TAPP
1 TAPP是什么
TAPP是一种新的k8s应用负载，基于CRD实现，它同时包含kubernetes
deployment和statefulset的大多数特性， 并且用户可以轻松在Kubernetes上运行遗留应用程序。当前许多用户希望采用Kubernetes，并将其旧版应用程序迁移到Kubernetes。 但是他们不能直接使用Kubernetes的工作负载（例如，deployment，statefulset），因此将这些应用程序转换为微服务将需要大量的努力。 Tapp可以解决这些问题。
2 TAPP特性
-
支持每个pod的唯一索引，(与
statefulset相同) -
支持在特定实例(pod)上执行启动/停止/升级操作
它更适合传统的操作和维护，例如当管理员想要停止一台计算机时，他可以停止该计算机上的实例，而不会影响其他计算机上的实例。
-
支持实例的就地更新
尽管许多无状态工作负载可以容忍非就地更新，但不排除某些应用很敏感，毕竟Pod重新启动是一种比较极端的方式，导致较低的可用性或较高的运行成本。
-
支持多种版本的实例
实例使用不同的镜像或不同的配置。
-
支持HPA，根据多种指标（例如CPU，内存，自定义指标）
-
支持滚动更新，回滚
3 TAPP使用
3.1 编译tapp-controller
~/go_workspace/src/tkestack.io/tapp# make build
hack/build.sh
Build success!
命令行运行tapp-controller
root@xiabingyao-LC0:~/go_workspace/src/tkestack.io/tapp# bin/tapp-controller --kubeconfig=/root/.kube/config
I0121 15:07:41.615902 3726361 controller.go:155] Setting up event handlers
I0121 15:07:41.644437 3726361 controller.go:196] Starting tapp controller
I0121 15:07:41.644466 3726361 controller.go:199] Waiting for informer caches to sync
E0121 15:07:41.646686 3726361 reflector.go:126] pkg/mod/k8s.io/client-go@v11.0.1-0.20191029005444-8e4128053008+incompatible/tools/cache/reflector.go:94: Failed to list *v1.TApp: the server could not find the requested resource (get tapps.apps.tkestack.io)
I0121 15:07:42.744715 3726361 controller.go:204] Starting workers
I0121 15:07:42.744775 3726361 controller.go:210] Started workers
创建tapp crd的过程也定义在tapp-controller代码中，一启动自动创建好了
3.2 创建tapp对象
3.2.1 使用同样的template
创建example-tapp tapp cr
kubectl apply -f - << EOF
apiVersion: apps.tkestack.io/v1
kind: TApp
metadata:
name: example-tapp
namespace: test
spec:
replicas: 3
template:
metadata:
labels:
app: example-tapp
spec:
containers:
- name: nginx
image: nginx:1.7.9
EOF
3.2.2 使用不一样的template
kubectl apply -f - << EOF
apiVersion: apps.tkestack.io/v1
kind: TApp
metadata:
name: example-tapp
namespace: test
spec:
replicas: 3
template:
metadata:
labels:
app: example-tapp
spec:
containers:
- name: nginx
image: nginx:1.8.1
templatePool:
"test":
metadata:
labels:
app: example-tapp
spec:
containers:
- name: nginx
image: nginx:1.7.9
templates:
"1": "test"
# defaultTemplateName: "test"
EOF
tapp新增了多个字段以支持更多的特性，spec.templatePools、spec.templates和defaultTemplateName
spec.templatePools来声明templatespec.templates指定pod使用哪个template，如果没指定，默认使用default template- 在
spec.templatePools中使用spec.DefaultTemplateName来设置default template， 如果未设置的话，spec.template作为default template.
3.3 查询tapp对象
查看tapp资源列表
~/go_workspace/src/tkestack.io/tapp# kubectl -n test get tapps.apps.tkestack.io
NAME AGE
example-tapp 72s
查看tapp资源描述
~/go_workspace/src/tkestack.io/tapp# kubectl descirbe tapp tapps.apps.tkestack.io example-tapp
查看pod
~/go_workspace/src/tkestack.io/tapp# kubectl -n test get pod
NAME READY STATUS RESTARTS AGE
example-tapp-0 1/1 Running 0 73s
example-tapp-1 1/1 Running 0 73s
example-tapp-2 1/1 Running 0 73s
3.4 更新tapp对象
如果仅更新容器的映像，则Tapp控制器将对Pod进行就地更新，否则它将删除Pod并重新创建它们.
