Documentation

Using Helm

The recommended way to deploy the Compute Gardener Scheduler is using Helm:

# Add the Helm repository
helm repo add compute-gardener https://elevated-systems.github.io/compute-gardener-scheduler
helm repo update

# Install the chart
helm install compute-gardener-scheduler compute-gardener/compute-gardener-scheduler \
  --namespace kube-system \
  --set carbonAware.electricityMap.apiKey=YOUR_API_KEY

Using YAML Manifests

Alternatively, you can deploy using the provided YAML manifests:

# First, update the API key in the manifest
# Then apply the manifest
kubectl apply -f manifests/compute-gardener-scheduler/compute-gardener-scheduler.yaml

Metrics Server: Highly recommended but not strictly required. Without Metrics Server, the scheduler won't be able to collect real-time node utilization data, resulting in less accurate energy usage estimates. Core carbon-aware and price-aware scheduling will still function using requested resources rather than actual usage.

Prometheus: Highly recommended but not strictly required. Without Prometheus, you won't be able to visualize scheduler performance metrics or validate carbon/cost savings. The scheduler will continue to function, but you'll miss valuable insights into its operation and won't have visibility into the actual emissions and cost reductions achieved.

The scheduler uses hardware-specific power profiles to accurately estimate and optimize energy consumption:

# Global PUE defaults
# Default datacenter PUE (typical range: 1.1-1.6)
defaultPUE: 1.1

# Default GPU-specific PUE for power conversion losses
defaultGPUPUE: 1.15

# CPU power profiles
cpuProfiles:
  "Intel(R) Xeon(R) Platinum 8275CL":
    # Idle power in watts
    idlePower: 10.5
    # Max power in watts
    maxPower: 120.0
  "Intel(R) Core(TM) i5-6500 CPU @ 3.20GHz":
    idlePower: 5.0
    maxPower: 65.0

# GPU power profiles with workload type coefficients
gpuProfiles:
  "NVIDIA A100":
    # Idle power in watts
    idlePower: 25.0
    # Max power in watts at 100% utilization
    maxPower: 400.0
    # Power coefficients for different workload types
    workloadTypes:
      # Inference typically uses ~60% of max power at 100% utilization
      inference: 0.6
      # Training uses full power
      training: 1.0
      # Rendering uses ~90% of max power at 100% utilization
      rendering: 0.9
  "NVIDIA GeForce GTX 1660":
    idlePower: 7.0
    maxPower: 125.0
    workloadTypes:
      inference: 0.5
      training: 0.9
      rendering: 0.8

# Memory power profiles
memProfiles:
  "DDR4-2666 ECC":
    # Idle power per GB in watts
    idlePowerPerGB: 0.125
    # Max power per GB in watts at full utilization
    maxPowerPerGB: 0.375
    # Base power overhead in watts
    baseIdlePower: 1.0

# Cloud instance mappings to hardware components
cloudInstanceMapping:
  aws:
    "m5.large":
      cpuModel: "Intel(R) Xeon(R) Platinum 8175M"
      memoryType: "DDR4-2666 ECC"
      numCPUs: 2
      # in MB
      totalMemory: 8192

The hardware detection system works through multiple layers:

• Hardware Profile Database: A ConfigMap containing power profiles for CPUs, GPUs, and memory types
• Cloud Instance Detection: Automatically maps cloud instance types to hardware components
• Hybrid Cloud Hardware Detection: Uses node labels or runtime detection to identify hardware
• Auto Labeling: Scripts provided to automatically label nodes with hardware information

The scheduler exports the following Prometheus metrics:

• scheduler_compute_gardener_carbon_intensity: Current carbon intensity (gCO2eq/kWh) for a given region
• scheduler_compute_gardener_electricity_rate: Current electricity rate ($/kWh) for a given location
• scheduler_compute_gardener_scheduling_attempt_total: Number of attempts to schedule pods by result
• scheduler_compute_gardener_pod_scheduling_duration_seconds: Latency for scheduling attempts
• scheduler_compute_gardener_estimated_savings: Estimated savings from scheduling (carbon, cost)
• scheduler_compute_gardener_price_delay_total: Number of scheduling delays due to price thresholds
• scheduler_compute_gardener_carbon_delay_total: Number of scheduling delays due to carbon intensity thresholds
• scheduler_compute_gardener_node_cpu_usage_cores: CPU usage on nodes (baseline, current, final)
• scheduler_compute_gardener_node_memory_usage_bytes: Memory usage on nodes (baseline, current, final)
• scheduler_compute_gardener_node_gpu_usage: GPU utilization on nodes (baseline, current, final)
• scheduler_compute_gardener_node_power_estimate_watts: Estimated node power consumption based on detected hardware profiles (baseline, current, final)
• scheduler_compute_gardener_metrics_samples_stored: Number of pod metrics samples currently stored in cache
• scheduler_compute_gardener_metrics_cache_size: Number of pods being tracked in metrics cache
• scheduler_compute_gardener_job_energy_usage_kwh: Estimated energy usage for completed jobs
• scheduler_compute_gardener_job_carbon_emissions_grams: Estimated carbon emissions for completed jobs
• scheduler_compute_gardener_scheduling_efficiency: Scheduling efficiency metrics (carbon/cost improvements)

Metrics Collection

The deployment includes both a Service and ServiceMonitor for Prometheus integration:

# Service exposes the metrics endpoint
apiVersion: v1
kind: Service
metadata:
  name: compute-gardener-scheduler-metrics
  namespace: kube-system
  labels:
    component: scheduler
    tier: control-plane
spec:
  ports:
  - name: https
    port: 10259
    targetPort: 10259
    protocol: TCP
  selector:
    component: scheduler
    tier: control-plane

# ServiceMonitor configures Prometheus to scrape metrics
apiVersion: monitoring.coreos.com/v1
kind: ServiceMonitor
metadata:
  name: compute-gardener-scheduler-monitor
  # Adjust to your Prometheus namespace
  namespace: cattle-monitoring-system
spec:
  selector:
    matchLabels:
      component: scheduler
      tier: control-plane
  endpoints:
  - port: https 
    scheme: https
    path: /metrics
    interval: 30s

Additionally, the Pod template includes Prometheus annotations for environments that use annotation-based discovery:

annotations:
  prometheus.io/scrape: 'true'
  prometheus.io/port: '10259'
  prometheus.io/scheme: 'https'
  prometheus.io/path: '/metrics'