Rails 8 Project Setup

1. Gemfile Essentials

The Solid Stack (solid_queue, solid_cache, solid_cable) plus propshaft, importmap-rails, kamal, and thruster — a complete Rails 8 setup with no Redis dependency.

# cooldown: gems must be public 3 days before resolving (supply-chain defense, Bundler 4.0.13+)
source "https://rubygems.org", cooldown: 3

gem "rails", "~> 8.1"

# Assets & front end
gem "importmap-rails"   # JavaScript without Node bundling
gem "propshaft"         # modern asset pipeline (replaces Sprockets)
gem "stimulus-rails"
gem "turbo-rails"

# Deployment and drivers
gem "bootsnap", require: false
gem "kamal", require: false    # zero-downtime deployment
gem "puma", ">= 5.0"
gem "solid_cable", ">= 3.0"    # database-backed ActionCable
gem "solid_cache", "~> 1.0"    # database-backed cache store
gem "solid_queue", "~> 1.2"    # database-backed job queue
gem "sqlite3", ">= 2.0"
gem "thruster", require: false # HTTP/2 proxy (replaces nginx for simple setups)
gem "trilogy", "~> 2.9"        # MySQL adapter, faster than mysql2

# Operations
gem "mission_control-jobs"     # web UI for Solid Queue

Bundler cooldown (supply-chain defense):

• cooldown: 3 on the source line means gems must have been public for 3 days before Bundler will resolve them, so freshly-hijacked releases can be vetted/yanked first (announcement). Needs Bundler 4.0.13+ (check BUNDLED WITH in Gemfile.lock; older versions silently ignore cooldown).
• Complement it machine-wide with bundle config set --global cooldown 4; the explicit per-Gemfile line documents the policy and survives global-config changes.
• Scope: own projects only — never add it to vendored gem Gemfiles, .ruby-lsp/Gemfile, or appraisal gemfiles.

Vite + pnpm apps (alternative front-end stack):

• importmap-rails/no-Node is the default for new apps, but apps with heavier JavaScript needs may run Vite (+ Tailwind, Stimulus) alongside Propshaft, with Propshaft still serving page-specific assets from app/assets. Don't migrate such an app to importmaps.
• When Node packages are involved, use pnpm — not npm, yarn, or bun.

Optional add-on gems (beyond the defaults)

The stack above is everything a new app needs. The toolkit blesses a short list of add-ons beyond it — reach for each only when the need is real; the linked skill covers it in depth.

# Gemfile — add only what the app actually uses
gem "simple_form"     # form-builder DSL (DRYs labels/wrappers); plain `form_with` is fine too
gem "friendly_id"     # SEO slugs in URLs — see [[rails-models]]
gem "pundit"          # per-resource authorization policies — see [[rails-security]]

group :development, :test do
  gem "stackprof"       # CPU/wall profiler — see [[rails-performance]] / [[rails-testing]]
  gem "test-prof"       # test-suite profiling (mainly inherited factory suites) — see [[rails-testing]]
end

group :test do
  gem "shoulda-matchers" # one-line model declaration checks — see [[rails-testing]]
  gem "shoulda-context"  # `should` macro for Minitest
end

Already standard elsewhere in the toolkit (don't re-decide these): pagy (pagination — [[rails-api]]), prosopite (N+1 detection — [[rails-performance]]), simplecov + simplecov_json_formatter (coverage — [[rails-testing]]), bundler-audit + brakeman (security audit — [[rails-audit]]).

Deliberately not recommended (a default already covers the need): kaminari → use pagy; bullet → use prosopite; good_job → use solid_queue; devise → Rails 8 built-in auth ([[rails-security]]); factory_bot/faker → fixtures-first ([[rails-testing]]); dry-rb/Sorbet/reek → against the vanilla-Rails grain ([[rails-philosophy]]).

2. Solid Queue Configuration

config/queue.yml defines dispatchers and workers, with process counts derived from CPU count:

default: &default
  dispatchers:
    - polling_interval: 1
      batch_size: 500
  workers:
    - queues: [ "default", "solid_queue_recurring" ]
      threads: 3
      processes: <%= Integer(ENV.fetch("JOB_CONCURRENCY") { Concurrent.physical_processor_count }) %>
      polling_interval: 0.1

development: *default
test: *default
production: *default

For queue naming (SLO tiers, worker-per-queue), see [[rails-jobs]] Pattern 5.

