DeltaFlow output transformation (preview)

DeltaFlow is a capability within Fabric Eventstreams that transforms raw Change Data Capture (CDC) events into a flattened, analytics-ready format. Instead of working with deeply nested Debezium JSON payloads, you get tabular rows that closely mirror the structure of your source database tables — enriched with metadata columns that describe each change.

This article describes the transformation process, the metadata columns DeltaFlow adds, and how different operation types (inserts, updates, and deletes) are represented in the output.

Raw Debezium CDC event structure

CDC connectors powered by Debezium emit change events as JSON envelopes. A typical raw event contains nested structures including the row data before and after the change, source metadata, and the operation type:

{
  "before": null,
  "after": {
    "OrderID": 1001,
    "CustomerName": "Contoso Ltd",
    "OrderTotal": 249.99,
    "OrderDate": "2026-03-01"
  },
  "source": {
    "version": "2.x",
    "connector": "sqlserver",
    "name": "fabsql",
    "ts_ms": 1741305600000,
    "db": "SalesDB",
    "schema": "dbo",
    "table": "Orders"
  },
  "op": "c",
  "ts_ms": 1741305600123
}

This nested format requires parsing and extraction before you can run analytics queries — for example, to access OrderTotal, you need to navigate into the after object and extract it. Querying this format directly with KQL or SQL is complex and error-prone.

DeltaFlow transformed output

DeltaFlow flattens the nested Debezium envelope into a single tabular row. Source table columns appear as top-level columns, and DeltaFlow appends metadata columns that describe the change:

You can query this output directly with KQL or other analytics tools — no parsing required.

Metadata columns

DeltaFlow adds the following metadata columns to every output row:

Operation types

DeltaFlow represents each type of database change as a distinct operation. The following table describes when each operation occurs:

Initial snapshot

When a CDC connector first starts, it performs an initial snapshot of the source table to capture all existing rows. Each row in the snapshot appears as an Insert operation in the DeltaFlow output:

After the snapshot completes, the connector switches to streaming mode and captures ongoing changes.

Insert

When a new row is added to the source table, DeltaFlow emits a single row with the operation Insert:

Update

When an existing row is modified, DeltaFlow emits two rows with the same timestamp — one for the state before the change (Pre_Update) and one for the state after (Post_Update). This pattern is similar to Delta change data feeds:

In this example, the OrderTotal changed from 249.99 to 299.99. Both the before and after states are preserved, which makes it possible to compute diffs, track field-level changes, or build audit trails.

Delete

When a row is deleted from the source table, DeltaFlow emits a single row with the operation Delete, containing the last known values of the row:

Schema and destination table mapping

When you route a DeltaFlow-enabled stream to a supported destination such as an Eventhouse, DeltaFlow automatically:

1. Creates destination tables that match the source table structure, with columns for each source column plus the five metadata columns.
2. Manages schema evolution — when source tables change (for example, columns are added or new tables are created), DeltaFlow detects the changes, updates the registered schemas, and adjusts the destination tables accordingly.

For more information on destination table management, see the DeltaFlow capabilities section in each CDC connector page.

Related content

• Eventstreams overview — DeltaFlow section
• Add Azure SQL Database CDC source
• Add Azure SQL Managed Instance CDC source
• Add SQL Server on VM CDC source
• Add PostgreSQL Database CDC source