LabelBot NeedSync needs to check if model is being trained.

* NeedsSync needs to check whether there is a model being trained or
  if there is a dataset being imported. Otherwise we end up launching
  multiple overlapping jobs because it takes a long time for the model
  to train. During which time the Tekton job will have finished.

* Related to kubeflow#178

Label_Microservice/auto-update/base/kustomization.yaml

@@ -4,6 +4,6 @@ resources:
 - deployment.yaml
 - service-account.yaml
 images:
-- digest: sha256:32ecb5f829796b37fb5b5ddc2c3d98c693a81728b285fa90756f2b4b8f1075ae
+- digest: sha256:2b7e759153baacf59514c98e50cb01eec0fc7aa2d7e2b271ac0a3c5cc516a74d
   name: gcr.io/issue-label-bot-dev/labelbot-diff
-  newName: gcr.io/issue-label-bot-dev/labelbot-diff:874d2f5-dirty
+  newName: gcr.io/issue-label-bot-dev/labelbot-diff:3697618-dirty

Label_Microservice/deployment/README.md

@@ -22,20 +22,22 @@ Deploying it
    skaffold build
    ```
 
-1. Edit the image
+1. Update the image
 
    ```
-   cd deployment/overlays/prod
-   kustomize edit set image gcr.io/issue-label-bot-dev/bot-worker=gcr.io/issue-label-bot-dev/bot-worker:${TAG}@${SHA}
+   cd ..
+   make update-diff-image
    ```
 
-1. Create the deployment
+1. Hydrate the GitOps manifests
 
    ```
-   cd Label_Microservice/deployment/overlays/prod
-   kustomize build | kubectl apply -f -
+   cd ..
+   make hydrate-prod
    ```
 
+1. Commit and push the manifests
+
 ## Staging/Dev
 
 There is a staging/dev instance running in a different namespace

Label_Microservice/go/cmd/automl/main.go

@@ -28,6 +28,9 @@ func init() {
 	rootCmd.AddCommand(serveCmd)
 	rootCmd.AddCommand(getCmd)
 	rootCmd.AddCommand(labelCmd)
+	rootCmd.AddCommand(isTrainCmd)
+
+	rootCmd.PersistentFlags().IntVarP(&Verbose, "verbose", "v", int(log.InfoLevel), "verbose output")
 
 	labelCmd.AddCommand(labelModelCmd)
 
@@ -46,6 +49,11 @@ func init() {
 	getCmd.Flags().StringVarP(&getOptions.location, "location", "", "us-central1", "Location to search for models")
 	getCmd.Flags().StringVarP(&getOptions.outputFile, "output", "", "", "(Optional) If supplied write the evaluation scores to this file in csv format.")
 	getCmd.MarkFlagRequired("model")
+
+
+	isTrainCmd.Flags().StringVarP(&getOptions.name, "name", "", "", "The model to get.")
+	isTrainCmd.Flags().StringVarP(&getOptions.project, "project", "", "issue-label-bot-dev", "Project to check")
+	isTrainCmd.Flags().StringVarP(&getOptions.location, "location", "", "us-central1", "Location to search for models")
 }
 
 type cliOptions struct {
@@ -66,6 +74,8 @@ type getCmdOptions struct {
 }
 
 var (
+	Verbose  int
+
 	options    = cliOptions{}
 	getOptions = getCmdOptions{}
 	rootCmd    = &cobra.Command{
@@ -78,6 +88,7 @@ var (
 		Short: "Start webserver.",
 		Long:  `starts the controller`,
 		Run: func(cmd *cobra.Command, args []string) {
+			log.SetLevel(log.Level(Verbose))
 			router := mux.NewRouter().StrictSlash(true)
 
 			interval, err := time.ParseDuration(options.retrainInterval)
@@ -110,7 +121,7 @@ var (
 		Short: "Get the specified model.",
 		Long:  `Get the specified model`,
 		Run: func(cmd *cobra.Command, args []string) {
-
+			log.SetLevel(log.Level(Verbose))
 			name := fmt.Sprintf("projects/%v/locations/%v/models/%v", getOptions.project, getOptions.location, getOptions.name)
 			model, err := automl.GetModel(name)
 
