Every classified article, newest first. Filters combine.
<p><strong>Schema-aware Eventstreams</strong> (Preview) let a single Eventstream recognise several event types while data is in motion, instead of treating a mixed feed as one wide, sparse schema. The data preview groups live events by type. You can inspect payload, headers and metadata, expand nested JSON, and export samples before building any processing logic.</p><p>Three maturity levels coexist in one stream:</p><ul><li><strong>Schematized events</strong> are CloudEvents whose <strong>ce_type</strong> header references a schema registered in <strong>Event SchemaSet</strong> (GA) or the Fabric Schema Registry.</li><li><strong>Classified events</strong> are split by a field you choose, such as deviceType or eventType, through the new Classifier workflow. They do not become registered contracts.</li><li><strong>Untyped events</strong> stay visible without any declared structure. Events with a null or missing classifier value also stay untyped.</li></ul><p>Supported operators such as Filter, and destinations such as Eventhouse, Lakehouse, custom endpoints and derived streams, can take one event type as input. They then see only that type's fields. One important behavior: events from schema-associated sources that fail validation are <strong>dropped</strong> and reported only in runtime logs, not routed as untyped. The post gives no cost or throughput figures.</p>
Preview RTI feature with useful behavior details on schema routing, validation drops.
<p><strong>Custom visuals with Copilot</strong> are in preview for Real-Time Dashboards. In Editing mode you choose <strong>Add visual > Describe a visual</strong> and describe the operational view you need. Copilot then generates a KQL query plus an <strong>HTML-based visualization</strong> and adds it as a new tile, for example entrances laid out spatially with congestion colouring, or a service-dependency health graph.</p><p>You refine the result iteratively in tile edit mode through the Copilot pane, adjusting the query or the visual, then apply and save. The setup requires a configured identity and a cloud connection on the dashboard's data sources. The post gives no limits, capacity cost, security model for the generated HTML, or accuracy guidance beyond advice to validate the output.</p>
Bare preview of Copilot-generated RTI visuals; no limits, cost or governance detail.
SQLBI
Oct 22, 2026
3/5 · Tier B
AI ReadinessSemantic Model
announcement
<p>The <strong>DAX Lib MCP server 0.1.0</strong> connects any local (stdio) MCP client, such as Claude, Codex, GitHub Copilot in VS Code or Cursor, to the DAX Lib catalog of <strong>DAX user-defined functions</strong> and to the Power BI Desktop models open on the same machine. Through the assistant you can search packages, read READMEs and versions, list open models and install a package's functions into a chosen model.</p><p>The server downloads the package and writes the original function code (formatting and comments intact) directly into the model. The assistant does not retype it. Functions with the same name are replaced and nothing else in the model changes. The file is not saved for you. The install tool is flagged as data-changing, so most clients ask for approval first. The server is MIT open source with no telemetry.</p><p>Limits:</p><ul><li>It runs on Windows x64 only and needs Node.js 18 or later.</li><li>It targets local Desktop models only.</li><li>The model needs <strong>compatibility level 1702 or higher</strong>, and the server will not raise it for you.</li><li>It cannot remove or update functions. You update by reinstalling the newer version.</li><li>If you omit the version, you get the latest one, which may be a prerelease.</li></ul>
Practical MCP tooling for installing DAX UDF libraries; clear limits, niche impact.
<p><strong>Incremental migration</strong> for Fabric Data Warehouse (Preview) lets Migration Assistant run repeatedly against the same destination Warehouse. Each run migrates a selected subset of object definitions, not every source object. Sources can be SQL Server, Synapse dedicated SQL pools or Teradata. The destination can be empty, previously migrated, or a Warehouse built by other means.</p><p>From the Warehouse <strong>Migration tab</strong>, <strong>Get more objects</strong> starts a new run. The selection dialog flags <strong>naming conflicts</strong> and auto-includes <strong>direct dependencies</strong>, so the final scope can be larger than what you picked. Non-conflicting objects are created immediately. Conflicting ones stay pending until you resolve each one in a side-by-side SQL diff:</p><ul><li><strong>Keep existing</strong> — retain the destination definition.</li><li><strong>Keep incoming</strong> — replace it with the new translated definition.</li><li><strong>Rename incoming</strong> — create the new object under a different name.</li></ul><p>Key limits:</p><ul><li>It migrates definitions only. There is no continuous row sync, and data loading is a separate step.</li><li>Replacing a table can force a data reload and revalidation of dependents.</li><li>Consolidating several sources into one Warehouse does not reconcile definitions or merge data.</li></ul>
Phased warehouse migration with conflict resolution; preview with concrete behavior and limits.
<p><strong>Fabric Migration Assistant</strong> for Data Warehouse now accepts Teradata sources in preview. You upload a ZIP of <strong>.sql</strong> and <strong>.bteq</strong> files, choose the objects in scope, and the tool translates them to Fabric T-SQL. It deploys eligible:</p><ul><li>tables</li><li>views</li><li>stored procedures</li><li>SQL functions</li><li>macros, which become stored procedures</li></ul><p>Volatile tables map to local temp tables. Untranslatable scripts land in <strong>Objects to fix</strong>, with source, translation comments and errors. Copilot <strong>Fix query errors</strong> can suggest corrections where it is enabled.</p><p>For <strong>BTEQ</strong>, only SQL and supported control flow become stored procedures. Imports, exports, scheduling and external operations must be rebuilt with <strong>COPY INTO</strong>, pipelines or other services.</p><p>Data is not moved by the tool. Three loading options are offered:</p><ul><li><strong>Copy job</strong> from the Copy data step: full copy, plus incremental where the source supports it.</li><li><strong>ADLS Gen2 staging plus COPY INTO</strong> for large historical loads.</li><li><strong>Pipelines</strong> for orchestrated or custom-recovery loads.</li></ul><p>The guidance separates three outcomes: translation, deployment and validated behavior.</p>
Teradata source support with concrete mappings, BTEQ handling and data-movement options.
<p><strong>Fabric Notebook Toolkit (FNTK)</strong> is a preview CLI that lets coding agents work against live Fabric notebooks. Supported agents include Claude Code, GitHub Copilot CLI and Codex. The agent can do three things:</p><ul><li>Discover workspaces, notebooks and lakehouses.</li><li>Read and edit cells.</li><li>Execute a single cell in a persistent session, then iterate without resubmitting the whole notebook.</li></ul><p>Diagnostics are exposed through <strong>fntk monitor</strong> and <strong>fntk logs</strong>. These return cell output, Spark job state, and driver, executor and Livy logs as structured JSON with suggested next actions.</p><p>Other command groups:</p><ul><li><strong>init</strong> installs agent workflow skills.</li><li><strong>notebook</strong>, <strong>cell</strong> and <strong>code</strong> manage notebooks, cells and execution.</li><li><strong>resource</strong> manages notebook resource files.</li><li><strong>server</strong> runs a local server.</li></ul><p>Execution runs on Fabric capacity. A Microsoft data team reports pairing FNTK with an Azure DevOps MCP server for user-story context. The post gives no cost, security-model or limits detail.</p>
Agent-driven live notebook execution and log access; useful but preview without cost details.
<p>Eventstreams Q3 2026 broadens Fabric's streaming layer from connectivity toward in-flight processing and operations.</p><p><strong>Generally available:</strong></p><ul><li>Connectors for SAP Datasphere, Cribl, Solace PubSub+ and MQTT.</li><li><strong>Mirrored Database Change Feed</strong> and <strong>MongoDB CDC</strong>, both feeding DeltaFlow transformations.</li><li><strong>Schema Registry and Event Schema Sets</strong>: versioned Avro contracts.</li><li>Activator rule management inside Eventstream.</li></ul><p><strong>In preview:</strong></p><ul><li><strong>Custom Stream Connector</strong>, which brings your own Kafka Connect source connectors.</li><li><strong>Schema-aware Eventstreams</strong>: multiple event types in one stream, classified by registered schema or a discriminator field.</li><li><strong>Reference Data with Delta Table</strong>: INNER and LEFT OUTER joins against lakehouse or warehouse tables, including shortcuts.</li><li>Workspace Monitoring and diagnostic logs.</li><li>Capacity Operation Events as a stream source.</li><li>Workspace-level Private Link.</li><li><strong>Variable Library</strong> support for environment promotion.</li></ul>
Quarterly RTI roundup: GA connectors, schema registry, reference joins, monitoring, Variable Library support.
<p><strong>GQL</strong> in Microsoft Fabric Graph gained a batch of capabilities between January and August 2026:</p><ul><li><strong>Traversal control:</strong> ANY SHORTEST paths, quantified edge patterns with per-step filters, path modes (WALK, TRAIL, SIMPLE, ACYCLIC, DIFFERENT EDGES) and unbounded variable-length paths.</li><li><strong>Path inspection:</strong> NODES, EDGES, INNER_NODES, ELEMENT_ID, SOURCE and DESTINATION.</li><li><strong>Query composition:</strong> NEXT for multistage queries, EXISTS and CALL/OPTIONAL CALL subqueries, and UNION DISTINCT.</li><li><strong>Aggregation:</strong> filtered and paged aggregates, nested aggregates and COLLECT_* functions.</li><li><strong>Values and text:</strong> dynamic nested data, CAST, RANGE, POWER, Unicode NORMALIZE and MSFT.REGEXP functions.</li></ul><p>The guidance is to start from selective nodes and prefer bounded traversals. Unbounded paths are not unlimited compute and need testing at realistic graph density. The post includes no cost or capacity figures.</p>
GQL syntax roundup for Fabric Graph; niche workload outside my priority areas.
<p>Fabric Data Warehouse adds four T-SQL language features in preview:</p><ul><li><strong>GROUP BY ALL</strong> groups by every non-aggregate expression in the SELECT list.</li><li><strong>ORDER BY ALL</strong> sorts by the SELECT columns in their listed order.</li><li><strong>QUALIFY</strong> filters rows after window functions are evaluated, removing the wrapper CTE for top-N and dedup patterns.</li><li><strong>FROM ... SELECT</strong> lets a query start with its source and define the selected columns afterward.</li></ul><p>The post includes no performance, limitations or SQL analytics endpoint support details. The goal is shorter, more maintainable queries.</p>
Bare preview of T-SQL syntax sugar; useful but no behavior or limits.
<p>Fabric Data Warehouse and the SQL analytics endpoint add new analytical functions in preview:</p><ul><li><strong>MEDIAN</strong> and <strong>QUANTILE</strong> for exact results.</li><li><strong>APPROX_MEDIAN</strong> and <strong>APPROX_QUANTILE</strong> for faster estimates on large datasets.</li><li><strong>Aggregate syntax</strong> for the existing PERCENTILE_CONT and PERCENTILE_DISC, alongside their window forms.</li></ul><p>Each function works as an aggregate or as a window function. The post does not give approximation error bounds or performance numbers.</p>
Preview MEDIAN/QUANTILE functions; handy for statistics but no accuracy or cost details.
Tabular Editor
Oct 20, 2026
3/5 · Tier B
AI ReadinessSemantic Model
tutorial
<p><strong>Agent skills</strong> are folders containing a <strong>SKILL.md</strong> Markdown file, plus optional references, scripts and assets folders. They follow an open standard that originated at Anthropic. The YAML <strong>frontmatter</strong> (name, description, optional allowed-tools) is always loaded into the agent's context. The instructions load only when the skill is invoked, either by a slash command or automatically when the description matches the task. Reference files load only when the instructions point the agent to them.</p><p>This <strong>progressive disclosure</strong> keeps always-loaded context small, which cuts token cost and protects output quality. A vague description means the agent never loads the skill. An overpromising one means it loads on the wrong tasks and steers the agent off course. Telling an agent to read a skill file is not the same as installing the skill, because you lose the harness integration and progressive loading.</p><p>Skills supply context, not tools, which sets them apart from <strong>MCP servers</strong>. They should encode business definitions and processes that agents cannot infer, not technical detail they can already figure out. Write them iteratively: run the task, record where the agent failed, then refine. Cut prescriptive text that confuses more than it helps. Version skills in Git next to the model code so they do not drift.</p>
Practical guidance on agent skills for semantic model work; no Fabric limits or cost.
<p>Fabric Data Warehouse can call <strong>Fabric User Data Functions</strong> in preview. You define a T-SQL <strong>external function</strong> with CREATE FUNCTION ... AS EXTERNAL FUNCTION that maps to a published Python function. You then call it from SELECT, INSERT or UPDATE, and each call is delegated to the UDF runtime.</p><p>Suggested uses include:</p><ul><li>Calling APIs and sending email or Teams messages.</li><li>Parsing files and reading or writing other Fabric stores.</li><li>Retrieving secrets from Key Vault or variable libraries.</li><li>Custom encryption or masking, and enrichment such as IBAN validation or ML scoring.</li></ul><p>The example derives IPv6 /48 prefixes with Python's ipaddress module inside an UPDATE.</p><p>The post does not state throughput, batching, CU cost or failure semantics.</p>
Preview: T-SQL external functions calling Python UDFs; no cost or limit details.
<p>A custom Date table can now behave like the hidden <strong>auto date/time</strong> table, keeping a <strong>continuous X-axis</strong> in line charts at year, quarter, month or week level. Month and quarter labels can stay as sorted text columns. The behavior comes from column metadata, not from special treatment of Power BI's own tables.</p><p>Four conditions must be met:</p><ul><li>The visual uses levels of a <strong>user hierarchy</strong>. Loose columns are not enough.</li><li>The lowest level is a date column with <strong>dataCategory: PaddedDateTableDates</strong>. This property is undocumented but required.</li><li>The top level has <strong>dataCategory: Years</strong>.</li><li>The visual includes levels starting from that Years level.</li></ul><p>The Date column must also be <strong>isKey</strong>, which marking the table as a date table sets. The table-level dataCategory: Time is ignored. Intermediate levels can be any date-categorized column.</p><p>Power BI Desktop's Data category dropdown does not offer these values. You have to set them in <strong>TMDL view</strong>, <strong>Tabular Editor</strong> or the <strong>Power BI Modeling MCP server</strong>. One cosmetic side effect: tooltips repeat the year when a level such as Year Month already contains it. Other visuals, slicers and filters behave as before.</p>
Concrete TMDL metadata recipe giving any Date table auto date/time's continuous axis.
Fabric Updates
Oct 19, 2026
4/5 · Tier A
AI ReadinessBreaking ChangeMigration
announcement
<p>The <strong>custom-connection integration</strong> that let Microsoft Foundry agents call a Fabric data agent as a custom tool is deprecated and stops being supported on <strong>6 November 2026</strong>. The replacement is the <strong>Fabric IQ (OneLake Catalog) tool</strong>, currently in Preview.</p><p>With the new integration, published data agents appear by name in the OneLake Catalog inside the Foundry portal, so no workspace or artifact IDs are needed. A single Foundry agent can call <strong>multiple data agents</strong>, where the old integration allowed only one. Each data agent is exposed as a tool through an <strong>MCP endpoint</strong>, and responses will soon include the generated query.</p><p>Migration runs either through the portal (add the Fabric IQ tool, pick the agent, rewrite instructions, remove the old tool) or through code. The code path creates a Foundry connection via ARM REST with <strong>authType UserEntraToken</strong>, category RemoteTool, the data agent MCP URL as target and audience api.fabric.microsoft.com. It uses <strong>isSharedToAll true</strong>, then references the full connection resource ID from the SDK. Calls run as the signed-in user, so Fabric permissions apply per user.</p>
Hard retirement deadline 11/6/2026 for Foundry custom-tool data agent integration; migration steps.
Fabric Updates
Oct 19, 2026
4/5 · Tier A
AI ReadinessSemantic Model
announcement
<p><strong>Advanced NL2SQL</strong> and <strong>advanced DAX generation</strong> in Fabric data agents are now GA. The core improvements roll into the standard runtime between <strong>2 and 10 November</strong>. Instead of generating a query in one pass, the agent uses sub-tools to explore the schema, test joins and look up real filter values (for example "California" mapping to "CA") before writing SQL or DAX. Each step and assumption is visible in the run steps.</p><p>Advanced NL2SQL works in three modes: <strong>example-query following</strong> (adapting an approved example), <strong>clarifying questions</strong>, and <strong>data exploration</strong>. Two new configurations apply to SQL sources on the preview runtime:</p><ul><li><strong>Schema descriptions</strong> add business meaning to tables and columns.</li><li><strong>Topics</strong> hold subject-specific Markdown guidance that the agent retrieves only when it is relevant, unlike data source instructions, which are always sent.</li></ul><p>On Microsoft's internal benchmark, smart-answer accuracy rose from 78.8% to 90.9%. P50 latency fell from 28.4s to 26.0s and P95 from 46.8s to 32.9s. Advanced DAX uses <strong>instance value indexing</strong> over real column values and aims for consistent answers across the data agent, Power BI and M365 Copilot.</p>
GA with benchmark accuracy/latency numbers and runtime rollout dates requiring agent re-testing.
<p><strong>Full-text indexing</strong> (September Power BI release) is enabled per text column and speeds up text search and filtering. It also powers the new <strong>TextContains()</strong> and <strong>TextSimilarity()</strong> DAX functions. The index rebuilds during a <strong>Full</strong> refresh or a refresh of type <strong>Indexes</strong>.</p><p>The build is visible in a Profiler trace as a <strong>Progress Report Begin/End</strong> pair with subclass <strong>Tabular Object Processing</strong>. Its TextData reads "Finished building persisted full-text index for column..." and the Duration and CPUTime columns give the cost.</p><p>The test case was a Street column with 73,206 distinct values in a 160,257-row table, built in 1,392 ms using 484 ms of CPU. Longer strings and higher cardinality will cost more. The docs warn that indexing too many text columns can significantly slow refresh.</p><p>Power BI reports cannot use these indexes yet. They are aimed at arbitrary-DAX front ends and "chat with your data" agents, for example finding feedback text that mentions air conditioning.</p>
Measured full-text index build cost in refresh; Profiler events to monitor it.
<p>7 features - 1 must-know - 4 worth knowing.</p><ul><li>More accurate surge protection with semantic models</li></ul>
1 must-know of 7 features
<p><strong>Fabric Policies</strong> are a preview governance layer that defines what actions are allowed, to whom and where, scoped to the tenant or to an individual capacity. Policies are grouped into <strong>Policy Sets</strong>. Only one Policy Set can be active per tenant or per capacity, and activating a replacement swaps the governing definitions.</p><p>Three areas are governed today:</p><ul><li><strong>Item creation by capacity</strong>: any Fabric item type, but Power BI items are not yet enforced.</li><li><strong>Security-sensitive workspace settings</strong>: outbound access protection, inbound access protection and CMK.</li><li><strong>External data sharing</strong>: coming soon, with conditions on sharer, source workspace, sensitivity label and recipient domain.</li></ul><p>Each policy is <strong>Allow all</strong>, <strong>Block all</strong> or <strong>Advanced</strong>. Advanced uses an allow-based model. All conditions in a rule must match, a match on any complete rule allows the action, and anything unmatched is blocked.</p><p>Policies never grant roles or permissions. They add a governance decision on top of existing access. Item creation must pass both the tenant item creation setting and the active policy. Editing a Policy Set follows workspace roles, while activation requires Fabric admin (tenant scope) or capacity admin (capacity scope). Management is available through Policies Center in the OneLake Catalog, the M365 Audit Log and REST APIs.</p>
Capacity-scoped item creation control changes rollout governance; preview with clear allow-based semantics.
<p>The <strong>Databricks ADBC driver</strong> is GA for the Databricks and Azure Databricks Power Query connectors. It replaces the embedded <strong>Simba Spark ODBC</strong> driver. The Power BI service will move from ODBC to ADBC through Q1 2027, so opt-in now means mandatory later.</p><p>ADBC is Arrow-native and fetches large result sets with minimal serialization or copying. It supports OAuth (user), OAuth M2M and Entra ID authentication.</p><p>There are three opt-in levels:</p><ul><li><strong>Per connection</strong>: add <strong>Implementation="2.0"</strong> to the Databricks.Catalogs source step. Use "1.0" to force ODBC for side-by-side comparison.</li><li><strong>Per workspace</strong>: a workspace admin override.</li><li><strong>Tenant-wide</strong>: the admin setting "Users can connect to data sources by using Apache Arrow database connectivity (ADBC)".</li></ul><p>Existing semantic models, dataflows and paginated reports remain unchanged until switched. Separately installed ODBC drivers are unaffected.</p>
Forced ODBC-to-ADBC driver switch for Databricks connectors by Q1 2027; validate now.
<p><strong>SQL database in Fabric</strong> now works in preview inside workspaces secured with <strong>workspace-level private link</strong>. The workspace maps to a specific VNet through Azure Private Link. When inbound public access is restricted, the workspace is reachable only through the approved private endpoint, while other workspaces in the tenant stay on public access.</p><p>Setup has two parts. First, create the Azure Private Link resources. Then, in Workspace settings > Inbound network, choose <strong>Allow connections from selected networks and workspace-level private links</strong>.</p><p>The post gives no limitations, performance notes or cost details beyond the standard private endpoint setup.</p>
Bare preview: SQL database now works in workspace-level private link workspaces.
<p><strong>File data ingestion with Materialized Lake Views</strong> is available in preview. A view can read CSV or Parquet files directly from a OneLake <strong>Files</strong> folder using <code>USING OneLake_Files</code>, materialize the result as a Delta table, and feed downstream bronze, silver, and gold views.</p><p>Recursive folder discovery, <code>__filepath__</code> source traceability, CSV header and delimiter options, and dependency-ordered refresh are built in. <strong>DYNAMIC</strong> schema mode accommodates evolving landing files, while <strong>FIXED</strong> mode rejects schema drift and protects downstream contracts. Current scope is limited to CSV and Parquet, and the capability remains preview.</p><p>Data quality constraints can be applied in downstream views, while lineage exposes the source folder, discovered files, processed-file counts, and added rows. A full refresh supports reprocessing an entire folder when incremental append behavior is insufficient.</p>
Substantive preview adds declarative CSV and Parquet ingestion, schema controls, lineage, quality rules, and refresh behavior.
<p><strong>Amazon Redshift connectors</strong> across Microsoft data products are transitioning to the Amazon Redshift ODBC 2.x driver. The change rolls out automatically by region and adds external-object metadata visibility, Redshift Data Sharing reads, newer Redshift feature support, and connection auditing.</p><p>The transition affects Fabric Copy jobs, Copy pipelines, Dataflows Gen2, semantic models, Power BI Desktop, gateways, and related integrations. Customers wanting earlier testing can enable preview features, but the extract provides no detailed compatibility matrix or rollback procedure.</p>
Automatic Redshift driver migration improves connectivity but requires validation across Fabric, Power BI, and gateway workloads.
<p><strong>Fabric App functions</strong> add TypeScript-based server-side logic to Fabric Apps without separately provisioning backend infrastructure. Functions can execute business rules, call external APIs, access protected resources, orchestrate AI workflows, and manage secrets outside the browser.</p><p>Application authentication lets a function access downstream services as the application rather than the signed-in user. Local debugging uses Azure Functions Core Tools, while deployment packages the app and functions together. The extract does not specify preview limitations, quotas, or separate execution costs.</p>
Preview server-side functions add secure identity, secrets, integrations, and AI orchestration to Fabric Apps.
<p><strong>My workspace governance controls</strong> are generally available, including in government clouds. Administrators can now grant a 24-hour temporary access window to a user’s personal workspace or revoke it early through REST APIs.</p><p>Fabric administrators or authorized service principals can call these APIs, enabling repeatable investigation, retention, and recovery workflows instead of manual Admin Portal operations. Deleted users’ orphaned My workspaces remain recoverable for 30 days; restoration requires a new workspace name and a newly assigned administrator.</p>
GA APIs make temporary My workspace administration programmable, including service-principal automation and orphaned workspace recovery.
<p><strong>CI/CD for report development</strong> is presented as an incremental maturity ladder rather than an enterprise-only practice. Moving from PBIX files toward PBIP, TMDL, Git, separate environments, and automated checks makes changes smaller, reviewable, reversible, and safer for human or AI-assisted development.</p><p>Deployment pipelines move content between workspaces but do not, by themselves, provide source history or configurable quality gates. Stronger practices add best-practice scans, measure validation, visual checks, regression testing, and deployment failures when quality checks do not pass.</p>
The maturity ladder connects PBIP, Git, environments, automated validation, and deployment practices to safer report delivery.
<p>8 features - 4 must-know - 4 worth knowing.</p><ul><li>Find the operations driving capacity pressure (Preview)</li><li>Turn business meaning into a governed event (Generally Available)</li><li>Process, act, and analyze without changing the publisher</li><li>Monitor capacity health as events occur (Generally Available)</li></ul>
4 must-know of 8 features
<p>The <strong>Capacity Metrics app</strong> preview adds redesigned health, compute, storage, and item-detail experiences for diagnosing capacity pressure and throttling.</p><p>Administrators can compare throttling time across capacities, inspect hourly utilization patterns, and identify the items contributing most to a selected period. Compute usage is shown in <strong>CU seconds</strong> alongside throttling state, with smoothed consumption highlighting the highest-impact items.</p><p>The item-details page adds a 14-day history of an individual item's capacity impact. Because metric behavior has changed, the preview installs alongside the existing generally available app rather than replacing it immediately.</p>
Preview metrics materially improve capacity diagnosis, throttling analysis, workload attribution, and operational capacity management.
<p><strong>Custom SQL Pools</strong> are generally available in Fabric Data Warehouse, giving administrators explicit control over how compute resources are allocated among workloads.</p><p>Separate pools can protect ingestion, reporting, and ad hoc analysis from one another by applying maximum resource-allocation percentages. Statement-type classification separates distributed <strong>SELECT</strong> requests from <strong>NONSELECT</strong> loading and modification statements; each starts with a 50% maximum, and the combined limits cannot exceed 100%.</p><p>Application-name and regular-expression classifiers remain available when statement type does not reflect the required workload boundary. Custom pools complement, rather than replace, Fabric Warehouse's autonomous workload-management behavior.</p>
General availability adds explicit workload isolation and statement classification for predictable Fabric Warehouse resource allocation.
Tabular Editor
Oct 12, 2026
3/5 · Tier B
AI ReadinessSemantic ModelGovernance
deep-dive
<p><strong>Claude Dashboards</strong> are beta interactive artifacts that use connectors to create HTML dashboards from extracts or live data queries. Fabric data can be reached through Microsoft's preview Power BI MCP server configured as a custom connector.</p><p>Live dashboards require a remotely hosted connector, an Entra app registration, relevant tenant settings, and user permissions. Queries run under the user's identity, while local MCP servers can provide data during authoring but cannot support live queries after publication.</p><p>The article positions Claude Dashboards as a lightweight front end rather than an enterprise replacement for Fabric apps. Semantic-model quality, row-level security, connector governance, and beta-product limitations remain decisive risks for production adoption.</p>
Substantive MCP and semantic-model analysis identifies live-query mechanics, beta limitations, and enterprise governance risks.
<p><strong>APPROXIMATEDISTINCTCOUNT</strong> is not universally faster than <strong>DISTINCTCOUNT</strong>. Direct Lake can apply optimizations to exact distinct counts that the approximate implementation cannot reproduce, particularly for low-cardinality columns.</p><p>In the tested 11.7-million-row taxi model, counting two VendorID values across 85,562 result rows took 188 ms with DISTINCTCOUNT versus 688 ms with APPROXIMATEDISTINCTCOUNT. CPU rose from 110 ms to 547 ms, while memory increased from approximately 8 MB to 410 MB. The approximate function can still be valuable for very high-cardinality columns with millions of distinct values, but its crossover point depends on the data and query shape.</p><p>The implementation is changing as optimizations roll out, so benchmark representative production queries before replacing existing measures.</p>
Measured Direct Lake results show when approximate distinct counts consume more time and memory than exact counts.
<p><strong>fabric-cicd v1.4.0</strong> adds deployment support for Graph Model and Cosmos DB items, alongside dynamic replacement variables in bulk publish mode and Python 3.14 compatibility.</p><p>Deployment behavior becomes safer and more predictable: dynamic variables are validated before deployment, paginated reports publish serially to reduce intermittent failures, and fixes address KQL Queryset cluster resolution, OneLake path bindings, Activator workspace IDs, environment-variable replacement, and parameterized Data Pipeline payloads.</p><p>The release also pins <strong>jsonpath-ng</strong> to avoid an upstream regression that could corrupt payloads during <code>key_value_replace</code> parameterization. An end-to-end deployment tutorial improves adoption, but teams should retest existing pipelines because new item types and variable validation can expose previously hidden configuration errors.</p>
New Fabric item support, dynamic deployment variables, Python compatibility, and deployment reliability materially expand fabric-cicd.
<p><strong>PBIP is generally available</strong> as Microsoft's endorsed text-based Power BI project format for source control, making structured semantic-model and report files preferable to opaque PBIX files in Git and AI-assisted workflows. The recap also describes GA Power BI authoring skills, the authoring MCP server, and Desktop Bridge.</p><p>Fabric Apps are positioned for broader use, including upcoming Power BI Pro and PPU availability, while Fabric SQL databases are included within stated limits. Microsoft is joining the vendor-neutral Apache Ossie effort, adding DAX support and a two-way converter for semantic models, although translated SQL and DAX expressions are not automatically guaranteed to remain equivalent.</p><p>Ontologies can use metrics imported from multiple semantic models, and semantic views are being developed to place business meaning closer to lakehouse data. Fabric Apps and AI-generated reports still introduce quality, security, governance, and data-literacy risks; preview availability and release timing must be verified before planning adoption.</p>
Reports substantive GA, licensing, semantic interoperability, ontology, and AI authoring changes beyond conference promotion.
<p><strong>DAX contracts</strong> define the business rules a measure must satisfy before AI generates or changes its code. They sit between vague model documentation and an implementation algorithm, specifying filter behavior, ranking population, tie handling, blanks, totals, and visual context.</p><p>The practical shift is to prompt AI to discover and formalize these rules first, then generate DAX against the contract. The contract becomes the reviewable artifact: when requirements change, update the contract and verify the measure rather than treating generated code as the source of truth.</p>
Provides a useful AI-era DAX specification practice, but lacks measured platform behavior or implementation detail.
<p><strong>Fabric Influencers Spotlight: September 2026</strong> curates 30 community submissions across Fabric administration, CI/CD, OneLake security, capacity governance, migration, Spark, warehousing, AI, real-time intelligence, and Power BI.</p><p>The collection points to potentially valuable practitioner material, including <strong>fabric-cicd</strong> deployment patterns, Direct Lake security behavior, Liquid Clustering, Workspace Identity, and Databricks interoperability. However, the spotlight itself is an index rather than a technical analysis: implementation limits, costs, benchmarks, and evidence remain in the linked contributions.</p>
Community roundup surfaces 30 Fabric submissions but provides little original technical depth or measured implementation detail.
<p><strong>ApproximateDistinctCount()</strong> is available for Import and Direct Lake semantic models, reducing query duration and CPU for distinct-count workloads. On a 1.4-billion-row Direct Lake fact table, the tested query fell from about 6 seconds to 4 seconds, while CPU decreased from 82 to 48 seconds.</p><p>The speed improvement increases memory use substantially in this test: approximately 206 MB for the exact count versus 1,063 MB for the approximate count. Results were about 1.6% away from the exact values, so the function improves performance by exchanging numerical precision for faster analysis.</p><p>A field parameter provides a practical compromise: default reports to the approximate measure for trend exploration, while allowing users to switch to the exact measure when reported values must be fully accurate.</p>
Measured Direct Lake performance gains expose accuracy and memory trade-offs for expensive distinct-count measures.
fabric.guru
Oct 2, 2026
4/5 · Tier A
AI ReadinessGovernanceAPI Automation
tutorial
<p><strong>Copilot access for semantic models</strong> can now be controlled separately for read-only users. Writers can enable or disable whether a model is discoverable and usable through Microsoft Copilot, report Copilot, and data agents.</p><p>The setting is enabled by default. The article demonstrates applying it from a Fabric notebook with SemPy and a REST client, but relies on an internal endpoint whose authentication and behavior may change. Restricting access prevents unprepared models from producing poorly grounded AI responses while leaving normal report consumption available.</p>
Programmatic Copilot gating improves AI governance, with concrete SemPy automation and an important unstable API warning.
<p><strong>Fabric-RLM</strong> connects a Fabric notebook to one or more Power BI semantic models and lets an LLM inspect schemas, measures, relationships, and period completeness before querying data. Its semantic-model skill can aggregate measures, execute DAX, correct generated-code errors, and record which model fields supported each result.</p><p>The workflow extends beyond conversational answers: a task can produce HTML or Markdown reports and Excel workbooks. Runs can be inspected step by step, including model responses, generated code, outputs, and metrics.</p><p>Accuracy depends on explicitly designed validation. Independent queries can recompute headline and total revenue, reject incomplete comparison periods, verify category names, enforce report structure, and return failed work to the model for repair. The example requires a paid Fabric capacity and uses GPT-5.1 with high reasoning effort, so execution and model costs require platform controls.</p>
Practical notebook automation combines semantic-model querying, LLM reasoning, artifact generation, lineage, and independent validation.
<p><strong>Apache Ossie</strong> defines a vendor-neutral format for exchanging semantic metadata across analytics, BI, and AI platforms. Microsoft and Snowflake are collaborating on converting one Ossie document into both a <strong>Snowflake Semantic View</strong> and a <strong>Power BI semantic model</strong>.</p><p>The integration is intended to preserve shared business meaning without moving or duplicating underlying data. Microsoft also plans to contribute semantic-model expertise, including potential recognition of <strong>DAX</strong> and support for ontologies. Ossie remains an Apache incubating project, so its specification, governance, tooling, and compatibility are still evolving.</p>
Apache Ossie enables portable semantic definitions across Power BI and Snowflake, but provides limited implementation detail.
<p>139 features - 57 must-know - 76 worth knowing.</p><ul><li>Policies in Fabric (Preview)</li><li>Fabric CI/CD - Deployment Plan (Preview)</li><li>Restrict access to sensitive data with data loss prevention policies (Generally Available)</li><li>Bulk export / import API (Generally Available)</li><li>OneLake Outbound Access Protection with external shortcuts (Generally Available)</li><li>Capacity Operation Events in Real-time hub (Preview)</li><li>Build and query ontologies conversationally</li><li>Bind faster and query at source scale</li><li>Apply enterprise security and network controls</li><li>New enhancements for Mapping Data Flow Transforms in Dataflow Gen2 (Preview)</li><li>Upgrade to Fabric Data Factory with confidence</li><li>On-demand billing and Zero-provisioned (F0) Capacity (Preview)</li><li>On-demand billing for Fabric Data Warehouse</li><li>Discover tables and more with the improved OneLake Catalog Search API (Preview)</li><li>Fabric Core MCP Server (Generally Available)</li><li>Branch workspace admin profile (Preview)</li><li>Compare and commit changes (Generally Available)</li><li>Branch out with selective branching (Generally Available)</li><li>OneLake Table Read API (Preview)</li><li>New OneLake security UX with Members and Data Tabs (Generally Available)</li><li>Custom Live Pools for Fabric Data Engineering (Generally Available)</li><li>Spark connector for SQL databases (Generally Available)</li><li>High Concurrency Support for the Fabric Livy API (Generally Available)</li><li>High-concurrency support in the Microsoft ADO.NET driver (Generally Available)</li><li>High-concurrency mode for the Microsoft JDBC driver (Generally Available)</li><li>High-concurrency support in the Microsoft ODBC driver (Generally Available)</li><li>Modernize Teradata Workloads Faster with Migration Assistant (Preview)</li><li>Fabric warehouse Custom SQL pools (Generally Available)</li><li>Distributed Bitmap Filters (Generally Available)</li><li>Automate Every Deployment with Confidence: Introducing Pre/Post Deployment Support (Preview) – Pradeep Srikakolapu</li><li>Automate Fabric Warehouse Lifecycle Management with Item definition support in CRUD APIs</li><li>Business Events (Generally Available)</li><li>Capacity Overview Events in Real-time hub (Generally Available)</li><li>Trigger UDF as an action (Generally Available)</li><li>Stateful conditions in Activator for Event based triggers (Generally Available)</li><li>Fabric IQ MCP Server (Generally Available)</li><li>Microsoft Fabric data agent now supports Model Context Protocol tasks (Generally Available)</li><li>Bring Power BI DAX measures into the Ontology</li><li>Capture business logic with natural-language rules</li><li>Change Event Streaming for SQL database in Microsoft Fabric (Preview)</li><li>Lakehouse maintenance activity (Generally Available)</li><li>Fabric Actions activity (Preview)</li><li>Mirroring for Google Big Query (Generally Available)</li><li>Extended capabilities for Mirroring (Generally Available)</li><li>Change data capture in Copy job (Generally Available)</li><li>SCD Type 2 in Copy job (Generally Available)</li><li>Copy job integration with Eventstream as a source and destination (Preview)</li><li>Copy job supports SQL CDC custom capture instance names (Generally Available)</li><li>Custom staging database and schema for CDC and SCD Type 2 in Copy job (Generally Available)</li><li>CDC from read-only Snowflake databases in Copy job (Generally Available)</li><li>Optimized copy to Lakehouse (Generally Available)</li><li>Expanded support for the modern evaluator in Dataflow Gen2</li><li>Dynamic schema support for the Fabric Data Warehouse destination (Preview)</li><li>New enhancements for upgrading Power BI Dataflows Gen1 to Dataflows Gen2 (CI/CD) (Preview)</li><li>Data connection policies: Authentication allowlist (Generally Available)</li><li>Data connection policies: Tenant allowlist (Generally Available)</li><li>Admin APIs for cloud connections (Generally Available)</li></ul>
57 must-know of 139 features
<p>Fabric adds a guided path for moving <strong>Azure Data Factory</strong> environments into Fabric Data Factory without immediately disrupting production. Existing ADF assets remain the source of truth while teams view pipelines in Fabric, assess activity compatibility, map connections, and upgrade supported pipelines incrementally.</p><p>The <strong>Synapse Spark migration assistant</strong> moves pools, notebooks, and job definitions while mapping lake databases through OneLake shortcuts. Artifacts are copied while data remains in place, allowing existing pipelines to continue operating. OBOS reports migrating more than 600 notebooks and pipelines, consolidating over 4,500 objects into 87 lakehouses, reducing operational costs by 20%, and accelerating processing by 30%.</p>
Provides guided ADF, Synapse Spark, and warehouse migration paths with compatibility assessment and measured customer outcomes.
