Automated Anomaly Detection

What is an Anomaly in Performance Test Results?

An anomaly is any pattern in test results that stands out from what is expected or from a baseline. Normally, anomalies come in a few forms:

• Sudden spikes - response time increases suddenly at a specific point in the test then recovers.
• Gradual degradation - response time slowly increases over the duration of the test, indicating a memory leak or connection pool exhaustion.
• Error bursts - a transaction that was succeeding starts failing repeatedly.
• Throughput plateau - sending more load produces no more throughput, indicating a saturation point and the bottleneck.
• Transaction-specific degradation - most transactions are stable but one specific endpoint shows 5x worse response time.
• Samples which stand out - a number of requests take, for example, 10x the time compared with the average response time.

Finding these manually means staring at charts for a long time. With AI, you either describe or provide the data and ask it to find the most “interesting” parts.