kubernetes资源调度

调度

创建一个pod的工作流程
kubernetes基于list-watch机制的控制器架构，实现组件间交互的解耦。
其他组件监控自己负责的资源，当这些资源发生变化时，kube-apiserver会通知这些组件，这个过程类似于发布与订阅。

• 用户提交创建Pod的请求，可以通过API Server的REST API ，也可用Kubectl命令行工具，支持Json和Yaml两种格式；
• API Server 处理用户请求，存储Pod数据到Etcd；
• Schedule通过和 API Server的watch机制，查看到新的pod，尝试为Pod绑定Node；
• 过滤主机：调度器用一组规则过滤掉不符合要求的主机，比如Pod指定了所需要的资源，那么就要过滤掉资源不够的主机；
• 主机打分：对第一步筛选出的符合要求的主机进行打分，在主机打分阶段，调度器会考虑一些整体优化策略，比如把一个Replication Controller的副本分布到不同的主机上，使用最低负载的主机等；
• 选择主机：选择打分最高的主机，进行binding操作，结果存储到Etcd中；
• kubelet根据调度结果执行Pod创建操作： 绑定成功后，会启动container, docker run, scheduler会调用API Server的API在etcd中创建一个bound pod对象，描述在一个工作节点上绑定运行的所有pod信息。运行在每个工作节点上的kubelet也会定期与etcd同步bound pod信息，一旦发现应该在该工作节点上运行的bound pod对象没有更新，则调用Docker API创建并启动pod内的容器。

nodeSelector

nodeSelector是最简单的约束方式。nodeSelector是pod.spec的一个字段

通过–show-labels可以查看指定node的labels

[root@master ~]# kubectl get node node1 --show-labels
NAME    STATUS     ROLES    AGE     VERSION   LABELS
node1   NotReady   <none>   4d21h   v1.20.0   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/arch=amd64,kubernetes.io/hostname=node1,kubernetes.io/os=linux

//如果没有额外添加 nodes labels，那么看到的如上所示的默认标签。我们可以通过 kubectl label node 命令给指定 node 添加 labels：
[root@master ~]# kubectl label node node1 disktype=ssd
node/node1 labeled
[root@master ~]# kubectl get node node1 --show-labels
NAME    STATUS     ROLES    AGE     VERSION   LABELS
node1   NotReady   <none>   4d21h   v1.20.0   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,disktype=ssd,kubernetes.io/arch=amd64,kubernetes.io/hostname=node1,kubernetes.io/os=linux

//也可以通过 kubectl label node 删除指定的 labels（标签 key 接 - 号即可）
[root@master ~]# kubectl label node node1 disktype-
node/node1 labeled
[root@master ~]# kubectl get node node1 --show-labels
NAME    STATUS     ROLES    AGE     VERSION   LABELS
node1   NotReady   <none>   4d21h   v1.20.0   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/arch=amd64,kubernetes.io/hostname=node1,kubernetes.io/os=linux

创建测试 pod 并指定 nodeSelector 选项绑定节点：

[root@master mainfest]# kubectl label node node1 disktype=ssd
node/node1 labeled
[root@master mainfest]# kubectl get nodes node1 --show-labels
NAME    STATUS   ROLES    AGE     VERSION   LABELS
node1   Ready    <none>   4d21h   v1.20.0   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,disktype=ssd,kubernetes.io/arch=amd64,kubernetes.io/hostname=node1,kubernetes.io/os=linux

[root@master mainfest]# vim test.yaml
apiVersion: v1
kind: Pod
metadata:
  name: test
  labels:
    env: test
spec:
  containers:
  - name: test
    image: nginx
    imagePullPolicy: IfNotPresent
  nodeSelector:
    disktype: ssd

//查看pod调度的节点
[root@master mainfest]# kubectl get pod -o wide
NAME   READY   STATUS    RESTARTS   AGE   IP            NODE    NOMINATED NODE   READINESS GATES
t1     1/1     Running   0          73s   10.244.1.31   node1   <none>           <none>
test   1/1     Running   0          6s    10.244.1.32   node1   <none>           <none>
test这个pod被强制调度到了有disktype=ssd这个label的node上。

nodeAffinity

nodeAffinity意为node亲和性调度策略。是用于替换nodeSelector的全新调度策略。目前有两种节点亲和性表达：

1、RequiredDuringSchedulingIgnoredDuringExecution:
必须满足制定的规则才可以调度pode到Node上。相当于硬限制；
2、PreferredDuringSchedulingIgnoreDuringExecution:
强调优先满足制定规则，调度器会尝试调度pod到Node上，但并不强求，相当于软限制。多个优先级规则还可以设置权重值，以定义执行的先后顺序。
IgnoredDuringExecution的意思是：

