A supervisor for docker-compose apps.

# Harbormaster

Harbormaster is a small utility that lets you easily deploy multiple Docker-Compose applications on a single host.

It does this by taking a list of git repository URLs that contain Docker Compose files and running the Compose apps they contain. It will also handle updating/restarting the apps when the repositories change.

## Rationale

Do you have a home server you want to run a few apps on, but don't want everything to break every time you upgrade the OS? Do you want automatic updates but don't want to buy an extra 4 servers so you can run Kubernetes?

Do you have a work server that you want to run a few small services on, but don't want to have to manually manage it? Do you find that having every deployment action be in a git repo more tidy?

Harbormaster is for you.

At its core, Harbormaster takes a YAML config file with a list of git repository URLs containing Docker Compose files, clones/pulls them, and starts the services they describe.

You run Harbormaster on a timer, pointing it to a directory, and it updates all the repositories in its configuration, and restarts the Compose services if they have changed. That's it!

It also cleanly stores data for all apps in a single data/ directory, so you always have one directory that holds all the state, which you can easily back up and restore.

## Installation

You can run Harbormaster directly from Docker, without installing anything. Skip down to the Docker installation section.

Installing Harbormaster is simple. You can use pipx (recommended):

$pipx install docker-harbormaster  Or pip (less recommended): $ pip install docker-harbormaster


You need to also make sure you have git installed on your system.

You can also download a standalone executable for Linux from the pipelines page.

## Docker installation

You can run Harbormaster by using just Docker. You need to follow a few simple steps to set up your configuration and SSH:

• Your harbormaster.yml configuration file should be in a git repository. Check that repository out into some directory, that we're going to call your "config" directory.
• If you need an SSH key to pull the Harbormaster configuration file and the various repositories, copy the private key into your config directory, to a file called ssh_private_key (make sure it's not protected with a passphrase).
• Make a directory for Harbormaster to work in somewhere. All your apps' data is going to reside in that directory.
• Run the Harbormaster image:
docker run -d \
--restart unless-stopped \
-v /var/run/docker.sock:/var/run/docker.sock \
-v <the path to your config directory>:/config \
-v <the path to your Harbormaster working directory>:/main \
stavros/harbormaster


Harbormaster will now run every five minutes, pull your config repository (from whatever remote it has), and run the apps in the config.

If you want to run it immediately at some point, you can use the following command:

\$ docker exec -i -t <container id> /usr/bin/run-harbormaster


## High-level architecture overview

At its core, Harbormaster works very simply: It takes a YAML file containing a list of repositories, pulls/clones them as necessary, messes with their docker-compose.yml files in the way you specify, and tells Compose to start, stop, or restart them, as needed.

That's all it does.

## Usage

Harbormaster uses a single YAML configuration file that's basically a list of repositories containing docker-compose.yml files/apps to deploy:

config:
# Prune *all unused* system images after a run. Good for saving space on the host.
# Careful, if you run this on a system with other Docker images, it will delete them.
prune: true
apps:
myapp:
# The git repository URL to clone.
url: https://github.com/someuser/somerepo.git
# Which branch to deploy.
branch: main
# The environment variables to run Compose with.
environment:
FOO: bar
MYVAR: 1
# A file to load environment variables from. The file must consist of lines
# in the form of key=value. The filename is relative to the Harbormaster
# config file (this file). This can also be a YAML file with the .yml extension,
# containing a single YAML collection of string values.
# Variables in the environment key above take precedence over variables
# in the file.
# Make sure all these variable names appear in the environment section of the
# app's docker-compose.yml file.
environment_file: "somefile.txt"
otherapp:
url: https://gitlab.com/otheruser/otherrepo.git
# The Compose config filename, if it's not docker-compose.yml, or if you
# want to use Harbormaster-specific overrides:
compose_config:
- docker-compose.yml
- docker-compose.harbormaster.yml
# A dictionary of replacements (see below).
replacements:
MYVOLUMENAME: volume
# A file containing replacements. Works in the exact same way as the
# environment_file above.
replacements_file: "otherfile.txt"
# A YAML environment file.
environment_file: "somefile.yml"
oldapp:
# This is an old app, so it shouldn't be run.
enabled: false
# Two apps can use the same repo.
url: https://gitlab.com/otheruser/otherrepo.git


Then, just execute harbormaster run in the same directory as that configuration file. Harbormaster will parse the file, automatically download the repositories mentioned in it (and keep them up to date).

