简介:KubeStack is an OpenStack network provider for kubernetes.KubeStack is devided into two functions:
- kubestack running on the same host with kube-controller-manager, which provides network management for openstack
- kubestack running on each minion host, which setups container's network interfaces
kubestack.conf的默认配置文件如下所示:
[Global] auth-url = http://${IF_IP}:5000/v2.0 username = admin password = admin tenant-name = admin region = RegionOne ext-net-id = ${EXT_NET_ID} [LoadBalancer] create-monitor = yes monitor-delay = 1m monitor-timeout = 30s monitor-max-retries = 3 [Plugin] plugin-name = ovs
// kubestack.go
1、func main()
1、调用config, err := os.Open(configFile)打开配置文件,configFile默认为/etc/kubestack.conf
2、调用openstack, err := common.NewOpenStack(config)
3、调用server := kubestack.NewKubeHandler(openstack)
4、最后,调用server.Serve(port)进行监听,端口号默认为4237
OpenStack数据结构如下所示:
// OpenStack is an implementation of network provider Interface for OpenStack type OpenStack struct { network *gophercloud.ServiceClient identity *gophercloud.ServiceClient provider *gophercloud.ProviderClient region string lbOpts LoadBalancerOpts pluginOpts PluginOpts ExtNetID string Plugin plugins.PluginInterface }
// pkg/common/openstack.go
2、func NewOpenStack(config io.Reader) (*OpenStack, error)
- 调用err := gcfg.ReadInto(&cfg, config)读取配置信息
- 调用provider, err := openstack.AuthenticatedClient(cfg.toAuthOptions()) 用于auth openstack
- 调用identity, err := openstack.NewIdentityV2(provider, gophercloud.EndpointOpts{Availability: gophercloud.AvailabilityAdmin})用于find identity endpoint
- 调用network, err := openstack.NewNetworkV2(provider, gophercloud.EndpointOpts{Region: cfg.Global.Region})
- 创建 os := OpenStack{
identity: identity,
network: network,
provider: provider,
region: cfg.Global.Region,
lbOpts: cfg.LoadBalancer,
pluginOpts: cfg.Plugin,
ExtNetID: cfg.Global.ExtNetID,
}
- 当cfg.Plugin.PluginName 不为nil时:
- integrationBriage := "br-int",若cfg.Plugin.IntegrationBridge不为"",则设置integrationBriage = cfg.Plugin.IntegrationBridge
- 调用plugin, _ := plugins.GetNetworkPlugin(cfg.Plugin.PluginName)
- 若plugin不为nil,则调用plugin.Init(integrationBriage),并且os.Plugin = plugin
- 最后,返回return &os, nil
// pkg/plugins/plugins.go
// GetNetworkPlugin creates an instance of the named network plugin, or nil if
// the name is unkown. The error return is only used if the named plugin
// was known but failed to initialize
3、func GetNetworkPlugin(name string) (PluginInterface, error)
该函数仅仅调用f, found := plugins[name]获取plugin的Factory函数f(),调用该f()函数生成相应的PluginInterface
(如果相应的Plugin被支持,则相应的包在初始化的时候就已经在init()函数中进行了注册,因此一定能在此时找到)
PluginInterface数据结构如下所示:
type PluginInterface interface { SetupInterface(podName, podInfraContainerID string, port *ports.Port, ipcidr, gateway string, dnsServers []string, containerRuntime string) error DestroyInterface(podName, podInfraContainerID string, port *ports.Port, containerRuntime string) error Init(integrationBridge string) error }
// pkg/kubestack/kubestack.go
4、func NewKubeHandler(driver *common.OpenStack) *kubeHandler
该函数仅仅只是初始化并注册grpc server,生成h := &KubeHandler{driver: driver, server: grpc.NewServer()}
kubeHandler的结构如下所示:
// KubeHandler forwards requests and responses between // the docker daemon and the plugin. type KubeHandler struct { driver *common.OpenStack server *grpc.Server }
-------------到此为止,kubestack 初始化完成,接下来为网络的创建过程-------------
// pkg/kubestack/kubestack.go
1、func (h *KubeHandler) CreateNetwork(c context.Context, req *provider.CreateNetworkRequest) (*provider.CommonResponse, error)
1、首先调用req.Network.TenantID = h.driver.ToTenantID(req.Network.TenantID),将tenantName转换为tenantID
2、再调用err := h.driver.CreateNetwork(req.Network)创建网络
Network数据结构如下所示:
type Network struct { Name string Uid string TenantID string SegmentID int32 Subnets []*Subnet // Status of network // Valid value: Initializing, Active, Pending, Failed, Terminating Status string }
Subnet数据结构如下所示:
type Subnet struct { Name string Uid string Cidr string Gateway string Tenantid string Dnsservers []string Routes []*Route }
// pkg/common/openstack.go
2、func (os *OpenStack) CreateNetwork(network *provider.Network) error
1、首先创建opts := networks.CreateOpts{
Name: network.Name,
AdminStateUp: &adminStateUp,
TenantID: network.TenantID,
}
再调用osNet, err := networks.Create(os.network, opts).Extract()创建网络
2、同样调用routerOpts := routers.CreateOpts{
Name: network.Name,
TenantID: network.TenantID,
