topic reset caused by ntp syncronization in Operating System - HP-UX

reset caused by ntp syncronization

WilliamSmith11 — Wed, 08 Apr 2009 15:33:05 GMT

Hi
I have a cluster with two rx6600 running hpux 11.23, one of the nodes lose connection with the NTP server and the rare condition seeing in the OLDsyslog is that the time go back by 192 second,

Exactly when the connection was lose with the NTP and when the system when back in the time by 192 second it happen a reset.

Can this issue cause a reset of the servers?

Can happen the same if I lose the NTP server?

Or the reset can happen because it when back in time?

Regards

W.S

Re: reset caused by ntp syncronization

James R. Ferguson — Wed, 08 Apr 2009 16:03:18 GMT

Hi:

I doubt that this happened because of a transient loss in connectivity to your NTP source(s). I suspsect that this adjustment occured naturally to correct your server's time. The choice osf _when_ to perform and how to perform such large adjustments is yours.

You should configure your NTP to set your server's time during startup. Thereafter, there should only be small increements and decrements in the clock rather than large jumps.

To do this, modify '/etc/rc.config.d/netdaemons' to specify a NTPDATE_SERVER value. Whether or not your server clocks steps small changes or slews them thereafter is controlled by the setting for 'XNTPD_ARGS'. Slewing occurs if it is set as '-x'. See the 'xntpd' manpages.

Regards!

...JRF...

Re: reset caused by ntp syncronization

kevin_m — Wed, 08 Apr 2009 22:26:50 GMT

Regarding your question about a reset of the server, that can happen if you're using Oracle RAC. If the time is off by over 60 seconds it's likely one of the nodes will be rebooted. I'm assuming by cluster you are referring to SG but I wanted to mention it in case.
- Kevin

Re: reset caused by ntp syncronization

WilliamSmith11 — Thu, 09 Apr 2009 01:04:32 GMT

Kevin , yes is a cluster, Real aplication Cluster.

Re: reset caused by ntp syncronization

kevin_m — Thu, 09 Apr 2009 01:37:21 GMT

You can verify what caused the reboot by checking /etc/shutdownlog. If RAC forced it there may be an entry for 'Oracle CRS TOC' or 'INIT'. Here is some information we received from Oracle support:

From the logs we can see the time was reset by 67.219555 seconds. The misscount time is 60 seconds. With a time change such as this one in larger increments we run the risk of rebooting the nodes since the cssd assumes it has been hung for that time -- here it was more than 1 minute. Pls ensure the time changes are made in smaller increments or only when the crs is down. You might consider the "ntp -x" option to ensure time change happens in smaller increments.