Microsoft Fabric Operations Agent Step by Step Walkthrough
Fabric Capacity and Workspace You need a Microsoft Fabric workspace backed by a paid capacity. Trial capacities are not supported for Operations Agent. Your capacity must be provisioned in a supported region. As of April 2026, Operations Agent is available in all Microsoft Fabric regions except South Central US and East US. If your capacity is outside the US or EU, you will also need to enable cross geo processing and storage for AI through the tenant settings. Your workspace must contain an Eventhouse with at least one KQL database. The Eventhouse is the telemetry backbone, and the KQL database holds the tables the agent will monitor. In the screenshot below, you can see a workspace named OperationAgent-WS that contains an Eventhouse (ops_eventhouse), two KQL databases (ops_db and ops_eventhouse), and a Lakehouse (ops_lakehouse). This is the environment used throughout this guide. Figure 1. Workspace contents showing the Eventhouse, KQL databases, and Lakehouse ready for the Operations Agent. Enabling the Operations Agent in the Admin Portal A Fabric administrator must enable the Operations Agent preview toggle in the Admin Portal before anyone in the organization can create an agent. Navigate to the Admin Portal, locate the section for Real Time Intelligence, and find the setting labeled Enable Operations Agents (Preview). Toggle it to Enabled for the entire organization or for specific security groups depending on your governance requirements. In addition to this toggle, ensure that Microsoft Copilot and Azure OpenAI Service are also enabled at the tenant level. The Operations Agent relies on Azure OpenAI to generate its playbook and to reason about data when conditions are met. Figure 2. The Admin Portal showing the Enable Operations Agents (Preview) toggle set to Enabled for the entire organization. Note that messages sent to Operations Agents are processed through the Azure AI Bot Service. If your capacity is outside the EU Data Boundary, data may be processed outside your geographic or national cloud boundary. Be sure to communicate this to your compliance stakeholders before enabling the feature in production tenants. Microsoft Teams Account Every person who will receive recommendations from the agent must have a Microsoft Teams account. The Operations Agent delivers its findings and action suggestions through a dedicated Teams app called Fabric Operations Agent. You can install this app from the Teams app store by searching for its name. Once installed, the agent will be able to send messages containing data summaries and recommended actions directly to the designated recipients. Creating and Configuring the Operations Agent With your prerequisites in place, you are ready to create the Operations Agent. The following steps walk you through the entire configuration process using the Fabric portal. Step 1: Create a New Operations Agent Open the Microsoft Fabric portal and navigate to your workspace. On the Fabric home page, select the ellipsis icon and then select Create. In the Create pane, scroll to the Real Time Intelligence section and select Operations Agent. A dialog will appear asking you to name your agent and select the target workspace. Choose a descriptive name that reflects the agent’s purpose. In this guide, the agent is named OperationsAgent_1 and is deployed to the OperationAgent-WS workspace. Step 2: Define Business Goals and Agent Instructions Once the agent is created, you are taken to the Agent Setup page. This page is divided into two halves. On the left side, you configure the agent’s behavior. On the right side, you see the generated Agent Playbook after saving. The first field is Business Goals, where you describe the high level objective the agent should accomplish. Write this in clear, outcome oriented language. In this demo, the business goal is set to: “Monitor data pipeline execution and alert on failures.” The second field is Agent Instructions, where you provide more specific guidance on how the agent should reason about the data. Think of this as a brief you would hand to an analyst who will be watching your systems overnight. Be explicit about the table name, the column to watch, and the condition that constitutes an alert. In this demo, the instruction reads: “Monitor pipeline_runs table. Alert when status is failed.” Together, the business goals and instructions give the underlying large language model enough context to generate an accurate playbook. The more specific your instructions, the more reliable the agent’s behavior will be. Figure 3. The Agent Setup page showing business goals, agent instructions, and the generated playbook on the right. On the right side of the screen, you can see the Agent Playbook that was generated after saving. The playbook includes a Business Term Glossary, which shows the business objects the agent inferred from your goals and data. In this case, it identified an object called PipelineRun, mapped to the pipeline_runs table, with two properties: status (the pipeline run status from the status column) and runId (the unique identifier from the run_id column). It also displays the Rules section, which contains the conditions the agent will evaluate. Review the playbook carefully. Since it is generated by an AI model, there may be occasional misinterpretations. Verify that every property maps to the correct column and that the rules reflect your intended thresholds. If something is off, update your goals or instructions and save again to regenerate the playbook. Step 3: Add a Knowledge Source Scroll down on the Agent Setup page to find the Knowledge section. This is where you connect the agent to the data it will monitor. When you first open this section, it will display a message indicating that no knowledge source has been added yet. Figure 4. The Knowledge section before any data source has been added. Select the Add Data button to browse the available data sources. A panel will appear listing the KQL databases and Eventhouses accessible within your Fabric environment. In this demo, three sources are available: ops_db in the OperationAgent-WS workspace, wms_eventhouse in the WMS-CDC-Demo workspace, and ops_eventhouse in the OperationAgent-WS workspace. Select the database that contains the table you want the agent to monitor. For this guide, select ops_db, which holds the pipeline_runs table referenced in the agent instructions. Figure 5. Selecting the knowledge source from available KQL databases and Eventhouses. Once the knowledge source is connected, the agent will be able to query this database at regular intervals (approximately every five minutes) to evaluate its rules. Make sure the table in your selected database is actively receiving data, especially if you plan to demonstrate the agent detecting a condition in real time. Step 4: Define Actions Actions are the responses the agent can recommend when it detects a condition that matches its rules. Scroll further down the Agent Setup page to find the Actions section. Select the Add Action button to define a new custom action. A dialog titled New Custom Action will appear. It has three fields. The Action Name is a short, descriptive label for the action. The Action Description explains the purpose of the action and gives the agent context about when to use it. The Parameters section allows you to define input fields that pass dynamic values (such as names, dates, or identifiers) into the Power Automate flow that will be triggered. Figure 6. The New Custom Action dialog where you define the action name, description, and optional parameters. In this demo, the action is named Send Email Alert with a description indicating that it should send an email notification when a pipeline failure is detected. Once created, you can see the action listed in the Actions section with a green status indicator showing that the action is successfully connected. Figure 7. The Actions section showing the Send Email Alert action with a connected status. Step 5: Configure the Custom Action with Power Automate After creating the action, you need to configure it by linking it to an activator item and a Power Automate flow. Select the action you just created to open the Configure Custom Action pane. In this pane, you will see several fields. First, select the Workspace where the activator item resides. In this demo, the workspace is OperationAgent-WS. Next, select the Activator, which is the Fabric item that bridges the Operations Agent and Power Automate. Here, the activator is named Email_Alert_Activator. Once the connection is created, a Connection String is generated. This string is a unique identifier that links the Operations Agent to the Power Automate flow. Select the Copy button to copy this connection string to your clipboard. You will need it in the next step. Below the connection string, you will find the Open Flow Builder button. Select this to launch the Power Automate flow designer where you will build the email notification flow. Figure 8. The Configure Custom Action pane showing the workspace, activator, connection string, and the button to open the flow builder. Step 6: Build the Power Automate Flow When you select Open Flow Builder, a new browser tab opens with the Power Automate designer. The flow is pre-configured with a trigger called When an Activator Rule is Triggered. This trigger fires whenever the Operations Agent approves an action. In the Parameters tab of the trigger, you will see a field labeled Connection String. Paste the connection string you copied from the previous step into this field. This is the critical link that connects the Power Automate flow back to your Operations Agent. If this string is incorrect or missing, the flow will not fire when the agent recommends the action. Figure 9. The Power Automate flow builder with the activator trigger and the Connection String field. Below the trigger, you can add any actions your workflow requires. For an email alert scenario, add an Office 365 Outlook action to send an email to the operations team. You can use dynamic content from the trigger to include details such as the pipeline run ID, the failure status, and any parameters passed through from the Operations Agent. Save the flow and return to the Fabric portal. Your action is now fully configured and ready to be triggered by the agent. Step 7: Generate the Playbook and Start the Agent With all configuration complete (business goals, instructions, knowledge source, and actions), select Save on the Agent Setup page. Fabric will use the underlying large language model to generate the agent’s playbook. The playbook is a structured summary of everything the agent knows: its goals, the properties it monitors, and the rules it evaluates. You can also select Generate Playbook at the top of the page to regenerate the playbook if you have made changes. Review the playbook one final time to confirm that properties map correctly to your table columns and that rules reflect the exact conditions you want to monitor. When you are satisfied, select Start in the toolbar at the top of the page. The agent will begin actively monitoring your data. It queries the knowledge source approximately every five minutes, evaluating the playbook rules against the latest data. If a condition is met, the agent uses the LLM to summarize the data, generate a recommendation, and send a message to the designated recipients through Microsoft Teams. To pause the agent at any time, select Stop. This is useful during demos when you want to control the timing of the demonstration. How the Agent Operates at Runtime Once started, the Operations Agent follows a continuous loop. Every five minutes, it queries the connected KQL database to evaluate the rules defined in the playbook. If no conditions are met, it continues silently. If a condition is matched (for example, a pipeline run with a status of "failed" appears in the pipeline_runs table), the agent proceeds through the following sequence. First, the agent uses the large language model to analyze the data that triggered the condition. It summarizes the context, identifies the relevant business object (such as a specific pipeline run), and determines which action to recommend. Second, the agent sends a message to the designated recipients through Microsoft Teams. This message contains a summary of the detected insight, the data context that triggered it, and a suggested action. Recipients can approve the action by selecting Yes or reject it by selecting No. If parameters are included (such as a run ID or a severity level), they can be reviewed and adjusted before final approval. Third, if the recipient approves the action, the agent executes it on behalf of the creator using the creator’s credentials. In this demo, approving the action would trigger the Power Automate flow that sends an email alert. It is important to note that if a recommendation is not responded to within three days, the operation is automatically canceled. After cancellation, the action can no longer be approved or interacted with.Approaches to Integrating Azure Databricks with Microsoft Fabric: The Better Together Story!
