Recently I rolled out a Ruby-on-Rails/Spree-based webshop on the budgethoster Site5. I thought to share some gotcha's, reasoning around this project. To debunk the idea that hosting for Rails is either complex and hard, or done on Heroku. To explain a bit about Spree as a good option for your e-commerce (or not) and to go a little into how I modified Spree.
My wife makes custom bags, smartphone- and tablet sleeves, and all sorts of leather, handcrafted jewelery. Obviously she wanted to sell some through the internet; So I made her a webshop.
Reasons for choosing for Spree
I decided to go for Spree, after some investigation. A few important reasons where:
- It's catchphrase It was designed to make customization and upgrades as simple as possible.
- It is built in Ruby on Rails, and that is what I do. (Although I probably know more about PHP, and Drupal, at that).
- We wanted a simple shop. Her needs are mostly simplicity; less features in the shop equals better. Spree promises that; as opposed to most ready-to-go shopping tools, that promise every feature you may wish. And get in your way.
- We needed flexibility. Simple means that it offers less options and choices. But also that the business-logic needs to handle more.
- I wanted the site to be ready for mobile and thus to be responsive. Rails is ready for this. HTML5 and all that. So is Spree.
Some reasons why I had thoughts about either writing my own e-commerce or going with a PHP-solution were:
- I had (and have, read on) my questions about how true the catchphrase It was designed to make customization... as simple as possible was, in reality.
- Spree is Rails. Rails requires professional hosting; not a problem for a project developed for over 1k; where hosting at $20 per month is peanuts. In this case $20 per month is far over budget; again, the insides of her exact business-choices are out of scope for this project. Good-enough PHP-hosters come for under the $5 per month.
- I was going to build it for free, obviously. So we could not afford me spending hundreds of hours on tweaking some tool, CSS and whatnot. A turnkey-solution like Magento would be a faster solution.
- We chose the Dutch bank and payment provider Omnikassa because they simply offered the best deal for her. There was no omnikassa payment plugin for Spree, but there were some for more famous frameworks.
I went for Spree after proof-of-concepting my two biggest challenges: offsite payment and (Zurb-foundation](http://foundation.zurb.com) as CSS/HTML framework, instead of the default Skeleton, came out quite alright. And a 30 days trial on Site5 proved me that hosting there was going to work out.
I replaced most views with my own HTML, so that I can use Zurb foundation. It offered just a little more features, such as a slider and more advanced responsive features; like hiding entire subtrees of elements on certain devises.
In the end, we decided to go Desktop-first; I am re-doing some of the views now, so they are prettier on mobile too. The reason we put this lower in our todo, is that the most-used payment system in the Netherlands, iDeal, trough our PSP omnikassa, is not mobile-friendly in the first place. So why offer an entire mobile-webshop when people cannot pay (properly) on a mobile?
For under €5 per month you get a server with git, ssh-access and Passenger to host your Rails application. A few Euro more for a static IP address, which I need for the SSL-certificate. It is an e-commerce, you need HTTPs.
The only problem I had was some asset-precompiling issue, where the compiler just died on me. After a support-call, the Site5-engineers upped some memory on the server and I could compile just fine. It turned out that some spree plugin came with several hundreds of (demo) asset-files like huge videos that needed to be "compiled" by the asset-pipeline. Cleaning up Spree Slider and removing its assets fixed the issue for good. But hey, I don't expect a budget-hoster to support me compiling hundreds of megabytes of video and other demo-stuff. Fair enough.
Also, their uptime is reasonable. Not enterprise-alike reliable (IMHO), so I
have nagios to check the
frontpage every several minutes for certain strings to occur (It looks
if the words "Anna Treurniet" occur in the
<title>). Every odd
month (or so) there is a restart or some short downtime. One time Site5
changed the MySQL-setup (location of the socket moved to elsewhere) so I
had over half a day of downtime until solved. And now and again the
application gets shut-down for no clear reason, so it needs to boot,
resulting in the webshop loading very slow for one or two visitors. The
nagios checking actually kind-of solves this too, since it acts as a
"visitor" opening the site every five minutes.
All in all, I am very happy with this host. It offers far (far!) more then one would expect from a €5/month. In my "enterprise"-jobs, I have to deal with €500-and-up-per-month hosters who have far worse deals, support, deployment and uptime.
Their absolute biggest downside is the way their bulkhosting environment holds them back from upgrading to Ruby 1.9.3. So, if you depend on a reasonable recent Ruby-version, bad luck.
In order to keep the deployment smooth and somewhat close to the
experience I have with Heroku, I use git-deploy. Git deploy consists of a few simple hooks that run on the server in post-receive hooks. So, after you push your changes to production with
git push production, the server runs a few commands, like (when needed) database-migrations, assets and cache-refreshing and then a restart of the Rails application. I have used this for other, PHP-based systems too.
Some problems, as mentioned above with asset-precompiling aside, it
works like a charm.
Obviously, when using git to manage the deployment, you need a good branching and releasing management. With that in place, I can fix and deploy changes within minutes. Yes. Minutes. Probably faster than most of you can log in over SSH, find the sourcecode on production, and hack a fix in with Vim.
Testing and TDD
Unfortunately, I did not manage to get good integration tests for Spree going. Most of the extensions lack any form of tests. Spree itself is is covered pretty well, but integrating Spree means changing configuration, overriding behaviours with Decorators. And I still have no idea how to tests these properly. The rest is mostly view-overrides, which often breaks Spree's own tests and requires me to rewrite all the spree-tests in my application. It mostly boils down to my inexperience with testing mostly, though.
A Spree Extension was written to allow offsite-payment with the Dutch payment-system Omnikassa. This is and was a mess. Spree had no (proper) support for offsite payments, so I had to hack into the entire checkout-workflow in order to get this payment-system going. The Omnikassa-extension somehow breaks the feature in Spree to allow discounts; it breaks certain orders in the back-end and whatnot. Spree 1.2.0 promises to have this checkout-workflow-inflexibility fixed, but an upgrade is rather hard, seeing all the customisations the application needed.
I don't think I will use Spree for a future e-commerce project. Despite its promise to be the most flexible solution, I found it making too much assumptions and being far too inflexible in areas. I'd rather roll my own, next time. The most important parts that Spree offers me, are either very easy to develop myself (products, categories, their views, content-management), not needed (credit card-handling, user-login, 3rd party statistics) or covered in solid Rails plugins (administration, editors).
Against small-things-made-hard, like changing the checkout-workflow (one-click-checkout?, offsite-payments), manually ordering the products on the frontpage, integrating a simple CMS for the few "static pages" and so on. Usually, in an Average Rails-projects these things take a few hours programming and deploying. Here they took me days of stepping through spree-core code in order to understand how my Decorator did (not) change some ordering or some menu-addition.
For me, Spree offers me too little benefits to overcome its downsides. Despites its promise, it is very much a ready-made application, which works according to various assumptions about workflow, features and even looks, that you can configure and beat into shape; mostly. The very same conceptual things that Drove me from Drupal.
But overall, rolling a simple, Spree-site on Site5, gives you a well-designed, ready-to-go e-commerce environment for under €10 per month. And with a few hacks and tools, you can make the deployment to that host really easy. Whether Spree is the best solution for your e-commerce needs, depends on how much (and what exactly) you need to customize.