Storage Foundation for Oracle® RAC 7.4.1 Administrator's Guide - Linux

Examining GAB seed membership

The number of systems that participate in the cluster is specified as an argument to the gabconfig command in /etc/gabtab. In the following example, two nodes are expected to form a cluster:

# cat /etc/gabtab

/sbin/gabconfig -c -n2

GAB waits until the specified number of nodes becomes available to automatically create the port "a" membership. Port "a" indicates GAB membership for an SF Oracle RAC cluster node. Every GAB reconfiguration, such as a node joining or leaving increments or decrements this seed membership in every cluster member node.

A sample port 'a' membership as seen in gabconfig -a is shown:

Port a gen 7e6e7e01 membership 01

In this case, 7e6e7e01 indicates the "membership generation number" and 01 corresponds to the cluster's "node map". All nodes present in the node map reflects the same membership ID as seen by the following command:

# gabconfig -a | grep "Port a"

The semi-colon is used as a placeholder for a node that has left the cluster. In the following example, node 0 has left the cluster:

# gabconfig -a | grep "Port a"

Port a gen 7e6e7e04 membership ;1

When the last node exits the port "a" membership, there are no other nodes to increment the membership ID. Thus the port "a" membership ceases to exist on any node in the cluster.

When the last and the final system is brought back up from a complete cluster cold shutdown state, the cluster will seed automatically and form port "a" membership on all systems. Systems can then be brought down and restarted in any combination so long as at least one node remains active at any given time.

The fact that all nodes share the same membership ID and node map certifies that all nodes in the node map participates in the same port "a" membership. This consistency check is used to detect "split-brain" and "pre-existing split-brain" scenarios.

Split-brain occurs when a running cluster is segregated into two or more partitions that have no knowledge of the other partitions. The pre-existing network partition is detected when the "cold" nodes (not previously participating in cluster) start and are allowed to form a membership that might not include all nodes (multiple sub-clusters), thus resulting in a potential split-brain.