OpenProject on Docker all-in-one container

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.

Supported architectures

Starting with OpenProject 12.5.6 we publish our containers for three architectures.

  1. AMD64 (x86)
  2. ARM64
  3. PPC64

The OpenProject BIM Edition is only supported on AMD64, however.

Limitations

Note that the docker container setup does not allow for integration of repositories within OpenProject. You can reference external repositories, but cannot set them up through OpenProject itself. For that feature to work, you need to use the packaged installation method.

Available containers

OpenProject provides two container flavors with a number of target tags:

  • slim: Contains just the application image and application server used for starting behind a proxy, or within the OpenProject for Docker compose or OpenProject Helm chart installation methods. We recommend using this image for production systems.
  • all-in-one: Starts internal PostgreSQL, memcached servers alongside the application in order to provide a quick-start for testing or getting started with OpenProject. The lower part of this documentation focused on this installation method.

OpenProject follows semantic versioning, and tags are pushed on Docker Hub openproject/openproject in the following way:

  • X.Y.Z, X.Y.Z-slim : non-floating tags to provide exactly one release of OpenProject, and you have to update these tags when a new release is made.
  • X.Y, X.Y-slim floating tags that get pushed whenever a new patch release is made. Please keep in mind that in case of urgent fixes, database migrations might still occur in patch releases, so you need to be able to run the seeder/migrations job. Refer to the installation method of your choice for information on how to do that.
  • X, X-slim floating tags that get pushed whenever a new patch or minor release is made. If you use these tags, you are aware that application changes will occur.
  • dev, dev-slim floating tag that gets pushed nightly with the latest development version. These tags are automatically deployed to our QA instances, and are useful for testing and early feedback. We try to keep these versions usable, but we strongly recommend against using them for anything with production data.

We recommend to use non-floating tags for production systems, and use the built-in version check, or our release notes (subscribe to them through GitHub, or release newsletters) to be informed of updates.

Installation overview

OpenProject’s docker setup can be launched in two ways:

One container per process (recommend)

This is the recommended approach for using OpenProject with Docker, where each component has a single container inside, orchestrated using a Compose file. Allows to easily choose which services you want to run, and simplifies scaling and monitoring aspects.

Please follow the OpenProject for Docker compose documentation for this installation method

Single docker container

This guide will show you to install OpenProject in one container with all the processes inside. This allows for a very quick start but is not recommended for production as it hinders upgradability of the different components such as the database.

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 OPENPROJECT_SECRET_KEY_BASE=secret \
  -e OPENPROJECT_HOST__NAME=localhost:8080 \
  -e OPENPROJECT_HTTPS=false \
  -e OPENPROJECT_DEFAULT__LANGUAGE=en \
  openproject/openproject:15

Explanation of the used configuration values:

  • -p 8080:80 binds the port 80 of the container to 8080 on the machine running docker.
  • OPENPROJECT_SECRET_KEY_BASE sets the secret key base for Rails. Please use a pseudo-random value for this and treat it like a password.
  • OPENPROJECT_HOST__NAME sets the host name of the application. This value is used for generating forms and links in emails, and needs to match the external request host name (The value users are seeing in their browsers).
  • OPENPROJECT_HTTPS=false disables the on-by-default HTTPS mode of OpenProject so you can access the instance over HTTP-only. For all production systems we strongly advise not to set this to false, and instead set up a proper TLS/SSL termination on your outer web server.
  • OPENPROJECT_DEFAULT__LANGUAGE does two things. It controls for the very first installation, in which language basic data (such as types, status names, etc.) and demo data is being created in. It also sets the default fallback language for new users.

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 OPENPROJECT_SECRET_KEY_BASE=secret \
  -e OPENPROJECT_HOST__NAME=localhost:8080 \
  -e OPENPROJECT_HTTPS=false \
  openproject/openproject:15

Note: We’ve had reports of people being unable to start OpenProject this way because of an issue regarding pseudo-TTY allocations and permissions to write to /dev/stdout. If you run into this, a workaround seems to be to add -t to your run command, even if you run in detached mode.

Using this container in production

The one-liner above is great to get started quickly, but we strongly advise against using this setup for production purposes, as it disables HTTPS mode and is insecure.

Also, if you want to run OpenProject in production you need 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 OPENPROJECT_HOST__NAME=openproject.example.com \
  -e OPENPROJECT_SECRET_KEY_BASE=secret \
  -v /var/lib/openproject/pgdata:/var/openproject/pgdata \
  -v /var/lib/openproject/assets:/var/openproject/assets \
  openproject/openproject:15

Please make sure you set the correct public facing hostname in OPENPROJECT_HOST__NAME. If you don’t have a load-balancing or proxying web server in front of your docker container, you will otherwise be vulnerable to HOST header injections, as the internal server has no way of identifying the correct host name. We strongly recommend you use an external load-balancing or proxying web server for termination of TLS/SSL and general security hardening.

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.

Note: MacOS users might encounter an “Operation not permitted” error on the mounted directories. The fix for this is to create the two directories in a user-owned directory of the host machine.

