BIND failure methods (NS)

[ddd]

For one I may think that the querying server really only needs one NS glue
record in order to find that given domain, then it queries that domain srever
for whatever records it needs. If it can't contact one NS/A given by parent,
then it could contact parent again for the next one...?  That's good though.
So what the parent answers with is _all_ NS records and associated glue
records it has for that queried domain ?
The Querying party then tries to contact each, and all nameservers will at
least fail over to the next one in the list, and/or use any available with the
least RTT ?
So it seems that the good news is that if we have 3 NS machines listed at our
parent, and we want to change the address of one of them, the other 2 will
answer all queries ?
What about if a server has ALREADY received 3 NS & A records from our parent,
and those TTL's are 24hrs, if all 3 become unavailable during the 24 hrs, will
the Querying party re-query the parent BEFORE the TTL has expired ?


LaMont Jones wrote:

> mikey78 at my-deja.com wrote:
> > If we have 3 NS records (and associated glue A records) at our parent,
> > and two of them become unavailable, will resolvers use the one working ?
>
> Other nameservers will tend to use the one with the best round trip time
> (RTT).  So, in general, the two down IP's will only be queried every so
> often, and the one that's up will get the bulk of the traffic.
>
> > Let's say we have ns1.ourdomain.com, ns2, ns3. Addresses are .1, .2, &
> > .3 let's say. So if the BIND server at cnn.com needs to find addresses
> > (on the behalf of a client) in ourdomain.com, will it get all 3 NS
> > records for ns1,ns2,ns3 from our parent at .com or only 1 ?
>
> The nameserver giving out the referral doesn't care if they're reachable
> or not, that's the referred application's problem...
>
> lamont