前言

上几篇文章介绍了 kubernetes 的核心数据结构 schemeKubernetes:kube-apiserver 的启动流程。在启动流程篇中重点关注的是启动的核心逻辑,并没有关注 kube-apiserver 和外部组件的交互。

而交互是非常必要的,其定义了边界和依赖。

kubernetes 架构图

Kubernetes 架构图可以看出,kube-apiserver 是唯一和 etcd 交互的组件。因此,这里将 kube-apiserveretcd 交互的部分单独拿出来加以介绍,做到知其然,知其所以然。

使用 etcd

既然是交互,首先需要了解的是怎么用交互的组件。这里同 kube-apiserver 交互的是大名鼎鼎的 etcd,不需要多介绍它。

仅给出示例:

 1package main
 2
 3import (
 4	"context"
 5	"fmt"
 6	"log"
 7	"time"
 8
 9	"go.etcd.io/etcd/clientv3"
10)
11
12func main() {
13	cli, err := clientv3.New(clientv3.Config{
14		Endpoints:   []string{"127.0.0.1:2379"},
15		DialTimeout: time.Second * 5,
16	})
17	if err != nil {
18		log.Fatal(err)
19	}
20	fmt.Println("connect to etcd success.")
21	defer cli.Close()
22
23	// lease with 5 second
24	resp, err := cli.Grant(context.TODO(), 5)
25	if err != nil {
26		log.Fatal(err)
27	}
28
29	// delete key:name after expire of lease
30	_, err = cli.Put(context.TODO(), "name", "hxia", clientv3.WithLease(resp.ID))
31	if err != nil {
32		log.Fatal(err)
33	}
34}

详细内容可参考 go-by-example: etcdQuickstart

kube-apiserver 和 etcd

顺序看 kube-apiserveretcd 的交互是非常复杂的,容易头晕。这里,逆序的看 kube-apiserveretcd 的交互。首先,找到它们在哪里交互的,接着从这一点开始发散,摸清整体脉络。

那么,它们在哪里交互的呢?这个问题不难回答,在 handler。作为 RESTful API 的处理单元,handler 内定义了 kube-apiserveretcd 的交互。

以处理 GEThandler 为例:

 1# kubernetes/vendor/k8s.io/apiserver/pkg/endpoints/installer.go
 2func (a *APIInstaller) registerResourceHandlers(path string, storage rest.Storage, ws *restful.WebService) (*metav1.APIResource, *storageversion.ResourceInfo, error) {
 3    switch action.Verb {
 4    case "GET": // Get a resource.
 5        var handler restful.RouteFunction
 6        if isGetterWithOptions {
 7            handler = restfulGetResourceWithOptions(getterWithOptions, reqScope, isSubresource)
 8        } else {
 9            handler = restfulGetResource(getter, reqScope)
10        }
11
12        route := ws.GET(action.Path).To(handler).
13            Doc(doc).
14            Param(ws.QueryParameter("pretty", "If 'true', then the output is pretty printed.")).
15            Operation("read"+namespaced+kind+strings.Title(subresource)+operationSuffix).
16            Produces(append(storageMeta.ProducesMIMETypes(action.Verb), mediaTypes...)...).
17            Returns(http.StatusOK, "OK", producedObject).
18            Writes(producedObject)
19
20        addParams(route, action.Params)
21        routes = append(routes, route)
22    }
23}

进入 restfulGetResourcehandler 是怎么创建的。

 1func restfulGetResource(r rest.Getter, scope handlers.RequestScope) restful.RouteFunction {
 2	return func(req *restful.Request, res *restful.Response) {
 3		handlers.GetResource(r, &scope)(res.ResponseWriter, req.Request)
 4	}
 5}
 6
 7// GetResource returns a function that handles retrieving a single resource from a rest.Storage object.
 8func GetResource(r rest.Getter, scope *RequestScope) http.HandlerFunc {
 9	return getResourceHandler(scope,
10		func(ctx context.Context, name string, req *http.Request) (runtime.Object, error) {
11			...
12			return r.Get(ctx, name, &options)
13		})
14}
15
16// Getter is an object that can retrieve a named RESTful resource.
17type Getter interface {
18	// Get finds a resource in the storage by name and returns it.
19	// Although it can return an arbitrary error value, IsNotFound(err) is true for the
20	// returned error value err when the specified resource is not found.
21	Get(ctx context.Context, name string, options *metav1.GetOptions) (runtime.Object, error)
22}

