Architecture¶

This page describes the architecture and data flow of the Time Series Visualization solution.

Solution Components¶

flowchart TB
    subgraph UserInputs["Power BI Report: User Inputs"]
        direction LR
        A[Filters] ~~~ B[Native Slicers] ~~~ C[Time Series Brush Slicer] ~~~ D[Field Parameters]
    end

    subgraph SemanticModel["Power BI Semantic Model"]
        direction LR
        F[Dynamic M Query Parameters] ~~~ G[Power Query Functions] ~~~ H[KQL Queries] ~~~ I[DirectQuery]
    end

    subgraph Backend["KQL Database"]
        direction LR
        J[Tables & Views] ~~~ K[Time Series Processing] ~~~ K2[Anomaly Detection]
    end

    subgraph Visualization["Power BI Visualizations"]
        direction LR
        L[Line Charts] ~~~ M[Statistics Tables]
    end

    UserInputs ==> SemanticModel
    SemanticModel ==> Backend
    SemanticModel ==> Visualization

Key Component Details¶

KQL Database ¶

The KQL Database (in Microsoft Fabric or Azure Data Explorer) provides:

Highly scalable storage for time series data
Performant query execution with native time series functions
Server-side processing for filtering, aggregation, and anomaly detection

Power BI Semantic Model ¶

The semantic model operates in DirectQuery mode, which means:

Queries are executed in real-time against the KQL database
No data is imported or cached in Power BI
Changes to source data are immediately reflected

Dynamic M Query Parameters ¶

Power BI's Dynamic M Query Parameters enable:

Passing user inputs from filters and slicers to Power Query
Constructing custom KQL queries based on user selections
Efficient filtering at the data source level

Power Query Functions ¶

Custom Power Query functions handle:

Parsing parameter values from the brush slicer
Computing optimal time bin sizes
Constructing well-formed KQL queries
Handling edge cases and data validation

Field Parameters ¶

Power BI Field Parameters allow end users to:

Customize chart layouts dynamically
Choose which metrics to display
Configure small multiples and legends

Time Series Brush Slicer ¶

The custom visual provides:

Interactive time range selection
Visual context for the full time period
Anomaly and marker highlighting
Smooth brushing experience

Power BI reports offer rich and highly-interactive data visualization experiences. - Wide variety of visuals, such as tables and line charts - Interactive cross filtering and cross highlighting - Intuitive layout and formatting options

End-to-End Interaction Flow¶

Interaction Flow Diagram¶

This diagram shows how user interactions with the Power BI report trigger interactions between key components of the solution. Considering the iterative nature of interactive analytics, this sequence of steps will be repeated as the user continues to change and refine the focus of their time series exploration.

sequenceDiagram
    autonumber
    actor User
    box rgba(100,149,237,0.1) Power BI Report        
        participant Filters as Filters &<br/>Native Slicers
        participant Brush as Time Series<br/>Brush Slicer
        participant Visuals as Other Visuals
    end

    box rgba(50,205,50,0.1) Semantic Model
        participant PQ as Power Query
    end

    box rgba(255,165,0,0.1) Data Platform
        participant KQL as KQL Database
    end

    rect rgba(200,220,255,0.3)
        Note over User,KQL: Initial Report Load
        User->>Filters: Select tags, time range<br/>& other options
        Filters->>PQ: Pass as Dynamic M Query Parameters
        PQ->>PQ: Parse parameters<br/>Compute time bins<br/>Build KQL queries
        PQ->>KQL: Submit KQL queries
        KQL->>KQL: Retrieve data,<br/>filter, aggregate,<br/>detect anomalies
        KQL-->>PQ: Return aggregated results
        PQ-->>Brush: Render time series overview
    end

    rect rgba(255,220,200,0.3)
        Note over User,KQL: Interactive Exploration with Time Brushing
        User->>Brush: Select portion of time range
        Brush->>PQ: Pass selected range<br/>as Dynamic M Query Parameter
        PQ->>PQ: Parse parameters<br/>Compute time bins<br/>Build KQL queries
        PQ->>KQL: Submit KQL queries
        KQL->>KQL: Retrieve data,<br/>filter, aggregate,<br/>detect anomalies
        KQL-->>PQ: Return aggregated results
        PQ-->>Visuals: Update charts & tables
        Visuals-->>User: Display detailed analysis
    end

Interaction Flow Details¶

This section provides additional details and context about each step in the interaction flow between the user, the report, and other components of the solution.

