@aws/ml-container-creator 0.6.1 → 0.8.0

package/templates/marketplace/deploy

@@ -0,0 +1,890 @@
+#!/bin/bash
+# Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Marketplace model package deployment script.
+# Deploys a pre-built AWS Marketplace model package using CreateModel with ModelPackageName.
+# No build, push, or submit steps — the vendor provides the container and weights.
+
+set -e
+set -u
+set -o pipefail
+
+# Parse flags
+FORCE_NEW=false
+FORCE_IC=false
+while [ $# -gt 0 ]; do
+    case "$1" in
+        --force) FORCE_NEW=true; shift ;;
+        --force-ic) FORCE_IC=true; shift ;;
+        --help|-h)
+            echo "Usage: ./do/deploy [--force] [--force-ic]"
+            echo ""
+            echo "Options:"
+            echo "  --force      Create a new deployment, even if one already exists."
+            echo "  --force-ic   Recreate the endpoint configuration on the existing endpoint."
+            echo ""
+            echo "Without flags, deploy resumes from the last run."
+            exit 0
+            ;;
+        *)
+            echo "❌ Unknown option: $1"
+            echo "   Run ./do/deploy --help for usage."
+            exit 1
+            ;;
+    esac
+done
+
+# Source configuration
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+source "${SCRIPT_DIR}/config"
+
+echo "🚀 Deploying Marketplace Model Package"
+echo "   Project: ${PROJECT_NAME}"
+echo "   Deployment config: marketplace"
+echo "   Region: ${AWS_REGION}"
+echo "   Model package: ${MODEL_PACKAGE_ARN}"
+echo "   Deployment target: ${DEPLOYMENT_TARGET}"
+echo "   Instance type: ${INSTANCE_TYPE}"
+<% if (deploymentTarget === 'async-inference') { %>
+echo "   S3 output: ${ASYNC_S3_OUTPUT_PATH}"
+echo "   SNS success: ${ASYNC_SNS_SUCCESS_TOPIC}"
+echo "   SNS error: ${ASYNC_SNS_ERROR_TOPIC}"
+<% if (asyncMaxConcurrentInvocations) { %>
+echo "   Max concurrent: ${ASYNC_MAX_CONCURRENT_INVOCATIONS}"
+<% } %>
+<% } else if (deploymentTarget === 'batch-transform') { %>
+echo "   Instance count: ${BATCH_INSTANCE_COUNT}"
+echo "   S3 input: ${BATCH_INPUT_PATH}"
+echo "   S3 output: ${BATCH_OUTPUT_PATH}"
+echo "   Split type: ${BATCH_SPLIT_TYPE}"
+echo "   Strategy: ${BATCH_STRATEGY}"
+<% } %>
+
+# Check AWS credentials
+echo "🔍 Validating AWS credentials..."
+if ! aws sts get-caller-identity &> /dev/null; then
+    echo "❌ AWS credentials not configured"
+    echo "   Run: aws configure"
+    echo "   Or set AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY environment variables"
+    exit 4
+fi
+
+AWS_ACCOUNT_ID=$(aws sts get-caller-identity --query Account --output text)
+echo "✅ AWS credentials validated (Account: ${AWS_ACCOUNT_ID})"
+
+# Source shared helpers
+source "${SCRIPT_DIR}/lib/wait.sh"
+source "${SCRIPT_DIR}/lib/endpoint-config.sh"
+
+# Validate execution role ARN
+if [ -z "${ROLE_ARN:-}" ]; then
+    echo "❌ Execution role ARN not provided"
+    echo ""
+    echo "Usage:"
+    echo "  export ROLE_ARN=arn:aws:iam::ACCOUNT_ID:role/YOUR_ROLE"
+    echo "  ./do/deploy"
+    echo ""
+    echo "Or set ROLE_ARN in do/config"
+    echo ""
+    echo "The execution role must have permissions for:"
+    echo "  • SageMaker model and endpoint management"
+    echo "  • Access to the Marketplace model package"
+    echo "  • CloudWatch Logs"
+    exit 3
+fi
+
+echo "   Using execution role: ${ROLE_ARN}"
+
+<% if (deploymentTarget === 'realtime-inference') { %>
+# ============================================================
+# SageMaker Real-Time Inference Deployment (Model-Based)
+# Marketplace packages use the classic model-based flow:
+# CreateModel(ModelPackageName) → CreateEndpointConfig → CreateEndpoint
+# ============================================================
+
+# ============================================================
+# Idempotency: check for existing deployment from a previous run
+# ============================================================
+SKIP_TO=""
+
+if [ "${FORCE_NEW}" = true ]; then
+    echo "🔄 --force: ignoring previous deployment, creating new resources."
+elif [ -n "${ENDPOINT_NAME:-}" ]; then
+    echo "🔍 Checking for existing deployment: ${ENDPOINT_NAME}"
+
+    EP_STATUS=$(_get_endpoint_status "${ENDPOINT_NAME}")
+
+    case "${EP_STATUS}" in
+        InService)
+            echo "✅ Endpoint already InService: ${ENDPOINT_NAME}"
+            echo ""
+            echo "📋 Deployment is already live. Nothing to do."
+            echo "   Endpoint: ${ENDPOINT_NAME}"
+            echo ""
+            echo "🧪 Test your endpoint:"
+            echo "   ./do/test"
+            echo ""
+            echo "🧹 Clean up when done:"
+            echo "   ./do/clean endpoint"
+            exit 0
+            ;;
+        Creating|Updating)
+            echo "⏳ Endpoint still ${EP_STATUS}: ${ENDPOINT_NAME}"
+            SKIP_TO="wait_endpoint"
+            ;;
+        Failed)
+            echo "⚠️  Previous endpoint failed: ${ENDPOINT_NAME}"
+            echo "   Creating a new deployment. Clean up the failed endpoint with:"
+            echo "   ./do/clean endpoint"
+            echo ""
+            ;;
+        "")
+            echo "   Previous endpoint not found (may have been cleaned up). Creating new deployment."
