In a two-node cluster that is not using grid interprocess communication (GIPC) and has only a single private network interface card (NIC), when the NIC fails on the node on which the Oracle Grid Infrastructure Management Repository database (mgmtdb) is running (node A, for example) after the other node (node B) was evicted, then when the private network is reestablished on node A, node B joins the cluster but the database instance on node A is unable to start.
To start the database instance:
Stop the Oracle Clusterware stack on both nodes, as follows:
crsctl stop cluster
Start the Oracle Clusterware stack on both nodes, as follows:
crsctl start cluster
Stop the database instance on node A, on which the NIC failed, as follows:
srvctl stop instance -db db_unique_name -node A
Start the database on both nodes by running the following commands one of the nodes:
srvctl start instance -db db_unique_name -node B srvctl start instance -db db_unique_name -node A
The database instance will fail to start on node A until after the instance starts on node B.