly and incrementally update a series of containers with a new version in a way that does not cause downtime for the entire application. One by one, UCP will update containers and check the application health for any issues. If the deployment begins to fail in any way, a rollback can be applied to return the app to the last known working configuration.
In the following steps we will update the pets_web service with a new image version. We will use a purposely broken image to simulate a bad deployment.
-
Click on the
pets_webservice. On the Details page change the image tochrch/docker-pets:broken. Make sure to click the green check so that the change is captured. -
On the Scheduling page update the following values:
- Update Parallelism
1 - Update Delay
5 - Failure Action
pause - Max Failure Ratio
0.2(%)
- Update Parallelism
These values mean that during a rolling update, containers will be updated 1 container at a time 5 seconds apart. If more than 20% of the new containers fail their health checks then UCP will pause the rollout and wait for administrator action.
-
The changes you made are now staged but have not yet been applied. Click Save Changes. This will start the rolling update.
-
Now view the Visualizer app in your browser. You will see that the
chrch/docker-pets:brokenimage is rolled out to a single container but it fails to pass it's health check. The secondchrch/docker-pets:brokencontainer will start failing afterwards which will trigger the rolling update to pause.
- In UCP, click on the
pets_webservice. You will see the status of the update is now paused because of a failed health check.
- In the Details page of
pets_webclick Actions / Rollback. This will automatically rollback thepets_webservice to the last working image.
-
Repeat step 1 but this time use the image
chrch/docker-pets:2.0. -
Repeat step 2 with the same values and click Save Changes.
-
Observe a successful rolling update in the Visualizer. You will start to see each container being updated with the new image and in good health. Now go to the
<node1-public-ip>:<port>that corresponds to the internal port5000. After a couple refreshes you should see that some of the containers have already updated. -
Before you go on to the next section, delete the
petsstack.
Layer 7 load balancing in UCP can be provided by a feature called the HTTP Routing Mesh (HRM). In this section of the lab we configure HRM for several different ports and endpoints of docker-pets.
This part of the lab requires external configuration of DNS. An external DNS provider like AWS or an internal DNS should be configured for this lab. The details of the DNS configuration is outside the scope of the tutorial.
Setup: Configure a wildcard DNS entry *.<yourdomain.com>. You will insert your URL for <yourdomain.com>. You can configure this DNS record to point to a load balancer that balances across multiple UCP nodes. For a simpler configuration, you can also use the IP address of any UCP node as this DNS record.
- In the UCP GUI go to Admin Settings / Routing Mesh. In this window, check the box to enable the routing mesh and ensure that it is serving HTTP traffic on port
80.
- Deploy the following compose file as the
petsstack. Input your configured URL for<yourdomain.com>. In this step we are configuring three URLs,vote.*,admin.*, andviz.*for three different services provided by the app. Go to these URLs and ensure that they are working correctly.
version: '3.1'
services:
web:
image: chrch/docker-pets:1.0
deploy:
replicas: 3
ports:
- 5000
- 7000
labels:
com.docker.ucp.mesh.http.5000: "external_route=http://vote.<yourdomain.com>,internal_port=5000"
com.docker.ucp.mesh.http.7000: "external_route=http://admin.<yourdomain.com>,internal_port=7000,sticky_sessions=paas_admin_id"
healthcheck:
interval: 10s
timeout: 2s
retries: 3
environment:
DB: 'db'
ADMIN_PASSWORD_FILE: '/run/secrets/admin_password'
networks:
- backend
- ucp-hrm
secrets:
- admin_password
db:
image: consul:0.7.2
command: agent -server -ui -client=0.0.0.0 -bootstrap-expect=3 -retry-join=db -retry-join=db -retry-join=db -retry-interval 5s
deploy:
replicas: 3
ports:
- 8500
environment:
CONSUL_BIND_INTERFACE: 'eth2'
CONSUL_LOCAL_CONFIG: '{"skip_leave_on_interrupt": true}'
networks:
- backend
visualizer:
image: manomarks/visualizer
ports:
- 8080
labels:
com.docker.ucp.mesh.http.8080: "external_route=http://viz.<yourdomain.com>,internal_port=8080"
deploy:
placement:
constraints: [node.role == manager]
volumes:
- /var/run/docker.sock:/var/run/docker.sock
networks:
- ucp-hrm
networks:
backend:
ucp-hrm:
external: true
secrets:
admin_password:
external: true