Service概念
Kubernetes Service定义了一个Pod的逻辑分组，一种可以访问它们的策略。这组Pod能被Service访问到，通常是通过label Selector。
Service能够提供负载均衡的能力，它只提供4层负载均衡的能力，而没有7层功能，有时我们可能需要更多的匹配规则来转发请求，这点上4层负载均衡是不支持的。
Service的类型
Service在K8S中有四种类型
ClusterIp：默认类型，自动分配一个仅Cluster内部可以访问的虚拟IP。NodePort：在ClusterIP基础上为Service在每台机器上绑定一个端口，这样就可以通过NodePort来访问服务。LoadBalancer：在NodePort的基础上，借助cloud provider创建一个外部负载均衡器，并将请求转发到NodePort。ExternalName：把集群外部的服务引入到集群内部来，在集群内部直接使用。没有任何类型代理被创建，只有Kubernetes1.7或更高版本的kube-dns才支持。
VIP和Service代理
在Kubernetes集群中，每个Node运行一个kube-proxy进程。kube-proxy负责为Service实现了一种VIP(虚拟IP)的形式，而不是ExternalName的形式。在Kubernetes v1.0版本，代理v完全在userspace。在Kubernetes v1.1版本，新增了iptables代理，但并不是默认的运行模式。从Kubernetes v1.2起，默认就是iptables代理。在Kubernetes v1.8.0-beta.0中，添加了ipvs代理，在Kubernetes 1.14版本开始默认使用ipvs代理。在Kubernetes v1.0版本，Service是“4层”(TCP/UDP over IP)的概念。在Kubernetes v1.1版本，新增了Ingress API（beta版），用来表示“7层”(HTTP)服务。
代理模式的分类
userspace代理模式
iptables代理模式
-
ipvs代理模式kube-proxy会监视KubernetesService对象和Endpoints对象，调用netlink接口以相应地创建ipvs规则并定期与KubernetesService对象和Endpoints对象同步ipvs规则，以确保ipvs状态与期望一致。访问服务时，流量将被重定向到其中一个后端Pod。与iptables类似，ipvs于netfilter的hook功能，但使用哈希表作为底层数据结构并在内核空间中工作。这意味着ipvs可以更快地重定向流量，并在同步代理规则时具有更好地性能。此外，ipvs为负载均衡算法提供了更多选项，例如：rr：轮询调度lc：最小连接数dh：目标哈希sh：源哈希sed：最短期望延迟nq：不排队调度
注意：ipvs模式假定在运行kube-proxy之前在节点上都已经安装了ipvs内核模块。当kube-proxy以ipvs代理模式启动时，kube-proxy将验证节点上是否安装了ipvs模块，如果未安装，则kube-proxy将回退到iptables代理模式。
Service的 spec.type
ClusterIP
ClusterIP主要在每个node节点使用iptables，将发向clusterIP对应端口的数据，转发到kube-proxy中。然后kube-proxy自己内部实现有负载均衡的方法，并可以查询到这个service下对应pod的地址和端口，进而把数据转发给对应的pod地址和端口。
- 用户通过
kubectl命令向apiserver发送创建service的命令，apiserver收到请求后将数据存储到etcd中。 kubernetes的每个节点中都有一个叫做kube-proxy的进程，这个进程负责感知service，pod的变化，并将变化的信息写入本地的iptables规则中。iptables使用NAT等技术将virtualIP的流量转至endpoint中。
创建myapp-deploy.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
name: myapp-deploy
labels:
app: myapp-deploy
spec:
replicas: 3
template:
metadata:
name: myapp-deploy
labels:
app: myapp-deploy
release: stabel
env: test
spec:
containers:
- name: myapp-deploy
image: chinda.com/library/myapp:v1
imagePullPolicy: IfNotPresent
ports:
- containerPort: 80
name: http
restartPolicy: Always
selector:
matchLabels:
app: myapp-deploy
release: stabel
创建myapp-service.yaml
apiVersion: v1
kind: Service
metadata:
name: myapp-service
spec:
selector:
app: myapp-service
release: stabel
ports:
- port: 80
targetPort: 80
name: http
type: ClusterIP
测试
[root@k8s-master01 ~]# kubectl get svc
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
kubernetes ClusterIP 10.96.0.1 <none> 443/TCP 13d
myapp-service ClusterIP 10.107.157.192 <none> 80/TCP 9s
[root@k8s-master01 ~]# curl 10.107.157.192
curl: (7) Failed connect to 10.107.157.192:80; Connection refused
[root@k8s-master01 ~]# ipvsadm -Ln
IP Virtual Server version 1.2.1 (size=4096)
