topic Re: MCSG and NTP in Operating System - HP-UX

MCSG and NTP

derek b smith_1 — Fri, 03 Feb 2006 12:39:36 GMT

I have an floating IP between an rp8400 and an n4000 97.6 and a system IP on the rp8400 of 97.210.
97.210 is also my configured IP that NTP uses as this is our master NTP server that talks to Ohio State U., MIT and GATECH.
If the cluster and package were to fail over to the N4000 then I would want the NTP stuff to move over as well. I know how to set this up as a cluster service, but what I am asking is how do I tell HPUX to use 97.6 as the NTP address to server my LAN clients instead of the system IP address?

thanks
derek

Re: MCSG and NTP

James R. Ferguson — Fri, 03 Feb 2006 12:47:17 GMT

Hi Derek:

You should run 'xntpd' on all nodes of the cluster. Then, regardless of where the application runs, time will be correct.

Regards!

...JRF...

Re: MCSG and NTP

Geoff Wild — Fri, 03 Feb 2006 12:52:52 GMT

I'm with James - and that's what we do as well - every node in all our clusters talk to NTP servers directly - that way all nodes are insync (sheesh - isn't that a band?)

Rgds...Geoff

Re: MCSG and NTP

Rick Garland — Fri, 03 Feb 2006 12:55:13 GMT

Run NTP on each indovidual node within the cluster.

No need to worry about what the cluster is doing then, the time will always be correct.

Re: MCSG and NTP

derek b smith_1 — Fri, 03 Feb 2006 12:55:45 GMT

Hello....yes we do run xntpd on all nodes in the 2 node cluster, but if the primary node fails at 97.210 then the 97.6 moves over to the secondary and the secondary system IP is not 97.210 rather it is 97.106. All my local LAN clients are pointing to 97.210 to get their time synced. Hope you see the delemma here?
It sounds like a MCSG environment is not a good mix with an NTP server???

Re: MCSG and NTP

Geoff Wild — Fri, 03 Feb 2006 13:05:51 GMT

Okay - I get it now - NTP server in MC/SG - you just need all your clients pointing at the floating ip of the NTP package - not the system ip.

Rgds...Geoff

Re: MCSG and NTP

Patrick Wallek — Fri, 03 Feb 2006 13:06:56 GMT

Just point all of your NTP clients to the 97.6 address rather than the 97.210 adress.

Re: MCSG and NTP

James R. Ferguson — Fri, 03 Feb 2006 13:07:57 GMT

Hi (again) Derek:

You could configure each node to be a broadcaster to you LAN.

http://docs.hp.com/en/B2355-90685/ch07.html

Regards!

...JRF...

Re: MCSG and NTP

derek b smith_1 — Fri, 03 Feb 2006 13:11:12 GMT

I have tried pointing some of my NTP clients to 97.6 (primary cluster server with system IP of 97.210) but ntp hangs says "trying again.....trying again...."

NTP is not configured to listen at 97.6 (floater IP) and this is what I need to do. : )

Re: MCSG and NTP

rick jones — Fri, 03 Feb 2006 21:05:21 GMT

when xntpd is running on your system, do a netstat -an | grep and see if it is binding to the wildcard IP ("*") or if it is binding to specific IPs. I would guess the former, but it is possible that it binds to the list of active IPs at the time xntpd is started, and if the package IP is not active at that time...

now, if you've packaged xntpd such that it is started whenever serviceguard does its thing rather than when the system does its boot thing (not sure if i'm ratholing there or not) then presumably the IP would be assigned by then, but who knows.

however, as already pointed-out, i'm pretty sure it isn't a case of telling the xntpd on which IP's it should serve time, but telling the clients from which IP's they should request time.

finally, since NTP clients have had "built-in" HA gong back to before the time when ServiceGuard was a gleam in HP's eye, you don't really need to make NTP a package at all. you simply need to configure your clients with the primary IP's of both your rp8400 and n4000, and preferably at least one other source of time, and let them deal with the rest. less muss, less fuss

same thing with DNS service

getting back to your clients reporting "trying trying" - can they ping your 97.6 address?

Re: MCSG and NTP

paolo barila — Sat, 04 Feb 2006 14:57:29 GMT

Hi,
once xntpd is up and running on both cluster nodes, I would check my ntp clients this way:

-stop the daemon for a while if it's running:
# /sbin/init.d/xntpd stop

-verify connection with ntp server:

# ntpdate -d ...97.210

# ntpdate -d ...97.106

if this doesn't work then your clients don't reach ntp server on port 123, (did you open it only for the floating IP?)

Then configure

#vi /etc/ntp.conf

server ...97.210
server ...97.106

-initial clock alignment:
# ntpdate ip_ntpserver

-start daemon
# /sbin/init.d/xntpd start

-check
# ntpq -p

Hope this helps

Pablo

Re: MCSG and NTP

Devender Khatana — Sat, 04 Feb 2006 15:38:11 GMT

Hi,

It is also possible to configure two NTP servers on all the clients one as primary and other as secondry. The best is to configure floating IP as NTP source allthough.

HTH,
Devender

Re: MCSG and NTP

paolo barila — Sun, 05 Feb 2006 12:42:03 GMT

...ah...check also in /etc/ntp.conf of
rp8400 and n4000 to be sure you don't restrict access from your clients to ntp server like this:

restrict default ignore
restrict x.y.z.w # your client?