Using Azure Automation to Shut Down a VM only if a SQL Agent Job is Not Running.

I have a client who uses MDS (Master Data Services) and SSIS (Integration Services) in an Azure VM. Since we only need to execute the SQL Agent job that runs the SSIS packages infrequently, we shut down the VM when it is not in use in order to save costs. We wanted to make sure that the Azure VM did not shut down when a specific SQL Agent job was still running, so I tackled this with some PowerShell runbooks in Azure Automation.

I split the job into two parts. The first runbook simply checks if a specified SQL Agent job is running and returns a text value that indicates whether it is running. A parent runbook checks if the VM is started. If the VM is started, it calls the child runbook to check if the job is running, and then shuts down the VM if the job is not running.

It’s fairly easy and convenient to have nested PowerShell runbooks in Azure Automation. There are two main ways to call a child runbook.

1. Inline execution
2. Using the Start-AzureRmAutomationRunbook cmdlet

It was less obvious to me how to call a child runbook when the parent runs in Azure and the child runs on a hybrid worker, especially when you need to use the output from the child runbook in the parent. A hyrid runbook worker allows us to access resources that are behind a VNET or on premises.

Travis Roberts has a nice video on just this topic that gave me the answers I needed.

Below is my parent runbook.

# Ensures you do not inherit an AzureRMContext in your runbook
Disable-AzureRmContextAutosave –Scope Process

$connection = Get-AutomationConnection -Name AzureRunAsConnection
Connect-AzureRmAccount -ServicePrincipal -Tenant $connection.TenantID `
-ApplicationID $connection.ApplicationID -CertificateThumbprint $connection.CertificateThumbprint

$rgName='MyResourceGroup'
$vmName='MyVM'
$SubID = 'XXXXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX'

$AzureContext = Select-AzureRmSubscription -SubscriptionId $SubID
'Check if VM is on'
$vm=((Get-AzureRmVM -ResourceGroupName $rgName -AzureRmContext $AzureContext -Name $vmName -Status).Statuses[1]).Code
 $vm 
 if ($vm -eq 'PowerState/running')
 {
    Do 
    {
        #if VM running call other runbook
        start-sleep -Seconds 60;
        'Check if job is running'
        $JobRunning = start-azureRMautomationrunbook -AutomationAccount 'ProgramsAutomation' -Name 'CheckRunningSQLJob' -ResourceGroupName $rgName -AzureRMContext $AzureContext -Runon 'Backups' -Wait;
        Write-Output $JobRunning
        
        
    } Until ($JobRunning -eq 'run0')
    
    'Stopping VM'
    stop-azurermvm -Name $VMname -ResourceGroupName $RgName -force
}

The runbook sets the Azure context to the appropriate subscription (especially important when you are a guest user in someone else’s tenant). Then it checks if the VM is started. If it is, it goes into a do-while loop. This task isn’t super time sensitive (it’s just to save money when the VM isn’t in use), so it’s waiting 60 seconds and then calling the child runbook to find out if my SQL Agent job is running. This makes sure that the child runbook is called at least once. If the result is that the job is not running, it stops the VM. If the job is running, the loop starts over, waiting 60 seconds before checking again. This loop is essentially polling the job status until it sees that the job is completed. One thing to note is the -Wait parameter on the end of that Start-AzureRmAutomationRunbook command. If you don’t specify the -Wait parameter, the command will immediately return a job object. If you specify the -Wait parameter, it waits for that child job to complete and returns the results of that job.

And here is my child runbook.

[OutputType([string])]

$SQLJobName = 'MySQLAgentJobName'
$SQLInstanceName = 'MySQLServer

$cred=Get-AutomationPSCredential -Name 'mycredential'
 
$server = Connect-DbaInstance -SqlInstance $SQLInstanceName -SqlCredential $cred
 
Get-DbaRunningJob -SqlInstance $server | Get-DbaRunningJob

$JobStatus = (Get-DbaRunningJob -SqlInstance $server).Name -match $SQLJobName

If ($JobStatus -ne $false) 
{
#job is running. Passing back a string because bits and ints were causing issues.
    $JobRunning = 'run1'
    Write-Output $JobRunning 
}
else 
{ 
#job is idle
    $JobRunning = 'run0'
    Write-Output $JobRunning 
}

I’m using dbatools to check if the job is running on the server. That is the Get-DBARunning Job command. The important part to note is that you have to use the Write-Output command for this output to be available to the parent runbook. I got some weird results when I tried to return an int or a boolean (it was returning an object rather than a single value), so I just went with a string. The string, while not the most efficient, works just fine. If you understand why this is, feel free to leave me a comment.

These runbooks have been in place for a couple of months now, and they are working great to shut down the VM to save money while making sure not to disturb an important SQL Agent job that might occasionally run late. I didn’t find much documentation nor many examples of using output from a child job that runs on a hybrid worker, so I wanted to get this published to help others that go searching.