cleanup.sh

#!/bin/bash
# =============================================================================
# Cleanup script for SageMaker MLflow Portal
# Destroys all CDK stacks and deletes the serverless MLflow App
# Run this to tear down the entire setup
# =============================================================================
set -e

STACK_PREFIX="sagemaker-infra"
REGION="${CDK_DEFAULT_REGION:-us-east-1}"

echo "============================================"
echo "  SageMaker MLflow Portal - Full Cleanup"
echo "============================================"
echo ""
echo "WARNING: This will destroy ALL resources including:"
echo "  - EC2 instance and ALB (Flask App stack)"
echo "  - Serverless MLflow App"
echo "  - SageMaker Domain and user profiles"
echo "  - VPC and networking resources"
echo "  - S3 helper scripts bucket (auto-deleted)"
echo ""
echo "  NOTE: The MLflow artifacts S3 bucket has RETAIN policy"
echo "        and will NOT be deleted. Delete it manually if needed."
echo ""
read -p "Are you sure you want to proceed? (yes/no): " CONFIRM
if [ "$CONFIRM" != "yes" ]; then
  echo "Cleanup cancelled."
  exit 0
fi
echo ""

# Derive MLflow App name from SageMaker Domain ID
echo "Resolving MLflow App name from SageMaker Domain stack..."
DOMAIN_ID=$(aws cloudformation describe-stacks \
  --stack-name "${STACK_PREFIX}-domain" \
  --region "$REGION" \
  --query "Stacks[0].Outputs[?OutputKey=='DomainId'].OutputValue" \
  --output text 2>/dev/null || true)

if [ -n "$DOMAIN_ID" ] && [ "$DOMAIN_ID" != "None" ]; then
  MLFLOW_APP_NAME="mlflow-app-${DOMAIN_ID}"
  echo "  Domain ID: $DOMAIN_ID"
  echo "  MLflow App Name: $MLFLOW_APP_NAME"
else
  echo "  WARNING: Could not resolve Domain ID from stack outputs."
  echo "  Falling back to legacy app name 'sagemaker-mlflow-app'."
  MLFLOW_APP_NAME="sagemaker-mlflow-app"
fi
echo ""

# Step 1: Destroy Flask App stack first (depends on others)
echo "[Step 1/5] Destroying Flask App stack..."
cdk destroy ${STACK_PREFIX}-flaskapp --force 2>/dev/null || echo "  Flask App stack not found or already deleted. Skipping."
echo ""

# Step 2: Delete the serverless MLflow App via AWS CLI
echo "[Step 2/5] Deleting serverless MLflow App..."
MLFLOW_APP_ARN=$(aws sagemaker list-mlflow-apps \
  --region "$REGION" \
  --no-paginate 2>/dev/null | python3 -c "
import json, sys
data = json.load(sys.stdin)
for app in data.get('Summaries', []):
    if app.get('Name') == '$MLFLOW_APP_NAME':
        print(app.get('Arn', ''))
        break
" 2>/dev/null || true)

if [ -n "$MLFLOW_APP_ARN" ]; then
  echo "  Found MLflow App: $MLFLOW_APP_ARN"
  aws sagemaker delete-mlflow-app \
    --arn "$MLFLOW_APP_ARN" \
    --region "$REGION" 2>/dev/null || echo "  Failed to delete MLflow App. You may need to delete it manually."
  echo "  MLflow App deletion initiated."
  echo "  Waiting 60 seconds for MLflow App to be fully deleted..."
  sleep 60
else
  echo "  No MLflow App found with name '$MLFLOW_APP_NAME'. Skipping."
fi
echo ""

# Step 3: Destroy MLflow Resources stack (IAM role + S3 bucket)
echo "[Step 3/5] Destroying MLflow Resources stack..."
cdk destroy ${STACK_PREFIX}-mlflow --force 2>/dev/null || echo "  MLflow stack not found or already deleted. Skipping."
echo ""

# Step 4: Destroy SageMaker Domain stack
echo "[Step 4/5] Destroying SageMaker Domain stack..."
cdk destroy ${STACK_PREFIX}-domain --force 2>/dev/null || echo "  Domain stack not found or already deleted. Skipping."
echo ""

# Step 5: Destroy Networking stack (VPC, subnets, NAT gateway)
echo "[Step 5/5] Destroying Networking stack..."
cdk destroy ${STACK_PREFIX}-networking --force 2>/dev/null || echo "  Networking stack not found or already deleted. Skipping."
echo ""

echo "============================================"
echo "  Cleanup Complete!"
echo "============================================"
echo ""
echo "All stacks have been destroyed."
echo ""
echo "Remaining manual cleanup (if needed):"
echo "  - MLflow artifacts S3 bucket (RETAIN policy — check S3 console)"
echo "  - Any CloudWatch log groups created by Lambda functions"
echo "  - CDK bootstrap stack (if you no longer need CDK in this account)"