硬件环境：
三台虚拟机：
192.168.99.129 master（kube-apiserver、kube-controller-manager、kube-proxy、kube-scheduler、kubelet、etcd、calico、docker）
192.168.99.130 slave1（kube-proxy、kubelet、etcd proxy、calico、docker、dns）
192.168.99.131 slave2（kube-proxy、kubelet、etcd proxy、calico、docker）
软件环境：
kubernetes 1.5.2
etcd 3.1.0
calico 0.23.1
【etcd】
calico需要每个node节点都要运行一个etcd proxy，所以master主机上部署一个etcd，其他node节点上部署etcd proxy。
master上etcd启动命令如下：（etcd新版本基本只使用2379这个端口了，但是有一些老的程序之前与etcd集成时使用的是4001端口，因此我同时监听2379和4001这两个端口）
etcd --name infra1 \
--data-dir /var/lib/etcd \
--listen-client-urls http://0.0.0.0:2379,http://0.0.0.0:4001 \
--advertise-client-urls http://192.168.99.129:2379,http://192.168.99.129:4001 \
--listen-peer-urls http://0.0.0.0:2380 \
--initial-advertise-peer-urls http://192.168.99.129:2380 \
--initial-cluster-token etcd-cluster \
--initial-cluster 'infra1=http://192.168.99.129:2380' \
--initial-cluster-state new \
--enable-pprof \
>> /var/log/etcd.log >& &
node上etcd proxy启动命令如下：
etcd --name infra-proxy1 \
--proxy=on \
--listen-client-urls http://0.0.0.0:2379 \
--initial-cluster 'infra1=http://192.168.99.130:2380' \
--enable-pprof \
>> /var/log/etcd.log >& &
etcd --name infra-proxy1 \
--proxy=on \
--listen-client-urls http://0.0.0.0:2379 \
--initial-cluster 'infra1=http://192.168.99.131:2380' \
--enable-pprof \
>> /var/log/etcd.log >& &
【kubernetes】
1、kube-apiserver和kubelet的启动脚本中添加--allow_privileged=true，如果不添加的话，下面在部署calico的时候，会以下错误：
The DaemonSet "calico-node" is invalid: spec.template.spec.containers[].securityContext.privileged: Forbidden: disallowed by policy
2、在kubelet的启动脚本中增加--network-plugin=cni和--network-plugin-dir=/etc/cni/net.d
kube-apiserver和kubelet的启动脚本如下：
kube-apiserver \
--logtostderr=true --v= \
--etcd-servers=http://k8s-master:4001 \
--insecure-bind-address=0.0.0.0 --insecure-port= \
--service-cluster-ip-range=10.254.0.0/ \
--allow_privileged=true \
>> /var/log/kube-apiserver.log >& &
kubelet \
--logtostderr=true --v= \
--address=0.0.0.0 \
--api-servers=http://k8s-master:8080 \
--pod-infra-container-image=index.tenxcloud.com/google_containers/pause-amd64:3.0 \
--cluster-dns=10.254.159.10 \
--cluster-domain=cluster.local \
--hostname-override=192.168.99.130 \
--allow_privileged=true \
--network-plugin=cni \
--network-plugin-dir=/etc/cni/net.d \
>> /var/log/kubelet.log >& &
3、下载 https://github.com/containernetworking/cni/releases/download/v0.4.0/cni-v0.4.0.tgz，解压之后，将loopback拷贝到/opt/cni/bin目录下，如果不做这步的话，创建pod时会抛错，说找不到loopback。
4、calico必须部署在master节点和所有的node节点上，如果master节点不部署calico，会出现容器内无法访问master的问题。因为calico是以dameonset部署的，所以在master节点上启动kubelet，calico就会部署在master节点上了。
【calico】
1、下载calico.yaml，地址为http://docs.projectcalico.org/v2.0/getting-started/kubernetes/installation/hosted/calico.yaml
2、修改calico.yaml文件中，etcd的地址
etcd_endpoints: "http://192.168.99.129:2379"
3、通过以下命令部署calico
