© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. The serverless journey of shop.LEGO.com S V S 3 2 0 Sheen Brisals Senior Application Engineer The LEGO Group Danilo Poccia Principal Evangelist Amazon Web Services

Agenda How it all started A journey through patterns Takeaways

© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.

No content

The “monolith” at the start of this journey Corporate data center Server contents Server contents Tax Products CRM Fulfillment Rewards Server contents

The “monolith” at the start of this journey VPC Node.js + React Elastic Beanstalk container Corporate data center Server contents Server contents Tax Products CRM Fulfillment Rewards Server contents

The “monolith” at the start of this journey VPC Node.js + React Elastic Beanstalk container Server contents Server contents Tax Products CRM Fulfillment Rewards VPC Gateway Corporate data center Server contents

Then, during Black Friday 2017 … VPC Node.js + React Elastic Beanstalk container Server contents Server contents Tax Products CRM Fulfillment Rewards VPC Gateway 503 Service Unavailable Corporate data center Server contents

Consolation in chaos …

September 2018 – Decoupling the backend VPC Node.js + React Elastic Beanstalk container Server contents Server contents Tax Products CRM Fulfillment Rewards AWS Cloud SaaS VPC Gateway Server contents Corporate data center Amazon API Gateway AWS Lambda function

September 2018 – First production Lambda function

And the serverless journey begins!

Black Friday 2018

July 10, 2019 shop.LEGO.com was switched to serverless on AWS

AWS services in use Amazon S3 Amazon SNS AWS Fargate AWS Lambda Amazon Kinesis Data Firehose Amazon DynamoDB Amazon API Gateway AWS Step Functions Amazon SQS Amazon EC2 Amazon ECS Amazon ECR Amazon VPC AWS WAF AWS KMS AWS Identity and Access Management (IAM) AWS X-Ray Amazon Route 53 AWS CloudTrail Amazon ElastiCache Amazon CloudWatch Amazon EventBridge AWS CloudFormation Parameter Store

Growing usage stats from production 165+ Lambda functions 35+ microservices 30+ API Gateway endpoints 10+ DynamoDB tables 5+ S3 buckets 20+ SNS topics 60+ SQS queues + DLQs 70+ Systems Manager params

© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.

Use case – Add item to shopping basket Customer Product Validation Basket

Pattern – Atomic request-response API Shopping app Add item API Business logic Basket data store Commerce platform Request Response Invoke Check/add • Performance is priority • Logic evaluation • No partial failure • Success or error response • Validation steps • Limit data relay

A journey through patterns Use case Pattern ✓ Add item to shopping basket Atomic request-response API

Use case – Status polling for long-running processes “What’s the status of my order?”

Pattern – CQRS* with status cache Shop POST GET Submit order API Submit order function Submit order queue Process order function Update status Order status API Status provider function Order cache (TTL) Order number Query status Payment providers New commerce platform Frontend visibility Processing Failed Complete Backend visibility Validating Validation failed Payment authorized Payment processing Card declined Sent to fulfillment * Command Query Responsibility Segregation

A journey through patterns Use case Pattern ✓ Add item to shopping basket Atomic request-response API ✓ Status polling for long-running processes CQRS with status cache

Use case – Voucher codes generation and notification Admin Voucher codes Voucher store Admin File download Email

Pattern – Email notification with signed URL Admin user Vouchers API Request handler Generate vouchers Voucher backup Voucher table Voucher topic Email notification Email with Amazon S3 signed URL Acknowledge POST 1. Create vouchers 2. Update voucher table 3. Store batch on Amazon S3 4. Generate Amazon S3 signed URL 5. Send out notification • Vouchers from 1 to 1M • Unique voucher codes • Secure download option • User notification GET Async Sync

A journey through patterns Use case Pattern ✓ Add item to shopping basket Atomic request-response API ✓ Status polling for long-running processes CQRS with status cache ✓ Voucher codes generation and notification Email notification with signed URL

Use case – User identity lookup in different systems Customer Customer identity Authenticate Authorize Rewards

What happens if identities don’t match? Customer Profile API Request handler LEGO identity Client app Rewards platform Commerce platform LEGO ID Rewards ID

Pattern – API authorizer with identity lookup Customer Profile API Request handler • Session must be valid • Session must belong to the user • Profile must exist • Profile ID must match LEGO identity Authorizer Commerce platform Rewards platform LEGO ID Rewards ID Client app

A journey through patterns Use case Pattern ✓ Add item to shopping basket Atomic request-response API ✓ Status polling for long-running processes CQRS with status cache ✓ Voucher codes generation and notification Email notification with signed URL ✓ User identity lookup in different systems API authorizer with identity lookup

