不背锅运维:K8S之污点、污点容忍
节点亲和性是Pod的一种属性,它使 Pod 被吸引到一类特定的节点 (这可能出于一种偏好,也可能是硬性要求)。 污点(Taint) 则相反——它使节点能够排斥一类特定的Pod。也就是说避免pod调度到特定node上,告诉默认的pod,我拒绝你的分配,是一种主动拒绝的行为。
写在开篇
❝
本篇分享k8s的污点、污点容忍,感谢持续关注我的盆友们。如有考k8s认证需求的盆友,可和我取得联系,了解K8S认证请猛戳:《K8s CKA+CKS认证实战班》2023版
❞
K8s CKA+CKS认证实战班》2023版:https://mp.weixin.qq.com/s/h1bjcIwy2enVD203o-ntlA
污点和污点容忍
什么是污点?
❝
节点亲和性是Pod的一种属性,它使 Pod 被吸引到一类特定的节点 (这可能出于一种偏好,也可能是硬性要求)。 污点(Taint) 则相反——它使节点能够排斥一类特定的Pod。也就是说避免pod调度到特定node上,告诉默认的pod,我拒绝你的分配,是一种主动拒绝的行为。
❞
什么是污点容忍?
❝
是应用于 Pod 上的。容忍度允许调度器调度带有对应污点的 Pod。 容忍度允许调度但并不保证调度。也就是说,允许pod调度到持有Taint的node上,希望pod能够分配到带污点的节点,增加了污点容忍,那么就会认可这个污点,就**「有可能」**分配到带污点的节点(如果希望pod可以被分配到带有污点的节点上,要在pod配置中添加污点容忍字段)。
❞
污点和容忍(Toleration)相互配合,可以用来避免Pod被分配到不合适的节点上,每个节点上都可以应用一个或多个污点,这表示对于那些不能容忍这些污点的Pod, 是不会被该节点接受的。
再用大白话理解一下,也就是说,基于节点标签的分配,是站在Pod的角度上,它是通过在pod上添加属性来确定pod是否要调度到指定的node上的。其实,也还可以站在Node的角度上,通过在node上添加污点属性,来避免pod被分配到不合适的节点上。
语法格式
节点添加污点的语法格式
kubectl taint node xxxx key=value:[effect]
effect(效果):
- NoSchedule:不能被调度
- PreferNoSchedule:尽量不要调度
- NoExecute:不但不会调度,还会驱逐Node上已有的pod
删除污点的语法格式
kubectl taint node xxxx key=value:[effect]-
实战
- 给test-b-k8s-node02节点打上污点,不干预调度到哪台节点,让k8s按自己的算法进行调度,看看这10个pod会不会分配到带有污点的节点上
# 打污点
kubectl taint node test-b-k8s-node02 disktype=sas:NoSchedule
# 查看node详情的Taints字段
tantianran@test-b-k8s-master:~$ kubectl describe node test-b-k8s-node02 | grep Taint
Taints: disktype=sas:NoSchedule
goweb-demo.yaml
apiVersion: v1
kind: Namespace
metadata:
name: test-a
---
apiVersion: apps/v1
kind: Deployment
metadata:
name: goweb-demo
namespace: test-a
spec:
replicas: 10
selector:
matchLabels:
app: goweb-demo
template:
metadata:
labels:
app: goweb-demo
spec:
containers:
- name: goweb-demo
image: 192.168.11.247/web-demo/goweb-demo:20221229v3
---
apiVersion: v1
kind: Service
metadata:
name: goweb-demo
namespace: test-a
spec:
ports:
- port: 80
protocol: TCP
targetPort: 8090
selector:
app: goweb-demo
type: NodePort
tantianran@test-b-k8s-master:~/goweb-demo$ kubectl get pod -n test-a -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
goweb-demo-b98869456-84p4b 1/1 Running 0 18s 10.244.240.50 test-b-k8s-node01 <none> <none>
goweb-demo-b98869456-cjjj8 1/1 Running 0 18s 10.244.240.13 test-b-k8s-node01 <none> <none>
goweb-demo-b98869456-fxgjf 1/1 Running 0 18s 10.244.240.12 test-b-k8s-node01 <none> <none>
goweb-demo-b98869456-jfdvl 1/1 Running 0 18s 10.244.240.43 test-b-k8s-node01 <none> <none>
