serverless.yml

service: aqts-capture-error-handler

provider:
  name: aws
  rolePermissionsBoundary:
    Fn::Sub:
      - "arn:aws:iam::${accountId}:policy/csr-Developer-Permissions-Boundary"
      - accountId:
         Ref: "AWS::AccountId"
  vpc: ${self:custom.vpc}
  region: ${opt:region, 'us-west-2'}
  stage: ${opt:stage, 'TEST'}
  runtime: python3.8
  memorySize: 128
  timeout: 45
  logRetentionInDays: 90
  deploymentBucket:
    name: ${opt:bucket, iow-cloud-applications}
  stackTags:
    "wma:project_id": "aqtscapture"
    "wma:subSystem": "aqtscapture"
    "wma:application_id": "Aquarius TS Capture Error Handler"
    "wma:contact": "Andrew Yan ayan@usgs.gov"
    "wma:role": etl
    "wma:system": NWIS
    "wma:environment": ${self:provider.stage}
    "wma:taggingVersion": 0.0.1
    "wma:costCenter": TBD
    "wma:organization": IOW
    "displayName": "Error Handler"
    commitIdentifier: ${git:sha1}

custom:
  exportGitVariables: false
  vpc:
    securityGroupIds: ${ssm:/iow/retriever-capture/${self:provider.stage}/securityGroupIds~split}
    subnetIds: ${ssm:/iow/aws/vpc/${self:provider.stage}/subnetIds~split}
  esLogs:
    endpoint: ${ssm:/iow/aws/vpc/${self:provider.stage}/elasticsearchEndpoint}
    index: ${self:service}
    retentionInDays: 90
    filterPattern: "?WARNING ?ERROR ?CRITICAL"
  prune:
    automatic: true
    number: 3
  mailingList: ${ssm:/aws/reference/secretsmanager/AQTS-CAPTURE-MAILING-LISTS~true}


functions:
  aqtsErrorHandler:
    handler: persist_error.handler.lambda_handler
    role:
      Fn::Sub:
        - arn:aws:iam::${accountId}:role/csr-Lambda-Role
        - accountId:
            Ref: AWS::AccountId
    onError:
      Ref: snsTopic
    reservedConcurrency: 290
    environment:
      AWS_SQS_QUEUE_URL:
        Ref: errorPersistQueue
      AWS_TERMINAL_QUEUE_URL:
        Ref: terminalErrorsQueue
      AWS_SNS_ARN:
        Ref: snsTopic
      AWS_DEPLOYMENT_REGION: ${self:provider.region}
      DEPLOY_STAGE: ${self:provider.stage}
      MAX_RETRIES: 6
      LOG_LEVEL: WARNING

resources:
  Resources:
    snsTopic:
      Type: AWS::SNS::Topic
      Properties:
        DisplayName: ${self:service}-${self:provider.stage}-topic
        TopicName: ${self:service}-${self:provider.stage}-topic
        Subscription:
          - Protocol: email
            Endpoint: ${self:custom.mailingList.TERMINAL_ERRORS_GROUP}
    invocationAlarm:
      Type: AWS::CloudWatch::Alarm
      Properties:
        AlarmName: ${self:service}-${self:provider.stage}-invocation-alarm
        AlarmDescription: Notify when invocations exceeds the specified threshold
        Namespace: 'AWS/Lambda'
        Dimensions:
          - Name: FunctionName
            Value:
              Ref: AqtsErrorHandlerLambdaFunction
        MetricName: Invocations
        Statistic: Sum
        ComparisonOperator: GreaterThanOrEqualToThreshold
        Threshold: 1000
        Period: 60
        EvaluationPeriods: 5
        AlarmActions:
          - Ref: snsTopic
    concurrencyAlarm:
      Type: AWS::CloudWatch::Alarm
      Properties:
        AlarmName: ${self:service}-${self:provider.stage}-concurrency-alarm
        AlarmDescription: Notify when concurrency exceeds the specified threshold
        Namespace: 'AWS/Lambda'
        Dimensions:
          - Name: FunctionName
            Value:
              Ref: AqtsErrorHandlerLambdaFunction
        MetricName: ConcurrentExecutions
        Statistic: Maximum
        ComparisonOperator: GreaterThanOrEqualToThreshold
        Threshold: 150
        Period: 60
        EvaluationPeriods: 1
        AlarmActions:
          - Ref: snsTopic
    errorAlarm:
      Type: AWS::CloudWatch::Alarm
      Properties:
        AlarmName: ${self:service}-${self:provider.stage}-error-alarm
        AlarmDescription: Notify when number of errors exceeds the specified threshold
        Namespace: 'AWS/Lambda'
        Dimensions:
          - Name: FunctionName
            Value:
              Ref: AqtsErrorHandlerLambdaFunction
        MetricName: Errors
        Statistic: Sum
        ComparisonOperator: GreaterThanOrEqualToThreshold
        Threshold: 10
        Period: 60
        EvaluationPeriods: 1
        AlarmActions:
          - Ref: snsTopic
    errorPersistQueue:
      Type: AWS::SQS::Queue
      Properties:
        QueueName: aqts-capture-error-queue-${self:provider.stage}
        MessageRetentionPeriod: 1209600
    terminalErrorsQueue:
      Type: AWS::SQS::Queue
      Properties:
        QueueName: aqts-capture-terminal-errors-queue-${self:provider.stage}
        MessageRetentionPeriod: 345600

plugins:
  - serverless-plugin-git-variables
  - serverless-python-requirements
  - serverless-es-logs
  - serverless-prune-plugin

package:
  exclude:
    - node_modules/**
    - Dockerfile
    - .dockerignore
    - Jenkinsfile
    - package.json
    - package-lock.json