change GetDisk error reporting to temporary in CreateVolume codepath

Author: leiyiz

pkg/gce-cloud-provider/compute/gce-compute.go

@@ -560,7 +560,7 @@ func (cloud *CloudProvider) insertRegionalDisk(
 		if IsGCEError(err, "alreadyExists") {
 			disk, err := cloud.GetDisk(ctx, project, volKey, gceAPIVersion)
 			if err != nil {
-				return err
+				return status.Error(codes.Unavailable, err.Error())
 			}
 			err = cloud.ValidateExistingDisk(ctx, disk, params,
 				int64(capacityRange.GetRequiredBytes()),
@@ -572,16 +572,18 @@ func (cloud *CloudProvider) insertRegionalDisk(
 			klog.Warningf("GCE PD %s already exists, reusing", volKey.Name)
 			return nil
 		}
-		return status.Error(codes.Internal, fmt.Sprintf("unknown Insert disk error: %v", err.Error()))
+		return fmt.Errorf("unknown Insert Regional disk error: %w", err)
 	}
 	klog.V(5).Infof("InsertDisk operation %s for disk %s", opName, diskToCreate.Name)
 
 	err = cloud.waitForRegionalOp(ctx, project, opName, volKey.Region)
+	// failed to wait for Op to finish, however, the Op possibly is still running as expected
+	// the error code returned should be non-final
 	if err != nil {
 		if IsGCEError(err, "alreadyExists") {
 			disk, err := cloud.GetDisk(ctx, project, volKey, gceAPIVersion)
 			if err != nil {
-				return err
+				return status.Errorf(codes.Unavailable, "error when getting disk: %v", err.Error())
 			}
 			err = cloud.ValidateExistingDisk(ctx, disk, params,
 				int64(capacityRange.GetRequiredBytes()),
@@ -593,7 +595,7 @@ func (cloud *CloudProvider) insertRegionalDisk(
 			klog.Warningf("GCE PD %s already exists after wait, reusing", volKey.Name)
 			return nil
 		}
-		return fmt.Errorf("unknown Insert disk operation error: %w", err)
+		return status.Errorf(codes.Unavailable, "unknown error when polling the operation: %v", err.Error())
 	}
 	return nil
 }
@@ -695,7 +697,7 @@ func (cloud *CloudProvider) insertZonalDisk(
 		if IsGCEError(err, "alreadyExists") {
 			disk, err := cloud.GetDisk(ctx, project, volKey, gceAPIVersion)
 			if err != nil {
-				return err
+				return status.Error(codes.Unavailable, err.Error())
 			}
 			err = cloud.ValidateExistingDisk(ctx, disk, params,
 				int64(capacityRange.GetRequiredBytes()),
@@ -714,10 +716,12 @@ func (cloud *CloudProvider) insertZonalDisk(
 	err = cloud.waitForZonalOp(ctx, project, opName, volKey.Zone)
 
 	if err != nil {
+		// failed to wait for Op to finish, however, the Op possibly is still running as expected
+		// the error code returned should be non-final
 		if IsGCEError(err, "alreadyExists") {
 			disk, err := cloud.GetDisk(ctx, project, volKey, gceAPIVersion)
 			if err != nil {
-				return err
+				return status.Errorf(codes.Unavailable, "error when getting disk: %v", err.Error())
 			}
 			err = cloud.ValidateExistingDisk(ctx, disk, params,
 				int64(capacityRange.GetRequiredBytes()),
@@ -729,7 +733,7 @@ func (cloud *CloudProvider) insertZonalDisk(
 			klog.Warningf("GCE PD %s already exists after wait, reusing", volKey.Name)
 			return nil
 		}
-		return fmt.Errorf("unknown Insert disk operation error: %w", err)
+		return status.Errorf(codes.Unavailable, "unknown error when polling the operation: %v", err.Error())
 	}
 	return nil
 }

pkg/gce-pd-csi-driver/controller.go

@@ -324,7 +324,7 @@ func (gceCS *GCEControllerServer) CreateVolume(ctx context.Context, req *csi.Cre
 	existingDisk, err := gceCS.CloudProvider.GetDisk(ctx, gceCS.CloudProvider.GetDefaultProject(), volKey, gceAPIVersion)
 	if err != nil {
 		if !gce.IsGCEError(err, "notFound") {
-			return nil, common.LoggedError("CreateVolume, failed to getDisk when validating: ", err)
+			return nil, common.LoggedError("CreateVolume, failed to getDisk when validating: ", status.Error(codes.Unavailable, err.Error()))
 		}
 	}
 	if err == nil {
@@ -1878,7 +1878,7 @@ func createRegionalDisk(ctx context.Context, cloudProvider gce.GCECompute, name
 	}
 	disk, err := cloudProvider.GetDisk(ctx, project, meta.RegionalKey(name, region), gceAPIVersion)
 	if err != nil {
-		return nil, fmt.Errorf("failed to get disk after creating regional disk: %w", err)
+		return nil, status.Errorf(codes.Unavailable, "failed to get disk after creating regional disk: %v", err.Error())
 	}
 	return disk, nil
 }
@@ -1900,7 +1900,7 @@ func createSingleZoneDisk(ctx context.Context, cloudProvider gce.GCECompute, nam
 	}
 	disk, err := cloudProvider.GetDisk(ctx, project, meta.ZonalKey(name, diskZone), gceAPIVersion)
 	if err != nil {
-		return nil, err
+		return nil, status.Errorf(codes.Unavailable, "failed to get disk after creating zonal disk: %v", err.Error())
 	}
 	return disk, nil
 }