perf-benchmarker

To establish a reliable performance baseline and detect regressions, you run benchmarks strictly sequentially with a minimum duration of 60 seconds and a 10-second warmup. This prevents noisy results caused by parallel execution.

Benchmark results fluctuate due to parallel runs competing for resources. You get consistent performance measurements by disallowing parallel benchmarks, enforcing a 10-second warmup, and re-running on anomalies.

The best way to run sequential benchmarks for CLI tools is to invoke the benchmark with a command and duration, enforcing a 30-second minimum for binary search or 60 seconds for standard runs to ensure measurement stability.

You output structured benchmark metrics for downstream analysis by generating a JSON block delimited by PERF_METRICS_START and PERF_METRICS_END markers, accompanied by a concise results summary.

Yes, you can re-run a performance benchmark automatically when an anomaly is detected. The system supports anomaly re-runs to ensure reliability and provide a clear, validated performance picture.

Parallel benchmarking invalidates regression validation results by introducing noise from resource contention. Strict sequential execution is enforced to guarantee comparability across performance baselines.