如果一个pod所在的节点在pod运行期间标签发生了变更，不在符合该pod的节点亲和性需求，则系统将忽略node上lable的变化，该pod能机选在该节点运行。

NodeAffinity 语法支持的操作符包括：

• In：label 的值在某个列表中
• NotIn：label 的值不在某个列表中
• Exists：某个 label 存在
• DoesNotExit：某个 label 不存在
• Gt：label 的值大于某个值
• Lt：label 的值小于某个值

nodeAffinity规则设置的注意事项如下：

如果同时定义了nodeSelector和nodeAffinity，name必须两个条件都得到满足，pod才能最终运行在指定的node上。
如果nodeAffinity指定了多个nodeSelectorTerms，那么其中一个能够匹配成功即可。
如果在nodeSelectorTerms中有多个matchExpressions，则一个节点必须满足所有matchExpressions才能运行该pod。

apiVersion: v1
kind: Pod
metadata:
  name: test1
  labels:
    app: nginx
spec:
  containers:
  - name: test1
    image: nginx
    imagePullPolicy: IfNotPresent
  affinity:
    nodeAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:		//硬策略
        nodeSelectorTerms:
        - matchExpressions:
          - key: disktype
            values:
            - ssd
            operator: In
      preferredDuringSchedulingIgnoredDuringExecution:		//软策略
      - weight: 10
        preference:
          matchExpressions:
          - key: name
            values:
            - test
            operator: In      
            
//给node2打上disktype=ssd的标签
[root@master mainfest]# kubectl label node node2 disktype=ssd
node/node2 labeled
[root@master mainfest]# kubectl get node node2 --show-labels
NAME    STATUS   ROLES    AGE     VERSION   LABELS
node2   Ready    <none>   4d22h   v1.20.0   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,disktype=ssd,kubernetes.io/arch=amd64,kubernetes.io/hostname=node2,kubernetes.io/os=linux

//给node1打上name=test的label和删除name=test的label并测试查看结果
[root@master ~]# kubectl label node node1 name=test
node/node1 labeled
[root@master ~]# kubectl get node node1 --show-labels
NAME    STATUS   ROLES    AGE     VERSION   LABELS
node1   Ready    <none>   4d22h   v1.20.0   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,disktype=ssd,kubernetes.io/arch=amd64,kubernetes.io/hostname=node1,kubernetes.io/os=linux,name=test

[root@master mainfest]# kubectl apply -f test1.yaml 
pod/test1 created
[root@master mainfest]# kubectl get pod -o wide
NAME    READY   STATUS    RESTARTS   AGE   IP            NODE    NOMINATED NODE   READINESS GATES
t1      1/1     Running   0          36m   10.244.1.31   node1   <none>           <none>
test    1/1     Running   0          35m   10.244.1.32   node1   <none>           <none>
test1   1/1     Running   0          3s    10.244.1.33   node1   <none>           <none>

//删除node1上name=test的label
[root@master mainfest]# kubectl label node node1 name-
node/node1 labeled
[root@master mainfest]# kubectl get node node1 --show-labels
NAME    STATUS   ROLES    AGE     VERSION   LABELS
node1   Ready    <none>   4d22h   v1.20.0   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,disktype=ssd,kubernetes.io/arch=amd64,kubernetes.io/hostname=node1,kubernetes.io/os=linux

[root@master mainfest]# kubectl apply -f test1.yaml 
pod/test1 created
[root@master mainfest]# kubectl get pod -o wide
NAME    READY   STATUS    RESTARTS   AGE   IP            NODE    NOMINATED NODE   READINESS GATES
t1      1/1     Running   0          40m   10.244.1.31   node1   <none>           <none>
test    1/1     Running   0          39m   10.244.1.32   node1   <none>           <none>
test1   1/1     Running   0          7s    10.244.2.44   node2   <none>           <none>

taint

Taints:避免Pod调度到特定Node上
Tolerations:允许Pod调度到持有Taints的Node上
应用场景:

• 专用节点:根据业务线将Node分组管理，希望在默认情况下不调度该节点，只有配置了污点容忍才允许分配

• 配备特殊硬件:部分Node配有SSD硬盘、GPU，希望在默认情况下不调度该节点，只有配置了污点容忍才允许分配

• 基于Taint的驱逐

#Taint(污点)
[root@master haproxy]# kubectl describe node master
Name:               master.example.com
Roles:              control-plane,master
Labels:             beta.kubernetes.io/arch=amd64
                    beta.kubernetes.io/os=linux
                    kubernetes.io/arch=amd64
                    kubernetes.io/hostname=master.example.com
                    kubernetes.io/os=linux
                    node-role.kubernetes.io/control-plane=
                    node-role.kubernetes.io/master=
                    node.kubernetes.io/exclude-from-external-load-balancers=
Annotations:        flannel.alpha.coreos.com/backend-data: {"VNI":1,"VtepMAC":"8e:50:ba:7a:30:2b"}
                    flannel.alpha.coreos.com/backend-type: vxlan
                    flannel.alpha.coreos.com/kube-subnet-manager: true
                    flannel.alpha.coreos.com/public-ip: 192.168.240.30
                    kubeadm.alpha.kubernetes.io/cri-socket: /var/run/dockershim.sock
                    node.alpha.kubernetes.io/ttl: 0
                    volumes.kubernetes.io/controller-managed-attach-detach: true
CreationTimestamp:  Sun, 19 Dec 2021 02:41:49 -0500
Taints:             node-role.kubernetes.io/master:NoSchedule  #aints:避免Pod调度到特定Node上
Unschedulable:      false


