Skip to main content

Project Bootstrap Guide

When we're starting up a new project, often there is a period of time where the design and research team is doing their initial work, we don't have any infrastructure set up, and engineers want to do some initial experimentation. This guide is intended as a resource for infrastructure engineers bootstrapping a new project from scratch. Not all projects will need all of the elements here; for instance, if you're working in an existing organization's AWS accounts or git repos, then you won't likely need to set up your own.

GitHub and Git Repos

The first thing you will likely need to do for your project is create a number of GitHub repos and possibly a new GitHub organization. This is the prerequisite for nearly everything else you need to do to set up your infrastructure, so doing it as early as possible is a good idea.

  • If your client does not have an organization for you to do this in, or if they would prefer you create a new one, then you will need to create a new GitHub organization. GitHub has docs that explain this process; you will probably want the "GitHub Team" product, and use the Truss admin credit card to handle this cost. It's also a good idea to make sure the leads for your project are owners, and probably your contact at the client organization as well. This will allow them to easily add or remove new members if necessary.
  • Once you have a GitHub organization, you will need to create a number of repositories:
    • myproject-infra - This is a repo which you will use to hold the Terraform code used to configure your project's AWS (or other cloud service) account(s).
    • application repo(s) - You should talk to the AppEng lead to see how they would like to organize the application and create repos they can use early on. Some projects, like MyMove, may want a single repo for all the application code, but other projects may want separate repos for the frontend and backend, for instance. Neither of these is inherently correct, so consult with the AppEng lead (if you have one) to make this decision.
    • myproject-infra-gov - If your project is going to need to have resources in GovCloud, as many of ours do, you will need another repo for the Terraform used for that. This is for two reasons; one, we generally want tighter permissions on who has access to this repo, and second, if we want to have Atlantis handling the infrastructure in these accounts, we need a separate repo.
  • If at all possible, you should be using Terraform to maintain the users, teams, and repos for your project. The catch being that you'll need your myproject-infra repo first.

1Password

It is highly likely that you will want to get a 1Password team for your project. This will allow you to store things like passwords and MFA codes for AWS accounts, GitHub "robot" users, and anything else that you need to give members of the team access to.

You can create a new account by going to https://1password.com and clicking on "sign in" in the upper right hand corner, and then instead of clicking on an existing account, click on "create a new account" at the bottom of the page, then click on "for my team" on the next page, and follow the creation dialogue. Use the Truss admin card for the charges.

You should only need to add people who actually need to access these credentials for the project to the team (since there's a charge per-user). This generally means all members of the infra team, the engineering lead, and the project lead -- but you can always add more people later if necessary.

Like the GitHub organization/repositories, this should probably be done as early in the process as possible, since you will want to store credentials for the other accounts you are creating.

AWS Organization and Accounts

Once you have a Git repo and 1Password set up, you will want to set up your AWS organization and/or accounts. Even if you don't have an actual application to stand up yet, you should get this bootstrapped so that things are ready to go.

If you are starting from scratch, you will need to open a new AWS account by going to https://aws.amazon.com and creating an AWS account, using the Truss admin card as the billing information. This account will be your org-root account that will start your AWS Organization. Once that is created, you should bootstrap Terraform in it and set up the admin-global namespace. Things you will want to set up in the org-root account:

  • AWS logs S3 bucket
  • IAM users for org-root users (infra users only)
  • AWS Config for the organization
  • AWS Cloudtrail for the organization
  • AWS GuardDuty for the organization

Once that is done, you'll also need to set up some other accounts using Terraform in the org-root account:

  • -id account for handling all your IAM users
  • -infra account for project-wide infrastructure
  • -dev account for your dev deployment of the application

You may want to set up others as well (-staging or -prod for instance), but these are probably the ones you want as soon as possible.

GovCloud

If you will need to be using GovCloud for this project, you should set up the parallel organization in GovCloud as well. See the (GovCloud Organization)[./aws/govcloud/gov-orgs.md] docs for an explanation of how to set this up.

If you will be working entirely in GovCloud, you will not need the -dev account in your commercial organization; you can leave your commercial organization with only the org-root, -id, and -infra accounts. These will be necessary because there are some things that can only be done in commercial AWS, such as public DNS.

Atlantis

If you plan to use Atlantis to maintain your Terraform (which we highly recommend), you should also consider setting this up early. We have an Atlantis doc that will help you set this up.

Setting up Atlantis early will get your engineers used to the workflow and encourage good Terraform habits from the start. This will mean less headaches later trying to figure out which changes are expected and which are not.

Placeholder service modules

Once the AppEng lead has identified, at a high level, the services they will be deploying -- such as a frontend, backend, and database -- you should start building Terraform modules to deploy these components. Even if, at this point, they are largely placeholders, you should start figuring out what you'll need to get these working -- things like ECS services, RDS instances, ALBs, VPCs, security groups, IAM roles, etc.

Even if there isn't much to actually put there, and the service will be dormant, working this out before you need to start demoing things for the client will help you work out any kinks and help developers get a tight feedback loop once they start working on the application in earnest.

CI/CD Pipeline

Setting up your CI/CD pipeline early is another piece of automation that will pay off in the long run. Doing so means that developers will get used to the idea that their changes will be put, at the very least, into a real running environment every time they push, which means that building in tests and operability are not optional components or things to be added later.

Truss tends to use either CircleCI or GitHub Actions for our projects; GitHub Actions are newer and as a result our patterns for them are not as well-developed, however they have a lower barrier to entry and you can use them on a limited basis for free even with private repos. It may also easier to use GitHub Actions in environments where clients are cool on using a SaaS product like CircleCI (since they are already likely using GitHub). Alternatively, you may need to use a CI/CD pipeline that the client already uses.

Regardless of which you pick, you should start wiring up the application repo(s) to your pipeline early, and making sure there is an automated deployment into the dev environment at the very least as close to the beginning of the project as possible. This means that your pipeline should at least do the following:

  • run pre-commit checks on the repo for each PR
  • run tests (and possibly output code coverage metrics as well)
  • build a docker image and push it to ECR
  • attempt to deploy the application, usually by updating the ECS task definition, and if at all possible, ensure that the correct version comes up and stays up, rolling back if it does not