Get Started¶
Omnia Deployment Flow¶
Choose your deployment path based on your cluster requirements, available hardware, and desired workload. Each path is a self-contained, end-to-end tutorial that takes you from a bare set of PowerEdge servers to a fully operational cluster.
Note
Before selecting a path, complete the Prerequisites Checklist to ensure your hardware, networking, and software environment are ready.
Select a catalog for the deployment path
Catalog selection is mandatory because it determines which functional groups, software groups, packages, operating-system versions, and node architectures Omnia builds and provisions.
- Slurm Quickstart requires Slurm controller and compute functional
layers, including the mandatory
slurm_custom_groupcomponent. - K8S Telemetry Only requires service Kubernetes control-plane and worker functional layers with the service Kubernetes and Telemetry components.
- Full Deployment requires a combined catalog containing both Slurm and service Kubernetes functional layers.
- BuildStreaM Deployment requires a catalog that matches every functional group requested by its automated workflow.
See Select or update the catalog for the shipped catalog choices. The Slurm Quickstart and K8S Telemetry Only paths provide the exact minimum layer names, required components, and verification commands.
Deployment Paths at a Glance¶
| Path | Name | Workload | Nodes | Time | Description |
|---|---|---|---|---|---|
| A | Slurm Quickstart | Traditional HPC (Slurm) | 4+ | ~2 hrs | Overview page with links to detailed Slurm deployment guides. Covers Slurm setup, GPU provisioning, node management, configuration backup, and HPC benchmarks. Ideal for first-time users and large-scale HPC workloads. |
| B | K8S Telemetry Only | Kubernetes + Telemetry (no Slurm) | 5 | ~2 hrs | Deploys a 3-control-plane + 1-worker Kubernetes cluster with the complete telemetry pipeline (For example: iDRAC metrics, LDMS, Kafka, VictoriaMetrics). No Slurm. Use this when you need infrastructure monitoring without a job scheduler. |
| C | Full Deployment | Slurm + Service K8s + Telemetry | 8 | ~4 hrs | Production-grade deployment with Slurm scheduling, a highly available 3-node Kubernetes service cluster, LDAP authentication, and full telemetry (For example: iDRAC, VictoriaMetrics). Best for teams running mixed HPC/AI workloads with monitoring requirements. |
| D | BuildStreaM Deployment | BuildStreaM (Catalog-Driven CI/CD) | 8+ | ~6 hrs | Automated, catalog-driven deployment using GitLab CI/CD pipelines. BuildStreaM reads a declarative catalog to provision and configure the entire cluster. Best for organizations with GitOps workflows or repeated, reproducible deployments at scale. |
Omnia deployment modules¶
Omnia uses a modular, capability-based architecture in which each deployment
module handles a specific part of cluster deployment. Modules exchange
documented input/output contracts and can be invoked separately using the
omnia.sh CLI. End-to-end paths run the required modules in dependency order.
| Deployment module | Description |
|---|---|
Repository Manager (repo_manager) |
Local repository creation and package management for air-gapped deployments |
Image Build Manager (image_build_manager) |
Diskless OS image building for each functional group |
Discovery (discovery) |
BMC discovery and PXE mapping generation through OME; manual mappings are supplied directly to Orchestrator |
Orchestrator (orchestrator) |
Node provisioning, boot configuration, and cluster setup |
Telemetry (telemetry) |
Telemetry pipeline deployment (iDRAC, LDMS, Kafka, VictoriaMetrics, VictoriaLogs) |
BuildStreaM (build_stream) |
GitLab CI/CD automation for catalog-driven build and deployment pipelines |
Utilities (utils) |
Utility operations including unattended OS installation and log collection |
Info
For detailed information on module execution order and dependencies, see Running Deployment Modules.
Which Path Should I Choose?¶
"I just want Slurm running as fast as possible." Start with Slurm Quickstart (Path A). You can always add Kubernetes and telemetry later.
"I only need telemetry dashboards -- no job scheduler." Choose K8S Telemetry Only (Path B). This gives you iDRAC-to-Victoria Metrics visibility without the overhead of Slurm.
"I need a production cluster with monitoring and authentication." Go with Full Deployment (Path C). This is the canonical Omnia deployment that exercises every major subsystem.
"I want CI/CD-driven, repeatable infrastructure." Use BuildStreaM Deployment (Path D). BuildStreaM automates the entire lifecycle through GitLab pipelines and a declarative catalog.
Before You Begin¶
Every path assumes you have completed the items in Prerequisites Checklist. That page covers:
- Supported hardware and firmware versions
- OIM (management node) requirements (RAM, OS, Podman, NICs)
- Network switch configuration (admin + BMC VLANs)
- NFS / storage preparation
- BIOS and iDRAC settings on target nodes
- Required RHEL subscriptions and Docker credentials
Tip
Print or bookmark the Prerequisites Checklist -- it doubles as a day-of-deployment runbook you can hand to a datacenter technician.