Skip to content

Latest commit

 

History

History
101 lines (84 loc) · 3.17 KB

README.md

File metadata and controls

101 lines (84 loc) · 3.17 KB
Raw

Mlflow on AWS Fargate

Description

This module runs Mlflow on AWS Fargate as a load-balanced Elastic Container Service.

By default, uses Elastic File System for backend storage and S3 for artifact storage. Optionally, a Relational Database Storage instance can be used for metadata.

Architecture

Mlflow on AWS Fargate Module Architecture

Inputs/Outputs

Input Parameters

Required

  • vpc-id: The VPC-ID that the ECS cluster will be created in.
  • subnet-ids: The subnets that the Fargate task will use.
  • ecr-repository-name: The name of the ECR repository to pull the image from.
  • artifacts-bucket-name: Name of the artifacts store bucket

Optional

  • ecs-cluster-name: Name of the ECS cluster.
  • service-name: Name of the service.
  • task-cpu-units: The number of cpu units used by the Fargate task.
  • task-memory-limit-mb: The amount (in MiB) of memory used by the Fargate task.
  • lb-access-logs-bucket-name: Name of the bucket to store load balancer access logs
  • lb-access-logs-bucket-prefix: Prefix for load balancer access logs
  • rds-hostname: Endpoint address of the RDS instance
  • rds-port: Port of the RDS instance
  • rds-credentials-secret-arn: RDS database credentials stored in SecretsManager
  • rds-security-group-id: Security group of the RDS instance
    • needed so that an inbound rule can be created to grant Fargate access to the database
  • efs-removal-policy: EFS removal policy. RETAIN by default. Allowed values: RETAIN, DESTROY.

Sample manifest declaration

name: mlflow-fargate
path: modules/mlflow/mlflow-fargate
parameters:
  - name: vpc-id
    valueFrom:
      moduleMetadata:
        group: networking
        name: networking
        key: VpcId
  - name: subnet-ids
    valueFrom:
      moduleMetadata:
        group: networking
        name: networking
        key: PrivateSubnetIds
  - name: ecr-repository-name
    valueFrom:
      moduleMetadata:
        group: storage
        name: ecr-mlflow
        key: EcrRepositoryName
  - name: artifacts-bucket-name
    valueFrom:
      moduleMetadata:
        group: storage
        name: buckets
        key: ArtifactsBucketName
  - name: rds-hostname
    valueFrom:
      moduleMetadata:
        group: database
        name: mlflow-mysql
        key: DatabaseHostname
  - name: rds-credentials-secret-arn
    valueFrom:
      moduleMetadata:
        group: database
        name: mlflow-mysql
        key: CredentialsSecretArn

Module Metadata Outputs

  • ECSClusterName: Name of the ECS cluster.
  • ServiceName: Name of the service.
  • LoadBalancerDNSName: Load balancer DNS name.
  • LoadBalancerAccessLogsBucketArn: Load balancer access logs bucket arn
  • EFSFileSystemId: EFS file system id.

Output Example

{
  "ECSClusterName": "mlops-mlops-mlflow-mlflow-fargate-EcsCluster97242B84-xxxxxxxxxxxx",
  "ServiceName": "mlops-mlops-mlflow-mlflow-fargate-MlflowLBServiceEBACC043-xxxxxxxxxxxx",
  "LoadBalancerDNSName": "xxxxxxxxxxxx.elb.us-east-1.amazonaws.com",
  "LoadBalancerAccessLogsBucketArn": "arn:aws:s3:::xxxxxxxxxxxx",
  "EFSFileSystemId": "fs-xxxxxxxxxxx",
}