-
Notifications
You must be signed in to change notification settings - Fork 38
Closed
Labels
enhancementNew feature or requestNew feature or request
Description
Summary
Implement kubectl grove compare command to compare AIConfigurator predictions to actual runtime performance.
Goal
Show whether deployed workload meets planned performance targets.
Usage
kubectl grove compare <podcliqueset-name> [-n namespace]Expected Output
Plan vs Actual Comparison: my-inference
┌─────────────────────┬──────────────┬──────────────┬────────────┐
│ Metric │ Planned │ Actual │ Status │
├─────────────────────┼──────────────┼──────────────┼────────────┤
│ Throughput │ 804 tok/s/gpu│ 756 tok/s/gpu│ ⚠ -6% │
│ TTFT (p99) │ ≤300ms │ 342ms │ ❌ EXCEED │
│ PlacementScore │ 1.0 │ 0.72 │ ⚠ -28% │
└─────────────────────┴──────────────┴──────────────┴────────────┘
Diagnosis:
PlacementScore 0.72 → prefill pods split across 2 racks
Recommendations:
1. Reschedule prefill pods to consolidate on rack-1
Prerequisites
- P1: kubectl grove plan (for stored plan)
- P3: kubectl grove metrics (for actual metrics)
Priority
P3 - Metrics Integration
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
enhancementNew feature or requestNew feature or request