<p>40 features - 12 must-know - 15 worth knowing.</p><ul><li>Deprecation of Old File Picker experience in Power BI Desktop</li><li>Power BI Projects (PBIP) (Generally Available)</li><li>New Purge Data option for Fabric APIs</li><li>Power BI Agentic Experiences (Generally Available)</li><li>Power BI data answers in Microsoft Copilot Chat (Generally Available)</li><li>Fabric IQ Plugin in Microsoft 365 Copilot Cowork (Generally Available)</li><li>New control for Copilot access to semantic models</li><li>Remote Power BI Authoring MCP Server (Preview)</li><li>Create and edit DirectQuery semantic models in the Power BI service</li><li>Approximate Distinct Count comes to Import and Direct Lake (Preview)</li><li>Accelerate text searches in your semantic models with String Indexing (Preview)</li><li>Introducing Full-Text Search in Power BI Semantic Models (Preview)</li></ul>
12 must-know of 40 features
<p>Fabric adds a broad set of platform capabilities for operating larger estates. Preview observability introduces a Monitor item, shared workspace telemetry in Eventhouse, refreshed Monitor Hub views, public APIs, and operations-agent investigations; Activator can alert on successful as well as failed jobs.</p><p><strong>Deployment plans</strong> provide a visual deployment sequence with YAML as the source of truth, including pre- and post-deployment actions. Bulk item-definition import and export APIs are generally available, with multi-item movement, deployment-order detection, and reference updates supporting workspace-as-code, CI/CD, and migrations.</p><p>Capacity administration expands with preview on-demand compute and zero-provisioned capacity, alongside generally available surge protection, overage, and larger-capacity support. New or expanded OneLake catalog policies, outbound access protection, and FQDN allowlists add governance controls, but preview coverage, quotas, API limits, workload support, and cost behavior require tenant-specific validation.</p>
Official platform update spans observability, deployment-as-code, capacity controls, APIs, and security with significant operational implications.
<p><strong>Fabric IQ</strong> is positioned as the governed business-context layer within Microsoft IQ, combining live operational signals, history, relationships, policies, and available actions. Real-Time Intelligence supplies continuously changing state, while Fabric IQ connects those signals to business meaning so people and agents can reason about operational consequences.</p><p>The article describes an observe-analyze-decide-act loop spanning OneLake, Real-Time Intelligence, governed entities, AI recommendations, and approved actions. It also introduces or previews Fabric Observability Insights, Investigate, and an operations agent performance view, but the extract does not specify detailed limits, pricing, data volumes, or production readiness.</p>
Presents Fabric as an operational AI foundation, but provides limited implementation detail, limits, or cost evidence.
<p>Tabular Editor 3.27.0 adds unsaved-change indicators in the TOM Explorer, using orange for modified, green for added and red for deleted objects. Red and green in the same pane are hard to tell apart for people with deuteranopia, which affects about 1 in 12 men.</p><p>The September 2026 release adds a <strong>Color blindness mode</strong> that switches the green to teal. The post also links to Microsoft's accessible report design guidance.</p>
Tabular Editor product note on color blindness mode; minor tool change.
Fabric Updates
Sep 29, 2026
4/5 · Tier A
AI ReadinessSemantic ModelGovernance
announcement
<p><strong>Fabric Planning</strong> connects budgets, forecasts, targets, and scenarios directly to existing governed semantic models. Planners can connect using their signed-in identity, retain existing permissions, write plan data back into the semantic foundation, and expose planning concepts to Fabric IQ's ontology and agents.</p><p>The preview <strong>Native Planning Engine</strong> uses Rust and Apache Arrow with OneLake integration, claiming up to 10x performance improvements and support for models with tens of millions of cells. Event triggers can start pipelines, Dataflow Gen2, notebook, refresh, Teams, or Outlook actions when writeback starts, succeeds, or fails, with writeback metadata passed as parameters.</p><p>The feature remains a planning workload with preview components and implied scale, governance, and writeback considerations. Organizations must validate calculation behavior, concurrency, permissions, trigger failure handling, and the operational cost of downstream automation before replacing a dedicated planning platform.</p>
Fabric Planning adds semantic-model writeback, ontology integration, scalable calculations, and event-driven automation with concrete preview limits.
<p>9 features - 2 must-know - 7 worth knowing.</p><ul><li>Fabric data connectors</li><li>Built to be governed</li></ul>
2 must-know of 9 features
<p>31 features - 8 must-know - 14 worth knowing.</p><ul><li>On-demand billing brings more flexibility to OneLake compute</li><li>Strengthen data protection with OneLake security</li><li>Connect Salesforce Data 360 to OneLake with bidirectional integration</li><li>Read and write OneLake data from Azure Databricks</li><li>Give agents, applications, and data engines secure access to OneLake tables</li><li>Build operational applications on OneLake</li><li>Manage and govern your data estate in the OneLake catalog</li><li>Apply more precise policies across Fabric</li></ul>
8 must-know of 31 features
<p>9 features - 4 must-know - 3 worth knowing.</p><ul><li>Unifying the data estate: Zero ETL, and now zero excuses</li><li>Transformation in the tools people already use</li><li>Orchestrating data with ease: From low code to pro-code</li><li>Upgrading the integration estate you already have</li></ul>
4 must-know of 9 features
<p><strong>Fabric IQ</strong> is generally available in Microsoft 365 Copilot Cowork and Copilot Chat, grounding answers in existing governed Power BI reports and semantic models rather than a separate AI data copy. Users can reuse established measures, relationships, and certified definitions, while enhanced controls govern which models Copilot may access.</p><p>The release also previews GPU-accelerated Fabric Data Warehouse query execution and introduces early semantic views for defining reusable business meaning directly alongside OneLake data. Fabric IQ MCP is generally available, allowing development environments and external clients to access governed semantic content, while Fabric Apps extend trusted models into applications that can accept input, write back data, and preserve shared state.</p>
Combines Fabric IQ GA, GPU warehouse preview, semantic views, MCP governance, and agentic application capabilities.
<p><strong>Fabric Apps</strong> bring application backends, governed Fabric data, authentication, and deployment into one experience. New capabilities include connectors for warehouses, lakehouses, SQL databases, and semantic models; PostgreSQL transactional storage; TypeScript backend functions; policy-controlled access; and Secret Store integration. Integrated OneLake file storage is identified as coming soon.</p><p><strong>Database Hub</strong> is in preview as a centralized control plane for SQL Server, Azure SQL, PostgreSQL, Cosmos DB, and Fabric SQL databases. Preview database agents for SQL and PostgreSQL analyze workload health, performance, security, availability, and cost signals while operating within RBAC, approval workflows, auditing, and observability controls.</p>
Introduces Fabric Apps, Database Hub, and governed database agents with concrete access, monitoring, and operational guardrails.
Power BI Updates
Sep 29, 2026
3/5 · Tier B
AI ReadinessSemantic ModelReport Authoring
announcement
Power BI Desktop is gaining a natural-language app-building experience powered by Fabric apps. Users can start from a trusted semantic model, generate and edit an application, publish it, accept inputs, write back data, preserve shared state, and support operational workflows.
The preview will be available to Pro, Premium Per User, and Fabric capacity customers. Semantic views aim to define governed metrics closer to OneLake data, while Fabric IQ in Microsoft 365 Copilot Cowork and Chat is generally available. Fabric IQ MCP is also entering general availability for conversational analytics in development environments and external MCP clients.
The announcement provides direction rather than detailed limits, capacity costs, performance characteristics, or governance controls. Apache Ossie and cross-platform semantic conversion remain strategic interoperability efforts.
Introduces natural-language Fabric apps, semantic views, Copilot integrations, and MCP availability, but provides limited implementation detail.
Crossjoin
Sep 27, 2026
2/5 · Tier B
Report AuthoringGovernance
announcement
<p><strong>Power BI Report Builder</strong> now presents Power Query datasets with a dedicated custom-property view. The view exposes the dataset's M code, configured parameters, and the ID of its Fabric cloud connection.</p><p>Copying the connection ID and searching for it in <strong>Manage Connections and Gateways</strong> makes it easier to locate the precise connection, edit it, or refresh credentials. The change improves supportability but does not introduce a new query engine or materially change report execution behavior.</p>
Useful Report Builder usability improvement exposes Power Query connection identifiers, but scope is narrow and operationally focused.
fabric.guru
Sep 27, 2026
3/5 · Tier B
AI ReadinessPySpark
deep-dive
<p><strong>Fabric-RLM</strong> improves deterministic tasks by giving a language model an executable Python workspace, a limited turn loop, and an output contract. Instead of asking the model to reason directly from token patterns, the harness lets it construct a list, run code, inspect the result, and submit a typed answer.</p><p>Across nine small and midsize models, only one answered the weekday-letter question correctly without the harness, while all nine succeeded through Fabric-RLM. The result demonstrates tool-assisted verification rather than increased model knowledge; it does not establish reliability for open-ended generation or complex data work.</p>
Measured model comparison demonstrates how executable tools and guardrails improve deterministic data tasks, though the example is intentionally simple.
fabric.guru
Sep 25, 2026
4/5 · Tier A
AI ReadinessAPI AutomationData Quality
deep-dive
<p><strong>Fabric-RLM</strong> runs recursive language-model tasks beside Fabric data instead of placing entire files into a prompt. The model writes Python, executes it in the notebook environment, inspects results, repairs failures, and returns only outputs matching a typed contract and optional validators.</p><p>Inputs can include Lakehouse files, Delta tables, and semantic models; skills provide reusable procedures for exploration, Excel, PDFs, Delta, and semantic-model work. Each run records generated code, outputs, errors, tokens, timings, and validation repairs, enabling inspection and replay.</p><p>A 140 MB IMF CSV containing 1.57 million rows was processed with DuckDB and openpyxl in 78 seconds, producing a validated workbook. The approach still depends on model-generated code, notebook execution limits, validator quality, and model-endpoint or external-model costs.</p>
Concrete Fabric-native RLM implementation includes measured performance, validation, cost comparison, and operational traceability.
<p>On-premises data gateway version 3000.334 introduces <strong>soft delete</strong>: deleted gateways are retained for 30 days and can be recovered, reducing disruption from accidental administrative deletion.</p><p>The release also updates the embedded Log4j library, addresses CVE-2026-18401, improves Windows Server 2016 sign-in and configurator accessibility, and aligns gateway execution with the September 2026 Power BI Desktop release. Most other changes are maintenance and security updates rather than new Fabric data capabilities.</p>
Gateway soft delete adds operational recovery, while the remaining release is mainly security, compatibility, and patch maintenance.
Tabular Editor
Sep 24, 2026
3/5 · Tier B
AI ReadinessSemantic ModelGovernance
deep-dive
<p><strong>Tabular Editor 3.27</strong> includes an MCP server that lets coding agents such as Claude Code, GitHub Copilot, Codex, and Cursor read and edit the model currently open in Tabular Editor, including unsaved changes.</p><p>Permissions separate metadata read/write, model-data access, Best Practice Analyzer execution, documents, and macros. Changes remain visible in the TOM Explorer and Properties pane, each script becomes an undo step, and scripts attempting operations outside the model require manual review. Agents can also use BPA rules, VertiPaq Analyzer statistics, DAX queries when permitted, and version-specific C# scripting references.</p>
Live MCP editing adds practical agent controls, permissions, auditability, and review workflows for semantic-model development.
<p><strong>Tabular Editor 3.27</strong> integrates agentic semantic-model development through a built-in MCP server and a generally available AI Assistant. Both can apply C#-scripted changes, but permissions, safety checks, visibility, selective revert, and undo controls keep edits reviewable. AI Assistant direct changes are opt-in and can use locally hosted models.</p><p>The release also adds unsaved change indicators, property-level revert, deleted-object restoration, improved Data Preview, user-context-aware calculated-column validation, and compatibility level 1706 support. A breaking change affects scripts: query and command calls on <code>Model.Database</code> must use built-in script helpers instead.</p>
The release combines governed AI model editing, granular change management, validation improvements, and a script-breaking API change.
SQLBI
Sep 22, 2026
2/5 · Tier B
AI ReadinessSemantic Model
opinion
<p>AI-generated DAX has become more useful in 2026, particularly for drafting measures, transforming models, and handling longer development tasks. The productivity gain is strongest when the request clearly states business intent and the developer can inspect the generated semantics.</p><p>AI still does not reliably resolve filter context, data-model design, edge cases, or performance without validation. The article is primarily an opinion and training-position update rather than a Fabric feature or measured production study.</p>
The opinion reports improving AI-generated DAX but emphasizes human validation, modeling judgment, and performance review.
<p>A daily average is not a single business definition. <strong>AVERAGEX</strong> excludes blank measure results but includes zeros, so averaging over visible dates can silently calculate an average over selling days rather than calendar days.</p><p>Calendar-day averages require a denominator that includes dates without sales, while the date range must represent the actual observation period. A calendar extending beyond transaction history understates results, and transaction first and last dates are only proxies unless data availability is known.</p><p>The article uses AI to draft measures and validation queries, but the important control remains testing filter direction, blank handling, date boundaries, and edge cases against explicit business requirements.</p>
The article exposes how blanks, calendar scope, and data-availability boundaries change daily-average DAX results.
<p><strong>Semantic Link Labs</strong> can identify unused tables, columns, and measures in Power BI semantic models from a Fabric notebook. The <code>find_unused_objects</code> function can inspect downstream report structure or analyze DAX queries captured by Workspace Monitoring.</p><p>The result can be displayed in an interactive widget or returned as a pandas dataframe, making the analysis suitable for repeatable model-maintenance workflows. Query-based results depend on the observation window and activity captured in Workspace Monitoring.</p>
Semantic Link Labs automates unused-object detection using report structure or Workspace Monitoring DAX queries.
Fabric Updates
Sep 17, 2026
4/5 · Tier A
⚠️
Breaking ChangeGovernance
announcement
<p>Fabric Data Factory stops allowing <strong>new legacy Teams and Outlook activities</strong> from October 30, 2026. Existing activity instances remain editable and continue running; there is no forced migration.</p><p>New pipelines should use the modern Teams and Outlook 365 Email activities. The modern Outlook activity supports <strong>Workspace Identity</strong> and service principal authentication, while Teams currently does not support those identity options.</p>
Legacy notification activities reach a creation cutoff, requiring new pipelines to use modern Teams and Outlook activities.
<p><strong>Gateway soft delete</strong> is in preview for standard-mode on-premises data gateways. Deleted gateway clusters and members are retained for 30 days, allowing tenant administrators and gateway administrators to restore the gateway with its previous metadata and permissions.</p><p>This is recovery from accidental resource deletion, not a backup system: it does not restore individual nodes, data sources, or partial permission sets. Personal-mode and virtual network data gateways are unsupported, and preview API and audit-log support are unavailable. Gateways are permanently deleted after the retention window unless an administrator removes them earlier.</p>
Preview recovery reduces gateway deletion risk, but unsupported gateway types and absent API or audit support limit governance value.
<p><strong>Mirroring for Google BigQuery</strong> is generally available and continuously replicates BigQuery datasets into OneLake without customer-built pipelines, schedules, or transformation logic. Fabric manages snapshots, change capture, and schema evolution, while the data lands as open Delta tables for use across Fabric workloads.</p><p>Mirrored data can support Direct Lake semantic models, SQL analytics endpoints, notebooks, Eventhouse scenarios, and cross-source warehouse analysis. Connectivity includes Virtual Network support and on-premises Data Gateway support at version 3000.286.6 or later. The announcement states no Fabric replication compute or network ingress fees, but downstream analytics still consumes compute, and current limitations and supported schema changes require review before production adoption.</p>
GA BigQuery mirroring adds supported cross-cloud replication into open Delta tables, enabling Fabric workloads without pipeline maintenance.
The semantic model settings pane is becoming the default configuration experience in the Power BI service. It opens beside the workspace, keeps context visible, groups refresh, data access, performance, and storage-related settings, and provides search across the available options.
The classic full-page settings experience remains accessible through Open settings page during the gradual rollout, but it is planned for retirement. Scheduled refresh and data connection settings are also expected to be consolidated in the pane. Preview behavior may change before general availability.
Improves semantic model administration workflow, but changes navigation rather than model behavior or governance capability.
<p>Power Query is moving Fabric and Power BI connections from legacy embedded <strong>ODBC</strong> drivers to <strong>ADBC</strong>, Apache Arrow Database Connectivity. Each connector has a default flip, where new connections use ADBC while existing connections continue on ODBC, followed by a cutover that disables ODBC in the service.</p><p>The <strong>pq-adbc-advisor</strong> scanner inventories affected semantic models, dataflows, and Data Pipelines, classifies them as <strong>Will fail</strong>, <strong>Needs review</strong>, or <strong>Ready</strong>, and recommends a per-item fix. It can scan a workspace or use the Fabric admin Scanner API for a tenant sweep, tracks progress across scans, and does not rewrite M, trigger refreshes, or transmit item contents. Anonymous adoption counts are collected by default, with an opt-out.</p>
ADBC cutovers can break refreshes, and the scanner inventories connector risk across workspaces and tenants before deadlines.
<p>Git-controlled semantic-model metadata creates a searchable history of measure, table, partition, and property changes instead of relying on dated PBIX copies. The recommended starting point is a plain-text format such as <strong>PBIP with TMDL</strong>, optionally alongside <strong>PBIR</strong> report files.</p><p>Power BI Desktop can save a complete project, while Tabular Editor and its CLI can save model-only TMDL structures. The article positions source control as the first step on a CI/CD ladder; later steps can add validation, deployment, and testing through pipelines and the Tabular Editor CLI.</p>
It provides a concrete first rung for Git-controlled semantic models, with PBIP, TMDL, PBIR, and tool-specific setup details.
<p>A Tableau to Power BI migration is framed as a <strong>UX and adoption project</strong> rather than a purely technical lift. Tableau is canvas-first. Power BI is model-first: relationships and DAX sit underneath, and the report canvas is the last layer. That difference creates friction for long-time Tableau users.</p><p>The core advice is to avoid a pixel-for-pixel twin. Use the move as a forced audit and drop charts nobody uses. Keep familiar navigation, naming and filter patterns so users' muscle memory still works. Improve everything else deliberately.</p><p>Automation tools such as BIChart are positioned as handling the bulk rebuild so teams can focus on adoption. There are no concrete mapping guidance, gotchas or numbers, for example for LOD-to-DAX translation or RLS. The post closes with a vendor plug.</p>
Adoption-focused Tableau migration framing; light on concrete steps, partly vendor promo.
<p>The <strong>SQL DW operations skill</strong> is generally available as an open-source Fabric skill for compatible AI coding tools such as GitHub Copilot CLI. It turns natural-language operational questions into bounded, read-only diagnostics across Query Insights, warehouse system views, SQL pool metrics, and the Capacity Metrics app.</p><p>Results separate diagnosis, evidence, ruled-out causes, recommendations, and validation steps. The skill distinguishes failed queries from cancellations, execution volume from per-run cost, and capacity consumption from warehouse CPU. Capacity Metrics operation identifiers are not joined directly to Query Insights statement identifiers, so the signals remain correlated by time rather than treated as equivalent measurements.</p>
GA read-only diagnostics correlate warehouse queries, capacity metrics, contention, and custom SQL pool candidates.
<p>Fabric event-driven architectures use <strong>Real-Time Hub</strong> as the common entry point for Business Events, Fabric Events, and Azure Events. The publisher determines the event pillar: business workloads publish business meaning, Fabric publishes platform activity, and Azure Storage supplies external storage activity.</p><p>Consumer choice follows the required outcome: <strong>Activator</strong> handles actions, <strong>Eventhouse</strong> provides queryable event history, and <strong>Eventstream</strong> transforms or routes streams. The article emphasizes event names, payloads, versions, ownership, and schemas as contracts so publishers and consumers can evolve independently, but it does not provide detailed throughput, latency, retention, or cost limits.</p>
Provides useful event-channel and consumer guidance, but remains conceptual adoption advice without operational limits or cost analysis.
<p><strong>Tabular Editor CLI 0.7.0</strong> is a limited public preview focused on safer defaults and consistent automation output. Deployment and refresh now perform dry runs unless <code>--execute</code> is supplied, and remote Power BI Service or Desktop models can be addressed without first exporting them.</p><p>Property assignment now uses <code>-p</code>, listing and dependency inspection converge on <code>get</code>, and several commands move under <code>util</code> or are removed. Validation, BPA, tests, and queries share a JSON findings structure, while change output groups before-and-after properties by object. Existing scripts parsing commands or JSON must be updated; the preview is not recommended for production CI/CD and expires after October 31, 2026.</p>
The preview release changes deployment safety, command syntax, JSON contracts, BPA parity, and refresh automation behavior.
<p><strong>DAX INFO functions</strong> expose metadata such as the last modification time of measures and other semantic-model objects. Querying <code>INFO.MEASURES()</code> in DAX Query View or DAX Studio lets you sort measures by modification time and compare model changes with the start of a performance problem.</p><p>A broader query can combine measures, calculated columns, calculation groups, roles, perspectives, and other objects. Newer metadata functions may fail on older compatibility levels, and adding a calculation group can unexpectedly update many measure timestamps, so the results are useful evidence rather than a perfect audit log.</p>
A concrete DAX metadata technique exposes modification timestamps for diagnosing unexplained semantic-model performance regressions.
<p><strong>semantic-link-labs 0.17.1</strong> adds APIs for creating, deleting, and listing workspace relationships. Workspace monitoring also accepts MDX queries from Excel, extending usage analysis beyond DAX workloads.</p><p><strong>lineage_view</strong> can analyze Excel files for broken semantic-model references, although sensitivity-labeled Excel files are unsupported. <strong>migrate_to_direct_lake</strong> now accepts different source schemas per table, improving migrations from heterogeneous lakehouse structures.</p>
Extends semantic-link-labs automation with workspace relationships, Excel lineage, schema-aware Direct Lake migration, and MDX monitoring.
<p>Business Events should be treated as durable product contracts rather than one-off triggers. Events describe factual business occurrences, use names such as <code>Retail.Inventory.LowStockThreshold</code>, and leave reactions to independently owned consumers.</p><p>The guidance recommends explicit <strong>dataVersion</strong> values, parallel operation during breaking schema migrations, and idempotent consumers because delivery retries can continue for up to 24 hours. Existing subscriptions remain on their selected version rather than upgrading automatically. Event chains should represent meaningful business stages; centrally controlled, multi-step workflows belong in orchestration instead.</p>
Practical Business Events guidance addresses fanout, schema evolution, retries, idempotency, and event-chain design.
<p>The <strong>Fabric item relations API</strong> exposes upstream and downstream lineage through two preview REST operations. Responses include related items, typed dependency edges, and workspace metadata, allowing applications to resolve dependencies across workspace boundaries.</p><p>Calls require <code>beta=true</code> and Read permission on the queried item; both users and service principals are supported. Relation types include Association, Shortcut, PushData, Orchestration, Datasource, and parent-child or hidden dependencies. The API is preview-only and not recommended for production use, so consumers must tolerate contract changes.</p>
Preview REST operations expose typed Fabric lineage for impact analysis, CI/CD checks, catalogs, and grounded AI agents.
Fabric Updates
Sep 10, 2026
4/5 · Tier A
Spark RuntimePySpark
announcement
<p>Fabric’s <strong>Native Execution Engine</strong> can now read and parse JSON through its vectorized native path in preview. JSON values can flow as columnar batches into eligible filters, projections, joins, and aggregations instead of first passing through the JVM-based reader.</p><p>Existing Spark DataFrame and SQL code remains unchanged. The benefit depends on operator support and plan shape; unsupported operations can still cause execution-path transitions. The practical target is end-to-end job improvement for telemetry, API payloads, metadata-driven pipelines, and JSON-to-Delta standardization, not an automatic guarantee for every JSON workload.</p>
Preview JSON support extends Fabric’s native vectorized Spark path to semi-structured ingestion and downstream transformations.
Power BI Updates
Sep 10, 2026
5/5 · Tier A
⚠️
Breaking ChangeAI ReadinessMigration
announcement
Power BI Q&A now has a February 2027 retirement date, extending the earlier December 2026 timeline. The retirement removes the Q&A visual, breaks reports that contain it, and ends the associated Q&A configuration experiences.
Copilot is available across Fabric capacities beginning at F2, and Fabric Copilot Capacity can also be created from F2. This gives Pro and PPU estates a smaller-capacity route to evaluate Copilot, but it does not provide a direct Copilot replacement for embedded PaaS Q&A. Copilot is also unavailable in sovereign clouds because of GPU availability.
Embedded customers must evaluate SaaS migration, narrative experiences, Fabric IQ MCP, or another analytical interface. Sovereign-cloud customers need a separate migration discussion because the proposed Copilot paths are not currently available there.
Extends Q&A retirement to February 2027 while exposing unsupported embedded and sovereign-cloud replacement paths.
MoonStory (Elena Drakulevska)
Sep 10, 2026
2/5 · Tier B
Report Authoring
tutorial
<p>Five low-profile Power BI Desktop settings and right-click behaviors:</p><ul><li><strong>Set as landing page</strong>: right-click a page tab to choose it (requires Desktop May 2026 or later). The report then always opens on that page instead of whichever page was active at last save.</li><li><strong>Page list popup</strong>: right-clicking the page navigation arrows lists every page, so you can jump directly to one.</li><li><strong>Reset to default</strong>: right-clicking almost any format pane setting restores its default value.</li><li><strong>Cross-filter by default</strong>: a report-level option switches the default interaction from cross-highlighting to cross-filtering. The path is Options, then Current file, then Report settings.</li><li><strong>Screen-adaptive tooltips</strong>: tooltip pages can resize to the viewer's screen.</li></ul>
Practical small report settings; landing page feature is genuinely useful for maintenance.
<p><strong>OneLake Catalog</strong> will return tables as standalone search results, starting in late September. Users can search by table name, description, or exact column name across semantic models, lakehouses, and mirrored databases without knowing the parent workspace.</p><p>Discovery requires <strong>Read</strong> permission on the parent item, but not Read Data or Read All. It does not grant data access, and tables protected by object-level security remain hidden. Administrators can disable the <strong>Users can find objects in search</strong> tenant setting to keep results at item level. The same permission-aware search is exposed through the Search API and Fabric MCP surfaces.</p>
Preview table discovery changes metadata visibility, tenant governance, API automation, and permission expectations.
<p>The Fabric SQL Query Editor now provides a more scalable web development experience for large warehouses. The GA data grid supports column resizing and larger LOB values, while the redesigned Object Explorer, pinned objects, and more responsive IntelliSense improve navigation and authoring across large schemas.</p><p>Query workflows also gain controllable autosave, bulk query management, .sql import and export, and preview sharing that preserves query context and results. GA integrations extend validated SQL work into OneLake analysis and the semantic layer, with a preview path to Fabric Activator operational workflows.</p>
GA improvements make warehouse SQL development scalable, portable, and connected to OneLake, semantic models, and operational workflows.
<p>The August Fabric Influencers Spotlight aggregates 42 community submissions across governance, engineering, warehousing, AI, Real-Time Intelligence, semantic models, and certifications. Relevant entries include cross-tenant private endpoints, RLS and Purview labeling, auditing and DLP, Fabric CI/CD decisions, CU-cost measurement, ADF migration, pipeline automation, and Delta Lake.</p><p>It is a discovery index rather than a technical analysis: the extract gives short descriptions but not the configurations, measurements, limitations, or evidence needed to make an architecture decision.</p>
Community roundup surfaces relevant governance, CI/CD, capacity, migration, and engineering links but provides little original technical detail.
<p><strong>Power BI now bundles Oracle Managed ODP.NET</strong> for Import and DirectQuery in Desktop, removing separate client installation for supported Oracle versions. Power BI Service also supports direct cloud connections to Oracle Autonomous Database and other Oracle cloud endpoints without an Oracle gateway.</p><p>On-premises Oracle sources still require an enterprise or VNET gateway, while personal gateways are unsupported for updated connections. Entra ID SSO remains gateway-only, and rebinding an existing model away from a personal gateway cannot be reversed.</p>
Preview connectivity reduces Oracle client administration but has important gateway, SSO, and irreversible binding limitations.
MoonStory (Elena Drakulevska)
Sep 8, 2026
3/5 · Tier B
AI ReadinessReport AuthoringCI/CD
announcement
<p>The <strong>Power BI Report Authoring Skill</strong> is in preview through the open-source <strong>Skills for Fabric</strong> marketplace, shipped as the powerbi-authoring bundle. It lets AI coding agents create and edit reports directly in <strong>PBIP</strong> project files. Supported agents include GitHub Copilot CLI and Claude Code.</p><p>The skill can:</p><ul><li>add pages, visuals, slicers and filters</li><li>apply themes and formatting</li><li>modernize old visuals</li><li>validate report structure</li></ul><p>It works alongside companion skills:</p><ul><li><strong>Report Planner</strong> gathers requirements.</li><li><strong>Report Design</strong> handles layout and visual choice.</li><li><strong>Desktop Bridge</strong> connects the agent to Power BI Desktop.</li></ul><p>The Desktop Bridge closes the loop. The agent can reload the open report in Desktop, capture screenshots and inspect the rendered state, then iterate. The cycle becomes edit, reload, screenshot, review, rather than editing blind.</p><p>The post covers no limitations, cost, accuracy data or governance controls. Installation and changelog details are deferred to the Skills for Fabric repo.</p>
Preview agent skills editing PBIP reports with Desktop Bridge feedback loop; no limits detailed.
Tabular Editor
Sep 8, 2026
2/5 · Tier B
AI ReadinessCI/CD
tutorial
<p><strong>Context management</strong> determines an agent’s relevance, reliability, latency, and token cost. Context includes prompts, conversation history, tool descriptions, retrieved documents, memory files, skills, and metadata such as PBIP, TMDL, and PBIR content.</p><p>The practical approach is to provide only the information needed for the current task, progressively disclose detailed references, prune stale conversation state, and curate persistent instructions rather than allowing the agent to write all its own memory. Larger context windows do not eliminate distraction, quality degradation, or token costs.</p>
Useful agent-context principles, but lacks Fabric-specific limits, costs, governance, or implementation detail.
<p>DAX totals can be numerically valid while being semantically wrong when a business rule is evaluated above the grain where its inputs have unambiguous meaning. The article distinguishes the source-table grain, business-rule grain, evaluation grain, and requested output grain.</p><p>For the lemonade example, ingredient rates belong to recipes while water belongs to batches. Iterating batches is correct, but iterating recipes is more efficient: aggregate water by recipe, apply each recipe’s rate, and aggregate only afterward. The resulting totals are 38 sugar spoons and 3.20 liters of lemon juice, rather than values based on unweighted average rates.</p>
The article gives a rigorous grain-based method for preventing plausible but incorrect DAX totals and avoiding unnecessary iteration.
<p>A new comparative paper evaluates Power BI storage choices for semantic models over Azure Databricks data. The results are not universal: <strong>Direct Lake on OneLake</strong> performs best in most tested scenarios, while a composite model with a DirectQuery fact table, Dual dimensions, and Import aggregations is stronger for larger data volumes.</p><p>The analysis focuses on performance and does not evaluate cost or governance. Its practical value is the decision framework: reproduce representative tests against your own data, model shape, report queries, and volume before selecting a storage mode.</p>
Comparative testing identifies when Direct Lake wins and when composite storage modes better handle larger Databricks volumes.
<p>Reliable medallion architecture depends on operating discipline: batch writes, rerunnable transformations, consumer-oriented Gold models, evidence-based tuning, and monitoring by layer.</p><p>The guidance favors bulk loading with <strong>COPY INTO</strong> or pipelines, set-based CTAS, INSERT...SELECT, and MERGE patterns, with 100 MB to 1 GB files as a starting point for file-based ingestion. Silver should enforce quality rules and idempotent processing, while Gold should expose explicit-grain facts, dimensions, marts, or aggregates shaped around analytical use.</p><p>Fabric Warehouse defaults and V-Order should generally remain enabled unless measurements justify change. Query Insights and warehouse monitoring views should be combined with pipeline metrics such as duration, accepted and rejected records, refresh time, and critical Gold query performance.</p>
Gives practical medallion operating guidance with batch-size ranges, rerunnable Silver transformations, Gold modeling, and layer monitoring.
<p><strong>Workspace Outbound Access Protection (OAP)</strong> applies default-deny outbound policies to Operations Agent and Fabric Maps. Workspace administrators approve destinations through Data Connection Rules, while blocked actions appear in notifications and the Operations Agent Activity Log.</p><p>Fabric Maps validates references during save and load, then validates map tiles, geospatial features, and query requests at runtime. Lakehouses in the same workspace remain allowed; cross-workspace Lakehouses require rules. Approved WMS, WMTS, and WFS endpoints are supported, while cross-workspace Kusto databases and Ontologies remain blocked. Policy failures deny access rather than permitting unvalidated connections.</p>
Preview adds default-deny outbound controls, allow lists, audit visibility, and fail-closed behavior for agents and maps.
<p><strong>Connection Recency</strong> exposes creation time, last binding to an item, and last credential use for Fabric connections. The signals distinguish a connection merely linked to an item from one whose credentials were actually used.</p><p>The REST APIs support tenant-scale inventory work: identify potentially stale connections, compare duplicate configurations, and find connections with only one user owner. Gateway environments have a limit of <strong>1,000 connections</strong>. Credential-use data can lag by up to 30 minutes and should support governance review rather than real-time monitoring.</p><p>Connections created before recency was introduced can show null binding dates, so the example filters older records. API visibility also depends on permissions: cloud connections are limited to those the caller owns, while gateway administrators can see gateway connections. Deletion should remain a reviewed action, not an automatic consequence of age.</p>
Adds actionable REST signals for stale, duplicate, and single-owner connections, including a 90-day review pattern and gateway limits.
<p><strong>Capacity Operation Events</strong> are available in Real-Time Hub as a preview, adding operation-level telemetry to the broader Capacity Overview Events. Administrators can identify which operations, workspaces, and artifacts are contributing to capacity consumption, throttling, or performance pressure.</p><p>Events include workspace and item details, capacity-unit consumption, operation duration, start time, and smoothing-window attribution. The telemetry can feed Eventhouse, custom dashboards, alerts, and Microsoft Activator workflows for near-real-time notification or remediation.</p><p>The feature complements rather than replaces the Capacity Metrics App: overview events answer what is happening to capacity, while operation events help explain which workloads caused it. Preview schemas and event availability should be validated before production automation depends on them.</p>
Exposes operation-level capacity telemetry for consumption, throttling, workspaces, artifacts, and near-real-time automation.
A benchmark compares four ways to expose the same Delta tables from Azure Databricks to Power BI: Direct Lake on OneLake, Direct Lake over mirrored Unity Catalog tables, DirectQuery against a Databricks SQL warehouse, and a composite model combining DirectQuery with Import aggregations.
Direct Lake on OneLake performs well across the broadest range of tested situations and typical repeatedly used workloads. At very large volumes, the composite model with aggregations is fastest and most consistent when queries resolve against the aggregation tables, remaining below 100 milliseconds in those cases. Queries that fall through to DirectQuery do not receive that benefit.
Results depend on data volume, cache state, filters, and query shape. The comparison measures end-user Power BI experience rather than raw database execution speed and reflects testing from June and July 2026, so it should guide—not replace—testing on the target workload.
Benchmarks four Databricks serving patterns and reports measurable workload-dependent performance differences, including sub-100ms aggregation queries.
<p><strong>Fabric Pipelines and Copy Jobs</strong> can move data to or from Snowflake without exposing the workspace, staging storage, or Snowflake transfer path publicly.</p><p>The topology combines <strong>Workspace-level Private Link</strong> for inbound workspace access, an external Azure Blob staging account with Trusted Workspace Access, and Snowflake storage integration configured with <strong>USE_PRIVATELINK_ENDPOINT = TRUE</strong>. Fabric and Snowflake use separate identities, requiring Blob Data Reader and Blob Data Contributor permissions respectively.</p><p>OneLake cannot currently provide the required private staging path in this configuration, so external Azure Blob is necessary. Workspace-level Private Link requires supported capacity and region settings, while Snowflake-managed outbound private connectivity requires Business Critical edition or higher. The Snowflake private endpoint also requires approval on the storage account.</p>
Provides a concrete private-network topology, identity model, prerequisites, and Snowflake edition limitation for regulated workloads.
Power BI Updates
Sep 2, 2026
2/5 · Tier B
Report AuthoringSemantic Model
announcement
<p>Microsoft has refreshed its Power BI sample reports now that the new <strong>modern visual defaults</strong> are generally available. The samples include Competitive Marketing, Revenue Opportunities, Regional Sales, Employee Hiring, Corporate Spend, AI and AdventureWorks.</p><p>Semantic model improvements:</p><ul><li>Clear naming and descriptions</li><li>Common measures</li><li>Consistent formats and sort-by columns</li><li>Hidden technical columns</li><li>Reviewed relationships</li><li>Calculation groups for period comparisons and inactive relationships</li></ul><p>Report improvements:</p><ul><li>Latest base theme and larger default canvas</li><li>Mobile layouts and report page tooltips</li><li>Bing Maps and the Q&A visual replaced (Azure Maps replaces Bing Maps)</li><li>Fewer shapes, using built-in padding, titles and subtitles</li><li>New card visual, date picker, button slicer and field parameters</li></ul>
Reference samples demonstrating modern defaults and model hygiene; useful templates, little new.
