How software that runs Wikipedia is tested
by Željko Filipin
This is a new and shiny version of my How MediaWiki, software that runs Wikipedia, is tested blog post. I gave this talk several times at various meetups and conferences, but I never had the time to sit down and document it. There are some slides too.
The previous version of this talk (and blog post) was centered around tools. I would mention a tool and then talk about the problems it solves. The night before I went to Selenium conference I had an idea to center the talk about the problems, and then mention the tools.
What were the problems that we wanted to solve? Communication, maintainability, browser automation, browser/operating system/version combinatorial explosion, visibility/transparency, code reuse, speed, “Halp wanted!” and “Works at my machine!”. Let me say a few works about each problem and how we solved it.
Anybody surprised that the biggest problem in a software project is communication? Exactly. It is hard to know what to test, it is hard to let other people know what is tested and what is the result.
Cucumber. You can ask the people that know what needs to be tested to write examples in any natural language. When the test runs, Cucumber creates nice reports that anybody can read.
Were you a part of test automation project that failed because the test code was not maintainable? Exactly. Test code is like any other code, you have to make sure it easy to maintain.
The Page Object pattern. It is a simple solution. Every page in the application is represented by one class.
For the above example, this is how a login page could look like.
Code that connects Cucumber and the page could look like this.
Above code uses page-object Ruby gem.
Could this simple pattern be so useful? It could. For more information take a look at Cucumber & Cheese book.
Once upon a time, browser automation was a hard problem. I think it is solved now. There are plenty of open source and commercial tools that automate all major browsers on all major operating systems. Even mobile browsers and native applications.
Do you know how many combinations of major browsers, operating systems and their versions you should test? Do you know how hard it is to maintain computers or virtual machines with all those combinations? Exactly.
Sauce Labs. They support everything we need, and more.
We needed a way for our work to be visible to our coworkers, but also to anybody else. Yes, anybody.
There are several solutions to the problem.
Code is hosted in Gerrit and there are GitHub mirrors for almost all repositories. Of course, everything is completely open. We also use Gerrit for code review. Almost every commit is reviewed by somebody else before it gets merged into master.
Mailing list, bug tracker and continuous integration server (Jenkins) is open. Anybody can see our e-mail communication, bugs and Jenkins jobs. Every Jenkins job starts several Sauce Labs jobs. All of them are open too.
MediaWiki consists of core and extensions, a usual practice. Every extension has a separate Git repository. Several extensions have Selenium tests and there was code that was copy/pasted from one repository to the other. There has to be a better way, right?
Did you notice that Selenium tests take a long time to run? I did too. We have tried to speed up the test using several optimizations, but they are still slow. The only way to really speed them up is to run them in parallel. I had good results with it on several projects.
We are not there yet. When the time comes, we will probably use parallel_tests Ruby gem.
Are you surprised that an open source project is looking for help? Surprisingly (at least to me) more people would like to write production code instead of test code. If you would like to be on the testing side, read on.
I have left an interesting problem for the end. How many times a day do you hear people say: “works at my machine”? I do not think I should go into a lot of detail explaining the problem. It should be familiar to anybody involved in software development. Sometimes software works on one machine but not on the other. Sometimes it is hard to tell why.
Since this is not a simple problem, we have several solutions for it.
Jenkins is Single Source Of Truth. It does not matter if it works on your machine. If it fails at Jenkins machine, you should fix it.
Are you surprised that a piece of open source software is not trivial to install and configure to work correctly? Mediawiki-Vagrant provides an easy way to install a basic MediaWiki installation in a virutal machine, but also provides a way to easily configure it in different ways, including complete setup for running Selenium tests.
The last piece of the puzzle is Bundler.
Bundler uses a file called
Gemfile, located at the root folder of the project. It defines which Ruby gems the project needs.
Gemfile.lock file that goes into more detail.
Why did we pick Ruby to run Selenium tests?
Can you believe that all the problems that we wanted to solve were already solved by a mature Ruby library? Other languages had libraries that solved some of the problems. If a language had all or most of the tools already available, then the tools were not as mature as in Ruby.tags: code - event - featured - ruby - selenium - speaker - testing - testival - wikimedia