# config/puma.rb — run Solid Queue inside Puma, no separate worker process
unless ENV["SOLID_QUEUE_IN_PUMA"] == "false"
  plugin :solid_queue
end

3. Solid Cache Configuration

# config/cache.yml
default_options: &default_options
  store_options:
    max_age: <%= 60.days.to_i %>
    namespace: <%= Rails.env %>   # isolate cache per environment

default_connection: &default_connection
  database: cache

default: &default
  <<: *default_connection
  <<: *default_options

development: *default
test: *default_options
production: *default

4. Solid Cable Configuration

# config/cable.yml
cable: &cable
  adapter: solid_cable
  connects_to:
    database:
      writing: cable
      reading: cable
  polling_interval: 0.1.seconds
  message_retention: 1.day

development: *cable
test:
  adapter: test
production: *cable

5. Database Setup (Multiple Databases)

Separate databases for primary data, cache, queue, and cable prevent contention and allow independent scaling. Each gets its own migrations_paths.

# SQLite
default: &default
  adapter: sqlite3
  pool: 5
  timeout: 5000

production:
  primary:
    <<: *default
    database: storage/production.sqlite3
    schema_dump: schema_sqlite.rb
  cable:
    <<: *default
    database: storage/production_cable.sqlite3
    migrations_paths: db/cable_migrate
  cache:
    <<: *default
    database: storage/production_cache.sqlite3
    migrations_paths: db/cache_migrate
  queue:
    <<: *default
    database: storage/production_queue.sqlite3
    migrations_paths: db/queue_migrate

# MySQL via trilogy — note the much larger pool in production
default: &default
  adapter: trilogy
  host: <%= ENV.fetch("MYSQL_HOST", "127.0.0.1") %>
  port: <%= ENV.fetch("MYSQL_PORT", "3306") %>
  username: <%= ENV.fetch("MYSQL_USER", "root") %>
  password: <%= ENV["MYSQL_PASSWORD"] %>
  pool: 50
  timeout: 5000

production:
  primary:
    <<: *default
    database: app_production
  cable:
    <<: *default
    database: app_production_cable
    migrations_paths: db/cable_migrate
  queue:
    <<: *default
    database: app_production_queue
    migrations_paths: db/queue_migrate
  cache:
    <<: *default
    database: app_production_cache
    migrations_paths: db/cache_migrate

6. Kamal Deployment

config/deploy.yml: servers, automatic SSL via the proxy, secrets from .kamal/secrets, and persistent volumes for SQLite/Active Storage.

service: app
image: app

servers:
  web:
    - app.example.com

ssh:
  user: root

# Automatic SSL
proxy:
  ssl: true
  host: app.example.com

# Secrets come from .kamal/secrets; clear vars are plain
env:
  secret:
    - SECRET_KEY_BASE
    - SMTP_USERNAME
    - SMTP_PASSWORD
  clear:
    BASE_URL: https://app.example.com
    SOLID_QUEUE_IN_PUMA: true    # jobs run in the web container

registry:
  server: localhost:5555

aliases:
  console: app exec --interactive --reuse "bin/rails console"
  shell: app exec --interactive --reuse "bash"
  logs: app logs -f
  dbc: app exec --interactive --reuse "bin/rails dbconsole --include-password"

# Persistent storage for sqlite and Active Storage
volumes:
  - "app_storage:/rails/storage"

# Bridge assets between versions during zero-downtime deploys
asset_path: /rails/public/assets

builder:
  arch: amd64

7. Dockerfile Best Practices

Multi-stage build with jemalloc, bootsnap precompilation, and a non-root user:

# syntax=docker/dockerfile:1
# check=error=true

ARG RUBY_VERSION=3.4.7
FROM docker.io/library/ruby:$RUBY_VERSION-slim AS base

WORKDIR /rails

RUN apt-get update -qq && \
    apt-get install --no-install-recommends -y curl libjemalloc2 libvips sqlite3 libssl-dev && \
    ln -s /usr/lib/$(uname -m)-linux-gnu/libjemalloc.so.2 /usr/local/lib/libjemalloc.so && \
    rm -rf /var/lib/apt/lists /var/cache/apt/archives

