Kubernetes网络分析中如何实现Container间通信

这篇文章主要介绍Kubernetes网络分析中如何实现Container间通信，文中介绍的非常详细，具有一定的参考价值，感兴趣的小伙伴们一定要看完！

成都创新互联主要企业基础官网建设，电商平台建设，移动手机平台，重庆小程序开发公司等一系列专为中小企业按需求定制开发产品体系；应对中小企业在互联网运营的各种问题，为中小企业在互联网的运营中保驾护航。

1. 引言##

上图显示了Kubernetes的基本结构图。

• Master管理多个Slave节点

• Slave节点上面可以运行多个Pod

• Pod可以部署多个副本，多个副本可以运行在不同的Node上

• 一个Pod可以包含多个Container，一个Pod内的Container共享同样的网络地址空间

最重要的是最后一句话：一个Pod内的Container共享同样的网络地址空间。这是通过Mapped Container做到的。

2. Mapped Container##

基本描述为下：

• 容器 A 的网络模式为正常docker的网络模式

• 容器 B 的网络模式为应用容器A的网络模式

###2.1 共享网络模式###

下面是一个例子来验证，我这里创建了一个busybox的Pod。

[root@centos7-node-221 ~]$ kubectl get po
NAME      READY     STATUS    RESTARTS   AGE
busybox   1/1       Running   224        9d
[root@centos7-node-221 ~]$ kubectl describe po busybox
Name:				busybox
Namespace:			default
Image(s):			busybox
Node:				centos7-node-226/192.168.1.226
Labels:				
Status:				Running
Reason:				
Message:			
IP:				172.16.58.6
Replication Controllers:	
Containers:
  busybox:
    Image:			busybox
    State:			Running
      Started:			Thu, 08 Oct 2015 08:20:30 -0400
    Last Termination State:	Terminated
      Exit Code:		0
      Started:			Thu, 08 Oct 2015 07:20:26 -0400
      Finished:			Thu, 08 Oct 2015 08:20:26 -0400
    Ready:			True
    Restart Count:		224
    Variables:
Conditions:
  Type		Status
  Ready 	True 
Volumes:
  default-token-lv94w:
    Type:	Secret (a secret that should populate this volume)
    SecretName:	default-token-lv94w
Events:
  FirstSeen	LastSeen	Count	From				SubobjectPath			Reason	Message
  9d		37m		225	{kubelet centos7-node-226}	spec.containers{busybox}	pulled	Container image "busybox" already present on machine
  37m		37m		1	{kubelet centos7-node-226}	spec.containers{busybox}	Created	Created with docker id fc8580292210
  37m		37m		1	{kubelet centos7-node-226}	spec.containers{busybox}	Started	Started with docker id fc8580292210

我们去192.168.1.226看下这个Pod和其Container.

[root@centos7-node-226 ~]$ docker ps | grep busybox
fc8580292210        busybox                                               "sleep 3600"        37 minutes ago      Up 37 minutes                           k8s_busybox.62fa0587_busybox_default_86e98e8c-665f-11e5-af98-525400d7abb6_7f734c4d                                                
02d259dc8ab5        gcr.io/google_containers/pause:0.8.0                  "/pause"            9 days ago          Up 9 days                               k8s_POD.7be6d81d_busybox_default_86e98e8c-665f-11e5-af98-525400d7abb6_ff9224f5

发现有两个容器，一个是pause容器，一个是busybox容器。其中pause容器为主网络容器，其他容器都共享pause容器的网络模式。我们分别看下其网络模式。下面是两个容器的网络模式。

[root@centos7-node-226 ~]$ docker inspect 02d259dc8ab5  | grep NetworkMode
        "NetworkMode": "bridge",
[root@centos7-node-226 ~]$ docker inspect fc8580292210 | grep NetworkMode
        "NetworkMode": "container:02d259dc8ab59c1746d54d2df24d8733b2b9379a9fdfbfdc2066429b4a934a04", # 这个container的id号码就是上一个container的id的long形式

所以可以看到fc8580292210(busybox)使用的是pause容器的网络空间。

让我们进一步验证。

2.2 IP地址和hostname、网络IO###

下面我在 192.168.1.224 搭建了一个DNS的pod，里面有4个容器，共享一个网络空间，我们采用查看其ip地址、hostname和网络IO的方式来鉴定。 下面是容器的id号

