Openkruise/rollouts 源码解读

Openkruise/rollouts 源码解读

最近因为工作需要的原因,无意中接触到了 rollouts 这个开源项目。它是由阿里巴巴开源的一个k8s相关的一个工具。

其主要作用是工作负载(workload) 批量发布与回滚

当前支持的工作负载(workload)有:

  • Deployment
  • CloneSet
  • StatefulSet

同时也支持 Nginx/ALB/Istio 的流量路由控制

常规情况下,我们升级一个工作负载,以Deployment为例,我们更改Deployment 镜像的版本之后,Deployment的控制器会重新去创建新的副本,然后替换掉原来的副本。
这种情况下,如果我们创建出来的副本程序本身是Ready的,但业务不是Ready的,控制器本身仍然会继续将没有更新的副本更新完成,即使我们后续发现了问题,将其回滚,那它的影响面也会非常大。
所以我们需要一定的反悔时间,留给工作人员去观察业务是否正常,正常之后,再继续灰度下去,而不是ALL IN,然后ALL GET OUT。而 rollouts 正是帮我们做这个事情的。

rollout 状态流转

rollout 状态流转

简单使用

参考 基础使用文档

源码解读

CRD设计

Rollout

WorkloadRef 定义了我们要去监听的workload对象,当这个对象 PodTemplate 发生变更后,就会执行更新策略。

// WorkloadRef holds a references to the Kubernetes object
type WorkloadRef struct {
    // API Version of the referent
    APIVersion string `json:"apiVersion"`
    // Kind of the referent
    Kind string `json:"kind"`
    // Name of the referent
    Name string `json:"name"`
}

CanaryStrategy 定义了我们的更新策略,其中CanaryStep则定义了我们每一步的动作:更新的副本数,更新完成后暂停多久,流量的权重。
对于流量相关的我这里就不做详解了,有兴趣的可以自己看。

// CanaryStrategy defines parameters for a Replica Based Canary
type CanaryStrategy struct {
    // Steps define the order of phases to execute release in batches(20%, 40%, 60%, 80%, 100%)
    // +optional
    Steps []CanaryStep `json:"steps,omitempty"`
    // TrafficRoutings hosts all the supported service meshes supported to enable more fine-grained traffic routing
    TrafficRoutings []*TrafficRouting `json:"trafficRoutings,omitempty"`
}

// CanaryStep defines a step of a canary workload.
type CanaryStep struct {
    // SetWeight sets what percentage of the canary pods should receive
    // +optional
    Weight *int32 `json:"weight,omitempty"`
    // Replicas is the number of expected canary pods in this batch
    // it can be an absolute number (ex: 5) or a percentage of total pods.
    Replicas *intstr.IntOrString `json:"replicas,omitempty"`
    // Pause defines a pause stage for a rollout, manual or auto
    // +optional
    Pause RolloutPause `json:"pause,omitempty"`
}

上面这部分定义了我们要观察的对象已经我们的动作。
下面这部分则定义了rollout的状态, rollout 在执行过程中,它的状态就会记录在这里面。

type RolloutStatus struct {
    // INSERT ADDITIONAL STATUS FIELD - define observed state of cluster
    // Important: Run "make" to regenerate code after modifying this file
    
    // observedGeneration is the most recent generation observed for this Rollout.
    ObservedGeneration int64 `json:"observedGeneration,omitempty"`
    // CanaryRevision the hash of the canary pod template
    // +optional
    //CanaryRevision string `json:"canaryRevision,omitempty"`
    // StableRevision indicates the revision pods that has successfully rolled out
    StableRevision string `json:"stableRevision,omitempty"`
    // Conditions a list of conditions a rollout can have.
    // +optional
    Conditions []RolloutCondition `json:"conditions,omitempty"`
    // Canary describes the state of the canary rollout
    // +optional
    CanaryStatus *CanaryStatus `json:"canaryStatus,omitempty"`
    // +optional
    //BlueGreenStatus *BlueGreenStatus `json:"blueGreenStatus,omitempty"`
    // Phase is the rollout phase.
    Phase RolloutPhase `json:"phase,omitempty"`
    // Message provides details on why the rollout is in its current phase
    Message string `json:"message,omitempty"`
}

