MySQL resource agent (mysql-wss) may elect and bootstrap two different cluster members at the same time

[Alexander Dobdin]

This results in a Primary Component collision that usually leads to split-brain situation, where one node remains in an isolated state. Occasionally, Galera detects this collision and orders an abort on one of the bootstrapped node, but usually that node stuck in this aborting phase and never exits. During that state it

Steps to reproduce:
1. Check the state of the Galera cluster members on all the controllers using the following MySQL statements:
> SELECT VARIABLE_VALUE FROM information_schema.GLOBAL_STATUS WHERE VARIABLE_NAME = "wsrep_local_state"
output should be: 4
> SELECT VARIABLE_VALUE FROM information_schema.GLOBAL_STATUS WHERE VARIABLE_NAME = "wsrep_cluster_size"
output should be the number of the controllers: 3
2. Disable the entire p_mysqld resource:
# pcs resource disable p_mysqld
3. Wait until all the MySQL server daemon shut down properly on all the controllers.
4. Enable the p_mysqld resource:
# pcs resource enable p_mysqld

Fault frequency is around ~8% (10 occurrence out of 123 attempts)

[Michael Polenchuk (mpolenchuk)]

Not reproduced, please make sure you've got the latest ocf script https://github.com/openstack/fuel-library/blob/stable/mitaka/files/fuel-ha-utils/ocf/mysql-wss