On the 19th and 20th of September CITCON (pronounced "kit-con") took place in Zagreb, Croatia. CITCON is dedicated to continuous integration and testing. It brings together some of the most interesting people of the European testing and continuous integration community. These people also determine the topics of the conference.

They can do this because CITCON is an Open Space conference. If you're not familiar with the concept of Open Space, check out Wikipedia. On Friday evening, attendees can pitch their proposals. Through dot voting and (constant) shuffling of the schedule, the attendees create their conference program.

In this post we'll wrap up a few topics that were discussed.


Peter Zsoldos (@zsepi) went into his most recent brain-spin: polytesting. If I have a set of requirements, is it feasible to apply those requirements at different levels of my application; say, component, integration and UI level. This sounds very appealing because you can perform ATDD at different levels.
This approach is particularly interesting because it has the potential to keep you focused on the required functionality all the way. You'll need good, concrete requirements for this to work.


Microservices are a hot topic nowadays. The promise of small, isolated units with clear interfaces is tempting. There are generally two types of architectures that can be applied. The most common one is where there is no central entity, and services communicate to each other directly.

Douglas Squirrel (@douglasquirrel) explained an alternative architecture by using a central pub-sub "database" to which each service is allowed to publish "facts". Douglas deliberately used the term facts to describe single items that are considered true at a specific point in time ("events"" is too generic a term).

The second model comes closer to mechanisms such as event sourcing (or even ESBs if you take it to the extreme). One of the advantages of this approach is that, because facts are stored, it's possible to construct new functionality based on existing facts. For example, you could use this functionality in a game to create leader boards and, at a later stage, create leaderboards per continent, country, or whatever seems fit.

Unit testing

Arjan Molenaar introduced a flaming hot topic this year: "unit testing is a waste". Inspired by recent discussions of DHH, Martin Fowler, and Kent Beck, Arjan tried to find out the opinions of the CITCON crowd. Most of the people contributing to the discussion must have been working in consultancy, because the main conclusion was "It depends".

Whether unit testing is worth the effort mainly depends on the goals that people try to achieve when writing their unit tests. Some people write them from a TDD perspective. They use tests to guide themselves through development cycles, making sure they haven't made little errors. If this helps you, then please keep doing it! If it does not really help, well ...

Other people write unit tests from a regression perspective, or at least maintain them for regression testing. This part caused the most discussion. How useful are unit tests for regression testing purposes? Are you really catching regression if you isolate a single unit?

The growing interest in microservices also sheds new light on this discussion. In the future, when microservices will be widely adopted, we will be working with much smaller codebases. They might be so small and clear that unit tests are no longer required to guide us through the development process.

CI scaling

Another trending topic was scaling CI systems. It was good to see that the ideas we have at Xebia were consistent with the ideas we heard at CITCON. First off, the solution for everything (and world peace, it seems) is microservices. Unfortunately, some of us, for the time being, must deal with monolithic codebases. Luckily there are still options for your growing CI system, even though for now it remains one big chunk of code.

The staged pipeline: you test the things most likely to break first. Basically, you break your test suite up into multiple test suites and run them at separate stages in the pipeline.

But how do you determine what is most likely to break and what to test where? Tests that are most likely to break are those that are closely linked to the code changes, so run them first. Also, determine high-risk areas for your application. These areas can be identified based on trends (in failing tests) or simply through human analysis. To determine where to run the different test suites is mainly a matter of speed versus confidence. You want fast feedback so you don't want to push all your tests to the end of the pipeline. But you also don't want to wait forever before you know you can move on to the next thing.

Beer brewing for process refinement

Who isn't interested in beer brewing? Tom Denley (@scarytom) proposed a session on home brewing and the analogy. Because Arjan is a homebrewer himself, this seemed like an obvious session for him.

In addition to Tom explaining the process of brewing, we discussed how we got into brewing. In both cases, the first brew was made with a can of hopped malt syrup, adding yeast, and producing a beer from there. For your second beer, you replace the can of syrup with malt extract powder and dark malt (for flavour). At a later stage, you can replace the malt extract with ground malt.

What we basically do is start with the end in mind. If you're starting with continuous delivery, it is considered good practice to do the same: get your application deployed in production as soon as possible and optimise your process from your deployed app back to source code.

Again, it was a good conference with some nice take-aways. Next year's episode will most likely take place in Finland. The year after that... The Netherlands?