跳转至
DEPRECATED: This documentation is deprecated and will be removed in a future version.
DEPRECATED

Kepler Metrics

This document describes the metrics exported by Kepler for monitoring energy consumption at various levels (node, container, process, VM, pod).

Overview

Kepler exports metrics in Prometheus format that can be scraped by Prometheus or other compatible monitoring systems. All metrics follow the Prometheus naming conventions and are prefixed with kepler_.

Metric Types

  • COUNTER: A cumulative metric that only increases over time (e.g., total energy consumption)
  • GAUGE: A metric that can increase and decrease (e.g., current power consumption)

Zone-Based Organization

Kepler organizes energy metrics by zones, which correspond to different hardware components:

  • package: CPU socket-level energy (includes cores and uncore components)
  • core: CPU core-level energy (available on some processors)
  • uncore: Uncore components (last-level cache, integrated GPU, memory controller)
  • dram: Memory energy consumption

Metrics Reference

Node Metrics

These metrics provide energy and power information at the node level, representing the total consumption of the entire system.

kepler_node_cpu_joules_total

  • Type: COUNTER
  • Description: Total energy consumption of CPU at node level in joules
  • Labels:
  • zone: Energy zone (package, core, uncore, dram)
  • path: Hardware sensor path
  • Constant Labels:
  • node_name: Name of the node

kepler_node_cpu_watts

  • Type: GAUGE
  • Description: Current power consumption of CPU at node level in watts
  • Labels:
  • zone: Energy zone (package, core, uncore, dram)
  • path: Hardware sensor path
  • Constant Labels:
  • node_name: Name of the node

kepler_node_cpu_active_joules_total

  • Type: COUNTER
  • Description: Energy consumption of CPU in active state at node level in joules
  • Labels:
  • zone: Energy zone (package, core, uncore, dram)
  • path: Hardware sensor path
  • Constant Labels:
  • node_name: Name of the node

kepler_node_cpu_active_watts

  • Type: GAUGE
  • Description: Power consumption of CPU in active state at node level in watts
  • Labels:
  • zone: Energy zone (package, core, uncore, dram)
  • path: Hardware sensor path
  • Constant Labels:
  • node_name: Name of the node

kepler_node_cpu_idle_joules_total

  • Type: COUNTER
  • Description: Energy consumption of CPU in idle state at node level in joules
  • Labels:
  • zone: Energy zone (package, core, uncore, dram)
  • path: Hardware sensor path
  • Constant Labels:
  • node_name: Name of the node

kepler_node_cpu_idle_watts

  • Type: GAUGE
  • Description: Power consumption of CPU in idle state at node level in watts
  • Labels:
  • zone: Energy zone (package, core, uncore, dram)
  • path: Hardware sensor path
  • Constant Labels:
  • node_name: Name of the node

kepler_node_cpu_usage_ratio

  • Type: GAUGE
  • Description: CPU usage ratio of a node (value between 0.0 and 1.0)
  • Constant Labels:
  • node_name: Name of the node

kepler_node_cpu_info

  • Type: GAUGE
  • Description: CPU information from procfs
  • Labels:
  • processor: Processor number
  • vendor_id: CPU vendor ID
  • model_name: CPU model name
  • physical_id: Physical CPU ID
  • core_id: Core ID

Container Metrics

These metrics provide energy and power information for individual containers.

kepler_container_cpu_joules_total

  • Type: COUNTER
  • Description: Total energy consumption of CPU at container level in joules
  • Labels:
  • container_id: Container identifier
  • container_name: Container name
  • runtime: Container runtime (docker, containerd, cri-o, podman)
  • state: Container state (running, terminated)
  • zone: Energy zone (package, core, uncore, dram)
  • pod_id: Pod identifier (if running in Kubernetes)
  • Constant Labels:
  • node_name: Name of the node

kepler_container_cpu_watts

  • Type: GAUGE
  • Description: Current power consumption of CPU at container level in watts
  • Labels:
  • container_id: Container identifier
  • container_name: Container name
  • runtime: Container runtime (docker, containerd, cri-o, podman)
  • state: Container state (running, terminated)
  • zone: Energy zone (package, core, uncore, dram)
  • pod_id: Pod identifier (if running in Kubernetes)
  • Constant Labels:
  • node_name: Name of the node

Process Metrics

These metrics provide energy and power information for individual processes.

kepler_process_cpu_joules_total

  • Type: COUNTER
  • Description: Total energy consumption of CPU at process level in joules
  • Labels:
  • pid: Process ID
  • comm: Process command name
  • exe: Process executable path
  • type: Process type (container, vm, regular)
  • state: Process state (running, terminated)
  • container_id: Container ID (if process runs in container)
  • vm_id: VM ID (if process runs in VM)
  • zone: Energy zone (package, core, uncore, dram)
  • Constant Labels:
  • node_name: Name of the node

kepler_process_cpu_watts

  • Type: GAUGE
  • Description: Current power consumption of CPU at process level in watts
  • Labels:
  • pid: Process ID
  • comm: Process command name
  • exe: Process executable path
  • type: Process type (container, vm, regular)
  • state: Process state (running, terminated)
  • container_id: Container ID (if process runs in container)
  • vm_id: VM ID (if process runs in VM)
  • zone: Energy zone (package, core, uncore, dram)
  • Constant Labels:
  • node_name: Name of the node

kepler_process_cpu_seconds_total

  • Type: COUNTER
  • Description: Total user and system CPU time at process level in seconds
  • Labels:
  • pid: Process ID
  • comm: Process command name
  • exe: Process executable path
  • type: Process type (container, vm, regular)
  • container_id: Container ID (if process runs in container)
  • vm_id: VM ID (if process runs in VM)
  • Constant Labels:
  • node_name: Name of the node

