Merge pull request #406 from jaypume/feature-knowledge-base

Support showing knowledge base

build/crds/sedna.io_lifelonglearningjobs.yaml

@@ -21187,6 +21187,56 @@ spec:
                   - type
                   type: object
                 type: array
+              knowledgeBase:
+                properties:
+                  AIClasses:
+                    properties:
+                      listOfAIClasses:
+                        items:
+                          type: string
+                        type: array
+                      numberOfAIClasses:
+                        type: integer
+                    required:
+                    - numberOfAIClasses
+                    type: object
+                  AIModels:
+                    properties:
+                      listOfAIModels:
+                        items:
+                          properties:
+                            averagePrecision:
+                              type: number
+                            modelID:
+                              type: string
+                            numberOfTrainSamples:
+                              type: integer
+                          required:
+                          - averagePrecision
+                          - modelID
+                          - numberOfTrainSamples
+                          type: object
+                        type: array
+                      numberOfAIModels:
+                        type: integer
+                    required:
+                    - numberOfAIModels
+                    type: object
+                  samples:
+                    properties:
+                      numberOfLabeledUnseenSample:
+                        type: integer
+                      numberOfUnseenSample:
+                        type: integer
+                    required:
+                    - numberOfLabeledUnseenSample
+                    - numberOfUnseenSample
+                    type: object
+                required:
+                - AIClasses
+                - AIModels
+                - samples
+                type: object
               startTime:
                 description: Represents time when the job was acknowledged by the
                   job controller. It is not guaranteed to be set in happens-before

pkg/apis/sedna/v1alpha1/lifelonglearningjob_types.go

@@ -86,9 +86,47 @@ type LLDeploySpec struct {
 
 // LifelongLearningJobList is a list of LifelongLearningJobs.
 type LifelongLearningJobList struct {
-	metav1.TypeMeta `json:",inline"`
-	metav1.ListMeta `json:"metadata"`
-	Items           []LifelongLearningJob `json:"items"`
+	metav1.TypeMeta  `json:",inline"`
+	metav1.ListMeta  `json:"metadata"`
+	Items            []LifelongLearningJob `json:"items"`
+	OverallKBStatus  OverallStatus         `json:"overallStatus"`
+	OverallKBDetails KnowledgeBase         `json:"overallKBStatus"`
+}
+
+type OverallStatus struct {
+	NumberOfUnseenSample    int     `json:"numberOfUnseenSample"`
+	NumberOfNewUnseenSample int     `json:"numberOfNewUnseenSample"`
+	NumberOfAIModels        int     `json:"numberOfAIModels"`
+	PercentOfNewAIModels    float32 `json:"percentOfNewAIModels"`
+	NumberOfClasses         int     `json:"numberOfClasses"`
+	PercentOfNewClasses     float32 `json:"percentOfNewClasses"`
+}
+
+type KnowledgeBase struct {
+	AIModels  AIModels  `json:"AIModels"`
+	AIClasses AIClasses `json:"AIClasses"`
+	Samples   Samples   `json:"samples"`
+}
+
+type AIModels struct {
+	NumberOfAIModels int       `json:"numberOfAIModels"`
+	ListOfAIModels   []AIModel `json:"listOfAIModels,omitempty"`
+}
+
+type AIModel struct {
+	ModelID              string  `json:"modelID"`
+	AveragePrecision     float32 `json:"averagePrecision"`
+	NumberOfTrainSamples int     `json:"numberOfTrainSamples"`
+}
+
+type AIClasses struct {
+	NumberOfAIClasses int      `json:"numberOfAIClasses"`
+	ListOfAIClasses   []string `json:"listOfAIClasses,omitempty"`
+}
+
+type Samples struct {
+	NumberOfLabeledUnseenSample int `json:"numberOfLabeledUnseenSample"`
+	NumberOfUnseenSample        int `json:"numberOfUnseenSample"`
 }
 
 // LLJobStatus represents the current state of a lifelonglearning job
@@ -97,6 +135,8 @@ type LLJobStatus struct {
 	// +optional
 	Conditions []LLJobCondition `json:"conditions,omitempty"`
 
+	KnowledgeBase KnowledgeBase `json:"knowledgeBase,omitempty"`
+
 	// Represents time when the job was acknowledged by the job controller.
 	// It is not guaranteed to be set in happens-before order across separate operations.
 	// It is represented in RFC3339 form and is in UTC.

