Delta Funnel

Delta Lake to SQL Server. No Spark. No JDBC/ODBC bottleneck.

DataFusion SQL in.
Native TDS bulk load out.

Observed: 13.4M rows in ~14 minutes vs. a ~2 hour Spark/JDBC path.
Read the Delta Funnel documentation.

When To Use It

Use Delta Funnel when:

• SQL Server writes are the bottleneck.
• Spark is too much machinery for a focused export.
• You want SQL transforms over Delta Lake without a cluster.
• You want Rust or Python orchestration with reports and tracing.

Install

For Rust, add the delta-funnel crate:

cargo add delta-funnel

For Python, add the deltafunnel package:

uv add deltafunnel

Python Quickstart

from deltafunnel import Session

ado_connection_string = (
    "server=tcp:localhost,1433;"
    "database=warehouse;"
    "User ID=etl_user;"
    "Password=REPLACE_ME;"
    "encrypt=true;"
    "TrustServerCertificate=yes"
)

session = Session(default_mssql_connection_string=ado_connection_string)

# Register the Delta table as "orders" so SQL can reference it.
orders = session.delta_lake("file:///path/to/orders-delta", name="orders")

# Build a lazy DataFusion SQL query. No rows are read yet.
daily_orders = session.table_from_sql("""
    select customer_id, order_date, total_amount
    from orders
    where order_date >= date '2026-01-01'
""")

# Preview executes the DataFusion query with a limit; notebooks render it as a table.
daily_orders.preview(limit=20)

# Write executes the query and loads the result into SQL Server.
report = daily_orders.write_to_mssql(
    schema="dbo",
    table="daily_orders",
    load_mode="create_and_load",  # "replace" also supports a missing target
    # dry_run=True,  # validate the load plan without writing rows
)

For private S3 sources, SQL Server load modes, dry runs, and reports, see the private S3 sources, SQL Server guide, and dry runs and reports.

Multi-output Writes

For workflows that write several related tables in one run, use Table.to_mssql(...) to create output specs and Session.write_all(...) to execute them together. Shared lazy SQL dependencies can be cached during the workflow so common upstream work is not repeated for each output.

See the multiple outputs and shared caching guide for setup, dry runs, cache options, and failure behavior.

Rust Quickstart

use delta_funnel::{
    DeltaFunnelRuntime, DeltaFunnelSession, DeltaSourceConfig, LoadMode,
    MssqlConnectionConfig, MssqlOutputTarget, MssqlTargetConfig,
    MssqlTargetTable, OutputWritePlan, RunMode, SessionOptions,
};

fn main() -> Result<(), Box<dyn std::error::Error>> {
    let ado_connection_string = concat!(
        "server=tcp:localhost,1433;",
        "database=warehouse;",
        "User ID=etl_user;",
        "Password=REPLACE_ME;",
        "encrypt=true;",
        "TrustServerCertificate=yes",
    );

    let default_connection = MssqlConnectionConfig::new(ado_connection_string)?
        .with_display_label("warehouse");
    let mut session = DeltaFunnelSession::new(
        SessionOptions::new().with_default_mssql_connection(default_connection),
    )?;
    let runtime = DeltaFunnelRuntime::new()?;

    // Register the Delta table as "orders" so SQL can reference it.
    let _orders = session.delta_lake(DeltaSourceConfig::new(
        "orders",
        "file:///path/to/orders-delta",
    ))?;

    // Build a lazy DataFusion SQL query. No rows are read yet.
    let daily_orders = runtime.table_from_sql(
        &mut session,
        r#"
        select customer_id, order_date, total_amount
        from orders
        where order_date >= date '2026-01-01'
        "#,
    )?;

    // Preview executes the DataFusion query with a limit.
    let preview = runtime.preview_table(&session, &daily_orders, 20)?;
    println!("{}", preview.text());

    // Write executes the query and loads the result into SQL Server.
    let target = MssqlTargetConfig::new(MssqlTargetTable::new("dbo", "daily_orders")?)
        .with_load_mode(LoadMode::CreateAndLoad);
    let output = OutputWritePlan::new(
        daily_orders,
        MssqlOutputTarget::new("daily_orders", target, RunMode::Execute),
    );
    let report = runtime.write_to_mssql(&session, &output)?;

    println!("wrote output {}", report.output_name());
    Ok(())
}

For a guided version of this workflow, see the Rust quickstart. For the full Rust API, see docs.rs/delta-funnel and the query_load_dry_run example.

Development

For local builds and test setup, see the local development guide.