3.4.1 更新特定的pod
kubectl apply -f - << EOF
apiVersion: apps.tkestack.io/v1
kind: TApp
metadata:
name: example-tapp
namespace: test
spec:
replicas: 3
template:
metadata:
labels:
app: example-tapp
spec:
containers:
- name: nginx
image: nginx:latest
templatePool:
"test2":
metadata:
labels:
app: example-tapp
spec:
containers:
- name: nginx
image: nginx:1.8.1
templates:
"1": "test2"
EOF
指定pod-xxx-1使用test2的spec.templatePool, pod是并发更新的
3.4.2 滚动更新
kubectl apply -f - << EOF
apiVersion: apps.tkestack.io/v1
kind: TApp
metadata:
name: example-tapp
namespace: test
spec:
replicas: 3
template:
metadata:
labels:
app: example-tapp
spec:
containers:
- name: nginx
image: nginx:latest
templatePool:
"test2":
metadata:
labels:
app: example-tapp
spec:
containers:
- name: nginx
image: nginx:1.7.9
templates:
"1": "test2"
"2": "test2"
updateStrategy:
template: test2
maxUnavailable: 1
EOF
- 使用
spec.updateStrategy指定滚动更新template - 使用
spec.updateStrategy.maxUnavailable指定更新过程中最大可容忍不可用的pod数量，可以是整数或百分比; 默认值为1
3.5 杀死特定pod
kubectl apply -f - << EOF
apiVersion: apps.tkestack.io/v1
kind: TApp
metadata:
name: example-tapp
namespace: test
spec:
replicas: 3
template:
metadata:
labels:
app: example-tapp
spec:
containers:
- name: nginx
image: nginx:latest
templatePool:
"test2":
metadata:
labels:
app: example-tapp
spec:
containers:
- name: nginx
image: nginx:1.7.9
templates:
"1": "test2"
statuses:
"1": "Killed"
EOF
使用spec.statuses可以指定pod的状态，上面例子指定pod-xxx-1被tapp controller kill掉；如果想再次运行pod-xxx-1，移除spec.statuses即可
3.6 扩缩容tapp
使用什么样的template，跟上面【创建tapp对象】【使用不一样的template】章节方式一致，kubectl scale命令同样适用于
tapp
~/go_workspace/src/tkestack.io/tapp# kubectl -n test scale --replicas=4 tapps.apps.tkestack.io example-tapp
3.7 Headless service
Tapp支持statefulset headless service方式，每个pod的subdomain: $(podname).$(namespace).svc.cluster.local, 如cluster domain是"cluster.local"
kubectl apply -f - << EOF
apiVersion: v1
kind: Service
metadata:
name: nginx
namespace: test
labels:
app: example-tapp
spec:
ports:
- port: 80
name: web
clusterIP: None
selector:
app: example-tapp
---
apiVersion: apps.tkestack.io/v1
kind: TApp
metadata:
name: example-tapp
namespace: test
spec:
replicas: 3
serviceName: "nginx"
template:
metadata:
labels:
app: example-tapp
spec:
containers:
- name: nginx
image: nginx:1.7.9
EOF
3.8 删除unused template
通过设置spec.autoDeleteUnusedTemplate为true，可以删除未被使用的template，来使tapp对象内容更简洁
kubectl apply -f - << EOF
apiVersion: apps.tkestack.io/v1
kind: TApp
metadata:
name: example-tapp
namespace: test
spec:
replicas: 3
autoDeleteUnusedTemplate: true
template:
metadata:
labels:
app: example-tapp
spec:
containers:
- name: nginx
image: nginx:1.7.9
templatePool:
"test1":
metadata:
labels:
app: example-tapp
spec:
containers:
- name: nginx
image: nginx:1.7.9
"test2":
metadata:
labels:
app: example-tapp
spec:
containers:
- name: nginx
image: nginx:1.7.8
"test3":
metadata:
labels:
app: example-tapp
spec:
containers:
- name: nginx
image: nginx:1.7.7
"test4":
metadata:
labels:
app: example-tapp
spec:
containers:
- name: nginx
image: nginx:1.7.7
templates:
"1": "test1"
updateStrategy:
template: "test3"
maxUnavailable: 1
DefaultTemplateName: "test4"
EOF
3.9 tapp with volume
kubectl apply -f - << EOF
apiVersion: apps.tkestack.io/v1
kind: TApp
metadata:
name: example-tapp
namespace: test
spec:
replicas: 2