@@ -131,7 +142,7 @@ var (
 			e, err := yaml.Marshal(evaluation)
 
 			if err != nil {
-				log.Fatalf("Error marshiling the evaluation to yaml %v; error: %v", getOptions.name, err)
+				log.Fatalf("Error marshaling the evaluation to yaml %v; error: %v", getOptions.name, err)
 			}
 
 			fmt.Printf(string(e) + "\n")
@@ -148,6 +159,7 @@ var (
 		Short: "Label the specified model.",
 		Long:  `Label the specified model`,
 		Run: func(cmd *cobra.Command, args []string) {
+			log.SetLevel(log.Level(Verbose))
 			if len(args) < 2 {
 				log.Fatalf("Error usage is label models <model> label1=value1 label2=value2")
 			}
@@ -174,6 +186,22 @@ var (
 
 		},
 	}
+
+	isTrainCmd = &cobra.Command{
+		Use:   "isTraining",
+		Short: "Check if a model is being trained.",
+		Run: func(cmd *cobra.Command, args []string) {
+			log.SetLevel(log.Level(Verbose))
+			isTraining, err := automl.IsTraining(getOptions.project, getOptions.location)
+
+			if err != nil {
+				log.Fatalf("Error checking if model %v is being trained; error: %v", getOptions.name, err)
+			}
+
+			fmt.Printf("is training: %v", isTraining)
+		},
+	}
+
 )
 
 func main() {

Label_Microservice/go/cmd/automl/pkg/automl/automl.go

@@ -6,11 +6,13 @@ import (
 	"context"
 	"encoding/csv"
 	"fmt"
+	"github.com/go-yaml/yaml"
 	"github.com/golang/protobuf/proto"
 	"github.com/pkg/errors"
 	log "github.com/sirupsen/logrus"
 	"google.golang.org/api/iterator"
 	automlpb "google.golang.org/genproto/googleapis/cloud/automl/v1"
+	lropb "google.golang.org/genproto/googleapis/longrunning"
 	"google.golang.org/genproto/protobuf/field_mask"
 	"os"
 )
@@ -102,9 +104,6 @@ func GetLatestDeployed(projectID string, location string, modelName string) (*au
 // TODO(jlewi): We should really filter on labels; they don't appear to show up in the UI but they
 // are in the API: https://cloud.google.com/automl/docs/reference/rest/v1/projects.locations.models#Model
 func GetLatestTrained(projectID string, location string, modelName string) (*automlpb.Model, error) {
-	// projectID := "my-project-id"
-	// location := "us-central1"
-
 	ctx := context.Background()
 	client, err := automl.NewClient(ctx)
 	if err != nil {
@@ -193,11 +192,6 @@ func GetModelEvaluation(name string, outputFile string) (*automlpb.ModelEvaluati
 		//log.Infof("Evaluation: %+v, %+v", e, m.GetAuRoc())
 	}
 
-	//
-	//if err != nil {
-	//	return nil, err
-	//}
-
 	return nil, nil
 }
 
@@ -240,3 +234,92 @@ func LabelModel(name string, labels map[string]string) error {
 
 	return err
 }
+
+func toString(v interface{}) string {
+	output, err := yaml.Marshal(v)
+
+	if err != nil {
+		return err.Error()
+	}
+
+	return string(output)
+
+}
+// IsTraining determines if there is an import job or model training currently in progress
+//
+// TODO(jlewi): We should really only look at models and datasets matching some labels or other criterion.
+// Its not clear how to link LRO's for import jobs to the corresponding dataset. So right now
+// we just check if there are any operations still in progress.
+func IsTraining(projectID string, location string) (bool, error) {
+	ctx := context.Background()
+	client, err := automl.NewClient(ctx)
+	if err != nil {
+		return false, fmt.Errorf("NewClient: %v", err)
+	}
+	defer client.Close()
+
+	req := &lropb.ListOperationsRequest{
+		Name: fmt.Sprintf("projects/%s/locations/%s", projectID, location),
+	}
+
+	it := client.LROClient.ListOperations(ctx, req)
+
+	for {
+		op, err := it.Next()
+		if err == iterator.Done {
+			break
+		}
+		if err != nil {
+			return false, fmt.Errorf("ListOperations.Next: %v", err)
+		}
+
+		if op.GetDone() {
+			continue
+		}
+
+		// Since the OP is still running check if its an ImportJob or model training.
+		opMeta := &automlpb.OperationMetadata{}
+
+		err = op.GetMetadata().UnmarshalTo(opMeta)
+		if err != nil {
+			log.Errorf("Could not unmarshal metadata for: %v; error: %v", op.GetName(), err)
+		}
+
+		// Creating a dataset is not actually a long running operation; when we create a dataset we are just
+		// defining the name essentionally.
+		if opMeta.GetCreateDatasetDetails() != nil {
+			details := opMeta.GetCreateDatasetDetails()
+
+			if details != nil {
+				log.Infof("Found running op dataset pp: %v\n%v",op.Name, toString(details))
+			}
+			return true, nil
+		}
+
+		// ImportData is a long running operation can take hours
+		// TODO(jlewi): How do we associate a running ImportData with its dataset? Neither Metadata, nor details appears
+		// to contain the dataset id.
+		if opMeta.GetImportDataDetails() != nil {
+			log.Infof("ImportData Op: %v; metadata:\n%v", op.Name, toString(opMeta))
+			details := opMeta.GetImportDataDetails()
+
+			if details != nil {
+				log.Infof("ImportDataDetails Op: %v; Details:\n%v",op.Name, toString(details))
+			}
+
+			return true, nil
+		}
+
+		if opMeta.GetCreateModelDetails() != nil {
+			details := opMeta.GetCreateModelDetails()
+
+			if details != nil {
+				log.Infof("Create Model Op: %v; Details:\n%v",op.Name, toString(details))
+			}
+
+			return true, nil
+		}
+	}
+
+	return false, nil
+}

