>>>>> On Sun, 11 Feb 2001 12:44:32 -0500,
>>>>> Andy Hubbell, <[EMAIL PROTECTED]> (ah) writes:

ah> Should the master process exec a new lmtpd for each delivery?  This
ah> would seem somewhat wasteful, especially on a busy system with lots
ah> of disk IO...  Nothing else is talking lmtp to this server so it
ah> should only be getting lmtp connections on the socket.

You'll probably want to tweak some settings further.  Here are a few
things I've got going:

    /etc/postfix/main.cf:

        # Be careful here.  Make sure your box has the resources to
        # handle what ever you set this to!
        lmtp_destination_recipient_limit = 3000

        # Allow a service to be used this many times before recycling.
        max_use = 200

    /etc/postfix/master.cf:

        # Setting max number of lmtp processes to 10.
        lmtp      unix  -       -       n       -       10      lmtp


Something Wietse points out is that you want to avoid the
destination host from switching too much.  Here's from the
LMTP_README in snapshot-20010128:

You can prevent the LMTP client from switching between servers by
configuring a separate mail delivery transport for each LMTP server:

    /etc/postfix/master.cf:

        lmtp1      unix  -       -       n       -       -       lmtp
        lmtp2      unix  -       -       n       -       -       lmtp
          .         .    .       .       .       .       .        .

Configure transport table entries such that the lmtp1 mail delivery
transport is used for all deliveries to the LMTP server #1, the
mail lmtp2 transport for the LMTP server #2, and so on.

    /etc/postfix/transport:

        foo.com lmtp1:lmtp1host
        bar.com lmtp2:lmtp2host


Obviously, for UNIX domain sockets, just change the "lmtp1:lmtp1host"
to the appropriate thing, such as "lmtp1:unix:/path/name".

The reason why I set the lmtp service to a particular number is so
that I can throttle things back a bit.  That way if the SMTP host
gets a flood of mail, it won't blast the Cyrus server so hard that
performance crumbles.  Personally, it *seemed* like this also helped
connection caching by limiting the likelihood that another lmtp
connection would be kicked off, but I suspect Wietse's analysis is
more accurate.  It's been working fine as is, so I didn't dig
further.  Only so much time to play, alas.

-- 
Amos

Reply via email to