Fabric Updates
Sep 2, 2026
3/5 · Tier B
AI ReadinessGovernance
announcement
<p><strong>Fabric Data Agents</strong> are generally available in Copilot Studio as tools through <strong>Fabric IQ Data MCP</strong>, replacing the earlier connected-agent experience. Copilot Studio can invoke the data agent alongside connectors, workflows, SharePoint knowledge, and other tools.</p><p>The Fabric data agent continues executing in Fabric and respects permissions on the underlying data sources. Agents can now be published to <strong>Microsoft Teams</strong> and <strong>Microsoft 365 Copilot</strong>, allowing users to ask questions over governed enterprise data inside their normal work surfaces.</p><p>The integration improves composition and reuse, but answer quality still depends on the data agent’s definitions, source permissions, and orchestration decisions. The extract does not provide specific consumption pricing, quotas, or accuracy benchmarks.</p>
GA integration makes governed Fabric data reusable in Copilot Studio, with permission inheritance and broader publishing options.
<p><strong>Power BI and Fabric are transitioning eight connectors from embedded ODBC drivers to Apache Arrow Database Connectivity (ADBC)</strong>, including Snowflake, BigQuery, Databricks, Spark, Impala, and Dremio. ADBC uses Arrow-oriented transport to reduce serialization and copying for large result sets, while improving driver security and memory safety.</p><p>Existing connections continue using ODBC until edited or explicitly configured with <strong>Implementation="2.0"</strong>. During the validation window, administrators can enable ADBC for individual connections, a workspace, or the tenant. ODBC becomes unavailable in the service after the connector-specific cutover, currently expected in early 2027 for the listed connectors, with no tenant opt-out afterward.</p>
GA driver transition has mandatory cutover dates, validation controls, connector-specific behavior, and production migration risk.
<p>The <strong>List Networking Communication Policies Admin API</strong> is generally available for tenant-wide visibility into workspace networking policies. It exposes inbound controls such as public access, IP firewall rules, and default behavior, alongside outbound restrictions, connection and gateway rules, Git communication policies, and managed private endpoints.</p><p>Results are paginated and support filters for selected inbound and outbound conditions. Fabric administrators or authorized service principals can incorporate the response into governance dashboards, SIEM monitoring, security reviews, and compliance evidence without inspecting workspaces individually.</p>
A GA tenant-level Admin API enables automated auditing of workspace network controls at enterprise scale.
<p>New guidance connects Fabric REST APIs, Git integration, deployment pipelines, Variable library, Fabric CLI, Terraform, and the <strong>fabric-cicd</strong> library into end-to-end delivery patterns. Tutorials cover Terraform-provisioned environments, local deployment, Azure DevOps approval gates, parameterized environment values, and Git-driven Bulk Import and Export API deployments.</p><p>The dependency guidance is particularly important: some references bind through portable logical IDs, while others retain workspace-specific object IDs and require parameterization. An open-source generic Git sample extends the pattern to GitLab, Bitbucket, GitHub Enterprise Server, Azure DevOps Server, and multi-tenant deployments.</p>
The resources provide concrete Fabric promotion patterns, automation tooling, dependency handling, and multi-provider Git support.
Fabric Updates
Sep 1, 2026
3/5 · Tier B
AI ReadinessSemantic Model
announcement
<p><strong>Advanced DAX generation</strong> is available in the Fabric data agent Preview runtime. Instead of translating a question in one pass, the agent can inspect semantic-model metadata, use tools, review intermediate results, refine its query, and execute the final DAX.</p><p>Instance value indexing helps resolve informal user wording to actual column values, improving filters for company names, products, locations, and categories. The feature remains preview-only, and the runtime affects more than DAX generation. Clear names, focused AI schemas, instructions, verified answers, and performant measures remain necessary for reliable results.</p>
Preview agentic DAX generation improves value resolution and iterative reasoning, but accuracy still depends on model preparation.
Tabular Editor
Sep 1, 2026
1/5 · Tier C
AI ReadinessCI/CD
tutorial
<p><strong>Markdown</strong> provides a lightweight, human-readable format for agent instructions, documentation, skills, and memory files. Its simple syntax, broad tooling support, and low formatting overhead make it convenient for repositories and AI workflows.</p><p>The article contrasts Markdown with PBIP artifact files and recommends storing durable context in Git, including alongside report or semantic-model definitions. It remains an introductory format guide rather than a discussion of agent accuracy, security, deployment behavior, or Fabric-specific automation.</p>
Beginner Markdown tutorial offers limited architectural value despite mentioning PBIP and agent context files.
<p>The new preview guidance compares four dbt patterns: Warehouse-only, Lakehouse landing with Warehouse transformation, Lakehouse refinement with Warehouse serving, and Lakehouse-only Delta-first architecture. The choice depends on where raw data lands, which engine performs transformations, ownership boundaries, and how curated data is consumed.</p><p>It separates architecture from orchestration. dbt can own transformation logic, tests, dependencies, and materialization, while native scheduling or Fabric pipelines can orchestrate broader workflow dependencies. The guidance is architectural rather than a performance benchmark, so engine behavior and operational cost still require workload-specific validation.</p>
Preview guidance compares practical dbt medallion ownership patterns across Lakehouse and Warehouse architectures.
<p><strong>Eventstream can retrieve streaming data from private MQTT, Kafka, CDC, and messaging sources through an approved virtual network path.</strong> A customer-owned Azure VNet connects to the private source using VPN or ExpressRoute, while a delegated subnet hosts the managed streaming connector.</p><p>The workspace identity needs permission to use the VNet, and Fabric represents the network through a streaming virtual network data gateway. Publishing the Eventstream injects the connector into the subnet, allowing outbound retrieval without exposing the source publicly. Network routing, DNS, firewalls, credentials, certificates, source availability, and subnet IP consumption remain customer responsibilities.</p>
Detailed private-network Eventstream architecture covers routing, identity permissions, connector placement, and subnet capacity implications.
<p><strong>Fabric Migration Assistant for SQL database is generally available.</strong> Its new <strong>Validate</strong> step analyzes an uploaded DACPAC before a Fabric SQL database is provisioned, identifying compatible and incompatible objects, failure reasons, and dependency chains without consuming a target or committing capacity.</p><p>After assessment, the guided workflow provisions the database, deploys the schema with interactive fix suggestions, copies data through Fabric Copy Jobs, and supports on-premises access through Fabric Data Gateway. Pre- and post-deployment scripts can handle load-specific constraint changes before validation and cutover.</p>
GA migration workflow adds pre-provisioning DACPAC validation, dependency-aware remediation, schema deployment, and integrated data movement.
<p><strong>Real-Time Dashboards can now embed a published Fabric Map as a reusable tile in preview.</strong> The dashboard stores a reference to the map item rather than copying its queries, layers, filters, styling, or data-source configuration.</p><p>Viewers can pan, zoom, change basemaps, toggle layers, and adjust unlocked filters temporarily. Locked filters enforce a baseline scope. Map creation and editing remain outside the dashboard in this phase, and viewers need permission to the referenced map and its data source; unavailable dependencies surface an error.</p>
Preview integration embeds reusable Fabric Maps in Real-Time Dashboards, with clear reference, filter, permission, and authoring boundaries.
<p><strong>Five scenarios show a common Fabric Business Events pattern:</strong> a workload detects a meaningful condition, publishes a schema-defined event, Real-Time Hub routes it, and Activator or Eventhouse reacts. Publishers include notebooks, User Data Functions, Eventstream, and Activator.</p><p>The scenarios distinguish immediate action from durable analysis. Activator handles alerts and chained automation, while Eventhouse preserves queryable event history for dashboards and audits. The reusable boundary keeps detection with the publisher and action, retention, or analysis with independent consumers.</p>
Concrete Business Event scenarios demonstrate reusable publisher-schema-consumer boundaries across Activator, Eventhouse, notebooks, and Eventstream.
Crossjoin
Aug 30, 2026
3/5 · Tier B
Report AuthoringGovernance
tutorial
<p><strong>Power BI reports can reference images and other files stored in OneLake.</strong> A measure can construct an authenticated OneLake URL dynamically, then use the <strong>Image URL</strong> data category to render context-specific images without embedding binary content in the semantic model.</p><p>The same approach supports files such as GeoJSON for Shape Map visuals. Replacing the file in OneLake updates the referenced asset without editing the report, although Azure Maps support is not available yet and is expected later.</p>
Secure OneLake-hosted images and GeoJSON remove model-bloat and public-URL workarounds.
<p>Gateway version 3000.330 aligns on-premises connectivity with the August 2026 Power BI Desktop release and includes stated security, authentication, diagnostics, dependency, and reliability updates. Matching gateway and Desktop runtimes helps maintain consistent query execution for gateway-backed refreshes.</p><p>The post provides no specific breaking changes, performance measurements, CVE identifiers, or configuration changes beyond upgrading the deployment.</p>
The gateway update mainly provides routine compatibility, dependency, security, and reliability maintenance without behavior-changing technical detail.
<p><strong>Activator</strong> handles immediate responses to published Business Events, such as alerts, pipelines, notebooks, flows, or User Data Functions. <strong>Eventhouse</strong> stores events as queryable history for KQL analysis, operational dashboards, recurrence analysis, and machine-learning features.</p><p>The recommended pattern is fanout: publish once, then let Activator and Eventhouse consume the signal independently. This separates detection from operational response and preserves an analytical record without coupling publisher logic to downstream teams.</p>
The article gives a practical event-driven decision pattern for immediate actions, durable history, or both.
<p><strong>Fabric Data Warehouse</strong> now generally supports using a workspace identity as the credential for <strong>COPY INTO</strong>. The initiating user still needs permission to load the target table, while the workspace identity separately authenticates to the source files. Both authorization checks must succeed.</p><p>This removes the need to grant ingestion users direct access to restricted raw storage or distribute SAS tokens, account keys, or service-principal secrets. Administrators can grant the workspace identity narrowly scoped source access while warehouse administrators retain table-level control. Existing storage, Fabric, and network controls continue to apply.</p>
GA workspace-identity ingestion clearly separates table authorization from restricted source access, with executable syntax and regulated scenarios.
<p>45 features - 19 must-know - 24 worth knowing.</p><ul><li>Git Integration and Deployment Pipeline – Item level permission restriction</li><li>Fabric Runtime 2.0 (Generally Available)</li><li>Data Agent is migrating from Assistants API to Responses API</li><li>CI/CD 2.0 with DacFx (Preview)</li><li>OneLake security improvements for SQL analytics endpoints (Generally Available)</li><li>Git Integration – Workspace Relation API (Preview)</li><li>Resource instance rules for OneLake (Generally Available)</li><li>Native Execution Engine performance improvements</li><li>Data Agent orchestrator upgraded to GPT 5.1</li><li>Identity columns with identity insert (Generally Available)</li><li>GPU Query Acceleration (Preview)</li><li>Secure data ingestion with COPY INTO and Workspace Identity (Generally Available)</li><li>Eventstream MQTT connector (Generally Available)</li><li>Reference data enrichment in Eventstream (Preview)</li><li>Eventstream observability in Workspace Monitoring re-enabled with per-Eventstream control (Preview)</li><li>Secure Azure Event Hubs Connections in Eventstream with Workspace Identity (Preview)</li><li>Workspace Outbound Access Protection (OAP) for Operations Agent (Preview)</li><li>Upgrade Dataflow Gen1 to Dataflow Gen2 (CI/CD) using the Upgrade Wizard (Preview)</li><li>Migration Assistant for SQL database in Fabric (Generally Available)</li></ul>
19 must-know of 45 features
<p><strong>IDENTITY columns</strong> are generally available in Fabric Data Warehouse, with support for <strong>IDENTITY_INSERT</strong> and reseeding. Tables can generate system-managed numeric surrogate keys during normal inserts, while migration or controlled backfill operations can explicitly preserve existing identity values.</p><p>Identity values are unique but not guaranteed to be sequential or gap-free because Fabric Warehouse generates them during distributed, parallel execution. This preserves load throughput but makes them unsuitable for assumptions about contiguous numbering.</p>
Fabric Data Warehouse identity insert and reseeding reach GA, directly improving migration and surrogate-key loading patterns.
<p>The post lists SQLCon and FabCon Europe 2026 sessions, Microsoft SQL team appearances, migration workshops, and booth activities in Barcelona. It provides event logistics and session identifiers rather than technical content or implementation guidance.</p>
This is conference promotion and session scheduling, not substantive Fabric guidance.
<p><strong>Anonymous data access</strong> for Fabric Apps is available in preview. A tenant administrator first enables the capability for the organization or selected security groups; an app’s data model then explicitly grants the anonymous role operations such as read or create.</p><p>The layered design supports focused public experiences while keeping other entities and operations authenticated. Anonymous users have no identity claims, so the model must expose only minimum necessary fields and operations. Public endpoints also require protection against untrusted input, spam, automation, and unexpected traffic.</p><p>The capability is unsuitable for personal, confidential, financial, or internal business data. Because it is preview, I would validate behavior, monitoring, abuse controls, and rejected operations before production use.</p>
Preview anonymous access changes Fabric App security boundaries, with explicit tenant, model-role, and operation controls.
<p><strong>OneLake Security</strong> is becoming more practical for enterprise SQL analytics endpoint deployments. In user identity mode, the endpoint passes the signed-in identity through to OneLake, where lakehouse policies govern row, column, and object access.</p><p>The improvements address nested Microsoft Entra groups, shortcut-backed hub-and-spoke lakehouses, column-level security inherited through groups, and broader service-principal scenarios, including service-principal-owned lakehouses. Security synchronization still requires operational attention: renamed or deleted tables and columns can cause configuration errors, and propagation speed remains an area of ongoing optimization.</p>
OneLake Security gains nested-group, shortcut, column-security, and service-principal support for SQL analytics endpoints.
Power BI Updates
Aug 25, 2026
5/5 · Tier A
Report AuthoringAI ReadinessGovernance
release-notes
<p>25 features - 0 must-know - 15 worth knowing.</p>
0 must-know of 25 features
<p><strong>Custom SQL Pools are preview workload-governance boundaries for Fabric Data Warehouse.</strong> They cap the share of warehouse resources available to classified workloads, allowing administrators to trade query latency for more predictable contention and consumption.</p><p>A bursty query scaling to 10 vNodes for 10 seconds can represent roughly 600 billed vNode-seconds under a one-minute minimum. A pool limiting it to five vNodes may extend runtime to 20 seconds while reducing the comparable allocation to roughly 300 billed vNode-seconds. This is not a spending cap or monthly budget limit.</p><p>Disabling <strong>Optimize for Reads</strong> may reduce cache-related consumption because cached resources otherwise remain allocated for at least five minutes, but query performance must be measured afterward. Background workloads also benefit from 24-hour smoothing behavior.</p>
Preview SQL Pools provide quantified workload isolation and consumption trade-offs, including billing-window and cache-retention implications.
Tabular Editor
Aug 25, 2026
1/5 · Tier C
AI Readiness
tutorial
<p>Agentic workloads trade off model capability, behavior, speed, hosting, configuration, and token cost. Higher-capability models and larger reasoning budgets can improve difficult coding or long-horizon tasks, but they increase latency and spend.</p><p>The article recommends matching model strength and effort to task complexity, using faster or cheaper models for iterative work, and avoiding unnecessary model switches that can invalidate cached context. The guidance is conceptual and does not provide measured workloads, provider comparisons, or enterprise governance requirements.</p>
General model-selection guidance mentions cost and effort, but provides no Fabric-specific AI limits or governance.
<p><strong>Mirrored Google Lakehouse runtime catalog</strong> is available in preview for bringing Google-managed Apache Iceberg catalog metadata into OneLake. Fabric reflects the catalog and creates the necessary shortcuts while the underlying data remains in its existing cloud storage location.</p><p>This avoids a copy-based ingestion pipeline and can expose supported tables to SQL, Power BI, Spark, Data Science, Data Warehousing, and Real-Time Intelligence workloads. It also enables joins and discovery alongside Fabric-managed and other mirrored data.</p><p>The benefit is reduced duplication and faster onboarding, not automatic migration. Support for Iceberg features, cross-cloud connectivity, query performance, permissions, and preview reliability must be validated before replacing established ingestion paths.</p>
Preview catalog mirroring enables zero-copy multi-cloud Iceberg access and changes Fabric migration and governance architecture.
<p>AI can help identify representative data cases and generate repeatable <strong>DAX tests</strong> for semantic-model measures. The approach separates business validation from automation: humans define the expected behavior, while AI explores the model, creates filter contexts, and produces queries that verify the measure repeatedly.</p><p>The example tests a dynamic relative new-customers measure with positive, negative, false-positive, and false-negative cases. It specifically checks that multiple selected brands are treated as one union rather than calculating a separate first purchase for each brand. Reusable DAX functions return standardized pass/fail rows, and a final <code>UNION</code> combines the test results. AI can generate cases, but it cannot establish the business requirement or guarantee that the selected cases provide complete coverage.</p>
The tutorial turns AI-assisted case discovery into repeatable DAX regression tests for filter-sensitive semantic-model logic.
<p><strong>Direct Lake on OneLake</strong> semantic models now support calculated columns in preview. Authors can define DAX columns in Desktop or web modeling without changing storage mode or pushing every derivation upstream into Delta tables.</p><p>These columns use <strong>User Context</strong> expression context, are evaluated at query time, and respect RLS and OLS. They cannot be relationship keys, do not respond to filter context, and are not materialized. Relationship keys should remain upstream, while calculations that must react to slicers belong in measures. The preview also introduces an important security distinction from standard Import calculated columns, where derived results can expose protected information.</p>
Direct Lake calculated columns add a requested modeling capability while exposing distinctive query-time, relationship, and security semantics.
<p>The <strong>Dataflows Upgrade Wizard</strong> provides a guided preview migration from Power BI Dataflows Gen1 to Fabric Dataflows Gen2 (CI/CD) for eligible workspaces on Fabric capacity. It assesses each dataflow, preserves its ID, schedule, queries, and connections, and allows multiple items from one workspace to be upgraded together.</p><p>The replacement is <strong>irreversible</strong>. Incremental refresh, Bring Your Own Lake storage, linked dataflows, possible DirectQuery consumers, naming changes, and legacy connector usage require follow-up. After migration, refresh each item, reopen downstream dataflows, and move consumers to the modern Power Platform Dataflows connector. Gen2 also adds Git, deployment workflows, monitoring, destinations, and compute options, but published benchmarks and lower-rate standard compute pricing must be validated against representative workloads.</p>
Guided Gen1 migration preserves identities but exposes irreversible upgrades, connector gaps, dependencies, and concrete performance and cost considerations.
<p><strong>Semantic Link Labs</strong> now exposes several notebook tools with interactive user interfaces, reducing the amount of code needed for common Power BI analysis. Examples include semantic-model and report lineage exploration, broken-visual discovery, and a VertiPaq Analyzer experience.</p><p>The workflow installs the library in a Fabric notebook, connects to a workspace and semantic model, and launches the relevant tool from a notebook cell. The extract does not provide capacity measurements or deployment guarantees, so compatibility with the notebook runtime and package version should be treated as an operational dependency.</p>
Demonstrates usable Semantic Link Labs interfaces for lineage, broken-visual discovery, and VertiPaq analysis inside notebooks.
<p>27 features - 6 must-know - 11 worth knowing.</p><ul><li>Deprecation of Old File Picker experience in Power BI Desktop</li><li>Updates to required semantic model permissions for Fabric Apps</li><li>More control over Semantic Model refresh in Power BI Service</li><li>Convert Direct Lake Tables to Import Storage Mode in Web Modeling (Preview)</li><li>Direct Lake Calculated Columns (Preview)</li><li>Faster PBIP development with instant reloads and VS Code integration</li></ul>
6 must-know of 27 features
Power BI Updates
Aug 20, 2026
2/5 · Tier B
AI ReadinessSemantic Model
opinion
<p><strong>Fabric IQ</strong> can use existing Power BI semantic models as business context for AI agents and Microsoft 365 Copilot. Measures, relationships, descriptions, synonyms, sample questions, and report terminology provide definitions that help AI answer using approved business meaning rather than ambiguous natural-language guesses.</p><p>The practical message is that organizations do not necessarily need to rebuild their reporting foundation for AI. However, the article does not quantify Fabric IQ cost, explain grounding limits or evaluation behavior, or address permissions and governance in enough detail to support a production rollout.</p>
Useful AI-grounding perspective, but it is largely opinion and omits operational limits, governance, accuracy, and cost evidence.
<p><strong>Eventstream workspace identity authentication for Azure Event Hubs</strong> is available in preview. Eventstream uses the Fabric workspace identity to obtain a Microsoft Entra ID token instead of storing an Event Hubs shared access key or connection string.</p><p>An administrator grants the workspace identity the <strong>Azure Event Hubs Data Receiver</strong> role through Azure RBAC. Developers then select Workspace identity when configuring the Eventstream source. Permissions can be changed at the Event Hubs namespace without recreating the connection, while separate workspaces retain separate access boundaries.</p><p>The feature removes manual secret rotation, but it remains preview functionality and requires administrators to configure both the workspace identity and Azure resource permissions.</p>
Preview identity-based Eventstream authentication reduces secret management and enables workspace-scoped Azure authorization.
<p>The <strong>KPI visualization</strong> is generally available in Fabric Real-Time Dashboards. A numeric result from a KQL query can be displayed as a Number, Bar, Gauge, or Donut, with configurable minimum and maximum scales.</p><p>Authors can define healthy, warning, and critical ranges, reverse the threshold direction for metrics such as latency, format values and units, and add reference lines for SLAs or operating limits. The visual requires Fabric-enabled capacity and Editor permissions on the dashboard.</p>
GA KPI visualization adds configurable ranges, targets, formatting, and four display modes for operational dashboards.
<p>The preview <strong>Fabric Data Warehouse MCP Server</strong> gives compatible agents a managed way to execute T-SQL against Fabric Warehouses and SQL analytics endpoints. It supports a global endpoint or an item-scoped endpoint, with Microsoft Entra ID OAuth 2.0 authentication and the user’s existing Fabric and SQL permissions.</p><p><strong>Skills for Fabric</strong> provide task-specific instructions for warehouse authoring, consumption, and operations. Skills guide the agent; the MCP server executes queries and returns live results, allowing iterative validation, error correction, schema exploration, and query-performance investigation through queryinsights.</p><p>The feature is preview and exposes execution against live data. Existing permissions still constrain the agent, but organizations must govern which MCP clients and users can perform DDL, DML, or diagnostic queries.</p>
Preview remote MCP execution and Fabric-specific skills let agents query, author, validate, and diagnose warehouses under existing permissions.
<p>A maintainable Fabric Data Warehouse medallion design gives each layer one job: <strong>Bronze</strong> preserves traceable source data, <strong>Silver</strong> cleanses and conforms it, and <strong>Gold</strong> shapes trusted business outputs for analytics.</p><p>The implementation uses patterns such as <strong>COPY INTO</strong> for bulk Bronze ingestion, CTAS and INSERT...SELECT for Silver transformations, MERGE for incremental updates, and fact, dimension, view, or aggregate structures in Gold. Metadata such as ingestion timestamps and source filenames supports replay and auditing.</p><p>The main operational risks are cleansing too early, non-repeatable transformations, tiny trickle loads, uncontrolled schema changes, and business logic leaking into reports. Sensitive fields should be protected or removed before Gold consumption.</p>
Concrete T-SQL medallion patterns separate raw ingestion, cleansing, quality gates, and business-ready warehouse outputs.
<p><strong>Delta Lake 4.4.0</strong> defaults Delta Spark to Apache Spark 4.2 while retaining support for Spark 4.1.0 and 4.0.1. The release adds SQL identity-column support, preserves generated-column expressions through Unity Catalog, and supports <code>SHOW PARTITIONS</code> on Delta tables.</p><p>It also formalizes preservation of <strong>VOID</strong> columns across reads and time travel, and extends catalog-managed Delta tables with metric views when paired with the Unity Catalog 0.6.0 connector. Metric views centralize dimensions and measures instead of duplicating aggregation logic across consumers.</p><p>The release spans Delta Spark, Kernel, UniForm, Sharing, and Flink. These are upstream ecosystem capabilities, so Fabric adoption depends on the Spark runtime, Delta version, catalog integration, and supported protocol features available in the target tenant.</p>
Substantive Delta 4.4 release coverage includes Spark 4.2 compatibility, catalog-managed tables, identity columns, and governed metric views.
<p><strong>Fabric Planning</strong> now supports PowerTable and Intelligence-only provisioning without Planning sheets. In that configuration, editors use the Stakeholder billing persona rather than the higher Planner persona, while read-only users remain Viewers.</p><p>Billing remains activity- and session-based against Fabric capacity, with 30-day session meters of <strong>37 CU for Viewers</strong>, <strong>168 CU for Stakeholders</strong>, and <strong>847 CU for Planners</strong>. A session dynamically upgrades when a user performs a higher-tier activity, and a user with multiple roles in one capacity is billed once at the highest applicable role.</p><p>Creating and accessing a Planning sheet triggers the Planner tier with user notification. Other Fabric resources, including SQL, OneLake, and the Power BI XMLA API, meter separately; the article recommends retaining approximately 30% capacity buffer.</p>
GA billing behavior provides concrete CU session rates, role upgrades, and highest-role attribution for Fabric Planning usage.
<p><strong>Fabric workspace CMK APIs are generally available.</strong> Administrators can assign or rotate an Azure Key Vault key, inspect encryption state, reset a workspace to Microsoft-managed encryption, and list tenant workspaces with encryption details.</p><p>The APIs use versionless key URIs, so key rotation uses the same assignment operation. Microsoft Entra authentication and existing workspace or tenant-admin permissions govern access, while status values support auditing and drift detection.</p>
GA APIs make workspace CMK provisioning, rotation, auditing, and tenant-wide encryption governance automatable.
<p><strong>Fabric integration workloads use three primary connectivity patterns:</strong> a Microsoft-managed VNet data gateway, a customer-managed on-premises data gateway, or a direct cloud connection without a gateway.</p><p>Choose based on where the source resides and whether Fabric must cross a private network boundary. VNet gateways suit private endpoints and ExpressRoute-reachable sources without gateway virtual machines; standard gateways suit on-premises or privately hosted sources requiring customer-managed clustering and operations; direct connections suit publicly reachable cloud services.</p><p>This guidance covers Dataflow Gen2, pipelines, Copy Job, Mirroring, semantic models, and paginated reports. Notebooks and Spark workloads use Fabric Managed VNet and follow a different model.</p>
The guide clarifies gateway selection and private-network trade-offs across Fabric integration workloads, but adds limited new behavior.
<p><strong>FabCon + SQLCon 2027 is accepting session submissions</strong> for its March 8–12 Atlanta conference. Tracks span Fabric governance, developer experiences, warehousing, data engineering, integration, real-time intelligence, AI agents, and Power BI.</p><p>Submissions receive blind review, so titles and descriptions should make the audience, technical depth, and expected takeaways clear.</p>
This is a conference call-for-content announcement without substantive Fabric implementation or platform behavior.
<p><strong>Power BI Desktop</strong> now detects changes made externally to Power BI Project files and prompts authors to apply them without restarting Desktop. The update covers edits from Visual Studio Code, external tools, scripts, automation, and AI-assisted development workflows.</p><p>A new <strong>Open in VS Code</strong> command launches the project folder as the active workspace. The feature reduces friction when alternating between Desktop and file-based report or semantic-model development, but it remains preview functionality and does not remove the need to manage conflicting or invalid file edits.</p>
PBIP projects now support immediate external-file reloads and direct VS Code launch, materially improving code-first development workflows.
Fabric Updates
Aug 18, 2026
2/5 · Tier B
AI ReadinessReport Authoring
announcement
<p><strong>Fabric data agents now render conversational charts with Fabric Visuals in preview.</strong> Responses can use line, bar, stacked bar, pie, scatter, area, and stacked-area charts across lakehouses, warehouses, and Power BI semantic models.</p><p>Visuals are enabled by default and can be requested explicitly or guided through agent instructions. The change improves formatting, legends, tooltips, markers, and axis scaling, but the extract does not describe preview limitations, cost, accessibility coverage, or accuracy controls.</p>
Preview data-agent visuals improve rendering consistency, but the announcement provides few limitations, costs, or governance details.
Tabular Editor
Aug 18, 2026
1/5 · Tier C
AI Readiness
deep-dive
<p>Large language models process text as <strong>tokens</strong>, generate output sequentially, and use attention to relate tokens within a context window. Input, output, tool, memory, and reasoning tokens affect cost, speed, and the amount of information an agent can process.</p><p>Models are probabilistic and can produce subtle errors even when their language and code look plausible. Reasoning and tool use can improve reliability, but they do not guarantee correctness; deterministic validation remains necessary for precise artifacts and business-critical results.</p>
LLM fundamentals are accurate and useful, but introductory and disconnected from actionable Fabric AI architecture.
<p><strong>Eventstream observability</strong> now uses a per-eventstream <strong>Log Eventstream activity</strong> toggle instead of enabling monitoring across an entire workspace. Monitoring is off by default, so teams can select only critical pipelines.</p><p>Enabled streams emit throughput and error metrics approximately every minute, plus node-status data approximately every six hours, into three monitoring KQL tables. Workspace Monitoring and its monitoring Eventhouse must already be provisioned. Diagnostic logs and broader connector coverage are not yet available.</p>
Per-eventstream observability adds practical monitoring controls and KQL metrics, but remains preview with limited diagnostics.
Fabric Updates
Aug 17, 2026
4/5 · Tier A
AI ReadinessSemantic ModelReal-Time Intelligence
deep-dive
<p><strong>Data source routing</strong> is generally available for Fabric data agents connected to multiple sources. The orchestrator selects the most likely lakehouse, warehouse, semantic model, or KQL database before invoking the corresponding NL2SQL, DAX, or KQL query tool.</p><p>Routing uses selected schema, descriptions, and example queries, often from a metadata subset rather than the full schema. Large or overlapping sources can therefore route confidently to the wrong place. Narrow schemas, purposeful descriptions, representative examples, and explicit routing rules improve accuracy; run steps expose when the dedicated routing tool was needed.</p>
GA routing behavior, metadata sampling, source-specific query tools, and concrete tuning guidance materially affect agent reliability.
<p><strong>GPU-powered Query Acceleration</strong> is a Fabric Data Warehouse preview that automatically offloads eligible scans, filters, joins, and aggregations. Existing T-SQL, DirectQuery models, and applications require no query rewrites; enabling the feature uses a workspace toggle.</p><p>The optimizer can accelerate an entire query, selected plan fragments, or nothing, depending on operators, data formats, functions, and query shape. Partitioned execution handles data larger than GPU memory, while unsupported or resource-constrained fragments fall back to CPU execution. The extract provides no benchmark figures or separate consumption pricing, so measured workload testing is essential.</p>
GPU execution could materially improve warehouse concurrency, but preview coverage, workload variance, and economics remain unspecified.
<p><strong>Business Events</strong> should be published by the Fabric workload that first detects the meaningful business condition and owns the authoritative context. Notebooks suit batch detection, User Data Functions suit incoming webhooks, Eventstreams suit continuous thresholds, and Activator suits derived business states.</p><p>The publisher emits a schema-valid, versioned payload through the Business Events API and does not directly send alerts, write analytical tables, or coordinate consumers. Retries and diagnostics are necessary to avoid silent loss or duplicate actions, while incomplete payloads and broad exception handling hide contract failures.</p>
Useful event-publishing design guidance and code, but it is a conceptual series article without measured limits or production evidence.
<p><strong>Business Events</strong> should begin with a schema contract describing what happened, which entity was involved, when it happened, and which facts consumers need to act. Specific fields such as <code>store_id</code>, <code>actual_transactions</code>, <code>deviation_pct</code>, <code>severity</code>, and <code>window_start</code> are more reusable than generic message fields.</p><p>Real-Time Hub supplies the CloudEvents envelope and platform metadata, leaving publishers responsible for the business payload. Namespace event names for discovery and ownership, publish an explicit data version, prefer additive changes, and treat field removal or renaming as a breaking change.</p>
Provides concrete schema and versioning practices for Business Events, but remains conceptual rather than implementation or scale tested.
<p>Choose the Fabric event pillar by identifying who publishes the signal. <strong>Business Events</strong> represent business conditions emitted by workloads; <strong>Fabric Events</strong> represent platform activity such as jobs, capacity, items, or OneLake; and <strong>Azure Events</strong> represent Azure Storage activity.</p><p>All three flow through Real-Time Hub, but their consumers differ. Business Events can feed Activator and Eventhouse, while Fabric and Azure Events can feed Activator and Eventstream. That distinction affects whether the target is business-signal analysis, automated action, or stream processing.</p>
Clarifies event-source selection and consumer compatibility, offering useful Real-Time Hub architecture guidance without implementation depth.
<p><strong>Real-Time Hub</strong> provides a common entry point for signals published by Fabric workloads, Fabric itself, or Azure Storage. Consumers such as Activator, Eventhouse, User Data Functions, Notebooks, Spark Jobs, Dataflows Gen2, and real-time dashboards can react independently, replacing scheduled polling where low latency matters.</p><p>The article illustrates a Business Event named <strong>Retail.Sales.VolumeAlert</strong>, with fields for store, expected and actual transactions, deviation, and severity. A notebook publishes the event when volume falls more than 30% below baseline, while Activator alerts operations and Eventhouse stores the history. It is a conceptual introduction rather than a detailed treatment of schemas, delivery guarantees, replay, scaling, or cost.</p>
Introduces Real-Time Hub event-driven concepts and a useful example, but provides little implementation or operational detail.
<p>Fabric Data Agent integrations using the Python SDK must migrate from the retiring <strong>Assistants API</strong> to the <strong>Responses API</strong>; the article cites an August 26, 2026 migration deadline. The RANCH helper converts uploaded notebooks by scanning code cells and proposing regex-based replacements, without using an LLM.</p><p>The migration has an important evaluation limitation: Responses API notebooks cannot be used for evaluation from outside Fabric. External evaluation must use the <strong>MCP endpoint</strong>, and the article points to sample Responses API notebooks for comparison.</p>
Provides a concrete notebook migration aid for the Fabric Data Agent move from Assistants API to Responses API.
Crossjoin
Aug 16, 2026
3/5 · Tier B
Report AuthoringSemantic Model
tutorial
<p>A Power BI report can partially expose a measure’s <strong>Detail Rows Expression</strong> by embedding a paginated report visual. The pattern uses a published semantic model, a paginated report connected to it, multi-value Customer and Product parameters, and a drillthrough page that passes the selected values.</p><p>The paginated dataset applies those selections through <strong>RSCustomDaxFilter</strong> and returns the detail table defined by the measure. This remains a workaround rather than native report support: it is hard-coded to one measure because Power BI drillthrough does not pass the measure clicked by the user.</p>
Provides a detailed paginated-report workaround for exposing Detail Rows Expressions, including multi-value parameter handling and a major measure limitation.
Fabric Updates
Aug 13, 2026
2/5 · Tier B
AI ReadinessAPI AutomationWarehouse
announcement
<p><strong>Skills for Fabric</strong> adds preview guidance for SQL databases to compatible AI coding tools including GitHub Copilot CLI, Claude Code, Cursor, and Windsurf. The skills supply Fabric-specific instructions for authoring schemas, consuming data, discovering items and endpoints, authenticating, and operating resources through APIs or CLI commands.</p><p>The workflow is planning-first: the agent proposes schemas, relationships, indexes, queries, commands, and validation steps before execution. It can also suggest vector columns for AI scenarios. The skills improve contextual accuracy but do not replace identity permissions, code review, or validation of generated changes.</p>
SQL database skills provide useful AI-assisted Fabric workflows, but the preview is primarily an enablement announcement without measured safeguards.
<p><strong>Fabric Lakehouse</strong> is now available as a target for dbt jobs in preview. Teams can connect an existing OneLake Lakehouse or create one while configuring a dbt project.</p><p>The integration brings dbt models, business logic, and tests closer to Lakehouse data. Supported adapters, runtime versions, setup requirements, and preview limitations remain documentation-dependent.</p>
Preview dbt integration expands Lakehouse transformation options but provides limited runtime, cost, and compatibility detail.
<p><strong>Capacity Overview Events</strong> are generally available in Real-Time Hub. <strong>Capacity Summary</strong> events arrive every 30 seconds with utilization and health information, while <strong>Capacity State</strong> events report changes such as pause, resume, or throttling-related conditions.</p><p>Events can drive threshold alerts, operational workflows, dashboards, and retention in Eventhouse for trend analysis and capacity planning. The optional community Capacity Events Accelerator supplies templates and dashboards, but it is not part of the core capability.</p>
GA capacity events expose 30-second utilization signals and state changes for alerting, automation, and historical analysis.
Fabric Updates
Aug 12, 2026
5/5 · Tier A
Spark RuntimeDelta & ParquetBreaking Change
announcement
<p><strong>Fabric Runtime 2.0</strong> is generally available and can be selected at workspace or environment scope. It upgrades the foundation to Apache Spark 4.1, Delta Lake 4.2, Python 3.13, Java 21, Scala 2.13, and Azure Linux 3.0.</p><p>The runtime also expands the <strong>Native Execution Engine</strong> for supported Spark SQL and DataFrame workloads without additional cost or required code changes. Runtime 2.0 is not immediately the default, but Microsoft plans to make it the default for the experience, new workspaces, and new environment items in late September 2026. Existing workloads therefore need compatibility and performance validation before adoption.</p>
Runtime 2.0 is GA with major Spark, Delta, Python, Java, and OS upgrades plus a near-term default-runtime transition.
<p><strong>SQL Audit Logs</strong> for Fabric Data Warehouse and SQL Analytics Endpoint now support generally available identity-based predicate exclusions. Administrators can prevent events from selected users or service principals from being generated, reducing storage, export, query, and SIEM-processing volume.</p><p>Filtering occurs before persistence, so excluded events cannot be recovered later. The control is available through the API and user experience, making it suitable for repeatable governance automation but requiring explicit policy review. Workspace administrators and users with the applicable audit permission can manage it.</p>
GA identity-based audit exclusions reduce SQL audit noise at ingestion, but permanently remove excluded events from retained evidence.
<p>Gateway version <strong>3000.326</strong> aligns gateway query execution with the July 2026 Power BI Desktop release.</p><p>The practical change is compatibility: reports published from the matching Desktop release should use the same query runtime during gateway refreshes. The extract provides no measurable performance, behavioral, or capacity changes.</p>
Desktop compatibility maintenance provides little Fabric architecture or operational insight beyond version alignment.