template:
metadata:
labels:
app: example-tapp
spec:
containers:
- name: nginx
image: nginx:1.7.9
volumeMounts:
- name: www
mountPath: /usr/share/nginx/html
volumeClaimTemplates:
- metadata:
name: www
labels:
app: example-tapp
spec:
accessModes: [ "ReadWriteOnce" ]
storageClassName: rbd
resources:
requests:
storage: 10Gi
EOF
3.10 删除tapp
# kubectl -n test delete tapps.apps.tkestack.io example-tapp
3.11 清理tapp
# kubectl delete crd tapps.apps.tkestack.io
4 代码分析
4.1 结构体
4.1.1 TApp
// TApp represents a set of pods with consistent identities.
type TApp struct {
metav1.TypeMeta `json:",inline"`
metav1.ObjectMeta `json:"metadata,omitempty"`
// Spec defines the desired identities of pods in this tapp.
Spec TAppSpec `json:"spec,omitempty"`
// Status is the current status of pods in this TApp. This data
// may be out of date by some window of time.
Status TAppStatus `json:"status,omitempty"`
}
4.1.2 TAppSpec
// A TAppSpec is the specification of a TApp.
type TAppSpec struct {
// Replicas is the desired number of replicas of the given Template.
// These are replicas in the sense that they are instantiations of the
// same Template, but individual replicas also have a consistent identity.
Replicas int32 `json:"replicas"`
// Selector is a label query over pods that should match the replica count.
// If empty, defaulted to labels on the pod template.
// More info: http://releases.k8s.io/release-1.4/docs/user-guide/labels.md#label-selectors
Selector *metav1.LabelSelector `json:"selector,omitempty"`
// Template is the object that describes the pod that will be initial created/default scaled
// it should be added to TemplatePool
Template corev1.PodTemplateSpec `json:"template"`
// TemplatePool stores a map whose key is template name and value is podTemplate
TemplatePool map[string]corev1.PodTemplateSpec `json:"templatePool,omitempty"`
// Statuses stores desired instance status instanceID --> desiredStatus
Statuses map[string]InstanceStatus `json:"statuses,omitempty"`
// Templates stores instanceID --> template name
Templates map[string]string `json:"templates,omitempty"`
// UpdateStrategy indicates the TappUpdateStrategy that will be
// employed to update Pods in the TApp
UpdateStrategy TAppUpdateStrategy `json:"updateStrategy,omitempty"`
// ForceDeletePod indicates whether force delete pods when it is being deleted because of NodeLost.
// Default values is false.
ForceDeletePod bool `json:"forceDeletePod,omitempty"`
// AutoDeleteUnusedTemplate indicates whether auto delete templates when it is unused.
// Default values is false.
AutoDeleteUnusedTemplate bool `json:"autoDeleteUnusedTemplate,omitempty"`
// NeverMigrate indicates whether to migrate pods. If it is true, pods will never be migrated to
// other nodes, otherwise it depends on other conditions(e.g. pod restart policy).
NeverMigrate bool `json:"neverMigrate,omitempty"`
// volumeClaimTemplates is a list of claims that pods are allowed to reference.
// The StatefulSet controller is responsible for mapping network identities to
// claims in a way that maintains the identity of a pod. Every claim in
// this list must have at least one matching (by name) volumeMount in one
// container in the template. A claim in this list takes precedence over
// any volumes in the template, with the same name.
