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.

Lately, I have been using SSIS execution frameworks and Biml created by other people to populate data marts and data warehouses. It has taught me a few things and helped me clarify what I like and dislike compared to my usual framework. I’ve got the beginning of my preferences list started below. There are probably situations where I would want to deviate from my preferences, but I think they make a good starting point.

Populating Data

• For self-service BI environments, a date dimension that doesn’t go out much further than the greatest date in your data. This can be a view or stored procedure that limits and updates dates rather than a static date dimension that goes out until the end of time.
• Unknown values are included in normal dimension loads, not in separate scripts that must be run on deployment. This way, if an unknown value is ever left out or deleted, it will be added in the next data load rather than requiring a special execution of a script.
• Every table should have InsertDateTime and UpdateDateTime columns. The UpdateDateTime column should be populated with the same value as the InsertDateTime column upon creation of the row, rather than being left null.
• Whatever you use to create tables, include primary keys, foreign keys, and indexes with your table definitions. Provide explicit constraint names to simplify database comparisons. You can disable your foreign keys, but they need to be there to provide that metadata.
• Separate your final dimensional/reporting tables from audit tables and staging tables. This can be done with separate schemas or even separate databases.

Data Integration Process

• There should be consistent error handling in each layer (staging, dims, facts, etc.). If you write errors to another location (flat file, database table), have a process that notifies the right people that errors occurred. The process of consuming corrected data must be built, tested, and integrated into the existing process.
• Make your error handling process reflect what end users need to see when an error occurs. Does it make sense to have a partial load when there is an issue? Or should it be all or nothing?
• Have smart master packages that determine which packages to run. Don’t check whether the package should run inside of the package itself – do that in the master package.
• Master packages should execute child packages in parallel as much as possible rather than defaulting to sequential execution.
• Have an audit log with one row per package. Include the SSIS ServerExecutionID in the audit log – not the package -specific ID but the execution ID for the entire run. If there are incremental loads, the where clause used to filter the load should be captured in the audit table. Include row counts as well as package start and stop time in your audit log.
• Add an AuditLogID column on your dimension, fact, and staging tables so you can trace each row back to the process that populated it.
• For dims and facts, perform change detection/deduplication of records, usually through hash values and either SSIS lookups or SQL queries with WHERE NOT EXISTS.
• Avoid T-SQL MERGE statements. Write individual insert/update/delete statements. This avoid any bugs in MERGE and makes your SQL easier to understand and troubleshoot.
• Use consistent naming of tasks, source, destinations, packages, connection managers, etc. Connection managers pointing to databases should have names that refer to the database rather than the server.
• If you are downloading files, move the files to an archive folder once files are processed. You can have rules in place if you have retention limits. But you probably need to keep files from at least the last load for audit and troubleshooting purposes. This could change if you are importing very sensitive data.
• Even if you need to copy all columns from a table, write a select statement for database sources that explicitly names fields rather than using SELECT *. or just selecting the table or view.
• SSIS lookups should use an explicit query rather than referencing an entire table.
• Implement restartability at the package level for most packages (you should have single-purpose packages executed by a master package). Checkpoints are ineffective within a package. If you build your audit log table correctly, you can get the list of packages that have not run in the last X minutes/hours and feed that to your master package.
• Send email from your scheduling tool rather than within an SSIS package.
• Track data lineage in your tables. This can be as simple as having a table that lists all of your data sources with an ID column and including that ID value in each row of your staging, fact, and dimension tables.
• Dims and facts are not truncated. Data should be inserted and updated (and deleted, if necessary).
• Connection strings used in multiple packages should be project-level connection strings.

Biml Specifics

• Understand whether you need a flexible Biml Framework or just an accelerator for a current project. If you need flexibility, don’t hardcode connection strings and other things that change when you add/change sources and destinations. If you just need to accelerate development of a simple data mart, total flexibility may be overkill and actually cause more work.
• Have a single place where you add synthetic metadata, as much as possible. BimlScript gets messy and difficult to understand when you have some extended properties that are read in, some annotations added directly, and some variables defined in your code. This is why I like synthetic metadata stored in a database. Also, extended properties don’t exist in Azure SQL Data Warehouse, so if you need your framework to work there you can’t go that route.
• Don’t repeat your code in multiple files. If you have some logic that gets reused, move it to a separate file and reference it from other files.

What Do You Think?

What’s on your SSIS preferences list? Do you disagree with one of my preferences and want to share your knowledge? Let’s chat in the comments.

I wrote about the violin plot custom visual by Daniel Marsh-Patrick back in February. I thought it was a good visual then, but version 1.3 has recently been released with some nice enhancements.

First, the violin plot is now a certified custom visual. This means that it has been tested by the Power BI team to ensure it meets certain requirements, one of which is that the visual does not access external services or resources. You can be confident your data isn’t being sent externally when you use the violin plot.

As for the functional enhancements, a new legend has been added. This is a great addition to make the chart clearer and more easily read, especially for audiences that may not be familiar with how the violin plot works. The customizable legend calls out what markers are used for mean, median, and quartiles.

Another good enhancement is the new column option for the combo plot. It allows you to have your plot show as a range column chart where the bar spans from the minimum value to the maximum value for each category. I chose to show only the mean and median in the example below, but you can also add quartiles.

The barcode plot also has a nice enhancement in the tooltip. Now when you hover over a bar, you can see the number of samples with the highlighted value.