+            ;;
+        *)
+            echo "   Endpoint in unexpected state: ${EP_STATUS}. Creating new deployment."
+            ;;
+    esac
+fi
+
+# ============================================================
+# Create resources (skip if resuming from wait)
+# ============================================================
+if [ -z "${SKIP_TO}" ]; then
+    TIMESTAMP=$(date +%s)
+    MODEL_NAME_SM="${PROJECT_NAME}-mkt-model-${TIMESTAMP}"
+    ENDPOINT_CONFIG_NAME="${PROJECT_NAME}-mkt-epc-${TIMESTAMP}"
+    ENDPOINT_NAME="${PROJECT_NAME}-mkt-ep-${TIMESTAMP}"
+
+    _update_config_var "ENDPOINT_NAME" "${ENDPOINT_NAME}"
+    _update_config_var "ENDPOINT_CONFIG_NAME" "${ENDPOINT_CONFIG_NAME}"
+    _update_config_var "SAGEMAKER_MODEL_NAME" "${MODEL_NAME_SM}"
+
+    # Step 1: Create SageMaker model from Marketplace model package
+    echo "📦 Creating SageMaker model from Marketplace package: ${MODEL_NAME_SM}"
+    if ! aws sagemaker create-model \
+        --model-name "${MODEL_NAME_SM}" \
+        --primary-container "{\"ModelPackageName\":\"${MODEL_PACKAGE_ARN}\"}" \
+        --execution-role-arn "${ROLE_ARN}" \
+        --region "${AWS_REGION}"; then
+
+        echo "❌ Failed to create model from package ARN. Check IAM permissions and subscription status."
+        echo "   Check that:"
+        echo "   • The model package ARN is correct: ${MODEL_PACKAGE_ARN}"
+        echo "   • Your Marketplace subscription is active"
+        echo "   • The execution role has permission to access the model package"
+        exit 4
+    fi
+
+    echo "✅ SageMaker model created: ${MODEL_NAME_SM}"
+
+    # Record model in manifest (non-blocking)
+    MODEL_ARN="arn:aws:sagemaker:${AWS_REGION}:${AWS_ACCOUNT_ID}:model/${MODEL_NAME_SM}"
+    ./do/manifest add \
+        --type sagemaker-model \
+        --id "${MODEL_ARN}" \
+        --project "${PROJECT_NAME}" \
+        --meta "{\"modelName\":\"${MODEL_NAME_SM}\",\"modelPackageArn\":\"${MODEL_PACKAGE_ARN}\",\"region\":\"${AWS_REGION}\"}" \
+        2>/dev/null || true
+
+    # Step 2: Create endpoint configuration
+    # Set MODEL_NAME_SM so endpoint-config.sh uses model-based flow (no --execution-role-arn on epc)
+    VARIANT_JSON="[{\"VariantName\":\"AllTraffic\",\"ModelName\":\"${MODEL_NAME_SM}\",\"InstanceType\":\"${INSTANCE_TYPE}\",\"InitialInstanceCount\":1}]"
+
+    echo "⚙️  Creating endpoint configuration: ${ENDPOINT_CONFIG_NAME}"
+    if ! aws sagemaker create-endpoint-config \
+        --endpoint-config-name "${ENDPOINT_CONFIG_NAME}" \
+        --production-variants "${VARIANT_JSON}" \
+        --region "${AWS_REGION}"; then
+
+        echo "❌ Failed to create endpoint configuration"
+        echo "   Check that:"
+        echo "   • The instance type is valid: ${INSTANCE_TYPE}"
+        echo "   • The instance type is available in region: ${AWS_REGION}"
+        echo "   • You have sufficient service quota for the instance type"
+        exit 4
+    fi
+
+    echo "✅ Endpoint configuration created: ${ENDPOINT_CONFIG_NAME}"
+
+    # Record endpoint config in manifest (non-blocking)
+    ENDPOINT_CONFIG_ARN="arn:aws:sagemaker:${AWS_REGION}:${AWS_ACCOUNT_ID}:endpoint-config/${ENDPOINT_CONFIG_NAME}"
+    ./do/manifest add \
+        --type sagemaker-endpoint-config \
+        --id "${ENDPOINT_CONFIG_ARN}" \
+        --project "${PROJECT_NAME}" \
+        --meta "{\"endpointConfigName\":\"${ENDPOINT_CONFIG_NAME}\",\"instanceType\":\"${INSTANCE_TYPE}\",\"region\":\"${AWS_REGION}\"}" \
+        2>/dev/null || true
+
+    # Step 3: Create endpoint
+    echo "🚀 Creating endpoint: ${ENDPOINT_NAME}"
+    if ! aws sagemaker create-endpoint \
+        --endpoint-name "${ENDPOINT_NAME}" \
+        --endpoint-config-name "${ENDPOINT_CONFIG_NAME}" \
+        --region "${AWS_REGION}"; then
+
+        echo "❌ Failed to create endpoint"
+        echo "   Check that:"
+        echo "   • Your IAM credentials have sagemaker:CreateEndpoint permission"
+        echo "   • You have sufficient service quota in region: ${AWS_REGION}"
+        exit 4
+    fi
+
+    echo "✅ Endpoint creation initiated: ${ENDPOINT_NAME}"
+
+    # Record endpoint in manifest (non-blocking)
+    ENDPOINT_ARN="arn:aws:sagemaker:${AWS_REGION}:${AWS_ACCOUNT_ID}:endpoint/${ENDPOINT_NAME}"
+    ./do/manifest add \
+        --type sagemaker-endpoint \
+        --id "${ENDPOINT_ARN}" \
+        --project "${PROJECT_NAME}" \
+        --meta "{\"endpointName\":\"${ENDPOINT_NAME}\",\"instanceType\":\"${INSTANCE_TYPE}\",\"region\":\"${AWS_REGION}\"}" \
+        2>/dev/null || true
+fi
+
+# ============================================================
+# Wait for endpoint
+# ============================================================
+if [ -z "${SKIP_TO}" ] || [ "${SKIP_TO}" = "wait_endpoint" ]; then
+    echo "⏳ Waiting for endpoint to reach InService status..."