[root@centos7-node-224 ~]$ docker ps | grep dns
b00a08d078d6        dockerimages.yinnut.com:15043/skydns:2015-03-11-001   "/skydns -machines=h   8 hours ago         Up 8 hours                              k8s_skydns.c878079e_kube-dns-v9-y05vd_kube-system_12725077-64c0-11e5-9309-525400d7abb6_46f95e60                                   
4e843585b938        dockerimages.yinnut.com:15043/exechealthz:1.0         "/exechealthz '-cmd=   11 days ago         Up 11 days                              k8s_healthz.8ab20f84_kube-dns-v9-y05vd_kube-system_12725077-64c0-11e5-9309-525400d7abb6_f7c469e5                                  
296ff779abb2        dockerimages.yinnut.com:15043/kube2sky:1.11           "/kube2sky -domain=c   11 days ago         Up 11 days                              k8s_kube2sky.2a46d768_kube-dns-v9-y05vd_kube-system_12725077-64c0-11e5-9309-525400d7abb6_349c7246                                 
f0118fac6952        dockerimages.yinnut.com:15043/etcd:2.0.9              "/usr/local/bin/etcd   11 days ago         Up 11 days                              k8s_etcd.64e02c2f_kube-dns-v9-y05vd_kube-system_12725077-64c0-11e5-9309-525400d7abb6_9235054b                                     
f281dbf1ec41        gcr.io/google_containers/pause:0.8.0                  "/pause"               11 days ago         Up 11 days                              k8s_POD.6e934112_kube-dns-v9-y05vd_kube-system_12725077-64c0-11e5-9309-525400d7abb6_a8ea96d0

我们查看前三个 b00a08d078d6 4e843585b938 296ff779abb2 的上述属性。

2.2.1 dns设置和hostname####

[root@centos7-node-224 ~]$ for id in b00a08d078d6 4e843585b938 296ff779abb2 ; do echo $id; docker exec $id cat /etc/hosts ; docker exec $id cat /etc/resolv.conf ; echo  "" ; done
b00a08d078d6
172.16.60.4	kube-dns-v9-y05vd
127.0.0.1	localhost
::1	localhost ip6-localhost ip6-loopback
fe00::0	ip6-localnet
ff00::0	ip6-mcastprefix
ff02::1	ip6-allnodes
ff02::2	ip6-allrouters
nameserver 192.168.1.208
search 8.8.8.8
options ndots:5

4e843585b938
172.16.60.4	kube-dns-v9-y05vd
127.0.0.1	localhost
::1	localhost ip6-localhost ip6-loopback
fe00::0	ip6-localnet
ff00::0	ip6-mcastprefix
ff02::1	ip6-allnodes
ff02::2	ip6-allrouters
nameserver 192.168.1.208
search 8.8.8.8
options ndots:5

296ff779abb2
172.16.60.4	kube-dns-v9-y05vd
127.0.0.1	localhost
::1	localhost ip6-localhost ip6-loopback
fe00::0	ip6-localnet
ff00::0	ip6-mcastprefix
ff02::1	ip6-allnodes
ff02::2	ip6-allrouters
nameserver 192.168.1.208
search 8.8.8.8
options ndots:5

2.2.2 IP地址####

[root@centos7-node-224 ~]$ for id in b00a08d078d6 4e843585b938 296ff779abb2 ; do echo $id; docker exec $id ip a  ; echo  "" ; done
b00a08d078d6
1: lo:  mtu 65536 qdisc noqueue state UNKNOWN group default 
    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
    inet 127.0.0.1/8 scope host lo
       valid_lft forever preferred_lft forever

4e843585b938
1: lo:  mtu 65536 qdisc noqueue state UNKNOWN group default 
    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
    inet 127.0.0.1/8 scope host lo
       valid_lft forever preferred_lft forever

296ff779abb2
1: lo:  mtu 65536 qdisc noqueue state UNKNOWN group default 
    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
    inet 127.0.0.1/8 scope host lo
       valid_lft forever preferred_lft forever

