Episode 328 – Fabric June 2026 Feature Summary (Part 2)
Episode 328 wraps up the Microsoft Fabric June 2026 Feature Summary — recorded live and in person from the European Power Platform Conference in Copenhagen, sessions finished, apple cider in hand, and a hotel lobby slowly filling up around the microphones. Part 1 covered the platform, OneLake, Data Engineering, Data Science, and Conversational Analytics. This one takes the back half in a single pass: GPUs arrive in Data Warehouse, Real-Time Intelligence gets an AI-first dashboard editor, Business Events get a proper coming-out party, Kafka goes GA, and Project Rayfin brings real applications to Fabric.
A quick note on pace: with a July update already on the horizon and Power BI news stacking up behind it, this episode deliberately stays at the top of the waves rather than going three sessions deep. Seventy-four degrees in Copenhagen apparently feels a lot warmer than it sounds, and shorts and flip-flops were in play for every minute the hosts weren’t presenting.
Data Warehouse: GPUs, DacFx, and Monitoring That’s Actually Front and Center
The headline is GPU-accelerated Fabric Data Warehouse. It brings GPU-powered execution into the warehouse so agents, applications, and reports can keep issuing complex analytical queries without the usual tradeoffs between performance, concurrency, and complexity. The open question on the show is whether this eventually reaches other workloads — and, less enthusiastically, what “GPU-powered” means for CU consumption. Faster is good. Keeping an eye on the meter is still wise.
CI/CD support for the SQL analytics endpoint (preview) arrives via DacFx — the Data-tier Application Framework, the .NET library Microsoft has long used for database application lifecycle work. Endpoint definitions can be managed as DacFx database projects alongside other Fabric items in Git, with incremental schema changes deployed through Fabric Deployment Pipelines. Alongside it, pre- and post-deployment scripts for Fabric Warehouse (preview) let teams validate prerequisites and schema readiness before a deployment, then apply permissions, seed data, and run validation afterward — DBA-flavored work, now wired into the pipeline instead of someone’s checklist.
ALTER COLUMN (preview) brings in-place column changes — length, precision, scale — through standard T-SQL, with no underlying data rewrite, which cuts down on CTAS workarounds in migrations. Surprising to learn it wasn’t already there.
The most interesting item for John is Datawarehouse Monitor (preview), previously known as Query Activity. It puts running and historical queries in one place with per-query metrics like CPU time, data scanned, and cache usage, so you can spot expensive or failed queries and cancel runaway ones — no T-SQL required. It’s the same kind of hidden-but-high-value monitoring he’s been living in on the Eventhouse side, now pre-baked into the warehouse UI instead of buried. There’s no notification system in the box, but the primitives are clearly there to build one.
Three more Delta-flavored wins. New enhanced metadata sync for the SQL analytics endpoint (preview) targets a 30-second data freshness SLO, so changes in storage show up in the endpoint within half a minute — enabled per workspace via a setting. Configurable retention in Fabric Warehouse (preview) replaces the hard 30-day history window with anything from 1 to 120 days. And Time Travel via the SQL analytics endpoint (preview) extends the OPTION (FOR TIMESTAMP AS OF ...) clause that warehouse users already had out to Lakehouse data — one clause, one timestamp, applied consistently across the whole query. That one prompted a brief on-air debate, and the resolution is worth repeating: Time Travel has been in Warehouse for a while; what’s new is reaching Lakehouse data through the endpoint.
Real-Time Intelligence: Change Feeds, an AI-First Tile Editor, and Live Refresh GA
Mirrored database change feeds can now stream straight into Eventstreams (preview). Any mirrored database with Delta Change Data Feed enabled through Extended Capabilities becomes a source for row-level inserts, updates, and deletes — Azure SQL, Cosmos DB, Oracle, PostgreSQL, Snowflake, and open mirroring partners included. No custom Spark notebooks polling for incremental updates. And because Activator works on any streaming data, alerting on those changes comes along for the ride.
If you’ve opened a Real-Time Dashboard tile editor recently and done a double take, that’s the redesigned AI-first editing experience (preview). Pick a visual type, describe what you want in natural language, and Copilot generates the visualization for review — no KQL expertise required — with a per-visual chat history you can revert to and the option to fine-tune the KQL manually afterward. Worth the licensing callout the hosts made: this is Copilot inside Fabric, so it needs a paid capacity, not a trial. Using Fabric skills from GitHub Copilot in VS Code to author dashboards — which is how John does most of his authoring, deploying straight out to the service — runs on your own tokens and works on a trial.
