Kubernetes K8S之affinity親和性與反親和性詳解與示例
阿新 • • 發佈：2020-11-03
Kubernetes K8S之Node節點親和性與反親和性以及Pod親和性與反親和性詳解與示例
主機配置規劃
| 伺服器名稱(hostname) | 系統版本 | 配置 | 內網IP | 外網IP(模擬) |
|---|---|---|---|---|
| k8s-master | CentOS7.7 | 2C/4G/20G | 172.16.1.110 | 10.0.0.110 |
| k8s-node01 | CentOS7.7 | 2C/4G/20G | 172.16.1.111 | 10.0.0.111 |
| k8s-node02 | CentOS7.7 | 2C/4G/20G | 172.16.1.112 | 10.0.0.112 |
親和性和反親和性
nodeSelector提供了一種非常簡單的方法，將pods約束到具有特定標籤的節點。而親和性/反親和性極大地擴充套件了可表達的約束型別。關鍵的增強是：
1、親和性/反親和性語言更具表達性。除了使用邏輯AND操作建立的精確匹配之外，該語言還提供了更多的匹配規則；
2、可以指示規則是優選項而不是硬要求，因此如果排程器不能滿足，pod仍將被排程；
3、可以針對節點（或其他拓撲域）上執行的pods的標籤進行約束，而不是針對節點的自身標籤，這影響哪些Pod可以或不可以共處。
親和特性包括兩種型別：node節點親和性/反親和性 和 pod親和性/反親和性。pod親和性/反親和性約束針對的是pod標籤而不是節點標籤。
拓撲域是什麼：多個node節點，擁有相同的label標籤【節點標籤的鍵值相同】，那麼這些節點就處於同一個拓撲域。★★★★★
node節點親和性
當前有兩種型別的節點親和性，稱為requiredDuringSchedulingIgnoredDuringExecution
前者表示Pod要排程到的節點必須滿足規則條件，不滿足則不會排程，pod會一直處於Pending狀態；後者表示優先排程到滿足規則條件的節點，如果不能滿足再排程到其他節點。
名稱中的 IgnoredDuringExecution 部分意味著，與nodeSelector的工作方式類似，如果節點上的標籤在Pod執行時發生更改，使得pod上的親和性規則不再滿足，那麼pod仍將繼續在該節點上執行。
在未來，會計劃提供requiredDuringSchedulingRequiredDuringExecution，類似requiredDuringSchedulingIgnoredDuringExecution。不同之處就是pod執行過程中如果節點不再滿足pod的親和性，則pod會在該節點中逐出。
節點親和性語法支援以下運算子：In，NotIn，Exists，DoesNotExist，Gt，Lt。可以使用NotIn和DoesNotExist實現節點的反親和行為。
運算子關係：
- In：label的值在某個列表中
- NotIn：label的值不在某個列表中
- Gt：label的值大於某個值
- Lt：label的值小於某個值
- Exists：某個label存在
- DoesNotExist：某個label不存在
1、如果同時指定nodeSelector和nodeAffinity，則必須滿足兩個條件，才能將Pod排程到候選節點上。
2、如果在nodeAffinity型別下指定了多個nodeSelectorTerms物件【物件不能有多個，如果存在多個只有最後一個生效】，那麼只有最後一個nodeSelectorTerms物件生效。
3、如果在nodeSelectorTerms下指定了多個matchExpressions列表，那麼只要能滿足其中一個matchExpressions，就可以將pod排程到某個節點上【針對節點硬親和】。
4、如果在matchExpressions下有多個key列表，那麼只有當所有key滿足時，才能將pod排程到某個節點【針對硬親和】。
5、在key下的values只要有一個滿足條件，那麼當前的key就滿足條件
6、如果pod已經排程在該節點，當我們刪除或修該節點的標籤時，pod不會被移除。換句話說，親和性選擇只有在pod排程期間有效。
7、preferredDuringSchedulingIgnoredDuringExecution中的weight（權重）欄位在1-100範圍內。對於每個滿足所有排程需求的節點(資源請求、RequiredDuringScheduling親和表示式等)，排程器將通過迭代該欄位的元素來計算一個總和，如果節點與相應的匹配表示式匹配，則向該總和新增“權重”。然後將該分數與節點的其他優先順序函式的分數結合起來。總得分最高的節點是最受歡迎的。
node節點親和性示例
準備事項
給node節點打label標籤
1 ### --overwrite覆蓋已存在的標籤資訊 2 kubectl label nodes k8s-node01 disk-type=ssd --overwrite 3 kubectl label nodes k8s-node01 cpu-num=12 4 5 kubectl label nodes k8s-node02 disk-type=sata 6 kubectl label nodes k8s-node02 cpu-num=24
查詢所有節點標籤資訊
1 [root@k8s-master ~]# kubectl get node -o wide --show-labels 2 NAME STATUS ROLES AGE VERSION INTERNAL-IP EXTERNAL-IP OS-IMAGE KERNEL-VERSION CONTAINER-RUNTIME LABELS 3 k8s-master Ready master 43d v1.17.4 172.16.1.110 <none> CentOS Linux 7 (Core) 3.10.0-1062.el7.x86_64 docker://19.3.8 beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/arch=amd64,kubernetes.io/hostname=k8s-master,kubernetes.io/os=linux,node-role.kubernetes.io/master= 4 k8s-node01 Ready <none> 43d v1.17.4 172.16.1.111 <none> CentOS Linux 7 (Core) 3.10.0-1062.el7.x86_64 docker://19.3.8 beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,cpu-num=12,disk-type=ssd,kubernetes.io/arch=amd64,kubernetes.io/hostname=k8s-node01,kubernetes.io/os=linux 5 k8s-node02 Ready <none> 43d v1.17.4 172.16.1.112 <none> CentOS Linux 7 (Core) 3.10.0-1062.el7.x86_64 docker://19.3.8 beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,cpu-num=24,disk-type=sata,kubernetes.io/arch=amd64,kubernetes.io/hostname=k8s-node02,kubernetes.io/os=linux
參見上面，給k8s-node01打了cpu-num=12,disk-type=ssd標籤；給k8s-node02打了cpu-num=24,disk-type=sata標籤。
node硬親和性示例
必須滿足條件才能排程，否則不會排程
要執行的yaml檔案
1 [root@k8s-master nodeAffinity]# pwd 2 /root/k8s_practice/scheduler/nodeAffinity 3 [root@k8s-master nodeAffinity]# cat node_required_affinity.yaml 4 apiVersion: apps/v1 5 kind: Deployment 6 metadata: 7 name: node-affinity-deploy 8 labels: 9 app: nodeaffinity-deploy 10 spec: 11 replicas: 5 12 selector: 13 matchLabels: 14 app: myapp 15 template: 16 metadata: 17 labels: 18 app: myapp 19 spec: 20 containers: 21 - name: myapp-pod 22 image: registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 23 imagePullPolicy: IfNotPresent 24 ports: 25 - containerPort: 80 26 affinity: 27 nodeAffinity: 28 requiredDuringSchedulingIgnoredDuringExecution: 29 nodeSelectorTerms: 30 - matchExpressions: 31 # 表示node標籤存在 disk-type=ssd 或 disk-type=sas 32 - key: disk-type 33 operator: In 34 values: 35 - ssd 36 - sas 37 # 表示node標籤存在 cpu-num且值大於6 38 - key: cpu-num 39 operator: Gt 40 values: 41 - "6"