Fabric Updates
Aug 11, 2026
1/5 · Tier C
AI ReadinessGovernance
announcement
<p><strong>SQL Server 2025</strong> can invoke approved external AI services through T-SQL REST calls and governed external model objects. REST invocation is disabled by default and requires administrative enablement, endpoint permissions, approved authentication, and destination controls.</p><p><strong>CREATE EXTERNAL MODEL</strong> provides an explicit database object for embedding and vector-search integrations, with separate permissions for creation, alteration, and execution. SQL Server Audit and Extended Events provide visibility into configuration changes and endpoint or model usage.</p><p>The security boundary remains privilege and endpoint authorization: managed identities, least-privilege grants, outbound allowlists, endpoint-side RBAC, and behavioral monitoring are required. These controls apply to SQL Server 2025 and should not be mistaken for Fabric feature behavior.</p>
Useful SQL Server security guidance, but it is not a Fabric capability or Fabric integration change.
<p><strong>Capacity Metrics</strong> shows warehouse-level CU consumption over time, while <strong>Query Insights</strong> identifies the queries, users, and workloads active in the same period.</p><p>The proposed method joins those views analytically: sum query <strong>vCore seconds</strong> for a time window, then allocate the warehouse’s total CU consumption proportionally. A query responsible for 25% of vCore seconds receives an approximate 25% share of consumption.</p><p>This remains weighted attribution, not billing-grade per-query measurement. System activity, concurrency, time-window alignment, and the difference between allocated CPU and total warehouse consumption can affect the result.</p>
Provides a concrete, query-level method for attributing warehouse CU consumption using Capacity Metrics and Query Insights.
Tabular Editor
Aug 11, 2026
2/5 · Tier B
AI ReadinessCI/CDSemantic Model
tutorial
<p><strong>Agentic development</strong> uses an LLM-enabled application to inspect and modify BI artifacts through tools such as the Tabular Editor CLI, scripts, and source-controlled project files. The examples extend beyond semantic models to reports, documentation, ontologies, and end-to-end data-platform work.</p><p>Reusable skills, organizational context, memory files, and validation prompts can make agents more useful than unstructured prompting. The approach also introduces risks around incorrect code, data loss, security, and cost; the article demonstrates getting started rather than defining production-grade controls.</p>
Concrete agentic BI examples use Tabular Editor and Fabric artifacts, but remain introductory without production controls or measurements.
<p><strong>TMDL View on the Web</strong> brings code-first semantic-model editing into the browser. Developers can script published model objects, inspect properties hidden from the standard UI, use IntelliSense and bulk search-and-replace, and apply changes without downloading model files or opening Desktop.</p><p>The editor supports advanced properties such as partition definitions and <strong>isAvailableInMdx</strong>, reusable scripts, and version history containing model metadata and data for recovery. It is especially useful for targeted bulk changes, but the preview requires disciplined permissions and change control because browser edits can alter production-published models directly.</p>
Browser-based TMDL editing enables advanced metadata changes and reuse, with version history reducing recovery risk; preview limitations remain unspecified.
<p><strong>AI-assisted DAX refactoring</strong> can identify duplicated logic across measures and move the shared algorithm into a user-defined function (UDF). The example extracts the new-customer calculation, first as a generic function with multiple column arguments and then as a simpler model-dependent function.</p><p>The technique improves maintainability, but AI may silently change algorithms while refactoring—for example, replacing <strong>COUNTROWS</strong> logic with a filtered <strong>CALCULATE</strong> expression or reusing another measure. Generated code still requires semantic validation, performance testing, and review of filter-context behavior before production.</p>
Shows AI-assisted DAX refactoring with user-defined functions, reuse trade-offs, and explicit production-validation requirements.
Fabric Updates
Aug 10, 2026
3/5 · Tier B
AI ReadinessGovernance
tutorial
<p><strong>Build agent with AI</strong> helps authors explore schemas, inspect query history, generate instructions and example queries, and validate proposed queries before applying configuration changes.</p><p>The preview supports Lakehouse SQL analytics endpoints, Warehouses, Mirrored Databases, SQL Databases, and Eventhouses. It runs read-only validation queries, but authors still review and approve generated instructions.</p><p>Generated guidance can encode incorrect joins or business logic, especially when query history reflects inconsistent practices. The supported-source boundary and preview status therefore matter for production governance.</p>
Preview assistant improves Data Agent configuration and validation, with useful read-only safeguards and source limitations.
<p><strong>Delegated OneLake Shortcuts</strong> are in preview as an alternative to default pass-through authentication. A shortcut uses a configured organizational account or service principal to access the source, so downstream users do not each need direct permissions in the producer environment.</p><p>Security can be applied on both producer and consumer sides, with permissions combined by intersection. The preview supports table- and column-level security, including cross-tenant OneLake sharing. External Data Sharing remains the better fit when consumers have no identity in the producer tenant. Existing shortcuts cannot be switched in place; they must be deleted and recreated with the other authentication method.</p>
Details a preview security capability for scalable and cross-tenant zero-copy sharing, including supported controls and recreation limitations.
<p><strong>fabric-cicd 1.3.0</strong> adds <code>FABRIC_CICD_RETRY_API_MAX_DURATION_SECONDS</code> for controlling API request execution, connection retries, and long-running-operation polling; the default maximum is 300 seconds. Mirrored Database deployments also gain <code>$sqlendpoint</code> and <code>$sqlendpointid</code> replacement variables.</p><p>The release adds feature-flag discovery and removal functions, fixes Spark pool item-name filtering, and prevents deployment failures when asynchronously provisioned item attributes are not yet available. File-path filtering and repository-folder refreshes are also optimized.</p>
fabric-cicd 1.3 adds configurable API retry duration, SQL endpoint variables, feature-flag discovery/removal, and asynchronous provisioning fixes.
SQLBI
Aug 10, 2026
2/5 · Tier B
Report Authoring
tutorial
<p><strong>Fluent 2</strong>, the new default visual design style in Power BI, removes alternating row colors from Table and Matrix visuals. The banding feature still exists: select the visual and, in the Format pane under <strong>Visual > Style presets</strong>, change Style from <strong>None</strong> (the Fluent 2 default for new matrices) to <strong>Default</strong>.</p><p>The post covers fixing a single visual. It does not cover applying the change across a whole report.</p>
Practical fix for Fluent 2 default removing matrix banding; small but useful.
<p>A semantic model can keep an aggregated <strong>SalesSummary</strong> table in Import mode while hiding a transaction-level <strong>SalesDetail</strong> table in DirectQuery mode. A measure’s Detail Rows Expression can return rows from SalesDetail, allowing users to inspect transactions without exposing the detail table for ordinary browsing.</p><p>Shared Customer and Product dimensions propagate the current filter context to both fact tables. This can reduce memory footprint and capacity requirements compared with importing all detail, while giving the model author control over returned columns and row-volume safeguards. It is less flexible than aggregations, supports only one Import aggregation path per use case, and is primarily useful through Excel PivotTables; Power BI reports require a workaround.</p>
Explains a controlled composite-model detail pattern, its performance and capacity benefits, and important limitations versus aggregations.
<p>From <strong>August 16, 2026</strong>, tenants without an explicit Item Recovery setting move to enabled, with a three-day default retention window. Existing explicit enablement, retention, or disablement choices remain unchanged.</p><p>Supported deleted items enter a soft-deleted state and can be restored during retention; after the window, they are hard-deleted. Admins can choose between three and 90 days or disable the feature.</p><p>Item Recovery adds no separate meter, but retained soft-deleted items consume standard OneLake storage. The feature does not cover every Fabric item type.</p>
Changes the tenant default for item recovery and explains retention, supported types, administrative control, and OneLake storage impact.
<p>The <strong>modern evaluation engine</strong> is an opt-in preview for Mashup evaluations executed through VNet data gateways. It preserves the existing private-network gateway architecture while introducing an updated execution path.</p><p>Potential benefits include faster supported semantic model and Dataflow Gen2 refreshes and improved DirectQuery cold-start behavior. Results depend on connectors, source performance, gateway configuration, concurrency, and workload design; no benchmark numbers are provided.</p><p>Administrators can enable the setting on selected gateways, compare refresh duration, query responsiveness, reliability, and capacity utilization, then expand gradually.</p>
Preview opt-in changes Mashup execution for VNet gateways and recommends representative refresh, DirectQuery, reliability, and capacity testing.
<p><strong>Scheduled User Data Functions</strong> can run parameterized business logic on a recurring cadence through the Fabric job scheduler. The function owns the code; scheduling, monitoring, execution history, and failure notifications remain within Fabric.</p><p>Useful workloads include data-quality checks, feedback enrichment, synchronization, threshold alerts, and maintenance logic. The article does not specify preview limitations, execution pricing, concurrency behavior, retry semantics, or permissions, so those require validation before production adoption.</p>
Scheduled User Data Functions simplify recurring code-based automation, but preview limits and operational costs are not covered.
<p>Fabric medallion design should begin with the amount of <strong>Spark</strong> required, rather than treating Warehouse and Lakehouse as mutually exclusive choices. A SQL-centric team handling structured data can place Bronze, Silver, and Gold in Warehouse schemas or separate warehouses.</p><p>A hybrid pattern lands raw or semi-structured data in a Lakehouse for Spark and Python processing, then serves Gold from Fabric Data Warehouse. Delta tables in OneLake allow the two experiences to work over shared storage without necessarily copying data. Separating layers into items or workspaces improves security and change isolation, although the article does not address workload sizing or measured performance.</p>
Provides practical Warehouse-versus-Lakehouse architecture guidance, but lacks implementation, performance, and cost measurements.
<p><strong>Blob leases</strong> provide storage-managed mutual exclusion for a logical Fabric job. A persistent marker file identifies the process, while a finite 15–60 second lease grants one runner exclusive ownership.</p><p>The pattern resolves the OneLake URI through <strong>fsspec</strong>, uses the authenticated Azure filesystem client, acquires the lease, writes owner metadata, and renews the lease with a heartbeat. If the process dies, renewal stops and the lease eventually expires, avoiding permanent locks.</p><p>The lock URI must represent the logical operation—not a run, notebook, or pipeline name. The pattern adds complexity and should complement, rather than replace, idempotent processing, transactional writes, and normal pipeline concurrency settings.</p>
Practical cross-orchestrator locking pattern addresses duplicate Fabric jobs and unsafe concurrent Delta writes.
Power BI Updates
Aug 4, 2026
2/5 · Tier B
Report Authoring
tutorial
<p><strong>Conditional formatting for lines and legends</strong> dynamically controls colors across supported Power BI visuals using rules, gradients, or field values. Line, area, and stacked-area charts can format complete lines, segments, shading, markers, and series labels.</p><p>Field-driven colors centralize report styling in measures or columns, while visual calculations can supply logic directly within a visual. The feature also supports legend gradients and consistent colors for field parameters, reducing manual updates as periods, categories, or selected measures change.</p>
Useful visual-formatting capability with practical scenarios, but limited architectural impact and no performance or governance analysis.
<p><strong>Mirrored Azure Monitor</strong> and <strong>Mirrored AWS Glue catalog</strong> are preview capabilities that expose external data through OneLake without conventional ingestion and duplicate storage.</p><p>Azure Monitor mirrors selected Log Analytics tables, while AWS Glue mirrors metadata for supported Apache Iceberg tables stored in Amazon S3. Fabric uses connections, mirrored metadata, and OneLake shortcuts so Fabric workloads can analyze the data while it remains in its source environment.</p><p>The zero-copy model reduces pipeline and storage overhead, but preview support, authentication, source permissions, and Iceberg compatibility determine whether it is suitable for production.</p>
Preview mirroring reduces cross-cloud integration work, but source support, security, and operational limits require validation.
<p><strong>Change Event Streaming (CES)</strong> captures committed inserts, updates, and deletes from Fabric SQL database and continuously sends structured events to Fabric Eventstream or Azure Event Hubs. It uses transaction-log capture, includes operation and key information plus before-and-after values, and is configured with T-SQL stored procedures.</p><p>Eventstream decouples consumers such as Eventhouse, Real-Time Dashboards, Activator, and Lakehouse pipelines. SQL tracks delivery, retries transient failures, and splits large events for destination limits. CES is preview and also listed for Azure SQL Database, Azure SQL Managed Instance, and SQL Server 2025, so support boundaries and production behavior need verification.</p>
Preview CES offers concrete near-real-time SQL change streaming architecture, but availability and production limits remain unspecified.
<p><strong>Eventhouse</strong> combines streaming ingestion, time-series storage, and near-real-time querying in a database environment optimized for high-volume events. Automatic indexing and time-based partitioning reduce the manual tuning normally required for continuously arriving data, while KQL supports structured, semi-structured, and free-text analysis.</p><p>Enabling <strong>OneLake availability</strong> exposes a logical Delta representation to other Fabric engines, including Power BI Direct Lake, Lakehouse, Warehouse, and notebooks. This reduces pipeline duplication, but it does not remove the need to design ingestion permissions, workspace boundaries, retention, and query-cost controls. Eventhouse can also provide a query endpoint over existing Lakehouse or Warehouse data.</p>
Practical Eventhouse architecture overview explains ingestion, KQL, OneLake availability, and real-time analytics trade-offs.
Crossjoin
Aug 2, 2026
3/5 · Tier B
AI ReadinessReport AuthoringSemantic Model
deep-dive
<p><strong>Copilot in Excel</strong> can query a Power BI semantic model and create worksheets containing extracted fact data and analysis results. In the demonstrated ABC analysis, Copilot generated a DAX query, copied transaction data into Excel, and performed much of the remaining logic with Excel formulas.</p><p>The result was correct and accessible to Excel users, but the extracted worksheet was static rather than a live-connected table. The workflow therefore resembles an AI-assisted <strong>Export to Excel</strong>, not a continuously synchronized semantic-model analysis. Copilot also did not reuse the model’s DAX user-defined function in this test.</p>
Shows practical Copilot analysis against semantic models, with important static-export and formula-generation limitations.
<p>This monthly spotlight aggregates community articles, videos, and practical guidance across Fabric administration, engineering, governance, AI, Real-Time Intelligence, and Power BI. Its value is discovery rather than technical analysis: the extract supplies short descriptions and links to many unrelated resources, without presenting implementation details, measurements, limitations, or validated outcomes.</p>
Community roundup links to potentially useful articles but provides little substantive guidance itself.
<p>22 features - 2 must-know - 11 worth knowing.</p><ul><li>DAX user-defined functions (Generally Available)</li><li>Power BI Enhanced Report Format (PBIR) Default-On update</li></ul>
2 must-know of 22 features
Fabric Updates
Jul 31, 2026
5/5 · Tier A
Spark RuntimePySpark
announcement
<p><strong>Fabric Runtime Release Channels</strong> add an early-access Spark runtime alongside the default production channel. Teams can run representative notebooks and Spark Job Definitions against upcoming libraries, dependencies, security patches, and operating-system changes before Microsoft promotes them to default.</p><p>Opt-in uses <code>spark.fabric.pools.skipStarterPools=true</code> and <code>spark.computeConf.runtime.releaseChannel=earlyAccess</code> in an Environment item. Channel-specific release notes and the session’s VHD name provide evidence of what was tested. Removing the setting returns workloads to the default channel, making this a practical validation path without permanently changing production configuration.</p>
Runtime release channels provide a concrete production safeguard against untested Spark library, dependency, security, and operating-system changes.
<p>31 features - 18 must-know - 10 worth knowing.</p><ul><li>Fabric Runtime Release Channels (Preview)</li><li>Fabric-CICD tool v1.2.0 – new bulk publish mode (Preview)</li><li>Microsoft Fabric Runtime 2.0 (Preview)</li><li>Spark Diagnostic Emitter: Spark 4.1 Runtime Support and New Log Ingestion API</li><li>Efficient Scaledown with the remote shuffle manager (Preview)</li><li>Customer-managed key encryption for Spark shuffle data on disk (Generally Available)</li><li>Analytics and Insights for Materialized lake views (Generally Available)</li><li>Introducing Event-Driven Refresh for Materialized Lake Views (Preview)</li><li>AI functions: new models, no package dependency, better usage stats (Generally Available)</li><li>Lakehouse table health check (Generally Available)</li><li>Usage-based resource estimations (Generally Available)</li><li>Eventstream connector private network support (Generally Available)</li><li>Azure Event Hubs source in Eventstream now supports workspace identity authentication (Preview)</li><li>Custom CA and mTLS support in Eventstream streaming connectors (Generally Available)</li><li>Introducing the Oracle CDC connector for Eventstream (Preview)</li><li>Eventhouse update policies now support referencing accelerated shortcuts in update policies (Generally Available)</li><li>Shortcut Tables in Eventhouse Now Automatically Synchronize Schema Changes</li><li>Cross-domain intelligence with Azure Monitor data in Microsoft Fabric (Preview)</li></ul>
18 must-know of 31 features
Fabric Updates
Jul 30, 2026
5/5 · Tier A
⚠️
Breaking ChangeAI ReadinessGovernance
announcement
<p>The integration between <strong>Copilot in Power BI</strong> and Fabric data agents retires on <strong>August 26, 2026</strong>. After that date, Copilot in Power BI will no longer route user questions to Fabric data agents, although the underlying data-agent experience remains available.</p><p>The change results from retirement of the OpenAI Assistant API used by the integration; Microsoft is not replacing this specific connection. Supported alternatives include the Fabric data-agent experience, Microsoft Foundry, Copilot Studio, Microsoft 365 Copilot, and an MCP server endpoint. Existing permissions and configured security controls continue to apply through the selected replacement experience, but access behavior must be validated in that host.</p>
A dated retirement removes a Power BI integration and requires migration planning for existing data-agent workflows.
<p>Microsoft positions Fabric as a unified, AI-enabled lakehouse spanning OneLake, Power BI, Spark, Warehouse, Real-Time Intelligence, and agent experiences. The post highlights open formats, shortcuts, mirroring, cross-engine governance, and customer consolidation examples, including LSEG’s reported 33-petabyte estate and UNC Health’s claimed care-gap review improvement.</p><p>These statements are strategic positioning and customer evidence rather than a technical evaluation of architecture, capacity cost, security implementation, or migration complexity. The reported benefits should therefore be treated as directional claims requiring independent validation against a client’s workload and operating model.</p>
Vendor recognition and customer claims provide strategic positioning but limited independently actionable Fabric guidance.
<p><strong>Workspace Outbound Access Protection (OAP)</strong> makes the workspace an outbound security boundary for Eventstream and supported Real-Time Intelligence items. When enabled, outbound connections are blocked unless administrators approve them through data connection rules. Rules can permit trusted external sources, cross-workspace destinations, or other specified endpoints.</p><p>OAP does not affect processing inside the workspace: filtering, aggregation, grouping, and similar Eventstream operators continue to run. Same-workspace Fabric destinations generally require no additional approval, while external and cross-workspace connections do. The feature is in preview and complements, rather than replaces, identity, workspace permissions, and private networking controls.</p>
Preview outbound controls materially change Eventstream connectivity, workspace architecture, and approval requirements for sensitive streaming data.
<p><strong>User Data Functions</strong> provide serverless Python execution for APIs, integrations, orchestration, AI calls, and Fabric data access. Combined with <strong>Business Events</strong>, they form an action layer that reacts to meaningful signals such as inventory shortages, abandoned carts, or shipment delays.</p><p>The pattern uses Real-Time Hub for event discovery and subscriptions. A function publishes or consumes a typed payload through an Event Schema Set; Fabric creates the CloudEvents envelope and handles routing, deduplication, and delivery to subscribers.</p><p>The article is an applied architecture guide rather than a capacity or reliability analysis. Production designs still need explicit handling for retries, idempotency, downstream failures, event ordering, authentication, and function execution limits.</p>
It presents a practical event-driven UDF pattern with typed payloads, integrations, and realistic automation scenarios.
<p><strong>Real-Time Dashboard</strong> is generally available for monitoring operational data from Eventhouse. Live refresh can run at intervals as low as 10 seconds, with dashboard pages composed of query-backed tiles.</p><p>Users can filter, drill into data, and explore underlying results. Tile thresholds can trigger Activator notifications or workflows, while Copilot can generate queries and visuals from natural-language descriptions; that Copilot authoring experience remains in preview.</p><p>The feature closes the gap between scheduled reporting and operational monitoring, but freshness depends on the Eventhouse ingestion path, query performance, and the selected refresh interval.</p>
Real-Time Dashboard is generally available with ten-second refresh, alerts, exploration, and preview Copilot authoring.
<p><strong>Fabric Embed</strong> is in preview for embedding Real-Time Dashboards in JavaScript or TypeScript web applications through the Fabric Embed SDK. The initial experience uses delegated Microsoft Entra ID authentication and supports Fabric F SKUs and eligible trial capacity.</p><p>Authorization continues to use the signed-in user’s existing workspace, item, data-source, and Microsoft Entra permissions. Embedding does not create an additional entitlement or grant access to users who cannot open the dashboard directly in Fabric.</p><p>The preview does not support other Fabric item types, service-principal authentication, or app-owns-data embedding. SDK packages, APIs, events, and configuration contracts may change before general availability.</p>
The preview enables delegated embedding of Real-Time Dashboards, with clear authentication, item-support, and access boundaries.
<p><strong>Capacity-level Spark job concurrency and queue monitoring</strong> is generally available alongside the existing workspace-level view. Administrators can compare each workspace’s maximum capacity-unit consumption across 24-hour, 7-day, or 30-day windows.</p><p>The combined views distinguish workspace pool limits, capacity saturation, and contention caused by other workspaces. They also make HTTP 429 throttling diagnosable instead of leaving teams to infer the cause from queued jobs.</p><p>Usage trends support concrete actions: adjust workspace pool limits, reschedule overlapping pipelines, increase capacity limits, enable autoscale, or right-size the SKU. Maximum usage highlights peaks, so it should be interpreted alongside concurrency timing and workload duration.</p>
GA capacity-level Spark concurrency monitoring exposes workspace peaks, contention, queue causes, and actionable sizing signals.
<p><strong>Azure Monitor Logs mirroring into Fabric</strong> is available in preview. Log Analytics telemetry becomes available in OneLake alongside ERP, CRM, and other operational data without duplicating the logs, with near-real-time availability.</p><p>The mirrored data can feed <strong>Eventhouse</strong>, Real-Time Dashboards, Spark, Power BI, and Operations Agent workflows. The article states that Analytics, Basic, and Auxiliary Azure Monitor log tiers are supported, but it provides few implementation limits, pricing details, or measured latency figures.</p>
Preview mirroring connects Azure Monitor telemetry with business data for real-time cross-domain decisions.
<p><strong>Customer-managed key encryption</strong> now extends to temporary data written by Fabric Spark jobs. Local-disk spill data and shuffle data retained during execution use the customer-supplied key, in addition to encryption already protecting Fabric data at rest.</p><p>Fabric references the key through its identifier and connects it to cluster storage with a <strong>disk encryption set</strong>. The protection is infrastructure-level, so existing notebooks, Spark jobs, and pipelines require no code changes. Revoking or removing access to the key can prevent protected data from being read, making key availability an operational dependency.</p>
Customer-managed keys now encrypt Spark local-disk and shuffle data, closing an important processing-stage compliance gap.
<p><strong>Planning in Microsoft Fabric IQ</strong> is generally available. Planning connects budgets, forecasts, targets, and scenarios to governed enterprise data and Power BI semantic models, with planning data written back to Fabric SQL.</p><p>The release adds public CRUD and listing APIs, Git and deployment-pipeline integration, Variable Library support, service-principal authentication, direct connectivity to semantic models and Fabric SQL, and IBCS-certified reporting capabilities. PowerTable adds governed reference-data workflows with row- and column-level permissions, approvals, history, and writeback; detailed capacity and metering mechanics are not provided in the extract.</p>
Fabric Planning reaches general availability with APIs, Git, deployment pipelines, semantic-model connectivity, and governed writeback.
Fabric Updates
Jul 28, 2026
3/5 · Tier B
AI ReadinessAPI Automation
announcement
<p>Fabric data agents now connect to Microsoft Foundry through <strong>Model Context Protocol (MCP)</strong>, appearing as tools that Foundry agents can invoke for governed OneLake data. Builders can discover agents through the OneLake Catalog instead of entering workspace and artifact IDs manually, and one Foundry agent can use multiple specialized Fabric data agents.</p><p>Foundry Observability adds traces and logs for tool calls, timing, inputs, outputs, and status. This improves troubleshooting as agents move toward production, although the announcement does not specify quotas, latency expectations, token costs, or answer-quality controls. The rollout is regional and expected to complete over the stated release window.</p>
Fabric data agents gain MCP discovery, multi-agent composition, and tracing, but operational limits and costs remain unspecified.
Fabric Updates
Jul 28, 2026
4/5 · Tier A
Spark RuntimePySparkCapacity
deep-dive
<p><strong>Fabric’s Native Execution Engine</strong> adds automatic optimizations for Spark workloads. Broadcast hash tables can be reused per executor, ranking window functions such as <code>rank</code> and <code>dense_rank</code> remain in the native execution path, and reused common table expressions are materialized automatically.</p><p>The features are enabled by default when the supported runtime and engine are available, requiring no query changes. Ranking-intensive benchmarks improved by up to 16 percent, while CTE-heavy and repeated-join workloads may benefit differently. Faster execution can reduce elapsed time and capacity consumption, but the extract gives no universal workload-level savings guarantee.</p>
Native Execution Engine optimizations reduce repeated work, keep ranking native, and report up to sixteen-percent benchmark gains.
<p><strong>OneLake security</strong> is now generally available across all existing and new OneLake items; the previous opt-in experience is removed. Row-level security now extends to <strong>Eventhouse</strong> and Graph Database, while the redesigned column-level security interface supports searching, filtering, and bulk permission edits.</p><p>All five OneLake security APIs are GA, including bulk role updates. SQL Analytics Endpoint now defaults to <strong>User’s identity mode</strong>, reducing latency, improving nested-group handling, and addressing service-principal limitations. The catalog’s Secure tab also performs better with large numbers of secured items and roles.</p>
OneLake security reaches GA with broader engine coverage, security APIs, identity sync, and mandatory rollout.
<p><strong>semantic-link-labs 0.17.0</strong> adds interactive tools for identifying unused semantic-model objects and migrating Import or DirectQuery models to Direct Lake with guided compatibility checks.</p><p>The migration workflow supports in-place conversion when unsupported features are absent, and can target either a Lakehouse or Warehouse. New analysis features expose Delta statistics for Direct Lake tables, while lineage and VertiPaq tooling receive usability and rule updates.</p>
Adds interactive Direct Lake migration, usage-based model cleanup, Delta analysis, and support for lakehouse or warehouse sources.
Fabric Updates
Jul 27, 2026
4/5 · Tier A
Spark RuntimePySpark
announcement
<p>Fabric Spark’s <strong>Native Execution Engine</strong> now reads CSV files through a purpose-built Velox parser instead of falling back to Spark’s default reader. SIMD parsing and structural indexing keep CSV data in the vectorized execution pipeline, eliminating row-to-columnar conversion overhead.</p><p>Benchmarks show a 20 GB CSV read/write workload falling from 56 seconds to 34–38 seconds, with up to 2× gains on TPC-DS, approximately 35% on TPC-H, and around 20% end-to-end. Results depend on schema complexity, file size, and data distribution. Fabric Runtime 3.5 or later and an enabled Native Execution Engine are required.</p>
Native Execution Engine adds a CSV reader with measured speedups, runtime prerequisites, and automatic acceleration.
SQLBI
Jul 27, 2026
4/5 · Tier A
Report AuthoringSemantic Model
deep-dive
<p><strong>Visual calculations</strong> can improve performance when a calculation reuses values already returned by the visual’s source query. In the example, a year-over-year distinct-customer calculation uses <strong>PREVIOUS</strong> over the visual table instead of issuing another model query for each year, reducing execution time from approximately 13 seconds to 6 seconds on a 23-million-row Contoso model.</p><p>The benefit is not universal. Visual calculations add a <strong>VISUAL SHAPE</strong> and may trigger densification, which restores otherwise omitted blank combinations. With 67 stores, 2,517 products, and 10 years, the virtual table can reach 1,686,390 rows, making formula-engine processing substantially more expensive.</p><p>Use them when a small visual table contains expensive base measures and the follow-on calculation can operate locally; test wide, sparse matrices carefully.</p>
Quantifies visual-calculation gains and regressions, including 13-to-6-second improvement and densification costs at 1.7 million rows.
Fabric Updates
Jul 27, 2026
5/5 · Tier A
⚠️
Breaking ChangeAI ReadinessAPI Automation
release-notes
<p>OpenAI is scheduled to retire the <strong>Assistants API</strong> on August 26, 2026. Applications that directly create assistants, manage threads, and submit runs for Fabric data agents must migrate to the supported data-agent MCP endpoint; the Assistants API is not compatible as a drop-in replacement for the Responses API.</p><p>Fabric portal users require no action, and SDK integrations are being moved internally to the Responses API. Direct REST or Assistants API consumers must obtain the published agent’s MCP server URL and tool name, then validate the complete integration before retiring the old path. Agent configuration remains unchanged, but internally migrated experiences may have a one-time conversation-history reset.</p>
Direct Assistants API integrations must migrate to MCP before August 26, 2026, making this an actionable breaking change.
<p><strong>VNET Data Gateway Autoscaling</strong> is in Preview and dynamically adjusts active gateway nodes between administrator-defined minimum and maximum limits. The service evaluates workload-specific signals every five minutes, scales out after sustained pressure, and scales in after sustained low demand; the scaling frequency cannot be tuned.</p><p>Administrators can combine autoscaling with start, stop, restart, and inactivity-timeout controls. A gateway may start at its configured maximum size, then scale based on demand. When it stops after the configured idle period, gateway billing pauses. Maximum node count is also constrained by available subnet IP addresses, so network capacity remains part of sizing.</p>
Introduces demand-based VNET gateway scaling with measurable sampling, node caps, shutdown controls, and billing implications.
<p><strong>DAX user-defined functions</strong> can encapsulate complex analytical operations, including table-valued ABC analyses, for reuse by M365 Copilot. When the semantic model’s AI instructions explicitly identify the function and explain when to use it, Copilot generated the expected DAX consistently.</p><p>This matters because Copilot may otherwise produce syntactically valid but inefficient queries, add unnecessary measures, or interpret vague requests differently between users. The approach does not guarantee optimal performance or eliminate the need to test generated queries, and describing the function only in its metadata was insufficient in the demonstrated case.</p>
Demonstrates how DAX UDFs constrain Copilot behavior, improve consistency, and potentially control query performance and CU cost.
<p>A streaming pattern captures operational cart changes, flattens CDC records with DeltaFlow, and processes warm-abandoned carts in a Structured Streaming notebook. <strong>AI Functions for PySpark</strong> extracts intent, product attributes, urgency, and price sensitivity into columns that downstream subscribers can consume.</p><p>The enriched row is written to Gold and emitted as a Business Event, allowing email, on-site personalization, CDP updates, and merchandising monitoring to react from the same event. A Direct Lake report can read the enriched Gold data without scheduled refresh. The design assumes consent, suppression, and PII-minimization policies, but provides little detail about AI operator latency, failure handling, throughput, or preview limitations.</p>
Practical real-time pattern combines CDC, Spark AI enrichment, Business Events, and Direct Lake, but omits limits and costs.
<p><strong>Incremental Liquid Clustering</strong> in Fabric Spark Runtime 2.0 changes OPTIMIZE from broad partial-Z-Cube rewrites to targeted maintenance. The engine selects unclustered, undersized, poorly clustered, deletion-vector-heavy, or overlapping files while leaving healthy files untouched.</p><p>This addresses the severe write amplification possible in Runtime 1.3, where appending a tiny batch could repeatedly rewrite a much larger partial Z-Cube. Normal <code>OPTIMIZE</code> uses incremental maintenance; <code>OPTIMIZE FULL</code> remains available for broader layout resets.</p><p>Auto Reclustering detects overlapping file ranges and can include existing clustered files in a rewrite. Its defaults are four offending files and a 0.75 overlap threshold; lowering either value improves layout sooner but increases write cost. Incremental clustering and Auto Reclustering are enabled by default.</p>
Runtime 2.0 materially reduces liquid-clustering write amplification with explicit thresholds and maintenance behavior.
Power BI Updates
Jul 23, 2026
5/5 · Tier A
⚠️
Breaking ChangeMigrationAPI Automation
announcement
<p>The Power BI integration that created reports through SharePoint’s <strong>Visualize the List</strong> and <strong>Visualize the Library</strong> features is retired. No new reports were allowed from November 11, 2025, and previously published reports became unavailable on December 11, 2025.</p><p>Replacement paths are creating a semantic model directly from the SharePoint list for quick reporting, or connecting from Power BI Desktop for advanced transformation and modelling. Administrators can use the Admin Groups API to locate integration-created workspaces, commonly identified by names ending in <code>SPList</code>.</p>
The integration is retired, existing reports become unavailable, and admins receive an API-based discovery path for migration.
<p>26 features - 11 must-know - 6 worth knowing.</p><ul><li>Live editing Direct Lake semantic models with Power BI Desktop (Generally Available)</li><li>TMDL view (Generally Available)</li><li>Standalone Copilot to default-on</li><li>Auto-selection of Copilot workspaces</li><li>Enhanced DAX Time Intelligence (Preview)</li><li>Editing semantic models in the Power BI Service (Generally Available)</li><li>Download PBIX of XMLA altered semantic models</li><li>Fabric Notebooks for Power BI: Best Practices Analyzer and Memory Analyzer (Generally Available)</li><li>Direct Lake on OneLake + import in web modeling (Preview)</li><li>Direct Lake on OneLake + import in Desktop</li><li>Refresh data or schema options in Power BI Desktop</li></ul>
11 must-know of 26 features
Power BI Updates
Jul 23, 2026
5/5 · Tier A
⚠️
Breaking ChangeMigrationReport Authoring
release-notes
<p>16 features - 1 must-know - 8 worth knowing.</p><ul><li>Bing Map deprecation plan & Update in Power BI</li></ul>
1 must-know of 16 features
Power BI Updates
Jul 23, 2026
2/5 · Tier B
Report Authoring
announcement
<p>The new Card visual receives re-engineered responsive rendering before general availability. Resizing now preserves section proportions and prevents lower sections from being cut off, although existing short-height cards may change appearance.</p><p>GA also exposes callout sizing and background formatting independently of the divider, alongside hero images, collage layouts, rounded corners, and expanded section styling. Existing reports need deliberate preparation because the new rendering and defaults can alter established layouts.</p>
The article prepares authors for Card visual GA behavior, but offers limited detail beyond presentation and layout changes.
Power BI Updates
Jul 23, 2026
5/5 · Tier A
⚠️
Breaking ChangeReport Authoring
release-notes
<p>21 features - 1 must-know - 13 worth knowing.</p><ul><li>Announcing deprecation of R and Python visuals in Embed for your customers solution</li></ul>
1 must-know of 21 features
Power BI Updates
Jul 23, 2026
5/5 · Tier A
⚠️
Breaking ChangeMigrationSemantic Model
release-notes
<p>16 features - 5 must-know - 6 worth knowing.</p><ul><li>Deprecation of Power BI Q&A</li><li>Power BI Enhanced Report Format (PBIR) as Default Format</li><li>SQL Server Reporting Services Report Viewer Web Part Support Discontinuation</li><li>Updates on Editing Semantic Models in the Power BI Service</li><li>Chinese coded character set (GB18030-2022) support</li></ul>
5 must-know of 16 features
Power BI Updates
Jul 23, 2026
5/5 · Tier A
⚠️
Breaking ChangeGovernanceMigration
release-notes
<p>21 features - 6 must-know - 5 worth knowing.</p><ul><li>Deprecation of legacy Import Excel & CSV experience in Power BI Service</li><li>Deprecation of Hierarchies in Power BI scorecards</li><li>Deprecation of SSRS, PBIRS and SSAS Management Packs in SCOM</li><li>Deprecation of the Simba Vertica ODBC driver</li><li>TABLEOF and NAMEOF DAX functions</li><li>Fabric Copilot capacity tenant setting default update</li></ul>
6 must-know of 21 features
<p><strong>OneLake resource instance rules</strong> are generally available for allowing specific Azure resource instances to access a workspace over the public endpoint. Workspace administrators add trusted resource ARM IDs, and OneLake validates the calling resource identity before evaluating data permissions.</p><p>The rules complement rather than replace <strong>Private Link</strong>, IP firewall rules, and identity-based permissions. They reduce dependence on changing IP ranges and avoid requiring private networking for every Azure Databricks, Azure SQL, Data Factory, Event Grid, or Machine Learning integration, while preserving a layered security model.</p>
GA resource identity rules materially change how Azure services can access OneLake without broad IP allowlists or private networking.
Fabric Updates
Jul 23, 2026
1/5 · Tier C
AI ReadinessGovernance
announcement
<p>Fabric is positioned as a unified foundation for relational, analytical, real-time, document, graph, and vector workloads, with shared governance across those experiences. The post cites Forrester recognition and customer outcome figures, including reduced latency and manual work, but does not explain the evaluated configurations, workload boundaries, or reproducible implementation details.</p><p>It also presents Fabric IQ, data agents, AI Functions, and Rayfin as parts of an AI-oriented platform direction. The material is strategic positioning rather than a technical change log, sizing guide, or architecture deep dive.</p>
The post is a vendor recognition announcement with broad positioning and customer claims, but little actionable Fabric behavior or implementation detail.
