Sending Passbook Push Notifications with Houston and Sidekiq

The Nomad collection of iOS/Ruby tools are a great resource. I recently switched how Goldstar is sending out push notifications from Grocer to Houston.

Grocer has gone unmaintained for most of 2014, and it was obvious how to recover from a problem we were having with it where Apple would close the persistent connection we were using to send notifications, and so any notification sent after that would be lost. So I used Josh Symonds’ technique (with some tweaks I’ll mention later) to send them in a more robust way.

The one thing that Grocer does out of the box that Houston didn’t that we really needed was a way to send Passbook Push Notifications. These just like regular Push Notifications, however they are totally blank, sent to a token the pass registers with your server, and only work against the Production push notification gateway, using a different SSL certificate than your standard cert. This causes some problems because Houston assumes the Rails development environment should send pushes against the dev apple gateway.

Continue…

Expanding on the Law of Demeter

Recently, my friend Avdi Grim posted a great explanation of the Law of Demeter (LoD). It’s been tweeted and posted all over the web, so hopefully you saw it and read it. I’m totally in agreement with his conclusions in that post, and LoD is a code smell I see in every MVC project I have ever worked in.

I’m going to pick on Avdi’s post just a little, but this is more-so criticism of every LoD explanation I’ve ever read. I think all of these explanations of LoD are actually quite contrite and not addressing the real problem. Avdi (and others) seem to focus on “Structrual Coupling” or exposing the “internal” (actually external*) structure of your objects to other objects that don’t need to know it.

An Example:

def user_info(user) "Name: #{user.name}. Boss: #{user.department.try(:head).try(:name)}" end

Avdi does a beautiful job illustrating why this code is a problem but I want to go a little deeper.

The more insidious LoD violation see in codebases is what I’d call “behavioral” LoD violations, or, when the behavior of a class relies on the behavior of objects that are distant in the object graph.

My example needs a little setup. Lets say you have Vehicles which has_many Platforms which has_many Models which has_many Versions (and Platforms has many Version through models). Pretty simple 4 level deep object graph that a car manufacturer might have. A very common set of objects you might want is to know which Platforms are actually in production. Some our outdated, some are future, but you want all the platforms that are currently in production.

Now, I know how I’d do this, but I commonly see code like this:

class Vehicles
  def current_platforms(options = {})
    platforms = platforms.find(:all,
                        :conditions => 
                          ["platforms.on_date >= ? AND platforms.status IN ('open', 'cancelled') AND platforms.cached_inventory_status != ?",
                            Date.today.to_s(:db), 
                            Platform.cached_inventory_status(:no_inventory)],
                        :order => 'on_date asc, time_note asc',
                        :include => {:models => :versions})
  
    if options[:skip_some_special_case]
      platforms.select { |platform| platform.versions.any? { |offer| !version.special_case} }
    elsif options[:skip_some_other_case]
      platforms.select { |platform| platform.versions.any?(&:other_case?) }
    else
      platforms
    end
  end
end

This code is a great example of my problem with how most folks end up simplifying LoD violations. No where are we chaining methods “too deep” or as it gets simplified to are we “using too many periods”. We’ve only used 1 period (because of a has_many :through =>). However, in our if statements, we are tightly coupling our code to the behavior of a class that is 3 siblings away from the one we’re working in. Nasty. Never should a Platform do operations on, or even know about, a Version.

So why is my example actually worse than something.try(:something_else).try(:a_third)? Well, lets get back to that * I left on “(actually external [structure])”.

Most people seem to classify the active_record generated attribute accessors of Rails model classes as “internal structure” (Avdi doesn’t, I should state for correctness). These are most certainly external structure. You can override them and return whatever you’d like and attributes[:foo] is actually the “internal” accessor for all active_record classes. External structure, in every project I’ve ever worked in, has been the most consistent, least volatile part of the codebases. External structure almost *never* changes, and when it does, it’s typically *very* easy to change (it’s not even a refactoring, it’s just changing the messages exchanged between classes).

My example relies on behavior of a class 3 classes away. If there’s anything I’ve learned, it’s that behavior is always more volatile than structure, and as Avdi points out, LoD really starts hurting as code churns and whatever your LoD relies on changes and becomes more and more complex.

So the big question is, how do we fix this?