goweb-demo-b98869456-k6krp 1/1 Running 0 18s 10.244.240.41 test-b-k8s-node01 <none> <none>
goweb-demo-b98869456-kcpsz 1/1 Running 0 18s 10.244.240.6 test-b-k8s-node01 <none> <none>
goweb-demo-b98869456-lrkzc 1/1 Running 0 18s 10.244.240.49 test-b-k8s-node01 <none> <none>
goweb-demo-b98869456-nqr2j 1/1 Running 0 18s 10.244.240.33 test-b-k8s-node01 <none> <none>
goweb-demo-b98869456-pt5zk 1/1 Running 0 18s 10.244.240.28 test-b-k8s-node01 <none> <none>
goweb-demo-b98869456-s9rt5 1/1 Running 0 18s 10.244.240.42 test-b-k8s-node01 <none> <none>
tantianran@test-b-k8s-master:~/goweb-demo$
❝
发现全部都在test-b-k8s-node01节点,test-b-k8s-node01节点有污点,因此拒绝承载pod。
❞
- test-b-k8s-node01节点已经有污点了,再通过nodeSelector强行指派到该节点,看看会不会分配到带有污点的节点上
goweb-demo.yaml
apiVersion: v1
kind: Namespace
metadata:
name: test-a
---
apiVersion: apps/v1
kind: Deployment
metadata:
name: goweb-demo
namespace: test-a
spec:
replicas: 10
selector:
matchLabels:
app: goweb-demo
template:
metadata:
labels:
app: goweb-demo
spec:
nodeSelector:
disktype: sas
containers:
- name: goweb-demo
image: 192.168.11.247/web-demo/goweb-demo:20221229v3
---
apiVersion: v1
kind: Service
metadata:
name: goweb-demo
namespace: test-a
spec:
ports:
- port: 80
protocol: TCP
targetPort: 8090
selector:
app: goweb-demo
type: NodePort
查看pod的创建情况
tantianran@test-b-k8s-master:~/goweb-demo$ kubectl get pod -n test-a
NAME READY STATUS RESTARTS AGE
goweb-demo-54bc765fff-2gb98 0/1 Pending 0 20s
goweb-demo-54bc765fff-67c56 0/1 Pending 0 20s
goweb-demo-54bc765fff-6fdvx 0/1 Pending 0 20s
goweb-demo-54bc765fff-c2bgd 0/1 Pending 0 20s
goweb-demo-54bc765fff-d55mw 0/1 Pending 0 20s
goweb-demo-54bc765fff-dl4x4 0/1 Pending 0 20s
goweb-demo-54bc765fff-g4vb2 0/1 Pending 0 20s
goweb-demo-54bc765fff-htjkp 0/1 Pending 0 20s
goweb-demo-54bc765fff-s76rh 0/1 Pending 0 20s
goweb-demo-54bc765fff-vg6dn 0/1 Pending 0 20s
tantianran@test-b-k8s-master:~/goweb-demo$
❝
尴尬的局面很显然了,该节点明明存在污点,又非得往上面指派,因此让所有Pod处于在了Pending的状态,也就是待分配的状态,那如果非要往带有污点的Node上指派pod,怎么办?看下面的例子
❞
- 非要往带有污点的Node上指派pod,保留nodeSelector,直接增加污点容忍,pod是不是肯定会分配到带有污点的节点上?测试下便知 goweb-demo.yaml
apiVersion: v1
kind: Namespace
metadata:
name: test-a
---
apiVersion: apps/v1
kind: Deployment
metadata:
name: goweb-demo
namespace: test-a
spec:
replicas: 10
selector:
matchLabels:
app: goweb-demo
template:
metadata:
labels:
app: goweb-demo
spec:
nodeSelector:
disktype: sas
tolerations:
- key: "disktype"
operator: "Equal"
value: "sas"
effect: "NoSchedule"
containers:
- name: goweb-demo
