“Psst… Hey buddy, want to keep your plain text password in your source control?”

Hopefully, you answer is “No”, but if you’d been following along with this JSON/ARM template series of posts, you will have noticed that I’ve defaulted to deployments with straight up, plain text, username/passwords in my template examples. So before we go any further, lets secure this up a bit!

Goal: Deploy a Windows Nano VM using an ARM template in GitHub without having to put the Admin username or password anywhere in plain text.

To do this, we’ll need a place to hide the secrets. In this case, we’ll use Azure Key Vault. With Key Vault, you create a resource group and vault (using PowerShell or an ARM template) and then load up some secrets or keys. You can hide basic passwords or load certificates.

I’m going to create my Key Vault using PowerShell. I’m purposely creating my vault in its own Resource Group. You can create the Key Vault in the same Resource Group as the rest of your deployment, but then if you delete the whole group for during testing, you’ll have to recreate the vault each time.

1. Log into Azure
2. Get your Subscription details, copy the GUID of the subscription you want to use for the key vault into the $VSSubID variable.
3. Customize the variables for the Resource Group name, Vault name and deployment region.
4. Once the vault is created, customize the “SuperSecretPassword” and “FancyAdminName” to your liking and add those secrets to the vault.

Login-AzureRmAccount
Get-AzureRmSubscription
$VSSubID = {subcription_guid}  
Set-AzureRmContext -SubscriptionID $VSSubID

$RGName = "KeyVaultRG"
$VaultName = "MyVaultName"
$Location = "West US"

New-AzureRmResourceGroup -Name $RGName -Location $Location

New-AzureRmKeyVault -VaultName $VaultName -ResourceGroupName $RGName -Location $Location

Set-AzureRmKeyVaultAccessPolicy -VaultName $VaultName -EnabledForTemplateDeployment -ResourceGroupName $RGName

$secretvalue = ConvertTo-SecureString 'SuperSecretPassword' -AsPlainText -Force
$secret = Set-AzureKeyVaultSecret -VaultName $VaultName -Name 'ServerAdminPassword' -SecretValue $secretvalue

$secretvalue = ConvertTo-SecureString 'FancyAdminName' -AsPlainText -Force
$secret = Set-AzureKeyVaultSecret -VaultName $VaultName -Name 'ServerAdminName' -SecretValue $secretvalue

Now that my vault is crated, I’ll need to edit my ARM template to grab the secrets from the Key Vault when my VM is deployed. Make the change to your parameters.json file, not your deployment template. This way you can use the same template deployment with different subscriptions and vaults easily. Instead of using the basic string type for your adminUsername and adminPassword, reference the secrets you added to the vault.

"adminUsername": {
      "reference": {
      "keyVault": {
        "id": "/subscriptions/{guid}/resourceGroups/KeyVaultRG/providers/Microsoft.KeyVault/vaults/ImperfectKeyVault"
      }, 
      "secretName": "ServerAdminName" 
       } 
    },
   "adminPassword": {
      "reference": {
      "keyVault": {
        "id": "/subscriptions/{guid}/resourceGroups/KeyVaultRG/providers/Microsoft.KeyVault/vaults/ImperfectKeyVault"
      }, 
      "secretName": "ServerAdminPassword" 
       } 
    }

Now, when you run your deployment, the admin username and password will be pulled from the vault and passed to your VM during creation. If you go to log in that server directly, you’ll need to provide the plain text versions of those secrets just like usual.  It’s worth noting that not anyone can pull keys out of your vault.  By default only the subscription admin will be able to access the contents of the vault, but you can add additional permissions to a vault to allow for other users and automation tools to access the same secrets.

You can find the example of this template in my GitHub repo.

Learn more about Azure Key Vault at https://azure.microsoft.com/en-us/documentation/articles/key-vault-get-started/

So here I am, in JSON-land, working on my deployment for VMs. But so far, I’ve only deployed one VM in my network. But one VM is boring. I want to be able to rack and stack more than one machine at a time! 🙂  Enter the copy object and the copyIndex function!  You can read the official documentation about it, but let see how this works within my existing template. Things to note:

• You can only reference the copyIndex function in RESOURCES. Don’t try to build variables with them, you’ll just be sad.
• You can’t use a copy loop in a nested resource (this blog series hasn’t encountered this yet).
• You can’t use them to loop items that are defined as properties within a resource, like multiple data disks.

