automaticdavid / demo_cat_azure

Is it cat ? Azure version.

This repo sets up your exisitng Ansible Tower install to deploy the "Is It Cat?" application in Azure.

"Is It Cat" is a simple image claissification Python app that comes with 2 (pre-trained) model versions: one that only recognizes cars (and thinks that some cats are cars), and one that recognizes cars and cats.

What you need

A Tower setup with an isolated node in Azure. This node needs git to be installed and the jmespath pip module in the default venv.

Go to the lab/ directory and complete up the lab-credentials.yml.txt file.

Rename it to lab-credentials.secret.yml
Encrypt it with ansible-vault if you feel like it.

Run: ansible-playbook --vault-password-file=vault-password.secret.txt tower_setup.yml

This will create an organization called "meteor", a project, an inventory, credential types, credentials and the template needed for the demo.

Run the demo

Just run the tower job template named: ML Cat - Azure (dev) - Provision

This will create the needed Azure components: ressource groups, virtnet, storage account and blob storage, loadbalancer, nics, and (by default, 2 VMs). You can use extra_vars to override the default values found in the default of the Azure role that is used (see: requirements.yml).

It will then download and install the specified version of the "Is it cat" application and its appropriate TensorFlow model on both VMs.

The extra_vars configured in the template are:
app: the url for the "Is it cat" app
version: the git tag of the app that will be deployed
az_count: the number of instances you want (scale up an existing deployement)
az_vm_size: the azure size of the VMs
object_store: tells the app which storage backend to use, for azure it's blob
scope: the prefix that will get use in azure for all the VMs