aws-sdk-sagemaker 1.356.0 → 1.357.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 3f0b9d6eec27903cf17198c529d6a3afafb0173ccb84d4975f8c44b4fe354d52
-  data.tar.gz: cd19f6247b5bd223b05f1a73e34aae8527c0e069257b24fb9fccc233b6579959
+  metadata.gz: 7f316e630f07094ff5167bc4cc33094f38cd54a4bc12ab9115ed3b92ac53a187
+  data.tar.gz: c769b634cb5498cac7f8cbff132d978204d47b44d8ca9a0f374984fa1f3fd4f1
 SHA512:
-  metadata.gz: e25ef94ab127188f40f966af619066bf328b81f18c9906bf6aa0cb30ec43263fd153be41519867f22561614247d58d55ec1e5a66faf23469bf09d33c213a2075
-  data.tar.gz: 2fd039d0bedadee4e5edc2b58af02e39bce099b81c676498348d541a585d8738adaa557c7f0ceb2395332ce1db1813f57bd66a7acaf005f13795d6171bd57a5a
+  metadata.gz: d23dbe4e149951a9edb1418bcce5b9e0c730a138f422ad6f63dcdb4469f8dadc14f754e570133107a9af62a8db737a3d0de69428d0892b23de324e7338875bce
+  data.tar.gz: 5d47d51c8ae0d509a97c0ee14d65090b0252bfc52d40efc37f314b7da7e590503ff44e2fa1a2d6deb10124342be8a6e672bae327204ed4a5d04da0eebedf0147

data/CHANGELOG.md

@@ -1,6 +1,11 @@
 Unreleased Changes
 ------------------
 
+1.357.0 (2026-03-30)
+------------------
+
+* Feature - Added support for placement strategy and consolidation for SageMaker inference component endpoints. Customers can now configure how inference component copies are distributed across instances and availability zones (AZs), and enable automatic consolidation to optimizes resource utilization.
+
 1.356.0 (2026-03-26)
 ------------------
 

