These errors were seen on a two node VCS 5.0 cluster running on Solaris 10 whenever the MultiNICB resource was onlined. The MultiNICB resource was configured in Base mode and was using Intel Pro e1000g interfaces.
The VCS configuration looked as follows:
group aa (
SystemList = { svr1 = 0, svr2 = 1 }
AutoStartList = { svr1 }
)
MultiNICB ddd (
Critical = 0
Device = { e1000g0 = 0, e1000g4 = 1 }
)
The hostname.e1000g# files had the following entries:
# cat /etc/hostname.e1000g0
svr1 netmask + broadcast + deprecated -failover up addif svr netmask + broadcast + up
# cat /etc/hostname.e1000g4
svr3 netmask + broadcast + deprecated -failover up
As soon as the MultiNICB resource was onlined,the FAILED flag was set.It was also observed in the 'ifconfig -a' output that another interface outside of the mnicb group had the same subnet mask and whenever both nodes in the cluster were rebooted,all interfaces belonging to the mnicb group had the FAILED flag set.
4. Resolution.
As one of the requirements of MultiNICB agent is that the MultiNICB resource must include ALL of the interfaces that belong to the same IP subnet. If some interfaces on the same IP subnet are outside of MultiNICB control,
this can lead to complications in the event of device failures.
As there was a requirement in this particular configuration that the MultiNICB resource had to be configured in Base Mode,the above error messages could be resolved in two ways and full High Availability of the network interfaces would still be provided:
1) Use IPMP mode for monitoring the devices. It is a safe and reliable way to test for network connectivity. Minimal changes to the configuration required.
2) Use Link test only: In this configuration, the agent does not do any ICMP pings and determines the state of the interface by testing the link status of the interface.
So with the above MultiNiCB ddd resource would looks as follows with the Use Link test only:
MultiNICB ddd (
IgnoreLinkStatus = 0
LinkTestRatio=02
Device = { e1000g0 = 0, e1000g4 = 1 }
In this instance it was decided that the 'Link test only' option best suited the requirements of this configuration.