可以看到:

  1. restfulGetResource 返回一个路由函数,路由函数内包含传递给 restfulGetResourcegetter 对象。
  2. 返回的路由函数内,调用的是 getterGet 方法获取资源对象 runtime.Object。这里的 getter 是实现 Getter 接口的对象。

基于上述分析,现在重点就变成 getter 调用的 Get 具体做了什么。通过逐级向上追溯,找到了 Getter 接口的实例对象 customResourceDefinitionStorage

 1# kubernetes/vendor/k8s.io/apiextensions-apiserver/pkg/apiserver/apiserver.go
 2func (c completedConfig) New(delegationTarget genericapiserver.DelegationTarget) (*CustomResourceDefinitions, error) {
 3    if resource := "customresourcedefinitions"; apiResourceConfig.ResourceEnabled(v1.SchemeGroupVersion.WithResource(resource)) {
 4        // 调用 NetREST 创建资源实体 customResourceDefinitionStorage
 5        customResourceDefinitionStorage, err := customresourcedefinition.NewREST(Scheme, c.GenericConfig.RESTOptionsGetter)
 6        if err != nil {
 7            return nil, err
 8        }
 9        storage[resource] = customResourceDefinitionStorage
10        storage[resource+"/status"] = customresourcedefinition.NewStatusREST(Scheme, customResourceDefinitionStorage)
11    }
12}
13
14func NewREST(scheme *runtime.Scheme, optsGetter generic.RESTOptionsGetter) (*REST, error) {
15	strategy := NewStrategy(scheme)
16
17	store := &genericregistry.Store{
18		...
19	}
20	options := &generic.StoreOptions{RESTOptions: optsGetter, AttrFunc: GetAttrs}
21	if err := store.CompleteWithOptions(options); err != nil {
22		return nil, err
23	}
24	return &REST{store}, nil
25}

接口的实例对象找到了,继续看实例对象的 Get 做了什么。

 1# kubernetes/vendor/k8s.io/apiextensions-apiserver/pkg/registry/customresourcedefinition/etcd.go
 2// rest implements a RESTStorage for API services against etcd
 3type REST struct {
 4	*genericregistry.Store
 5}
 6
 7# kubernetes/vendor/k8s.io/apiserver/pkg/registry/generic/registry/store.go
 8type Store struct {
 9    Storage DryRunnableStorage
10}
11
12# kubernetes/vendor/k8s.io/apiserver/pkg/registry/generic/registry/dryrun.go
13type DryRunnableStorage struct {
14	Storage storage.Interface
15	Codec   runtime.Codec
16}
17
18# kubernetes/vendor/k8s.io/apiserver/pkg/registry/generic/registry/store.go
19// Get retrieves the item from storage.
20func (e *Store) Get(ctx context.Context, name string, options *metav1.GetOptions) (runtime.Object, error) {
21	obj := e.NewFunc()
22	key, err := e.KeyFunc(ctx, name)
23	if err != nil {
24		return nil, err
25	}
26	if err := e.Storage.Get(ctx, key, storage.GetOptions{ResourceVersion: options.ResourceVersion}, obj); err != nil {
27		return nil, storeerr.InterpretGetError(err, e.qualifiedResourceFromContext(ctx), name)
28	}
29	if e.Decorator != nil {
30		e.Decorator(obj)
31	}
32	return obj, nil
33}
34
35# kubernetes/vendor/k8s.io/apiserver/pkg/registry/generic/registry/dryrun.go
36func (s *DryRunnableStorage) Get(ctx context.Context, key string, opts storage.GetOptions, objPtr runtime.Object) error {
37	return s.Storage.Get(ctx, key, opts, objPtr)
38}

REST 对象包含 *genericregistry.Store,其继承了 StoreGet 方法。在 Store.Get 方法内,通过 e.Storage.Get 调用 DryRunnableStorageGet 方法。实际是通过 DryRunnableStorage 内的 Storage 存储接口调用 Get 方法,从而访问 etcd

DryRunnableStorage.Storage 是一个接口,它的实体对象是什么呢?