// TODO: Define the behavior if a claim already exists with the same name.
VolumeClaimTemplates []corev1.PersistentVolumeClaim `json:"volumeClaimTemplates,omitempty"`
// ServiceName is the name of the service that governs this TApp.
// This service must exist before the TApp, and is responsible for
// the network identity of the set. Pods get DNS/hostnames that follow the
// pattern: pod-specific-string.serviceName.default.svc.cluster.local
// where "pod-specific-string" is managed by the TApp controller.
ServiceName string `json:"serviceName,omitempty"`
//DefaultTemplateName is the default template name for scale
DefaultTemplateName string `json:"defaultTemplateName"`
}
4.1.3 TAppUpdateStrategy
// TApp update strategy
type TAppUpdateStrategy struct {
// Following fields are rolling update related configuration.
// Template is the rolling update template name
Template string `json:"template,omitempty"`
// MaxUnavailable is the max unavailable number when tapp is rolling update, default is 1.
MaxUnavailable *intstr.IntOrString `json:"maxUnavailable,omitempty"`
// Following fields are force update related configuration.
ForceUpdate ForceUpdateStrategy `json:"forceUpdate,omitempty"`
}
4.1.4 ForceUpdateStrategy
type ForceUpdateStrategy struct {
// MaxUnavailable is the max unavailable number when tapp is forced update, default is 100%.
MaxUnavailable *intstr.IntOrString `json:"maxUnavailable,omitempty"`
}
4.1.5 InstanceStatus
type InstanceStatus string
const (
InstanceNotCreated InstanceStatus = "NotCreated"
InstancePending InstanceStatus = "Pending"
InstanceRunning InstanceStatus = "Running"
InstanceUpdating InstanceStatus = "Updating"
InstancePodFailed InstanceStatus = "PodFailed"
InstancePodSucc InstanceStatus = "PodSucc"
InstanceKilling InstanceStatus = "Killing"
InstanceKilled InstanceStatus = "Killed"
InstanceFailed InstanceStatus = "Failed"
InstanceSucc InstanceStatus = "Succ"
InstanceUnknown InstanceStatus = "Unknown"
)
4.1.6 TAppStatus
type AppStatus string
const (
AppPending AppStatus = "Pending"
AppRunning AppStatus = "Running"
AppFailed AppStatus = "Failed"
AppSucc AppStatus = "Succ"
AppKilled AppStatus = "Killed"
)
// TAppStatus represents the current state of a TApp.
type TAppStatus struct {
// most recent generation observed by controller.
ObservedGeneration int64 `json:"observedGeneration,omitempty"`
// Replicas is the number of actual replicas.
Replicas int32 `json:"replicas"`
// ReadyReplicas is the number of running replicas
ReadyReplicas int32 `json:"readyReplicas"`
// ScaleSelector is a label for query over pods that should match the replica count used by HPA.
ScaleLabelSelector string `json:"scaleLabelSelector,omitempty"`
// AppStatus describe the current TApp state
AppStatus AppStatus `json:"appStatus,omitempty"`
// Statues stores actual instanceID --> InstanceStatus
Statuses map[string]InstanceStatus `json:"statuses,omitempty"`
}
4.1.7 TAppList
// TAppList is a collection of TApp.
type TAppList struct {
metav1.TypeMeta `json:",inline"`
metav1.ListMeta `json:"metadata,omitempty"`
Items []TApp `json:"items"`
}
4.1.8 Instance
// instance is the control block used to transmit all updates about a single instance.