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

Configuration

Please see the advanced configuration guide’s docker paragraphs

Disabling HTTPS mode

By default, OpenProject will expect a HTTPS request in production systems. In most cases, you will have an external web server or load balancer terminating the SSL/TLS connection and proxy/reverse-proxy to the docker container. You will then have to set up the web server to forward the protocol information (usually, this is X-Forwarded-Proto but depends on your web server).

NOTE: This does not imply the docker container itself is running on SSL.

If you really want to disable HTTPS responses by OpenProject, you will need to add the environment variable OPENPROJECT_HTTPS=false. Note that this will disable secure cookies for session cookies, and is strongly discouraged for any production system.

Disabling HSTS headers and insecure request upgrade

If OpenProject is running as HTTPS, it will output HTTP Strict Transport Security (HSTS) headers by default, which will force the browser to upgrade a request to HTTPS even before calling the web server. On the server side, if a request is still issued through HTTP, OpenProject will return a redirect to HTTPS.

In most cases, you will want to leave this enabled. If you disabled HTTPS mode above, this flag will also be disabled.

If you really want to disable HSTS headers and request upgrades, you will need to set the setting OPENPROJECT_HSTS=false.

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

Reverse Proxy Setup

The containers above are not meant as public facing endpoints. Always use an existing proxying web server or load balancer to provide access to OpenProject.

There are two ways to run OpenProject. We’ll cover each configuration in a separate of the following sections. Moreover we’re going to give basic configurations for both the Apache and nginx web servers.

Apache

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.

Nginx

The nginx configuration will go into /etc/nginx/conf.d/openproject.conf.

Assumptions

All 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 accordingly under Administration -> System Settings or set it directly during startup by setting OPENPROJECT_HOST__NAME.

NOTE: There is another example for external SSL/TLS termination for packaged installations

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.

Let’s assume we want OpenProject to be accessed under https://openproject.example.com.

The apache configuration for this looks as follows.

<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>

The nginx counterpart can be seen below.

server {
    listen 80;

    server_name openproject.example.com;

    return 301 https://$host$request_uri;
}

server {
  listen 443 ssl;
  server_name openproject.example.com;

  ssl_certificate     /etc/ssl/crt/server.crt;
  ssl_certificate_key /etc/ssl/crt/server.key;

  proxy_redirect    off;

  location / {
    proxy_set_header Host $host;
    proxy_set_header X-Real-IP  $remote_addr;
    proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
    proxy_set_header X-Forwarded-Proto https;

    proxy_pass http://127.0.0.1:8080;
  }
}

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

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>

The equivalent nginx configuration looks as follows.

server {
    listen 80;

    server_name example.com;

    location /openproject {
      return 301 https://$host$request_uri;
    }
}

server {
  listen 443 ssl;
  server_name example.com;

  ssl_certificate     /etc/ssl/crt/server.crt;
  ssl_certificate_key /etc/ssl/crt/server.key;

  proxy_redirect    off;

  location /openproject {
    proxy_set_header Host $host;
    proxy_set_header X-Real-IP  $remote_addr;
    proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
    proxy_set_header X-Forwarded-Proto https;

    proxy_pass http://127.0.0.1:8080/openproject;
  }
}

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: "dev"
end

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

FROM openproject/openproject:15

# If installing a local plugin (using `path:` in the `Gemfile.plugins` above),
# you will have to copy the plugin code into the container here and use the
# path inside of the container. Say for `/app/vendor/plugins/openproject-slack`:
# COPY /path/to/my/local/openproject-slack /app/vendor/plugins/openproject-slack

COPY Gemfile.plugins /app/

# If the plugin uses any external NPM dependencies you have to install them here.
# RUN npm add npm <package-name>*

RUN bundle config unset deployment && bundle install && bundle config set deployment 'true'
RUN ./docker/prod/setup/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.

slim image

If you are using the -slim tag you will need to do the following to add your plugin.

FROM openproject/openproject:15 AS plugin

# If installing a local plugin (using `path:` in the `Gemfile.plugins` above),
# you will have to copy the plugin code into the container here and use the
# path inside of the container. Say for `/app/vendor/plugins/openproject-slack`:
# COPY /path/to/my/local/openproject-slack /app/vendor/plugins/openproject-slack

COPY Gemfile.plugins /app/

# If the plugin uses any external NPM dependencies you have to install them here.
# RUN npm add npm <package-name>*

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

FROM openproject/openproject:15-slim

COPY --from=plugin /usr/bin/git /usr/bin/git
COPY --chown=$APP_USER:$APP_USER --from=plugin /app/vendor/bundle /app/vendor/bundle
COPY --chown=$APP_USER:$APP_USER --from=plugin /usr/local/bundle /usr/local/bundle
COPY --chown=$APP_USER:$APP_USER --from=plugin /app/public/assets /app/public/assets
COPY --chown=$APP_USER:$APP_USER --from=plugin /app/config/frontend_assets.manifest.json /app/config/frontend_assets.manifest.json
COPY --chown=$APP_USER:$APP_USER --from=plugin /app/Gemfile.* /app/

4. Build the image

To actually build the docker image run:

docker build --pull -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 here). You just have to use your chosen tag instead of openproject/openproject:15. 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.