还是从资源实体入手,看 REST{store} 是如何实例化的。

 1// NewREST returns a RESTStorage object that will work against API services.
 2func NewREST(scheme *runtime.Scheme, optsGetter generic.RESTOptionsGetter) (*REST, error) {
 3	strategy := NewStrategy(scheme)
 4
 5	store := &genericregistry.Store{
 6		...
 7	}
 8	options := &generic.StoreOptions{RESTOptions: optsGetter, AttrFunc: GetAttrs}
 9    // 进入 CompleteWithOptions
10	if err := store.CompleteWithOptions(options); err != nil {
11		return nil, err
12	}
13	return &REST{store}, nil
14}
15
16func (e *Store) CompleteWithOptions(options *generic.StoreOptions) error {
17    if e.Storage.Storage == nil {
18        e.Storage.Storage, e.DestroyFunc, err = opts.Decorator(
19            opts.StorageConfig,
20            prefix,
21            keyFunc,
22            e.NewFunc,
23            e.NewListFunc,
24            attrFunc,
25            options.TriggerFunc,
26            options.Indexers,
27        )
28    }
29}

看到这里,已经知道哪里实例化的 storage.Interface 对象了。这里的 opts.Decorator 是一个装饰函数。接着,继续探案,看这个装饰函数干了什么,知道它干了什么就能挖出来最关键的一环,存储接口是怎么访问 etcd 的。

 1# kubernetes/vendor/k8s.io/apiserver/pkg/registry/generic/registry/store.go
 2func (e *Store) CompleteWithOptions(options *generic.StoreOptions) error {
 3	// 通过 options.RESTOptions.GetRESTOptions 实例化 opts
 4	// options.RESTOptions 是满足 RESTOptionsGetter 接口的实例
 5	opts, err := options.RESTOptions.GetRESTOptions(e.DefaultQualifiedResource)
 6	if err != nil {
 7		return err
 8	}
 9}
10
11# kubernetes/vendor/k8s.io/apiserver/pkg/registry/generic/options.go
12type RESTOptionsGetter interface {
13	GetRESTOptions(resource schema.GroupResource) (RESTOptions, error)
14}
15
16func NewREST(scheme *runtime.Scheme, optsGetter generic.RESTOptionsGetter) (*REST, error) {
17	strategy := NewStrategy(scheme)
18
19	store := &genericregistry.Store{
20		...
21	}
22	// 创建 options
23	options := &generic.StoreOptions{RESTOptions: optsGetter, AttrFunc: GetAttrs}
24	// 将 options 作为参数传递给 Store.CompleteWithOptions
25	if err := store.CompleteWithOptions(options); err != nil {
26		return nil, err
27	}
28	return &REST{store}, nil
29}
30
31func (c completedConfig) New(delegationTarget genericapiserver.DelegationTarget) (*CustomResourceDefinitions, error) {
32	if resource := "customresourcedefinitions"; apiResourceConfig.ResourceEnabled(v1.SchemeGroupVersion.WithResource(resource)) {
33		customResourceDefinitionStorage, err := customresourcedefinition.NewREST(Scheme, c.GenericConfig.RESTOptionsGetter)
34		if err != nil {
35			return nil, err
36		}
37		storage[resource] = customResourceDefinitionStorage
38		storage[resource+"/status"] = customresourcedefinition.NewStatusREST(Scheme, customResourceDefinitionStorage)
39	}
40}

可以看到,c.GenericConfig.RESTOptionsGetter 即为 optsGetter,调用 c.GenericConfig.RESTOptionsGetterGetRESTOptions 得到 RESTOptions

c.GenericConfig.RESTOptionsGetter 在哪里实例化的呢?