data/VERSION

@@ -1 +1 @@
-1.356.0
+1.357.0

data/lib/aws-sdk-sagemaker/client.rb

@@ -4412,6 +4412,11 @@ module Aws::SageMaker
     #           status: "ENABLED", # accepts ENABLED, DISABLED
     #           min_instance_count: 1,
     #           max_instance_count: 1,
+    #           scale_in_policy: {
+    #             strategy: "IDLE_RELEASE", # required, accepts IDLE_RELEASE, CONSOLIDATION
+    #             maximum_step_size: 1,
+    #             cooldown_in_minutes: 1,
+    #           },
     #         },
     #         routing_config: {
     #           routing_strategy: "LEAST_OUTSTANDING_REQUESTS", # required, accepts LEAST_OUTSTANDING_REQUESTS, RANDOM
@@ -4517,6 +4522,11 @@ module Aws::SageMaker
     #           status: "ENABLED", # accepts ENABLED, DISABLED
     #           min_instance_count: 1,
     #           max_instance_count: 1,
+    #           scale_in_policy: {
+    #             strategy: "IDLE_RELEASE", # required, accepts IDLE_RELEASE, CONSOLIDATION
+    #             maximum_step_size: 1,
+    #             cooldown_in_minutes: 1,
+    #           },
     #         },
     #         routing_config: {
     #           routing_strategy: "LEAST_OUTSTANDING_REQUESTS", # required, accepts LEAST_OUTSTANDING_REQUESTS, RANDOM
@@ -5963,6 +5973,13 @@ module Aws::SageMaker
     #       data_cache_config: {
     #         enable_caching: false, # required
     #       },
+    #       scheduling_config: {
+    #         placement_strategy: "SPREAD", # required, accepts SPREAD, BINPACK
+    #         availability_zone_balance: {
+    #           enforcement_mode: "PERMISSIVE", # required, accepts PERMISSIVE
+    #           max_imbalance: 1,
+    #         },
+    #       },
     #     },
     #     runtime_config: {
     #       copy_count: 1, # required
@@ -15256,6 +15273,9 @@ module Aws::SageMaker
     #   resp.production_variants[0].managed_instance_scaling.status #=> String, one of "ENABLED", "DISABLED"
     #   resp.production_variants[0].managed_instance_scaling.min_instance_count #=> Integer
     #   resp.production_variants[0].managed_instance_scaling.max_instance_count #=> Integer
+    #   resp.production_variants[0].managed_instance_scaling.scale_in_policy.strategy #=> String, one of "IDLE_RELEASE", "CONSOLIDATION"
+    #   resp.production_variants[0].managed_instance_scaling.scale_in_policy.maximum_step_size #=> Integer
+    #   resp.production_variants[0].managed_instance_scaling.scale_in_policy.cooldown_in_minutes #=> Integer
     #   resp.production_variants[0].routing_config.routing_strategy #=> String, one of "LEAST_OUTSTANDING_REQUESTS", "RANDOM"
     #   resp.production_variants[0].capacity_reservation_config.ml_reservation_arn #=> String
     #   resp.production_variants[0].capacity_reservation_config.capacity_reservation_preference #=> String, one of "capacity-reservations-only"
@@ -15326,6 +15346,9 @@ module Aws::SageMaker
     #   resp.pending_deployment_summary.production_variants[0].managed_instance_scaling.status #=> String, one of "ENABLED", "DISABLED"
     #   resp.pending_deployment_summary.production_variants[0].managed_instance_scaling.min_instance_count #=> Integer
     #   resp.pending_deployment_summary.production_variants[0].managed_instance_scaling.max_instance_count #=> Integer
+    #   resp.pending_deployment_summary.production_variants[0].managed_instance_scaling.scale_in_policy.strategy #=> String, one of "IDLE_RELEASE", "CONSOLIDATION"
+    #   resp.pending_deployment_summary.production_variants[0].managed_instance_scaling.scale_in_policy.maximum_step_size #=> Integer
+    #   resp.pending_deployment_summary.production_variants[0].managed_instance_scaling.scale_in_policy.cooldown_in_minutes #=> Integer
     #   resp.pending_deployment_summary.production_variants[0].routing_config.routing_strategy #=> String, one of "LEAST_OUTSTANDING_REQUESTS", "RANDOM"
     #   resp.pending_deployment_summary.start_time #=> Time
     #   resp.pending_deployment_summary.shadow_production_variants #=> Array
@@ -15353,6 +15376,9 @@ module Aws::SageMaker
     #   resp.pending_deployment_summary.shadow_production_variants[0].managed_instance_scaling.status #=> String, one of "ENABLED", "DISABLED"
     #   resp.pending_deployment_summary.shadow_production_variants[0].managed_instance_scaling.min_instance_count #=> Integer
     #   resp.pending_deployment_summary.shadow_production_variants[0].managed_instance_scaling.max_instance_count #=> Integer
+    #   resp.pending_deployment_summary.shadow_production_variants[0].managed_instance_scaling.scale_in_policy.strategy #=> String, one of "IDLE_RELEASE", "CONSOLIDATION"
+    #   resp.pending_deployment_summary.shadow_production_variants[0].managed_instance_scaling.scale_in_policy.maximum_step_size #=> Integer
+    #   resp.pending_deployment_summary.shadow_production_variants[0].managed_instance_scaling.scale_in_policy.cooldown_in_minutes #=> Integer
     #   resp.pending_deployment_summary.shadow_production_variants[0].routing_config.routing_strategy #=> String, one of "LEAST_OUTSTANDING_REQUESTS", "RANDOM"
     #   resp.explainer_config.clarify_explainer_config.enable_explanations #=> String
     #   resp.explainer_config.clarify_explainer_config.inference_config.features_attribute #=> String
@@ -15400,6 +15426,9 @@ module Aws::SageMaker
     #   resp.shadow_production_variants[0].managed_instance_scaling.status #=> String, one of "ENABLED", "DISABLED"
     #   resp.shadow_production_variants[0].managed_instance_scaling.min_instance_count #=> Integer
     #   resp.shadow_production_variants[0].managed_instance_scaling.max_instance_count #=> Integer
+    #   resp.shadow_production_variants[0].managed_instance_scaling.scale_in_policy.strategy #=> String, one of "IDLE_RELEASE", "CONSOLIDATION"
+    #   resp.shadow_production_variants[0].managed_instance_scaling.scale_in_policy.maximum_step_size #=> Integer
+    #   resp.shadow_production_variants[0].managed_instance_scaling.scale_in_policy.cooldown_in_minutes #=> Integer
     #   resp.shadow_production_variants[0].routing_config.routing_strategy #=> String, one of "LEAST_OUTSTANDING_REQUESTS", "RANDOM"
     #   resp.shadow_production_variants[0].capacity_reservation_config.ml_reservation_arn #=> String
     #   resp.shadow_production_variants[0].capacity_reservation_config.capacity_reservation_preference #=> String, one of "capacity-reservations-only"
@@ -15480,6 +15509,9 @@ module Aws::SageMaker
     #   resp.production_variants[0].managed_instance_scaling.status #=> String, one of "ENABLED", "DISABLED"
     #   resp.production_variants[0].managed_instance_scaling.min_instance_count #=> Integer
     #   resp.production_variants[0].managed_instance_scaling.max_instance_count #=> Integer
+    #   resp.production_variants[0].managed_instance_scaling.scale_in_policy.strategy #=> String, one of "IDLE_RELEASE", "CONSOLIDATION"
+    #   resp.production_variants[0].managed_instance_scaling.scale_in_policy.maximum_step_size #=> Integer
+    #   resp.production_variants[0].managed_instance_scaling.scale_in_policy.cooldown_in_minutes #=> Integer
     #   resp.production_variants[0].routing_config.routing_strategy #=> String, one of "LEAST_OUTSTANDING_REQUESTS", "RANDOM"
     #   resp.production_variants[0].inference_ami_version #=> String, one of "al2-ami-sagemaker-inference-gpu-2", "al2-ami-sagemaker-inference-gpu-2-1", "al2-ami-sagemaker-inference-gpu-3-1", "al2-ami-sagemaker-inference-neuron-2", "al2023-ami-sagemaker-inference-gpu-4-1"
     #   resp.production_variants[0].capacity_reservation_config.capacity_reservation_preference #=> String, one of "capacity-reservations-only"
@@ -15546,6 +15578,9 @@ module Aws::SageMaker
     #   resp.shadow_production_variants[0].managed_instance_scaling.status #=> String, one of "ENABLED", "DISABLED"
     #   resp.shadow_production_variants[0].managed_instance_scaling.min_instance_count #=> Integer
     #   resp.shadow_production_variants[0].managed_instance_scaling.max_instance_count #=> Integer
+    #   resp.shadow_production_variants[0].managed_instance_scaling.scale_in_policy.strategy #=> String, one of "IDLE_RELEASE", "CONSOLIDATION"
+    #   resp.shadow_production_variants[0].managed_instance_scaling.scale_in_policy.maximum_step_size #=> Integer
+    #   resp.shadow_production_variants[0].managed_instance_scaling.scale_in_policy.cooldown_in_minutes #=> Integer
     #   resp.shadow_production_variants[0].routing_config.routing_strategy #=> String, one of "LEAST_OUTSTANDING_REQUESTS", "RANDOM"
     #   resp.shadow_production_variants[0].inference_ami_version #=> String, one of "al2-ami-sagemaker-inference-gpu-2", "al2-ami-sagemaker-inference-gpu-2-1", "al2-ami-sagemaker-inference-gpu-3-1", "al2-ami-sagemaker-inference-neuron-2", "al2023-ami-sagemaker-inference-gpu-4-1"
     #   resp.shadow_production_variants[0].capacity_reservation_config.capacity_reservation_preference #=> String, one of "capacity-reservations-only"
@@ -16470,6 +16505,9 @@ module Aws::SageMaker
     #   resp.specification.compute_resource_requirements.max_memory_required_in_mb #=> Integer
     #   resp.specification.base_inference_component_name #=> String
     #   resp.specification.data_cache_config.enable_caching #=> Boolean
+    #   resp.specification.scheduling_config.placement_strategy #=> String, one of "SPREAD", "BINPACK"
+    #   resp.specification.scheduling_config.availability_zone_balance.enforcement_mode #=> String, one of "PERMISSIVE"
+    #   resp.specification.scheduling_config.availability_zone_balance.max_imbalance #=> Integer
     #   resp.runtime_config.desired_copy_count #=> Integer
     #   resp.runtime_config.current_copy_count #=> Integer
     #   resp.creation_time #=> Time
@@ -30312,6 +30350,13 @@ module Aws::SageMaker
     #       data_cache_config: {
     #         enable_caching: false, # required
     #       },
+    #       scheduling_config: {
+    #         placement_strategy: "SPREAD", # required, accepts SPREAD, BINPACK
+    #         availability_zone_balance: {
+    #           enforcement_mode: "PERMISSIVE", # required, accepts PERMISSIVE
+    #           max_imbalance: 1,
+    #         },
+    #       },
     #     },
     #     runtime_config: {
     #       copy_count: 1, # required
@@ -32495,7 +32540,7 @@ module Aws::SageMaker
         tracer: tracer
       )
       context[:gem_name] = 'aws-sdk-sagemaker'
-      context[:gem_version] = '1.356.0'
+      context[:gem_version] = '1.357.0'
       Seahorse::Client::Request.new(handlers, context)
     end
 

data/lib/aws-sdk-sagemaker/client_api.rb

@@ -194,6 +194,8 @@ module Aws::SageMaker
     Autotune = Shapes::StructureShape.new(name: 'Autotune')
     AutotuneMode = Shapes::StringShape.new(name: 'AutotuneMode')
     AvailabilityZone = Shapes::StringShape.new(name: 'AvailabilityZone')
+    AvailabilityZoneBalanceEnforcementMode = Shapes::StringShape.new(name: 'AvailabilityZoneBalanceEnforcementMode')
+    AvailabilityZoneBalanceMaxImbalance = Shapes::IntegerShape.new(name: 'AvailabilityZoneBalanceMaxImbalance')
     AvailabilityZoneId = Shapes::StringShape.new(name: 'AvailabilityZoneId')
     AvailableInstanceCount = Shapes::IntegerShape.new(name: 'AvailableInstanceCount')
     AvailableSpareInstanceCount = Shapes::IntegerShape.new(name: 'AvailableSpareInstanceCount')
@@ -1332,6 +1334,7 @@ module Aws::SageMaker
     InUseInstanceCount = Shapes::IntegerShape.new(name: 'InUseInstanceCount')
     IncludeNodeLogicalIdsBoolean = Shapes::BooleanShape.new(name: 'IncludeNodeLogicalIdsBoolean')
     InferenceComponentArn = Shapes::StringShape.new(name: 'InferenceComponentArn')
+    InferenceComponentAvailabilityZoneBalance = Shapes::StructureShape.new(name: 'InferenceComponentAvailabilityZoneBalance')
     InferenceComponentCapacitySize = Shapes::StructureShape.new(name: 'InferenceComponentCapacitySize')
     InferenceComponentCapacitySizeType = Shapes::StringShape.new(name: 'InferenceComponentCapacitySizeType')
     InferenceComponentComputeResourceRequirements = Shapes::StructureShape.new(name: 'InferenceComponentComputeResourceRequirements')
@@ -1344,9 +1347,11 @@ module Aws::SageMaker
     InferenceComponentMetadata = Shapes::StructureShape.new(name: 'InferenceComponentMetadata')
     InferenceComponentName = Shapes::StringShape.new(name: 'InferenceComponentName')
     InferenceComponentNameContains = Shapes::StringShape.new(name: 'InferenceComponentNameContains')
+    InferenceComponentPlacementStrategy = Shapes::StringShape.new(name: 'InferenceComponentPlacementStrategy')
     InferenceComponentRollingUpdatePolicy = Shapes::StructureShape.new(name: 'InferenceComponentRollingUpdatePolicy')
     InferenceComponentRuntimeConfig = Shapes::StructureShape.new(name: 'InferenceComponentRuntimeConfig')
     InferenceComponentRuntimeConfigSummary = Shapes::StructureShape.new(name: 'InferenceComponentRuntimeConfigSummary')
+    InferenceComponentSchedulingConfig = Shapes::StructureShape.new(name: 'InferenceComponentSchedulingConfig')
     InferenceComponentSortKey = Shapes::StringShape.new(name: 'InferenceComponentSortKey')
     InferenceComponentSpecification = Shapes::StructureShape.new(name: 'InferenceComponentSpecification')
     InferenceComponentSpecificationSummary = Shapes::StructureShape.new(name: 'InferenceComponentSpecificationSummary')
@@ -1660,8 +1665,11 @@ module Aws::SageMaker
     MLflowConfiguration = Shapes::StructureShape.new(name: 'MLflowConfiguration')
     MaintenanceStatus = Shapes::StringShape.new(name: 'MaintenanceStatus')
     MajorMinorVersion = Shapes::StringShape.new(name: 'MajorMinorVersion')
+    ManagedInstanceScalingCooldownInMinutes = Shapes::IntegerShape.new(name: 'ManagedInstanceScalingCooldownInMinutes')
     ManagedInstanceScalingMaxInstanceCount = Shapes::IntegerShape.new(name: 'ManagedInstanceScalingMaxInstanceCount')
+    ManagedInstanceScalingMaximumStepSize = Shapes::IntegerShape.new(name: 'ManagedInstanceScalingMaximumStepSize')
     ManagedInstanceScalingMinInstanceCount = Shapes::IntegerShape.new(name: 'ManagedInstanceScalingMinInstanceCount')
+    ManagedInstanceScalingScaleInStrategy = Shapes::StringShape.new(name: 'ManagedInstanceScalingScaleInStrategy')
     ManagedInstanceScalingStatus = Shapes::StringShape.new(name: 'ManagedInstanceScalingStatus')
     MapString2048 = Shapes::MapShape.new(name: 'MapString2048')
     MaxAutoMLJobRuntimeInSeconds = Shapes::IntegerShape.new(name: 'MaxAutoMLJobRuntimeInSeconds')
@@ -2115,6 +2123,7 @@ module Aws::SageMaker
     ProductionVariantInstanceType = Shapes::StringShape.new(name: 'ProductionVariantInstanceType')
     ProductionVariantList = Shapes::ListShape.new(name: 'ProductionVariantList')
     ProductionVariantManagedInstanceScaling = Shapes::StructureShape.new(name: 'ProductionVariantManagedInstanceScaling')
+    ProductionVariantManagedInstanceScalingScaleInPolicy = Shapes::StructureShape.new(name: 'ProductionVariantManagedInstanceScalingScaleInPolicy')
     ProductionVariantModelDataDownloadTimeoutInSeconds = Shapes::IntegerShape.new(name: 'ProductionVariantModelDataDownloadTimeoutInSeconds')
     ProductionVariantRoutingConfig = Shapes::StructureShape.new(name: 'ProductionVariantRoutingConfig')
     ProductionVariantSSMAccess = Shapes::BooleanShape.new(name: 'ProductionVariantSSMAccess')
@@ -7683,6 +7692,10 @@ module Aws::SageMaker
     ImportHubContentResponse.add_member(:hub_content_arn, Shapes::ShapeRef.new(shape: HubContentArn, required: true, location_name: "HubContentArn"))
     ImportHubContentResponse.struct_class = Types::ImportHubContentResponse
 
+    InferenceComponentAvailabilityZoneBalance.add_member(:enforcement_mode, Shapes::ShapeRef.new(shape: AvailabilityZoneBalanceEnforcementMode, required: true, location_name: "EnforcementMode"))
+    InferenceComponentAvailabilityZoneBalance.add_member(:max_imbalance, Shapes::ShapeRef.new(shape: AvailabilityZoneBalanceMaxImbalance, location_name: "MaxImbalance"))
+    InferenceComponentAvailabilityZoneBalance.struct_class = Types::InferenceComponentAvailabilityZoneBalance
+
     InferenceComponentCapacitySize.add_member(:type, Shapes::ShapeRef.new(shape: InferenceComponentCapacitySizeType, required: true, location_name: "Type"))
     InferenceComponentCapacitySize.add_member(:value, Shapes::ShapeRef.new(shape: CapacitySizeValue, required: true, location_name: "Value"))
     InferenceComponentCapacitySize.struct_class = Types::InferenceComponentCapacitySize
@@ -7729,12 +7742,17 @@ module Aws::SageMaker
     InferenceComponentRuntimeConfigSummary.add_member(:current_copy_count, Shapes::ShapeRef.new(shape: InferenceComponentCopyCount, location_name: "CurrentCopyCount"))
     InferenceComponentRuntimeConfigSummary.struct_class = Types::InferenceComponentRuntimeConfigSummary
 
+    InferenceComponentSchedulingConfig.add_member(:placement_strategy, Shapes::ShapeRef.new(shape: InferenceComponentPlacementStrategy, required: true, location_name: "PlacementStrategy"))
+    InferenceComponentSchedulingConfig.add_member(:availability_zone_balance, Shapes::ShapeRef.new(shape: InferenceComponentAvailabilityZoneBalance, location_name: "AvailabilityZoneBalance"))
+    InferenceComponentSchedulingConfig.struct_class = Types::InferenceComponentSchedulingConfig
+
     InferenceComponentSpecification.add_member(:model_name, Shapes::ShapeRef.new(shape: ModelName, location_name: "ModelName"))
     InferenceComponentSpecification.add_member(:container, Shapes::ShapeRef.new(shape: InferenceComponentContainerSpecification, location_name: "Container"))
     InferenceComponentSpecification.add_member(:startup_parameters, Shapes::ShapeRef.new(shape: InferenceComponentStartupParameters, location_name: "StartupParameters"))
     InferenceComponentSpecification.add_member(:compute_resource_requirements, Shapes::ShapeRef.new(shape: InferenceComponentComputeResourceRequirements, location_name: "ComputeResourceRequirements"))
     InferenceComponentSpecification.add_member(:base_inference_component_name, Shapes::ShapeRef.new(shape: InferenceComponentName, location_name: "BaseInferenceComponentName"))
     InferenceComponentSpecification.add_member(:data_cache_config, Shapes::ShapeRef.new(shape: InferenceComponentDataCacheConfig, location_name: "DataCacheConfig"))
+    InferenceComponentSpecification.add_member(:scheduling_config, Shapes::ShapeRef.new(shape: InferenceComponentSchedulingConfig, location_name: "SchedulingConfig"))
     InferenceComponentSpecification.struct_class = Types::InferenceComponentSpecification
 
     InferenceComponentSpecificationSummary.add_member(:model_name, Shapes::ShapeRef.new(shape: ModelName, location_name: "ModelName"))
@@ -7743,6 +7761,7 @@ module Aws::SageMaker
     InferenceComponentSpecificationSummary.add_member(:compute_resource_requirements, Shapes::ShapeRef.new(shape: InferenceComponentComputeResourceRequirements, location_name: "ComputeResourceRequirements"))
     InferenceComponentSpecificationSummary.add_member(:base_inference_component_name, Shapes::ShapeRef.new(shape: InferenceComponentName, location_name: "BaseInferenceComponentName"))
     InferenceComponentSpecificationSummary.add_member(:data_cache_config, Shapes::ShapeRef.new(shape: InferenceComponentDataCacheConfigSummary, location_name: "DataCacheConfig"))
+    InferenceComponentSpecificationSummary.add_member(:scheduling_config, Shapes::ShapeRef.new(shape: InferenceComponentSchedulingConfig, location_name: "SchedulingConfig"))
     InferenceComponentSpecificationSummary.struct_class = Types::InferenceComponentSpecificationSummary
 
     InferenceComponentStartupParameters.add_member(:model_data_download_timeout_in_seconds, Shapes::ShapeRef.new(shape: ProductionVariantModelDataDownloadTimeoutInSeconds, location_name: "ModelDataDownloadTimeoutInSeconds"))
@@ -10446,8 +10465,14 @@ module Aws::SageMaker
     ProductionVariantManagedInstanceScaling.add_member(:status, Shapes::ShapeRef.new(shape: ManagedInstanceScalingStatus, location_name: "Status"))
     ProductionVariantManagedInstanceScaling.add_member(:min_instance_count, Shapes::ShapeRef.new(shape: ManagedInstanceScalingMinInstanceCount, location_name: "MinInstanceCount"))
     ProductionVariantManagedInstanceScaling.add_member(:max_instance_count, Shapes::ShapeRef.new(shape: ManagedInstanceScalingMaxInstanceCount, location_name: "MaxInstanceCount"))
+    ProductionVariantManagedInstanceScaling.add_member(:scale_in_policy, Shapes::ShapeRef.new(shape: ProductionVariantManagedInstanceScalingScaleInPolicy, location_name: "ScaleInPolicy"))
     ProductionVariantManagedInstanceScaling.struct_class = Types::ProductionVariantManagedInstanceScaling
 
+    ProductionVariantManagedInstanceScalingScaleInPolicy.add_member(:strategy, Shapes::ShapeRef.new(shape: ManagedInstanceScalingScaleInStrategy, required: true, location_name: "Strategy"))
+    ProductionVariantManagedInstanceScalingScaleInPolicy.add_member(:maximum_step_size, Shapes::ShapeRef.new(shape: ManagedInstanceScalingMaximumStepSize, location_name: "MaximumStepSize"))
+    ProductionVariantManagedInstanceScalingScaleInPolicy.add_member(:cooldown_in_minutes, Shapes::ShapeRef.new(shape: ManagedInstanceScalingCooldownInMinutes, location_name: "CooldownInMinutes"))
+    ProductionVariantManagedInstanceScalingScaleInPolicy.struct_class = Types::ProductionVariantManagedInstanceScalingScaleInPolicy
+
     ProductionVariantRoutingConfig.add_member(:routing_strategy, Shapes::ShapeRef.new(shape: RoutingStrategy, required: true, location_name: "RoutingStrategy"))
     ProductionVariantRoutingConfig.struct_class = Types::ProductionVariantRoutingConfig
 

data/lib/aws-sdk-sagemaker/types.rb

@@ -28799,6 +28799,40 @@ module Aws::SageMaker
       include Aws::Structure
     end
 
+    # Configuration for balancing inference component copies across
+    # Availability Zones.
+    #
+    # @!attribute [rw] enforcement_mode
+    #   Determines how strictly the Availability Zone balance constraint is
+    #   enforced.
+    #
+    #   PERMISSIVE
+    #
+    #   : The endpoint attempts to balance copies across Availability Zones
+    #     but proceeds with scheduling even if balance can't be achieved
+    #     due to available capacity or instance distribution across
+    #     Availability Zones.
+    #   @return [String]
+    #
+    # @!attribute [rw] max_imbalance
+    #   The maximum allowed difference in the number of inference component
+    #   copies between any two Availability Zones. This parameter applies
+    #   only when the endpoint has instances across two or more Availability
+    #   Zones. A copy placement is allowed if it reduces imbalance or the
+    #   resulting imbalance is within this value.
+    #
+    #   Default value: `0`.
+    #   @return [Integer]
+    #
+    # @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/InferenceComponentAvailabilityZoneBalance AWS API Documentation
+    #
+    class InferenceComponentAvailabilityZoneBalance < Struct.new(
+      :enforcement_mode,
+      :max_imbalance)
+      SENSITIVE = []
+      include Aws::Structure
+    end
+
     # Specifies the type and size of the endpoint capacity to activate for a
     # rolling deployment or a rollback strategy. You can specify your
     # batches as either of the following:
@@ -29096,6 +29130,40 @@ module Aws::SageMaker
       include Aws::Structure
     end
 