// RolloutCondition describes the state of a rollout at a certain point.
type RolloutCondition struct {
    // Type of rollout condition.
    Type RolloutConditionType `json:"type"`
    // Phase of the condition, one of True, False, Unknown.
    Status corev1.ConditionStatus `json:"status"`
    // The last time this condition was updated.
    LastUpdateTime metav1.Time `json:"lastUpdateTime,omitempty"`
    // Last time the condition transitioned from one status to another.
    LastTransitionTime metav1.Time `json:"lastTransitionTime,omitempty"`
    // The reason for the condition's last transition.
    Reason string `json:"reason"`
    // A human readable message indicating details about the transition.
    Message string `json:"message"`
}
// CanaryStatus status fields that only pertain to the canary rollout
type CanaryStatus struct {
    // observedWorkloadGeneration is the most recent generation observed for this Rollout ref workload generation.
    ObservedWorkloadGeneration int64 `json:"observedWorkloadGeneration,omitempty"`
    // ObservedRolloutID will record the newest spec.RolloutID if status.canaryRevision equals to workload.updateRevision
    ObservedRolloutID string `json:"observedRolloutID,omitempty"`
    // RolloutHash from rollout.spec object
    RolloutHash string `json:"rolloutHash,omitempty"`
    // CanaryService holds the name of a service which selects pods with canary version and don't select any pods with stable version.
    CanaryService string `json:"canaryService"`
    // CanaryRevision is calculated by rollout based on podTemplateHash, and the internal logic flow uses
    // It may be different from rs podTemplateHash in different k8s versions, so it cannot be used as service selector label
    // +optional
    CanaryRevision string `json:"canaryRevision"`
    // pod template hash is used as service selector label
    PodTemplateHash string `json:"podTemplateHash"`
    // CanaryReplicas the numbers of canary revision pods
    CanaryReplicas int32 `json:"canaryReplicas"`
    // CanaryReadyReplicas the numbers of ready canary revision pods
    CanaryReadyReplicas int32 `json:"canaryReadyReplicas"`
    // CurrentStepIndex defines the current step of the rollout is on. If the current step index is null, the
    // controller will execute the rollout.
    // +optional
    CurrentStepIndex int32           `json:"currentStepIndex"`
    CurrentStepState CanaryStepState `json:"currentStepState"`
    Message          string          `json:"message,omitempty"`
    LastUpdateTime   *metav1.Time    `json:"lastUpdateTime,omitempty"`
}

其中整个rollout有这些阶段:

const (
    // RolloutPhaseInitial indicates a rollout is Initial
    RolloutPhaseInitial RolloutPhase = "Initial"
    // RolloutPhaseHealthy indicates a rollout is healthy
    RolloutPhaseHealthy RolloutPhase = "Healthy"
    // RolloutPhaseProgressing indicates a rollout is not yet healthy but still making progress towards a healthy state
    RolloutPhaseProgressing RolloutPhase = "Progressing"
    // RolloutPhaseTerminating indicates a rollout is terminated
    RolloutPhaseTerminating RolloutPhase = "Terminating"
)

每个CanaryStep则有这些状态:

const (
    CanaryStepStateUpgrade         CanaryStepState = "StepUpgrade"
    CanaryStepStateTrafficRouting  CanaryStepState = "StepTrafficRouting"
    CanaryStepStateMetricsAnalysis CanaryStepState = "StepMetricsAnalysis"
    CanaryStepStatePaused          CanaryStepState = "StepPaused"
    CanaryStepStateReady           CanaryStepState = "StepReady"
    CanaryStepStateCompleted       CanaryStepState = "Completed"
)

每个Conditions则有这些类型

const (
    RolloutConditionProgressing RolloutConditionType = "Progressing"
    // Progressing Reason
    ProgressingReasonInitializing = "Initializing"
    ProgressingReasonInRolling    = "InRolling"
    ProgressingReasonFinalising   = "Finalising"
    ProgressingReasonSucceeded    = "Succeeded"
    ProgressingReasonCancelling   = "Cancelling"
    ProgressingReasonCanceled     = "Canceled"
    ProgressingReasonPaused       = "Paused"
    
