PowerShell module for Databricks on Azure and AWS : Avaiilable via PowerShell Gallery: DatabricksPS Over the last year I worked a lot with Databricks on Azure and I have to say that I was (and still am) very impressed how well it works and how it integrates with other services of the Microsoft Azure Data Platform like Data Lake Store, Data Factory, etc. Some of the projects I worked on also included CI/CD like pipelines using Azure DevOps where Databricks did not really shine so bright in the beginning. There are no native tasks for it or anything. But this is OK as for those scenarios, where you need Continue reading PowerShell module for Databricks on Azure and AWS→
Storing Images in a PowerBI/Analysis Services Data Models : As some of you probably remember, when PowerPivot was still only available in Excel and Power Query did not yet exist, it was possible to load images from a database (binary column) directly into the data model and display them in PowerView. Unfortunately, this feature did not work anymore in PowerBI Desktop and the only way to display images in a visual was to provide the URL of the image which is public accessible. The visual would then grab the image on-the-fly from the URL and render it. This of course has various drawbacks: The image needs to be available Continue reading Storing Images in a PowerBI/Analysis Services Data Models→
Deploying an Azure Data Factory project as ARM Template : In my last post I wrote about how to Debug Custom .Net Activities in Azure Data Factory locally. This fixes one of the biggest issues in Azure Data Factory at the moment for developers. The next bigger problem that you will run into is when it comes to deploying your Azure Data Factory project. At the moment, you can only do it manually from Visual Studio which, for bigger projects, can take quite some time. So I extended and advanced the code from my CustomActivityDebugger. Well, actually I rewrote some major parts of it and moved it into a new Continue reading Deploying an Azure Data Factory project as ARM Template→
Error-handling in Power Query : Data is the daily bread-and-butter for any analyst. In order to provide good results you also need good data. Sometimes this data is very well prepared beforehand and you can use it as it is but it is also very common that you need to prepare and transform the data on your own. To do this Microsoft has introduced Power Query (on tool of the Power BI suite). Power Query can be used to extract, transform and load data into Excel and/or Power Pivot directly. When using any data you usually know what the data looks like and what to Continue reading Error-handling in Power Query→

Reading Delta Lake Tables natively in PowerBI

Posted on by — 39 Comments ↓

I also contributed the connector described in this post to the official delta.io Connectors page and repo (link). You will find the most recent updates in my personal repo which are then merged to the official repo once it has been tested thoroughly!

Working with analytical data platforms and big data on a daily basis, I was quite happy when Microsoft finally announced a connector for Parquet files back in November 2020. The Parquet file format is developed by the Apache foundation as an open-source project and has become a fundamental part of most data lake systems nowadays.

“Apache Parquet is a columnar storage format available to any project in the Hadoop ecosystem, regardless of the choice of data processing framework, data model or programming language.”

However, Parquet is just a file format and does not really support you when it comes to data management. Common data manipulation operations (DML)  like updates and deletes still need to be handled manually by the data pipeline. This was one of the reasons why Delta Lake (delta.io) was developed besides a lot of other features like ACID transactions, proper meta data handling and a lot more. If you are interested in the details, please follow the link above.

So what is a Delta Lake table and how is it related to Parquet? Basically a Delta Lake table is a folder in your Data Lake (or wherever you store your data) and consists of two parts:

  1. Delta log files (in the sub-folder _delta_log)
  2. Data files (Parquet files in the root folder or sub-folders if partitioning is used)

The Delta log persists all transactions that modified the data or meta data in the table. For example, if you execute an INSERT statement, a new transaction is created in the Delta log and a new file is added to the data files which is referenced by the Delta log. If a DELETE statement is executed, a particular set of data files is (logically) removed from the Delta log but the data file still resides in the folder for a certain time. So we cannot just simply read all Parquet files in the root folder but need to process the Delta log first so we know which Parquet files are valid for the latest state of the table.

These logs are usually stored as JSON files (actually JSONL files to be more precise). After 10 transactions, a so-called checkpoint-file is created which is in Parquet format and stores all transactions up to that point in time. The relevant logs for the final table are then the combination of the last checkpoint-file and the JSON files that were created afterwards. If you are interested in all the details on how the Delta Log works, here is the full Delta Log protocol specification.

From those logs we get the information which Parquet files in the main folder must be processed to obtain the final table. The content of those Parquet files can then simply be combined and loaded into PowerBI.

