Deploy Content to Multiple Servers (EC2)

Question

I’ve been working on a cloud based (AWS EC2 ) PHP Web Application, and I’m struggling with one issue when it comes to working with multiple servers (all under an AWS Elastic Load Balancer). On one server, when I upload the latest files, they’re instantly in production across the entire application. But this isn’t true when using multiple servers – you have to upload files to each of them, every time you commit a change. This could work alright if you don’t update anything very often, or if you just have one or two servers. But what if you update the system multiple times in one week, across ten servers?

What I’m looking for is a way to ‘commit’ changes from our dev or testing server and have it ‘pushed’ out to all of our production servers immediately. Ideally the update would be applied to only one server at a time (even though it just takes a second or two per server) so the ELB will not send traffic to it while files are changing so as not to disrupt any production traffic that may be flowing to the ELB . What is the best way of doing this? One of my thoughts would be to use SVN on the dev server, but that doesn’t really ‘push’ to the servers. I’m looking for a process that takes just a few seconds to commit an update and subsequently begin applying it to servers. Also, for those of you familiar with AWS , what’s the best way to update an AMI with the latest updates so the auto-scaler always launches new instances with the latest version of the software?

There have to be good ways of doing this….can’t really picture sites like Facebook, Google, Apple, Amazon, Twitter, etc. going through and updating hundreds or thousands of servers manually and one by one when they make a change.

Thanks in advance for your help. I’m hoping we can find some solution to this problem….what has to be at least 100 Google searches by both myself and my business partner over the last day have proven unsuccessful for the most part in solving this problem.

Alex

Answer 1

We use scalr.net to manage our web servers and load balancer instances. It worked pretty well until now. we have a server farm for each of our environments (2 production farms, staging, sandbox). We have a pre configured roles for a web servers so it's super easy to open new instances and scale when needed. the web server pull code from github when it boots up.

We haven't completed all the deployment changes we want to do, but basically here's how we deploy new versions into our production environment:

1. we use phing to update the source code and deployment on each web service. we created a task that execute a git pull and run database changes (dbdeploy phing task). http://www.phing.info/trac/

2. we wrote a shell script that executes phing and we added it to scalr as a script. Scalr has a nice interface to manage scripts.

   #!/bin/sh
   
   cd /var/www
   phing -f /var/www/build.xml -Denvironment=production deploy
   

3. scalr has an option to execute scripts on all the instances in a specific farm, so each release we just push to the master branch in github and execute the scalr script.

We want to create a github hook that deploys automatically when we push to the master branch. Scalr has api that can execute scripts, so it's possible.