Time Series Visualization (preview) is a purpose-built visual for multivariate time series: legend search, next/prev navigation through logical groups, pinning and overlaying series for comparison, multiple panels with synchronized time frames, flexible Y-axis scaling, crosshairs, and log scales. Share Copilot Exploration Insights lets you hand colleagues a link to the query and visual you just discovered, and they can run it, adjust it, or save it to a KQL queryset.
Real-Time Dashboards Live Refresh is now generally available, and it’s a genuine model change rather than a faster timer. Instead of polling on a schedule, lightweight background queries detect when new data is ingested and refresh only the affected visuals — the closest Power BI analogy being change detection rather than automatic page refresh. Quiet periods cost you nothing, and viewers can pause refresh while they investigate a data point.
Eventstream observability through workspace monitoring (preview) adds three tables to your monitoring Eventhouse: node status, throughput metrics, and error metrics, all queryable in KQL with zero configuration beyond the workspace toggle. More signals is good. The standing gripe stands too: workspace monitoring still has to be turned on workspace by workspace, and monitoring a platform from inside that same platform leaves an obvious gap when things go sideways. A dedicated monitoring capacity helps, but a single point of failure for all monitoring is a single point of failure no matter where it lives.
Business Events Get a Coming-Out Party
Business Events in Fabric is a unified pub-sub model that turns raw signals into governed, structured business events other teams can discover, consume, and act on. Eventstream becomes a publisher (preview), applying filtering, enrichment, and correlation to turn low-level changes like CDC rows into events such as OrderCreated or HighValueOrderDetected. Activator becomes a publisher too (preview), so a condition detected in a Power BI report, a Real-Time Dashboard, a KQL query, or a warehouse SQL query can emit a governed event. Published events land in Eventhouse automatically, each mapped to its own KQL table, so historical analysis and Real-Time Dashboards come free — land it in Eventhouse and you can visualize it.
Most of Business Events is in preview. The piece that’s generally available, naturally, is the one that explains how you’ll be charged. Capacity consumption works on two operation types: event operations charged per event (publish charged to the Event Schema Set item, filtering and delivery to the consumer’s capacity) and an event listener charged per hour to the consumer. For what it’s worth, real-time in Fabric is relatively cheap compared to other platforms — but depending on how you build your eventstreams, the meter can still move faster than you expect.
Network governance shows up across the rest of the section. Azure Blob Storage events now work under a tenant-level private links policy by allowlisting the Azure Event Grid system topic as a trusted resource through Resource Instance Rules — the same mechanism that already controls inbound access to OneLake — rather than relaxing the policy. Azure and Fabric events support workspace outbound access protection, so cross-workspace consumption is blocked by default until a Real-Time Events connector is added to the workspace’s data connection rules. And the MQTT source connector picks up custom CA and mTLS support (preview) using certificates from Azure Key Vault, which matters for anyone pulling IoT data off brokers that only accept trusted certificates.
The enhanced Azure IoT Hub source connector (preview), built on Kafka Connect, finally preserves event metadata — system properties and user-defined application properties alike — by copying them into the user metadata section, with system keys prefixed so you can tell them apart. That unlocks per-device dashboards, ingestion latency calculations, and routing based on device identity. Eventstream connectors for Apache Kafka and Azure Service Bus went generally available, with the Kafka connector adding SASL_SSL plus custom CA and mTLS. Kafka is very much a first-class citizen in Fabric now, and more of the platform is being built on it over time. Rounding out the section, the HTTP connector supports pagination (preview) — page-based and cursor-based — which is exactly the gap that used to force custom code. John’s weather sensor pipeline went Power Automate flow, then Logic App, then a custom Azure Function, and finally landed on the Eventstream HTTP connector calling the sensors every minute. Small payloads, so he doesn’t need pagination, but he loves that it’s there.
Databases Apps: Project Rayfin Arrives
New section, new naming puzzle. Fabric Apps (preview) are applications built on top of Fabric databases — SQL and Cosmos DB today, with Horizon on the way — using the open-source Rayfin SDK and CLI. You get an enterprise-grade backend with a built-in database, authentication, and front-end hosting, integrated with existing Fabric data sources starting with semantic models, and your schema, services, and access policies are defined in code so an agent has the context to build the app logic for you. Answering the oldest question in databases — “great, now how do I get a form in front of this thing?” — without reaching outside the platform.
A naming tip from hands-on experience: the blog post heads this up as Databases Apps and the project is called Rayfin, but when talking to GitHub Copilot CLI, “Fabric apps” is the term that gets you where you’re going — ask about Rayfin and it’ll politely check whether you mean a Fabric app. Jason got it spun up locally using the local dev model, which requires Docker, and tried to port an existing anonymous session-feedback form into a Rayfin app. The sticking point: Rayfin apps require authentication, so an anonymous form isn’t a straight swap, and the service isn’t live yet in every data center. Alex Powers has already built a Rayfin app on top of the Fabric roadmap that looks considerably better than the plain site — a good preview of where this goes. Watch the space.
