ml_ai

GenAI & ML inference project deployment

This directory contains the Terraform code to spawn a Google Cloud project with all the necessary infrastructure and configuration required for running the GenAI & ML inference solution guide.

These deployment scripts are part of the Dataflow Gen AI & ML solution guide.

Bill of resources created by this script

The scripts will create the following resources

Resource	Name	Description
Project	Set by user	Optional, you may reuse an existing project. If the project is created by Terraform, it will enable the APIs for Cloud Build, Dataflow, Monitoring, Pub/Sub, Dataflow Autoscaling, and Artifact Registry
Docker registry	dataflow-containers	An Artifact Registry Docker repo for the custom Dataflow container used in the pipeline. The Cloud Build service agent is granted admin role in this repository. The Dataflow service account is granted reader role. By default, only the 3 latest versions of each image are kept in the repo.
Bucket	Same as project id	Using the standard storage class, this is a regional bucket in the region specified by the user.
Pub/Sub topic	messages	The input Pub/Sub topic for the sample pipeline.
Pub/Sub topic	predictions	The output Pub/Sub topic for the sample pipeline.
Pub/Sub subscription	messages-sub	The subscription to the messages topic that is actually used by the Dataflow pipeline.
Pub/Sub subscription	predictions-sub	The subscription to the predictions topic, useful to visualize the messages produced by the pipeline.
Service account	my-dataflow-sa	Dataflow worker servive account. It has storage admin, Dataflow worker, metrics writer and Pub/Sub editor roles assigned at project level.
VPC network	dataflow-net	If the project is created from scratch, the default network is removed and this network is re-created with a single regional sub-network.
Cloud NAT	dataflow-nat	Cloud NAT in the region specified by the user, in case the Dataflow workers need to reach the Internet. This is not necessary for the sample pipeline provided.
Firewall rules	Several rules	Ingress and egress rules to remove unnecessary traffic, and to ensure the traffice required by Dataflow. If you want to access a VM using SSH, apply the network tag ssh to that instance. Same for http-server and for https-server

Configuration variables

This deployment accepts the following configuration variables:

Variable	Type	Description
billing_account	string	Optional. Billing account to be used if the project is created from scratch.
organization	string	Optional. Organization (or folder) number, where the project will be created. Only required if you are creating the project from scratch. Use the format organizations/XXXXXXX or folder/XXXXXXXX.
project_create	boolean	Set to false to reuse an existing project. Or to true to create a new project from scratch.
project_id	string	Project Id.
region	string	Region to be used for all the resources. The VPC will contain only a single sub-network in this region.
destroy_all_resources	bool	Optional. Default true. Destroy buckets and the Spanner instance with tf destroy .
network-prefix	string	Optional. Default "dataflow". Add a prefix to the network net, subnet, NAT

The default values of all the optional configuration variables are set for development projects. For a production project, you should change destroy_all_resources to false.

How to deploy

1. Set the configuration variables:
   • Create a file named terraform.tfvars in the current directory.
   • Add the following configuration variables to the file, replacing the values with your own:

     billing_account = "YOUR_BILLING_ACCOUNT"
     organization = "YOUR_ORGANIZATION_ID"
     project_create = true/false
     project_id = "YOUR_PROJECT_ID"
     region = "YOUR_REGION"
     

   • If this is a production deployment, make sure you change also the optional variables.
2. Initialize Terraform:
   • Run the following command to initialize Terraform:

     terraform init

3. Apply the configuration:
   • Run the following command to apply the Terraform configuration:

     terraform apply

4. Wait for the deployment to complete:
   • Terraform will output the status of the deployment. Wait for it to complete successfully.
5. Access the deployed resources: You are now ready to launch the sample pipeline in this solution guide.

How to remove

The setup will be continuously consuming as this is a streaming architecture, running without stop.

BEWARE: THE COMMAND BELOW WILL DESTROY AND REMOVE ALL THE RESOURCES.

To destroy and stop all the resources, run:

terraform destroy