MRSK deploys web apps anywhere from bare metal to cloud VMs using Docker with zero downtime. It uses the dynamic reverse-proxy Traefik to hold requests while the new application container is started and the old one is stopped. It works seamlessly across multiple hosts, using SSHKit to execute commands. It was built for Rails applications, but works with any type of web app that can be containerized with Docker.

Watch the screencast: https://www.youtube.com/watch?v=LL1cV2FXZ5I

Install MRSK globally with gem install mrsk. Then, inside your app directory, run mrsk init (or mrsk init --bundle within Rails apps where you want a bin/mrsk binstub). Now edit the new file config/deploy.yml. It could look as simple as this:

service: hey
image: 37s/hey
servers:
  - 192.168.0.1
  - 192.168.0.2
registry:
  username: registry-user-name
  password:
    - MRSK_REGISTRY_PASSWORD
env:
  secret:
    - RAILS_MASTER_KEY

Then edit your .env file to add your registry password as MRSK_REGISTRY_PASSWORD (and your RAILS_MASTER_KEY for production with a Rails app).

Now you're ready to deploy to the servers:

mrsk deploy

This will:

1. Connect to the servers over SSH (using root by default, authenticated by your ssh key)
2. Install Docker on any server that might be missing it (using apt-get)
3. Log into the registry both locally and remotely
4. Build the image using the standard Dockerfile in the root of the application.
5. Push the image to the registry.
6. Pull the image from the registry onto the servers.
7. Ensure Traefik is running and accepting traffic on port 80.
8. Ensure your app responds with 200 OK to GET /up.
9. Start a new container with the version of the app that matches the current git version hash.
10. Stop the old container running the previous version of the app.
11. Prune unused images and stopped containers to ensure servers don't fill up.

Voila! All the servers are now serving the app on port 80. If you're just running a single server, you're ready to go. If you're running multiple servers, you need to put a load balancer in front of them.

In the past decade+, there's been an explosion in commercial offerings that make deploying web apps easier. Heroku kicked it off with an incredible offering that stayed ahead of the competition seemingly forever. These days we have excellent alternatives like Fly.io and Render. And hosted Kubernetes is making things easier too on AWS, GCP, Digital Ocean, and elsewhere. But these are all offerings that have you renting computers in the cloud at a premium. If you want to run on your own hardware, or even just have a clear migration path to do so in the future, you need to carefully consider how locked in you get to these commercial platforms. Preferably before the bills swallow your business whole!

MRSK seeks to bring the advance in ergonomics pioneered by these commercial offerings to deploying web apps anywhere. Whether that's low-cost cloud options without the managed-service markup from the likes of Digital Ocean, Hetzner, OVH, etc, or it's your own colocated bare metal. To MRSK, it's all the same. Feed the config file a list of IP addresses with vanilla Ubuntu servers that have seen no prep beyond an added SSH key, and you'll be running in literally minutes.

This approach gives you enormous portability. You can have your web app deployed on several clouds at ease like this. Or you can buy the baseline with your own hardware, then deploy to a cloud before a big seasonal spike to get more capacity. When you're not locked into a single provider from a tooling perspective, there are a lot of compelling options available.

Ultimately, MRSK is meant to compress the complexity of going to production using open source tooling that isn't tied to any commercial offering. Not to zero, mind you. You're probably still better off with a fully managed service if basic Linux or Docker is still difficult, but as soon as those concepts are familiar, you'll be ready to go with MRSK.

MRSK basically is Capistrano for Containers, without the need to carefully prepare servers in advance. No need to ensure that the servers have just the right version of Ruby or other dependencies you need. That all lives in the Docker image now. You can boot a brand new Ubuntu (or whatever) server, add it to the list of servers in MRSK, and it'll be auto-provisioned with Docker, and run right away. Docker's layer caching also speeds up deployments with less mucking about on the server. And the images built for MRSK can be used for CI or later introspection.

Kubernetes is a beast. Running it yourself on your own hardware is not for the faint of heart. It's a fine option if you want to run on someone else's platform, either transparently like Render or explicitly on AWS/GCP, but if you'd like the freedom to move between cloud and your own hardware, or even mix the two, MRSK is much simpler. You can see everything that's going on, it's just basic Docker commands being called.

Docker Swarm is much simpler than Kubernetes, but it's still built on the same declarative model that uses state reconciliation. MRSK is intentionally designed around imperative commands, like Capistrano.

