jesusgollonet / rde Goto Github PK

There are 3 main types of secrets that need to be managed:

• an ssh key to get into the instance
• access to github (could be the ssh key or github access tokens)
• API keys for projects.

First idea is to try to use one password cli (op) for as much as we can, but otherwise can look into doppler or other managed services if needed.

• Docker
• Node.js
• Terraform
• AWS CLI
• SSH key
• packer
• jq
• op (1password)
• zsh
• fzf.sh
• z
• pure prompt
• gh

The ami id is hardcoded in instances/variables.tfvars. Since the AMI id is
expected to change while we iterate we should parameterize the value

This should be executed on a GH action instead of my local machine. That way I
can ensure it doesn't rely on any magic knowledge already available on my
machine (ssh keys, AMI id...)

At the moment we're creating an AMI with a given hardcoded name (rde-ami).
To make the script idempotent, we want to make sure that image is deleted if it
exists.

Explain:

• [] What this is
• [] What problem it solves
• [] Structure (directories, tools)
• [] How to run
• [] Status

The end goal is to have a cloud machine that I can spin up and down on demand
with all my tools and configuration ready. Some thoughts:

• I should be able to turn instance on and off, ideally from the ipad (so can't assume cli). I wonder if this can implemented intuitively with github actions, or maybe a lambda and a simple web form.
• On accessing, I should be able to access any gh repos, machines... etc that I'm working on. We could take this a step further and play with the idea of pre-built repos from gitpod.
• All state should be kept outside of the image. This includes secrets (ssh keys, API keys...), assets. I should be able to recover state at any point.
• Image configuration should ideally be automated, as in if I install a tool or change something in my machine, this should trigger a packer build so it's ready for next time.

At the moment I'm working with local state, but even at this stage it's proving
inconvenient as by definition this project needs to work on different machines.
Move state to Terraform Cloud