mlops-practice

Practice repo for MLOps on Azure using terraform.

Terraform Concepts:

Basic commands:

• terraform init
• terraform plan
• terraform apply
• terraform destroy

Resources:

resource "<type>: "<name>" {
    ...
}

resource "azurerm_resource_group" "rg" {
  name      = "${var.resource_group_name}_${var.environment}"
  location  = var.location
}

resource "azurerm_virtual_network" "vnet" {
  name                = "vnet"
  location            = var.location
  address_space       = ["10.0.0.0/16"]
  resource_group_name = azurerm_resource_group.rg.name
  dns_servers         = []
}

variables can be cross refences by resource_type.resource_name.key pr resource_type.resource_name.id (some use IDs)

Data blocks:

If a resource already exists in the statefile, or if you want to reference that exists but your didn't create you can do that in a data block.

Providers:

• providers.tf

Helper functions the terraform uses to interact with external sercices/resouces. IE: local docker server, Cloud provider (Azure).

provider "aws" {
  alias  = "west"
  region = "us-west-2"
}

Variables: Variables are common values that can be declared and referenced elsewhere

• variables.tf

variable "location" { }

terraform.tfvars: terraform.tfvars holds the common values, is referrenced in variables.tf

location = "centralus"

Statefile:

• Terraform statefile only keeps readings of what it did last time.

• terraform.tfstate (Make sure it's kept in the .gitignore)

• terraform destroy can only destroy resources it created, by looking in the statefile.

• terraform.tfstate.backup is a backup of the statefile, you can use this to undo a destroy-

Terraform Modules:

• Way to package up terraform code to be used by other people.
• Directory with tf files in it.

module "name" {
  source = "./some_path"
  servers = 5
}

Terraform graph:

• terraform graph | dot -Tsvg > graph.svg

Terraform Import:

• Let's say that you have created any resources in Azure Cloud by some other means(manually or using azure CLI) before you started using Terraform. You can import them under Terraform management instead of destroying and re-creating them from scratch. The terraform import command can be used to import existing resources.
• only one resource at a time.

How-To's:

Azure CLI

• az login --use-device-code
• az account set --subscription "" if on more than 1 subscription
• az ad sp create-for-rbac --name aw-sp-1 --role Admin --scopes /subscriptions/df854c79-25cc-463f-9b9d-b5c918678a91
  • Take the output and add it as secrets in the github repo. This will be references in the workflow file.

Github Setup

• Add Terraform Workflow in actions
• From your Terraform Cloud User Settings, click on Tokens and generate an API token named GitHub Actions.
• Add the token to your Github repository as a secret. Name the secret TF_API_TOKEN.
• set on to workload_dispatch
• run to check connectivity and provisioning of resources to your Azure subscription.

Terraform Docs

• terraform-docs markdown table --output-file terraform_docs.md --output-mode inject .

Links

• IaC Overview
• terraform-docs
• Managing your machine learning infrastructure as code with Terraform

Deploying Azure Functions

zip -r test_app.zip ./test_app/
az functionapp deployment source config-zip   -g aw-rg-2-dev   -n awazfunctionapp-dev   --src ./test_app.zip

https://learn.microsoft.com/en-us/answers/questions/1401268/azure-functions-with-fastapi-and-openai-streaming