From Instances to Functions: Going Serverless

Serverless is a hot topic in the software architecture world. A serverless architecture approach replaces long-running virtual machines with ephemeral compute power that comes into existence on request and disappears immediately after use. Pay for the hamburger, not the cow!

Transcript

1. From Instances to Functions
   Going Serverless
   Yos Riady
   yos.io
   goo.gl/hvrw9s
   

   View Slide

2. View Slide

3. Serverless:
   Moving away from servers and
   infrastructure concerns & allowing
   developers to primarily focus on code
   

   View Slide

4. View Slide

5. How We Got Here
   Pros & Cons of the
   Serverless Architecture
   Background Why Concepts Code Next Steps
   Intro to Serverless Summary
   Walkthrough of a
   real Serverless
   application
   

   View Slide

6. How We Got Here
   

   View Slide

7. View Slide

8. View Slide

9. Cloud Computing Evolution
   

   View Slide

10. Software Architecture Evolution
    

    View Slide

11. Functions as a Service
    ● FaaS / Serverless platforms are compute services that
    runs your code in response to events.
    

    View Slide

12. View Slide

13. Anatomy of a Lambda Function
    exports.myFunction = function(event, context, callback) {
    console.log("value1 = " + event.key1);
    console.log("value2 = " + event.key2);
    var result = event.key1 + event.key2;
    callback(null, { result: result, success: true });
    }
    

    View Slide

14. View Slide

15. How do FaaS platforms execute your code?
    ● When a lambda function is invoked,
    ○ A container (execution environment) is provisioned
    ● It takes time to set up the container
    ○ “Cold start”
    ● After execution, the container is maintained for some time (e.g. 5 minutes)
    ● “Warm” container reuse for subsequent invocations
    ○ Significantly faster
    

    View Slide

16. View Slide

17. How We Got Here
    Pros & Cons of the
    Serverless Architecture
    Background Why Concepts Code Next Steps
    Intro to Serverless Summary
    Walkthrough of a
    real Serverless
    application
    

    View Slide

18. Why Serverless
    

    View Slide

19. View Slide

20. View Slide

21. View Slide

22. View Slide

23. Benefits of Serverless
    ● Scalability baked in and high availability
    ○ No need for capacity planning
    ● Less Ops
    ○ No infrastructure to manage
    ● Low cost
    ○ Billed in 100ms increments
    ○ Pay only for what you use!
    

    View Slide

24. Drawbacks of Serverless
    ● Vendor lock-in
    ● Limited degree of customization & control
    ○ Memory
    ○ Timeouts
    ○ Disk space
    ● Decentralization
    ○ Debugging / Monitoring challenges
    ○ Integration testing can be hard
    

    View Slide

25. Serverless Use Cases
    ● Web / Mobile backend APIs
    ● Image / Asset Processing
    ● Webhooks
    ● DevOps automation
    ● Log processing
    ● ETL / Batch processing
    ● Big data / Machine learning
    ● Bots
    ● Web scraping & automation
    

    View Slide

26. Serverless Backend
    

    View Slide

27. Serverless Asset Processing
    

    View Slide

28. How We Got Here
    Pros & Cons of the
    Serverless Architecture
    Background Why Concepts Code Next Steps
    Intro to Serverless Summary
    Walkthrough of a
    real Serverless
    application
    

    View Slide

29. View Slide

30. Introduction to the Serverless Framework
    ● CLI that helps you develop and deploy your AWS Lambda functions, along with
    the AWS infrastructure resources they require. You get:
    ○ Structure & best practices
    ○ Automation
    ○ Plugin ecosystem
    ● Provider and runtime agnostic
    ○ AWS
    ○ Azure
    ○ OpenWhisk
    ○ GCP
    

    View Slide

31. Events Functions Resources
    

    View Slide

32. Deployable units of code
    Single Responsibility Principle
    (think Unix Philosophy)
    Languages and runtimes vary
    Functions
    Executable snippets of code
    

