feat: add confidence threshold to workload scaling policy

Author: UOndro

.gitignore

@@ -1,4 +1,5 @@
 .idea/
+.vscode/
 
 # built binary
 terraform-provider-castai

castai/resource_workload_scaling_policy.go

@@ -25,16 +25,18 @@ const (
 	minApplyThresholdValue          = 0.01
 	maxApplyThresholdValue          = 2.5
 	defaultApplyThresholdPercentage = 0.1
+	defaultConfidenceThreshold      = 0.9
 	minNumeratorValue               = 0.0
 	maxExponentValue                = 1.
 	minExponentValue                = 0.
 )
 
 const (
-	FieldLimitStrategy           = "limit"
-	FieldLimitStrategyType       = "type"
-	FieldLimitStrategyMultiplier = "multiplier"
-
+	FieldLimitStrategy                             = "limit"
+	FieldLimitStrategyType                         = "type"
+	FieldLimitStrategyMultiplier                   = "multiplier"
+	FieldConfidence                                = "confidence"
+	FieldConfidenceThreshold                       = "threshold"
 	DeprecatedFieldApplyThreshold                  = "apply_threshold"
 	FieldApplyThresholdStrategy                    = "apply_threshold_strategy"
 	FieldApplyThresholdStrategyType                = "type"
@@ -119,6 +121,26 @@ func resourceWorkloadScalingPolicy() *schema.Resource {
 					},
 				},
 			},
+			FieldConfidence: {
+				Type:        schema.TypeList,
+				Optional:    true,
+				MaxItems:    1,
+				Description: "Defines the confidence settings for applying recommendations.",
+				DiffSuppressFunc: func(k, old, new string, d *schema.ResourceData) bool {
+					return suppressConfidenceThresholdDefaultValueDiff(FieldConfidence, old, new, d)
+				},
+				Elem: &schema.Resource{
+					Schema: map[string]*schema.Schema{
+						FieldConfidenceThreshold: {
+							Type:             schema.TypeFloat,
+							Optional:         true,
+							Default: 		  0.9,
+							Description:      "Defines the confidence threshold for applying recommendations. The smaller number indicates that we require fewer metrics data points to apply recommendations.",
+							ValidateDiagFunc: validation.ToDiagFunc(validation.FloatBetween(0, 1)),
+						},
+					},
+				},
+			},
 			"downscaling": {
 				Type:     schema.TypeList,
 				Optional: true,
@@ -381,6 +403,8 @@ func resourceWorkloadScalingPolicyCreate(ctx context.Context, d *schema.Resource
 		}
 		req.RecommendationPolicies.Memory = memory
 	}
+	
+	req.RecommendationPolicies.Confidence = toConfidence(toSection(d, FieldConfidence))
 
 	req.RecommendationPolicies.Startup = toStartup(toSection(d, "startup"))
 
@@ -452,6 +476,9 @@ func resourceWorkloadScalingPolicyRead(ctx context.Context, d *schema.ResourceDa
 	if err := d.Set("startup", toStartupMap(sp.RecommendationPolicies.Startup)); err != nil {
 		return diag.FromErr(fmt.Errorf("setting startup: %w", err))
 	}
+	if err := d.Set(FieldConfidence, toConfidenceMap(sp.RecommendationPolicies.Confidence)); err != nil {
+		return diag.FromErr(fmt.Errorf("setting confidence: %w", err))
+	}
 	if err := d.Set("downscaling", toDownscalingMap(sp.RecommendationPolicies.Downscaling)); err != nil {
 		return diag.FromErr(fmt.Errorf("setting downscaling: %w", err))
 	}
@@ -483,6 +510,7 @@ func resourceWorkloadScalingPolicyUpdate(ctx context.Context, d *schema.Resource
 		"downscaling",
 		"memory_event",
 		"anti_affinity",
+		FieldConfidence,
 	) {
 		tflog.Info(ctx, "scaling policy up to date")
 		return nil
@@ -509,6 +537,7 @@ func resourceWorkloadScalingPolicyUpdate(ctx context.Context, d *schema.Resource
 			Downscaling:      toDownscaling(toSection(d, "downscaling")),
 			MemoryEvent:      toMemoryEvent(toSection(d, "memory_event")),
 			AntiAffinity:     toAntiAffinity(toSection(d, "anti_affinity")),
+			Confidence: 	 toConfidence(toSection(d, FieldConfidence)),
 		},
 	}
 