2.2.3 网络连接IO####

[root@centos7-node-224 ~]$ for id in b00a08d078d6 4e843585b938 296ff779abb2 ; do echo $id; docker exec $id netstat -lan  ; echo  "" ; done
b00a08d078d6
Active Internet connections (servers and established)
Proto Recv-Q Send-Q Local Address           Foreign Address         State       
tcp        0      0 127.0.0.1:4001          0.0.0.0:*               LISTEN      
tcp        0      0 127.0.0.1:2379          0.0.0.0:*               LISTEN      
tcp        0      0 127.0.0.1:2380          0.0.0.0:*               LISTEN      
tcp        0      0 127.0.0.1:7001          0.0.0.0:*               LISTEN      
tcp        0      0 172.16.60.4:48582       10.254.0.1:443          ESTABLISHED 
tcp        0      0 127.0.0.1:4001          127.0.0.1:35394         ESTABLISHED 
tcp        0      0 172.16.60.4:48584       10.254.0.1:443          ESTABLISHED 
tcp        0      0 127.0.0.1:60161         127.0.0.1:2379          ESTABLISHED 
tcp        0      0 127.0.0.1:51445         127.0.0.1:4001          ESTABLISHED 
tcp        0      0 127.0.0.1:4001          127.0.0.1:51445         ESTABLISHED 
tcp        0      0 127.0.0.1:35550         127.0.0.1:4001          ESTABLISHED 
tcp        0      0 127.0.0.1:35394         127.0.0.1:4001          ESTABLISHED 
tcp        0      0 127.0.0.1:2379          127.0.0.1:60161         ESTABLISHED 
tcp        0      0 127.0.0.1:4001          127.0.0.1:51433         ESTABLISHED 
tcp        0      0 127.0.0.1:51433         127.0.0.1:4001          ESTABLISHED 
tcp        0      0 127.0.0.1:4001          127.0.0.1:35550         ESTABLISHED 
Active UNIX domain sockets (servers and established)
Proto RefCnt Flags       Type       State         I-Node Path

4e843585b938
Active Internet connections (servers and established)
Proto Recv-Q Send-Q Local Address           Foreign Address         State       
tcp        0      0 127.0.0.1:4001          0.0.0.0:*               LISTEN      
tcp        0      0 127.0.0.1:2379          0.0.0.0:*               LISTEN      
tcp        0      0 127.0.0.1:2380          0.0.0.0:*               LISTEN      
tcp        0      0 127.0.0.1:7001          0.0.0.0:*               LISTEN      
tcp        0      0 172.16.60.4:48582       10.254.0.1:443          ESTABLISHED 
tcp        0      0 127.0.0.1:4001          127.0.0.1:35394         ESTABLISHED 
tcp        0      0 172.16.60.4:48584       10.254.0.1:443          ESTABLISHED 
tcp        0      0 127.0.0.1:60161         127.0.0.1:2379          ESTABLISHED 
tcp        0      0 127.0.0.1:51445         127.0.0.1:4001          ESTABLISHED 
tcp        0      0 127.0.0.1:4001          127.0.0.1:51445         ESTABLISHED 
tcp        0      0 127.0.0.1:35550         127.0.0.1:4001          ESTABLISHED 
tcp        0      0 127.0.0.1:35394         127.0.0.1:4001          ESTABLISHED 
tcp        0      0 127.0.0.1:2379          127.0.0.1:60161         ESTABLISHED 
tcp        0      0 127.0.0.1:4001          127.0.0.1:51433         ESTABLISHED 
tcp        0      0 127.0.0.1:51433         127.0.0.1:4001          ESTABLISHED 
tcp        0      0 127.0.0.1:4001          127.0.0.1:35550         ESTABLISHED 
Active UNIX domain sockets (servers and established)
Proto RefCnt Flags       Type       State         I-Node Path

296ff779abb2
Active Internet connections (servers and established)
Proto Recv-Q Send-Q Local Address           Foreign Address         State       
tcp        0      0 127.0.0.1:4001          0.0.0.0:*               LISTEN      
tcp        0      0 127.0.0.1:2379          0.0.0.0:*               LISTEN      
tcp        0      0 127.0.0.1:2380          0.0.0.0:*               LISTEN      
tcp        0      0 127.0.0.1:7001          0.0.0.0:*               LISTEN      
tcp        0      0 172.16.60.4:48582       10.254.0.1:443          ESTABLISHED 
tcp        0      0 127.0.0.1:4001          127.0.0.1:35394         ESTABLISHED 
tcp        0      0 172.16.60.4:48584       10.254.0.1:443          ESTABLISHED 
tcp        0      0 127.0.0.1:60161         127.0.0.1:2379          ESTABLISHED 
tcp        0      0 127.0.0.1:51445         127.0.0.1:4001          ESTABLISHED 
tcp        0      0 127.0.0.1:4001          127.0.0.1:51445         ESTABLISHED 
tcp        0      0 127.0.0.1:35550         127.0.0.1:4001          ESTABLISHED 
tcp        0      0 127.0.0.1:35394         127.0.0.1:4001          ESTABLISHED 
tcp        0      0 127.0.0.1:2379          127.0.0.1:60161         ESTABLISHED 
tcp        0      0 127.0.0.1:4001          127.0.0.1:51433         ESTABLISHED 
tcp        0      0 127.0.0.1:51433         127.0.0.1:4001          ESTABLISHED 
tcp        0      0 127.0.0.1:4001          127.0.0.1:35550         ESTABLISHED 
Active UNIX domain sockets (servers and established)
Proto RefCnt Flags       Type       State         I-Node Path