    // Terminating condition
    RolloutConditionTerminating RolloutConditionType = "Terminating"
    // Terminating Reason
    TerminatingReasonInTerminating = "InTerminating"
    TerminatingReasonCompleted     = "Completed"
)
BatchRelease

BatchRelease的Spec部分我就步过多介绍,它和 Rollout 差别不大。主要区别在与Status的记录。

BatchReleaseCanaryStatus记录了每个批次的发布状态,其中CurrentBatchState 3个状态:Upgrading,Verifying,Ready

type BatchReleaseCanaryStatus struct {
    // CurrentBatchState indicates the release state of the current batch.
    CurrentBatchState BatchReleaseBatchStateType `json:"batchState,omitempty"`
    // The current batch the rollout is working on/blocked, it starts from 0
    CurrentBatch int32 `json:"currentBatch"`
    // BatchReadyTime is the ready timestamp of the current batch or the last batch.
    // This field is updated once a batch ready, and the batches[x].pausedSeconds
    // relies on this field to calculate the real-time duration.
    BatchReadyTime *metav1.Time `json:"batchReadyTime,omitempty"`
    // UpdatedReplicas is the number of upgraded Pods.
    UpdatedReplicas int32 `json:"updatedReplicas,omitempty"`
    // UpdatedReadyReplicas is the number upgraded Pods that have a Ready Condition.
    UpdatedReadyReplicas int32 `json:"updatedReadyReplicas,omitempty"`
}

const (
    // UpgradingBatchState indicates that current batch is at upgrading pod state
    UpgradingBatchState BatchReleaseBatchStateType = "Upgrading"
    // VerifyingBatchState indicates that current batch is at verifying whether it's ready state
    VerifyingBatchState BatchReleaseBatchStateType = "Verifying"
    // ReadyBatchState indicates that current batch is at batch ready state
    ReadyBatchState BatchReleaseBatchStateType = "Ready"
)

上述两个CRD就是整个项目的核心,它们的各种状态转换,构成了完整的升级流程;其中 Rollout 负责整体流程的把控,由用户创建;BatchRelease 负责具体的各个workload的发布,由Rollout进行创建。

代码解读

main函数我就不详细介绍了,其主要是启动了两个控制器,来分别管理Rollout和BatchRelease两个CRD,同时启动一个webhook,来暂停workload,使得k8s控制器,暂时不做升级。

// main.go
func main() {
	...
	// Rollout的管理器
	mgr, err := ctrl.NewManager(cfg, ctrl.Options{
		Scheme:                 scheme,
		MetricsBindAddress:     metricsAddr,
		Port:                   9443,
		HealthProbeBindAddress: probeAddr,
		LeaderElection:         enableLeaderElection,
		LeaderElectionID:       "71ddec2c.kruise.io",
		NewClient:              utilclient.NewClient,
	})
	...
    // BatchRelease的管理器
	if err = br.Add(mgr); err != nil {
		setupLog.Error(err, "unable to create controller", "controller", "BatchRelease")
		os.Exit(1)
	}
	...
    // workload 和 rollout 的 webhook
	if err = webhook.SetupWithManager(mgr); err != nil {
		setupLog.Error(err, "unable to setup webhook")
		os.Exit(1)
	}
    ...
}

Rollout 的调协过程
type RolloutReconciler struct {
	client.Client
	Scheme *runtime.Scheme

	Recorder record.EventRecorder //用于事件记录
	Finder   *util.ControllerFinder //用于寻找对应workload的控制器
}

我们可以详细来看看 Rollout Reconcile的主流程,
大概可以概括为以下几步:

  1. 监听rollout变更
  2. 获取一个动态的Watcher来监听对应的workload更新
  3. 处理finalizer:如果Rollout是新创建的添加finalizer,如果是Rollout需要删除就移除finalizer
  4. 更新rollout status,主要处理阶段为:RolloutPhaseInitial->RolloutPhaseHealthy->RolloutPhaseProgressing->RolloutPhaseHealthy->RolloutPhaseTerminating
  5. 处理RolloutPhaseProgressing阶段更新流程
  6. 处理RolloutPhaseTerminating阶段终止流程
func (r *RolloutReconciler) Reconcile(ctx context.Context, req ctrl.Request) (ctrl.Result, error) {
	// 1.监听rollout变更
	rollout := &rolloutv1alpha1.Rollout{}
	err := r.Get(context.TODO(), req.NamespacedName, rollout)
	if err != nil {
		if errors.IsNotFound(err) {
			return ctrl.Result{}, nil
		}
		return ctrl.Result{}, err
	}
    // 2. 获取一个动态的Watcher来监听对应的workload更新
	workloadRef := rollout.Spec.ObjectRef.WorkloadRef
	workloadGVK := util.GetGVKFrom(workloadRef)
	_, exists := watchedWorkload.Load(workloadGVK.String())
	if workloadRef != nil && !exists {
		succeeded, err := util.AddWatcherDynamically(runtimeController, workloadHandler, workloadGVK)
		if err != nil {
			return ctrl.Result{}, err
		} else if succeeded {
			watchedWorkload.LoadOrStore(workloadGVK.String(), struct{}{})
			klog.Infof("Rollout controller begin to watch workload type: %s", workloadGVK.String())
			return ctrl.Result{}, nil
		}
	}
	// 3.处理finalizer
	err = r.handleFinalizer(rollout)
	if err != nil {
		return ctrl.Result{}, err
	}
	// 4.更新rollout status的
	done, err := r.updateRolloutStatus(rollout)
	if err != nil {
		return ctrl.Result{}, err
	} else if !done {
		return ctrl.Result{}, nil
	}

	var recheckTime *time.Time
	switch rollout.Status.Phase {
	case rolloutv1alpha1.RolloutPhaseProgressing:
		// 5. 处理RolloutPhaseProgressing阶段更新流程
		recheckTime, err = r.reconcileRolloutProgressing(rollout)
	case rolloutv1alpha1.RolloutPhaseTerminating:
		// 6. 处理RolloutPhaseTerminating阶段终止流程
		recheckTime, err = r.reconcileRolloutTerminating(rollout)
	}
	if err != nil {
		return ctrl.Result{}, err
	} else if recheckTime != nil {
		return ctrl.Result{RequeueAfter: time.Until(*recheckTime)}, nil
	}
	return ctrl.Result{}, nil
}