Ultimately, there are a myriad of ways to deploy web apps, but this is the toolkit we're using at 37signals to bring HEY home from the cloud without losing the advantages of modern containerization tooling.

MRSK uses dotenv to automatically load environment variables set in the .env file present in the application root. This file can be used to set variables like MRSK_REGISTRY_PASSWORD or database passwords. But for this reason you must ensure that .env files are not checked into Git or included in your Dockerfile! The format is just key-value like:

MRSK_REGISTRY_PASSWORD=pw
DB_PASSWORD=secret123

If you're using a centralized secret store, like 1Password, you can create .env.erb as a template which looks up the secrets. Example of a .env.erb file:

<% if (session_token = `op signin --account my-one-password-account --raw`.strip) != "" %># Generated by mrsk envify
GITHUB_TOKEN=<%= `gh config get -h github.com oauth_token`.strip %>
MRSK_REGISTRY_PASSWORD=<%= `op read "op://Vault/Docker Hub/password" -n --session  #{session_token}` %>
RAILS_MASTER_KEY=<%= `op read "op://Vault/My App/RAILS_MASTER_SECRET" -n --session #{session_token}` %>
MYSQL_ROOT_PASSWORD=<%= `op read "op://Vault/My App/MYSQL_ROOT_PASSWORD" -n --session #{session_token}` %>
<% else raise ArgumentError, "Session token missing" end %>

This template can safely be checked into git. Then everyone deploying the app can run mrsk envify when they setup the app for the first time or passwords change to get the correct .env file.

If you need separate env variables for different destinations, you can set them with .env.destination.erb for the template, which will generate .env.staging when run with mrsk envify -d staging.

If you are using open source secret store like bitwarden, you can create .env.erb as a template which looks up the secrets.

You can store SOME_SECRET in a secure note in bitwarden vault.

$ bw list items --search SOME_SECRET | jq
? Master password: [hidden]

[
  {
    "object": "item",
    "id": "123123123-1232-4224-222f-234234234234",
    "organizationId": null,
    "folderId": null,
    "type": 2,
    "reprompt": 0,
    "name": "SOME_SECRET",
    "notes": "yyy",
    "favorite": false,
    "secureNote": {
      "type": 0
    },
    "collectionIds": [],
    "revisionDate": "2023-02-28T23:54:47.868Z",
    "creationDate": "2022-11-07T03:16:05.828Z",
    "deletedDate": null
  }
]

and extract the id of SOME_SECRET from the json above and use in the erb below.

Example .env.erb file:

<% if (session_token=`bw unlock --raw`.strip) != "" %># Generated by mrsk envify
SOME_SECRET=<%= `bw get notes 123123123-1232-4224-222f-234234234234 --session #{session_token}` %>
<% else raise ArgumentError, "session_token token missing" end %>

Then everyone deploying the app can run mrsk envify and mrsk will generate .env

The default registry is Docker Hub, but you can change it using registry/server:

registry:
  server: registry.digitalocean.com
  username: registry-user-name
  password: <%= ENV.fetch("MRSK_REGISTRY_PASSWORD") %>

The default SSH user is root, but you can change it using ssh/user:

ssh:
  user: app

If you need to connect to server through a proxy host, you can use ssh/proxy:

ssh:
  proxy: "192.168.0.1" # defaults to root as the user

Or with specific user:

ssh:
  proxy: "[email protected]"

You can inject env variables into the app containers using env:

env:
  DATABASE_URL: mysql2://db1/hey_production/
  REDIS_URL: redis://redis1:6379/1

If you have env variables that are secret, you can divide the env block into clear and secret:

env:
  clear:
    DATABASE_URL: mysql2://db1/hey_production/
    REDIS_URL: redis://redis1:6379/1
  secret:
    - DATABASE_PASSWORD
    - REDIS_PASSWORD

The list of secret env variables will be expanded at run time from your local machine. So a reference to a secret DATABASE_PASSWORD will look for ENV["DATABASE_PASSWORD"] on the machine running MRSK. Just like with build secrets.

If the referenced secret ENVs are missing, the configuration will be halted with a KeyError exception.

Note: Marking an ENV as secret currently only redacts its value in the output for MRSK. The ENV is still injected in the clear into the container at runtime.

