Flash card
Amazon RDS Performance Insights: database load decomposed by wait event, SQL statement, host and user.
- Plots database load as average active sessions against the instance’s vCPU count, so the line crossing max vCPU is the saturation signal.
- Breaks that load down by wait event, which separates a CPU-bound workload from one blocked on locks or on I/O.
- Also slices by top SQL, host, user and application, so the specific statement responsible is identifiable rather than inferred.
- Available across RDS and Aurora engines, with a free retention window and a longer paid one.
- Turning it on needs no schema change, no extension, and no agent inside the database.
Pick it when
Pick it over CloudWatch metrics when you know the database is the constraint but not why. CloudWatch reports CPU, connections and IOPS on the instance; Performance Insights attributes that load to the wait events and statements causing it.
It's the wrong answer when
It is the wrong answer when the bottleneck is not in the database at all, and when the scenario describes a connection-count problem rather than a query problem. Thousands of short-lived connections exhausting the instance is a pooling answer, not a tuning one.