Managing a Dual Master Situation

If a daisy chain is broken, or if a ring is broken in two places, it is possible to form two separate active stack topologies. This results in a dual master situation, as shown in the following example .

Click to expand in new window
Example of a Split Stack that Results in a Dual Master Situation
../Graphics/BD_161_Nodes_drawing_1.svg
P6 Node 6 is powered off
M Master nodes
B Backup nodes
S Standby nodes
X Indicates the broken link

In the example, a link was broken while a node in the ring was powered off. The broken link formerly connected the original master (M1) and backup (M2) nodes of a single active topology.

All nodes in the stack except the powered-off node are in the active topology and all nodes are configured to be master-capable. Nodes 1, 7 and 8 form an active topology and nodes 2, 3, 4, and 5 form another active topology. Node M2 immediately transitions from backup to master node role. Nodes B8 and B3 are elected in their respective active topologies as backup nodes.

If the backup node is on one stack and the master node is on the other, the backup node becomes a master node because the situation is similar to that of master failure. Because both stacks are configured to operate as a single stack, there is confusion in your networks. For example, all of the switch‘s configured IP addresses appear to be duplicated. The management IP address also appears to be duplicated since that address applies to the entire original stack.

To help mitigate the dual master problem, you can configure master-capability so as to prevent some nodes in the stack from operating in backup or master node roles. In addition, you can force all nodes in the (broken) stack topology to restart and come up as not master-capable for the life of that restart. The save configuration {primary | secondary | existing-config | new-config} command saves the configuration on all nodes in the active topology.

Standby nodes that exist in a severed stack segment that does not contain either the original master or backup node do not attempt to become the master node. Instead, these nodes reboot. After rebooting, however, a master election process occurs among the nodes on this broken segment, resulting in a dual master situation.

Dual master conditions are also possible when two non-adjacent nodes in a ring or a single (middle) node in a daisy chain reboot.

For a period of time, a rebooting node does not advertise itself to its neighbors, resulting in temporary stacking link failures. This can cause node isolation, and the nodes that are isolated perform as a severed stack segment depending on the circumstances of the severance:
  • If the backup node is on the broken portion, it becomes a (dual) master.
  • If the backup node is on the same portion as the master, all nodes on the (other) broken portion reboot.

When the rebooting nodes have sufficiently recovered, or when a severed stack is rejoined, the dual master condition is resolved, resulting in the reboot of one of the master nodes. All standby and backup nodes that had been acquired by the losing master node also reboot.

You can avoid a dual master possibility during configuration by:
  • Configuring the stack in a ring topology.
  • Avoiding having too many master-capable nodes when configuring larger stacks.
  • Placing the master-capable nodes that provide stack redundancy in such a way that broken stacking links are unlikely.