QCon day three

Monkeys in lab coats: Applying Failure Testing Research @Netflix

This keynote was presented by the duo Kolton Andrus and Peter Alvaro and it was a pretty interesting one. The key point here is the importance of mixing Academia and Industry together when we need to tackle hard problems. The use case was the development of a tool to identify potential bugs in Netflix services. The nice thing about this talk was the way they evolve the ideas and the amazing results that can arise when people from different backgrounds and/or specialities get together with the same mindset, to solve a problem. Another interesting point is that even when their interests were different, one in the theoretical aspects and the other in the practicability of the solution, both converged in a way that solving the problem would be beneficial to both, however, by different reasons. The talk was very amusing and, despite of the sleepy morning, this was an interesting moment.

Test-Driven Microservices: System Confidence

Presented by Russ Miles and was a bad presentation (sorry for that). The reason is that the contents were not fitted for the audience. He talked about System Tests in the context of Microservices and the need to adapt the test philosophy to match the problem at stake. He put some points in a interesting way, for example the concept of microservice as Loosely coupled service oriented with bounded contexts, he points out for the need to increase the focus in confidence instead in testing and this pretty much sums up the juice of the talk. On his defense he was a pretty funny guy. He started the talk playing on his electric guitar the music Highway to Hell from AC/DC (I must say he plays guitar wonderfully) with the title Microservices from Hell. In some parts of the talk he threw some jokes, like the problem with Microservices is the size, and gents all we know that size... doesn't matter and In California there is sun, for those who are British is that thing that lives in the sky. But in the end, funniess aside, this was a poor moment in a very cool conference.

Hacking bank mobile apps

This talk was presented by Stevie Graham and was a pretty amazing demonstration of perseverance. It was basically an overview about the inner working of how he unlocked the private API banks use to enable mobile applications to make the usual financial operations. The use case target OSX enviroment applications and the techniques used were not new but the work was pretty overwhelming as he spends six months just to figure how to decompile a obfuscated part and to construct the process of creating a token based on 3DES with RSA encryption. The steps to unlock the API consisted in method hooking to log important information and the construction of the authentication tokens to be able to process the requests and invoke the banking API. As I said the steps are simple, however the amount of work is enormous because it is needed circumvent protections a several layers. I loved the presentation and also the questions part because this one triggered some infuriated behaviors from banking responsible people.

Successful Go program design, 6 years on

This talk was presented by Peter Bourgon and was all about best practices in the Go programming language. For me was a little bit disappointing because I was expecting a higher overview about the Go in the context of their problems. I thought that it would be demonstrated how a complexity of a industry level project is managed when we use Go as the main language. Turns out that the focus of the presentation was directed for low level best practices that pretty much all programers already use, in most of the other languages. Topics like dependency injection, and the need to develop oriented to contracts (most known as interfaces), refactoring constructors to group arguments into a complex structure as a good practice, factories and replacing the nil value by default implementations, were some of the many tiny advices given by Peter. I agree with all those points, that's for sure, I just was expecting a very different approach in this talk/lecture. I agree that my expectations are my responsibility nevertheless by the amount of people leaving I pretty much believe that I was not the only one living in a dreamy world created by myself.

Applied supervised learning predicting recidivism

Here the star of the show was Michelle Lee. Unfortunately this was not a very pleasant show. It was pretty much an overview about all the typical steps needed to do machine learning in supervised algorithms. The several phases of data gathering, pre-processing, analysis of the statistical tools used and several variations and alternatives regarding the statistical meaning of the data and the results were pretty much all we could get from this talk. Personally I think that the level of analysis is not proper for the audience. At the same level it was not possible to discuss with detail the algorithms given and, on the other hand, it was not given a satisfying context of the nature of the problem because, as she argue for respect regarding the charities envolved I will not share real data. That's fine by me. What's not so fine is that the mocked data she created was not also shown and, as a consequence, she did not discuss clearly the purpose to achieve in the end. By doing this all the statistical tools and analysis were pretty much hard to understand since the original problem to tackle was not explained clearly. I hope I'm not being to harsh on this case because, honestly I was not with high expectations, nevertheless the few I've got were not even met. In the end I asked if she would find some usefulness in the use of market basket algorithms and, apparently she didn't understand my question and respond to something else. I this case I think I'm also responsible because for what my friend told me it is perfectly plausible that she didn't understand my terrible english, but that didn't mean I was less frustrated by that.