Failover, Lost Writes and Split-Brain

A MySQL primary with asynchronous replicas is monitored by an orchestrator that promotes the most up-to-date replica when it misses three heartbeats (heartbeat interval 5 s). During a network glitch, the primary becomes unreachable from the orchestrator but stays up and reachable from half of the application servers for another 40 seconds.

1. Describe, step by step, what goes wrong in this scenario.
2. Explain how fencing prevents the worst outcome and give two concrete fencing mechanisms.
3. Would switching to semi-synchronous replication have prevented the problem? What would it have changed?