+    echo "   This may take several minutes..."
+    echo "   If this times out, re-run ./do/deploy to resume."
+
+    wait_endpoint "${ENDPOINT_NAME}"
+fi
+
+echo "✅ Deployment complete!"
+echo ""
+echo "📋 Deployment Details:"
+echo "   Endpoint: ${ENDPOINT_NAME}"
+echo "   Endpoint Config: ${ENDPOINT_CONFIG_NAME}"
+echo "   Model: ${SAGEMAKER_MODEL_NAME:-${MODEL_NAME_SM:-N/A}}"
+echo "   Model Package: ${MODEL_PACKAGE_ARN}"
+echo "   Region: ${AWS_REGION}"
+echo "   Instance Type: ${INSTANCE_TYPE}"
+echo ""
+echo "📋 What's next?"
+echo "   • Test your endpoint:         ./do/test"
+<% if (typeof includeBenchmark !== 'undefined' && includeBenchmark) { %>
+echo "   • Benchmark performance:      ./do/benchmark"
+<% } %>
+echo "   • View endpoint status:       ./do/status"
+echo "   • Register this deployment:   ./do/register"
+echo "   • View logs:                  ./do/logs"
+echo "   • Clean up when done:         ./do/clean endpoint"
+
+<% } else if (deploymentTarget === 'async-inference') { %>
+# ============================================================
+# SageMaker Async Inference Deployment (Model-Based)
+# Marketplace packages use: CreateModel(ModelPackageName) → CreateEndpointConfig(AsyncInferenceConfig) → CreateEndpoint
+# ============================================================
+
+# ============================================================
+# Bootstrap async infrastructure (S3 bucket + SNS topics)
+# ============================================================
+
+# Extract bucket name from S3 output path
+ASYNC_S3_BUCKET=$(echo "${ASYNC_S3_OUTPUT_PATH}" | sed 's|s3://||' | cut -d'/' -f1)
+
+<% if (!asyncS3OutputPath) { %>
+# Bootstrap default S3 bucket (check-and-create)
+echo "🔍 Checking if S3 bucket exists: ${ASYNC_S3_BUCKET}"
+if ! aws s3api head-bucket --bucket "${ASYNC_S3_BUCKET}" --region "${AWS_REGION}" 2>/dev/null; then
+    echo "📦 Creating S3 bucket: ${ASYNC_S3_BUCKET}"
+    if [ "${AWS_REGION}" = "us-east-1" ]; then
+        if ! aws s3api create-bucket \
+            --bucket "${ASYNC_S3_BUCKET}" \
+            --region "${AWS_REGION}"; then
+            echo "❌ Failed to create S3 bucket: ${ASYNC_S3_BUCKET}"
+            exit 4
+        fi
+    else
+        if ! aws s3api create-bucket \
+            --bucket "${ASYNC_S3_BUCKET}" \
+            --region "${AWS_REGION}" \
+            --create-bucket-configuration LocationConstraint="${AWS_REGION}"; then
+            echo "❌ Failed to create S3 bucket: ${ASYNC_S3_BUCKET}"
+            exit 4
+        fi
+    fi
+    echo "✅ S3 bucket created: ${ASYNC_S3_BUCKET}"
+else
+    echo "✅ S3 bucket exists: ${ASYNC_S3_BUCKET}"
+fi
+<% } else { %>
+# Custom S3 output path provided — skip bucket creation
+echo "✅ Using custom S3 output path: ${ASYNC_S3_OUTPUT_PATH}"
+<% } %>
+
+# Extract topic name from SNS success topic ARN
+ASYNC_SNS_SUCCESS_TOPIC_NAME=$(echo "${ASYNC_SNS_SUCCESS_TOPIC}" | awk -F: '{print $NF}')
+
+<% if (!asyncSnsSuccessTopic) { %>
+# Bootstrap default SNS success topic (check-and-create)
+echo "🔍 Checking if SNS success topic exists: ${ASYNC_SNS_SUCCESS_TOPIC_NAME}"
+if ! aws sns get-topic-attributes --topic-arn "${ASYNC_SNS_SUCCESS_TOPIC}" --region "${AWS_REGION}" 2>/dev/null; then
+    echo "📦 Creating SNS success topic: ${ASYNC_SNS_SUCCESS_TOPIC_NAME}"
+    if ! aws sns create-topic \
+        --name "${ASYNC_SNS_SUCCESS_TOPIC_NAME}" \
+        --region "${AWS_REGION}" > /dev/null; then
+        echo "❌ Failed to create SNS success topic"
+        exit 4
+    fi
+    echo "✅ SNS success topic created: ${ASYNC_SNS_SUCCESS_TOPIC_NAME}"
+else
+    echo "✅ SNS success topic exists: ${ASYNC_SNS_SUCCESS_TOPIC_NAME}"
+fi
+<% } else { %>
+# Custom SNS success topic ARN provided — skip topic creation
+echo "✅ Using custom SNS success topic: ${ASYNC_SNS_SUCCESS_TOPIC}"
+<% } %>
+
+# Extract topic name from SNS error topic ARN
+ASYNC_SNS_ERROR_TOPIC_NAME=$(echo "${ASYNC_SNS_ERROR_TOPIC}" | awk -F: '{print $NF}')
+
+<% if (!asyncSnsErrorTopic) { %>
+# Bootstrap default SNS error topic (check-and-create)
+echo "🔍 Checking if SNS error topic exists: ${ASYNC_SNS_ERROR_TOPIC_NAME}"
+if ! aws sns get-topic-attributes --topic-arn "${ASYNC_SNS_ERROR_TOPIC}" --region "${AWS_REGION}" 2>/dev/null; then
+    echo "📦 Creating SNS error topic: ${ASYNC_SNS_ERROR_TOPIC_NAME}"
+    if ! aws sns create-topic \
+        --name "${ASYNC_SNS_ERROR_TOPIC_NAME}" \
+        --region "${AWS_REGION}" > /dev/null; then
+        echo "❌ Failed to create SNS error topic"
+        exit 4
+    fi
+    echo "✅ SNS error topic created: ${ASYNC_SNS_ERROR_TOPIC_NAME}"
+else
+    echo "✅ SNS error topic exists: ${ASYNC_SNS_ERROR_TOPIC_NAME}"
+fi
+<% } else { %>
+# Custom SNS error topic ARN provided — skip topic creation
+echo "✅ Using custom SNS error topic: ${ASYNC_SNS_ERROR_TOPIC}"
+<% } %>
+
+# ============================================================
+# Idempotency: check for existing deployment from a previous run
+# ============================================================
+SKIP_TO=""
+
+if [ "${FORCE_NEW}" = true ]; then
+    echo "🔄 --force: ignoring previous deployment, creating new resources."
