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-deterministic-device-claims

Kubernetes DRA Deterministic Device Claim and Topology Preservation Standard

Enforces deterministic DeviceClass constraints, strict CEL selector evaluation, and pod-scoped lifecycle management for Dynamic Resource Allocation (DRA) in Kubernetes AI inference infrastructure. Eliminates GPU memory fragmentation, non-NVLink interconnect allocations, and orphaned hardware slices across heterogeneous accelerator nodes.

Prohibited Anti-Pattern: Ad-Hoc Unbounded Device Slicing and Standalone Unmanaged Resource Claims

Export Ready-to-Drop Configuration

Select format below to copy code or download file directly.

---
description: Kubernetes DRA Deterministic Device Claim and Topology Preservation Standard - Enforces deterministic DeviceClass constraints, strict CEL selector evaluation, and pod-scoped lifecycle management for Dynamic Resource Allocation (DRA) in Kubernetes AI inference infrastructure. Eliminates GPU memory fragmentation, non-NVLink interconnect allocations, and orphaned hardware slices across heterogeneous accelerator nodes.
globs: deployments/**/*.yaml, k8s/**/*.yaml, helm/**/*.yaml, manifests/**/*.yaml
alwaysApply: true
verified_by: "agentrules.org"
license: "CC-BY-4.0"
source: "https://agentrules.org/rules/kubernetes-dra-deterministic-device-claims"
---

# Kubernetes DRA Deterministic Device Claim and Topology Preservation Standard

> **Severity**: CRITICAL | **Category**: devops
> **Anti-Pattern Warning**: Ad-Hoc Unbounded Device Slicing and Standalone Unmanaged Resource Claims

## System Constraints
You are a Principal Infrastructure Architect enforcing the Kubernetes Dynamic Resource Allocation (DRA) Deterministic Device Claim Standard. Mandate that all accelerator allocations utilize resource.k8s.io specifications with strict CEL device selectors, topology-aware constraints, and Pod-managed ResourceClaimTemplates to guarantee deterministic slicing and zero-leak cleanup lifecycles.

## Strict Guidelines
- Engineers MUST ALWAYS define accelerator claims using pod-level `spec.resourceClaimTemplates` rather than standalone cluster-scoped `ResourceClaim` manifests to enforce synchronous pod lifecycle teardown.
- Engineers MUST ALWAYS enforce strict Common Expression Language (CEL) selector rules on DeviceClasses and Claims to explicitly evaluate NUMA alignment, interconnect bandwidth (e.g., NVLink domain), and compute slice dimensions.
- Engineers MUST NOT rely on implicit default device drivers or legacy `resources.requests["nvidia.com/gpu"]` when orchestrating heterogeneous GPU topologies under DRA.
- Device claims MUST specify `count` and topology policies explicitly; multi-GPU allocations MUST mandate `allInSame: ["nvlink-domain"]` or equivalent interconnect boundaries.
- ResourceClaimTemplates MUST NEVER declare persistent allocation lifecycles for ephemeral model inference workloads to prevent resource deadlocks and slice fragmentation.
- Operators and controllers MUST validate DRA driver CRDs against schema versions matching or exceeding Kubernetes 1.31 (resource.k8s.io/v1beta1).

## Bad vs Good Implementations
### Ad-Hoc Unmanaged Device Claim with Missing Interconnect Bounds (Anti-Pattern)

#### Bad Antipattern (AI Common Hallucination)
> **Why it fails**: Referencing an external standalone ResourceClaim decouples allocation lifecycle from the Pod, resulting in leaked resource claims upon pod failure. It also lacks CEL device filtering and topology affinity, risking allocation of disjoint GPUs lacking NVLink interconnect.
```yaml
apiVersion: v1
kind: Pod
metadata:
  name: llm-inference-worker
spec:
  # ANTI-PATTERN: References standalone claim not tied to Pod lifecycle
  # Missing strict CEL matchers and interconnect topology constraints
  resourceClaims:
    - name: gpu-claim
      resourceClaimName: shared-h100-cluster-claim
  containers:
    - name: vllm-engine
      image: vllm/vllm-openai:latest
      resources:
        claims:
          - name: gpu-claim
```

