Kubernetes K8S之affinity亲和性与反亲和性详解与示例
Kubernetes K8S之Node节点亲和性与反亲和性以及Pod亲和性与反亲和性详解与示例
主机配置规划
| 服务器名称(hostname) | 系统版本 | 配置 | 内网IP | 外网IP(模拟) |
|---|---|---|---|---|
| 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 |
亲和性和反亲和性
nodeSelector提供了一种非常简单的方法，将pods约束到具有特定标签的节点。而亲和性/反亲和性极大地扩展了可表达的约束类型。关键的增强是：
1、亲和性/反亲和性语言更具表达性。除了使用逻辑AND操作创建的精确匹配之外，该语言还提供了更多的匹配规则；
2、可以指示规则是优选项而不是硬要求，因此如果调度器不能满足，pod仍将被调度；
3、可以针对节点（或其他拓扑域）上运行的pods的标签进行约束，而不是针对节点的自身标签，这影响哪些Pod可以或不可以共处。
亲和特性包括两种类型：node节点亲和性/反亲和性 和 pod亲和性/反亲和性。pod亲和性/反亲和性约束针对的是pod标签而不是节点标签。
拓扑域是什么：多个node节点，拥有相同的label标签【节点标签的键值相同】，那么这些节点就处于同一个拓扑域。★★★★★
node节点亲和性
当前有两种类型的节点亲和性，称为requiredDuringSchedulingIgnoredDuringExecution和 preferredDuringSchedulingIgnoredDuringExecution，可以将它们分别视为“硬”【必须满足条件】和“软”【优选满足条件】要求。
前者表示Pod要调度到的节点必须满足规则条件，不满足则不会调度，pod会一直处于Pending状态；后者表示优先调度到满足规则条件的节点，如果不能满足再调度到其他节点。
名称中的 IgnoredDuringExecution 部分意味着，与nodeSelector的工作方式类似，如果节点上的标签在Pod运行时发生更改，使得pod上的亲和性规则不再满足，那么pod仍将继续在该节点上运行。
在未来，会计划提供requiredDuringSchedulingRequiredDuringExecution，类似requiredDuringSchedulingIgnoredDuringExecution。不同之处就是pod运行过程中如果节点不再满足pod的亲和性，则pod会在该节点中逐出。
节点亲和性语法支持以下运算符：In，NotIn，Exists，DoesNotExist，Gt，Lt。可以使用NotIn和DoesNotExist实现节点的反亲和行为。
运算符关系：
- In：label的值在某个列表中
- NotIn：label的值不在某个列表中
- Gt：label的值大于某个值
- Lt：label的值小于某个值
- Exists：某个label存在
- DoesNotExist：某个label不存在
其他重要说明：
1、如果同时指定nodeSelector和nodeAffinity，则必须满足两个条件，才能将Pod调度到候选节点上。
2、如果在nodeAffinity类型下指定了多个nodeSelectorTerms对象【对象不能有多个，如果存在多个只有最后一个生效】，那么只有最后一个nodeSelectorTerms对象生效。
3、如果在nodeSelectorTerms下指定了多个matchExpressions列表，那么只要能满足其中一个matchExpressions，就可以将pod调度到某个节点上【针对节点硬亲和】。
4、如果在matchExpressions下有多个key列表，那么只有当所有key满足时，才能将pod调度到某个节点【针对硬亲和】。
5、在key下的values只要有一个满足条件，那么当前的key就满足条件
6、如果pod已经调度在该节点，当我们删除或修该节点的标签时，pod不会被移除。换句话说，亲和性选择只有在pod调度期间有效。
7、preferredDuringSchedulingIgnoredDuringExecution中的weight（权重）字段在1-100范围内。对于每个满足所有调度需求的节点(资源请求、RequiredDuringScheduling亲和表达式等)，调度器将通过迭代该字段的元素来计算一个总和，如果节点与相应的匹配表达式匹配，则向该总和添加“权重”。然后将该分数与节点的其他优先级函数的分数结合起来。总得分最高的节点是最受欢迎的。
node节点亲和性示例
准备事项
给node节点打label标签
### --overwrite覆盖已存在的标签信息
kubectl label nodes k8s-node01 disk-type=ssd --overwrite
kubectl label nodes k8s-node01 cpu-num=12
kubectl label nodes k8s-node02 disk-type=sata
kubectl label nodes k8s-node02 cpu-num=24
查询所有节点标签信息
[root@k8s-master ~]# kubectl get node -o wide --show-labels
NAME STATUS ROLES AGE VERSION INTERNAL-IP EXTERNAL-IP OS-IMAGE KERNEL-VERSION CONTAINER-RUNTIME LABELS
k8s-master Ready master 43d v1.17.4 172.16.1.110 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=
k8s-node01 Ready 43d v1.17.4 172.16.1.111 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
k8s-node02 Ready 43d v1.17.4 172.16.1.112 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
参见上面，给k8s-node01打了cpu-num=12,disk-type=ssd标签；给k8s-node02打了cpu-num=24,disk-type=sata标签。
node硬亲和性示例
必须满足条件才能调度，否则不会调度
要运行的yaml文件
[root@k8s-master nodeAffinity]# pwd
/root/k8s_practice/scheduler/nodeAffinity
[root@k8s-master nodeAffinity]# cat node_required_affinity.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
name: node-affinity-deploy
labels:
app: nodeaffinity-deploy
spec:
replicas: 5
selector:
matchLabels:
app: myapp
template:
metadata:
labels:
app: myapp
spec:
containers:
- name: myapp-pod
image: registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1
imagePullPolicy: IfNotPresent
ports:
