AgentRules.org
RulesStacksClientsAuditorPrunerSelf-HealRFCsBenchmarkCLICI/CDAuthorsSubmit
This specification is maintained by the AgentRules Working Group. Have enhancements or want to claim maintainership?Submit Specification →
devopscritical PriorityID: k8s-dra-cdi-gpu-slice-isolation-standard

Kubernetes Dynamic Resource Allocation (DRA) CDI GPU Slice Isolation Standard

Enforces strict hardware-level noisy-neighbor isolation for multi-tenant inference workloads by migrating from legacy Kubernetes device plugins to Dynamic Resource Allocation (DRA) using Container Device Interface (CDI) resource claims. Mandates explicit slice partitioning (such as NVIDIA MIG profiles), device-level resource claim templates, and isolated CDI container specs to eliminate spatial and temporal contention across co-located tenants.

Prohibited Anti-Pattern: Deploying Multi-Tenant GPU Workloads via Legacy Device Plugins and Host-Shared Device Nodes

Export Ready-to-Drop Configuration

Select format below to copy code or download file directly.

---
description: Kubernetes Dynamic Resource Allocation (DRA) CDI GPU Slice Isolation Standard - Enforces strict hardware-level noisy-neighbor isolation for multi-tenant inference workloads by migrating from legacy Kubernetes device plugins to Dynamic Resource Allocation (DRA) using Container Device Interface (CDI) resource claims. Mandates explicit slice partitioning (such as NVIDIA MIG profiles), device-level resource claim templates, and isolated CDI container specs to eliminate spatial and temporal contention across co-located tenants.
globs: deploy/**/*.yaml, k8s/**/*.yaml, helm/**/*.yaml, templates/**/*.yaml, infra/**/*.ts
alwaysApply: true
verified_by: "agentrules.org"
license: "CC-BY-4.0"
source: "https://agentrules.org/rules/k8s-dra-cdi-gpu-slice-isolation-standard"
---

# Kubernetes Dynamic Resource Allocation (DRA) CDI GPU Slice Isolation Standard

> **Severity**: CRITICAL | **Category**: devops
> **Anti-Pattern Warning**: Deploying Multi-Tenant GPU Workloads via Legacy Device Plugins and Host-Shared Device Nodes

## System Constraints
You are a Principal Software Architect enforcing Kubernetes Dynamic Resource Allocation (DRA) and Container Device Interface (CDI) GPU slice isolation for mission-critical multi-tenant AI/ML inference infrastructure. You must reject legacy device plugin allocations (`nvidia.com/gpu` or arbitrary time-slicing) and enforce fine-grained hardware-isolated MIG/CDI resource claims with strict memory and compute bandwidth isolation boundaries.

## Strict Guidelines
- Engineers MUST ALWAYS provision GPU slices using Kubernetes Dynamic Resource Allocation (`resource.k8s.io/v1beta1` or `resource.k8s.io/v1alpha3`) `ResourceClaimTemplate` objects instead of legacy `resources.limits['nvidia.com/gpu']`.
- Engineers MUST NOT deploy multi-tenant inference workloads on unpartitioned GPUs using software time-slicing; hardware-level spatial isolation (e.g., MIG 1g.10gb, 2g.20gb) MUST be declared via structured CDI device selectors.
- Pods MUST declare `resourceClaims` at the pod spec level and map them directly into container `resources.claims` targeting specific CDI slice identifiers.
- Components NEVER mount raw `/dev/nvidia*` host paths, set privileged security contexts, or bypass CDI driver injection routines.
- Services MUST declare explicit DRA device selection matchers specifying GPU architecture, interconnect topology, and maximum VRAM allocation limits per tenant claim.
- Engineers MUST ensure DRA `DeviceClass` and CDI configuration enforce exclusive device instance binding to prevent concurrent cross-tenant context execution.

