tchan2/packet-gpu

A first step example of creating an environment that allows for the implementation of deep learning in a GPU supported Packet server using Terraform, Docker, and Jupyter.

packet-gpu's Issues

Inheriting variables is broken

packet-gpu/packet-gpu.tf

Line 9 in 1cb1743

name = "Project 1"

The first run of the terraform will always create a "Project 1" in the user account, even when a variable is defined. Maybe need to figure out a single source of truth and define defaults there?

Handling package installation success check

I like that you're handling this:

packet-gpu/tf_userdata.sh.tpl

Line 33 in 3bb40a1

docker

and while the script will exit if docker isn't in the $PATH (as you intended), I'd recommend just adding some additional handling so when checking the cloudinit output, there can be some more robust messaging, like:

if ! type docker; then echo "Docker not installed..."; exit 1; fi

this also allows you add some retry behavior down the road if you wanted for the other dependencies as well.

Consider making `facility` variable a list

I'd recommend making the facility variable, itself, a list, rather than providing it as a string into a list (now capped at that single value for length):

packet-gpu/packet-gpu.tf

Line 20 in 3bb40a1

facilities = ["${var.facility}"]

so something like this:

packet-gpu/variables.tf

Lines 23 to 26 in 3bb40a1

	variable "facility" {
	description = "The facility your Packet server is in. This information can be found in the 'Location' part of your preferred server's details."
	default = "dfw2"
	}

could be expressed like:

variable "facility" {
   default = ["dfw2"]
}

and then referenced like:

   facilities = "${var.facility}"

so even if you only plan to use a single facility, this project can scale if there are more GPU-equipped facilities one would like to use.