Kubernetes - Core Concepts 2

Scheduling

How scheduling works

pod-definition.yaml

apiVersion:
kind:
metadata:
  name: nginx
  labels:
    name: nginx-name
spec:
  containers:
    - name: nginx-container
      image: nginx
      ports:
        - containerPort: 8080
      nodeName:                   (일반적으로 생략되고, 자동작성해줌)

kubectl get pod -o wide

If No scheduler

• manually schedule pods

pod-bind-definition.yaml

apiVersion:
kind: Binding
metadata:
  name: nginx
  labels:
    name: nginx-name
target:
  apiVersion: v1
  kind: Node
  ** nodeName: node02 **

Labels and Selectors

replicaset-definition.yaml

apiVersion: apps/v1
kind: ReplicaSet
metadata:
  name: rs-1

spec:
  replicas: 2
  selector:
    matchLabels:
      name: nginx-name
      class: mammal  
      kind: cat  
      color: green

    template:
      metadata:
        labels:
          name: nginx-name
          class: Mammal  
          kind: cat  
          color: green
    spec:
      containers:
        - name: nginx-container
          image: nginx
          ports:
            - containerPort: 8080

kubectl get po --show-labels
kubectl get po -l color=green
kubectl get po -l color=green,kind=cat
kubectl get po -l color=green --no-headers | wc -l

Taints and Tolerations (오염과 관용)

어떤 모기는 모기약뿌린 사람한테 안가지만, 어떤 슈퍼모기는 모기약뿌린 사람한테도 감

• 사람은 node, 모기는 pod, 슈퍼모기는 tolerant 된 pod

kubectl taint nodes [node-name] [key=value:taint-effect]
- taint-effect = [NoSchedule | PreferNoSchedule | NoExecute]
  - NoExecute: node에
  - NoSchedule:
  - PreferNoSchedule:

$ kubectl taint nodes node1 app=blue:NoSchedule

pod-definition.yml

spec:
  containers:
    - name:
      image:
    tolerations:
      key: app
      operator: Equal
      value: blue
      effect: NoSchedule

controlplane node에 taint가 설정되어서 pod들이 controlplane node 에 가지 않음

Node Selectors

set limitation on pods

node 1: size 10 node 2: size 5 node 3: size 5 위와 같은 구성일 때

kubectl label nodes <node-name> <label-key>=<label-value>

pod-nodeselector-definition.yml

...
spec:
  containers:
    - name:
      image:
  nodeSelector:
    ????????????

kubectl explain pod –recursive

Node Affinity (친밀감, 관련성)

pod host on particuler node

pod-nodeselector-definition.yml

...
spec:
  containers:
    - name:
      image:
  affinity:
    nodeAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:     (라벨이 없으면 안함)
      preferredDuringSchedulingIgnoredDuringExecution:    (라벨이 없어도 가능)
      preferredDuringSchedulingRequiredDuringExecution:    (라벨이 없으면 앱 삭제)
        nodeSelectorTerms:
        - matchExpressions:
          - key: size
            operator: In / NotIn / Exists / ...
            values:
            - Large

Resoruce Requirements and Limits

apiVersion: v1
kind: LimitRange
metadata:
  name: cpu-limit-range
spec:
  limits:
  - default:
      cpu: 1
    defaultRequest:
      cpu: 0.5
    type: Container

DeamonSets

one pod copy of ALL each node

kube-apiserver에 의해 생성됨

• use case
  • monitoring agent
  • kube-proxy
  • weave-net

daemon-set-definition.yaml (similar to replicaset)

apiVersion: apps/v1
kind: DeamonSet
metadata:
  name: monitoring-agent
spec:
  selector:
    matchLabels:
      app: monitoring-agent
  template:
    metadata:
      labels:
        app: monitoring-agent
    spec:
      containers:
      - name: monitoring-agent
        image: monitoring-agent

kubectl create -f daemon-set-definition.yaml

after v1.12

Static Pods

/etc/kubernetes/manifests/.yaml

자동으로 읽고 pod 생성/수정/삭제

1. service
   • –pod-manifest-path=/etc/kubernetes/manifest
2. kubelet.service
   • –config= > staticPodPath: /etc/kubernetes/manifest

static pod 는 각각의 kubelet에 의해 생성됨

kube-apiserver 에서는 read-only

kubeadm 은 control-plane에 static pod 형태로 구성

ps -ef

grep kubelet > find “–config”

Multiple Scheduler

Custom Scheduler

pod-scheduler-definition.yaml

...
spec:
  containers:
    - name:
      image:
  schedulerName: my-custom-scheduler

kubectl get events

.