It’s actually a theoretically simple fix, and similar to the fixes in Avdi’s post: delegate the behavior down the object graph!

class Vehicle
  def current_platforms(options={})
    if options[:skip_some_special_case]
      platforms.reject(&:special_case?)
    elsif options[:skip_some_other_case]
      platforms.select(&:other_case?)
    else
      platforms.select(&:regular_case)
    end
  end
end

class Platform
  def regular_case
    ...
  end
  def special_case?
    models.special_case?
  end
  def other_case?
    models.other_case?
  end
end

class Model
  def special_case?
    versions.special_case?
  end
  def other_case?
    versions.other_case?
  end
end

class Version
  def special_case?
    ...
  end
  def other_case?
    ...
  end
end

Now, Platform knows nothing about versions, versions can easily change what defines the “cases” without any changes to Platform and ironically, when I made this refactoring to the original source I got this derived example from, it’s actually *more* efficient. Removing the explicit find that started the original method means that if we’ve already loaded this object graph, even outside of this method, we can keep reusing it without going to the database.

I hope this helps, and for the love of Demeter, can we stop describing it as “using too many periods”?

Pro Tip: Encapsulate logic in Rails Views for maintainability

A Rails view anti-pattern is that they very quickly turn into a nasty tangle of conditional html blocks.

Example:


<% if logged_in? && @user == current_user %>

Simple example, but what is the intention of that code? Why are you really checking for? Give that state a name.

Change that to:


def viewing_own_profile(user)
  logged_in? && user == current_user
end

<% if viewing_own_profile(@user) %>

Now you can test that helper method without resorting to rendering a view and parsing it to check if elements exist, the state you care about is named and the intention is clear.

From Zero to Rails Hero in 11 Easy Steps (Installing Rails on OS X 10.6.5)

This is the easiest, fastest and best-to-develop-in way to get a rails dev stack up and running on OS X 10.6 (Snow Leopard).

I actually find that it’s pretty easy, if everything is done in the right order.

  1. Install XCode
  2. Install Homebrew
    $ ruby -e "$(curl -fsSL https://gist.github.com/raw/323731/install_homebrew.rb)"
  3. Install git using homebrew
    $ brew install git
  4. Install mysql using homebrew
    $ brew install mysql
  5. Install rvm (and follow the instructions to set up bash and don’t forget to open a new terminal window)
    $ bash < <( curl http://rvm.beginrescueend.com/releases/rvm-install-head )
  6. Install Ruby Enterprise Edition
    $ rvm install ree
  7. Make REE the default Ruby
    $ rvm use ree --default
  8. Tell rubygems to not install rdocs
    Put the following lines in ~/.gemrc

    install: --no-rdoc --no-ri
    update: --no-rdoc --no-ri
  9. Install Passenger
    $ gem install passenger

    Now follow that up by running the install script (and follow the directions)

    $ ./passenger-install-apache2-module

    At the end of the install process it will give you instructions to add 3 lines to the “Apache configuration file”, but it won’t tell you *where* your apache config file is. It’s at /etc/apache2/httpd.conf

  10. Restart Apache
    At this point, if you’re using 10.6.5, you might do a sudo /etc/apachectl restart and run into the following error:
    /usr/sbin/apachectl: line 82: ulimit: open files: cannot modify limit: Invalid argumentTo fix this, edit /usr/sbin/apachectl and change ULIMIT_MAX_FILES to an empty string.

    Now you can restart apache.

  11. Install PassengerPane 1.3 www.fngtps.com/files/2/2009/09/PassengerPane-1.3.tgz

BOOM! Done.

Well, for me, I had to `gem install isolate` and Rubymine and then I was done, but you get the idea. Just load up the Passenger Preference Pane, point at site at a rails project on your hard drive, and load it up in your browser. Assuming your rails app will boot, you should be good to go!

Bundler: Oh the fail I know.

First, a disclaimer: I don’t want to be negative. I am only writing this because I feel it deserves attention by the community and because I love Rails so much that the fact that something is making me lose faith in it truly makes me a little sad.

I first started using Bundler 0.7.6 in December ’09 in a Rails 2.3.5 project. We also use Bundler (not sure what version we started with) in our main Rails 2.3.x app that powers www.Goldstar.com. In the 3+ months that we have been using bundler, it has, by far, been the largest pain point and time sink of any part of our toolset to the point that today, I proclaimed on twitter, than in it’s current state, I can’t recommend Rails 3.0 to anyone starting or upgrading a product.

