AWSTemplateFormatVersion: '2010-09-09'
Transform: 'AWS::Serverless-2016-10-31'
Description: 'Meter Data Lake prediction pipeline configuration. (qs-1r18anahd)'
Metadata:
  cfn-lint:
    config:
      configure_rules:
        RuleId: E9007
      ignore_checks:
        - E9007
      ignore_reasons:
        - "A combination of Serverless Transform and metadata not being passed through (yet) means that we need to globally exclude E9007 until an upstream workaround is available."
Globals:
  Function:
    Runtime: python3.7
    Timeout: 240

Resources:
  MachineLearningPipelineConfig:
    Type: AWS::Serverless::SimpleTable
    Properties:
      PrimaryKey:
        Name: name
        Type: String

  InitializeConfigTableFunction:
    Type: AWS::Serverless::Function
    Properties:
      Handler: index.lambda_handler
      InlineCode: |
          import json
          import boto3
          import cfnresponse
          import logging
          import uuid

          client = boto3.resource('dynamodb')
          sagemaker = boto3.client('sagemaker')

          def get_config(name):
            response = DYNAMODB.Table(CONFIG_TABLE_NAME).get_item(
                Key={'name': name}
            )
            return response['Item']["value"]

          def lambda_handler(event, context):
              table_name = event['ResourceProperties']['DynamoTableName']
              table = client.Table(table_name)
              status = cfnresponse.SUCCESS

              if event['RequestType'] == 'Create':
                  items = [
                    { "name": "Data_start", "value": "2013-06-01" },
                    { "name": "Data_end", "value": "2014-01-01" },
                    { "name": "Forecast_period", "value": 7 },
                    { "name": "Training_samples", "value": 50 },
                    { "name": "Training_instance_type", "value": "ml.c5.2xlarge" },
                    { "name": "Endpoint_instance_type", "value": "ml.m5.xlarge" },
                    { "name": "ML_endpoint_name", "value": "ml-endpoint-{}".format(str(uuid.uuid4())) },
                    { "name": "Meter_start", "value": 1 },
                    { "name": "Meter_end", "value": 100 },
                    { "name": "Batch_size", "value": 25 }
                  ]

                  try:
                    with table.batch_writer() as batch:
                          for item in items:
                              batch.put_item(Item=item)
                  except Exception as e:
                      logging.error('Exception: %s' % e, exc_info=True)
                      status = cfnresponse.FAILED
                  finally:
                      cfnresponse.send(event, context, status, {}, None)

              elif event['RequestType'] == 'Delete':
                  # deleting sagemaker endpoint
                  endpoint_name = table.get_item(
                      Key={'name': 'ML_endpoint_name'}
                  )['Item']['value']

                  try:
                    sagemaker.delete_endpoint(EndpointName=endpoint_name)
                  except Exception as e:
                      logging.error('Exception: %s' % e, exc_info=True)
                  finally:
                      cfnresponse.send(event, context, status, {}, None)

      Policies:
        - Statement:
          - Sid: SageMakerDeleteEndpoint
            Effect: Allow
            Action: 'sagemaker:DeleteEndpoint'
            Resource: !Sub 'arn:${AWS::Partition}:sagemaker:::*'
        - DynamoDBCrudPolicy:
            TableName: !Ref MachineLearningPipelineConfig

  InitializeDynamoDB:
    Type: Custom::InitializeConfigTableFunction
    Properties:
      ServiceToken: !GetAtt 'InitializeConfigTableFunction.Arn'
      DynamoTableName:
        Ref: MachineLearningPipelineConfig

Outputs:
  MachineLearningPipelineConfigTable:
    Value: !Ref MachineLearningPipelineConfig