[BUGS] ntpdate/ntpd at startup (was: Dovecot not starting on startup)

John Marshall john.marshall at riverwillow.com.au
Thu Apr 24 21:14:27 EST 2008


On Thu, 24 Apr 2008, 10:23 +1000, Andrew Reilly wrote:
> On Thu, 24 Apr 2008 10:13:41 +1000
> Jerahmy Pocott <quakenet1 at optusnet.com.au> wrote:
> 
> > I'm just going to revert to using ntpdate instead of the -g flag to  
> > ntpd like I always used to..
> > 
> > If ntpdate is really going to disappear we need something better than  
> > the current state of things..
> 
> When I read about ntpdate -b going away, I read up on ntpd, and
> have just switched over to using that as intended: a PLL that
> keeps the clock in synch.  Any particular reason not to go that
> way too?

I think we all use ntpd. The issue here is not how to keep the clock
synchronized, it's how to set the clock in the first place - at startup time
- before the server applications start (which is what ntpdate -b does) - so
that software doesn't see the time go backwards a few minutes later when
ntpd has acquired sufficient confidence to set the clock.

It was noted in an earlier post that the server configuration "iburst"
option can reduce this initial synch time from several minutes to about 10
seconds. At present, that looks like being the closest we can get to what
ntpdate does.

> My only problem with ntpd is that it gives up too easily.  I run
> djbdns/tinydns under supervise for name service, and supervise
> starts more or less last.  After ntpd, anyway.  So ntpd starts
> up, can't find any of the pool servers, and just sits there,
> doing nothing, forever.  I have to remember to /etc/rc.d/ntpd
> restart.  Hmm.  I should just put that in rc.local...  Yes,
> there are clearly lots of ways to "fix" this, but having
> ntpd try again every now and then would be a big win, in my
> opinion...

-- 
John Marshall


More information about the BUGS mailing list