I encapsulated all this logic into a custom Power Query function which takes the folder listing of the Delta table folder as input and returns the content of the Delta table. The folder listing can either come from an Azure Data Lake Store, a local folder, or an Azure Blob Storage. The mandatory fields/columns are [Content], [Name] and [Folder Path]. There is also an optional parameter which allows you the specify further options for reading the Delta table like the Version  if you want to use time-travel. However, this is still experimental and if you want to get the latest state of the table, you can simply omit it.

The most current M-code for the function can be found in my Github repository for PowerBI: fn_ReadDeltaTable.pq and will also be constantly updated there if I find any improvement.
The repository also contains an PowerBI desktop file (.pbix) where you can see the single steps that make up for the final function.

Once you have added the function to your PowerBI / Power Query environment you can call it like this:

I would further recommend to nest your queries and separate the access to the storage (e.g. Azure Data Lake Store) and the reading of the table (execution of the function). If you are reading for an ADLS, it is mandatory to also specify [HierarchicalNavigation = false] !
If you are reading from a blob storage, the standard folder listing is slightly different and needs to be changed.

Right now the connector/function is still experimental and performance is not yet optimal. But I hope to get this fixed in the near future to have a native way to read and finally visualize Delta lake tables in PowerBI.

After some thorough testing the connector/function finally reached a state where it can be used without any major blocking issues, however there are still some known limitations:

  • Partitioned tables
    • currently columns used for partitioning will always have the value NULL FIXED!
    • values for partitioning columns are not stored as part of the parquet file but need to be derived from the folder path FIXED!
  • Performance
    • is currently not great but this is mainly related to the Parquet connector as it seems
    • very much depends on your data – please test on your own!
  • Time Travel
    • currently only supports “VERSION AS OF”
    • need to add “TIMESTAMP AS OF”
  • Predicate Pushdown / Partition Elimination
    • currently not supported – it always reads the whole table FIXED!

Any feedback is welcome!

Special thanks also goes to Imke Feldmann (@TheBIccountant, blog) and Chris Webb (@cwebb_bi, blog) who helped me writing and tuning the PQ function!

Downloads: fn_ReadDeltaTable.pq (M-code)

PowerBI & Big Data – Using pre-calculated Aggregations of Semi- and Non-Additive Measures

Posted on by — 2 Comments ↓

Calculating and visualizing semi- and non-additive measures like distinct count in Power BI is usually not a big deal. However, things can become challenging if your data volume grows and exceeds the limits of Power BI!

In one of my recent projects we wanted to visualize data from the customers analytical platform based on Azure Databricks in Power BI. The connection between those two tools works pretty flawless which I also described in my previous post but the challenge was the use-case and the calculations. We wanted to display the distinct customers across various aggregations levels over a billion rows fact table. We came up with different potential solutions all having their pros and cons:

  1. load all data into Power BI (import mode) and do the aggregations there
  2. use Power BI with direct query and let the back-end do the heavy lifting
  3. load only necessary pre-aggregated data into Power BI (import mode)

Please keep in mind that we are dealing with a distinct count measure here. Semi- and Non-additive measure like this cannot easily be aggregated from lower levels to higher levels without having all the detail data available!

Option 1. has the obvious drawback that data model would be huge in size as we were dealing with billions of transactions. This would have exceeded our current size limits for Power BI data models.

Option 2. would usually work fine, but again, for the amount of data we were dealing with the back-end was just no able to provide sub-second latency that was required.

So we went for Option 3. and did the various aggregations on the different levels in Azure Databricks and loaded only the final results to Power BI. First we wanted to use Power BI Aggregations and Composite Models. Unfortunately, this did not work out for us as we were not in control which aggregation table (we had multiple for the different aggregation levels) was used by the engine which potentially resulted in wrong results when additional aggregation was done in Power BI. Also, when slicing for random aggregation levels, Power BI was querying the details in direct query mode causing very poor query performance.

After some further thinking we came up with a new solution which was also based on pre-calculated aggregations but not realized using built-in aggregation tables but having a combined table for all aggregations and some very straight-forward DAX to select the row we wanted! In the end the whole solution consisted of one SQL view using COUNT(DISTINCT xxx) aggregation and GROUP BY GROUPING SETS (T-SQL, Databricks, … supported in all major SQL engines) and a very simple DAX measure!

Here is a little example that illustrates the approach. Assume you want to calculate the distinct customers that bought certain products in a subcategory/category by year. The first step is to create a view that provides this information:

Please note that when we have a natural relationship between hierarchy levels (= only 1:n relationships) we need to specify the current level and also all upper levels to allow a proper drill-down later on! E.g. ProductCategory (1 -> n) ProductSubcategory

This calculates all the different aggregation levels we need. Columns with NULL mean they were not filtered/grouped by when calculating the aggregation.
Rows 80-84 contain the aggregations grouped by Year only whereas rows 77-79 contain only aggregates by ProductCategoryKey. The rows 75-76 were aggregated by Year AND ProductCategoryKey.
Depending on your final report layout, you may not need all of them and you should consider removing those that are not needed!