Azure Databricks and Microsoft Fabric can be combined to create a unified and scalable analytics ecosystem. This document outlines eight distinct integration approaches, each accompanied by step-by-step implementation guidance and key design considerations. These methods are not prescriptive—your cloud architecture team can choose the integration strategy that best aligns with your organization’s governance model, workload requirements and platform preferences. Whether you prioritize centralized orchestration, direct data access, or seamless reporting, the flexibility of these options allows you to tailor the solution to your specific needs.Announcing Fabric Mirroring integration for Azure Database for MySQL - Public Preview at FabCon 2026
At FabCon 2026, we’re excited to announce the Public Preview of Microsoft Fabric Mirroring integration for Azure Database for MySQL. This integration makes it easier than ever to analyze MySQL operational data using Fabric’s unified analytics platform, without building or maintaining ETL pipelines. This milestone brings near real-time data replication from Azure Database for MySQL into Microsoft Fabric OneLake, unlocking powerful analytics, reporting, and AI scenarios while keeping transactional workloads isolated and performant. Why Fabric integration for Azure Database for MySQL? MySQL is widely used to power business‑critical applications, but operational databases aren’t optimized for analytics. Traditionally, teams rely on complex ETL pipelines, custom connectors, or batch exports — adding cost, latency, and operational overhead. With Fabric integration, Azure Database for MySQL now connects directly to Microsoft Fabric, enabling: Zero‑ETL analytics on MySQL operational data Near real-time synchronization into OneLake Analytics‑ready open formats for BI, data engineering, and AI A unified experience across Power BI, Lakehouse, Warehousing, and notebooks All without impacting your production workloads. What’s new in the Public Preview? The Public Preview introduces a first‑class integration between Azure Database for MySQL and Microsoft Fabric, designed for simplicity and scale. It introduces a solid set of core operational and enterprise‑readiness capabilities, enabling end-users to confidently get started and scale their analytics scenarios. Core replication operations Start, monitor, and stop replication directly from the integrated experience Support for both initial data load and continuous change data capture (CDC) to keep data in sync with minimal latency. Network and security Firewall and gateway support, enabling replication from secured MySQL environments. Support for Azure Database for MySQL servers configured with customer‑managed keys (BYOK), aligning with enterprise security and compliance requirements. Broader data coverage and troubleshooting Ability to mirror tables containing previously unsupported data types, expanding schema compatibility and reducing onboarding friction. Support for up to 1,000 tables per server, enabling larger and more complex databases to participate in Fabric analytics. Basic error messaging and visibility to help identify replication issues and monitor progress during setup and ongoing operations. What scenarios does it unlock? With Fabric integration in place, you can now analyze data in Azure Database for MySQL without impacting production, combine it with other data in Fabric for richer reporting, and use Fabric’s built‑in analytics and AI tools to get insights faster. Learn more about exploring replicated data in Fabric in Explore data in your mirrored database using Microsoft Fabric. How does it work (high level)? Fabric integration for Azure Database for MySQL follows a simple but powerful pattern: Enable and Configure - Enable replication in the Azure portal, then use the Fabric portal to provide connection details and select MySQL tables to mirror. Initial snapshot - Fabric takes a bulk snapshot of the selected tables, converts the data to Parquet, and writes it to a Fabric landing zone. Continuous change capture - Ongoing inserts, updates, and deletes are captured from MySQL binlogs and continuously written as incremental Parquet files. Analytics‑ready in Fabric - The Fabric Replicator processes snapshot and change files and applies them to Delta tables in OneLake, keeping data in sync and ready for analytics. This design ensures low overhead on the source, while providing fresh data for analytics and AI workloads. Below is a more detailed workflow illustrating how this works: Getting started with the Public Preview To try Fabric integration for Azure Database for MySQL during Public Preview, you’ll need: An Azure Database for MySQL instance An active Microsoft Fabric capacity (trial or paid) Access to a Fabric workspace Once enabled, you can select the MySQL databases and tables you want to replicate and begin analyzing data in Fabric. For step-by-step tutorial, please refer to - https://learn.microsoft.com/azure/mysql/integration/fabric-mirroring-mysql The demo video below showcases how the mirroring integration works and walks through the end-to-end experience of mirroring MySQL data into Microsoft Fabric. Stay Connected We’re thrilled to share this milestone at FabCon 2026 and can’t wait to see how you use Fabric integration for Azure Database for MySQL to simplify analytics and unlock new insights. We welcome your feedback and invite you to share your experiences or suggestions at AskAzureDBforMySQL@service.microsoft.com Try the Public Preview, share your feedback, and help shape what’s next. Thank you for choosing Azure Database for MySQL!Azure Databricks & Fabric Disaster Recovery: The Better Together Story
Author's: Amudha Palani amudhapalani, Oscar Alvarado oscaralvarado, Eric Kwashie ekwashie, Peter Lo PeterLo and Rafia Aqil Rafia_Aqil Disaster recovery (DR) is a critical component of any cloud-native data analytics platform, ensuring business continuity even during rare regional outages caused by natural disasters, infrastructure failures, or other disruptions. Identify Business Critical Workloads Before designing any disaster recovery strategy, organizations must first identify which workloads are truly business‑critical and require regional redundancy. Not all Databricks or Fabric processes need full DR protection; instead, customers should evaluate the operational impact of downtime, data freshness requirements, regulatory obligations, SLAs, and dependencies across upstream and downstream systems. By classifying workloads into tiers and aligning DR investments accordingly, customers ensure they protect what matters most without over‑engineering the platform. Azure Databricks Azure Databricks requires a customer‑driven approach to disaster recovery, where organizations are responsible for replicating workspaces, data, infrastructure components, and security configurations across regions. Full System Failover (Active-Passive) Strategy A comprehensive approach that replicates all dependent services to the secondary region. Implementation requirements include: Infrastructure Components: Replicate Azure services (ADLS, Key Vault, SQL databases) using Terraform Deploy network infrastructure (subnets) in the secondary region Establish data synchronization mechanisms Data Replication Strategy: Use Deep Clone for Delta tables rather than geo-redundant storage Implement periodic synchronization jobs using Delta's incremental replication Measure data transfer results using time travel syntax Workspace Asset Synchronization: Co-deploy cluster configurations, notebooks, jobs, and permissions using CI/CD Utilize Terraform and SCIM for identity and access management Keep job concurrencies at zero in the secondary region to prevent execution Fully Redundant (Active-Active) Strategy The most sophisticated approach where all transactions are processed in multiple regions simultaneously. While providing maximum resilience, this strategy: Requires complex data synchronization between regions Incurs highest operational costs due to duplicate processing Typically needed only for mission-critical workloads with zero-tolerance for downtime Can be implemented as partial active-active, processing most workload in primary with subset in secondary Enabling Disaster Recovery Create a secondary workspace in a paired region. Use CI/CD to keep Workspace Assets Synchronized continuously. Requirement Approach Tools Cluster Configurations Co-deploy to both regions as code Terraform Code (Notebooks, Libraries, SQL) Co-deploy with CI/CD pipelines Git, Azure DevOps, GitHub Actions Jobs Co-deploy with CI/CD, set concurrency to zero in secondary Databricks Asset Bundles, Terraform Permissions (Users, Groups, ACLs) Use IdP/SCIM and infrastructure as code Terraform, SCIM Secrets Co-deploy using secret management Terraform, Azure Key Vault Table Metadata Co-deploy with CI/CD workflows Git, Terraform Cloud Services (ADLS, Network) Co-deploy infrastructure Terraform Update your orchestrator (ADF, Fabric pipelines, etc.) to include a simple region toggle to reroute job execution. Replicate all dependent services (Key Vault, Storage accounts, SQL DB). Implement Delta “Deep Clone” synchronization jobs to keep datasets continuously aligned between regions. Introduce an application‑level “Sync Tool” that redirects: data ingestion compute execution Enable parallel processing in both regions for selected or all workloads. Use bi‑directional synchronization for Delta data to maintain consistency across regions. For performance and cost control, run most workloads in primary and only subset workloads in secondary to keep it warm. Implement Three-Pillar DR Design Primary Workspace: Your production Databricks environment running normal operations Secondary Workspace: A standby Databricks workspace in a different(paired) Azure region that remains ready to take over if the primary fails. This architecture ensures business continuity while optimizing costs by keeping the secondary workspace dormant until needed. The DR solution is built on three fundamental pillars that work together to provide comprehensive protection: 1. Infrastructure Provisioning (Terraform) The infrastructure layer creates and manages all Azure resources required for disaster recovery using Infrastructure as Code (Terraform). What It Creates: Secondary Resource Group: A dedicated resource group in your paired DR region (e.g., if primary is in East US, secondary might be in West US 2) Secondary Databricks Workspace: A standby Databricks workspace with the same SKU as your primary, ready to receive failover traffic DR Storage Account: An ADLS Gen2 storage account that serves as the backup destination for your critical data Monitoring Infrastructure: Azure Monitor Log Analytics workspace and alert action groups to track DR health Protection Locks: Management locks to prevent accidental deletion of critical DR resources Key Design Principle: The Terraform configuration references your existing primary workspace without modifying it. It only creates new resources in the secondary region, ensuring your production environment remains untouched during setup. 