kubelet注册csi driver的相关功能代码与kubelet的pluginManager有关，所以接下来对pluginManager进行分析。分析将分为pluginManager的初始化分析以及pluginManager的运行（处理逻辑）分析。

基于tag v1.17.4

https://github.com/kubernetes/kubernetes/releases/tag/v1.17.4

kubelet注册csi driver的原理

kubelet的pluginManager会监听某个特定目录，而负责向kubelet注册csi driver的组件Node Driver Registrar会创建暴露服务的socket在该目录下（每个plugin会对应一个Node Driver Registrar组件，也就是说，一个Node Driver Registrar只负责一个plugin的注册工作），pluginManager通过Node Driver Registrar组件暴露的socket获取plugin信息（包括plugin的socket地址、plugin名称等），从而最终做到根据该目录下socket文件的新增/删除来做相应的plugin注册/取消注册操作。

该图是Node Driver Registrar向kubelet注册csi driver的步骤流程图，这里大概看一下，具体请看这篇博客：Node Driver Registrar源码分析，结合本篇博客一起理解kubelet注册csi driver的整个过程。

plugin注册完成后，后续kubelet将通过plugin暴露的socket与plugin进行通信，做存储挂载/解除挂载等操作。

Node Driver Registrar

Node Driver Registrar在前面的文章中介绍过，它是一个sidecar容器，通过Kubelet的插件注册机制将CSI plugin（csi driver，两个名词意义一样）注册到Kubelet，让kubelet做volume的mount/umount操作时知道怎么调用相应的csi plugin。

1 pluginManager的初始化

调用NewMainKubelet()初始化kubelet的时候，会调用pluginmanager.NewPluginManager来初始化pluginManager，所以把NewMainKubelet()作为分析入口。

NewMainKubelet()

NewMainKubelet()中调用了pluginmanager.NewPluginManager来初始化pluginManager。

这里留意klet.getPluginsRegistrationDir()，调用该方法实际会返回plugins_registry，而该sockDir会传参进入pluginManager的desiredStateOfWorldPopulator结构体当中，相当于pluginManager会监听plugins_registry目录（负责向kubelet注册csi driver的组件Node Driver Registrar会创建暴露服务的socket在该目录下），pluginManager通过Node Driver Registrar组件暴露的socket获取plugin信息（包括plugin的socket地址、plugin名称等），从而最终做到根据该目录下socket文件的新增/删除来做相应的plugin注册/取消注册操作。

// pkg/kubelet/kubelet.go
// NewMainKubelet instantiates a new Kubelet object along with all the required internal modules.
// No initialization of Kubelet and its modules should happen here.
func NewMainKubelet(kubeCfg *kubeletconfiginternal.KubeletConfiguration,
    ...
    nodeStatusMaxImages int32) (*Kubelet, error) {
    ...
    klet.pluginManager = pluginmanager.NewPluginManager(
        klet.getPluginsRegistrationDir(), /* sockDir */
        kubeDeps.Recorder,
    )
    ...
}


// pkg/kubelet/pluginmanager/plugin_manager.go
// NewPluginManager returns a new concrete instance implementing the
// PluginManager interface.
func NewPluginManager(
    sockDir string,
    recorder record.EventRecorder) PluginManager {
    asw := cache.NewActualStateOfWorld()
    dsw := cache.NewDesiredStateOfWorld()
    reconciler := reconciler.NewReconciler(
        operationexecutor.NewOperationExecutor(
            operationexecutor.NewOperationGenerator(
                recorder,
            ),
        ),
        loopSleepDuration,
        dsw,
        asw,
    )

    pm := &pluginManager{
        desiredStateOfWorldPopulator: pluginwatcher.NewWatcher(
            sockDir,
            dsw,
        ),
        reconciler:          reconciler,
        desiredStateOfWorld: dsw,
        actualStateOfWorld:  asw,
    }
    return pm
}

klet.getPluginsRegistrationDir()

调用klet.getPluginsRegistrationDir()会返回plugins_registry。

// pkg/kubelet/kubelet_getters.go
// getPluginsRegistrationDir returns the full path to the directory under which
// plugins socket should be placed to be registered.
// More information is available about plugin registration in the pluginwatcher
// module
func (kl *Kubelet) getPluginsRegistrationDir() string {
    return filepath.Join(kl.getRootDir(), config.DefaultKubeletPluginsRegistrationDirName)
}


// pkg/kubelet/config/defaults.go
const (
    ...
    DefaultKubeletPluginsRegistrationDirName = "plugins_registry"
    ...
)

2 pluginManager struct

再来看到pluginManager结构体，pluginManager结构体与volumeManager结构体类似，都有actualStateOfWorld与desiredStateOfWorld两个属性。

kubelet pluginManager监听的socket注册目录每增加/删除一个socket文件，都会写入desiredStateOfWorld中/从desiredStateOfWorld中删除。

// pkg/kubelet/pluginmanager/plugin_manager.go
// pluginManager implements the PluginManager interface
type pluginManager struct {
    // desiredStateOfWorldPopulator (the plugin watcher) runs an asynchronous
    // periodic loop to populate the desiredStateOfWorld.
    desiredStateOfWorldPopulator *pluginwatcher.Watcher

    // reconciler runs an asynchronous periodic loop to reconcile the
    // desiredStateOfWorld with the actualStateOfWorld by triggering register
    // and unregister operations using the operationExecutor.
    reconciler reconciler.Reconciler

    // actualStateOfWorld is a data structure containing the actual state of
    // the world according to the manager: i.e. which plugins are registered.
    // The data structure is populated upon successful completion of register
    // and unregister actions triggered by the reconciler.
    actualStateOfWorld cache.ActualStateOfWorld

