wjlove / docker-mosquitto

Easily configurable eclipse MQTT broker

mosquitto

These Docker images behave almost exactly as the official Eclipse mosquitto server images, while providing easy configuration of most parameters through environment variables. They are also ready for TLS connections using official root certificates out-of-the-box. Finally, whenever relevant files pointed at by the Mosquitto configuration change, this implementation will arrange for mosquitto to reload its configuration to take the changes into account.

Simpler Cases

For all known options present in the default configuration file, it is possible to set their value through creating an environment variable starting with MOSQUITTO_ and continuing with the same name as the option, but in upper case. So, for example, to change the retry interval to 10 seconds, an interval controlled by the configuration option retry_interval, you would set the variable MOSQUITTO_RETRY_INTERVAL to 10.

Sub-Sectioning

This simple heuristic will not work all the time, as Mosquitto divides configuration files into logical sub-sections and several options within these sub-sections share the same name. This image provides a workaround based on the fact that each sub-section is prefixed with a specific comment header in the default configuration file, a file that is made part of the original image upon which this image is based. The implementation is currently able to detect the following sections:

• default will contain options for the default listener.
• extra will contain options for an additional listener. There is currently no support for further listeners.
• persistence will contain options for persistence.
• logging will contain options for logging.
• security will contain options for security, authorisations, ACLs, etc.
• bridges will contain options for bridging mosquitto installations together for improved scalability.

For example, the content of the default section starts with the following header in the file (up to the next header):

# =================================================================
# Default listener
# =================================================================

To start making use of this sectioning of the configuration file, you have to specify an include directory. This can be achieved using a configuration file with a given value for include_dir, or by specifying the environment variable MOSQUITTO_INCLUDE_DIR. Whenever the directory is specified, the main configuration file will be automatically sliced into a number of configuration files, named as described above and created in the include directory. A backup of the original configuration file will be made and a new configuration file where all sub-sections have been removed will be created.

In this context, it becomes possible to use specially crafted environment variables to address options in the different sub-sections. These variables should start with MOSQUITTO__ (note the double underscore), followed by the name of the section (as in the list above) in uppercase, followed by __ (two underscores again), followed by the name of the option in upper case. So, to specify a different port for the default listener, you could for example, set MOSQUITTO_INCLUDE_DIR to /mosquitto/config/sections/ to trigger this mechanism and then set MOSQUITTO__DEFAULT__PORT to 2883.

Note that while logical sections makes it easier to grasp the entire configuration, some of them are not entirely necessary. For example, the persistence section does not seem to have option names that clash with other sections. If you are sure about this, you can set variables directly using "single" underscores. Consequently, there are, for example two ways of turning on persistence:

• You could set the variable MOSQUITTO_PERSISTENCE to true and set the variable MOSQUITTO_PERSISTENCE_LOCATION to /mosquitto/data/
• You could also reason with sections and set the variables MOSQUITTO__PERSISTENCE__PERSISTENCE and MOSQUITTO__PERSISTENCE__PERSISTENCE_LOCATION, respectively.

In general, sub-sectioning is more deterministic.

Using from Compose

These images makes configuration much more visible (and flexible) from compose files whenever deploying mosquitto containers. Below is a somewhat constructed example:

volumes:
  persistence:
    driver: local

services:
  mosquitto:
    image: efrecon/mosquitto:1.5.5
    volumes:
      - persistence:/mosquitto/data
    environment:
      - MOSQUITTO_INCLUDE_DIR=/mosquitto/config/sections/
      - MOSQUITTO__PERSISTENCE__AUTOSAVE_ON_CHANGES=true
      - MOSQUITTO__PERSISTENCE__AUTOSAVE_INTERVAL=100
      - MOSQUITTO__PERSISTENCE__PERSISTENCE=true
      - MOSQUITTO__PERSISTENCE__PERSISTENCE_LOCATION=/mosquitto/data/
      - MOSQUITTO__LOGGING__LOG_DEST=stderr
    ports:
      -
        target: 1883
        published: 1883
        protocol: tcp
        mode: host
    logging:
      driver: "json-file"
      options:
        max-size: "1m"
        max-file: "10"
    deploy:
      restart_policy:
        delay: 10s
        max_attempts: 10
        window: 60s
      replicas: 1

Jumping into the container with docker exec would show a modified version of the regular configuration file at /mosquitto/config/mosquitto.conf and several configuration files in the directory /mosquitto/config/sections/, one for each of the sections supported by this implementation.