image: 192.168.11.247/web-demo/goweb-demo:20221229v3
---
apiVersion: v1
kind: Service
metadata:
name: goweb-demo
namespace: test-a
spec:
ports:
- port: 80
protocol: TCP
targetPort: 8090
selector:
app: goweb-demo
type: NodePort
查看Pod创建状态
tantianran@test-b-k8s-master:~/goweb-demo$ kubectl get pod -n test-a
NAME READY STATUS RESTARTS AGE
goweb-demo-68cf558b74-6qddp 0/1 Pending 0 109s
goweb-demo-68cf558b74-7g6cm 0/1 Pending 0 109s
goweb-demo-68cf558b74-f7g6t 0/1 Pending 0 109s
goweb-demo-68cf558b74-kcs9j 0/1 Pending 0 109s
goweb-demo-68cf558b74-kxssv 0/1 Pending 0 109s
goweb-demo-68cf558b74-pgrvb 0/1 Pending 0 109s
goweb-demo-68cf558b74-ps5dn 0/1 Pending 0 109s
goweb-demo-68cf558b74-rb2w5 0/1 Pending 0 109s
goweb-demo-68cf558b74-tcnj4 0/1 Pending 0 109s
goweb-demo-68cf558b74-txqfs 0/1 Pending 0 109s
❝
在上面的yaml中,tolerations字段为污点容忍,经过测试就可以回答刚才的问题:保留nodeSelector,直接增加污点容忍,pod是不是肯定会分配到带有污点的节点上?经过测试后,给出的答案是:不是。
❞
那怎么办呢?继续测试,现在把nodeSelector去掉,只留下污点容忍,看看pod会不会有可能分配到打了污点的节点上?
goweb-demo.yaml
apiVersion: v1
kind: Namespace
metadata:
name: test-a
---
apiVersion: apps/v1
kind: Deployment
metadata:
name: goweb-demo
namespace: test-a
spec:
replicas: 10
selector:
matchLabels:
app: goweb-demo
template:
metadata:
labels:
app: goweb-demo
spec:
tolerations:
- key: "disktype"
operator: "Equal"
value: "sas"
effect: "NoSchedule"
containers:
- name: goweb-demo
image: 192.168.11.247/web-demo/goweb-demo:20221229v3
---
apiVersion: v1
kind: Service
metadata:
name: goweb-demo
namespace: test-a
spec:
ports:
- port: 80
protocol: TCP
targetPort: 8090
selector:
app: goweb-demo
type: NodePort
查看pod创建情况
tantianran@test-b-k8s-master:~/goweb-demo$ kubectl get pod -n test-a -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
goweb-demo-55ff5cd68c-287vw 1/1 Running 0 110s 10.244.222.57 test-b-k8s-node02 <none> <none>
goweb-demo-55ff5cd68c-7s7zb 1/1 Running 0 110s 10.244.222.24 test-b-k8s-node02 <none> <none>
goweb-demo-55ff5cd68c-84jww 1/1 Running 0 110s 10.244.240.24 test-b-k8s-node01 <none> <none>
goweb-demo-55ff5cd68c-b5l9m 1/1 Running 0 110s 10.244.240.15 test-b-k8s-node01 <none> <none>
goweb-demo-55ff5cd68c-c2gfp 1/1 Running 0 110s 10.244.222.3 test-b-k8s-node02 <none> <none>
goweb-demo-55ff5cd68c-hpjn4 1/1 Running 0 110s 10.244.240.62 test-b-k8s-node01 <none> <none>
goweb-demo-55ff5cd68c-j5bvc 1/1 Running 0 110s 10.244.222.43 test-b-k8s-node02 <none> <none>
