fix(savepoint): missing savepoint path env and not handled savepoint errors

internal/managed_job/run.go

@@ -3,7 +3,8 @@ package managed_job
 import (
 	"errors"
 	"flink-kube-operator/internal/crd/v1alpha1"
-	"flink-kube-operator/internal/jar"
+	"strings"
+	"time"
 
 	"flink-kube-operator/pkg"
 
@@ -11,70 +12,64 @@ import (
 	"go.uber.org/zap"
 )
 
-// upload jar file and set the jarId for later usages
-func (job *ManagedJob) upload() error {
-	jarFile, err := jar.NewJarFile(job.def.Spec.JarURI)
-	if err != nil {
-		pkg.Logger.Debug("[main] error on download jar", zap.Error(err))
+// restore the job from savepoint and jarId in managedJob
+func (job *ManagedJob) restore() error {
+	var savepointPath string
+	if job.def.Status.LastSavepointPath == nil {
+		pkg.Logger.Error("[managed-job] [restore]", zap.Error(v1alpha1.ErrNoSavepointPath))
+		//return v1alpha1.ErrNoSavepointPath
+	} else {
+		savepointPath = *job.def.Status.LastSavepointPath
+	}
+	if job.def.Status.JarId == nil {
+		err := errors.New("missing  jar id")
+		pkg.Logger.Error("[managed-job] [restore]", zap.Error(err))
 		return err
 	}
-	fileName, err := jarFile.Upload(job.client)
-	if err != nil {
-		pkg.Logger.Debug("[main] error on upload jar", zap.Error(err))
-		return err
+	pkg.Logger.Info("[managed-job] [restore] restoring job", zap.String("name", job.def.GetName()), zap.String("savepointPath", savepointPath))
+	var jobId *string
+	for {
+		runJarResp, err := job.client.RunJar(api.RunOpts{
+			JarID:                 *job.def.Status.JarId,
+			AllowNonRestoredState: true,
+			EntryClass:            job.def.Spec.EntryClass,
+			SavepointPath:         savepointPath,
+		})
+		if err != nil {
+			if strings.ContainsAny(err.Error(), ".jar does not exist") {
+				err := job.upload()
+				if err != nil {
+					job.crd.Patch(job.def.UID, map[string]interface{}{
+						"status": map[string]interface{}{
+							"error": "[upload-error] " + err.Error(),
+						},
+					})
+					return nil
+				}
+				continue
+			}
+			pkg.Logger.Error("[managed-job] [restore]", zap.Error(err))
+			return err
+		}
+		jobId = &runJarResp.JobId
+		pkg.Logger.Debug("[main] after run jar", zap.Any("run-jar-resp", runJarResp))
+		break
 	}
-	err = jarFile.Delete()
-	if err != nil {
-		pkg.Logger.Debug("[main] error on delete jar", zap.Error(err))
-	}
-	pkg.Logger.Debug("[main] after upload jar", zap.Any("upload-jar-resp", fileName))
 
-	job.def.Status.JarId = &fileName
+	// job.def.Status.JobId = &runJarResp.JobId
+	// job.def.Status.JobStatus = v1alpha1.JobStatusCreating
+	// job.def.Status.Error = nil
 	job.crd.Patch(job.def.UID, map[string]interface{}{
 		"status": map[string]interface{}{
-			"jarId": job.def.Status.JarId,
+			"jobId":                             jobId,
+			"jobStatus":                         v1alpha1.JobStatusCreating,
+			"lifeCycleStatus":                   v1alpha1.LifeCycleStatusRestoring,
+			"lastRestoredSavepointDate":         job.def.Status.LastSavepointDate,
+			"restoredCount":                     job.def.Status.RestoredCount + 1,
+			"lastRestoredSavepointRestoredDate": time.Now().Format(time.RFC3339),
+			"error":                             nil,
 		},
 	})
+
 	return nil
 }
-
-// run the job from saved jarId in managedJob
-func (job *ManagedJob) run() error {
-	if job.def.Status.JarId == nil {
-		err := errors.New("missing jar id")
-		pkg.Logger.Error("[managed-job] [run]", zap.Error(err))
-		return err
-	}
-	pkg.Logger.Info("[managed-job] [run] starting job", zap.String("name", job.def.GetName()))
-	runJarResp, err := job.client.RunJar(api.RunOpts{
-		JarID:                 *job.def.Status.JarId,
-		AllowNonRestoredState: true,
-		EntryClass:            job.def.Spec.EntryClass,
-	})
-	if err != nil {
-		pkg.Logger.Error("[managed-job] [run]", zap.Error(err))
-		return err
-	}
-	pkg.Logger.Debug("[main] after run jar", zap.Any("run-jar-resp", runJarResp))
-
-	// if job.state == nil {
-	// 	job.state = &jobState{}
-	// }
-	// job.state.JobId = &runJarResp.JobId
-	// job.state.Status = v1alpha1.JobStatusCreating
-	// job.updateState(*job.state)
-	// job.crd.SetJobStatus(job.def.UID, v1alpha1.FlinkJobStatus{
-	// 	JobId: job.state.JobId,
-	// })
-	job.crd.Patch(job.def.UID, map[string]interface{}{
-		"status": map[string]interface{}{
-			"jobId":           &runJarResp.JobId,
-			"jobStatus":       v1alpha1.JobStatusCreating,
-			"lifeCycleStatus": v1alpha1.LifeCycleStatusInitializing,
-			"error":           nil,
-		},
-	})
-	//job.updateState(jobState{JobId: &runJarResp.JobId, Status: JobStatusCreating})
-
-	return err
-}