The takeaways:
1. Using Postgres for a high-throughput system requires significant investment from query optimization to the underlying infrastructure.
2. If you’re an RDS user, make sure to monitor your ReadIOPS, WriteIOPS, and QueueDepth CloudWatch metrics closely; even a minor load increase can trigger an I/O sharp decrease.
3. Replacing a GP2 with a GP3 or IO2 might do it, but what you really need is to reassess your architecture to evaluate optimizing (queries, instance) vs. replatforming (PlanetScale Postgres Metal).