執行yaml檔案並檢視狀態
1 [root@k8s-master nodeAffinity]# kubectl apply -f node_required_affinity.yaml 2 deployment.apps/node-affinity-deploy created 3 [root@k8s-master nodeAffinity]# 4 [root@k8s-master nodeAffinity]# kubectl get deploy -o wide 5 NAME READY UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTOR 6 node-affinity-deploy 5/5 5 5 6s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp 7 [root@k8s-master nodeAffinity]# 8 [root@k8s-master nodeAffinity]# kubectl get rs -o wide 9 NAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTOR 10 node-affinity-deploy-5c88ffb8ff 5 5 5 11s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp,pod-template-hash=5c88ffb8ff 11 [root@k8s-master nodeAffinity]# 12 [root@k8s-master nodeAffinity]# kubectl get pod -o wide 13 NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES 14 node-affinity-deploy-5c88ffb8ff-2mbfl 1/1 Running 0 15s 10.244.4.237 k8s-node01 <none> <none> 15 node-affinity-deploy-5c88ffb8ff-9hjhk 1/1 Running 0 15s 10.244.4.235 k8s-node01 <none> <none> 16 node-affinity-deploy-5c88ffb8ff-9rg75 1/1 Running 0 15s 10.244.4.239 k8s-node01 <none> <none> 17 node-affinity-deploy-5c88ffb8ff-pqtfh 1/1 Running 0 15s 10.244.4.236 k8s-node01 <none> <none> 18 node-affinity-deploy-5c88ffb8ff-zqpl8 1/1 Running 0 15s 10.244.4.238 k8s-node01 <none> <none>
由上可見，再根據之前打的標籤，很容易推斷出當前pod只能排程在k8s-node01節點。
即使我們刪除原來的rs，重新生成rs後pod依舊會排程到k8s-node01節點。如下：
1 [root@k8s-master nodeAffinity]# kubectl delete rs node-affinity-deploy-5c88ffb8ff 2 replicaset.apps "node-affinity-deploy-5c88ffb8ff" deleted 3 [root@k8s-master nodeAffinity]# 4 [root@k8s-master nodeAffinity]# kubectl get rs -o wide 5 NAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTOR 6 node-affinity-deploy-5c88ffb8ff 5 5 2 4s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp,pod-template-hash=5c88ffb8ff 7 [root@k8s-master nodeAffinity]# 8 [root@k8s-master nodeAffinity]# kubectl get pod -o wide 9 NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES 10 node-affinity-deploy-5c88ffb8ff-2v2tb 1/1 Running 0 11s 10.244.4.241 k8s-node01 <none> <none> 11 node-affinity-deploy-5c88ffb8ff-gl4fm 1/1 Running 0 11s 10.244.4.240 k8s-node01 <none> <none> 12 node-affinity-deploy-5c88ffb8ff-j26rg 1/1 Running 0 11s 10.244.4.244 k8s-node01 <none> <none> 13 node-affinity-deploy-5c88ffb8ff-vhzmn 1/1 Running 0 11s 10.244.4.243 k8s-node01 <none> <none> 14 node-affinity-deploy-5c88ffb8ff-xxj8m 1/1 Running 0 11s 10.244.4.242 k8s-node01 <none> <none>
node軟親和性示例
優先排程到滿足條件的節點，如果都不滿足也會排程到其他節點。
要執行的yaml檔案
1 [root@k8s-master nodeAffinity]# pwd 2 /root/k8s_practice/scheduler/nodeAffinity 3 [root@k8s-master nodeAffinity]# cat node_preferred_affinity.yaml 4 apiVersion: apps/v1 5 kind: Deployment 6 metadata: 7 name: node-affinity-deploy 8 labels: 9 app: nodeaffinity-deploy 10 spec: 11 replicas: 5 12 selector: 13 matchLabels: 14 app: myapp 15 template: 16 metadata: 17 labels: 18 app: myapp 19 spec: 20 containers: 21 - name: myapp-pod 22 image: registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 23 imagePullPolicy: IfNotPresent 24 ports: 25 - containerPort: 80 26 affinity: 27 nodeAffinity: 28 preferredDuringSchedulingIgnoredDuringExecution: 29 - weight: 1 30 preference: 31 matchExpressions: 32 # 表示node標籤存在 disk-type=ssd 或 disk-type=sas 33 - key: disk-type 34 operator: In 35 values: 36 - ssd 37 - sas 38 - weight: 50 39 preference: 40 matchExpressions: 41 # 表示node標籤存在 cpu-num且值大於16 42 - key: cpu-num 43 operator: Gt 44 values: 45 - "16"
執行yaml檔案並檢視狀態
