Test - Runtime Verification¶

Overview¶

The test command provides runtime verification tests for bazzite-ai installations. It validates that GPU access, services, containers, and network connectivity are working correctly.

Key Concept: These are runtime tests that run on an installed bazzite-ai system using ujust test. For development overlay management, see the /bazzite-ai-dev:overlay skill.

Quick Reference¶

Quick Tests¶

Individual Tests¶

Pod Testing (default INSTANCE=90)¶

K3d Cluster Testing (default INSTANCE=90)¶

Portainer + K3d Testing (default INSTANCE=91)¶

VM & Install Testing¶

Parameters¶

ujust test ACTION [SUBACTION] [OPTIONS...]

Quick Tests¶

Quick Test (~30s)¶

ujust test quick

Runs essential checks:

1. GPU detection
2. CDI configuration
3. Running services status

Use when: Quick sanity check after installation or reboot.

Full Test Suite (~2min)¶

ujust test all

Runs comprehensive tests:

1. GPU and CUDA validation
2. PyTorch in container
3. All service health checks
4. Network connectivity

Use when: Full validation after major changes.

GPU Testing¶

GPU Detection¶

ujust test gpu

Checks:

• GPU vendor detection (NVIDIA, AMD, Intel)
• Driver loaded
• CDI configuration present

CUDA Test¶

ujust test cuda

Runs CUDA tests inside nvidia container:

• nvidia-smi output
• CUDA version
• GPU memory info

PyTorch Test¶

ujust test pytorch

Runs PyTorch GPU tests inside jupyter container:

• torch.cuda.is_available()
• GPU tensor operations
• Memory allocation

Pod Lifecycle Testing¶

Test pods use isolated instances (default: 90) to avoid interfering with user configurations.

Full Pod Lifecycle¶

ujust test pods all

Runs complete lifecycle:

1. config - Configure test pods
2. start - Start all pods
3. status - Verify running
4. stop - Stop all pods
5. delete - Clean up configs

Custom Instance¶

# Use different instance number
ujust test pods all INSTANCE=50

K3d Cluster Testing¶

Tests k3d Kubernetes cluster functionality.

Full K3d Lifecycle¶

ujust test k3d all

Tests:

1. Cluster creation on bazzite-ai network
2. Node health
3. GPU support (if NVIDIA)
4. Network connectivity to other pods
5. Ollama inference from k8s
6. Cleanup

Network Connectivity¶

ujust test k3d network

Verifies k8s pods can reach bazzite-ai network services (ollama, jupyter, etc.)

Common Workflows¶

After Installation¶

# Quick sanity check
ujust test quick

# If issues, run full suite
ujust test all

GPU Troubleshooting¶

# Check GPU detection
ujust test gpu

# Test CUDA
ujust test cuda

# Test PyTorch
ujust test pytorch

Service Validation¶

# Check all services
ujust test services

# Individual service
ujust test ollama
ujust test jupyter

Before Development¶

# Full validation
ujust test all

# Enable overlay mode for development
just overlay refresh

Troubleshooting¶

GPU Test Fails¶

Symptom: ujust test gpu reports no GPU

Check:

# Host GPU
nvidia-smi  # or lspci | grep -i vga

# CDI configuration
ls /etc/cdi/

Fix:

# Regenerate CDI
ujust config gpu setup

CUDA Test Fails¶

Symptom: CUDA not available in container

Cause: CDI not configured or driver mismatch

Fix:

# Rebuild CDI spec
sudo nvidia-ctk cdi generate --output=/etc/cdi/nvidia.yaml

Service Test Fails¶

Symptom: Service reports unhealthy

Check:

# Service status
systemctl --user status <service>

# Logs
journalctl --user -u <service> -n 50

Pod Test Cleanup Failed¶

Symptom: Test pods still exist after failure

Fix:

# Manual cleanup
ujust test pods delete INSTANCE=90

Cross-References¶

• Overlay Development: /bazzite-ai-dev:overlay skill for development mode
• GPU Setup: ujust config gpu setup for GPU configuration
• Services: Individual service skills for detailed management