So in my example, let’s extend the template to deploy 3 VMs in the same subnet and put them all in the same NSG (network security group). First, you must declare in the parameters that you’d like count something:

   "count": { 
      "type": "int", 
      "defaultValue": 3 
    },

Now, whenever you want to reference the “current iteration” of the loop elsewhere in your template, you’ll use copyIndex(). Another thing to note – computers like to count starting at zero, humans like to count starting at 1. If we stick with copyIndex(), we’ll get VMs numbered 0, 1, 2. If you’d like them to be numbered 1, 2, 3 use copyIndex(1), which will offset the numbering by 1.

Since this template is going to create 3 VMs, we will need to make sure to account for other resources that VMs have a one to one relationship with and be sure to loop those as well. The only items that we are required to loop are the network interfaces and the public IP addresses (which have individual DNS names).  We don’t have to loop things like the storage account (which can be used by multiple VMs) or the network, since we only need one in this case.

In the resource for the publicIPAddresses, you’ll see the reference to the copyIndex() in the name, and in the DNS domain label. In each case, the incremented number will be tacked on at the end of each. Then finally, you’ll see the specific reference to looping this resource added, with the name of “pubipcopy”, pulling the count from the parameter you set early.

     {
      "type": "Microsoft.Network/publicIPAddresses",
      "apiVersion": "2015-06-15",
      "name": "[concat(parameters('publicIPAddressName'), parameters('dnsNameForPublicIP'),copyIndex())]", 
      "location": "[parameters('location')]",
      "properties": {
        "publicIPAllocationMethod": "[variables('publicIPAddressType')]",
        "idleTimeoutInMinutes": 4,
        "dnsSettings": {
          "domainNameLabel": "[concat(parameters('dnsNameForPublicIP'),copyIndex())]"
        }
      },
       "copy": {
        "name": "pubipcopy",
        "count": "[parameters('count')]"
      }
     },

One of the reasons you need to name each resource copy object is so they can be used in places where dependencies are required.

For the network interface, you’ll see I’ve made similar references to the copyIndex() in the name. Also a network interface is dependent on the existence of a public IP address, so you’ll see “pubipcopy” referenced under “dependsOn” as well as in the IP configuration properties.

   {
      "apiVersion": "2015-06-15",
      "type": "Microsoft.Network/networkInterfaces",
      "name": "[concat(parameters('nicName'), copyIndex())]",
      "location": "[parameters('location')]",
    
      "dependsOn": [
        "pubipcopy",
        "[concat('Microsoft.Network/virtualNetworks/', parameters('vnetName'))]",
        "[resourceId('Microsoft.Network/networkSecurityGroups', parameters('networkSecurityGroupName'))]"
      ],
      "properties": {
        "ipConfigurations": [
          {
            "name": "ipconfig1",
            "properties": {
             "privateIPAllocationMethod": "Dynamic",
             "publicIPAddress": {
                "id": "[resourceId('Microsoft.Network/publicIPAddresses', concat(parameters('publicIPAddressName'),parameters('dnsNameForPublicIP'),copyIndex()))]"
              },
              "subnet": {
                "id": "[variables('subnetRef')]"
              }
              }
          }
        ],
       "dnsSettings": {
         "dnsServers": []
       },
       "enableIPForwarding": false,
       "networkSecurityGroup": {
               "id": "[resourceId('Microsoft.Network/networkSecurityGroups', parameters('networkSecurityGroupName'))]"
            }
      },
       "copy": {
        "name": "niccopy",
        "count": "[parameters('count')]"
      }
    },

This all comes together when you get to the VM resource where it refers to the NIC in the network profile and using the copyIndex() when creating the VM name, computer name and OS disk.