    // desiredStateOfWorld is a data structure containing the desired state of
    // the world according to the plugin manager: i.e. what plugins are registered.
    // The data structure is populated by the desired state of the world
    // populator (plugin watcher).
    desiredStateOfWorld cache.DesiredStateOfWorld
}

actualStateOfWorld

actualStateOfWorld结构体中存放的是已经完成了plugin注册操作的Node Driver Registrar组件暴露的socket相关信息。

// ActualStateOfWorld defines a set of thread-safe operations for the kubelet
// plugin manager's actual state of the world cache.
// This cache contains a map of socket file path to plugin information of
// all plugins attached to this node.
type ActualStateOfWorld interface {

    // GetRegisteredPlugins generates and returns a list of plugins
    // that are successfully registered plugins in the current actual state of world.
    GetRegisteredPlugins() []PluginInfo

    // AddPlugin add the given plugin in the cache.
    // An error will be returned if socketPath of the PluginInfo object is empty.
    // Note that this is different from desired world cache's AddOrUpdatePlugin
    // because for the actual state of world cache, there won't be a scenario where
    // we need to update an existing plugin if the timestamps don't match. This is
    // because the plugin should have been unregistered in the reconciller and therefore
    // removed from the actual state of world cache first before adding it back into
    // the actual state of world cache again with the new timestamp
    AddPlugin(pluginInfo PluginInfo) error

    // RemovePlugin deletes the plugin with the given socket path from the actual
    // state of world.
    // If a plugin does not exist with the given socket path, this is a no-op.
    RemovePlugin(socketPath string)

    // PluginExists checks if the given plugin exists in the current actual
    // state of world cache with the correct timestamp
    PluginExistsWithCorrectTimestamp(pluginInfo PluginInfo) bool
}

// NewActualStateOfWorld returns a new instance of ActualStateOfWorld
func NewActualStateOfWorld() ActualStateOfWorld {
    return &actualStateOfWorld{
        socketFileToInfo: make(map[string]PluginInfo),
    }
}

type actualStateOfWorld struct {

    // socketFileToInfo is a map containing the set of successfully registered plugins
    // The keys are plugin socket file paths. The values are PluginInfo objects
    socketFileToInfo map[string]PluginInfo
    sync.RWMutex
}

desiredStateOfWorld

desiredStateOfWorld结构体中存放的是在pluginManager监听目录下存在的，希望完成plugin注册操作的Node Driver Registrar组件暴露的socket相关信息。

// DesiredStateOfWorld defines a set of thread-safe operations for the kubelet
// plugin manager's desired state of the world cache.
// This cache contains a map of socket file path to plugin information of
// all plugins attached to this node.
type DesiredStateOfWorld interface {
    // AddOrUpdatePlugin add the given plugin in the cache if it doesn't already exist.
    // If it does exist in the cache, then the timestamp of the PluginInfo object in the cache will be updated.
    // An error will be returned if socketPath is empty.
    AddOrUpdatePlugin(socketPath string) error

    // RemovePlugin deletes the plugin with the given socket path from the desired
    // state of world.
    // If a plugin does not exist with the given socket path, this is a no-op.
    RemovePlugin(socketPath string)

    // GetPluginsToRegister generates and returns a list of plugins
    // in the current desired state of world.
    GetPluginsToRegister() []PluginInfo

    // PluginExists checks if the given socket path exists in the current desired
    // state of world cache
    PluginExists(socketPath string) bool
}

// NewDesiredStateOfWorld returns a new instance of DesiredStateOfWorld.
func NewDesiredStateOfWorld() DesiredStateOfWorld {
    return &desiredStateOfWorld{
        socketFileToInfo: make(map[string]PluginInfo),
    }
}

type desiredStateOfWorld struct {

    // socketFileToInfo is a map containing the set of successfully registered plugins
    // The keys are plugin socket file paths. The values are PluginInfo objects
    socketFileToInfo map[string]PluginInfo
    sync.RWMutex
}

3 pluginManager的运行

上面介绍了pluginManager的初始化，接下来介绍pluginManager的运行也即Run方法进行分析，分析一下pluginManager的处理逻辑。

因为调用逻辑比较复杂，这里直接跳过了调用过程的分析，直接进入kl.pluginManager.Run()的分析，下面只给出该方法的一个调用链：

kubelet的Run()方法（pkg/kubelet/kubelet.go） --> kl.updateRuntimeUp()（pkg/kubelet/kubelet.go） --> kl.initializeRuntimeDependentModules()（pkg/kubelet/kubelet.go） --> kl.pluginManager.Run()

kl.pluginManager.Run

下面直接看到kl.pluginManager.Run的代码。

该方法主要逻辑有两个：

（1）pm.desiredStateOfWorldPopulator.Start()：持续监听plugin的socket注册目录的变化事件，将Node Driver Registrar的socket信息写入desiredStateOfWorld中/从desiredStateOfWorld中删除；

（2）pm.reconciler.Run()。

// pkg/kubelet/pluginmanager/plugin_manager.go
func (pm *pluginManager) Run(sourcesReady config.SourcesReady, stopCh <-chan struct{}) {
    defer runtime.HandleCrash()

    pm.desiredStateOfWorldPopulator.Start(stopCh)
    klog.V(2).Infof("The desired_state_of_world populator (plugin watcher) starts")

    klog.Infof("Starting Kubelet Plugin Manager")
    go pm.reconciler.Run(stopCh)

    metrics.Register(pm.actualStateOfWorld, pm.desiredStateOfWorld)
    <-stopCh
    klog.Infof("Shutting down Kubelet Plugin Manager")
}

3.1 pm.desiredStateOfWorldPopulator.Start()

跑一个goroutine，持续监听plugin的socket注册目录的变化事件：

（1）当变化事件为新增事件时，即socket目录下多了文件，则调用w.handleCreateEvent，将该socket加入到desiredStateOfWorld中；

（2）当变化事件为删除事件时，即socket目录下删除了文件，则调用w.handleDeleteEvent，将该socket从desiredStateOfWorld中删除。

// pkg/kubelet/pluginmanager/pluginwatcher/plugin_watcher.go
// Start watches for the creation and deletion of plugin sockets at the path
func (w *Watcher) Start(stopCh <-chan struct{}) error {
    klog.V(2).Infof("Plugin Watcher Start at %s", w.path)

    w.stopped = make(chan struct{})