1 [root@k8s-master nodeAffinity]# kubectl apply -f node_preferred_affinity.yaml 2 deployment.apps/node-affinity-deploy created 3 [root@k8s-master nodeAffinity]# 4 [root@k8s-master nodeAffinity]# kubectl get deploy -o wide 5 NAME READY UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTOR 6 node-affinity-deploy 5/5 5 5 9s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp 7 [root@k8s-master nodeAffinity]# 8 [root@k8s-master nodeAffinity]# kubectl get rs -o wide 9 NAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTOR 10 node-affinity-deploy-d5d9cbc8d 5 5 5 13s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp,pod-template-hash=d5d9cbc8d 11 [root@k8s-master nodeAffinity]# 12 [root@k8s-master nodeAffinity]# kubectl get pod -o wide 13 NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES 14 node-affinity-deploy-d5d9cbc8d-bv86t 1/1 Running 0 18s 10.244.2.243 k8s-node02 <none> <none> 15 node-affinity-deploy-d5d9cbc8d-dnbr8 1/1 Running 0 18s 10.244.2.244 k8s-node02 <none> <none> 16 node-affinity-deploy-d5d9cbc8d-ldq82 1/1 Running 0 18s 10.244.2.246 k8s-node02 <none> <none> 17 node-affinity-deploy-d5d9cbc8d-nt74q 1/1 Running 0 18s 10.244.4.2 k8s-node01 <none> <none> 18 node-affinity-deploy-d5d9cbc8d-rt5nb 1/1 Running 0 18s 10.244.2.245 k8s-node02 <none> <none>
由上可見，再根據之前打的標籤，很容易推斷出當前pod會【優先】排程在k8s-node02節點。
node軟硬親和性聯合示例
硬親和性與軟親和性一起使用
要執行的yaml檔案
1 [root@k8s-master nodeAffinity]# pwd 2 /root/k8s_practice/scheduler/nodeAffinity 3 [root@k8s-master nodeAffinity]# cat node_affinity.yaml 4 apiVersion: apps/v1 5 kind: Deployment 6 metadata: 7 name: node-affinity-deploy 8 labels: 9 app: nodeaffinity-deploy 10 spec: 11 replicas: 5 12 selector: 13 matchLabels: 14 app: myapp 15 template: 16 metadata: 17 labels: 18 app: myapp 19 spec: 20 containers: 21 - name: myapp-pod 22 image: registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 23 imagePullPolicy: IfNotPresent 24 ports: 25 - containerPort: 80 26 affinity: 27 nodeAffinity: 28 requiredDuringSchedulingIgnoredDuringExecution: 29 nodeSelectorTerms: 30 # 表示node標籤存在 cpu-num且值大於10 31 - matchExpressions: 32 - key: cpu-num 33 operator: Gt 34 values: 35 - "10" 36 preferredDuringSchedulingIgnoredDuringExecution: 37 - weight: 50 38 preference: 39 matchExpressions: 40 # 表示node標籤存在 disk-type=ssd 或 disk-type=sas 41 - key: disk-type 42 operator: In 43 values: 44 - ssd 45 - sas
執行yaml檔案並檢視狀態
1 [root@k8s-master nodeAffinity]# kubectl apply -f node_affinity.yaml 2 deployment.apps/node-affinity-deploy created 3 [root@k8s-master nodeAffinity]# 4 [root@k8s-master nodeAffinity]# kubectl get deploy -o wide 5 NAME READY UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTOR 6 node-affinity-deploy 5/5 5 5 9s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp 7 [root@k8s-master nodeAffinity]# 8 [root@k8s-master nodeAffinity]# kubectl get rs -o wide 9 NAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTOR 10 node-affinity-deploy-f9cb9b99b 5 5 5 13s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp,pod-template-hash=f9cb9b99b 11 [root@k8s-master nodeAffinity]# 12 [root@k8s-master nodeAffinity]# kubectl get pod -o wide 13 NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES 14 node-affinity-deploy-f9cb9b99b-8w2nc 1/1 Running 0 17s 10.244.4.10 k8s-node01 <none> <none> 15 node-affinity-deploy-f9cb9b99b-csk2s 1/1 Running 0 17s 10.244.4.9 k8s-node01 <none> <none> 16 node-affinity-deploy-f9cb9b99b-g42kq 1/1 Running 0 17s 10.244.4.8 k8s-node01 <none> <none> 17 node-affinity-deploy-f9cb9b99b-m6xbv 1/1 Running 0 17s 10.244.4.7 k8s-node01 <none> <none> 18 node-affinity-deploy-f9cb9b99b-mxbdp 1/1 Running 0 17s 10.244.2.253 k8s-node02 <none> <none>
由上可見，再根據之前打的標籤，很容易推斷出k8s-node01、k8s-node02都滿足必要條件，但當前pod會【優先】排程在k8s-node01節點。
Pod親和性
與節點親和性一樣，當前有Pod親和性/反親和性都有兩種型別，稱為requiredDuringSchedulingIgnoredDuringExecution和preferredDuringSchedulingIgnoredDuringExecution，分別表示“硬”與“軟”要求。對於硬要求，如果不滿足則pod會一直處於Pending狀態。