This table is then loaded into Power BI. You can either use a custom SQL query like above in Power BI directly or create a view in the back-end system which would be my preferred solution. Alternatively you can also create all these grouping sets using Power Query/M. The incredible Imke Feldmann (t, b) came up with a solution that allows you to specify the grouping sets in a similar way as in SQL and do all this magic within Power BI directly! I hope she will blog about it pretty soon!
(The sample workbook at the end of this post also contains a little preview of this M-magic.)

Now that we have all the data we need in Power BI, we need to display the right values for the selections in the report which of course can be dynamic. That’s a bit tricky but once you understand the concept, it is pretty straight forward. First of all, the table containing the aggregations must not be related to any other table as we build them on the fly within our DAX measure. The table itself can also be hidden.

And this is the final DAX for our measure:

The first part is to get all the selected values of the lookup/dimension tables the user selects on the report. These are all the _sel_XXX variables. SELECTEDVALUE() returns the selected value if only one item is in the current filter context and BLANK()/NULL otherwise. We then use TREATAS() to apply those filters (either a single item or NULL) to our aggregations table. This should usually only return a table with a single row so we can use MAXX() to get our actual value from that one row. I also added a check in case multiple rows are returned which can potentially happen if you use multi-selects in your filters and instead of showing wrong values I’d rather indicate that there is something wrong with the calculation.

The measure can then be sliced and diced by our pre-defined aggregation levels as if it would be a regular measure but instead of having to process those expensive calculations on the fly we use the pre-calculated aggregates!

One thing to be aware of is that it will produce wrong results if multiple items for any of the aggregation levels are selected so it is highly recommended to set all slicers/filters to single select only or ensure that the filtered aggregation levels are also used in the chart. In this case only the grand total will show wrong values or NULL then.
This could also be fixed in the DAX measure by checking how many rows are actually selected for each level and throw an error in case it is used in a filter and the count of values is > 1.

I did some further thinking and this approach could probably also be used to mimic custom roll-ups and unary operators we know from Analysis Services Multidimensional cubes. If I find some proper examples and this turns out to be feasibly I will write another blog post about it!

Download: Custom_Aggregations_NonAdditive_Measure.pbix

Connecting Power BI to Azure Databricks

Posted on by — 23 Comments ↓

I work a lot with Azure Databricks and a topic that always comes up is reporting on top of the data that is processed with Databricks. Even though notebooks offer some great ways to visualize data for analysts and power users, it is usually not the kind of report the top-management would expect. For those scenarios, you still need to use a proper reporting tool, which usually is Power BI when you are already using Azure and other Microsoft tools.

So, I am very happy that there is finally an official connector in PowerBI to access data from Azure Databricks! Previously you had to use the generic Spark connector (docs) which was rather difficult to configure and did only support authentication using a Databricks Personal Access Token.

With the new connector you can simply click on “Get Data” and then either search for “Azure Databricks” or go the “Azure” and scroll down until you see the new connector:

The next dialog that pops up will ask you for the hostname and HTTP path – this is very similar to the Spark connector. You find all the necessary information via the Databricks Web UI. As this connection is always bound to an existing cluster you need to go the clusters details page and check the Advanced Tab “JDBC/ODBC” as described here:
(NOTE: you can simply copy the Server Hostname and the HTTP Path from the cluster page)

The last part is then the authentication. As mentioned earlier the new connector now also supports Azure Active Directory authentication which allows you to use the same user that you use to connect to the Databricks Web UI!
Personal Access Tokens are also still supported and there is also Basic authentication using username/password.

Once you are connected, you can choose the tables that you want to import/connect and start building your report!

Here is also a quick overview which features are supported by the Spark and the Azure Databricks connector as there are some minor but important differences:

Feature ComparisonSpark ConnectorDatabricks Connector
Power BI DesktopYESYES
Power BI ServiceYESYES *
Direct Query (Desktop)YESYES
Direct Query (Service)YESYES *
Import ModeYESYES
Manual Refresh (Service)YESYES *
Scheduled Refresh (Service)YESYES *
Azure Active Directory (AAD) AuthenticationNOYES
Personal Access Token AuthenticationYESYES
Username/Password AuthenticationYESYES
General AvailableYESYES
Performacne Improvements with Spark 3.xNO *YES *
Supports On-Premises data gatewayYESNO
Features supported by Spark and Databricks Connector for PowerBI

*) Updated 2020-10-06: the new Databricks Connector for PowerBI now supports all features also in the PowerBI service!

