[pkg-ntp-maintainers] Bug#484974: Bug#484974: ntpdate's ip-up script runs even for virtual interfaces

[Josip Rodin]

On Mon, Jun 16, 2008 at 12:10:59AM +0200, Peter Eisentraut wrote:
> Josip Rodin wrote:
> > /etc/network/if-up.d/ntpdate runs even if you bring up a virtual interface
> > like eth0:5; I don't see any reason why it should ever do that.
> 
> Unless you can provide a concrete reason why the current behavior is really 
> harmful, I am hesitant to add more special rules based on singular demand.  

If the purpose of having if-up.d/ntpdate is to synchronize time with a new
network interface, which is usually a hardware interface, then it makes no
sense to run it when a virtual interface is brought up. Technically one
could argue that a particular new virtual interface could be brought up that
has e.g. specific connectivity or specific firewall setup that allows
ntpdate where a default hardware interface doesn't - but who does that? :)

> Everyone's network is different, so we should keep the default simple.  
> Usually, if we make changes like this, someone else will soon not like it and 
> request the opposite behavior.

Well, I agree that the default should be simple. By that measure,
the current default of running ntpdate all the time, is just wrong.
Why should changing the system time automatically be done while
running ifup eth0:5?

If we went through the trouble of removing the init script for ntpdate
(and that's a conventional way of automatically running things),
why force it upon the ifupdown system?

Which brings up the question... where is the rationale for doing that,
anyway? I see in the changelog:

  * npdate is no longer started from an init script but instead by ifup
    (closes: #56499, #245338, #312576)

The first bug has Tomas Pospisek asking if you could add an optional
/etc/ppp/ip-up.d setup for ntpdate in the package. Its duplicate has
Francesco Potorti` suggesting that his /etc/ppp/ip-up.d/zz90-ntpdate would
be nice if it came standard with ntpdate. In the other duplicate, Rahul Jain
suggests that an /etc/ppp/ip-up.d/ntpdate script would be good for desktops.

(The other two bugs relate to the old init.d script.)

So, where is the demand for ntpdate to be run on eth* interfaces? All I see
is people asking for it with PPP connections (typically, dialups and/or
desktops). Why aren't we doing this for ppp interfaces only?

> Also, in the iproute world, one doesn't use virtual interfaces but instead 
> adds multiple IP addresses to one interface.  I am not sure how that should 
> be handled under your theory.

That's just another reason why ntpdate shouldn't be excessively intertwined
with ifupdown - it just can't foresee such consequences...

-- 
     2. That which causes joy or happiness.