Power BI Updates
Jul 23, 2026
3/5 · Tier B
Report Authoring
deep-dive
<p>A new <strong>Customize theme</strong> pane (preview) appears when no visual is selected. It sets report-wide defaults for:</p><ul><li><strong>Theme settings</strong>: import/export custom themes and choose the base theme</li><li><strong>Colors</strong>: data, structural, sentiment and divergent colors</li><li><strong>Text</strong>: title, callout, label and general styles</li><li><strong>Visual properties</strong>: border, corner rounding, shadow, background, padding, header icons and tooltips</li><li><strong>Page</strong>: default size, background and wallpaper</li><li><strong>Filter pane</strong> styling</li></ul><p>New visuals inherit these settings, as do visuals reset to default. Visual-level overrides are left untouched.</p><p>Built-in color-only themes become <strong>color palettes</strong> that can layer on any base theme. Existing reports do not pick up new default updates automatically; you apply them with <strong>Update theme</strong>. A 'Save base theme selection' toggle pins the base theme in exported JSON. To use the pane, enable it with on-object interaction under Preview features.</p>
Report-wide theme customization changes how branded report estates are maintained.
Fabric Updates
Jul 22, 2026
3/5 · Tier B
Spark Runtime
announcement
<p><strong>Lakehouse Query Explorer</strong> is generally available as an integrated Spark SQL editor inside the Lakehouse explorer. It keeps schemas, tables, shortcuts, and lakehouse objects visible while users inspect data, test joins, compare query tabs, review results, and create views.</p><p>It complements rather than replaces notebooks or the SQL analytics endpoint. Query Explorer suits interactive validation and lightweight SQL; notebooks remain appropriate for PySpark, multi-step transformations, scheduling, and orchestration. Query tabs and history do not persist, so useful logic should be saved as a view or moved into code-managed assets.</p>
GA Query Explorer improves in-flow Spark SQL validation and view creation, with clear boundaries around notebooks, scheduling, and SQL endpoint consumption.
Fabric Updates
Jul 22, 2026
2/5 · Tier B
AI ReadinessSemantic Model
opinion
<p>Conversational analytics is positioned around using Power BI semantic models as the trusted business context for AI answers. Fabric IQ exposes that context to Microsoft 365 Copilot and other experiences, while Prep for AI instructions can add semantic-model-specific guidance.</p><p>Fabric data agents provide a more specialized option: teams can combine semantic models, lakehouses, mirrored databases, shortcuts, graph data, and ontologies, then evaluate responses for domain-specific use cases. The post describes an intended convergence of Power BI Copilot with Microsoft 365 Copilot, but does not specify migration timing, licensing, response-quality thresholds, or detailed permission behavior.</p>
It outlines Fabric IQ, Copilot, and data-agent direction, but offers few measurable limits, costs, accuracy controls, or deployment mechanics.
<p>22 features - 3 must-know - 8 worth knowing.</p><ul><li>Deprecation of Old File Picker experience in Power BI Desktop</li><li>CRUD APIs (REST) for paginated reports</li><li>TMDL View on the Web</li></ul>
3 must-know of 22 features
Power BI Updates
Jul 21, 2026
3/5 · Tier B
AI ReadinessSemantic ModelGovernance
announcement
<p><strong>Power BI semantic models</strong> can provide governed answers directly in Microsoft 365 Copilot Chat, so users can ask business questions without opening a report. Responses are grounded in model measures, relationships, and business definitions rather than raw-table interpretation.</p><p>Microsoft 365 Copilot Chat requires its own license, while Copilot Cowork supports longer-running analysis and follow-up actions on a pay-per-use basis. The Power BI Standalone Copilot preview is expected to move toward the Microsoft 365 Copilot experience, while Report Copilot inside Power BI remains supported.</p><p>The practical constraint is model quality: unclear measures, weak relationships, missing descriptions, and ambiguous business terms reduce answer reliability. Preview status, permission handling, sensitivity labels, processing scope, and consumption economics still require tenant-specific validation.</p>
Describes governed Power BI answers in Microsoft 365 Copilot, licensing boundaries, preview experiences, and planned Copilot consolidation.
<p><strong>semantic-link-labs 0.16.0</strong> adds gateway restart and shutdown operations, OneLake tier management, and an interactive semantic-model lineage view. The lineage interface identifies broken downstream report visuals and filters, applies bulk fixes, and supports rebinding reports to another semantic model.</p><p>The release reorganizes OneLake functions under <strong>sempy_labs.onelake</strong> and removes <strong>add_table_to_direct_lake_semantic_model</strong>; callers must use <strong>add_direct_lake_tables</strong> instead. The removed function is the main compatibility concern for notebook automation and CI/CD code.</p>
Adds operational APIs, interactive lineage remediation, and removes a previously available Direct Lake function.
<p><strong>Snowflake Iceberg mirroring</strong> brings externally stored Iceberg tables into OneLake without copying their Parquet data. Fabric mirrors table metadata and creates a shortcut to customer-managed Amazon S3, Google Cloud Storage, or Azure Storage.</p><p>OneLake virtualizes the Iceberg tables as Delta Lake, allowing the SQL analytics endpoint, Spark, Power BI Direct Lake, and other Fabric engines to consume them. Snowflake-managed storage is not yet supported, and external, transient, temporary, and dynamic tables are excluded.</p><p>Setup requires Snowflake permissions, reachable storage, and a virtual network or on-premises data gateway when network isolation blocks Fabric. Mirroring compute is free, but Fabric query compute, applicable storage beyond the free allowance, and Snowflake compute remain chargeable. A mirrored database supports up to 1,000 tables, and Snowflake-granted object security must be recreated in Fabric.</p>
Snowflake Iceberg mirroring exposes important storage, security, networking, table-support, and cross-engine behavior.
<p><strong>Fabric Planning</strong> uses existing Fabric capacity rather than a separate subscription, but meters usage according to planning roles, time-bound user sessions, and automation jobs. Viewers, stakeholders, and planners contribute differently, while successful synchronization and sheet-update jobs use job-based billing.</p><p>A session associates a user, role, capacity, and planning artifact, then runs for 30 days, described as 730 hours. During preview, the Capacity Metrics app exposes usage by Planning experience, role, and automation job, but shows no charges; billing begins at General Availability, scheduled for July 2026.</p>
Preview documents role, session, job, and CU billing mechanics with measurable monitoring behavior and a planned GA transition.
<p><strong>Fabric Operations Agents</strong> can monitor Power BI activity exposed through Workspace Monitoring and alert on events such as failed DAX queries. The demonstrated agent queried the Workspace Monitoring KQL database, detected a non-cancelled visual error, and sent a notification to Teams.</p><p>The pattern required several iterations of agent instructions and manual KQL testing. Although the generated query returned the requested username, timestamps, error details, and object identifiers, the Teams message omitted most of that context. This makes instruction tuning and output validation essential before treating the agent as an operational control.</p>
Provides a concrete Operations Agent monitoring pattern, but the alert output still failed to include requested diagnostic identifiers.
<p><strong>OneLake Storage Tiers and Lifecycle Management</strong> are generally available. Data can remain online while moving between hot, cool, and cold tiers according to access patterns, reducing long-term storage cost for retained data.</p><p>Policies support multiple rules, filters, and conditions. <strong>AutoTierFromCoolToHot</strong> can move data to cool storage after inactivity and return it to hot storage when accessed, but cooler tiers introduce higher access and transaction costs.</p><p>Workspace defaults can place historical data directly into cool or cold storage. Storage reporting, diagnostic logs, and the OneLake MCP server help identify candidates; policies should reflect actual read patterns because an unexpected analytical scan can create avoidable consumption.</p>
GA storage tiering changes OneLake cost management and retention architecture, with practical access and transaction trade-offs.
<p><strong>OneLake item-size reporting</strong> is generally available in the Fabric portal. Workspace administrators can refresh a report showing storage by item, with sorting and searching by size.</p><p>The report separates visible data, system data, and soft-deleted data, and shows billing status. This gives administrators a more complete view than inspecting individual lakehouses or relying on workload-specific views.</p><p>The report is available under Workspace settings, OneLake, and Storage report, and requires workspace-admin access. It identifies oversized items and supports targeted investigation or lifecycle-management decisions, but it is reporting rather than automatic remediation.</p>
GA item-level reporting provides actionable OneLake storage attribution for workspace administration and lifecycle policy design.
Power BI Updates
Jul 16, 2026
3/5 · Tier B
Report Authoring
tutorial
<p>The preview <strong>date picker slicer</strong> combines relative ranges, manual ranges, single-date selection, and a movable range slider. Relative ranges can anchor to the first date, last date, or today with offsets, and they advance automatically as refreshed data changes.</p><p>The slicer distinguishes the selected relative range from the dates actually available in the data. When a date table extends beyond fact data, filtering the slicer to nonblank measure values prevents future or incomplete periods from entering the comparison. Relative behavior also persists when a visual is pinned to a dashboard.</p><p>Authors can make the control compact by hiding the header, slider, and summary, but should provide an alternative clear-all-slicers control. Creation is preview-only in Desktop; published reports can be viewed and edited in the service.</p>
Adds dynamic relative date behavior, full/manual range selection, dashboard persistence, and explicit handling for incomplete or future dates.
<p><strong>User Data Functions</strong> can act as governed API gateways for Fabric REST operations. A caller invokes the function, which retrieves a service-principal secret from Azure Key Vault, creates a scoped Fabric client, and triggers the permitted operation.</p><p>The pattern centralizes orchestration and keeps credentials out of applications, notebooks, pipelines, and reports. The service principal still defines the effective access boundary, so workspace, item, capacity, and API permissions must be intentionally scoped.</p><p>The example triggers a pipeline through the Job Scheduler API, but the same design can create workspaces and items, manage metadata, run jobs, and perform supported capacity operations. Key Vault and the Fabric generic connection become operational dependencies; incorrect workspace or vault permissions are the likely first failure points.</p>
Concrete User Data Function pattern centralizes Fabric REST automation, service-principal access, and Key Vault secret handling.
<p>Fabric Data Factory adds <strong>L2 workspace monitoring</strong>, an <strong>Operations Agent for Pipelines</strong>, and a <strong>Refresh Materialized Lake View activity</strong>, all in Preview. L2 monitoring extends pipeline-level status with activity-level execution details, diagnostics, and failure drill-down in the centralized monitoring event store.</p><p>The Operations Agent observes pipeline health, analyzes run data, and recommends remediation. RMLV activity lets orchestration refresh materialized lake views as part of an end-to-end pipeline rather than relying on separate manual configuration. Preview status means behavior, permissions, and operational coverage require validation before production standardization.</p>
Adds activity-level observability, AI-assisted troubleshooting, and materialized lake view orchestration to Fabric pipelines.
<p>A real-time pattern combines database CDC, DeltaFlow, Eventstream, Spark Structured Streaming, and <strong>AI Functions for PySpark</strong>. CDC captures inserts, updates, and deletes; DeltaFlow reshapes raw Debezium envelopes into source-table-shaped rows; a Spark notebook classifies free-text fields inside each micro-batch.</p><p>The enriched rows can fan out to Business Events, Eventhouse, and operational dashboards. Classification becomes an ordinary output column that downstream systems can filter, alert on, aggregate, or route. The design removes model-serving infrastructure, but inference latency, throughput, classification accuracy, label taxonomy, and downstream action reliability still need measurement before high-volume production use.</p>
Demonstrates a concrete CDC-to-action architecture combining Eventstream, Spark micro-batches, AI Functions, and operational destinations.
<p>5 features - 2 must-know - 1 worth knowing.</p><ul><li>Use Workspace Identity and Service Principals for production deployments</li><li>Get fresher data with low-latency sync</li></ul>
2 must-know of 5 features
SQLBI
Jul 13, 2026
3/5 · Tier B
Report AuthoringSemantic Model
tutorial
<p><strong>ISINSCOPE</strong> and <strong>ISATLEVEL</strong> both support different formatting rules at different hierarchy levels, but they operate in different layers. <strong>ISINSCOPE</strong> belongs in a semantic-model measure and detects query group-by columns; <strong>ISATLEVEL</strong> belongs in a visual calculation and inspects the visual hierarchy.</p><p>The measure approach is reusable across reports but requires semantic-model authoring rights. The visual-calculation approach leaves the model unchanged and lets report authors modify the rule locally, using visual-aware functions such as <strong>COLLAPSE</strong> and <strong>COLLAPSEALL</strong>. Conditions must be tested from the most specific hierarchy level to the most general.</p>
Explains a useful semantic-model versus report-layer design choice for hierarchy-aware formatting with concrete DAX patterns.
<p>OneLake Architectural Guidance organizes enterprise modernization around five patterns: data access unification, medallion architecture, governed data mesh, platform simplification, and external data sharing. The framework connects strategic outcomes to architectural patterns and then to Fabric capabilities.</p><p>Its central recommendation is to virtualize data where possible, standardize processing, support governed domain ownership, consolidate analytics workloads, and share data securely across organizational boundaries. The guidance is a useful framing tool, but the extract provides few concrete configuration choices, cost figures, security boundaries, or failure modes.</p>
Provides useful OneLake architecture patterns, but offers strategic guidance with limited implementation detail, limits, or measured evidence.
<p><strong>Direct Lake</strong> moves much of the refresh burden from semantic-model ingestion to preparing optimized Delta tables in OneLake. Framing is typically fast and consumes minimal capacity, but the cost of loading, transforming, partitioning, and optimizing the gold layer must be included in the comparison with Import refresh.</p><p>The main architectural benefits are avoiding Import refresh memory and timeout limits, reusing the same OneLake tables across multiple semantic models, and allowing SQL, Spark, notebooks, and Direct Lake to consume one governed data layer. A cited example loaded a 20-million-row fact table in Spark for roughly half the CU cost of an equivalent Import refresh, although results depend on workload design.</p><p>Direct Lake is not automatically superior: query performance and CU behavior differ from Import, and existing Import models that work well do not necessarily justify migration. The decision should compare end-to-end gold-layer processing, query load, security, and refresh frequency.</p>
Explains Direct Lake’s refresh, storage, reuse, cost, and workload trade-offs with concrete CU evidence and architectural consequences.
Power BI Updates
Jul 10, 2026
0/5 · Tier C
AI Readiness
opinion
<p>Microsoft highlights its 2026 Gartner recognition and positions Power BI semantic models as a foundation for governed analytics and AI experiences. The article cites adoption figures and references Copilot, Fabric IQ, agent skills, TMDL views, and Rayfin, but provides no implementation limits, costs, or independently measured behavior.</p>
This is vendor recognition and positioning content without implementation detail, limits, costs, or measurable platform behavior.
<p>The <strong>Oracle Database CDC source connector for Fabric Eventstream</strong> is in Preview. It captures inserts, updates, and deletes as structured events, including before-and-after row values, instead of repeatedly extracting complete table snapshots.</p><p>On first connection, the connector takes a consistent initial snapshot, then reads Oracle archived redo logs through <strong>LogMiner</strong>. Events can route in parallel to Eventhouse, Lakehouse, and downstream actions such as Activator alerts. Oracle Database 12c and later are supported, including 19c, 21c, and Autonomous Database. Production readiness depends on redo-log retention, source permissions, event ordering, schema evolution, and connector preview behavior.</p>
Adds preview Oracle CDC ingestion with LogMiner, initial snapshots, before-after records, and direct real-time Fabric routing.
<p>10 features - 3 must-know - 7 worth knowing.</p><ul><li>SQL operator (Generally Available)</li><li>Mirrored database change feed connector (Preview)</li><li>Eventstream as a Business Events publisher (Preview)</li></ul>
3 must-know of 10 features
<p><strong>Mirrored Database Change Feeds</strong> expose row-level inserts, updates, and deletes from mirrored Delta tables through Change Data Feed. Eventstream consumes those changes, filters and reshapes them with SQL operators, then routes the results to downstream Fabric services.</p><p>A manufacturing example keeps failed inspections, groups them into one-minute tumbling windows, and sends clusters to <strong>Activator</strong> for Teams, email, ticketing, or webhook actions. The raw stream separately feeds <strong>Eventhouse</strong> so analysts retain both passes and failures for yield calculations. The demonstrated end-to-end response is roughly 90 seconds, but preview connector availability and change-feed semantics require validation for each mirrored source.</p>
Concrete mirrored CDC-to-action pattern with SQL, windowing, Activator, Eventhouse, and approximately 90-second response.
SQLBI
Jul 8, 2026
1/5 · Tier C
AI Readiness
opinion
<p><strong>SQLBI’s 2026 AI guidance</strong> favors using generative AI to improve productivity while retaining human review and transparency around AI-generated technical content. The examples include subtitle translation, DAX analysis, testing, demonstrations, and recurring code-generation tasks.</p><p>The guidance emphasizes that DAX knowledge remains important because structured prompts and human validation are needed to correct inaccurate or inefficient output. It is an editorial policy and experience report rather than a Fabric implementation guide, with no product-specific limits, pricing, security model, or operational controls.</p>
Offers organizational AI-use guidance and disclosure practices, but provides no Fabric-specific limits, costs, governance, or implementation detail.
<p><strong>Fabric Data Warehouse GPU query acceleration</strong> is enabled with a workspace toggle and executes queries over OneLake Parquet data using GPUs. In this 100 GB TPC-H test on F64, single-user query latency stayed below 500 milliseconds, while high-concurrency throughput was the more significant result.</p><p>At 1,000 concurrent users, GPU execution delivered 101.4 queries per second versus 16.9 on CPU, with p95 latency of 15.1 seconds versus 97.7 seconds. The feature is in limited availability, depends on GPU capacity and regional support, and is less compelling for lightly used warehouses. The results are an independent experiment rather than a guaranteed customer benchmark.</p>
Measured GPU warehouse testing reports latency, concurrency, throughput, limited availability, and when acceleration justifies capacity economics.
Power BI Updates
Jul 7, 2026
3/5 · Tier B
Report Authoring
deep-dive
<p><strong>Shape Map</strong> is generally available for visualizing data against built-in or custom geographic boundaries. Authors can use built-in regions, upload TopoJSON or GeoJSON files, or reference centrally hosted map files through a URL measure.</p><p>Location-only maps support manual colors, rule or gradient conditional formatting, and field-driven colors. Adding a legend assigns categorical colors, while Color saturation shades regions by a numeric value. Projection choices such as Albers USA, Equirectangular, Mercator, and Orthographic affect geographic distortion and presentation.</p><p>Custom maps can be downloaded from the visual, edited, and uploaded again. The article does not cover file-size limits, refresh governance for remote URLs, or performance behavior for complex boundaries, so those remain deployment considerations.</p>
GA custom-boundary mapping adds practical geographic flexibility, URL reuse, projections, and data-driven coloring beyond standard map visuals.
Fabric Updates
Jul 6, 2026
2/5 · Tier B
AI ReadinessGovernanceAPI Automation
announcement
<p><strong>Rayfin</strong> is positioned as a Fabric application backend for experiences combining analytics, operational data, writeback, workflows, and AI. The proposed model combines semantic models, Fabric SQL databases, modern web experiences, and policy-based application authorization.</p><p>Most of the material describes planned investments rather than generally available behavior: connectors for existing Fabric sources, external Functions, notebook and pipeline invocation, richer Entra integration, TypeScript, version control, and secret management. Regional availability is currently limited, with West US 2 and North Central US identified as supported provisioning locations.</p>
Rayfin’s AMA outlines promising application, security, connector, and Functions roadmap items, but mostly describes future capabilities.
Power BI Updates
Jul 6, 2026
3/5 · Tier B
Report Authoring
deep-dive
<p><strong>Power BI tooltip options</strong> are generally available with progressively richer customization: extra tooltip fields, curated tooltip-only fields, sentence templates, chart-specific context, and fully custom report-page tooltips.</p><p>Sentence templates substitute field values, support field parameters, and can use a hierarchy-aware measure so text remains correct while users drill. Report-page tooltips inherit the hovered point’s filters but are read-only; interactive exploration requires drillthrough instead.</p>
GA tooltip controls provide practical visual-context customization, including curated fields, sentence templates, and hierarchy-aware hover text.
Fabric Updates
Jul 6, 2026
3/5 · Tier B
AI ReadinessGovernance
tutorial
<p><strong>Sensitivity labels</strong> from Microsoft Purview can provide AI skills with context about whether data is General, Confidential, Highly Confidential, or custom-classified. The label remains a protection and access-control mechanism, but the proposed pattern also uses it to prioritize sources, limit detail, or decline analysis.</p><p>A report-analysis skill might fully analyze General reports, summarize Confidential reports only for authorized users, and restrict Highly Confidential material to explicitly cleared executives. The approach improves relevance and reduces inappropriate context mixing, but it depends on consistent labeling and explicit skill rules; labels do not replace permissions or guarantee accurate AI responses.</p>
Sensitivity labels are proposed as AI context signals, improving source selection and response handling beyond access enforcement.
<p>The <strong>“database was evicted to balance the CPU load”</strong> error indicates that a Power BI service node experienced excessive CPU pressure and moved a semantic model to another node. Queries or refreshes running on that model are cancelled, even when that model is not the primary source of the pressure.</p><p>Isolated occurrences may be transient, particularly during aggressive load testing. Repeated occurrences during normal production usage point toward excessive CPU demand from the workload or semantic model; semantic-model scale-out may reduce concurrency pressure, but DAX and model optimization remain the underlying remedy.</p>
Describes a rare production failure where node CPU pressure evicts semantic models and cancels active queries or refreshes.
<p>The <strong>UC Delta API</strong> is a versioned, discoverable, intent-based REST interface for Delta table operations, introduced in open source Unity Catalog 0.5. Engines negotiate protocol versions and required table features through the catalog before reading or writing.</p><p>For managed tables, the client stages table metadata, receives catalog requirements and short-lived storage credentials, writes Delta data and commit files, then submits an intent list with preconditions for atomic catalog ratification. The wire format remains Delta-native rather than flattening schema and protocol details into a generic API.</p><p>The local examples use Spark and DuckDB against OSS Unity Catalog. This is an ecosystem and integration direction, not a Fabric tenant feature announcement; production adoption depends on catalog-provider support, API version, authentication, and managed-table compatibility.</p>
Important open ecosystem architecture introduces versioned catalog-mediated Delta commits, but Fabric applicability remains indirect.
<p><strong>Workspace Outbound Access Protection</strong> is in preview for Real-Time Intelligence workloads. It enables an administrator-controlled, default-deny posture for connections leaving a workspace, reducing data-exfiltration risk while retaining approved Fabric-native paths.</p><p>Behavior differs by item. Eventstream continues routing to supported items in the same workspace but blocks cross-workspace and external destinations. Eventhouse retains selected Event Hubs, OneLake, follower-database, and configured shortcut paths, while blocking unsupported external connections. KQL Queryset authoring and investigation continue, but Copilot-generated KQL is unavailable when protection is enabled.</p>
Provides concrete allow and block behavior for outbound protection across Eventstream, Eventhouse, KQL, dashboards, and Copilot.
<p><strong>OneLake shortcuts</strong> support two security patterns for zero-copy distribution. Passthrough shortcuts evaluate the signed-in user against the target, preserving source-owner control and per-user row or column restrictions.</p><p><strong>Delegated shortcuts</strong> access the target through an organizational identity or service principal. Security configured for the delegated identity and consuming shortcut combines through intersection, so consumers cannot exceed the delegated identity’s privileges. This reduces provisioning overhead for large audiences and supports cross-tenant or external storage scenarios.</p>
Explains passthrough versus delegated shortcut authorization, intersection behavior, scale, and cross-tenant governance implications.
Power BI Updates
Jul 2, 2026
4/5 · Tier A
Report AuthoringSemantic Model
deep-dive
<p><strong>Visual calculations</strong> are generally available for calculations defined directly on a visual rather than in the semantic model. They operate over the visual’s already-aggregated data, which simplifies patterns such as running sums, moving averages, previous-value comparisons, and percent-of-parent calculations.</p><p>Custom totals for tables and matrices use visual calculations to provide Sum, Average, Min, Max, Count, distinct count, or no total. The logic stays local to the visual and is supported in Desktop, the service, and embedded scenarios, but it is not reusable model logic like a measure.</p>
GA visual calculations add visual-scoped DAX, running patterns, and custom totals with concrete behavior and deployment implications.
<p><strong>semantic-link-labs 0.15.2</strong> refreshes the VertiPaq and Delta Analyzer interfaces with consistent styling and optional dark mode. DAX formatting now uses the newer daxformatter.com API, while several identity-resolution and dependency bugs are fixed.</p><p>The new <strong>ui-styling SKILL.md</strong> helps an AI generate Fabric Notebook-compatible interfaces matching the library’s tools. It is useful for internal tooling consistency, but the release contains little change to Fabric runtime behavior or workload economics.</p>
Mostly delivers UI polish, bug fixes, and a reusable notebook styling skill rather than platform behavior changes.
<p>6 features - 3 must-know - 3 worth knowing.</p><ul><li>Private network (vNet/On-Prem) support for connectors (Generally Available)</li><li>Apache Kafka and Azure Service Bus connectors (Generally Available)</li><li>Custom CA and mTLS support (Generally Available)</li></ul>
3 must-know of 6 features
Power BI Updates
Jul 1, 2026
1/5 · Tier C
AI ReadinessSemantic Model
opinion
<p>This is a positioning post. <strong>Power BI reports</strong> remain the drag-and-drop BI experience on semantic models. <strong>Fabric Apps</strong> (preview), built with <strong>Rayfin</strong>, offer an AI-first way to build managed web apps in TypeScript and CSS on the same semantic models, hosted in Fabric.</p><p>It also mentions the modeling MCP server, agentic report authoring through PBIP/PBIR, and the new <strong>Power BI Desktop bridge</strong> that shows external local edits live in Desktop. It gives no limits, cost or security detail.</p>
Positioning piece on Fabric Apps/Rayfin; no limits, cost or behavior detail.
<p>The <strong>Fabric data agent API</strong> is public, allowing developers to create, configure, update, and publish data agents from local code, CI/CD pipelines, services, portals, containers, and functions. The SDK uses Fabric authentication and the public management-plane API.</p><p>Automation covers agent instructions, data sources, example queries, and lifecycle operations. Git and deployment pipelines remain responsible for workspace-level promotion, while the API provides programmatic provisioning and updates. After publication, the <strong>MCP endpoint</strong> is the runtime consumption surface rather than the management API.</p>
Makes data-agent lifecycle management available outside Fabric with SDK examples, authentication, staging, publishing, and MCP separation.
<p><strong>Item Recovery</strong> is generally available for supported Fabric items. Deleted items enter a soft-deleted state and can be restored from the Workspace Recycle bin or REST API with their content, metadata, and permissions during a tenant-configured retention window.</p><p>Administrators can set retention up to 90 days and adjust it as policy changes. Coverage is limited to currently supported item types, and items are permanently deleted after the retention period expires. REST APIs and audit events can connect recovery to operational and compliance workflows.</p>
Adds GA soft-delete recovery, 90-day configurable retention, REST access, tenant policy, and supported-item coverage.
<p><strong>Power BI Desktop Bridge</strong> is a preview local server that lets external tools and agents inspect Power BI Desktop, reload on-disk PBIP/PBIR changes, and capture report-page screenshots. Its manifest exposes available capabilities, while state inspection identifies the open file and unsaved changes.</p><p>The CLI provides commands such as <strong>manifest</strong>, <strong>status</strong>, <strong>reload</strong>, and <strong>screenshot-all</strong>. The bridge is enabled by default but can be controlled through Desktop preview settings. It creates a fast edit–reload–visual-validation loop, though it remains a local Desktop automation mechanism rather than a service deployment API.</p>
The preview exposes concrete local APIs and CLI workflows for PBIR editing, reload, state inspection, and screenshot validation.
<p><strong>fabric-cicd v1.2.0</strong> adds Map and Paginated Report item support, bulk publishing through Fabric’s bulk import API, and multiple connection IDs for semantic-model binding.</p><p>Dynamic replacement variables and case-insensitive matching make <strong>find_replace</strong> more flexible. Logging is now opt-in through <strong>FABRIC_CICD_FILE_LOGGING_ENABLED</strong>, while sensitive files receive owner-only permissions on POSIX systems.</p>
Adds bulk deployment, semantic-model binding, new item types, and safer logging for Fabric CI/CD automation.
<p><strong>AI-assisted Synapse and Pipeline Migration skills</strong> move Spark pools, lake databases, notebooks, Spark job definitions, and orchestration pipelines into Fabric from the command line through REST APIs. The migration is idempotent, performs dependency-aware sequencing, validates results, and produces a migration report.</p><p>Lift-and-shift prioritizes speed by applying automatic mappings and skipping refactoring. Migrate-and-modernize offers decisions for environments, Lakehouse topology, Delta conversion, OneLake shortcuts, bindings, and code changes such as <strong>mssparkutils</strong> to <strong>notebookutils</strong>. The target runtime upgrades to Spark 3.5 Runtime 1.3.</p><p>Important gaps remain: GPU pools and C#/.NET notebooks have no direct equivalent, pipeline triggers must be recreated, and external <strong>abfss://</strong> paths require deliberate shortcut or code decisions.</p>
Detailed preview migration workflow covers mappings, refactoring, blockers, idempotency, runtime changes, and REST automation.
<p><strong>DAX user-defined functions</strong> expand like macros, allowing a function to effectively return <strong>REMOVEFILTERS</strong> when invoked as a <strong>CALCULATE</strong> filter argument. This encapsulates the columns that should be cleared from a calendar without duplicating the filter list throughout measures.</p><p>The pattern fixes a subtle calendar-based time-intelligence error: a weekday slicer changes <strong>MAX(Date[Date])</strong>, preventing <strong>ENDALIGNED</strong> from identifying the intended month end. The function must be called inside <strong>CALCULATE</strong>; it is not independently valid as a table expression or scalar result.</p>
Advanced DAX UDF pattern fixes calendar filter-keep errors and exposes macro-expansion behavior.
<p><strong>Outbound access protection (OAP)</strong> now governs Azure and Fabric event consumption across workspaces. When enabled, a consumer workspace cannot reach event sources in other workspaces unless administrators allow the <strong>Real-Time Events connector</strong> through data connection rules. Same-workspace consumption remains allowed.</p><p>OAP is enforced on the consumer workspace, not the source workspace. Existing cross-workspace consumers are placed into a paused state when the policy blocks them; events are retained for up to seven days and then dropped. Restoring delivery requires the connector rule plus deleting and recreating the consumer configuration.</p><p>The capability complements workspace and tenant private links, but it creates an operational dependency between network policy changes and event-flow continuity.</p>
Introduces enforceable cross-workspace event controls with a seven-day retention consequence when configurations pause.
<p><strong>DirectQuery</strong> becomes a credible default for very large Fabric models when fact tables exceed a couple of billion rows, dimensions exceed a couple of million rows, or high-cardinality distinct counts dominate. The proposed design keeps fact tables in the Warehouse or Lakehouse SQL endpoint, uses <strong>Dual</strong>-mode dimensions, and adds aggregation tables.</p><p>This can outperform Import or Direct Lake for aggregation-heavy queries because native aggregations currently depend on DirectQuery. The Warehouse engine may also handle distinct counts and sustained concurrency better than VertiPaq. Capacity economics matter: SQL generated for DirectQuery fact access is treated as background work and smoothed over 24 hours, while Import and Direct Lake query work is interactive and smoothed over shorter windows.</p><p>The trade-off is operational complexity. Tuning requires semantic-model, DirectQuery, and Warehouse expertise, and poorly designed relationships, aggregations, or generated SQL can erase the advantage. Load testing remains essential before using the pattern to avoid a capacity upgrade.</p>
Specific composite-model guidance connects billion-row scale, concurrency, engine behavior, and Fabric capacity economics.
Fabric Updates
Jun 26, 2026
5/5 · Tier A
Spark RuntimePySparkCapacity
deep-dive
<p><strong>Efficient Scaledown</strong> decouples Spark shuffle data from executor lifetime. Fabric can route large shuffle blocks to remote Azure Blob Storage, migrate blocks during executor decommissioning, keep small shuffles local, and use Adaptive Query Execution to adjust shuffle writes.</p><p>The preview requires the Native Execution Engine and Fabric Spark Runtime 3.5 or later. Microsoft reports a TPC-DS benchmark reduction from 3,724 to 2,121 VM-minutes, or 43%, and up to 57% runtime improvement for smart routing versus sending every shuffle remotely. The results are workload-dependent, and remote storage introduces an additional data path.</p>
Preview Spark capability includes configuration, runtime prerequisites, resilience behavior, and measured 43% compute reduction.
Fabric Updates
Jun 26, 2026
3/5 · Tier B
AI ReadinessGovernance
deep-dive
<p><strong>Fabric data protection</strong> combines Microsoft Purview sensitivity labels, protection policies, DLP, audit logs, OneLake Catalog insights, and DSPM for Fabric to reduce oversharing risks when Copilot and agents access analytics content.</p><p>Sensitivity labels can be manual, default, or mandatory, and protection can persist through exports to Office files and PBIX. DLP can detect sensitive information in semantic models and structured OneLake items, initially in simulation mode before alerts, policy tips, or the preview <strong>restrict access</strong> action is enabled.</p><p>These controls improve AI readiness by constraining AI responses through the user’s existing permissions, but they do not replace permission design or guarantee that classifications and detection rules are complete.</p>
Connects Fabric AI usage to Purview labels, DLP enforcement, access controls, auditability, and exposure assessments.
<p><strong>BCP API</strong> provides direct client-side bulk ingestion into Fabric Data Warehouse tables through C# <strong>SqlBulkCopy</strong>, Java <strong>SQLServerBulkCopy</strong>, the <strong>bcp.exe</strong> utility, and compatible integration connectors.</p><p>It is intended for data already held in an application, process memory, local file, or integration runtime where staging files first is impractical. Bulk batches avoid the overhead of row-by-row INSERT statements. When files can be staged, server-side <strong>COPY INTO</strong> remains the recommended managed loading pattern.</p><p>The capability is preview functionality and the article provides no throughput or cost benchmarks, so workload-specific testing is necessary before replacing an established ingestion path.</p>
Adds direct client-side bulk loading for Fabric Warehouse while clearly positioning COPY INTO as the preferred staged path.
<p>The June 2026 gateway release, version <strong>3000.322</strong>, adds Windows Web Account Manager authentication, updates Apache Log4j to 2.25.4, and introduces consent-based diagnostic uploads surfaced in Dataflow Gen2 runs. It also aligns gateway query behavior with the June 2026 Power BI Desktop release.</p><p>The changes improve authentication, vulnerability posture, and troubleshooting, but the release contains no material Fabric workload or architecture change.</p>
Provides routine gateway maintenance, security updates, and diagnostics improvements without major Fabric architecture or behavior changes.
<p>This monthly spotlight curates community content across Data Engineering, Data Science, governance, Real-Time Intelligence, databases, and AI. The entries range from beginner pipeline guidance to more relevant developments such as Unity Catalog tables in OneLake and Azure DevOps deployment for Fabric apps.</p><p>It provides pointers rather than detailed implementation evidence, limits, performance measurements, or cost analysis. Its main value is discovering potentially useful follow-up material from MVPs and Fabric Super Users.</p>
Community roundup highlights contributors, with limited detail beyond linked tutorials and emerging Fabric app content.
<p>The SQL analytics endpoint now exposes <strong>sp_get_table_health_metrics</strong> as a generally available T-SQL diagnostic for Lakehouse tables. It reports anomaly types, checkpoint and snapshot versions, physical and deleted rows, and file, row, and deleted-row distributions.</p><p>The procedure supports a check-then-act maintenance pattern: inspect table health, run <strong>OPTIMIZE</strong> only when anomalies such as small files or excessive deleted rows appear, and otherwise skip the Spark operation. The SQL analytics endpoint remains read-only, so remediation still occurs through Spark or the Lakehouse engine.</p>
GA T-SQL diagnostics enable conditional Lakehouse maintenance, reducing unnecessary OPTIMIZE compute and capacity consumption.
<p>3 features - 0 must-know - 0 worth knowing.</p>
0 must-know of 3 features
<p><strong>Protection policies</strong> are generally available for Power BI semantic models. A Microsoft Purview sensitivity label can specify which users or groups retain access, with enforcement occurring at runtime rather than through workspace-by-workspace permissions.</p><p>The semantic model is the enforcement point: blocking access prevents downstream reports from returning data, while labels can inherit from upstream Fabric items and propagate to connected reports. Reports and dashboards are not directly protected by the policy, so model lineage and label placement matter.</p>
GA protection policies enforce sensitivity-based access centrally through semantic models and downstream reports.
<p><strong>Entra ID SSO for the Denodo connector</strong> lets each DirectQuery viewer authenticate to Denodo with their own Microsoft Entra identity. Power BI forwards the user token through the on-premises data gateway, and Denodo validates the token directly.</p><p>This removes the Kerberos Constrained Delegation chain, domain-joined gateways, synchronized Active Directory identities, and high-privilege gateway service accounts. Denodo can apply its own row-level and column-level policies to the actual report user.</p><p>The capability is <strong>DirectQuery-only</strong>; Import models do not forward query identities. Tenant administrators must enable Denodo SSO and gateway Entra SSO, and the Denodo server and gateway must support the preview authentication flow.</p>
Preview adds end-to-end Entra identity for Denodo DirectQuery, removing Kerberos and Active Directory dependencies.
<p><strong>Fabric Data Factory</strong> provides pipelines and activities for orchestrating data across Snowflake, Databricks, BigQuery, Salesforce, SFTP, and Azure sources. A single workflow can combine cross-cloud extraction, Databricks notebook execution, and loading or merging into another warehouse.</p><p>The article also positions shortcuts, mirroring, copy jobs, lineage, and OneLake as options for reducing movement or centralizing governance. These patterns are useful architecture directions, but connector-specific limitations, egress costs, throughput, authentication boundaries, and operational ownership are not detailed.</p>
GA orchestration patterns connect major cloud platforms, but the article offers broad capability claims without implementation limits or measured costs.
Fabric Updates
Jun 22, 2026
5/5 · Tier A
Spark RuntimePySparkDelta & Parquet
deep-dive
<p><strong>Fabric’s Native Execution Engine</strong> now keeps Python and Scala UDFs, arrays, maps, and structs in its optimized native columnar execution path. Existing Spark notebooks and jobs require no syntax changes when the engine is enabled.</p><p>Reported gains reach <strong>5.76× for vectorized UDFs</strong>, 1.08–2.5× for complex UDFs, and 2.35× for selected TPC-DS workloads. Actual improvements depend on UDF shape, nesting depth, data volume, and cluster configuration; fallback alerts should still be checked.</p>
Measured Native Execution Engine gains address Python UDF, complex-type, and Spark execution bottlenecks.
