kube-proxy当前支持三种方式实现负载均衡，分别是: userspace, Iptables, IPVS. 但前两者随着Service的数量增长，存在性能的瓶颈，在生产环境是不能接受的。所以本篇文章主要对IPVS模式进行源码分析。

代码版本: release-1.15

kube-proxy 整体逻辑结构


kube-proxy

这张时序图描述了kube-proxy的整体逻辑结构，由于kub-proxy组件和其它的kube-* 组件一样都是使用pflag和cobra库去构建命令行应用程序。所以先简单介绍下该包的基本使用方式:

1
2
3
4
5
6
7
8
9
10
11
12
13
func main() {
  command := &cobra.Command{
    Use:   "echo [string to echo]",
    Short: "Echo anything to the screen",
    Long: `echo is for echoing anything back.Echo works a lot like print, except it has a child command.`,
    Args: cobra.MinimumNArgs(1),
    Run: func(cmd *cobra.Command, args []string) {
      fmt.Println("Print: " + strings.Join(args, " "))
    },
  }

  command.Execute()
}

上面这段代码就是使用cobra包的一个最简单的例子，首先初始化Command结构，其中该结构中的Run就是最终要执行的真正逻辑。当初始化完成Command之后，通过commnad.Execute去启动应用程序。

现在看上面的图就能比较直观的理解程序的启动机制了，这张图的整体过程就是对Commnad结构中的Run进行核心逻辑实现。也就是说kube-proxy核心逻辑入口就是从这里开始(Command.Run)。😋

在Command.Run中主要做了如下几件事,看下面的代码:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
// Run runs the specified ProxyServer.
func (o *Options) Run() error {
	defer close(o.errCh)
	
	//....
	proxyServer, err := NewProxyServer(o)
	if err != nil {
		return err
	}

	if o.CleanupAndExit {
		return proxyServer.CleanupAndExit()
	}

	o.proxyServer = proxyServer
	return o.runLoop()
}

1.对ProxyServer实例进行初始化。
2.如果在启动kube-proxy服务时，CleanupAndExit参数设置为true,则会将userspace, iptables, ipvs三种模式之前设置的所有规则清除掉，然后直接退出。
3.如果在启动kube-proxy服务时，CleanupAndExit参数设置为flase,则会调用runLoop来启动ProxyServer服务。

首先先来看看ProxyServer的结构定义:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
type ProxyServer struct {
	Client                 clientset.Interface 
	EventClient            v1core.EventsGetter
	IptInterface           utiliptables.Interface
	IpvsInterface          utilipvs.Interface
	IpsetInterface         utilipset.Interface
	execer                 exec.Interface
	Proxier                proxy.ProxyProvider
	Broadcaster            record.EventBroadcaster
	Recorder               record.EventRecorder
	ConntrackConfiguration kubeproxyconfig.KubeProxyConntrackConfiguration
	Conntracker            Conntracker // if nil, ignored
	ProxyMode              string
	NodeRef                *v1.ObjectReference
	CleanupIPVS            bool
	MetricsBindAddress     string
	EnableProfiling        bool
	OOMScoreAdj            *int32
	ConfigSyncPeriod       time.Duration
	HealthzServer          *healthcheck.HealthzServer
}

在ProxyServer结构中:

