Kubernetes中的网络解析——以flannel为例

我们当初使用kubernetes-vagrant-centos-cluster安装了拥有三个节点的kubernetes集群，节点的状态如下所述。

  1. [root@node1 ~]# kubectl get nodes -o wide
  2. NAME      STATUS    ROLES     AGE       VERSION   EXTERNAL-IP   OS-IMAGE                KERNEL-VERSION               CONTAINER-RUNTIME
  3. node1     Ready     <none>    2d        v1.9.1    <none>        CentOS Linux 7 (Core)   3.10.0-693.11.6.el7.x86_64   docker://1.12.6
  4. node2     Ready     <none>    2d        v1.9.1    <none>        CentOS Linux 7 (Core)   3.10.0-693.11.6.el7.x86_64   docker://1.12.6
  5. node3     Ready     <none>    2d        v1.9.1    <none>        CentOS Linux 7 (Core)   3.10.0-693.11.6.el7.x86_64   docker://1.12.6

当前Kubernetes集群中运行的所有Pod信息：

  1. [root@node1 ~]# kubectl get pods --all-namespaces -o wide
  2. NAMESPACE     NAME                                              READY     STATUS    RESTARTS   AGE       IP            NODE
  3. kube-system   coredns-5984fb8cbb-sjqv9                          1/1       Running   0          1h        172.33.68.2   node1
  4. kube-system   coredns-5984fb8cbb-tkfrc                          1/1       Running   1          1h        172.33.96.3   node3
  5. kube-system   heapster-v1.5.0-684c7f9488-z6sdz                  4/4       Running   0          1h        172.33.31.3   node2
  6. kube-system   kubernetes-dashboard-6b66b8b96c-mnm2c             1/1       Running   0          1h        172.33.31.2   node2
  7. kube-system   monitoring-influxdb-grafana-v4-54b7854697-tw9cd   2/2       Running   2          1h        172.33.96.2   node3

当前etcd中的注册的宿主机的pod地址网段信息：

  1. [root@node1 ~]# etcdctl ls /kube-centos/network/subnets
  2. /kube-centos/network/subnets/172.33.68.0-24
  3. /kube-centos/network/subnets/172.33.31.0-24
  4. /kube-centos/network/subnets/172.33.96.0-24

而每个node上的Pod子网是根据我们在安装flannel时配置来划分的，在etcd中查看该配置：

  1. [root@node1 ~]# etcdctl get /kube-centos/network/config
  2. {"Network":"172.33.0.0/16","SubnetLen":24,"Backend":{"Type":"host-gw"}}

我们知道Kubernetes集群内部存在三类IP，分别是：

  • Node IP：宿主机的IP地址
  • Pod IP：使用网络插件创建的IP（如flannel），使跨主机的Pod可以互通
  • Cluster IP：虚拟IP，通过iptables规则访问服务

在安装node节点的时候，节点上的进程是按照flannel -> docker -> kubelet -> kube-proxy的顺序启动的，我们下面也会按照该顺序来讲解，flannel的网络划分和如何与docker交互，如何通过iptables访问service。

Flannel

Flannel是作为一个二进制文件的方式部署在每个node上，主要实现两个功能：

  • 为每个node分配subnet，容器将自动从该子网中获取IP地址
  • 当有node加入到网络中时，为每个node增加路由配置

下面是使用host-gw backend的flannel网络架构图：

flannel网络架构（图片来自openshift）

注意：以上IP非本示例中的IP，但是不影响读者理解。

Node1上的flannel配置如下:

  1. [root@node1 ~]# cat /usr/lib/systemd/system/flanneld.service
  2. [Unit]
  3. Description=Flanneld overlay address etcd agent
  4. After=network.target
  5. After=network-online.target
  6. Wants=network-online.target
  7. After=etcd.service
  8. Before=docker.service
  10. [Service]
  11. Type=notify
  12. EnvironmentFile=/etc/sysconfig/flanneld
  13. EnvironmentFile=-/etc/sysconfig/docker-network
  14. ExecStart=/usr/bin/flanneld-start $FLANNEL_OPTIONS
  15. ExecStartPost=/usr/libexec/flannel/mk-docker-opts.sh -k DOCKER_NETWORK_OPTIONS -d /run/flannel/docker
  16. Restart=on-failure
  18. [Install]
  19. WantedBy=multi-user.target
  20. RequiredBy=docker.service