pkg/globalmanager/controllers/lifelonglearning/downstream.go

@@ -31,6 +31,7 @@ import (
 func (c *Controller) syncToEdge(eventType watch.EventType, obj interface{}) error {
 	job, ok := obj.(*sednav1.LifelongLearningJob)
 	if !ok {
+		klog.V(4).Infof("get job %s failed, stop to sync to edge", job.Name)
 		return nil
 	}
 
@@ -54,6 +55,7 @@ func (c *Controller) syncToEdge(eventType watch.EventType, obj interface{}) erro
 	var deployNodeName string
 
 	getAnnotationsNodeName := func(nodeName sednav1.LLJobStage) string {
+		klog.V(4).Infof("getAnnotationsNodeName return %s", job.Name)
 		return runtime.AnnotationsKeyPrefix + string(nodeName)
 	}
 	ann := job.GetAnnotations()
@@ -88,6 +90,7 @@ func (c *Controller) syncToEdge(eventType watch.EventType, obj interface{}) erro
 	jobStage := latestCondition.Stage
 
 	syncJobWithNodeName := func(nodeName string) {
+		klog.V(4).Infof("syncJobWithNodeName nodeName is %s, job name is %s ", nodeName, job.Name)
 		if err := c.sendToEdgeFunc(nodeName, eventType, job); err != nil {
 			klog.Warningf("Error to sync lifelong learning job %s to node %s in stage %s: %v",
 				job.Name, nodeName, jobStage, err)
@@ -108,10 +111,13 @@ func (c *Controller) syncToEdge(eventType watch.EventType, obj interface{}) erro
 	doJobStageEvent := func(nodeName string) {
 		switch currentType {
 		case sednav1.LLJobStageCondWaiting:
+			klog.V(4).Infof("LLJobStageCondWaiting, dataset nodeName is %s", nodeName)
 			syncJobWithNodeName(dsNodeName)
 		case sednav1.LLJobStageCondRunning:
+			klog.V(4).Infof("LLJobStageCondRunning, nodeName is %s", nodeName)
 			syncJobWithNodeName(nodeName)
 		case sednav1.LLJobStageCondCompleted, sednav1.LLJobStageCondFailed:
+			klog.V(4).Infof("LLJobStageCondCompleted, nodeName is %s", nodeName)
 			if !isJobResidentNode(nodeName) {
 				// delete LC's job from nodeName that's different from dataset node when worker's status
 				// is completed or failed.

pkg/globalmanager/controllers/lifelonglearning/lifelonglearningjob.go

@@ -283,6 +283,7 @@ func (c *Controller) sync(key string) (bool, error) {
 
 	if needUpdated {
 		if err := c.updateJobStatus(&job); err != nil {
+			klog.V(4).Infof("the job needUpdated, err is %s", err)
 			return forget, err
 		}
 
@@ -293,13 +294,13 @@ func (c *Controller) sync(key string) (bool, error) {
 
 		forget = true
 	}
-
 	return forget, err
 }
 
 // setWorkerNodeNameOfJob sets the worker nodeName of the specified job
 // which is used for downstream to sync job info to the specified LC located in nodeName.
 func (c *Controller) setWorkerNodeNameOfJob(job *sednav1.LifelongLearningJob, jobStage string, nodeName string) error {
+	klog.V(4).Infof("setWorkerNodeNameOfJob job name is %s, job Stage is %s, nodeName is %s", job.Name, jobStage, nodeName)
 	key := runtime.AnnotationsKeyPrefix + jobStage
 
 	return c.addJobAnnotations(job, key, nodeName)
@@ -368,6 +369,8 @@ func (c *Controller) transitJobState(job *sednav1.LifelongLearningJob) (bool, er
 	currentType := latestCondition.Type
 	newConditionType = currentType
 
+	//klog.Infof("==== stage is %s, type is %s, name is %s", jobStage, currentType, job.Name)
+
 	switch currentType {
 	case initialType:
 		newConditionType = sednav1.LLJobStageCondWaiting
@@ -780,8 +783,8 @@ func (c *Controller) createInferPod(job *sednav1.LifelongLearningJob) error {
 		"NAMESPACE":   job.Namespace,
 		"JOB_NAME":    job.Name,
 		"WORKER_NAME": "inferworker-" + utilrand.String(5),
-
-		"LC_SERVER": c.cfg.LC.Server,
+		"LC_SERVER":   c.cfg.LC.Server,
+		"OUTPUT_URL":  job.Spec.OutputDir,
 	}
 
 	workerParam.WorkerType = runtime.InferencePodType