kubectl apply -f calico.yaml
【部署centos和redis】
1、部署centos，指定部署在192.168.99.130节点上，centos-rcd.yaml如下：
apiVersion: v1
kind: ReplicationController
metadata:
name: centos
labels:
name: centos
spec:
replicas:
template:
metadata:
labels:
name: centos
spec:
containers:
- name: centos
image: index.tenxcloud.com/tenxcloud/docker-centos
ports:
- containerPort:
nodeSelector:
kubernetes.io/hostname: "192.168.99.130"
2、部署redis，指定部署在192.168.99.131节点上，redis-rc.yaml如下：
apiVersion: v1
kind: ReplicationController
metadata:
name: redis
labels:
k8s-app: redis
spec:
replicas:
selector:
k8s-app: redis
template:
metadata:
labels:
k8s-app: redis
spec:
containers:
- name: redis
image: 10.10.30.166/public/redis:v1
ports:
- containerPort:
name: redis-tcp
protocol: TCP
nodeSelector:
kubernetes.io/hostname: "192.168.99.131"
redis-svc.yaml如下：
apiVersion: v1
kind: Service
metadata:
name: redis
spec:
selector:
k8s-app: redis
clusterIP: 10.254.159.20
ports:
- name: ""
port:
protocol: TCP
3、部署情况如下：
[root@master redis]# kubectl get pods -o wide
NAME READY STATUS RESTARTS AGE IP NODE
centos-bpzkc / Running 23h 192.168.140.197 192.168.99.130
dns-99cqq / Running 1d 192.168.140.196 192.168.99.130
redis-c7wk3 / Running 4m 192.168.140.82 192.168.99.131
[root@master redis]# kubectl get svc -o wide
NAME CLUSTER-IP EXTERNAL-IP PORT(S) AGE SELECTOR
dns 10.254.159.10 <none> /UDP,/TCP 1d k8s-app=dns
kubernetes 10.254.0.1 <none> /TCP 2d <none>
redis 10.254.159.20 <none> /TCP 4m k8s-app=redis
master主机上的路由：
[root@master redis]# route -n
Kernel IP routing table
Destination Gateway Genmask Flags Metric Ref Use Iface
0.0.0.0 192.168.99.2 0.0.0.0 UG eno16777736
172.17.0.0 0.0.0.0 255.255.0.0 U docker0
192.168.99.0 0.0.0.0 255.255.255.0 U eno16777736
192.168.122.0 0.0.0.0 255.255.255.0 U virbr0
192.168.140.64 192.168.99.131 255.255.255.192 UG eno16777736
192.168.140.192 192.168.99.130 255.255.255.192 UG eno16777736
slave1主机上的路由：
[root@slave1 bin]# route -n
Kernel IP routing table
Destination Gateway Genmask Flags Metric Ref Use Iface
0.0.0.0 192.168.99.2 0.0.0.0 UG eno16777736
172.17.0.0 0.0.0.0 255.255.0.0 U docker0
192.168.99.0 0.0.0.0 255.255.255.0 U eno16777736
192.168.122.0 0.0.0.0 255.255.255.0 U virbr0
192.168.140.64 192.168.99.131 255.255.255.192 UG eno16777736
192.168.140.192 0.0.0.0 255.255.255.192 U *
192.168.140.196 0.0.0.0 255.255.255.255 UH cali12b26626b64
192.168.140.197 0.0.0.0 255.255.255.255 UH calic477824fb70
slave2主机上的路由：
[root@slave2 bin]# route -n
Kernel IP routing table
Destination Gateway Genmask Flags Metric Ref Use Iface
0.0.0.0 192.168.99.2 0.0.0.0 UG eno16777736
172.17.0.0 0.0.0.0 255.255.0.0 U docker0
192.168.99.0 0.0.0.0 255.255.255.0 U eno16777736
192.168.122.0 0.0.0.0 255.255.255.0 U virbr0
192.168.140.64 0.0.0.0 255.255.255.192 U *
192.168.140.82 0.0.0.0 255.255.255.255 UH calieb567fc0b5e
