It’s very unclear what level of serializability MySQL’s “serializable” level actually provides; it is seldom used in practice because the performance cost is extreme (essentially table locks everywhere for both reads and writes).
I would be surprised if either isolation level is correctly enforced across distributed replicas, even with semi-synchronous row-based replication. My MySQL operational experience was strongly characterized by fighting replication divergence.
It’s very unclear what level of serializability MySQL’s “serializable” level actually provides; it is seldom used in practice because the performance cost is extreme (essentially table locks everywhere for both reads and writes).
I would be surprised if either isolation level is correctly enforced across distributed replicas, even with semi-synchronous row-based replication. My MySQL operational experience was strongly characterized by fighting replication divergence.