Demystifying the Power BI XMLA Endpoint

When the Power BI product team began promoting the new XMLA endpoint connectivity for Power BI I thought that this could be a game changer for IT-driven, enterprise class BI solutions. Now that I have used it in a few real solutions and actively working with clients, I’m sharing my experience about how it works and what you can do with it. The read-only endpoint has been in GA for Premium capacities and the read/write endpoint is currently in preview.

Before my long-winded introduction, I’ll get to the point:
Using the XMLA endpoint, the Power BI Premium service now includes the capabilities of SQL Server Analysis & Azure Analysis Services combined with newer data modeling capabilities of Power BI. Data models published to the Power BI service now support version control, scripted builds and team application life cycle management, enterprise IT tooling and scripted object management.

…wait, what? How?

Let’s just start by reviewing some of the challenges that have existed in the Power BI platform prior to the availability of this capability:

  • SSAS/AAS enterprise features
    Buried deep within the Power BI cloud service is the SQL Server Analysis Services (SSAS) Tabular model “Vertipaq” in-memory analytics database engine. The SSAS engine & on-prem product itself has many useful features not exposed in the Power BI implementation of data models and datasets. These “enterprise class” features are numerous including object-level scripting for deployments, source control, application life cycle, continuous integration and build management, data partitioning, perspectives, translations, currency conversion, KPI definitions, measure groups and calculation groups.
  • SSAS/AAS redundant services & costs
    At one point, Azure Analysis Services was a superset of data modeling features but now many exclusive features are available in Power BI datasets in a Premium capacity workspace so this choice isn’t so clear. Power BI and AAS also have separate costs. As the Power BI services continues to evolve, many new an compelling features are available only in Power BI datasets and not in Analysis Services; such as hybrid mixed-mode data models, aggregations, incremental refresh policies, dataflows and AutoML.
  • Source control, Application life cycle, builds & Team data model development
    Power BI Desktop is a convenient tool for creating an entire Power BI solution from soup to nuts but isn’t optimized for IT scale projects. A single PBIX file contains connections, queries, data model objects, measures and report pages. A PBIX file structure cannot be easily parsed, compared, branched or merged by established source control and build management tools. Analysis Services projects on the other hand support Microsoft development tools like Visual Studio, Azure DevOps and SQL Server Management Studio. Several mature third-party development tools like Tabular Editor, DAX Studio and the ALM Toolkit enhance or exceed the features of Microsoft’s development suite.

The XMLA endpoint in Power BI Premium bridges this divide by exposing the underlying Analysis Services database instances in the Power BI service. It really is as simple as that. Each workspace is in fact an SSAS instance and each Power BI dataset is an SSAS database. What you can do with traditional SSAS through management and development tools, TMSL or XMLA script, PowerShell or API automation; you can do with Power BI.

Use a Premium capacity workspace

If you have Premium capacity setup in your tenant, you are good to go. Several of our consulting clients do but to experiment and test new features, I use my own private tenant. In lieu of paying for Premium capacity, I can setup an embedded capacity in the Azure portal. Sizing the embedded capacity to A4 is the same as a P1 premium capacity but it can be paused when I don’t need to use it. The cost is about $8.50 per hour so MAKE SURE TO PAUSE WHEN DONE.

Here’s the Power BI Embedded capacity I created in my Azure tenant, sized to A4 and currently paused. It takes about a minute to start or pause the embedded service.

After starting the capacity, I jump over to the Power BI Admin Portal and click the capacity name to change settings. Note that because my capacity was added as an embedded service, it shows up in the Power BI Embedded page but the settings are the same for a Premium capacity.

Enable the Read/Write endpoint

By default, the XMLA endpoint will be set to Read Only. Under the Workloads group on the Capacity settings page, switch XMLA endpoint to Read Write and then save the changes.

Now any workspace set to use Premium capacity can be accessed with an XMLA endpoint address. I’ve created a test workspace in my tenant to test migrating an SSAS tabular database to a Power BI dataset. I select the workspace in the Power BI Portal and in the workspace settings, make sure that Dedicated capacity is switched on. I know that it is because of the diamond icon next to the workspace name. The Workspace Connection address is below. Click the Copy button to get the address on my clipboard.

Now, I can use that address in any tool that knows how to connect to SSAS tabular in compatibility level 1450 or above. Let’s try to connect using SQL Server Management Studio. I need a newer version of SSMS, 18.4 or higher.

Connect to SQL Server Analysis Services and paste the XMLA endpoint address for the Server name. You need to use Azure Active Directory authentication. If your organization uses MFA, you can use that option but I will choose regular AAD authentication.

…and viola! Connected.

I can run queries and scripts here in DAX, MDX, XMLA or TMSL. Most but currently not all capabilities are supported in the preview. In particular, if you have RLS roles, the members must be dropped and then added back in the Power BI Portal.

So far, I have scripted existing Power BI datasets and migrated them to Analysis Services projects in Visuals Studio, and then deployed to a new dataset from Visual Studio. The learning here is that migration is a one-way street from Desktop to Visual Studio. Whether development starts in Power BI Desktop or Visual Studio, there is no going back to Desktop. Ongoing development must be in Visual Studio.

Definitions: “Report”, “Database” and “Dataset

In the self-service Power BI world, the term “Report” has been used to mean at least two different things. With the queries and data model managed separately and deployed as a Power BI dataset, the term “Report” in this context means only report pages with visuals (don’t get me started talk about “dashboards”). A data model originating from a Power BI Desktop file is published as a dataset. Now that we see these objects through the lens of Analysis Services, a Power BI dataset is a Database.

In migrated projects, continue to author and manage reports in Desktop connected to the deployed dataset as an Analysis Services connection. You can actually switch the connection between an SSAS instance, an AAS instance or a published dataset using the XMLA endpoint address. I usually use query parameters for the Server and Database to easily change these values.

There’s a lot more that I will need to cover in later posts, but I’ll mention a few things briefly.

Best practices: After working with the XMLA endpoint, the choices now seem pretty clear to me but there was some confusion until I got to that point. Best practices will continue to emerge. In light of this and other recent announcements, I can say that I have a much clearer vision for how to plan and manage solutions (and to delineate between self-service and enterprise BI projects) than I did a year ago. If you have questions, please post them in the comments and I’ll do my best to address them in future posts.

Combining & querying datasets: The ability to use the the endpoint to query one data model from another model enables some very compelling composite model scenarios – but planning these solutions is important.

Large models & storage: The size and scale limitations are similar to those in AAS and generally only limited by your chosen Premium capacity. Since models are compressed and typically only include column data needed for analytic reporting, it is unusual to see datasets larger than a few gigabytes but Premium will support model sizes up to 400 GB.

By default, datasets deployed using the endpoint are stored in the Azure data centers using single file storage. This is fine for small models but after creating larger partitioned models, using large file storage will improve development and deployment performance by managing objects in multiple files on the instance backend. There will eventually be a UI for this setting but it is currently available only through the management API or via PowerShell.