HA behaviour with clustered switches.

Hey everyone,

we have a newly designed infoblox environment with two sites and an HA pair on each side. Every node is a hardware appliance.

Let's consider one HA pair, with appliances A and B. Every appliance has three ports connected: LAN1, LAN2 and HA.

LAN1 and LAN2 are connected as a bond. On the other end are two clustered Switches, which have configured Port-Channels. Therefore, Port 1 on Switch 1 and 2 are considered as Port-Channel 1 and so on. We also configured the LAN1 and LAN2 redundancy to use the bonding mechanism.

The HA port is connected to either Switch 1 (Appliance A) or Switch 2 (Appliance B).

If I'm not mistaken, the active node sends the vrrp packets over its HA port to the LAN1 interfache of the passive node. Let's consider that the switch cluster is updated. In this case, LAN1 and HA from appliance A (active) will be down and LAN2 is still up if Switch 1 is updated. For appliance B, LAN1 is down and LAN2 and HA are still up. The question: How does the HA mechanism behave?

We assume that appliance B does not receive vrrp messages on its LAN1 interface, since the HA from appliance A is down. So, appliance B will be active or at least tries to active. Nonetheless, appliance A is still reachable since LAN2 is still up. This seems to be split brain situation.

Do we understand that right or do we miss something? How is it supposed to be solved? Is there a best practise?

Tagged: