Tungsten Fabric入门宝典系列文章，来自技术大牛倾囊相授的实践经验，由TF中文社区为您编译呈现，旨在帮助新手深入理解TF的运行、安装、集成、调试等全流程。如果您有相关经验或疑问，欢迎与我们互动，并与社区极客们进一步交流。更多TF技术文章，请点击公号底部按钮＞学习＞文章合集。

作者：Tatsuya Naganawa 译者：TF编译组

在多个编排器之间共享控制平面有很多好处，包括routing/bridging、DNS、security等。

下面我来描述每种情况的使用方法和配置。

K8s+OpenStack

Kubernetes + OpenStack的组合已经涵盖并且运行良好。

https://github.com/Juniper/contrail-ansible-deployer/wiki/Deployment-Example:-Contrail-and-Kubernetes-and-Openstack

另外，Tungsten Fabric支持嵌套安装（nested installation）和非嵌套安装（non-nested installation），因此你可以选择其中一个选项。

https://github.com/Juniper/contrail-kubernetes-docs

K8s+K8s

将多个Kubernetes集群添加到一个Tungsten Fabric中，是一种可能的安装选项。

由于kube-manager支持cluster_name参数，该参数修改了将要创建的租户名称（默认为“k8s”），因此这应该是可行的。不过，我在上次尝试该方法时效果不佳，因为有些对象被其它kube-manager作为陈旧对象（stale object）删除了。

https://github.com/Juniper/contrail-container-builder/blob/master/containers/kubernetes/kube-manager/entrypoint.sh#L28

在将来的版本中，可能会更改此行为。

注意：

从R2002及更高版本开始，这个修补程序解决了该问题，并且不再需要自定义修补程序。

https://review.opencontrail.org/c/Juniper/contrail-controller/+/55758

注意：应用这些补丁，似乎可以将多个kube-master添加到一个Tungsten Fabric集群中。

diff --git a/src/container/kube-manager/kube_manager/kube_manager.py b/src/container/kube-manager/kube_manager/kube_manager.py
index 0f6f7a0..adb20a6 100644
--- a/src/container/kube-manager/kube_manager/kube_manager.py
+++ b/src/container/kube-manager/kube_manager/kube_manager.py
@@ -219,10 +219,10 @@ def main(args_str=None, kube_api_skip=False, event_queue=None,

     if args.cluster_id:
         client_pfx = args.cluster_id + '-'
-        zk_path_pfx = args.cluster_id + '/'
+        zk_path_pfx = args.cluster_id + '/' + args.cluster_name
     else:
         client_pfx = ''
-        zk_path_pfx = ''
+        zk_path_pfx = '' + args.cluster_name

     # randomize collector list
     args.random_collectors = args.collectors
diff --git a/src/container/kube-manager/kube_manager/vnc/vnc_namespace.py b/src/container/kube-manager/kube_manager/vnc/vnc_namespace.py
index 00cce81..f968cae 100644
--- a/src/container/kube-manager/kube_manager/vnc/vnc_namespace.py
+++ b/src/container/kube-manager/kube_manager/vnc/vnc_namespace.py
@@ -594,7 +594,8 @@ class VncNamespace(VncCommon):
                 self._queue.put(event)

     def namespace_timer(self):
-        self._sync_namespace_project()
+        # self._sync_namespace_project() ## temporary disabled
+        pass

     def _get_namespace_firewall_ingress_rule_name(self, ns_name):
         return "-".join([vnc_kube_config.cluster_name(),

由于kube-master创建的pod-network都在同一个Tungsten Fabric controller上，因此在它们之间实现路由泄漏（route-leak）是可能的:)

由于cluster_name将成为Tungsten Fabric的fw-policy中的标签之一，因此在多个Kubernetes集群之间也可以使用相同的标签

172.31.9.29 Tungsten Fabric controller
172.31.22.24 kube-master1 (KUBERNETES_CLUSTER_NAME=k8s1 is set)
172.31.12.82 kube-node1 (it belongs to kube-master1)
172.31.41.5 kube-master2(KUBERNETES_CLUSTER_NAME=k8s2 is set)
172.31.4.1 kube-node2 (it belongs to kube-master2)

[root@ip-172-31-22-24 ~]# kubectl get node
NAME                                              STATUS     ROLES    AGE   VERSION
ip-172-31-12-82.ap-northeast-1.compute.internal   Ready      <none>   57m   v1.12.3
ip-172-31-22-24.ap-northeast-1.compute.internal   NotReady   master   58m   v1.12.3
[root@ip-172-31-22-24 ~]# 

[root@ip-172-31-41-5 ~]# kubectl get node
NAME                                             STATUS     ROLES    AGE   VERSION
ip-172-31-4-1.ap-northeast-1.compute.internal    Ready      <none>   40m   v1.12.3
ip-172-31-41-5.ap-northeast-1.compute.internal   NotReady   master   40m   v1.12.3
[root@ip-172-31-41-5 ~]# 

[root@ip-172-31-22-24 ~]# kubectl get pod -o wide
NAME                                 READY   STATUS    RESTARTS   AGE     IP              NODE                                              NOMINATED NODE
cirros-deployment-75c98888b9-7pf82   1/1     Running   0          28m     10.47.255.249   ip-172-31-12-82.ap-northeast-1.compute.internal   <none>
cirros-deployment-75c98888b9-sgrc6   1/1     Running   0          28m     10.47.255.250   ip-172-31-12-82.ap-northeast-1.compute.internal   <none>
cirros-vn1                           1/1     Running   0          7m56s   10.0.1.3        ip-172-31-12-82.ap-northeast-1.compute.internal   <none>
[root@ip-172-31-22-24 ~]# 

[root@ip-172-31-41-5 ~]# kubectl get pod -o wide
NAME                                 READY   STATUS    RESTARTS   AGE     IP              NODE                                            NOMINATED NODE
cirros-deployment-75c98888b9-5lqzc   1/1     Running   0          27m     10.47.255.250   ip-172-31-4-1.ap-northeast-1.compute.internal   <none>
cirros-deployment-75c98888b9-dg8bf   1/1     Running   0          27m     10.47.255.249   ip-172-31-4-1.ap-northeast-1.compute.internal   <none>
cirros-vn2                           1/1     Running   0          5m36s   10.0.2.3        ip-172-31-4-1.ap-northeast-1.compute.internal   <none>
[root@ip-172-31-41-5 ~]# 

/ # ping 10.0.2.3
PING 10.0.2.3 (10.0.2.3): 56 data bytes
64 bytes from 10.0.2.3: seq=83 ttl=63 time=1.333 ms
64 bytes from 10.0.2.3: seq=84 ttl=63 time=0.327 ms
64 bytes from 10.0.2.3: seq=85 ttl=63 time=0.319 ms
64 bytes from 10.0.2.3: seq=86 ttl=63 time=0.325 ms
^C
--- 10.0.2.3 ping statistics ---
87 packets transmitted, 4 packets received, 95% packet loss
round-trip min/avg/max = 0.319/0.576/1.333 ms
/ # 
/ # ip -o a
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue qlen 1000\    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
1: lo    inet 127.0.0.1/8 scope host lo\       valid_lft forever preferred_lft forever
18: eth0@if19: <BROADCAST,MULTICAST,UP,LOWER_UP,M-DOWN> mtu 1500 qdisc noqueue \    link/ether 02:b9:11:c9:4c:b1 brd ff:ff:ff:ff:ff:ff
18: eth0    inet 10.0.1.3/24 scope global eth0\       valid_lft forever preferred_lft forever
/ #
 -> 在属于不同kubernetes 集群的Pod之间ping，工作良好

[root@ip-172-31-9-29 ~]# ./contrail-introspect-cli/ist.py ctr route show -t default-domain:k8s1-default:vn1:vn1.inet.0 

default-domain:k8s1-default:vn1:vn1.inet.0: 2 destinations, 2 routes (1 primary, 1 secondary, 0 infeasible)

10.0.1.3/32, age: 0:06:50.001343, last_modified: 2019-Jul-28 18:23:08.243656
    [XMPP (interface)|ip-172-31-12-82.local] age: 0:06:50.005553, localpref: 200, nh: 172.31.12.82, encap: ['gre', 'udp'], label: 50, AS path: None

10.0.2.3/32, age: 0:02:25.188713, last_modified: 2019-Jul-28 18:27:33.056286
    [XMPP (interface)|ip-172-31-4-1.local] age: 0:02:25.193517, localpref: 200, nh: 172.31.4.1, encap: ['gre', 'udp'], label: 50, AS path: None
[root@ip-172-31-9-29 ~]# 
[root@ip-172-31-9-29 ~]# ./contrail-introspect-cli/ist.py ctr route show -t default-domain:k8s2-default:vn2:vn2.inet.0 

default-domain:k8s2-default:vn2:vn2.inet.0: 2 destinations, 2 routes (1 primary, 1 secondary, 0 infeasible)