Pod的親和性與反親和性是基於Node節點上已經執行pod的標籤(而不是節點上的標籤)決定的，從而約束哪些節點適合排程你的pod。
規則的形式是：如果X已經運行了一個或多個符合規則Y的pod，則此pod應該在X中執行(如果是反親和的情況下，則不應該在X中執行）。當然pod必須處在同一名稱空間，不然親和性/反親和性無作用。從概念上講，X是一個拓撲域。我們可以使用topologyKey來表示它，topologyKey 的值是node節點標籤的鍵以便系統用來表示這樣的拓撲域。當然這裡也有個隱藏條件，就是node節點標籤的鍵值相同時，才是在同一拓撲域中；如果只是節點標籤名相同，但是值不同，那麼也不在同一拓撲域。★★★★★
也就是說：Pod的親和性/反親和性排程是根據拓撲域來界定排程的，而不是根據node節點。★★★★★
注意事項1、pod之間親和性/反親和性需要大量的處理，這會明顯降低大型叢集中的排程速度。不建議在大於幾百個節點的叢集中使用它們。
2、Pod反親和性要求對節點進行一致的標記。換句話說，叢集中的每個節點都必須有一個匹配topologyKey的適當標籤。如果某些或所有節點缺少指定的topologyKey標籤，可能會導致意外行為。
requiredDuringSchedulingIgnoredDuringExecution中親和性的一個示例是“將服務A和服務B的Pod放置在同一區域【拓撲域】中，因為它們之間有很多交流”；preferredDuringSchedulingIgnoredDuringExecution中反親和性的示例是“將此服務的 pod 跨區域【拓撲域】分佈”【此時硬性要求是說不通的，因為你可能擁有的 pod 數多於區域數】。
Pod親和性/反親和性語法支援以下運算子：In，NotIn，Exists，DoesNotExist。
原則上，topologyKey可以是任何合法的標籤鍵。但是，出於效能和安全方面的原因，topologyKey有一些限制：
1、對於Pod親和性，在requiredDuringSchedulingIgnoredDuringExecution和preferredDuringSchedulingIgnoredDuringExecution中topologyKey都不允許為空。
2、對於Pod反親和性，在requiredDuringSchedulingIgnoredDuringExecution和preferredDuringSchedulingIgnoredDuringExecution中topologyKey也都不允許為空。
3、對於requiredDuringSchedulingIgnoredDuringExecution的pod反親和性，引入了允許控制器LimitPodHardAntiAffinityTopology來限制topologyKey的kubernet.io/hostname。如果你想讓它對自定義拓撲可用，你可以修改許可控制器，或者乾脆禁用它。
4、除上述情況外，topologyKey可以是任何合法的標籤鍵。
Pod 間親和通過 PodSpec 中 affinity 欄位下的 podAffinity 欄位進行指定。而 pod 間反親和通過 PodSpec 中 affinity 欄位下的 podAntiAffinity 欄位進行指定。
Pod親和性/反親和性的requiredDuringSchedulingIgnoredDuringExecution所關聯的matchExpressions下有多個key列表，那麼只有當所有key滿足時，才能將pod排程到某個區域【針對Pod硬親和】。
pod親和性與反親和性示例
為了更好的演示Pod親和性與反親和性，本次示例我們會將k8s-master節點也加入進來進行演示。
準備事項
給node節點打label標籤
1 # 刪除已存在標籤 2 kubectl label nodes k8s-node01 cpu-num- 3 kubectl label nodes k8s-node01 disk-type- 4 kubectl label nodes k8s-node02 cpu-num- 5 kubectl label nodes k8s-node02 disk-type- 6 7 8 ### --overwrite覆蓋已存在的標籤資訊 9 # k8s-master 標籤新增 10 kubectl label nodes k8s-master busi-use=www --overwrite 11 kubectl label nodes k8s-master disk-type=ssd --overwrite 12 kubectl label nodes k8s-master busi-db=redis 13 14 # k8s-node01 標籤新增 15 kubectl label nodes k8s-node01 busi-use=www 16 kubectl label nodes k8s-node01 disk-type=sata 17 kubectl label nodes k8s-node01 busi-db=redis 18 19 # k8s-node02 標籤新增 20 kubectl label nodes k8s-node02 busi-use=www 21 kubectl label nodes k8s-node02 disk-type=ssd 22 kubectl label nodes k8s-node02 busi-db=etcd
查詢所有節點標籤資訊
1 [root@k8s-master ~]# kubectl get node -o wide --show-labels 2 NAME STATUS ROLES AGE VERSION INTERNAL-IP EXTERNAL-IP OS-IMAGE KERNEL-VERSION CONTAINER-RUNTIME LABELS 3 k8s-master Ready master 28d v1.17.4 172.16.1.110 <none> CentOS Linux 7 (Core) 3.10.0-1062.el7.x86_64 docker://19.3.8 beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,busi-db=redis,busi-use=www,disk-type=ssd,kubernetes.io/arch=amd64,kubernetes.io/hostname=k8s-master,kubernetes.io/os=linux,node-role.kubernetes.io/master= 4 k8s-node01 Ready <none> 28d v1.17.4 172.16.1.111 <none> CentOS Linux 7 (Core) 3.10.0-1062.el7.x86_64 docker://19.3.8 beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,busi-db=redis,busi-use=www,disk-type=sata,kubernetes.io/arch=amd64,kubernetes.io/hostname=k8s-node01,kubernetes.io/os=linux 5 k8s-node02 Ready <none> 28d v1.17.4 172.16.1.112 <none> CentOS Linux 7 (Core) 3.10.0-1062.el7.x86_64 docker://19.3.8 beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,busi-db=etcd,busi-use=www,disk-type=ssd,kubernetes.io/arch=amd64,kubernetes.io/hostname=k8s-node02,kubernetes.io/os=linux
如上所述：k8s-master添加了disk-type=ssd,busi-db=redis,busi-use=www標籤
k8s-node01添加了disk-type=sata,busi-db=redis,busi-use=www標籤
k8s-node02添加了disk-type=ssd,busi-db=etcd,busi-use=www標籤
通過deployment執行一個pod，或者直接執行一個pod也可以。為後續的Pod親和性與反親和性測驗做基礎。
