On 21 January 2010 19:22, Alastair Sherringham <sherringham@???> wrote:
> Looks like this has been a problem in the past. Digging briefly
> (first/second hit) pinted me to a Debian bug + patch :
> Bug#517808: dovecot-common: please call ntp-wait before starting Dovecot
> http://www.mail-archive.com/debian-bugs-closed@lists.debian.org/msg218224.html
> Fixed in dovecot-common 1.1.11-3 (Squeeze/Backports). A patch to the init
> script to do an ntp-wait.
Sorry for following up so late. More information ...
To re-cap, I have had problems where the clock on my VP is 30 mins in
the future on boot. This behaviour started after a dist-upgrade to
Debian Lenny a few weeks ago. Correcting this (ntpdate,ntpd) causes
dovecot to die, and since Dovecot is what does AUTH for Postfix, both
Dovecot and Postfix are useless without a manual restart.
Linux calliope 2.6.26-2-xen-686 #1 SMP Wed Feb 10 11:28:19 UTC 2010
i686 GNU/Linux
This still happens and I have been testing all morning - rebooting the
VPS numerous times ...
ntpd wasn't helping
ntpdate wasn't helping
What now happens is that *sometimes* there is no clock skew. Sometimes
there is. Also - disabling ntpd (update-rc.d) and reinstalling ntpdate
has changed behaviour.
I now have ntpdate installed again - there is an
/etc/cron.daily/ntpdate containing :
ntpdate -s ntp0.sfo.bitfolk.com ntp0.lon.bitfolk.com
0.debian.pool.ntp.org 1.pool.ntp.org
There are no other crontab ntp calls (or @reboot jobs).
From boot, sometimes my clock is good, sometimes in the future.
However, now, ntpdate *fixes* this before Dovecot dies, so I am
happier.
ntpdate[1036]: step time server 194.29.130.252 offset -2089.601967 sec
It is always ~30 mins.
I sit and reboot - login and check - reboot ...
OK no skew - dove OK
BAD - skew FIXED - dove OK
BAD - skew FIXED - dove OK
OK no skew - dove OK
OK no skew - dove OK (fsck ..)
BAD - skew FIXED - dove OK
OK no skew - dove OK
BAD - skew FIXED - dove OK
Odd?
Cheers,
--
Alastair Sherringham
