Node selector란?
- Worker node에 할당된 label을 이용해 node를 선택
- node label 설정
- kubectl label nodes <NODE_NAME> <LABEL_KEY>=<LABEL_VALUE>
- kubectl label node1.example.com gpu=true
- kubectl get nodes -L <LABEL_KEY>
- definition
apiVersion: v1
kind: Pod
metadata:
name: tensorflow-gpu
spec:
containers:
- name: tensorflow
image: tensorflow/tensorflow:nightly-jupyter
ports:
- containerPort: 8888
protocol: TCP
nodeSelector:
gpu: "true"
[ Hands - on ]
- node2에 gpu=true이라는 key, value 값 label
kubectl label node node2.example.com gpu=true
kubectl get nodes -L gpu
NAME STATUS ROLES AGE VERSION GPU
master.example.com Ready control-plane 17h v1.24.1
node1.example.com Ready <none> 17h v1.24.1
node2.example.com Ready <none> 17h v1.24.1 true
- gpu=true이라는 key, value 값 label 되어 있는 node에서 pod 실행
vi tensorflow-gpu.yaml
apiVersion: v1
kind: Pod
metadata:
name: tensorflow-gpu
spec:
containers:
- name: tensorflow
image: tensorflow/tensorflow:nightly-jupyter
ports:
- containerPort: 8888
protocol: TCP
nodeSelector:
gpu: "true"
kubectl get pods -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
tensorflow-gpu 1/1 Running 0 79s 10.44.0.1 node2.example.com <none> <none>
Affinity & antiAffinity
nodeAffinity
- node의 특정 집합에만 pod를 스케줄 하도록 지시
- nodeSelector : selector field에 명시된 모든 label이 포함되어야 배치됨
- nodeAffinity : 특정 node에만 pod가 실행되도록 유도
- nodeAffinity 요구 조건
- 엄격한 요구(필수 충족 조건) : requiredDuringSchedulingIgnoredDuringExecution
- 선호도 요구(가중치) : preferredDuringSchedulingIgnoredDuringExecution
- definition
apiVersion: v1
kind: Pod
metadata:
name: redis-ssd
spec:
containers:
- name: redis
image: redis
affinity:
nodeAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
nodeSelectorTerms:
- matchExpressions:
- {key: disktype, operator: In, values: ["ssd"]}
[ Hands - on ]
- node2에 disktype=ssd이라는 key, value 값 label
kubectl label node node2.example.com disktype=ssd
kubectl get nodes -L gpu,disktype
NAME STATUS ROLES AGE VERSION GPU DISKTYPE
master.example.com Ready control-plane 18h v1.24.1
node1.example.com Ready <none> 18h v1.24.1
node2.example.com Ready <none> 18h v1.24.1 true ssd
vi tensorflow-gpu-ssd.yaml
apiVersion: v1
kind: Pod
metadata:
name: tensorflow-gpu-ssd
spec:
containers:
- name: tensorflow
image: tensorflow/tensorflow:nightly-jupyter
ports:
- containerPort: 8888
protocol: TCP
affinity:
nodeAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
nodeSelectorTerms:
- matchExpressions:
- {key: disktype, operator: Exists}
preferredDuringSchedulingIgnoredDuringExecution:
- weight: 10
preference:
matchExpressions:
- {key: gpu, operator: In, values: ["true"]}
- {key: disktype, operator: In, values: ["ssd"]}
kubectl create -f tensorflow-gpu-ssd.yaml
kubectl get pods -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
tensorflow-gpu-ssd 1/1 Running 0 4s 10.44.0.1 node2.example.com <none> <none>
Pod Affinity
- Pod 스케줄링
- pod affinity : pod를 더 가깝게 배치
- pod antiAffinity : pod를 더 멀리 배치
- Pod Affinity 요구조건
- 엄격한 요구 : requiredDuringSchedulingIgnoredDuringExecution
- 선호도 요구 : preferredDuringSchedulingIgnoredDuringExecution
- topologyKey
- node label을 이용해 pod의 affinity를 설정할 수 있는 또 하나의 기준
- K8s는 pod를 스케줄링 할 때 먼저 pod의 label을 기준으로 대상 노드를 찾고, 이후 topologyKey 필드를 확인해 해당 node가 원하는 node인지 확인
[ Hands - on ]
1. Affinity
kubectl run backend --labels app=backend --image=busybox -- sleep 9999999
kubectl get pods -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
backend 1/1 Running 0 2m44s 10.36.0.1 node1.example.com <none> <none>
kubectl get pods --show-labels
NAME READY STATUS RESTARTS AGE LABELS
backend 1/1 Running 0 18s app=backend
- app=backend로 label 된 pod가 실행되어 있는 node에 배치되는 pod 생성 및 확인
vi pod-affinity.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
name: frontend
spec:
replicas: 5
selector:
matchLabels:
app: frontend
template:
metadata:
labels:
app: frontend
spec:
affinity:
podAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
- labelSelector:
matchLabels:
app: backend
topologyKey: kubernetes.io/hostname
containers:
- name: main
image: busybox
args:
- sleep
- "99999"
kubectl apply -f pod-affinity.yaml
kubectl get pods -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
backend 1/1 Running 0 4m2s 10.36.0.1 node1.example.com <none> <none>
frontend-7c4546cff-6gmdz 1/1 Running 0 16s 10.36.0.2 node1.example.com <none> <none>
frontend-7c4546cff-m9ltp 1/1 Running 0 16s 10.36.0.5 node1.example.com <none> <none>
frontend-7c4546cff-tvs4r 1/1 Running 0 16s 10.36.0.4 node1.example.com <none> <none>
frontend-7c4546cff-z9g4k 1/1 Running 0 16s 10.36.0.6 node1.example.com <none> <none>
frontend-7c4546cff-zht2v 1/1 Running 0 16s 10.36.0.3 node1.example.com <none> <none>
2. antiaffinity
- app=backend로 label 된 pod가 실행되어 있지 않은 node에 배치되는 pod 생성 및 확인
vi pod-antiaffinity.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
name: frontend
spec:
replicas: 5
selector:
matchLabels:
app: frontend
template:
metadata:
labels:
app: frontend
spec:
affinity:
podAntiAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
- labelSelector:
matchLabels:
app: backend
topologyKey: kubernetes.io/hostname
containers:
- name: main
image: busybox
args:
- sleep
- "99999"
kubectl apply -f pod-antiaffinity.yaml
kubectl get pods -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
backend 1/1 Running 0 7m55s 10.36.0.1 node1.example.com <none> <none>
frontend-5ddc8895bc-7tgrq 1/1 Running 0 16s 10.44.0.3 node2.example.com <none> <none>
frontend-5ddc8895bc-fk8rl 1/1 Running 0 16s 10.44.0.2 node2.example.com <none> <none>
frontend-5ddc8895bc-m7ktc 1/1 Running 0 16s 10.44.0.4 node2.example.com <none> <none>
frontend-5ddc8895bc-q25kn 1/1 Running 0 16s 10.44.0.1 node2.example.com <none> <none>
frontend-5ddc8895bc-x577z 1/1 Running 0 16s 10.44.0.5 node2.example.com <none> <none>
참고 : [따배쿠/멤버십] 12-1. Pod Scheduling
댓글