Update 2020-10-06: So from the current point of view the new Databricks Connector is a superset of old Spark Connector with additional options for authentication and better performance with the latest Spark versions. So it is highly recommended to use the new Databricks Connector unless you have very specific reasons to use the Spark connector! Actually the only reason why I would still use the Spark connector is the support for the On-Premises data gateway in case your Spark or Databricks cluster is hosted in a private VNet.

So currently the generic Spark connector still looks superior simply for the support in the Power BI Service. However, I am quite sure that it will be fully supported also by the Power BI Service in the near future. I will update this post accordingly!
On the other hand, Azure Active Directory authentication is a huge plus for the native Azure Databricks connector as you do not have to mess around with Databricks Personal Access Tokens (PAT) anymore!

Another thing that I have not yet tested but would be very interesting is whether Pass-Through security works with this new connector. So you log in with your AAD credentials in Power BI, they get passed on to Databricks and from there to the Data Lake Store. For Databricks Table Access Control I assume this will just work as it does for PAT as it is not related to AAD authentication.

Azure Data Factory, dynamic JSON and Key Vault references

Posted on by — 11 Comments ↓

Paul Andrews (b, t) recently blogged about HOW TO USE ‘SPECIFY DYNAMIC CONTENTS IN JSON FORMAT’ IN AZURE DATA FACTORY LINKED SERVICES. He shows how you can modify the JSON of a given Azure Data Factory linked service and inject parameters into settings which do not support dynamic content in the GUI. What he shows with Linked Services and parameters also applies to Key Vault references – sometimes the GUI allows you to reference a value from the Key Vault instead of hard-coding it but for other settings the GUI only offers a simple text box:

As You can see, the setting “AccessToken” can use a Key Vault reference whereas settings like “Databricks Workspace URL” and “Cluster” do not support them. This is usually fine because the guys at Microsoft also thought about this and support Key Vault references for the settings that are actually security relevant or sensitive. Also, providing the option to use Key Vault references everywhere would flood the GUI. So this is just fine.

But there can be good reasons where you want to get values from the Key Vault also for non-sensitive settings, especially when it comes to CI/CD and multiple environments. From my experience, when you implement a bigger ADF project, you will probably have a Key Vault for your sensitive settings and all other values are provided during the deployment via ARM parameters.

So you will end up with a mix of Key Vault references and ARM template parameters which very likely will be derived from the Key Vault at some point anyway. To solve this, you can modify the JSON of an ADF linked service directly and inject KeyVault references into almost every property!
Lets have a look at the JSON of the Databricks linked service from above:

As you can see in lines 8-15, the property “accessToken” references the secret “Databricks-Accesstoken” from the Key Vault linked service “KV_001” and the actual value is populated at runtime.

After reading all this, you can probably guess what we are going to do next –
We also replace the other properties by Key Vault references:

You now have a linked service that is configured solely by the Key Vault. If you think one step further, you can replace all values which are usually sourced by ARM parameters with Key Vault references instead and you will end up with an ARM template that only has two parameters – the name of the Data Factory and the URI of the Key Vault linked service! (you may even be able to derive the Key Vaults URI from the Data Factory name if the names are aligned!)

The only drawback I could find so far was that you cannot use the GUI anymore but need to work with the JSON from now on – or at least until you remove the Key Vault references again so that the GUI can display the JSON properly again. But this is just a minor thing as linked services usually do not change very often.

I also tried using the same approach to inject Key Vault references into Pipelines and Dataset but unfortunately this did not work 🙁
This is probably because Pipelines and Datasets are evaluated at a different stage and hence cannot dynamically reference the Key Vault.

DatabricksPS and Azure AD Authentication

Posted on by — 1 Comment ↓

Avaiilable via PowerShell Gallery: DatabricksPS

Databricks recently announced that it is now also supporting Azure Active Directory Authentication for the REST API which is now in public preview. This may not sound super exciting but is actually a very important feature when it comes to Continuous Integration/Continuous Delivery pipelines in Azure DevOps or any other CI/CD tool. Previously, whenever you wanted to deploy content to a new Databricks workspace, you first needed to manually create a user-bound API access token. As you can imagine, manual steps are also bad for otherwise automated processes like a CI/CD pipeline. With Databricks REST API finally supporting Azure Active Directory Authentication of regular users and service principals, this last manual step is finally also gone!