<p><strong>Fabric Activator</strong> can now invoke a <strong>Copy job</strong> directly when a monitored condition occurs, and the capability is generally available. The direct connection removes the need for a pipeline solely to orchestrate event-driven copying.</p><p>Rules can react to events such as a new file, a table update, or a metric threshold. This avoids empty scheduled runs and reduces latency-versus-cost trade-offs, although the extract does not specify trigger deduplication, retry behavior, concurrency limits, or event volume costs.</p>
GA Activator-to-Copy Job triggering enables event-driven data movement without an intermediate pipeline.
<p><strong>Extreme DAX, Second Edition</strong> focuses on advanced, real-world DAX problems after covering familiar modeling and calculation fundamentals. Its strongest material reportedly includes difficult topics such as Auto-Exist and newer language capabilities including user-defined functions.</p><p>This is a resource recommendation rather than a Fabric or Power BI feature change, benchmark, or actionable implementation analysis.</p>
The post recommends an advanced DAX book but provides no implementation detail or platform change.
<p><strong>DataFrameWriterV2</strong> makes table intent explicit through operations such as <code>create()</code>, <code>append()</code>, <code>replace()</code>, <code>createOrReplace()</code>, <code>overwrite()</code>, and <code>overwritePartitions()</code>. This avoids encoding several meanings through <code>mode("overwrite")</code> and side-channel options.</p><p>V2 separates <strong>table properties</strong> from per-write options. Properties become persisted Delta metadata, while options such as schema evolution, transaction identifiers, and predicates apply only to the current write. The API also better represents catalog-managed tables and newer table capabilities.</p><p>Compatibility remains important: Spark exposes writer-level <code>clusterBy</code>, but Delta support is not yet wired through that surface in the described versions. Existing V1 code can therefore remain appropriate where provider behavior, path writes, or deployed runtime support is uncertain.</p>
Detailed API comparison clarifies table-operation semantics, metadata handling, schema evolution, and Fabric compatibility boundaries.
Power BI Updates
Jun 18, 2026
5/5 · Tier A
⚠️
Breaking ChangeMigrationCI/CD
release-notes
<p><strong>Power BI and Fabric are replacing embedded Simba ODBC drivers with ADBC drivers</strong> across Databricks, Dremio, BigQuery, Hive, Impala, Snowflake, and Spark connectors. The selected implementation can be controlled per connection, at workspace level, or through a tenant setting.</p><p><strong>Implementation="2.0" selects ADBC</strong>, while <strong>Implementation="1.0" selects ODBC</strong>; explicit connection settings override workspace and tenant defaults. ODBC execution in the service requires a gateway from September 2026, and ODBC drivers stop shipping with Desktop and Gateway in March 2027, making February 2027 the final retention release.</p>
Planned ODBC removal has explicit service, Desktop, gateway, and September 2026 migration milestones.
<p><strong>Fabric Data Warehouse</strong> adds preview functions for approximate string comparison: <strong>EDIT_DISTANCE</strong>, <strong>EDIT_DISTANCE_SIMILARITY</strong>, <strong>JARO_WINKLER_DISTANCE</strong>, and <strong>JARO_WINKLER_SIMILARITY</strong>. The distance functions measure difference, while similarity functions provide scores useful for filtering and ranking candidates.</p><p>The preview also adds ANSI-style <strong>||</strong> concatenation, <strong>||=</strong> variable appending, and <strong>UNISTR</strong> Unicode escape handling. Thresholds are application-specific, so fuzzy matches should produce review or mapping candidates rather than silently overwrite mastered values.</p>
Preview fuzzy-matching functions add practical SQL support for typo detection, standardization, and near-duplicate discovery.
<p><strong>Eventstream AI Skills</strong> can generate and deploy a multi-source Eventstream from a natural-language prompt, using Fabric REST APIs behind the scenes. The demonstrated topology ingests weather feeds for 11 cities, filters humidity events, and lands results in an Eventhouse table.</p><p>The example reduced more than 42 portal interactions per source to a reported deployment time under four minutes. The extract does not establish production guarantees for generated definitions, prompt accuracy, permissions, retries, schema drift, or consumption cost, so AI-generated topology requires review before promotion.</p>
Concrete Eventstream automation demonstrates multi-source real-time deployment, but provides limited governance, reliability, and cost detail.
<p>Microsoft is moving Power Query connectors toward Microsoft-owned, in-house implementations and defining a lifecycle of <strong>Preview</strong>, <strong>GA</strong>, transparent migration, and retirement. The stated goals are stronger supply-chain control, predictable support, and more consistent security and versioning across Power BI, Dataflows, and Fabric.</p><p>Recent GA examples include Snowflake, Google BigQuery, and Amazon Redshift V2 connectors, alongside Bring Your Own Driver experiences for Vertica and IBM Netezza. The announcement does not provide specific retirement dates, compatibility matrices, or migration actions for each legacy connector, so estates need source-by-source assessment.</p>
Connector ownership and lifecycle changes affect enterprise security posture, migration planning, and long-term Power Query support.
<p>The new <strong>Starburst secured by Entra ID</strong> connector enables Microsoft Entra single sign-on for DirectQuery semantic models through an on-premises data gateway. Starburst evaluates the actual report viewer's identity, including its row-level and column-level access policies, instead of using one shared connection account.</p><p>Two tenant settings and a connector-level option are required. The feature also supports Import models, but identity passthrough is relevant only to DirectQuery; existing reports using the original Starburst connector continue to work without migration.</p>
Preview connector SSO passes each viewer identity to Starburst for source-side row and column security.
<p><strong>Workspace outbound access protection</strong> now covers Power BI reports in preview. A report in a protected workspace can bind only to a semantic model in that same workspace, preventing cross-workspace query paths that could otherwise cross a regulated data boundary.</p><p>The restriction is automatic and has no report-level allow list or exception. It does not control many browser-side features, such as custom visuals, maps, URLs, R and Python visuals, or exports. Service-rendered subscriptions and PDF exports require separate tenant controls, and configuration changes can take roughly 15 minutes to propagate.</p>
Preview outbound protection blocks cross-workspace report bindings and documents important client-side and export limitations.
<p>The <strong>ServiceNow OneLake connector</strong> provides preview, zero-copy access to enterprise data stored in OneLake. ServiceNow workflows, analytics, and AI experiences can query governed data in place rather than maintaining a separate ETL copy.</p><p>The integration uses OneLake Table APIs compatible with the <strong>Apache Iceberg REST Catalog</strong> standard. This enables ServiceNow and other engines such as Trino, DuckDB, and Spark to discover and query lakehouse data through open interfaces, although the extract does not specify supported table types, security boundaries, query performance, or preview limitations.</p>
Preview partner integration exposes OneLake data through Iceberg APIs, but provides limited Fabric implementation detail.
<p>The <strong>Fabric Spark Operations Skill</strong> provides preview, read-only, AI-assisted diagnosis for Spark workloads. Plain-English requests can identify failed notebooks, pipeline-linked Spark sessions, workspace-wide failure patterns, performance problems, and stuck sessions without manually navigating the Fabric portal.</p><p>The skill combines Spark Monitoring APIs, logs, pipeline runs, and session activity into a ranked report with recommended fixes and links to the relevant run. Spark session data is retained for up to 30 days; older runs fall back to available failed-run snapshots. Deeper analysis can copy event logs to OneLake and launch a local Spark History Server, so authentication, log access, and local tooling remain part of the operating model.</p>
Open-source automation combines Spark monitoring APIs, logs, pipeline lineage, and capacity diagnostics into severity-ranked troubleshooting.
<p><strong>Delta 4.3</strong> routes catalog-managed table operations through Unity Catalog Delta APIs, including loads, creation, replacement, schema evolution, and metadata-changing writes. The catalog validates commit intent and rejects malformed or conflicting changes before they alter table state.</p><p><strong>UniForm</strong> now converts Iceberg metadata atomically and incrementally, avoiding full-snapshot regeneration on every commit. Experimental IcebergCompatV3 allows deletion vectors and UniForm together, while Spark streaming and Change Data Feed support catalog-managed tables and shared Delta tables.</p>
Delta 4.3 adds catalog-validated operations, atomic UniForm, deletion-vector compatibility, streaming, and CDF for managed tables.
Fabric Updates
Jun 16, 2026
4/5 · Tier A
Spark RuntimePySpark
announcement
<p>Fabric’s <strong>Spark History Server</strong> now loads application details incrementally from backend snapshots instead of parsing the complete event log before displaying the UI. Large applications with hundreds of thousands of tasks or event logs above 10 GB can therefore render their first data in seconds or minutes rather than taking up to an hour or failing entirely.</p><p>The release also supports compressed and rolling event logs through <code>spark.eventLog.compress</code> and <code>spark.eventLog.rolling.enabled</code>. This is particularly useful for long-running Spark Streaming workloads, where previously enabling either setting could prevent the History Server from rendering.</p>
Snapshot loading addresses documented enterprise-scale History Server failures with improvements from hour-long waits to seconds or minutes.
<p>Fabric adds preview visibility for service issues through in-product banners, a dedicated <strong>Service Health</strong> page in the Fabric admin portal, and optional Teams and email notifications.</p><p>Admins enable in-product banners and external notifications independently. Detailed health messages are restricted to users who can create Microsoft support cases, and coverage depends on supported scenarios, configuration, and available service data.</p>
Preview service-health notifications improve operational awareness, but provide limited platform behavior or cost detail.
Power BI Updates
Jun 16, 2026
3/5 · Tier B
Report AuthoringGovernance
announcement
<p><strong>Org apps with audiences</strong> are generally available in Power BI and Fabric. A workspace can contain multiple org app items, while each app can define audiences with different navigation, landing experiences, and visible content.</p><p>Audience membership can be assigned during sharing, and audience tabs may be shown, hidden, or used as the primary navigation. Org apps also integrate with Git and deployment pipelines and can distribute reports, paginated reports, notebooks, real-time dashboards, and other supported items.</p><p>Access propagation now applies consistently to content and underlying semantic models, helping prevent lingering model access after an app user or item is removed.</p>
GA org apps combine multiple app items and audience-specific navigation for governed, tailored distribution experiences.
<p><strong>Copy job for SAP with the Microsoft ABAP Add-On</strong> runs extraction inside SAP and transfers data through Fabric Data Factory. It supports SAP tables, views, and ABAP CDS views across on-premises or cloud-hosted ECC and S/4HANA systems.</p><p>Parallel execution targets high-throughput ingestion, with full snapshots for initial loads and watermark-based incremental extraction for ongoing changes. The preview reduces dependence on external extraction frameworks, but the extract still requires installing and operating a Microsoft-provided ABAP component inside the SAP landscape; the article provides no detailed sizing, licensing, or preview limitations.</p>
Preview Copy job integration adds high-throughput SAP extraction, incremental loading, and direct OneLake delivery across ECC and S/4HANA.
<p><strong>Azure Databricks</strong> can now read OneLake data generally, while storing Unity Catalog-managed tables directly in OneLake is available in beta. This enables Fabric and Databricks workloads to operate on a shared Delta-based storage layer instead of maintaining duplicated copies.</p><p>The architecture combines OneLake shortcuts and mirroring with Databricks processing, Fabric analytics, and catalog-based discovery. A single storage copy can reduce movement and duplication, but the practical governance model still spans Unity Catalog, Fabric permissions, identities, and workload-specific access behavior.</p><p>The article positions OneLake as an AI foundation, but it provides little quantified cost, performance, or concurrency evidence for choosing it over existing ADLS Gen2 storage.</p>
Describes substantive OneLake and Databricks interoperability, but much of the AI framing remains high-level.
<p><strong>Azure Databricks</strong> can store Unity Catalog-managed tables directly in OneLake through beta external-location support. A Unity Catalog external location maps to a OneLake path, making OneLake the primary storage destination for managed tables and related asset types.</p><p><strong>Publish to Fabric</strong> is in preview and creates mirrored Azure Databricks Catalog items from Databricks Catalog Explorer. Fabric polls Unity Catalog metadata, reflects schema and table changes at synchronization, and allows a manual refresh; published tables remain read-only in Fabric because Databricks is the source of truth.</p><p>The workflow requires identities, Fabric workspace roles, storage credentials, Unity Catalog permissions, and a Fabric connection. Catalogs can be published independently to multiple Fabric workspaces, but the beta and preview status require compatibility and operational testing before production adoption.</p>
Provides concrete beta and preview mechanics for shared Databricks-Fabric Delta storage, publishing, synchronization, and read-only behavior.
<p><strong>DAX user-defined functions</strong> can declare optional parameters with default expressions. Callers may omit trailing parameters or leave an empty positional argument to skip an earlier optional parameter while supplying a later one.</p><p>Optional parameters should normally come last, because a mandatory parameter after an optional one forces awkward empty argument positions. Using <strong>BLANK()</strong> as a default lets a function detect a missing value through <strong>ISBLANK</strong>, but omitted and explicitly blank arguments remain indistinguishable.</p>
Useful advanced UDF syntax improves reusable DAX APIs, with practical calling rules and an important blank-value limitation.
<p><strong>V-Order</strong> is a write-time Parquet optimization that increases write cost but can improve later reads, including Power BI Direct Lake scans. Dataflows Gen2 can now disable V-Order compression for staged data when Fast Copy is not used.</p><p>In a 5.58 GB, 17.6-million-row test, refresh time fell from 1:59 to 1:32. Staging improved from 1:31 to 1:13, while the group-by improved from 12 seconds to 7 seconds, showing that staging overhead can dominate when staged data is rarely queried.</p>
Measured testing shows disabling staged-data V-Order cut refresh time by 27 seconds without slowing the downstream group-by.
<p><strong>Real-Time Dashboard</strong> tile editing now offers AI-assisted, manual, and hybrid KQL authoring workflows in preview. Copilot generates a visual from a prompt, preserves iteration history, and lets editors inspect results, schema, and data before applying changes.</p><p>The redesigned interface gives advanced users more room for code and preview work, while parameters can be inserted and tested directly in the editor. The extract does not specify Copilot licensing, query limitations, data-access behavior, or preview-region constraints.</p>
Preview improves Real-Time Dashboard authoring, but provides limited operational detail, constraints, or cost analysis.
<p><strong>Time Series Visualization</strong> adds dedicated exploration for multivariate time-series data in Real-Time Dashboards. Users can search and select series, navigate hierarchical entities, synchronize time ranges across measures, and hide or reorder data without changing the underlying query.</p><p>Editors can choose global, separate, or adaptive Y-axis scaling, remove outlier influence, assign colors, zoom, and switch between linear and logarithmic axes. The extract does not describe supported query shapes, series-count limits, or rendering performance for large sensor populations.</p>
Preview adds useful time-series exploration, but lacks implementation limits, performance measurements, and operational cost detail.
<p><strong>Live Refresh</strong> is generally available for Fabric Real-Time Dashboards. Instead of executing visual queries on a fixed schedule, a lightweight background query detects ingestion and refreshes affected visuals only when new data arrives.</p><p>Editors can configure Live Refresh or manual refresh, including a fallback interval for visuals that lack ingestion detection and a minimum polling interval. Viewers can pause refresh while investigating a point in time, avoiding unexpected visual changes.</p><p>The design can reduce query load during quiet periods, but active high-volume sources can still generate substantial dashboard compute usage because detected ingestion triggers visual refreshes.</p>
GA event-driven dashboard refresh reduces unnecessary polling and exposes fallback and minimum-polling controls.
<p><strong>Rayfin</strong> is an open-source SDK and CLI for defining an application backend in code and deploying a hosted site as a Fabric item. The intended workflow turns evolving markdown or agent-generated content into a stable URL with organizational sign-in and workspace-governed access.</p><p>Captured site data lands in a SQL database in Fabric and can therefore feed OneLake, Power BI, and notebooks without a separate integration pipeline. The extract does not establish production support boundaries, deployment isolation, database sizing, or the precise permission model inherited by the deployed site.</p>
Rayfin connects agent-generated applications to Fabric identity, SQL storage, and governance through an open-source deployment model.
<p><strong>Azure Data Factory</strong> pipelines can now be migrated from a Fabric workspace by mounting an ADF instance, selecting pipelines, mapping connections, and creating Fabric pipeline items. Mounting is read-only and does not move or alter the source factory until migration begins.</p><p>Supported authentication types and services can receive automatically created Fabric connections; unsupported types require an existing or newly configured connection. Migrated pipelines are placed in a source-prefixed folder, schedules remain disabled, and activities using unmapped connections are deactivated. Production cutover therefore still requires connection validation, trigger configuration, and end-to-end testing.</p>
Fabric-first ADF migration adds concrete connection mapping and post-migration safeguards, including disabled triggers and deactivated activities.
<p><strong>OneLake catalog</strong> can now assign, edit, and remove compatible workspace and OneLake security-role memberships across multiple workspaces from its Secure tab in preview. Administrators can filter managed workspaces and apply bulk membership changes without opening each workspace.</p><p>The workflow targets onboarding, quarterly access reviews, environment rollouts, and standardized data products. It should reduce configuration drift, but the extract does not define role compatibility, delegation boundaries, audit details, or how bulk changes interact with existing row- and column-level protections.</p>
Preview centralizes cross-workspace role membership, reducing access drift and making estate-wide reviews materially more practical.
<p><strong>SSMS 22.7</strong> adds preview support for graphical schema comparison, configurable SQL formatting, and SQLCMD variables in the publish dialog.</p><p>Schema compare works across connected databases, SQL projects, and DACPAC files, with reusable <strong>.scmp</strong> definitions and selective deployment actions. SQL formatting supports format-on-save, while SQLCMD variables parameterize environment-specific deployment values for interactive or SqlPackage-based releases.</p>
Useful SQL DevOps tooling, but it targets SSMS rather than Fabric platform behavior.
Power BI Updates
Jun 8, 2026
3/5 · Tier B
Report AuthoringCI/CDAI Readiness
announcement
<p><strong>Power BI authoring skills let an AI agent create and modify PBIR report definitions</strong> from natural-language prompts or reference images. The workflow can generate pages, visuals, themes, navigation, and layouts, then reload Power BI Desktop and capture screenshots for iterative refinement.</p><p>The capability integrates with the Modeling MCP Server and semantic-model authoring skills for an end-to-end model-to-report workflow. It is preview functionality, so generated definitions still require validation of business logic, accessibility, security, and deployment behavior.</p>
Preview agent skills generate PBIR reports and iteratively validate rendered output, but production limitations remain unspecified.
<p><strong>SharePoint and OneDrive Shortcuts in OneLake</strong> are generally available. Fabric lakehouses can reference SharePoint folders and OneDrive locations in place, allowing files to remain in Microsoft 365 while Fabric workloads consume the latest content without a traditional copy pipeline.</p><p>Supported shortcut transformations convert CSV, Parquet, and JSON files into synchronized Delta tables. GA also adds <strong>Service Principal</strong> and <strong>Workspace Identity</strong> authentication through Microsoft Entra ID, including supported cross-tenant scenarios, and caches SharePoint metadata to reduce API calls, throttling, and folder-enumeration overhead. The extract does not define all file-size, refresh-latency, permission-propagation, or query-performance limits.</p>
GA shortcuts add in-place SharePoint access, Delta transformations, identity authentication, and metadata caching for production workloads.
<p><strong>Partitioned Compute</strong> allowed a Dataflows Gen2 query combining 244 SharePoint Excel files to evaluate partitions concurrently. Average refresh time fell from 28:08 without it to 18:37 at concurrency 4, 9:42 at 8, and 5:24 at 16.</p><p>Increasing the limit beyond 16 produced no meaningful improvement: concurrency 250 took 5:21 and no limit took 5:16. One unlimited run failed with SharePoint HTTP 429 throttling, demonstrating that source limits can make excessive parallelism harmful.</p>
Production-like testing quantifies concurrency gains and shows unlimited parallelism can trigger SharePoint 429 failures.
<p><strong>Copilot in web modeling can analyze and modify semantic models conversationally</strong>, including renaming tables and columns, creating relationships, and generating DAX measures. It operates from the Power BI Service Model view in Editing mode.</p><p>Admins, Members, and Contributors can use it on the active model after granting session permission. Each session creates a restore checkpoint through semantic version history, providing recovery from incorrect changes, but the preview does not replace review of model logic or security behavior.</p>
Preview Copilot can apply model changes with permissions and restore checkpoints, directly affecting governed semantic-model authoring.
Power BI Updates
Jun 4, 2026
2/5 · Tier B
AI ReadinessSemantic ModelReport Authoring
announcement
<p>Power BI positions the semantic model as the governed foundation for agent-driven analytics. Previewed <strong>Agent Skills</strong> can create semantic models and reports from natural-language instructions or screenshots, while Fabric Apps let coding agents build web applications over existing semantic models.</p><p><strong>Org apps</strong> are approaching general availability with multiple apps per workspace and audience-specific item visibility. DAX user-defined functions provide reusable business logic, but this article offers no meaningful limits, pricing, or operational guidance for the previews.</p>
Build roundup announces agentic Power BI capabilities but provides limited implementation, cost, and governance detail.
<p><strong>Fabric Data Warehouse</strong> receives a broad set of preview, coming-soon, and generally available capabilities. GPU query acceleration is intended to offload supported analytical operations without SQL rewrites, with reported benchmarks of up to 7x improvement and near-flat response times across concurrency levels; those figures are early product benchmarks, not a guaranteed workload result.</p><p>Configurable data-history retention allows one-to-120-day windows for Time Travel, point-in-time table clones, snapshots, and in-place restore. Longer retention supports recovery and audit scenarios but increases storage cost. Additional changes expand T-SQL with string-similarity and date-bucketing functions, procedural scalar UDFs, transactional ALTER TABLE support, and ALTER COLUMN preview support.</p><p>The update also covers enterprise web tooling, cache cooldown, table health checks, statement routing, distributed bitmap filters, bulk copy, and pro-developer APIs. CI/CD pre- and post-deployment scripts and SQL Analytics Endpoint automation are especially relevant to repeatable warehouse promotion, although many capabilities remain preview or forthcoming.</p>
Build coverage includes warehouse performance, retention, T-SQL, APIs, CI/CD, and capacity-relevant behavior changes.
<p><strong>Dataflow Gen2</strong> uses several complementary performance paths rather than one universal acceleration switch. Staging materializes intermediate results for reuse, Fast Copy targets high-throughput Lakehouse ingestion, the Modern Query Evaluator improves Power Query M execution for transformation-heavy workloads, and Partitioned Compute parallelizes suitable file-based processing.</p><p>The recommended progression is to start with defaults, then select the optimization matching the bottleneck. Fast Copy works best when ingestion transformations are lightweight; richer shaping can follow in a separate query over staged data. Partitioned Compute requires partition-friendly sources and typically benefits from retaining partition keys created by the Combine Files experience.</p><p>The benchmark framing links shorter execution time to lower capacity cost and positions Dataflow Gen2 with Lakehouse and Direct Lake as the preferred analytical setup. Actual savings still depend on source access, transformation compatibility, partitionability, concurrency, and whether staging creates useful reuse rather than additional writes.</p>
Benchmark-focused guidance connects Dataflow Gen2 execution options with measured performance, cost, OneLake, and Direct Lake architecture.
<p><strong>Migration Assistant for Fabric Data Warehouse</strong> now supports live connectivity as a generally available input path, removing the requirement to prepare a DACPAC before beginning discovery and migration. SQL-file bundles are available in preview for offline or disconnected workflows, providing an alternative when direct source access or DACPAC tooling is unavailable.</p><p>The assistant also improves dependency awareness by identifying unsupported patterns, relationships between objects, and likely downstream issues. Progress views show migration stages, migrated versus pending objects, and object-level results by type and dependency, making remediation more actionable during large schema migrations.</p><p>The changes cover metadata translation, issue resolution, data movement, and connection updates, but they do not eliminate source-specific gaps. The reported customer example—300 pipelines, 500 tables, and 450 GB with approximately 66% less effort—should be treated as an observed case rather than a sizing guarantee.</p>
Migration Assistant adds direct connectivity, SQL-file workflows, dependency analysis, and concrete enterprise migration progress visibility.
<p><strong>Run Mapping Data Flow transforms in Dataflow Gen2</strong> brings a visual, low-code authoring model for advanced joins, splits, aggregations, derived columns, and schema shaping, with execution backed by Spark. It targets larger transformation workloads while keeping ingestion, transformation, orchestration, and monitoring within Fabric Data Factory.</p><p>The capability also provides a migration path for Mapping Data Flows built in Azure Data Factory or Synapse. <strong>My Queries</strong> adds personal reuse of saved Power Query logic across Dataflow Gen2 items, while the modern Power Query Get Data experience provides a more consistent connection flow in Power BI Desktop. The listed capabilities are previews or forthcoming, so connector and transformation compatibility remain material adoption questions.</p>
Preview Spark-backed Mapping Data Flows improve migration and scale, but the article provides limited constraints or measured results.
Fabric Updates
Jun 4, 2026
3/5 · Tier B
AI ReadinessReal-Time IntelligenceGovernance
announcement
<p><strong>Operations agent</strong> monitors Fabric data through Real-Time Intelligence, investigates anomalies, recommends actions, and—when authorized—runs pipelines, notebooks, user data functions, or Power Automate workflows.</p><p>Agents use natural-language instructions grounded in Fabric IQ Ontology, ask clarifying questions instead of silently guessing, and provide previews before rules are committed. Teams notifications, natural-language status queries, decision tracing, authorization records, and Microsoft Agent 365 and Entra ID integration improve operational visibility and governance.</p><p>The release is described as generally available, although the experience remains marked Preview until the regional rollout completes later in June. The article provides no capacity, execution, action-volume, or licensing details, so production cost and scale still require validation.</p>
GA operations agents combine real-time monitoring, authorized Fabric actions, Teams collaboration, tracing, and centralized governance.
<p><strong>Fabric Data Factory</strong> adds human approval, conditional retries, and native refresh activities for Materialized Lake Views and SQL analytics endpoints. Approval pauses a pipeline until a decision, while conditional retries inspect error codes or messages with AND/OR logic, allowing transient failures to retry without repeatedly executing non-retriable work.</p><p>Variable Libraries now support connection and item references, allowing the same pipeline or Airflow DAG to resolve environment-specific resources through active value sets. The SQL Endpoint refresh activity is generally available; approval, Materialized Lake View refresh, conditional retries, the modern canvas, and Airflow Variable Library support are preview features.</p><p>An in-workspace ADF migration entry point provides a no-cost assessment before migration, but assessment availability does not guarantee that every ADF connector, trigger, expression, or integration runtime behavior transfers unchanged.</p>
Builds materially stronger orchestration with approvals, conditional retries, refresh activities, promotion variables, Airflow support, and ADF migration.
Power BI Updates
Jun 3, 2026
2/5 · Tier B
AI ReadinessSemantic Model
announcement
<p><strong>Agent Skills for Power BI aim to automate semantic-model and report development</strong> from natural-language prompts, screenshots, or existing solutions. The workflow uses Power BI Projects and builds on the Modeling MCP capability.</p><p><strong>Fabric Apps for Semantic Models</strong> extend the same governed data foundation into AI-generated web applications, potentially combining analytics, operational workflows, and external integrations. The announcement describes scenarios and architecture direction but does not establish production limits, pricing, accuracy controls, or deployment governance.</p>
The announcement introduces preview agent skills and Fabric Apps but provides little evidence about limits, cost, or governance.
<p><strong>Graph in Fabric</strong> is generally available for modeling, visualizing, querying, and analyzing complex relationships directly over OneLake data. Its distributed architecture targets large graphs, while GQL supports pattern matching, path traversal, and aggregations.</p><p>Graph makes dependencies explicit for scenarios such as fraud networks, supply chains, cybersecurity blast-radius analysis, and customer recommendations. Fabric IQ Ontology adds shared business meaning so agents can reason over entities and relationships rather than infer connections from isolated tables.</p><p>Natural-language-to-GQL lowers the query barrier, and role-based experiences serve business users, analysts, engineers, data scientists, and developers. The announcement supplies no measured performance, graph modeling constraints, or capacity economics, so billion-relationship claims should not substitute for workload testing.</p>
Graph reaches GA with OneLake integration, GQL traversal, ontology support, and relationship-focused agent reasoning.
<p><strong>Business Events</strong> can now flow automatically into Eventhouse and Real-Time Dashboards in Preview. Each event definition maps to a dedicated KQL table, and published events are retained for historical analysis, near-real-time monitoring, cross-event correlation, and machine-learning workloads.</p><p>Events can originate from Notebooks, User Data Functions, Eventstreams, or Activator. A Real-Time Dashboard connects directly to the Eventhouse table, avoiding an additional ingestion pipeline or configuration step for the basic analytical path.</p><p>Eventhouse retention, event volume, query cost, schema evolution, and dashboard latency are not quantified. Those omissions matter when high-frequency telemetry or long retention turns a convenient default into a material capacity and storage commitment.</p>
Business Events now persist automatically in Eventhouse for KQL history, cross-event analysis, and live dashboards.
<p><strong>Activator</strong> can now publish Business Events in Preview when conditions are met in Power BI reports, Real-Time Dashboards, KQL queries, or Warehouse SQL queries. Instead of producing only a notification or workflow, an alert becomes a structured signal that other Fabric components can discover, consume, analyze, and act upon.</p><p>Published events are stored in Eventhouse by default. Real-Time Hub centralizes event definitions, publishers, consumers, and payload previews, reducing format drift between teams. Activator can also consume the resulting events to coordinate notifications, workflows, and escalations.</p><p>The release does not specify delivery guarantees, duplicate handling, ordering, throttling, or event-volume costs. Those behaviors are important when one condition drives financial, operational, or compliance-sensitive automation.</p>
Activator can publish governed, persistent Business Events, enabling shared definitions, historical analysis, and centralized responses.
<p><strong>GPU-accelerated Fabric Data Warehouse</strong> is an early-access preview that can be enabled at workspace level for eligible warehouses and SQL Analytics Endpoints. The SQL surface and application architecture remain unchanged while the optimizer selectively pushes joins, aggregations, scans, and arithmetic to GPUs; ineligible plan sections continue on CPUs.</p><p>The article reports up to 5x customer improvements, up to 7x benchmark improvements, and a 22-query, 100 GB workload completing in approximately five seconds at one and 64 concurrent users. These figures are workload-dependent, and the extract does not explain GPU eligibility rules, capacity consumption, pricing, quota effects, or production preview constraints.</p>
Early-access GPU execution claims substantial concurrency improvements, but eligibility, cost, limits, and reproducible benchmark details remain unclear.
<p><strong>Fabric Skills</strong> package Microsoft-authored instructions, API patterns, authentication guidance, and workload-specific recipes for GitHub Copilot, Claude Code, Cursor, Windsurf, and compatible tools. Skills cover authoring, read-only consumption, operations, migration, and end-to-end architectures.</p><p>The skills select appropriate Fabric APIs, CLIs, token audiences, query patterns, and schema conventions instead of relying on generic SQL or Python knowledge. The repository is MIT-licensed and supports focused bundles or organization-specific extensions; the initial release has limited coverage for Power BI, Data Science, and Real-Time Intelligence.</p>
Microsoft-authored skills provide workload-specific Fabric APIs, authentication audiences, and architecture guidance for coding agents.
<p>The <strong>Mirrored Database Change Feed connector</strong> is in Preview and streams Delta Change Data Feed updates from Fabric Mirrored Databases into Eventstreams. It carries row-level inserts, updates, and deletes with operation metadata and timestamps, preserving source-change fidelity without custom Spark polling jobs.</p><p>After discovery through Real-Time Hub, Eventstreams can filter, transform, aggregate, and route changes to Eventhouse, Activator, Lakehouse, or other destinations. Supported mirrored sources include Azure SQL, Snowflake, Cosmos DB, Oracle, and Open Mirroring partners, provided change tracking is enabled through the relevant extended capabilities.</p><p>The connector removes ingestion plumbing but does not eliminate design concerns around ordering, deletes, schema evolution, replay, latency, or downstream deduplication. Preview behavior and source-specific limitations should be tested before replacing a production change-processing pipeline.</p>
A managed connector streams Mirrored Database Delta CDF changes into Eventstreams, removing custom polling for low-latency reactions.
<p>46 features - 18 must-know - 26 worth knowing.</p><ul><li>Extended Lineage for Materialized Lake Views</li><li>Fabric Git Integration – GitHub Enterprise Cloud with Data Residency (Generally Available)</li><li>Optimize your storage costs with OneLake storage tiers and lifecycle management (Preview)</li><li>AI functions: New models, no package dependency, better usage stats (Generally Available)</li><li>Fabric data agents to support Service Principals (Preview)</li><li>CI/CD Support for SQL analytics endpoint (Preview)</li><li>Pre & Postscripts for Fabric Warehouse Deployments (Preview)</li><li>New enhanced metadata sync in SQL analytics endpoint (Preview)</li><li>Configurable Retention in Fabric Warehouse (Preview)</li><li>Time Travel using SQL analytics endpoint (Preview)</li><li>Stream Mirrored Database change feeds into Eventstreams (Preview)</li><li>Real-Time Dashboards Live Refresh (Generally Available)</li><li>CDC with SQL estates in Copy job (Generally Available)</li><li>Edit Copy Job via JSON Payloads for Maximum Flexibility</li><li>Connection and Item Reference Support for Variable Libraries in Pipelines</li><li>Workspace Identity authentication for SharePoint</li><li>Snowflake now supports Secret-less Authentication using Microsoft Fabric Workspace Identity</li><li>Mirroring in Fabric advancing network security controls</li></ul>
18 must-know of 46 features
<p><strong>CoddSpeed</strong> is a GPU-accelerated execution engine for Fabric Data Warehouse, derived from the TQP research prototype and planned for early access preview. It introduces two hardware abstractions: the Coprocessor Abstraction Layer offloads query fragments, while the Data Abstraction Layer provides unified caching and shuffle across interconnects.</p><p>Reported results include 7.9× warm and 4.7× cold speedups on a single A100 at TPC-H scale factor 100, 14.9× warm and 9.7× cold on a 300 GB customer workload, and 27.1× with eight H100 GPUs and NVLink versus 2.1× without the data-movement layer. A 16-H100, two-node test reached 30.4×, but results use selected workloads, warm-cache runs, and hardware that customers may not possess.</p><p>Unsupported or oversized plan fragments can fall back to CPU, and operator coverage determines how much of a query benefits. Transfer overhead also means larger pushed-down fragments matter more than accelerating isolated operators.</p>
Provides measured GPU warehouse architecture and performance results, including cold/warm behavior, scaling, data movement, and fallback.
<p>Fabric Data Factory adds capabilities for secure ingestion, diagnostics, transformation, orchestration, and code-driven promotion. Enhanced Dataflow Gen2 diagnostics provide richer logs across on-premises data gateways, while Copy job and Copy activity are generally available for workspace-staged ingestion into private-link-protected Fabric Warehouses.</p><p>Generally available additions include the Dataflow Execute Query streaming API, SQL analytics endpoint refresh activity, and CDC for Azure SQL Database, Azure SQL Managed Instance, and SQL Server. Preview capabilities include private mirroring, Mapping Data Flow transformations, pipeline conditional retries, approval activities, Airflow workspace identity, and Copilot-assisted DAG authoring.</p><p>dbt job pipeline activities, APIs, and project export extend orchestration and CI/CD options. JSON editing for Copy jobs and connection or item references can improve portability, but promotion behavior and network prerequisites remain important implementation constraints.</p>
Build roundup contains production-ready CDC, private-network ingestion, dbt APIs, pipeline controls, and deployment-oriented references.
Fabric Updates
Jun 2, 2026
2/5 · Tier B
AI ReadinessGovernanceWarehouse
announcement
<p>Build 2026 adds agent-assisted database development across VS Code and SSMS. Schema Designer can generate database structures and migration-oriented ORM code, Data API builder can expose REST, GraphQL, and MCP endpoints, and SQL Notebooks bring interactive SQL and Python workflows into VS Code.</p><p>Security and operations updates include Entra server principals, fixed server roles, AES-backed customer-managed TDE preview, regex-based masking preview, local time zones, and Change Event Streaming. Most capabilities target Azure SQL Database, Hyperscale, SQL Server, or SSMS rather than core Fabric analytics.</p><p>Fabric-specific references include forthcoming local time-zone support and Change Event Streaming for SQL database in Fabric, plus a broader Fabric Apps direction. The article is therefore useful for adjacent Microsoft database architecture, but not a primary Fabric platform change log.</p>
Most content concerns Azure SQL and developer tooling; Fabric relevance is limited to SQL database and future integration references.
Fabric Updates
Jun 2, 2026
4/5 · Tier A
AI ReadinessGovernanceSemantic Model
announcement
<p><strong>Fabric IQ Ontology</strong> is expanding as a governed business-context layer for agents. Ontology can expose entities, relationships, rules, and mappings through an MCP server, provide context to Foundry IQ and Copilot Studio, and support Operations Agent scenarios.</p><p>Modeling improvements include untyped properties, metadata enrichment, and repeated relationship or property names. Item-level access control supports finer boundaries for shared ontology ownership, while UX and query reliability improvements target larger models.</p><p><strong>Graph in Fabric</strong> is generally available, with relationship-first modeling, GQL support, natural-language-to-GQL, and scale-out processing for large relationship sets. The extract emphasizes broad agent integration, but many surrounding integrations remain preview or Frontier experiences and require careful permission testing.</p>
Fabric IQ adds ontology integrations, MCP access, item-level controls, and generally available graph modeling for governed agent context.
Fabric Updates
Jun 2, 2026
4/5 · Tier A
Spark RuntimeWarehouseAI Readiness
announcement
<p>Fabric Analytics is adding capabilities across Spark, Warehouse, Power BI, and Fabric IQ for agent-driven workloads. <strong>Runtime 2.0</strong> includes Apache Spark 4.1, Delta Lake 4.1, and Python 3.13, while the Native Execution Engine expands support for complex types and Python, Scala, and Java UDFs.</p><p>Efficient Scaledown uses Remote Shuffle Manager to offload shuffle data to Azure Blob storage, allowing executors to release sooner for bursty workloads. Fabric Warehouse is also introducing preview GPU query acceleration, reportedly delivering up to 7x faster performance in internal comparisons; early adopters saw agent response times reduced by up to 50%.</p><p>Cache cooldown controls expose a performance-versus-cost setting for retaining warm query results. Several capabilities remain preview or forthcoming, so the stated benchmarks should be validated against representative concurrency, query shapes, and data sizes.</p>
Build roundup includes Runtime 2.0, native Spark execution, GPU warehouse acceleration, and measured agent response improvements.