1~3步比较简单,我就不多说,我们详细讲解一下4,5,6三步。

4. 更新rollout status

部分代码已经删减

func (r *RolloutReconciler) updateRolloutStatus(rollout *rolloutv1alpha1.Rollout) (done bool, err error) {
	... 
	// 1.如果是删除CRD, rollout status Phase 就由其他阶段转换为(终止阶段)RolloutPhaseTerminating,同时添加 Condition:RolloutConditionTerminating=false.
	// 注意这里要特别注意Reason 因为Condition也状态转换条件之一
	// 2.如果rollout status Phase 没有,就默认是初始化阶段(RolloutPhaseInitial)
	if !rollout.DeletionTimestamp.IsZero() && newStatus.Phase != rolloutv1alpha1.RolloutPhaseTerminating {
		newStatus.Phase = rolloutv1alpha1.RolloutPhaseTerminating
		cond := util.NewRolloutCondition(rolloutv1alpha1.RolloutConditionTerminating, corev1.ConditionFalse, rolloutv1alpha1.TerminatingReasonInTerminating, "Rollout is in terminating")
		util.SetRolloutCondition(&newStatus, *cond)
	} else if newStatus.Phase == "" {
		newStatus.Phase = rolloutv1alpha1.RolloutPhaseInitial
	}
	// 3.获取rollout 对于的workload 的信息,没有获取到重置rollout status Phase 为RolloutPhaseInitial
	workload, err := r.Finder.GetWorkloadForRef(rollout.Namespace, rollout.Spec.ObjectRef.WorkloadRef)
	if err != nil {
		klog.Errorf("rollout(%s/%s) get workload failed: %s", rollout.Namespace, rollout.Name, err.Error())
		return
	} else if workload == nil {
		if rollout.DeletionTimestamp.IsZero() {
			resetStatus(&newStatus)
			klog.Infof("rollout(%s/%s) workload not found, and reset status be Initial", rollout.Namespace, rollout.Name)
		}
		done = true
		return
	}
	...
    // 4 rollout status Phase 状态转换,每次调协,如果条件成立就转换一次,转换顺序为:
	// RolloutPhaseInitial --> RolloutPhaseHealthy
	// RolloutPhaseHealthy --> RolloutPhaseProgressing
	// RolloutPhaseProgressing --> RolloutPhaseHealthy
	// 其中RolloutPhaseHealthy --> RolloutPhaseProgressing 阶段 会将Condition的Type设置为RolloutConditionProgressing,同时修改其Reason为ProgressingReasonInitializing
	// RolloutPhaseProgressing --> RolloutPhaseHealthy 阶段 则需要Condition的Type为RolloutConditionProgressing的Reason为ProgressingReasonSucceeded或ProgressingReasonCanceled.
	// 或者没有Type为RolloutConditionProgressing的Condition
	switch newStatus.Phase {
	case rolloutv1alpha1.RolloutPhaseInitial:
		klog.Infof("rollout(%s/%s) status phase from(%s) -> to(%s)", rollout.Namespace, rollout.Name, rolloutv1alpha1.RolloutPhaseInitial, rolloutv1alpha1.RolloutPhaseHealthy)
		newStatus.Phase = rolloutv1alpha1.RolloutPhaseHealthy
		newStatus.Message = "rollout is healthy"
	case rolloutv1alpha1.RolloutPhaseHealthy:
		if workload.InRolloutProgressing {
			// from healthy to progressing
			klog.Infof("rollout(%s/%s) status phase from(%s) -> to(%s)", rollout.Namespace, rollout.Name, rolloutv1alpha1.RolloutPhaseHealthy, rolloutv1alpha1.RolloutPhaseProgressing)
			newStatus.Phase = rolloutv1alpha1.RolloutPhaseProgressing
			cond := util.NewRolloutCondition(rolloutv1alpha1.RolloutConditionProgressing, corev1.ConditionFalse, rolloutv1alpha1.ProgressingReasonInitializing, "Rollout is in Progressing")
			util.SetRolloutCondition(&newStatus, *cond)
		} else if workload.IsInStable && newStatus.CanaryStatus == nil {
			newStatus.CanaryStatus = &rolloutv1alpha1.CanaryStatus{
				CanaryReplicas:             workload.CanaryReplicas,
				CanaryReadyReplicas:        workload.CanaryReadyReplicas,
				ObservedRolloutID:          getRolloutID(workload, rollout),
				ObservedWorkloadGeneration: workload.Generation,
				PodTemplateHash:            workload.PodTemplateHash,
				CanaryRevision:             workload.CanaryRevision,
				CurrentStepIndex:           int32(len(rollout.Spec.Strategy.Canary.Steps)),
				CurrentStepState:           rolloutv1alpha1.CanaryStepStateCompleted,
			}
			newStatus.Message = "workload deployment is completed"
		}
	case rolloutv1alpha1.RolloutPhaseProgressing:
		cond := util.GetRolloutCondition(newStatus, rolloutv1alpha1.RolloutConditionProgressing)
		if cond == nil || cond.Reason == rolloutv1alpha1.ProgressingReasonSucceeded || cond.Reason == rolloutv1alpha1.ProgressingReasonCanceled {
			newStatus.Phase = rolloutv1alpha1.RolloutPhaseHealthy
		}
	}
	done = true
	return
}
5. 处理RolloutPhaseProgressing阶段更新流程

