Pizza Pie Perfect Pricing Problem Preserves Purity and Promotes Practical Progress

Transcript

1. Pizza Pie Perfect Pricing Problem Preserves Purity & Promotes Practical Progress Mattias Andersson
2. Point Packages  Perfectly Pricing Pizza Pies  Case study: How did we scalably connect 1000s of stores more reliably & securely?  Preserving Purity  What benefits did our pure serverless architecture give us?  Promoting Practical Progress  What can we take away from this serverless problem-solving process?
3. Perfectly Pricing Pizza Pies
5. Context  Established Point of Sale (POS) software vendor  Many existing restaurant customers, big & small  Flexible pricing capability a core product differentiator
6. Opportunity  Entering new market: directly offer online ordering service Had previously only enabled other integrations via XML-based API  Quickly build an online store Multi-tenant, scalable Opting to build serverless system on AWS (full story another time) API-based to support enterprise integrations
7. Basic System Structure Amazon API Gateway Client AWS Lambda Amazon CloudFront Amazon DynamoDB AWS Certificate Manager (ACM) Amazon Route 53 Amazon S3
8. Particular Problem: Perfectly Pricing Pizza Pies  Business needed the new online system to price online orders always exactly matching in-store orders  Pricing parity a critical core business requirement for online offering Too low is bad! Too high is bad!  Short timeline  But should still build to align with further dev & product plans
9. Context  Highly-variable pizza/ticket pricing  Prices change by location and order context Regional, store-level, customer-level  Prices even change based on time or in-store situation Happy hour, delivery fee rates change, deals expire, toppings out of stock, etc.  Existing pricing logic to match is  15+ years of complicated option-driven business logic  locked in hardware and software running in each restaurant  with store-level configuration, pricing, and status information.  None of these are (yet) available to our serverless system, online
10. High-Level Options  Run existing pricing logic in the cloud Challenging to configure, expensive to run (1+ VM per store), no status  Rewrite all existing pricing logic to run in the cloud Extremely challenging to code and configure; would require years to do right  Rewrite some subset of existing pricing logic in the cloud Too limiting for the business  Connect the cloud to existing pricing logic already running in stores Higher latency, but business’s preferred approach
11. Technical Problem Amazon API Gateway Client AWS Lambda In-Store Server Firewall
12. General Idea — Meet in the Middle AWS Lambda In-Store Server Firewall AWS Cloud
13. Store General Idea — Meet in the Middle Pricing In-Store Server Firewall AWS Cloud
14. Pricing Pricing Pricing Pricing Pricing Pricing Pricing Brief Aside — Lambda Instances Pricing Pricing Pricing
15. Store Technical Solution — First Contact Pricing In-Store Server AWS Cloud Amazon Simple Queue Service (Amazon SQS)
16. Store Technical Solution — First Contact Pricing In-Store Server AWS Cloud Store Config Store Request Queue ?
17. Store Technical Solution — The Voyage Home Pricing In-Store Server AWS Cloud Store Config Store Request Queue Lambda Response Queue $
18. Response Queue Leasing  Build a new service to maintain a pool of SQS queues  When asked for a queue, it looks for an existing queue with no current lease  If there is no available queue, it will make a new one and add it to the pool  It will assign a timed queue lease to the calling Lambda instance  It stores its data in a simple DynamoDB table  DynamoDB writes to assign/claim a lease will fail if the record was changed since lookup  Then it just retries everything  Only returns once the lease has been confirmed AWS Lambda Amazon DynamoDB Amazon Simple Queue Service (Amazon SQS)
19. Long Polling  To get the fastest response from SQS, use Long-Polling  We call SQS asking for a message, but tell it to wait if there is none  Wait time can be up to 20 seconds  When SQS gets a new message, it immediately sends it to a waiting worker  Compared to checking every 2s in a loop:  This saves 90% on cost  This saves ~1s of latency (0-2s random distribution) Amazon Simple Queue Service (Amazon SQS)
20. Put More Terminals to Work  Store servers have historically been failure-prone  A well-meaning employee might turn off the designated Primary  Inbound connections can only go to one server  Unless proxied/load-balanced  Price requests are single-threaded and block the next request  Several idle In-Store Terminals could easily process requests  Take advantage: Make multiple In-Store Terminals watch for pricing requests
21. Secure Responses  Lambda instance handling a Pricing Request will: Generate a Request ID (UUID or Hash) Create response-queue-restricted IAM credentials via STS Include these in the pricing request  In-Store server must include Request ID with the response  Pricing Lambda will ignore any non-matching responses (presumed stale) AWS Simple Token Service (STS)
22. Store Technical Solution — The Final Frontier Pricing In-Store Server AWS Cloud Queue Lease Store Request Queue ? Lambda Response Queue Store Config In-Store Server In-Store Server AWS STS $