包含了与kube-apiserver通信的Client。

操作Iptables的IptInterface:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
type Interface interface {
	// GetVersion returns the "X.Y.Z" version string for iptables.
	GetVersion() (string, error)
	// EnsureChain checks if the specified chain exists and, if not, creates it.  If the chain existed, return true.
	EnsureChain(table Table, chain Chain) (bool, error)
	// FlushChain clears the specified chain.  If the chain did not exist, return error.
	FlushChain(table Table, chain Chain) error
	// DeleteChain deletes the specified chain.  If the chain did not exist, return error.
	DeleteChain(table Table, chain Chain) error
	// EnsureRule checks if the specified rule is present and, if not, creates it.  If the rule existed, return true.
	EnsureRule(position RulePosition, table Table, chain Chain, args ...string) (bool, error)
	// DeleteRule checks if the specified rule is present and, if so, deletes it.
	DeleteRule(table Table, chain Chain, args ...string) error
	// IsIpv6 returns true if this is managing ipv6 tables
	IsIpv6() bool
	// SaveInto calls `iptables-save` for table and stores result in a given buffer.
	SaveInto(table Table, buffer *bytes.Buffer) error
	// Restore runs `iptables-restore` passing data through []byte.
	// table is the Table to restore
	// data should be formatted like the output of SaveInto()
	// flush sets the presence of the "--noflush" flag. see: FlushFlag
	// counters sets the "--counters" flag. see: RestoreCountersFlag
	Restore(table Table, data []byte, flush FlushFlag, counters RestoreCountersFlag) error
	// RestoreAll is the same as Restore except that no table is specified.
	RestoreAll(data []byte, flush FlushFlag, counters RestoreCountersFlag) error
	// AddReloadFunc adds a function to call on iptables reload
	AddReloadFunc(reloadFunc func())
	// Destroy cleans up resources used by the Interface
	Destroy()
}

操作IPVS的IpvsInterface:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
type Interface interface {
	// Flush clears all virtual servers in system. return occurred error immediately.
	Flush() error
	// AddVirtualServer creates the specified virtual server.
	AddVirtualServer(*VirtualServer) error
	// UpdateVirtualServer updates an already existing virtual server.  If the virtual server does not exist, return error.
	UpdateVirtualServer(*VirtualServer) error
	// DeleteVirtualServer deletes the specified virtual server.  If the virtual server does not exist, return error.
	DeleteVirtualServer(*VirtualServer) error
	// Given a partial virtual server, GetVirtualServer will return the specified virtual server information in the system.
	GetVirtualServer(*VirtualServer) (*VirtualServer, error)
	// GetVirtualServers lists all virtual servers in the system.
	GetVirtualServers() ([]*VirtualServer, error)
	// AddRealServer creates the specified real server for the specified virtual server.
	AddRealServer(*VirtualServer, *RealServer) error
	// GetRealServers returns all real servers for the specified virtual server.
	GetRealServers(*VirtualServer) ([]*RealServer, error)
	// DeleteRealServer deletes the specified real server from the specified virtual server.
	DeleteRealServer(*VirtualServer, *RealServer) error
	// UpdateRealServer updates the specified real server from the specified virtual server.
	UpdateRealServer(*VirtualServer, *RealServer) error
}

操作IpSet的IpsetInterface:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
// Interface is an injectable interface for running ipset commands.  Implementations must be goroutine-safe.
type Interface interface {
	// FlushSet deletes all entries from a named set.
	FlushSet(set string) error
	// DestroySet deletes a named set.
	DestroySet(set string) error
	// DestroyAllSets deletes all sets.
	DestroyAllSets() error
	// CreateSet creates a new set.  It will ignore error when the set already exists if ignoreExistErr=true.
	CreateSet(set *IPSet, ignoreExistErr bool) error
	// AddEntry adds a new entry to the named set.  It will ignore error when the entry already exists if ignoreExistErr=true.
	AddEntry(entry string, set *IPSet, ignoreExistErr bool) error
	// DelEntry deletes one entry from the named set
	DelEntry(entry string, set string) error
	// Test test if an entry exists in the named set
	TestEntry(entry string, set string) (bool, error)
	// ListEntries lists all the entries from a named set
	ListEntries(set string) ([]string, error)
	// ListSets list all set names from kernel
	ListSets() ([]string, error)
	// GetVersion returns the "X.Y" version string for ipset.
	GetVersion() (string, error)
}

以及通过ProxyMode参数获取基于userspace, iptables, ipvs三种方式中的哪种使用的Proxier:

1
2
3
4
5
6
7
8
9
10
11
12
// ProxyProvider is the interface provided by proxier implementations.
type ProxyProvider interface {
	config.EndpointsHandler
	config.ServiceHandler

	// Sync immediately synchronizes the ProxyProvider's current state to proxy rules.
	Sync()
	// SyncLoop runs periodic work.
	// This is expected to run as a goroutine or as the main loop of the app.
	// It does not return.
	SyncLoop()
}

接下来重点介绍基于ipvs模式实现的Proxier, 在ipvs模式下Proxier结构的定义:

注意:由于Proxier定义的内容太多,我这里把接下来介绍的内容保留，其它不准备介绍的内容删除掉，省的占用太多的篇幅。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
type Proxier struct {
	endpointsChanges *proxy.EndpointChangeTracker
	serviceChanges   *proxy.ServiceChangeTracker

	//...
	serviceMap   proxy.ServiceMap
	endpointsMap proxy.EndpointsMap
	portsMap     map[utilproxy.LocalPort]utilproxy.Closeable
	
	//...
	syncRunner      *async.BoundedFrequencyRunner // governs calls to syncProxyRules

	//...
	iptables       utiliptables.Interface
	ipvs           utilipvs.Interface
	ipset          utilipset.Interface
	exec           utilexec.Interface
	
	//...
	ipvsScheduler  string
}

在Proxier结构中，先介绍下async.BoundedFrequencyRunner,其它的在介绍ProxyServer.Run的时候介绍:

BoundedFrequencyRunner的定义结构如下:

1
2
3
4
5
6
7
8
9
10
11
12
13
type BoundedFrequencyRunner struct {
	name        string        // the name of this instance
	minInterval time.Duration // the min time between runs, modulo bursts
	maxInterval time.Duration // the max time between runs

	run chan struct{} // try an async run

	mu      sync.Mutex  // guards runs of fn and all mutations
	fn      func()      // function to run
	lastRun time.Time   // time of last run
	timer   timer       // timer for deferred runs
	limiter rateLimiter // rate limiter for on-demand runs
}

BoundedFrequencyRunner结构中的run会异步的去定期的执行任务fn，比如定期的执行proxier.syncProxyRules去创建或者更新VirtuaServer和RealServer并将VirtualServer的VIP绑定到dummy interface(kube-ipvs0)。

下面是在NewProxier方法中初始化BoundedFrequencyRunner对象的示例:

1
2
proxier.syncRunner = async.NewBoundedFrequencyRunner(
    "sync-runner", proxier.syncProxyRules, minSyncPeriod, syncPeriod, burstSyncs)

其中:

  • minSyncPeriod: 规则最小的更新时间
  • syncPeriod: 规则最大更新时间
  • proxier.syncProxyRules: 同步规则的实现函数(也是kube-proxy基于ipvs同步规则的核心实现)

接下来介绍下ProxyServer.Run的逻辑实现部分:

ProxyServer启动流程


kube-proxy

上面这张图是ProxyServer启动的整个流程图。在启动过程中，主要做了下面这几件事儿:

1.启动健康检查服务HealthzServer.
2.启动暴露监控指标的MetricsServer.
3.如果需要调整系统的conntrack相关参数,则对系统的conntrack进行参数调整.
4.创建一个informerFactory实例，后面去通过informerFactory获取kubernetes的各类资源数据.
5.创建一个ServiceConfig实例，这个实例主要作用是实时的WATCH kubernetes Service资源的变化，并加入队列中，用于后续对变化的Service进行规则同步。
6.注册servier event hander到Proxier.
7.启动serviceConfig.

接下来详细的介绍下[4-7]这几步的流程:

ServiceConfig的结构定义如下:

1
2
3
4
type ServiceConfig struct {
	listerSynced  cache.InformerSynced
	eventHandlers []ServiceHandler
}

ServiceHandler的结构定义如下:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
type ServiceHandler interface {
	// OnServiceAdd is called whenever creation of new service object
	// is observed.
	OnServiceAdd(service *v1.Service)
	// OnServiceUpdate is called whenever modification of an existing
	// service object is observed.
	OnServiceUpdate(oldService, service *v1.Service)
	// OnServiceDelete is called whenever deletion of an existing service
	// object is observed.
	OnServiceDelete(service *v1.Service)
	// OnServiceSynced is called once all the initial even handlers were
	// called and the state is fully propagated to local cache.
	OnServiceSynced()
}

