I think the point was that profile guided optimization relies on the workload staying relatively fixed. If the workload suddenly changes (like promoting a reado

I think the point was that profile guided optimization relies on the workload staying relatively fixed. If the workload suddenly changes (like promoting a readonly slave to be the writable master) the assumptions made by PGO may not be valid, and performance could be worse than if no modifications were made in the first place. I think you'd just have to measure scenarios before using it in production.