1 ### yaml檔案 2 [root@k8s-master podAffinity]# pwd 3 /root/k8s_practice/scheduler/podAffinity 4 [root@k8s-master podAffinity]# cat web_deploy.yaml 5 apiVersion: apps/v1 6 kind: Deployment 7 metadata: 8 name: web-deploy 9 labels: 10 app: myweb-deploy 11 spec: 12 replicas: 1 13 selector: 14 matchLabels: 15 app: myapp-web 16 template: 17 metadata: 18 labels: 19 app: myapp-web 20 version: v1 21 spec: 22 containers: 23 - name: myapp-pod 24 image: registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 25 imagePullPolicy: IfNotPresent 26 ports: 27 - containerPort: 80 28 [root@k8s-master podAffinity]# 29 ### 執行yaml檔案 30 [root@k8s-master podAffinity]# kubectl apply -f web_deploy.yaml 31 deployment.apps/web-deploy created 32 [root@k8s-master podAffinity]# 33 ### 檢視pod標籤 34 [root@k8s-master podAffinity]# kubectl get pod -o wide --show-labels 35 NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES LABELS 36 web-deploy-5ccc9d7c55-kkwst 1/1 Running 0 15m 10.244.2.4 k8s-node02 <none> <none> app=myapp-web,pod-template-hash=5ccc9d7c55,version=v1
當前pod在k8s-node02節點；其中pod的標籤app=myapp-web,version=v1會在後面pod親和性/反親和性示例中使用。
pod硬親和性示例
要執行的yaml檔案
1 [root@k8s-master podAffinity]# pwd 2 /root/k8s_practice/scheduler/podAffinity 3 [root@k8s-master podAffinity]# cat pod_required_affinity.yaml 4 apiVersion: apps/v1 5 kind: Deployment 6 metadata: 7 name: pod-podaffinity-deploy 8 labels: 9 app: podaffinity-deploy 10 spec: 11 replicas: 6 12 selector: 13 matchLabels: 14 app: myapp 15 template: 16 metadata: 17 labels: 18 app: myapp 19 spec: 20 # 允許在master節點執行 21 tolerations: 22 - key: node-role.kubernetes.io/master 23 effect: NoSchedule 24 containers: 25 - name: myapp-pod 26 image: registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 27 imagePullPolicy: IfNotPresent 28 ports: 29 - containerPort: 80 30 affinity: 31 podAffinity: 32 requiredDuringSchedulingIgnoredDuringExecution: 33 - labelSelector: 34 # 由於是Pod親和性/反親和性；因此這裡匹配規則寫的是Pod的標籤資訊 35 matchExpressions: 36 - key: app 37 operator: In 38 values: 39 - myapp-web 40 # 拓撲域 若多個node節點具有相同的標籤資訊【標籤鍵值相同】，則表示這些node節點就在同一拓撲域 41 # 請對比如下兩個不同的拓撲域，Pod的排程結果 42 #topologyKey: busi-use 43 topologyKey: disk-type
執行yaml檔案並檢視狀態
1 [root@k8s-master podAffinity]# kubectl apply -f pod_required_affinity.yaml 2 deployment.apps/pod-podaffinity-deploy created 3 [root@k8s-master podAffinity]# 4 [root@k8s-master podAffinity]# kubectl get deploy -o wide 5 NAME READY UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTOR 6 pod-podaffinity-deploy 6/6 6 6 48s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp 7 web-deploy 1/1 1 1 22h myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp-web 8 [root@k8s-master podAffinity]# 9 [root@k8s-master podAffinity]# kubectl get rs -o wide 10 NAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTOR 11 pod-podaffinity-deploy-848559bf5b 6 6 6 52s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp,pod-template-hash=848559bf5b 12 web-deploy-5ccc9d7c55 1 1 1 22h myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp-web,pod-template-hash=5ccc9d7c55 13 [root@k8s-master podAffinity]# 14 [root@k8s-master podAffinity]# kubectl get pod -o wide 15 NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES 16 pod-podaffinity-deploy-848559bf5b-8kkwm 1/1 Running 0 54s 10.244.0.80 k8s-master <none> <none> 17 pod-podaffinity-deploy-848559bf5b-8s59f 1/1 Running 0 54s 10.244.2.252 k8s-node02 <none> <none> 18 pod-podaffinity-deploy-848559bf5b-8z4dv 1/1 Running 0 54s 10.244.2.253 k8s-node02 <none> <none> 19 pod-podaffinity-deploy-848559bf5b-gs7sb 1/1 Running 0 54s 10.244.0.79 k8s-master <none> <none> 20 pod-podaffinity-deploy-848559bf5b-sm6nz 1/1 Running 0 54s 10.244.0.78 k8s-master <none> <none> 21 pod-podaffinity-deploy-848559bf5b-zbr6v 1/1 Running 0 54s 10.244.2.251 k8s-node02 <none> <none> 22 web-deploy-5ccc9d7c55-khhrr 1/1 Running 3 22h 10.244.2.245 k8s-node02 <none> <none>
由上可見，yaml檔案中為topologyKey: disk-type；雖然k8s-master、k8s-node01、k8s-node02都有disk-type標籤；但是k8s-master和k8s-node02節點的disk-type標籤值為ssd；而k8s-node01節點的disk-type標籤值為sata。因此k8s-master和k8s-node02節點屬於同一拓撲域，Pod只會排程到這兩個節點上。
pod軟親和性示例
要執行的yaml檔案