创建ServiceConfig实例对象的具体实现如下:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
func NewServiceConfig(serviceInformer coreinformers.ServiceInformer, resyncPeriod time.Duration) *ServiceConfig {
	result := &ServiceConfig{
		listerSynced: serviceInformer.Informer().HasSynced,
	}

	serviceInformer.Informer().AddEventHandlerWithResyncPeriod(
		cache.ResourceEventHandlerFuncs{
			AddFunc:    result.handleAddService,
			UpdateFunc: result.handleUpdateService,
			DeleteFunc: result.handleDeleteService,
		},
		resyncPeriod,
	)

	return result
}
  • 首先通过执行serviceInformer.Informer().HasSynced来将kubernetes下的所有Service资源同步到缓存listerSynced中。
  • 其次为AddEventHandlerWithResyncPeriod添加针对Service对象，添加，更新，删除的事件触发函数。当Service有相应的触发动作，就会调用相应的函数:
1
2
3
handleAddService
handleUpdateService
handleDeleteService

我们看看handleAddService触发函数的实现逻辑，具体代码如下:

1
2
3
4
5
6
7
8
9
10
11
func (c *ServiceConfig) handleAddService(obj interface{}) {
	service, ok := obj.(*v1.Service)
	if !ok {
		utilruntime.HandleError(fmt.Errorf("unexpected object type: %v", obj))
		return
	}
	for i := range c.eventHandlers {
		klog.V(4).Info("Calling handler.OnServiceAdd")
		c.eventHandlers[i].OnServiceAdd(service)
	}
}

当watch到kubernetes集群中有新的Service被创建之后，会触发handleAddService函数，并在该函数中遍历eventHandlers分别去调用OnServiceAdd来对proxier结构中的serviceChanages进行更新并去同步相应的规则。

OnServiceAdd的具体实现逻辑如下:

1
2
3
4
5
6
7
8
9
10
11
// OnServiceAdd is called whenever creation of new service object is observed.
func (proxier *Proxier) OnServiceAdd(service *v1.Service) {
	proxier.OnServiceUpdate(nil, service)
}

// OnServiceUpdate is called whenever modification of an existing service object is observed.
func (proxier *Proxier) OnServiceUpdate(oldService, service *v1.Service) {
	if proxier.serviceChanges.Update(oldService, service) && proxier.isInitialized() {
		proxier.syncRunner.Run()
	}
}

ServiceChangeTracker的结构定义如下:

1
2
3
4
5
6
7
8
9
10
11
12
13
// ServiceChangeTracker carries state about uncommitted changes to an arbitrary number of
// Services, keyed by their namespace and name.
type ServiceChangeTracker struct {
	// lock protects items.
	lock sync.Mutex
	// items maps a service to its serviceChange.
	items map[types.NamespacedName]*serviceChange
	// makeServiceInfo allows proxier to inject customized information when processing service.
	makeServiceInfo makeServicePortFunc
	// isIPv6Mode indicates if change tracker is under IPv6/IPv4 mode. Nil means not applicable.
	isIPv6Mode *bool
	recorder   record.EventRecorder
}

serviceChanage的结构定义如下:

1
2
3
4
5
6
7
// serviceChange contains all changes to services that happened since proxy rules were synced.  For a single object,
// changes are accumulated, i.e. previous is state from before applying the changes,
// current is state after applying all of the changes.
type serviceChange struct {
	previous ServiceMap
	current  ServiceMap
}

到这里在回过头来看上面的基于IPVS实现的Proxier的整体流程就完全通了，ProxyServer.Run函数在启动时，通过kubernetes LIST/WATCH机制去实时的感知kubernetes集群Service资源的变化，然后不断的在更新Proxier结构中的ServiceChanges，然后将变化的Service保存在ServiceChanges结构中的ServiceMap中，给后续的async.BoundedFrequencyRunner去执行同步规则函数syncProxyRules来使用。

8.endpointConfig的实现机制和serviceConfig的机制完全一样，这里就不在详细的介绍了。
9.上面做的所有预处理工作，会在informerFactory.Start这步生效。
10.birthCry的作用就是通过event的方式通知kubernetes, kube-proxy这边的所有准备工作都处理好了，我要启动了。

1
2
3
func (s *ProxyServer) birthCry() {
	s.Recorder.Eventf(s.NodeRef, api.EventTypeNormal, "Starting", "Starting kube-proxy.")
}

11.最终通过SyncLoop启动kube-proxy服务，并立刻执行syncProxyRules先来一遍同步在说.之后便会通过异步的方式定期的去同步IPVS, Iptables, Ipset的规则通过syncProxyRules函数:

而syncProxyRules函数是kube-proxy实现的核心。主体逻辑是遍历ServiceMap并遍历ServiceMap下的endpointsMap及创建的Service类型(如: CLusterIP, Loadbalancer, NodePort)去分别创建相应的IPVS规则。

syncProxyRules的函数实现定义如下:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
func (proxier *Proxier) syncProxyRules() {
	//.....