其中有两个环境变量文件的配置如下：

  1. [root@node1 ~]# cat /etc/sysconfig/flanneld
  2. # Flanneld configuration options
  3. FLANNEL_ETCD_ENDPOINTS="http://172.17.8.101:2379"
  4. FLANNEL_ETCD_PREFIX="/kube-centos/network"
  5. FLANNEL_OPTIONS="-iface=eth2"

上面的配置文件仅供flanneld使用。

  1. [root@node1 ~]# cat /etc/sysconfig/docker-network
  2. # /etc/sysconfig/docker-network
  3. DOCKER_NETWORK_OPTIONS=

还有一个ExecStartPost=/usr/libexec/flannel/mk-docker-opts.sh -k DOCKER_NETWORK_OPTIONS -d /run/flannel/docker，其中的/usr/libexec/flannel/mk-docker-opts.sh脚本是在flanneld启动后运行，将会生成两个环境变量配置文件：

  • /run/flannel/docker
  • /run/flannel/subnet.env

我们再来看下/run/flannel/docker的配置。

  1. [root@node1 ~]# cat /run/flannel/docker
  2. DOCKER_OPT_BIP="--bip=172.33.68.1/24"
  3. DOCKER_OPT_IPMASQ="--ip-masq=true"
  4. DOCKER_OPT_MTU="--mtu=1500"
  5. DOCKER_NETWORK_OPTIONS=" --bip=172.33.68.1/24 --ip-masq=true --mtu=1500"

如果你使用systemctl命令先启动flannel后启动docker的话，docker将会读取以上环境变量。

我们再来看下/run/flannel/subnet.env的配置。

  1. [root@node1 ~]# cat /run/flannel/subnet.env
  2. FLANNEL_NETWORK=172.33.0.0/16
  3. FLANNEL_SUBNET=172.33.68.1/24
  4. FLANNEL_MTU=1500
  5. FLANNEL_IPMASQ=false

以上环境变量是flannel向etcd中注册的。

Docker

Node1的docker配置如下：

  1. [root@node1 ~]# cat /usr/lib/systemd/system/docker.service
  2. [Unit]
  3. Description=Docker Application Container Engine
  4. Documentation=http://docs.docker.com
  5. After=network.target rhel-push-plugin.socket registries.service
  6. Wants=docker-storage-setup.service
  7. Requires=docker-cleanup.timer
  9. [Service]
  10. Type=notify
  11. NotifyAccess=all
  12. EnvironmentFile=-/run/containers/registries.conf
  13. EnvironmentFile=-/etc/sysconfig/docker
  14. EnvironmentFile=-/etc/sysconfig/docker-storage
  15. EnvironmentFile=-/etc/sysconfig/docker-network
  16. Environment=GOTRACEBACK=crash
  17. Environment=DOCKER_HTTP_HOST_COMPAT=1
  18. Environment=PATH=/usr/libexec/docker:/usr/bin:/usr/sbin
  19. ExecStart=/usr/bin/dockerd-current \
  20.           --add-runtime docker-runc=/usr/libexec/docker/docker-runc-current \
  21.           --default-runtime=docker-runc \
  22.           --exec-opt native.cgroupdriver=systemd \
  23.           --userland-proxy-path=/usr/libexec/docker/docker-proxy-current \
  24.           $OPTIONS \
  25.           $DOCKER_STORAGE_OPTIONS \
  26.           $DOCKER_NETWORK_OPTIONS \
  27.           $ADD_REGISTRY \
  28.           $BLOCK_REGISTRY \
  29.           $INSECURE_REGISTRY\
  30.           $REGISTRIES
  31. ExecReload=/bin/kill -s HUP $MAINPID
  32. LimitNOFILE=1048576
  33. LimitNPROC=1048576
  34. LimitCORE=infinity
  35. TimeoutStartSec=0
  36. Restart=on-abnormal
  37. MountFlags=slave
  38. KillMode=process
  40. [Install]
  41. WantedBy=multi-user.target