## Bad vs Good Implementations
### Kubernetes GPU Claim Migration: Legacy Device Plugin to DRA CDI MIG Slice

#### Bad Antipattern (AI Common Hallucination)
> **Why it fails**: Using legacy device plugins (`nvidia.com/gpu`) without Dynamic Resource Allocation allows unpartitioned GPU execution where high-load tenants starve co-located tenants of PCIe bandwidth, L2 cache, and SM execution cycles. Privileged access bypasses Container Device Interface (CDI) isolation entirely.
```yaml
# Anti-Pattern: Unpartitioned GPU sharing via legacy device plugin
apiVersion: v1
kind: Pod
metadata:
  name: tenant-inference-llm
  namespace: tenant-alpha
spec:
  containers:
  - name: inference-engine
    image: vllm/vllm-openai:latest
    resources:
      limits:
        nvidia.com/gpu: "1" # Legacy device plugin; suffers noisy-neighbor memory and compute contention
    securityContext:
      privileged: true # Dangerous bypass to access device nodes
```

#### Production Standard (Correct)
> **Standard**: Utilizes Dynamic Resource Allocation (DRA) with CEL-based device selectors matching an exact CDI-managed NVIDIA Multi-Instance GPU (MIG) profile (1g.10gb) on an H100 GPU. Guarantees hardware-level memory protection, private SM allocation, and zero noisy-neighbor cross-contamination between tenant workloads.
```yaml
# Production Standard: Strict hardware isolation via DRA and CDI ResourceClaims
apiVersion: v1
kind: Pod
metadata:
  name: tenant-inference-llm
  namespace: tenant-alpha
spec:
  containers:
  - name: inference-engine
    image: vllm/vllm-openai:latest
    resources:
      claims:
      - name: dedicated-gpu-slice
    securityContext:
      allowPrivilegeEscalation: false
      capabilities:
        drop: ["ALL"]
      readOnlyRootFilesystem: true
  resourceClaims:
  - name: dedicated-gpu-slice
    resourceClaimTemplateName: isolated-mig-1g10gb-template
---
apiVersion: resource.k8s.io/v1beta1
kind: ResourceClaimTemplate
metadata:
  name: isolated-mig-1g10gb-template
  namespace: tenant-alpha
spec:
  spec:
    devices:
      requests:
      - name: mig-slice
        deviceClassName: gpu.nvidia.com
        selectors:
        - cel:
            expression: >-
              device.attributes['gpu.nvidia.com'].model == 'NVIDIA-H100-80GB-HBM3' &&
              device.attributes['gpu.nvidia.com'].migProfile == '1g.10gb'
        allocationMode: ExactCount
        count: 1
```
Recommended MCP Servers (mcp-serverlist.com)

Equip agent with system tools to verify code for this stack.

Kubernetes Resource & Cluster MCP Server
Inspects active DRA ResourceClasses, ResourceClaims, and CDI device configurations in real-time.
Recommended Agent Skills (askills.dev)

Ready-to-use executable skill functions tailored to this stack.

Kubectl DRA CDI Plugin CLI
Validates, describes, and benchmarks Dynamic Resource Allocation claims and CDI hardware slice topology.
Architectural Specification & Invariant Rationale

Architectural Invariant Design: Kubernetes Dynamic Resource Allocation (DRA) CDI GPU Slice Isolation Standard

Kubernetes Dynamic Resource Allocation (DRA) CDI GPU Slice Isolation Standard defines an industrial-grade behavioral standard for devops engineering. Enforces strict hardware-level noisy-neighbor isolation for multi-tenant inference workloads by migrating from legacy Kubernetes device plugins to Dynamic Resource Allocation (DRA) using Container Device Interface (CDI) resource claims. Mandates explicit slice partitioning (such as NVIDIA MIG profiles), device-level resource claim templates, and isolated CDI container specs to eliminate spatial and temporal contention across co-located tenants. Modern software architectures require strict adherence to deterministic patterns to guarantee high availability, maintainability, and zero security vulnerabilities.