+elif [ -n "${ENDPOINT_NAME:-}" ]; then
+    echo "🔍 Checking for existing deployment: ${ENDPOINT_NAME}"
+
+    EP_STATUS=$(_get_endpoint_status "${ENDPOINT_NAME}")
+
+    case "${EP_STATUS}" in
+        InService)
+            echo "✅ Async endpoint already InService: ${ENDPOINT_NAME}"
+            echo ""
+            echo "📋 Deployment is already live. Nothing to do."
+            echo "   Endpoint: ${ENDPOINT_NAME}"
+            echo ""
+            echo "🧪 Test your async endpoint:"
+            echo "   ./do/test"
+            echo ""
+            echo "🧹 Clean up when done:"
+            echo "   ./do/clean endpoint"
+            exit 0
+            ;;
+        Creating|Updating)
+            echo "⏳ Endpoint still ${EP_STATUS}: ${ENDPOINT_NAME}"
+            SKIP_TO="wait_endpoint"
+            ;;
+        Failed)
+            echo "⚠️  Previous endpoint failed: ${ENDPOINT_NAME}"
+            echo "   Creating a new deployment. Clean up the failed endpoint with:"
+            echo "   ./do/clean endpoint"
+            echo ""
+            ;;
+        "")
+            echo "   Previous endpoint not found (may have been cleaned up). Creating new deployment."
+            ;;
+        *)
+            echo "   Endpoint in unexpected state: ${EP_STATUS}. Creating new deployment."
+            ;;
+    esac
+fi
+
+# ============================================================
+# Create async resources (skip if resuming from wait)
+# ============================================================
+if [ -z "${SKIP_TO}" ]; then
+    TIMESTAMP=$(date +%s)
+    MODEL_NAME_SM="${PROJECT_NAME}-mkt-async-model-${TIMESTAMP}"
+    ENDPOINT_CONFIG_NAME="${PROJECT_NAME}-mkt-async-epc-${TIMESTAMP}"
+    ENDPOINT_NAME="${PROJECT_NAME}-mkt-async-ep-${TIMESTAMP}"
+
+    _update_config_var "ENDPOINT_NAME" "${ENDPOINT_NAME}"
+    _update_config_var "ENDPOINT_CONFIG_NAME" "${ENDPOINT_CONFIG_NAME}"
+    _update_config_var "SAGEMAKER_MODEL_NAME" "${MODEL_NAME_SM}"
+
+    # Step 1: Create SageMaker model from Marketplace model package
+    echo "📦 Creating SageMaker model from Marketplace package: ${MODEL_NAME_SM}"
+    if ! aws sagemaker create-model \
+        --model-name "${MODEL_NAME_SM}" \
+        --primary-container "{\"ModelPackageName\":\"${MODEL_PACKAGE_ARN}\"}" \
+        --execution-role-arn "${ROLE_ARN}" \
+        --region "${AWS_REGION}"; then
+
+        echo "❌ Failed to create model from package ARN. Check IAM permissions and subscription status."
