Ed W <li...@wildgooses.com> wrote on 10/31/2012 09:03:14:
> 
> On 31/10/2012 10:36, Tomalak Geret'kal wrote:
> > On 31/10/2012 10:35, Miroslav Lichvar wrote:
> >> On Wed, Oct 31, 2012 at 10:14:01AM +0000, Tomalak Geret'kal wrote:
> >>> Again, chrony doesn't need the TTL. Caching is handled by the
> >>> resolver.
> >>> getaddrinfo() blocking is a more concrete problem to solve - good
> >>> spot.
> >> I don't think getaddrinfo() does any caching, that has to be done in
> >> nscd or a local DNS server.
> >>
> > I didn't claim otherwise!
> >
> > Tom
> 
> Then you need to be very careful that millions of routers out there 
> don't suddenly start issuing DNS requests every few seconds because 
> someone upgraded a firmware to include this new version of chrony and 
> didn't notice the change in behaviour (not all chinese/taiwanese router 
> board builders will read the english release notes...)

Please don't let me get blamed for starting that avalanche!  :-)

> Remember unlike windows its much more normal to not have any dns caching 

> on linux and rely on the nearest upstream (hopefully inside your 
> infrastructure)

My distros (Fedora and previously RHL) have done local caching for as long 
as I can remember.  I'd be surprised to see one that doesn't actually.

> I see that this is a very dangerous change.

I'm inclined to agree with this now.  I posted the Q because chrony's 
behavior seemed at odds with most network services, but I can clearly see 
now there are some very good reasons for that.  In our case, we've got our 
own internal radio-clock server so we're not using pool.ntp.org, but I 
certainly realize many installs our going to be pointing to some 
round-robin DNS setup.

IMHO I think the best strategy forward would be just leave things as they 
are, but perhaps add some warnings to the documentation and perhaps even 
run-time logs that the name resolution is a one-time deal.  Run-time 
warnings could even point the user to a specific part of the docs 
explaining what was learned in this discussion.

In any case, I very much appreciate everyone's effort to make it better 
and/or explain the devil in the details.  Thank you all.
--
John Florian

Reply via email to