Custom Script Extensions for ML Model Deployment Automation

Question

Pulumi · Accepted Answer

To automate ML model deployment with custom scripts on Azure, you can utilize Pulumi with the Azure Native provider. Pulumi allows you to define infrastructure as code using familiar programming languages such as Python, making it easier to automate tasks like machine learning model deployments.

Below is a Pulumi program in Python that demonstrates how to create an Azure Virtual Machine (VM), then use the Azure Native `VirtualMachineExtension` to deploy a custom script extension to the VM. This custom script could be used to automate the deployment of a machine learning model.

First, you need to create a virtual machine. We'll assume that you've already got a Resource Group set up. Then, after the VM is created, you can add the `VirtualMachineExtension` which allows you to run a custom script. This script might be something that installs necessary dependencies, downloads your ML model, and runs it.

Here's a basic program that demonstrates these steps:

```python
import pulumi
import pulumi_azure_native as azure_native

# Create a new virtual machine
vm = azure_native.compute.VirtualMachine("myVirtualMachine",
    resource_group_name="myResourceGroup",
    # Define hardware, image, OS, and admin details
    hardware_profile=azure_native.compute.HardwareProfileArgs(
        vm_size="Standard_DS1_v2",
    ),
    storage_profile=azure_native.compute.StorageProfileArgs(
        image_reference=azure_native.compute.ImageReferenceArgs(
            publisher="Canonical",
            offer="UbuntuServer",
            sku="18.04-LTS",
            version="latest",
        ),
        os_disk=azure_native.compute.OSDiskArgs(
            caching=azure_native.compute.CachingTypes.READ_WRITE,
            managed_disk=azure_native.compute.ManagedDiskParametersArgs(
                storage_account_type=azure_native.compute.StorageAccountTypes.PREMIUM_LRS,
            ),
            name="myOsDisk",
            create_option="FromImage",
        ),
    ),
    os_profile=azure_native.compute.OSProfileArgs(
        admin_username="testadmin",
        computer_name="myVM",
        admin_password="SecurePassw0rd",
    ),
    network_profile=azure_native.compute.NetworkProfileArgs(
        # You need an existing network interface ID ready here
        network_interfaces=[
            azure_native.compute.NetworkInterfaceReferenceArgs(
                id="/subscriptions/{subscription_id}/resourceGroups/myResourceGroup/providers/Microsoft.Network/networkInterfaces/myNIC",
                primary=True,
            ),
        ],
    ),
    location="eastus",
)

# Custom script extension to setup ML model
vm_extension = azure_native.compute.VirtualMachineExtension("myVMExtension",
    # The name of the extension should be unique within the VM
    vm_extension_name="CustomMLModelSetupScript",
    resource_group_name="myResourceGroup",
    publisher="Microsoft.Azure.Extensions",
    type="CustomScript",
    type_handler_version="2.0",
    # Assumes you have this script available in a location accessible by Azure
    settings={
        "fileUris": ["https://storageaccountname.blob.core.windows.net/scripts/setup_ml_model.sh"],
        "commandToExecute": f"bash setup_ml_model.sh"
    },
    virtual_machine_name=vm.name,
    location="eastus",
)

pulumi.export('virtual_machine_name', vm.name)
pulumi.export('vm_extension_name', vm_extension.name)
```

In this code:
- We are creating an Azure virtual machine with a specified size, image, OS, and admin details within a given resource group and location.
- The `azure_native.compute.VirtualMachineExtension` is used to run a custom script on the virtual machine. This script is expected to be a shell script hosted at the provided URL.
- The `type` property for the extension is set to `CustomScript`, which is supported by Microsoft Azure to run custom scripts on virtual machines.
- The `commandToExecute` specifies the exact command to run once the script is fetched by the VM.

Please replace placeholders like `"myResourceGroup"`, `"myVM"`, `"SecurePassw0rd"`, `"/subscriptions/{subscription_id}/resourceGroups/myResourceGroup/providers/Microsoft.Network/networkInterfaces/myNIC"`, and `"https://storageaccountname.blob.core.windows.net/scripts/setup_ml_model.sh"` with your actual Resource Group, VM name, Administrator Username and Password, the network interface ID, and the storage location of your script respectively.

Finally, `pulumi.export` is used to output the name of the VM and the VM extension, which could be useful for cross-reference or querying it later.

Make sure you have Pulumi and Azure CLI installed and configured with the necessary permissions, then you can run the program using the Pulumi CLI commands `pulumi up` to deploy or update the infrastructure as per the defined program.