3. Container 之通信##

下面分析下最复杂的Container之间的通信。

3.1 Pod 之内 Container 通信

先说最简单的, Pod内的Container通信，由于共享网络地址空间，直接访问127.0.0.1即可。

3.2 跨机器之间Contianer通信###

3.2.1 背景引导####

例子： 192.168.1.224的fluentd-elasticsearch容器要连接192.168.1.223的elasticsearch-logging容器。

• 192.168.1.223的elasticsearch-logging容器及其IP地址:

[root@centos7-node-223 ~]$ docker ps |grep elasticsearch-logging
667cfd84c979        dockerimages.yinnut.com:15043/elasticsearch:1.7       "/run.sh"              12 days ago          Up 12 days                              k8s_elasticsearch-logging.89fda9f_elasticsearch-logging-v1-i8x6q_kube-system_8b558d2c-62a3-11e5-9d7b-525400d7abb6_2a02a2c8        
5201c8cbdebd        gcr.io/google_containers/pause:0.8.0                  "/pause"               12 days ago          Up 12 days                              k8s_POD.8ecd2043_elasticsearch-logging-v1-i8x6q_kube-system_8b558d2c-62a3-11e5-9d7b-525400d7abb6_4022db35                         
[root@centos7-node-223 ~]$ docker inspect 5201c8cbdebd |grep IPAddress
        "IPAddress": "172.16.77.4", #IP地址
        "SecondaryIPAddresses": null,

可以看到elasticsearch-logging的容器的Pod的IP地址为172.16.77.4

• 192.168.1.223的fluentd-elasticsearch容器及其IP地址:

[root@centos7-node-224 ~]$ docker ps |grep fluentd-elasticsearch
d326d81468b5        gcr.io/google_containers/fluentd-elasticsearch:1.11   "td-agent -q"          12 days ago         Up 12 days                              k8s_fluentd-elasticsearch.27a08aa3_fluentd-elasticsearch-centos7-node-224_kube-system_7dcc6ce562f3742190a876fda85e2359_58c54ef3   
f9b76639d241        gcr.io/google_containers/pause:0.8.0                  "/pause"               12 days ago         Up 12 days                              k8s_POD.7be6d81d_fluentd-elasticsearch-centos7-node-224_kube-system_7dcc6ce562f3742190a876fda85e2359_333e52c0                     
[root@centos7-node-224 ~]$ docker inspect f9b76639d241 | grep IPAddress
        "IPAddress": "172.16.60.2",
        "SecondaryIPAddresses": null,

可以看到fluentd-elasticsearch的容器的Pod的IP地址为172.16.60.2

• 我们看下 fluentd-elasticsearch 的网络连接情况

[root@centos7-node-224 ~]$ docker exec d326d81468b5 netstat -nla | grep 172.16.77.4
tcp        0      0 172.16.60.2:56354       172.16.77.4:9200        TIME_WAIT  
tcp        0      0 172.16.60.2:56350       172.16.77.4:9200        TIME_WAIT  
tcp        0      0 172.16.60.2:56347       172.16.77.4:9200        TIME_WAIT  
tcp        0      0 172.16.60.2:56357       172.16.77.4:9200        TIME_WAIT  
tcp        0      0 172.16.60.2:56344       172.16.77.4:9200        TIME_WAIT  
tcp        0      0 172.16.60.2:56352       172.16.77.4:9200        TIME_WAIT

可以看到其的确是连接了 172.16.77.4 的9200端口。 而对方 elasticsearch-logging 容器的确开启了9200端口

[root@centos7-node-223 ~]$ docker exec  667cfd84c979  ss -l|grep LISTEN
tcp    LISTEN     0      50                  :::9200                 :::*       
tcp    LISTEN     0      50                  :::9300                 :::*

那么这个过程是如何完成的呢？？？

3.2.2 Container间通信流程####

192.168.1.224/fluentd-elasticsearch -> 192.168.1.223/elasticsearch-logging

192.168.1.224/fluentd-elasticsearch 需要连接到elasticsearch-logging容器.

