operator: add zone aware API spec validation

operator/CHANGELOG.md

@@ -1,5 +1,6 @@
 ## Main
 
+- [9378](https://github.com/grafana/loki/pull/9378) **aminesnow**: Add zone aware API spec validation
 - [9408](https://github.com/grafana/loki/pull/9408) **JoaoBraveCoding**: Add PodAntiAffinity overwrites per component
 - [9429](https://github.com/grafana/loki/pull/9429) **aminesnow**: Add default TopologySpreadContraints to Gateway
 - [9418](https://github.com/grafana/loki/pull/9418) **JoaoBraveCoding**: Add default TopologySpreadContaints to Querier

operator/apis/loki/v1/lokistack_types.go

@@ -829,6 +829,7 @@ type ReplicationSpec struct {
 	Factor int32 `json:"factor,omitempty"`
 
 	// Zones defines an array of ZoneSpec that the scheduler will try to satisfy.
+	// IMPORTANT: Make sure that the replication factor defined is less than or equal to the number of available zones.
 	//
 	// +optional
 	// +kubebuilder:validation:Optional

operator/bundle/community-openshift/manifests/loki-operator.clusterserviceversion.yaml

@@ -150,7 +150,7 @@ metadata:
     categories: OpenShift Optional, Logging & Tracing
     certified: "false"
     containerImage: docker.io/grafana/loki-operator:main-ac1c1fd
-    createdAt: "2023-05-09T09:43:36Z"
+    createdAt: "2023-05-11T08:04:29Z"
     description: The Community Loki Operator provides Kubernetes native deployment
       and management of Loki and related logging components.
     operators.operatorframework.io/builder: operator-sdk-unknown
@@ -446,8 +446,9 @@ spec:
         path: replication.factor
         x-descriptors:
         - urn:alm:descriptor:com.tectonic.ui:number
-      - description: Zones defines an array of ZoneSpec that the scheduler will try
-          to satisfy.
+      - description: 'Zones defines an array of ZoneSpec that the scheduler will try
+          to satisfy. IMPORTANT: Make sure that the replication factor defined is
+          less than or equal to the number of available zones.'
         displayName: Zones Spec
         path: replication.zones
       - description: MaxSkew describes the maximum degree to which Pods can be unevenly

operator/bundle/community-openshift/manifests/loki.grafana.com_lokistacks.yaml

@@ -344,8 +344,10 @@ spec:
                     minimum: 1
                     type: integer
                   zones:
-                    description: Zones defines an array of ZoneSpec that the scheduler
-                      will try to satisfy.
+                    description: 'Zones defines an array of ZoneSpec that the scheduler
+                      will try to satisfy. IMPORTANT: Make sure that the replication
+                      factor defined is less than or equal to the number of available
+                      zones.'
                     items:
                       description: ZoneSpec defines the spec to support zone-aware
                         component deployments.

operator/bundle/community/manifests/loki-operator.clusterserviceversion.yaml

@@ -150,7 +150,7 @@ metadata:
     categories: OpenShift Optional, Logging & Tracing
     certified: "false"
     containerImage: docker.io/grafana/loki-operator:main-ac1c1fd
-    createdAt: "2023-05-09T09:44:16Z"
+    createdAt: "2023-05-11T08:04:26Z"
     description: The Community Loki Operator provides Kubernetes native deployment
       and management of Loki and related logging components.
     operators.operatorframework.io/builder: operator-sdk-unknown
@@ -446,8 +446,9 @@ spec:
         path: replication.factor
         x-descriptors:
         - urn:alm:descriptor:com.tectonic.ui:number
-      - description: Zones defines an array of ZoneSpec that the scheduler will try
-          to satisfy.
+      - description: 'Zones defines an array of ZoneSpec that the scheduler will try
+          to satisfy. IMPORTANT: Make sure that the replication factor defined is
+          less than or equal to the number of available zones.'
         displayName: Zones Spec
         path: replication.zones
       - description: MaxSkew describes the maximum degree to which Pods can be unevenly

operator/bundle/community/manifests/loki.grafana.com_lokistacks.yaml

@@ -344,8 +344,10 @@ spec:
                     minimum: 1
                     type: integer
                   zones:
-                    description: Zones defines an array of ZoneSpec that the scheduler
-                      will try to satisfy.
+                    description: 'Zones defines an array of ZoneSpec that the scheduler
+                      will try to satisfy. IMPORTANT: Make sure that the replication
+                      factor defined is less than or equal to the number of available
+                      zones.'
                     items:
                       description: ZoneSpec defines the spec to support zone-aware
                         component deployments.