ENV RAILS_ENV="production" \
    BUNDLE_DEPLOYMENT="1" \
    BUNDLE_PATH="/usr/local/bundle" \
    BUNDLE_WITHOUT="development:test" \
    LD_PRELOAD="/usr/local/lib/libjemalloc.so"

# Build stage
FROM base AS build

RUN apt-get update -qq && \
    apt-get install --no-install-recommends -y build-essential git libyaml-dev pkg-config && \
    rm -rf /var/lib/apt/lists /var/cache/apt/archives

COPY Gemfile Gemfile.lock vendor ./
RUN bundle install && \
    rm -rf ~/.bundle/ "${BUNDLE_PATH}"/ruby/*/cache "${BUNDLE_PATH}"/ruby/*/bundler/gems/*/.git && \
    bundle exec bootsnap precompile -j 1 --gemfile

COPY . .

RUN bundle exec bootsnap precompile -j 1 app/ lib/
# SECRET_KEY_BASE_DUMMY=1 allows asset precompilation without real secrets
RUN SECRET_KEY_BASE_DUMMY=1 ./bin/rails assets:precompile

# Final stage
FROM base

RUN groupadd --system --gid 1000 rails && \
    useradd rails --uid 1000 --gid 1000 --create-home --shell /bin/bash
USER 1000:1000

COPY --chown=rails:rails --from=build "${BUNDLE_PATH}" "${BUNDLE_PATH}"
COPY --chown=rails:rails --from=build /rails /rails

ENTRYPOINT ["/rails/bin/docker-entrypoint"]

EXPOSE 80
CMD ["./bin/thrust", "./bin/rails", "server"]

#!/bin/bash -e
# bin/docker-entrypoint — auto-migrate on deploy when starting the server
if [ "${1}" == "./bin/thrust" ] && [ "${2}" == "./bin/rails" ] && [ "${3}" == "server" ]; then
  MIGRATE=1 ./bin/rails db:prepare
fi

exec "${@}"

8. Recurring Jobs Configuration

Scheduled tasks live in config/recurring.yml:

production: &production
  cleanup_magic_links:
    command: "MagicLink.cleanup"
    schedule: every 4 hours

Full pattern (command vs class, Solid Queue maintenance entries, schedule syntax): [[rails-jobs]] Pattern 4.

9. Puma Configuration for Production

Workers per CPU, single thread per worker (optimal for SQLite's low I/O wait), copy-on-write warmup, and GC deferred to between requests:

# config/puma.rb
port ENV.fetch("PORT", 3000)
plugin :tmp_restart

unless ENV["SOLID_QUEUE_IN_PUMA"] == "false"
  plugin :solid_queue
end

if !Rails.env.local?
  # 1 worker per CPU, 1 thread per worker (optimal for SQLite)
  workers Integer(ENV.fetch("WEB_CONCURRENCY") { Concurrent.physical_processor_count })
  threads 1, 1

  # Optimize for copy-on-write
  before_fork do
    Process.warmup
  end

  # Defer major GC until after request handling
  before_worker_boot do
    GC.config(rgengc_allow_full_mark: false)
  end

  out_of_band do
    GC.start if GC.latest_gc_info(:need_major_by)
  end
end

Quick Reference

New Project Checklist

1. Add Solid Stack gems to Gemfile (with cooldown: 3 on the source line)
2. Create config/queue.yml, config/cache.yml, config/cable.yml
3. Configure multiple databases in config/database.yml
4. Add plugin :solid_queue to config/puma.rb
5. Create config/deploy.yml for Kamal
6. Create production Dockerfile with jemalloc and bootsnap
7. Set up config/recurring.yml for scheduled tasks
8. Configure bin/docker-entrypoint for auto-migrations

Environment Variables

# Required secrets (in .kamal/secrets)
SECRET_KEY_BASE=...
RAILS_MASTER_KEY=...

# Database (for MySQL)
MYSQL_HOST=127.0.0.1
MYSQL_PASSWORD=...

# Concurrency tuning
WEB_CONCURRENCY=4        # Puma workers
JOB_CONCURRENCY=4        # Solid Queue processes
SOLID_QUEUE_IN_PUMA=true # Run jobs in web process