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: kubernetes-dra-gpu-partitioning-protocol

Kubernetes Dynamic Resource Allocation (DRA) GPU Partitioning Protocol

Enforces Kubernetes Dynamic Resource Allocation (DRA) v1beta1/v1alpha3 structured parameters (resource.k8s.io) for multi-tenant GPU hardware slicing. Eliminates legacy integer-based overcommit, uncontrolled memory contention, and driver-level preemption by validating deterministic hardware claim bindings, Common Expression Language (CEL) selectors, and device class partitioning.

Prohibited Anti-Pattern: Deploying fractional or partitioned GPU workloads via legacy extended resources or manual environment injections without DRA Structured Parameters

Export Ready-to-Drop Configuration

Select format below to copy code or download file directly.

---
description: Kubernetes Dynamic Resource Allocation (DRA) GPU Partitioning Protocol - Enforces Kubernetes Dynamic Resource Allocation (DRA) v1beta1/v1alpha3 structured parameters (resource.k8s.io) for multi-tenant GPU hardware slicing. Eliminates legacy integer-based overcommit, uncontrolled memory contention, and driver-level preemption by validating deterministic hardware claim bindings, Common Expression Language (CEL) selectors, and device class partitioning.
globs: deploy/**/*.{ts,js,yaml,yml}, infra/**/*.{ts,js,yaml,yml}, src/k8s/**/*.{ts,js,yaml,yml}
alwaysApply: true
verified_by: "agentrules.org"
license: "CC-BY-4.0"
source: "https://agentrules.org/rules/kubernetes-dra-gpu-partitioning-protocol"
---

# Kubernetes Dynamic Resource Allocation (DRA) GPU Partitioning Protocol

> **Severity**: CRITICAL | **Category**: devops
> **Anti-Pattern Warning**: Deploying fractional or partitioned GPU workloads via legacy extended resources or manual environment injections without DRA Structured Parameters

## System Constraints
You are a Principal Cloud-Native Hardware Infrastructure Architect. Enforce deterministic, production-grade Kubernetes Dynamic Resource Allocation (DRA) protocols for partitioned accelerator workloads. Reject legacy extended resource time-slicing (e.g., nvidia.com/gpu integer limits with manual CUDA_VISIBLE_DEVICES overrides) and mandate structured CEL selectors, isolated MIG/hardware profile claims, and WaitForFirstConsumer claim templates.

## Strict Guidelines
- Engineers MUST declare sliced GPU allocations using `resource.k8s.io/v1beta1` `ResourceClaim` or `ResourceClaimTemplate` objects instead of opaque extended resource limits (e.g., `nvidia.com/gpu`).
- Configurations MUST define Common Expression Language (CEL) rules within `device.selectors[].cel` to deterministically match explicit hardware capabilities, slice sizes, and memory boundaries.
- Engineers MUST specify `allocationMode: ExactCount` and enforce `WaitForFirstConsumer` in `ResourceClaimTemplate` to prevent cross-NUMA socket fragmentation and scheduling deadlocks.
- Engineers MUST NOT permit multi-tenant workloads to rely on cooperative software-only time-slicing without hardware-enforced memory isolation (e.g., strict Multi-Instance GPU partitioning).
- Container specifications MUST explicitly reference the claim using `resources.claims` to ensure kubelet DRA runtime plugins inject verified device cgroup boundaries.

## Bad vs Good Implementations
### CDK8s/TypeScript: Partitioned GPU Claim via DRA Structured Parameters

#### Bad Antipattern (AI Common Hallucination)
> **Why it fails**: Legacy extended resource requests coupled with unverified environment variable flags bypass cluster-wide hardware topology checks, allow cross-tenant memory exhaustion, and risk driver-level kernel preemption under burst loads.
```typescript
import { ApiObject } from 'cdk8s';
import { Construct } from 'constructs';

export class UnsafeGpuWorkload extends Construct {
  constructor(scope: Construct, id: string) {
    super(scope, id);

    // CRITICAL ANTI-PATTERN: Opaque integer resource request with cooperative
    // time-slicing environment variables, prone to noisy-neighbor memory panics.
    new ApiObject(this, 'UnsafePod', {
      apiVersion: 'v1',
      kind: 'Pod',
      metadata: { name: 'llm-sliced-inference' },
      spec: {
        containers: [{
          name: 'inference-engine',
          image: 'vllm/vllm-openai:latest',
          resources: {
            limits: { 'nvidia.com/gpu': '1' } // Unpartitioned whole-device request
          },
          env: [
            { name: 'NVIDIA_VISIBLE_DEVICES', value: '0' },
            { name: 'CUDA_MPS_ACTIVE_THREAD_PERCENTAGE', value: '50' }
          ]
        }]
      }
    });
  }
}
```