查看Node1上的docker启动参数：

  1. [root@node1 ~]# systemctl status -l docker
  2. ● docker.service - Docker Application Container Engine
  3.    Loaded: loaded (/usr/lib/systemd/system/docker.service; enabled; vendor preset: disabled)
  4.   Drop-In: /usr/lib/systemd/system/docker.service.d
  5.            └─flannel.conf
  6.    Active: active (running) since Fri 2018-02-02 22:52:43 CST; 2h 28min ago
  7.      Docs: http://docs.docker.com
  8.  Main PID: 4334 (dockerd-current)
  9.    CGroup: /system.slice/docker.service
  10.            ‣ 4334 /usr/bin/dockerd-current --add-runtime docker-runc=/usr/libexec/docker/docker-runc-current --default-runtime=docker-runc --exec-opt native.cgroupdriver=systemd --userland-proxy-path=/usr/libexec/docker/docker-proxy-current --selinux-enabled --log-driver=journald --signature-verification=false --bip=172.33.68.1/24 --ip-masq=true --mtu=1500

我们可以看到在docker在启动时有如下参数：--bip=172.33.68.1/24 --ip-masq=true --mtu=1500。上述参数flannel启动时运行的脚本生成的，通过环境变量传递过来的。

我们查看下node1宿主机上的网络接口：

  1. [root@node1 ~]# ip addr
  2. 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN qlen 1
  3.     link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
  4.     inet 127.0.0.1/8 scope host lo
  5.        valid_lft forever preferred_lft forever
  6.     inet6 ::1/128 scope host
  7.        valid_lft forever preferred_lft forever
  8. 2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP qlen 1000
  9.     link/ether 52:54:00:00:57:32 brd ff:ff:ff:ff:ff:ff
  10.     inet 10.0.2.15/24 brd 10.0.2.255 scope global dynamic eth0
  11.        valid_lft 85095sec preferred_lft 85095sec
  12.     inet6 fe80::5054:ff:fe00:5732/64 scope link
  13.        valid_lft forever preferred_lft forever
  14. 3: eth1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP qlen 1000
  15.     link/ether 08:00:27:7b:0f:b1 brd ff:ff:ff:ff:ff:ff
  16.     inet 172.17.8.101/24 brd 172.17.8.255 scope global eth1
  17.        valid_lft forever preferred_lft forever
  18. 4: eth2: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast state UP qlen 1000
  19.     link/ether 08:00:27:ef:25:06 brd ff:ff:ff:ff:ff:ff
  20.     inet 172.30.113.231/21 brd 172.30.119.255 scope global dynamic eth2
  21.        valid_lft 85096sec preferred_lft 85096sec
  22.     inet6 fe80::a00:27ff:feef:2506/64 scope link
  23.        valid_lft forever preferred_lft forever
  24. 5: docker0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP
  25.     link/ether 02:42:d0:ae:80:ea brd ff:ff:ff:ff:ff:ff
  26.     inet 172.33.68.1/24 scope global docker0
  27.        valid_lft forever preferred_lft forever
  28.     inet6 fe80::42:d0ff:feae:80ea/64 scope link
  29.        valid_lft forever preferred_lft forever
  30. 7: veth295bef2@if6: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue master docker0 state UP
  31.     link/ether 6a:72:d7:9f:29:19 brd ff:ff:ff:ff:ff:ff link-netnsid 0
  32.     inet6 fe80::6872:d7ff:fe9f:2919/64 scope link
  33.        valid_lft forever preferred_lft forever