// It serves as the manifest for a single instance. Users must populate the pod
// and parent fields to pass it around safely.
type Instance struct {
// pod is the desired pod.
pod *corev1.Pod
// id is the identity index of this instance.
id string
// parent is a pointer to the parent tapp.
parent *tappv1.TApp
}
4.2 核心逻辑
4.2.1 主函数
主函数流程图：
sequenceDiagram
main ->> + k8s: 初始化kube client，与k8s对象交互
k8s ->> - main: 返回kube client
main ->> + tapp: 初始化tapp client，与tapp crd对象交互
tapp ->> - main: 返回tapp client
main ->> + k8s: 初始化extensions client， 与k8s crd对象交互
k8s ->> - main: 返回extensions client
main ->> + kubeInformerFactory: 初始化kubeInformer工厂
kubeInformerFactory ->> - main: 返回kubeInformer工厂
main ->> + tappInformerFactory: 初始化tappInformer工厂
tappInformerFactory ->> - main: 返回tappInformer工厂
main ->> + controller: NewController，初始化controller对象
controller ->> + kubeInformerFactory: 初始化podInformer, 并设置Event触发回调函数
kubeInformerFactory -->> - controller:
controller ->> + tappInformerFactory: 初始化tappInformer, 并设置Event触发回调函数
tappInformerFactory -->> - controller:
controller ->> - main: 返回controller对象
main ->> + leaderElection: 获取leaderElection对象
leaderElection -->> - main: 根据leaderElection.LeaderElect值是否进行leader选举
alt LeaderElect != true: 没启用leader选举
leaderElection ->> controller: 直接进入run函数,最终调用controller对象的Run函数
else LeaderElect == true: 启用leader选举
leaderElection ->> controller: 初始化资源锁,设置选举回调函数为run,最终调用controller对象的Run函数
end
4.2.2 Sync函数
main函数最终调用的是run函数，run函数调用流程：
run -> controller.Run -> c.runWorker -> controller.Sync
run := func(ctx context.Context) {
stop := ctx.Done()
//createCRD默认为true，tapp控制器启动时自动创建tapp crd，如果存在则更新
if createCRD {
wait.PollImmediateUntil(time.Second*5, func() (bool, error) { return tapp.EnsureCRDCreated(extensionsClient) }, stop)
}
//registerAdmission默认为false，是否注册一个ValidatingWebhook，用于校验tapp对象的合法性
if registerAdmission {
wait.PollImmediateUntil(time.Second*5, func() (bool, error) {
return admission.Register(kubeClient, namespace, tlsCAfile)
}, stop)
server, err := admission.NewServer(listenAddress, tlsCertFile, tlsKeyFile)
if err != nil {
klog.Fatalf("Error new admission server: %v", err)
}
go server.Run(stop)
}
//运行所有🕐已注册的kubeInformer
//Start函数已经是非阻塞，在一个独立的goroutine启动所有已注册informer, 无需`go kubeInformerFactory.Start`
go kubeInformerFactory.Start(stop)
//运行所有🕐已注册的tappInformer
go tappInformerFactory.Start(stop)
//设置deletePodAfterAppFinish变量，默认为true
tapp.SetDeletePodAfterAppFinish(deletePodAfterAppFinish)
//
if err = controller.Run(worker, stop); err != nil {
klog.Fatalf("Error running controller: %s", err.Error())
}
}
// worker runs a worker thread that just dequeues items, processes them, and marks them done.
// It enforces that the syncHandler is never invoked concurrently with the same key.
func (c *Controller) runWorker() {
for {
func() {
key, quit := c.workqueue.Get()
if quit {
return
}
defer c.workqueue.Done(key)
if err := c.syncHandler(key.(string)); err != nil {
klog.Errorf("Error syncing TApp %v, re-queuing: %v", key.(string), err)
c.workqueue.AddRateLimited(key)
} else {
c.workqueue.Forget(key)
}
}()
}
}
核心处理函数是c.syncHandler，而controller.syncHandler = controller.Sync
// Sync syncs the given tapp.
func (c *Controller) Sync(key string) error {
startTime := time.Now()
defer func() {
klog.V(4).Infof("Finished syncing tapp %s(%v)", key, time.Now().Sub(startTime))
}()
//controller.podStoreSynced = podInformer.Informer().HasSynced
if !c.podStoreSynced() {
klog.V(2).Infof("Pod store is not synced, skip syncing tapp %s", key)
// Sleep to give the pod reflector goroutine a chance to run.
