Gravis

Runs your heavy Solid Queue jobs on machines that exist only while there is work.

Some jobs need a big machine (video rendering, ffmpeg, ML) but run only a few times a day. Keeping a big worker running 24/7 for them is wasted money. With gravis, enqueueing such a job starts a machine of the right size, and the machine is stopped when the queue is empty. Between jobs you pay nothing.

class RenderVideoJob < ApplicationJob
  include Gravis::Job
  gravis cpu: 8, memory: 16   # vCPU / GB

  def perform(video)
    # your code, unchanged
  end
end

RenderVideoJob.perform_later(video)   # enqueue as always

Everything else stays normal: jobs live in Solid Queue, show up in Mission Control, retry as usual. The machine runs your app's own Docker image.

When to use it: heavy jobs where nobody waits on the result — renders, transcodes, batch work. The first job of a burst waits 1–2 minutes for the machine to start. When not to: anything a user is waiting on. Keep those on your normal worker.

• Rails 8+ with Solid Queue
• Your app deployed as a Docker image
• One always-on Solid Queue process (a worker service, or Rails 8's Puma plugin)
• An account with a supported provider — currently AWS ECS Fargate via gravis-ecs

1. Add the gem for your provider:

gem "gravis-ecs"

Not on rubygems.org yet — until then install from git:

gem "gravis",     github: "zarpay/gravis", glob: "gravis/*.gemspec"
gem "gravis-ecs", github: "zarpay/gravis", glob: "gravis-ecs/*.gemspec"

2. Mark your heavy jobs:

class RenderVideoJob < ApplicationJob
  include Gravis::Job
  gravis cpu: 8, memory: 16   # optional — default is 2 vCPU / 8 GB
end

Don't set queue_as on these jobs — gravis routes them itself. Your other jobs, workers, and queue.yml need no changes.

3. Set up the AWS side — one task definition and one env var. Copy-paste recipes for Terraform and CDK: gravis-ecs setup and docs/infrastructure.md.

That's it. In development and test nothing is set up, so gravis stays off and these jobs run inline like any other job.

Sizes are vCPU / GB pairs. Invalid pairs fail when the class loads, so you find out in development, not in production. Valid pairs for Fargate: size table.

Scheduled jobs work: RenderVideoJob.set(wait: 2.hours).perform_later(video) — the machine starts when the job is due, not before.

Every boot logs one line telling you where gravis stands:

[gravis] enabled — provider: ecs, cluster: myapp-production, task_definition: myapp-gravis
[gravis] disabled — GRAVIS_TARGET not set; gravis jobs run inline

Optional — bin/rails generate gravis:install writes this file with everything commented out:

• How it works inside: docs/architecture.md
• AWS setup recipes: docs/infrastructure.md
• Developing the gem, writing a new provider: CONTRIBUTING.md

MIT