    // Creating the directory to be watched if it doesn't exist yet,
    // and walks through the directory to discover the existing plugins.
    if err := w.init(); err != nil {
        return err
    }

    fsWatcher, err := fsnotify.NewWatcher()
    if err != nil {
        return fmt.Errorf("failed to start plugin fsWatcher, err: %v", err)
    }
    w.fsWatcher = fsWatcher

    // Traverse plugin dir and add filesystem watchers before starting the plugin processing goroutine.
    if err := w.traversePluginDir(w.path); err != nil {
        klog.Errorf("failed to traverse plugin socket path %q, err: %v", w.path, err)
    }

    go func(fsWatcher *fsnotify.Watcher) {
        defer close(w.stopped)
        for {
            select {
            case event := <-fsWatcher.Events:
                //TODO: Handle errors by taking corrective measures
                if event.Op&fsnotify.Create == fsnotify.Create {
                    err := w.handleCreateEvent(event)
                    if err != nil {
                        klog.Errorf("error %v when handling create event: %s", err, event)
                    }
                } else if event.Op&fsnotify.Remove == fsnotify.Remove {
                    w.handleDeleteEvent(event)
                }
                continue
            case err := <-fsWatcher.Errors:
                if err != nil {
                    klog.Errorf("fsWatcher received error: %v", err)
                }
                continue
            case <-stopCh:
                // In case of plugin watcher being stopped by plugin manager, stop
                // probing the creation/deletion of plugin sockets.
                // Also give all pending go routines a chance to complete
                select {
                case <-w.stopped:
                case <-time.After(11 * time.Second):
                    klog.Errorf("timeout on stopping watcher")
                }
                w.fsWatcher.Close()
                return
            }
        }
    }(fsWatcher)

    return nil
}

w.handleCreateEvent()

w.handleCreateEvent()主要逻辑：

（1）判断新增事件是否为文件，且是否是socket文件；

（2）是socket文件，则调用w.handlePluginRegistration做处理，主要是将该socket加入到desiredStateOfWorld中。

// pkg/kubelet/pluginmanager/pluginwatcher/plugin_watcher.go
func (w *Watcher) handleCreateEvent(event fsnotify.Event) error {
    klog.V(6).Infof("Handling create event: %v", event)

    fi, err := os.Stat(event.Name)
    if err != nil {
        return fmt.Errorf("stat file %s failed: %v", event.Name, err)
    }

    if strings.HasPrefix(fi.Name(), ".") {
        klog.V(5).Infof("Ignoring file (starts with '.'): %s", fi.Name())
        return nil
    }

    if !fi.IsDir() {
        isSocket, err := util.IsUnixDomainSocket(util.NormalizePath(event.Name))
        if err != nil {
            return fmt.Errorf("failed to determine if file: %s is a unix domain socket: %v", event.Name, err)
        }
        if !isSocket {
            klog.V(5).Infof("Ignoring non socket file %s", fi.Name())
            return nil
        }

        return w.handlePluginRegistration(event.Name)
    }

    return w.traversePluginDir(event.Name)
}

func (w *Watcher) handlePluginRegistration(socketPath string) error {
    if runtime.GOOS == "windows" {
        socketPath = util.NormalizePath(socketPath)
    }
    //TODO: Implement rate limiting to mitigate any DOS kind of attacks.
    // Update desired state of world list of plugins
    // If the socket path does exist in the desired world cache, there's still
    // a possibility that it has been deleted and recreated again before it is
    // removed from the desired world cache, so we still need to call AddOrUpdatePlugin
    // in this case to update the timestamp
    klog.V(2).Infof("Adding socket path or updating timestamp %s to desired state cache", socketPath)
    err := w.desiredStateOfWorld.AddOrUpdatePlugin(socketPath)
    if err != nil {
        return fmt.Errorf("error adding socket path %s or updating timestamp to desired state cache: %v", socketPath, err)
    }
    return nil
}

w.handleDeleteEvent()

w.handleDeleteEvent()主要逻辑：

（1）将socket从desiredStateOfWorld中删除。

// pkg/kubelet/pluginmanager/pluginwatcher/plugin_watcher.go
func (w *Watcher) handleDeleteEvent(event fsnotify.Event) {
    klog.V(6).Infof("Handling delete event: %v", event)

    socketPath := event.Name
    klog.V(2).Infof("Removing socket path %s from desired state cache", socketPath)
    w.desiredStateOfWorld.RemovePlugin(socketPath)
}

3.2 pm.reconciler.Run()

pm.reconciler.Run()主要逻辑为对比desiredStateOfWorld与actualStateOfWorld做调谐，做plugin的注册操作/取消注册操作。具体逻辑如下：

（1）对比actualStateOfWorld，如果desiredStateOfWorld中没有该socket信息，或者desiredStateOfWorld中该socket的Timestamp值与actualStateOfWorld中的不相等（即plugin更新了），则说明该plugin需要取消注册（更新的plugin需先取消注册，然后再次注册），调用rc.operationExecutor.UnregisterPlugin做plugin取消注册操作；

（2）对比desiredStateOfWorld，如果actualStateOfWorld中没有该socket信息，则调用rc.operationExecutor.RegisterPlugin做plugin注册操作。

// pkg/kubelet/pluginmanager/reconciler/reconciler.go
func (rc *reconciler) Run(stopCh <-chan struct{}) {
    wait.Until(func() {
        rc.reconcile()
    },
        rc.loopSleepDuration,
        stopCh)
}

func (rc *reconciler) reconcile() {
    // Unregisterations are triggered before registrations