time.Sleep(PodStoreSyncedPollPeriod)
return fmt.Errorf("waiting for pods controller to sync")
}
namespace, name, err := cache.SplitMetaNamespaceKey(key)
if err != nil {
return err
}
tapp, err := c.tappLister.TApps(namespace).Get(name)
if errors.IsNotFound(err) {
klog.Infof("TApp has been deleted %v", key)
return nil
}
if err != nil {
klog.Errorf("Unable to retrieve tapp %s from store: %v", util.GetTAppFullName(tapp), err)
return err
}
//1. 为tapp对象设置默认值
//2. 根据tapp.Spec.AutoDeleteUnusedTemplate开关，是否移除未被使用的template
//3. 为tapp设置labelSelector
//4. 为tapp设置scaleLabelSelector
//5. reflect反射对比前后tapp的Spec内容，有差异则更新tapp
//6. reflect反射对比前后tapp的Status内容，有差异则更新tapp Status
err = c.preprocessTApp(tapp)
if err != nil {
klog.Errorf("Failed to preprocess tapp %s: %v", util.GetTAppFullName(tapp), err)
return err
}
//1. 获取tapp的labelSelector
//2. 根据labelSelector过滤pod列表并返回
pods, err := c.getPodsForTApp(tapp)
if err != nil {
klog.Errorf("Failed to get pods for tapp %s: %v", util.GetTAppFullName(tapp), err)
return err
}
//Tapp finish判断依据：tapp状态处于Failed或Succ或Killed
if isTAppFinished(tapp) && tapp.Generation == tapp.Status.ObservedGeneration &&
tapp.Spec.Replicas == tapp.Status.Replicas && len(pods) == 0 {
klog.Errorf("Tapp %s has finished, replica: %d, status: %s", util.GetTAppFullName(tapp),
tapp.Spec.Replicas, tapp.Status.AppStatus)
return nil
}
//同步tapp分为三个环节：
//1. 计算出要添加/删除/强制删除/更新的instance，instance~=pod，多了parent父指针指向tapp和id
//2. 同步pod的Conditions
//3. 同步instances
c.syncTApp(tapp, pods)
//更新tapp的status字段
if err := c.updateTAppStatus(tapp, pods); err != nil {
klog.Errorf("Failed to update tapp %s's status: %v", util.GetTAppFullName(tapp), err)
return err
}
return nil
}
所有pod的处理逻辑最终都在syncTApp函数里
func (c *Controller) syncTApp(tapp *tappv1.TApp, pods []*corev1.Pod) {
klog.V(4).Infof("Syncing tapp %s with %d pods", util.GetTAppFullName(tapp), len(pods))
//1. a. 设置template的templateHash和uniqHash，包括TemplatePool；
// b. template hash: 对spec.template进行hash；
// c. uniq hash: 对container和initContainer但不包括image内容进行hash；
//2. 返回一个key为pod索引序号，值为pod对象的podMap map数据结构；
//3. a. tapp.Spec.Statuses存放的是pod索引和Instance状态的map，这里instance~=pod的概念；
// b. 遍历tapp.Spec.Statuses计算出正在运行的pod放入running sets和Instance处于killed状态的pod放入completed sets；
// c. 如果启用deletePodAfterAppFinish，遍历tapp.Status.Statuses把Instance处于Failed/Succ状态的pod放入completed sets；
//4. a. 同步Running的pod，通过遍历第三步的running sets，再和第二步的podMap对比；
// 返回podActions map a1，podActions存放的pod索引id和pod的下一步执行动作(CREATE/UPDATE/RECREATE/DELETE)；
// b. 同步Completed的pod，通过遍历第三步的completed sets，再和第二步的podMap对比；如果存在于podMap，即把podAction置为DELETE；
// 返回podActions map a2；
// c. 同步不合法的pod，通过遍历🥈第二步的podMap，如果pod索引id>=tapp.Spec.Replicas, 即把podAction置为DELETE；