Fabric Updates
Jun 2, 2026
2/5 · Tier B
AI ReadinessGovernance
announcement
<p><strong>Rayfin</strong> is presented as an open-source SDK and CLI for defining application data models, APIs, business logic, identity, and access policies in code, then deploying the backend into Microsoft Fabric. The proposed Fabric deployment places application data alongside OneLake and makes it available to analytics, Power BI, notebooks, and AI workloads.</p><p>The announcement emphasizes inherited governance and security, but does not specify the underlying Fabric artifacts, supported transactional behavior, deployment model, networking, identity integration, cost model, or separation between application and analytical workloads. Those omissions make this an early platform announcement rather than an actionable replacement decision for established backend architectures.</p>
Rayfin proposes code-first application backends deployable to Fabric, but provides few implementation limits, controls, or operational details.
Fabric Updates
Jun 2, 2026
2/5 · Tier B
AI ReadinessReal-Time Intelligence
announcement
<p><strong>Fabric IQ</strong> positions OneLake, semantic models, ontologies, graphs, planning, and agents as a shared business-context layer for real-time applications. Real-Time Intelligence supplies live signals, while Fabric IQ supplies the entities, relationships, rules, and meaning needed to interpret those signals consistently.</p><p>Operations Agents and Graph are described as generally available, Planning is expected to reach general availability later in June, and Ontology remains in preview. The announcement emphasizes an observe-understand-decide-act loop, but does not provide meaningful limits, capacity consumption, accuracy measurements, or deployment guidance.</p>
Broad Build announcement positions Fabric IQ and RTI as shared context for agents, but provides limited implementation detail.
<p><strong>DAX user-defined functions</strong> are production-ready reusable semantic-model objects. Functions can be called from measures, calculated columns, visual calculations, and other functions, with documentation comments, IntelliSense, typed parameters, optional defaults, and automatic dependency tracking when model objects are renamed.</p><p>Functions can be authored in Desktop, web modeling, DAX Query View, TMDL, SSMS, the XMLA endpoint, or Semantic Link Labs. Persisted model functions appear in <strong>INFO.USERDEFINEDFUNCTIONS()</strong>; functions defined only inside a query remain query-scoped and are not catalogued. TMDL and XMLA make versioning and repeatable deployment practical across models.</p>
DAX UDFs are generally available with typed parameters, model integration, TMDL authoring, dependency tracking, and deployment automation.
<p><strong>Admin consent for gateway diagnostics</strong> gives tenant administrators organization-wide control over whether on-premises gateway diagnostic data is sent to the cloud. Gateway administrators can enable diagnostics at individual gateways, while tenant administrators can revoke consent; revocation stops all gateways and terminates ongoing transfers.</p><p>Gateway logs now appear alongside Dataflow Gen2 run details, helping separate dataflow logic, connectivity, authentication, and downstream failures without switching tools. The release also aligns the gateway with the May 2026 Power BI Desktop runtime, reducing version drift between authoring and service refresh execution.</p>
Adds consent-controlled gateway diagnostics and unified Dataflow Gen2 troubleshooting, with direct implications for sensitive logs and administration.
<p><strong>ALL</strong>, <strong>ALLSELECTED</strong>, <strong>ALLEXCEPT</strong>, and <strong>REMOVEFILTERS</strong> are not interchangeable despite their shared filter-removal reputation. <strong>REMOVEFILTERS</strong> is the clearest choice when <strong>CALCULATE</strong> should clear filters; <strong>ALL</strong> is required when a table expression must be returned.</p><p><strong>ALLSELECTED</strong> preserves outer selections, making it appropriate for visual totals, while <strong>ALLEXCEPT</strong> preserves a specified grouping grain. The distinction between modifying filter context and returning a table matters in iterators and nested calculations.</p>
Clarifies high-impact DAX filter-context choices, including visual totals and table-expression differences that cause calculation errors.
<p>6 features - 1 must-know - 4 worth knowing.</p><ul><li>Understand Business Events capacity (Generally Available)</li></ul>
1 must-know of 6 features
Fabric Updates
Jun 1, 2026
3/5 · Tier B
Report AuthoringAI Readiness
tutorial
<p><strong>Translytical task flows</strong> can turn notifications into report data rather than email messages. A user submits an alert through a writeback report, a User Data Function stores it in a Fabric SQL database, a Lakehouse shortcut exposes it, and a Direct Lake semantic model surfaces report-specific active alerts.</p><p>The pattern provides a persistent alert badge and detail panel inside each report. Direct Lake models can see changes immediately through the shortcut; non-Direct Lake models require additional Data Activator and refresh-pipeline work, introducing more latency and operational complexity.</p>
A concrete translytical task-flow pattern embeds persistent, targeted alerts inside Power BI reports instead of email.
<p><strong>Partitioned Compute</strong> can parallelize evaluations within a single Dataflows Gen2 query, not only separate queries. In a ten-row test with a 60-second delay per row, default concurrency completed the refresh in 1:47.</p><p>Setting concurrency to 1 forced sequential evaluation and increased refresh time to 11:31. The default level is variable, so the feature provides an important tuning control when transformation work can safely run in parallel.</p>
Controlled testing demonstrates that Dataflows Gen2 concurrency can parallelize one query, with a major speed-versus-resource trade-off.
<p>The May <strong>Fabric Influencers Spotlight</strong> curates community content across Power BI, data engineering, data science, governance, databases, and capacity administration. Most entries are brief descriptions of external tutorials or videos rather than technical findings.</p><p>One highlighted benchmark reports that decomposing a high-cardinality decimal column reduced VertiPaq scan time by 45% and total query duration by 44% on a 500-million-row Import model. Other items cover schema-enabled Lakehouses, Data Agent cost optimization, capacity automation, and SharePoint mirroring, but the extract provides insufficient implementation detail to validate them.</p>
The community roundup contains useful links, but only one summarized benchmark provides substantial actionable Fabric engineering evidence.
<p><strong>Incremental Liquid Clustering</strong> is available by default in Fabric Runtime 2.0 with Delta 4.1 and Spark 4.1. <code>OPTIMIZE</code> now processes newly unclustered files, undersized files, and files with sufficient deletion vectors instead of repeatedly rewriting healthy, well-clustered data.</p><p>Auto Reclustering selectively repairs overlapping or degraded file groups as data evolves. Microsoft reports average clustering speedups of 8.9x for non-overlapping streaming appends, 5.5x for full-overlap appends, and 4.9x for partial-overlap MERGE workloads. A full recluster remains available with <code>OPTIMIZE FULL</code>, such as after changing clustering keys.</p>
Runtime 2.0 makes Liquid Clustering incremental, with measured 4.9x–8.9x faster OPTIMIZE operations and comparable skipping.
<p>The enhanced preview <strong>Azure IoT Hub connector</strong> preserves both system and custom event metadata when messages enter Fabric Eventstream. System properties such as device ID, enqueue time, and authentication method receive a <code>___src__</code> prefix; custom properties retain their existing names.</p><p>Users enable <strong>Extended features</strong>, extract fields with Eventstream’s SQL operator and <code>GETMETADATAPROPERTYVALUE</code>, then route enriched events to Eventhouse for KQL analysis. The article gives no throughput, retention, pricing, or preview reliability limits, so production sizing remains unresolved.</p>
The preview connector preserves IoT Hub system metadata, enabling device-level latency, routing, and security analysis in Eventstream.
<p><strong>fabric-cicd v1.1.0</strong> adds support for DataBuildToolJob items, workspace-name replacement variables, and per-workspace Fabric API endpoints for private-link-enabled workspaces.</p><p>Environment deployments now use the item definition API for Spark compute settings, correcting incomplete configuration deployment. The release also fixes post-publish filtering and prevents an empty item list from unintentionally bypassing item selection.</p>
Adds private-link endpoint configuration and fixes Environment deployment behavior, filtering, and workspace-variable substitution.
<p><strong>Excel to Delta Shortcut Transformations</strong> brings low-code ingestion of .xlsx and .xls files into Fabric lakehouses. The transformation discovers workbook sheets, supports explicit names, indexes, and wildcards, and writes either one combined Delta table or separate tables per sheet.</p><p>Schema inference, cross-sheet validation, intelligent removal of banners and empty rows, sheet-name sanitization, and per-file/per-sheet monitoring reduce the custom Spark logic traditionally required. Combined output includes filepath and sheetname metadata for lineage, while continuous sync detects new or modified source files.</p><p>The feature remains in preview, and inconsistent workbook layouts remain a practical risk: mismatched sheets can be skipped with warnings or cause the transform to fail. Complex business formatting or unstable schemas still require testing before production adoption.</p>
Preview adds practical zero-code Excel ingestion with sheet selection, schema handling, continuous sync, and traceability.
<p><strong>Copy job</strong> now provides generally available Change Data Capture replication across key SQL estate sources and destinations, including SQL Server, Azure SQL, Azure SQL Managed Instance, Fabric SQL, and Snowflake. Additional Lakehouse, Oracle, BigQuery, and warehouse connectors remain preview or are progressing toward GA.</p><p>Native <strong>SCD Type 2</strong> support also extends to Fabric Warehouse and Synapse SQL Pool. Copy job preserves historical versions with effective dating and represents source deletes as inactive destination rows rather than physically removing them.</p>
Copy job CDC for SQL estate reaches GA, while native SCD Type 2 expands to Fabric Warehouse and Synapse SQL Pool.
Fabric Updates
May 26, 2026
4/5 · Tier A
Spark RuntimeCapacityGovernance
announcement
<p><strong>Custom Live Pools</strong> provide scheduled, ready-to-run Spark capacity integrated with Fabric Environment artifacts. Workspace and capacity administrators can configure activation windows, monitor hydration and readiness, and observe utilization through Fabric monitoring experiences.</p><p>Environment library publishing does not automatically refresh active or idle pooled sessions. Administrators must update the libraries so the next pool warm-up uses the new package set. The preview is also intended to work with workspaces using <strong>Managed Private Endpoints</strong>.</p>
Preview Custom Live Pools schedule warm Spark capacity, expose lifecycle monitoring, and define important library-refresh and networking behavior.
<p><strong>Copy job</strong> now supports JSON-payload editing generally, allowing advanced users to define, automate, and bulk-update configurations beyond the visual authoring experience. This is useful for repeatable deployment and consistent settings across environments.</p><p>Copy jobs can also switch between full and incremental modes without recreation. Incremental mode still begins with an initial full copy, while auto-partitioning now extends in preview to Oracle, SAP HANA, and Fabric Lakehouse tables.</p>
GA JSON editing and mode switching improve repeatable data movement; auto-partitioning adds preview performance coverage.
<p><strong>Dataflows Gen2</strong> can limit how many queries evaluate concurrently during a refresh. In a ten-query test, setting concurrency to 1 increased refresh duration to 12 minutes 29 seconds, while allowing up to 250 concurrent evaluations reduced it to approximately 1 minute 33 seconds.</p><p>Default concurrency is variable, and higher concurrency is not universally better. Parallel evaluations can compete for available resources, so the useful setting depends on query duration, source throttling, destination load and capacity contention.</p>
Measured concurrency tests show substantial refresh-time differences and warn that maximum parallelism can increase contention.
<p>Power BI tenant migration covers three materially different patterns: side-by-side cross-tenant migration, tenant split, and tenant remap. Cross-tenant moves support controlled coexistence but require duplicate licensing and capacity; tenant splits must untangle users, workspaces, gateways, shared models, and dependencies; tenant remap preserves the Microsoft 365 tenant identity but deletes and recreates the Power BI tenant in another region.</p><p>There is no one-click migration tool. Admin APIs and Git can help export or rehydrate selected definitions, but apps, dashboards, RLS memberships, refresh schedules, governance settings, Lakehouse tables, and schemas require additional rebuilding or data migration. Remap downtime may be three to 24 hours, followed by restoration and validation.</p>
The post details tenant migration patterns, downtime, unsupported artifacts, Fabric gaps, backup validation, and realistic operational trade-offs.
<p><strong>SQL Analytics Endpoint metadata sync</strong> introduces a new preview architecture for parsing Delta logs and separating data changes from schema changes. On-query refresh is intended to expose newly landed lakehouse data in seconds rather than minutes.</p><p><strong>Time travel</strong> is also available in preview, allowing T-SQL queries against historical table states within the lakehouse retention period. This supports audit validation and report investigation without maintaining copied historical tables; retention configuration therefore determines how far back the endpoint can query.</p>
Preview SQL Analytics Endpoint updates target near-real-time metadata visibility and historical Delta querying without duplicate datasets.
Power BI Updates
May 20, 2026
2/5 · Tier B
Report Authoring
announcement
<p>The preview redesigns Power BI Desktop's <strong>Power Query Get Data</strong> experience around faster connector discovery, simpler connection configuration, and improved accessibility. Home, New, Recent data, OneLake Catalog, Blank Table, and Blank Query provide clearer entry points, while authentication and connection settings are consolidated.</p><p>The experience includes keyboard navigation and dark mode. It must be enabled under Desktop preview features and restarted, so teams should treat it as a user-interface trial rather than a production migration capability.</p>
The preview improves connector discovery and accessibility, but changes user workflow rather than platform architecture or governance.
<p><strong>OneLake item-size reporting</strong> adds a workspace-settings report showing storage by item, including visible, hidden, system, and soft-deleted data. Workspace administrators can sort and search the results, then refresh the report when they need current figures.</p><p>Each refresh scans and aggregates item contents, so runtime and <strong>CU consumption</strong> vary with workspace size. Results are cached, allowing other workspace administrators to inspect the latest report without triggering another scan. The capability is currently in preview.</p>
Preview reporting exposes item-level OneLake storage, including hidden and soft-deleted data, with refresh compute costs.
<p><strong>Notebook Export Control</strong> adds tenant-level and workspace-level governance over notebook downloads and data exports from rich DataFrame previews. When disabled at the tenant level, notebook downloads, table downloads, and notebook-rendered local exports are removed, including for workspace administrators.</p><p>Notebook execution, OneLake reads and writes, and governed downstream consumption continue to work. The setting is configured under Fabric Admin Portal tenant settings and applies immediately across affected workspaces.</p>
Tenant-enforced notebook export restrictions directly reduce sensitive-data exfiltration paths in regulated Fabric environments.
<p><strong>Fabric Warehouse connection handling</strong> now becomes deterministic. A missing or <code>master</code> Initial Catalog connects to <code>master</code>; an invalid warehouse fails instead of routing to a random warehouse. Explicit warehouse names or IDs continue to connect directly.</p><p>Auditing is database-scoped, so a session that starts on <code>master</code> has no warehouse context and its initial authentication event is absent from that warehouse’s audit log. Queries issued after <code>USE [warehouse]</code> can be audited. Complete traceability requires correlating database audit logs with Fabric Platform Logs.</p>
GA connectivity behavior changes affect deterministic routing and the completeness of warehouse audit trails.
<p><strong>Copilot-generated insights</strong> in Real-Time Dashboards can now be shared through links. A recipient sees the query results and optional visualization in a read-only view, then can rerun the query against current data.</p><p>Recipients can save the query to a KQL Queryset, modify the visual, reshare it, or add it to a dashboard. The capability requires a Fabric-enabled capacity and a real-time dashboard connected to a single data source.</p>
Useful preview for sharing live Copilot findings, but provides limited governance, cost, or operational detail.
<p>Eligible Fabric and Power BI users can create support tickets directly from the Fabric Help Pane. The flow can include limited session metadata, such as tenant location and artifact identifiers, giving support engineers context from the page where the problem occurred.</p><p>The tenant’s <strong>Publish “Get Help” information</strong> setting controls whether users see Contact Support or are directed to an internal help desk. Ticket creation also depends on eligible support-plan entitlements and user consent for advanced telemetry access.</p>
In-product support captures artifact context and exposes a tenant-controlled help-desk path, with limited architectural impact.
<p><strong>Outbound Access Protection</strong> for semantic models is available in preview as a workspace-level deny-by-default network policy. Import refreshes and DirectQuery queries are evaluated at the bound data connection, so Power Query expressions and parameters cannot bypass the rule. Composite models are also controlled because filter values can otherwise cross source boundaries and appear in another system's query logs.</p><p>Enabling the policy blocks destinations unless explicit connection exceptions exist. Even a lakehouse in the same workspace may require SQL and ADLS Gen2 exceptions for SQL endpoint and Direct Lake access. The policy requires an F SKU, tenant enablement, approximately 15 minutes to propagate, and a workspace containing only supported artifacts; reports and several other items prevent activation.</p>
Semantic-model outbound protection changes network behavior and documents enforcement, propagation delay, exceptions, and unsupported workspace contents.
<p><strong>Fabric Jumpstart</strong> consolidates Microsoft-sponsored and community-built accelerators, tutorials, demos, and reference implementations into a searchable catalog. Solutions can be installed through a Python package, including end-to-end scenarios such as healthcare billing and platform monitoring.</p><p>The catalog is valuable for rapid exploration and proof-of-concept work, but the extract does not describe support boundaries, version pinning, security review, or production-readiness guarantees for community solutions.</p>
A useful catalog of runnable demos and accelerators supports evaluation, but the announcement provides limited production evidence.
<p><strong>SSMS 22.5</strong> can create, rename, and delete Fabric Warehouses from Object Explorer, and can create a SQL database project from a warehouse connection. Connecting now requires item-level permissions rather than workspace-level permissions.</p><p>The database-project workflow moves warehouse development closer to version-controlled deployment, although several related capabilities—such as object script generation and government-cloud support—remain on the roadmap.</p>
SSMS 22.5 improves warehouse administration and database-project workflows, with meaningful item-permission and migration implications.
<p><strong>semantic-link-labs 0.15.1</strong> expands notebook automation around Direct Lake, mirrored catalogs, OneLake lifecycle policies, and semantic-model security. Direct Lake generation can now include lakehouse descriptions, while <strong>list_data_access_roles</strong> exposes both row-level and column-level security details.</p><p>New functions manage mirrored Azure Databricks catalogs, export or import OneLake lifecycle policies, and synchronize source descriptions into Direct Lake models. Lakehouse maintenance also gains Z-ordering and deletion-vector cleanup controls, which can affect maintenance duration and storage behavior.</p>
Adds Direct Lake metadata automation, mirrored-catalog operations, security visibility, and OneLake lifecycle-policy APIs.
Fabric Updates
May 18, 2026
4/5 · Tier A
Spark RuntimeCapacityDelta & Parquet
announcement
<p><strong>Resource Profiles</strong> provide workload-aware Spark configurations instead of requiring teams to tune many individual Spark properties. Available profiles include write-heavy, read-heavy for Spark, and read-heavy for Power BI, with settings optimized for different read, write, and consumption patterns.</p><p>Profiles can be configured at the environment level and overridden when required. New workspaces default to a write-optimized profile, while existing workloads are intended to remain unchanged unless profiles are applied. The extract provides no measured performance or CU-cost results, so the claimed price-performance benefits still require workload testing.</p>
Workload-aware Spark profiles change configuration behavior and target performance and cost variability across medallion workloads.
<p><strong>User-aware calculated columns</strong> use the <strong>Expression Context</strong> property set to <strong>User Context</strong> to evaluate at query time under the current user’s security context. Unlike standard calculated columns, they are not materialized in memory, even in Import models.</p><p>The feature supports localized values through <strong>USERCULTURE()</strong>, row-level virtual calculations, and selected security scenarios. Values are evaluated row by row in an otherwise empty filter context, remain consistent across visuals for the same user, and can still participate in filters and slicers.</p><p>Virtual columns introduce query-time cost and limitations. High-cardinality dependencies can increase execution work, and localized columns should use stable numeric keys through <strong>Group By Columns</strong> so selections survive language changes.</p>
Introduces user-aware virtual columns with materialization, localization, security, and query-time behavior implications.
<p><strong>Fabric Data Warehouse</strong> is positioned as a progressive modernization target for Synapse dedicated SQL pool customers, rather than a forced rip-and-replace migration. Existing ADLS Gen2 data can remain in place through <strong>OneLake shortcuts</strong>, while warehouse objects and pipelines move incrementally into Fabric.</p><p>The proposed benefits include online scaling, fewer manually managed concurrency controls, intelligent distribution, autonomous workload management, shared OneLake storage, virtualization, and near-real-time mirroring. These claims describe the intended operating model, but the extract provides no workload benchmarks, capacity sizing data, migration compatibility matrix, or detailed limits for stored procedures, distribution behavior, or custom SQL pools.</p><p>The migration mapping is straightforward: ADLS Gen2 becomes shortcut-accessible, Synapse SQL objects move to Fabric Warehouse, and ADF or Synapse pipelines move to Fabric Data Factory. Security and governance also shift toward shared Fabric controls, including OneSecurity and workspace or capacity isolation.</p>
Provides a concrete Synapse dedicated SQL pool modernization path, migration mappings, and important architectural trade-offs.
<p>ClickHouse integrates the Rust <strong>Delta Kernel</strong> to centralize Delta transaction-log parsing, snapshot resolution, schema handling, and data skipping. ClickHouse retains control of performance-critical Parquet reading while the Kernel supplies table metadata, statistics, deletion vectors, and required logical-to-physical transformations.</p><p>The integration adds transactional writes, schema evolution, time travel, partition pruning, and statistics-based pruning without duplicating the evolving Delta protocol inside ClickHouse. The reported query example processed 100 million rows and 14.82 GB in 3.878 seconds, using 9.16 GiB of peak memory, although that result is workload-specific.</p>
Delta Kernel gives ClickHouse maintained protocol semantics, transactional writes, schema evolution, time travel, and pruning across engines.
<p>Published <strong>model refresh parallelism</strong> limits are guaranteed concurrency levels, not absolute hard caps. Chris Webb’s F2 tests show two and six simultaneous 120-second refreshes can succeed despite F2 documenting support for one concurrent refresh.</p><p>Under heavier contention, refreshes are retried automatically with increasing delays. Fifteen simultaneous refreshes eventually succeeded, but some took 305 seconds; sustained contention can still produce final failures. The limit is distinct from instantaneous CU consumption, although concurrency protects capacity from large resource spikes.</p><p>Use the capacity Refresh Schedule, Monitoring Hub, Workspace Monitoring, and Log Analytics to identify overlaps. Reduce unnecessary refresh frequency, shorten refresh duration through query and model tuning, use incremental refresh, or redistribute work across capacities.</p>
Measured concurrency tests expose refresh limits, retry behavior, scheduling effects, and concrete capacity-remediation options.
<p><strong>semantic-link-labs 0.15.0</strong> adds substantial Fabric automation for schema-enabled lakehouses, mirrored Azure Databricks catalogs, Direct Lake models, semantic-model perspectives, workspace tags, and SQL database recovery. It can create schemas and materialized lake views, query mirrored catalogs, and manage their lifecycle.</p><p>Direct Lake generation now supports lakehouses, warehouses, SQL databases, mirrored Azure Databricks catalogs, and mirrored databases. Semantic-model deployment can create a filtered mini model from a perspective, but that capability is limited to single-lakehouse-sourced Direct Lake models.</p><p>The release also exposes tenant-setting and sensitivity-label metadata, adds CMK revalidation, and makes TOM add operations return the created object, improving notebook-driven deployment workflows.</p>
Introduces broad source support, mirrored Databricks catalog management, materialized lake views, and semantic-model deployment controls.
<p>14 features - 1 must-know - 10 worth knowing.</p><ul><li>Deprecating the Built-In Netezza ODBC Driver</li></ul>
1 must-know of 14 features
<p><strong>Custom authorization for Fabric API for GraphQL</strong> uses an Authorizer User Data Function before a GraphQL operation returns data. The function receives request context such as identity, roles, or tenant information and decides whether the request is allowed, restricted, or denied.</p><p>The pattern moves business-specific authorization into programmable policy logic instead of relying only on the API’s built-in permission model. It suits multi-tenant APIs and role-sensitive fields, but the extract does not specify execution limits, failure behavior, caching, latency, auditing details, or how field-level restrictions are represented.</p><p>Because the capability is in public preview, policy behavior and operational guarantees should be treated as subject to change before using it as the sole control for regulated data.</p>
Preview authorization hooks add programmable tenant and role policies to Fabric GraphQL APIs, directly affecting application security design.
Fabric Updates
May 13, 2026
3/5 · Tier B
AI ReadinessReal-Time Intelligence
announcement
<p><strong>Fabric Data Agent</strong> can now discover and use Eventhouse user-defined functions, materialized views, and shortcut tables. This extends agent grounding beyond raw KQL tables: validated functions can supply domain logic, materialized views can answer common aggregations efficiently, and shortcuts can expose data stored outside the Eventhouse.</p><p>Eventhouse functions help preserve time-zone handling, filtering, deduplication, and other business rules that an agent might otherwise recreate incorrectly. Materialized views can reduce scans over high-volume telemetry, while shortcut tables provide federated access without copying the underlying data.</p><p>The extract describes automatic discovery but provides no accuracy measurements, query-routing guarantees, freshness limits, cross-source join constraints, or additional capacity cost. Metadata quality and careful review of generated answers therefore remain important.</p>
Data Agent can now use validated KQL logic, precomputed views, and external shortcut tables, improving grounded real-time answers.
<p><strong>Fabric Eventstream</strong> can now publish Business Events directly as a native destination. Teams can filter, aggregate, and apply thresholds on streaming data, then emit a governed, discoverable event from the same no-code pipeline.</p><p>A Business Event is a schema-defined signal with organizational meaning, rather than raw telemetry. Publishing it through Real-Time Hub decouples the producer from consumers such as alerts, dashboards, analytics, and operational workflows. Notebooks and User Data Functions remain better choices when the pipeline needs complex transformations, machine-learning scoring, or multi-stream joins.</p><p>The extract does not describe event delivery guarantees, ordering, deduplication, schema evolution, retention, or capacity consumption. Those omissions matter when a signal drives safety or financial operations.</p>
Eventstream gains a no-code path for publishing governed business signals, enabling reusable real-time contracts for downstream consumers.
<p><strong>SQL database in Fabric</strong> now supports a Git-based workflow for managing database schema alongside application code. Schema files can be developed in VS Code, committed to branches, reviewed through pull requests, and applied to the Fabric workspace after approval.</p><p>The workflow provides history and change traceability instead of relying on direct production edits or manually executed scripts. SQL Database Projects can structure and validate the schema before publication, while Git integration supports GitHub and Azure DevOps repositories.</p>
Fabric SQL Database schema integrates with GitHub and Azure DevOps, enabling reviewed, traceable, rollback-ready database changes.
Fabric Updates
May 13, 2026
4/5 · Tier A
AI ReadinessAPI AutomationGovernance
announcement
<p><strong>Fabric Data Agents</strong> now support service-principal authentication in preview. A custom application or backend can obtain an Entra ID application token and call a published data agent without requiring every end user to authenticate directly to Fabric.</p><p>The service principal receives workspace access, which the data agent inherits, and can support both direct application calls and Microsoft Foundry agent integrations. This replaces delegated user-token flows for application-owned workloads and enables auditable, centrally managed non-human identities.</p><p>Support for data agents connected to KQL databases is not yet available and is described as coming soon. The extract also does not detail token scopes, row-level security behavior, throttling, secretless credential options outside Azure Web Apps, or preview service-level guarantees.</p>
Service-principal authentication enables production applications to call Fabric Data Agents without delegated user sign-in, with explicit KQL limitations.
<p><strong>OneLake storage tiers</strong> add hot, cool, and cold options for balancing storage cost against access cost. Cool and cold reduce monthly storage charges but introduce higher transaction costs, retrieval charges, and substantially greater capacity consumption when large datasets are read.</p><p><strong>Lifecycle management</strong> automatically changes file tiers using creation, modification, or access conditions. Policies can move aging bronze data to cooler tiers and return frequently accessed files to hot storage. Workspace administrators can configure policies through workspace settings or use the lifecycle-management APIs.</p><p>The example shows the tradeoff clearly: storing 10 TB for 60 months costs $13,800 hot, $7,500 cool, or $2,400 cold, while a 1 TB read consumes approximately 2,600, 206,500, or 665,000 CU seconds respectively. Tiering therefore requires workload-based cost modelling, not storage-price comparison alone.</p>
Preview storage tiers include concrete storage, retrieval, transaction, and capacity-consumption tradeoffs.
<p><strong>Configurable Data Retention</strong> for Fabric Warehouse is in preview, allowing each warehouse to retain historical data versions for 1 to 120 days through T-SQL. The retention window supports time travel, point-in-time clones, restore points, and warehouse snapshots.</p><p>Lowering retention permanently removes history outside the new window through managed garbage collection; increasing it later cannot recover deleted versions. This setting is separate from dropped-warehouse retention, which defaults to seven days and can be configured from 7 to 90 days.</p>
Preview retention controls provide concrete recovery windows and include a critical irreversible cleanup behavior.
<p><strong>Data Warehouse Monitor</strong> provides a unified preview experience for active and completed Fabric Warehouse queries. It combines live workload visibility, historical execution comparisons, performance-pattern analysis, and one-click cancellation of long-running queries.</p><p>The experience is powered by <strong>Query Insights</strong>, which supplies execution-level telemetry for identifying regressions and recurring resource-intensive workloads. The article does not provide capacity-cost measurements, retention limits, permissions detail, or quantified performance improvements.</p>
Unified active and historical query monitoring improves Fabric Warehouse troubleshooting and enables direct cancellation of problematic queries.
<p>The <strong>maximum allowable memory allocation</strong> error can occur while refreshing a semantic model or rendering a report when its assigned service node cannot provide the required resources. The condition is often temporary because other models on the node may be refreshing or executing expensive queries.</p><p>A single occurrence may be transient, but repeated failures suggest memory spikes from the model or its DAX workloads. Report failures require inspecting visual-generated queries and measure design; refresh failures call for reducing model size and eliminating memory-heavy calculated columns or tables.</p><p>The error does not necessarily mean the model exceeded a documented per-model limit. It can reflect the changing memory and CPU pressure of co-located models, so diagnosis should compare query or refresh memory consumption with concurrent activity.</p>
Rare memory-allocation failures reveal node-level contention and provide targeted remediation for query and refresh workloads.
<p>The <strong>Execute DAX Queries REST API</strong> runs DAX against Power BI semantic models and returns results as Apache Arrow IPC rather than JSON. Columnar binary output preserves native data types, supports multiple <code>EVALUATE</code> statements, and is designed for high-volume extraction.</p><p>The API is currently available only for semantic models hosted on Power BI Premium or Fabric capacities. Callers must process Arrow streams, and Fabric notebooks can authenticate through <strong>notebookutils</strong> using the current user or workspace identity without an explicit app registration.</p><p>Results can be converted to pandas and persisted to Delta. Clean column names and apply V-Order when the output will later feed Direct Lake, but treat extraction as a capacity workload and govern query volume, result size, and permissions.</p>
Preview REST API adds Arrow-based DAX extraction, multi-statement queries, and Fabric notebook automation with important capacity limits.
<p>DuckDB’s Delta extension now supports <strong>INSERT</strong>, time travel, and snapshot pinning, while its Unity Catalog extension supports both open-source and Databricks catalog endpoints. Multiple inserts inside one transaction become a single atomic Delta version; UPDATE, MERGE, and DELETE remain unsupported.</p><p>Catalog Managed Tables add catalog-coordinated commits, making Unity Catalog the arbiter for concurrent writes instead of leaving the catalog’s metadata, audit trail, and statistics disconnected from direct Delta-log changes. The update is useful for lightweight cross-engine access, but it is not a complete replacement for a production transformation engine.</p>
DuckDB’s stable Delta and Unity Catalog extensions add writes, time travel, and coordinated catalog commits for open-table interoperability.
<p><strong>Fabric Livy API High Concurrency sessions</strong> allow multiple Spark workloads to run in parallel within a managed underlying session. Each request receives an isolated REPL, while Fabric handles session packing, reuse, monitoring, and billing.</p><p>Clients supply a <strong>sessionTag</strong> so Fabric can route work to reusable sessions. Up to five REPLs can run per underlying Livy session, and failures or cancellations are isolated between requests. High-concurrency jobs appear in Monitoring Hub with an HC-specific identifier.</p><p>The feature is in preview. It reduces session-start and idle-allocation overhead, but the five-REPL ceiling, session state, capacity availability, and preview API contracts need to be included in orchestration design.</p>
Preview Livy high concurrency enables parallel isolated Spark automation with explicit session reuse and a five-REPL limit.
Fabric Updates
May 6, 2026
3/5 · Tier B
AI ReadinessGovernance
announcement
<p><strong>OneLake catalog</strong> is now natively available in Microsoft Foundry knowledge workflows. AI builders can browse governed Fabric assets, review ownership, endorsement, sensitivity, and location, and select a source without manually copying URLs or identifiers.</p><p>The integration creates a path from governed discovery to grounded AI, but it requires a Fabric lakehouse, a Foundry project, an Azure AI Search service at Basic tier or higher, same-tenant placement, and a managed identity with appropriate role-assignment permissions.</p><p>The feature improves source selection and reduces integration friction; it does not remove the need to verify whether the chosen asset is suitable, current, permissioned, and accurate for agent grounding.</p>
GA OneLake catalog integration simplifies governed data discovery in Foundry, but requires Azure AI Search and managed-identity setup.
<p>The <strong>semantic model settings pane</strong> keeps configuration in context instead of navigating away from the workspace. Collapsible sections and search make a large collection of settings easier to reach, including refresh history, credentials, gateways, sensitivity labels, endorsements, and query caching.</p><p>The preview is opt-in from the existing settings page and is currently off by default. Microsoft plans to make it default, retain a fallback to the classic page temporarily, and eventually retire that page. Editing still requires ownership or Write permission, while credentials and scheduled refresh remain owner-only settings.</p>
Preview settings redesign improves navigation but changes administration workflow without materially changing semantic-model behavior.
<p><strong>OneLake security</strong> is generally available as a fine-grained access-control layer for OneLake data. Roles can secure items, folders, tables, rows, and columns across Spark, Power BI, Fabric data agents, and external applications such as Excel.</p><p>New role creation supports <strong>RLS</strong> and <strong>CLS</strong> during authoring, with inline validation and recommendations. Cross-region shortcuts, item-level sharing without Viewer permission, workspace private links, and identity-mode defaults for new SQL analytics endpoints also improve administration.</p><p>New items have security enabled by default, while existing supported items are automatically upgraded during the May rollout. Existing permissions are not changed, but administrators gain the ability to create and edit OneLake security roles. Single-role management APIs are also available.</p>
GA security model, default enablement, cross-engine enforcement, role APIs, and rollout behavior materially affect platform governance.
<p><strong>Direct Lake on SQL</strong> connects a semantic model to a Fabric Warehouse’s SQL analytics endpoint while loading Delta Parquet data into VertiPaq memory on demand. Loaded data delivers import-like query performance without a separate import copy, and remains resident until eviction from inactivity, data changes, or memory pressure.</p><p>The model can fall back to DirectQuery when Direct Lake guardrails or unsupported conditions are encountered. Warehouse-managed file consolidation, statistics, V-Order, and clustering reduce the manual maintenance burden associated with lakehouse storage and help avoid file-related fallback conditions.</p><p>Performance remains strongly dependent on memory consumption and column cardinality. Integer keys, focused star schemas, right-sized data types, dimension separation, and measures instead of calculated columns reduce the amount of data that must be transcoded and held in memory. Direct Lake on OneLake is presented as the longer-term strategic path, especially for multi-source and OneLake Security scenarios.</p>
The deep dive explains Direct Lake on SQL internals, fallback, cardinality-driven memory, and warehouse storage optimizations.
<p><strong>My queries</strong> is a Dataflow Gen2 preview feature that stores reusable query definitions in a personal library. Users save a query from the Queries pane, then import its M code into another dataflow through the modern Get Data experience.</p><p>Fabric stores only the <strong>M code</strong>; credentials, destinations, and other connection metadata remain local to each dataflow. Saved items live in a My queries folder in My Workspace, and duplicate names receive numeric suffixes.</p>
Preview productivity feature reduces duplicated M authoring but lacks enterprise governance, deployment, or measured performance detail.
<p>The <strong>Fabric capacity metrics app</strong> now provides a generally available health page covering capacities by region, utilization, and interactive, background, and total throttling signals. Capacity cards include average and 95th-percentile metrics with drill-down into usage patterns.</p><p><strong>Timepoint summary</strong> shows which workspaces, items, and operations consumed compute during a selected window, separated by interactive or background activity. Timepoint detail adds user identity where available, operation IDs, burndown, and overage context for root-cause analysis.</p><p>The generally available <strong>Chargeback app</strong> breaks usage down by workspace, SKU, and workload type, with filtering and export. It supports internal allocation decisions, but the organization still must define whether costs follow workspaces, capacities, departments, unused allocation, or reservation benefits.</p>
GA capacity health, timepoint diagnostics, and chargeback materially improve throttling investigation, attribution, and SKU decisions.
<p>A slicer cannot filter a measure directly because measures have no inherent filter granularity. The correct pattern lets a slicer set a threshold, then evaluates a measure over an explicit entity such as Product, Customer, or Store.</p><p>A reusable <strong>DAX UDF</strong> can wrap <strong>CALCULATE</strong>, <strong>FILTER</strong>, and <strong>KEEPFILTERS</strong>, while a calculation group exposes the behavior across measures. The pattern can be extended with disconnected tables so users select both the threshold and the entity granularity.</p>
Provides a reusable calculation-group and UDF pattern for parameterized entity filtering at explicit granularity.
The <strong>Approval activity</strong> adds human-in-the-loop control to Fabric pipelines in preview. A pipeline can pause for approval or rejection, branch on the outcome, handle timeouts, and notify reviewers through Outlook or Teams.