operator/bundle/openshift/manifests/loki-operator.clusterserviceversion.yaml

@@ -150,7 +150,7 @@ metadata:
     categories: OpenShift Optional, Logging & Tracing
     certified: "false"
     containerImage: quay.io/openshift-logging/loki-operator:v0.1.0
-    createdAt: "2023-05-09T09:43:37Z"
+    createdAt: "2023-05-11T08:04:32Z"
     description: |
       The Loki Operator for OCP provides a means for configuring and managing a Loki stack for cluster logging.
       ## Prerequisites and Requirements
@@ -459,8 +459,9 @@ spec:
         path: replication.factor
         x-descriptors:
         - urn:alm:descriptor:com.tectonic.ui:number
-      - description: Zones defines an array of ZoneSpec that the scheduler will try
-          to satisfy.
+      - description: 'Zones defines an array of ZoneSpec that the scheduler will try
+          to satisfy. IMPORTANT: Make sure that the replication factor defined is
+          less than or equal to the number of available zones.'
         displayName: Zones Spec
         path: replication.zones
       - description: MaxSkew describes the maximum degree to which Pods can be unevenly

operator/bundle/openshift/manifests/loki.grafana.com_lokistacks.yaml

@@ -344,8 +344,10 @@ spec:
                     minimum: 1
                     type: integer
                   zones:
-                    description: Zones defines an array of ZoneSpec that the scheduler
-                      will try to satisfy.
+                    description: 'Zones defines an array of ZoneSpec that the scheduler
+                      will try to satisfy. IMPORTANT: Make sure that the replication
+                      factor defined is less than or equal to the number of available
+                      zones.'
                     items:
                       description: ZoneSpec defines the spec to support zone-aware
                         component deployments.

operator/config/crd/bases/loki.grafana.com_lokistacks.yaml

@@ -327,8 +327,10 @@ spec:
                     minimum: 1
                     type: integer
                   zones:
-                    description: Zones defines an array of ZoneSpec that the scheduler
-                      will try to satisfy.
+                    description: 'Zones defines an array of ZoneSpec that the scheduler
+                      will try to satisfy. IMPORTANT: Make sure that the replication
+                      factor defined is less than or equal to the number of available
+                      zones.'
                     items:
                       description: ZoneSpec defines the spec to support zone-aware
                         component deployments.

operator/config/manifests/community-openshift/bases/loki-operator.clusterserviceversion.yaml

@@ -359,8 +359,9 @@ spec:
         path: replication.factor
         x-descriptors:
         - urn:alm:descriptor:com.tectonic.ui:number
-      - description: Zones defines an array of ZoneSpec that the scheduler will try
-          to satisfy.
+      - description: 'Zones defines an array of ZoneSpec that the scheduler will try
+          to satisfy. IMPORTANT: Make sure that the replication factor defined is
+          less than or equal to the number of available zones.'
         displayName: Zones Spec
         path: replication.zones
       - description: MaxSkew describes the maximum degree to which Pods can be unevenly

operator/config/manifests/community/bases/loki-operator.clusterserviceversion.yaml

@@ -359,8 +359,9 @@ spec:
         path: replication.factor
         x-descriptors:
         - urn:alm:descriptor:com.tectonic.ui:number
-      - description: Zones defines an array of ZoneSpec that the scheduler will try
-          to satisfy.
+      - description: 'Zones defines an array of ZoneSpec that the scheduler will try
+          to satisfy. IMPORTANT: Make sure that the replication factor defined is
+          less than or equal to the number of available zones.'
         displayName: Zones Spec
         path: replication.zones
       - description: MaxSkew describes the maximum degree to which Pods can be unevenly

operator/config/manifests/openshift/bases/loki-operator.clusterserviceversion.yaml

