Orion stands as a powerful command-line tool/daemon designed for identifying regressions within perf-scale CPT runs, leveraging metadata provided during the process. The detection mechanism relies on hunter.
$ podman build -f Dockerfile -t orion
# Needed env vars.
# ES/OpenSearch Server where the results live
$ export ES_SERVER='my-opensearch.perf.com'
# Version of OpenShift
$ export version=4.19
# Index where the benchmark data is stored
$ export es_benchmark_index=ripsaw-kube-burner*
# Index where you store the run metadata
$ export es_metadata_index=perf_scale_ci*
$ podman run --env-host orion orion cmd --config orion/examples/trt-external-payload-node-density.yaml --hunter-analyze$ git clone <repository_url>
$ python3.11 -m venv venv
$ source venv/bin/activate
$ pip install -r requirements.txt
$ pip install .# Command-line mode
$ orion cmd --hunter-analyze
# Daemon mode
$ orion daemon- Regression Detection: Identify performance regressions using advanced statistical methods
- Multiple Algorithms: Support for Hunter, CMR, and anomaly detection
- Flexible Configuration: YAML-based configuration with extensive customization options
- Command-line & Daemon Modes: Use as a CLI tool or run as a service
- Multiple Output Formats: JSON, CSV, and JUnit XML output support
- Installation Guide - Detailed setup and build instructions
- Configuration - Configuration format and metrics options
- Usage Guide - Command-line options, examples, and configurations
- Daemon Mode - API documentation and daemon setup
Orion currently supports Python version 3.11.x. Please be aware that using other Python versions might lead to dependency conflicts. Python 3.12.x may result in errors due to the removal of distutils.
Please read CONTRIBUTING.md for details on our code of conduct and the process for submitting pull requests.
This project is licensed under the terms specified in the LICENSE file.
