Skip to content

Latest commit

 

History

History
206 lines (176 loc) · 5.21 KB

kubernetes.md

File metadata and controls

206 lines (176 loc) · 5.21 KB
Raw

Kubernetes Metrics Backend

Description

The Kubernetes metrics backend interfaces with the Kubernetes API to expose allocation metrics gathered by querying for pod resource requests.

Configuration

Since the Kubernetes metrics backend implementation uses an in-cluster configuration, no additional configuration is required.

Example

apiVersion: cerebral.containership.io/v1alpha1
kind: MetricsBackend
metadata:
  name: kubernetes
spec:
  type: kubernetes

Available Metrics

CPU Percent Allocation

Description

Returns the percent of allocated CPUs across the nodes in the autoscaling group by summing the total CPU requests of the pods, and dividing by the nodes' total allocatable CPUs.

Metric

cpu_percent_allocation

Configuration

No configuration is available for this metric type.

Example

apiVersion: cerebral.containership.io/v1alpha1
kind: AutoscalingPolicy
metadata:
  name: cpu-example-policy
spec:
  metric: cpu_percent_allocation
  metricConfiguration: {}
  metricsBackend: kubernetes
  pollInterval: 15
  samplePeriod: 300
  scalingPolicy:
    scaleDown:
      adjustmentType: absolute
      adjustmentValue: 1
      comparisonOperator: <=
      threshold: 30
    scaleUp:
      adjustmentType: absolute
      adjustmentValue: 2
      comparisonOperator: '>='
      threshold: 70

GPU Percent Allocation

Description

Returns the percent of allocated GPUs across the nodes in the autoscaling group by summing the total GPU requests of the pods, and dividing by the nodes' total allocatable GPUs.

Note: Both amd.com/gpu and nvidia.com/gpu are summed together when calculating requested and allocatable GPU totals

Metric

gpu_percent_allocation

Configuration

No configuration is available for this metric type.

Example

apiVersion: cerebral.containership.io/v1alpha1
kind: AutoscalingPolicy
metadata:
  name: gpu-example-policy
spec:
  metric: gpu_percent_allocation
  metricConfiguration: {}
  metricsBackend: kubernetes
  pollInterval: 15
  samplePeriod: 300
  scalingPolicy:
    scaleDown:
      adjustmentType: absolute
      adjustmentValue: 1
      comparisonOperator: <=
      threshold: 30
    scaleUp:
      adjustmentType: absolute
      adjustmentValue: 2
      comparisonOperator: '>='
      threshold: 70

Memory Percent Allocation

Description

Returns the percent of allocated memory across the nodes in the autoscaling group by summing the total memory requests of the pods, and dividing by the nodes' total allocatable memory.

Metric

memory_percent_allocation

Configuration

No configuration is available for this metric type.

Example

apiVersion: cerebral.containership.io/v1alpha1
kind: AutoscalingPolicy
metadata:
  name: memory-example-policy
spec:
  metric: memory_percent_allocation
  metricConfiguration: {}
  metricsBackend: kubernetes
  pollInterval: 15
  samplePeriod: 300
  scalingPolicy:
    scaleDown:
      adjustmentType: absolute
      adjustmentValue: 1
      comparisonOperator: <=
      threshold: 30
    scaleUp:
      adjustmentType: absolute
      adjustmentValue: 2
      comparisonOperator: '>='
      threshold: 70

Ephemeral Storage Percent Allocation

Description

Returns the percent of allocated ephemeral storage across the nodes in the autoscaling group by summing the total ephemeral storage requests of the pods, and dividing by the nodes' total allocatable ephemeral storage.

Metric

ephemeral_storage_percent_allocation

Configuration

No configuration is available for this metric type.

Example

apiVersion: cerebral.containership.io/v1alpha1
kind: AutoscalingPolicy
metadata:
  name: ephemeral-storage-example-policy
spec:
  metric: ephemeral_storage_percent_allocation
  metricConfiguration: {}
  metricsBackend: kubernetes
  pollInterval: 15
  samplePeriod: 300
  scalingPolicy:
    scaleDown:
      adjustmentType: absolute
      adjustmentValue: 1
      comparisonOperator: <=
      threshold: 30
    scaleUp:
      adjustmentType: absolute
      adjustmentValue: 2
      comparisonOperator: '>='
      threshold: 70

Pod Percent Allocation

Description

Returns the percent of allocated pods across the nodes in the autoscaling group by summing the total number of pods running on the nodes, and dividing by the nodes' total allocatable pods.

Metric

pod_percent_allocation

Configuration

No configuration is available for this metric type.

Example

apiVersion: cerebral.containership.io/v1alpha1
kind: AutoscalingPolicy
metadata:
  name: pod-example-policy
spec:
  metric: pod_percent_allocation
  metricConfiguration: {}
  metricsBackend: kubernetes
  pollInterval: 15
  samplePeriod: 300
  scalingPolicy:
    scaleDown:
      adjustmentType: absolute
      adjustmentValue: 1
      comparisonOperator: <=
      threshold: 30
    scaleUp:
      adjustmentType: absolute
      adjustmentValue: 2
      comparisonOperator: '>='
      threshold: 70