    View Slide

33. AWS Lambda Runtimes (May 2017)
    ● Node.js – v4.3.2 and 6.10
    ● Java – Java 8
    ● Python – Python 3.6 and 2.7
    ● .NET Core – .NET Core 1.0.1 (C#)
    

    View Slide

34. Anatomy of an AWS Lambda Node.js Function
    exports.myHandler = function(event, context, callback) {
    console.log("value1 = " + event.key1);
    console.log("value2 = " + event.key2);
    callback(null, "some success message");
    // or
    // callback("some error type");
    }
    

    View Slide

35. HTTP request
    SNS notification
    DB events
    S3 events
    Scheduled events
    On-demand invocation
    Events
    Triggers functions
    

    View Slide

36. AWS Lambda Events (May 2017)
    ● HTTP (API Gateway)
    ● Kinesis
    ● DynamoDB
    ● S3
    ● SNS
    ● Alexa Skill
    ● IoT
    ● CloudWatch Event
    ● CloudWatch Log
    ● Schedule
    

    View Slide

37. NoSQL / SQL Databases
    File storage
    REST APIs
    Third-party APIs
    Legacy services
    etc.
    Resources
    Supporting cloud services
    

    View Slide

38. serverless.yml
    # serverless.yml
    service: users
    functions: # Your "Functions"
    usersCreate:
    events: # The "Events" that trigger this function
    - http: post users/create
    usersDelete:
    events:
    - http: delete users/delete
    resources: # The "Resources" your "Functions" use. Raw AWS CloudFormation goes in here.
    

    View Slide

39. Serverless Development Workflow
    1. serverless create to bootstrap a Serverless project
    2. Write your functions
    3. serverless deploy to deploy the current state of the project
    4. serverless invoke to test the live function
    5. serverless logs to view function logs
    6. Write and run unit tests for your functions locally with mocks
    

    View Slide

40. Hands-On
    

    View Slide

41. How We Got Here
    Pros & Cons of the
    Serverless Architecture
    Background Why Concepts Code Next Steps
    Intro to Serverless Summary
    Walkthrough of a
    real Serverless
    application
    

    View Slide

42. Kids these days
    

    View Slide

43. What we’re building
    ● An image processing backend for a photo filter app
    ○ Users supply an image URL
    ○ Images are analyzed and processed to add relevant emojis
    ○ Users can view the processed images
    

    View Slide

44. What we’re building it with
    ● AWS and the Serverless framework
    ○ Lambda: Compute service
    ○ S3: File Storage
    ○ DynamoDB: NoSQL Database
    ○ Rekognition: Image Recognition API
    

    View Slide

45. View Slide

46. Problem decomposition
    ● Divide-and-conquer: Serverless applications are composed of many granular
    functions, each solving a distinct sub-problem.
    ○ downloadImage
    ○ analyzeImage
    ○ processImage
    ○ viewImage
    

    View Slide

47. Events Functions Resources
    

    View Slide

48. HTTP Event AWS S3
    downloadImage
    Function #0: Download Image
    

    View Slide

49. Function #0: Download Image
    ● Sub-problem: How do we save the image supplied by a user?
    ● Solution: Create a new downloadImage function
    ○ Triggered by an HTTP event
    ○ Saves image from URL to an S3 bucket
    ○ Returns an ID back to the client
    

    View Slide

50. S3 Event AWS Rekognition
    AWS DynamoDB
    analyzeImage
    Function #1: Analyze Image
    

    View Slide

51. Function #1: Analyze Image
    ● Sub-problem: How do we analyze what’s in the image?
    ● Solution: Create a new analyzeImage function
    ○ Triggered by an S3 event
    ○ Calls the AWS Rekognition DetectFaces API
    ○ Writes analysis results to DynamoDB
    