Use case – On-demand customer data migration Customer New site New platform Notify Process New commerce platform Fetch Update

Pattern – Publish-subscribe sync • On-demand data sync • Real-time migration • Data in JSON format • No impact to user experience Customer topic New commerce platform New customer handler Customer created Fetch data Update data

A journey through patterns Use case Pattern ✓ Add item to shopping basket Atomic request-response API ✓ Status polling for long-running processes CQRS with status cache ✓ Voucher codes generation and notification Email notification with signed URL ✓ User identity lookup in different systems API authorizer with identity lookup ✓ On-demand customer data migration Publish-subscribe sync

Use case – Product catalog import and update Commerce platform Feeds transform Products feeds

Pattern – Event-driven data pipeline with buffering Transform feed Transformed feeds queue Feeds store Commerce platform API Update data Data feeds Trigger Push Trigger Update • Feeds in JSON format • Independent feed flow • Immediate handling of feeds • Decoupled processing • Failure retry process • Failure notification DLQ Thousands per day processed as soon as possible Fine-grained control on throughput and errors Observability

Same pattern works for multiple object types Commerce platform API Data feeds Trigger Push Trigger Update Push Trigger Push Trigger Product Pricing SKU Observability DLQs

A journey through patterns Use case Pattern ✓ Add item to shopping basket Atomic request-response API ✓ Status polling for long-running processes CQRS with status cache ✓ Voucher codes generation and notification Email notification with signed URL ✓ User identity lookup in different systems API authorizer with identity lookup ✓ On-demand customer data migration Publish-subscribe sync ✓ Product catalog import and update Event-driven data pipeline with buffering

© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. “Code is a liability. Write less functions.” On Twitter Lots of People

Use case – API-driven data ingestion Processing Data store Data ingestion pipeline Event producer Ingest API

A first solution… Can we do better? Lambda function receive event & send to Kinesis Kinesis Data Firehose buffering API Gateway receive events Processing application S3 bucket store events Event producer

Pattern – Codeless data ingestion Kinesis Data Firehose API Gateway S3 bucket Event producer • API to stream directly • Less point of failures • Native integration • Less compute costs • Less code to maintain • Fully managed and scalable Transformation and validation Backup Fan-out function Processing functions Trigger Async Errors

https://bit.ly/2qhsBYH

A journey through patterns Use case Pattern ✓ Add item to shopping basket Atomic request-response API ✓ Status polling for long-running processes CQRS with status cache ✓ Voucher codes generation and notification Email notification with signed URL ✓ User identity lookup in different systems API authorizer with identity lookup ✓ On-demand customer data migration Publish-subscribe sync ✓ Product catalog import and update Event-driven data pipeline with buffering ✓ API-driven data ingestion Codeless data ingestion

Use case – Unique order ID generation Checkout flow Loyalty customers Order number Customer ID

Pattern – Codeless sequence generator Sequences table Sequence number API Sequence item • No Lambda function • Repeatable pattern • Fully managed • Stateful resource • Atomic counter • Single table partition Counter attribute Any microservice

DynamoDB Atomic Counter { "TableName": ”sequences", "Key": { "id": { ”S": ”order" } }, "ExpressionAttributeValues": { ":one": { "N": 1 } }, "UpdateExpression": "SET num = num + :one", "ReturnValues": "UPDATED_NEW" } Atomic update

https://bit.ly/33z1G8W

A journey through patterns Use case Pattern ✓ Add item to shopping basket Atomic request-response API ✓ Status polling for long-running processes CQRS with status cache ✓ Voucher codes generation and notification Email notification with signed URL ✓ User identity lookup in different systems API authorizer with identity lookup ✓ On-demand customer data migration Publish-subscribe sync ✓ Product catalog import and update Event-driven data pipeline with buffering ✓ API-driven data ingestion Codeless data ingestion ✓ Unique order number generation Codeless sequence generator

August 15, 2019 – Unified LEGO.com

Use case – Website migration with URL changes Old site URL New site Identify destination Content provider

Pattern – URL redirects cached by CDN URL redirect function Content store URL rules Cache on /tmp Application Load Balancer CDN • 512 MB /tmp storage • 128MB – 3GB memory

A journey through patterns Use case Pattern ✓ Add item to shopping basket Atomic request-response API ✓ Status polling for long-running processes CQRS with status cache ✓ Voucher codes generation and notification Email notification with signed URL ✓ User identity lookup in different systems API authorizer with identity lookup ✓ On-demand customer data migration Publish-subscribe sync ✓ Product catalog import and update Event-driven data pipeline with buffering ✓ API-driven data ingestion Codeless data ingestion ✓ Unique order number generation Codeless sequence generator ✓ Website migration with URL changes URL redirects cached by CDN

