In my sixth year with Rails the Ruby community has been experiencing a sea change: gem innovation has peaked and leading Rails developers have been moving on to JavaScript 6 Node, Dockerized microservices in Go, functional Clojure on the JVM or highly concurrent Elixir Phoenix on the fault tolerant Erlang VM. Despite these defections Ruby adoption has resurged (as, coincidentally, has Java) given new versions that have improved garbage collection and performance. Rails 5 has embraced concurrent-ruby and websockets, while HTTP/2 looks set to drive further significant change to all web application frameworks in the near future.

Meanwhile our in-house multi-currency e-commerce solution has grown to 45kloc, 2,100 tests and nearly 100 gem dependencies thanks to ever expanding integration with fraud detection services, fulfillment providers, payment gateways, email marketing solutions and an autodialer. Between daily business requirements we have been addressing technical debt by straggling to Ruby 2.2, Rails 4.2, RSpec 3.4 and ActiveAdmin 1.0.0.pre2. We have adopted CBRA and gemified the persistence core of the application. New gems adopted have included VCR, Wisper and React on Rails.

On the tools side this year has seen widespread adoption of Slack for distributed teams, but we have stuck with Atlasssian HipChat, configuring integration with Airbrake for production exceptions, Assembla for tickets, BitBucket for code reviews, CodeShip for CI and Coveralls for test coverage. Increasingly global sales have raised expectations of operations to provide 24×7 uptime, monitored by PagerDuty and requiring MySQL hot backups.

Why Docker? Because it promises to make dev-ops easier. Dockerizing an existing production environment is not trivial, moving development to Docker is perhaps a useful stepping stone. It might also make developer on-boarding easier but I haven’t tried that yet.

Docker can be installed on OS X in various ways, most notably Docker Toolbox or homebrew. I was previously using docker-machine with Parallels but recently I discovered Dlite which simplifies using docker by creating a local OS X docker.sock that forwards requests to the docker VM. Dlite also sets up a an NFS server that exports /Users to the VM. Instead of VirtualBox or Parallels Dlite leverages a lightweight hypervisor called xhyve which requires Yosemite and hardware virtualization support (sysctl kern.hv_support). Note that as of writing Docker for Mac and Windows in public beta is using a similar approach.

With Dlite started I can get a usable bash session with:

docker run -v /Users:/home -it ruby:2.2 bash

I can then cd to the project directory and bundle.

To run browser based tests I was previously building a docker image with Xvfb, Chrome and Selenium Server installed. Recently, however, I found a different approach that uses standalone selenium-server as another docker service. On a Mac workstation installing selenium-server as a service is as simple as

brew install selenium-server-standalone chromedriver
selenium-server

Once tests are passing with a local selenium-server then adding selenium/standalone-chrome to docker-compose.yml and running browser based tests inside the container is quite achievable. The location of the selenium-server is exported to the app container as SELENIUM_PORT, eg. tcp://172.0.0.5:4444

Wednesday. Day 1. Keynote introduction by Jeremy Daer, nee Kemper, DHH thankfully racing. Heroku presenation on Kafka, Shopify presentations on shipit-engine, request_interceptor and BuildKite, Rafael França on Sprockets, and well attended presentation by Akira Matsuda with ideas for assorted Rails performance hacks. Closing keynote from Nickolas Means with a history of Lockheed’s original skunkworks.

Thursday. Day 2. Ruby Heroes including Avdi Grimm, Charles Nutter, Akira Matsuda and Richard Schneeman. Surprisingly few vendor booths, including Indeed Prime, Hired, Skylight, RubyMine, Stitch Fix and Codacy. Rails 5 presentations on ActionCable, Turbolinks and Sean Griffin on miscellaneous enhancements. Richard Schneeman on maintaining Sprockets and David Copeland of Stitch Fix on emergent software architecture. Lightening talks on Scrivito, RocketJob, Appfolio Farcy, Railscamp, Bundler Gemstash, multi_zip, Brakeman, RubyBookClub, Reinteractive Wallaby and Benjamin Fleischer on JSONAPI.

Friday. Day 3. Aaron Patterson, now at GitHub, workshops on Hashicorp tools and RedPotion and presentations on Percy and Sandi Metz on refactoring. Paul Lamere from Spotify with a wonderful presentation on music data analytics… in Python.

P.S. I never guessed the most interesting presentation on Rails would be by Justin Searls.

In year five with Rails I finally started contributing to high profile open source projects (ActiveAdmin, ActiveMerchant), albeit with mixed success. Understanding complex libraries written by others reacquainted me with debugging in RubyMine. In addition I published the first gems of my own using jeweler, taking my public GitHub contributions over 200 for the year. Working on OSS exposed me to Colloquy IRC and Travis CI. I contributed slightly more to StackOverflow and project related Google Groups.

At my employer our ‘mature’ Rails codebase has grown from 20 to 30KLOC, monitored soberly by Coveralls (87%), CodeClimate (2.8), Codeship CI, Rubocop, Gemnasium and Bliss. I remain committed to SublimeText, now with GitGutter and SublimeGit, and to RSpec, influenced heavily by a RSpec workshop at RailsConf 2014. New gems adopted this year included (tentatively) interactors, oink and thor. Documentation has been mildly improved using YARD.

After ten years of development Rails is a mature framework, and a recurring theme at conferences is how to build and maintain Rails applications that are increasingly complex. Rails achieved early success as an opinionated framework that emphasizes MVC and convention over configuration, but as developers have tackled ever more ambitious projects there has been divergence as to how to structure them.

What is generally agreed is that business logic should reside in models, away from controllers, views and API implementations. This has led to ‘fat’ models and discussion of strategies to manage them such as this blog post from CodeClimate titled ‘7 Patterns to Refactor Fat ActiveRecord Models‘. An important point often missed when first learning Rails is that app/models is not intended exclusively for classes persisted with ActiveRecord as it may include other domain objects. In addition to domain objects other popular patterns used to structure code include form objects, presenters, decorators and service objects. Presenters and decorators intermediate between controllers and views, whereas service objects intermediate between controllers and models.

Presenters and decorators are probably less controversial in the Rails community and have done much to replace view helpers. Service objects, however, have been disparaged by DHH in favor of concerns, but many developers with a traditional OOP background are unconvinced by his approach. One example project that uses service objects is GitHub itself. More recently Collective Idea have published an ‘interactor’ gem that elaborates this approach.

These strategies all result in more classes, and probably the first step in managing a codebase with many classes is to introduce namespaces. The Rails class loader automatically searches sub-folders matching namespaces so source files can be arranged similarly to the namespace hierarchy. Given most developers nowadays navigate code using global search or ‘jump anywhere’ features in their text editor the neater project directory layout is of limited benefit, but namespacing can reduce name conflicts with libraries or between different areas of an application.

The next step beyond namespacing, suggested by Stephan Hagemann is to divide a project into internal engines or gems, an approach labelled Component Based Rails Architecture. An example project is provided at https://github.com/shageman/the_next_big_thing Each component has its own test suite so the scope of regression testing of any feature change is more limited. If a component proves reusable it can be split out from the original project altogether.

Beyond a component based architecture the final step to scaling a complex application is to break it apart into separate services interacting with an internal API, an approach referred to as Service Oriented Architecture (SOA). I will defer comment on the benefits and challenges of such an approach to a later post.

See also

• Railsconf ’14 presentation ‘Concerns, Decorators, Presenters, Service Objects, Helpers, Help Me Decide!’
• Railscast on Form objects
• Railscast on Presenters
• Railscast on Service Objects