Skip to content

Kube Scheduler

By monitoring Kube Scheduler metrics to help configure and optimize Kube Scheduler, you can improve cluster resource utilization and application performance

Config¶

Precondition¶

Installed DataKit

Collector configuration¶

ConfigMap resource in datakit.yaml to collect metric data for Kube Scheduler

apiVersion: v1
kind: ConfigMap
metadata:
  name: datakit-conf
  namespace: datakit
data:
   kubernetesprometheus.conf: |-  
   # The following configuration is not static, please modify it according to the actual situation
        [[inputs.kubernetesprometheus.instances]]
          role       = "pod"
          namespaces = ["kube-system"]
          selector   = "component=Kube Scheduler,tier=control-plane"      
          scrape   = "true"
          scheme   = "https"
          port     = "10259"
          path     = "/metrics"
          interval = "30s"

         [inputs.kubernetesprometheus.instances.custom]
           measurement        = "Kube Scheduler"
           job_as_measurement = false
           [inputs.kubernetesprometheus.instances.custom.tags]
             cluster_name_k8s = "k8s-test"           
             node_name        = "__kubernetes_pod_node_name"
             instance         = "__kubernetes_mate_instance"

         [inputs.kubernetesprometheus.instances.auth]
           bearer_token_file = "/var/run/secrets/kubernetes.io/serviceaccount/token"
           [inputs.kubernetesprometheus.instances.auth.tls_config]
             insecure_skip_verify = true
             ca_certs = []
             cert     = ""
             cert_key = ""

Restart DataKit

kubectl rollout restart ds  datakit -n datakit

Metric¶

kube scheduler metric set¶

The Kube Scheduler metric is located under the kube-scheduler metric set. Here is the relevant description of the Kube Scheduler metric

Metrics	Description	Unit
`scheduler_scheduler_cache_size`	`Number of nodes, Pods, and AssumedPods (assuming Pods to be scheduled) in the scheduler cache`	count
`scheduler_pending_pods`	`Number of Pending Pods`	count
`process_cpu_seconds_total`	`The kube-scheduler process Total CPU usage time`	s
`rest_client_request_duration_seconds_bucket`	`Analysis of HTTP request latency from method (Verb) and URL dimensions`	ms
`rest_client_requests_total`	`Analyze the number of HTTP requests from the Status Code, Method and Host dimensions`	count
`rest_client_requests_total`	`Total number of HTTP requests made by kube-scheduler to kube-apiserver`	count