Harbormaster only ever writes to the working directory you specify, and nowhere else. All the data for each Compose app is under <workdir>/data/<appname>, so you can easily back up the entire data directory in one go.

WARNING: Make sure the Compose config in each of the repos does not use the container_name directive, otherwise Harbormaster might not always be able to terminate your apps when necessary.

If you want to trigger Harbormaster via a webhook (perhaps whenever the config file repository changes), you can use Captain Webhook.

## Testing

When developing Harbormaster-compatible Compose files for our app, you usually need to test them. The naive way requires you to write a Harbormaster configuration file, create a repository for your app, commit the app's Compose file in it, and run Harbormaster with the configuration file to test.

To make this easier, Harbormaster includes the harbormaster test command, which will create a temporary directory for the data/cache/etc directories, and run the Compose files directly from your repository, without committing or writing a Harbormaster configuration file.

Run harbormaster test --help to see the available options.

## Recommended deployment

Note: The Harbormaster Docker image mentioned in "Docker installation" is still relatively new, but it's a very convenient way to deploy Harbormaster without installing anything. That may become the recommended way to deploy Harbormaster in the future.

The recommended way to run Harbormaster is on a timer. You can use systemd, with two files. Put the Harbormaster configuration YAML in a repository, and clone it somewhere. Then, use the two files to run Harbormaster in that repository.

/etc/systemd/system/harbormaster.service:

[Unit]
Description=Run the Harbormaster updater
Wants=harbormaster.timer

[Service]
ExecStart=/usr/local/bin/harbormaster run
ExecStartPre=/usr/bin/git pull
WorkingDirectory=<the repository directory>

[Install]
WantedBy=multi-user.target


/etc/systemd/system/harbormaster.timer:

[Unit]
Description=Run Harbormaster every few minutes.
Requires=harbormaster.service

[Timer]
Unit=harbormaster.service
OnUnitInactiveSec=5m

[Install]
WantedBy=timers.target


Then, run:

sudo systemctl daemon-reload
sudo systemctl enable harbormaster

# To run Harbormaster immediately:
sudo service harbormaster start

# To check the Harbormaster run logs:
sudo journalctl -fu harbormaster


This will run Harbormaster every five minutes, pulling your configuration repository before the run.

## Recommended repository layout

Usually, you will have one repository per app. However, for small apps, like ones that already have a Docker container on the Docker Hub (and thus just need a Compose file), it might be more convenient to store the Compose file(s) in the same repository as the Harbormaster config, one branch per app.

That way, you can pull the Harbormaster configuration and all the app definitions in the same way, from the same repository, and specify the branch to load the app from with the branch directive in the config.

## Recommended secrets handling

The recommended way for handling secrets is to add plaintext files to a secrets/ subdirectory of the repository (e.g. secrets/myservice.txt) and use git-crypt to encrypt them. That way, it's easy to add more secrets to the repository, but also only authorized people and the deployment server has access to the files.

## Handling data directories

Due to the way Compose files work, you need to do some extra work to properly tell Harbormaster about your volumes.

Harbormaster provides two kinds of directories: Data and cache.

Data is anything that you want to keep. Data directories will never be deleted, if you remove an app later on, its corresponding data directory will be moved under the archives/ directory and renamed to <appname>-<deletion date>.

Cache is anything you don't care about. When you remove an app from the config, the cache dir is deleted.

Harbormaster will look for a file called docker-compose.yml at the root of the repo, and look for some specific strings (you read more about this in the "replacements" section below).

The built-in strings to be replaced are:

• {{ HM_DATA_DIR }} - The app's data that you want to persist. Will be stored in the data/ directory, under the main Harbormaster working directory.
• {{ HM_CACHE_DIR }} - Any data you don't want to keep. Will be stored in the cache/ directory, under the main Harbormaster working directory.
• {{ HM_REPO_DIR }} - The app's repository. Use this if you want to mount the app's directory itself, for example to access some code that you don't want to copy into the container.

