MC/ServiceGuard subnet polling interval

[Harri Muttilainen]

I have noticed that something is pinging our network at about 2-3 minutes interval. I suspect it would be the SG, but I have not seen any documented value for the subnet polling interval. Would anyone know it?

[John Palmer]

Hi,

The 'Network Polling Interval' is configured in the cluster configuration file as:-

NETWORK_POLLING_INTERVAL

The default is 2 seconds.

Regards,
John

[MARTINACHE]

Your cluster config file should something like :
/etc/cmcluster/CLUSTERNAME.ascii

grep NETWORK /etc/cmcluster/CLUSTERNAME.ascii

The answer shoul be :
NETWORK_POLLING_INTERVAL 2000000

This means 2 seconds interval.

regards,

Patrice.

[Harri Muttilainen]

Thanks, John,

but I meant the subnet monitoring that the package is doing. The network_polling_interval is for the linkloop polling interval for the network interfaces of the cluster (right?). The package is also monitoring the subnet and I think this is done by pinging it. Do you know about that?

[John Palmer]

Yes, network_polling_interval is the interval between checking LAN cards.

I don't believe that there is any network monitoring built into a Serviceguard package by default.

However, what you build into your package monitor script is up to you. It's possible that this has been configured to ping on a regular basis. I suggest that you have a look at your scripts.

Regards,
John

[MARTINACHE]

Your package configuration file should be something like :

/etc/cmcluster/PKGNAME/PKGNAME.ascii

grep POLLING /etc/cmcluster/PKGNAME/PKGNAME.ascii

You will see the polling interval of resources that are monitored by the package.

You you need to know which SUBNET are monitored by package :

grep SUBNET /etc/cmcluster/PKGNAME/PKGNAME.ascii

[Harri Muttilainen]

Thanks John and Patrice!

That could be the answer. The confusing part is still that the 60 seconds default for the resource monitoring seems a little bit too frequent according to what I've noticed. The other confusing point is that we have actually two packages running on separate machines, but only the other one seems to do the polling although the package scripts are almost identical.

[Stephen Doud]

SUBNET polling is not performed directly.
ServiceGuard uses the NETWORK_POLLING_INTERVAL to determine when to test each LAN card. When both the PRIMARY and STANDBY LANs are determined to be uncommunicative, the SUBNET is marked DOWN and if the SUBNET was loaded in the package configuration file when CMAPPLYCONF was run, ServiceGuard will fail the package (an adoptive node may pick it up if it's SUBNET is UP).