@@ -740,6 +769,17 @@ func suppressThresholdStrategyDefaultValueDiff(resource, oldValue, newValue stri
 	return oldValue == newValue
 }
 
+func suppressConfidenceThresholdDefaultValueDiff(resource, oldValue, newValue string, d *schema.ResourceData) bool {
+	isConfidenceUnset := newValue == "0" || newValue == ""
+	if isConfidenceUnset {
+		confidenceThreshold := d.Get(fmt.Sprintf("%s.0.%s", resource, FieldConfidenceThreshold))
+		// Suppress diff if configuration saved from API equals to default
+		return confidenceThreshold == defaultConfidenceThreshold
+	}
+
+	return oldValue == newValue
+}
+
 func toWorkloadResourceLimit(obj map[string]any) (*sdk.WorkloadoptimizationV1ResourceLimitStrategy, error) {
 	if len(obj) == 0 {
 		return nil, nil
@@ -852,6 +892,32 @@ func applyThresholdStrategyToMap(s *sdk.WorkloadoptimizationV1ApplyThresholdStra
 	return []map[string]any{m}
 }
 
+func toConfidence(confidence map[string]any) *sdk.WorkloadoptimizationV1ConfidenceSettings {
+	if len(confidence) == 0 {
+		return nil
+	}
+
+	result := &sdk.WorkloadoptimizationV1ConfidenceSettings{}
+
+	if v, ok := confidence[FieldConfidenceThreshold].(float64); ok {
+		result.Threshold = &v
+	}
+
+	return result
+}
+
+func toConfidenceMap(s *sdk.WorkloadoptimizationV1ConfidenceSettings) []map[string]any {
+	if s == nil {
+		return nil
+	}
+
+	m := map[string]any{
+		FieldConfidenceThreshold: s.Threshold,
+	}
+
+	return []map[string]any{m}
+}
+
 func toStartup(startup map[string]any) *sdk.WorkloadoptimizationV1StartupSettings {
 	if len(startup) == 0 {
 		return nil

castai/resource_workload_scaling_policy_test.go

@@ -52,6 +52,7 @@ func TestAccResourceWorkloadScalingPolicy(t *testing.T) {
 					resource.TestCheckResourceAttr(resourceName, "memory.0.limit.0.type", "MULTIPLIER"),
 					resource.TestCheckResourceAttr(resourceName, "memory.0.limit.0.multiplier", "1.8"),
 					resource.TestCheckResourceAttr(resourceName, "memory.0.management_option", "READ_ONLY"),
+					resource.TestCheckResourceAttr(resourceName, "confidence.0.threshold", "0.4"),
 				),
 			},
 			{
@@ -88,6 +89,7 @@ func TestAccResourceWorkloadScalingPolicy(t *testing.T) {
 					resource.TestCheckResourceAttr(resourceName, "startup.0.period_seconds", "123"),
 					resource.TestCheckResourceAttr(resourceName, "downscaling.0.apply_type", "DEFERRED"),
 					resource.TestCheckResourceAttr(resourceName, "memory_event.0.apply_type", "DEFERRED"),
+					resource.TestCheckResourceAttr(resourceName, "confidence.0.threshold", "0.6"),
 					resource.TestCheckResourceAttr(resourceName, "anti_affinity.0.consider_anti_affinity", "true"),
 				),
 			},
@@ -112,6 +114,9 @@ func scalingPolicyConfig(clusterName, projectID, name string) string {
 		cluster_id			= castai_gke_cluster.test.id
 		apply_type			= "IMMEDIATE"
 		management_option	= "READ_ONLY"
+		confidence {
+			threshold = 0.4
+		}
 		cpu {
 			function 		= "QUANTILE"
 			overhead 		= 0.05
@@ -195,6 +200,9 @@ func scalingPolicyConfigUpdated(clusterName, projectID, name string) string {
 		anti_affinity {
 			consider_anti_affinity = true
 		}
+		confidence {
+			threshold = 0.6
+		}
 	}`, updatedName)
 
 	return ConfigCompose(testAccGKEClusterConfig(name, clusterName, projectID), cfg)

docs/resources/workload_scaling_policy.md

@@ -39,4 +39,7 @@ resource "castai_workload_scaling_policy" "services" {
   anti_affinity {
     consider_anti_affinity = false
   }
+  confidence {
+    threshold = 0.9
+  }
 }