goweb-demo-55ff5cd68c-r95f6 1/1 Running 0 110s 10.244.240.16 test-b-k8s-node01 <none> <none>
goweb-demo-55ff5cd68c-rhvmw 1/1 Running 0 110s 10.244.240.60 test-b-k8s-node01 <none> <none>
goweb-demo-55ff5cd68c-rl8nh 1/1 Running 0 110s 10.244.222.8 test-b-k8s-node02 <none> <none>
❝
从上面的分配结果可以看出,有些Pod分配到了打了污点容忍的test-b-k8s-node02节点上。
❞
- 再玩个小例子,让它容忍任何带污点的节点,master默认也是有污点的(二进制搭建的除外),那pod会不会有可能跑master去哦?测试一下便知
先看看master的污点情况
tantianran@test-b-k8s-master:~/goweb-demo$ kubectl describe node test-b-k8s-master | grep Taint
Taints: node-role.kubernetes.io/control-plane:NoSchedule
goweb-demo.yaml
apiVersion: v1
kind: Namespace
metadata:
name: test-a
---
apiVersion: apps/v1
kind: Deployment
metadata:
name: goweb-demo
namespace: test-a
spec:
replicas: 10
selector:
matchLabels:
app: goweb-demo
template:
metadata:
labels:
app: goweb-demo
spec:
tolerations:
- effect: "NoSchedule"
operator: "Exists"
containers:
- name: goweb-demo
image: 192.168.11.247/web-demo/goweb-demo:20221229v3
---
apiVersion: v1
kind: Service
metadata:
name: goweb-demo
namespace: test-a
spec:
ports:
- port: 80
protocol: TCP
targetPort: 8090
selector:
app: goweb-demo
type: NodePort
查看pod创建情况
tantianran@test-b-k8s-master:~/goweb-demo$ kubectl get pod -n test-a -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
goweb-demo-65bbd7b49c-5qb5m 0/1 ImagePullBackOff 0 20s 10.244.82.55 test-b-k8s-master <none> <none>
goweb-demo-65bbd7b49c-7qqw8 1/1 Running 0 20s 10.244.222.13 test-b-k8s-node02 <none> <none>
goweb-demo-65bbd7b49c-9tflk 1/1 Running 0 20s 10.244.240.27 test-b-k8s-node01 <none> <none>
goweb-demo-65bbd7b49c-dgxhx 1/1 Running 0 20s 10.244.222.44 test-b-k8s-node02 <none> <none>
goweb-demo-65bbd7b49c-fbmn5 1/1 Running 0 20s 10.244.240.1 test-b-k8s-node01 <none> <none>
goweb-demo-65bbd7b49c-h2nnz 1/1 Running 0 20s 10.244.240.39 test-b-k8s-node01 <none> <none>
goweb-demo-65bbd7b49c-kczsp 1/1 Running 0 20s 10.244.240.40 test-b-k8s-node01 <none> <none>
goweb-demo-65bbd7b49c-ms768 1/1 Running 0 20s 10.244.222.45 test-b-k8s-node02 <none> <none>
goweb-demo-65bbd7b49c-pbwht 0/1 ErrImagePull 0 20s 10.244.82.56 test-b-k8s-master <none> <none>
goweb-demo-65bbd7b49c-zqxlt 1/1 Running 0 20s 10.244.222.18 test-b-k8s-node02 <none> <none>
居然还真的有2个Pod跑master节点去了(test-b-k8s-master),master节点本地没有对应的image,且harbor服务器也没开机,所以拉取image失败了。这个不是关心的问题。关心的问题是,居然有pod跑master节点去了,是因为k8s的这个机制:如果一个容忍度的key为空且operator为Exists,表示这个容忍度与任意的key、value和effect都匹配,即这个容忍度能容忍任何污点。了解了这个Exists的机制后,就知道为啥会跑master去了吧?
