On Thu, Apr 08, 2021 at 07:14:03PM +0200, Thomas Bohl wrote: > Hi, > > > In the event that the local delivery fails with a 5xx error, OpenSMTPD > > queues the message for retry (and eventually supplies the sender with > > a "delayed" delivery notification, etc). My naive expectation is that > > SMTP 5xx means permanent failure, so an immediate bounce would be more > > appropriate. > (if I read it right) according to RFC 1893 both codes should be treated as > permanent failures. I don't see anything in RFC 2033 that would negate that > for LMTP. > > Having said that: Accepting a message as an MX only to immediately let it > bounce is bad practice. I have last seen such a setup 15 years ago. Where > the MX would accept mails for subdomains it had no mailbox- knowledge. > Before the system was finally overhauled 90% of the data centres internet > upload was bounce messages! Insane! > > [...] > You should seriously bring the mailbox status to the front!
Hi Thomas, You are right -- this is exactly the kind of insight I was hoping for! Now I understand the logic behind these retries, and it makes perfect sense. Clearly, the ideal solution then is to not accept mail for delivery (reject it at the RCPT TO stage) on the system boundary, unless we are willing to do whatever it takes (spend extra resources if needed) to deliver it to a mailbox. I can easily see how I could reject mails for nonexistent recipients at this stage (thanks to Reio's earlier email). I will need to look into the case of the full mailbox, so that gets inbound mail rejected up-front as well. Ideally, this would be more nuanced than taking the mailbox completely off the table (pun intended), so that the rejection would mention "mailbox full", rather than "no such user", since that might confuse senders in a way that is bad for PR... (: Many thanks, Tom
