dokku-logging-supervisord is a plugin for dokku that injects supervisord to run applications and redirects stdout & stderr to app/process specific log files (rather than the Docker default per-container JSON files).

This plugin uses the docker-args hook to inject the data volume argument. As such, at the moment it only works with the development version of Dokku.

# Create the directory to house the log files:
sudo mkdir -p /var/log/dokku
sudo chown dokku:dokku /var/log/dokku

# Install the plugin:
git clone https://github.com/sehrope/dokku-logging-supervisord.git /var/lib/dokku/plugins/logging-supervisord

All future deployments will use this plugin to start all processes and all log output will be in /var/log/dokku/$APP/.

Normally, dokku only runs the web process within Procfile. The dokku-logging-supervisord plugin will run all process types (web, worker, etc.) and will restart crashed applications.

Additionally, it creates and binds a shared directory for each app from /var/log/dokku/$APP on the host machine to /var/log/app in the app's container. The supervisord config is setup to have each process in your Procfile send it's stdout and stderr to a separate file in that directory named $PROCESS_NAME.$PROCESS_NUM.log. Output for the supervisord process itself (startup/shutdown notices, etc) will be logged to a file named supervisor.log in the same log directory.

If you have an app myapp with a Procfile that looks like this:

web: node web.js
worker: node worker.js

And you push your app with Dokku like this:

$ git push [email protected]:myapp master

Then upon starting it you would have log files at:

/var/log/dokku/myapp/supervisor.log
/var/log/dokku/myapp/web.00.log
/var/log/dokku/myapp/worker.00.log

This plugin supports running multiple of the same process type. At start it checks for a file in the apps home diretory named SCALE. The file should be a series of lines of the form name=<num> where name is the process name and <num> is the number of processes of that type to start.

Example:

web=1
worker=5
clock=1

If the file does not exist then a single process of each type will be created for each process type in Procfile. Additional lines in the file ignored.

Logs for each process will go to separate log file in /var/log/dokku/$APP/process.<num>.log

Note: All the processes will run in same Docker container. They do not run in separate containers. This means that if you have multiple "web" processes they will each try to listen on the same PORT environment variable. For this to work properly you should use the socket option SO_REUSEPORT. If that is not available then you will need to stick with a single web process.

Rather than editing the file manually you can use the command:

dokku scale myapp web=1 worker=6

This will generate a new SCALE file and then deploy the app. An app rebuild will not happen. It will just kill and restart your application.

Adding the SCALE file is done by copying it into the container. This adds another layer to the container's AUFS. As there is a max number of layers you may need to occasionally run a rebuild (try dokku rebuild myapp) to rebase the container.

• Better handle log file rotation
• Add date/time to log output
• Have the application runner see the scale file on the host so we don't have to copy it (volume mount?)

Thanks to dokku-supervisord and dokku-persistent-storage as this plugin is really a combination of those two.

This plugin is released under the MIT license. See the file LICENSE.