1.4 Master-Election Algorithm

Novell Cluster Services for OES 11 SP1 and later introduces some intelligence in the master election process when the master leaves a cluster (voluntarily or involuntarily). The new algorithm substantially reduces the time needed for master election in some cases.

During an upgrade, a master-election dispute can rarely occur in a larger mixed-mode cluster (4 or more nodes) that consists of mixed NetWare 6.5 SP8 and OES nodes. If multiple nodes, including the master, concurrently leave the cluster, the different master-election algorithms might result in a dispute where equal-sized groups of old nodes and new nodes each elects its own master. SBD resolves the dispute by fencing one of the masters.

The fencing does not result in data loss. You can reboot the nodes that are poison-pilled, and they will re-join the cluster.

If you have a larger cluster (4 or more nodes), you can avoid the master-election dispute/fencing problem by upgrading the NetWare nodes with higher IP addresses first, in descending order from highest to lowest. This helps ensure that in master node election disputes between a NetWare node and an OES node, the OES node has the higher IP address and becomes the master.

For example, assume that you have a four-node NetWare 6.5 SP8 cluster with IP address assignments as follows:

  • Node1, IP: 192.168.99.10
  • Node2, IP: 192.168.99.11
  • Node3, IP: 192.168.99.12
  • Node4, IP: 192.168.99.13

You should first upgrade Node4 because it has the highest IP address. Upgrade Node3 second, then Node2, and finally Node1.