Oracle Database Appliance
 

PLA-Support>Eng Systems>Exadata/ODA/SSC>Oracle Database Appliance>DB: ODA_EST
 

Applies to:

Symptoms

 2-Node RAC cluster using GI running in Guest vm on oda:

On the physical servers (dom0), we have 4 NICs and we do couple every 2 of them to build up 2 bonded interfaces bond0 and bond1. 2 XEN bridges were created using these bonded interfaces: xenbr0 and xenbr1. The guests (RAC nodes) get two virtual NICs eth0 and eth1, mapped to the xenbr's. GI installed and configured inside the guests uses 1 NIC for public network and one for interconnect.

Problem:

In a situation if we lose both NICs behind the same bond in dom0 (i.e. the whole bond becomes disfunctional), this is processed normally by GI in the guest if it's the bond behind the interconnect.
It's not detected by GI if the bond exists behind the guest's public interface so GI does not initiate an VIP-failover, thus compromising client access to the cluster.

Changes

 The nic attached to public or private inrterconnect goes down.

Cause

The GI checks for an link down event in order to determine if there is an interface failure. In this case the down link even is not detected inside the guest because this event is not being transferrred from the physically failed hardware of dom0.

This issue is beimng tracked in the following bug:

BUG 13104709 - VIP DIDN'T FAILOVER IN GI 11.2 RUNNING ON OVM 

Solution

 In 12.1.0.2 there is a new feature called PING TARGET was introduced that allows to overcome this issue. Users can set that using: 

Refer to the following note for more info on the pingtarget feature:

12.1.0.2 Ping Target for Oracle Clusterware VIP (Doc ID 1958241.1)

References

This solution has no attachment