GatewayInfo: &routers.GatewayInfo{NetworkID: os.ExtNetID},
}
再调用osRouter, err := routers.Create(os.network, routerOpts).Extract()创建router
3、最后,遍历network.Subnets创建子网
- 首先创建 subnetOpts := subnets.CreateOpts{
NetworkID: networkID, //其实就是osNet.ID
CIDR: sub.Cidr,
Name: sub.Name,
IPVersion: gophercloud.IPv4,
TenantID: network.TenantID,
GatewaryIP: &sub.Gateway,
DNSNameserver: sub.Dnsservers,
}
再调用s, err := subnets.Create(os.network, subnetOpts).Extract()创建子网
// 将子网加入router中
2. 创建opts := routers.AddInterfaceOpts{SubnetID: s.ID},
再调用routers.AddInterface(os.network, osRouter.ID, opts).Extract()
---------------------------------------创建pod-----------------------------------------
// pkg/kubestack/kubestack.go
1、func (h *KubeHandler) SetupPod(c context.Context, req *provider.SetupPodRequest) (*provider.CommonResponse, error)
该函数仅仅是调用h.driver.SetupPod(req.PodName, req.Namespace, req.PodInfraContainerID, req.Network, req.ContainerRuntime)
SetupPodRequest数据结构如下所示:
type SetupPodRequest struct { PodName string Namespace string ContainerRuntime string PodInfraContainerID string Network *Network }
// pkg/common/openstack.go
2、func (os *OpenStack) SetupPod(podName, namespace, podInfraContainerID string, network *provider.Network, containerRuntime string)
1、调用portName := os.BuildPortName(podName, namespace, network.Uid),该函数的作用仅仅是将podNamePrefix和上述三个参数连成一个字符串而已
2、获取dns server的ip,调用networkPorts, err := os.ListPorts(network.Uid, "network:dhcp"),再遍历networkPorts,调用dnsServers = append(dnsServers, p.FixedIPs[0].IPAddress)
3、调用port, err := os.GetPort(portName),从openstack中获取port,如果port不存在,则调用portWithBinding, err := os.CreatePort(network.Uid, network.TenantID, portName, podHostname)创建一个
4、如果port.DeviceOwner和实际的deviceOwner不符,则更新hostname
5、调用subnet, err := os.getProviderSubnet(port.FixedIPs[0].SubnetID)获取子网和网关
6、最后,调用os.Plugin.SetupInterface(podName+"_"+namespace, podInfraContainerID, port, fmt.Sprintf("%s/%d", port.FixedIPs[0].IPAddress, prefixSize), subnet.Gateway, dnsServers, containerRuntime)为pod设置网卡
-----------------------------------OVS plugin实现---------------------------------------
OVSPlugin的数据结构很简单,如下所示:
type OVSPlugin struct { IntegrationBridage string }
// pkg/plugins/openvswitch/openvswitch.go
1、func (p *OVSPlugin) Init(integrationBridge string) error
该函数仅仅将p.IntegrationBridage赋值为integrationBridge
// pkg/plugins/openvswitch/openvswitch.go
2、func (p *OVSPlugin) SetupInterface(podName, podInfraContainerID string, port *ports.Port, ipcidr, gatewary string, dnsServers []string, containerRuntime string) error
1、先调用err := p.SetupOVSInterface(podName, podInfraContainerID, port, ipcidr, gatewary, containerRuntime)创建OVS interface
2、再根据containerRuntime的不同,分别调用p.SetupDockerInterface(podName, podInfraContainerID, port, ipcidr, gateway)或者p.SetupHyperInterface(podName, podInfraContainerID, port, ipcidr, gateway, dnsServers)
// pkg/plugins/openvswitch/openvswitch.go
3、func (p *OVSPlugin) SetupOVSInterface(podName, podInfraContainerID string, port *ports.Port, ipcidr, gateway string, containerRuntime string)
1、该函数调用exec.RunCommand函数创建一个veth pair以及一个bridge,并将其中一个veth加入bridge中
2、调用ovs-vsctl,对另一个veth进行操作
// pkg/plugins/openvswitch/openvswitch.go
4、func (p *OVSPlugin) SetupDockerInterface(podName, podInfraContainerID string, port *ports.Port, ipcidr, gateway string) error
1、先创建一个veth对,tapName和vifName,并且将tapName加入之前创建的bridge中
2、将vifName的MAC地址设置为port.MACAddress
3、根据podInfraContainerID容器的pid找到对应的net ns,并把它软链接到/var/run/netns目录,从而能用ip命令对其直接进行操作。
4、将vifName加入net ns中,再删除其中的eth0网卡,再将vifName重命名为eth0
5、调用`ip addr add dev eth0 ipcidr`添加ip,以及`ip route add default via gateway`创建默认路由
// pkg/plugins/openvswitch/openvswitch.go
5、func (p *OVSPlugin) SetupHyperInterface(podName, podInfraContainerID string, port *ports.Port, ipcidr, gateway string, dnsServers []string)
1、首先获取bridge和tapName
2、创建interfaceSpec := map[string]string{
"bridge": bridge,
"ifname": tapName,
"mac": port.MACAddress,
"ip": ipcidr,
"gateway": gateway,
} 生成network interface的配置
3、调用podSpec, err := p.getHyperPodSpec(podName)获取原始的hyper配置,并解码至specData中
4、调用specData["interfaces"] = []map[string]string{interfaceSpec}将network interface的配置加入其中
5、在specData["dns"]的基础之上添加dnsServers
6、最后调用newPodSpec, err := json.Marshal(specData)和p.saveHyperPodSpec(string(newPodSpec), podName)将新的hyper配置写入文件中
注:Neutron Port代表的是一个逻辑交换机中的一个虚拟交换机端口。虚拟机实例将自己的网卡和这些端口相连,这些逻辑端口同时定义了接入到它们中的接口的MAC地址和IP地址。当一个IP地址和一个端口关联起来的时候,这意味着这个端口和一个子网相连,因为这个IP地址就是从一个特定子网的allocation pool中获取的。