>> We have discovered an issue in our postfix configuration that appears to >> limit the amount of recipients permitted in a single message to 100. > >This is not the case by default.
I thought it must be but what I'm seeing corresponds to a limit of 100 somewhere, false theory debunked thanks. > Perhaps you're sending invalid recipients and triggering error > delays. Or you have a slow virtual alias table lookup table > (SQL, ...) and by the 100th recipient the pipeline between > smtpd(8) and cleanup(8) stalls because cleanup recipient > rewriting is not keeping up. ... Interesting, I'll see what info I can find on this. As you might see from the logs I've just submitted (http://pastebin.centos.org/36256/, http://pastebin.centos.org/36261/) I was using mailinator addresses to test which I believe just accepts mail to its domain. > Which determines the maximum number of possible concurrent SMTP > connections, not the number of recipients per message. Again, makes perfect sense. > >> and also, according to the Tuning Guide the >> smtpd_recipient_limit option configures “The maximal number of recipients >> that the Postfix SMTP server accepts per message delivery request.” > > You're making this part up. It is simply not true.[ > Not made up! Quoted from http://www.postfix.org/postconf.5.html#smtpd_recipient_limit > >This sounds like a timeout making a connection, not a timeout >mid-connection. Was Thunderbird configured to make a separate >connection for each recipient? Is this some sort of "mail-merge" >application with personalized messages to each recipient? Connection is timed out after the 100th RCPT TO according to the tcpdump log http://pastebin.centos.org/36261/ I'm not sure if Thunderbird is configured this way I'll have to dig into this but it doesn't seem so based on the SMTP conversation in the packet trace. >smaller personal mail-servers. Beefier servers with lots memory, >CPU, disk and network may well be able to handle 1000 or more >concurrent connections. > >That said, your sending software does not benefit from opening a >huge number of connections, around 20 is usually quite enough for Getting pretty convinced now that this is not a multiple connection issue I'll have to dig into how to print anvil stats to see if I can prove it. Thanks, Matt