+        echo "   Check that:"
+        echo "   • The model package ARN is correct: ${MODEL_PACKAGE_ARN}"
+        echo "   • Your Marketplace subscription is active"
+        echo "   • The execution role has permission to access the model package"
+        exit 4
+    fi
+
+    echo "✅ SageMaker model created: ${MODEL_NAME_SM}"
+
+    # Record model in manifest (non-blocking)
+    MODEL_ARN="arn:aws:sagemaker:${AWS_REGION}:${AWS_ACCOUNT_ID}:model/${MODEL_NAME_SM}"
+    ./do/manifest add \
+        --type sagemaker-model \
+        --id "${MODEL_ARN}" \
+        --project "${PROJECT_NAME}" \
+        --meta "{\"modelName\":\"${MODEL_NAME_SM}\",\"modelPackageArn\":\"${MODEL_PACKAGE_ARN}\",\"region\":\"${AWS_REGION}\"}" \
+        2>/dev/null || true
+
+    # Step 2: Build production variant and AsyncInferenceConfig
+    VARIANT_JSON="[{\"VariantName\":\"AllTraffic\",\"ModelName\":\"${MODEL_NAME_SM}\",\"InstanceType\":\"${INSTANCE_TYPE}\",\"InitialInstanceCount\":1}]"
+
+    ASYNC_CONFIG="{\"OutputConfig\":{\"S3OutputPath\":\"${ASYNC_S3_OUTPUT_PATH}\",\"NotificationConfig\":{\"SuccessTopic\":\"${ASYNC_SNS_SUCCESS_TOPIC}\",\"ErrorTopic\":\"${ASYNC_SNS_ERROR_TOPIC}\"}}"
+    if [ -n "${ASYNC_MAX_CONCURRENT_INVOCATIONS:-}" ]; then
+        ASYNC_CONFIG="${ASYNC_CONFIG},\"ClientConfig\":{\"MaxConcurrentInvocationsPerInstance\":${ASYNC_MAX_CONCURRENT_INVOCATIONS}}"
+    fi
+    ASYNC_CONFIG="${ASYNC_CONFIG}}"
+
+    # Step 3: Create endpoint configuration with AsyncInferenceConfig
+    echo "⚙️  Creating async endpoint configuration: ${ENDPOINT_CONFIG_NAME}"
+    if ! aws sagemaker create-endpoint-config \
+        --endpoint-config-name "${ENDPOINT_CONFIG_NAME}" \
+        --production-variants "${VARIANT_JSON}" \
+        --async-inference-config "${ASYNC_CONFIG}" \
+        --region "${AWS_REGION}"; then
+
+        echo "❌ Failed to create async endpoint configuration"
+        echo "   Check that:"
+        echo "   • The S3 output path is accessible: ${ASYNC_S3_OUTPUT_PATH}"
+        echo "   • The IAM role has s3:PutObject permission on the output path"
+        echo "   • The instance type is valid: ${INSTANCE_TYPE}"
+        echo "   • You have sufficient service quota for the instance type"
+        exit 4
+    fi
+
+    echo "✅ Async endpoint configuration created: ${ENDPOINT_CONFIG_NAME}"
+
+    # Record endpoint config in manifest (non-blocking)
+    ENDPOINT_CONFIG_ARN="arn:aws:sagemaker:${AWS_REGION}:${AWS_ACCOUNT_ID}:endpoint-config/${ENDPOINT_CONFIG_NAME}"
+    ./do/manifest add \
+        --type sagemaker-endpoint-config \
+        --id "${ENDPOINT_CONFIG_ARN}" \
+        --project "${PROJECT_NAME}" \
+        --meta "{\"endpointConfigName\":\"${ENDPOINT_CONFIG_NAME}\",\"instanceType\":\"${INSTANCE_TYPE}\",\"region\":\"${AWS_REGION}\"}" \
+        2>/dev/null || true
+
+    # Step 4: Create endpoint
+    echo "🚀 Creating async endpoint: ${ENDPOINT_NAME}"
+    if ! aws sagemaker create-endpoint \
+        --endpoint-name "${ENDPOINT_NAME}" \
+        --endpoint-config-name "${ENDPOINT_CONFIG_NAME}" \
+        --region "${AWS_REGION}"; then
+
+        echo "❌ Failed to create async endpoint"
+        echo "   Check that:"
+        echo "   • Your IAM credentials have sagemaker:CreateEndpoint permission"
+        echo "   • You have sufficient service quota in region: ${AWS_REGION}"
+        exit 4
+    fi
+
+    echo "✅ Async endpoint creation initiated: ${ENDPOINT_NAME}"
+
+    # Record endpoint in manifest (non-blocking)
+    ENDPOINT_ARN="arn:aws:sagemaker:${AWS_REGION}:${AWS_ACCOUNT_ID}:endpoint/${ENDPOINT_NAME}"
+    ./do/manifest add \
+        --type sagemaker-endpoint \
+        --id "${ENDPOINT_ARN}" \
+        --project "${PROJECT_NAME}" \
+        --meta "{\"endpointName\":\"${ENDPOINT_NAME}\",\"instanceType\":\"${INSTANCE_TYPE}\",\"region\":\"${AWS_REGION}\"}" \
+        2>/dev/null || true
+fi
+
+# ============================================================
+# Wait for endpoint
+# ============================================================
+if [ -z "${SKIP_TO}" ] || [ "${SKIP_TO}" = "wait_endpoint" ]; then
+    echo "⏳ Waiting for async endpoint to reach InService status..."
+    echo "   This may take several minutes..."
+    echo "   If this times out, re-run ./do/deploy to resume."
+
+    wait_endpoint "${ENDPOINT_NAME}"
+fi
+
+echo "✅ Async deployment complete!"
+echo ""
+echo "📋 Deployment Details:"
+echo "   Endpoint: ${ENDPOINT_NAME}"
+echo "   Endpoint Config: ${ENDPOINT_CONFIG_NAME}"
+echo "   Model: ${SAGEMAKER_MODEL_NAME:-${MODEL_NAME_SM:-N/A}}"
+echo "   Model Package: ${MODEL_PACKAGE_ARN}"
+echo "   Region: ${AWS_REGION}"
+echo "   Instance Type: ${INSTANCE_TYPE}"
+echo "   S3 Output: ${ASYNC_S3_OUTPUT_PATH}"
+echo "   SNS Success: ${ASYNC_SNS_SUCCESS_TOPIC}"
+echo "   SNS Error: ${ASYNC_SNS_ERROR_TOPIC}"
+echo ""
+echo "📋 What's next?"