This extends pipelines from data movement into auditable business workflows, including semantic model refresh sign-off and regulated validation steps. The article provides no detailed limits, licensing behavior, timeout semantics, or operational cost information.
Approval activity brings human gates into pipelines, but the article is primarily a preview announcement without implementation limits or cost detail.
<p><strong>Binding hints</strong> let a semantic model identify the service connection it should use when published. The hint stores a JSON data-source signature containing the connection kind and path, plus the Shareable Cloud Connection’s connection ID.</p><p>The feature requires compatibility level 1608 or higher and is configured in TMDL View. Multiple hints can support multiple source connections or repeated source types. The Fabric list connections API is the preferred way to obtain the kind and path metadata.</p><p>After publishing, the model can already be linked to the intended connection and refreshed without a separate manual binding step or API call. This is especially useful for programmatically generated models, but the connection must already exist and be accessible in the target environment.</p>
Binding hints enable publish-time connection binding, removing manual or API rebinding in generated semantic-model deployments.
The <strong>OneLake Catalog Search API</strong> provides cross-workspace discovery using free-text metadata matching, item-type filters, and lightweight catalog entries. Results expose stable identifiers but do not grant access to item content.
The same capability is available through the Fabric Core MCP Server and Fabric CLI. <strong>fab find</strong> supports filtering, detailed output, paging, and JMESPath queries, making discovery composable in scripts and agent workflows.
The preview currently searches cataloged items and accessible metadata; richer entity coverage and filtering are planned.
Programmatic cross-workspace discovery enables catalog applications, CLI automation, and agent workflows with stable item identifiers.
<p>19 features - 6 must-know - 12 worth knowing.</p><ul><li>Semantic Link (SemPy) 0.14.0: admin APIs and more</li><li>Deprecating the Built-In Netezza ODBC Driver</li><li>Manage ML experiments and models across workspaces and platforms</li><li>ALTER TABLE in explicit user transaction (Generally Available)</li><li>Stream Mirrored Database change feeds into Eventstreams (Preview)</li><li>Custom CA and mTLS support in Eventstream connectors (Preview)</li></ul>
6 must-know of 19 features
<strong>Eventstream workspace monitoring</strong> automatically creates three tables in a Fabric-managed monitoring Eventhouse: node status, flow metrics, and error metrics. KQL queries, dashboards, and alerts can use the resulting telemetry without custom instrumentation.
Flow and error metrics update approximately every minute, while node status is emitted approximately every six hours. Existing Eventstreams must be republished after monitoring is enabled, because the configuration is applied at publish time. Diagnostic logs and more frequent status updates are planned but not included in the preview.
Workspace monitoring creates queryable Eventstream health, volume, latency, and error telemetry with concrete update intervals and setup behavior.
Power BI Updates
Apr 30, 2026
2/5 · Tier B
Report Authoring
announcement
<p>The <strong>new Card visual</strong> is generally available with hero images, updated default styling, collage layouts, and configurable ordering for callout areas, reference labels, and images. Images can be uploaded, supplied by URL, or sourced from a data column.</p><p>New reports use the modern defaults automatically. Existing reports can opt in through the report theme, while the legacy Card visual continues working in existing reports and can be restored through the visual context menu.</p>
The GA Card visual adds useful layout and image controls, but remains a presentation-level report authoring change.
Power BI Updates
Apr 30, 2026
2/5 · Tier B
Report Authoring
announcement
<p>The Power BI <strong>Image visual</strong> now supports state-based formatting for all, default, hover, and pressed states. Authors can apply different backgrounds, borders, corner shapes, and image effects to create interactive navigation or branded report elements.</p><p>Images can come from local uploads, URLs, or data columns, with Fit, Fill, Center, and Stretch options plus exposure, contrast, saturation, and blur controls. Authors can also replace an image in place instead of recreating the visual.</p>
GA-style image states and sources expand report design, but provide limited architectural or operational impact.
Power BI Updates
Apr 30, 2026
2/5 · Tier B
Report Authoring
announcement
<p><strong>Modern visual tooltips</strong> are generally available across Power BI Desktop, the service, mobile, Teams, and embedded reports. The Actions footer lets consumers drill down, drill up, or drill through directly from a data point tooltip.</p><p>New reports use the modern experience by default, with styling inherited from the report theme. Existing reports retain their current appearance until authors reset visuals to the default styling; individual formatting changes affect only the selected visual.</p>
GA tooltip interactions improve report navigation, but the change is primarily authoring and presentation focused.
<p><strong>Direct Lake and import composite models</strong> combine tables stored in Direct Lake on OneLake with imported tables from Power Query sources. This allows a model to keep large analytical tables close to Delta storage while importing smaller dimensions or operational targets that need calculated columns, hierarchies, or connector support.</p><p>The preview supports regular relationships between Direct Lake and import tables, improving on older DirectQuery/import limitations. Web modeling can add or convert tables, refresh import data, reframe Direct Lake tables, and synchronize schema; Desktop live editing remains available with fewer source-selection options.</p><p>A practical design is to retain high-volume fact tables in Direct Lake and import small, slowly changing tables such as targets or manually maintained mappings. Import tables still require refresh, while Direct Lake performance remains dependent on Delta layout, metadata, and fallback behavior.</p>
Deep dive demonstrates preview composite models combining Direct Lake and import tables, relationships, web modeling, and practical migration trade-offs.
The April 2026 gateway release adds <strong>admin-triggered auto-update</strong> as a generally available capability. Gateway administrators can initiate upgrades during approved maintenance windows instead of relying only on background updates.
Version 3000.314 aligns gateway execution with the April Power BI Desktop release and its connectivity changes. The operational benefit is predictability for scheduled refreshes; the release does not describe new Fabric-native connectivity or performance behavior.
Admin-triggered gateway updates improve maintenance control, but the release mainly adds operational lifecycle management.
<p>The Fabric monitoring hub now provides a <strong>Schedule failures</strong> page for viewing and managing failure-notification recipients across scheduled items. Users can configure, edit, or remove notifications without opening each pipeline, notebook, or Dataflow Gen2 item.</p><p>The page shares configuration with each item’s scheduler. Contributor access or item Write permission is required, and manually triggered runs do not generate these notifications.</p>
Centralized scheduled-failure notification management reduces operational effort, but applies only to scheduled runs.
The <strong>Fabric Eventstreams SQL operator</strong> is generally available as a code-first transformation and routing component. One SQL definition can send processed events in parallel to Eventhouse, Lakehouse, Activator, or downstream Eventstreams.
GA adds built-in testing for individual outputs and event-time processing with controls for late and out-of-order events. This centralizes logic that previously required duplicated transformations or more complicated topologies, while preserving Eventstream’s no-code options for simpler scenarios.
The SQL operator reaches GA with fan-out, event-time handling, late-data controls, and built-in multi-output testing.
<p><strong>Mirrored Dremio catalog</strong> brings Dremio-managed Apache Iceberg tables into OneLake without copying data or building ETL pipelines. Fabric connects to the Dremio Iceberg REST Catalog, mirrors catalog metadata, and creates shortcuts that expose selected tables across Fabric.</p><p>Dremio credential vending can avoid separately managing storage credentials, while automatic table inclusion and ongoing metadata synchronization keep the Fabric view current. The feature is preview, and on-premises gateway connectivity is not yet available, so network-restricted deployments may need to wait.</p>
Preview zero-copy Dremio Iceberg access broadens Fabric integration, but operational limitations and production evidence are sparse.
<p><strong>Delta 4.2</strong> expands Delta Kernel with an experimental Apache Flink connector supporting catalog-managed tables and transactionally consistent writes. Kernel also adds geospatial, collation, and Variant support, with Variant shredding graduating from preview.</p><p>Catalog-managed tables gain atomic <strong>REPLACE TABLE AS SELECT</strong> and Dynamic Partition Overwrite, so failed operations leave the prior table state untouched. SQL <strong>INSERT BY NAME</strong> can evolve schemas when auto-merge is enabled, and UniForm generates Iceberg metadata synchronously at commit time. Legacy Hive Metastore support for UniForm is deprecated.</p>
Delta 4.2 adds a Kernel Flink connector, atomic catalog operations, SQL schema evolution, synchronous UniForm, and new data types.
Crossjoin
Apr 26, 2026
2/5 · Tier B
AI Readiness
tutorial
<p><strong>FabricAI.Prompt()</strong> in Dataflows Gen2 can generate synthetic tabular content, not just process one prompt per input row. A prompt requests CSV with defined columns, values, row count, and headers; <strong>CSV.Document</strong> then parses the returned text into a table.</p><p>This provides a lightweight way to create mock data for demos, transformation development, and pipeline prototyping. The approach depends on the model following the requested schema, so generated values should not be treated as statistically controlled test data without validation.</p>
Practical FabricAI.Prompt sample-generation technique is useful, but lacks limits, cost, governance, or production reliability analysis.
<p>This monthly spotlight aggregates community posts, videos, and presentations covering Fabric Warehouse, Power BI, data engineering, governance, ingestion, and Real-Time Intelligence. The entries range from SQL AI-function experiments and Custom Totals benchmarking to mirroring, metadata-driven ingestion, and Eventstream tutorials.</p><p>It is useful as a discovery index, but the roundup itself does not provide enough implementation detail, measurements, or limitations to support an architecture decision.</p>
Monthly influencer roundup amplifies varied community content but provides little original technical detail or validated implementation evidence.
<p><strong>Nested folder support</strong> is generally available for shortcut transformations that convert structured CSV, Parquet, or JSON files into Delta tables. A shortcut can target a top-level folder, recursively discover files and new partitions, apply one transformation configuration, and preserve relative directory structure.</p><p>Cycle prevention excludes shortcuts nested inside the target folder, limiting traversal to physical folders and files. Existing shortcuts retain single-level behavior unless configured otherwise, while new shortcuts enable subfolder processing by default. The article does not specify file-count limits, schema-evolution behavior, or transformation processing latency.</p>
Generally available recursive shortcut transformations automatically process new nested partitions while preserving source hierarchy and preventing shortcut cycles.
<p><strong>Resource instance rules for OneLake</strong> allow workspace administrators to permit inbound access from explicitly trusted Azure resource instances. Requests arriving through the public endpoint are checked against approved Azure resource identities before normal data permissions are evaluated.</p><p>The rules address managed services with dynamic or unknown outbound IP addresses and can coexist with Private Link and IP firewall rules. Tenant administration must enable the relevant setting, after which workspace admins configure trusted Azure resource ARM IDs. The capability remains in preview.</p>
Preview identity-based network controls change how managed Azure services can securely access OneLake without static IP allowlists.
<p><strong>Cross-workspace MLflow logging</strong> is generally available and lets notebooks or external environments write experiments, metrics, parameters, and registered models directly to a selected Fabric workspace. The <strong>synapseml-mlflow</strong> plugin changes the MLflow tracking URI; standard MLflow APIs remain unchanged.</p><p>This supports Development-to-Test-to-Production separation, importing assets from Databricks, Azure Machine Learning, or local environments, and training where protected data resides while serving from another workspace. Workspaces with Outbound Access Protection require a cross-workspace managed private endpoint; same-workspace logging and external access without that restriction need no equivalent setup.</p>
GA cross-workspace MLflow logging enables environment-separated MLOps and external asset consolidation, with private-endpoint requirements under OAP.
<p><strong>Change Event Streaming</strong> in SQL Server 2025, Azure SQL Database, and Azure SQL Managed Instance pushes inserts, updates, and deletes directly to a Fabric Eventstream custom endpoint. Unlike polling-based CDC, it emits CloudEvents messages from the database change stream without intermediate change tables.</p><p>Eventstream provisions an Event Hubs-compatible endpoint supporting AMQP and Kafka, with Microsoft Entra managed identity, SAS token, or SAS key authentication. An Eventstream SQL operator can parse the nested JSON payload with <strong>json_parse()</strong>, filter operation types, and route refined data to derived streams, Eventhouse, alerts, or other destinations. SQL Server 2025 requires the FULL recovery model.</p>
Practical SQL-to-Eventstream integration provides near-real-time change delivery, authentication options, and downstream Eventhouse patterns.
<p><strong>Workspace outbound access protection (OAP)</strong> is generally available for Data Factory workloads, including Pipelines, Copy Job, and Dataflows. Workspace administrators define trusted outbound endpoints; connections to public internet destinations and other unapproved endpoints are blocked.</p><p>OAP applies at workspace level, allowing different controls for business units, environments, or data domains. Exploratory pipeline and Copy Job APIs, Data Warehouse destinations, notebook and Spark Job Definition connections, Machine Learning Models, and Experiments are supported. Data Agents and Eventstreams remain preview capabilities, while Power BI semantic models and reports are planned.</p><p>Combining outbound protection with inbound protection creates a stronger workspace boundary, but the practical limitation is endpoint allow-list coverage: every required external system must be explicitly reachable before production pipelines run.</p>
GA outbound controls materially improve exfiltration prevention and determine which Data Factory connections remain deployable.
<p><strong>Customer-managed keys (CMK) for Eventhouse</strong> are available in preview. Organizations can control the encryption keys used to protect Eventhouse data, supporting regulatory requirements that require customer ownership or lifecycle control of encryption material.</p><p>The extract provides no details about key rotation, regional availability, recovery behavior, service dependencies, performance impact, or pricing. Those limitations make this an early governance signal rather than a production-readiness guide.</p>
Preview CMK support matters for regulated Eventhouse deployments, but the announcement provides no configuration limits or cost detail.
<p><strong>Fabric Local MCP</strong> is generally available as an open-source server that gives compatible AI clients grounded access to Fabric API specifications, OneLake file and table operations, item creation, and Fabric CLI execution. Integrated authentication, retry handling, telemetry, and diagnostics reduce the custom integration work normally required for agent tooling.</p><p><strong>Fabric Remote MCP</strong> is in preview and provides a cloud-hosted endpoint for authenticated operations in a Fabric environment. Requests flow through Entra ID, Fabric RBAC, REST APIs, and audit logging, allowing agents to manage supported resources without local installation. Local MCP is primarily a developer and code-generation interface; Remote MCP is intended for operational agents and automation.</p><p>The key constraint is autonomy: an MCP client can invoke tools within the identity and permissions it receives. Preview Remote MCP therefore requires the same least-privilege and change-control treatment as any automation that can create items, alter workspaces, or manage permissions.</p>
Local MCP is GA and Remote MCP preview exposes Fabric APIs, OneLake operations, authentication, and agent execution boundaries.
<p><strong>DAX user-defined functions</strong> support specialized reference parameter types: <strong>MEASUREREF</strong>, <strong>COLUMNREF</strong>, <strong>TABLEREF</strong>, and <strong>CALENDARREF</strong>. These constrain callers to specific model objects instead of accepting arbitrary expressions through <strong>ANYREF</strong>.</p><p>The distinction affects semantics as well as documentation. A measure reference predictably triggers context transition inside an iterator, whereas an arbitrary expression may require an explicit <strong>CALCULATE</strong>. Specialized types also improve validation and IntelliSense, while rejecting incompatible scalar expressions at authoring time.</p>
Defines specialized DAX UDF references, improving validation and making context-transition behavior predictable.
<p><strong>fabric-cicd v1.0.0</strong> introduces breaking authentication changes: callers must provide an explicit token credential, and implicit notebook authentication and identity logging are removed.</p><p>The release adds Git-diff-based changed-item detection, Ontology support, hard-delete control, long-running-operation timeouts, and improved deployment reporting. It also prevents accidental GUID replacement and rejects non-YAML files during YAML parameterization, avoiding file corruption.</p>
Introduces explicit authentication requirements and selective deployment while fixing parameterization, notebook ordering, and deletion behavior.
<p>Agentic Fabric applications can use <strong>Fabric REST APIs</strong> for repeatable day-zero provisioning across development, test, and production. Encoding workspace items and configuration in scripts reduces manual setup and configuration drift while making infrastructure changes reviewable and version-controlled.</p><p>The updated architecture adds an optional <strong>Fabric Data Agent</strong> for read-only questions over explicitly configured Lakehouses, semantic models, and other supported data assets. Task agents reason and execute actions; the Data Agent answers questions using governed data and cannot create, update, or delete records. Its responses may use SQL, DAX, or KQL behind the scenes.</p><p>This separation improves security review and evaluation, but it does not make raw data automatically safe for natural-language access. The configured models, permissions, and exposed tables still determine what users can discover and query.</p>
REST-based deployment and read-only Data Agents provide concrete automation and governance improvements for production agentic applications.
<p><strong>dbt job activity</strong> is available in preview for Fabric data pipelines. It lets a pipeline select or create a dbt job, chain dbt execution with upstream and downstream activities, and monitor the dbt status in pipeline run history.</p><p>All settings support dynamic content, enabling runtime selection of jobs, models, or execution behavior. Advanced options include dbt commands such as build, run, compile, snapshot, and test, along with select, exclude, full refresh, and fail-fast controls. Teams can also configure Teams or email notifications after runs complete.</p><p>This consolidates orchestration in Fabric, but the extract does not describe preview service limits, retry semantics, concurrency, connection scope, or deployment behavior. Those details should be tested before replacing an established scheduler.</p>
Preview dbt orchestration adds useful pipeline integration, parameterization, and monitoring, but production limitations are unspecified.
<p><strong>Eventstream</strong> separates network security decisions by traffic direction. Internal Fabric connections inherit Fabric authentication, workspace permissions, and encryption, while inbound connections are secured with tenant- or workspace-level Private Link.</p><p>Outbound connectivity depends on the source: <strong>Managed Private Endpoints</strong> support Azure Event Hubs and IoT Hub, while preview VNet injection supports third-party brokers, cloud services, and database CDC sources. The key limitation is that each feature addresses a different connection pattern rather than providing one universal network boundary.</p>
Practical Eventstream network-security decision guidance covering traffic direction and private connectivity options.
<p><strong>Translytical task flows</strong> are generally available for turning report interactions into actions. A user selects records, enters values, and invokes a Fabric User Data Function that can insert, update, or delete data or call an external API.</p><p>The pattern supports annotations, data-quality corrections, approvals, notifications, and write-back directly from Power BI. Native User Data Function connections currently include Fabric SQL databases, Fabric warehouses, and lakehouse files. Updated data can appear in the same report after the operation.</p><p>The capability moves Power BI beyond read-only analysis, so authorization, validation, idempotency, auditability, and failure handling become part of the report design rather than optional workflow concerns.</p>
GA task flows connect report context to write-back and external actions, creating useful workflows but expanding transactional governance requirements.
<p><strong>Refresh History</strong> now reports warnings when a semantic model refresh completes despite broken calculated columns, measures, or calculated tables. The refresh succeeds, but affected objects remain unusable in reports.</p><p>Typical causes include deleting a referenced table or column. Power BI Desktop already surfaces these problems with warning indicators; the Service now exposes them after UI-triggered refreshes. XMLA-triggered refreshes do not currently produce the same warning behavior.</p>
Refresh warnings expose previously tolerated broken calculations without failing refresh, affecting semantic-model quality and deployment checks.
Fabric Updates
Apr 16, 2026
4/5 · Tier A
AI ReadinessAPI AutomationGovernance
announcement
<p><strong>OneLake MCP</strong> is generally available through the Fabric MCP Server and gives AI agents workspace discovery, table metadata, and file-system tools. The 19 commands can browse, read, and write OneLake data across lakehouses, mirrored databases, KQL databases, and other supported items.</p><p>Agents authenticate through the user's Azure identity and inherit existing Fabric permissions, so MCP does not create an independent authorization boundary. The example demonstrates schema discovery, physical-file inspection, replication-health checks, and optimization assessment without opening a notebook or issuing a query.</p>
GA OneLake MCP exposes 19 permission-aware tools for conversational discovery, inspection, and file operations.
<p><strong>Fabric Data Factory</strong> retains the core ADF execution engine and connector library while adding Fabric-native experiences such as Mirroring, Copy Jobs, Airflow, dbt, unified CI/CD, and Copilot-assisted development. Existing pipeline logic and operational knowledge generally transfer without a rewrite.</p><p>The strongest migration case is reducing integration sprawl: Mirroring handles continuous inserts, updates, and deletes, while Copy Jobs support bulk movement, incremental synchronization, CDC, and SCD Type 2 history. However, the guidance is largely strategic; it does not establish workload-specific feature parity, capacity costs, or a detailed migration runbook.</p>
Migration-oriented advisory compares ADF and Fabric Data Factory, but provides few concrete gaps, costs, or execution steps.
fabric.guru
Apr 14, 2026
2/5 · Tier B
AI Readiness
tutorial
<p><strong>Harrier</strong> provides multilingual text embeddings in 270M, 0.6B, and 27B sizes. The smaller models can run inside a Fabric Python notebook, with the model stored in a lakehouse and loaded for local inference.</p><p>The reported benchmark improves R1 from 0.68 with <strong>text-embedding-ada-002</strong> to 0.76 with the 270M model, while a Harrier-plus-BM25 pipeline reaches an R3 score of 99%. The article claims no additional Fabric CU consumption beyond notebook Python execution, but model loading time, notebook compute duration, and production orchestration still need measurement.</p>
Practical embedding benchmark with latency and retrieval results, but limited governance, operational, and cost analysis.
<p><strong>Eventhouse</strong> brings anomaly detection, data agents, SQL endpoints, and notebooks onto event and time-series data without duplicating it. Teams can investigate live signals, combine KQL, SQL, Python, and Spark, and expose the same data to agent-driven workflows.</p><p>The post describes the intended experience rather than implementation boundaries. It does not quantify detection behavior, SQL endpoint performance, agent grounding limits, notebook constraints, or the capacity and storage costs of these integrations.</p>
Preview Eventhouse integration overview highlights multiple experiences but supplies little implementation detail, limits, or cost information.
<p><strong>Capacity Scheduler</strong> lets Eventhouse define a recurring seven-day schedule for minimum capacity. Each day is divided into 60-minute blocks, and each block can specify a baseline or no minimum while autoscale remains available.</p><p>This changes the commitment model from one static weekly minimum to workload-aligned guarantees. It can reduce paid baseline resources during quiet periods, but scheduled minimums still need to fit available capacity; the interface warns when upcoming commitments exceed capacity and shows the next 24 hours before saving.</p>
Preview scheduler introduces explicit weekly minimum-capacity controls, directly affecting Eventhouse commitments and operational cost planning.
<p>Fabric is standardizing the <strong>Analyze data with</strong> menu across Lakehouse, Warehouse, and Eventhouse. The menu exposes SQL Endpoint, Eventhouse Endpoint, and new or existing Spark notebooks from a consistent location.</p><p>Eventhouse Endpoint can create child Eventhouse and KQL Database artifacts for Lakehouse or Warehouse data, with backend schema synchronization and no manual data duplication. The announcement does not quantify synchronization latency, operational overhead, permissions behavior, or query costs.</p>
Preview unifies analytics entry points, but provides limited behavioral, performance, or cost detail.
<p><strong>OneLake File Explorer</strong> is generally available for Windows. It exposes OneLake workspaces and files through Windows File Explorer, allowing users to drag local CSV, Excel, Parquet, image, or model files directly into OneLake.</p><p>Files become available to Fabric pipelines, notebooks, and downstream analytics after upload. The announcement does not describe file-size limits, synchronization behavior, offline handling, auditing, or controls for preventing unmanaged local data from entering governed workspaces.</p>
GA file movement improves usability, but the announcement omits governance, synchronization, and operational limits.
<p><strong>ALTER TABLE</strong> in Fabric Data Warehouse can now run inside an explicit <strong>BEGIN TRAN … COMMIT</strong> transaction for supported operations. Multiple column and constraint changes commit atomically, and failures roll back the entire transaction.</p><p>Supported cases include adding nullable columns, dropping columns, adding or dropping NOT ENFORCED primary, unique, and foreign-key constraints, and altering distributed temporary tables. Adding non-nullable columns, ALTER COLUMN, and ALTER on non-distributed temporary tables remain unsupported.</p>
Transactional ALTER TABLE enables atomic Warehouse schema migrations with explicit supported and unsupported operations.
<p>Fabric SQL Database now supports selecting any Azure SQL Database collation when creating a database, using the REST API, PowerShell, or Fabric CLI. Collation controls how textual values are sorted and compared, including language, case, and accent behavior.</p><p>The setting helps migrate applications with regional or multilingual requirements, but it applies to the Fabric SQL database itself. Replicated data in the SQL analytics endpoint does not inherit this collation, which must be handled independently.</p>
Preview collation selection improves SQL database compatibility, but replication and endpoint behavior remain separate.
<p><strong>semantic-link-labs 0.14.3</strong> adds APIs for listing, recovering, permanently deleting, and bulk exporting or importing Fabric items. Tenant administrators can enable item recovery, making workspace recovery and controlled migration more automatable.</p><p>The redesigned VertiPaq Analyzer now exports profiling data to a selected lakehouse, workspace, or schema and exposes Direct Lake type, source, workspace, and table details. It also calculates column cardinality for supported Import and Direct Lake sources, while missing-row analysis remains Import-only.</p>
Adds tenant item-recovery and bulk-transfer APIs, while materially improving Direct Lake diagnostics and storage analysis.
<p>Power BI does not broadly support using third-party semantic models as interchangeable semantic layers. XMLA compatibility can make a vendor model connect through the Analysis Services connector, but that workaround is not fully supported or reliable.</p><p>DirectQuery introduces additional problems: Power BI expects relational star-schema behavior, generates SQL joins, and may recalculate metrics incorrectly at subtotals or grand totals. Importing third-party metrics is also unsafe because many metrics must be evaluated after aggregation. A supported alternative is synchronizing or reproducing the calculations in a Power BI semantic model, typically with DAX.</p>
The article explains structural limits when Power BI sits over third-party metrics stores and recommends supported calculation synchronization.
<p>10 features - 2 must-know - 6 worth knowing.</p><ul><li>Stream real-time data from private networks into Eventstreams (Preview)</li><li>Secure data streaming: Custom CA and mTLS in Fabric Eventstream connectors (Preview)</li></ul>
2 must-know of 10 features
<p>Fabric Lakehouses and eligible Eventstreams can now associate a user, service principal, or managed identity with the item through a preview REST API. The associated identity replaces the creator or owner as the dependency for operations that previously relied on that user principal.</p><p>This addresses failures caused by staff departures, role changes, or expired credentials. The initial scope excludes Eventstreams using Azure Events or Fabric Events as a source. Assignment requires Write permission on the item and affected dependent items; administrators can inspect identities through admin item APIs.</p>
Preview identity association removes owner-account dependencies and adds REST-based remediation for Lakehouses and Eventstreams.
<p><strong>Item Recovery</strong> places deleted supported Fabric items into a soft-deleted state instead of removing them immediately. Items remain recoverable for a tenant-configured <strong>7–90 day retention window</strong>, with a workspace recycle bin and REST API access.</p><p>Contributors, Members, and Admins can restore items; Workspace Admins can permanently delete them. Restoration preserves permissions, data, lineage, endorsements, and sensitivity labels, but coverage depends on the supported-item list. Tenant administrators must enable the preview capability.</p>
Preview item-level recovery adds configurable retention, recycle-bin restoration, API support, and forensic value.
<p><strong>Dropped warehouse recovery</strong> retains deleted Fabric warehouses in the Workspace Recycle Bin instead of immediately removing them. A Workspace Administrator can restore the warehouse during a configurable <strong>7–90 day retention period</strong>, with seven days as the default.</p><p>Recovery includes the warehouse’s tables, schemas, data, snapshots, permissions, saved queries, views, and stored procedures. This removes the need for manual rebuilds or copy-based restoration, but it is a preview capability and does not eliminate the need to understand retention and supported recovery behavior.</p>
Preview warehouse recovery restores dropped warehouses, snapshots, security, and objects during a 7–90 day retention window.
<p><strong>Shortcut transformations</strong> are generally available for converting structured files referenced through OneLake shortcuts into Delta tables without custom code or pipelines. Fabric can process CSV, Parquet, and JSON, infer schemas, handle nested folders, and evolve schemas as new columns appear.</p><p>The resulting tables are written to the Lakehouse Tables area and remain synchronized with source changes through incremental processing. This can reduce orchestration overhead and avoid always-on pipeline infrastructure, although compute still occurs when source changes are detected and schema inference requires validation for production data.</p>
GA shortcut transformations incrementally convert external CSV, Parquet, and JSON files into synchronized Delta tables without pipelines.
<p><strong>Eventstream</strong> now supports custom Certificate Authorities and mutual TLS for Kafka-based sources, including Apache Kafka, Amazon MSK, Confluent Cloud, and Confluent Schema Registry. Certificates and private keys are stored in <strong>Azure Key Vault</strong>, then retrieved by connectors at runtime.</p><p>The capability supports enterprises whose brokers use internally issued certificates or require client authentication. Private-network deployments may require vNet injection and a private Key Vault endpoint. Certificates must be supplied in PEM format, and the feature remains preview, so connector coverage and operational behavior should be validated before production adoption.</p>
Preview Eventstream support for custom CA certificates and mTLS enables governed Kafka connectivity for private or regulated sources.
<p><strong>semantic-link-labs 0.14.2</strong> updates <strong>sempy_labs</strong> for Python 3.12, which is now the default Python version in Fabric notebooks. The release also includes a bug fix.</p><p>This is primarily a compatibility release rather than a new automation capability, but older notebooks or pinned environments may fail without the update.</p>
Python 3.12 compatibility matters because it is now the default runtime for Fabric notebooks.
<p><strong>Migration Assistant for SQL database in Fabric</strong> provides a wizard-driven path from SQL Server or Azure SQL Database into Fabric. It imports a <strong>DACPAC</strong>, assesses unsupported objects, reports compatibility issues, and provides remediation guidance before schema deployment.</p><p>After provisioning the target database, integrated Fabric Copy Jobs move the data, with Data Gateway support for on-premises sources and pre- or post-deployment scripts for constraint handling. The workflow preserves familiar T-SQL, SSMS, and VS Code practices, but compatibility assessment remains essential because unsupported SQL features still require redesign or removal.</p>
Preview Migration Assistant combines DACPAC schema assessment, guided remediation, Copy Jobs, and gateway connectivity for SQL Server migrations.
<p><strong>ANY_VALUE()</strong> is generally available in Fabric Data Warehouse. It returns an arbitrary value from each group or partition, allowing a query to group by the true business key while projecting descriptive columns that are constant for that key.</p><p>For example, grouping taxi revenue by <strong>GeographyID</strong> can return city, state, and country through <strong>ANY_VALUE()</strong> rather than adding every column to <strong>GROUP BY</strong> or misusing <strong>MIN()</strong> and <strong>MAX()</strong>. The function is only safe when the selected value is genuinely constant within the group; varying values require an aggregate or windowing rule that reflects business meaning.</p>
GA warehouse function simplifies functionally dependent descriptive columns in grouped T-SQL queries.
SQLBI
Apr 6, 2026
3/5 · Tier B
Report AuthoringSemantic Model
tutorial
<p>A visual calculation can highlight the matrix row whose final visible-column value is highest, without adding a formatting-only measure to the semantic model. The technique navigates the visual lattice with <strong>ROWS</strong>, <strong>COLUMNS</strong>, <strong>EXPAND</strong>, and <strong>COLLAPSEALL</strong>.</p><p>The reusable version passes the visual’s matrix content and relevant columns to a function, so the core logic no longer hard-codes Brand, Sales Amount, or Year. This keeps the calculation with the visual, although changing the visual structure still requires careful adjustment of the function call and lattice navigation.</p>
Shows reusable visual-calculation logic for highlighting matrix rows without model-only formatting measures.
SQLBI
Apr 6, 2026
1/5 · Tier C
Report Authoring
community-event
<p>The SQLBI+ session demonstrates visual-calculation lattice navigation with <strong>ROWS</strong>, <strong>COLUMNS</strong>, <strong>EXPAND</strong>, and <strong>COLLAPSEALL</strong>. It also covers reusable functions that encapsulate this navigation logic.</p>
Promotes a paid SQLBI+ session about visual-calculation navigation rather than documenting a platform change.
<p>Fabric Data Warehouse preview functions process unstructured text directly in T-SQL. Functions such as <strong>ai_extract()</strong>, <strong>ai_analyze_sentiment()</strong>, <strong>ai_classify()</strong>, <strong>ai_summarize()</strong>, <strong>ai_translate()</strong>, and <strong>ai_fix_grammar()</strong> support extraction, labeling, analysis, and transformation without a separate application-layer pipeline.</p><p><strong>ai_generate_response()</strong> enables custom prompt-based processing, including bespoke extraction or masking rules. Applying these functions in set-based queries can simplify large-scale enrichment, but the preview status means teams must validate output quality, latency, throughput, model behavior, and handling of sensitive text before production use.</p>
Preview T-SQL AI functions bring extraction, classification, sentiment, and transformation of text into warehouse workloads.
<p>Workspace item menus now reposition to remain visible, and number keys no longer switch tabs unless the intended shortcut is used. Semantic-model menu handling is still being addressed; open-in-new-tab, tab grouping, workspace closing, and dark mode remain planned or preview items.</p>
Minor workspace UX fixes and planned interface requests provide little architectural or operational value.
Fabric Updates
Apr 6, 2026
3/5 · Tier B
Spark RuntimeAPI Automation
announcement
<p>The preview <strong>Microsoft ADO.NET Driver for Fabric Data Engineering</strong> lets .NET applications connect to and manage Fabric Spark workloads through familiar <strong>DbConnection</strong>, <strong>DbCommand</strong>, and <strong>DbDataReader</strong> abstractions. It supports Microsoft Entra authentication, parameterized Spark SQL, complex Spark data types, asynchronous prefetch, connection pooling, session reuse, and automatic reconnection.</p><p>The driver is built on Fabric’s Livy APIs and can use lakehouse data, Fabric environments, and Spark configurations. Pooling and session reuse may reduce startup overhead, but teams should benchmark preview behavior against job startup latency, concurrent sessions, result sizes, and failure recovery.</p>
Preview ADO.NET connectivity exposes Fabric Spark SQL and session management to .NET applications through Livy APIs.
<p>A Direct Lake on OneLake model can expose multiple copies of a dimension without duplicating its underlying Delta table. The workaround uses TMDL to duplicate the model expression and table metadata, remove lineage tags, repoint the copy to the new expression, and force the name change to be recognized.</p><p>The copied table initially shows as unrefreshed and requires a <strong>Schema and Data</strong> refresh before relationships can be created. The technique remains a workaround for bugs affecting the normal role-playing-dimension pattern, so scripts should be tested against the current Desktop and Service behavior.</p>
A concrete TMDL workaround enables role-playing dimensions in Direct Lake without duplicating the underlying Delta data.
<p>Workspace-level <strong>Customer-Managed Keys (CMK)</strong> can now be enabled on Fabric capacities that already use Power BI <strong>Bring Your Own Key (BYOK)</strong>. BYOK protects Power BI semantic models at the capacity level, while CMK protects other Fabric workspace data at the workspace level.</p><p>Organizations can reuse the same Azure Key Vault or Managed HSM key or choose separate keys for isolation. This removes the previous need for dedicated capacities solely to combine these controls, reducing deployment complexity while preserving separate encryption boundaries. The capability remains in preview, so key access, rotation, recovery, and service-permission behavior require careful validation.</p>
Workspace CMK now interoperates with capacity BYOK, removing dedicated-capacity requirements for layered encryption controls.
<p>This monthly spotlight amplifies Fabric community posts across Power BI, data engineering, data science, governance, Real-Time Intelligence, and integration. The extract points to potentially useful material, including capacity benchmarking, Dataflow Gen2 partitioned compute, Open Mirroring, Data Agents, and Copy Jobs, but does not provide enough detail to evaluate those techniques directly.</p>
Monthly influencer roundup highlights varied community content without consolidating enough technical evidence for architecture decisions.
<p>A Fabric notebook can call the Power BI remote <strong>MCP server</strong> and retrieve a semantic model’s Prep for AI configuration through the <strong>GetSemanticModelSchema</strong> tool. The response includes grounding instructions and the projections behind verified answers.</p><p>The retrieved verified-answer definition contains the tables, columns, measures, and filters used by a visual, rather than the visual’s DAX query. The API is read-only in this scenario: Prep for AI configurations cannot be changed programmatically, so the technique supports inspection, verification, monitoring, and best-practice analysis rather than deployment.</p>
Shows authenticated MCP retrieval of Prep for AI metadata, verified answers, and a concrete monitoring use case with clear write limitations.
<p>SQLBits 2026 takes place from 22–25 April at ICC Wales and includes Microsoft-led sessions, workshops, AI and SQL training, and product leadership panels. The post is promotional and does not document Fabric behavior, limits, costs, or implementation patterns.</p>
Conference promotion offers attendance details but no durable Fabric implementation knowledge or technical evidence.
<p><strong>Dataflows Gen1</strong> is supported but has entered a legacy state: new innovation is moving exclusively to <strong>Dataflows Gen2</strong>, and Gen1 will receive only a narrow set of high-impact fixes. Retirement dates are still being finalized, with at least 12 months' notice promised for Gen1 workloads running on Premium capacity.</p><p>Gen2 retains the Power Query authoring model while adding broader destinations, Fabric runtime integration, parallel execution, Fast Copy, larger-scale processing, diagnostics, and tiered pricing. Destinations include Fabric Lakehouse and Warehouse, Azure storage and SQL, SharePoint, OneDrive, Snowflake, and other cloud databases.</p><p>Premium customers with Fabric access are directed toward Gen2, while Pro and PPU users may continue using Gen1 where it remains the best current fit. GCC Premium customers are expected to receive Gen2 support before future transition milestones, and more granular administrative controls are planned.</p>
Gen1 enters legacy status with no innovation, pending retirement, migration guidance, and material licensing and administration implications.
<p>The third edition of the Mastering DAX video course starts from foundational DAX concepts and adds modules on <strong>user-defined functions</strong> and calendar-based time intelligence. Optimization and VertiPaq internals are excluded because they are covered separately.</p>
Markets a DAX training course and mentions new modules without providing platform behavior or production guidance.
Real-Time IntelligenceData Quality
announcement
