Add 'updateStrategy' field to RayServiceSpec, allowing users to disab…

…le/enable zero-downtime update.

docs/reference/api.md

@@ -207,12 +207,26 @@ _Appears in:_
 | `serviceUnhealthySecondThreshold` _integer_ | Deprecated: This field is not used anymore. ref: https://github.com/ray-project/kuberay/issues/1685 |  |  |
 | `deploymentUnhealthySecondThreshold` _integer_ | Deprecated: This field is not used anymore. ref: https://github.com/ray-project/kuberay/issues/1685 |  |  |
 | `serveService` _[Service](https://kubernetes.io/docs/reference/generated/kubernetes-api/v1.28/#service-v1-core)_ | ServeService is the Kubernetes service for head node and worker nodes who have healthy http proxy to serve traffics. |  |  |
+| `upgradeStrategy` _[RayServiceUpgradeStrategy](#rayserviceupgradestrategy)_ | UpgradeStrategy represents the strategy used when upgrading the RayService. Currently supports `NewCluster` and `None` | NewCluster |  |
 | `serveConfigV2` _string_ | Important: Run "make" to regenerate code after modifying this file<br />Defines the applications and deployments to deploy, should be a YAML multi-line scalar string. |  |  |
 | `rayClusterConfig` _[RayClusterSpec](#rayclusterspec)_ |  |  |  |
 
 
 
 
+#### RayServiceUpgradeStrategy
+
+_Underlying type:_ _string_
+
+
+
+
+
+_Appears in:_
+- [RayServiceSpec](#rayservicespec)
+
+
+
 #### ScaleStrategy
 
 

ray-operator/apis/ray/v1/rayservice_types.go

@@ -20,6 +20,15 @@ const (
 	FailedToUpdateService            ServiceStatus = "FailedToUpdateService"
 )
 
+type RayServiceUpgradeStrategy string
+
+const (
+	// During upgrade, NewCluster strategy will create new upgraded cluster and switch to it when it becomes ready
+	NewCluster RayServiceUpgradeStrategy = "NewCluster"
+	// Having None as the strategy will mean down-time while it wait for new cluster to be ready when performing upgrade
+	None RayServiceUpgradeStrategy = "None"
+)
+
 // These statuses should match Ray Serve's application statuses
 // See `enum ApplicationStatus` in https://sourcegraph.com/github.com/ray-project/ray/-/blob/src/ray/protobuf/serve.proto for more details.
 var ApplicationStatusEnum = struct {
@@ -57,6 +66,9 @@ type RayServiceSpec struct {
 	DeploymentUnhealthySecondThreshold *int32 `json:"deploymentUnhealthySecondThreshold,omitempty"`
 	// ServeService is the Kubernetes service for head node and worker nodes who have healthy http proxy to serve traffics.
 	ServeService *corev1.Service `json:"serveService,omitempty"`
+	// UpgradeStrategy represents the strategy used when upgrading the RayService. Currently supports `NewCluster` and `None`
+	// +kubebuilder:default:=NewCluster
+	UpgradeStrategy RayServiceUpgradeStrategy `json:"upgradeStrategy,omitempty"`
 	// Important: Run "make" to regenerate code after modifying this file
 	// Defines the applications and deployments to deploy, should be a YAML multi-line scalar string.
 	ServeConfigV2  string         `json:"serveConfigV2,omitempty"`

ray-operator/controllers/ray/rayservice_controller.go

@@ -248,6 +248,12 @@ func validateRayServiceSpec(rayService *rayv1.RayService) error {
 	if headSvc := rayService.Spec.RayClusterSpec.HeadGroupSpec.HeadService; headSvc != nil && headSvc.Name != "" {
 		return fmt.Errorf("spec.rayClusterConfig.headGroupSpec.headService.metadata.name should not be set")
 	}
+
+	if upgradeStrategy := rayService.Spec.UpgradeStrategy; upgradeStrategy != "" {
+		if upgradeStrategy != rayv1.NewCluster && upgradeStrategy != rayv1.None {
+			return fmt.Errorf("spec.UpgradeStrategy is invalid, valid options are %s and %s", rayv1.NewCluster, rayv1.None)
+		}
+	}
 	return nil
 }
 
@@ -424,10 +430,19 @@ func (r *RayServiceReconciler) reconcileRayCluster(ctx context.Context, rayServi
 	if clusterAction == RolloutNew {
 		// For LLM serving, some users might not have sufficient GPU resources to run two RayClusters simultaneously.
 		// Therefore, KubeRay offers ENABLE_ZERO_DOWNTIME as a feature flag for zero-downtime upgrades.
+		zeroDowntimeEnvVar := os.Getenv(ENABLE_ZERO_DOWNTIME)
+		rayServiceSpecUpgradeStrategy := rayServiceInstance.Spec.UpgradeStrategy
+		// There are two ways to enable zero downtime upgrade. Through ENABLE_ZERO_DOWNTIME env var or setting Spec.UpgradeStrategy.
+		// If no fields are set, zero downtime upgrade by default is enabled.
+		// Spec.UpgradeStrategy takes precedence over ENABLE_ZERO_DOWNTIME.
 		enableZeroDowntime := true
-		if s := os.Getenv(ENABLE_ZERO_DOWNTIME); strings.ToLower(s) == "false" {
-			enableZeroDowntime = false
+		if zeroDowntimeEnvVar != "" {
+			enableZeroDowntime = strings.ToLower(zeroDowntimeEnvVar) != "false"
 		}
+		if rayServiceSpecUpgradeStrategy != "" {
+			enableZeroDowntime = rayServiceSpecUpgradeStrategy == rayv1.NewCluster
+		}
+
 		if enableZeroDowntime || !enableZeroDowntime && activeRayCluster == nil {
 			// Add a pending cluster name. In the next reconcile loop, shouldPrepareNewRayCluster will return DoNothing and we will
 			// actually create the pending RayCluster instance.

