named not listening at all - was Re: named is listening on wrong device

[D. Stussy]

Well, if the interface scan is only about once per hour, maybe we should have
that changed if "named" finds no interfaces to once per minute (or so).....

One of the other reasons I have the daemons load before bringing the interfaces
up is simply because I want all services to [already] be available when the
machine becomes reachable.  This order of operations implies this result.

On 12 Sep 1999, Jim Reid wrote:
> >>>>> ">" == D Stussy <kd6lvw at bde-arc.ampr.org> writes:
>     >> I like to start all my servers first (just to give them low
>     >> process numbers - look's neat).  Then I define the interfaces
>     >> (via "ipconfig").  "GateD" has no problem finding that I have
>     >> brought up an interface (it's found within a few seconds of
>     >> being up), but "NameD" just sits there until I tell it via an
>     >> "ndc reload".  The syslog initially reports something like
>     >> "name server started ...  but no interfaces."  OK, no problem,
>     >> but why does it then take a reload for it to go and look again?
> 
> Telling named to reload means it re-reads its config files, reload any
> zone files, refresh any slave zones and scan for network interfaces
> that have come up or gone down since the last time it scanned them.
> Think of it as a global reset and reconfiguration of the name server.
> So if you do that after the network interfaces have come alive....
> 
>     >> Should I force NameD to listen on [0.0.0.0].53 (i.e. wildcard
>     >> the IP address) instead of the default of a socket on each
>     >> interface's IP address, at least when starting it?  ...
> 
> I don't think that'll work. If there are no interfaces up when named
> does an ioctl(.. SIOCGIFCONF ..), then there's nothing available to
> set up a port 53 listener apart from the loopback interface. In
> networking terms, that's not very interesting because only things on
> the local system can use that interface. In fact I think [0.0.0.0]
> will be treated by named as a synonym for the loopback interface -
> "this host". I've not bothered to check this, so I could be wrong.
> 
>     >> Or just forget about it and either change the order so the interfaces
>     >> are before the daemons or have a nameserver reload in my system
>     >> initialization scripts (to re-read the kernel interface list)?
> 
> Well I'd just forget about it. Who cares what PID a process gets and
> why should it matter? It also seems very strange to want to start
> network services like gated and named before any network interfaces
> are switched on.
> 
> BTW, the interface-interval clause in BIND8's options{} statement
> allows you to tell named how often it should scan for network
> interfaces. The default for this interval is 1 hour which is "good
> enough" for most people. You could configure named to scan more
> frequently, though I don't see the point of this unless network
> interfaces get enabled and disabled very frequently on your
> system. [Even with a dialup link, the system's network interfaces
> should not be changing *that* often.] IIRC gated checks the status of
> network interfaces much more often because that is a reasonable thing
> for a routing daemon to do. If an interface suddenly comes up, the
> routing table probably needs to change and a bunch of RIP or OSPF
> updates have to be sent. For a name server, the state of any network
> interfaces should hopefully be fairly stable.
> 
> 
> 
>