1 [root@k8s-master podAffinity]# pwd 2 /root/k8s_practice/scheduler/podAffinity 3 [root@k8s-master podAffinity]# 4 [root@k8s-master podAffinity]# cat pod_preferred_affinity.yaml 5 apiVersion: apps/v1 6 kind: Deployment 7 metadata: 8 name: pod-podaffinity-deploy 9 labels: 10 app: podaffinity-deploy 11 spec: 12 replicas: 6 13 selector: 14 matchLabels: 15 app: myapp 16 template: 17 metadata: 18 labels: 19 app: myapp 20 spec: 21 # 允許在master節點執行 22 tolerations: 23 - key: node-role.kubernetes.io/master 24 effect: NoSchedule 25 containers: 26 - name: myapp-pod 27 image: registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 28 imagePullPolicy: IfNotPresent 29 ports: 30 - containerPort: 80 31 affinity: 32 podAffinity: 33 preferredDuringSchedulingIgnoredDuringExecution: 34 - weight: 100 35 podAffinityTerm: 36 labelSelector: 37 # 由於是Pod親和性/反親和性；因此這裡匹配規則寫的是Pod的標籤資訊 38 matchExpressions: 39 - key: version 40 operator: In 41 values: 42 - v1 43 - v2 44 # 拓撲域 若多個node節點具有相同的標籤資訊【標籤鍵值相同】，則表示這些node節點就在同一拓撲域 45 topologyKey: disk-type
執行yaml檔案並檢視狀態
1 [root@k8s-master podAffinity]# kubectl apply -f pod_preferred_affinity.yaml 2 deployment.apps/pod-podaffinity-deploy created 3 [root@k8s-master podAffinity]# 4 [root@k8s-master podAffinity]# kubectl get deploy -o wide 5 NAME READY UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTOR 6 pod-podaffinity-deploy 6/6 6 6 75s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp 7 web-deploy 1/1 1 1 25h myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp-web 8 [root@k8s-master podAffinity]# 9 [root@k8s-master podAffinity]# kubectl get rs -o wide 10 NAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTOR 11 pod-podaffinity-deploy-8474b4b586 6 6 6 79s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp,pod-template-hash=8474b4b586 12 web-deploy-5ccc9d7c55 1 1 1 25h myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp-web,pod-template-hash=5ccc9d7c55 13 [root@k8s-master podAffinity]# 14 [root@k8s-master podAffinity]# kubectl get pod -o wide 15 NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES 16 pod-podaffinity-deploy-8474b4b586-57gxh 1/1 Running 0 83s 10.244.2.4 k8s-node02 <none> <none> 17 pod-podaffinity-deploy-8474b4b586-kd5l4 1/1 Running 0 83s 10.244.2.3 k8s-node02 <none> <none> 18 pod-podaffinity-deploy-8474b4b586-mlvv7 1/1 Running 0 83s 10.244.0.84 k8s-master <none> <none> 19 pod-podaffinity-deploy-8474b4b586-mtk6r 1/1 Running 0 83s 10.244.0.86 k8s-master <none> <none> 20 pod-podaffinity-deploy-8474b4b586-n5jpj 1/1 Running 0 83s 10.244.0.85 k8s-master <none> <none> 21 pod-podaffinity-deploy-8474b4b586-q2xdl 1/1 Running 0 83s 10.244.3.22 k8s-node01 <none> <none> 22 web-deploy-5ccc9d7c55-khhrr 1/1 Running 3 25h 10.244.2.245 k8s-node02 <none> <none>
由上可見，再根據k8s-master、k8s-node01、k8s-node02的標籤資訊；很容易推斷出Pod會優先排程到k8s-master、k8s-node02節點。
pod硬反親和性示例
要執行的yaml檔案
1 [root@k8s-master podAffinity]# pwd 2 /root/k8s_practice/scheduler/podAffinity 3 [root@k8s-master podAffinity]# 4 [root@k8s-master podAffinity]# cat pod_required_AntiAffinity.yaml 5 apiVersion: apps/v1 6 kind: Deployment 7 metadata: 8 name: pod-podantiaffinity-deploy 9 labels: 10 app: podantiaffinity-deploy 11 spec: 12 replicas: 6 13 selector: 14 matchLabels: 15 app: myapp 16 template: 17 metadata: 18 labels: 19 app: myapp 20 spec: 21 # 允許在master節點執行 22 tolerations: 23 - key: node-role.kubernetes.io/master 24 effect: NoSchedule 25 containers: 26 - name: myapp-pod 27 image: registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 28 imagePullPolicy: IfNotPresent 29 ports: 30 - containerPort: 80 31 affinity: 32 podAntiAffinity: 33 requiredDuringSchedulingIgnoredDuringExecution: 34 - labelSelector: 35 # 由於是Pod親和性/反親和性；因此這裡匹配規則寫的是Pod的標籤資訊 36 matchExpressions: 37 - key: app 38 operator: In 39 values: 40 - myapp-web 41 # 拓撲域 若多個node節點具有相同的標籤資訊【標籤鍵值相同】，則表示這些node節點就在同一拓撲域 42 topologyKey: disk-type
執行yaml檔案並檢視狀態