还记得前面创建通用配置的 BuildGenericConfig 吗?在该函数内,实例化了 c.GenericConfig.RESTOptionsGetter

 1# kubernetes/pkg/controlplane/apiserver/config.go
 2func BuildGenericConfig(
 3	s controlplaneapiserver.CompletedOptions,
 4	schemes []*runtime.Scheme,
 5	getOpenAPIDefinitions func(ref openapicommon.ReferenceCallback) map[string]openapicommon.OpenAPIDefinition,
 6){
 7	storageFactoryConfig := kubeapiserver.NewStorageFactoryConfig()
 8	storageFactoryConfig.APIResourceConfig = genericConfig.MergedResourceConfig
 9	storageFactory, lastErr = storageFactoryConfig.Complete(s.Etcd).New()
10	if lastErr != nil {
11		return
12	}
13	if lastErr = s.Etcd.ApplyWithStorageFactoryTo(storageFactory, genericConfig); lastErr != nil {
14		return
15	}
16}
17
18func (s *EtcdOptions) ApplyWithStorageFactoryTo(factory serverstorage.StorageFactory, c *server.Config) error {
19	c.RESTOptionsGetter = s.CreateRESTOptionsGetter(factory, c.ResourceTransformers)
20	return nil
21}
22
23func (s *EtcdOptions) CreateRESTOptionsGetter(factory serverstorage.StorageFactory, resourceTransformers storagevalue.ResourceTransformers) generic.RESTOptionsGetter {
24	if resourceTransformers != nil {
25		factory = &transformerStorageFactory{
26			delegate:             factory,
27			resourceTransformers: resourceTransformers,
28		}
29	}
30	return &StorageFactoryRestOptionsFactory{Options: *s, StorageFactory: factory}
31}

过程也不复杂,可以看到,RESTOptionsGetter 接口的实例化对象是 &StorageFactoryRestOptionsFactory

调用 c.GenericConfig.RESTOptionsGetterGetRESTOptions 实际调用的是 StorageFactoryRestOptionsFactory.GetRESTOptions

 1func (f *StorageFactoryRestOptionsFactory) GetRESTOptions(resource schema.GroupResource) (generic.RESTOptions, error) {
 2	ret := generic.RESTOptions{
 3		StorageConfig:             storageConfig,
 4		Decorator:                 generic.UndecoratedStorage,
 5		DeleteCollectionWorkers:   f.Options.DeleteCollectionWorkers,
 6		EnableGarbageCollection:   f.Options.EnableGarbageCollection,
 7		ResourcePrefix:            f.StorageFactory.ResourcePrefix(resource),
 8		CountMetricPollPeriod:     f.Options.StorageConfig.CountMetricPollPeriod,
 9		StorageObjectCountTracker: f.Options.StorageConfig.StorageObjectCountTracker,
10	}
11
12	return ret, nil
13}

RESTOptions 中包含了 Decorator 的创建,这里我们的重点是 Decorator,进入 generic.UndecoratedStorage 看它是怎么一个函数。