Label_Microservice/go/cmd/automl/pkg/server/server.go

@@ -60,6 +60,18 @@ func (s *Server) NeedsSync(w http.ResponseWriter, r *http.Request) {
 	}
 
 	getErr := func() error {
+		isTraining, err := automl.IsTraining(s.Project, s.Location)
+
+		if err != nil {
+			appendError(response, fmt.Sprintf("Error checking if model is being trained; %v", err))
+			return err
+		}
+
+		if isTraining {
+			log.Infof("There is model currently being trained; no sync needed.")
+			return nil
+		}
+
 		latest, err := automl.GetLatestDeployed(s.Project, s.Location, s.Name)
 
 		if err != nil {

Label_Microservice/labelbot-diff.image.json

@@ -1 +1 @@
-{"builds":[{"imageName":"gcr.io/issue-label-bot-dev/labelbot-diff","tag":"gcr.io/issue-label-bot-dev/labelbot-diff:874d2f5-dirty@sha256:32ecb5f829796b37fb5b5ddc2c3d98c693a81728b285fa90756f2b4b8f1075ae"}]}
+{"builds":[{"imageName":"gcr.io/issue-label-bot-dev/labelbot-diff","tag":"gcr.io/issue-label-bot-dev/labelbot-diff:3697618-dirty@sha256:2b7e759153baacf59514c98e50cb01eec0fc7aa2d7e2b271ac0a3c5cc516a74d"}]}

acm-repos/issue-label-bot/namespaces/label-bot-prod/apps_v1_deployment_labelbot-diff.yaml

@@ -22,7 +22,7 @@ spec:
         - --kptFile=/src/code-intelligence.git/Label_Microservice/deployment/Kptfile
         - --port=8080
         - --retrainInterval=24h
-        image: gcr.io/issue-label-bot-dev/labelbot-diff:874d2f5-dirty@sha256:32ecb5f829796b37fb5b5ddc2c3d98c693a81728b285fa90756f2b4b8f1075ae
+        image: gcr.io/issue-label-bot-dev/labelbot-diff:3697618-dirty@sha256:2b7e759153baacf59514c98e50cb01eec0fc7aa2d7e2b271ac0a3c5cc516a74d
         name: diff
         ports:
         - containerPort: 8080

acm-repos/issue-label-bot/namespaces/label-bot-prod/automl.cloudai.kubeflow.org_v1alpha1_modelsync_labelbot-train.yaml

@@ -0,0 +1,30 @@
+apiVersion: automl.cloudai.kubeflow.org/v1alpha1
+kind: ModelSync
+metadata:
+  name: labelbot-train
+  namespace: label-bot-prod
+spec:
+  failedPipelineRunsHistoryLimit: 10
+  needsSyncUrl: http://labelbot-diff.label-bot-prod/needsTrain
+  pipelineRunTemplate:
+    spec:
+      params:
+      - name: notebook-path
+        value: Label_Microservice/notebooks/automl.ipynb
+      - name: requirements
+        value: Label_Microservice/requirements.train.txt
+      - name: output
+        value: gs://issue-label-bot-dev_public/label-bot/training/runs
+      pipelineRef:
+        name: run-notebook
+      resources:
+      - name: notebook-repo
+        resourceSpec:
+          params:
+          - name: url
+            value: https://github.com/kubeflow/code-intelligence.git
+          - name: revision
+            value: master
+          type: git
+      serviceAccountName: default-editor
+  successfulPipelineRunsHistoryLimit: 10