2. Data Synchronization (Delta Notebooks) The data synchronization layer ensures your critical data is continuously backed up to the secondary region. How It Works: The solution uses a Databricks notebook that runs in your primary workspace on a scheduled basis. This notebook: Connects to Backup Storage: Uses Unity Catalog with Azure Managed Identity for secure, credential-free authentication to the secondary storage account Identifies Critical Tables: Reads from a configuration list you define (sales data, customer data, inventory, financial transactions, etc.) Performs Deep Clone: Uses Delta Lake's native CLONE functionality to create exact copies of your tables in the backup storage Tracks Sync Status: Logs each synchronization operation, tracks row counts, and reports on data freshness Authentication Flow: The synchronization process leverages Unity Catalog's managed identity capabilities: An existing Access Connector for Unity Catalog is granted "Storage Blob Data Contributor" permissions on the backup storage. Storage credentials are created in Databricks that reference this Access Connector. The notebook uses these credentials transparently—no storage keys or secrets are required. What Gets Synced: You define which tables are critical to your business operations. The notebook creates backup copies including: Full table data and schema Table partitioning structure Delta transaction logs for point-in-time recovery 3. Failover Automation (Python Scripts) The failover automation layer orchestrates the switch from primary to secondary workspace when disaster strikes. Microsoft Fabric Microsoft Fabric provides built‑in disaster recovery capabilities designed to keep analytics and Power BI experiences available during regional outages. Fabric simplifies continuity for reporting workloads, while still requiring customer planning for deeper data and workload replication. Power BI Business Continuity Power BI, now integrated into Fabric, provides automatic disaster recovery as a default offering: No opt-in required: DR capabilities are automatically included. Azure storage geo-redundant replication: Ensures backup instances exist in other regions. Read-only access during disasters: Semantic models, reports, and dashboards remain accessible. Always supported: BCDR for Power BI remains active regardless of OneLake DR setting. Microsoft Fabric Fabric's cross-region DR uses a shared responsibility model between Microsoft and customers: Microsoft's Responsibilities: Ensure baseline infrastructure and platform services availability Maintain Azure regional pairings for geo-redundancy. Provide DR capabilities for Power BI as default. Customer Responsibilities: Enable disaster recovery settings for capacities Set up secondary capacity and workspaces in paired regions Replicate data and configurations Enabling Disaster Recovery Organizations can enable BCDR through the Admin portal under Capacity settings: Navigate to Admin portal → Capacity settings Select the appropriate Fabric Capacity Access Disaster Recovery configuration Enable the disaster recovery toggle Critical Timing Considerations: 30-day minimum activation period: Once enabled, the setting remains active for at least 30 days and cannot be reverted. 72-hour activation window: Initial enablement can take up to 72 hours to become fully effective. Azure Databricks & Microsoft Fabric DR Considerations Building a resilient analytics platform requires understanding how disaster recovery responsibilities differ between Azure Databricks and Microsoft Fabric. While both platforms operate within Azure’s regional architecture, their DR models, failover behaviors, and customer responsibilities are fundamentally different. Recovery Procedures Procedure Databricks Fabric Failover Stop workloads, update routing, resume in secondary region. Microsoft initiates failover; customers restore services in DR capacity. Restore to Primary Stop secondary workloads, replicate data/code back, test, resume production. Recreate workspaces and items in new capacity; restore Lakehouse and Warehouse data. Asset Syncing Use CI/CD and Terraform to sync clusters, jobs, notebooks, permissions. Use Git integration and pipelines to sync notebooks and pipelines; manually restore Lakehouses. Business Considerations Consideration Databricks Fabric Control Customers manage DR strategy, failover timing, and asset replication. Microsoft manages failover; customers restore services post-failover. Regional Dependencies Must ensure secondary region has sufficient capacity and services. DR only available in Azure regions with Fabric support and paired regions. Power BI Continuity Not applicable. Power BI offers built-in BCDR with read-only access to semantic models and reports. Activation Timeline Immediate upon configuration. DR setting takes up to 72 hours to activate; 30-day wait before changes allowed.Smart Pipelines Orchestration: Designing Predictable Data Platforms on Shared Spark
Introduction In mature data platforms, scaling compute is rarely the primary challenge. Shared, elastic Spark pools already provide sufficient processing capacity for most workloads. The harder problem is achieving predictable execution when multiple pipelines compete for the same resources. In Azure Synapse, Spark pools are commonly shared across pipelines to optimize cost and utilization. While this model is efficient, it introduces a key limitation: execution order is determined by scheduling behavior, not business priority. This post describes an orchestration pattern that makes priority explicit, allowing critical workloads to run predictably on shared Spark compute without modifying Spark code, configuration, or cluster capacity. Goal This work does not aim to optimize Spark performance. Its goal is to ensure that, when pipelines share a Spark pool: latency-sensitive workloads run first heavy backfills do not delay critical pipelines execution order is deterministic under contention All of this needed to be achieved without changes to Spark configuration, notebook logic, or cluster size. Why This Problem Occurs In a naïve orchestration model, pipelines are triggered in parallel. From Spark’s perspective: all jobs are equivalent all jobs attempt to acquire executors at the same time scheduling decisions are based on availability and timing As a result, priority is implicit and often incorrect. A heavy workload may acquire executors before a lightweight but critical one simply because it requests more resources earlier. This behavior is expected from Spark. The issue lies in orchestration, not in compute. Core Concept: Priority as Execution Ordering In shared Spark platforms, priority is enforced through execution ordering, not compute tuning. The orchestration layer controls when workloads are admitted to shared compute. Once execution begins, Spark processes each workload normally. This preserves Spark’s execution model while providing deterministic workload ordering. Step 1: Workload Classification In the demo presented in this blog, workloads are classified during configuration based on business impact: Category Description Priority example Light (critical) SLA sensitive dashboard and downstream consumers High priority , low resource weight(data volume) Medium (High) Core reporting workloads Medium priority Heavy(Best Effort) Backfills and historical computes Low priority, high resource weight(data volume) This classification is external to Spark and external to code. It represents business intent, not implementation. As a future phase, classification can be automated,for example, an agent may adjust priority based on observed failure rates or execution stability. Workload classification is expressed as orchestration metadata, for example: [ {"name":"ExecDashboard","pipeline":"PL_Light_ExecDashboard","weight":1,"tier":"Critical"}, {"name":"FinanceReporting","pipeline":"PL_Medium_FinanceReporting","weight":3,"tier":"High"}, {"name":"Backfill","pipeline":"PL_Heavy_Backfill","weight":8,"tier":"BestEffort"} ] What Runs in Each Workload Category All pipelines execute on the same shared Spark pool, but the work they perform differs in scope, data volume, and sensitivity to contention. Light workloads power SLA-sensitive dashboards and downstream consumers. Their notebooks perform targeted reads with strong filtering, limited joins, and small aggregations. Execution time is short, and overall pipeline duration is dominated by executor availability rather than computation. Medium workloads represent core reporting and analytics logic. These notebooks process larger datasets, perform joins across multiple sources, and apply aggregations that are more expensive than Light workloads but still time-bounded and business-critical. Heavy workloads are best-effort pipelines such as backfills and historical recomputation. Their notebooks scan large data volumes, apply expensive transformations, and are optimized for throughput rather than responsiveness. These workloads tolerate delay but place significant pressure on shared compute when admitted concurrently. All workloads use the same Spark pool, executor configuration, and runtime. The distinction reflects business intent and execution characteristics, not Spark tuning. Example notebooks for each category are available in the accompanying GitHub repository. Step 2: Naïve Orchestration (Baseline) The following pipeline run illustrates the baseline behavior when all workloads are triggered in parallel against a shared Spark pool. All Light, Medium, and Heavy pipelines are admitted concurrently. Executor acquisition and execution order depend on timing rather than business priority, resulting in non-deterministic behavior under contention. Although Light workloads require minimal compute, they are delayed by executor contention caused by Medium and Heavy pipelines entering the Spark pool at the same time. Step 3: Smart Orchestration (Priority-Aware) Orchestration Model The same child pipelines and notebooks are reused. The parent pipeline enforces admission order: Light (Critical) Medium (High) Heavy (Best Effort) Dependencies control admission to the Spark pool. Parallelism is preserved within a priority class. Effect on Shared Spark Light workloads enter the Spark pool without contention Medium workloads run after Light completes Heavy workloads are intentionally delayed Executor acquisition aligns with business priority Light pipelines execute first and complete before medium pipelines are admitted. Heavy workloads run last by design. No Spark configuration changes are introduced. The Spark pool, notebooks, and executor configuration are identical to the naïve run. Only the orchestration graph differs. Step 4: Impact on Light Workloads Light workloads are particularly sensitive to orchestration because their runtime is dominated by queueing time, not computation. Comparing the naïve and priority-aware runs shows that Spark execution time is unchanged, but pipeline duration improves due to earlier admission to the Spark pool and immediate executor access Naïve Execution Spark execution time: short and unchanged Pipeline duration: minutes under contention Delay caused by executor unavailability Smart Execution Spark execution time: unchanged Pipeline duration closely matches compute time Immediate access to executors The improvement comes from removing admission contention, not from increasing resources. Results and Performance Compared to naïve orchestration, priority-aware orchestration ensures that Light workloads complete in minutes rather than tens of minutes under contention, while Spark execution time itself remains unchanged. Heavy workloads no longer delay latency-sensitive pipelines, and execution order is deterministic across runs. These improvements are achieved solely by controlling admission to the shared Spark pool, without modifying Spark configuration, notebook logic, or cluster capacity. Next Steps: 1. Optimizing Heavy Workloads Once heavy workloads are isolated by priority, they can be optimized independently: retries with backoff tolerance for transient failures increased executor counts or larger pools Without admission control, these optimizations increase contention, with smart orchestration, they do not impact critical pipelines. 