+    # The scheduling configuration that determines how inference component
+    # copies are placed across available instances when copies are added or
+    # removed.
+    #
+    # @!attribute [rw] placement_strategy
+    #   The strategy for placing inference component copies across available
+    #   instances. If you also set `AvailabilityZoneBalance`, this strategy
+    #   applies to placement within each Availability Zone.
+    #
+    #   SPREAD
+    #
+    #   : Distributes copies evenly across available instances for better
+    #     resilience.
+    #
+    #   BINPACK
+    #
+    #   : Packs copies onto fewer instances to optimize resource
+    #     utilization.
+    #   @return [String]
+    #
+    # @!attribute [rw] availability_zone_balance
+    #   Configuration for balancing inference component copies across
+    #   Availability Zones.
+    #   @return [Types::InferenceComponentAvailabilityZoneBalance]
+    #
+    # @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/InferenceComponentSchedulingConfig AWS API Documentation
+    #
+    class InferenceComponentSchedulingConfig < Struct.new(
+      :placement_strategy,
+      :availability_zone_balance)
+      SENSITIVE = []
+      include Aws::Structure
+    end
+
     # Details about the resources to deploy with this inference component,
     # including the model, container, and compute resources.
     #
@@ -29149,6 +29217,12 @@ module Aws::SageMaker
     #   Settings that affect how the inference component caches data.
     #   @return [Types::InferenceComponentDataCacheConfig]
     #
+    # @!attribute [rw] scheduling_config
+    #   The scheduling configuration that determines how inference component
+    #   copies are placed across available instances when copies are added
+    #   or removed.
+    #   @return [Types::InferenceComponentSchedulingConfig]
+    #
     # @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/InferenceComponentSpecification AWS API Documentation
     #
     class InferenceComponentSpecification < Struct.new(
@@ -29157,7 +29231,8 @@ module Aws::SageMaker
       :startup_parameters,
       :compute_resource_requirements,
       :base_inference_component_name,
-      :data_cache_config)
+      :data_cache_config,
+      :scheduling_config)
       SENSITIVE = []
       include Aws::Structure
     end
@@ -29193,6 +29268,12 @@ module Aws::SageMaker
     #   Settings that affect how the inference component caches data.
     #   @return [Types::InferenceComponentDataCacheConfigSummary]
     #
+    # @!attribute [rw] scheduling_config
+    #   The scheduling configuration that determines how inference component
+    #   copies are placed across available instances when copies are added
+    #   or removed.
+    #   @return [Types::InferenceComponentSchedulingConfig]
+    #
     # @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/InferenceComponentSpecificationSummary AWS API Documentation
     #
     class InferenceComponentSpecificationSummary < Struct.new(
@@ -29201,7 +29282,8 @@ module Aws::SageMaker
       :startup_parameters,
       :compute_resource_requirements,
       :base_inference_component_name,
-      :data_cache_config)
+      :data_cache_config,
+      :scheduling_config)
       SENSITIVE = []
       include Aws::Structure
     end
@@ -43874,12 +43956,60 @@ module Aws::SageMaker
     #   it scales up to accommodate an increase in traffic.
     #   @return [Integer]
     #
+    # @!attribute [rw] scale_in_policy
+    #   Configures the scale-in behavior for managed instance scaling.
+    #   @return [Types::ProductionVariantManagedInstanceScalingScaleInPolicy]
+    #
     # @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ProductionVariantManagedInstanceScaling AWS API Documentation
     #
     class ProductionVariantManagedInstanceScaling < Struct.new(
       :status,
       :min_instance_count,
-      :max_instance_count)
+      :max_instance_count,
+      :scale_in_policy)
+      SENSITIVE = []
+      include Aws::Structure
+    end
+
+    # Configures the scale-in behavior for managed instance scaling.
+    #
+    # @!attribute [rw] strategy
+    #   The strategy for scaling in instances.
+    #
+    #   IDLE\_RELEASE
+    #
+    #   : Releases instances that have no hosted inference component copies.
+    #
+    #   CONSOLIDATION
+    #
+    #   : Consolidates inference component copies onto fewer instances to
+    #     release more instances. Consolidation honors the scheduling
+    #     configuration of each inference component. For example, if an
+    #     inference component specifies Availability Zone balance,
+    #     consolidation only proceeds when the resulting distribution does
+    #     not increase the imbalance.
+    #   @return [String]
+    #
+    # @!attribute [rw] maximum_step_size
+    #   The maximum number of instances that the endpoint can terminate at a
+    #   time during a consolidation scale-in operation.
+    #
+    #   Default value: `1`.
+    #   @return [Integer]
+    #
+    # @!attribute [rw] cooldown_in_minutes
+    #   The cooldown period, in minutes, after the last endpoint operation
+    #   before the endpoint evaluates consolidation scale-in opportunities.
+    #
+    #   Default value: `20`.
+    #   @return [Integer]
+    #
+    # @see http://docs.aws.amazon.com/goto/WebAPI/sagemaker-2017-07-24/ProductionVariantManagedInstanceScalingScaleInPolicy AWS API Documentation
+    #
+    class ProductionVariantManagedInstanceScalingScaleInPolicy < Struct.new(
+      :strategy,
+      :maximum_step_size,
+      :cooldown_in_minutes)
       SENSITIVE = []
       include Aws::Structure
     end

data/lib/aws-sdk-sagemaker.rb

@@ -55,7 +55,7 @@ module Aws::SageMaker
   autoload :EndpointProvider, 'aws-sdk-sagemaker/endpoint_provider'
   autoload :Endpoints, 'aws-sdk-sagemaker/endpoints'
 