1 [root@k8s-master podAffinity]# kubectl apply -f pod_required_AntiAffinity.yaml 2 deployment.apps/pod-podantiaffinity-deploy created 3 [root@k8s-master podAffinity]# 4 [root@k8s-master podAffinity]# kubectl get deploy -o wide 5 NAME READY UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTOR 6 pod-podantiaffinity-deploy 6/6 6 6 68s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp 7 web-deploy 1/1 1 1 25h myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp-web 8 [root@k8s-master podAffinity]# 9 [root@k8s-master podAffinity]# kubectl get rs -o wide 10 NAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTOR 11 pod-podantiaffinity-deploy-5fb4764b6b 6 6 6 72s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp,pod-template-hash=5fb4764b6b 12 web-deploy-5ccc9d7c55 1 1 1 25h myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp-web,pod-template-hash=5ccc9d7c55 13 [root@k8s-master podAffinity]# 14 [root@k8s-master podAffinity]# kubectl get pod -o wide 15 NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES 16 pod-podantiaffinity-deploy-5fb4764b6b-b5bzd 1/1 Running 0 75s 10.244.3.28 k8s-node01 <none> <none> 17 pod-podantiaffinity-deploy-5fb4764b6b-b6qjg 1/1 Running 0 75s 10.244.3.23 k8s-node01 <none> <none> 18 pod-podantiaffinity-deploy-5fb4764b6b-h262g 1/1 Running 0 75s 10.244.3.27 k8s-node01 <none> <none> 19 pod-podantiaffinity-deploy-5fb4764b6b-q98gt 1/1 Running 0 75s 10.244.3.24 k8s-node01 <none> <none> 20 pod-podantiaffinity-deploy-5fb4764b6b-v6kpm 1/1 Running 0 75s 10.244.3.25 k8s-node01 <none> <none> 21 pod-podantiaffinity-deploy-5fb4764b6b-wtmm6 1/1 Running 0 75s 10.244.3.26 k8s-node01 <none> <none> 22 web-deploy-5ccc9d7c55-khhrr 1/1 Running 3 25h 10.244.2.245 k8s-node02 <none> <none>
由上可見，由於是Pod反親和測驗，再根據k8s-master、k8s-node01、k8s-node02的標籤資訊；很容易推斷出Pod只能排程到k8s-node01節點。
pod軟反親和性示例
要執行的yaml檔案
1 [root@k8s-master podAffinity]# pwd 2 /root/k8s_practice/scheduler/podAffinity 3 [root@k8s-master podAffinity]# 4 [root@k8s-master podAffinity]# cat pod_preferred_AntiAffinity.yaml 5 apiVersion: apps/v1 6 kind: Deployment 7 metadata: 8 name: pod-podantiaffinity-deploy 9 labels: 10 app: podantiaffinity-deploy 11 spec: 12 replicas: 6 13 selector: 14 matchLabels: 15 app: myapp 16 template: 17 metadata: 18 labels: 19 app: myapp 20 spec: 21 # 允許在master節點執行 22 tolerations: 23 - key: node-role.kubernetes.io/master 24 effect: NoSchedule 25 containers: 26 - name: myapp-pod 27 image: registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 28 imagePullPolicy: IfNotPresent 29 ports: 30 - containerPort: 80 31 affinity: 32 podAntiAffinity: 33 preferredDuringSchedulingIgnoredDuringExecution: 34 - weight: 100 35 podAffinityTerm: 36 labelSelector: 37 # 由於是Pod親和性/反親和性；因此這裡匹配規則寫的是Pod的標籤資訊 38 matchExpressions: 39 - key: version 40 operator: In 41 values: 42 - v1 43 - v2 44 # 拓撲域 若多個node節點具有相同的標籤資訊【標籤鍵值相同】，則表示這些node節點就在同一拓撲域 45 topologyKey: disk-type
執行yaml檔案並檢視狀態
1 [root@k8s-master podAffinity]# kubectl apply -f pod_preferred_AntiAffinity.yaml 2 deployment.apps/pod-podantiaffinity-deploy created 3 [root@k8s-master podAffinity]# 4 [root@k8s-master podAffinity]# kubectl get deploy -o wide 5 NAME READY UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTOR 6 pod-podantiaffinity-deploy 6/6 6 6 9s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp 7 web-deploy 1/1 1 1 26h myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp-web 8 [root@k8s-master podAffinity]# 9 [root@k8s-master podAffinity]# kubectl get rs -o wide 10 NAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTOR 11 pod-podantiaffinity-deploy-54d758ddb4 6 6 6 13s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp,pod-template-hash=54d758ddb4 12 web-deploy-5ccc9d7c55 1 1 1 26h myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp-web,pod-template-hash=5ccc9d7c55 13 [root@k8s-master podAffinity]# 14 [root@k8s-master podAffinity]# kubectl get pod -o wide 15 NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES 16 pod-podantiaffinity-deploy-54d758ddb4-58t9p 1/1 