当rollout status Phase的状态处于RolloutPhaseProgressing后,后续的状态转换则主要是Type为RolloutConditionProgressing的Condition的Reason转换。

func (r *RolloutReconciler) reconcileRolloutProgressing(rollout *rolloutv1alpha1.Rollout) (*time.Time, error) {
	//1.获取Type 为 RolloutConditionProgressing 的 Condition 和 workload
	cond := util.GetRolloutCondition(rollout.Status, rolloutv1alpha1.RolloutConditionProgressing)
	workload, err := r.Finder.GetWorkloadForRef(rollout.Namespace, rollout.Spec.ObjectRef.WorkloadRef)
	if err != nil {
		return nil, err
	} else if workload == nil {
		return nil, nil
	} else if !workload.IsStatusConsistent {
		return nil, nil
	}
	
	var recheckTime *time.Time
	newStatus := rollout.Status.DeepCopy()
	// Condition Reason 状态转换,每次调协,如果条件成立就转换一次,转换顺序为:
	// ProgressingReasonInitializing --> ProgressingReasonInRolling
	// ProgressingReasonInRolling --> ProgressingReasonCancelling
	// ProgressingReasonInRolling --> ProgressingReasonPaused
	// ProgressingReasonInRolling --> ProgressingReasonInitializing
	// ProgressingReasonInRolling --> ProgressingReasonFinalising
	// ProgressingReasonFinalising --> ProgressingReasonSucceeded
	// ProgressingReasonPaused --> ProgressingReasonCancelling
	// ProgressingReasonPaused --> ProgressingReasonInRolling
	// ProgressingReasonCancelling --> ProgressingReasonCanceled
	switch cond.Reason {
	case rolloutv1alpha1.ProgressingReasonInitializing:
		newStatus.CanaryStatus = &rolloutv1alpha1.CanaryStatus{} // 初始化CanaryStatus
		done, _, err := r.doProgressingInitializing(rollout, newStatus)
		if done {
			progressingStateTransition(newStatus, corev1.ConditionFalse, rolloutv1alpha1.ProgressingReasonInRolling, "Rollout is in Progressing")
		}
	case rolloutv1alpha1.ProgressingReasonInRolling:
		recheckTime, err = r.doProgressingInRolling(rollout, workload, newStatus)
	case rolloutv1alpha1.ProgressingReasonFinalising:
		var done bool
		done, recheckTime, err = r.doFinalising(rollout, newStatus, true)
		if done {
			progressingStateTransition(newStatus, corev1.ConditionTrue, rolloutv1alpha1.ProgressingReasonSucceeded, "Rollout has been completed, and succeed")
		}
	case rolloutv1alpha1.ProgressingReasonPaused:
		if workload.IsInRollback {
			newStatus.CanaryStatus.CanaryRevision = workload.CanaryRevision
			r.Recorder.Eventf(rollout, corev1.EventTypeNormal, "Progressing", "workload has been rollback, then rollout is canceled")
			progressingStateTransition(newStatus, corev1.ConditionFalse, rolloutv1alpha1.ProgressingReasonCancelling, "The workload has been rolled back and the rollout process will be cancelled")
		} else if !rollout.Spec.Strategy.Paused {
			progressingStateTransition(newStatus, corev1.ConditionFalse, rolloutv1alpha1.ProgressingReasonInRolling, "")
		}

	case rolloutv1alpha1.ProgressingReasonCancelling:
		var done bool
		done, recheckTime, err = r.doFinalising(rollout, newStatus, false)
		if done {
			progressingStateTransition(newStatus, corev1.ConditionFalse, rolloutv1alpha1.ProgressingReasonCanceled, "")
		}
	case rolloutv1alpha1.ProgressingReasonSucceeded, rolloutv1alpha1.ProgressingReasonCanceled:
	}

	err = r.updateRolloutStatusInternal(rollout, *newStatus)
	if err != nil {
		return nil, err
	}
	return recheckTime, nil
}

其主要状态变更为:

  1. ProgressingReasonInitializing --> ProgressingReasonInRolling
  2. ProgressingReasonInRolling --> ProgressingReasonCancelling
  3. ProgressingReasonInRolling --> ProgressingReasonPaused
  4. ProgressingReasonInRolling --> ProgressingReasonInitializing
  5. ProgressingReasonInRolling --> ProgressingReasonFinalising
  6. ProgressingReasonFinalising --> ProgressingReasonSucceeded
  7. ProgressingReasonPaused --> ProgressingReasonCancelling
  8. ProgressingReasonPaused --> ProgressingReasonInRolling
  9. ProgressingReasonCancelling --> ProgressingReasonCanceled