我们分类来解释下该虚拟机中的网络接口。

  • lo：回环网络，127.0.0.1
  • eth0：NAT网络，虚拟机创建时自动分配，仅可以在几台虚拟机之间访问
  • eth1：bridge网络，使用vagrant分配给虚拟机的地址，虚拟机之间和本地电脑都可以访问
  • eth2：bridge网络，使用DHCP分配，用于访问互联网的网卡
  • docker0：bridge网络，docker默认使用的网卡，作为该节点上所有容器的虚拟交换机
  • veth295bef2@if6：veth pair，连接docker0和Pod中的容器。veth pair可以理解为使用网线连接好的两个接口，把两个端口放到两个namespace中，那么这两个namespace就能打通。参考linux 网络虚拟化： network namespace 简介。

我们再看下该节点的docker上有哪些网络。

  1. [root@node1 ~]# docker network ls
  2. NETWORK ID          NAME                DRIVER              SCOPE
  3. 940bb75e653b        bridge              bridge              local
  4. d94c046e105d        host                host                local
  5. 2db7597fd546        none                null                local

再检查下bridge网络940bb75e653b的信息。

  1. [root@node1 ~]# docker network inspect 940bb75e653b
  2. [
  3.     {
  4.         "Name": "bridge",
  5.         "Id": "940bb75e653bfa10dab4cce8813c2b3ce17501e4e4935f7dc13805a61b732d2c",
  6.         "Scope": "local",
  7.         "Driver": "bridge",
  8.         "EnableIPv6": false,
  9.         "IPAM": {
  10.             "Driver": "default",
  11.             "Options": null,
  12.             "Config": [
  13.                 {
  14.                     "Subnet": "172.33.68.1/24",
  15.                     "Gateway": "172.33.68.1"
  16.                 }
  17.             ]
  18.         },
  19.         "Internal": false,
  20.         "Containers": {
  21.             "944d4aa660e30e1be9a18d30c9dcfa3b0504d1e5dbd00f3004b76582f1c9a85b": {
  22.                 "Name": "k8s_POD_coredns-5984fb8cbb-sjqv9_kube-system_c5a2e959-082a-11e8-b4cd-525400005732_0",
  23.                 "EndpointID": "7397d7282e464fc4ec5756d6b328df889cdf46134dbbe3753517e175d3844a85",
  24.                 "MacAddress": "02:42:ac:21:44:02",
  25.                 "IPv4Address": "172.33.68.2/24",
  26.                 "IPv6Address": ""
  27.             }
  28.         },
  29.         "Options": {
  30.             "com.docker.network.bridge.default_bridge": "true",
  31.             "com.docker.network.bridge.enable_icc": "true",
  32.             "com.docker.network.bridge.enable_ip_masquerade": "true",
  33.             "com.docker.network.bridge.host_binding_ipv4": "0.0.0.0",
  34.             "com.docker.network.bridge.name": "docker0",
  35.             "com.docker.network.driver.mtu": "1500"
  36.         },
  37.         "Labels": {}
  38.     }
  39. ]

我们可以看到该网络中的Config与docker的启动配置相符。

Node1上运行的容器：

  1. [root@node1 ~]# docker ps
  2. CONTAINER ID        IMAGE                                                                                               COMMAND                  CREATED             STATUS              PORTS               NAMES
  3. a37407a234dd        docker.io/coredns/coredns@sha256:adf2e5b4504ef9ffa43f16010bd064273338759e92f6f616dd159115748799bc   "/coredns -conf /etc/"   About an hour ago   Up About an hour                        k8s_coredns_coredns-5984fb8cbb-sjqv9_kube-system_c5a2e959-082a-11e8-b4cd-525400005732_0
  4. 944d4aa660e3        docker.io/openshift/origin-pod                                                                      "/usr/bin/pod"           About an hour ago   Up About an hour                        k8s_POD_coredns-5984fb8cbb-sjqv9_kube-system_c5a2e959-082a-11e8-b4cd-525400005732_0