You can add custom volumes into the app containers using volumes:

volumes:
  - "/local/path:/container/path"

If your application uses separate hosts for running jobs or other roles beyond the default web running, you can specify these hosts in a dedicated role with a new entrypoint command like so:

servers:
  web:
    - 192.168.0.1
    - 192.168.0.2
  job:
    hosts:
      - 192.168.0.3
      - 192.168.0.4
    cmd: bin/jobs

Note: Traefik will only by default be installed and run on the servers in the web role (and on all servers if no roles are defined). If you need Traefik on hosts in other roles than web, add traefik: true:

servers:
  web:
    - 192.168.0.1
    - 192.168.0.2
  web2:
    traefik: true
    hosts:
      - 192.168.0.3
      - 192.168.0.4

You can specialize the default Traefik rules by setting labels on the containers that are being started:

labels:
  traefik.http.routers.hey.rule: Host(\`app.hey.com\`)

Note: The escaped backticks are needed to ensure the rule is passed in correctly and not treated as command substitution by Bash!

This allows you to run multiple applications on the same server sharing the same Traefik instance and port. See https://doc.traefik.io/traefik/routing/routers/#rule for a full list of available routing rules.

The labels can also be applied on a per-role basis:

servers:
  web:
    - 192.168.0.1
    - 192.168.0.2
  job:
    hosts:
      - 192.168.0.3
      - 192.168.0.4
    cmd: bin/jobs
    labels:
      my-label: "50"

If you're developing on ARM64 (like Apple Silicon), but you want to deploy on AMD64 (x86 64-bit), you can use multi-architecture images. By default, MRSK will setup a local buildx configuration that does this through QEMU emulation. But this can be quite slow, especially on the first build.

If you want to speed up this process by using a remote AMD64 host to natively build the AMD64 part of the image, while natively building the ARM64 part locally, you can do so using builder options:

builder:
  local:
    arch: arm64
    host: unix:///Users/<%= `whoami`.strip %>/.docker/run/docker.sock
  remote:
    arch: amd64
    host: ssh://[email protected]

Note: You must have Docker running on the remote host being used as a builder. This instance should only be shared for builds using the same registry and credentials.

If you're developing on ARM64 (like Apple Silicon), want to deploy on AMD64 (x86 64-bit), but don't need to run the image locally (or on other ARM64 hosts), you can configure a remote builder that just targets AMD64. This is a bit faster than building with multi-arch, as there's nothing to build locally.

builder:
  remote:
    arch: amd64
    host: ssh://[email protected]

If you're developing on the same architecture as the one you're deploying on, you can speed up the build by forgoing both multi-arch and remote building:

builder:
  multiarch: false

This is also a good option if you're running MRSK from a CI server that shares architecture with the deployment servers.

Some images need a secret passed in during build time, like a GITHUB_TOKEN, to give access to private gem repositories. This can be done by having the secret in ENV, then referencing it in the builder configuration:

builder:
  secrets:
    - GITHUB_TOKEN

This build secret can then be referenced in the Dockerfile:

# Copy Gemfiles
COPY Gemfile Gemfile.lock ./

# Install dependencies, including private repositories via access token (then remove bundle cache with exposed GITHUB_TOKEN)
RUN --mount=type=secret,id=GITHUB_TOKEN \
  BUNDLE_GITHUB__COM=x-access-token:$(cat /run/secrets/GITHUB_TOKEN) \
  bundle install && \
  rm -rf /usr/local/bundle/cache

You can customize the traefik command line:

traefik:
  args:
    accesslog: true
    accesslog.format: json

This will start the traefik container with --accesslog=true accesslog.format=json.

Build arguments that aren't secret can also be configured:

builder:
  args:
    RUBY_VERSION: 3.2.0

This build argument can then be used in the Dockerfile:

# Private repositories need an access token during the build
ARG RUBY_VERSION
FROM ruby:$RUBY_VERSION-slim as base

You can manage your accessory services via MRSK as well. The services will build off public images, and will not be automatically updated when you deploy:

accessories:
  mysql:
    image: mysql:5.7
    host: 1.1.1.3
    port: 3306
    env:
      clear:
        MYSQL_ROOT_HOST: '%'
      secret:
        - MYSQL_ROOT_PASSWORD
    volumes:
      - /var/lib/mysql:/var/lib/mysql
  redis:
    image: redis:latest
    host: 1.1.1.4
    port: "36379:6379"
    volumes:
      - /var/lib/redis:/data

Now run mrsk accessory start mysql to start the MySQL server on 1.1.1.3. See mrsk accessory for all the commands possible.