Virtual Machine Metrics

These metrics provide energy and power information for virtual machines.

kepler_vm_cpu_joules_total

  • Type: COUNTER
  • Description: Total energy consumption of CPU at VM level in joules
  • Labels:
  • vm_id: Virtual machine identifier
  • vm_name: Virtual machine name
  • hypervisor: Hypervisor type (kvm, qemu)
  • state: VM state (running, terminated)
  • zone: Energy zone (package, core, uncore, dram)
  • Constant Labels:
  • node_name: Name of the node

kepler_vm_cpu_watts

  • Type: GAUGE
  • Description: Current power consumption of CPU at VM level in watts
  • Labels:
  • vm_id: Virtual machine identifier
  • vm_name: Virtual machine name
  • hypervisor: Hypervisor type (kvm, qemu)
  • state: VM state (running, terminated)
  • zone: Energy zone (package, core, uncore, dram)
  • Constant Labels:
  • node_name: Name of the node

Pod Metrics

These metrics provide energy and power information for Kubernetes pods.

kepler_pod_cpu_joules_total

  • Type: COUNTER
  • Description: Total energy consumption of CPU at pod level in joules
  • Labels:
  • pod_id: Pod identifier
  • pod_name: Pod name
  • pod_namespace: Pod namespace
  • state: Pod state (running, terminated)
  • zone: Energy zone (package, core, uncore, dram)
  • Constant Labels:
  • node_name: Name of the node

kepler_pod_cpu_watts

  • Type: GAUGE
  • Description: Current power consumption of CPU at pod level in watts
  • Labels:
  • pod_id: Pod identifier
  • pod_name: Pod name
  • pod_namespace: Pod namespace
  • state: Pod state (running, terminated)
  • zone: Energy zone (package, core, uncore, dram)
  • Constant Labels:
  • node_name: Name of the node

Build Information Metrics

kepler_build_info

  • Type: GAUGE
  • Description: A metric with a constant '1' value labeled with version information
  • Labels:
  • arch: CPU architecture
  • branch: Git branch
  • revision: Git revision
  • version: Kepler version
  • goversion: Go version used to build

Power Attribution Algorithm

Kepler distributes energy consumption from hardware sensors to workloads using the following algorithm:

  1. Read Hardware Sensors: Collect total energy from RAPL (Running Average Power Limit) zones
  2. Calculate Node Usage: Determine total CPU time and usage ratio from /proc/stat
  3. Split Node Energy: Divide total energy into Active and Idle based on CPU usage ratio
  4. Process Attribution: Calculate CPU time deltas for each workload
  5. Distribute Active Energy: Allocate Active energy proportionally based on CPU utilization ratios
  6. Aggregate Workloads: Sum process energy to container/VM/pod levels
  7. Handle Edge Cases: Manage idle power, new/terminated processes, and counter wraparound

Zone-Based Energy Sources

RAPL (Running Average Power Limit)

RAPL is Intel's power capping mechanism that provides energy readings at different levels:

  • Package: Complete CPU socket energy (cores + uncore)
  • Core (PP0): CPU cores energy (processor model dependent)
  • Uncore (PP1): Uncore components (last-level cache, integrated GPU, memory controller)
  • DRAM: Memory energy consumption

Zone Availability by Architecture

Microarchitecture Package CORE (PP0) UNCORE (PP1) DRAM
Haswell ✓ ✓/✗ ✓/✗ ✓
Broadwell ✓ ✓/✗ ✓/✗ ✓
Skylake ✓ ✓ ✓/✗ ✓
Kaby Lake ✓ ✓ ✓/✗ ✓

✓ = Available, ✗ = Not available (varies by consumer/server grade)

Prometheus Queries

Getting Power Consumption (Watts)

To get current power consumption in watts, use the gauge metrics directly:

# Container power consumption
kepler_container_cpu_watts{container_name="nginx"}

# Node power consumption by zone
kepler_node_cpu_watts{zone="package"}

Getting Energy Consumption Rate (Watts from Counters)

To calculate power consumption from energy counters using rate functions:

# Container power consumption rate
rate(kepler_container_cpu_joules_total[1m])

# Node power consumption rate by zone
rate(kepler_node_cpu_joules_total{zone="package"}[1m])

Aggregating Pod Energy

# Total pod energy consumption
sum by (pod_name, pod_namespace) (rate(kepler_pod_cpu_joules_total[1m]))

Node Energy Breakdown

# Energy consumption by zone
sum by (zone) (rate(kepler_node_cpu_joules_total[1m]))

Configuration

Energy metrics can be controlled through Kepler configuration:

Metrics Levels

Configure which metric levels to export:

exporter:
  prometheus:
    metricsLevel:
      - node      # Node-level metrics
      - process   # Process-level metrics  
      - container # Container-level metrics
      - vm        # VM-level metrics
      - pod       # Pod-level metrics (requires Kubernetes)

RAPL Zones

Filter which RAPL zones to monitor:

rapl:
  zones: ["package", "core", "dram"]  # Empty array enables all available zones

For more configuration options, see the Configuration Guide.

RAPL Power Domain

RAPL power domains supported in some recent Intel microarchitecture (consumer-grade/server-grade):

Microarchitecture Package CORE (PP0) UNCORE (PP1) DRAM
Haswell Y/Y Y/N Y/N Y/Y
Broadwell Y/Y Y/N Y/N Y/Y
Skylake Y/Y Y/Y Y/N Y/Y
Kaby Lake Y/Y Y/Y Y/N Y/Y

Legend: Y = Available, N = Not available (varies by consumer/server grade)

Copyright Contributors to the Kepler's project.

The Linux Foundation® (TLF) has registered trademarks and uses trademarks. For a list of TLF trademarks, see Trademark Usage.