其中Reason为ProgressingReasonInRolling则是核心,它负责处理工作负载在更新过程中碰到的各种问题。

  1. 如果workload需要回滚,则进入 ProgressingReasonCancelling 态。
  2. 如果workload需要暂停,则进入 ProgressingReasonPaused 态。
  3. 如果workload连续发布,就删除掉已经创建的batchRelease,然后重新进入 ProgressingReasonInitializing 态。
  4. 如果rollout发生改变,则重新计算当前的处于计划的第几步,然后更新Status.CanaryStatus。
  5. 如果Status.CanaryStatus.CurrentStepState的状态为CanaryStepStateCompleted,即已经更新完成,则进入 ProgressingReasonFinalising 态。
  6. 正常进行更新。

接下来,我们继续看如何进行正常的更新,其主要状态变更则转移为 Status.CanaryStatus 变更,接下来都以 CanaryStatus 进行表述。

其核心代码下面部分,为了方便阅读,我保留了注释和核心部分,日志则被我移除了:

// canary.go
func (r *rolloutContext) runCanary() error {
	canaryStatus := r.newStatus.CanaryStatus
	// init canary status
	if canaryStatus.CanaryRevision == "" {
		canaryStatus.CurrentStepState = rolloutv1alpha1.CanaryStepStateUpgrade
		canaryStatus.CanaryRevision = r.workload.CanaryRevision
		canaryStatus.CurrentStepIndex = 1
		canaryStatus.RolloutHash = r.rollout.Annotations[util.RolloutHashAnnotation]
	}

	// update canary status
	batch, err := r.batchControl.FetchBatchRelease()
	if err != nil {
		canaryStatus.CanaryReplicas = r.workload.CanaryReplicas
		canaryStatus.CanaryReadyReplicas = r.workload.CanaryReadyReplicas
	} else {
		canaryStatus.CanaryReplicas = batch.Status.CanaryStatus.UpdatedReplicas
		canaryStatus.CanaryReadyReplicas = batch.Status.CanaryStatus.UpdatedReadyReplicas
	}

	switch canaryStatus.CurrentStepState {
	case rolloutv1alpha1.CanaryStepStateUpgrade:
		done, err := r.doCanaryUpgrade()
		if err != nil {
			return err
		} else if done {
			canaryStatus.CurrentStepState = rolloutv1alpha1.CanaryStepStateTrafficRouting
			canaryStatus.LastUpdateTime = &metav1.Time{Time: time.Now()}
		}

	case rolloutv1alpha1.CanaryStepStateTrafficRouting:
		done, err := r.doCanaryTrafficRouting()
		if err != nil {
			return err
		} else if done {
			canaryStatus.LastUpdateTime = &metav1.Time{Time: time.Now()}
			canaryStatus.CurrentStepState = rolloutv1alpha1.CanaryStepStateMetricsAnalysis
		}
		expectedTime := time.Now().Add(time.Duration(defaultGracePeriodSeconds) * time.Second)
		r.recheckTime = &expectedTime

	case rolloutv1alpha1.CanaryStepStateMetricsAnalysis:
		done, err := r.doCanaryMetricsAnalysis()
		if err != nil {
			return err
		} else if done {
			canaryStatus.CurrentStepState = rolloutv1alpha1.CanaryStepStatePaused
		}

	case rolloutv1alpha1.CanaryStepStatePaused:
		done, err := r.doCanaryPaused()
		if err != nil {
			return err
		} else if done {
			canaryStatus.LastUpdateTime = &metav1.Time{Time: time.Now()}
			canaryStatus.CurrentStepState = rolloutv1alpha1.CanaryStepStateReady
		}

	case rolloutv1alpha1.CanaryStepStateReady:
		// run next step
		if len(r.rollout.Spec.Strategy.Canary.Steps) > int(canaryStatus.CurrentStepIndex) {
			canaryStatus.LastUpdateTime = &metav1.Time{Time: time.Now()}
			canaryStatus.CurrentStepIndex++
			canaryStatus.CurrentStepState = rolloutv1alpha1.CanaryStepStateUpgrade
		} else {
			canaryStatus.LastUpdateTime = &metav1.Time{Time: time.Now()}
			canaryStatus.CurrentStepState = rolloutv1alpha1.CanaryStepStateCompleted
		}
		// canary completed
	case rolloutv1alpha1.CanaryStepStateCompleted:
	}

	return nil
}

通过阅读上面代码,我们可以发现它的主要工作就是处理 CanaryStatus 的状态转换。
0. 在 上一段代码里 当Reason处于ProgressingReasonInitializing时,CanaryStatus 会被初始化,方便在这里进行初始化设置。

