NETWORK_FAILURE_DETECTION Parameter

[J Helgeson]

I think I understand how this parameter is supposed to work with regards to how ServiceGuard fails to a standby card when the inbound and outbound counters stop incrementing, but my question is what counters are used to determine health of the card. We recently performed a reset on a supervisor card thinking that a failover would occur,but it did not. We currently have NETWORK_FAILURE_DETECTION set to INOUT.

Thanks,
Jack Helgeson

[Ganesan R]

Hi,

netstat -i will give you the in/out packet counters.

I think reset will not stop the growth of these counters. Even if you clear the counters it will continue to grow.

Network failure will happen only if any in/out counters stop growing for the pre-determined period.

[John Bigg]

The Serviceguard network sensor monitors the unicast packet counters. You can use the lanadmin display command to see the unicast counters.

But there are two problems with your test. Firstly we only check the counters if we do not receive our poll messages. If we receive the poll messages there is no need to check the counters since by definition we have both sent and received ok.

Secondly, we check the counters over a number of poll events so resetting these counters will not result in a failure unless you continuously reset the counters for a number of seconds so Serviceguard cannot detect them increasing over a period of time.

I think you are going to have to consider another method of testing for lan failures like actually breaking a network connection or with 11.18 using cmmodnet -d to simulate this.