@@ -371,8 +371,9 @@ spec:
         path: replication.factor
         x-descriptors:
         - urn:alm:descriptor:com.tectonic.ui:number
-      - description: Zones defines an array of ZoneSpec that the scheduler will try
-          to satisfy.
+      - description: 'Zones defines an array of ZoneSpec that the scheduler will try
+          to satisfy. IMPORTANT: Make sure that the replication factor defined is
+          less than or equal to the number of available zones.'
         displayName: Zones Spec
         path: replication.zones
       - description: MaxSkew describes the maximum degree to which Pods can be unevenly

operator/internal/validation/lokistack.go

@@ -68,6 +68,11 @@ func (v *LokiStackValidator) validate(ctx context.Context, obj runtime.Object) e
 		allErrs = append(allErrs, errors...)
 	}
 
+	errors = v.validateReplicationSpec(ctx, stack.Spec)
+	if len(errors) != 0 {
+		allErrs = append(allErrs, errors...)
+	}
+
 	if v.ExtendedValidator != nil {
 		allErrs = append(allErrs, v.ExtendedValidator(ctx, stack)...)
 	}
@@ -83,6 +88,27 @@ func (v *LokiStackValidator) validate(ctx context.Context, obj runtime.Object) e
 	)
 }
 
+func (v LokiStackValidator) validateReplicationSpec(ctx context.Context, stack lokiv1.LokiStackSpec) field.ErrorList {
+	if stack.Replication == nil {
+		return nil
+	}
+
+	var allErrs field.ErrorList
+
+	// nolint:staticcheck
+	if stack.Replication != nil && stack.ReplicationFactor > 0 {
+		allErrs = append(allErrs, field.Invalid(
+			field.NewPath("spec", "replicationFactor"),
+			stack.ReplicationFactor,
+			lokiv1.ErrReplicationSpecConflict.Error(),
+		))
+
+		return allErrs
+	}
+
+	return nil
+}
+
 // ValidateSchemas ensures that the schemas are in a valid format
 func ValidateSchemas(v *lokiv1.ObjectStorageSpec, utcTime time.Time, status lokiv1.LokiStackStorageStatus) field.ErrorList {
 	var allErrs field.ErrorList

operator/internal/validation/lokistack_test.go

@@ -286,6 +286,70 @@ var ltt = []struct {
 			},
 		),
 	},
+	{
+		desc: "valid replication zones",
+		spec: lokiv1.LokiStack{
+			Spec: lokiv1.LokiStackSpec{
+				Storage: lokiv1.ObjectStorageSpec{
+					Schemas: []lokiv1.ObjectStorageSchema{
+						{
+							Version:       lokiv1.ObjectStorageSchemaV12,
+							EffectiveDate: "2020-10-11",
+						},
+					},
+				},
+				Replication: &lokiv1.ReplicationSpec{
+					Zones: []lokiv1.ZoneSpec{
+						{
+							TopologyKey: "zone",
+						},
+					},
+					Factor: 1,
+				},
+			},
+		},
+	},
+	{
+		desc: "using both replication and replicationFactor",
+		spec: lokiv1.LokiStack{
+			Spec: lokiv1.LokiStackSpec{
+				Storage: lokiv1.ObjectStorageSpec{
+					Schemas: []lokiv1.ObjectStorageSchema{
+						{
+							Version:       lokiv1.ObjectStorageSchemaV12,
+							EffectiveDate: "2020-10-11",
+						},
+					},
+				},
+				ReplicationFactor: 2,
+				Replication: &lokiv1.ReplicationSpec{
+					Zones: []lokiv1.ZoneSpec{
+						{
+							TopologyKey: "zone",
+						},
+						{
+							TopologyKey: "region",
+						},
+						{
+							TopologyKey: "planet",
+						},
+					},
+					Factor: 1,
+				},
+			},
+		},
+		err: apierrors.NewInvalid(
+			schema.GroupKind{Group: "loki.grafana.com", Kind: "LokiStack"},
+			"testing-stack",
+			field.ErrorList{
+				field.Invalid(
+					field.NewPath("spec", "replicationFactor"),
+					2,
+					lokiv1.ErrReplicationSpecConflict.Error(),
+				),
+			},
+		),
+	},
 }
 
 func TestLokiStackValidationWebhook_ValidateCreate(t *testing.T) {