Complex Cases

For even more complex cases, you will have to provide your own configuration file by hand. If this is the case, there is probably little point in preferring this image over the official Eclipse image. However, you might still want to mix behaviours. As long as you provide a configuration that contains the same sections (in comments) as the original configuration file, you should be able to benefit from this implementation. Note that the entrypoint requires the file to be located at /mosquitto/config/mosquitto.conf.

Security

Secrets

For security reasons, you would probably not want to expose the value of secrets using environment variables. A possible workaround is to copy the default configuration file inside your own project, modify sensitive data in the file and mount it/copy it into the container. As long as your copy keeps track of all possible options and their default values as lines that are commented away, you would still be able to tune the configuration externally as explained in the previous paragraph. Note that most secret information in mosquitto is offloaded to other files than the main configuration file, so in most cases you will be safe to point at these files through environment variables; while still making sure that the files containing secret information are present within the container.

TLS

In order to provide TLS encryption, you will have to add an extra listener to mosquitto. Provided you have official key and cert for a host, you could place them in a volume (tls in the example below) and adapt the following compose file:

volumes:
  persistence:
    driver: local
  tls:
    driver: local

services:
  mosquitto:
    image: efrecon/mosquitto:1.5.5
    volumes:
      - persistence:/mosquitto/data
      - tls:/mosquitto/config/security
    environment:
      - MOSQUITTO_INCLUDE_DIR=/mosquitto/config/sections/
      - MOSQUITTO__PERSISTENCE__AUTOSAVE_ON_CHANGES=true
      - MOSQUITTO__PERSISTENCE__AUTOSAVE_INTERVAL=100
      - MOSQUITTO__PERSISTENCE__PERSISTENCE=true
      - MOSQUITTO__PERSISTENCE__PERSISTENCE_LOCATION=/mosquitto/data/
      - MOSQUITTO__LOGGING__LOG_DEST=stderr
      - MOSQUITTO__EXTRA__LISTENER=8883
      - MOSQUITTO__EXTRA__CAPATH=/etc/ssl/certs/
      - MOSQUITTO__EXTRA__KEYFILE=/mosquitto/config/security/yourhostname.key
      - MOSQUITTO__EXTRA__CERTFILE=/mosquitto/config/security/yourhostname.crt
    ports:
      -
        target: 8883
        published: 8883
        protocol: tcp
        mode: host
    logging:
      driver: "json-file"
      options:
        max-size: "1m"
        max-file: "10"
    deploy:
      restart_policy:
        delay: 10s
        max_attempts: 10
        window: 60s
      replicas: 1

This can be combined with an auto-renewing reverse proxy such as caddy to protect access to your mosquitto container using certificates from Let's Encrypt. You will then have to share the volume where caddy stores handshaked certificates with your mosquitto container and adapt the path to access the key and certificate.

Configuring the Entrypoint

The entrypoint itself can be configured either by command-line options, or through environment variables, all starting with MQ_. Command-line options have precedence over environment variables. The image default is to provide feedback about the transformations applied by this implementation through setting the --verbose flag.

Reloading

When running with sectioning, which is triggered by setting the environment variable MOSQUITTO_INCLUDE_DIR or setting the configuration parameter include_dir, this implementation is able to detect changes in files that are pointed at by the configuration, e.g. password file, server certificate or key, etc. This is controlled by the command-line option --watcher to the entrypoint (or the environment variable MQ_WATCHER); a good default that should work in most cases is provided.

The implementation will look into the sectioned configuration files for the values of known configuration parameters and, for each, start a process that will watch the file pointed at by the parameter for changes. When the file changes, the SIGHUP process is sent to mosquitto, which will then reload its configuration (see section about signals in the manual).

File watching is implemented using watch.sh, signalling using signal.sh. The signalling implementation will actively look for a probable process if the PID file does not exist.

Automated Builds

Builds will happen automatically for all current and future versions of the official images by the way of the scripts in the hooks directory. This means that versioning and tagging of these images will match the official Docker library. Currently, version discovery is based on reg-tags, a shell library present as a submodule to this repository.

Implementation

All substitution and slicing occurs from docker-entrypoint.sh. Substitution in the various configuration files occurs at the shell level. Slicing the main configuration file into sub-sections required a more complex algorithm and is implemented in the Tcl script slicer.tcl. The script is called twice, once for detecting the location of the include directory, and a second time to create the various sub-section files.

Configuration of the entrypoint can occur through environment variables starting with MQ_. The value of these variables is also used in the processes that are started from the entrypoint to watch relevant files for changes. The entire process tree forming the implementation is placed under the control of tini to ease garbage collection of processes when containers are killed or stopped.