Query about time difference between two node cluster

[Md. Minhaz Khan]

Dear Concern

We have two nodes configured with MC/Service Guard.My query is if time difference between two nodes becomes 20 minutes or 1 hour while cluster is running then is there any problem? I mean cluster status will abnormal or not?

Thanks in Advance

Minhaz

[Suraj K Sankari]

Hi,
>>My query is if time difference between two nodes becomes 20 minutes or 1 hour while cluster is running then is there any problem?

Please elaborate your question.
Which time difference you talking about?

Suraj

[Md. Minhaz Khan]

Actually i meant system time difference between two nodes differ when cluster is running.

Say present time of the two node is:

Node1: Mon Jun 8 10:48:34
Node2: Mon Jun 8 10:48:55


After synchronization time of Node1 with NTP server if Node1 time becomes as follows when cluster is running:

Node1: Mon Jun 8 11:48:34

and

Node2 time still is:Mon Jun 8 10:48:55


Then there may could occur any abnormality in the cluster or at first we need to stop the cluster then after synchronization time of node1 and node2 with NTP server, start the cluster again.


Thanks

Minhaz

[Michal Kapalka (mikap)]

Hi,

the problem could be present if you are running som app like a oracle, that is time depend on data consistence.

The logs will have also the difference, syslog,
and the cluster and also the packages logs.

mikap

your way how to correct the prblem with the time is correct, stop the cluster, settup of NTP server and use ( date or ntpdate -b ntp_server ) and them start the cluster.

[Michal Kapalka (mikap)]

Hi,

the problem could be present if you are running som app like a oracle, that is time depend on data consistence.

The logs will have also the difference, syslog,
and the cluster and also the packages logs.

mikap

your way how to correct the problem with the time is correct, stop the cluster, settup of NTP server and use ( date or ntpdate -b ntp_server ) and them start the cluster.

[Md. Minhaz Khan]

Dear Michal

Our cluster does not care about Oracle because oaracle 10g has its own CRS. Our cluster package only configured with a vg that exclusively mount by the active node. So in this scenario if i change the time of one node then cluster will abnormal or not?

I know that best method is always stop the cluster then sync nodes with NTP server then start the cluster.


Thanks
Minhaz

[Nido]

Hi Minhaz,

Check the ntp configuration and NTP drift setting to fix the time differences between the nodes.

Follow the URL:
http://docs.hp.com/en/B2355-90147/ch07s06.html?btnNext=next%A0%BB

Cheers!!

[Michal Kapalka (mikap)]

hi,

if your app is independ on the time, it should go to change the time on one node online, but i'm not able to say,that nothing will happend.

The best way is to stop the cluster, or risk the cluster failure.

mikap

[sujit kumar singh]

Hi Larger time differences in the two nodes can cause the improper functionalty of the cmcld daemon and other cluster services. always use an NTP server for the cluster nodes.

from application point of view also that is necessary as if the time diff is too high and the application fails over to the other node, there can be usually time mismatch on the application files.

regards
sujit

[Duncan Edmonstone]

Minhaz,

> Our cluster does not care about Oracle because oaracle 10g has its own CRS.

Are you saying you have a Serviceguard cluster running across these 2 nodes, and then you *also* have an Oracle clusterware cluster (CRS) running across these 2 nodes???

The *only* configuration in which this would be supported is when you have Oracle RAC integtrated with Serviceguard using Serviceguard Extension for RAC... in that situation Serviceguard starts and stops CRS and is actually responsible for cluster membership.

If you are just running the 2 cluster products seperately on one set of nodes, that *sin't* supported.

And yes, I agree with the other posters here - get the 2 nodes synched together using NTP.

HTH

Duncan

[Md. Minhaz Khan]

Thanks a lot to everybody to help me.

Thanks
Minhaz

[John Bigg]

I know this thread is closed, but I have to add comments to correct some misinformation included here.

> Hi Larger time differences in the two nodes can cause the
> improper functionalty of the cmcld daemon and other cluster
> services.

This is simply not true.

Serviceguard does not use clock time at all. It is what is known as wildtime safe. i.e. it can handle the clock going backwards, forwards and the time on different nodes being vastly different. Timers are based on the number of ticks since the node was booted and all time on a single node is based on this. This value does not change if the clock time is adjusted. The communication between the nodes uses a heartbeat sequence number and since the period between heartbeat is a known fixed interval, no time needs to be passed between nodes.

Therefore, as far as Serviceguard (and cmcld, the main cluster daemon) is concerned, the clock can be adjusted without problem.

Note though that the same cannot be said of applications and HP-UX itself. HP-UX does not support the clock being turned backards. If the clock is ahead the NTP should be used to slow things down until the clock has caught back otherwise you could see some strange effects, make failures for example. Furthermore, applications may not like having the clock set differently on different nodes. What happens in your database if you fail over a package to another node which is set back in time and then update rows based on time?

Lastly, I'd encourage you to keep time in sync simply because it is a nightmare troubleshooting cluster problems when the timestamps in the cluster flight recorder logs (which are real clock times) and in syslog are not in sync. You have to end up processing the logs to adjust the time to make inter node comparisons possible without great difficulty.

So, to summarise. Serviceguard does not care about the time, but you need to carefully consider you log files, HP-UX and any applications you may run.