Episode 321 – Microsoft Fabric February 2026 Feature Summary
Fresh off back-to-back recording sessions — and a working trip together in Hawaii that apparently didn’t leave enough time to actually record a podcast — John and Jason are back to dig into the Microsoft Fabric February 2026 Feature Summary. Jason voted in the Texas primary between episodes. John is still in Hawaii but not for long. And despite Jason’s ongoing frustration with the Fabric blog’s formatting (it remains, as he puts it, “far and away inferior” to the Power BI blog), there’s a solid collection of updates to work through this month spanning the platform, data engineering, data science, warehouse, real-time intelligence, data factory, and the VS Code extension.
Oh — and Fabric Databases is conspicuously absent from this month’s post. More on that at the end.
Platform: The OneLake Catalog Is Becoming the Center of Everything
Workspace Apps Now Surface in the OneLake Catalog
Workspace apps — which John and Jason spent a moment untangling from Power BI apps, org apps, and apps v2 (the naming situation is genuinely confusing) — now appear in the OneLake catalog. The practical effect: if you want to find content across your tenant, you no longer have to bounce workspace by workspace. The catalog is becoming the one-stop shop for surfacing everything in Fabric, regardless of type. John’s take: the OneLake catalog should eventually be the be-all, end-all for content discovery. This is a step in that direction.
Streamlined Item Details View
When you open any Fabric item from the catalog, you now get a consistent preview experience before diving in — showing you what’s inside before you click “Open.” Jason hates it. Or rather, hated it — because it rolled out before the blog post dropped and caught him mid-demo unprepared. He described it as “they moved my cheese, John,” complete with a Swiss cheese metaphor that admittedly landed. His actual verdict after reflection: it makes sense coming from the catalog, it just takes some getting used to. Give it a month.
Workspace Identity Limit: 1,000 → 10,000
Workspace identities in Fabric — essentially service principals in Entra ID created specifically for Fabric workspaces — had a hard limit of 1,000 per tenant. That limit has been increased to 10,000. You can also set a custom limit below 10,000 if you want to stay well clear of Entra ID’s overall tenant ceiling of 30,000 entities. (Beyond that, you’d need to file a support ticket to have the limit raised further.) John had a use case for around 40 workspace identities that made this relevant personally. Jason pointed out this is one of those limits that feels fine until it isn’t — and for larger enterprises, a thousand goes fast.
Horizontal Tab Improvements: Adaptive Names + Overflow Menu
Two quality-of-life wins for the new Fabric UI’s horizontal tab experience. First, tab names can now be truncated adaptively — so instead of your tab text eating your entire screen, Fabric can shorten it intelligently while still keeping it identifiable. Whether AI is involved in the truncation is unclear from the blog post, but John thinks the “adaptive” label is a hint it’s more than a simple character count. Second, the overflow behavior has been improved: instead of that tiny arrow you could never quite click at the edge of the tab strip, there’s now a proper overflow icon that drops a vertical list of all your open items. Jason liked this one without hesitation.
Data Engineering: Notebooks Get Smarter
Notebook Version History Now Captures All Changes — Including CI/CD
Fabric notebooks already supported version history when editing in the web UI, auto-saving in the background so you could roll back. What wasn’t being properly captured: changes made through Git, CI/CD pipelines, or IDE plugins outside the browser. That gap is now closed. All external edits now surface in the version history alongside UI edits. Jason called this one out as completing the story — before, version history felt like it only told part of the truth.
%run Directive Support for Python Notebooks
Python notebooks in Fabric now support the %run directive, allowing you to execute one notebook from within another. The practical use case: put all your library imports and variable declarations in a shared notebook, then call it from multiple others rather than duplicating that setup everywhere. It runs in the context of the calling notebook. Jason confirmed this is specific to Python notebooks — PySpark and T-SQL notebooks aren’t in scope here.
Full-Size Cell Mode in Notebooks
One button click now expands a notebook cell to fill the full screen. If you’ve ever been on a laptop trying to read a cell that’s drowning in scroll bars — three layers deep, competing for vertical space — you know exactly why this matters. Jason flagged this as particularly valuable for anyone not working on a wide-screen curved monitor with plenty of real estate. Basically: everyone demoing or working remotely.
Private Link Support for GraphQL API + Full CI/CD Support
Fabric’s API for GraphQL now supports tenant-level private link connections. It’s niche — neither John nor Jason have personally used GraphQL in Fabric yet — but for organizations with compliance requirements around network isolation, this is a meaningful unlock. CI/CD support for GraphQL in Fabric has also been added, continuing the pattern of making every workload fully compatible with modern DevOps pipelines.