Now, if you know me, you know that this is pretty out-of-left-field. I’m a HUGE Rails evangelist. And I truly believe that it’s right for many projects in all kinds of scenarios.

Anyway, on to the fail. . .

Bundler 0.7.x had lots of issues, too numerous to count. However, it did have the option to pass a yaml file to it via –build-options that would allow you to pass a few arguments into the various gems it needed to build. This was good enough for us, and we followed the convention we use with database.yml to have all the correct options in the various environments we use to dev/deploy our rails apps.

Builder 0.8 soon followed suit (I believe in January) and messed us up by changing the path inside the bundler folder that all the gems where kept at to be name-spaced to what version of ruby you were using. I’m not sure what this was meant to fix, but it broke a few things for us and it broke Rubymine (my personal editor of choice). But hey, we upgraded.

I’m going to leave out all the pain that was causes by upgrading to 0.8 and just say that we are currently using Bundler 0.8 in dev and production, except that a bug, where older version of already cached gems won’t upgrade unless you rm -rf the bundler directory (json-pure-1.2.2 to json-pure-1.2.4 fwiw) is breaking our deploys to production. This means that every deploy to production, we can’t be sure the bug isn’t going to break it, so we must rebuild every gem in the app (of which there are MANY) every time we deploy. This just isn’t going to work for us.

Now, we’d like to move to Bundler 0.9 because it promises to fix our issue (bundle install correctly updates gems as required). However, –build-options has been removed from 0.9. Because –build-options was removed, bundler has moved to a don’t-vendor-some-of-your-gems solution where it will use gems from the system if they aren’t in the bundle. This allows you to pass different incantations to `gem install` for different gems (mysql, pg, etc) and it will use those already built gems when you later do `bundle install`. But that means we’re back to managing gem installs on all of our different systems in different ways, which was one of the few things that Bundler did significantly better than config.gem!

Our systems developer, Steven Baker, is currently working on re-implimenting a solution to the –build-options issue, which he first ran by “Carlhuda” in hopes that they might include it and we’re not going down a road that will just end in us playing by ourselves and having to re-impliment our Gemfile or build_options.yml file for whatever solution is, but this brings me to my point:

Rails 3.0, because it relies on Bundler, is not in anyway stable/useable/suggested unless you want to pretty continuously fix/re-impliment/cry yourself to sleep because of Bundler. Carl L. has posted a roadmap which shows at least 2 more revisions as well as closing 20+ currently open issues. I have no idea what the issue velocity is on the project, but to me, it looks like it’s in the months and not days or weeks before Bundler is going to be stable. And, as with every version before it, there have been major downsides to the implementation, that have forced pretty major breaking changes. In other words, every version of bundler is a new experiment (not to mention the various gem management techniques that merb used which were also a source of headaches), and I have little confidence that just slapping 0.9 or 0.10 is going to change that.

Bottom line: A significant piece of Rails 3 is still in the “spike” phase, and shows little to no promise that it will become solid enough to trust for _months_.

Please skip the “Patches are accepted” (we’re working on one kthxbye) and “Why are you even using bundler?” (troll). I’m not here to indict or blame anyone, just to report our experience and clarify why I’ve lost significant confidence in Rails because numerous people have asked.

Sidenote: I’m putting this at the bottom, because it’s secondary to the point of the post, but why, precisely, do we *have* to use Bundler to make Rails 3 work? Who made this decision? If it was the same people that wrote Bundler, doesn’t anyone see a problem with that (NIH)? I’m assuming there were discussions in the rails-high-council private campfire room at some point, but obviously us peons aren’t privy to those. Config.gem wasn’t the easiest thing to use, but I’ve successfully used it for close to a year with very few issues and had it’s eccentricities worked out pretty well. To have it completely removed from Rails 3, and replaced by an experiment, seems a bit premature.

Rails App Template

I start a lot of projects. More than I’d like to admit. I finish and deploy very few, but I like cranking out the first iteration of an app in a weekend. Sometimes I do it just to learn some new feature, or try out an architecture idea. I don’t like to do this in the context of whatever existing app I’m working on, because I don’t want any baggage of whatever I was thinking last week. Last week is too long ago.

Continue…