@Gargron if the failure is in the old primary, and you can quantify the problem automatically eg hardware failure etc then no

However you have to engineer a failover trigger to be resistant to network events. That’s the risk: network partitions (depending on your setup), transient events (short network hiccups)

Failover (needlessly) to a new primary is not free in the best case as the old primary has to be reinitialized on the new timeline