  1. 初始化 CanaryStatus 的 CurrentStepState,CanaryRevision,RolloutHash
  2. 更新 CanaryStatus 的 CanaryReplicas 和 CanaryReadyReplicas信息
  3. CanaryStatus.CurrentStepState状态转换:
  4. 如果当前这步工作负载升级完成,则 CanaryStepStateUpgrade --> CanaryStepStateTrafficRouting
  5. CanaryStepStateTrafficRouting --> CanaryStepStateMetricsAnalysis --> CanaryStepStatePaused
  6. 如果当前时间在允许允许的时间之前,则CanaryStepStatePaused --> CanaryStepStateReady,如果 duration 为 nil,就需要手动进入CanaryStepStateReady
  7. 如果还有步骤没有执行,则 CanaryStepStateReady --> CanaryStepStateUpgrade ,索引+1
  8. 如果所有步骤已经执行,则 CanaryStepStateReady --> CanaryStepStateCompleted

这里面工作工作负载更新,核心阶段为 CanaryStepStateUpgrade
核心代码为:

func (r *rolloutContext) doCanaryUpgrade() (bool, error) {
	// verify whether batchRelease configuration is the latest
	steps := len(r.rollout.Spec.Strategy.Canary.Steps)
	canaryStatus := r.newStatus.CanaryStatus
	isLatest, err := r.batchControl.Verify(canaryStatus.CurrentStepIndex)
	if err != nil {
		return false, err
	} else if !isLatest {
		return false, nil
	}

	// fetch batchRelease
	batch, err := r.batchControl.FetchBatchRelease()
	if err != nil {
		return false, err
	} else if batch.Status.ObservedReleasePlanHash != util.HashReleasePlanBatches(&batch.Spec.ReleasePlan) ||
		batch.Generation != batch.Status.ObservedGeneration {
		return false, nil
	}
	batchData := util.DumpJSON(batch.Status)	
	cond := util.GetRolloutCondition(*r.newStatus, rolloutv1alpha1.RolloutConditionProgressing)
	cond.Message = fmt.Sprintf("Rollout is in step(%d/%d), and upgrade workload new versions", canaryStatus.CurrentStepIndex, steps)
	r.newStatus.Message = cond.Message
	// promote workload next batch release
	if !batchrelease.IsPromoted(r.rollout, batch, r.workload.IsInRollback) {
		r.recorder.Eventf(r.rollout, corev1.EventTypeNormal, "Progressing", fmt.Sprintf("start upgrade step(%d) canary pods with new versions", canaryStatus.CurrentStepIndex))
		klog.Infof("rollout(%s/%s) will promote batch from(%d) -> to(%d)", r.rollout.Namespace, r.rollout.Name, *batch.Spec.ReleasePlan.BatchPartition+1, canaryStatus.CurrentStepIndex)
		return r.batchControl.Promote(canaryStatus.CurrentStepIndex, r.workload.IsInRollback, false)
	}

	// check whether batchRelease is ready
	if batch.Status.CanaryStatus.CurrentBatchState != rolloutv1alpha1.ReadyBatchState ||
		batch.Status.CanaryStatus.CurrentBatch+1 < canaryStatus.CurrentStepIndex {
		return false, nil
	}
	r.recorder.Eventf(r.rollout, corev1.EventTypeNormal, "Progressing", fmt.Sprintf("upgrade step(%d) canary pods with new versions done", canaryStatus.CurrentStepIndex))
	return true, nil
}

未完待续

6. 处理RolloutPhaseTerminating阶段终止流程
  • 1
    点赞
  • 1
    收藏
    觉得还不错? 一键收藏
  • 0
    评论

“相关推荐”对你有帮助么?

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值