Subject: Re: rc.d: time synchronization issues at boot time
To: None <tech-userlevel@NetBSD.org>
From: Alan Barrett <apb@cequrux.com>
List: tech-userlevel
Date: 03/16/2005 19:30:04
On Wed, 16 Mar 2005, der Mouse wrote:
> I don't think that just retarting named after starting ntpd is enough.
> You really don't want to kick named until ntpd has synced up and done
> any initial time step it's going to.

Yes, but we were talking about ntpdate, not ntpd. ntpdate does its
initial time step "immediately", and it's unusual for ntpd to want to do
another time step if you have run ntpdate before starting ntpd.  Even if
ntpd does perform another step soon, it will be a small step (probably
less than 1 second), which will not hurt named.

> Alternatively, I wonder if we should provide a time-since-boot
> interface

There is something a lot like that, but named doesn't use it.  Also,
the code in named that expires records due to absolute timeouts (e.g.
signatures reaching their end of life) and the code that expires records
due to relative timeouts (e.g. TTL counting down) both rely on the same
lower level functions, so separating the two notions of time would not
be trivial.

--apb (Alan Barrett)