You can use a specific container to run your Cron jobs:

servers:
  cron:
    hosts:
      - 192.168.0.1
    cmd:
      bash -c "cat config/crontab | crontab - && cron -f"

This assumes the Cron settings are stored in config/crontab.

If you'd like to broadcast audits of deploys, rollbacks, etc to a chatroom or elsewhere, you can configure the audit_broadcast_cmd setting with the path to a bin file that will be passed the audit line as the first argument:

audit_broadcast_cmd:
  bin/audit_broadcast

The broadcast command could look something like:

#!/usr/bin/env bash
curl -q -d content="[My App] ${1}" https://3.basecamp.com/XXXXX/integrations/XXXXX/buckets/XXXXX/chats/XXXXX/lines

That'll post a line like follows to a preconfigured chatbot in Basecamp:

[My App] [dhh] Rolled back to version d264c4e92470ad1bd18590f04466787262f605de

MRSK defaults to checking the health of your application again /up on port 3000. You can tailor both with the healthcheck setting:

healthcheck:
  path: /healthz
  port: 4000

This will ensure your application is configured with a traefik label for the healthcheck against /healthz and that the pre-deploy healthcheck that MRSK performs is done against the same path on port 4000.

You can execute one-off commands on the servers:

# Runs command on all servers
mrsk app exec 'ruby -v'
App Host: 192.168.0.1
ruby 3.1.3p185 (2022-11-24 revision 1a6b16756e) [x86_64-linux]

App Host: 192.168.0.2
ruby 3.1.3p185 (2022-11-24 revision 1a6b16756e) [x86_64-linux]

# Runs command on primary server
mrsk app exec --primary 'cat .ruby-version'
App Host: 192.168.0.1
3.1.3

# Runs Rails command on all servers
mrsk app exec 'bin/rails about'
App Host: 192.168.0.1
About your application's environment
Rails version             7.1.0.alpha
Ruby version              ruby 3.1.3p185 (2022-11-24 revision 1a6b16756e) [x86_64-linux]
RubyGems version          3.3.26
Rack version              2.2.5
Middleware                ActionDispatch::HostAuthorization, Rack::Sendfile, ActionDispatch::Static, ActionDispatch::Executor, Rack::Runtime, Rack::MethodOverride, ActionDispatch::RequestId, ActionDispatch::RemoteIp, Rails::Rack::Logger, ActionDispatch::ShowExceptions, ActionDispatch::DebugExceptions, ActionDispatch::Callbacks, ActionDispatch::Cookies, ActionDispatch::Session::CookieStore, ActionDispatch::Flash, ActionDispatch::ContentSecurityPolicy::Middleware, ActionDispatch::PermissionsPolicy::Middleware, Rack::Head, Rack::ConditionalGet, Rack::ETag, Rack::TempfileReaper
Application root          /rails
Environment               production
Database adapter          sqlite3
Database schema version   20221231233303

App Host: 192.168.0.2
About your application's environment
Rails version             7.1.0.alpha
Ruby version              ruby 3.1.3p185 (2022-11-24 revision 1a6b16756e) [x86_64-linux]
RubyGems version          3.3.26
Rack version              2.2.5
Middleware                ActionDispatch::HostAuthorization, Rack::Sendfile, ActionDispatch::Static, ActionDispatch::Executor, Rack::Runtime, Rack::MethodOverride, ActionDispatch::RequestId, ActionDispatch::RemoteIp, Rails::Rack::Logger, ActionDispatch::ShowExceptions, ActionDispatch::DebugExceptions, ActionDispatch::Callbacks, ActionDispatch::Cookies, ActionDispatch::Session::CookieStore, ActionDispatch::Flash, ActionDispatch::ContentSecurityPolicy::Middleware, ActionDispatch::PermissionsPolicy::Middleware, Rack::Head, Rack::ConditionalGet, Rack::ETag, Rack::TempfileReaper
Application root          /rails
Environment               production
Database adapter          sqlite3
Database schema version   20221231233303

# Run Rails runner on primary server
mrsk app exec -p 'bin/rails runner "puts Rails.application.config.time_zone"'
UTC

You can run interactive commands, like a Rails console or a bash session, on a server (default is primary, use --hosts to connect to another):

