Install OpenProject with Docker

Docker is a way to distribute self-contained applications easily. We provide a Docker image for the Community Edition that you can very easily install and upgrade on your servers. However, contrary to the manual or package-based installation, your machine needs to have the Docker Engine installed first, which usually requires a recent operating system. Please see the Docker Engine installation page if you don’t have Docker installed.

OpenProject with Docker can be launched in two ways:

  1. Multiple containers (recommended), each with a single process inside, using a Compose file. Allows to easily choose which services you want to run, and simplifies scaling and monitoring aspects.
  2. One container with all the processes inside. Easy but not recommended for production. This is the legacy behaviour.

Quick Start

First, you must clone the OpenProject repository:

git clone --depth=1 --branch=stable/10 https://github.com/opf/openproject

Then, go into the OpenProject folder and you can launch all the services required by OpenProject with docker-compose:

docker-compose up -d

After some time, you will be able to access OpenProject on http://localhost:8080. The default username and password is login: admin, and password: admin.

Note that the official docker-compose.yml file present in the repository can be adjusted to your convenience. For instance you could mount specific configuration files, override environment variables, or switch off services you don’t need. Please refer to the official docker-compose documentation for more details.

You can stop the Compose stack by running:

docker-compose down

All-in-one container

Quick Start

The fastest way to get an OpenProject instance up and running is to run the following command:

docker run -it -p 8080:80 -e SECRET_KEY_BASE=secret openproject/community:10

This will take a bit of time the first time you launch it, but after a few minutes you should see a success message indicating the default administration password (login: admin, password: admin).

You can then launch a browser and access your new OpenProject installation at http://localhost:8080. Easy!

To stop the container, simply hit CTRL-C.

Note that the above command will not daemonize the container and will display the logs to your terminal, which helps with debugging if anything goes wrong. For normal usage you probably want to start it in the background, which can be achieved with the -d flag:

docker run -d -p 8080:80 -e SECRET_KEY_BASE=secret openproject/community:10

The one-liner above is great to get started quickly, but if you want to run OpenProject in production you will likely want to ensure that your data is not lost if you restart the container.

To achieve this, we recommend that you create a directory on your host system where the Docker Engine is installed (for instance: /var/lib/openproject) where all this data will be stored.

You can use the following commands to create the local directories where the data will be stored across container restarts, and start the container with those directories mounted:

sudo mkdir -p /var/lib/openproject/{pgdata,assets}

docker run -d -p 8080:80 --name openproject -e SECRET_KEY_BASE=secret \
  -v /var/lib/openproject/pgdata:/var/openproject/pgdata \
  -v /var/lib/openproject/assets:/var/openproject/assets \
  openproject/community:10

Note: Make sure to replace secret with a random string. One way to generate one is to run head /dev/urandom | tr -dc A-Za-z0-9 | head -c 32 ; echo '' if you are on Linux.

Since we named the container, you can now stop it by running:

docker stop openproject

And start it again:

docker start openproject

If you want to destroy the container, run the following commands

docker stop openproject
docker rm openproject

Initial configuration

OpenProject is usually configured through a YAML file, but with the Docker image you need to pass all configuration through environment variables. You can overwrite any of the values usually found in the standard YAML file by using environment variables.

Environment variables can be either passed directly on the command-line to the Docker Engine, or via an environment file:

docker run -d -e KEY1=VALUE1 -e KEY2=VALUE2 ...
# or
docker run -d --env-file path/to/file ...

For more advanced configuration, please have a look at the Advanced configuration section.

Apache Reverse Proxy Setup

Often there will be an existing web server through which you want to make OpenProject acccessible. There are two ways to run OpenProject. We’ll cover each configuration in a separate of the following sections.

For both configurations the following Apache mods are required:

  • proxy
  • proxy_http
  • rewrite
  • ssl (optional)

In each case you will create a file /usr/local/apache2/conf/sites/openproject.conf with the contents as described in the respective sections.

Both configuration examples are based on the following assumptions:

  • the site is accessed via https
  • certificate and key are located under /etc/ssl/crt/server.{crt, key}
  • the OpenProject docker container’s port 80 is mapped to the docker host’s port 8080

Important: Once OpenProject is running make sure to also set the host name and protocol accordingly under Administration -> System Settings.

