This sets up Terraform, takes temporary slip notes, and allows me to experiment with the system while I work through the documentation and setup the data lab.
The first example was to setup a basic instance. I created one, changed it, and destroyed it. I confirmed the system was in the state Terraform reported in the AWS console between steps.
I hadn't used the AWS CLI before, and I didn't have useful I AM Roles available. I created one, david, with programatic access and permissions for a data scientist. That may do the job. The information for all of that is stored in LastPass.
One thing that was surprising was that I would need to know the location-specific AMI before I could deploy an EC2 instance. It makes sense, but it's a tighter-abstraction to what I'm building than I thought. What that tells me is creating documentation, slips, and storing them in a git repo is critical.
I'm going to be using a GPU on AWS for training my code. That means I'll create local port forwarding over SSH to connect port 8888 to a P2 instance that is ideal for building a deep learning module. In order to use a P2 instance, a limit increase must be setup. That's been done with Case ID 6728534471.
As I work through more of the Terraform documentation, I'll keep this in mind so I know how to store/use/lookup whatever I need.
I made a mistake, and need to get P instances in us-west-2 instead of us-west-1. They gave me the resources, but in the region I asked for, the wrong region. The new case is 6730418791.
I created a separate directory for the dependencies, because I wanted to compare and think about things. This is not normal practice. Git is enough for that. Also, the terraform init command must be called in the new directory, which brings down a binary file for the provider. For AWS, that's 158M. Also, this confirms that I was right to ignore the .terraform directory in git.
There was an error setting up the Elastic IP address, which I think is tied to the permissions granted a data scientist with the I AM role I used, as that is not what a data scientist typically does. Either way, I destroyed the instances and left a clean slate once I got that far.
- How can I stop or hybernate instances from Terraform?
- How can I interpret state to mean, "all shut down, you can go home now?"
- How can I integrate cost management and limitations with my practice?