Data Factory: CDC Grows Up, and Pipelines Learn to Wait for a Human
CDC with SQL estates in Copy job is generally available, covering Azure SQL Database, SQL Server, and Azure SQL MI, with Copy job detecting CDC-enabled tables and applying inserts, updates, and deletes downstream without custom incremental logic. Extended SCD Type 2 support in Copy job now reaches Fabric Warehouse (preview), preserving change history with effective dating and soft deletes — the transaction-log pattern that gives you dimension history over time. Auto partition support extends to Oracle, Fabric Lakehouse, and SAP HANA (preview), parallelizing large reads and writes instead of grinding through them serially. Copy jobs can also be edited directly via JSON payloads, which is the Data Factory habit of getting under the hood finally opening up in Fabric — think Power Query’s Advanced Editor versus the wizard. And SAP data can be integrated through Copy job with an ABAP Add-On (preview), a plug-in installed on the SAP server for extraction scenarios the standard APIs can’t handle.
The eyebrow-raiser is business workflow management in Fabric Pipelines, headlined by a new Approval activity (preview). Pipeline execution can now pause and wait for a human decision — an Outlook-based approval request, approve or reject, with branches on each side — before continuing. The use cases are real: approve before promoting data or models to production, gate a deletion that’s about to remove more than it should, handle exceptions that need eyes on them. The jury is out on whether pipelines are the right home for it, though. Power Automate has owned business workflow while pipelines owned data workflow, and this blurs the line. The counterargument is reasonable: if the bulk of the work is data, you don’t want to stand up a separate tool for one approval step; if the bulk of the work is business process, stay in a flow. Just be careful about introducing a second approval experience alongside the one your users already know.
Two new activities round out the orchestration story: Refresh Materialized Lake View (preview), where Fabric decides whether a full, incremental, or no refresh is needed, and Refresh SQL Endpoint, now generally available. The modern pipeline node experience (preview) adds color and context to the canvas — you can see at a glance where success and failure paths lead instead of squinting at a check mark and an X — though the blog notes the new UI slipped to end of July. Conditional activity retries (preview) are quietly valuable: define retry behavior by error code, error message, or both, with AND/OR logic, so transient failures retry and catastrophic ones fail fast instead of burning compute.
Variable Libraries keep spreading, as they should — connection and item reference support in pipelines, plus Variable Library support in Apache Airflow jobs, so DAGs resolve environment-specific values at runtime instead of being edited per environment. On the connectivity side, workspace identity authentication for SharePoint arrives as Azure ACS is retired, which means shortcuts and connections to SharePoint lists can stop leaning on user identities. Snowflake adds secret-less authentication using Fabric workspace identity (Snowflake is deprecating username/password anyway), the Google BigQuery V2 connector moves from ODBC to ADBC for better throughput, and mirroring network security controls expand to Azure SQL Database, SAP, SQL Server 2016–2022, and SharePoint for workspaces restricted to selected networks and private links.
That’s a Wrap on June
Two episodes, one feature summary, and a July update already queued up to do it all over again in a couple of weeks. The Power BI announcements that stacked up in the meantime are on the list for an upcoming show. Thanks to everyone who said hello in Copenhagen — and apologies for any lobby noise. That’s what happens when you record at a conference.
Links
Microsoft Blog Posts & Documentation
- Fabric June 2026 Feature Summary
- Time Travel in Fabric Data Warehouse
- Configurable Data Retention in Fabric Warehouse
- SQL Analytics Endpoint Metadata Sync
- Extended Capabilities in Mirroring (Change Data Feed)
- Real-Time Dashboards – Create and Edit Tiles
- Monitor Eventstream Items with Workspace Monitoring
- Analyze Business Events in Eventhouse
- HTTP Source in Fabric Eventstream
- Fabric Apps and the Rayfin SDK
- Change Data Capture (CDC) in Copy Job
- Approval Activity in Fabric Pipelines
Previous Episodes
- Episode 327 – Microsoft Fabric June 2026 Feature Summary (Part 1)
- Episode 326 – Microsoft Fabric & Power BI April 2026 Feature Summary
- Episode 322 – FabCon Atlanta 2026 Keynote Recap
- Episode 321 – Microsoft Fabric February 2026 Feature Summary
Subscribe: SoundCloud | iTunes | Spotify | TuneIn | Amazon Music

