/dockerfile-rails

Provides a Rails generator to produce Dockerfiles and related files.

Primary LanguageDockerfileMIT LicenseMIT

Overview

Provides a Rails generator to produce Dockerfiles and related files. This is being proposed as the generator to be included in Rails 7.1, and a substantial number of pull requests along those lines have already been merged. This repository contains fixes and features beyond those pull requests. Highlights:

  • Supports all Rails supported releases, not just Rails 7.1, and likely works with a number of previous releases.
  • Can be customized using flags on the generate dockerfile command, and rerun to produce a custom tailored dockerfile based on detecting the actual features used by your application.
  • Will set .node_version, packageManager and install gems if needed to deploy your application.
  • Can produce a docker-compose.yml file for locally testing your configuration before deploying.

For more background:

  • Motivation - why this generator was created and what problems it is meant to solve
  • Demos - scripts to copy and paste into an empty directory to launch demo apps
  • Test Results - expected outputs for each test

Usage

Install from the root of your Rails project by running the following.

bundle add dockerfile-rails --optimistic --group development
bin/rails generate dockerfile

The --optimistic flag will make sure you always get the latest dockerfile-rails gem when you run bundle update && rails g dockerfile.

General option:

  • --force - overwrite existing files
  • --skip - keep existing files

If neither are specified, you will be prompted if a file exists with different contents. If both are specified, --force takes precedence.

Runtime Optimizations:

Build optimizations:

  • --cache - use build caching to speed up builds
  • --parallel - use multi-stage builds to install gems and node modules in parallel

Add/remove a Feature:

  • --ci - include test gems in deployed image
  • --compose - generate a docker-compose.yml file
  • --max-idle=n - exit afer n seconds of inactivity. Supports iso 8601 and sleep syntaxes. Uses passenger for now, awaiting puma support.
  • --nginx - serve static files via nginx. May require --root on some targets to access /dev/stdout
  • --no-link - don't add --link to COPY statements. Some tools (like at the moment, buildah) don't yet support this feature.
  • --no-lock - don't add linux platforms, set BUNDLE_DEPLOY, or --frozen-lockfile. May be needed at times to work around a rubygems bug.
  • --sudo - install and configure sudo to enable sudo -iu rails access to full environment

Error Tracking & Alerting:

  • --rollbar - install gem and a default initializer for Rollbar
  • --sentry - install gems and a default initializer for Sentry

Add a Database:

Generally the dockerfile generator will be able to determine what dependencies you are actually using. But should you be using DATABASE_URL, for example, at runtime additional support may be needed:

  • --litefs - use LiteFS
  • --mysql - add mysql libraries
  • --postgresql - add postgresql libraries
  • --redis - add redis libraries
  • --sqlite3 - add sqlite3 libraries
  • --sqlserver - add SQL Server libraries

Add a package/environment variable/build argument:

Not all of your needs can be determined by scanning your application. For example, I like to add vim and procps.

  • --add package... - add one or more debian packages
  • --arg=name:value - add a build argument
  • --env=name:value - add an environment variable
  • --remove package... - remove package from "to be added" list

Args and environment variables can be tailored to a specific build phase by adding -base, -build, or -deploy after the flag name (e.g --add-build freetds-dev --add-deploy freetds-bin). If no such suffix is found, the default for arg is -base, and the default for env is -deploy. Removal of an arg or environment variable is done by leaving the value blank (e.g --env-build=PORT:).

Configuration:

  • --bin-cd - adjust binstubs to set current working directory autocrlf enabled or may not be able to set bin stubs as executable.
  • --label=name:value - specify docker label. Can be used multiple times. See LABEL for detail
  • --no-prepare - omit db:prepare. Useful for cloud platforms with release phases
  • --passenger - use Phusion Passenger under nginx
  • --platform=s - specify target platform. See FROM for details
  • --variant=s - dockerhub ruby variant, defaults to slim. See docker official images for list.
  • --precompile=defer - may be needed when your configuration requires access to secrets that are not available at build time. Results in larger images and slower deployments.
  • --root - run application as root
  • --windows - make Dockerfile work for Windows users that may have set git config --global core.autocrlf true
  • --private-gemserver-domain=gems.example.com - set the domain name of your private gemserver. This is used to tell bundler for what domain to use the credentials of a private gemserver provided via a docker secret

Advanced Customization:

There may be times where feature detection plus flags just aren't enough. As an example, you may wish to configure and run multiple processes.

  • --instructions=path - a dockerfile fragment to be inserted into the final document.
  • --migration=cmd - a replacement (generally a script) for db:prepare/db:migrate.
  • --procfile=path - a Procfile to use in place of launching Rails directly.

Like with environment variables, packages, and build args, --instructions can be tailored to a specific build phase by adding -base, -build, or -deploy after the flag name, with the default being -deploy.

Additionaly, if the instructions start with a shebang instead the file being treated as a Dockerfile fragment, the file is treated as a script and a RUN statement is added to your Dockerfile instead.


Options are saved between runs into config/dockerfile.yml. To invert a boolean options, add or remove a no- prefix from the option name.

Testing

A single invocation of rake test:all will run all of the tests defined. dockerfile-rails has are three types of tests:

  • rake test:rubocop runs rubocop using the same options as the Rails codebase.
  • rake test:system creates a new esbuild application, generates a dockerfile, builds and runs it. As this is time consuming, only one application is tested this way at this time, and a --javascript example was selected as it exercises a large portion of the features.
  • rake test runs integration tests, as described below

The current integration testing strategy is to run rails new and generate dockerfile with various configurations and compare the generated artifacts with expected results. ARG values in Dockerfiles are masked before comparison.

Running all integration tests, or even a single individual test can be done as follows:

rake test
ruby test/test_minimal.rb

To assist with this process, outputs of tests can be captured automatically. This is useful when adding new tests and when making a change that affects many tests. Be sure to inspect the output (e.g., by using git diff) before committing.

rake test:capture

If you are running a single test, the following environment variables settings may be helpful:

  • RAILS_ENV=TEST will match the environment used to produce the captured outputs.
  • TEST_CAPTURE=1 will capture test results.
  • TEST_KEEP=1 will leave the test app behind for inspection after the test completes.

Historical Links

The following links relate to the coordination between this package and Rails 7.1.

Parallel efforts for Hanami: