For Oracle RAC node eviction caused by heartbeat Network (not instance eviction), oraclerac
Node eviction of Oracle RAC caused by heartbeat Network (not instance eviction)
Q:
Suppose the following scenario: for the four node rac, the heartbeat line uses a m network switch. If the M network switch loses power, I want to know the algorithm of the crs eviction node?
Oracle Dalian GCS reply:
If the heartbeat network is disconnected, the oracle cluster is split into several small groups, which we call cohort (you can find in ocssd. log.
Oracle determines which sub-cluster to leave based on the number of nodes contained in each cohort. The basic principle is:
1. If the number of nodes is large, the group is left behind and the number of nodes is small.
2. If the number of nodes in each sub-cluster is the same, the nodes containing the minimum node number will be retained.
This is why you see two nodes in the cluster. If there is a problem with the heartbeat network, it must be because Node 2 is restarted.
In addition:
Node eviction is not related to VF (that is, voting disk, which many people say.
The main reason is that many people confuse the instance eviction at the database level with the definition at the cluster level.
That's why we have come to this misunderstanding.
The database instance-level eviction is indeed related to the control file, because the control file plays the role of VF at the database level.