2. Moving Beyond Static Classification In this implementation, workload classification is static and configuration-driven, which is sufficient for stabilization. A next phase is adaptive classification: collect execution metrics and failure rates detect unstable pipelines reclassify pipelines that exceed thresholds (e.g., >20% failures in a rolling window) This prevents unstable workloads from impacting critical execution paths and makes the pipeline reliable with minimal maintenance. 3. Assisted Classification with Copilot agent At scale, priority decisions benefit from automation. A Copilot-style agent can use historical execution data to recommend classification changes, grounding decisions in observed behavior while keeping engineers in control. Example: Changing workload classification from Light to Medium Consider a pipeline initially classified as Light because it powers an SLA-sensitive dashboard and typically executes quickly with minimal resource usage. Over time, execution telemetry shows a change in behavior: The pipeline fails in 4 of the last 10 runs due to transient Spark errors Average duration has increased by 3×, even when admitted early Retry attempts amplify contention for other Light workloads Based on these signals, an automated agent flags the workload as unstable and recommends reclassifying it from Light to Medium. After reclassification: The pipeline is admitted after Light workloads but before Heavy workloads It no longer blocks latency-critical paths when retries occur Execution remains predictable, while instability is isolated from critical workloads The notebook logic and Spark configuration remain unchanged, only the workload’s admission priority is updated via orchestration metadata. This approach allows the platform to adapt to changing workload characteristics while preserving deterministic execution for critical pipelines. Conclusion Parallel execution is a default, not a strategy. In shared environments, orchestration must explicitly encode business intent rather than relying on scheduler behavior. Enforcing priority at the orchestration layer restores predictability without sacrificing efficiency and provides a foundation for adaptive, policy-driven execution as platforms evolve. Links Orchestrating data movement and transformation in Azure Data Factory - Training | Microsoft Learn How to Optimize Spark Jobs for Maximum Performance: A Complete Guide GitHub repo for notebook reference: sallydabbahmsft/Smart-pipelines-orchestration Feedback: Sally Dabbah | LinkedInTableau to Power BI Migration: Semantic Layer-First Approach for Cloud Architects
Author's: Lavanya Sreedhar LavanyaSreedhar, Peter Lo PeterLo, Aryan Anmol aryananmol, Shreya Harvu shreyaharvu and Rafia Aqil Rafia_Aqil In this guide, we provide practical guidance for migrating from Tableau to Power BI, with a focus on technical best practices and architecture. Unifying business intelligence on the Microsoft Fabric platform, enterprises gain closer integration with Microsoft 365 (Teams, Copilot, Excel). For cloud solution architects and BI developers, a successful migration is not just about rebuilding dashboards in a new tool. It requires thoughtful architectural planning and a shift to a more model-centric approach to BI. Why Semantic Layer-First Architecture Matters The Traditional Migration Challenge Most Tableau to Power BI migrations follow a dashboard-centric approach: teams attempt to replicate existing Tableau workbooks, calculated fields, and LOD (Level of Detail) expressions directly into Power BI reports. While this may seem efficient initially, it creates significant downstream challenges: Duplicated logic: Each report embeds its own calculations and business rules, leading to conflicting KPIs across the organization Maintenance overhead: Changes to business logic require updating dozens or hundreds of individual reports Governance gaps: Without centralized definitions, semantic drift occurs—different teams calculate "Revenue" or "Active Customer" differently Scalability issues: As data volumes grow, report-level transformations become performance bottlenecks The Semantic Layer-First Alternative Microsoft's recommended approach centers on semantic models (formerly called datasets)—centralized, governed data models that separate business logic from visualization. In this architecture: The payoff is substantial: when data evolves or business rules change, you update the semantic model once, and all dependent reports automatically reflect the changes—no manual redesign required. Understanding Migration Complexity: Simple to Very Complex Dashboards Not all Tableau dashboards are created equal. The migration strategy should align with dashboard complexity, and the semantic layer approach becomes increasingly valuable as complexity grows. Follow a Step-by-Step Migration Strategy Migrating from Tableau to Power BI is not a one-click effort – it requires a mix of automated and manual refactoring, plus a sound change management plan. Below are key strategies and best practices for a successful migration: Audit your Tableau estate: Start by taking inventory of all existing Tableau workbooks, data sources, and dashboards. Determine what needs to be migrated (focus on high-value, widely used reports first) and identify any redundant or obsolete content that can be retired rather than converted. Conduct a proof-of-concept (PoC): Before migrating everything, pick a representative complex dashboard (or a subset of your data) and perform a pilot migration. This will help you validate that Power BI can connect to your data (e.g. setting up the Power BI gateways for on-premises sources), test performance (Import vs DirectQuery modes), and experiment with replicating key visuals or calculations. Use the PoC to uncover any surprises early – for example, test that any Level of Detail expressions or table calculations in Tableau can be re-created in DAX. The lessons learned here should inform your overall project plan. Use a phased migration approach: Plan to run Tableau and Power BI in parallel for some period, rather than switching everything at once. Migrate in waves – for example, by business unit or subject area – and incorporate user feedback as you go. This phased approach reduces risk and allows your team to improve the process with each iteration. It also gives end users time to adjust gradually. Migrate high-impact dashboards first: Prioritize the migration of key reports and dashboards that are critical to the business or have the most usage. Delivering these early wins will not only surface any technical challenges to solve but will also help demonstrate the value of Power BI’s capabilities to stakeholders. Early success builds buy-in and momentum for the rest of the migration. Reimagine (don’t just replicate) the experience: It’s rarely possible – or desirable – to exactly re-create every Tableau visualization pixel-for-pixel in Power BI. Embrace the opportunity to focus on business questions and improve user experience with Power BI’s features. For example, rather than replicating a complex Tableau workaround, you might implement a cleaner solution in Power BI using native features (like bookmarks, drilldowns, or simpler navigation between pages). Engage business users and subject matter experts during this redesign to ensure the new reports meet their needs. Enable dataset reusability: One major benefit of the Power BI approach is the ability to create shared datasets and dataflows. As you migrate, look for opportunities to create central semantic models (datasets) that can serve multiple reports. For instance, if several Tableau workbooks are all using similar data about sales, you can create one central Sales dataset in Power BI. Report creators across the organization can then build different Power BI reports on that single dataset without duplicating data or logic. This reduces maintenance and promotes a “build once, reuse often” strategy. Provide training and support: Expect a learning curve for teams moving to Power BI – especially those who are very fluent in Tableau. Plan for user upskilling and training programs. Establish a support community or office hours where new users can ask questions and get help. If possible, identify Power BI champions or recruit a Power BI Center of Excellence (COE) team who can guide others. During the transition, ensure there are subject matter experts (SMEs) available to address questions and validate that the new reports are correct. Manage change and expectations: It’s important to communicate why the organization is moving to Power BI (e.g. benefits like deeper integration, lower TCO, better governance) to get buy-in from end users. Some users may be resistant to change, especially if they’ve invested a lot of time in mastering Tableau. Prepare to handle varying responses – emphasize the personal benefits (like improved performance, new capabilities, or career growth with popular skills) to encourage adoption. Also, involve influential business users early and gather their feedback, so they feel ownership in the new solution. Establish governance from Day 1: Don’t wait until after migration to think about governance. Use this chance to set up Power BI governance aligned to best practices. Decide on important aspects such as workspace naming conventions, who can create or publish content, how you’ll monitor usage and costs, and how to manage data access and security (for example, designing a strategy for RLS/OLS/CLS, and deciding when to use per-user datasets vs. organizational semantic models). Good governance will ensure your shiny new Power BI environment doesn’t sprawl into chaos over time. Allow time for adjustment and iteration: Finally, be patient and iterative. Depending on the scale of your organization and the number of Tableau assets, a full migration can take months or even a year or more. Plan realistic transition periods where both systems might coexist. Continuously refine your approach with each wave of migration. Power BI’s frequent update cadence (monthly releases) means new features may emerge even during your project – stay updated, as new capabilities could simplify your migration (for example, the introduction of field parameters or Copilot might let you modernize certain Tableau features more easily). Reimagine (don’t just replicate) the experience (Step 5): Phase 1: Assessment and Planning 1. Audit Your Tableau Estate Inventory all workbooks, data sources, and calculated fields Identify high-traffic dashboards (prioritize for early migration) Categorize by complexity (Simple/Medium/Complex/Very Complex) 2. Design Your Semantic Architecture Map Tableau data sources to Power BI data sources (DirectQuery, Import, or Direct Lake) Plan star schema for fact/dimension tables Identify shared calculations that should live in semantic models vs. report-specific logic 3. Choose Storage Modes Source Type Recommended Mode Rationale Databricks Delta Lake Direct Lake Real-time analytics, no refresh lag Azure SQL Database DirectQuery or Import Based on data volume and refresh SLAs On-Premises SQL Server Import (via Gateway) Network latency considerations Excel/CSV files Import Small reference data Phase 2: Build the Semantic Layer 1. Create Star Schema Data Models Tableau often relies on flat, denormalized datasets. Power BI performs best with star schemas: Fact tables: Transactional data (sales, orders, events) with foreign keys to dimensions Dimension tables: Descriptive attributes (customers, products, dates) with primary keys Relationships: One-to-many from dimension to fact, leveraging bidirectional filtering sparingly 2. Migrate Calculations to DAX Measures Convert Tableau calculated fields to DAX measures in the semantic model: --Example of DAX: -- Define as measure: Total Revenue = SUMX( 'Sales', 'Sales'[Quantity] * 'Sales'[Unit Price] ) 2.1 Use Copilot to Accelerate DAX Development Leverage Copilot in Power BI Desktop to generate and validate DAX: Describe the calculation in natural language Copilot suggests DAX syntax Review, test, and refine 2.2 Document your Semantic Model Invest in creating an AI-ready foundation for your semantic model. AI systems need to understand unique business contexts in order to prioritize correct information to provide consistent and reliable responses to your end users. Name Tables and Columns Clearly: Avoid ambiguity in your semantic model. Use human-readable, business-friendly names. Avoid abbreviations, acronyms, or technical terms. This improves Copilot’s ability to interpret user intent. Create Meaningful Measures: Define reusable DAX measures for key business metrics (e.g., Revenue, Profit Margin). AI features rely on these to generate insights and summaries. Document Semantic Model objects: Add descriptions and synonyms to your Tables, Columns and measures. This enhances natural language querying and improves Copilot’s contextual understanding. Build an AI Data Schema: prepare your semantic model for AI by utilizing tooling features such as Prep data for AI. Phase 3: Understanding Migration Complexity: Simple to Very Complex Dashboards Not all Tableau dashboards are created equal. The migration strategy should align with dashboard complexity, and the semantic layer approach becomes increasingly valuable as complexity grows. 1. Dashboard Conversion Best Practices Think in "pages" not "sheets": Power BI reports combine multiple visuals per page; group related visuals logically Use slicers for interactivity: Replace Tableau filters with Power BI slicers and filter pane Leverage bookmarks for navigation: Create dynamic report experiences with show/hide containers Simple Complexity Level Category Tableau Feature Power BI Equivalent Microsoft Fabric Enhancements Best Practice Notes Data Model Single custom SQL Power Query for data shaping and ETL. OneLake Shortcuts for unified data access. Use star schema for optimized performance; push logic into the semantic layer rather than visuals. Calculations Basic IF/ELSE, SUM Data Analysis Expressions (DAX) for measures and calculated columns. Copilot for Power BI to assist with DAX creation. Fabric IQ for natural language queries. Centralize calculations in semantic models for consistency and governance. Medium Complexity Level Category Tableau Feature Power BI Equivalent Fabric Enhancements Best Practice Notes Data Model Multiple custom SQL (up to 3) Connect live to databases (Azure Databricks): DirectQuery in Power BI Connect with cloud data sources: Power BI data sources OneLake Shortcuts for unified access without databricks compute cost. Semantic Models can combine multiple sources. Optimize with star schema; Prefer OneLake Shortcuts for performance; avoid heavy transformations in visuals. Calculations Nested IFs, CASE Data Analysis Expressions (DAX) for measures and calculated columns. Copilot for Power BI to assist with DAX creation. Fabric Data Agent for conversational BI. Fabric IQ for natural language queries: Fabric IQ Centralize logic in semantic models; use Copilot for automation and validation; keep calculations reusable. Reporting Tooltip format in Bar and Map visuals Select All/Clear option for Single Select dropdown Standard tooltips offer help tooltips, text, and background formatting. Dynamic tooltip will be able to create the Tooltip page and reuse it in multiple visuals The customization is so much better than the OOB tooltips Create report tooltip pages in Power BI - Power BI | Microsoft Learn Use Clear All Slicers Button. Disable Single Select, Add Clear All Slicers button, Customize the Button and Use the Button Complex Complexity Level Category Tableau Feature Power BI Equivalent Fabric Enhancements Best Practice Notes Data Model Multiple sources Create relationship using more than one column Composite Models in Power BI (DirectQuery + Import) for combining multiple sources, also connect to various cloud services. Dataflows for pre-processing. Power BI allows a relationship between 2 tables based on only one active column. OneLake Shortcuts for unified access without Azure Databricks compute cost; Microsoft Fabric Dataflows Gen2 offers multiple ways to ingest, transform, and load data efficiently. Consolidate sources into semantic models; use Direct Lake for performance; Plan and design data model to comply with star schema supported by Power BI Relationship DAX USERELATIONSHIP DAX for activating relationships in Power BI for a specific calculation Calculations LOD, window functions Data Analysis Expressions (DAX) for measures and calculated columns. Copilot to assist with complex DAX. Fabric IQ Ontology for semantic alignment. Change how visuals interact in a Power BI report. Centralize calculations in semantic layer; use variables in DAX for readability and performance. Fabric Data Agent for a conversational BI. Very Complex Complexity Level Category Tableau Feature Power BI Equivalent Fabric Enhancements Best Practice Notes Data Model Multi-source, Excel, SQL Composite Models in Power BI (DirectQuery + Import) for combining multiple sources, also connect to various cloud services. Dataflows for pre-processing. OneLake Shortcuts for unified access; Connector overview build-in support. Mirroring for real-time sync. Combine multiple sources into well-structured semantic models for consistency and optimized performance. Calculations Predictive logic Data Analysis Expressions (DAX) for measures and calculated columns. Fabric AutoML, ML models, AI Insights, Python/R, Notebook‑based ML (Spark/Scikit‑Learn), Fabric AI Functions, Fabric IQ Ontology Fabric Data Agent for a conversational BI. Centralize logic in semantic models; leverage Copilot for automation and parameter-driven workflows. Prepare for Copilot. 2. Tableau Feature Equivalents Tableau Feature Power BI Equivalent Microsoft Learn Link Calculated Fields DAX Measures DAX Documentation Parameters Field Parameters / Bookmarks Use report readers to change visuals Actions Drillthrough / Bookmarks Drillthrough Tableau Prep Power Query / Dataflows Differences between Dataflow Gen1 and Dataflow Gen2 Tableau Server Power BI Service What is Power BI? Overview of Components and Benefits Phase 4: Governance and Deployment Workspace Planning (Dev / Test / Prod Separation) A proper workspace strategy is essential for governed deployments in Fabric and Power BI. Fabric supports separate Development, Test, and Production stages using Deployment Pipelines, enabling controlled promotions of semantic models, reports, dataflows, notebooks, lakehouses, and other items. You can assign each workspace to a pipeline stage (Dev → Test → Prod) to ensure safe lifecycle management. Sensitivity Labeling (Microsoft Purview Information Protection) Sensitivity labels allow governed classification and protection of data across Fabric items. Sensitivity labels can be applied directly to Fabric items (semantic models, reports, dataflows, etc.) through the item's header flyout or the item settings. Labels from Microsoft Purview Information Protection enforce data access rules and help organizations meet compliance requirements. Endorsement & Certification (Promoted, Certified, Master Data) Endorsement improves discoverability and trust in shared organizational content. Promoted: Item creators mark content as recommended for broader use. Certified: Administrators or authorized reviewers validate content meets organizational quality standards. Master Data: Indicates authoritative single‑source‑of‑truth items such as semantic models or lakehouses. All Fabric items except dashboards can be promoted or certified; data‑containing items can be designated as Master Data. Monitoring & Capacity Planning Determine the appropriate size for fabric capacity when migrating from Tableau to PowerBI. The Fabric SKU Estimator can generate a SKU recommendation (estimate) for your capacity requirements. Ensuring performance and cost efficiency requires ongoing monitoring of your Fabric capacity. Microsoft recommends evaluating workloads using Fabric Capacity Metrics and planning SKU sizes based on real usage. Fabric uses bursting and smoothing to handle spikes while enforcing capacity limits. Monitoring helps identify high compute usage, background refreshes, and interactive workloads to optimize performance. Fabric Data Source Connections (OneLake+ Manage Connections) Microsoft Fabric is designed as an end‑to‑end analytics platform that integrates data from many different source systems into a unified environment powered by OneLake, Data Factory, Real‑Time Analytics, Dataflows , Lakehouses, Warehouses, and Mirrored Databases. The Strategic Advantage: Semantic Layer + Fabric IQ The semantic layer-first approach sets the foundation for the next evolution in enterprise analytics. Fabric IQ (announced at Ignite 2025) is Microsoft's semantic intelligence platform that auto-elevates semantic models into ontologies—structured knowledge graphs that power AI agents, Copilot experiences, and cross-domain data reasoning. What this means for your migration: Semantic models you build today become the foundation for AI-driven analytics tomorrow Data Agents can reason across multiple semantic models, answering questions that span domains Business users transition from "report consumers" to "data explorers" via natural language interfaces Conclusion: Build for the Future, Not Just for Today Migrating from Tableau to Power BI is more than a technology swap—it's an opportunity to re-architect your analytics strategy for the cloud-native, AI-powered era. The semantic layer-first approach requires upfront investment in data modeling, DAX expertise, and Fabric platform adoption. But the payoff is transformative: Consistency: Single source of truth for all business metrics Scalability: Semantic models that serve hundreds of reports and thousands of users Agility: Changes to business logic propagate instantly across the enterprise Future-readiness: Foundation for Fabric IQ, Data Agents, and AI-driven insights Start your migration with the end in mind: not just convert dashboards, but a modern, governed, AI-ready analytics platform that scales with your business. Addressing Key Migration Concerns (1) Why a semantic‑layered model approach is better than recreating Tableau dashboards A semantic‑layered modeling approach is the optimal strategy for migration and is significantly more effective than attempting to recreate Tableau dashboards exactly as they exist. By contrast, Power BI and Fabric encourage a semantic model–first architecture, where all business rules, relationships, calculations, and transformations are centralized in a governed model that serves many dashboards. The approach not only provides consistency and reuse across the enterprise but also ensures that report authors build on a single certified version of the truth. (2) How semantic-layered model approach reduces the constant redesign caused by changing data needs. A semantic‑layered modeling approach directly addresses concern about constant changes and frequent redesigns of dashboards when data evolves. With a semantic layer, changes are absorbed in the model layer—so the logic is updated once and flows automatically into all dependent reports. Combined with Fabric features like OneLake shortcuts, Direct Lake mode, and centralized governance, the semantic layer drastically reduces breakage, minimizes rework, and ensures scalability as data continues to grow and shift. Additional Resources Direct Lake in Microsoft Fabric Create Fabric Data Agents OneLake Shortcuts Write DAX queries with Copilot - DAX Prepare Your Data for AI - Power BI | Microsoft LearnBuilding a Reliable Real Time Data Pipeline with Microsoft Fabric
