feat(sagemaker): add support for serverless inference endpoints

enhanced_integ_test.ts

@@ -0,0 +1,131 @@
+import * as path from 'path';
+import * as ec2 from 'aws-cdk-lib/aws-ec2';
+import * as cdk from 'aws-cdk-lib';
+import { IntegTest } from '@aws-cdk/integ-tests-alpha';
+import * as sagemaker from '../lib';
+
+/*
+ * Stack verification steps:
+ * aws sagemaker describe-endpoint-config --endpoint-config-name <endpoint config name>
+ *
+ * For instance-based endpoint config, the above command will result in the following output:
+ *   {
+ *     "EndpointConfigName": "EndpointConfig...",
+ *     "EndpointConfigArn": "arn:aws:sagemaker:...",
+ *     "ProductionVariants": [
+ *         {
+ *             "VariantName": "firstVariant",
+ *             "ModelName": "ModelWithArtifactAndVpcModel...",
+ *             "InitialInstanceCount": 1,
+ *             "InstanceType": "ml.m5.large",
+ *             "InitialVariantWeight": 1.0
+ *         },
+ *         {
+ *             "VariantName": "secondVariant",
+ *             "ModelName": "ModelWithArtifactAndVpcModel...",
+ *             "InitialInstanceCount": 1,
+ *             "InstanceType": "ml.t2.medium",
+ *             "InitialVariantWeight": 1.0
+ *         },
+ *         {
+ *             "VariantName": "thirdVariant",
+ *             "ModelName": "ModelWithoutArtifactAndVpcModel...",
+ *             "InitialInstanceCount": 1,
+ *             "InstanceType": "ml.t2.medium",
+ *             "InitialVariantWeight": 2.0
+ *         }
+ *     ],
+ *     "CreationTime": "..."
+ *   }
+ *
+ * For serverless endpoint config, the command will show:
+ *   {
+ *     "EndpointConfigName": "ServerlessEndpointConfig...",
+ *     "EndpointConfigArn": "arn:aws:sagemaker:...",
+ *     "ProductionVariants": [
+ *         {
+ *             "VariantName": "serverlessVariant",
+ *             "ModelName": "ModelWithoutArtifactAndVpcModel...",
+ *             "InitialVariantWeight": 1.0,
+ *             "ServerlessConfig": {
+ *                 "MaxConcurrency": 10,
+ *                 "MemorySizeInMB": 2048,
+ *                 "ProvisionedConcurrency": 5
+ *             }
+ *         }
+ *     ],
+ *     "CreationTime": "..."
+ *   }
+ */
+
+const app = new cdk.App();
+const stack = new cdk.Stack(app, 'aws-cdk-sagemaker-endpointconfig');
+
+const image = sagemaker.ContainerImage.fromAsset(path.join(__dirname, 'test-image'));
+const modelData = sagemaker.ModelData.fromAsset(path.join(__dirname, 'test-artifacts', 'valid-artifact.tar.gz'));
+
+const modelWithArtifactAndVpc = new sagemaker.Model(stack, 'ModelWithArtifactAndVpc', {
+    containers: [{ image, modelData }],
+    vpc: new ec2.Vpc(stack, 'VPC', { restrictDefaultSecurityGroup: false }),
+});
+const modelWithoutArtifactAndVpc = new sagemaker.Model(stack, 'ModelWithoutArtifactAndVpc', {
+    containers: [{ image }],
+});
+
+// Test instance-based endpoint configuration
+const endpointConfig = new sagemaker.EndpointConfig(stack, 'EndpointConfig', {
+    instanceProductionVariants: [
+        {
+            model: modelWithArtifactAndVpc,
+            variantName: 'firstVariant',
+            instanceType: sagemaker.InstanceType.M5_LARGE,
+        },
+        {
+            model: modelWithArtifactAndVpc,
+            variantName: 'secondVariant',
+        },
+    ],
+});
+endpointConfig.addInstanceProductionVariant({
+    model: modelWithoutArtifactAndVpc,
+    variantName: 'thirdVariant',
+    initialVariantWeight: 2.0,
+});
+
+// Test serverless endpoint configuration with all properties
+new sagemaker.EndpointConfig(stack, 'ServerlessEndpointConfig', {
+    serverlessProductionVariant: {
+        model: modelWithoutArtifactAndVpc,
+        variantName: 'serverlessVariant',
+        maxConcurrency: 10,
+        memorySizeInMB: 2048,
+        provisionedConcurrency: 5,
+        initialVariantWeight: 1.0,
+    },
+});
+
+// Test serverless endpoint configuration with minimal properties
+new sagemaker.EndpointConfig(stack, 'MinimalServerlessEndpointConfig', {
+    serverlessProductionVariant: {
+        model: modelWithoutArtifactAndVpc,
+        variantName: 'minimalServerlessVariant',
+        maxConcurrency: 1,
+        memorySizeInMB: 1024,
+        // No provisionedConcurrency - testing optional property
+    },
+});
+
+// Test serverless endpoint configuration with boundary values
+new sagemaker.EndpointConfig(stack, 'BoundaryServerlessEndpointConfig', {
+    serverlessProductionVariant: {
+        model: modelWithoutArtifactAndVpc,
+        variantName: 'boundaryServerlessVariant',
+        maxConcurrency: 200, // Maximum allowed
+        memorySizeInMB: 6144, // Maximum allowed
+        provisionedConcurrency: 200, // Maximum allowed (equal to maxConcurrency)
+    },
+});
+
+new IntegTest(app, 'integtest-endpointconfig', {
+    testCases: [stack],
+});

packages/@aws-cdk/aws-sagemaker-alpha/README.md

@@ -214,6 +214,38 @@ const endpointConfig = new sagemaker.EndpointConfig(this, 'EndpointConfig', {
 });
 ```
 
+### Serverless Inference
+
+Amazon SageMaker Serverless Inference is a purpose-built inference option that makes it easy for you to deploy and scale ML models. Serverless endpoints automatically launch compute resources and scale them in and out depending on traffic, eliminating the need to choose instance types or manage scaling policies.
+
+To create a serverless endpoint configuration, use the `serverlessProductionVariant` property:
+
+```typescript
+import * as sagemaker from '@aws-cdk/aws-sagemaker-alpha';
+
+declare const model: sagemaker.Model;
+
+const endpointConfig = new sagemaker.EndpointConfig(this, 'ServerlessEndpointConfig', {
+  serverlessProductionVariant: {
+    model: model,
+    variantName: 'serverlessVariant',
+    maxConcurrency: 10,
+    memorySizeInMB: 2048,
+    provisionedConcurrency: 5, // optional
+  },
+});
+```
+
+Serverless inference is ideal for workloads with intermittent or unpredictable traffic patterns. You can configure:
+
+- `maxConcurrency`: Maximum concurrent invocations (1-200)
+- `memorySizeInMB`: Memory allocation in 1GB increments (1024, 2048, 3072, 4096, 5120, or 6144 MB)
+- `provisionedConcurrency`: Optional pre-warmed capacity to reduce cold starts
+
+**Note**: Provisioned concurrency incurs charges even when the endpoint is not processing requests. Use it only when you need to minimize cold start latency.
+
+You cannot mix serverless and instance-based variants in the same endpoint configuration.
+
 ### Endpoint
 
 When you create an endpoint from an `EndpointConfig`, Amazon SageMaker launches the ML compute