Use case – Keeping website sitemaps updated Timer Process Store sitemaps Publish Search engines

Pattern – Scheduled workflow Step Functions Trigger rule • Amazon S3 for storage • Amazon S3 as origin server • Runs daily – off peak • Sitemaps in XML Amazon CloudWatch Events

Pattern – Scheduled workflow /work /release CDN

A journey through patterns Use case Pattern ✓ Add item to shopping basket Atomic request-response API ✓ Status polling for long-running processes CQRS with status cache ✓ Voucher codes generation and notification Email notification with signed URL ✓ User identity lookup in different systems API authorizer with identity lookup ✓ On-demand customer data migration Publish-subscribe sync ✓ Product catalog import and update Event-driven data pipeline with buffering ✓ API-driven data ingestion Codeless data ingestion ✓ Unique order number generation Codeless sequence generator ✓ Website migration with URL changes URL redirects cached by CDN ✓ Keeping website sitemaps updated Scheduled workflow

Use case – Checkout event processing Login Checkout Payment Order Data sync Shipping

Use case – Hub-and-spoke event bus Login Checkout Payment Order Data sync Shipping

July 10, 2019 shop.LEGO.com was switched to serverless on AWS Flashback!

July 11, 2019 The day after, Amazon EventBridge was launched

Pattern Customer login Login Shipping Send order to SAP Data sync Customer, VIP, wishlist sync Checkout Submit order Payment Authorize payment Commerce platform Order Process order Order and customer updates Event relay Customer login Invoke every minute Events Order complete Customer login Payment authorized Order submit Order complete EventBridge FIFO queue

Pattern – Hub-and-spoke event bus { "version": "0", "id": "6a7e8feb-b491-4cf7-a9f1-bf3703467718", "detail-type": "State change Notification", "source": "service-order-submit-dev", "account": "111122223333", "time": "2019-08-29T12:10:21Z", "region": "eu-central-1", "resources": [ "arn:aws:events:event-bus/checkout-bus" ], "detail": { } } Customer-specific data goes in the “detail”

Pattern – Hub-and-spoke event bus { "version": "0", "id": "6a7e8feb-b491-4cf7-a9f1-bf3703467718", "detail-type": "State change Notification", "source": "service-order-submit-dev", "account": "123456789012", "time": "2019-08-29T12:10:21Z", "region": "eu-central-1", "resources": ["arn:aws:events:event-bus/checkout-bus"], "detail": { "event": { "meta_data": { "site_id": "LEGO Shop", "type": "CHECKOUT", "subtype": "ORDER", "status": "COMPLETE" }, "data": { "order_number": "T123456789", "customer_id": "bf3703467718-29T12-6a7e8feb" } } } } Standard syntax across multiple services Custom for each service

A journey through patterns Use case Pattern ✓ Add item to shopping basket Atomic request-response API ✓ Status polling for long-running processes CQRS with status cache ✓ Voucher codes generation and notification Email notification with signed URL ✓ User identity lookup in different systems API authorizer with identity lookup ✓ On-demand customer data migration Publish-subscribe sync ✓ Product catalog import and update Event-driven data pipeline with buffering ✓ API-driven data ingestion Codeless data ingestion ✓ Unique order number generation Codeless sequence generator ✓ Website migration with URL changes URL redirects cached by CDN ✓ Keeping website sitemaps updated Scheduled workflow ✓ Checkout event processing Hub-and-spoke event bus

A journey through patterns Use case Pattern ✓ Add item to shopping basket Atomic request-response API ✓ Status polling for long-running processes CQRS with status cache ✓ Voucher codes generation and notification Email notification with signed URL ✓ User identity lookup in different systems API authorizer with identity lookup ✓ On-demand customer data migration Publish-subscribe sync ✓ Product catalog import and update Event-driven data pipeline with buffering ✓ API-driven data ingestion Codeless data ingestion ✓ Unique order number generation Codeless sequence generator ✓ Website migration with URL changes URL redirects cached by CDN ✓ Keeping website sitemaps updated Scheduled workflow ✓ Checkout event processing Hub-and-spoke event bus

© 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved.

Takeaways Look for something simple to begin with Implement automated integration tests Separate dev, test, prod accounts Architect in “set pieces” No throwaway PoCs Leverage patterns

Thank you! © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. Sheen Brisals @sheenbrisals Danilo Poccia @danilop