Docker containers for Visual Regression Testing setup

Packing it all inside docker containers

In the previous post we looked at a possible solution for doing Visual Regression Testing. But stepping back and looking at it, I think there was tedious for the reader to actually try it out, because it required him to install some extra dependencies like ImageMagick, phantomjs( which in turn needs nodejs) to be able to run the example himself(also not considering docker installation). This would have gotten even more complicated if we wanted to add more browser beside phantomjs, but Docker containers can also be used as "build containers" not just to run applications.
Doing away with extra configuration of dependencies and just run your application or tests inside a container build from a Dockerfile recipe(or a prebuilt one available from a repository) is one of the "features" of working with containers.

Quite a few cloud services have appeared recently that offer to run docker containers. Mix them up with some git hooks and this opens the way to run unit tests and integration tests inside the containers on each commit or pull request before merging into your codebase. Something like what rultor is doing. What a great time to be alive :).


We're not going to talk about what Docker is, as we already did it here some time ago.

The plan

Just a little reminder of what we planned on doing. We have our web application pippo-demo.war file which was already built by Jenkins in another step. Part of our build setup we also have separate integration tests and visual regression tests to run for the web application.

Running our integration tests inside a prebuilt docker container with all the dependencies already baked in.

We want to deploy the pippo-demo.war inside an "actual" Tomcat instance and run integration tests using Selenium Webdriver against the deployed web app.

To keep things simple we just want to be able to run our integration tests through a single command.

We needed ImageMagick along with phantomjs and nodejs to do the visual regression tests.

  • So we'll create a new ui-tests docker image with these dependencies already installed. We base this ui-tests image upon a simple maven3 docker image because it's inside this image we checkout our integration tests suite.

Our integration tests are organised as a Maven project so we can run them with just

mvn clean verify  

We have prepared a generic reusable Tomcat7 container into which we deploy pippo-demo.war.

  • pre-integration-test phase we use the cargo-maven2-plugin to deploy the application inside the Tomcat7 container.
  • maven-failsafe-plugin runs our plain JUnit tests with Webdriver against the endpoints which we already discussed in article.

Build container

All the dockerfiles necessary to create the images are under docker folder.

PS: They are there as examples and for should you want to create your own dockerfiles. Be aware that there is an official docker repo location for maven containers just know it's built upon an open-jdk image(might be a legal issue with distributing the oracle jdk).

That is the reason why I've not pushed the images myself and will instead maybe change the example to openjdk images.

Docker Compose - a way of "orchestrating" containers

Sometimes when you have a more complicated setup, for ex. your web application for which you do integration testing also requires a database, we can also pack that database inside a container.

It makes sense to have separate services like databases, web containers defined and kept in separate containers to keep things clear, configurable(for example to be able to test with a different database) and play around with the "big building blocks" sort of like legos to swap one component for another.

Orchestrating those containers can get messy: specifying which volume directories are shared with the host or between containers themselves, what ports are to be exposed for each container and how they should be linked.

This is where Docker Compose(previously named fig) can help us.
Through a file called docker-compose.yml you can describe the whole setup. Let's look at the file we've used for the project:

  image: balamaci/tomcat7
    - /tmp/tomcatlog:/opt/tomcat/logs

  image: balamaci/ui-tests
    - .:/usr/src/app 
    - ~/.m2:/root/.m2 
    - tomcat
    - 8080:8080

We defined two services tomcat and uitests.

Our uitests container is linked to the tomcat container. This means there is an entry added into /etc/hosts inside the uitests container and to access the Tomcat web app server we can just do http://tomcat:8080 (from uitests which is our actual container in which we run the tests).


    - .:/usr/src/app 

The whole current directory - if we checked out the project - source files for the integration tests are being shared with the uitests container through a docker volume, they are not "burned" inside the image(like with the ADD command for Dockerfile). This way we can always checkout new tests and just reuse the same uitests container.

while with

    - ~/.m2:/root/.m2 

We shared the .m2 folder from the host with the container. That is because we don't want everytime we run our integration tests to pull all of maven dependencies from the internet again and again.

For more detailed help with every command, or what else you can configure be found here

If we have added in the config file a command line to be executed implicitly then we just could have done

docker-compose up  

but instead we'll run the mvn command explicit

docker-compose run uitests mvn clean verify  

which will also start the linked tomcat container

Stopping all the containers is done with:

docker-compose stop  

to also remove the containers so we have a clean slate:

docker-compose rm  

PS: We could have done this setup I think with all the dependencies and Tomcat server baked into a single image and container, but we kept things more clear this way and we also get a chance to play with docker-compose.
PS 2: docker-compose it's just a tool, a facilitator over docker. The created containers are in everyway "plain" docker containers, and all docker commands still apply to them.

The coderun

Installing docker and docker-compose:

wget -qO- | sh

#docker-compose installing
curl -L`uname -s`-`uname -m` > /usr/local/bin/docker-compose  
chmod +x /usr/local/bin/docker-compose  

clone the repo:

git clone  

build the images from the Dockerfiles


run the tests:

docker-compose run uitests mvn verify  

stopping and removing the containers:

docker-compose stop  
docker-compose rm  


Containers are awesome, I'm really buying into the hype and think they will definitely play a major role at least in helping reaching the next level of automation for testing that require more complex setups.

We've managed to simply our whole setup as we no longer force the reader to run tedious steps to configure the testing environment.

If we'd have pushed our uitests image to the docker image registry which offers free hosting for public repos, we could the whole would have been
checkout the github repo with the integration tests and run

docker-compose run uitests mvn verify  

With all the services popping up offering the possibility to deploy and run docker containers, while docker alternatives like rocket are being developed, it's definitely worth to keep an eye on evolution of container based packaging and deployments.


Docker Compose

comments powered by Disqus