Problems with $forward_method and list of peer socket specifications

[Mark Martinec]

Michael,

> I have some problems with $forward_method and list of peer socket
> specifications:
> 
> 1) The randomization of the destinations does not work correctly. If
> you have 2 values in the list, always the first one is used. The
> correction for this error is, changing
>        my($j) = int(rand($n_candidates-1));
> to
>        my($j) = int(rand($n_candidates));
> because rand(X) is always less than X.

A bug indeed. What was I thinking?!
Will be fixed in 2.7.1 and 2.8.0.

> 2) There is a case where the amavisd process hangs, when the
> destination MTA is gone away:
> 
> Apr 12 15:44:53 lxmhs36a amavis[32200]: (32200-01-5) about to connect
> to (smtp:mailrelay2.mail.lrz.de:625, smtp:mailrelay1.mail.lrz.de:625),
> FWD from <conferences at mail.elsevier-alerts.com> ->
> <felix.sedlmeyer at wzw.tum.de>
> Apr 12 15:44:53 lxmhs36a amavis[32200]: (32200-01-5) get_deadline
> fwd_init - deadline in 1198.5 s, set to 1199.000 s
> Apr 12 15:44:53 lxmhs36a amavis[32200]: (32200-01-5) smtp session reuse
> (smtp:[mailrelay1.mail.lrz.de]:625), 4 transactions so far
> Apr 12 15:44:53 lxmhs36a amavis[32200]: (32200-01-5)
> establish_or_refresh, state: quitsent
> Apr 12 15:44:53 lxmhs36a amavis[32200]: (32200-01-5) rw_loop:
> needline=0, flush=1, wr=1, timeout=1
> Apr 12 15:44:53 lxmhs36a amavis[32200]: (32200-01-5) rw_loop: receiving
> Apr 12 15:44:53 lxmhs36a amavis[32200]: (32200-01-5) rw_loop read: got
> eof
> Apr 12 15:44:53 lxmhs36a amavis[32200]: (32200-01-5) rw_loop: sending
> Apr 12 15:44:53 lxmhs36a amavis[32200]: (32200-01-5) mail_via_smtp:
> session failed: Error writing to socket: Broken pipe at
> /usr/local/sbin/amavisd line 6884.
> ...
> Apr 12 15:44:54 lxmhs36a amavis[32200]: (32200-01-6) smtp session reuse
> (smtp:[mailrelay1.mail.lrz.de]:625), 4 transactions so far
> Apr 12 15:44:54 lxmhs36a amavis[32200]: (32200-01-6)
> establish_or_refresh, state: quitsent
> Apr 12 15:44:54 lxmhs36a amavis[32200]: (32200-01-6) rw_loop:
> needline=0, flush=1, wr=1, timeout=1
> Apr 12 15:44:54 lxmhs36a amavis[32200]: (32200-01-6) rw_loop: receiving
> Apr 12 15:44:54 lxmhs36a amavis[32200]: (32200-01-6) rw_loop read: got
> eof
> Apr 12 15:44:54 lxmhs36a amavis[32200]: (32200-01-6) rw_loop: sending
> Apr 12 15:44:54 lxmhs36a amavis[32200]: (32200-01-6) mail_via_smtp:
> session failed: Error writing to socket: Broken pipe at
> /usr/local/sbin/amavisd line 6884.
> 
> and so on till the process is killed.
> 
> Call chain:
> 
> establish_or_refresh -> $smtp_handle->close -> $self->flush ->
> $self->rw_loop(0,1) -> die "Error writing to socket ..."
> 
> If the '$self->flush' is evaled, the error disappears, but I do not
> know if this is the correct solution.

Will prepare a bug fix, haven't done so yet.
 
> 3) If a crashed MTA comes back, it takes too long, if it is used again,
> if there is a high load. I think the reason is, that a cached connection
> is used indefinitely, if it is not idle or the process ends its life. It
> would be nice if we could limit the lifetime of such a connection, for
> example if we could specify a maximal lifetime and/or the maximal number
> of transactions.

Makes sense, will do something about it.

Thanks a lot!

  Mark