+echo "   • Test your async endpoint:   ./do/test"
+echo "   • Check async output:         aws s3 ls ${ASYNC_S3_OUTPUT_PATH}"
+<% if (typeof includeBenchmark !== 'undefined' && includeBenchmark) { %>
+echo "   • Benchmark performance:      ./do/benchmark"
+<% } %>
+echo "   • Register this deployment:   ./do/register"
+echo "   • View logs:                  ./do/logs"
+echo "   • Clean up when done:         ./do/clean endpoint"
+
+<% } else if (deploymentTarget === 'batch-transform') { %>
+# ============================================================
+# SageMaker Batch Transform Deployment
+# Marketplace packages use: CreateModel(ModelPackageName) → CreateTransformJob
+# ============================================================
+
+# Validate S3 input path
+if [ -z "${BATCH_INPUT_PATH:-}" ]; then
+    echo "❌ S3 input path not provided"
+    echo ""
+    echo "Set BATCH_INPUT_PATH in do/config or provide via CLI:"
+    echo "  export BATCH_INPUT_PATH=s3://my-bucket/input/"
+    echo "  ./do/deploy"
+    exit 3
+fi
+
+if [[ "${BATCH_INPUT_PATH}" != s3://* ]]; then
+    echo "❌ S3 input path must start with s3://"
+    echo "   Current value: ${BATCH_INPUT_PATH}"
+    exit 3
+fi
+
+# Validate S3 output path
+if [ -z "${BATCH_OUTPUT_PATH:-}" ]; then
+    echo "❌ S3 output path not provided"
+    echo ""
+    echo "Set BATCH_OUTPUT_PATH in do/config or provide via CLI:"
+    echo "  export BATCH_OUTPUT_PATH=s3://my-bucket/output/"
+    echo "  ./do/deploy"
+    exit 3
+fi
+
+if [[ "${BATCH_OUTPUT_PATH}" != s3://* ]]; then
+    echo "❌ S3 output path must start with s3://"
+    echo "   Current value: ${BATCH_OUTPUT_PATH}"
+    exit 3
+fi
+
+# ============================================================
+# Bootstrap S3 buckets for batch transform
+# ============================================================
+
+BATCH_INPUT_BUCKET=$(echo "${BATCH_INPUT_PATH}" | sed 's|s3://||' | cut -d'/' -f1)
+BATCH_OUTPUT_BUCKET=$(echo "${BATCH_OUTPUT_PATH}" | sed 's|s3://||' | cut -d'/' -f1)
+
+<% if (!batchInputPath) { %>
+# Bootstrap default S3 input bucket (check-and-create)
+echo "🔍 Checking if S3 input bucket exists: ${BATCH_INPUT_BUCKET}"
+if ! aws s3api head-bucket --bucket "${BATCH_INPUT_BUCKET}" --region "${AWS_REGION}" 2>/dev/null; then
+    echo "📦 Creating S3 input bucket: ${BATCH_INPUT_BUCKET}"
+    if [ "${AWS_REGION}" = "us-east-1" ]; then
+        if ! aws s3api create-bucket \
+            --bucket "${BATCH_INPUT_BUCKET}" \
+            --region "${AWS_REGION}"; then
+            echo "❌ Failed to create S3 input bucket: ${BATCH_INPUT_BUCKET}"
+            exit 4
+        fi
+    else
+        if ! aws s3api create-bucket \
+            --bucket "${BATCH_INPUT_BUCKET}" \
+            --region "${AWS_REGION}" \
+            --create-bucket-configuration LocationConstraint="${AWS_REGION}"; then
+            echo "❌ Failed to create S3 input bucket: ${BATCH_INPUT_BUCKET}"
+            exit 4
+        fi
+    fi
+    echo "✅ S3 input bucket created: ${BATCH_INPUT_BUCKET}"
+else
+    echo "✅ S3 input bucket exists: ${BATCH_INPUT_BUCKET}"
+fi
+
+# Upload sample input file if the input prefix is empty
+EXISTING_OBJECTS=$(aws s3 ls "${BATCH_INPUT_PATH}" --region "${AWS_REGION}" 2>/dev/null | head -1 || true)
+if [ -z "${EXISTING_OBJECTS}" ]; then
+    echo "📄 Uploading sample input file to ${BATCH_INPUT_PATH}"
+    echo '{"inputs": "What is machine learning?", "parameters": {"max_new_tokens": 50}}' | aws s3 cp - "${BATCH_INPUT_PATH}sample.jsonl" --region "${AWS_REGION}"
+    echo "✅ Sample input uploaded: ${BATCH_INPUT_PATH}sample.jsonl"
+    echo "   ⚠️  Replace this with your actual input data before running production jobs"
+fi
+<% } else { %>
+# Custom S3 input path provided — skip bucket creation
+echo "✅ Using custom S3 input path: ${BATCH_INPUT_PATH}"
+<% } %>
+
+<% if (!batchOutputPath) { %>
+# Bootstrap default S3 output bucket (check-and-create, may be same as input)
+if [ "${BATCH_OUTPUT_BUCKET}" != "${BATCH_INPUT_BUCKET}" ]; then
+    echo "🔍 Checking if S3 output bucket exists: ${BATCH_OUTPUT_BUCKET}"
+    if ! aws s3api head-bucket --bucket "${BATCH_OUTPUT_BUCKET}" --region "${AWS_REGION}" 2>/dev/null; then
+        echo "📦 Creating S3 output bucket: ${BATCH_OUTPUT_BUCKET}"
+        if [ "${AWS_REGION}" = "us-east-1" ]; then
+            if ! aws s3api create-bucket \
+                --bucket "${BATCH_OUTPUT_BUCKET}" \
+                --region "${AWS_REGION}"; then
+                echo "❌ Failed to create S3 output bucket: ${BATCH_OUTPUT_BUCKET}"
+                exit 4
+            fi
+        else
+            if ! aws s3api create-bucket \
+                --bucket "${BATCH_OUTPUT_BUCKET}" \
+                --region "${AWS_REGION}" \
+                --create-bucket-configuration LocationConstraint="${AWS_REGION}"; then
+                echo "❌ Failed to create S3 output bucket: ${BATCH_OUTPUT_BUCKET}"
+                exit 4
+            fi
+        fi
+        echo "✅ S3 output bucket created: ${BATCH_OUTPUT_BUCKET}"
+    else
+        echo "✅ S3 output bucket exists: ${BATCH_OUTPUT_BUCKET}"
+    fi
+else
+    echo "✅ S3 output bucket same as input: ${BATCH_OUTPUT_BUCKET}"
+fi
+<% } else { %>
+# Custom S3 output path provided — skip bucket creation
+echo "✅ Using custom S3 output path: ${BATCH_OUTPUT_PATH}"
+<% } %>
+
+# ============================================================
+# Check for previous transform job still running
+# ============================================================
+if [ "${FORCE_NEW}" != true ] && [ -n "${TRANSFORM_JOB_NAME:-}" ]; then
+    echo "🔍 Checking previous transform job: ${TRANSFORM_JOB_NAME}"
+    PREV_JOB_STATUS=$(aws sagemaker describe-transform-job \
+        --transform-job-name "${TRANSFORM_JOB_NAME}" \
+        --region "${AWS_REGION}" \
+        --query "TransformJobStatus" \
+        --output text 2>/dev/null || echo "")
+
+    case "${PREV_JOB_STATUS}" in
+        InProgress)
+            echo "⚠️  Previous transform job is still running: ${TRANSFORM_JOB_NAME}"
+            echo "   Wait for it to complete, or stop it with:"
+            echo "   aws sagemaker stop-transform-job --transform-job-name ${TRANSFORM_JOB_NAME} --region ${AWS_REGION}"
+            echo ""
+            echo "   Use --force to create a new job anyway."