The Two Pillars That Determine Success Data Quality Cannot Be an Afterthought The most common mistake we see is treating data quality as something to address after the pipeline is running. This approach creates technical debt that compounds over time and erodes trust in your data. Your CDC pipeline will ingest millions of events daily. Without proper validation at each layer, small issues become major problems. A single source system changing a column from integer to string can silently corrupt downstream analytics for days before anyone notices. What you need to implement from day one: Validation at the Bronze layer should focus on structural integrity. Every record landing in your raw layer needs verification that required fields exist, timestamps are valid, and CDC operation types are recognized. The Silver layer is where business validation happens. Here you check referential integrity, apply domain specific rules, and flag anomalies. A customer ID that does not exist in your customer master table needs to be caught here. Schema drift detection deserves special attention. Source systems change without warning. Your pipeline needs to detect these changes before they break downstream processes. The quality score approach works well in practice. Rather than binary pass or fail checks, calculate a quality score for each batch. A score above 95 percent proceeds normally. Between 90 and 95 percent sends a warning. Below 90 percent halts processing. Replication Lag Requires Active Management The second pillar is understanding and managing replication lag. In a real time pipeline, the value of data degrades rapidly with age. A five minute delay might be acceptable for daily reporting but catastrophic for fraud detection or inventory management. Lag accumulates at multiple points in your pipeline. There is capture lag between when a change occurs in the source database and when the CDC mechanism detects it. Processing lag occurs within Eventstream as events are transformed and routed. Ingestion lag happens between Eventstream and your destination tables. Each component adds latency, and under load, these delays compound. Building effective lag management: Monitor each stage independently. Knowing your total lag is useful, but knowing where lag accumulates is actionable. Establish baselines before setting alerts. Collect at least two weeks of baseline metrics before configuring alert thresholds. Implement automatic recovery procedures. When lag exceeds acceptable thresholds, your system should respond without waiting for human intervention. Operational Foundations You Cannot Skip Capacity Planning Prevents Expensive Surprises Microsoft Fabric uses a capacity unit model where all workloads draw from a shared pool. Underprovisioning leads to throttling and failed jobs. Overprovisioning wastes budget. Start with realistic estimates based on your data volumes. The F4 SKU handles most development and small production workloads comfortably. Medium deployments with 10 to 25 sources typically need F8. Large enterprise deployments should start at F16 and scale based on observed utilization. Watch for sustained utilization above 70 percent as a signal to consider scaling up. Network Security Shapes Your Architecture For production deployments handling sensitive data, network isolation is not optional. Private endpoints keep traffic on the Microsoft backbone network, eliminating exposure to the public internet. Plan your network architecture before building pipelines. Retrofitting private connectivity into an existing deployment is significantly more complex than designing it from the start. Logging Enables Troubleshooting When something goes wrong at 2 AM, your ability to diagnose the problem depends entirely on what information you captured beforehand. Centralized logging using Eventhouse gives you a queryable record of everything that happened across your pipeline. Log more than you think you need initially. Storage is inexpensive compared to the cost of troubleshooting without adequate information. Key Decisions You Need to Make Before proceeding with implementation, your team should align on several important decisions. Data retention requirements affect storage costs and query performance. How long do you need to keep Bronze layer data versus aggregated Gold layer data? Recovery time objectives determine how you architect for resilience. If the pipeline can be down for four hours without business impact, your approach differs from a scenario where even 15 minutes causes significant problems. Who owns data quality shapes how you design validation and alerting. If source system teams are responsible, your pipeline detects and reports. If your team owns quality, you implement correction and enrichment. Moving Forward The patterns and practices in this guide reflect lessons learned from real implementations. Every organization has unique requirements, but the fundamentals of data quality, lag management, capacity planning, and operational readiness apply universally. Start with the foundations. A pipeline that handles one source reliably is more valuable than one that theoretically handles fifty but fails unpredictably. Build observability from day one. Automate responses to common problems. Document what you learn. Your data is a strategic asset. The pipeline that delivers it reliably deserves the same careful engineering you would apply to any critical business system.A Technical Implementation Guide for Multi-Store Retail Environments
Understanding the Problem Space When organizations first approach multi-source data ingestion, they typically start with explicit configuration. Each database connection is defined individually, each table mapping is specified, and each destination partition is created manually. This works reasonably well for five or ten sources. It becomes painful at twenty. It becomes nearly unmanageable at fifty or more. The operational cost manifests in several ways. Every new store requires a ticket to the data engineering team. Someone must configure the connection, verify the schema compatibility, set up the ingestion pipeline, create the destination structures, and validate the data flow. In a fast-growing retail operation, this creates a bottleneck that delays time-to-insight for new locations. Beyond the immediate operational burden, there is also the risk of configuration drift. When each source is configured individually, small inconsistencies creep in over time. One store might have slightly different table names. Another might be ingesting an extra column that was added during a schema migration. These inconsistencies compound, making the overall system harder to maintain and debug. The solution presented here eliminates most of this manual work by implementing two complementary patterns: automatic source detection through regex-based CDC configuration, and dynamic partition creation through Delta Lake's native capabilities. Architecture Overview The proposed architecture places Debezium as the CDC engine, reading from PostgreSQL databases and publishing change events to Azure Event Hubs. Fabric EventStream consumes these events and writes them to a Delta Lake table in the Lakehouse. The key insight is that neither Debezium nor Delta Lake requires explicit enumeration of every data source. Both can operate on patterns rather than explicit lists. At the source layer, Debezium connects to PostgreSQL and monitors the write-ahead log for changes. Rather than configuring a separate connector for each store database, a single connector is configured with a regex pattern that matches all store databases. When a new database is created that matches this pattern, Debezium automatically begins capturing its changes without any configuration update. At the destination layer, Delta Lake tables are defined with partition columns but without explicit partition values. When a record arrives with a previously unseen partition value, Delta Lake automatically creates the necessary directory structure and begins writing data to the new partition. No DDL statement is required, no manual intervention is needed. The combination of these two behaviors creates an end-to-end pipeline where adding a new store database is as simple as creating the database itself. Everything downstream happens automatically. Implementation Details Configuring Debezium for Automatic Source Detection The critical configuration element in Debezium is the database include list parameter. Rather than specifying each database explicitly, this parameter accepts a regular expression that defines which databases should be monitored. For a retail environment where store databases follow a naming convention such as store_001, store_002, and so forth, the configuration would specify a pattern like store_.* as the include list. This pattern matches any database whose name begins with store_ followed by any characters. When the DBA creates store_058, Debezium detects this new database during its periodic metadata refresh and automatically begins capturing changes from it. The connector configuration should also specify which tables within each database to monitor. In most retail scenarios, the schema is standardized across all stores, so this can be a fixed list such as public.transactions, public.inventory, and public.products. If schema variations exist between stores, additional filtering logic may be required, but this is generally a sign that the source systems need standardization rather than accommodation of inconsistency. The topic routing configuration is equally important. All CDC events should be routed to a single Event Hubs topic, with the store identifier included in the message payload. This allows a single EventStream to process all store data while preserving the ability to identify which store each record originated from. The source metadata that Debezium includes with each change event contains the database name, which serves as the natural store identifier. Initial snapshot behavior must be considered carefully. When Debezium detects a new database, it will perform an initial snapshot to capture the current state before beginning to track incremental changes. For a store with substantial historical data, this snapshot may take anywhere from a few minutes to several hours. During this period, the connector is occupied with the snapshot and may exhibit increased latency for change events from other databases. Scheduling new store database creation during off-peak hours helps mitigate this impact. Configuring Delta Lake for Dynamic Partitioning Delta Lake supports dynamic partition creation as a default behavior. When a table is created with partition columns specified, any INSERT operation that includes a previously unseen partition value will automatically create the corresponding partition directory. The table definition should include the store identifier as the primary partition column. A secondary partition on date is typically advisable for managing data lifecycle and optimizing query performance. The combination of store_id and event_date provides a natural organization that supports both store-specific queries and time-range queries efficiently. The