problem with DNS resolution on HP-UX 11i

[Randall L. Kline]

I have a set of new servers (4) that are running 11i. I have three nameservers defined in resolv.conf, two of which are local system running localized DNS/named. The third entry is the corporate DNS server, which is accessed for anything outside of the local servers. We have 4 other systems running HP-UX 11.0, with identical configuration files resolv.conf, nsswitch.conf, and they work as expected. The new servers will only reference the first nameserver in resolv.conf, and then go to NIS, and files. The additional nameservers are never being referenced.

The line for hosts from nsswitch.conf is as follows:

hosts: dns [NOTFOUND=continue] nis [NOTFOUND=continue] files

Is there something different in 11i that causes this malfunction? If I put the corporate nameserver first in the list in resolv.conf, the nslookups work, but the local lookups fail. Any insight would be appreciated. I've looked at the other forum discussions, but they aren't helping.

[Sanjay_6]

Hi Randall,

I've not see this situation in 11.11.

Take a look at this thread from itrc. It may be of some help,

http://www1.itrc.hp.com/service/cki/docDisplay.do?docLocale=en_US&docId=200000069154812

The itrc doc id is UARPAKBRC00011629.

Hope this helps.

Regds

[rick jones]

If I am reading between the lines correctly, you expect the different servers in resolv.conf to be accessed if the preceeding ones do not have an answer . Your DNS is split between local and remote.

I do not recall (but could be wrong) the resolv.conf file having that semantic. My recollection is that the multiple servers in resolv.conf is only meant to deal with server reachability - the servers are supposed to be otherwise equal in tems of what they can answer.

If that is correct, you were simply lucky with the 11.0 systems. If I am incorrect then this has all just been a bunch of hot air :) Are you sure that on your 11.0 systems there aren't local names in the /etc/hosts file?

I supsect you need to configure your local servers to forward queries to the corporate, or return a redirect.

Still, it might be good to get some verbose, snaplen 536 byte tcpdump traces of both an 11.0 and an 11i (which 11i?) system doing both a local and a non-local resolution.

[Patrick Wallek · Accepted solution]

My understanding is the same as Rick's.

The /etc/reslov.conf is just for DNS failover (for lack of a better term). If the first is not accessable then it moves to the next server in the list.

Your DNS servers themselves should be set up such that if you query one for an address and that server doesn't know about it, then it forwards the request on to another server. That is how BIND/DNS was designed and how all DNS servers on the Internet function.

[RAC_1]

Are the nsswitch.conf file same on the server on which it works and on which it does not work.
My understanding is that if first dns server fails/unavailabe it goes to second and so fourth. The third dns server in nsswitch.conf should be able to resolve the adresses which it is supposed to do.

Anil

[Randall L. Kline]

Actually, after looking at the existing systems that do work, I found that there was a forwarders entry in /etc/named.conf. I duplicated these entries, but it still isn't working. This is a different version of BIND, so maybe the syntax for forwarders is different?

[rick jones]

ok, now I'm a bit (more :) confused - are you trying to run caching-only nameservers on the new servers or are you expecting them to strictly act as dns "clients?"

if you are running a caching only name server on the new/old servers then i would think the resolv.conf file on those systems would point to 127.0.0.1 first.

and if you are trying to run a named on those system, you might check the logs on the "actual" name servers and see if they are denying access based on IP or somesuch etc

[Randall L. Kline]

Let me try to clear this up. In the existing cluster running 11.0, there is a three-node SG cluster, which is running an application that has it's own domain defined, and is running two server with named configured, for redundancy. The problem is that with this configuration, there is no support for connectivity to devices outside of this cluster. We need the external resolution for standard reasons. Because we have two named instances, we have those two servers in the resolv.conf. There is no problem with access on the corporate DNS servers, because I can put those server's addresses in the top of the list in resolv.conf, and resolution is successful. Basically, we have two nameserver entries for the two cluster servers running named, as well as forwarders entries in named.conf pointing at the corporate DNS servers.

Clear as mud?

[rick jones]

ok, i _might_ be seeing through the mud :)

basically, leave the /etc/resolv.conf file pointing at the named's in the cluster, and take-out the corporate ones. i believe you are on the right track editing the named.conf on the two named's to "teach" them about the corporate DNS, but I must confess that I'm useless wrt the syntax you need to use - I only know enough to get a benchmark going :)

either someone else here will need to add that, or you can get one of the DNS books by Cricket Liu (former HP), or go to www.isc.org and look at the docs there or perhaps netnews for comp.protocols.dns.named (iirc) some of the True Experts (tm) in DNS hang-out there.

[Randall L. Kline]

I found the problem. The person that entered the forwarders entries didn't reinitialize named, so those entries weren't being used. Once I HUPed named on the local nameservers, resolution worked. Thanks to all of you for your guidance. Points have been assigned.