I have created (3) node MySQL InnoDB Cluster using version 8.0.20. I simulate a Primary node (node3) failure by doing the following:
- On Primary node, add RestartSec=30 to /usr/lib/systemd/system/mysqld.service [Service] section
- Run systemctl daemon-reload to reload the configuration
Then I killed mysqld process: kill -9 $(pidof mysqld)
I ran the query to find out who will be the new Primary. It is another read only node (node2). After 30 seconds, the node3 mysqld process starts, and per checking cluster.status(), node 3 already joined the cluster and becomes a R/O node.
I expect I would need to run rejoinInstance() for node 3 upon mysql restart. It turns out it is not necessary.
What kind of failure would require to run rejoinInstance?