Autoscaling IaaS VMs in ARM mode from a template - azure

Question

I've created an template-based deployment that over-provisions a number of Linux VMs. I'd like to autoscale them as per classic instances, where Azure will turn on/turn off instances according to CPU load.
Is this possible with ARM mode? And if not, is there a suggested alternative method? The only examples I can find are around using Application Insights and PaaS functionality. I've got a Python app running in Docker on Ubuntu hosts.

Answer 1

For IaaS, you must use virtual machine scale sets to use autoscale, else you need to stick with PaaS (web apps).

Answer 2

For this you would first need to create an availability group for the VMs. The resource decleration in the ARM template looks something like this:
{
"type": "Microsoft.Compute/availabilitySets",
"name": "[variables('availabilitySetName')]",
"apiVersion": "2015-05-01-preview",
"location": "[parameters('location')]",
"properties": {
"platformFaultDomainCount": "2"
}
}
Then for the virtual machine resource the decliration in the ARM Template would look something like this:
{
"apiVersion": "2015-05-01-preview",
"type": "Microsoft.Compute/virtualMachines",
"name": "[concat(variables('vmName'), '0')]",
"location": "[parameters('location')]",
"dependsOn": [
"[concat('Microsoft.Storage/storageAccounts/', parameters('newStorageAccountName'))]",
"[concat('Microsoft.Network/networkInterfaces/', variables('nicName'), '0')]",
"[concat('Microsoft.Compute/availabilitySets/', variables('availabilitySetName'))]"
],
"properties": {
"availabilitySet": {
"id": "[resourceId('Microsoft.Compute/availabilitySets', variables('availabilitySetName'))]"
},
...},
The quckstart templates are a good ref: https://raw.githubusercontent.com/azure/azure-quickstart-templates/master/201-2-vms-2-FDs-no-resource-loops/azuredeploy.json
Once you have two or more VMs of the same size in an availability set, you would configure autoscale using microsoft.insights/autoscalesettings, which I beleive you referenced in the question. This is done at the cloud service so it will work similar to PaaS... like so:
{
"apiVersion": "2014-04-01",
"name": "[concat(variables('vmName'), '-', resourceGroup().name)]",
"type": "microsoft.insights/autoscalesettings",
"location": "East US",
...},
A pretty good example is here: https://raw.githubusercontent.com/Azure/azure-quickstart-templates/6abc9f320e39d9d75dffb60846e88ab80d3ff33a/201-web-app-sql-database/azuredeploy.json
I also setup autoscale using the portal first and reviewed ARMExplorer to get a better idea of how things should look in my code. ARMExplorer is here: Azure Resource Explorer