    {
      "apiVersion": "2015-06-15",
      "type": "Microsoft.Compute/virtualMachines",
      "name": "[concat('NanoVM-', copyIndex())]", 
      "location": "[parameters('location')]",
      "copy": {
        "name": "vmcopy",
        "count": "[parameters('count')]"
        },
      "dependsOn": [
        
        "[concat('Microsoft.Storage/storageAccounts/', parameters('newStorageAccountName'))]",
        "niccopy"
      ],
      "properties": {
        "hardwareProfile": {
          "vmSize": "[parameters('vmSize')]"
        },
        "osProfile": {
          "computerName": "[concat('NanoVM-', copyIndex())]",
          "adminUsername": "[parameters('adminUsername')]",
          "adminPassword": "[parameters('adminPassword')]"
        },
        "storageProfile": {
          "imageReference": {
            "publisher": "[variables('imagePublisher')]",
            "offer": "[variables('imageOffer')]",
            "sku": "[parameters('windowsOSVersion')]",
            "version": "latest"
          },
          "osDisk": {
            "name": "[concat('osdisk_', concat('NanoVM-', copyIndex()))]" ,
            "vhd": {
              "uri": "[concat('http://',parameters('newStorageAccountName'),'.blob.core.windows.net/',variables('vmStorageAccountContainerName'),'/NanoVM-',copyIndex(),uniqueString(resourceGroup().id),'-osdisk.vhd')]"
            },
            "caching": "ReadWrite",
            "createOption": "FromImage"
          }
          },
            
          "networkProfile": {
          "networkInterfaces": [             
        {
              "id": "[resourceId('Microsoft.Network/networkInterfaces',concat(parameters('nicName'), copyIndex()))]"
            }
          ]
        },
          "diagnosticsProfile": {
          "bootDiagnostics": {
             "enabled": "true",
             "storageUri": "[concat('https://',parameters('newStorageAccountName'),'.blob.core.windows.net')]"
          }
        }
        }

      }

You can find my full template here: https://github.com/techbunny/Templates/tree/master/multiple_nano_server

There is more than one way to deploy an ARM template once you’ve created or edited on to your liking. The first way is to copy the code into a blank template deployment in the Azure portal. This has a few benefits, primarily that it gives you the option to edit any parameters in the template to your liking before deployment and saves you from having to write a separate “azuredeploy.parameters.json” file.

However the deployment blade in the portal doesn’t really give you much in the way of error messages when the template fails. Often, you are give something along the lines of “error exists on line 1: position 10,374” or some such information. But your template is clearly more than one line in length and I have no intention of counting out thousands of characters to find the error location. The full text of the message may or may not be helpful depending on what point in the deployment the error occurred.

This is where deploying the template via PowerShell shines. As much as the red text is heartbreaking, it will give you the proper line number and position count to line up with your code. Also included is some often useable error text to point you on your way.

This single line of PowerShell will deploy a template that is hosted online (like in GitHub) and the -Verbose switch will report on the status of each of your resources as they start and complete deployment.

New-AzureRmResourceGroupDeployment -ResourceGroupName "<YourResourceGroupName>" -TemplateURI $templateFileURI -TemplateParameterURI $parameterFileURI -Verbose

If that doesn’t solve your problem or point you in the right direction, it’s time to dig deeper in the Azure Audit Logs from within the portal. They are chronological, so after a few clicks, you can drill down into the details of the most recent error. You can also access those logs via PowerShell:

Get-AzureRmLog -Status Failed -ResourceGroup "YourResourceGroupName" -DetailedOutput

You can also access these logs by digging down into the deployment in the portal. Just to “Audit Logs” and you’ll find all the activity in your subscriptions. Looked for the ones related to failed deployments or validations of deployments and as you drill into the details you’ll find what’s going wrong. Usually it’s an incorrect location of a parameters file, maybe a storage account name that’s already in use, or a missing resource group.

Happy Hunting!

If you’ve come from the Service Manager days of Azure, you’ll be familiar with the endpoints that came with each Cloud Service, handling the port forwarding needed to RDP to your server or to connect with PowerShell (port 5986). You could add or remove endpoints as need to provide access to whatever service you desired (after making the corresponding firewall change on the server itself.)

This is like parking your car in a garage.  You can choose to lock or unlock the doors (aka the OS firewall) and you can choose to provide access to the garage (aka Azure endpoints).

With ARM, when you deploy a server, you’ll be prompted to confirm the naming of your network security group, this network security group takes the place of the endpoints: providing some default rules, including the RDP access to port 3389.

