notes from @ajsharp

gistfile1.txt

twitter mobile engineer

async requests
kestrel: scala port of starling

## sms code

* twitter mobile site was first deployed with unicorn
* twitter mobile site now uses rainbows (for concurrency reasons)

    class Tweet
      include Validatable
      include Objectification
    
      exposes :id, :name, ..., :more_attrs
    
      objectifies 'user' => User,
        'created_at' => lambda { |val| Time.parse(val) }
        'place' => Place
    end

## Scaling

* frontend cpu
  * use yajl for json
    * order of magnitude faster than ruby C json library
  * use gd2 for images
* object allocations
  * be careful with object allocation, GC is expensive
  * temp string creation
* prioritize concurrency
  * i/o bound (i.e. database)
  * 1.8 threads are cheap
  * typheous is GIL friendly
  ** i.e. it wont pause your threads while waiting for api requests


## explicit block i/o

    # all of these calls return proxy objects
    @user = find_user
    @trends = current_user.find_current_trends
    @tweet = current_user.find_tweet

    # drop down to typhoeus and do requests
    current_user.fire_requests!

# 3. Best Practices

company culture is like cement

small agile teams: 1pm, 1 designer, 3 devs

iterative development
continuous shipping
* avg 3 deploys a day
* 10-30 branches each
* test suite alone would take about an hour to run, but tests are distributed
  across many servers and run in about 2 mins
* write a feature in the morning, ship in the afternoon
* weekly iteration planning
* weekly internal releases
* strong belief in generalist developers
* specific testing practices are left up to the team, though unit testing is
  highly encouraged
* little-a agile
* pair programming not company policy, left up to team, but highly encouraged
* no manual QA team clicking on buttons
* developers "own" their products