#### Production Standard (Correct)
> **Standard**: Leverages the Kubernetes DRA v1beta1 framework with CEL expressions and exact hardware slice profiles, ensuring deterministic hardware isolation, NUMA alignment, and zero cross-tenant memory contention.
```typescript
import { ApiObject } from 'cdk8s';
import { Construct } from 'constructs';

export class PartitionedGpuWorkload extends Construct {
  constructor(scope: Construct, id: string) {
    super(scope, id);

    // PRODUCTION INVARIANT: Structured DRA claim with deterministic CEL validation
    // guarantees strict hardware-level memory and compute boundary separation.
    new ApiObject(this, 'SecureDraPod', {
      apiVersion: 'v1',
      kind: 'Pod',
      metadata: { name: 'llm-sliced-inference' },
      spec: {
        resourceClaims: [{
          name: 'isolated-gpu',
          resourceClaimTemplateName: 'gpu-mig-slice-claim'
        }],
        containers: [{
          name: 'inference-engine',
          image: 'vllm/vllm-openai:latest',
          resources: {
            claims: [{ name: 'isolated-gpu' }]
          }
        }]
      }
    });

    new ApiObject(this, 'GpuClaimTemplate', {
      apiVersion: 'resource.k8s.io/v1beta1',
      kind: 'ResourceClaimTemplate',
      metadata: { name: 'gpu-mig-slice-claim' },
      spec: {
        spec: {
          devices: {
            requests: [{
              name: 'mig-partition',
              deviceClassName: 'gpu.nvidia.com',
              allocationMode: 'ExactCount',
              count: 1,
              selectors: [{
                cel: {
                  expression: "device.attributes['gpu.nvidia.com'].profile == '3g.40gb' && device.capacity['gpu.nvidia.com'].memory >= quantity('40Gi')"
                }
              }]
            }]
          }
        }
      }
    });
  }
}
```
Recommended MCP Servers (mcp-serverlist.com)

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

Kubernetes Cluster Inspector MCP
Inspects cluster DeviceClasses, ResourceClaims, and nodes for Dynamic Resource Allocation readiness.
Recommended Agent Skills (askills.dev)

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

Kubectl DRA Resource Validator
CLI skill to query `resource.k8s.io` claim statuses, device capacities, and driver allocations directly from the API server.
Architectural Specification & Invariant Rationale

Architectural Invariant Design: Kubernetes Dynamic Resource Allocation (DRA) GPU Partitioning Protocol

Kubernetes Dynamic Resource Allocation (DRA) GPU Partitioning Protocol defines an industrial-grade behavioral standard for devops engineering. Enforces Kubernetes Dynamic Resource Allocation (DRA) v1beta1/v1alpha3 structured parameters (resource.k8s.io) for multi-tenant GPU hardware slicing. Eliminates legacy integer-based overcommit, uncontrolled memory contention, and driver-level preemption by validating deterministic hardware claim bindings, Common Expression Language (CEL) selectors, and device class partitioning. 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 fractional or partitioned GPU workloads via legacy extended resources or manual environment injections without DRA Structured Parameters 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 fractional or partitioned GPU workloads via legacy extended resources or manual environment injections without DRA Structured ParametersCritical
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 declare sliced GPU allocations using `resource.k8s.io/v1beta1` `ResourceClaim` or `ResourceClaimTemplate` objects instead of opaque extended resource limits (e.g., `nvidia.com/gpu`).

Engineers MUST declare sliced GPU allocations using `resource.k8s.io/v1beta1` `ResourceClaim` or `ResourceClaimTemplate` objects instead of opaque extended resource limits (e.g., `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

Configurations MUST define Common Expression Language (CEL) rules within `device.selectors[].cel` to deterministically match explicit hardware capabilities, slice sizes, and memory boundaries.

Configurations MUST define Common Expression Language (CEL) rules within `device.selectors[].cel` to deterministically match explicit hardware capabilities, slice sizes, and memory boundaries.

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

Engineers MUST specify `allocationMode

Engineers MUST specify `allocationMode: ExactCount` and enforce `WaitForFirstConsumer` in `ResourceClaimTemplate` to prevent cross-NUMA socket fragmentation and scheduling deadlocks.

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

Engineers MUST NOT permit multi-tenant workloads to rely on cooperative software-only time-slicing without hardware-enforced memory isolation (e.g., strict Multi-Instance GPU partitioning).

Engineers MUST NOT permit multi-tenant workloads to rely on cooperative software-only time-slicing without hardware-enforced memory isolation (e.g., strict Multi-Instance GPU partitioning).

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

Container specifications MUST explicitly reference the claim using `resources.claims` to ensure kubelet DRA runtime plugins inject verified device cgroup boundaries.

Container specifications MUST explicitly reference the claim using `resources.claims` to ensure kubelet DRA runtime plugins inject verified device cgroup boundaries.

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) GPU Partitioning Protocol critical for AI coding assistants?

Enforces Kubernetes Dynamic Resource Allocation (DRA) v1beta1/v1alpha3 structured parameters (resource.k8s.io) for multi-tenant GPU hardware slicing. Eliminates legacy integer-based overcommit, uncontrolled memory contention, and driver-level preemption by validating deterministic hardware claim bindings, Common Expression Language (CEL) selectors, and device class partitioning. 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 fractional or partitioned GPU workloads via legacy extended resources or manual environment injections without DRA Structured Parameters) 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