+            exit 4
+            ;;
+        Completed)
+            echo "✅ Previous transform job completed: ${TRANSFORM_JOB_NAME}"
+            echo "   Creating a new job. Results from the previous job are in:"
+            echo "   ${BATCH_OUTPUT_PATH}"
+            echo ""
+            ;;
+        *)
+            # Failed, Stopped, or not found — proceed with new job
+            ;;
+    esac
+fi
+
+# Generate unique names with timestamp
+TIMESTAMP=$(date +%s)
+MODEL_NAME_SM="${PROJECT_NAME}-mkt-batch-model-${TIMESTAMP}"
+TRANSFORM_JOB_NAME="${PROJECT_NAME}-mkt-batch-job-${TIMESTAMP}"
+
+_update_config_var "TRANSFORM_JOB_NAME" "${TRANSFORM_JOB_NAME}"
+_update_config_var "SAGEMAKER_MODEL_NAME" "${MODEL_NAME_SM}"
+
+# Step 1: Create SageMaker model from Marketplace model package
+echo "📦 Creating SageMaker model from Marketplace package: ${MODEL_NAME_SM}"
+if ! aws sagemaker create-model \
+    --model-name "${MODEL_NAME_SM}" \
+    --primary-container "{\"ModelPackageName\":\"${MODEL_PACKAGE_ARN}\"}" \
+    --execution-role-arn "${ROLE_ARN}" \
+    --region "${AWS_REGION}"; then
+
+    echo "❌ Failed to create model from package ARN. Check IAM permissions and subscription status."
+    echo "   Check that:"
+    echo "   • The model package ARN is correct: ${MODEL_PACKAGE_ARN}"
+    echo "   • Your Marketplace subscription is active"
+    echo "   • The execution role has permission to access the model package"
+    exit 4
+fi
+
+echo "✅ SageMaker model created: ${MODEL_NAME_SM}"
+
+# Record model in manifest (non-blocking)
+MODEL_ARN="arn:aws:sagemaker:${AWS_REGION}:${AWS_ACCOUNT_ID}:model/${MODEL_NAME_SM}"
+./do/manifest add \
+    --type sagemaker-model \
+    --id "${MODEL_ARN}" \
+    --project "${PROJECT_NAME}" \
+    --meta "{\"modelName\":\"${MODEL_NAME_SM}\",\"modelPackageArn\":\"${MODEL_PACKAGE_ARN}\",\"region\":\"${AWS_REGION}\"}" \
+    2>/dev/null || true
+
+# Step 2: Build transform job JSON
+TRANSFORM_JOB_JSON="{
+    \"TransformJobName\": \"${TRANSFORM_JOB_NAME}\",
+    \"ModelName\": \"${MODEL_NAME_SM}\",
+    \"TransformInput\": {
+        \"DataSource\": {
+            \"S3DataSource\": {
+                \"S3DataType\": \"S3Prefix\",
+                \"S3Uri\": \"${BATCH_INPUT_PATH}\"
+            }
+        },
+        \"ContentType\": \"application/json\",
+        \"SplitType\": \"${BATCH_SPLIT_TYPE}\"
+    },
+    \"TransformOutput\": {
+        \"S3OutputPath\": \"${BATCH_OUTPUT_PATH}\"
+        $([ "${BATCH_JOIN_SOURCE:-None}" = "Input" ] && echo ",\"Accept\": \"application/json\", \"AssembleWith\": \"${BATCH_SPLIT_TYPE}\"")
+    },
+    \"TransformResources\": {
+        \"InstanceType\": \"${INSTANCE_TYPE}\",
+        \"InstanceCount\": ${BATCH_INSTANCE_COUNT}
+    },
+    \"MaxConcurrentTransforms\": ${BATCH_MAX_CONCURRENT_TRANSFORMS:-1},
+    \"MaxPayloadInMB\": ${BATCH_MAX_PAYLOAD_IN_MB:-6},
+    \"BatchStrategy\": \"${BATCH_STRATEGY}\"
+    $([ "${BATCH_JOIN_SOURCE:-None}" = "Input" ] && echo ",\"DataProcessing\": { \"JoinSource\": \"Input\" }")
+}"
+
+# Step 3: Create transform job
+echo "🚀 Creating transform job: ${TRANSFORM_JOB_NAME}"
+if ! aws sagemaker create-transform-job \
+    --cli-input-json "${TRANSFORM_JOB_JSON}" \
+    --region "${AWS_REGION}"; then
+
+    echo "❌ Failed to create transform job"
+    echo "   Check that:"
+    echo "   • The S3 input path exists and is accessible: ${BATCH_INPUT_PATH}"
+    echo "   • The S3 output path is writable: ${BATCH_OUTPUT_PATH}"
+    echo "   • The instance type is valid: ${INSTANCE_TYPE}"
+    echo "   • You have sufficient service quota for the instance type"
+    exit 4
+fi
+
+echo "✅ Transform job created: ${TRANSFORM_JOB_NAME}"
+
+# Record transform job in manifest (non-blocking)
+TRANSFORM_JOB_ARN="arn:aws:sagemaker:${AWS_REGION}:${AWS_ACCOUNT_ID}:transform-job/${TRANSFORM_JOB_NAME}"
+./do/manifest add \
+    --type sagemaker-transform-job \
+    --id "${TRANSFORM_JOB_ARN}" \
+    --project "${PROJECT_NAME}" \
+    --meta "{\"transformJobName\":\"${TRANSFORM_JOB_NAME}\",\"modelName\":\"${MODEL_NAME_SM}\",\"instanceType\":\"${INSTANCE_TYPE}\",\"region\":\"${AWS_REGION}\"}" \
+    2>/dev/null || true
+
+# Step 4: Poll transform job status until completion or failure
+echo "⏳ Waiting for transform job to complete..."