- containerPort: 80
affinity:
nodeAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
nodeSelectorTerms:
- matchExpressions:
# 表示node标签存在 disk-type=ssd 或 disk-type=sas
- key: disk-type
operator: In
values:
- ssd
- sas
# 表示node标签存在 cpu-num且值大于6
- key: cpu-num
operator: Gt
values:
- "6"
运行yaml文件并查看状态
[root@k8s-master nodeAffinity]# kubectl apply -f node_required_affinity.yaml
deployment.apps/node-affinity-deploy created
[root@k8s-master nodeAffinity]#
[root@k8s-master nodeAffinity]# kubectl get deploy -o wide
NAME READY UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTOR
node-affinity-deploy 5/5 5 5 6s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp
[root@k8s-master nodeAffinity]#
[root@k8s-master nodeAffinity]# kubectl get rs -o wide
NAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTOR
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
[root@k8s-master nodeAffinity]#
[root@k8s-master nodeAffinity]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
node-affinity-deploy-5c88ffb8ff-2mbfl 1/1 Running 0 15s 10.244.4.237 k8s-node01
node-affinity-deploy-5c88ffb8ff-9hjhk 1/1 Running 0 15s 10.244.4.235 k8s-node01
node-affinity-deploy-5c88ffb8ff-9rg75 1/1 Running 0 15s 10.244.4.239 k8s-node01
node-affinity-deploy-5c88ffb8ff-pqtfh 1/1 Running 0 15s 10.244.4.236 k8s-node01
node-affinity-deploy-5c88ffb8ff-zqpl8 1/1 Running 0 15s 10.244.4.238 k8s-node01
由上可见，再根据之前打的标签，很容易推断出当前pod只能调度在k8s-node01节点。
即使我们删除原来的rs，重新生成rs后pod依旧会调度到k8s-node01节点。如下：
[root@k8s-master nodeAffinity]# kubectl delete rs node-affinity-deploy-5c88ffb8ff
replicaset.apps "node-affinity-deploy-5c88ffb8ff" deleted
[root@k8s-master nodeAffinity]#
[root@k8s-master nodeAffinity]# kubectl get rs -o wide
NAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTOR
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
[root@k8s-master nodeAffinity]#
[root@k8s-master nodeAffinity]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
node-affinity-deploy-5c88ffb8ff-2v2tb 1/1 Running 0 11s 10.244.4.241 k8s-node01
node-affinity-deploy-5c88ffb8ff-gl4fm 1/1 Running 0 11s 10.244.4.240 k8s-node01
node-affinity-deploy-5c88ffb8ff-j26rg 1/1 Running 0 11s 10.244.4.244 k8s-node01
node-affinity-deploy-5c88ffb8ff-vhzmn 1/1 Running 0 11s 10.244.4.243 k8s-node01
node-affinity-deploy-5c88ffb8ff-xxj8m 1/1 Running 0 11s 10.244.4.242 k8s-node01
node软亲和性示例
优先调度到满足条件的节点，如果都不满足也会调度到其他节点。
要运行的yaml文件
[root@k8s-master nodeAffinity]# pwd
/root/k8s_practice/scheduler/nodeAffinity
[root@k8s-master nodeAffinity]# cat node_preferred_affinity.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
name: node-affinity-deploy
labels:
app: nodeaffinity-deploy
spec:
replicas: 5
selector:
matchLabels:
app: myapp
template:
metadata:
labels:
app: myapp
spec:
containers:
- name: myapp-pod
image: registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1
imagePullPolicy: IfNotPresent
ports:
- containerPort: 80
affinity:
nodeAffinity:
preferredDuringSchedulingIgnoredDuringExecution:
- weight: 1
preference:
matchExpressions:
# 表示node标签存在 disk-type=ssd 或 disk-type=sas
- key: disk-type
operator: In
values:
- ssd
- sas
- weight: 50
preference:
matchExpressions:
# 表示node标签存在 cpu-num且值大于16
- key: cpu-num
operator: Gt
values:
- "16"
运行yaml文件并查看状态
[root@k8s-master nodeAffinity]# kubectl apply -f node_preferred_affinity.yaml