Framework Paradigm Shift & Drift MechanicsLLMs are trained on millions of open-source repositories containing outdated code, legacy idioms, and unpatched antipatterns. When coding assistants generate implementations for this stack, they default to common statistical tokens: Deploying Multi-Tenant GPU Workloads via Legacy Device Plugins and Host-Shared Device Nodes Without an explicit system instruction file, developers spend hours debugging silent failures.
Failure Modes & Production Blast Radius

Hazard Analysis: What Breaks Without This Rule

The following failure modes have been cataloged from hundreds of hours of production code reviews where AI assistants produced non-compliant implementations:

Failure ModeAI Symptom (The Anti-Pattern)SeverityProduction Impact & Remedy
AI Hallucination & Deprecation DriftDeploying Multi-Tenant GPU Workloads via Legacy Device Plugins and Host-Shared Device NodesCritical
Violating this industrial standard introduces regression bugs, security risks, and unexpected runtime failures in production.
Enforce explicit behavioral system directives across all AI coding assistants in your repository.
Silent Architecture InconsistencyAI assistants produce mixed paradigms across files, creating fragmented codebase conventions.Medium
Increases cognitive load during code reviews and leads to maintenance debt across engineering teams.
Pin project conventions in .cursor/rules, CLAUDE.md, and AGENTS.md so all team agents follow identical guardrails.
Architectural Directives in Detail

Step-by-Step Production Directives

Each rule constraint below is injected into the AI agent's system prompt to establish deterministic boundaries:

1

Engineers MUST ALWAYS provision GPU slices using Kubernetes Dynamic Resource Allocation (`resource.k8s.io/v1beta1` or `resource.k8s.io/v1alpha3`) `ResourceClaimTemplate` objects instead of legacy `resources.limits['nvidia.com/gpu']`.

Engineers MUST ALWAYS provision GPU slices using Kubernetes Dynamic Resource Allocation (`resource.k8s.io/v1beta1` or `resource.k8s.io/v1alpha3`) `ResourceClaimTemplate` objects instead of legacy `resources.limits['nvidia.com/gpu']`.

Rationale:AI models frequently generate unconstrained code here because statistical language distributions lack project-specific architectural context. Mandating this directive ensures that generated code adheres to verified industrial production requirements.
Enforcement:Integrated into the Agent Pack rule file. When Cursor, Claude Code, or Codex processes this prompt, it applies this constraint before emitting syntax.
2

Engineers MUST NOT deploy multi-tenant inference workloads on unpartitioned GPUs using software time-slicing

Engineers MUST NOT deploy multi-tenant inference workloads on unpartitioned GPUs using software time-slicing; hardware-level spatial isolation (e.g., MIG 1g.10gb, 2g.20gb) MUST be declared via structured CDI device selectors.

Rationale:AI models frequently generate unconstrained code here because statistical language distributions lack project-specific architectural context. Mandating this directive ensures that generated code adheres to verified industrial production requirements.
Enforcement:Integrated into the Agent Pack rule file. When Cursor, Claude Code, or Codex processes this prompt, it applies this constraint before emitting syntax.
3

Pods MUST declare `resourceClaims` at the pod spec level and map them directly into container `resources.claims` targeting specific CDI slice identifiers.

Pods MUST declare `resourceClaims` at the pod spec level and map them directly into container `resources.claims` targeting specific CDI slice identifiers.

Rationale:AI models frequently generate unconstrained code here because statistical language distributions lack project-specific architectural context. Mandating this directive ensures that generated code adheres to verified industrial production requirements.
Enforcement:Integrated into the Agent Pack rule file. When Cursor, Claude Code, or Codex processes this prompt, it applies this constraint before emitting syntax.
4

Components NEVER mount raw `/dev/nvidia*` host paths, set privileged security contexts, or bypass CDI driver injection routines.

Components NEVER mount raw `/dev/nvidia*` host paths, set privileged security contexts, or bypass CDI driver injection routines.

Rationale:AI models frequently generate unconstrained code here because statistical language distributions lack project-specific architectural context. Mandating this directive ensures that generated code adheres to verified industrial production requirements.
Enforcement:Integrated into the Agent Pack rule file. When Cursor, Claude Code, or Codex processes this prompt, it applies this constraint before emitting syntax.
5

Services MUST declare explicit DRA device selection matchers specifying GPU architecture, interconnect topology, and maximum VRAM allocation limits per tenant claim.

Services MUST declare explicit DRA device selection matchers specifying GPU architecture, interconnect topology, and maximum VRAM allocation limits per tenant claim.