    // Ensure plugins that should be unregistered are unregistered.
    for _, registeredPlugin := range rc.actualStateOfWorld.GetRegisteredPlugins() {
        unregisterPlugin := false
        if !rc.desiredStateOfWorld.PluginExists(registeredPlugin.SocketPath) {
            unregisterPlugin = true
        } else {
            // We also need to unregister the plugins that exist in both actual state of world
            // and desired state of world cache, but the timestamps don't match.
            // Iterate through desired state of world plugins and see if there's any plugin
            // with the same socket path but different timestamp.
            for _, dswPlugin := range rc.desiredStateOfWorld.GetPluginsToRegister() {
                if dswPlugin.SocketPath == registeredPlugin.SocketPath && dswPlugin.Timestamp != registeredPlugin.Timestamp {
                    klog.V(5).Infof(registeredPlugin.GenerateMsgDetailed("An updated version of plugin has been found, unregistering the plugin first before reregistering", ""))
                    unregisterPlugin = true
                    break
                }
            }
        }

        if unregisterPlugin {
            klog.V(5).Infof(registeredPlugin.GenerateMsgDetailed("Starting operationExecutor.UnregisterPlugin", ""))
            err := rc.operationExecutor.UnregisterPlugin(registeredPlugin.SocketPath, rc.getHandlers(), rc.actualStateOfWorld)
            if err != nil &&
                !goroutinemap.IsAlreadyExists(err) &&
                !exponentialbackoff.IsExponentialBackoff(err) {
                // Ignore goroutinemap.IsAlreadyExists and exponentialbackoff.IsExponentialBackoff errors, they are expected.
                // Log all other errors.
                klog.Errorf(registeredPlugin.GenerateErrorDetailed("operationExecutor.UnregisterPlugin failed", err).Error())
            }
            if err == nil {
                klog.V(1).Infof(registeredPlugin.GenerateMsgDetailed("operationExecutor.UnregisterPlugin started", ""))
            }
        }
    }

    // Ensure plugins that should be registered are registered
    for _, pluginToRegister := range rc.desiredStateOfWorld.GetPluginsToRegister() {
        if !rc.actualStateOfWorld.PluginExistsWithCorrectTimestamp(pluginToRegister) {
            klog.V(5).Infof(pluginToRegister.GenerateMsgDetailed("Starting operationExecutor.RegisterPlugin", ""))
            err := rc.operationExecutor.RegisterPlugin(pluginToRegister.SocketPath, pluginToRegister.Timestamp, rc.getHandlers(), rc.actualStateOfWorld)
            if err != nil &&
                !goroutinemap.IsAlreadyExists(err) &&
                !exponentialbackoff.IsExponentialBackoff(err) {
                // Ignore goroutinemap.IsAlreadyExists and exponentialbackoff.IsExponentialBackoff errors, they are expected.
                klog.Errorf(pluginToRegister.GenerateErrorDetailed("operationExecutor.RegisterPlugin failed", err).Error())
            }
            if err == nil {
                klog.V(1).Infof(pluginToRegister.GenerateMsgDetailed("operationExecutor.RegisterPlugin started", ""))
            }
        }
    }
}

3.2.1 rc.operationExecutor.UnregisterPlugin()

rc.operationExecutor.UnregisterPlugin()主要逻辑：做plugin取消注册操作。

那plugin取消注册操作具体做了什么呢？继续往下分析。

plugin取消注册操作方法调用链

kl.pluginManager.Run --> pm.desiredStateOfWorldPopulator.Start() --> pm.reconciler.Run() --> rc.reconcile() --> rc.operationExecutor.UnregisterPlugin() --> oe.operationGenerator.GenerateUnregisterPluginFunc() --> handler.DeRegisterPlugin() --> nim.UninstallCSIDriver() --> nim.updateNode()

下面来对plugin取消注册操作的部分关键方法进行分析。

GenerateUnregisterPluginFunc

下面来分析下GenerateUnregisterPluginFunc的逻辑，主要是定义并实现一个plugin取消注册的方法，然后返回。plugin取消注册方法主要逻辑如下：

（1）检测Node Driver Registrar组件socket的连通性；

（2）通过Node Driver Registrar组件socket获取plugin信息；

（3）从actualStateOfWorld中删除该Node Driver Registrar组件的socket信息；

（4）调用handler.DeRegisterPlugin做进一步的plugin取消注册操作。

所以接下来会对handler.DeRegisterPlugin方法进行分析。

// pkg/kubelet/pluginmanager/operationexecutor/operation_executor.go
func (oe *operationExecutor) UnregisterPlugin(
    socketPath string,
    pluginHandlers map[string]cache.PluginHandler,
    actualStateOfWorld ActualStateOfWorldUpdater) error {
    generatedOperation :=
        oe.operationGenerator.GenerateUnregisterPluginFunc(socketPath, pluginHandlers, actualStateOfWorld)

    return oe.pendingOperations.Run(
        socketPath, generatedOperation)
}


// pkg/kubelet/pluginmanager/operationexecutor/operation_generator.go
func (og *operationGenerator) GenerateUnregisterPluginFunc(
    socketPath string,
    pluginHandlers map[string]cache.PluginHandler,
    actualStateOfWorldUpdater ActualStateOfWorldUpdater) func() error {

    unregisterPluginFunc := func() error {
        client, conn, err := dial(socketPath, dialTimeoutDuration)
        if err != nil {
            return fmt.Errorf("UnregisterPlugin error -- dial failed at socket %s, err: %v", socketPath, err)
        }
        defer conn.Close()

        ctx, cancel := context.WithTimeout(context.Background(), time.Second)
        defer cancel()

        infoResp, err := client.GetInfo(ctx, &registerapi.InfoRequest{})
        if err != nil {
            return fmt.Errorf("UnregisterPlugin error -- failed to get plugin info using RPC GetInfo at socket %s, err: %v", socketPath, err)
        }

        handler, ok := pluginHandlers[infoResp.Type]
        if !ok {
            return fmt.Errorf("UnregisterPlugin error -- no handler registered for plugin type: %s at socket %s", infoResp.Type, socketPath)
        }

