Automatic modular rightsizing of Azure VM’s with special focus on Azure Virtual Desktop

It has long annoyed me that all the scaling options in Azure just add and remove hosts. They never target the host itself. Hosts are either under or overutilized in 84% of the case.

And this is especially relevant for AVD personal hostpools where users each have their own personal “VDI”.

So I’m releasing a custom PowerShell module called “ADDRS” (Azure Data Driven Right Sizing) that grabs mem/cpu performance of the VM or all VM’s in a resource group you tell it to check. It will then do some smart voodoo magic to determine what size out of an allowlist best fits.

Instructions / Example:

  • Use -WhatIf if you don’t want it to resize the VM
  • Use -Force if you want to resize a VM even if it is online (which will cause it to be shut down!)
  • Use -Boot if you want the VM to be started after resizing (by default it will stay deallocated)
  • Use -domain with your domain if your VM is domain joined
  • Use -region if your region is not westeurope
  • Use -Verbose if you want the full output incl financial projection
  • Use -Report if you want to output data to csv. Can be used together with -WhatIf
  • Modify minMemoryGB, maxMemoryGB, minvCPUs, maxvCPUs as desired for your usecase
  • You can adjust the preconfigured $allowedVMTypes array to allow specific VM types
  • use -maintenanceWindowStartHour, -maintenanceWindowLengthInHours and –maintenanceWindowDay if you want to ignore performance data during a maintenance window (e.g. for patching) as that isn’t representative
  • Set an Azure Tag called LCRightSizeConfig with the value disabled on machines you want to ignore
  • Set an Azure Tag called LCRightSizeConfig with a machine type value (e.g. “Standard_D4ds_v4“) if you want to lock a specific size for that machine, this can be useful if you want the script to resize from current to target automatically when it runs while the VM has been deallocated.

Example -Verbose output of two VM’s being resized:

Requirements:

The module requires that you’ve added the % Processor Time and Available MBytes performance counters to Azure Monitor:

and that your host(s) have the Azure Monitor agent installed.

The module will check if there is sufficient data about the machine in Azure Monitor, if not, no action will be taken. You can determine how far back the function looks by modifying $measurePeriodHours

Download / Installation

Option 1: Install-Module ADDRS

Option 2: get relevant functions/code from Git

and run the set-vmRightSize or set-rsgRightSize function, e.g.:

set-vmRightSize -targetVMName azvm01 -workspaceId 7ccd0949-2fd4-414e-b58c-c013cc6e445d

set-rsgRightSize -targetRSG1 rg-avd-we-01 -workspaceId 7ccd0949-2fd4-414e-b58c-c013cc6e445d

Right Sizing Frequency

It is recommended to match job schedules to the lookback period, or at least not run multiple times in the same lookback period. Otherwise, the data that is being used for sizing may not be representative if the machine had already been resized in an earlier run.

Issues

  • Make sure you’ve got enough data in Log Analytics
  • Make sure the allowedVMTypes list contains only VM types that you can actually upgrade to. If e.g. your VM has an ephemeral disk, and your allowList has types that do not, the resize will fail with an error message (but no harm will be done to the existing VM)
  • I’ve only tested the maintenance window parameters using UTC time, if you’re using different timezones your results in excluding data generated during the maintenance window may vary from mine

Keyvault RBAC model ARM role assignment

Yes, using ARM, not Bicep, I know, it’s bad!

Ran into a whole bunch of constrains and issue trying to assign an array of principals vs roles on keyvault using the RBAC access method, so sharing my working solution here as I couldn’t find a single good example on google:

        {
            "type": "Microsoft.KeyVault/vaults/providers/roleAssignments",
            "apiVersion": "2018-09-01-preview",
            "copy": {
                "name": "rbac-access-policy-loop",
                "count": "[length(parameters('accessPolicies'))]"
            },            
            "name": "[concat(variables('vaultName'),'/Microsoft.Authorization/',guid(concat(variables('vaultName'), parameters('accessPolicies')[copyIndex('rbac-access-policy-loop')].objectId, parameters('accessPolicies')[copyIndex('rbac-access-policy-loop')].roleId)))]",
            "dependsOn": [
                "[resourceId('Microsoft.KeyVault/vaults', variables('vaultName'))]"
            ],
            "properties": {
                "roleDefinitionId": "[concat('/providers/Microsoft.Authorization/roledefinitions/',parameters('accessPolicies')[copyIndex('rbac-access-policy-loop')].roleId)]",
                "principalId": "[parameters('accessPolicies')[copyIndex('rbac-access-policy-loop')].objectId]",
                "scope": "[resourceId('Microsoft.KeyVault/vaults', variables('vaultName'))]",
                "principalType": "Group"
            }
        }   

An example param would then look like this:

        "accessPolicies": {
            "value": [
                {
                    "roleId": "b86a8fe4-44ce-4948-aee5-eccb2c155cd7",
                    "objectId": "2d9cbd23-20b1-4921-a8e4-54b55161ad04"
                }                
            ]
        }  

ODBC SQL driver error

If you know how to clear the ODBC Driver 17’s token cache, please comment 🙂


Ran into the following error today and putting it here for searches as literally nothing came up:

Run-time error '-2147467259 (80004005)':
[Microsoft][ODBC Driver 17 for SQL Server][SQL Server]User option must be specified, if Authentication option is 'ActiveDirectoryInteractive'.

This was through an excel Macro, connecting to an Azure PaaS database. All other users were just getting prompted, but one kept having this issue.

We managed to ‘solve’ the issue by going into the macro properties and adding User ID=xxx@xxx.com to the connection string where xxx@xxx.com was the user’s actual UPN.