Prot LocalAddress:Port Scheduler Flags
-> RemoteAddress:Port Forward Weight ActiveConn InActConn
TCP 10.96.0.1:443 rr
-> 192.168.0.150:6443 Masq 1 3 0
TCP 10.96.0.10:53 rr
-> 10.244.0.43:53 Masq 1 0 0
-> 10.244.0.44:53 Masq 1 0 0
TCP 10.96.0.10:9153 rr
-> 10.244.0.43:9153 Masq 1 0 0
-> 10.244.0.44:9153 Masq 1 0 0
TCP 10.107.157.192:80 rr
UDP 10.96.0.10:53 rr
-> 10.244.0.43:53 Masq 1 0 0
-> 10.244.0.44:53 Masq 1 0 0
[root@k8s-master01 ~]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
myapp-deploy-6f87986465-2pcgp 1/1 Running 0 14m 10.244.1.123 k8s-node01 <none> <none>
myapp-deploy-6f87986465-7rhrq 1/1 Running 0 14m 10.244.1.124 k8s-node01 <none> <none>
myapp-deploy-6f87986465-scz2v 1/1 Running 0 14m 10.244.2.110 k8s-node02 <none> <none>
拒绝连接，ipvs检测service下没有挂载任何的pod，原因是deployment和service的spec.selector命名不相同。
修改后
修改后myapp-deploy.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
name: myapp-deploy
labels:
app: myapp-deploy
spec:
replicas: 3
template:
metadata:
name: myapp
labels:
app: myapp
release: stabel
env: test
spec:
containers:
- name: myapp-deploy
image: chinda.com/library/myapp:v1
imagePullPolicy: IfNotPresent
ports:
- containerPort: 80
name: http
restartPolicy: Always
selector:
matchLabels:
app: myapp
release: stabel
修改后myapp-service.yaml
apiVersion: v1
kind: Service
metadata:
name: myapp-service
spec:
selector:
app: myapp
release: stabel
ports:
- port: 80
targetPort: 80
name: http
type: ClusterIP
检测
[root@k8s-master01 ~]# kubectl get svc
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
kubernetes ClusterIP 10.96.0.1 <none> 443/TCP 13d
myapp-service ClusterIP 10.104.50.207 <none> 80/TCP 10s
[root@k8s-master01 ~]# ipvsadm -Ln
IP Virtual Server version 1.2.1 (size=4096)
Prot LocalAddress:Port Scheduler Flags
-> RemoteAddress:Port Forward Weight ActiveConn InActConn
TCP 10.96.0.1:443 rr
-> 192.168.0.150:6443 Masq 1 3 0
TCP 10.96.0.10:53 rr
-> 10.244.0.43:53 Masq 1 0 0
-> 10.244.0.44:53 Masq 1 0 0
TCP 10.96.0.10:9153 rr
-> 10.244.0.43:9153 Masq 1 0 0
-> 10.244.0.44:9153 Masq 1 0 0
TCP 10.104.50.207:80 rr
-> 10.244.1.125:80 Masq 1 0 0
-> 10.244.1.126:80 Masq 1 0 0
-> 10.244.2.111:80 Masq 1 0 0
UDP 10.96.0.10:53 rr
-> 10.244.0.43:53 Masq 1 0 0
-> 10.244.0.44:53 Masq 1 0 0
[root@k8s-master01 ~]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
myapp-deploy-677ccc888c-m42jt 1/1 Running 0 119s 10.244.1.125 k8s-node01 <none> <none>
myapp-deploy-677ccc888c-rgsn4 1/1 Running 0 119s 10.244.2.111 k8s-node02 <none> <none>
myapp-deploy-677ccc888c-w6hzg 1/1 Running 0 119s 10.244.1.126 k8s-node01 <none> <none>
[root@k8s-master01 ~]# curl 10.104.50.207
Hello MyApp | Version: v1 | <a href="hostname.html">Pod Name</a>
Headless Service