10.0.1.3/32, age: 0:02:36.482764, last_modified: 2019-Jul-28 18:27:33.055702
    [XMPP (interface)|ip-172-31-12-82.local] age: 0:02:36.489419, localpref: 200, nh: 172.31.12.82, encap: ['gre', 'udp'], label: 50, AS path: None

10.0.2.3/32, age: 0:04:37.126317, last_modified: 2019-Jul-28 18:25:32.412149
    [XMPP (interface)|ip-172-31-4-1.local] age: 0:04:37.133912, localpref: 200, nh: 172.31.4.1, encap: ['gre', 'udp'], label: 50, AS path: None
[root@ip-172-31-9-29 ~]#
 -> 基于以下的网络策略，每一个kube-master的每一个虚拟网络有路由去其他的kube-master 的Pod

(venv) [root@ip-172-31-9-29 ~]# contrail-api-cli --host 172.31.9.29 ls -l virtual-network
virtual-network/f9d06d27-8fc1-413d-a6d6-c51c42191ac0  default-domain:k8s2-default:vn2
virtual-network/384fb3ef-247b-42e6-a628-7111fe343f90  default-domain:k8s2-default:k8s2-default-service-network
virtual-network/c3098210-983b-46bc-b750-d06acfc66414  default-domain:k8s1-default:k8s1-default-pod-network
virtual-network/1ff6fdbd-ac2e-4601-b08c-5f7255466312  default-domain:default-project:ip-fabric
virtual-network/d8d95738-0a00-457f-b21e-60304859d1f9  default-domain:k8s2-default:k8s2-default-pod-network
virtual-network/0c075b76-4219-4f79-a4f5-1b4e6729f16e  default-domain:k8s1-default:k8s1-default-service-network
virtual-network/985b3b5f-84b7-4810-a54d-abd09a37f525  default-domain:k8s1-default:vn1
virtual-network/23782ea7-4000-491f-b20d-01c6ab9e2ba8  default-domain:default-project:default-virtual-network
virtual-network/90cce352-ef9b-4358-81b3-ef87a9cb63e8  default-domain:default-project:__link_local__
virtual-network/0292810c-c511-4147-89c0-9fdd571ccce8  default-domain:default-project:dci-network
(venv) [root@ip-172-31-9-29 ~]# 

(venv) [root@ip-172-31-9-29 ~]# contrail-api-cli --host 172.31.9.29 ls -l network-policy
network-policy/134d38b2-79e2-4a3e-a2f7-a3d61ceaf5e2  default-domain:k8s1-default:vn1-to-vn2  <-- route-leak between to kubernetes cluster
network-policy/8e5c5c4a-14eb-4fc4-9b46-81a5b923bbe0  default-domain:k8s1-default:k8s1-default-ip-fabric-np
network-policy/544d5076-3dff-45a1-bb47-8aec5e1e5a37  default-domain:k8s1-default:k8s1-default-pod-service-np
network-policy/33884d88-6492-4e0f-934c-080a794ce132  default-domain:k8s2-default:k8s2-default-ip-fabric-np
network-policy/232beb43-2008-4df3-969f-a4eee653ff46  default-domain:k8s2-default:k8s2-default-pod-service-np
network-policy/a6ee02bd-ad0d-4393-be60-66da8032237a  default-domain:k8s2-default:k8s2-default-service-np
network-policy/a9cedd67-127a-40fd-9f44-78890dc3cfe4  default-domain:k8s1-default:k8s1-default-service-np
(venv) [root@ip-172-31-9-29 ~]#

OpenStack+OpenStack

我还没有尝试将两个OpenStack集群添加到一个Tungsten Fabric controller中，但如果它们不使用相同的租户名称，那么应该是可行的。

K8s+vCenter

Kubernetes和vCenter的组合可以同时使用。用例类似于Kubernetes + OpenStack。

OpenStack+vCenter

OpenStack和vCenter的组合有点奇怪，因为OpenStack仪表盘可能用作vCenter网络的管理UI。

根据我的尝试，vcenter-plugin会检查所有可用租户下的所有virtual-network，而不是“vCenter”租户下的virtual-network，因此，如果创建了virtual-network或其它Neutron组件，也可以在ESXi上的vRouterVM上使用。通过此设置，vCenter用户可以自己实现网络功能，就像使用EC2 / VPC一样。