Now it’s time to deploy your app.
Because you are a hot new start-up, you will of course use AWS. You spend a few hours trying to figure out how many of which instance type you will need. You decide you need a load balancer, a bunch of app servers, a database server and a worker server. That should about cover you. Next up is Chef. You spend a few days learning about provisioning a new VM with Chef. You are so excited about how you can automatically provision a new app server when your start-up hits the front page of TechCrunch. You spend a week configuring and testing your Chef scripts. And then the big day comes. You deploy your application. Everything works the first time and you are overjoyed. First users start to appear and your architecture performs really well. You hit the front page of HackerNews and get tens of thousands of hits on your site. You find a small bug in your cache invalidation logic which you fix really quickly. You deploy your fix and what do you know? That script that installs your project’s dependencies when you deploy just pulled in a new version of a library that deploys your static files to S3. And of course that new version contains backwards incompatible changes and completely breaks your app. You are frantically trying to figure out what happened. You think, OK, I will put up a maintenance page while I sort this out. You try to SSH into your load balancer and try to change the routing settings. What a mess. What do you do when your database starts swapping? What happens when your Redis server runs out of memory and dies? You barely understand how to install MySQL, let alone how to get the most out of it.
Single command deployments. Fully-managed database service. Automatic load-balancing. On-demand scaling. No manual SSH to see what’s wrong.
A request comes in, the load balancer picks it up and sends it to a suitable
app server instance. The app server instance talks to a highly optimized
database instance. Redis is sitting on a super beefy machine with lots of RAM
and you never have to worry about it. When you are ready to make a deployment,
the load balancer bounces each app server in turn to make sure you don’t lose
any requests. All you need to do is
git push heroku master.
Sure, Heroku is a bit more expensive than EC2. But when you think of how many smart people are taking care of your app, it’s really cheap. It’s like having your own sys admin.
You need it.
Disclaimer: I’m in no way connected to Heroku. This posts is an attempt to convince myself that I shouldn’t be using Chef for my side-project-start-up-ideas. And I’m not a Ruby guy.
This article was first published on May 21, 2012. As you can see, there are no comments. I invite you to email me with your comments, criticisms, and other suggestions. Even better, write your own article as a response. Blogging is awesome.