Over on the Fediverse, I had a question :

To the sysadmin population of the Fediverse: do people have any numbers on how long common mail senders will retry sending mail if your MX is unreachable? Once upon a time people retried for many days, but my impression is that quite a few places now stop trying and bounce the email after quite short intervals, like a day.

(Boosts and practical experiences welcome, like "my MX was down for three days and I still got all that email sent from GMail".)

The context for my sudden curiosity is that there's a scheduled all-weekend, whole building power outage at the start of May for the building with our machine room. It seems likely that basically all of our systems will be down for roughly two and a half days, and longer if things go wrong, and this obviously includes our incoming email gateway.

As I mentioned, in the old days you could definitely expect mail systems to retry for more than a long weekend and so we wouldn't really worry about it. But I'm not sure about that in practice any more, hence my sudden curiosity. Based on replies to that post (and some additional research), common Unix MTA software still seems reasonably okay on retry durations; postfix and sendmail default to five days , while exim more or less defaults to four . RFC 5321 recommends four to five days in section 4.5.4.1 , for what an RFC on SMTP mail is worth these days.

Unfortunately, what matters in practice is how the dominant sources of your email behave, and generally those aren't going to be people running normal Unix mailers in normal configurations. A lot of our email comes from GMail and Office 365, who are obviously using custom mail systems. Office365 covers email from both people at other universities or organizations that use Office 365 and people using the university's central email system to send email to people in my department . It's also possible that configurations vary between organizations using Office365.

There are also all of the people and organizations sending out newsletters, notifications, and so on through the various mailing list service providers, like Amazon SES. These organizations may well have shorter retry times than for individual human-generated email from GMail, Office365, and so on. Another category of email is activity notification emails from places like Github, which people may also want to (eventually) get.

(We have access to an alternate location with a different network and power setup, so we could deploy a backup MX machine to there. There are some potential drawbacks to that, but we may do it as a precaution.)