Import self-signed root certificate

If you want to connect OpenProject to an external server as example SMTP-Server or a Nextcloud-Server that uses a self-signed certificate, you need to import the root certificate that was used to create the self-signed certificate. There are two ways to archive this.

The first way is to mount the root certificate via the --mount option into the container and add the certificate to the SSL_CERT_FILE variable.

sudo docker run -it -p 8080:80 \
  -e OPENPROJECT_SECRET_KEY_BASE=secret \
  -e OPENPROJECT_HOST__NAME=localhost:8080 \
  -e OPENPROJECT_HTTPS=false \
  -e OPENPROJECT_DEFAULT__LANGUAGE=en \
  --mount type=bind,source=$(pwd)/my_root.crt,target=/tmp/my_root.crt \ #mount my_root.crt to /tmp
  -e SSL_CERT_FILE=/tmp/my_root.crt \ #set the SSL_CERT_FILE to the path of my_root.crt
  openproject/openproject:15

The second way would be to build a new image of the openproject/openproject:15 or the -slim image.

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

2. Put your root SSL certificate into the folder. In this example, we will name it my_root.crt.

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

FROM openproject/openproject:15

COPY ./my_root.crt /usr/local/share/ca-certificates/
RUN update-ca-certificates

If you are using the -slim tag, you will need to do the following to import your root certificate:

FROM openproject/openproject:15-slim

USER root
COPY ./smtp.local_rootCA.crt /usr/local/share/ca-certificates/
RUN update-ca-certificates
USER $APP_USER

4. Build the image

docker build --pull -t openproject-with-custom-ca .

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 here). You just have to use your chosen tag instead of openproject/openproject:15

Offline/air-gapped installation

It’s possible to run the docker image on an a system with no internet access using docker save and docker load. The installation works the same as described above. The only difference is that you don’t download the image the usual way.

1) Save the image

On a system that has access to the internet run the following.

docker pull openproject/openproject:15 && docker save openproject/openproject:15 | gzip > openproject-stable.tar.gz

This creates a compressed archive containing the latest OpenProject docker image. The file will have a size of around 700mb.

2) Transfer the file onto the system

Copy the file onto the target system by any means that works. This could be sftp, scp or even via a USB stick in case of a truly air-gapped system.

3) Load the image

Once the file is on the system you can load it like this:

gunzip openproject-12.tar.gz && docker load -i openproject-12.tar

This extracts the archive and loads the contained image layers into docker. The .tar file can be deleted after this.

4) Proceed with the installation

After this both installation and later upgrades work just as usual. You only replaced docker-compose pull or the normal, implicit download of the image with the steps described here.

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
AWS S3

If you want to use S3 you will need to add the following configuration to the app section of stack.yml.

x-op-app: &app
  <<: *image
  <<: *restart_policy
  environment:
    ...
    OPENPROJECT_ATTACHMENTS__STORAGE: "fog"
    OPENPROJECT_FOG_DIRECTORY: "«s3-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»" # Must be the region that you created your bucket in
MinIO S3

If you want to use MinIO as a self-hosted S3-compliant storage backend you will need to add the following configuration to the app section of stack.yml.

x-op-app: &app
  <<: *image
  <<: *restart_policy
  environment:
    ...
    OPENPROJECT_ATTACHMENTS__STORAGE: "fog"
    OPENPROJECT_FOG_DIRECTORY: "«s3-bucket-name»"
    OPENPROJECT_FOG_CREDENTIALS_PROVIDER: "aws" # Minio is S3 compliant, so we can use the AWS provider
    OPENPROJECT_FOG_CREDENTIALS_ENDPOINT: "«https://minio-host.domain.tld»" # URI for your MinIO instance
    OPENPROJECT_FOG_CREDENTIALS_AWS__ACCESS__KEY__ID: "«access-key-id»"
    OPENPROJECT_FOG_CREDENTIALS_AWS__SECRET__ACCESS__KEY: "«secret-access-key»"
    OPENPROJECT_FOG_CREDENTIALS_PATH__STYLE: "true"
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.yml 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/openproject:15
cvgd4c4at61i        openproject_db       replicated          1/1                 postgres:13
uvtfnc9dnlbn        openproject_proxy    replicated          1/1                 openproject/openproject:15   *:8080->80/tcp
g8e3lannlpb8        openproject_seeder   replicated          0/1                 openproject/openproject:15
canb3m7ilkjn        openproject_web      replicated          1/1                 openproject/openproject:15
7ovn0sbu8a7w        openproject_worker   replicated          1/1                 openproject/openproject:15

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/openproject:15
cvgd4c4at61i        openproject_db       replicated          1/1                 postgres:10
uvtfnc9dnlbn        openproject_proxy    replicated          2/2                 openproject/openproject:15   *:8080->80/tcp
g8e3lannlpb8        openproject_seeder   replicated          0/1                 openproject/openproject:15
canb3m7ilkjn        openproject_web      replicated          6/6                 openproject/openproject:15
7ovn0sbu8a7w        openproject_worker   replicated          1/1                 openproject/openproject:15

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.