Skip to content

Dockerfile Deployment

New as of 0.3.15

While Dokku normally defaults to using heroku buildpacks for deployment, you can also use docker's native Dockerfile system to define a container.

Dockerfile support is considered a Power User feature. By using Dockerfile-based deployment, you agree that you will not have the same comfort as that enjoyed by Buildpack users, and Dokku features may work differently. Differences between the two systems will be documented here.

To use a dockerfiles for deployment, commit a valid Dockerfile to the root of your repository and push the repository to your Dokku installation. If this file is detected, Dokku will default to using it to construct containers except in the following two cases:

  • The application has a BUILDPACK_URL environment variable set via the dokku config:set command or in a committed .env file. In this case, Dokku will use your specified buildpack.
  • The application has a .buildpacks file in the root of the repository. In this case, Dokku will use your specified buildpack(s).

Exposed ports

Changed as of 0.5.0

Dokku will extract all tcp ports exposed using the EXPOSE directive (one port per line) and setup nginx to proxy the same port numbers to listen publicly. If you would like to change the exposed port, you should do so within your Dockerfile and app.

NOTE: Nginx does not support proxying UDP. UDP ports can be exposed by disabling the nginx proxy with dokku proxy:disable myapp

If you do not explicitly EXPOSE a port in your Dockerfile, dokku will configure the nginx proxy to listen on port 80 (and 443 for TLS) and forward traffic to your app listening on port 5000 inside the container. Just like buildpack apps, you can also use the $PORT environment variable in your app to maintain portability.

When ports are exposed through the default nginx proxy, they are proxied externally as HTTP ports. At this time, in no case do we proxy plain TCP or UDP ports. If you'd like to investigate alternative proxy methods, please refer to our proxy documentation.

Customizing the run command

By default no arguments are passed to docker run when deploying the container and the CMD or ENTRYPOINT defined in the Dockerfile are executed. You can take advantage of docker ability of overriding the CMD or passing parameters to your ENTRYPOINT setting $DOKKU_DOCKERFILE_START_CMD. Let's say for example you are deploying a base nodejs image, with the following ENTRYPOINT:

ENTRYPOINT ["node"]

You can do:

dokku config:set APP DOKKU_DOCKERFILE_START_CMD="--harmony server.js"

To tell docker what to run.

Setting $DOKKU_DOCKERFILE_CACHE_BUILD to true or false will enable or disable docker's image layer cache. Lastly, for more granular build control, you may also pass any docker build option to docker, by setting $DOKKU_DOCKER_BUILD_OPTS.

Procfiles and Multiple Processes

New as of 0.5.0

You can also customize the run command using a Procfile, much like you would on Heroku or with a buildpack deployed app. The Procfile should contain one or more lines defining process types and associated commands. When you deploy your app a Docker image will be built, the Procfile will be extracted from the image (it must be in the folder defined in your Dockerfile as WORKDIR or /app) and the commands in it will be passed to docker run to start your process(es). Here's an example Procfile:

web: bin/run-prod.sh
worker: bin/run-worker.sh

And Dockerfile:

FROM debian:jessie
WORKDIR /app
COPY . ./
CMD ["bin/run-dev.sh"]

When you deploy this app the web process will automatically be scaled to 1 and your Docker container will be started basically using the command docker run bin/run-prod.sh. If you want to also run a worker container for this app, you can run dokku ps:scale worker=1 and a new container will be started by running docker run bin/run-worker.sh (the actual docker run commands are a bit more complex, but this is the basic idea). If you use an ENTRYPOINT in your Dockerfile, the lines in your Procfile will be passed as arguments to the ENTRYPOINT script instead of being executed.