This container does the job of receiving an e-mail for a specified domain and spawning an instance of another container to do "something" with the e-mail. That's it. All very simple and straightforward. You would think...

Minimal configuration requires you to specify the domain you're receiving mail for, and how to connect to your Discourse instance (URL, API key, etc). This involves setting the following environment variables:

• MAIL_DOMAIN -- the domain name(s) to accept mail for and relay to Discourse. Any number of space-separated domain names can be listed here.

• DISCOURSE_BASE_URL -- the base URL for this Discourse instance. This will be whatever your Discourse site URL is. For example, https://discourse.example.com. If you're running a subfolder setup, be sure to account for that (ie https://example.com/forum).

• DISCOURSE_API_KEY -- the API key which will be used to authenticate to Discourse in order to submit mail. The value to use is shown in the "API" tab of the site admin dashboard.

• DISCOURSE_API_USERNAME -- (optional) the user whose identity and permissions will be used to make requests to the Discourse API. This defaults to system and should be OK for 99% of cases. The remaining 1% of times is where someone has (ill-advisedly) renamed the system user in Discourse.

For a straightforward setup, the above environment variables should be enough to get you up and running. If you have a desire for a more complicated setup, the following subsections may provide you with the power you need.

You can setup any Postfix configuration variables you need by setting env vars of the form POSTCONF_<var> with the value of the variable you want. For example, if you wanted to add a pre-delivery milter, you might use:

-e POSTCONF_smtpd_milters=192.0.2.42:12345

These checks are enabled by default in the latest image discourse/mail-receiver:release.

You may also decide whether or not to set up a Postfix server that has SPF, DKIM and DMARC checks configured by setting the build argument INCLUDE_DMARC to true or false when building the docker container for mail-receiver:

docker build --build-arg INCLUDE_DMARC=true -t local_discourse/mail-receiver:latest .

Configurations for these checks are stored in their respective configuration files policyd-spf.conf, opendkim.conf, opendmarc.conf in this repository.

The BLACKLISTED_SENDER_DOMAINS environment variable accepts a space-separated list of domain names. Mail messages from these senders will be fast-failed with SMTP code 554.

Postfix loves to log everything to syslog. In fact, that's really all it supports. Since, by default, Docker is not known for its superlative out-of-the-box syslog integration, this container runs a tiny script which reads all syslog data and dumps it to the container's stderr (which is then examinable by docker logs).

If, by some chance, you want to process your Postfix logs more extensively, you can set SOCKETEE_RELAY_SOCKET and all syslog messages will also be sent to that socket for further processing.

Every e-mail that is received is delivered to a custom discourse service. That service, which is a small Ruby program, makes a POST request to the admin interface on the specified URL (DISCOURSE_BASE_URL), with the key and username specified. Discourse itself stands ready to receive that e-mail and process it into the discussion, in exactly the same way as an e-mail received via POP3 polling.

Before delivery to the discourse service, a Postfix policy handler runs, asks Discourse if either the sender and/or recipient are invalid, and if so, rejects the incoming mail during the SMTP transaction, to prevent Discourse later sending out reply emails due to incoming spam ("backscatter"). Legitimate users will be notified of the failure by their MTA, and obvious spam just gets dropped without reply. This step is just about being a good citizen of the Internet and not full spam filtering.

When changing files in the lib/ directory, make sure to publish a new rubygems version once you are done by bumping the version in discourse_mail_receiver.gemspec and tagging a version starting with v in git e.g. v1.0.4, then push.