topic Re: Name Server Timeout in Operating System - HP-UX

Name Server Timeout

Magic_Wand — Wed, 12 Mar 2003 20:12:58 GMT

Folks,

I'm faced with a dillema:

We host a series of L-Class, 11.00 HP-UX boxes. Meanwhile, we have two Linux DNS servers which also serve about 200 other servers.

When sending mail, sometimes, I rec'v a NAME SERVER TIMEOUT, but it's not all the time, and it is with different domain names.

My question is that while it looks like bind496 was the default on 11.00, is there a newer bind client? Also, can this be attributed to a different problem other than the bind client?

I should mention that we are running on a GigE network and it's not saturated, so speed is not an issue.

Here's the message I get when I try to force sendmail:

# sendmail -q -v

Running /var/spool/mqueue/h2BMi2Y03372 (sequence 1 of 1)
blahblah.com: Name server timeout
... Transient parse error -- message queued for future delivery

But then as soon as I do the nslookup on blahblah.com, and another sendmail -q -v, then it goes through without a problem.

Thanks in advance.

Re: Name Server Timeout

John Bolene — Wed, 12 Mar 2003 20:30:32 GMT

Not sure where to point you to look either.

I run 2 C110 DNS 11.0 machines and they can resolve at least 100 requests per second over a 100 Meg fast ethernet network with only about 15% cpu utilization.

The only times we have any DNS timeouts is when the primary machine is down for maintenance or patches, which is about once a year.

All the 300 clients are 11.0 or 11i machines which have DNS caching setup on them to support the 500 or so terminals connected to each of them.

Re: Name Server Timeout

Michael Tully — Wed, 12 Mar 2003 20:53:04 GMT

One way that may help you would be to utilise a local hosts copy and DNS as a secondary.
Have your nsswitch.conf file look first in /etc/hosts and then DNS. We have a small Linux box that does DNS, that is looked after by our network div. We update our UX hosts files once a week.

Re: Name Server Timeout

Magic_Wand — Wed, 12 Mar 2003 20:54:30 GMT

The /etc/hosts file won't work, because these are outside domain names, and I have no way of knowing which domain names are going be sent mail to.

Re: Name Server Timeout

Ron Kinner — Wed, 12 Mar 2003 21:38:47 GMT

Speed could still be an issue. When the local DNS server does not know the IP then he has to query the next one up the line which is out on the internet. You could have some problems up that way with lost packets or congestion or slow machines. If the next upstream DNS doesn't know then it goes further.

you might also run
nslookup
set type=ns
.

and see if you have some bad rootservers.

Ron

Re: Name Server Timeout

Mark Landin — Wed, 12 Mar 2003 22:05:55 GMT

Remember that sendmail does not use nsswitch.conf like nslookup does. Thus, you may be able to resolve with nslookup while sendmail cannot.

That said ... are the domain names that are failing to resolve something that you would expect your DNS servers to have in cache? (i.e., are they domains that you send email to often?) It just may be that your upstream DNS servers are the one having the problem. Perhaps it's taking 17 seconds for them to reply to your Linux box, but sendmail is only waiting 15 seconds. By the time you notice and check it yourself, your Linux box has received the answer.

Re: Name Server Timeout

Magic_Wand — Wed, 12 Mar 2003 22:17:57 GMT

Unfortunately, the DNS lookup is not a problem, my DNS servers are responding in less than 1 second for names that I *know* are not cached.

One thing I'd like to know is that how can I tell sendmail NOT to lookup names, but send it DIRECTLY to the mailhub.

Re: Name Server Timeout

Chris Vail — Wed, 12 Mar 2003 22:32:46 GMT

To send mail directly to your mail hub, put an entry in the /etc/hosts with the name "mail", and change your nsswitch.conf so that it looks at files before it goes to DNS.

Chris

Re: Name Server Timeout

Mark Landin — Thu, 13 Mar 2003 13:45:12 GMT

Sending to a mailhub is easy. If your mailhub is named foo, do this:

1. Modify /etc/rc.config.d/mailservs as follows so that the following line appears:

export SENDMAIL_SERVER_NAME=foo

If your system does not RECEIVE mail, only sends mail to the hub, then you can disable the sendmail server process by changing the line

export SENDMAIL_SERVER=1

to

export SENDMAIL_SERVER=

2. Now you need to create a "nullclient" sendmail.cf file. This can be done using the pre-built .m4 macros HP provides in /usr/newconfig/etc/mail. Now, everytime I've had to do this I've had to relearn how to so the m4 stuff. If you have the O'Reilly sendmail books then they can provide examples which are almost correct. Unfortunately, I can't remember exactly how to do it, but the files you need are in that directory. Just create the sendmail.cf file using the nullclient.m4 macro. Set a few variables like DS, DM, and DR, all pointing to host foo, and you should be alright.