I recently made a change to how Healthchecks sends transactional email. Before:
The Healthchecks Django app is directly connecting to a 3rd-party SMTP relay (think AWS SES, SendGrid, Mailgun, but in our specific case it is Elastic Email), and sends SMTP commands over a TLS-encrypted connection. If the send operation fails, the Django app retries a couple times, then gives up and the email is lost.
A local OpenSMTPD instance runs on the same machine as the Django app. It accepts connections from local clients only, and relays all received messages to the external SMTP relay operated by, in our case, Elastic Email.
In this setup, the Django app can quickly hand off the outgoing emails to OpenSMTPD, and OpenSMTPD retries failed sends for minutes, hours or even days. If the 3rd-party SMTP relay has an outage, emails are not lost, just delayed. At least that’s the theory – we shall see how well this works in practice.
The OpenSMTPD configuration for this use case is surprisingly compact:
table secrets file:/etc/mail/secrets listen on lo inet4 port 25 action "relay" relay host smtp+tls://firstname.lastname@example.org:587 auth <secrets> match for any action "relay"
I also experimented with Postfix (as recommended here), and it gets the job done too. I also considered more lightweight relay-only MTAs: dma and nullmailer. Neither supports listening on port 25, instead you enqueue emails by piping data to
/usr/sbin/sendmail. This complicates integration with the Django app somewhat. I ultimately went with OpenSMTPD because it seemed to have the right balance of features and simplicity.