有时不需要或不想要负载均衡，以及单独的Service IP。遇到这种情况，可以通过指定Cluster IP(spec.clusterIP)的值为“None”来创建Headless Service。这类Service并不会分配Cluster IP，kube-proxy不会处理他们，而且平台也不会为它们进行负载均衡和路由。
创建myapp-headless-svc.yaml
apiVersion: v1
kind: Service
metadata:
name: myapp-headless
spec:
selector:
app: myapp
ports:
- port: 80
targetPort: 80
clusterIP: None
解析coredns
# svc 创建成功会写入到coredns中
[root@k8s-master01 ~]# kubectl get svc
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
kubernetes ClusterIP 10.96.0.1 <none> 443/TCP 13d
myapp-headless ClusterIP None <none> 80/TCP 6s
myapp-service ClusterIP 10.104.50.207 <none> 80/TCP 16m
[root@k8s-master01 ~]# kubectl get pod -n kube-system -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
coredns-5c98db65d4-dhrw6 1/1 Running 21 13d 10.244.0.44 k8s-master01 <none> <none>
coredns-5c98db65d4-jhvv2 1/1 Running 21 13d 10.244.0.43 k8s-master01 <none> <none>
etcd-k8s-master01 1/1 Running 22 13d 192.168.0.150 k8s-master01 <none> <none>
kube-apiserver-k8s-master01 1/1 Running 22 13d 192.168.0.150 k8s-master01 <none> <none>
kube-controller-manager-k8s-master01 1/1 Running 22 13d 192.168.0.150 k8s-master01 <none> <none>
kube-flannel-ds-amd64-bm5px 1/1 Running 28 13d 192.168.0.150 k8s-master01 <none> <none>
kube-flannel-ds-amd64-cfhhg 1/1 Running 23 13d 192.168.0.152 k8s-node02 <none> <none>
kube-flannel-ds-amd64-mlqgl 1/1 Running 20 13d 192.168.0.151 k8s-node01 <none> <none>
kube-proxy-4f8dk 1/1 Running 19 13d 192.168.0.152 k8s-node02 <none> <none>
kube-proxy-lg875 1/1 Running 19 13d 192.168.0.151 k8s-node01 <none> <none>
kube-proxy-zwpmh 1/1 Running 22 13d 192.168.0.150 k8s-master01 <none> <none>
kube-scheduler-k8s-master01 1/1 Running 22 13d 192.168.0.150 k8s-master01 <none> <none>
# 安装dig命令
[root@k8s-master01 ~]# yum -y install bind-utils
# 访问svc {service-name}.{namespace}.svc.cluster.local
[root@k8s-master01 ~]# dig -t A myapp-headless.default.svc.cluster.local. @10.244.0.44
; <<>> DiG 9.11.4-P2-RedHat-9.11.4-16.P2.el7_8.2 <<>> -t A myapp-headless.default.svc.cluster.local. @10.244.0.44
;; global options: +cmd
;; Got answer:
;; WARNING: .local is reserved for Multicast DNS
;; You are currently testing what happens when an mDNS query is leaked to DNS
;; ->>HEADER<<- opcode: QUERY, status: NOERROR, id: 6438
;; flags: qr aa rd; QUERY: 1, ANSWER: 3, AUTHORITY: 0, ADDITIONAL: 1
;; WARNING: recursion requested but not available
;; OPT PSEUDOSECTION:
; EDNS: version: 0, flags:; udp: 4096
;; QUESTION SECTION:
;myapp-headless.default.svc.cluster.local. IN A
;; ANSWER SECTION:
myapp-headless.default.svc.cluster.local. 30 IN A 10.244.1.125
myapp-headless.default.svc.cluster.local. 30 IN A 10.244.1.126
myapp-headless.default.svc.cluster.local. 30 IN A 10.244.2.111
;; Query time: 0 msec
;; SERVER: 10.244.0.44#53(10.244.0.44)
;; WHEN: Fri May 01 09:57:24 CST 2020
