Tuesday June 14th, 2022 NCLUG Meeting

[Bob Proulx]

Bob Proulx wrote:
> I had researched this problem and found that the wisdom of the net for
> "slow recipients" would be to create a custom transport for "turtles"
> (which I found particularly amusing was Google) and then deliver
> messages to turtles more slowly.

The best way to find an answer on the Internet is to post an incorrect
answer.  I posted this solution to the postfix list and then got
corrections as to how it should have been done.  :-)

> I modified the turtle examples changing the names from turtle to gmail
> and basically set up this configuration.
>
>     gmail unix - - y - - smtp
>         -o gmail_destination_concurrency_limit=2
>         -o gmail_destination_rate_delay=1s
>         -o gmail_destination_recipient_limit=2

Viktor Dukhovni (authoritative resource) wrote:

    No, those settings are used by the queue manager to schedule deliveries
    assigned to various delivery agent processes, the delivery agents
    themselves only see one message at a time and can do little to affect
    concurrency, and related limits.

    So these settings go in main.cf:

        gmail_destination_concurrency_limit = 2
        gmail_destination_rate_delay = 1s
        gmail_destination_recipient_limit = 2

I moved those settings to the main.cf file as instructed.

> Using this transport mapping for that transport.
>
>     gmail.com gmail:

    Sure, but the trailing ":" is unnecessary.

Noted.  Removed.

Upon the next queue run Google then accepted the exact messages that
it has previously not been accepting.  The three messages that Google
has not been accepting since Monday have now been accepted using the
newly defined slow transport above.  Will let this configuration
simmer on the stove and slow cook for a while but this looks to be a
good recipe for delivering mail to Google.

What is actually needed is a transport map that matches against the MX
destination.  I don't think such a feature exists.  I have asked about
it and will see what response to that question is returned.  But until
then this seems to be a good improvement.

Bob