Default Arguments for User-Defined Functions (UDFs)
Fabric User-Defined Functions now support default arguments. Previously, if a value wasn’t passed in, the function wouldn’t run. Now, if a value isn’t specified, the function runs with the default. Supported input types include strings, booleans, floats, arrays, and objects. Clean, useful, and exactly what you’d expect from a maturing function system.
ODBC Driver for Fabric Data Engineering (Preview)
John’s response to this one: “Welcome to 1995.” But then immediately acknowledged — if you’ve got a tool that talks ODBC and you want it to connect to Fabric Lakehouse data, you now have a driver for that. It’s powered through the Fabric REST APIs and supports flexible Spark SQL connectivity for .NET, Python, and other ODBC-compatible BI tools. Niche? Yes. Needed by someone? Absolutely.
Customer-Managed Key (CMK) Encryption for Notebook Code
Organizations that require customer-managed key encryption before they’ll commit certain workloads to a platform now have CMK support for notebook code. Short mention on the show — if your security team has been waiting on this, the wait is over.
Data Science: Semantic Link 0.13 and Real-Time Scoring Monitoring
Semantic Link has shipped version 0.13. Jason’s hot take: at version 13, maybe it’s time to drop the “0.” and call it version 13. John concurred. The library remains the primary way to query Fabric semantic models programmatically using natural language — and it keeps getting better with each release.
For teams building and running machine learning models in Fabric, real-time scoring endpoints now have a proper monitoring experience. You can track request volume, error rates, and latency as models run in production, compare metrics across endpoint versions, and make rollback or rollout decisions based on real usage data. John hit his stated limit on ML and data science depth, but the capability is there if you’re working in that space.
Data Warehouse: Migration Gets Easier, Pool Monitoring Gets Deeper
Export Migration Summary
The Migration Assistant in Fabric Data Warehouse now lets you export your full migration results — as Excel or CSV — directly from the summary or full-screen view. It sounds like a small thing, but having a shareable, portable artifact of your migration state is the kind of thing that unblocks a lot of enterprise approvals and handoffs.
SQL Pool Insights
Understanding why warehouse workloads slow down has historically required poking around at the individual query level, without much visibility into resource pressure at the pool level. SQL Pool Insights extends the existing Query Insights experience with pool-level telemetry — letting you see pressure events, resource allocation, and isolation between read-optimized and write-optimized workloads. John summed it up well: a lot of this was opaque before. Now it’s not.
Real-Time Intelligence: John’s Favorite Section
Get Data Menu Consolidation
Two menu items — “Data Sources” and “Azure Sources” — have been collapsed into a single “Get Data” entry point in the Real-Time Intelligence UI. John’s summary: “I threaded it so you don’t have to.” Two pages of blog post, one practical takeaway. Simpler navigation.
Private Network Support for Eventstream Connectors
Eventstream connectors can now reach into private networks. Jason clarified the important distinction here: this isn’t the on-premises data gateway that Power BI users know well. This is proper Azure private networking — ExpressRoute gateways and site-to-site VPN configurations that connect your Fabric environment to on-premises data sources at the network level. More complex to configure than dropping in a gateway server, but more capable and appropriate for enterprise-grade event streaming scenarios. This has been asked for for a while.
Real-Time Dashboard Performance: Up to 10x Faster
The Real-Time Dashboard — which runs live queries against Eventhouse data — has received significant performance improvements, with rendering up to 10x faster in some scenarios. John initially misquoted this as 10% before correcting himself. Jason noted he’s always in support of faster pie charts. John, who genuinely loves the Real-Time Dashboard feature, was clearly pleased. This is a meaningful improvement for operational monitoring use cases.
Data Factory: Quality-of-Life Wins Across the Board
Recent Sources in Data Factory
When creating a new data flow or pipeline in Data Factory, you can now access your recently used data sources directly — from the home tab or a dedicated “recent data” module. No more hunting through the full connector list when you’re repeatedly working with the same sources. Jason called this one “super useful” without hesitation. John agreed. It’s one of those features that makes you wonder how it wasn’t there from the start.
Dataflow Gen 2: No More Variable Retrieval Limits
There used to be a cap on how many variables a Dataflow Gen 2 could retrieve from a variable library. That cap has been removed entirely. No increase — just gone. And in the Power Query editor, variables are now evaluated in real time as you work, which is exactly the behavior you’d want when building and testing transformations.
Relative Referencing for Fabric Items in Dataflow Gen 2
This one has real implications for CI/CD and deployment pipelines. Previously, when a Dataflow Gen 2 referenced another Fabric item, it stored an absolute reference — the full URL to that specific item in a specific workspace. Now you can use relative references, which means “the current workspace” rather than a hardcoded workspace name. Move a dataflow through a deployment pipeline from dev to test to production, and the references follow correctly without any variable overrides needed. Jason said his team is going to be excited about this one.