❝
警告:要注意了,master之所以默认会打上污点,是因为master是管理节点、考虑到安全的问题,所以master节点是不建议跑常规的pod(或者说是不建议跑业务pod)。
❞
- 打了污点的节点,到底有没有办法可以强行分配到该节点上?我们试试
节点test-b-k8s-node02是打了污点的
tantianran@test-b-k8s-master:~/goweb-demo$ kubectl describe node test-b-k8s-node02 | grep Taint
Taints: disktype=sas:NoSchedule
goweb-demo.yaml
apiVersion: v1
kind: Namespace
metadata:
name: test-a
---
apiVersion: apps/v1
kind: Deployment
metadata:
name: goweb-demo
namespace: test-a
spec:
replicas: 10
selector:
matchLabels:
app: goweb-demo
template:
metadata:
labels:
app: goweb-demo
spec:
nodeName: test-b-k8s-node02
containers:
- name: goweb-demo
image: 192.168.11.247/web-demo/goweb-demo:20221229v3
---
apiVersion: v1
kind: Service
metadata:
name: goweb-demo
namespace: test-a
spec:
ports:
- port: 80
protocol: TCP
targetPort: 8090
selector:
app: goweb-demo
type: NodePort
❝
在上面的配置中,注意nodeName字段,nodeName指定节点名称,用于将Pod调度到指定的Node上,它的机制是**「不经过调度器」**
❞
查看pod创建情况
tantianran@test-b-k8s-master:~/goweb-demo$ kubectl get pod -n test-a -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
goweb-demo-dd446d4b9-2zdnx 1/1 Running 0 13s 10.244.222.39 test-b-k8s-node02 <none> <none>
goweb-demo-dd446d4b9-4qbg9 1/1 Running 0 13s 10.244.222.6 test-b-k8s-node02 <none> <none>
goweb-demo-dd446d4b9-67cpl 1/1 Running 0 13s 10.244.222.63 test-b-k8s-node02 <none> <none>
goweb-demo-dd446d4b9-fhsgf 1/1 Running 0 13s 10.244.222.53 test-b-k8s-node02 <none> <none>
goweb-demo-dd446d4b9-gp9gj 1/1 Running 0 13s 10.244.222.49 test-b-k8s-node02 <none> <none>
goweb-demo-dd446d4b9-hzvs2 1/1 Running 0 13s 10.244.222.9 test-b-k8s-node02 <none> <none>
goweb-demo-dd446d4b9-px598 1/1 Running 0 13s 10.244.222.22 test-b-k8s-node02 <none> <none>
goweb-demo-dd446d4b9-rkbm4 1/1 Running 0 13s 10.244.222.40 test-b-k8s-node02 <none> <none>
goweb-demo-dd446d4b9-vr9mq 1/1 Running 0 13s 10.244.222.17 test-b-k8s-node02 <none> <none>
goweb-demo-dd446d4b9-wnfqc 1/1 Running 0 13s 10.244.222.16 test-b-k8s-node02 <none> <none>
❝
会发现,所有Pod都分配到了test-b-k8s-node02节点,怎么不会分一些到test-b-k8s-node01节点?原因就是,它的机制是不经过调度器的。nodeName这个字段建议在生产环境中还是少用,所有Pod都在一个节点上,这就存在单点故障了。其实,测试环境下还是可以用的嘛!
❞
最后
❝
关于本篇的分享就到这里,后面还会有更多的小案例分享给大家,欢迎持续关注。想要考K8S认证的盆友请猛戳了解:https://mp.weixin.qq.com/s/h1bjcIwy2enVD203o-ntlA
本文转载于(喜欢的盆友关注我们):https://mp.weixin.qq.com/s/qJ8gr4xyuTjXkA6p9Yrp7g
更多推荐
所有评论(0)