# Starts a bash session in a new container made from the most recent app image
mrsk app exec -i bash

# Starts a bash session in the currently running container for the app
mrsk app exec -i --reuse bash

# Starts a Rails console in a new container made from the most recent app image
mrsk app exec -i 'bin/rails console'

You can see the state of your servers by running mrsk details:

Traefik Host: 192.168.0.1
CONTAINER ID   IMAGE     COMMAND                  CREATED          STATUS          PORTS                               NAMES
6195b2a28c81   traefik   "/entrypoint.sh --pr…"   30 minutes ago   Up 19 minutes   0.0.0.0:80->80/tcp, :::80->80/tcp   traefik

Traefik Host: 192.168.0.2
CONTAINER ID   IMAGE     COMMAND                  CREATED          STATUS          PORTS                               NAMES
de14a335d152   traefik   "/entrypoint.sh --pr…"   30 minutes ago   Up 19 minutes   0.0.0.0:80->80/tcp, :::80->80/tcp   traefik

App Host: 192.168.0.1
CONTAINER ID   IMAGE                                                                         COMMAND                  CREATED          STATUS          PORTS      NAMES
badb1aa51db3   registry.digitalocean.com/user/app:6ef8a6a84c525b123c5245345a8483f86d05a123   "/rails/bin/docker-e…"   13 minutes ago   Up 13 minutes   3000/tcp   chat-6ef8a6a84c525b123c5245345a8483f86d05a123

App Host: 192.168.0.2
CONTAINER ID   IMAGE                                                                         COMMAND                  CREATED          STATUS          PORTS      NAMES
1d3c91ed1f55   registry.digitalocean.com/user/app:6ef8a6a84c525b123c5245345a8483f86d05a123   "/rails/bin/docker-e…"   13 minutes ago   Up 13 minutes   3000/tcp   chat-6ef8a6a84c525b123c5245345a8483f86d05a123

You can also see just info for app containers with mrsk app details or just for Traefik with mrsk traefik details.

If you've discovered a bad deploy, you can quickly rollback by reactivating the old, paused container image. You can see what old containers are available for rollback by running mrsk app containers. It'll give you a presentation similar to mrsk app details, but include all the old containers as well. Showing something like this:

App Host: 192.168.0.1
CONTAINER ID   IMAGE                                                                         COMMAND                  CREATED          STATUS                      PORTS      NAMES
1d3c91ed1f51   registry.digitalocean.com/user/app:6ef8a6a84c525b123c5245345a8483f86d05a123   "/rails/bin/docker-e…"   19 minutes ago   Up 19 minutes               3000/tcp   chat-6ef8a6a84c525b123c5245345a8483f86d05a123
539f26b28369   registry.digitalocean.com/user/app:e5d9d7c2b898289dfbc5f7f1334140d984eedae4   "/rails/bin/docker-e…"   31 minutes ago   Exited (1) 27 minutes ago              chat-e5d9d7c2b898289dfbc5f7f1334140d984eedae4

App Host: 192.168.0.2
CONTAINER ID   IMAGE                                                                         COMMAND                  CREATED          STATUS                      PORTS      NAMES
badb1aa51db4   registry.digitalocean.com/user/app:6ef8a6a84c525b123c5245345a8483f86d05a123   "/rails/bin/docker-e…"   19 minutes ago   Up 19 minutes               3000/tcp   chat-6ef8a6a84c525b123c5245345a8483f86d05a123
6f170d1172ae   registry.digitalocean.com/user/app:e5d9d7c2b898289dfbc5f7f1334140d984eedae4   "/rails/bin/docker-e…"   31 minutes ago   Exited (1) 27 minutes ago              chat-e5d9d7c2b898289dfbc5f7f1334140d984eedae4

From the example above, we can see that e5d9d7c2b898289dfbc5f7f1334140d984eedae4 was the last version, so it's available as a rollback target. We can perform this rollback by running mrsk rollback e5d9d7c2b898289dfbc5f7f1334140d984eedae4. That'll stop 6ef8a6a84c525b123c5245345a8483f86d05a123 and then start e5d9d7c2b898289dfbc5f7f1334140d984eedae4. Because the old container is still available, this is very quick. Nothing to download from the registry.

Note that by default old containers are pruned after 3 days when you run mrsk deploy.

If you wish to remove the entire application, including Traefik, containers, images, and registry session, you can run mrsk remove. This will leave the servers clean.

This is beta software. Commands may still move around. But we're live in production at 37signals.

MRSK is released under the MIT License.