1) Virtual host root

The default scenario is to have OpenProject serve the whole virtual host. This requires no further configuration for the docker container beyond what is described above.

Assuming the desired server name is openproject.example.com the configuration will look like this:

<VirtualHost *:80>
    ServerName openproject.example.com

    RewriteEngine on
    RewriteCond %{HTTPS} !=on
    RewriteRule ^/?(.*)$ https://%{SERVER_NAME}/$1 [R,L]
</VirtualHost>

<VirtualHost *:443>
    ServerName openproject.example.com

    SSLEngine on
    SSLCertificateFile /etc/ssl/crt/server.crt
    SSLCertificateKeyFile /etc/ssl/crt//server.key

    RewriteEngine on
    RewriteRule "^$" "/" [R,L]

    ProxyRequests off

    <Location "/">
      RequestHeader set X-Forwarded-Proto 'https'

      ProxyPreserveHost On
      ProxyPass http://127.0.0.1:8080/
      ProxyPassReverse http://127.0.0.1:8080/
    </Location>
</VirtualHost>

2) Location (subdirectory)

Let’s assume you want OpenProject to run on your host with the server name example.com under the subdirectory /openproject.

If you want to run OpenProject in a subdirectory on your server, first you will need to configure OpenProject accordingly by adding the following options to the docker run call:

-e OPENPROJECT_RAILS__RELATIVE__URL__ROOT=/openproject \
-e OPENPROJECT_RAILS__FORCE__SSL=true \

The force ssl option can be left out if you are not using HTTPS.

The apache configuration for this configuration then looks like this:

<VirtualHost *:80>
    ServerName example.com

    RewriteEngine on
    RewriteCond %{HTTPS} !=on
    RewriteRule ^/?(openproject.*)$ https://%{SERVER_NAME}/$1 [R,L]
</VirtualHost>

<VirtualHost *:443>
    ServerName example.com

    SSLEngine on
    SSLCertificateFile /etc/ssl/crt/server.crt
    SSLCertificateKeyFile /etc/ssl/crt/server.key

    RewriteEngine on
    RewriteRule "^/openproject$" "/openproject/" [R,L]

    ProxyRequests off

    <Location "/openproject/">
      RequestHeader set X-Forwarded-Proto 'https'

      ProxyPreserveHost On
      ProxyPass http://127.0.0.1:8080/openproject/
      ProxyPassReverse http://127.0.0.1:8080/openproject/
    </Location>
</VirtualHost>

OpenProject plugins

The docker image itself does not support plugins. But you can create your own docker image to include plugins.

1. Create a new folder with any name, for instance custom-openproject. Change into that folder.

2. Create the file Gemfile.plugins in that folder. In the file you declare the plugins you want to install. For instance:

group :opf_plugins do
  gem "openproject-slack", git: "https://github.com/opf/openproject-slack.git", branch: "release/10.0"
end

3. Create the Dockerfile in the same folder. The contents have to look like this:

FROM openproject/community:10

COPY Gemfile.plugins /app/

RUN bundle config unset deployment && bundle install && bundle config set deployment 'true'
RUN bash docker/precompile-assets.sh

The file is based on the normal OpenProject docker image. All the Dockerfile does is copy your custom plugins gemfile into the image, install the gems and precompile any new assets.

4. Build the image

To actually build the docker image run:

docker build -t openproject-with-slack .

The -t option is the tag for your image. You can choose what ever you want.

5. Run the image

You can run the image just like the normal OpenProject image (as shown earlier). You just have to use your chosen tag instead of openproject/community:10. To just give it a quick try you can run this:

docker run -p 8080:80 --rm -it openproject-with-slack

After which you can access OpenProject under http://localhost:8080.

Docker Swarm

If you need to serve a very large number of users it’s time to scale up horizontally. One way to do that is to use your orchestration tool of choice such as Kubernetes or Swarm. Here we’ll cover how to scale up using the latter.

1) Setup Swarm

Here we will go through a simple setup of a Swarm with a single manager. For more advanced setups and more information please consult the docker swarm documentation.

