> Anyone know what the reason for this might be? Is this a general problem > of the postfix-queue plugin?
The postfix-queue plugin uses a socket to the cleanup daemon. I'm not sure if the connection over the socket is slow (the cleanup daemon might be slow reading data from it) or if there is some additional processing which takes time. An alternative path would be to drop the mail into the postdrop queue and the notify the pickup-daemon. This would have the advantage that qpsmtpd would not need to wait for an answer. A disavantage would be that if something goes wrong after putting the mail in the queue (e.g. pickup-daemon not accepting it), it would not reflect on the SMTP return code. Another (minor) disavantage would be that doing it this way, Postfix would insert another Received-Line (but who cares). I have just written a very stupid queue plugin which uses Mail::Postfix::Postdrop to inject the mail. Interestingly enough it seems that the code from our Postfix-queue Plugin was copied from there (we e.g. have a comment which still talks about speaking to the pickup daemon). I tested my plugin and while it still took 16 seconds to queue the mail (10 MB attachemnet) , this is certainly much better than the more than two minutes of the current queue plugin. Therefore I suggest that we change either the current plugin (which for reasons unknown to me uses a library which was made directly part of qpsmtpd) or add a second one (the internal library would need to be slightly modified). Regards Michael
