Read part 1 of the report.
Before I continue to part 2, a word about timing. Some of you may ask why it took me so long to publish my impressions of an event that happened last week (which, in Internet time, is ‘way in the past’). The thing is, I prefer to live the reality, rather than be outside it and look into it through the phone or tablet screen. As Louis CK would say, ‘the resolution on reality is amazing, it is super HD’, which is way better than even the new iPhone 6. Hence I make records of events as quickly as I can and get back to, you know, be in them.
On with the event. Day 2 of nodeconf.eu started with the front-end track, with Alex Liu from Netflix dazzling us with the radial approach they took with Node.js and Dust.js when it comes to A/B testing various designs. What is unique in this approach is that A/B testing is not done at a load ballancer level. A traditional approach would see two apps serving the same routes and the ballancer (such as Nginx) routing requests to instance A or B using some routing rule, and then measuring the outcome. Netflix took Dust.js partials to a completely new level by testing out many different combinations of designs within the same app or page, and testing more than A and B (there is C and D and E etc.). Of course, common sense suggest this is very hard to manage manually, hence their approach to use a registry that applies packaging rules and puts together distinct combinations of parts that form a particular test (they are served from CDN for the next person participating in it to cut on the processing latency). That, plus I got a kick of Netflix being among the users of Dust.js that we are also using.
You may know Matteo Collina already from my blog posts because he wrote the MQTT NPM module we are using in our own code. This time around, Matteo talked about the new work that is attempting at make communication between micro-services possible without the message brokers. This work is inspired by libchan by Docker team that is written in Go. Matteo and my internet buddy Adrian Rossouw are collaborating on providing a Node.js version as part of the project Graft.
Jake Verbaten shared the nitty-gritty of writing Node.js services in Uber, where everything you do needs to be up all the time, and run on a multitude of machines. He highlighted all the extra steps that separate anything you make from the day it is ‘productized’. Jake showed the tools Uber uses for this process, including the tool they call ‘potter’ (will be open-sourced at some point). It gets all the scaffolding going for a project, including the repo, continuous integration server, and monitoring.
Thorsen Lorentz went into the details of Chrome JavaScript engine (V8) and JavaScript performance. This is not something you normally thing right away when writing JavaScript, but can come in handy when doing performance tuning. However, his examples left me with mixed feelings. We like our abstractions and the fact that declaring variables, and later assigning values (as opposed to initializing them in the first statement) can have detrimental effect on performance filled me with dread. I don’t want to break the black box approach to Node.js if possible. Don’t get me wrong, it is perfectly understandable that V8 will work better if helped by certain coding pattern, but it means that I need to become aware of the actual JavaScript engine running my code, which further breaks the Node.js promise of front end and back end unity (unless you restrict yourself to only serving your Chrome clients, that is). Of course, all this is not Thorsen’s fault – we should not shoot the messenger.
After the break, Bert Belder from StrongLoop kicked off a Node Core track. StrongLoop is a company build around helping others succeed with Node.js, and has the most corporate contributions to the upcoming 0.12 version, as well as the current 0.10 (outside of Joyent, of course). Bert has invited the audience to ask for fixes in node, express, node inspector and other areas.
Fedor Indutny held up to the stereotype that Russians are scary good at math by taking us down to follow the white rabbit of TLS encryption. His credentials (see what I did there): he is the author of the TLS module for Node.js. Of course, most of us just want encryption to work and don’t care how, but it was fun to peek behind the curtain at RFC 5246 and find out more about it. He spent the rest of the talk walking us through code snippets of setting up the server with tls.js and exchanging encrypted hello’s.
For the second time at this conference, TJ Fontaine went into the dark allays of tracing Node and finding out what is happening when you launch ‘node app.js’. Apart from replicating Node graphics on T-shirts and the backs of his laptops, TJ’s favorite pastime is debugging v8 core dumps on OSX and Linux using lldb-v8. I got to see more V8 detritus than I saw in a – ever? Golden takeaway – please, please, please name your functions – debugging anonymous functions postmortem is no fun.
After lunch, there were more workshops – more debugging with TJ and hands on with NearForm’s own nscale deployment solution. Unfortunately, my jet lag finally caught up with me and I had to crash.
Our afternoon event involved vising a local hurling club (an Irish sport involving an ash bat and a ball, although hitting a fellow player instead is fine too). We got to make a nice group photo on the green.
Our evening entertainment took us to Waterford on the sea shore. Lots of Guinness (I swear it tastes better here than from a can back home!), and some traditional Irish music, including wonderfully quirky Irish pipes.
Continue on to the third and final installment of the report.
© Dejan Glozic, 2014
Dejan Glozic 