        // We remove the plugin to the actual state of world cache before calling a plugin consumer's Unregister handle
        // so that if we receive a register event during Register Plugin, we can process it as a Register call.
        actualStateOfWorldUpdater.RemovePlugin(socketPath)

        handler.DeRegisterPlugin(infoResp.Name)
        return nil
    }
    return unregisterPluginFunc
}

handler.DeRegisterPlugin()

handler.DeRegisterPlugin()方法里逻辑比较简单，主要是调用了unregisterDriver()方法。

unregisterDriver()方法主要逻辑：

（1）从csiDrivers变量中删除该plugin信息（后续kubelet调用csi plugin进行存储的挂载/解除挂载操作，将通过plugin名称从csiDrivers变量中拿到socket地址并进行通信，所以取消注册plugin时，需要从csiDrivers变量中把该plugin信息去除）；

（2）调用nim.UninstallCSIDriver()做进一步处理。

// pkg/volume/csi/csi_plugin.go
// DeRegisterPlugin is called when a plugin removed its socket, signaling
// it is no longer available
func (h *RegistrationHandler) DeRegisterPlugin(pluginName string) {
    klog.Info(log("registrationHandler.DeRegisterPlugin request for plugin %s", pluginName))
    if err := unregisterDriver(pluginName); err != nil {
        klog.Error(log("registrationHandler.DeRegisterPlugin failed: %v", err))
    }
}

func unregisterDriver(driverName string) error {
    csiDrivers.Delete(driverName)

    if err := nim.UninstallCSIDriver(driverName); err != nil {
        return errors.New(log("Error uninstalling CSI driver: %v", err))
    }

    return nil
}

nim.UninstallCSIDriver()

接下来看到nim.UninstallCSIDriver()方法的分析。

nim.UninstallCSIDriver()中主要看到nim.uninstallDriverFromCSINode()、removeMaxAttachLimit()与removeNodeIDFromNode()3个方法，主要逻辑都在其中：

（1）nim.uninstallDriverFromCSINode()：更新CSINode对象，从中去除取消注册的plugin的相关信息。

（2）removeMaxAttachLimit()：更新node对象，从node.Status.Capacity及node.Status.Allocatable中去除取消注册的plugin的相关信息。

（3）removeNodeIDFromNode()：更新node对象，从node对象的annotation中key为csi.volume.kubernetes.io/nodeid的值中去除取消注册的plugin信息。

node对象的annotation示例：

csi.volume.kubernetes.io/nodeid: '{"cephfs.csi.ceph.com":"192.168.1.10","rbd.csi.ceph.com":"192.168.1.10"}'

CSIDriver对象示例：

apiVersion: storage.k8s.io/v1
kind: CSINode
metadata:
  name: 192.168.1.10
spec:
  drivers:
  - name: cephfs.csi.ceph.com
    nodeID: 192.168.1.10
    topologyKeys: null
  - name: rbd.csi.ceph.com
    nodeID: 192.168.1.10
    topologyKeys: null

nim.UninstallCSIDriver()源码：

// pkg/volume/csi/nodeinfomanager/nodeinfomanager.go
// UninstallCSIDriver removes the node ID annotation from the Node object and CSIDrivers field from the
// CSINode object. If the CSINOdeInfo object contains no CSIDrivers, it will be deleted.
// If multiple calls to UninstallCSIDriver() are made in parallel, some calls might receive Node or
// CSINode update conflicts, which causes the function to retry the corresponding update.
func (nim *nodeInfoManager) UninstallCSIDriver(driverName string) error {
    if utilfeature.DefaultFeatureGate.Enabled(features.CSINodeInfo) {
        err := nim.uninstallDriverFromCSINode(driverName)
        if err != nil {
            return fmt.Errorf("error uninstalling CSI driver from CSINode object %v", err)
        }
    }

    err := nim.updateNode(
        removeMaxAttachLimit(driverName),
        removeNodeIDFromNode(driverName),
    )
    if err != nil {
        return fmt.Errorf("error removing CSI driver node info from Node object %v", err)
    }
    return nil
}

func (nim *nodeInfoManager) updateNode(updateFuncs ...nodeUpdateFunc) error {
    var updateErrs []error
    err := wait.ExponentialBackoff(updateBackoff, func() (bool, error) {
        if err := nim.tryUpdateNode(updateFuncs...); err != nil {
            updateErrs = append(updateErrs, err)
            return false, nil
        }
        return true, nil
    })
    if err != nil {
        return fmt.Errorf("error updating node: %v; caused by: %v", err, utilerrors.NewAggregate(updateErrs))
    }
    return nil
}

（1）nim.uninstallDriverFromCSINode()：更新CSINode对象，从中去除取消注册的plugin的相关信息。

// pkg/volume/csi/nodeinfomanager/nodeinfomanager.go
func (nim *nodeInfoManager) uninstallDriverFromCSINode(
    csiDriverName string) error {

    csiKubeClient := nim.volumeHost.GetKubeClient()
    if csiKubeClient == nil {
        return fmt.Errorf("error getting CSI client")
    }

    var updateErrs []error
    err := wait.ExponentialBackoff(updateBackoff, func() (bool, error) {
        if err := nim.tryUninstallDriverFromCSINode(csiKubeClient, csiDriverName); err != nil {
            updateErrs = append(updateErrs, err)
            return false, nil
        }
        return true, nil
    })
    if err != nil {
        return fmt.Errorf("error updating CSINode: %v; caused by: %v", err, utilerrors.NewAggregate(updateErrs))
    }
    return nil
}

func (nim *nodeInfoManager) tryUninstallDriverFromCSINode(
    csiKubeClient clientset.Interface,
    csiDriverName string) error {

    nodeInfoClient := csiKubeClient.StorageV1().CSINodes()
    nodeInfo, err := nodeInfoClient.Get(string(nim.nodeName), metav1.GetOptions{})
    if err != nil && errors.IsNotFound(err) {
        return nil
    } else if err != nil {
        return err
    }

    hasModified := false
    // Uninstall CSINodeDriver with name csiDriverName
    drivers := nodeInfo.Spec.Drivers[:0]
    for _, driver := range nodeInfo.Spec.Drivers {
        if driver.Name != csiDriverName {
            drivers = append(drivers, driver)
        } else {
            // Found a driver with name csiDriverName
            // Set hasModified to true because it will be removed
            hasModified = true
        }
    }

    if !hasModified {
        // No changes, don't update
        return nil
    }
    nodeInfo.Spec.Drivers = drivers