#### Production Standard (Correct)
> **Standard**: Uses an inline ResourceClaimTemplate scoped to the Pod's lifecycle for zero-leak teardown, enforces exact H100 hardware characteristics via CEL, and requires all 2 requested GPUs to reside within the same NVLink mesh domain to eliminate interconnect bottlenecks.
```yaml
apiVersion: v1
kind: Pod
metadata:
  name: llm-inference-worker
spec:
  restartPolicy: OnFailure
  containers:
    - name: vllm-engine
      image: vllm/vllm-openai:latest
      resources:
        claims:
          - name: gpu-claim
  # PRODUCTION INVARIANT: Pod-scoped template guarantees deterministic cleanup
  resourceClaims:
    - name: gpu-claim
      resourceClaimTemplateName: h100-nvlink-pair-template
---
apiVersion: resource.k8s.io/v1beta1
kind: ResourceClaimTemplate
metadata:
  name: h100-nvlink-pair-template
spec:
  spec:
    devices:
      requests:
        - name: gpu-request
          deviceClassName: gpu.nvidia.com
          count: 2
          selectors:
            - cel:
                expression: >-
                  device.attributes['gpu.nvidia.com'].productName.matches('^NVIDIA H100') &&
                  device.attributes['gpu.nvidia.com'].migCapable == false
          allInSame:
            - attribute: nvlink-mesh-id
```
Recommended MCP Servers (mcp-serverlist.com)

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

Kubernetes Resource Context Provider
Inspects running cluster DRA DeviceClasses, ResourceSlices, and active ResourceClaims to validate schema compliance and topology health.
Recommended Agent Skills (askills.dev)

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

Kubectl DRA CLI Extension
CLI tool to query, debug, and trace structured Dynamic Resource Allocation allocations, device slice state, and CEL selector matches.
Architectural Specification & Invariant Rationale

Architectural Invariant Design: Kubernetes DRA Deterministic Device Claim and Topology Preservation Standard

Kubernetes DRA Deterministic Device Claim and Topology Preservation Standard defines an industrial-grade behavioral standard for devops engineering. Enforces deterministic DeviceClass constraints, strict CEL selector evaluation, and pod-scoped lifecycle management for Dynamic Resource Allocation (DRA) in Kubernetes AI inference infrastructure. Eliminates GPU memory fragmentation, non-NVLink interconnect allocations, and orphaned hardware slices across heterogeneous accelerator nodes. 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: Ad-Hoc Unbounded Device Slicing and Standalone Unmanaged Resource Claims 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 DriftAd-Hoc Unbounded Device Slicing and Standalone Unmanaged Resource ClaimsCritical
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 define accelerator claims using pod-level `spec.resourceClaimTemplates` rather than standalone cluster-scoped `ResourceClaim` manifests to enforce synchronous pod lifecycle teardown.

Engineers MUST ALWAYS define accelerator claims using pod-level `spec.resourceClaimTemplates` rather than standalone cluster-scoped `ResourceClaim` manifests to enforce synchronous pod lifecycle teardown.

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 ALWAYS enforce strict Common Expression Language (CEL) selector rules on DeviceClasses and Claims to explicitly evaluate NUMA alignment, interconnect bandwidth (e.g., NVLink domain), and compute slice dimensions.

Engineers MUST ALWAYS enforce strict Common Expression Language (CEL) selector rules on DeviceClasses and Claims to explicitly evaluate NUMA alignment, interconnect bandwidth (e.g., NVLink domain), and compute slice dimensions.

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 NOT rely on implicit default device drivers or legacy `resources.requests["nvidia.com/gpu"]` when orchestrating heterogeneous GPU topologies under DRA.

Engineers MUST NOT rely on implicit default device drivers or legacy `resources.requests["nvidia.com/gpu"]` when orchestrating heterogeneous GPU topologies under DRA.

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

Device claims MUST specify `count` and topology policies explicitly

Device claims MUST specify `count` and topology policies explicitly; multi-GPU allocations MUST mandate `allInSame: ["nvlink-domain"]` or equivalent interconnect 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.
5

ResourceClaimTemplates MUST NEVER declare persistent allocation lifecycles for ephemeral model inference workloads to prevent resource deadlocks and slice fragmentation.

ResourceClaimTemplates MUST NEVER declare persistent allocation lifecycles for ephemeral model inference workloads to prevent resource deadlocks and slice fragmentation.

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

Operators and controllers MUST validate DRA driver CRDs against schema versions matching or exceeding Kubernetes 1.31 (resource.k8s.io/v1beta1).

Operators and controllers MUST validate DRA driver CRDs against schema versions matching or exceeding Kubernetes 1.31 (resource.k8s.io/v1beta1).

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 DRA Deterministic Device Claim and Topology Preservation Standard critical for AI coding assistants?

Enforces deterministic DeviceClass constraints, strict CEL selector evaluation, and pod-scoped lifecycle management for Dynamic Resource Allocation (DRA) in Kubernetes AI inference infrastructure. Eliminates GPU memory fragmentation, non-NVLink interconnect allocations, and orphaned hardware slices across heterogeneous accelerator nodes. 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 (Ad-Hoc Unbounded Device Slicing and Standalone Unmanaged Resource Claims) 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