So how does this differ from an endpoint?  One difference is that there is no port forwarding.  With a cloud service endpoint, the external port wasn’t 3389, because you could have multiple servers behind the single IP address assigned to the cloud service sharing the load balancer.   With ARM, the VM is assigned its own IP address without a load balancer, so the standard port number can be used. If you need to open an additional port, you simply add a new rule to the NSG and open the corresponding rule on the server firewall.

The second difference is that without a NSG (or the cloud service load balancer endpoints) to act as a gateway, your VM is basically just “hanging out there” with only its internal firewall to keep it safe.  So you really need to be sure you’ve checked that firewall configuration and you know what’s open to the public network.  By default, a Windows VM created with an Azure image has RDP and PowerShell/WinRM ports open on the operating system.

If you’ve been following along with my JSON template project, you’ll know that VMs can be created in ARM without the NSG.  And this fine if you are good with having your server parked out on the Internet with a few doors unlocked, so to speak.

But if you want to take advantage of the power of Network Security Groups to limit the sources of connections or to customize the way traffic flows between tiers of servers on your network, you’ll need to include the NSG resource in your template.

First you’ll  need to add a parameter to declare the Network Security Group.

  "networkSecurityGroupName": {
            "defaultValue": "NanoVMnsg",
            "type": "string"
    },

Then you’ll to configure the resource itself to include the security rules.

{
      "type": "Microsoft.Network/networkSecurityGroups",
      "name": "[parameters('networkSecurityGroupName')]",
      "apiVersion": "2015-06-15",
      "location": "[parameters('location')]",
      "properties": {
            "securityRules": [
                {
                    "name": "winrm",
                    "properties": {
                        "protocol": "Tcp",
                        "sourcePortRange": "*",
                        "destinationPortRange": "5985-5986",
                        "sourceAddressPrefix": "*",
                        "destinationAddressPrefix": "*",
                        "access": "Allow",
                        "priority": 110,
                        "direction": "Inbound"
                    }
                  },
                  {
                    "name": "rdp",
                    "properties": {
                        "protocol": "Tcp",
                        "sourcePortRange": "*",
                        "destinationPortRange": "3389",
                        "sourceAddressPrefix": "*",
                        "destinationAddressPrefix": "*",
                        "access": "Allow",
                        "priority": 120,
                        "direction": "Inbound"
                    }
                 }
                ]
            },
            "dependsOn": []
        },

Finally, you’ll need to adjust your network interface resource to depend on the NSG.

{
      "apiVersion": "2015-06-15",
      "type": "Microsoft.Network/networkInterfaces",
      "name": "[variables('nicName')]",
      "location": "[parameters('location')]",
      "dependsOn": [
        "[concat('Microsoft.Network/publicIPAddresses/', variables('publicIPAddressName'))]",
        "[concat('Microsoft.Network/virtualNetworks/', parameters('vnetName'))]",
        "[resourceId('Microsoft.Network/networkSecurityGroups', parameters('networkSecurityGroupName'))]"
      ],
      "properties": {
        "ipConfigurations": [
          {
            "name": "ipconfig1",
            "properties": {
             "privateIPAllocationMethod": "Dynamic",
             "publicIPAddress": {
                "id": "[resourceId('Microsoft.Network/publicIPAddresses',variables('publicIPAddressName'))]"
              },
              "subnet": {
                "id": "[variables('subnetRef')]"
              }
              }
          }
        ],
       "dnsSettings": {
         "dnsServers": []
       },
       "enableIPForwarding": false,
       "networkSecurityGroup": {
               "id": "[resourceId('Microsoft.Network/networkSecurityGroups', parameters('networkSecurityGroupName'))]"
            }
      }
    },

Go to my GitHub repository to find the full version of this template. By default, it will deploy a Nano Server VM.

Now here’s a little gotcha when it comes to NSGs and remote server management (as of this writing): If you created a VM from the Azure Portal and accept all the defaults (which include an NSG), that NSG doesn’t open the ports for WinRM by default.  It only opens RDP.  The OS firewall is open to accept WinRM and PowerShell, but the NSG blocks it.  You need to edit the NSG to include TCP ports 5985 (http) and/or 5986 (https) for remote management.

You might not notice this issue if you are connecting to full GUI servers with RDP, but if you are trying out Nano Server there is no RDP, so you must be able to connect with PowerShell.