deployment.apps/node-affinity-deploy created
[root@k8s-master nodeAffinity]#
[root@k8s-master nodeAffinity]# kubectl get deploy -o wide
NAME READY UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTOR
node-affinity-deploy 5/5 5 5 9s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp
[root@k8s-master nodeAffinity]#
[root@k8s-master nodeAffinity]# kubectl get rs -o wide
NAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTOR
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
[root@k8s-master nodeAffinity]#
[root@k8s-master nodeAffinity]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
node-affinity-deploy-d5d9cbc8d-bv86t 1/1 Running 0 18s 10.244.2.243 k8s-node02
node-affinity-deploy-d5d9cbc8d-dnbr8 1/1 Running 0 18s 10.244.2.244 k8s-node02
node-affinity-deploy-d5d9cbc8d-ldq82 1/1 Running 0 18s 10.244.2.246 k8s-node02
node-affinity-deploy-d5d9cbc8d-nt74q 1/1 Running 0 18s 10.244.4.2 k8s-node01
node-affinity-deploy-d5d9cbc8d-rt5nb 1/1 Running 0 18s 10.244.2.245 k8s-node02
由上可见，再根据之前打的标签，很容易推断出当前pod会【优先】调度在k8s-node02节点。
node软硬亲和性联合示例
硬亲和性与软亲和性一起使用
要运行的yaml文件
[root@k8s-master nodeAffinity]# pwd
/root/k8s_practice/scheduler/nodeAffinity
[root@k8s-master nodeAffinity]# cat node_affinity.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
name: node-affinity-deploy
labels:
app: nodeaffinity-deploy
spec:
replicas: 5
selector:
matchLabels:
app: myapp
template:
metadata:
labels:
app: myapp
spec:
containers:
- name: myapp-pod
image: registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1
imagePullPolicy: IfNotPresent
ports:
- containerPort: 80
affinity:
nodeAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
nodeSelectorTerms:
# 表示node标签存在 cpu-num且值大于10
- matchExpressions:
- key: cpu-num
operator: Gt
values:
- "10"
preferredDuringSchedulingIgnoredDuringExecution:
- weight: 50
preference:
matchExpressions:
# 表示node标签存在 disk-type=ssd 或 disk-type=sas
- key: disk-type
operator: In
values:
- ssd
- sas
运行yaml文件并查看状态
[root@k8s-master nodeAffinity]# kubectl apply -f node_affinity.yaml
deployment.apps/node-affinity-deploy created
[root@k8s-master nodeAffinity]#
[root@k8s-master nodeAffinity]# kubectl get deploy -o wide
NAME READY UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTOR
node-affinity-deploy 5/5 5 5 9s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp
[root@k8s-master nodeAffinity]#
[root@k8s-master nodeAffinity]# kubectl get rs -o wide
NAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTOR
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
[root@k8s-master nodeAffinity]#
[root@k8s-master nodeAffinity]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
node-affinity-deploy-f9cb9b99b-8w2nc 1/1 Running 0 17s 10.244.4.10 k8s-node01
node-affinity-deploy-f9cb9b99b-csk2s 1/1 Running 0 17s 10.244.4.9 k8s-node01
node-affinity-deploy-f9cb9b99b-g42kq 1/1 Running 0 17s 10.244.4.8 k8s-node01
node-affinity-deploy-f9cb9b99b-m6xbv 1/1 Running 0 17s 10.244.4.7 k8s-node01
node-affinity-deploy-f9cb9b99b-mxbdp 1/1 Running 0 17s 10.244.2.253 k8s-node02
由上可见，再根据之前打的标签，很容易推断出k8s-node01、k8s-node02都满足必要条件，但当前pod会【优先】调度在k8s-node01节点。
Pod亲和性
与节点亲和性一样，当前有Pod亲和性/反亲和性都有两种类型，称为requiredDuringSchedulingIgnoredDuringExecution和 preferredDuringSchedulingIgnoredDuringExecution，分别表示“硬”与“软”要求。对于硬要求，如果不满足则pod会一直处于Pending状态。