    _, err = nodeInfoClient.Update(nodeInfo)

    return err // do not wrap error

}

（2）removeMaxAttachLimit()：更新node对象，从node.Status.Capacity及node.Status.Allocatable中去除取消注册的plugin的相关信息。

// pkg/volume/csi/nodeinfomanager/nodeinfomanager.go
func removeMaxAttachLimit(driverName string) nodeUpdateFunc {
    return func(node *v1.Node) (*v1.Node, bool, error) {
        limitKey := v1.ResourceName(util.GetCSIAttachLimitKey(driverName))

        capacityExists := false
        if node.Status.Capacity != nil {
            _, capacityExists = node.Status.Capacity[limitKey]
        }

        allocatableExists := false
        if node.Status.Allocatable != nil {
            _, allocatableExists = node.Status.Allocatable[limitKey]
        }

        if !capacityExists && !allocatableExists {
            return node, false, nil
        }

        delete(node.Status.Capacity, limitKey)
        if len(node.Status.Capacity) == 0 {
            node.Status.Capacity = nil
        }

        delete(node.Status.Allocatable, limitKey)
        if len(node.Status.Allocatable) == 0 {
            node.Status.Allocatable = nil
        }

        return node, true, nil
    }
}

（3）removeNodeIDFromNode()：更新node对象，从node对象的annotation中key为csi.volume.kubernetes.io/nodeid的值中去除取消注册的plugin信息。

// pkg/volume/csi/nodeinfomanager/nodeinfomanager.go
// removeNodeIDFromNode returns a function that removes node ID information matching the given
// driver name from a Node object.
func removeNodeIDFromNode(csiDriverName string) nodeUpdateFunc {
    return func(node *v1.Node) (*v1.Node, bool, error) {
        var previousAnnotationValue string
        if node.ObjectMeta.Annotations != nil {
            previousAnnotationValue =
                node.ObjectMeta.Annotations[annotationKeyNodeID]
        }

        if previousAnnotationValue == "" {
            return node, false, nil
        }

        // Parse previousAnnotationValue as JSON
        existingDriverMap := map[string]string{}
        if err := json.Unmarshal([]byte(previousAnnotationValue), &existingDriverMap); err != nil {
            return nil, false, fmt.Errorf(
                "failed to parse node's %q annotation value (%q) err=%v",
                annotationKeyNodeID,
                previousAnnotationValue,
                err)
        }

        if _, ok := existingDriverMap[csiDriverName]; !ok {
            // Value is already missing in node annotation, nothing more to do
            return node, false, nil
        }

        // Delete annotation value
        delete(existingDriverMap, csiDriverName)
        if len(existingDriverMap) == 0 {
            delete(node.ObjectMeta.Annotations, annotationKeyNodeID)
        } else {
            jsonObj, err := json.Marshal(existingDriverMap)
            if err != nil {
                return nil, false, fmt.Errorf(
                    "failed while trying to remove key %q from node %q annotation. Existing data: %v",
                    csiDriverName,
                    annotationKeyNodeID,
                    previousAnnotationValue)
            }

            node.ObjectMeta.Annotations[annotationKeyNodeID] = string(jsonObj)
        }

        return node, true, nil
    }
}

3.2.2 rc.operationExecutor.RegisterPlugin()

rc.operationExecutor.RegisterPlugin()主要逻辑：做plugin注册操作。

那plugin注册操作具体做了什么呢？继续往下分析。

plugin注册操作方法调用链

kl.pluginManager.Run --> pm.desiredStateOfWorldPopulator.Start() --> pm.reconciler.Run() --> rc.reconcile() --> rc.operationExecutor.RegisterPlugin() --> oe.operationGenerator.GenerateRegisterPluginFunc() --> handler.RegisterPlugin() --> nim.InstallCSIDriver() --> nim.updateNode()

下面来对plugin注册操作的部分关键方法进行分析。

GenerateRegisterPluginFunc

下面来分析下GenerateRegisterPluginFunc的逻辑，主要是定义并实现一个plugin注册的方法，然后返回。plugin注册方法主要逻辑如下：

（1）检测Node Driver Registrar组件socket的连通性；

（2）通过Node Driver Registrar组件socket获取plugin信息（包括plugin的socket地址、plugin名称等）；

（3）调用handler.ValidatePlugin()，检查已注册的plugin中是否有比该需要注册的plugin同名的的更高的版本，如有，则返回注册失败，并通知plugin注册失败；

（4）向actualStateOfWorld中增加该Node Driver Registrar组件的socket信息；

（5）调用handler.RegisterPlugin()做进一步的plugin注册操作;

（6）调用og.notifyPlugin，通知plugin，已经向kubelet注册成功/注册失败。

所以接下来会对handler.RegisterPlugin()方法进行分析。

// pkg/kubelet/pluginmanager/operationexecutor/operation_executor.go
func (oe *operationExecutor) RegisterPlugin(
    socketPath string,
    timestamp time.Time,
    pluginHandlers map[string]cache.PluginHandler,
    actualStateOfWorld ActualStateOfWorldUpdater) error {
    generatedOperation :=
        oe.operationGenerator.GenerateRegisterPluginFunc(socketPath, timestamp, pluginHandlers, actualStateOfWorld)

    return oe.pendingOperations.Run(
        socketPath, generatedOperation)
}


// pkg/kubelet/pluginmanager/operationexecutor/operation_generator.go
func (og *operationGenerator) GenerateRegisterPluginFunc(
    socketPath string,
    timestamp time.Time,
    pluginHandlers map[string]cache.PluginHandler,
    actualStateOfWorldUpdater ActualStateOfWorldUpdater) func() error {