#Tolerations(污点容忍)
[root@master ~]# kubectl describe pod httpd1-57c7b6f7cb-sk86h 
Name:         httpd1-57c7b6f7cb-sk86h
Namespace:    default
·····
Tolerations:                 node.kubernetes.io/not-ready:NoExecute op=Exists for 300s
                             node.kubernetes.io/unreachable:NoExecute op=Exists for 300s ##Tolerations(污点容忍)允许Pod调度到持有Taints的Node上	

Events:
  Type     Reason                  Age                From     Message
  ----     ------                  ----               ----     -------
  Warning  FailedCreatePodSandBox  12m                kubelet  Failed to create pod sandbox: rpc error: code = Unknown desc = failed to set up sandbox container "2126da4117ba6ce45ddff8a2e0b9de59bac65f05c7be343249d50edea2cacf37" network for pod "httpd1-57c7b6f7cb-sk86h": networkPlugin cni failed to set up pod "httpd1-57c7b6f7cb-sk86h_default" network: open /run/flannel/subnet.env: no such file or directory
  Warning  FailedCreatePodSandBox  12m                kubelet  Failed to create pod "best2001/httpd"
  Normal   Pulled                  11m                kubelet  Successfully pulled image "best2001/httpd" in 16.175310708s
  Normal   Created                 11m                kubelet  Created container httpd1
  Normal   Started                 11m                kubelet  Started container httpd1

节点添加污点

格式: kubectl taint node [node] key=value:[effect]
例如: kubectl taint node k8s-node1 gpu=yes:NoSchedule验证: kubectl describe node k8s-node1 |grep Taint

其中[effect]可取值:

NoSchedule :一定不能被调度
PreferNoSchedule:尽量不要调度，非必须配置容忍
NoExecute:不仅不会调度，还会驱逐Node上已有的Pod
添加污点容忍(tolrations)字段到Pod配置中

#添加污点disktype
[root@master haproxy]# kubectl taint node node1.example.com disktype:NoSchedule
node/node1.example.com tainted

#查看
[root@master haproxy]# kubectl describe node node1.example.com
Name:               node1.example.com
Roles:              <none>
Labels:             beta.kubernetes.io/arch=amd64
                    beta.kubernetes.io/os=linux
                    kubernetes.io/arch=amd64
                    kubernetes.io/hostname=node1.example.com
                    kubernetes.io/os=linux
Annotations:        flannel.alpha.coreos.com/backend-data: {"VNI":1,"VtepMAC":"12:9e:43:99:21:bd"}
                    flannel.alpha.coreos.com/backend-type: vxlan
                    flannel.alpha.coreos.com/kube-subnet-manager: true
                    flannel.alpha.coreos.com/public-ip: 192.168.240.50
                    kubeadm.alpha.kubernetes.io/cri-socket: /var/run/dockershim.sock
                    node.alpha.kubernetes.io/ttl: 0
                    volumes.kubernetes.io/controller-managed-attach-detach: true
CreationTimestamp:  Sun, 19 Dec 2021 03:27:16 -0500
Taints:             disktype:NoSchedule  #污点添加成功

#测试创建一个容器
[root@master haproxy]# kubectl apply -f nginx.yml 
deployment.apps/nginx1 created
service/nginx1 created
[root@master haproxy]# kubectl get pods -o wide
NAME                      READY   STATUS    RESTARTS      AGE   IP            NODE                NOMINATED NODE   READINESS GATES
nginx1-7cf8bc594f-8j8tv   1/1     Running   0             14s   10.244.1.92   node2.example.com   <none>           <none>  #因为node1上面有污点，所以创建的容器会在node2上面跑

去掉污点:
kubectl taint node [node] key:[effect]-

[root@master haproxy]# kubectl taint node node1.example.com disktype-  
node/node1.example.com untainted

[root@master haproxy]# kubectl describe node node1.example.com
Name:               node1.example.com
Roles:              <none>
Labels:             beta.kubernetes.io/arch=amd64
                    beta.kubernetes.io/os=linux
                    kubernetes.io/arch=amd64
                    kubernetes.io/hostname=node1.example.com
                    kubernetes.io/os=linux
Annotations:        flannel.alpha.coreos.com/backend-data: {"VNI":1,"VtepMAC":"12:9e:43:99:21:bd"}
                    flannel.alpha.coreos.com/backend-type: vxlan
                    flannel.alpha.coreos.com/kube-subnet-manager: true
                    flannel.alpha.coreos.com/public-ip: 192.168.240.50
                    kubeadm.alpha.kubernetes.io/cri-socket: /var/run/dockershim.sock
                    node.alpha.kubernetes.io/ttl: 0
                    volumes.kubernetes.io/controller-managed-attach-detach: true
CreationTimestamp:  Sun, 19 Dec 2021 03:27:16 -0500
Taints:             <none>   #污点已经删除成功
Unschedulable:      false