As I had this issue at many of my customers where we had already fully automated the deployment of our data platform based on Azure and Databricks, I also wanted to use this new feature there. The deployment of regular Databricks objects (clusters, notebooks, jobs, …) was already implemented in the CI/CD pipeline using my PowerShell module DatabricksPS and of course I did not want to rewrite any of those steps. So, I simply extend the module’s authentication methods to also support Azure Active Directory Authentication. The only thing that actually changed was the call to Set-DatabricksEnvironment which now supports additional parameter sets and parameters:

The first thing you will realize is that it is now necessary to specify the Databricks Workspace explicitly either using SubscriptionID/ResourceGroupName/WorkspaceName to uniquely identify the Databricks workspace within Azure or using the OrganizationID that you see displayed in the URL of your Databricks Workspace. For the actual authentication the parameters -ClientID, -TenantID, -Credential and the switch -ServicePrincipal are used.

Regardless of whether you use regular username/password authentication with an AAD user or an AAD service principal, the first thing you need to do in both cases is to create an AAD Application as described in the official docs from Databricks:
Using Azure Active Directory Authentication Library
Using a service principal

Once you have ensured all prerequisites exist, you can use the samples below to authenticate with your AAD username/password with DatabricksPS:

Here is another sample using a regular service principal authentication and the OrganizationID with DatabricksPS:

As you can see, once the environment is set up using the new authentication methods, the rest of the script stays the same and there is not much more you need to do fully automate your CI/CD pipeline with DatabricksPS!

I have not yet fully tested all cmdlets of the module so if you experience any issues, please contact me or open a ticket in the GIT repository.

Professional Development for Databricks with Visual Studio Code

Posted on by — 36 Comments ↓

When working with Databricks you will usually start developing your code in the notebook-style UI that comes natively with Databricks. This is perfectly fine for most of the use cases but sometimes it is just not enough. Especially nowadays, where a lot of data engineers and scientists have a strong background also in regular software development and expect the same features that they are used to from their original Integrated Development Environments (IDE) also in Databricks.

For those users Databricks has developed Databricks Connect (Azure docs) which allows you to work with your local IDE of choice (Jupyter, PyCharm, RStudio, IntelliJ, Eclipse or Visual Studio Code) but execute the code on a Databricks cluster. This is awesome and provides a lot of advantages compared to the standard notebook UI. The two most important ones are probably the proper integration into source control / git and the ability to extend your IDE with tools like automatic formatters, linters, custom syntax highlighting, …

While Databricks Connect solves the problem of local execution and debugging, there was still a gap when it came to pushing your local changes back to Databricks to be executed as part of a regular ETL or ML pipeline. So far you had to either “deploy” your changes by manually uploading them via the Databricks UI again or write a script that uploads it via the REST API (Azure docs).

NOTE: I also published a PowerShell module that eases the automation/scripting of these tasks also as part of CI/CD pipeline. It is available from the PowerShell gallery DatabricksPS and integrates very well with this VSCode extension too!

However, this is not really something you would call a “seamless experience” so I also started working on an extension for Visual Studio Code to work more efficiently with Databricks. It has been in the VS Code gallery (Databricks VSCode) for about a month now and I received mostly positive feedback so far. Now I am at a stage where I want to get more people to use it – hence this blog post to announce it officially. The extension is currently published under GPLv3 license and is free to use for everyone. The GIT repository is also linked in the VS Code gallery if you want to participate or have any issues with the extension.

It currently supports the following features:

  • Workspace browser
    • Up-/download of notebooks and whole folders
    • Compare/Diff of local vs online notebook (currently only supported for raw files but not for notebooks)
    • Execution of local code and notebooks against a Databricks Cluster (via Databricks-Connect)
  • Cluster manager
    • Start/stop clusters
    • Script cluster definition as JSON
  • Job browser
    • Start/stop jobs
    • View job-run history + status
    • Script job definition as JSON
    • Script job-run output as JSON
  • DBFS browser
    • Upload files
    • Download files
    • (also works with mount points!)
  • Secrets browser
    • Create/delete secret scopes
    • Create/delete secrets
  • Support for multiple Databricks workspaces (e.g. DEV/TEST/PROD)
  • Easy configuration via standard VS Code settings

More features to come in the future but these will be mainly based on the requests that come from users or my personal needs. So your feedback is highly appreciated – either directly here or using the feedback section in the GIT repository.

I will also write some follow up post to show you how to work in the most efficient way using this new VSCode extension in combination with your Databricks workspace so stay tuned!

