chicken egg problem

classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

chicken egg problem

richard lucassen
Hello list,

According to "man dhcpd.conf" i can use *hostnames* as DNS:

option domain-name-servers ns1.example.com, ns2.example.com;

That's very nice, but can someone explain how the ns1.example.com and
ns2.example.com can be resolved by a new client?

What I initially was looking for is a way to randomly assign multiple
DNS servers to clients. With a simple:

option domain-name-servers 10.0.0.1, 10.0.0.2;

the 10.0.0.1 gets all the queries from all clients. Of course I can
assign the nameservers the other way round for some ranges:

option domain-name-servers 10.0.0.2, 10.0.0.1;

But I was wondering if it would be feasible to configure such a random
assignment of domain name servers. Of course it is not a show stopper
if it is not feasible, one DNS can handle all queries smoothly.

R.

--
richard lucassen
http://contact.xaq.nl/
_______________________________________________
dhcp-users mailing list
[hidden email]
https://lists.isc.org/mailman/listinfo/dhcp-users
Reply | Threaded
Open this post in threaded view
|

Re: chicken egg problem

Simon Hobson
richard lucassen <[hidden email]> wrote:

> According to "man dhcpd.conf" i can use *hostnames* as DNS:
>
> option domain-name-servers ns1.example.com, ns2.example.com;
>
> That's very nice, but can someone explain how the ns1.example.com and
> ns2.example.com can be resolved by a new client?

They can't, and they aren't. They are resolved by the *server* when it starts and the resulting IPs are given to the clients.

> What I initially was looking for is a way to randomly assign multiple
> DNS servers to clients. With a simple:
>
> option domain-name-servers 10.0.0.1, 10.0.0.2;
>
> the 10.0.0.1 gets all the queries from all clients. Of course I can
> assign the nameservers the other way round for some ranges:
>
> option domain-name-servers 10.0.0.2, 10.0.0.1;

You'll find it's more subtle than that !
Assuming a preponderance of Windows clients, try assigning "10.0.0.1, 10.0.0.2" and then take 10.0.0.1 offline for a while. When you turn it back on, you'll probably find that few clients will use it again - until 10.0.0.2 goes offline.

I know of people who've empirically found that if they specify a local server before external resolvers, they can get a form of split DNS - where the Windows clients check with the local server first and then go to an external resolver if the first one doesn't have the answer. They then found that this breaks if the internal server ever goes offline - because then the clients put it to the end of the list "permanently" and it only gets promoted if others above it fail.

IIRC there's also reports of Windows machines not changing settings (I suspect DNS servers was included) when the values change in the offers. Only if they lose their connection (lease expired, moved network, old lease no longer valid and NACKed) will they get a new lease and the new settings.

_______________________________________________
dhcp-users mailing list
[hidden email]
https://lists.isc.org/mailman/listinfo/dhcp-users
Reply | Threaded
Open this post in threaded view
|

Re: chicken egg problem

richard lucassen
In reply to this post by richard lucassen
On Thu, 2 Jun 2016 12:04:40 +0200
richard lucassen <[hidden email]> wrote:

> According to "man dhcpd.conf" i can use *hostnames* as DNS:
>
> option domain-name-servers ns1.example.com, ns2.example.com;
>
> That's very nice, but can someone explain how the ns1.example.com and
> ns2.example.com can be resolved by a new client?

Ok, I see what happens. It's the dhcp server that resolves the
hostname when it is started. Although the DNS is round robining the
answers, the dhcp server assigns the DNS servers as found when started.

--
richard lucassen
http://contact.xaq.nl/
_______________________________________________
dhcp-users mailing list
[hidden email]
https://lists.isc.org/mailman/listinfo/dhcp-users
Reply | Threaded
Open this post in threaded view
|

Re: chicken egg problem

richard lucassen
In reply to this post by Simon Hobson
On Thu, 2 Jun 2016 12:04:42 +0100
Simon Hobson <[hidden email]> wrote:

[random assignment of DNS servers]

> You'll find it's more subtle than that !
> Assuming a preponderance of Windows clients, try assigning "10.0.0.1,
> 10.0.0.2" and then take 10.0.0.1 offline for a while. When you turn
> it back on, you'll probably find that few clients will use it again -
> until 10.0.0.2 goes offline.
>
> I know of people who've empirically found that if they specify a
> local server before external resolvers, they can get a form of split
> DNS - where the Windows clients check with the local server first and
> then go to an external resolver if the first one doesn't have the
> answer. They then found that this breaks if the internal server ever
> goes offline - because then the clients put it to the end of the list
> "permanently" and it only gets promoted if others above it fail.

Sounds like a particular interpretation of an RFC. I've seen that
before :)

> IIRC there's also reports of Windows machines not changing settings
> (I suspect DNS servers was included) when the values change in the
> offers. Only if they lose their connection (lease expired, moved
> network, old lease no longer valid and NACKed) will they get a new
> lease and the new settings.

I think I'll stick to the simple idea to inverse DNS settings to some
ranges. Thnx for your time anyway!

R.

--
richard lucassen
http://contact.xaq.nl/
_______________________________________________
dhcp-users mailing list
[hidden email]
https://lists.isc.org/mailman/listinfo/dhcp-users