The opendatahub-operator is a Kubernetes operator written in Go that serves as the primary control plane for Open Data Hub, a platform designed to manage and deploy data science applications and infrastructure components. The operator uses the DataScienceCluster custom resource definition to declaratively deploy and configure applications such as Jupyter Notebooks, data science pipelines, and other machine learning tools within Kubernetes clusters.
The operator functions as a comprehensive orchestration layer for a multi-tenant, cloud-native data science platform. It manages the lifecycle of numerous integrated components including KServe for model serving, Ray for distributed computing, Training Operator for ML job management, Feast for feature stores, Model Registry for model versioning, TrustyAI for AI explainability, the ODH Dashboard for web-based management, Workbenches for notebook environments, and AI Pipelines for ML workflow orchestration. These components are automatically integrated based on DataScienceCluster configuration rather than requiring separate installation, streamlining deployment complexity.
The platform supports deployment on OpenShift 4.19 or higher and can be installed directly from the community-operators catalog on OperatorHub, with the latest releases available through the Fast channel. The operator also supports multiple deployment methods including Cloud Manager for multi-cloud provisioning and RHAII Mode for specific provider integrations. Installation requires creating a DSCInitialization custom resource followed by a DataScienceCluster resource to enable desired components.
The operator supports extensive configuration through environment variables and flags, with comprehensive documentation covering prerequisites, platform requirements, namespace configuration, and resource allocation. Optional external operators can be installed for advanced functionality including certificate management via OpenShift Cert Manager, job queueing through Red Hat build of Kueue, distributed tracing with OpenTelemetry and Tempo operators, and enhanced observability through Cluster Observability and Perses operators. GPU support is available through NVIDIA GPU Operator and DCGM Exporter for GPU-accelerated workloads.
The repository includes detailed developer documentation covering local development setup, manifest customization, component addition procedures, and comprehensive testing infrastructure including functional tests, end-to-end tests, integration tests via Jenkins pipeline, and Prometheus unit tests for alerts. The documentation provides guidance on upgrade testing, release workflows, troubleshooting, and runtime logging level adjustments. The operator's extensible architecture allows customization of manifest sources for both local development and production operator image builds, supporting diverse deployment scenarios and organizational requirements.