-  GEM_VERSION = '1.356.0'
+  GEM_VERSION = '1.357.0'
 
 end
 

data/sig/client.rbs

@@ -1815,7 +1815,12 @@ module Aws
                                         managed_instance_scaling: {
                                           status: ("ENABLED" | "DISABLED")?,
                                           min_instance_count: ::Integer?,
-                                          max_instance_count: ::Integer?
+                                          max_instance_count: ::Integer?,
+                                          scale_in_policy: {
+                                            strategy: ("IDLE_RELEASE" | "CONSOLIDATION"),
+                                            maximum_step_size: ::Integer?,
+                                            cooldown_in_minutes: ::Integer?
+                                          }?
                                         }?,
                                         routing_config: {
                                           routing_strategy: ("LEAST_OUTSTANDING_REQUESTS" | "RANDOM")
@@ -1920,7 +1925,12 @@ module Aws
                                         managed_instance_scaling: {
                                           status: ("ENABLED" | "DISABLED")?,
                                           min_instance_count: ::Integer?,
-                                          max_instance_count: ::Integer?
+                                          max_instance_count: ::Integer?,
+                                          scale_in_policy: {
+                                            strategy: ("IDLE_RELEASE" | "CONSOLIDATION"),
+                                            maximum_step_size: ::Integer?,
+                                            cooldown_in_minutes: ::Integer?
+                                          }?
                                         }?,
                                         routing_config: {
                                           routing_strategy: ("LEAST_OUTSTANDING_REQUESTS" | "RANDOM")
@@ -2606,6 +2616,13 @@ module Aws
                                           base_inference_component_name: ::String?,
                                           data_cache_config: {
                                             enable_caching: bool
+                                          }?,
+                                          scheduling_config: {
+                                            placement_strategy: ("SPREAD" | "BINPACK"),
+                                            availability_zone_balance: {
+                                              enforcement_mode: ("PERMISSIVE"),
+                                              max_imbalance: ::Integer?
+                                            }?
                                           }?
                                         },
                                         ?runtime_config: {
@@ -10156,6 +10173,13 @@ module Aws
                                           base_inference_component_name: ::String?,
                                           data_cache_config: {
                                             enable_caching: bool
+                                          }?,
+                                          scheduling_config: {
+                                            placement_strategy: ("SPREAD" | "BINPACK"),
+                                            availability_zone_balance: {
+                                              enforcement_mode: ("PERMISSIVE"),
+                                              max_imbalance: ::Integer?
+                                            }?
                                           }?
                                         },
                                         ?runtime_config: {

data/sig/types.rbs

@@ -5919,6 +5919,12 @@ module Aws::SageMaker
       SENSITIVE: []
     end
 
+    class InferenceComponentAvailabilityZoneBalance
+      attr_accessor enforcement_mode: ("PERMISSIVE")
+      attr_accessor max_imbalance: ::Integer
+      SENSITIVE: []
+    end
+
     class InferenceComponentCapacitySize
       attr_accessor type: ("COPY_COUNT" | "CAPACITY_PERCENT")
       attr_accessor value: ::Integer
@@ -5987,6 +5993,12 @@ module Aws::SageMaker
       SENSITIVE: []
     end
 
+    class InferenceComponentSchedulingConfig
+      attr_accessor placement_strategy: ("SPREAD" | "BINPACK")
+      attr_accessor availability_zone_balance: Types::InferenceComponentAvailabilityZoneBalance
+      SENSITIVE: []
+    end
+
     class InferenceComponentSpecification
       attr_accessor model_name: ::String
       attr_accessor container: Types::InferenceComponentContainerSpecification
@@ -5994,6 +6006,7 @@ module Aws::SageMaker
       attr_accessor compute_resource_requirements: Types::InferenceComponentComputeResourceRequirements
       attr_accessor base_inference_component_name: ::String
       attr_accessor data_cache_config: Types::InferenceComponentDataCacheConfig
+      attr_accessor scheduling_config: Types::InferenceComponentSchedulingConfig
       SENSITIVE: []
     end
 
@@ -6004,6 +6017,7 @@ module Aws::SageMaker
       attr_accessor compute_resource_requirements: Types::InferenceComponentComputeResourceRequirements
       attr_accessor base_inference_component_name: ::String
       attr_accessor data_cache_config: Types::InferenceComponentDataCacheConfigSummary
+      attr_accessor scheduling_config: Types::InferenceComponentSchedulingConfig
       SENSITIVE: []
     end
 
@@ -9312,6 +9326,14 @@ module Aws::SageMaker
       attr_accessor status: ("ENABLED" | "DISABLED")
       attr_accessor min_instance_count: ::Integer
       attr_accessor max_instance_count: ::Integer
+      attr_accessor scale_in_policy: Types::ProductionVariantManagedInstanceScalingScaleInPolicy
+      SENSITIVE: []
+    end
+
+    class ProductionVariantManagedInstanceScalingScaleInPolicy
+      attr_accessor strategy: ("IDLE_RELEASE" | "CONSOLIDATION")
+      attr_accessor maximum_step_size: ::Integer
+      attr_accessor cooldown_in_minutes: ::Integer
       SENSITIVE: []
     end
 

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: aws-sdk-sagemaker
 version: !ruby/object:Gem::Version
-  version: 1.356.0
+  version: 1.357.0
 platform: ruby
 authors:
 - Amazon Web Services