Manual Scheduling

  • Set node (modify nodeName will not work)

    1
    2
    3
    # Ex. Pod
    spec:
      nodeName: node01

  • Create a Binding object instead, add pass the binding to api in data parameter

Label Selector

1
kubectl get pods --selector app=my-app
  • Use annotations to store non-selector metadata

Taints and Tolerations

  • Taints mark nodes

    1
    2
    3
    4
    # add
    kubectl taint nodes node-name key=value:taint-effect
    # remove
    kubectl taint nodes node-name key=value:taint-effect-

  • taint-effect: NoSchedule, PreferNoSchedule, NoExecute

  • Tolerations mark pods

    1
    2
    3
    4
    5
    6
    7
    # Ex. Pod
    spec:
      tolerations:
      - key: key
        operator: Equal
        value: value
        effect: taint-effect

Node Selectors

  • Label nodes

    1
    kubectl label nodes node-name key=value

  • Configure node selector

    1
    2
    3
    4
    # Ex. Pod
    spec:
      nodeSelector:
        key: value

Node Affinity

1
2
3
4
5
6
7
8
9
10
# Ex. pod
spec:
  affinity:
    nodeAffinity:
      <type>:
        nodeSelectorTerms:
        - matchExpressions:
          - key: key
            operator: In/NotIn/Exists/etc.
            values: values
  • Types
    • requireDuringSchedulingIgnoreDuringExecution
    • preferredDuringSchedulingIgnoredDuringExecution
    • requiredDuringSchedulingRequiredDuringExecution

Resource Requirements and Limits

1
2
3
4
5
6
7
8
9
10
# Ex. Pod
spec:
  containers:
    resources:
      requests:
        memory: 256Mi
        cpu: 0.5
      limits:
        memory: 512Mi
        cpu: 1
  • Can create LimitRange object to overwrite default requests and limits
    1
    2
    3
    4
    5
    6
    7
    8
    9
    10
    11
    12
    13
    apiVersion: v1
    kind: LimitRange
    metadata:
      name: new-limit-range
    spec:
      limits:
      - default:
          memory: 1Gi
          cpu: 2
        defaultRequest:
          memory: 512Gi
          cpu: 1
        type: Container

Daemon Sets

  • Make sure that a specify pod always runs on each node
    1
    2
    3
    4
    5
    6
    7
    8
    9
    10
    11
    12
    13
    14
    15
    16
    apiVersion: apps/v1
    kind: DaemonSet
    metadata:
      name: my-daemon-set
    spec:
      selector:
        matchLabels:
          app: my-daemon
      template:
        metadata:
          labels:
            app: my-daemon
        spec:
          containers:
          - name: daemon-name
            image: daemon-image

Static Pods

  • Without master node, worker node can also create pods based on pod definition in the manifest directory
  • manifest directory path can be configured in kubelet.service
    1
    2
    3
    4
    5
    ExecStart= ... \\
      --pod-manifest-path=/etc/Kubernetes/manifests \\
      # or
      --config=kubeconfig.yaml \\
      ...
1
2
# kubeconfig.yaml
staticPodPath: /etc/Kubernetes/manifests
  • Use docker ps command to confirm pods if there is not kube-api-server
  • Or use kubectl get pods to get both pods and static pods. But you cannot use kubectl to modify static pods

Multiple Schedulers

  • Set default scheduler

    1
    2
    3
    4
    # kube-scheduler.service
    ExecStart= ... \\
      --scheduler-name=custom-scheduler \\
      ...

  • Create a new scheduler

    1
    2
    3
    4
    5
    6
    7
    # Copy from kube-scheduler
    # add
    spec:
      containers:
      - command:
        - --scheduler-name=custom-scheduler
        - --lock-object-name=custom-scheduler

  • Use scheduler

    1
    2
    3
    4
    # Ex. Pod
    spec:
      containers:
      - schedulerName: customer-scheduler