table should be created with automatic optimization enabled. The autoOptimize.optimizeWrite property causes Delta Lake to automatically coalesce small files during write operations, reducing the small file problem that frequently plagues streaming ingestion workloads. The autoOptimize.autoCompact property enables background compaction of files that have accumulated between optimization runs. When EventStream writes a record with store_id equal to store_058 and this value has never been seen before, Delta Lake creates the partition directory structure automatically. The first write creates the directory, and subsequent writes append to files within that directory. From the perspective of downstream queries, the new store's data is immediately available without any schema changes or administrative intervention. EventStream Processing Logic The EventStream configuration bridges the gap between Event Hubs and the Lakehouse. It must parse the Debezium change event format, extract the relevant fields including the store identifier, and route the data to the appropriate Delta table. The transformation logic should extract the store identifier from the source metadata section of the Debezium payload. This is typically found at a path like source.db within the JSON structure. The operation type indicating whether the change is an insert, update, or delete should be preserved for downstream CDC merge processing. A derived column for the event date should be computed from the event timestamp. This serves as the secondary partition key and enables time-based data management. The date extraction should use the source system timestamp rather than the ingestion timestamp to ensure that data is partitioned based on when the business event occurred rather than when it was processed. The destination configuration specifies the Delta table and the partition columns. EventStream handles the actual write operations, and Delta Lake handles the partition management automatically based on the values present in each record. Best Practices for Production Deployment Naming Convention Enforcement The automatic detection pattern is only as reliable as the naming convention it depends on. Before implementing this architecture, establish and enforce a strict naming convention for store databases. Document the convention, communicate it to all teams that provision databases, and implement validation checks in the database provisioning process. The naming convention should be simple and unambiguous. A pattern like store_NNN where NNN is a zero-padded three-digit number provides clear structure and allows for up to 999 stores without format changes. Avoid conventions that might conflict with other databases or that include characters with special meaning in regex patterns. Schema Standardization Automatic source detection assumes that all detected sources share a compatible schema. If store_058 has different table structures than store_001, the downstream processing will fail or produce incorrect results. Schema standardization must be enforced at the source system level before relying on automatic detection. Implement schema validation as part of the store database provisioning process. When a new store database is created, it should be created from a template that guarantees schema compatibility. If schema migrations are necessary, they should be applied uniformly across all store databases before being reflected in the CDC configuration. Partition Key Selection The choice of partition keys has significant implications for both storage efficiency and query performance. The store identifier is the natural first-level partition because it provides the strongest cardinality and aligns with common query patterns such as analyzing a specific store's performance. Date as a secondary partition enables efficient time-range queries and simplifies data lifecycle management. Retention policies can be implemented by dropping old date partitions rather than scanning and deleting individual records. However, the combination of store and date partitions can produce a large number of partition directories. With 100 stores and 365 days of retained data, the table would have 36,500 partitions. While Delta Lake handles this reasonably well, query planning overhead increases with partition count. Consider using month rather than date as the secondary partition if the total partition count becomes problematic. This reduces the partition count by a factor of roughly 30 while still enabling reasonably efficient time-based queries and lifecycle management. Monitoring and Alerting Automatic detection reduces operational burden but does not eliminate the need for monitoring. Implement alerting for several key scenarios. First, monitor for new store detection. While the system handles new stores automatically, operations teams should be notified when a new store begins ingesting data. This serves as a sanity check that the detection is working and provides visibility into the growth of the system. Second, monitor for schema compatibility failures. If a new database is detected but its schema does not match expectations, the CDC process may fail or produce malformed data. Alerting on processing errors helps catch these issues quickly. Third, monitor for snapshot completion. When a new store database triggers an initial snapshot, track the snapshot progress and completion. Extended snapshot times may indicate unusually large source tables or performance issues that warrant investigation. Fourth, monitor partition growth. While dynamic partitioning is convenient, runaway partition creation can indicate a problem such as incorrect store identifiers being generated. Alert if the number of distinct store partitions grows faster than expected based on the known rate of new store openings. Initial Snapshot Planning The initial snapshot that occurs when a new database is detected can be resource-intensive for both the source database and the CDC infrastructure. Plan for this by establishing a new store onboarding window during off-peak hours when the impact of snapshot processing is minimized. Consider implementing a two-phase onboarding process for stores with large historical datasets. In the first phase, configure the database but exclude it from the Debezium include pattern. Perform a bulk historical load using batch processing, which can be throttled and scheduled more flexibly than the streaming snapshot. In the second phase, add the database to the include pattern to begin capturing incremental changes. This approach reduces the load on the streaming infrastructure while still achieving complete data capture. Capacity Planning Dynamic detection and partitioning enable easy scaling in terms of configuration, but the underlying infrastructure must still be sized appropriately. Event Hubs throughput units must accommodate the aggregate event volume from all stores. Fabric capacity units must handle the combined processing load of EventStream and Delta Lake operations. Develop a capacity model that estimates resource requirements per store. Multiply by the current store count plus a growth buffer to determine infrastructure sizing. Review and adjust this model as actual usage patterns become clear. Risk Assessment and Mitigation Uncontrolled Source Proliferation The convenience of automatic detection carries a risk of unintended sources being captured. If the naming convention is not strictly enforced, or if the regex pattern is too broad, databases that should not be ingested may be detected and processed. Mitigation involves implementing strict naming convention governance and using precise regex patterns. The pattern should be as specific as possible while still accommodating legitimate variations. Consider implementing a whitelist in addition to the pattern match, where new databases matching the pattern are flagged for approval before ingestion begins. This adds a manual step but provides a safety checkpoint. Schema Drift Between Sources Over time, individual store databases may drift from the standard schema due to local modifications, failed migrations, or version inconsistencies. When the CDC process encounters unexpected schema elements, it may fail or produce incorrect data. Mitigation requires implementing schema validation at both the source and destination. At the source, periodic schema audits should compare each store database against the canonical schema and flag deviations. At the destination, schema evolution policies in Delta Lake should be configured to reject incompatible changes rather than silently accepting them. The merge schema option should be used cautiously and only when schema evolution is intentional. Partition Explosion Dynamic partition creation can lead to an excessive number of partitions if the partition key has unexpectedly high cardinality or if erroneous data introduces spurious partition values. A misconfigured pipeline might create thousands of partitions, degrading query performance and complicating data management. Mitigation involves implementing partition count monitoring with alerts at defined thresholds. Additionally, validate partition key values before writing to Delta Lake. If a store identifier does not match the expected format, reject the record or route it to an error table for investigation rather than creating an erroneous partition. CDC Lag During Snapshot When a new database triggers an initial snapshot, the Debezium connector dedicates resources to reading the full table contents. During this period, change events from other databases may experience increased latency. In severe cases, the replication slot lag may grow to problematic levels. Mitigation involves scheduling new database provisioning during low-activity periods, sizing the CDC infrastructure with headroom for snapshot operations, and monitoring replication slot lag with alerts at defined thresholds. For very large initial loads, consider the two-phase onboarding approach described earlier. Event Hubs Partition Affinity Event Hubs uses partitions to parallelize message processing. If all messages are routed to a single partition, throughput is limited to what that partition can handle. If messages are distributed across partitions without regard to ordering requirements, related events may be processed out of order. Mitigation involves configuring the Debezium producer to use the store identifier as the partition key. This ensures that all events for a given store are routed to the same Event Hubs partition, preserving ordering within each store while distributing load across partitions for different stores. The number of Event Hubs partitions should be set high enough to accommodate the expected number of stores with room for growth. Unlike some properties, partition count cannot be increased after the Event Hub is created. Orphaned Replication Slots If a store database is decommissioned but the replication slot is not cleaned up, PostgreSQL continues to retain write-ahead log segments for the orphaned slot. Over time, this can fill the disk and cause database outages. Mitigation requires implementing a decommissioning procedure that includes replication slot cleanup. Monitor for inactive replication slots and alert when a slot has not been read from in an extended period. Consider implementing automatic slot cleanup for slots that have been inactive beyond a defined threshold, though this should be done cautiously to avoid accidentally removing slots that are temporarily inactive due to maintenance. Operational Procedures Adding a New Store The procedure for adding a new store is intentionally minimal. The DBA creates