Re: [fetchmail-users] strange polling errors

[Matthias A. <mat...@gm...>]

Am 28.02.2011 11:43, schrieb Jonathan Buschmann:
> Hi Matthias,
> Your right that it is somehow a syslog problem. --nodetach --nosyslog
> behaves correctly with -vv
> OTOH I don't see anymore useful information compared with -v, as I sent
> earlier.
> 
> FWIW my syslog is configed at mail.debug... must be a bug in my syslog
> daemon (IHTA my OS does suffer from some bitrot.)

Basically the problem appears to be that the IDLE-ing connection gets
somehow closed at socket level prematurely.

How exactly needs to be established on a lower network layer, possibly
with tcpdump, wireshark, or similar, and see if the connection gets
reset, by whom, and when, or if there is a regular FIN/ACK sequence (in
that case, Exchange would be the suspect).

Possible reasons include:

- some firewall on either end expiring stateful rules associated to the
idling connection

- some masquerade/NAT device (possibly router, usually on the client
side) expiring forwarding rules prematurely

- the Exchange server closing the connection after 5 minutes of idle
time when it should allow 30 minutes.  I've never operated an Exchange
server and know nothing about configuration options.


If you can't find the reason, you can locally modify the imap.c source,
look for the figure 1680 (28 * 60 seconds), lower that to 270 (4.5 * 60
seconds), recompile and reinstall; however, I am not considering such a
change upstream.

Hope that helps
Matthias