First initialize your swarm on the host you wish to be the swarm manager.

docker swarm init
# You may need or want to specify the advertise address.
# Say your node manager host's IP is 10.0.2.77:
#
#   docker swarm init --advertise-addr=10.0.2.77

The host will automatically also join the swarm as a node to host containers.

Add nodes

To add worker nodes run docker swarm join-token worker. This will print the necessary command (which includes the join token) which you need to run on the host you wish to add as a worker node. For instance:

docker swarm join --token SWMTKN-1-2wnvro17w7w2u7878yflajyjfa93e8b2x58g9c04lavcee93eb-abig91iqb6e5vmupfvq2f33ni 10.0.2.77:2377

Where 10.0.2.77 is your swarm manager’s (advertise) IP address.

2) Setup shared storage

Note: This is only relevant if you have more than 1 node in your swarm.

If your containers run distributed on multiple nodes you will need a shared network storage to store OpenProject’s attachments. The easiest way for this would be to setup an NFS drive that is shared among all nodes and mounted to the same path on each of them. Say /mnt/openproject/.

Alternatively, if using S3 is an option, you can use S3 attachments instead. We will show both possibilities later in the configuration.

3) Create stack

To create a stack you need a stack file. The easiest way is to just copy OpenProject’s docker-compose.yml. Just download it and save it as, say, openproject-stack.yml.

Configuring storage

Note: This is only necessary if your swarm runs on multiple nodes.

Attachments

NFS

If you are using NFS to share attachments use a mounted docker volume to share the attachments folder.

Per default the YAML file will include the following section:

x-op-app: &app
  <<: *image
  <<: *restart_policy
  environment:
    # ... 
  volumes:
    - "opdata:/var/openproject/assets"
  depends_on:
    # ...

As you can see it already mounts a local directory by default. You can either change this to a path in your mounted NFS folder or just create a symlink:

ln -s /mnt/openproject/assets /var/openproject/assets

S3

If you want to use S3 you will have to add the respective configuration to the stack.yml’s environment section for the app.

x-op-app: &app
  <<: *image
  <<: *restart_policy
  environment:
    # ...
    # ADD THIS FOR S3 attachments substituting the respecive credentials:
    - "OPENPROJECT_ATTACHMENTS__STORAGE=fog"
    - "OPENPROJECT_FOG_DIRECTORY="<bucket-name>"
    - "OPENPROJECT_FOG_CREDENTIALS_PROVIDER=AWS"
    - "OPENPROJECT_FOG_CREDENTIALS_AWS__ACCESS__KEY__ID=<access-key-id>"
    - "OPENPROJECT_FOG_CREDENTIALS_AWS__SECRET__ACCESS__KEY=<secret-access-key>"
    - "OPENPROJECT_FOG_CREDENTIALS_REGION=us-east-1"
Database

The database’s data directory should also be shared so that the database service can be moved to another node in case the original node fails. The easiest way to do this would again be a shared NFS mount present on each node. This is also the easiest way to persist the database data so it remains even if you shutdown the whole stack.

You could either use a new mounted NFS folder or use a sub-folder in the one we will use for attachments. Along the same lines as attachments you could adjust the pgdata volume in the openproject-stack.yml so it would look something like this:

x-op-app: &app
  <<: *image
  <<: *restart_policy
  environment:
    # ... 
  volumes:
    - "pgdata:/mnt/openproject/pgdata"
    - "opdata:/mnt/openproject/assets"
  depends_on:
    # ...

Disclaimer: This may not be the best possible solution, but it is the most straight-forward one.

OpenProject Configuration

Any additional configuration of OpenProject happens in the environment section (like for S3 above) of the app inside of the openproject-stack.yml. For instance should you want to disable an OpenProject module globally, you would add the following:

x-op-app: &app
  <<: *image
  <<: *restart_policy
  environment:
    # ...
    - "OPENPROJECT_DISABLED__MODULES='backlogs meetings'"

Please refer to our documentation on the configuration and environment variables for further information on what you can configure and how.