+echo "   This may take several minutes depending on dataset size..."
+echo "   If this times out, check status with:"
+echo "   aws sagemaker describe-transform-job --transform-job-name ${TRANSFORM_JOB_NAME} --region ${AWS_REGION}"
+echo ""
+
+while true; do
+    JOB_STATUS=$(aws sagemaker describe-transform-job \
+        --transform-job-name "${TRANSFORM_JOB_NAME}" \
+        --region "${AWS_REGION}" \
+        --query "TransformJobStatus" \
+        --output text 2>&1) || {
+        if echo "${JOB_STATUS}" | grep -qi "expired\|token"; then
+            echo ""
+            echo "⚠️  Credentials expired, but the transform job is still running."
+            echo "   Refresh your credentials and check status with:"
+            echo "   aws sagemaker describe-transform-job --transform-job-name ${TRANSFORM_JOB_NAME} --region ${AWS_REGION} --query TransformJobStatus"
+            exit 4
+        fi
+        echo "❌ Failed to describe transform job: ${TRANSFORM_JOB_NAME}"
+        echo "   Error: ${JOB_STATUS}"
+        exit 4
+    }
+
+    case "${JOB_STATUS}" in
+        Completed)
+            echo "✅ Transform job completed successfully!"
+            break
+            ;;
+        Failed)
+            FAILURE_REASON=$(aws sagemaker describe-transform-job \
+                --transform-job-name "${TRANSFORM_JOB_NAME}" \
+                --region "${AWS_REGION}" \
+                --query "FailureReason" \
+                --output text 2>/dev/null || echo "Unknown")
+            echo "❌ Transform job failed"
+            echo "   Reason: ${FAILURE_REASON}"
+            echo ""
+            echo "   Check CloudWatch Logs for details:"
+            echo "   https://console.aws.amazon.com/cloudwatch/home?region=${AWS_REGION}#logsV2:log-groups/log-group//aws/sagemaker/TransformJobs"
+            exit 4
+            ;;
+        Stopped)
+            echo "⚠️  Transform job was stopped"
+            exit 4
+            ;;
+        InProgress)
+            echo "   $(date +%H:%M:%S) Job status: InProgress..."
+            sleep 30
+            ;;
+        *)
+            echo "   $(date +%H:%M:%S) Job status: ${JOB_STATUS}..."
+            sleep 30
+            ;;
+    esac
+done
+
+echo ""
+echo "📋 Deployment Details:"
+echo "   Transform Job: ${TRANSFORM_JOB_NAME}"
+echo "   Model: ${MODEL_NAME_SM}"
+echo "   Model Package: ${MODEL_PACKAGE_ARN}"
+echo "   Region: ${AWS_REGION}"
+echo "   Instance Type: ${INSTANCE_TYPE}"
+echo "   Instance Count: ${BATCH_INSTANCE_COUNT}"
+echo "   S3 Input: ${BATCH_INPUT_PATH}"
+echo "   S3 Output: ${BATCH_OUTPUT_PATH}"
+echo "   Split Type: ${BATCH_SPLIT_TYPE}"
+echo "   Strategy: ${BATCH_STRATEGY}"
+echo ""
+
+# Download results locally
+LOCAL_OUTPUT_DIR="${SCRIPT_DIR}/../batch-output"
+mkdir -p "${LOCAL_OUTPUT_DIR}"
+echo "📥 Downloading results to ${LOCAL_OUTPUT_DIR}/"
+if aws s3 sync "${BATCH_OUTPUT_PATH}" "${LOCAL_OUTPUT_DIR}/" --region "${AWS_REGION}"; then
+    DOWNLOADED=$(ls -1 "${LOCAL_OUTPUT_DIR}" 2>/dev/null | wc -l | tr -d ' ')
+    echo "✅ Downloaded ${DOWNLOADED} file(s) to ${LOCAL_OUTPUT_DIR}/"
+    echo ""
+
+    # Display first output file preview
+    FIRST_FILE=$(ls -1 "${LOCAL_OUTPUT_DIR}" 2>/dev/null | head -1)
+    if [ -n "${FIRST_FILE}" ]; then
+        echo "📄 Sample output (${FIRST_FILE}):"
+        head -5 "${LOCAL_OUTPUT_DIR}/${FIRST_FILE}"
+        LINES=$(wc -l < "${LOCAL_OUTPUT_DIR}/${FIRST_FILE}" | tr -d ' ')
+        if [ "${LINES}" -gt 5 ]; then
+            echo "   ... (${LINES} total lines)"
+        fi
+    fi
+else
+    echo "⚠️  Could not download output files"
+fi
+
+echo ""
+echo "📋 What's next?"
+echo "   • View results:               cat batch-output/"
+echo "   • Review results:             ./do/test"
+echo "   • Register this deployment:   ./do/register"
+echo "   • View logs:                  ./do/logs"
+echo "   • Clean up when done:         ./do/clean"
+
+<% } %>