    registerPluginFunc := func() error {
        client, conn, err := dial(socketPath, dialTimeoutDuration)
        if err != nil {
            return fmt.Errorf("RegisterPlugin error -- dial failed at socket %s, err: %v", socketPath, err)
        }
        defer conn.Close()

        ctx, cancel := context.WithTimeout(context.Background(), time.Second)
        defer cancel()

        infoResp, err := client.GetInfo(ctx, &registerapi.InfoRequest{})
        if err != nil {
            return fmt.Errorf("RegisterPlugin error -- failed to get plugin info using RPC GetInfo at socket %s, err: %v", socketPath, err)
        }

        handler, ok := pluginHandlers[infoResp.Type]
        if !ok {
            if err := og.notifyPlugin(client, false, fmt.Sprintf("RegisterPlugin error -- no handler registered for plugin type: %s at socket %s", infoResp.Type, socketPath)); err != nil {
                return fmt.Errorf("RegisterPlugin error -- failed to send error at socket %s, err: %v", socketPath, err)
            }
            return fmt.Errorf("RegisterPlugin error -- no handler registered for plugin type: %s at socket %s", infoResp.Type, socketPath)
        }

        if infoResp.Endpoint == "" {
            infoResp.Endpoint = socketPath
        }
        if err := handler.ValidatePlugin(infoResp.Name, infoResp.Endpoint, infoResp.SupportedVersions); err != nil {
            if err = og.notifyPlugin(client, false, fmt.Sprintf("RegisterPlugin error -- plugin validation failed with err: %v", err)); err != nil {
                return fmt.Errorf("RegisterPlugin error -- failed to send error at socket %s, err: %v", socketPath, err)
            }
            return fmt.Errorf("RegisterPlugin error -- pluginHandler.ValidatePluginFunc failed")
        }
        // We add the plugin to the actual state of world cache before calling a plugin consumer's Register handle
        // so that if we receive a delete event during Register Plugin, we can process it as a DeRegister call.
        err = actualStateOfWorldUpdater.AddPlugin(cache.PluginInfo{
            SocketPath: socketPath,
            Timestamp:  timestamp,
        })
        if err != nil {
            klog.Errorf("RegisterPlugin error -- failed to add plugin at socket %s, err: %v", socketPath, err)
        }
        if err := handler.RegisterPlugin(infoResp.Name, infoResp.Endpoint, infoResp.SupportedVersions); err != nil {
            return og.notifyPlugin(client, false, fmt.Sprintf("RegisterPlugin error -- plugin registration failed with err: %v", err))
        }

        // Notify is called after register to guarantee that even if notify throws an error Register will always be called after validate
        if err := og.notifyPlugin(client, true, ""); err != nil {
            return fmt.Errorf("RegisterPlugin error -- failed to send registration status at socket %s, err: %v", socketPath, err)
        }
        return nil
    }
    return registerPluginFunc
}

handler.RegisterPlugin()

handler.DeRegisterPlugin()方法主要逻辑：

（1）存储该plugin信息（主要是plugin名称与plugin的socket地址）到csiDrivers变量中（后续kubelet调用csi plugin进行存储的挂载/解除挂载操作，将通过plugin名称从此变量中拿到socket地址并进行通信）；

（2）检测Node Driver Registrar组件socket的连通性；

（3）通过plugin的socket获取plugin信息（包括plugin的NodeId、最大挂载数量限制、拓扑信息等）；

（4）调用nim.InstallCSIDriver，做进一步的plugin注册操作。

// pkg/volume/csi/csi_plugin.go

// TODO (verult) consider using a struct instead of global variables
// csiDrivers map keep track of all registered CSI drivers on the node and their
// corresponding sockets
var csiDrivers = &DriversStore{}

// RegisterPlugin is called when a plugin can be registered
func (h *RegistrationHandler) RegisterPlugin(pluginName string, endpoint string, versions []string) error {
    klog.Infof(log("Register new plugin with name: %s at endpoint: %s", pluginName, endpoint))

    highestSupportedVersion, err := h.validateVersions("RegisterPlugin", pluginName, endpoint, versions)
    if err != nil {
        return err
    }

    // Storing endpoint of newly registered CSI driver into the map, where CSI driver name will be the key
    // all other CSI components will be able to get the actual socket of CSI drivers by its name.
    csiDrivers.Set(pluginName, Driver{
        endpoint:                endpoint,
        highestSupportedVersion: highestSupportedVersion,
    })

    // Get node info from the driver.
    csi, err := newCsiDriverClient(csiDriverName(pluginName))
    if err != nil {
        return err
    }

    ctx, cancel := context.WithTimeout(context.Background(), csiTimeout)
    defer cancel()

    driverNodeID, maxVolumePerNode, accessibleTopology, err := csi.NodeGetInfo(ctx)
    if err != nil {
        if unregErr := unregisterDriver(pluginName); unregErr != nil {
            klog.Error(log("registrationHandler.RegisterPlugin failed to unregister plugin due to previous error: %v", unregErr))
        }
        return err
    }

    err = nim.InstallCSIDriver(pluginName, driverNodeID, maxVolumePerNode, accessibleTopology)
    if err != nil {
        if unregErr := unregisterDriver(pluginName); unregErr != nil {
            klog.Error(log("registrationHandler.RegisterPlugin failed to unregister plugin due to previous error: %v", unregErr))
        }
        return err
    }

    return nil
}

nim.InstallCSIDriver()

nim.InstallCSIDriver()中主要看到updateNodeIDInNode()与nim.updateCSINode()两个方法，主要逻辑都在其中：

（1）updateNodeIDInNode()：更新node对象，向node对象的annotation中key为csi.volume.kubernetes.io/nodeid的值中去增加注册的plugin信息。

