On 12/31/2017 01:30 PM, Jorge Bastos wrote:
It's that what I want!! Sequentially, my bad saying random!
The trickier parts of coding LDAs - with locking/semaphores and such - address the case of *(near) simultaneous* arrival of e-mails. In your case, you wouldn't need to lock "the mailbox" (if info@ even *has* one) but wherever the info "which address did we *last* forward to?" is stored. Do you actually need 100% correctness (a. of sequential selection, or at least b. of even distribution with 33% each) in that corner case? It's not like Received: headers would usually include timestamps with subsecond precision to base a retrospect analysis on ...
(Out of interest, if someone were to send an e-mail to you - say, order something from your company - and sends a reply-to-all to his own copy before anyone answered the original mail - say, "oops, I forgot to tell you the delivery address, here it is" -, do you really want those two mails to go to *different* recipients with 66+% probability?)
I'ld guess that such a function can be had in the final MTA, in dovecot acting as MDA, in the alias-resolving mechanism (if separate), maybe even in the mechanism generating out-of-office autoreplies (if tweakable). Deciding which route would be the *easiest* to take requires more details of your setup, though ...
Kind regards,
Jochen Bern Systemingenieur
www.binect.de