They will be replaced with the proper directory names (without trailing slashes), so the volumes section of your Compose file in your repository should look something like this:

volumes:
- {{ HM_DATA_DIR }}/my_data:/some_data_dir
- {{ HM_DATA_DIR }}/foo:/home/foo
- {{ HM_CACHE_DIR }}/my_cache:/some_cache_dir


### Replacements

Sometimes, the user needs to give access to paths that already exist on their system, or specify more parameters in the Dockerfile. This is where replacements come in.

Replacements are basically custom replacement strings (like the data directory strings) that you can specify yourself.

For example, if the user needs to specify a directory with their media, you can ask them to include a replacement called MEDIA_DIR in their Harbormaster config file, and then use the string {{ HM_MEDIA_DIR }} in your Compose file to mount the volume, like so:

volumes:
- {{ HM_MEDIA_DIR }}:/some_container_dir


Harbormaster will replace that string wherever in the file it finds it (not just the volumes section, and the user can specify it in their Harbormaster config like so:

someapp:
url: https://gitlab.com/otheruser/otherrepo.git
replacements:
MEDIA_DIR: /media/my_media


Keep in mind that if the variable is called VARNAME, the string that will end up being replaced is {{ HM_VARNAME }}. If the variable is not found, it will not be replaced or touched at all. This is to avoid messing with any unrelated templates in the Compose file.

Also, note that replacements will be written on disk, in the Compose config file. If, for some reason, you want to avoid that (e.g. if you have secrets you don't want exposed), try to use environment variables instead.

One experimental feature of replacements is the ability to specify defaults:

services:
app:
environment:
HTTP_PORT: {{ HM_PORT:80 }}
STACK: {{ HM_STACK:"production" }}


If you don't specify the PORT variable in the Harbormaster config file, the replacement will be replaced with 80.

This feature is still experimental and may change.

## Examples

This is an example of the configuration for a Harbormaster-compatible Compose app that adheres to some best practices.

We'll use two Compose files, mount volumes and pass secrets as environment variables.

The docker-compose.yml file is pretty straighforward, doesn't mount any volumes and uses an environment variable as a secret.

docker-compose.yml:

services:
main:
command: ./myscript
image: myapp
build: .
stdin_open: true
tty: true
restart: unless-stopped
environment:
- SOME_SECRET


The docker-compose.harbormaster.yml file is quite small, it overrides the command (so the script starts from the /state directory) and the volumes, so the /state directory maps to the host's data directory.

docker-compose.harbormaster.yml:

services:
main:
command: bash -c 'cd /state; /code/myscript'
volumes:
- {{ HM_DATA_DIR }}:/state/


The Harbormaster config file is very straightforward, it specifies a repo URL and the two Compose configuration files. The docker-compose.yml is specified first, and the Harbormaster override is second, so the command is overridden properly.

harbormaster.yml:

apps:
myapp:
url: https://github.com/myuser/myrepo.git
compose_config:
- docker-compose.yml
- docker-compose.harbormaster.yml


This is a good way to add Harbormaster configuration files with very few lines of configuration. Keep in mind that you unfortunately cannot override volumes with this technique, as Docker will complain that the volume has been specified twice.

It's better to define a different volume and change your command to use that directory, as we've done above.

## Bundled apps

Harbormaster includes some built-in apps in its repository, for your convenience. Check out the apps directory for the Compose files. You can include them in your Harbormaster config directly, with no other configuration.

Here's an example that includes the Plex media server and ZTNCUI:

apps:
plex:
url: https://gitlab.com/stavros/harbormaster.git
compose_config: apps/plex-bridge.yml
environment:
replacements:
MEDIA_DIR: "<your video directory on the host>"

ztncui:
url: https://gitlab.com/stavros/harbormaster.git
environment:
compose_config: apps/ztncui/docker-compose.harbormaster.yml

octoprint:
url: https://gitlab.com/stavros/harbormaster.git
compose_config: apps/octoprint/docker-compose.harbormaster.yml


## Project details

Uploaded source
Uploaded py3