VS Code gallery: paiqo.Databricks-VSCode
Github repository: Databricks-VSCode

How-To: Migrating Databricks workspaces

Posted on by — 56 Comments ↓

Foreword:
The approach described in this blog post only uses the Databricks REST API and therefore should work with both, Azure Databricks and also Databricks on AWS!

It recently had to migrate an existing Databricks workspace to a new Azure subscription causing as little interruption as possible and not loosing any valuable content. So I thought a simple Move of the Azure resource would be the easiest thing to do in this case. Unfortunately it turns out that moving an Azure Databricks Service (=workspace) is not supported:

Resource move is not supported for resource types ‘Microsoft.Databricks/workspaces’. (Code: ResourceMoveNotSupported)

I do not know what is/was the problem here but I did not have time to investigate but instead needed to come up with a proper solution in time. So I had a look what needs to be done for a manual export. Basically there are 5 types of content within a Databricks workspace:

  • Workspace items (notebooks and folders)
  • Clusters
  • Jobs
  • Secrets
  • Security (users and groups)

For all of them an appropriate REST API is provided by Databricks to manage and also exports and imports. This was fantastic news for me as I knew I could use my existing PowerShell module DatabricksPS to do all the stuff without having to re-invent the wheel again.
So I basically extended the module and added new Import and Export functions which automatically process all the different content types:

  • Export-DatabricksEnvironment
  • Import-DatabricksEnvironment

They can be further parameterized to only import/export certain artifacts and how to deal with updates to already existing items. The actual output of the export looks like this and of course you can also modify it manually to your needs – all files are in JSON except for the notebooks which are exported as .DBC file by default:

A very simple sample code doing and export and an import into a different environment could look like this:

Having those scripts made the whole migration a very easy task.
In addition, these new cmdlets can also be used in your Continuous Integration/Continuous Delivery (CI/CD) pipelines in Azure DevOps or any other CI/CD tool!

So just download the latest version from the PowerShell gallery and give it a try!

Data Virtualization in Microsoft Power BI – Part 2

Posted on by — 3 Comments ↓

In my previous post I showed how you can use Microsoft Power BI to create a Data Virtualization layer on top of multiple relational data sources querying them all at the same time through one common model. As I already mentioned in the post and what was also pointed out by Adam Saxton (b, t) in the comments is the fact, that this approach can cause serious performance problems at the data source and also on the Power BI side. So in this post we will have a closer look on what actually happens in the background and which queries are executed when you join different data sources on-the-fly.

We will use the same model as in the previous post (you can download it from there or at the end of this post) and run some basic queries against it so we can get a better understanding of the internals.
Here is our relationship diagram again as a reference. Please remember that each table comes from a different SQL server:

Relationships

In our test we will simply count the number of products for each Product Subcategory:

NumberOfProducts_by_SubCategory

Even though this query only touches two different data sources, it is a good way to analyze the queries sent to the data sources. To track these queries I used the built-in Performance Analyzer of Power BI desktop which can be enabled on the “View”-tab. It gives you detailed information about the performance of the report including the actual SQL queries (under “Direct query”) which were executed on the data sources. The plain text queries can also be copied using the “Copy queries” link at the bottom. In our case 3 SQL queries were executed against 2 different SQL databases:

Query 1:

Result:
Results_Query1

The query basically selects two columns from the DimProductSubcategory table:

  1. ProductSubcategoryKey – which is used in the join with DimProduct
  2. ProductSubcategoryName – which is the final name to be displayed in the visual

The inner sub-select (line 7-14) represents the original Power Query query. It selects all columns from the DimProductSubcategory table and renames [EnglishProductSubcagetoryName] to [ProductSubcategoryName] (line 10). Any other Power Query steps that are supported in direct query like aggregations, groupings, filters, etc. would also show up here.

Query 2 (shortened):

(The query was shortened at line 16 and line 29 as the removed columns/rows are not relevant for the purpose of this example.)

Similar to Query 1 above, the innermost sub-select (line 13-17) in the FROM clause returns the results of the Power Query query for DimProduct whereas the outer sub-select (line 7-20) groups the result by the common join-key [ProductSubcategoryKey].
This result is then joined with a static table which is made up from hard-coded SELECTs and UNION ALLs (line 24-30). If you take a closer look, you will realize that this table actually represents the original result of Query 1! Additionally it also includes a special NULL-item (line 30) that is used to handle non-matching entries.
The last step is to group the joined tables to obtain the final results.

Query 3 (shortened):

(The query was shortened at line 9 as the removed columns/rows are not relevant for the purpose of this example.)

The last query is necessary to display the correct grand total across all products and product sub-categories.

