Awesome place, old building, great transportation around. And guess what’s here: Banksy expo!
Node Interactive Day 1
by Mikeal Rogers @mikeal
- Fastest growing open source platform in theworld
- 400 new package every day
- 2.5 Million user in 2015 -> 5 Million users in 2016
- Why: Backend, API, Frontend, Mobile, tablets, desktops … IoT = The full stack -> Same tool for any platform (dev, patterns, libs, debugging)
- Mobile: Apache Cordova
- Desktop; Electron
- Dev tools
- Cloud: Open Shift, Joyent, Google cloud, AWS, Bluemix (IBM)…
- IoT: Jhonny5, on board stuff (Pi, arduino, Intel…)
- APIs and µServices
Node is very efficient in single process env which make sens to use it on IoT small, low energy hardware
Core state of the union
by James M Snell (IBM) @jasnell
Today: 85 contributores with commit access + 1084 individuals contributors
2700 commits since v4.0.0
Going for stability: Improving doc, update tooling, fixing bugs
Canary in the Goldmine: Project that continuously test modules against node major evolution.
Testing: Every non trivial commit runs tests on all platforms ~25
New major release every 6 Months.
- Even in April (going to LTS 30 month in October)
- Odd in October
Current LTS: v4.5.0 -> 04/2018
- Better Web Standards support (including HTTP/2)
- ES6/ES7 Support (Promises, Promise debugging, async await, ES6 Module see @bradleymeck, TC-39)
- Native module API: not linked to v8 -> to be independent from v8 updates
- Static error codes (consitency)
- better stream API
- Node.js VM agnostic with ChakraCore
- Improve Embedding (Electron, IoT)
- Internationalization i18n
- Debugging (nodereport, node –inspect)
by Ashley Williams @ag_dubs npm manager
Tech has a diversity problem, node has a diversity problem.
Teach node for everyone.
174 Students so far
The course is mainly using package motivation with cats ad the goal is to create an app from scratch. So Student can have a concrete product afterward.
Quotes from the speakers:
“The current problem in node is the environment… we need to work on that.”
“Not being a jerk is not enought, it should be more active welcoming. It means, listen to people and believe what they say.”
“The social problem is a system problem and system problems is what IT is good at fixing.”
Everything we need to know about Node.js Event loop
by Bert Belder @pisciaureus node.js - libuv - founder of StrongLoop
Most representation of the event loop on google image are wrong.
In reality, it starts with the program and finish with the process#exit between there are several steps.
Steps of the NodeJS loop
Every box as is own inside loop
It keeps looping till refs counter reach 0
The ref counter is ++ or – when system resources are opened and closed (network, disk …)
The unicorn function depending on the system is relying on epoll_wait() (Linux) or kenvent() (BSD) or GetQueuedCompletionStatusEx() (Windows)
NodeJS from monolith to µService
Netflix comes from a huge Monolithic system and want to move away from it.
In the new system, SemVer is used for API versionning, and there is a middle layer that get all the API requests and redirect them to the appropriate docker running the approriate version.
So one docker per version to isolate everything.
It’s easier for the developer too, who only have a VM with some docker on it. One docker = one service/API at a specific version.
Breaking the monolith
by @slashdotpeter Peter Marton
- Split in µServices: 1 DB - 1 Service
- Code by feature
- Do continuous delivery
µService is quite complex to setup for a MVP.
Security concern: node-http-signature
Dumping core in production is necessary when there is an issue but then you have to deal with the dump. The project PostMortem is there to help dealing with that and provide tools to works with the dump. They try also to have a common dump format to allow dumping and replaying with the best tools.
by Philipp Burckland
Why using JS: * it’s faster than R or Python (not C and not Python using C libs directly) * it could be plugged in realtime API running node
Unstructured data is 80% of data and from there with Latent Dirichet Allocation
You can use the figures to have Sentiment analysis and topic modeling for instance.
From Pterodatcyls and CActus to AI
by Ivan Seidel Gomes
Based on the game from google chrome with the dinosaure.
- Put sensors on the dinosaur
- Distance sensor (distance from the cactus)
- Size of the cactus (width)
- Speed sensor (speed increase during the game)
- Gamemanipulator: The sensor are reading image from the running game (RGB) and should find the world constraints
- Actuators: Send key events
IN: sensor OUT: Key stroke
Code in the middle … we just need to write code… that doesn’t work in machine learning.
Instead it connects input to output via neural network
One Neuron aggregate inputs and provide output ay + bz + x
You can add more neurons … what if we have 100 neurons … or more
the project use Synaptic library and RobotJS (keypress, read…)
Neurones –> Genetic algorythm (without connection + linear) like DNA
We can make random DNA: the same Genetic Algorythm with random values and run them agains the computer, we put them in life and test them and rank them and remove the bad ones (Natural Selection).
The best one are kept to make something that can be better: joining, crossing over, adding some randomness (mutations) to avoid exact replications.
When is it ready for production
Never… it can always been improved but are restraint by some other aspect (speed, size, time)
- Add logic to prevent the dumb algorythm to raise.
- Robot may really do things to your computer such as moving the mouse away all the time.
The Cost of Logging
Based on real life Performance consultancy report. The goal was to reduce latency and number of server.
The more you log, the better, it allows to diagnose the application and easier to debug. But there is a thread of.
Logging a line cost 0.22 ms and most of the time, you not only log, you transport the log (json, socket, write to disk): That’s the adapter problem.
Use the pino logger
Logging 10000 lines -> 250ms And in extreme mode it bufferise most of the stuff before sending: 10000 lines -> 125ms
How: using epoch instead of date, flatten string flatstr, fast safe stringify, quick-format
Transport is held in a separate process,
Instrumentation and Tracing in node.js
by Thomas Watson @wa7son
Traces are left all along the transaction from the incoming request to the final response. What if something goes wrong in one part of the program.
To intercept everything the idea is to patch every core operations (setTimeout, …) that’s complex. But there is AsyncHooks.
console.log() is getting into an infinite loop with AsyncHooks, use process._rawDebug() instead which is synchronous
Handles are used to transfert things from the JS world to the C/C++ world.
Redis doesn’t work with this but there is solution ongoing.