Launching

Once you made any necessary adjustments to the openproject-stack.yml you are ready to launch the stack.

docker stack deploy -c openproject-stack.yaml openproject

Once this has finished you should see something like this when running docker service ls:

docker service ls
ID                  NAME                 MODE                REPLICAS            IMAGE                      PORTS
kpdoc86ggema        openproject_cache    replicated          1/1                 memcached:latest           
qrd8rx6ybg90        openproject_cron     replicated          1/1                 openproject/community:10   
cvgd4c4at61i        openproject_db       replicated          1/1                 postgres:10                
uvtfnc9dnlbn        openproject_proxy    replicated          1/1                 openproject/community:10   *:8080->80/tcp
g8e3lannlpb8        openproject_seeder   replicated          0/1                 openproject/community:10   
canb3m7ilkjn        openproject_web      replicated          1/1                 openproject/community:10   
7ovn0sbu8a7w        openproject_worker   replicated          1/1                 openproject/community:10

You can now access OpenProject under http://0.0.0.0:8080. This endpoint then can be used in a apache reverse proxy setup as shown further up, for instance.

Don’t worry about one of the services (openproject_seeder) having 0/1 replicas. That is intended. The service will only start once to initialize the seed data and then stop.

Scaling

Now the whole reason we are using swarm is to be able to scale. This is now easily done using the docker service scale command.

We’ll keep the database and memcached at 1 which should be sufficient for any but huge amounts of users (several tens of thousands of users) assuming that the docker hosts (swarm nodes) are powerful enough. Even with the database’s data directory shared via NFS you cannot scale up the database in this setup. Scaling the database horizontally adds another level of complexity which we won’t cover here.

What we can scale is both the proxy, and most importantly the web service. For a couple of thousand users we may want to use 6 web service (openproject_web) replicas. The proxy processes (openproject_proxy) in front of the actual OpenProject process does not need as many replicas. 2 are fine here.

Also at least 2 worker (openproject_worker) replicas make sense to handle the increased number of background tasks. If you find that it takes too long for those tasks (such as sending emails or work package exports) to complete you may want to increase this number further.

docker service scale openproject_proxy=2 openproject_web=6 openproject_worker=2

This will take a moment to converge. Once done you should see something like the following when listing the services using docker service ls:

docker service ls
ID                  NAME                 MODE                REPLICAS            IMAGE                      PORTS
kpdoc86ggema        openproject_cache    replicated          1/1                 memcached:latest           
qrd8rx6ybg90        openproject_cron     replicated          1/1                 openproject/community:10   
cvgd4c4at61i        openproject_db       replicated          1/1                 postgres:10                
uvtfnc9dnlbn        openproject_proxy    replicated          2/2                 openproject/community:10   *:8080->80/tcp
g8e3lannlpb8        openproject_seeder   replicated          0/1                 openproject/community:10   
canb3m7ilkjn        openproject_web      replicated          6/6                 openproject/community:10   
7ovn0sbu8a7w        openproject_worker   replicated          1/1                 openproject/community:10

Docker swarm handles the networking necessary to distribute the load among the nodes. The application will still be accessible as before simply under http://0.0.0.0:8080 on each node, e.g. http://10.0.2.77:8080, the manager node’s IP.

Load balancer setup

Now as mentioned earlier you can simply use the manager node’s endpoint in a reverse proxy setup and the load will be balanced among the nodes. But that will be a single point of failure if the manager node goes down.

To make this more redundant you can use the load balancer directive in your proxy configuration. For instance for apache this could look like this:

<Proxy balancer://swarm>
    BalancerMember http://10.0.2.77:8080 # swarm node 1 (manager)
    BalancerMember http://10.0.2.78:8080 # swarm node 2
    BalancerMember http://10.0.2.79:8080 # swarm node 3, etc.

    ProxySet lbmethod=bytraffic
</Proxy>

# ...

ProxyPass "balancer://swarm/"
ProxyPassReverse "balancer://swarm/"

# instead of
#   ProxyPass http://127.0.0.1:8080/
#   ProxyPassReverse http://127.0.0.1:8080/
# shown in the reverse proxy configuration example further up

The application will be accessible on any node even if the process isn’t running on the node itself. In that case it will use swarm’s internal load balancing to route the request to a node that does run the service. So feel free to put all nodes into the load balancer configuration.