As you can see, most of the “magic” happens in Query 2. The virtual join or virtualization is done by hard-coding the results of the remote table/data source directly into the SQL query of the current table/data source. This works fine as long as the results of the remote query are small enough – both, in terms of numbers of rows and columns – but the more limiting factor is the number of rows. Roughly speaking, if you have more than thousand items that are joined this way, the queries tend to get slow. In reality this will very much depend on your data so I would highly recommend to test this with your own data!

I ran a simple test and created a join on the SalesOrderNumber which has about 27,000 distinct items. The query never returned any results and after having a look at the Performance Analyzer I realized, that the query similar to Query 2 above was never executed. I do not know yet whether this is because of the large number of items and the very long SQL query that is generated (27,000 times SELECT + UNION ALL !!!) or a bug.

At this point you may ask yourself if it makes sense to use Power BI for data virtualization or use another tool that was explicitly designed for this scenario. (Just google for “data virtualization”). These other tools may perform better even on higher volume data but they will also reach their limits if the joins get too big and, what is even more important, the are usually quite expensive.

So I think that Power BI is still a viable solution for data virtualization if you keep the following things in mind:
– keep the items in the join columns at a minimum
– use Power Query to pre-aggregate the data if possible
– don’t expect too much in terms of performance
– only use it when you know what you are doing 🙂

Downloads:

PowerBI_DataVirtualization_Part2.pbix
SQL_Query1.sql
SQL_Query2.sql
SQL_Query3.sql

Data Virtualization in Microsoft Power BI

Posted on by — 6 Comments ↓

Data Virtualization is actually a very new topic to me as I have barely seen it implemented in the real world or at any of my customers. But it becomes more and more interesting when working with big data where you cannot simply load all data into a single in-memory data model but still need to query across different data sources. So I decided to investigate how this could be done with my favorite reporting tool Power BI which I know is capable to connect to different data sources out of the box and also provides a rich set of visualizations that I need.

But let’s start slowly.

What is Data Virtualization?

According to Wikipedia, “Data virtualization is any approach to data management that allows an application to retrieve and manipulate data without requiring technical details about the data, such as how it is formatted at source, or where it is physically located,[1] and can provide a single customer view (or single view of any other entity) of the overall data.”

So basically, combining data from multiple sources and multiple formats into a common semantic layer which can be queried on-the-fly without the need of any ETL/ELT.

Sounds awesome – right?

The problem is that in reality the things are not as simple as they may sound, especially when it comes to joining across the different sources. While data virtualization usually works fine for small amounts of data that can be easily processed, it can be quite challenging  on large amounts of data which is where data virtualization would actually make sense to avoid lengthy and costly ETL/ELT.

What does Power BI have to do with this?

At first sight – nothing. But lets examine what we currently have in Power BI:

  • a semantic layer and data modelling capabilities
  • access to various data sources via Direct Query (remember, we do not want to load any data!)
  • ability to combine data from those sources

The last part is the most important one here and you may wonder what I am talking about. And you are right, by default a DQ model is only linked to one data source at a time but you can add other data sources manually in the Power Query editor!
This is where it get’s interesting and what this blog post is about.

Test-Case

To verify the statements from above I built a little test case that involves 2 local database and an Azure SQL database over which I want to create a semantic layer using Power BI. All three databases are actually the same AdventureWorksDW databases but for the purpose of this demonstration this is OK as it is just a technical feasibility study.

I started by creating a new Power BI file and connected it to my first data base in Direct Query mode.

Then I selected a single table, in my case I have chosen “FactResellerSales”.

The next step is to add a new table using the “Edit Queries” button on the ribbon:

EditQueries_AddNewDatasources

You will see the one table you have just selected before as a Power Query query which you can simply copy and rename. I renamed it to “DimProduct” as I want to load the DimProduct table from my second local database which can be accomplished by simply changing the connection to the SQL database in the first step “Source” of the query (I use my second local database AdventureWorksDW2014 – instead of AdventureWorksDW2012):

Several things to point out here:

ChangeLocalDatabaseAndTable
  1. once you change the name of the server or the database, you may get prompted for credentials
  2. when you click on Table in the Data column, Power BI asks you if you want to replace the next step – simply press [Yes]
  3. at the “Source” step, Power BI will complain that the results of the current steps are not valid in Direct Query mode – this is fine as the final result will be in the next step (“Navigation”)

I repeated the same procedure again and also added the table “DimProductSubcategory” from my Azure SQL database.
So right now we have 3 Direct Query tables pointing to 3 different databases and 3 different tables.