;; MSG SIZE rcvd: 237
[root@k8s-master01 ~]# kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
myapp-deploy-677ccc888c-m42jt 1/1 Running 0 48m 10.244.1.125 k8s-node01 <none> <none>
myapp-deploy-677ccc888c-rgsn4 1/1 Running 0 48m 10.244.2.111 k8s-node02 <none> <none>
myapp-deploy-677ccc888c-w6hzg 1/1 Running 0 48m 10.244.1.126 k8s-node01 <none> <none>
NodePort
NodePort的原理在于在Node上开了一个端口，将向该端口的流量导入kube-proxy，然后由kube-proxy进一步给对应的Pod。
创建nodeport.yaml
apiVersion: v1
kind: Service
metadata:
name: myapp-nodeport
spec:
selector:
app: myapp
release: stabel
ports:
- port: 80
targetPort: 80
type: NodePort
测试
[root@k8s-master01 ~]# kubectl get svc
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
kubernetes ClusterIP 10.96.0.1 <none> 443/TCP 13d
myapp-headless ClusterIP None <none> 80/TCP 36m
myapp-nodeport NodePort 10.105.209.63 <none> 80:31509/TCP 5s
myapp-service ClusterIP 10.104.50.207 <none> 80/TCP 52m
# 每个节点都会开放此端口
[root@k8s-master01 ~]# netstat -anp | grep 31509
tcp6 0 0 :::31509 :::* LISTEN 3338/kube-proxy
LoadBalancer
云供应商提供的服务，其实和NodePort是同一种方式。
ExternalName
这种类型的Service通过返回CNAME和它的值，可以将服务映射到externalName字段的内容(例如：www.123.com)。ExternalName Service是Service的特例，它没有selector，也没有定义任何的端口和Endpoint。相反的，对于运行在集群之外的服务，它通过返回该外部服务的别名这种方式来提供服务。
创建ex.yaml
apiVersion: v1
kind: Service
metadata:
name: myapp-ex
spec:
type: ExternalName
externalName: www.123.com
检测
[root@k8s-master01 ~]# kubectl get svc
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
kubernetes ClusterIP 10.96.0.1 <none> 443/TCP 13d
myapp-ex ExternalName <none> www.123.com <none> 6s
myapp-headless ClusterIP None <none> 80/TCP 99m
myapp-nodeport NodePort 10.105.209.63 <none> 80:31509/TCP 62m
myapp-service ClusterIP 10.104.50.207 <none> 80/TCP 115m
[root@k8s-master01 ~]# dig -t A myapp-ex.default.svc.cluster.local. @10.244.0.44
; <<>> DiG 9.11.4-P2-RedHat-9.11.4-16.P2.el7_8.2 <<>> -t A myapp-ex.default.svc.cluster.local. @10.244.0.44
;; global options: +cmd
;; Got answer:
;; WARNING: .local is reserved for Multicast DNS
;; You are currently testing what happens when an mDNS query is leaked to DNS
;; ->>HEADER<<- opcode: QUERY, status: NOERROR, id: 41068
;; flags: qr aa rd; QUERY: 1, ANSWER: 2, AUTHORITY: 0, ADDITIONAL: 1
;; WARNING: recursion requested but not available
;; OPT PSEUDOSECTION:
; EDNS: version: 0, flags:; udp: 4096
;; QUESTION SECTION:
;myapp-ex.default.svc.cluster.local. IN A
;; ANSWER SECTION:
myapp-ex.default.svc.cluster.local. 30 IN CNAME www.123.com.
www.123.com. 30 IN A 61.132.13.130
;; Query time: 68 msec
;; SERVER: 10.244.0.44#53(10.244.0.44)
;; WHEN: Fri May 01 11:08:54 CST 2020
;; MSG SIZE rcvd: 149
当查询主机 myapp-ex.default.svc.cluster.local时，集群的DNS服务将返回一个值www.123.comd的CNAME记录。访问这个服务的工作方式和其他的相同，唯一不同的时重定向发生在DNS层，而且不会进行代理或转发。