| name | env-deploy-and-evaluate |
| metadata | {"author":"rhoai3-demo","version":"2.0.0","platform-family":"rhoai","platform-baseline":"repo","ocp-baseline":"repo","skill-group":"Demo Environment"} |
| description | Deploy and validate the rhoai3-demo environment on the active OpenShift baseline once active scripts and stages exist. During the reimplementation, use this skill to rebuild the deployment workflow from legacy references. Use when the user asks to deploy the demo, set up a new AWS demo environment, evaluate an existing deployment, re-deploy a specific stage, check deployment status, run the ordered stage sequence, or asks "is my cluster ready?". Also use for planned ArgoCD sync and deployment reports. Do NOT use for chatbot UI/prompt changes (use rhoai-chatbot-customization), model evaluation or benchmarking (use rhoai-model-evaluation), or diagnosis after a specific failing component is observed (use env-troubleshoot).
|
Deploy and Evaluate RHOAI Demo
Orchestrates a controlled, stage-by-stage deployment of the RHOAI demo on the
active product baseline in docs/PLATFORM_BASELINE.md.
Reimplementation Status
The active implementation is being rewritten. Stages 110, 120, 210, and 220
have active deploy and validate wrappers. Later stages remain planned until
their root-level stage folders and Argo CD Applications are created.
Do not run scripts from backup/legacy-implementation-2026-06-09/ unless the
user explicitly asks to restore or inspect the legacy implementation.
When to Use
- Deploying the demo to a fresh cluster
- Evaluating an existing deployment for completeness
- Re-deploying a specific step after a fix
- Running the full step sequence, including optional edge paths
Instructions
Code and Documentation Must Be Aligned
NEVER update documentation without updating the corresponding code, and vice versa. Partial changes — updating a README design decision without changing the manifest, or changing a manifest without updating the README — are prohibited. Every change must be atomic: code + docs + SKILL knowledge in the same commit.
One Stage at a Time
NEVER advance to the next stage until the current stage fully validates.
- Read the active stage README once a stage exists.
- Deploy with the active
deploy.sh once recreated; it must apply the
ArgoCD Application as its first action.
- Wait — operators take minutes, GPU nodes take 5-10 min
- Diagnose — consult official docs for the active baseline; use the
env-troubleshoot skill for issues
- Validate with the active
validate.sh once recreated; confirm exit code
0 (exit 2 = warnings only, acceptable).
- Record — note result, move to the next stage
Deployment Sequence
Use the taxonomy in
.agents/skills/project-demo-stage-authoring/references/stage-taxonomy.md.
The active sequence is defined by existing root-level stage-YXX-slug/
folders and their Argo CD Applications.
Candidate flow for the reimplementation:
100 AI Platform Foundation
200 Production GenAI & Private Data
300 Agentic AI & Enterprise Integration
400 AI Operations, Evaluation & MLOps
Within each family, deploy lower stage identifiers before higher ones unless a
stage PLAN.md documents a different dependency.
Resource Configuration
| Node | GPUs | Active Model | Role |
|---|
| g6e.2xlarge | 1 L40S node, time-sliced to 4 units | nemotron-3-nano-30b-a3b (FP8 modelcar) in later stages | Private GenAI serving and benchmarking |
| External provider | 0 | OpenAI gpt-5.4-mini through MaaS using resource alias gpt-5-4-mini | Approved external model path when policy allows |
GitOps Deployment Pattern
Every future deploy.sh must apply an ArgoCD Application as its first action:
oc apply -f "$REPO_ROOT/gitops/argocd/app-of-apps/$STAGE_NAME.yaml"
Never apply manifests directly with oc apply -k for ArgoCD-managed resources.
Do not turn temporary live recovery for operator-managed pods into GitOps
patches for generated Deployments, generated image fields, copied CSVs, or
operand images.
Final E2E Validation
After the demo-flow script has been recreated and all stages pass:
./scripts/validate-demo-flow.sh
Evaluation Report Template
# RHOAI Demo Deployment Report
**Cluster:** <api-url> **Date:** <YYYY-MM-DD>
**GPU Config:** 1x g6e.2xlarge by default; 1 GPU per node
| Stage | Name | Status | Duration |
|------|------|--------|----------|
| 110 | RHOAI Base Platform | PASS | Xm |
| 120 | GPU-as-a-Service | PASS | Xm |
| ... | ... | ... | ... |
| E2E | Demo Flow | PASS | Xm |
Prerequisite Check
Before deploying to a fresh cluster, first read the Fresh Environment Checklist
in docs/OPERATIONS.md. Confirm local .env has the new environment's
KUBECONFIG, RHOAI_EXPECTED_API_SERVER, GIT_REPO_URL, GIT_REPO_BRANCH,
and required local credentials. Then run the prerequisite validation if that
skill helper is still present:
./.agents/skills/env-deploy-and-evaluate/scripts/validate-prerequisites.sh
Exit codes: 0 = ready, 1 = blocking failures, 2 = warnings only.
For per-stage deploy notes, known issues, and ArgoCD standards, read references/deploy-notes.md.