| name | rhoai-troubleshoot |
| metadata | {"author":"rhoai3-coding-demo","version":"1.1.0","platform-family":"rhoai","platform-baseline":"repo","ocp-baseline":"repo","skill-group":"Demo Environment"} |
| description | Diagnose and fix issues with the RHOAI demo deployment. Use when ANY deployment stage fails, validate.sh reports errors, pods are in CrashLoopBackOff/Pending/Error/ImagePullBackOff, ArgoCD shows OutOfSync or Degraded, operators are not installing, GPU nodes are not joining, InferenceServices are not Ready, or the user reports any problem with their OpenShift AI environment. Also use when the user asks "why isn't X working?" for any demo component. Do NOT use for resource scaling (use manage-resources skill).
|
RHOAI Troubleshooting
Structured diagnostic workflow for resolving issues with the RHOAI demo on OpenShift 4.20. The implemented manifests target OpenShift AI 3.4 with MaaS governance as the active model-access path.
When to Use
- A
validate.sh script reports failures
- A
deploy.sh script errors out or hangs
- Pods are stuck in CrashLoopBackOff, Pending, or Error
- ArgoCD shows OutOfSync, ComparisonError, or Degraded
- Operators are not installing or CSV not Succeeded
Instructions
Never guess. Every diagnosis must be backed by official documentation or observable cluster state.
Step 1: Run the Validation Script
./stages/NNN-<name>/validate.sh
Check exit code: 0 = pass, 1 = failures, 2 = warnings only.
Step 2: Consult Official Product Documentation
Use @RHOAI 3.3 and @OCP 4.20 docs in Cursor unless the manifest explicitly documents an early-access MaaS behavior. Focus on the relevant section:
| Steps | Doc Section |
|---|
| 01 | Installing and Uninstalling |
| GPU/NFD | Managing Resources |
| Model Serving | Deploying Models |
| Pipelines | Working with AI Pipelines |
| Model Registry | Enabling Model Registry |
| Evaluation | Evaluating AI Systems |
Step 3: Gather Cluster State
Run diagnostic commands from references/diagnostic-commands.md for the failing component.
Step 4: Match to Known Pattern
Check references/diagnostic-patterns.md for the symptom. Common patterns cover ArgoCD, operators, pods, and RHOAI-specific issues.
Step 5: Apply Fix and Verify
Execute the smallest change that fixes the issue. Re-run validate.sh to confirm.
Step 6: Update Knowledge Base
If this was a new issue, update docs/TROUBLESHOOTING.md with the symptom, cause, diagnostic commands, and recovery command. Keep stage READMEs educational.
Escalation Protocol
If unresolved after one diagnostic cycle:
- Document what was tried and observed
- Include exact error messages and
oc output
- Suggest manual checks
- Report to the user with full context