Whit Blauvelt:
> Hi,
>
> We have two ISPs, and a single Postfix server behind our Linux firewall that
> gets NATed to in the same way from IPs on each of those ISPs' lines, with
> both of those given in the MX records. After getting complaints that some
> incoming mail was getting delayed by 5 or 10 minutes - sometimes much longer
> - I sent a bunch of test mail from elsewhere, and saw that my test emails
> all arrived either instantly or precisely 5 minutes late.
>
> Turns out one of the incoming lines times out on the response, which gets
> logged on the sender but not the receiving system. Since most all mail is
> only 5 or 10 minutes late - assuming the sending system behaves well and
> goes to the next MX record in 5 or 10 minutes - it's not a particularly
> visible problem - except for some sending systems that don't behave well,
> which are out there.
>
> In telnetting to the two IPs, with the one that works the 220 line comes up
> right away. On the bad line, the "Connect to" and "Escape character is '^]'"
> lines come right up, but not the 220. If I then hit Enter the 220 line comes
> right up, but not without that.
>
> Turning off tcp_window_scaling on receiver, sender, or both doesn't change
> this. But the problem is specific to one ISP. Is this familiar to anyone?
Looks like some proxy is plugged in the wrong way (waiting for the
client to speak first). With SMTP the server speaks first.
Given zero details on Postfix configuration, I won't waste time
with random guesses.
Wietse
