Teaching vacancies - Encourage use of devcontainers for local development
Date: 2021-09-22
Status
Decided
Context and Problem Statement
Setting up the Teaching Vacancies app for local development is a frustrating process involving many steps and multiple dependencies driven by manually updated setup documentation. A workflow based on devcontainers would alleviate much of this setup pain, and provide a trivially reproducible environment for local development, benefitting both developers and non-developers on the team.
Decision Drivers
- Complex and time-consuming onboarding and "re-boarding" experience of the application
- Difficulties experiences by non-developers in getting the app set up locally, and getting it running again after major dependency changes (e.g. our recent addition of PostGIS)
- Increasing adoption of devcontainers as a de-facto standard in the wider development community including Ruby on Rails
- Possible use of cloud-based development environments such as Github Codespaces in the future to enable users on restricted organisation-managed devices to contribute to the application
Considered Options
- Provide devcontainer-based setup for Teaching Vacancies
- Do nothing and continue current local-based workflow
Decision Outcome
Add devcontainers as an option for now, with a view to iterate on it and improve it to the point where we can consider it the "official" default way of running Teaching Vacancies (while still allowing other development workflows for developers who prefer different ways of working).
Positive Consequences
- Drastically easier onboarding and "re-boarding" (e.g. on a new device or after an OS upgrade
causing developer tooling issues)
- Dependencies reduced to just Git, Docker, and VS Code
- A fully functioning development environment is ready in 10 minutes from scratch, with no user interaction beyond opening the repository in VS Code and selecting "Reopen in container"
- Moving entirety of development experience into a container fixes past Docker development workflow issues experienced on the team (where tasks and services where executed from the host instead of interacting with a shell and an editor from inside the container itself)
- Developers and other team members can develop on any host OS (macOS/Linux/Windows) but we only
need to support one single consistent environment
- Does away with all the Mac vs Linux vs WSL setup steps in our current documentation
- Reduces likelihood of "works on my machine" development environment issues
- "Leave no trace" on the host machine and complete isolation from other projects
- Removes possibility of "dependency hell" when working on multiple projects
- Removes need to clutter local environment with applications and dependencies that need to be
kept up to date and in sync (e.g. Google Chrome and
chromedriver
) - Removes need for language version managers (
rbenv
,nvm
)
- Provides executable documentation of project setup and dependencies
- Removes need for manually updated setup documentation that can become stale
- Experienced developers who have a different preferred workflow can get a clear, in-code view of setup steps and dependencies
- Good workflow for everyone, but excellent additional integration with Visual Studio Code
- Automatic passthrough of SSH and GPG keys
- Language extensions run within the container itself, and can be specified in the devcontainer configuration file for instant setup of useful extensions for new users
- Automatic bootstrapping of personal dotfiles
- Ability to easily move to cloud-based workflows in the future
- Ability to easily propagate new tools and improved configuration to all developers on the team
- Trivial rebuilds to a known good state when performing "dangerous" operations in the container
Negative Consequences
- Slightly reduced performance on some host OSs (non-Linux) due to Docker being Linux-native
technology (overhead of containers running in an intermediate VM)
- Somewhat mitigated by use of volumes for IO-intensive cache directories
- Can be worked around entirely by moving workspace root into the container, and we will continue to investigate before we fully agree on devcontainers as our default workflow
- Container layers need occasional pruning on the host as Docker can fill up disk space quickly
- Some duplication of Docker configuration between production and development configuration (but that is to be expected given that use cases are very different)