TAI and qmail

Jan Pisacka <[email protected]> Thu, 11 Jul 2002 17:32:41 +0200 (CEST)
Newsgroups gmane.comp.lib.libtai
Message-ID <[email protected]>
Hi,

I have a minor problem with Qmail installed on a Linux system, when it is
at the same time set up to use TAI. I have 2 similar machines, which
differ only that machine_1 uses POSIX:

cp /usr/share/zoneinfo/Europe/Prague /etc/localtime

while the machine_2 uses TAI:

cp /usr/share/zoneinfo/right/Europe/Prague /etc/localtime

Machine_1's clock is adjusted by ntpdate every 2 hours. It is never 
stepped: 
/usr/sbin/ntpdate -s -B -t 2 -p 8 -u `cat /etc/ntp/step-tickers` 
I guess that in this configuration time never skips back.  

Machine_2's clock is being taken care of by the clockspeed, set-up exactly 
according to the DJB's instructions.

A simple check -- "xclock -update 1" run on both machines and displayed
concurrently on anyone of them -- suggests that the time of the two is the
same.

Both machines use MTA qmail, configured to log through multilog, so that
all timestamps in the logs are in tai64n and can be converted to
human-readable form through tai64nlocal.

On machine_1 (POSIX), when I send a message to a user on the same machine,
there are no discrepancies between the timestamp in the message's header,
the timestamp in the log, and the ctime of the message file created in the
Maildir.

On the other hand, on machine_2 (TAI), the timestamp in the "Received: 
header is famous 22 seconds ahead of all the other timestamps.

If I send a message from machine_1 to machine_2 (and vice versa), all
timestamps in the logs on both machines agree to each other and to the
ctime of the message file in the recipient's Maildir. The only exception
is the "Received:" header recorded by the machine_2, which is always 22
seconds ahead.

Does anybody know where the error may come from?
Thanks,

Jan