shuttinng down DNS server creates problem

[Rahul Kumar Singh_1]

Hi everybody,
i have a 3 node cluster of HP rp3440 servers. one is database server, one is application server and the third one is failover server. the failover server is configured as DNS server also. if i stop the DNS service on failover server then there is no problem to do Telnet and FTP. But if i shutdown the failover server then or make down the primary lan card of this server then i am not able to do telnet or ftp with other servers.
is this problem related to DNS? can anybody tell me the solution.

regards
Rahul

[Steven E. Protter]

Your setup is the problem.

If you are relying on DNS name service for telnet and ftp, shutting down the DNS server is the cause.

example:

ftp failoverserver

# this fails

ftp 192.168.0.10 # ip address

# this works.

That would mean your DNS serer being down is the problem.

DNS stands for domain name service

So try it out with the ip address. If that works, you have isolated the problem.

Realistically, you have two choices to resolve this:

1) Make DNS a failover service thats part of your cluster
2) Take the DNS service outside the cluster so that when you shutdown the failover node you are not disabling your cluster.

SEP

[Mobeen_1]

Rahul,
Two ways really

1. Run your DNS outside the cluster as per the previous suggestion

or

2. Don't rely on DNS for name resolution. You could either have the host names and IP addresses in the hosts file or use IP

rgds
Mobeen

[Mel Burslan]

Rahul,

Serviceguard configuration should not depend on DNS at any time. All nodes in a cluster should know each others' simple name, fully qualified domain name and IP address by means of /etc/hosts file.

By using MCSG, you are trying to eliminate single point of failures, not adding more of them. And DNS is a SPOF if you rely on it as the timing of the response back from the DNS server can not be guaranteed depending on load of this server at any given time.

It is not advisable to run a whole company (or even god forbid world accesible external DNS) server running on a cluster member node function of which has nothing to do with the cluster itself. But even if you insist running DNS on this server, release yourself from dependency on this server by utilizing hosts file at the earliest possible moment. Make sure your /etc/nsswitch.conf file is configured to look up the hosts file firs, not the DNS.

Hope this helps

[Rahul Kumar Singh_1]

Hi everybody,
actually i am stoping the DNS service/shutting down the failover server after taking it out of the cluster. then why i am not able to telnet or ftp with other servers. please provide the suggestion.

regards
Rahul

[Mobeen_1]

Rahul,
You are mentioning that you are shutting down the failover server that is running DNS. But you don't mention anything about any other DNS servers running on other nodes?

If this is the only DNS server that you are running, then name resolution would not be possible, unless its up and running.

If this is not the only DNS server, then make sure that for all those servers which you want, there is a secondary DNS server IP address included.

regards
Mobeen

[Peter Godron]

Rahul,
I agree with Mobeen, make sure you have both DNS servers defined on all the machines and also ensure that DNS resolution is:
Primary server first
if no anser then use scondary server
if no answer then use loacl files
if no answer error

Regards

[Bill Hassell]

DNS is only critical to your operation if your name resolution depends on it. In a high-reliability system, you DON'T want to depend on it for most of the hostname/IP requirements. When you use telnet or ftp, the server attempts to verify the source by taking the incoming IP address and asking for a hostname (aka, a reverse DNS lookup). The proper way to setup such servers is to change the /etc/nsswitch.conf file to use files (/etc/hosts) first, and then dns. Change the line to read:

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

Now put the IP address and hostname for all systems that will connect to your servers and loss of DNS will not affect them. Note also that any additional DNS servers listed in /etc/resolv.conf will require 20-100 additional seconds for each request when the DNS services are not running on the network.