You can check out Daniel’s blog post to see the full list of enhancements for this release. Tweet me if you make something cool and shareable with the violin plot in Power BI.

Every once in a while, I come across a data warehouse where the data load uses a full truncate and reload pattern to populate a fact or dimension. While it may not be the end of the world for a small table, it does concern me and I usually recommend to redesign the load. My thoughts below on why this is an anti-pattern are true for using the actual TRUNCATE TABLE statement as well as executing a DELETE statement with no WHERE clause.

Surrogate Keys

Dimensional models use surrogate keys rather than natural keys. Surrogate keys are system-generated, meaningless values that are usually integers used to uniquely identify a record. They provide good performance for joins in queries, allow us to switch or use multiple source systems to feed the same tables, and facilitate the use of slowly changing dimensions. If you truncate a dimension table and then repopulate it, you will end up with different surrogate keys assigned to your dimension values. Let’s say we have a Geography dimension that looks like the table below.

If I truncate and reload this table, there is no guarantee that Denver will have a GeographyKey value of 1. I might reload the data and then have Paris as 1, Denver as 2, and Dallas as 3. And since we use surrogate keys as dimension lookups in fact tables and bridge tables (and snowflaked dimensions, if you use those), I would now need to update every table that references this dimension. That’s a lot of table updates without a good reason.

Error Handling

Another reason to avoid truncating the tables in your dimensional model is error handling. When you design your data load, you need to think about what should happen when it fails and where it is most likely to fail. Failures will often occur either in data transformation steps or upon inserting/updating values in the destination table due to data type/size conflicts.

Let’s say you truncate your fact table and then you encounter an ETL failure while performing the transformations to reload the table. Now you have an empty fact table. If you have error handling logic in your ETL, you may be able to redirect the error rows to another location to be handled later. But that still leaves you with an incomplete table at best. If instead of truncating and reloading, you were doing inserts and updates when you encountered an error, your table might reflect slightly stale data as of the previous load. You still might have a situation where your table is partially updated with the rows that were inserted before the error was encountered depending on how your ETL design. But having a fact table with old or partially updated data is usually (but not always) more preferable than having no data when a data load process fails.

Performance and Data Availability

When you truncate and reload a table, you are assuming you have access to all of the source data needed to fully repopulate the table. This may be true today, but what about next year when you switch CRM systems? Or what if your organization makes the decision to archive data in the source system that is older than 5 years? If any data needed for your fact or dimension becomes unavailable, your truncate and reload pattern will fail to serve your needs.

Even if you are sure your data will be available, you may want to consider your data load times. If you only have a handful of tables with a few hundred thousand rows max, you may load your data warehouse in a few minutes. But what happens when you have a few hundred million rows with some complicated transformation logic in your pipelines? You might be adding minutes to hours to your load times.

Why Do ETL Developers End Up Truncating Fact Tables?

Sometimes developers just don’t know better. But often there were understandable reasons for using the truncate and reload pattern. While I have never built a system where I truncated a dimension, I have had a couple were we at least started out with truncating the fact table. We usually built it to make data initially available with correct values, and then worked with stakeholders to find a different way to access the data or to have them generate the data differently to alleviate the problems that caused us to want to truncate and reload.

Often the cause is data quality issues. One one project I worked, we had dimension values that defined the granularity of the fact table that would change in the source systems, and the process to try to update them in the fact table was too complicated and took too long. So we made the decision to truncate and reload, understanding the risks of doing so. Having the data available and usable for analysis gave us more information on how useful the data was to users and helped us work to understand why and how our keys were changing. Over time, we were able to influence the way data was entered into source systems so that we didn’t have to go through the truncate and reload process and could perform incremental loads on our fact table.

Sometimes source systems allow hard deletion of data and don’t maintain a list of what was deleted. And depending on access methods and the size of the data, it can be difficult to compare the rows in a fact table with the source data to identify the deleted rows. I would argue this is a poor design of the source system, but we often can’t change that, and we still need to include the data in our data warehouse. So it’s understandable why someone in this situation would want to truncate and reload a fact table.

If your stakeholders are ok with the potential of long load times, empty tables when errors occur, and the assumption that source data will always be available, then there isn’t much problem with truncating and reloading a fact table. But that is often not the case. And that decision should be made explicitly rather than implicitly.

What To Do Instead of Truncating and Reloading a Fact Table

If you’ve been truncating and reloading a fact table because that just seemed like the simplest thing to do, you can change your load pattern.

My general load pattern is:

1. Truncate the update table
2. Insert new rows into the fact table and changed rows into the update table
3. Perform set-based updates on the fact table based upon the data in the update table

I’m able to determine what rows are new, changed, and unchanged by using hash values. I concatenate and hash the values that define the level of uniqueness of the row into one value I call HistoricalHash and the values from the remaining columns into a value I call ChangeHash. In other words, I load a transactional or periodic snapshot fact table in a manner similar to a Type 1 slowly changing dimension.

If you have data quality, data deletion, or other issues that prevent you from using a change detection pattern like the above, consider using a staging table and swapping it out with the fact table. Create a staging table that looks exactly like your fact table.

1. Truncate the staging table
2. Populate the staging table with all rows that should be in the fact table
3. Swap the staging table and the fact table (this usually involves renaming the tables)

Understand the Consequences of Your Design

If you are choosing to truncate a dimension or fact table, be sure that you understand the trade-offs. You may be unwittingly opting for simplicity of ETL over data availability and efficient data loads.