README.md

Deploy to AWS

Automation via GitHub Actions

A push to master will trigger infrastructure deployment to AWS using Github acitons defined in apply.yml

Also its possible to trigger the deployment manually using Github Actions

If the "Destroy the infra" option is selected the infra will be destroyed instead of applying the changes

Manual deployment to AWS from local

To deploy to AWS from local manually follow the create-aws-infra.sh script

Local development

to make localstack work with terraform we need to use terraform-local more info here https://docs.localstack.cloud/user-guide/integrations/terraform/

Also, for convenience we'll install awslocal command
https://docs.localstack.cloud/user-guide/integrations/aws-cli/

Conda environment created in the main repo needs to be activated (eg it has pipenv already installed and uses a proper python version):

conda activate capitalbikeshare-mlops

To have both tflocal and awslocal available:

pipenv install
pipenv shell

To test everything locally you can just run Important For this to run you need to have a docker image capitalbikeshare-service-prediction-service:latest defined in the service repo built locally
Also make sure that the port 80 is available

./test-infra.sh

It will execute all the following steps:

To be able to use localstack-pro image that supports ECR and ECS one needs to sign up for localstack free trial and have an environment variable LOCALSTACK_API_KEY set. On Mac/Linux:

export LOCALSTACK_API_KEY=<your_localstack_api_key>

Then we need to spin up container with localstack

docker compose up -d

First we need to make a bucket manually to hold the tfstate

Make bucket

awslocal s3 mb s3://terraform-state-mlops-zoomcamp

Check that the bucket is there just in case

awslocal s3 ls

Export environment name that we wish to work with. For instance, stg

export TF_ENV="stg"

To init tflocal with the corresponding s3 backend configuration (for staging/prod) Eg the following will do the init with the staging backend

tflocal init -reconfigure -backend-config=environments/${TF_ENV}/backend.config

First we need to create only ECR repo to be able to push an image

tflocal plan -target=aws_ecr_repository.prediction_service_ecr_repo -var-file="environments/${TF_ENV}/environment.tfvars" -out=ecr_repo.tfplan
tflocal apply ecr_repo.tfplan

Tag existing image (built in the other repo) and push to the ECR

image_name_with_tag="$(tflocal output -raw ecr_repository_url):latest"
docker tag capitalbikeshare-service-prediction-service:latest $image_name_with_tag
docker push $image_name_with_tag

See the plan

tflocal plan -var-file="environments/${TF_ENV}/environment.tfvars"

Apply (needs confirmation)

tflocal apply -var-file="environments/${TF_ENV}/environment.tfvars"

Run healthcheck

curl http://localhost/healthcheck

Run prediction

curl -X POST http://localhost/predict -H "Content-Type: application/json" -d '{"start_station_id": "31205","end_station_id": "31239","rideable_type": "docked_bike","member_casual": "member","started_at": "2020-04-06 07:54:59"}'

Destroy (needs confirmation)

tflocal destroy -var-file="environments/${TF_ENV}/environment.tfvars"

Requirements

The following requirements are needed by this module:

• terraform (>= 1.5)

• aws (~> 5.9)

Providers

The following providers are used by this module:

• aws (5.13.1)

Resources

The following resources are used by this module:

• aws_alb.application_load_balancer (resource)
• aws_default_subnet.default_subnet_a (resource)
• aws_default_subnet.default_subnet_b (resource)
• aws_default_vpc.default_vpc (resource)
• aws_ecr_repository.prediction_service_ecr_repo (resource)
• aws_ecs_cluster.ecs_cluster (resource)
• aws_ecs_service.app_service (resource)
• aws_ecs_task_definition.service_task (resource)
• aws_iam_role.ecsTaskExecutionRole (resource)
• aws_iam_role_policy_attachment.ecsTaskExecutionRole_policy (resource)
• aws_lb_listener.listener (resource)
• aws_lb_target_group.target_group (resource)
• aws_security_group.load_balancer_security_group (resource)
• aws_security_group.service_security_group (resource)

Required Inputs

The following input variables are required:

cpu

Description: ECS CPU to set to the tasks.

Type: number

desired_count

Description: How many ECS Tasks to initialy deploy per service.

Type: number

environment

Description: Environment name

Type: string

memory

Description: ECS memory to set to the tasks.

Type: number

region

Description: AWS Region where to create the resources.

Type: string

Optional Inputs

The following input variables are optional (have default values):

ecr_repo_name

Description: ECR repo name

Type: string

Default: "capitalbikeshare-dv-model-pipeline"

Outputs

The following outputs are exported:

app_url

Description: Log the load balancer app url

container_name

Description: n/a

ecr_repository_url

Description: n/a

ecs_cluster

Description: n/a

ecs_service

Description: n/a

task_definition_json

Description: n/a