the store database following the established naming convention. The database should be created from the standard template to ensure schema compatibility. Within minutes of database creation, Debezium detects the new database and begins the initial snapshot. Operations teams receive a notification of the new store detection. The snapshot progresses, with completion typically occurring within 30 minutes for a standard store data volume. Once the snapshot completes, incremental CDC begins. The first records arriving at the Lakehouse trigger automatic partition creation. From this point forward, the new store's data is fully integrated into the analytics platform with no additional intervention required. Removing a Store Store removal requires more deliberate action than store addition. First, stop ingesting new data by either renaming the database to no longer match the include pattern or by dropping the database entirely. Second, drop the replication slot associated with the store to prevent WAL retention issues. Third, decide whether to retain or purge historical data in the Lakehouse. If historical data should be retained, no action is needed at the Lakehouse level. The partition remains but simply receives no new data. If historical data should be purged, drop the partition using Delta Lake's partition drop capability. This removes the data files and the partition metadata in a single atomic operation. Handling Schema Changes Schema changes require coordination across all store databases and the downstream processing logic. Minor additive changes such as new nullable columns can often be handled through Delta Lake's schema evolution capabilities. The merge schema option allows new columns to be added automatically when encountered. Breaking changes such as column renames, type changes, or column removals require a more deliberate migration process. First, update the downstream processing logic to handle both the old and new schemas. Deploy this change and verify it works with existing data. Then, apply the schema change to source databases in a rolling fashion. Finally, once all sources have been migrated, remove support for the old schema from the processing logic. Disaster Recovery The architecture provides several recovery options depending on the failure scenario. If Event Hubs experiences an outage, Debezium buffers changes locally and resumes publishing when connectivity is restored. The replication slot ensures no changes are lost during the outage, though extended outages may cause WAL accumulation on the source database. If Fabric experiences an outage, events accumulate in Event Hubs up to the retention period. Once Fabric recovers, EventStream resumes processing from its last checkpoint, catching up on accumulated events. The Delta Lake table remains consistent due to its transactional nature. If a source database is lost, recovery depends on backup strategy. The Lakehouse contains a copy of all ingested data, which can serve as a read-only recovery source. Full database recovery requires restoring from PostgreSQL backups. Dynamic partitioning and automatic source detection transform multi-store data ingestion from an operational burden into a largely self-managing system. The combination of Debezium's pattern-based database detection with Delta Lake's dynamic partition creation eliminates most manual configuration work while maintaining the flexibility to accommodate growth. The implementation requires careful attention to naming conventions, schema standardization, and monitoring. The risks are real but manageable with appropriate governance and operational procedures. For organizations operating at scale with dozens or hundreds of similar data sources, this architecture provides a sustainable path to unified analytics without proportional growth in operational overhead. The key principle underlying this approach is that systems should adapt to data rather than requiring data to conform to rigid system configurations. By designing for automatic detection and dynamic accommodation, the data platform becomes a utility that business operations can leverage without constant engineering involvement. This shift from explicit configuration to pattern-based adaptation is essential for organizations seeking to derive value from data at scale.How to Join Microsoft Tech Community & Leina Future Data & AI Hub (Step-by-Step Guide) - Arabic
This session provides a step-by-step walkthrough on how to officially join the Microsoft Tech Community and become a member of the Leina Future Data & AI Hub – Microsoft User Group. The event is designed to help learners, trainers, and data professionals understand: • How to create and verify a Microsoft Tech Community profile • How to join an official Microsoft User Group • How community membership supports learning, networking, and professional growth • How participants can access recorded sessions, resources, and future events This session supports community enablement and knowledge sharing within the Microsoft ecosystem, with a focus on Data, AI, Power BI, and Microsoft Fabric learners. Target Audience: • Students and fresh graduates • Data analysts and Excel/Power BI learners • Trainers and professionals interested in Microsoft communities Session Format: • Recorded session • Practical walkthrough • Community-focused learning Speaker / Host Dr. Leina Nazar Abdelgalil Microsoft Certified Trainer (MCT) Founder – Leina Future Data & AI HubOverload to Optimal: Tuning Microsoft Fabric Capacity
Co-Authored by: Daya Ram, Sr. Cloud Solutions Architect and Rafia Aqil, Could Solutions Architect Optimizing Microsoft Fabric capacity is both a performance and cost exercise. By diagnosing workloads, tuning cluster and Spark settings, and applying data best practices, teams can reduce run times, avoid throttling, and lower total cost of ownership—without compromising SLAs. Use Fabric’s built-in observability (Monitoring Hub, Capacity Metrics, Spark UI) to identify hot spots and then apply cluster- and data-level remediations. Capacity Planning For capacity planning and sizing guidance, see Plan your capacity size. Selecting the wrong SKU can lead to two major issues: Over-provisioning: Paying for resources you don’t need. Under-provisioning: Struggling with performance bottlenecks and failed jobs. To simplify this process, Microsoft provides the Fabric SKU Estimator, a powerful tool designed to help organizations accurately size their capacity based on real-world usage patterns. Run the SKU Estimator before onboarding new workloads or scaling existing ones. Combine its recommendations with monitoring tools like Fabric Capacity Metrics to validate performance and adjust as needed. Options to Diagnose Capacity Issues 1) Monitoring Hub — Start with the Story of the Run What to use it for: Browse Spark activity across applications (notebooks, Spark Job Definitions, and pipelines). Quickly surface long‑running or anomalous runs; view read/write bytes, idle time, core allocation, and utilization. How to use it From the Fabric portal, open Monitoring (Monitor Hub). Select a Notebook or Spark Job Definition to run and choose Historical Runs. Inspect the Run Duration chart; click on a run to see read/write bytes, idle time, core allocation, overall utilization, and other Spark metrics. What to look for Use the guide: application detail monitoring to review and monitor your application. 2) Capacity Metrics App — Measure the Whole Environment What to use it for: Review capacity-wide utilization and system events (overloads, queueing); compare utilization across time windows and identify sustained peaks. How to use it Open the Microsoft Fabric Capacity Metrics app for your capacity. Review the Compute page (ribbon charts, utilization trends) and the System events tab to see overload or throttling windows. Use the Timepoint page to drill into a 30‑second interval and see which operations consumed the most compute. What to look for Use the Troubleshooting guide: Monitor and identify capacity usage to pinpoint top CU‑consuming items. 3) Spark UI — Diagnose at Deeper Level Why it matters: Spark UI exposes skew, shuffle, memory pressure, and long stages. Use it after Monitoring Hub/Capacity Metrics to pinpoint the problematic job. Key tabs to inspect Stages: uneven task durations (data skew), heavy shuffle read/write, large input/output volumes. Executors: storage memory, task time (GC), shuffle metrics. High GC or frequent spills indicate memory tuning is needed. Storage: which RDDs/cached tables occupy memory; any disk spill. Jobs: long‑running jobs and gaps in the timeline (driver compilation, non‑Spark code, driver overload). What to look for Set via environment Spark properties or session config. Data skew, Memory usage, High/Low Shuffles: Adjust Apache Spark settings: i.e. spark.ms.autotune.enabled, spark.task.cpus and spark.sql.shuffle.partitions. Remediation and Optimization Suggestions A) Cluster & Workspace Settings Runtime & Native Execution Engine (NEE) Use Fabric Runtime 1.3 (Spark 3.5, Delta 3.2) and enable the Native Execution Engine to boost performance; enable at the environment level under Spark compute → Acceleration. Starter Pools vs. Custom Pools Starter Pool: prehydrated, medium‑size pools; fast session starts, good for dev/quick runs. Custom Pools: size nodes, enable autoscale, dynamic executors. Create via workspace Spark Settings (requires capacity admin to enable workspace customization). High Concurrency Session Sharing Enable High Concurrency to share Spark Sessions across notebooks (and pipelines) to reduce session startup latency and cost; use session tags in pipelines to group notebooks. Autotune for Spark Enable Autotune (spark.ms.autotune.enabled = true) to auto‑adjust per‑query: spark.sql.shuffle.partitions Spark.sql.autoBroadcastJoinThreshold spark.sql.files.maxPartitionBytes. Autotune is disabled by default and is in preview; enable per environment or session. B) Data‑level best practices Microsoft Fabric offers several approaches to maintain optimal file sizes in Delta tables, review documentation here: Table Compaction - Microsoft Fabric. Intelligent Cache Enabled by default (Runtime 1.1/1.2) for Spark pools: caches frequently read files at node level for Delta/Parquet/CSV; improves subsequent read performance and TCO. OPTIMIZE & Z‑Order Run OPTIMIZE regularly to rewrite files and improve file layout. V‑Order V‑Order (disabled by default in new workspaces) can accelerate reads for read‑heavy workloads; enable via spark.sql.parquet.vorder.default = true. Vacuum Run VACUUM to remove unreferenced files (stale data); default retention is 7 days; align retention across OneLake to control storage costs and maintain time travel. Collaboration & Next Steps Engage Data Engineering Team to Define an Optimization Playbook Start with reviewing capacity sizing guidance, cluster‑level optimizations (runtime/NEE, pools, concurrency, Autotune) and then target data improvements (Z‑order, compaction, caching, query refactors). Triage: Monitor Hub → Capacity Metrics → Spark UI to map workloads and identify high‑impact jobs, and workloads causing throttling. Schedule: Operationalize maintenance: OPTIMIZE (full or selective) during off‑peak windows; enable Auto Compaction for micro‑batch/streaming writes; add VACUUM to your cadence with agreed retention. Add regular code review sessions to ensure consistent performance patterns. Fix: Adjust pool sizing or concurrency; enable Autotune; tune shuffle partitions; refactor problematic queries; re‑run compaction. Verify: Re‑run the job and change, i.e. reduced run time, lower shuffle, improved utilization.