    View Slide

52. AWS S3
    DynamoDB processImage
    Function #2: Process Image
    

    View Slide

53. Function #2: Process Image
    ● Sub-problem: How do we add emojis to the image?
    ● Solution: Create a new processImage function
    ○ Triggered by a DynamoDB event
    ○ Processes the image using ImageMagick
    ○ Saves the processed image to S3
    

    View Slide

54. Function #3: View Image
    AWS DynamoDB
    AWS S3
    HTTP viewImage
    

    View Slide

55. Event-Driven Pipelines
    downloadImage
    analyzeImage
    processImage
    viewImage
    

    View Slide

56. Result: A scalable image processing backend
    ● Less Ops: No servers to provision, faster development
    ● Scalable: AWS Lambda will execute as many functions as needed
    ● Cheap: You only pay for real customer usage
    

    View Slide

57. Serverless Architectures
    

    View Slide

58. Serverless Asset Processing
    

    View Slide

59. Serverless Backend
    

    View Slide

60. Serverless Node.js
    const serverless = require('serverless-http');
    const koa = require('koa');
    // construct your app as normal
    const app = koa();
    // register your middleware as normal
    app.use(/* ... */);
    // this is it!
    module.exports.handler = serverless(app);
    https://github.com/dougmoscrop/serverless-http
    

    View Slide

61. Migrating Legacy: Strangler Pattern
    

    View Slide

62. Serverless GraphQL
    

    View Slide

63. How We Got Here
    Pros & Cons of the
    Serverless Architecture
    Background Why Concepts Code Next Steps
    Intro to Serverless Summary
    Walkthrough of a
    real Serverless
    application
    

    View Slide

64. Next Steps
    ● Serverless is the next natural progression of cloud computing
    ● To get started, start small
    ○ Extract a single feature from your monolith
    ○ Rewrite it in Serverless
    

    View Slide

65. Further Learning
    

    View Slide

66. leanpub.com/going-serverless
    

    View Slide

67. meetup.com/Serverless-Singapore
    

    View Slide

68. Thanks
    Yos Riady
    yos.io
    

    View Slide

69. From Instances to Functions
    Going Serverless
    Yos Riady
    yos.io
    goo.gl/hvrw9s
    

    View Slide

70. Appendix A: AWS Lambda Pricing
    ● If you allocated 128MB of memory to your function, executed it 30 million times
    in one month, and it ran for 200ms each time:
    ○ $5.83 / month
    ● Requests
    ○ $0.20 per 1 million requests
    ● Duration (100ms increments)
    ○ $0.00001667 for every GB-second used
    ● Charges also depend on your Lambda’s allocated memory
    ○ 256MB lambdas are 2x as expensive as 128MB lambdas per 100ms
    ● API Gateway
    ○ $3.50 per million API calls received, plus the cost of data transfer out,
    

    View Slide

71. View Slide

72. Appendix B: FaaS Code Execution
    ● When a lambda function is invoked,
    ○ A container (execution environment) is provisioned
    ● It takes time to set up and initialize the container
    ○ “Cold start” ⛄
    ● After execution, the container is maintained for some time (e.g. 5 minutes)
    ● “Warm” container reuse for subsequent invocations
    ○ Significantly faster
    

    View Slide

73. Appendix D: Debugging Serverless
    ● It’s a challenge to debug distributed systems
    ● Helpful techniques:
    ○ Logs! Logs! Logs!
    ○ Correlation ID (AWS Request ID)
    ○ Unit tests
    ● Helpful tools:
    ○ AWS X-Ray
    

    View Slide

74. Appendix E: Serverless Auth
    ● How do you do Authentication and / or Authorization with Serverless?
    ● JSON Web Tokens
    ○ An authorization provider issues an identity token
    ○ Services can verify the token without communicating with an authorization
    server, via a shared secret
    

    View Slide

75. Appendix F: AWS Lambda Limits
    

    View Slide

76. Appendix H: AWS Step Functions
    

    View Slide