Pod的亲和性与反亲和性是基于Node节点上已经运行pod的标签(而不是节点上的标签)决定的，从而约束哪些节点适合调度你的pod。
规则的形式是：如果X已经运行了一个或多个符合规则Y的pod，则此pod应该在X中运行(如果是反亲和的情况下，则不应该在X中运行）。当然pod必须处在同一名称空间，不然亲和性/反亲和性无作用。从概念上讲，X是一个拓扑域。我们可以使用topologyKey来表示它，topologyKey 的值是node节点标签的键以便系统用来表示这样的拓扑域。当然这里也有个隐藏条件，就是node节点标签的键值相同时，才是在同一拓扑域中；如果只是节点标签名相同，但是值不同，那么也不在同一拓扑域。★★★★★
也就是说：Pod的亲和性/反亲和性调度是根据拓扑域来界定调度的，而不是根据node节点。★★★★★
注意事项
1、pod之间亲和性/反亲和性需要大量的处理，这会明显降低大型集群中的调度速度。不建议在大于几百个节点的集群中使用它们。
2、Pod反亲和性要求对节点进行一致的标记。换句话说，集群中的每个节点都必须有一个匹配topologyKey的适当标签。如果某些或所有节点缺少指定的topologyKey标签，可能会导致意外行为。
requiredDuringSchedulingIgnoredDuringExecution中亲和性的一个示例是“将服务A和服务B的Pod放置在同一区域【拓扑域】中，因为它们之间有很多交流”；preferredDuringSchedulingIgnoredDuringExecution中反亲和性的示例是“将此服务的 pod 跨区域【拓扑域】分布”【此时硬性要求是说不通的，因为你可能拥有的 pod 数多于区域数】。
Pod亲和性/反亲和性语法支持以下运算符：In，NotIn，Exists，DoesNotExist。
原则上，topologyKey可以是任何合法的标签键。但是，出于性能和安全方面的原因，topologyKey有一些限制：
1、对于Pod亲和性，在requiredDuringSchedulingIgnoredDuringExecution和preferredDuringSchedulingIgnoredDuringExecution中topologyKey都不允许为空。
2、对于Pod反亲和性，在requiredDuringSchedulingIgnoredDuringExecution和preferredDuringSchedulingIgnoredDuringExecution中topologyKey也都不允许为空。
3、对于requiredDuringSchedulingIgnoredDuringExecution的pod反亲和性，引入了允许控制器LimitPodHardAntiAffinityTopology来限制topologyKey的kubernet.io/hostname。如果你想让它对自定义拓扑可用，你可以修改许可控制器，或者干脆禁用它。
4、除上述情况外，topologyKey可以是任何合法的标签键。
Pod 间亲和通过 PodSpec 中 affinity 字段下的 podAffinity 字段进行指定。而 pod 间反亲和通过 PodSpec 中 affinity 字段下的 podAntiAffinity 字段进行指定。
Pod亲和性/反亲和性的requiredDuringSchedulingIgnoredDuringExecution所关联的matchExpressions下有多个key列表，那么只有当所有key满足时，才能将pod调度到某个区域【针对Pod硬亲和】。
pod亲和性与反亲和性示例
为了更好的演示Pod亲和性与反亲和性，本次示例我们会将k8s-master节点也加入进来进行演示。
准备事项
给node节点打label标签
# 删除已存在标签
kubectl label nodes k8s-node01 cpu-num-
kubectl label nodes k8s-node01 disk-type-
kubectl label nodes k8s-node02 cpu-num-
kubectl label nodes k8s-node02 disk-type-
### --overwrite覆盖已存在的标签信息
# k8s-master 标签添加
kubectl label nodes k8s-master busi-use=www --overwrite
kubectl label nodes k8s-master disk-type=ssd --overwrite
kubectl label nodes k8s-master busi-db=redis
# k8s-node01 标签添加
kubectl label nodes k8s-node01 busi-use=www
kubectl label nodes k8s-node01 disk-type=sata
kubectl label nodes k8s-node01 busi-db=redis
# k8s-node02 标签添加
kubectl label nodes k8s-node02 busi-use=www
kubectl label nodes k8s-node02 disk-type=ssd
kubectl label nodes k8s-node02 busi-db=etcd
查询所有节点标签信息
[root@k8s-master ~]# kubectl get node -o wide --show-labels
NAME STATUS ROLES AGE VERSION INTERNAL-IP EXTERNAL-IP OS-IMAGE KERNEL-VERSION CONTAINER-RUNTIME LABELS
k8s-master Ready master 28d v1.17.4 172.16.1.110 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=
k8s-node01 Ready 28d v1.17.4 172.16.1.111 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
k8s-node02 Ready 28d v1.17.4 172.16.1.112 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标签
k8s-node01添加了disk-type=sata,busi-db=redis,busi-use=www标签
k8s-node02添加了disk-type=ssd,busi-db=etcd,busi-use=www标签
通过deployment运行一个pod，或者直接运行一个pod也可以。为后续的Pod亲和性与反亲和性测验做基础。
### yaml文件
[root@k8s-master podAffinity]# pwd
/root/k8s_practice/scheduler/podAffinity
[root@k8s-master podAffinity]# cat web_deploy.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
name: web-deploy
labels:
app: myweb-deploy
spec:
replicas: 1
selector:
matchLabels:
app: myapp-web
template:
metadata:
labels:
app: myapp-web
version: v1
spec:
containers:
- name: myapp-pod
image: registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1
imagePullPolicy: IfNotPresent
ports:
- containerPort: 80
[root@k8s-master podAffinity]#
### 运行yaml文件
[root@k8s-master podAffinity]# kubectl apply -f web_deploy.yaml
deployment.apps/web-deploy created
[root@k8s-master podAffinity]#
### 查看pod标签
[root@k8s-master podAffinity]# kubectl get pod -o wide --show-labels
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES LABELS
web-deploy-5ccc9d7c55-kkwst 1/1 Running 0 15m 10.244.2.4 k8s-node02 app=myapp-web,pod-template-hash=5ccc9d7c55,version=v1
当前pod在k8s-node02节点；其中pod的标签app=myapp-web,version=v1会在后面pod亲和性/反亲和性示例中使用。
pod硬亲和性示例
要运行的yaml文件
[root@k8s-master podAffinity]# pwd
/root/k8s_practice/scheduler/podAffinity
[root@k8s-master podAffinity]# cat pod_required_affinity.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
name: pod-podaffinity-deploy
labels:
app: podaffinity-deploy
spec:
replicas: 6
selector:
matchLabels:
app: myapp
template:
metadata:
labels:
app: myapp
spec:
# 允许在master节点运行
tolerations:
- key: node-role.kubernetes.io/master
effect: NoSchedule
containers:
- name: myapp-pod
image: registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1
imagePullPolicy: IfNotPresent
ports:
- containerPort: 80
affinity:
podAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
- labelSelector:
# 由于是Pod亲和性/反亲和性；因此这里匹配规则写的是Pod的标签信息
matchExpressions:
- key: app
operator: In
values:
- myapp-web
# 拓扑域 若多个node节点具有相同的标签信息【标签键值相同】，则表示这些node节点就在同一拓扑域
# 请对比如下两个不同的拓扑域，Pod的调度结果
#topologyKey: busi-use
topologyKey: disk-type
运行yaml文件并查看状态
[root@k8s-master podAffinity]# kubectl apply -f pod_required_affinity.yaml
deployment.apps/pod-podaffinity-deploy created
[root@k8s-master podAffinity]#
[root@k8s-master podAffinity]# kubectl get deploy -o wide
NAME READY UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTOR
pod-podaffinity-deploy 6/6 6 6 48s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp
web-deploy 1/1 1 1 22h myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp-web
[root@k8s-master podAffinity]#
[root@k8s-master podAffinity]# kubectl get rs -o wide
NAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTOR
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
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
[root@k8s-master podAffinity]#
[root@k8s-master podAffinity]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
pod-podaffinity-deploy-848559bf5b-8kkwm 1/1 Running 0 54s 10.244.0.80 k8s-master
pod-podaffinity-deploy-848559bf5b-8s59f 1/1 Running 0 54s 10.244.2.252 k8s-node02
pod-podaffinity-deploy-848559bf5b-8z4dv 1/1 Running 0 54s 10.244.2.253 k8s-node02
pod-podaffinity-deploy-848559bf5b-gs7sb 1/1 Running 0 54s 10.244.0.79 k8s-master
pod-podaffinity-deploy-848559bf5b-sm6nz 1/1 Running 0 54s 10.244.0.78 k8s-master
pod-podaffinity-deploy-848559bf5b-zbr6v 1/1 Running 0 54s 10.244.2.251 k8s-node02
web-deploy-5ccc9d7c55-khhrr 1/1 Running 3 22h 10.244.2.245 k8s-node02
由上可见，yaml文件中为topologyKey: disk-type；虽然k8s-master、k8s-node01、k8s-node02都有disk-type标签；但是k8s-master和k8s-node02节点的disk-type标签值为ssd；而k8s-node01节点的disk-type标签值为sata。因此k8s-master和k8s-node02节点属于同一拓扑域，Pod只会调度到这两个节点上。
pod软亲和性示例
要运行的yaml文件
[root@k8s-master podAffinity]# pwd
/root/k8s_practice/scheduler/podAffinity
[root@k8s-master podAffinity]#
