On Sun, Nov 05, 2006 at 03:32:13PM +0300, Egor Ermakov wrote:
> 21:43:42 69670 SMTP<< 354 Go ahead
> 21:43:42 69670 SMTP>> writing message and terminating "."
> 21:43:42 69670 writing data block fd=8 size=8189 timeout=300
[...]
> 21:43:53 69670 writing data block fd=8 size=8189 timeout=300
> 21:43:53 69661 selecting on subprocess pipes
> 21:43:54 69670 writing data block fd=8 size=8189 timeout=300
[...]
> 21:44:00 69670 writing data block fd=8 size=8189 timeout=300
> 21:44:19 69670 writing error 32: Broken pipe
> 21:44:19 69670 ok=0 send_quit=0 send_rset=1 continue_more=0 yield=1
> first_address is NULL
> 21:44:19 69670 LOG: MAIN
> 21:44:19 69670 alt1.gmail-smtp-in.l.google.com [72.14.205.27]: Broken pipe
>
> It usually happens with medium and large message (letter with size
> less than 100kb is always delivered). As i can see from logs, first
> portion of data goes in instant and then delay appears. It may look at
> network problem, but there is 2 things:
ok, so what's happening here is that the delivery process
is seeing its connection to the remote server being reset
by the remote side (which is reported up to userspace as
EPIPE). Does this happen consistently for particular
messages? If so, does it happen after a particular part of
the message body is transmitted? Can you record the
conversation using something like strace or (better)
tcpdump, to see where the connection reset is occuring?
--
``Of course we are not patronising women. We are just going to explain to
them, in words of one syllable, what it is all about.'' (Olga Maitland)