192.168.140.192 192.168.99.130 255.255.255.192 UG eno16777736
master、slave1和slave2上redis的iptables规则如下，他们三个是一样的
iptables -S -t nat | grep redis
-A KUBE-SEP-XAJWX3SXEKZG2YR7 -s 192.168.140.82/ -m comment --comment "default/redis:1" -j KUBE-MARK-MASQ
-A KUBE-SEP-XAJWX3SXEKZG2YR7 -p tcp -m comment --comment "default/redis:1" -m tcp -j DNAT --to-destination 192.168.140.82:
-A KUBE-SERVICES -d 10.254.159.20/ -p tcp -m comment --comment "default/redis:1 cluster IP" -m tcp --dport -j KUBE-SVC-XXJ2TMJIYSJJDBZG
-A KUBE-SVC-XXJ2TMJIYSJJDBZG -m comment --comment "default/redis:1" -j KUBE-SEP-XAJWX3SXEKZG2YR7
从这个规则能够看出，redis的clusterIp 10.254.159.20:6379被dnat转换为192.168.140.82:6379，这里遇到一个奇怪的问题，目前不知道原因，现象是，当redis-rc.yaml中labels是k8s-app: redis时，iptables规则如上面显示，一切正常，但如果labels是name: redis，则只有下面这一条规则，这说明clusterip不会被转换成pod的IP，所以访问clusterIP肯定是不通的。
-A KUBE-SERVICES -d 10.254.159.20/ -p tcp -m comment --comment "default/redis:1 cluster IP" -m tcp --dport -j KUBE-SVC-XXJ2TMJIYSJJDBZG
【验证网络连通性】
1、在master主机上ping centos和redis的ip
[root@master redis]# ping 192.168.140.197
PING 192.168.140.197 (192.168.140.197) () bytes of data.
bytes from 192.168.140.197: icmp_seq= ttl= time=1.55 ms
bytes from 192.168.140.197: icmp_seq= ttl= time=0.487 ms
[root@master redis]# ping 192.168.140.82
PING 192.168.140.82 (192.168.140.82) () bytes of data.
bytes from 192.168.140.82: icmp_seq= ttl= time=0.317 ms
bytes from 192.168.140.82: icmp_seq= ttl= time=0.502 ms
2、在master主机上telnet redis的clusterip
[root@master redis]# telnet 10.254.159.20
Trying 10.254.159.20...
Connected to 10.254.159.20.
Escape character is '^]'.
3、在slave1上ping centos和redis的pod，访问redis的clusterip
[root@slave1 bin]# ping 192.168.140.197
PING 192.168.140.197 (192.168.140.197) () bytes of data.
bytes from 192.168.140.197: icmp_seq= ttl= time=0.329 ms
bytes from 192.168.140.197: icmp_seq= ttl= time=0.068 ms
[root@slave1 bin]# ping 192.168.140.82
PING 192.168.140.82 (192.168.140.82) () bytes of data.
bytes from 192.168.140.82: icmp_seq= ttl= time=0.291 ms
bytes from 192.168.140.82: icmp_seq= ttl= time=0.455 ms
[root@slave1 bin]# telnet 10.254.159.20
Trying 10.254.159.20...
Connected to 10.254.159.20.
Escape character is '^]'.
4、在centos容器内ping redis的pod
[root@centos-bpzkc /]# ping 192.168.140.82
PING 192.168.140.82 (192.168.140.82) () bytes of data.
bytes from 192.168.140.82: icmp_seq= ttl= time=0.951 ms
5、在centos容器内通过dns解析redis域名，并访问redis
[root@centos-bpzkc /]# nslookup redis
Server: 10.254.159.10
Address: 10.254.159.10# Name: redis.default.svc.cluster.local
Address: 10.254.159.20
[root@centos-bpzkc /]# telnet redis
Trying 10.254.159.20...
Connected to redis.
Escape character is '^]'.
6、在centos容器内访问master主机上的服务（kube-apiserver）
[root@centos-bpzkc /]# telnet 192.168.99.129
Trying 192.168.99.129...
Connected to 192.168.99.129.
Escape character is '^]'.