[root@k8s-master podAffinity]# cat pod_preferred_affinity.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
name: pod-podaffinity-deploy
labels:
app: podaffinity-deploy
spec:
replicas: 6
selector:
matchLabels:
app: myapp
template:
metadata:
labels:
app: myapp
spec:
# 允许在master节点运行
tolerations:
- key: node-role.kubernetes.io/master
effect: NoSchedule
containers:
- name: myapp-pod
image: registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1
imagePullPolicy: IfNotPresent
ports:
- containerPort: 80
affinity:
podAffinity:
preferredDuringSchedulingIgnoredDuringExecution:
- weight: 100
podAffinityTerm:
labelSelector:
# 由于是Pod亲和性/反亲和性；因此这里匹配规则写的是Pod的标签信息
matchExpressions:
- key: version
operator: In
values:
- v1
- v2
# 拓扑域 若多个node节点具有相同的标签信息【标签键值相同】，则表示这些node节点就在同一拓扑域
topologyKey: disk-type
运行yaml文件并查看状态
[root@k8s-master podAffinity]# kubectl apply -f pod_preferred_affinity.yaml
deployment.apps/pod-podaffinity-deploy created
[root@k8s-master podAffinity]#
[root@k8s-master podAffinity]# kubectl get deploy -o wide
NAME READY UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTOR
pod-podaffinity-deploy 6/6 6 6 75s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp
web-deploy 1/1 1 1 25h myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp-web
[root@k8s-master podAffinity]#
[root@k8s-master podAffinity]# kubectl get rs -o wide
NAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTOR
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
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
[root@k8s-master podAffinity]#
[root@k8s-master podAffinity]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
pod-podaffinity-deploy-8474b4b586-57gxh 1/1 Running 0 83s 10.244.2.4 k8s-node02
pod-podaffinity-deploy-8474b4b586-kd5l4 1/1 Running 0 83s 10.244.2.3 k8s-node02
pod-podaffinity-deploy-8474b4b586-mlvv7 1/1 Running 0 83s 10.244.0.84 k8s-master
pod-podaffinity-deploy-8474b4b586-mtk6r 1/1 Running 0 83s 10.244.0.86 k8s-master
pod-podaffinity-deploy-8474b4b586-n5jpj 1/1 Running 0 83s 10.244.0.85 k8s-master
pod-podaffinity-deploy-8474b4b586-q2xdl 1/1 Running 0 83s 10.244.3.22 k8s-node01
web-deploy-5ccc9d7c55-khhrr 1/1 Running 3 25h 10.244.2.245 k8s-node02
由上可见，再根据k8s-master、k8s-node01、k8s-node02的标签信息；很容易推断出Pod会优先调度到k8s-master、k8s-node02节点。
pod硬反亲和性示例
要运行的yaml文件
[root@k8s-master podAffinity]# pwd
/root/k8s_practice/scheduler/podAffinity
[root@k8s-master podAffinity]#
[root@k8s-master podAffinity]# cat pod_required_AntiAffinity.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
name: pod-podantiaffinity-deploy
labels:
app: podantiaffinity-deploy
spec:
replicas: 6
selector:
matchLabels:
app: myapp
template:
metadata:
labels:
app: myapp
spec:
# 允许在master节点运行
tolerations:
- key: node-role.kubernetes.io/master
effect: NoSchedule
containers:
- name: myapp-pod
image: registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1
imagePullPolicy: IfNotPresent
ports:
- containerPort: 80
affinity:
podAntiAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
- labelSelector:
# 由于是Pod亲和性/反亲和性；因此这里匹配规则写的是Pod的标签信息
matchExpressions:
- key: app
operator: In
values:
- myapp-web
# 拓扑域 若多个node节点具有相同的标签信息【标签键值相同】，则表示这些node节点就在同一拓扑域
topologyKey: disk-type
运行yaml文件并查看状态
[root@k8s-master podAffinity]# kubectl apply -f pod_required_AntiAffinity.yaml