// 返回podActions map a3；
// d. 对上面的podActions map a1, a2, a3进行merge操作并返回podActions
//5. a. 从running sets和podMap中过滤出ready的pod到availablePods sets，并返回
// b. 遍历podActions，匹配PodAction:
// c. 如果匹配到"DELETE"的PodAction，pod正处于删除状态且`tapp.Spec.ForceDeletePod`启用;
// 如果pod所在的节点不存在或节点Condition状态不为true，则把Instance放入forceDel []*Instance中;
// 如果pod不是正处于删除状态，则把Instance放入del []*Instance中;
// 从availablePods sets中删除这个pod；
// d. 如果匹配到"CREATE"的PodAction，创建新的Instance对象，其中设置pod的OwnerReferences为tapp对象；
// 并把这个Instance对象放入add []*Instance中；
// e. 设置maxUnavailableForceUpdate默认值为100%，minAvailablePods=(running sets的数量 - maxUnavailableForceUpdate)
// f. 再次遍历podActions，匹配PodAction:
// g. 如果匹配到"UPDATE"的PodAction，判断pod是否处于滚动更新(滚动更新判断依据：tapp.Spec.UpdateStrategy.Template和tapp.Spec.Templates[pod-id]是否一致)；
// 如果pod不处于滚动更新状态，则新建Instance对象，然后放入update []*Instance，并从availablePods删除该pod；
// 如果pod处于滚动更新状态，则把pod id直接放入rollingUpdateIds []string;
// h. 如果匹配到"RECREATE"的PodAction，
// 如果pod不处于滚动更新状态，则使用现在的pod构建Instance对象，然后放入del []*Instance，并从availablePods删除该pod；
// 如果pod处于滚动更新状态，则把pod id直接放入rollingUpdateIds []string;
// i. 设置maxUnavailable默认值为1，minAvailablePods=(running sets的数量 - maxUnavailable)
// j. rollingUpdateIds按pod id递增的顺序排列
// k. 遍历rollingUpdateIds，匹配podAction:
// l. 如果匹配到"UPDATE"的podAction，新建Instance对象，然后放入update []*Instance, 并从availablePods删除该pod；
// m. 如果匹配到"RECREATE"的PodAction，则使用现在的pod构建Instance对象，然后放入del []*Instance，并从availablePods删除该pod；
// x. 返回add/del/forceDel/update的[]*Instance;
add, del, forceDel, update := c.instanceToSync(tapp, pods)
//6. 同步pod的Conditions
c.syncPodConditions(pods, append(del, update...))
//7. a. 遍历add []*Instance，并发启动goroutine创建PersistentVolumeClaims/pod
// b. 遍历del []*Instance，并发启动goroutine删除pod
// c. 遍历forceDel []*Instance，并发启动goroutine强制删除pod
// d. 遍历update []*Instance，更新pod
c.syncer.SyncInstances(add, del, forceDel, update)
}
4.3 多版本实例实现原理
apiVersion: apps.tkestack.io/v1
kind: TApp
metadata:
name: example-tapp
namespace: test
spec:
replicas: 3
template:
metadata:
labels:
app: example-tapp
spec:
containers:
- name: nginx
image: nginx:1.8.1
templatePool:
"test":
metadata:
labels:
app: example-tapp
spec:
containers:
- name: nginx
image: nginx:1.7.9
"test2":
metadata:
labels:
app: example-tapp
spec:
containers:
- name: nginx
image: nginx:1.8.3
templates:
"1": "test"
"2": "test2"
看上面的例子，通过spec.templatePool定义template模板test和test2；
通过spec.templates指明pod-xxx-1使用test模板，pod-xxx-2使用test2模板
4.4 原地升级实现原理
pod标签中有两种hash，一种是template hash，另一种是uniq hash.
- template hash(针对spec.template内容)
- uniq hash(针对container和initContainer但不包括image内容)
原地升级触发条件是：当template hash发生改变，uniq hash没改变，则替换pod的镜像； 直接替换pod的镜像可以起到原地升级的效果.
参考链接
「真诚赞赏，手留余香」
爱折腾的工程师
真诚赞赏，手留余香
使用微信扫描二维码完成支付
