rust-ml-template.yml

AWSTemplateFormatVersion: '2010-09-09'
Transform: AWS::Serverless-2016-10-31
Description: >
  AWS Lambda Rust ML -- Example showing how to use Rust with AWS Lambda to build a customer churn predictor

Globals:
  Function:

Resources:
  S3ModelBucket:
    Type: AWS::S3::Bucket
    Properties:
      BucketName: !Sub
          - 'lambda-rust-ml-model-bucket-${AccountID}'
          - AccountID: !Ref AWS::AccountId
      VersioningConfiguration:
        Status: Enabled
  CustomerDataStream:
    Type: AWS::Kinesis::Stream
    Properties:
      ShardCount: 1
      RetentionPeriodHours: 24
  ResultQueue:
    Type: AWS::SQS::Queue
    Properties:
      VisibilityTimeout: 300
  CustomerChurnFunction:
    Type: AWS::Serverless::Function 
    Properties:
      CodeUri: target/lambda/aws-lambda-rust-ml
      MemorySize: 128
      Architectures: ["arm64"]
      Handler: bootstrap
      Runtime: provided.al2
      Timeout: 5
      Tracing: Active
      Environment:
        Variables:
          RUST_LOG: info
          S3_MODEL_LOCATION: !Ref S3ModelBucket

      Events:
        Stream:
          Type: Kinesis
          Properties:
            Stream: !GetAtt CustomerDataStream.Arn
            BatchSize: 100
            StartingPosition: LATEST
      Environment:
        Variables:
          STREAM_ARN: !Ref CustomerDataStream
          QUEUE_ARN: !Ref ResultQueue
      Policies:
        - KinesisStreamReadPolicy :
            StreamName: !Ref CustomerDataStream
        - S3ReadPolicy:
            BucketName: !Ref S3ModelBucket
    Metadata:
      BuildMethod: makefile
Outputs:
  CustomerChurnFunction:
    Description: "Customer Churn Prediction Function"
    Value: !GetAtt CustomerChurnFunction.Arn
  PutTable:
    Description: "Customer Profile Data Stream"
    Value: !GetAtt CustomerDataStream.Arn

Just in case the user wants to deploy this directly: add an AWS account ID to the end of the bucket name, otherwise the bucket name will not be unique.

Do we expect more lambda functions in the future? If not, simply move all the settings in the Global section into the function itself, so you don't need to explain what Global is.

Is it required to use an ARM CPU? What is the development platform you are going to use? An Apple M-series CPU? The cross-compilation might be a little tedious to explain.

Is it required to use an ARM CPU? What is the development platform you are going to use? An Apple M-series CPU? The cross-compilation might be a little tedious to explain.

I would like to use an ARM CPU to showcase Lambda cost reduction