Running 0 17s 10.244.3.31 k8s-node01 <none> <none> 17 pod-podantiaffinity-deploy-54d758ddb4-9ntd7 1/1 Running 0 17s 10.244.3.32 k8s-node01 <none> <none> 18 pod-podantiaffinity-deploy-54d758ddb4-9wr6p 1/1 Running 0 17s 10.244.2.5 k8s-node02 <none> <none> 19 pod-podantiaffinity-deploy-54d758ddb4-gnls4 1/1 Running 0 17s 10.244.3.30 k8s-node01 <none> <none> 20 pod-podantiaffinity-deploy-54d758ddb4-jlftn 1/1 Running 0 17s 10.244.3.29 k8s-node01 <none> <none> 21 pod-podantiaffinity-deploy-54d758ddb4-mvplv 1/1 Running 0 17s 10.244.0.87 k8s-master <none> <none> 22 web-deploy-5ccc9d7c55-khhrr 1/1 Running 3 26h 10.244.2.245 k8s-node02 <none> <none>
由上可見，由於是Pod反親和測驗，再根據k8s-master、k8s-node01、k8s-node02的標籤資訊；很容易推斷出Pod會優先排程到k8s-node01節點。
pod親和性與反親和性聯合示例
要執行的yaml檔案
1 [root@k8s-master podAffinity]# pwd 2 /root/k8s_practice/scheduler/podAffinity 3 [root@k8s-master podAffinity]# 4 [root@k8s-master podAffinity]# cat pod_podAffinity_all.yaml 5 apiVersion: apps/v1 6 kind: Deployment 7 metadata: 8 name: pod-podaffinity-all-deploy 9 labels: 10 app: podaffinity-all-deploy 11 spec: 12 replicas: 6 13 selector: 14 matchLabels: 15 app: myapp 16 template: 17 metadata: 18 labels: 19 app: myapp 20 spec: 21 # 允許在master節點執行 22 tolerations: 23 - key: node-role.kubernetes.io/master 24 effect: NoSchedule 25 containers: 26 - name: myapp-pod 27 image: registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 28 imagePullPolicy: IfNotPresent 29 ports: 30 - containerPort: 80 31 affinity: 32 podAffinity: 33 requiredDuringSchedulingIgnoredDuringExecution: 34 - labelSelector: 35 # 由於是Pod親和性/反親和性；因此這裡匹配規則寫的是Pod的標籤資訊 36 matchExpressions: 37 - key: app 38 operator: In 39 values: 40 - myapp-web 41 # 拓撲域 若多個node節點具有相同的標籤資訊【標籤鍵值相同】，則表示這些node節點就在同一拓撲域 42 topologyKey: disk-type 43 podAntiAffinity: 44 preferredDuringSchedulingIgnoredDuringExecution: 45 - weight: 100 46 podAffinityTerm: 47 labelSelector: 48 matchExpressions: 49 - key: version 50 operator: In 51 values: 52 - v1 53 - v2 54 topologyKey: busi-db
執行yaml檔案並檢視狀態
1 [root@k8s-master podAffinity]# kubectl apply -f pod_podAffinity_all.yaml 2 deployment.apps/pod-podaffinity-all-deploy created 3 [root@k8s-master podAffinity]# 4 [root@k8s-master podAffinity]# kubectl get deploy -o wide 5 NAME READY UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTOR 6 pod-podaffinity-all-deploy 6/6 6 1 5s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp 7 web-deploy 1/1 1 1 28h myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp-web 8 [root@k8s-master podAffinity]# 9 [root@k8s-master podAffinity]# kubectl get rs -o wide 10 NAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTOR 11 pod-podaffinity-all-deploy-5ddbf9cbf8 6 6 6 10s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp,pod-template-hash=5ddbf9cbf8 12 web-deploy-5ccc9d7c55 1 1 1 28h myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp-web,pod-template-hash=5ccc9d7c55 13 [root@k8s-master podAffinity]# 14 [root@k8s-master podAffinity]# kubectl get pod -o wide 15 NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES 16 pod-podaffinity-all-deploy-5ddbf9cbf8-5w5b7 1/1 Running 0 15s 10.244.0.91 k8s-master <none> <none> 17 pod-podaffinity-all-deploy-5ddbf9cbf8-j57g9 1/1 Running 0 15s 10.244.0.90 k8s-master <none> <none> 18 pod-podaffinity-all-deploy-5ddbf9cbf8-kwz6w 1/1 Running 0 15s 10.244.0.92 k8s-master <none> <none> 19 pod-podaffinity-all-deploy-5ddbf9cbf8-l8spj 1/1 Running 0 15s 10.244.2.6 k8s-node02 <none> <none> 20 pod-podaffinity-all-deploy-5ddbf9cbf8-lf22c 1/1 Running 0 15s 10.244.0.89 k8s-master <none> <none> 21 pod-podaffinity-all-deploy-5ddbf9cbf8-r2fgl 1/1 Running 0 15s 10.244.0.88 k8s-master <none> <none> 22 web-deploy-5ccc9d7c55-khhrr 1/1 Running 3 28h 10.244.2.245 k8s-node02 <none> <none>
由上可見，根據k8s-master、k8s-node01、k8s-node02的標籤資訊；很容易推斷出Pod只能排程到k8s-master、k8s-node02節點，且會優先排程到k8s-master節點。
相關閱讀
2、Kubernetes K8S排程器kube-scheduler詳解
完畢！
———END———
如果覺得不錯就關注下唄 (-^O^-) ！