• 域名到IP对应。 elasticsearch-logging -> 解析为10.254.24.205

root@fluentd-elasticsearch-centos7-node-224:/$ dig elasticsearch-logging

; <<>> DiG 9.9.5-3ubuntu0.5-Ubuntu <<>> elasticsearch-logging
;; global options: +cmd
;; Got answer:
;; ->>HEADER<<- opcode: QUERY, status: SERVFAIL, id: 39181
;; flags: qr rd ra; QUERY: 1, ANSWER: 0, AUTHORITY: 0, ADDITIONAL: 0

;; QUESTION SECTION:
;elasticsearch-logging.		IN	A

;; Query time: 1 msec
;; SERVER: 10.254.0.10#53(10.254.0.10)
;; WHEN: Fri Oct 09 06:04:35 UTC 2015
;; MSG SIZE  rcvd: 39

• 访问该IP地址10.254.24.205:9200端口。根据路由，请求将会到达网关172.16.60.1,也就是这个docker的宿主机的docker0网卡地址。

# 容器内
root@fluentd-elasticsearch-centos7-node-224:/$ ip route
default via 172.16.60.1 dev eth0 
172.16.60.0/24 dev eth0  proto kernel  scope link  src 172.16.60.2 
# 物理机
[root@centos7-node-224 ~]$ ifconfig docker0
docker0: flags=4163  mtu 1450
        inet 172.16.60.1  netmask 255.255.255.0  broadcast 0.0.0.0
        inet6 fe80::5484:7aff:fefe:9799  prefixlen 64  scopeid 0x20
        ether 56:84:7a:fe:97:99  txqueuelen 0  (Ethernet)
        RX packets 10182154  bytes 1777103288 (1.6 GiB)
        RX errors 0  dropped 0  overruns 0  frame 0
        TX packets 11195534  bytes 2271907616 (2.1 GiB)
        TX errors 0  dropped 0 overruns 0  carrier 0  collisions 0

• iptables负责转发请求到192.168.1.224:36967，而这个端口上kube-proxy进程在监听。

[root@centos7-node-224 ~]$ iptables-save  | grep 10.254.24.205 |grep 9200
-A KUBE-PORTALS-CONTAINER -d 10.254.24.205/32 -p tcp -m comment --comment "kube-system/elasticsearch-logging:" -m tcp --dport 9200 -j REDIRECT --to-ports 36967
-A KUBE-PORTALS-HOST -d 10.254.24.205/32 -p tcp -m comment --comment "kube-system/elasticsearch-logging:" -m tcp --dport 9200 -j DNAT --to-destination 192.168.1.224:36967
[root@centos7-node-224 ~]$ netstat -nlp|grep 36967
tcp6       0      0 :::36967                :::*                    LISTEN      930/kube-proxy

• 谁负责响应10.254.24.205:9200的请求？由上述分析，看起来是kube-proxy，那么kube-proxy进程看起来是个proxy，那么被转发给谁处理？当然给Pod啦。可以到这个服务的Selector是k8s-app=elasticsearch-logging

[root@centos7-node-224 ~]$ kubectl get svc --all-namespaces  | grep 10.254.24.205
kube-system   elasticsearch-logging   10.254.24.205            9200/TCP            k8s-app=elasticsearch-logging   14d

• 找到其对应的Pod为elasticsearch-logging-v1-gph5i和elasticsearch-logging-v1-i8x6q

[root@centos7-node-224 ~]$ kubectl get po -l k8s-app=elasticsearch-logging --all-namespaces
NAMESPACE     NAME                             READY     STATUS    RESTARTS   AGE
kube-system   elasticsearch-logging-v1-gph5i   1/1       Running   6          14d
kube-system   elasticsearch-logging-v1-i8x6q   1/1       Running   5          14d

• 我们查看其中的elasticsearch-logging-v1-i8x6q容器的IP地址，发现为172.16.77.4

[root@centos7-node-224 ~]$ kubectl describe po elasticsearch-logging-v1-i8x6q --namespace=kube-system | grep IP
IP:				172.16.77.4

• 故而很明显kube-proxy 会把部分请求转发给 其中的一个Pod来处理，而这个Pod的IP地址是172.16.77.4 . 而 172.16.77.4 这个Pod 在 192.168.1.223 机器上.

[root@centos7-node-223 ~]$ docker inspect 5201c8cbdebd | grep IPAddress
        "IPAddress": "172.16.77.4",
        "SecondaryIPAddresses": null,

• 那如何与172.16.77.4进行通信呢？跨机器之间通信则采用flannel等诸如此类的overlay网络或者ovs等L2网络。

以上是“Kubernetes网络分析中如何实现Container间通信”这篇文章的所有内容，感谢各位的阅读！希望分享的内容对大家有帮助，更多相关知识，欢迎关注创新互联行业资讯频道！

分享文章：Kubernetes网络分析中如何实现Container间通信
分享URL：http://bjjierui.cn/article/ijepsh.html