Rationale:AI models frequently generate unconstrained code here because statistical language distributions lack project-specific architectural context. Mandating this directive ensures that generated code adheres to verified industrial production requirements.
Enforcement:Integrated into the Agent Pack rule file. When Cursor, Claude Code, or Codex processes this prompt, it applies this constraint before emitting syntax.
6

Engineers MUST ensure DRA `DeviceClass` and CDI configuration enforce exclusive device instance binding to prevent concurrent cross-tenant context execution.

Engineers MUST ensure DRA `DeviceClass` and CDI configuration enforce exclusive device instance binding to prevent concurrent cross-tenant context execution.

Rationale:AI models frequently generate unconstrained code here because statistical language distributions lack project-specific architectural context. Mandating this directive ensures that generated code adheres to verified industrial production requirements.
Enforcement:Integrated into the Agent Pack rule file. When Cursor, Claude Code, or Codex processes this prompt, it applies this constraint before emitting syntax.
Technical Questions & Answers

Frequently Asked Technical Questions

Common questions developers face when configuring AI coding assistants for this stack:

Q:Why is Kubernetes Dynamic Resource Allocation (DRA) CDI GPU Slice Isolation Standard critical for AI coding assistants?

Enforces strict hardware-level noisy-neighbor isolation for multi-tenant inference workloads by migrating from legacy Kubernetes device plugins to Dynamic Resource Allocation (DRA) using Container Device Interface (CDI) resource claims. Mandates explicit slice partitioning (such as NVIDIA MIG profiles), device-level resource claim templates, and isolated CDI container specs to eliminate spatial and temporal contention across co-located tenants. Without explicit behavioral boundaries, AI coding assistants (Cursor, Claude Code, Windsurf, Codex) rely on training data averages, which frequently produce deprecated, inefficient, or vulnerable code patterns.

Q:How does adopting this rule prevent AI hallucinations?

This rule explicitly declares what the AI must avoid (Deploying Multi-Tenant GPU Workloads via Legacy Device Plugins and Host-Shared Device Nodes) and prescribes verified positive examples. Grounding the LLM with negative and positive pairs eliminates guesswork.

Q:Which AI coding environments are supported?

This specification is multi-platform compatible: Cursor (.cursor/rules/*.mdc & .cursorrules), Anthropic Claude Code (CLAUDE.md), Windsurf (.windsurfrules), Codex / Cline (AGENTS.md), and GitHub Copilot (.github/copilot-instructions.md).

Q:Can I combine this rule with other technology stacks?

Yes. Use the Interactive Stack Builder or download an Agent Pack to merge this rule with other frontend, backend, database, security, and DevOps rules without syntax collision.

Open Specification · CC BY 4.0·Free for Personal, Team & Commercial In-Repo Use

Citation & Open Source Attribution Guidelines

This specification is licensed under Creative Commons Attribution 4.0 International (CC BY 4.0). You are completely free to adapt and share this rule in your workspace. If citing, including in public AI benchmark datasets, or redistributing, attribution to AgentRules.org is required.

Related DEVOPS Rules

View all rules →
devops
86

Kubernetes Gateway API v1 HTTPRoute Canary and ReferenceGrant Invariants

Enforces strict conformance to Kubernetes Gateway API v1 specifications during migration from legacy Ingress controllers. Mandates explicit ReferenceGrant resources for cross-namespace routing, strict relative integer weighting for HTTPRoute canary releases, and explicit backendRef port declarations.

devops
118

ArgoCD ApplicationSet Matrix Generator Invariants & Cascade Deletion Guards

Enforces strict structural constraints on ArgoCD ApplicationSet Matrix generators combining cluster selectors and Git directories. Prevents catastrophic multi-cluster cascade pruning, accidental secret exposure through parameter interpolation, and runaway Application CR explosion via explicit sync policies, preservation finalizers, and parameter validation.

devops
40

Karpenter Resilient Topology, Interruption Queue, and Consolidation Budget Standard

Enforces deterministic disruption budgets, multi-instance-type diversification, and event-driven SQS interruption queue handling across Karpenter NodePools and EC2NodeClasses. Eliminates cascading eviction loops, instance starvation, and availability downtime during AWS Spot rebalances and underutilized node consolidation.

1 Rule in Stack

Kubernetes