micsapp Vast.ai Enterprise RBAC
Overview
Control access to Vast.ai GPU cloud instances and spending through team billing and API key management. Vast.ai uses a marketplace model with per-GPU-hour pricing that varies by GPU type (RTX 4090 ~$0.20/hr, A100 ~$1.50/hr, H100 ~$3.00/hr).
Prerequisites
- Vast.ai account with team billing enabled
- API key from cloud.vast.ai
- Understanding of GPU pricing tiers on the Vast.ai marketplace
Instructions
Step 1: Create Scoped API Keys per Team
# Key for the ML training team (high-end GPUs, high budget)
vastai set api-key --name "ml-training-team" \
--spending-limit 5000 \ # 5000: 5 seconds in ms
--allowed-gpu-types "A100,H100" \
--max-instances 10
# Key for the inference team (cost-efficient GPUs)
vastai set api-key --name "inference-prod" \
--spending-limit 1000 \ # 1000: 1 second in ms
--allowed-gpu-types "RTX_4090,RTX_3090,A6000" \
--max-instances 20
Step 2: Implement GPU Provisioning Policies
// vastai-policy.ts - Enforce rules before provisioning
interface ProvisionPolicy {
allowedGpuTypes: string[];
maxPricePerHour: number;
maxInstances: number;
requireSpotInstance: boolean;
}
const TEAM_POLICIES: Record<string, ProvisionPolicy> = {
training: { allowedGpuTypes: ['A100', 'H100'], maxPricePerHour: 4.00, maxInstances: 10, requireSpotInstance: false },
inference: { allowedGpuTypes: ['RTX_4090', 'RTX_3090'], maxPricePerHour: 0.50, maxInstances: 20, requireSpotInstance: true },
research: { allowedGpuTypes: ['RTX_4090'], maxPricePerHour: 0.30, maxInstances: 3, requireSpotInstance: true },
};
Step 3: Set Spending Alerts
# Configure spending alerts via the Vast.ai CLI
vastai set spending-alert --threshold 1000 --email "ops@company.com" # 1000: 1 second in ms
vastai set spending-alert --threshold 4000 --email "ops@company.com,finance@company.com" # 4000: dev server port
vastai set auto-stop --daily-limit 500 # Auto-destroy instances if daily spend exceeds $500
Step 4: Monitor Active Instances and Costs
# List all running instances with cost data
vastai show instances --raw | jq '.[] | {
id, gpu_name, num_gpus,
cost_per_hr: .dph_total,
hours_running: ((.end_date // now) - .start_date) / 3600, # 3600: timeout: 1 hour
total_cost: .total_dph
}'
# Get team spending summary
vastai show invoices --last 30 | jq '.total_cost'
Step 5: Auto-Terminate Idle Instances
# Cron job: destroy instances idle for more than 2 hours
vastai show instances --raw | \
jq -r '.[] | select(.gpu_utilization < 5 and .duration > 7200) | .id' | \ # 7200: timeout: 2 hours
xargs -I{} vastai destroy instance {}
Error Handling
| Issue | Cause | Solution |
|---|---|---|
insufficient_funds |
Account balance depleted | Add credits or enable auto-recharge |
| Instance won't start | GPU type unavailable in region | Try different region or GPU type |
| Spending limit hit | Daily cap reached | Increase limit or wait for next day |
| SSH connection refused | Instance still initializing | Wait 2-3 minutes after creation |
Examples
Basic usage: Apply vastai enterprise rbac to a standard project setup with default configuration options.
Advanced scenario: Customize vastai enterprise rbac for production environments with multiple constraints and team-specific requirements.
Output
- Configuration files or code changes applied to the project
- Validation report confirming correct implementation
- Summary of changes made and their rationale
Resources
- Official Vastai Enterprise Rbac documentation
- Community best practices and patterns
- Related skills in this plugin pack