Setup_Relationships

Now we need to connect our tables in the Relationship-view – similar as you would to with any other tables in a regular Direct Query setup:

Once the relationships are created, we can finally create our reports.

Are you excited? Well, I definitely was when I tried this setup the first time!

InteractiveReport

The visuals behave as if they were created on top of an Import Mode dataset or a Direct Query dataset that only connects to a single database. I think that is pretty awesome and again shows what Power BI is capable of!

We just used Power BI to create a semantic layer across different databases and tables which are now all joined and queried on-the-fly always showing the most recent data!

Conclusion

As you have seen, you Power BI allows us to combine multiple SQL databases in Direct Query mode and query them together as if they would be one single data source. So if you have a requirement where your data is distributed across databases and you it is too big to be loaded into memory or you need live data, you can give this approach a try. I have only tested it with regular Microsoft SQL databases but I assume that this works in a similar way with any other data source that supports Direct Query (e.g. SAP, Oracle, Spark, …) too. You can also do some basic transformations before joining the data in Direct Query mode which can also be very crucial when combining different data sources that might have slightly different formats. Again, I have not tested this thoroughly but at least everything that can be query folded should be  supported as a transformation.

In a follow-up post I will explain the technical details and what actually happens in the background when you use a setup like this so stay tuned!

Power BI – Dynamic TopN + Others with Drill-Down

Posted on by — 85 Comments ↓

A very common requirement in reporting is to show the Top N items (products, regions, customers, …) and this can also be achieved in Power BI quite easily.

But lets start from the beginning and show how this requirement usually evolves and how to solve the different stages.

The easiest thing to do is to simply resize the visual (e.g. table visual) to only who 5 rows and sort them descending by your measure:

This is very straight forward and I do not think it needs any further explanation.

The next requirement that usually comes up next is that the customer wants to control, how many Top items to show. So they implement a slicer and make the whole calculation dynamic as described here:
SQL BI – Use of RANKX in a Power BI measure
FourMoo – Dynamic TopN made easy with What-If Parameter

Again, this works pretty well and is explained in detail in the blog posts.

Once you have implemented this change the business users usually complain that Total is wrong. This depends on how you implemented the TopN measure and what the users actually expect. I have seen two scenarios that cause confusion:
1) The Total is the SUM of the TopN items only – not reflecting the actual Grand Total
2) The Total is NOT the SUM of the TopN items only – people complaining that Power BI does not sum up correctly

As I said, this pretty much depends on the business requirements and after discussing that in length with the users, the solution is usually to simply add an “Others” row that sums up all values which are not part of the TopN items. For regular business users this requirement sounds really trivial because in Excel the could just add a new row and subtract the values of the TopN items from the Grand Total.

However, they usually will not understand the complexity behind this requirement for Power BI. In Power BI we cannot simply add a new “Others” row on the fly. It has to be part of the data model and as the TopN calculations is already dynamic, also the calculation for “Others” has to be dynamic. As you probably expected, also this has been covered already:
Oraylis – Show TopN and rest in Power BI
Power BI community – Dynamic Top N and Others category

These work fine even if I do not like the DAX as it is unnecessarily complex (from my point of view) but the general approach is the same as the one that will I show in this blog post and follows these steps:
1) create a new table in the data model (either with Power Query or DAX) that contains all our items that we want to use in our TopN calculation and an additional row for “Others”
2) link the new table also to the fact table, similar to the original table that contains your items
3) write a measure that calculates the rank for each item, filters the TopN items and assigns the rest to the “Others” item
4) use the new measure in combination with the new table/column in your visual

Step 1 – Create table with “Others” row

I used a DAX calculated table that does a UNION() of the existing rows for the TopN calculation and a static row for “Others”. I used ROW() first so I can specify the new column names directly. I further use ALLNOBLANKROW() to remove to get rid of any blank rows.

Step 2 – Create Relationship

The new table is linked to the same table to which the original table was linked to. This can be the fact-table directly or an intermediate table that then filters the facts in a second step (as shown below)

Step 3 – Create DAX measure

That’s actually the tricky part about this solution, but I think the code is still very easy to read and understand:

Step 4 – Build Visual

One of the benefits of this approach is that it also allows you to use the “Others” value in slicers, for cross-filtering/-highlight and even in drill-downs. To do so we need to configure our visual with two levels. The first one is the column that contains the “Others” item and the second level is the original column that contains the items. The DAX measure will take care of the rest.

And that’s it! You can now use the column that contains the artificial “Others” in combination with the new measure wherever you like. In a slicer, in a chart or in a table/matrix!

The final PBIX workbook can also be downloaded: TopN_Others.pbix