Rally is a benchmarking tool developed by the Elasticsearch team to evaluate Elasticsearch's performance, allowing developers to reproduce performance numbers and create their own benchmarks without dealing with complex details. Originating from Python scripts used for nightly benchmarks, Rally aims to prevent gradual performance degradation and improve the performance of Elasticsearch by enabling developers to run benchmarks independently during development. It supports features like telemetry devices for detailed analysis, allowing inspection of Java flight recorder and JIT compiler behavior, and facilitates visualization of metrics data with Kibana. Rally currently supports single-machine benchmarks and is working towards allowing multi-machine benchmarks, improving measurement correctness, and separating benchmark definitions from the tool itself. It simplifies the process of running benchmarks, with the potential for more flexibility in benchmark scenarios and the elimination of current limitations such as coordinated omission in latency measurements.