23. Store Technical Solution — The Final Frontier Pricing In-Store Server AWS Cloud Queue Lease Store Request Queue ? Lambda Response Queue Store Config In-Store Server In-Store Server AWS STS $
24. Costing — Per Store Per Year Rate Normal store processes 200-1000 requests per day Very busy store may process 10,000 requests per day (Still <1RPS peak) Yearly store request rate from 50K to 4M, averaging ~200K Total Average store costs $3.15 to connect and $0.62 to process messages Very busy store costs $3.15 to connect and $12.40 to process messages Business finds this profitable
25. Results  Solved business problem (perfect-match pricing)  Technical solution was cost-effective and profitable  Versus old integration approach:  Easier setup (minutes instead of hours)  Better security (no open ports)  More robust (eliminated in-store SPoF)  Quicker responses (parallel calls)  Created helpful business analytics stream (real usage patterns)  Cleared a path for future dev (proven connection; source of truth)
26. Preserving (Serverless) Purity
27. The Shape of Serverless Scales: 1. Automatically 2. Linearly (or less) 3. Continuously 4. From zero 5. Based on usage 6. Infinitely 7. Transparently (i.e. expertly-managed) Value Cost
28. Sample Serverless Services (AWS) ‣ S3 ‣ DynamoDB ‣ Lambda ‣ API Gateway ‣ CloudFront ‣ Certificate Manager ‣ EventBridge ‣ StepFunctions ‣ Cognito ‣ SQS ‣ SNS ‣ Rekognition ‣ Transcribe ‣ Comprehend ‣ Translate ‣ Polly ‣ Lex ‣ Elastic Transcoder ‣ Kinesis Firehose ‣ Glue ‣ Athena ‣ SageMaker Serverless Inference ‣ KMS ‣ IAM ‣ CloudWatch ‣ CloudFormation ‣ …
29. Sample Serverless Services (GCP) ‣ Google Cloud Storage ‣ BigQuery ‣ Firestore ‣ Cloud Functions ‣ Cloud Run ‣ Pub/Sub ‣ Workflows ‣ Eventarc ‣ Cloud Translation ‣ Cloud Vision ‣ Text-to-Speech ‣ Speech-to-Text ‣ Document AI ‣ Vertex AI Recommendations Predictions ‣ Vertex AI Generative AI ‣ Cloud CDN ‣ Cloud Build ‣ Artifact Registry ‣ Cloud Shell ‣ Cloud Monitoring ‣ Cloud Logging ‣ Cloud Trace ‣ Error Reporting ‣ …
30. Sample Serverless Services (Azure) ‣ Azure Blob Storage ‣ Azure Functions ‣ Durable Functions ‣ Azure CDN ‣ Queue Storage ‣ Application Insights ‣ App Service Platform ‣ Azure Container Apps ‣ Azure Cosmos DB Serverless APIs ‣ Azure Repos ‣ Azure Artifacts ‣ Azure AI Speech ‣ Azure AI Language ‣ Azure AI Translator ‣ Azure AI Vision ‣ Azure AI Content Safety ‣ Azure AI Document Intelligence ‣ Azure OpenAI Service ‣ Azure Front Door Classic ‣ API Management v1 Consumption ‣ Azure Active Directory External Identities ‣ …
31. Benefits of Pure Serverlessness  Composable  Cost to business only rises when business value rises  Zero cost environments  Great fit with Immutable Infrastructure via Infrastructure as Code (IaC)  (Externalize data for long-term envs)  Dev environments to suit  New environment per push, for QA & forensics  But…
32. Warnings with Pure Serverlessness  Unexpected limits!  We hit the cap on API Gateways  Warming services  E.g. Lambda cold starts
33. Promoting Practical Progress
34. Compute/Storage/Network/… Bucketing Considered Harmful
35. Build Using Data Flow Primitives Offered Remember Transform Move That’s it!
36. Example: Whipping Up a Link Redirector  Heading into my Cloud Summit talk, I wanted some short links I could later update as needed  Quickly whipped up a link redirector  Control Tower & Organizations to make a new AWS account  Route 53 for domain registration (by far the slowest step!)  Route 53 for DNS (automatic)  AWS Certificate Manager for TLS cert  CloudFront for HTTPS & caching  API Gateway for endpoint handling  Later: DynamoDB for reading links  Even Later: Cognito for auth, writing links to DynamoDB
37. API Gateway is the Unsung Hero  Far more powerful than most realize!  Moves requests in & responses out  Remembers configuration  Transforms requests & responses Apache Velocity Template Language (VTL) Amazon API Gateway
38. Cloud Summit: Just Mocks
39. Example: Whipping Up a Link Redirector  Heading into my Cloud Summit talk, I wanted some short links I could later update as needed  Quickly whipped up a link redirector  Control Tower & Organizations to make a new AWS account  Route 53 for domain registration (by far the slowest step!)  Route 53 for DNS (automatic)  AWS Certificate Manager for TLS cert  CloudFront for HTTPS & caching  API Gateway for endpoint handling  Later: DynamoDB for reading links  Even Later: Cognito for auth, writing links to DynamoDB
40. AWS Community Day Organizing: Read DynamoDB
41. Example: Whipping Up a Link Redirector  Heading into my Cloud Summit talk, I wanted some short links I could later update as needed  Quickly whipped up a link redirector  Control Tower & Organizations to make a new AWS account  Route 53 for domain registration (by far the slowest step!)  Route 53 for DNS (automatic)  AWS Certificate Manager for TLS cert  CloudFront for HTTPS & caching  API Gateway for endpoint handling  Later: DynamoDB for reading links  Even Later: Cognito for auth, writing links to DynamoDB
42. Cheatsheets / AI Summit: Write DynamoDB
43. Point Packages  Perfectly Pricing Pizza Pies  Smoothly connected 1000s of stores reliably & securely via SQS & friends  Preserving Purity  Pure serverless architecture offered Immutable Infra & forensic envs  Promoting Practical Progress  Innovatively compose systems from service features that Remember, Transform, & Move data
44. Questions