deployment.apps/pod-podantiaffinity-deploy created
[root@k8s-master podAffinity]#
[root@k8s-master podAffinity]# kubectl get deploy -o wide
NAME READY UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTOR
pod-podantiaffinity-deploy 6/6 6 6 68s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp
web-deploy 1/1 1 1 25h myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp-web
[root@k8s-master podAffinity]#
[root@k8s-master podAffinity]# kubectl get rs -o wide
NAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTOR
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
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
[root@k8s-master podAffinity]#
[root@k8s-master podAffinity]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
pod-podantiaffinity-deploy-5fb4764b6b-b5bzd 1/1 Running 0 75s 10.244.3.28 k8s-node01
pod-podantiaffinity-deploy-5fb4764b6b-b6qjg 1/1 Running 0 75s 10.244.3.23 k8s-node01
pod-podantiaffinity-deploy-5fb4764b6b-h262g 1/1 Running 0 75s 10.244.3.27 k8s-node01
pod-podantiaffinity-deploy-5fb4764b6b-q98gt 1/1 Running 0 75s 10.244.3.24 k8s-node01
pod-podantiaffinity-deploy-5fb4764b6b-v6kpm 1/1 Running 0 75s 10.244.3.25 k8s-node01
pod-podantiaffinity-deploy-5fb4764b6b-wtmm6 1/1 Running 0 75s 10.244.3.26 k8s-node01
web-deploy-5ccc9d7c55-khhrr 1/1 Running 3 25h 10.244.2.245 k8s-node02
由上可见，由于是Pod反亲和测验，再根据k8s-master、k8s-node01、k8s-node02的标签信息；很容易推断出Pod只能调度到k8s-node01节点。
pod软反亲和性示例
要运行的yaml文件
[root@k8s-master podAffinity]# pwd
/root/k8s_practice/scheduler/podAffinity
[root@k8s-master podAffinity]#
[root@k8s-master podAffinity]# cat pod_preferred_AntiAffinity.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
name: pod-podantiaffinity-deploy
labels:
app: podantiaffinity-deploy
spec:
replicas: 6
selector:
matchLabels:
app: myapp
template:
metadata:
labels:
app: myapp
spec:
# 允许在master节点运行
tolerations:
- key: node-role.kubernetes.io/master
effect: NoSchedule
containers:
- name: myapp-pod
image: registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1
imagePullPolicy: IfNotPresent
ports:
- containerPort: 80
affinity:
podAntiAffinity:
preferredDuringSchedulingIgnoredDuringExecution:
- weight: 100
podAffinityTerm:
labelSelector:
# 由于是Pod亲和性/反亲和性；因此这里匹配规则写的是Pod的标签信息
matchExpressions:
- key: version
operator: In
values:
- v1
- v2
# 拓扑域 若多个node节点具有相同的标签信息【标签键值相同】，则表示这些node节点就在同一拓扑域
topologyKey: disk-type
运行yaml文件并查看状态
[root@k8s-master podAffinity]# kubectl apply -f pod_preferred_AntiAffinity.yaml
deployment.apps/pod-podantiaffinity-deploy created
[root@k8s-master podAffinity]#
[root@k8s-master podAffinity]# kubectl get deploy -o wide
NAME READY UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTOR
pod-podantiaffinity-deploy 6/6 6 6 9s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp
web-deploy 1/1 1 1 26h myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp-web
[root@k8s-master podAffinity]#
[root@k8s-master podAffinity]# kubectl get rs -o wide
NAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTOR
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
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
[root@k8s-master podAffinity]#
[root@k8s-master podAffinity]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
pod-podantiaffinity-deploy-54d758ddb4-58t9p 1/1 Running 0 17s 10.244.3.31 k8s-node01