（2）nim.updateCSINode()：创建或更新CSINode对象。

// pkg/volume/csi/nodeinfomanager/nodeinfomanager.go
// InstallCSIDriver updates the node ID annotation in the Node object and CSIDrivers field in the
// CSINode object. If the CSINode object doesn't yet exist, it will be created.
// If multiple calls to InstallCSIDriver() are made in parallel, some calls might receive Node or
// CSINode update conflicts, which causes the function to retry the corresponding update.
func (nim *nodeInfoManager) InstallCSIDriver(driverName string, driverNodeID string, maxAttachLimit int64, topology map[string]string) error {
    if driverNodeID == "" {
        return fmt.Errorf("error adding CSI driver node info: driverNodeID must not be empty")
    }

    nodeUpdateFuncs := []nodeUpdateFunc{
        updateNodeIDInNode(driverName, driverNodeID),
    }

    if utilfeature.DefaultFeatureGate.Enabled(features.CSINodeInfo) {
        nodeUpdateFuncs = append(nodeUpdateFuncs, updateTopologyLabels(topology))
    }

    err := nim.updateNode(nodeUpdateFuncs...)
    if err != nil {
        return fmt.Errorf("error updating Node object with CSI driver node info: %v", err)
    }

    if utilfeature.DefaultFeatureGate.Enabled(features.CSINodeInfo) {
        err = nim.updateCSINode(driverName, driverNodeID, maxAttachLimit, topology)
        if err != nil {
            return fmt.Errorf("error updating CSINode object with CSI driver node info: %v", err)
        }
    }
    return nil
}


// pkg/volume/csi/nodeinfomanager/nodeinfomanager.go
// updateNodeIDInNode returns a function that updates a Node object with the given
// Node ID information.
func updateNodeIDInNode(
    csiDriverName string,
    csiDriverNodeID string) nodeUpdateFunc {
    return func(node *v1.Node) (*v1.Node, bool, error) {
        existingDriverMap, err := buildNodeIDMapFromAnnotation(node)
        if err != nil {
            return nil, false, err
        }

        if val, ok := existingDriverMap[csiDriverName]; ok {
            if val == csiDriverNodeID {
                // Value already exists in node annotation, nothing more to do
                return node, false, nil
            }
        }

        // Add/update annotation value
        existingDriverMap[csiDriverName] = csiDriverNodeID
        jsonObj, err := json.Marshal(existingDriverMap)
        if err != nil {
            return nil, false, fmt.Errorf(
                "error while marshalling node ID map updated with driverName=%q, nodeID=%q: %v",
                csiDriverName,
                csiDriverNodeID,
                err)
        }

        if node.ObjectMeta.Annotations == nil {
            node.ObjectMeta.Annotations = make(map[string]string)
        }
        node.ObjectMeta.Annotations[annotationKeyNodeID] = string(jsonObj)

        return node, true, nil
    }
}


// pkg/volume/csi/nodeinfomanager/nodeinfomanager.go
func (nim *nodeInfoManager) updateCSINode(
    driverName string,
    driverNodeID string,
    maxAttachLimit int64,
    topology map[string]string) error {

    csiKubeClient := nim.volumeHost.GetKubeClient()
    if csiKubeClient == nil {
        return fmt.Errorf("error getting CSI client")
    }

    var updateErrs []error
    err := wait.ExponentialBackoff(updateBackoff, func() (bool, error) {
        if err := nim.tryUpdateCSINode(csiKubeClient, driverName, driverNodeID, maxAttachLimit, topology); err != nil {
            updateErrs = append(updateErrs, err)
            return false, nil
        }
        return true, nil
    })
    if err != nil {
        return fmt.Errorf("error updating CSINode: %v; caused by: %v", err, utilerrors.NewAggregate(updateErrs))
    }
    return nil
}

本节主要讲解了kubelet注册csi driver的原理，以及其代码的分析，也顺带提了一下Node Driver Registrar组件，下面来做个总结。

Node Driver Registrar

Node Driver Registrar在后面的文章中会介绍，Node Driver Registrar源码分析，它是一个sidecar容器，通过Kubelet的插件注册机制将CSI plugin（csi driver，两个名词意义一样）注册到Kubelet，让kubelet做volume的mount/umount操作时知道怎么调用相应的csi plugin。

kubelet注册csi driver的原理

kubelet的pluginManager会监听某个特定目录，而负责向kubelet注册csi driver的组件Node Driver Registrar会创建暴露服务的socket在该目录下（每个plugin会对应一个Node Driver Registrar组件，也就是说，一个Node Driver Registrar只负责一个plugin的注册工作），pluginManager通过Node Driver Registrar组件暴露的socket获取plugin信息（包括plugin的socket地址、plugin名称等），从而最终做到根据该目录下socket文件的新增/删除来做相应的plugin注册/取消注册操作。

plugin注册完成后，后续kubelet将通过plugin暴露的socket与plugin进行通信，做存储挂载/解除挂载等操作。

下面再来总结一下在kubelet的pluginManager中，plugin的注册/取消注册操作分别做了什么动作。

plugin注册操作

（1）存储该plugin信息（主要是plugin名称与plugin的socket地址）到csiDrivers变量中（后续kubelet调用csi plugin进行存储的挂载/解除挂载操作，将通过plugin名称从此变量中拿到socket地址并进行通信；csiDriver变量代码位置-pkg/volume/csi/csi_plugin.go）；

（2）更新node对象，向node对象的annotation中key为csi.volume.kubernetes.io/nodeid的值中去增加注册的plugin信息。

（3）创建或更新CSINode对象。

plugin取消注册操作

（1）从csiDrivers变量中删除该plugin信息（后续kubelet调用csi plugin进行存储的挂载/解除挂载操作，将通过plugin名称从csiDrivers变量（csiDriver变量代码位置-pkg/volume/csi/csi_plugin.go）中拿到socket地址并进行通信，所以取消注册plugin时，需要从csiDrivers变量中把该plugin信息去除）；

（2）更新CSINode对象，从中去除取消注册的plugin的相关信息。

（3）更新node对象，从node.Status.Capacity及node.Status.Allocatable中去除取消注册的plugin的相关信息。

（4）更新node对象，从node对象的annotation中key为csi.volume.kubernetes.io/nodeid的值中去除取消注册的plugin信息。