Running Code on EC2 Instantiation

In our last post, we looked at the basic setup for creating an EC2 instance using CloudFormation. That was a great first step, but didn’t really do much of anything. We have some compute resources ready and waiting, but mostly just existing out in the cloud, waiting for us to connect via SSH and tell it what to do.

While that’s amazing (seriously, requisitioning a new computing resource in mere seconds is a modern technical miracle), having to perform additional, manual setup to make use of those resources is non-ideal. We’d want our servers to be instantiated and automatically provisioned to do what we want to do.

We have a couple of choices. First, because we have the ability to “ssh” into the instance, we could just write a script that creates the AWS stack, waits and polls until the stack has been completely created, and then connect via “ssh” to execute commands and install software (and, honestly, this isn’t an awful idea). Alternatively, we could just have all of that done by CloudFormation and cut out the unnecessary moving parts.

CloudFormation allows us to specify a UserData block that will be executed automatically after the server has been instantiated. This allows us to write scripts that provision the server with all of the software that we need for it to be useful.

Note that the script needs to be base64-encoded (ignore the “!Sub” for now). I’m also executing a Bash script. The distro of Ubuntu that I’m using also comes with “python3” (I’m not sure if that’s by default or if it’s installed by Amazon to drive the other tools that Amazon uses when setting up these instances), so I could have written a Python script. Speaking of Python, I’m using the built-in HTTP server as a simple demonstration of running software at launch.

If you wanted robust software that will automatically restart on failure or have other behaviors, you will want to look into your distro’s system services APIs.

Since I’m running an HTTP server on the instance, I’d probably like to actually see it from the outside world. In order to do that, I’ll need to update the security group and expose port 80 (the default HTTP port). With that exposed, I can look up my IP address from the stack Outputs field, navigate to that IP address in my web browser, and see my “hello, world!” message!

If you’ve left your AWS stack running since our last post, you might need to bring it down and recreate it in order to have your UserData block executed.

You might also note that I’ve set up a very simple Makefile that provides some convenient shortcuts to common commands. This makes it easy to create a stack using “make up” and then bring it down later with “make down”.