pod-podantiaffinity-deploy-54d758ddb4-9ntd7 1/1 Running 0 17s 10.244.3.32 k8s-node01
pod-podantiaffinity-deploy-54d758ddb4-9wr6p 1/1 Running 0 17s 10.244.2.5 k8s-node02
pod-podantiaffinity-deploy-54d758ddb4-gnls4 1/1 Running 0 17s 10.244.3.30 k8s-node01
pod-podantiaffinity-deploy-54d758ddb4-jlftn 1/1 Running 0 17s 10.244.3.29 k8s-node01
pod-podantiaffinity-deploy-54d758ddb4-mvplv 1/1 Running 0 17s 10.244.0.87 k8s-master
web-deploy-5ccc9d7c55-khhrr 1/1 Running 3 26h 10.244.2.245 k8s-node02
由上可见，由于是Pod反亲和测验，再根据k8s-master、k8s-node01、k8s-node02的标签信息；很容易推断出Pod会优先调度到k8s-node01节点。
pod亲和性与反亲和性联合示例
要运行的yaml文件
[root@k8s-master podAffinity]# pwd
/root/k8s_practice/scheduler/podAffinity
[root@k8s-master podAffinity]#
[root@k8s-master podAffinity]# cat pod_podAffinity_all.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
name: pod-podaffinity-all-deploy
labels:
app: podaffinity-all-deploy
spec:
replicas: 6
selector:
matchLabels:
app: myapp
template:
metadata:
labels:
app: myapp
spec:
# 允许在master节点运行
tolerations:
- key: node-role.kubernetes.io/master
effect: NoSchedule
containers:
- name: myapp-pod
image: registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1
imagePullPolicy: IfNotPresent
ports:
- containerPort: 80
affinity:
podAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
- labelSelector:
# 由于是Pod亲和性/反亲和性；因此这里匹配规则写的是Pod的标签信息
matchExpressions:
- key: app
operator: In
values:
- myapp-web
# 拓扑域 若多个node节点具有相同的标签信息【标签键值相同】，则表示这些node节点就在同一拓扑域
topologyKey: disk-type
podAntiAffinity:
preferredDuringSchedulingIgnoredDuringExecution:
- weight: 100
podAffinityTerm:
labelSelector:
matchExpressions:
- key: version
operator: In
values:
- v1
- v2
topologyKey: busi-db
运行yaml文件并查看状态
[root@k8s-master podAffinity]# kubectl apply -f pod_podAffinity_all.yaml
deployment.apps/pod-podaffinity-all-deploy created
[root@k8s-master podAffinity]#
[root@k8s-master podAffinity]# kubectl get deploy -o wide
NAME READY UP-TO-DATE AVAILABLE AGE CONTAINERS IMAGES SELECTOR
pod-podaffinity-all-deploy 6/6 6 1 5s myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp
web-deploy 1/1 1 1 28h myapp-pod registry.cn-beijing.aliyuncs.com/google_registry/myapp:v1 app=myapp-web
[root@k8s-master podAffinity]#
[root@k8s-master podAffinity]# kubectl get rs -o wide
NAME DESIRED CURRENT READY AGE CONTAINERS IMAGES SELECTOR
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
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
[root@k8s-master podAffinity]#
[root@k8s-master podAffinity]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
pod-podaffinity-all-deploy-5ddbf9cbf8-5w5b7 1/1 Running 0 15s 10.244.0.91 k8s-master
pod-podaffinity-all-deploy-5ddbf9cbf8-j57g9 1/1 Running 0 15s 10.244.0.90 k8s-master
pod-podaffinity-all-deploy-5ddbf9cbf8-kwz6w 1/1 Running 0 15s 10.244.0.92 k8s-master
pod-podaffinity-all-deploy-5ddbf9cbf8-l8spj 1/1 Running 0 15s 10.244.2.6 k8s-node02
pod-podaffinity-all-deploy-5ddbf9cbf8-lf22c 1/1 Running 0 15s 10.244.0.89 k8s-master
pod-podaffinity-all-deploy-5ddbf9cbf8-r2fgl 1/1 Running 0 15s 10.244.0.88 k8s-master
web-deploy-5ccc9d7c55-khhrr 1/1 Running 3 28h 10.244.2.245 k8s-node02
由上可见，根据k8s-master、k8s-node01、k8s-node02的标签信息；很容易推断出Pod只能调度到k8s-master、k8s-node02节点，且会优先调度到k8s-master节点。
相关阅读
1、官网：将Pods分配给节点
2、Kubernetes K8S调度器kube-scheduler详解
完毕！