ray-operator/controllers/ray/rayservice_controller_unit_test.go

@@ -44,6 +44,13 @@ func TestValidateRayServiceSpec(t *testing.T) {
 		Spec: rayv1.RayServiceSpec{},
 	})
 	assert.NoError(t, err, "The RayService spec is valid.")
+
+	err = validateRayServiceSpec(&rayv1.RayService{
+		Spec: rayv1.RayServiceSpec{
+			UpgradeStrategy: "invalidStrategy",
+		},
+	})
+	assert.Error(t, err, "spec.UpgradeStrategy is invalid")
 }
 
 func TestGenerateHashWithoutReplicasAndWorkersToDelete(t *testing.T) {
@@ -762,12 +769,13 @@ func TestReconcileRayCluster(t *testing.T) {
 	}
 
 	tests := map[string]struct {
-		activeCluster           *rayv1.RayCluster
-		kubeRayVersion          string
-		updateRayClusterSpec    bool
-		enableZeroDowntime      bool
-		shouldPrepareNewCluster bool
-		updateKubeRayVersion    bool
+		activeCluster             *rayv1.RayCluster
+		rayServiceUpgradeStrategy rayv1.RayServiceUpgradeStrategy
+		kubeRayVersion            string
+		updateRayClusterSpec      bool
+		enableZeroDowntime        bool
+		shouldPrepareNewCluster   bool
+		updateKubeRayVersion      bool
 	}{
 		// Test 1: Neither active nor pending clusters exist. The `markRestart` function will be called, so the `PendingServiceStatus.RayClusterName` should be set.
 		"Zero-downtime upgrade is enabled. Neither active nor pending clusters exist.": {
@@ -790,8 +798,8 @@ func TestReconcileRayCluster(t *testing.T) {
 			enableZeroDowntime:      true,
 			shouldPrepareNewCluster: true,
 		},
-		// Test 4: The active cluster exists. Trigger the zero-downtime upgrade.
-		"Zero-downtime upgrade is disabled. The active cluster exists. Trigger the zero-downtime upgrade.": {
+		// Test 4: The active cluster exists. Zero-downtime upgrade is false, should not trigger zero-downtime upgrade.
+		"Zero-downtime upgrade is disabled. The active cluster exists. Does not trigger the zero-downtime upgrade.": {
 			activeCluster:           activeCluster.DeepCopy(),
 			updateRayClusterSpec:    true,
 			enableZeroDowntime:      false,
@@ -815,6 +823,37 @@ func TestReconcileRayCluster(t *testing.T) {
 			updateKubeRayVersion:    true,
 			kubeRayVersion:          "new-version",
 		},
+		// Test 7: Zero downtime upgrade is enabled, but is enabled through the RayServiceSpec
+		"Zero-downtime upgrade enabled. The active cluster exist. Zero-downtime upgrade is triggered through RayServiceSpec.": {
+			activeCluster:             activeCluster.DeepCopy(),
+			updateRayClusterSpec:      true,
+			enableZeroDowntime:        true,
+			shouldPrepareNewCluster:   true,
+			rayServiceUpgradeStrategy: rayv1.NewCluster,
+		},
+		// Test 8: Zero downtime upgrade is enabled. Env var is set to false but RayServiceSpec is set to NewCluster. Trigger the zero-downtime upgrade.
+		"Zero-downtime upgrade is enabled through RayServiceSpec and not through env var. Active cluster exist. Trigger the zero-downtime upgrade.": {
+			activeCluster:             activeCluster.DeepCopy(),
+			updateRayClusterSpec:      true,
+			enableZeroDowntime:        false,
+			shouldPrepareNewCluster:   true,
+			rayServiceUpgradeStrategy: rayv1.NewCluster,
+		},
+		// Test 9: Zero downtime upgrade is disabled. Env var is set to true but RayServiceSpec is set to None.
+		"Zero-downtime upgrade is disabled. Env var is set to true but RayServiceSpec is set to None.": {
+			activeCluster:             activeCluster.DeepCopy(),
+			updateRayClusterSpec:      true,
+			enableZeroDowntime:        true,
+			shouldPrepareNewCluster:   false,
+			rayServiceUpgradeStrategy: rayv1.None,
+		},
+		// Test 10: Zero downtime upgrade is enabled. Neither the env var nor the RayServiceSpec is set. Trigger the zero-downtime upgrade.
+		"Zero-downtime upgrade is enabled. Neither the env var nor the RayServiceSpec is set.": {
+			activeCluster:             nil,
+			updateRayClusterSpec:      true,
+			shouldPrepareNewCluster:   true,
+			rayServiceUpgradeStrategy: "",
+		},
 	}
 
 	for name, tc := range tests {
@@ -824,6 +863,9 @@ func TestReconcileRayCluster(t *testing.T) {
 			if !tc.enableZeroDowntime {
 				os.Setenv(ENABLE_ZERO_DOWNTIME, "false")
 			}
+			if tc.rayServiceUpgradeStrategy != "" {
+				rayService.Spec.UpgradeStrategy = tc.rayServiceUpgradeStrategy
+			}
 			runtimeObjects := []runtime.Object{}
 			if tc.activeCluster != nil {
 				// Update 'ray.io/kuberay-version' to a new version if kubeRayVersion is set.