Episode 296 – Developer-Focused Updates Arrive Post-Build

Recording just after wrapping their European Collaboration Summit sessions in Germany, John and Jason tackled the Microsoft Fabric May 2025 feature summary—a release heavily influenced by Microsoft Build that brought substantial developer-focused enhancements alongside expanding regional availability and database options.

Cross-Tenant DevOps Finally Arrives

While REST APIs for deployment pipelines and service principal support made headlines, Jason highlighted a buried gem: cross-tenant support for Azure DevOps. This functionality enables organizations to connect workspaces to different Azure DevOps tenants based on identity—critical for multinational organizations with multiple entities or third-party development scenarios.

“This means you can connect your workspaces to different Azure DevOps tenants,” Jason explained. “A service provider can build in their tenant and deploy to yours, then you can disconnect and reconnect to your own repo.” The capability also works in reverse: multiple Fabric tenants connecting to a single DevOps instance.

John noted Azure DevOps now supports service principals for connections, finally catching up to GitHub’s existing implementation. For organizations choosing between the platforms, Jason emphasized DevOps brings more to the table around boards, bugs, and task management—making it better suited for organizational versus communal development.

Cosmos DB Joins the Fabric Family

One of Build’s bigger announcements landed in Fabric: Cosmos DB in Microsoft Fabric, now in preview. Like SQL Database before it, Cosmos becomes a first-class Fabric citizen with data automatically mirrored into OneLake.

Currently requiring a preview interest form (not guaranteed access), the feature represents Fabric’s expanding database category. John noted it’s “just like SQL database in fabric—we now have more than one leaf in that category.”

The duo expects this integration will consume capacity units, though preview terms remain unclear. The addition continues Fabric’s strategy of bringing diverse data sources into a unified platform.

Fabric Data Agents Expand Their Reach

Fabric data agents (formerly AI Skills—a naming change Jason wishes came with “formerly known as” annotations for searchability) now integrate with both Power BI copilot and Copilot Studio. These agents ground large language models in specific datasets, enabling natural language queries across Lakehouse, Warehouse, Semantic Models, and KQL databases.

Jason’s actively experimenting with the capability, planning to enrich weather data with power consumption information for future sessions. John emphasized the flexibility: “If you’ve got Lakehouse, all you need is a shortcut and you can pick up SQL data, Event House data—everything surfaced in OneLake.”

The integration with Copilot Studio particularly matters, letting organizations deploy data agents anywhere Studio supports—extending beyond Fabric’s boundaries.

Dataflow Gen2 Reaches CICD Maturity

Dataflow Gen2’s CICD functionality hit general availability, eliminating the confusing dual-creation paths that plagued earlier previews. All new dataflows now include CICD capabilities with GitHub and DevOps integration.

John acknowledged uncertainty about whether existing pre-CICD dataflows automatically upgrade, noting a migration mechanism exists similar to Gen1-to-Gen2 paths. The UI delineation between versions remains murky—something users will need to navigate carefully.

Parameterization finally arrived in preview, letting users pass parameters into dataflows from pipelines or REST API calls. John hit an immediate limitation: parameters can’t form part of source file paths yet. “That’s one of the known issues,” he noted. “If you want to use it for that purpose, this isn’t going to do it for you.”

The bigger news for collaboration-focused audiences: SharePoint files as a destination. Dataflows can now write CSV files directly into SharePoint document libraries—the first semi-structured destination available.

“I can see huge uses for this,” Jason enthused. “Exfiltrating data into third party systems, intermediate staging areas, letting people review information without Power BI access.” He’d previously used paginated reports for similar purposes; dataflows provide a cleaner path.

Warehouse Snapshots & Event House Acceleration

Warehouse snapshots entered preview, providing read-only point-in-time views leveraging Delta Lake’s time travel capabilities. Data retains for 30 days (configurable retention coming), and snapshots can roll forward to supersede current state.

John wondered whether similar functionality might eventually reach Lakehouse, given both use Delta Lake underneath. “I’m also curious to see if we would have this in Lakehouse at some point,” he mused.

Event House accelerated OneLake shortcuts reached general availability—a significant performance milestone. The feature lets users create shortcuts from OneLake data into Event House with near-native performance, making storage location less critical.

“It kind of starts to not matter where you put your data first,” John explained. “It really has to do with how you’re processing it, who’s working with it, the nature of the data itself—less about platform capabilities.”

Users can toggle acceleration on/off per shortcut, though John suspects it consumes capacity units. Jason theorized turning it off during low-demand periods could help manage throttling.

SQL Server Mirroring Expands

Mirroring grew substantially, adding SQL Server 2016-2022 on-premises (via data gateway) and SQL Server 2025 as separate implementations. The version gap puzzled both hosts—no on-prem releases between 2022 and 2025 suggests Microsoft’s returned to three-year cycles.

Azure SQL Managed Instance gained private endpoint support and removed primary key requirements for mirrored tables. Users can now truncate source tables while mirroring stays active—important compatibility improvements.

Customizable retention periods rolled out across all mirroring, though John clarified this refers to snapshot/time-travel windows rather than data retention. “It’s controlling when the vacuum process runs to optimize Delta Lake,” he explained. “When you do that, you lose time travel capability.”

Regional Expansion & Minor Updates

User-defined functions expanded to Canada Central (finally prompting John’s “go Canada!”), plus regions across Australia, South America, Europe, and Asia. Only two US expansions (West US, North Central US) made the list, showing Microsoft’s geographic diversification priorities.

Other updates included:

  • Native CDC support in copy jobs for continuous data movement
  • Semantic model refresh activity now generally available in pipelines
  • Copilot for data pipelines generating pipeline designs from natural language descriptions
  • Incremental refresh for dataflows writing to Lakehouse destinations
  • Copilot column transformations in dataflows using natural language

The Mystery of Shortcut Transformations

One feature received minimal coverage: shortcut transformations announced at Build. “There’s no detail here,” John noted. “I don’t know what this represents because they are not yet available.” The preview links only to Build’s announcement paragraph with no additional documentation.

John confirmed awareness of details under NDA but couldn’t discuss publicly. “This is an absolutely exciting feature,” he teased, “but I don’t know anything more about it from a public standpoint.”

Looking Ahead

With Seattle Collaboration Summit approaching (where Jason presents solo while John attends his daughter’s graduation) and European SharePoint Conference in Dublin scheduled for December, the pair continue refining their session portfolio. They’re particularly excited about a brand-new reporting session debuting in Dublin.

May’s Fabric updates delivered on Build’s developer-focused promise—API expansions, CICD maturity, cross-tenant support, and new database integrations. As John summarized: “Evolutionary changes mostly, but a good update month.”


Links:

Subscribe: SoundCloud | iTunes | Spotify | TuneIn | Amazon Music


Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

AvePoint