Dataflow Gen 2: Run Without Publishing First
Remember when Dataflow Gen 2 first shipped and you couldn’t save the thing without publishing? That particular pain has been chipped away at over time, and here’s the latest improvement: you can now trigger a run directly without manually publishing first. Fabric publishes in the background. It’s one less step, but it’s the kind of friction removal that adds up.
New Query Evaluation Engine in Dataflow Gen 2 (GA)
The query evaluation engine — responsible for generating previews while you build dataflows — has received a complete overhaul and is now generally available. Performance has gone up significantly. If Dataflow Gen 2 felt sluggish when you first tried it, it’s worth revisiting. The experience has improved substantially from those early days.
Copy Job: Incremental Copy Now Supports Both CDF and Watermark Methods
Incremental copy from Fabric Lakehouse in the copy job now supports both the change data feed (CDF/delta-based) method and the traditional watermark method. John and Jason got briefly tangled in the CDC vs CDF acronym soup — change data capture vs. change data feed — but the upshot is: use delta change data feed when you can, it’s the more reliable approach and the one Microsoft recommends.
Additional Copy Job Improvements
A few more copy job updates rounded out the Data Factory section: SAP Datasphere outbound support for Amazon S3 and Google Cloud Storage; column mapping in CDC replication for copy jobs; row version support as an incremental column in the SQL database copy job; and service principal and workspace identity authentication now supported in copy jobs — continuing the platform-wide rollout of this auth pattern across Fabric workloads.
Parallel Read for Large CSV Files + Adaptive Performance Tuning (Preview)
Large CSV files can now be processed in parallel rather than sequentially. When multi-line behavior is explicitly defined, Fabric can identify record boundaries in large files, partition them into logical chunks, and process those chunks concurrently. Jason noted he’d always assumed CSV had to be processed as one big sequential blob — this changes that assumption. The related preview feature, adaptive performance tuning, dynamically adjusts copy throughput based on real execution conditions and available resources — essentially a burst-and-throttle approach that optimizes your capacity usage automatically.
VS Code Extension: MCP Support and Direct Item Editing
The Fabric VS Code extension has been enhanced with the ability to browse workspace folders, view and edit Fabric item definitions (the code representations that sync to Git — not the item contents themselves), and use the Fabric MCP server to create, edit, and delete items through natural language with AI tools like GitHub Copilot.
John went slightly off-script here to give an unprompted endorsement of the Fabric RTI MCP server: “Astoundingly good.” For anyone working with Eventhouse data, it can not only help build and configure ingestion policies through natural language — it can read the data and answer questions about it. Ask it how to approach a problem, and it’ll build you a materialized view. John uses it regularly. Worth trying if you haven’t.
The Elephant in the Room: Where’s Fabric Databases?
Jason flagged this at the end of the episode and it’s worth calling out: Fabric Databases — both SQL database and Cosmos — is absent from the February feature summary. No dedicated section, no feature updates. There was a wrap-up blog post from Anna on February 12th (“SQL database in Fabric, built for SaaS, ready for AI”), but nothing that reads as a monthly feature drop.
John’s read: sandbagging. Save it for FabCon. Jason’s read: also possibly sandbagging, and he’s hoping that means big news is coming. John noted there’s a dedicated SQL + Fabric session on the FabCon schedule, which supports the theory. If you’re deep in Fabric Databases, keep your eyes on the FabCon announcements. The absence might be the most interesting data point in this month’s update.
What’s Next
John is heading to FabCon Americas 2026 in Atlanta (March 16–20) and then to MVP Summit before reuniting with Jason in Cologne for the European Collaboration Summit. The plan is to record something from Atlanta — potentially bringing members of John’s team along. Jason is staying home and watching the FabCon keynote from his couch, which he acknowledged is “always a win.”
Expect the next episode to cover FabCon announcements. Based on this month’s omissions, there’s likely more to discuss than a typical monthly feature drop.
Links
Microsoft Blog Posts Referenced in This Episode
Previous Episodes
- Episode 318 – Power BI January 2026 Feature Update
- Episode 316 – 2024 Recap and 2026 Predictions: Looking Back and Looking Forward
Events
- European Collaboration Summit – Cologne, Germany
- TechCon365 Chicago
- European Power Platform Conference – Copenhagen
Subscribe: SoundCloud | iTunes | Spotify | TuneIn | Amazon Music
2 Replies to “Episode 321 – Microsoft Fabric February 2026 Feature Summary”