 1# kubernetes/vendor/k8s.io/apiserver/pkg/registry/generic/storage_decorator.go
 2func UndecoratedStorage(
 3	config *storagebackend.ConfigForResource,
 4	resourcePrefix string,
 5	keyFunc func(obj runtime.Object) (string, error),
 6	newFunc func() runtime.Object,
 7	newListFunc func() runtime.Object,
 8	getAttrsFunc storage.AttrFunc,
 9	trigger storage.IndexerFuncs,
10	indexers *cache.Indexers) (storage.Interface, factory.DestroyFunc, error) {
11	return NewRawStorage(config, newFunc, newListFunc, resourcePrefix)
12}
13
14func NewRawStorage(config *storagebackend.ConfigForResource, newFunc, newListFunc func() runtime.Object, resourcePrefix string) (storage.Interface, factory.DestroyFunc, error) {
15	return factory.Create(*config, newFunc, newListFunc, resourcePrefix)
16}
17
18# kubernetes/vendor/k8s.io/apiserver/pkg/storage/storagebackend/factory/factory.go
19func Create(c storagebackend.ConfigForResource, newFunc, newListFunc func() runtime.Object, resourcePrefix string) (storage.Interface, DestroyFunc, error) {
20	switch c.Type {
21	case storagebackend.StorageTypeETCD2:
22		return nil, nil, fmt.Errorf("%s is no longer a supported storage backend", c.Type)
23	case storagebackend.StorageTypeUnset, storagebackend.StorageTypeETCD3:
24		return newETCD3Storage(c, newFunc, newListFunc, resourcePrefix)
25	default:
26		return nil, nil, fmt.Errorf("unknown storage type: %s", c.Type)
27	}
28}
29
30# kubernetes/vendor/k8s.io/apiserver/pkg/storage/storagebackend/factory/etcd3.go
31func newETCD3Storage(c storagebackend.ConfigForResource, newFunc, newListFunc func() runtime.Object, resourcePrefix string) (storage.Interface, DestroyFunc, error) {
32	client, err := newETCD3Client(c.Transport)
33	if err != nil {
34		stopCompactor()
35		return nil, nil, err
36	}
37
38	client.KV = etcd3.NewETCDLatencyTracker(client.KV)
39
40	return etcd3.New(client, c.Codec, newFunc, newListFunc, c.Prefix, resourcePrefix, c.GroupResource, transformer, c.Paging, c.LeaseManagerConfig), destroyFunc, nil
41}
42
43// New returns an etcd3 implementation of storage.Interface.
44func New(c *clientv3.Client, codec runtime.Codec, newFunc, newListFunc func() runtime.Object, prefix, resourcePrefix string, groupResource schema.GroupResource, transformer value.Transformer, pagingEnabled bool, leaseManagerConfig LeaseManagerConfig) storage.Interface {
45	return newStore(c, codec, newFunc, newListFunc, prefix, resourcePrefix, groupResource, transformer, pagingEnabled, leaseManagerConfig)
46}
47
48func newStore(c *clientv3.Client, codec runtime.Codec, newFunc, newListFunc func() runtime.Object, prefix, resourcePrefix string, groupResource schema.GroupResource, transformer value.Transformer, pagingEnabled bool, leaseManagerConfig LeaseManagerConfig) *store {
49	s := &store{
50		client:              c,
51		codec:               codec,
52		versioner:           versioner,
53		transformer:         transformer,
54		pagingEnabled:       pagingEnabled,
55		pathPrefix:          pathPrefix,
56		groupResource:       groupResource,
57		groupResourceString: groupResource.String(),
58		watcher:             w,
59		leaseManager:        newDefaultLeaseManager(c, leaseManagerConfig),
60	}
61	return s
62}

上述代码基本都是函数的顺序调用,不用介绍太多。
可以看到 opts.Decorator 做的事情是实例化了一个访问 etcd 的接口实例 storestore 中存储了访问 etcdclientclient 是通过 newETCD3Client(c.Transport) 创建的。

到这里,基本破案了。访问 etcd 实际是通过 storeetcd 进行交互。这里的 store 只是存储了 client 并没有实际访问,实际访问在 handler

再回头看 DryRunnableStorage.Get 方法内的 s.Storage.Get 即可知道其调用的是 storeGet 方法。

 1# kubernetes/vendor/k8s.io/apiserver/pkg/registry/generic/registry/dryrun.go
 2func (s *DryRunnableStorage) Get(ctx context.Context, key string, opts storage.GetOptions, objPtr runtime.Object) error {
 3	return s.Storage.Get(ctx, key, opts, objPtr)
 4}
 5
 6# kubernetes/vendor/k8s.io/apiserver/pkg/storage/etcd3/store.go
 7// Get implements storage.Interface.Get.
 8func (s *store) Get(ctx context.Context, key string, opts storage.GetOptions, out runtime.Object) error {
 9	preparedKey, err := s.prepareKey(key)
10	if err != nil {
11		return err
12	}
13	startTime := time.Now()
14	// 通过 client 访问 key 对应的 value
15	getResp, err := s.client.KV.Get(ctx, preparedKey)
16
17	kv := getResp.Kvs[0]
18
19	data, _, err := s.transformer.TransformFromStorage(ctx, kv.Value, authenticatedDataString(preparedKey))
20	if err != nil {
21		return storage.NewInternalError(err.Error())
22	}
23
24	err = decode(s.codec, s.versioner, data, out, kv.ModRevision)
25	if err != nil {
26		recordDecodeError(s.groupResourceString, preparedKey)
27		return err
28	}
29	return nil
30}

最后,通过本文介绍了 kube-apiserveretcd 的交互。下一步将重点介绍 kube-apiserver 是怎么做鉴权,认证和准入机制的。