In this lesson, you will learn how to determine the precise threshold where your database infrastructure stops scaling and begins to degrade by identifying its saturation point.
Defining the Saturation Point
Saturation occurs when your database system reaches maximum utilization—whether in CPU, I/O, or connection availability—and can no longer handle additional requests without a disproportionate increase in latency. To identify this, we perform synthetic load testing, which involves simulating high-traffic scenarios to find the "knee" in your performance curve.
The Testing Approach
When testing, you must isolate the database from the application layer to avoid "noise." Use a benchmarking tool (like pgbench or sysbench) to flood the database with a representative workload that mimics your production traffic patterns, such as a mix of point selects, range scans, and transactional writes.
While applying this synthetic load, monitor two critical metrics: throughput (TPS) and latency. Initially, throughput will rise linearly with the number of concurrent connections. Eventually, you will hit a point where throughput plateaus while latency climbs exponentially. This is your saturation point.
Connection Pool Calibration
Often, the saturation point is not the database hardware itself, but the connection pool configuration. If your pool size is too small, you see artificial queueing (latency). If it is too large, you induce excessive context switching and lock contention inside the database engine.
-- Example: Using pgbench to stress test with 50 clients
-- and a scale factor of 100 to find the saturation knee.
pgbench -c 50 -j 4 -T 60 -f workload.sql my_database
By incrementally increasing the connection pool size during your synthetic tests, you can observe the exact moment when adding one more connection causes query times to spike due to resource contention. This allows you to tune your max_connections or pool settings to the "sweet spot" just before the saturation threshold, ensuring your application remains responsive under peak load without overwhelming the database engine.
Key Takeaway: Synthetic load testing identifies the saturation point by observing the trade-off between throughput and latency, allowing you to calibrate connection pools to maximize utilization without triggering performance degradation.