Use with the interaction flow diagram

These numbered steps correspond to the numbers shown on the preceding interaction flow diagram.

User configures initial report selections — The user selects assets/tags, defines a time range, and sets aggregation options using Power BI's native filters and slicers. These selections determine the scope of data to analyze.
Parameters flow to Power Query — Each slicer and filter is bound to a Dynamic M Query Parameter. When the user makes a selection, Power BI report passes these values to the Power Query engine in the semantic model.
Power Query constructs optimized KQL queries — Custom Power Query functions parse the incoming parameters and dynamically build KQL queries. This includes computing optimal time bin sizes based on the selected range (e.g., 1-minute bins for a 1-hour window, 1-hour bins for a 1-month window) to keep result sets manageable.
Power Query submits queries to the KQL database — Custom KQL queries constructed in the previous step are submitted to the KQL database.
KQL Database executes queries — The query runs entirely within the KQL database, leveraging native time series functions like make-series, series_decompose_anomalies, and summarize. Filtering, aggregation, and anomaly detection all happen at the source.
Aggregated results return to Power BI — Only the processed, summarized results are transferred to Power BI. For a million-point time series, this might be just a few hundred aggregated bins.
Time Series Brush Slicer renders the overview — The time series brush slicer displays data for the full time range, providing visual context and enabling the user to identify periods of interest.
User brushes to select a narrower time range — By clicking and dragging on the time series brush slicer, the user selects a specific portion of the time range for detailed analysis. The slicer outputs the start and end timestamps of the selected range as a formatted text string.
Selected range is passed to the Power Query engine — The time series brush slicer's output is bound to a Dynamic M Query Parameter, which triggers Power Query to rebuild relevant KQL queries with the narrower time window.
Power Query constructs optimized KQL queries — Power Query parses the incoming parameters and dynamically builds KQL queries that reflect the selected time window. Because the time range is smaller, Power Query calculates a finer time granularity (smaller bins), returning more detailed data for the selected period while still keeping the result set manageable.
Power Query submits queries to the KQL database — Custom KQL queries constructed in the previous step are submitted to the KQL database.
KQL Database executes queries — Queries for the selected time range are executed on the KQL database, leveraging native time series functions as well as filtering and aggregation capabilities.
Finer-grained data returns to Power BI — Summarized and enriched results for the selected time period return from the KQL database to Power BI
Updated data is delivered to report visuals — Line charts, statistics tables, and other visuals receive the updated finer-grained data.
Report visuals are rendered to enable detailed analysis — Power BI report visuals are re-rendered to show the detailed view of the selected time range, including any detected anomalies highlighted for investigation.
User continues exploration — The user can further refine selections, change tags, adjust time ranges, or drill into anomalies, with each interaction repeating this efficient interaction cycle.

Why This Architecture Is Fast¶

Optimization	Description
Filtered at source	The KQL database only processes data in your selected time range
Server-side processing	Aggregations, binning, and calculations happen in the highly-performant KQL database
Minimal data transfer	Only summarized results flow to Power BI
Automatic optimization	The report adjusts time granularity based on your selections

Scalability¶

This architecture can handle:

Billions of data points through intelligent binning
Multiple concurrent users by querying highly-scalable KQL Databases in DirectQuery mode
Real-time data without refresh delays
Large tag hierarchies with efficient metadata queries

Security Considerations¶

Authentication uses Microsoft Entra ID
Benefit from extensive security features in Microsoft Fabric
Row-level security can be implemented in the KQL database

Architecture¶

Solution Components¶

Key Component Details¶

KQL Database¶

Power BI Semantic Model¶

Dynamic M Query Parameters¶

Power Query Functions¶

Field Parameters¶

Time Series Brush Slicer¶

Power BI Report¶