rhoai-distributed-workloads

name	rhoai-distributed-workloads
metadata	{"author":"rhoai3-demo","version":"1.0.0","platform-family":"rhoai","platform-baseline":"repo","ocp-baseline":"repo","skill-group":"RHOAI Platform"}
description	Use when installing, documenting, reviewing, or rebuilding Red Hat OpenShift AI distributed workloads support: Red Hat build of Kueue Operator, cert-manager, DataScienceCluster component states for kueue, ray, trainingoperator, aipipelines, dashboard, and workbenches, GPU/RDMA prerequisites, CA bundle behavior, and validation of Kueue, KubeRay, and Kubeflow Training Operator pods. Do NOT use for Kueue day-2 queue enforcement, dashboard Kueue enablement, or embedded Kueue migration; use rhoai-kueue-workload-management. Do NOT use for ResourceFlavor, ClusterQueue, LocalQueue, RDMA, or Ray administrator troubleshooting; use rhoai-distributed-workload-operations. Do NOT use for user workflows that run Ray, PyTorchJob, or TrainJob workloads; use rhoai-distributed-workload-workflows. Do NOT use for general RHOAI installation, model serving, KFP authoring, or live troubleshooting; use rhoai-self-managed-installation, rhoai-kfp-pipeline-authoring, and env-* skills as appropriate. Do NOT use for Kubeflow Spark Operator activation or SparkApplication resources; use rhoai-kubeflow-spark-operator.

RHOAI Distributed Workloads

Use this skill to install and review the RHOAI distributed workloads component set for the active baseline in docs/PLATFORM_BASELINE.md.

Source Grounding

Read references/source-capture.md before using product configuration details. Official Red Hat documentation is product authority. This skill adapts the official component installation guidance to this repo's GitOps operating model.

Installation Model

Distributed workloads require both external Operators and DataScienceCluster component state:

• Red Hat build of Kueue Operator must be installed separately.
• cert-manager Operator must be installed.
• spec.components.kueue.managementState must be Unmanaged, so the Red Hat build of Kueue Operator manages Kueue.
• ray.managementState is Managed when Ray/KubeRay workloads are needed.
• trainingoperator.managementState is Managed when Kubeflow Training Operator workloads are needed.
• dashboard, aipipelines, and workbenches states depend on whether distributed workloads are launched from pipelines, workbenches, or both.

GitOps Policy

For this repo:

• Derive component state from the official table, but apply it through ArgoCD Applications and GitOps manifests.
• Keep prerequisite Operators in earlier or separate GitOps layers so DataScienceCluster does not enable distributed workload components before dependencies are available.
• Keep Kueue integration, namespace queue enforcement, dashboard Kueue enablement, and workload management policy in rhoai-kueue-workload-management.
• Keep ResourceFlavor, ClusterQueue, LocalQueue, RDMA, and Ray administrator troubleshooting in rhoai-distributed-workload-operations.
• Keep Ray, PyTorchJob, TrainJob, fine-tuning, checkpointing, monitoring, and user troubleshooting in rhoai-distributed-workload-workflows.
• Keep Kubeflow Spark Operator activation and SparkApplication resources in rhoai-kubeflow-spark-operator.
• If GPU-backed distributed workloads are part of the demo, pair this skill with accelerator/GPU skills and verify GPU support before enabling workloads.

Workflow

1. Confirm the active baseline in docs/PLATFORM_BASELINE.md.
2. Confirm RHOAI is installed with rhoai-self-managed-installation.
3. Verify prerequisites from references/official-doc-extraction.md.
4. Choose the launch surface:
   • pipelines only
   • workbenches only
   • pipelines and workbenches
5. Set the required DataScienceCluster component states from the official matrix.
6. Apply manifests through ArgoCD/GitOps.
7. Validate Kueue, KubeRay, and Kubeflow Training Operator deployments with references/validation-checklist.md.
8. Continue with rhoai-kueue-workload-management for queue enforcement, dashboard Kueue enablement, and migration.
9. Continue with rhoai-distributed-workload-operations for distributed workload quota resources, RDMA, and Ray administrator troubleshooting.
10. Continue with rhoai-distributed-workload-workflows for user-facing Ray, Training Operator, Kubeflow Trainer v2, checkpointing, and monitoring workflows.

References

• references/source-capture.md
• references/official-doc-extraction.md
• references/validation-checklist.md
• examples/dsc-distributed-workloads-states.md