	// Build IPVS rules for each service.
	for svcName, svc := range proxier.serviceMap {
		//......

		// Handle traffic that loops back to the originator with SNAT.
		for _, e := range proxier.endpointsMap[svcName] {
			//....
		}

		// Capture the clusterIP.
		// ipset call
		entry := &utilipset.Entry{
			IP:       svcInfo.ClusterIP().String(),
			Port:     svcInfo.Port(),
			Protocol: protocol,
			SetType:  utilipset.HashIPPort,
		}
		// add service Cluster IP:Port to kubeServiceAccess ip set for the purpose of solving hairpin.
		// proxier.kubeServiceAccessSet.activeEntries.Insert(entry.String())
		if valid := proxier.ipsetList[kubeClusterIPSet].validateEntry(entry); !valid {
			klog.Errorf("%s", fmt.Sprintf(EntryInvalidErr, entry, proxier.ipsetList[kubeClusterIPSet].Name))
			continue
		}
		proxier.ipsetList[kubeClusterIPSet].activeEntries.Insert(entry.String())
		// ipvs call
		serv := &utilipvs.VirtualServer{
			Address:   svcInfo.ClusterIP(),
			Port:      uint16(svcInfo.Port()),
			Protocol:  string(svcInfo.Protocol()),
			Scheduler: proxier.ipvsScheduler,
		}
		// Set session affinity flag and timeout for IPVS service
		if svcInfo.SessionAffinityType() == v1.ServiceAffinityClientIP {
			serv.Flags |= utilipvs.FlagPersistent
			serv.Timeout = uint32(svcInfo.StickyMaxAgeSeconds())
		}
		// We need to bind ClusterIP to dummy interface, so set `bindAddr` parameter to `true` in syncService()
		if err := proxier.syncService(svcNameString, serv, true); err == nil {
			activeIPVSServices[serv.String()] = true
			activeBindAddrs[serv.Address.String()] = true
			// ExternalTrafficPolicy only works for NodePort and external LB traffic, does not affect ClusterIP
			// So we still need clusterIP rules in onlyNodeLocalEndpoints mode.
			if err := proxier.syncEndpoint(svcName, false, serv); err != nil {
				klog.Errorf("Failed to sync endpoint for service: %v, err: %v", serv, err)
			}
		} else {
			klog.Errorf("Failed to sync service: %v, err: %v", serv, err)
		}

		// Capture externalIPs.
		for _, externalIP := range svcInfo.ExternalIPStrings() {
			//....
		}

		// Capture load-balancer ingress.
		for _, ingress := range svcInfo.LoadBalancerIPStrings() {
			//.....
		}

		if svcInfo.NodePort() != 0 {
			//....
		}
	}

	// sync ipset entries
	for _, set := range proxier.ipsetList {
		set.syncIPSetEntries()
	}

	// Tail call iptables rules for ipset, make sure only call iptables once
	// in a single loop per ip set.
	proxier.writeIptablesRules()

	// Sync iptables rules.
	// NOTE: NoFlushTables is used so we don't flush non-kubernetes chains in the table.
	proxier.iptablesData.Reset()
	proxier.iptablesData.Write(proxier.natChains.Bytes())
	proxier.iptablesData.Write(proxier.natRules.Bytes())
	proxier.iptablesData.Write(proxier.filterChains.Bytes())
	proxier.iptablesData.Write(proxier.filterRules.Bytes())

}

总结

kube-proxy的代码逻辑还是比较简洁的，整体的思想就是kube-proxy服务去watch kubernetes集群的Service和Endpoint对象，当这两个资源对象有状态变化时，会把它们保存在ServiceMap和EndPonintMap中，然后会通过async.BoundedFrequencyRunner去异步的执行syncProxyRules去下发规则。