我们可以看到当前已经有2个容器在运行。

Node1上的路由信息：

  1. [root@node1 ~]# route -n
  2. Kernel IP routing table
  3. Destination     Gateway         Genmask         Flags Metric Ref    Use Iface
  4. 0.0.0.0         10.0.2.2        0.0.0.0         UG    100    0        0 eth0
  5. 0.0.0.0         172.30.116.1    0.0.0.0         UG    101    0        0 eth2
  6. 10.0.2.0        0.0.0.0         255.255.255.0   U     100    0        0 eth0
  7. 172.17.8.0      0.0.0.0         255.255.255.0   U     100    0        0 eth1
  8. 172.30.112.0    0.0.0.0         255.255.248.0   U     100    0        0 eth2
  9. 172.33.68.0     0.0.0.0         255.255.255.0   U     0      0        0 docker0
  10. 172.33.96.0     172.30.118.65   255.255.255.0   UG    0      0        0 eth2

以上路由信息是由flannel添加的，当有新的节点加入到Kubernetes集群中后，每个节点上的路由表都将增加。

我们在node上来traceroute下node3上的coredns-5984fb8cbb-tkfrc容器，其IP地址是172.33.96.3，看看其路由信息。

  1. [root@node1 ~]# traceroute 172.33.96.3
  2. traceroute to 172.33.96.3 (172.33.96.3), 30 hops max, 60 byte packets
  3.  1  172.30.118.65 (172.30.118.65)  0.518 ms  0.367 ms  0.398 ms
  4.  2  172.33.96.3 (172.33.96.3)  0.451 ms  0.352 ms  0.223 ms

我们看到路由直接经过node3的公网IP后就到达了node3节点上的Pod。

Node1的iptables信息：

  1. [root@node1 ~]# iptables -L
  2. Chain INPUT (policy ACCEPT)
  3. target     prot opt source               destination
  4. KUBE-FIREWALL  all  --  anywhere             anywhere
  5. KUBE-SERVICES  all  --  anywhere             anywhere             /* kubernetes service portals */
  7. Chain FORWARD (policy ACCEPT)
  8. target     prot opt source               destination
  9. KUBE-FORWARD  all  --  anywhere             anywhere             /* kubernetes forward rules */
  10. DOCKER-ISOLATION  all  --  anywhere             anywhere
  11. DOCKER     all  --  anywhere             anywhere
  12. ACCEPT     all  --  anywhere             anywhere             ctstate RELATED,ESTABLISHED
  13. ACCEPT     all  --  anywhere             anywhere
  14. ACCEPT     all  --  anywhere             anywhere
  16. Chain OUTPUT (policy ACCEPT)
  17. target     prot opt source               destination
  18. KUBE-FIREWALL  all  --  anywhere             anywhere
  19. KUBE-SERVICES  all  --  anywhere             anywhere             /* kubernetes service portals */
  21. Chain DOCKER (1 references)
  22. target     prot opt source               destination
  24. Chain DOCKER-ISOLATION (1 references)
  25. target     prot opt source               destination
  26. RETURN     all  --  anywhere             anywhere
  28. Chain KUBE-FIREWALL (2 references)
  29. target     prot opt source               destination
  30. DROP       all  --  anywhere             anywhere             /* kubernetes firewall for dropping marked packets */ mark match 0x8000/0x8000
  32. Chain KUBE-FORWARD (1 references)
  33. target     prot opt source               destination
  34. ACCEPT     all  --  anywhere             anywhere             /* kubernetes forwarding rules */ mark match 0x4000/0x4000
  35. ACCEPT     all  --  10.254.0.0/16        anywhere             /* kubernetes forwarding conntrack pod source rule */ ctstate RELATED,ESTABLISHED
  36. ACCEPT     all  --  anywhere             10.254.0.0/16        /* kubernetes forwarding conntrack pod destination rule */ ctstate RELATED,ESTABLISHED
  38. Chain KUBE-SERVICES (2 references)
  39. target     prot opt source               destination

从上面的iptables中可以看到注入了很多Kuberentes service的规则。

参考