Transformations: Reshape data after loading

dlthub transformations let you build new tables or full datasets from datasets that have already been ingested with dlt. dlt transformations are written and run in a very similar fashion to dlt source and resources. dlt transformations require you to have loaded data to a location, for example a local duckdb database, a bucket or a warehouse on which the transformations may be executed. dlt transformations are fully supported for all of our sql destinations including all filesystem and bucket formats.

You create them with the @dlt.hub.transformation decorator, which has the same signature as the @dlt.resource decorator but yields a SQL query, including the resulting column schema, rather than data items. dlt transformations support the same write_dispositions per destination as dlt resources do.

Motivations

A few real-world scenarios where dlt transformations can be useful:

• Build one-stop reporting tables – Flatten and enrich raw data into a wide table that analysts can pivot, slice, and dice without writing SQL each time.
• Clean data – Remove irrelevant columns or anonymize sensitive information before sending it to a layer with lower privacy protections.
• Normalize JSON into 3-NF – Break out repeating attributes from nested JSON so updates are consistent and storage isn't wasted.
• Create dimensional (star-schema) models – Produce fact and dimension tables so BI users can drag-and-drop metrics and break them down by any dimension.
• Generate task-specific feature sets – Deliver slim tables tailored for personalization, forecasting, or other ML workflows.
• Apply shared business definitions – Encode rules such as "a sale is a transaction whose status became paid this month," ensuring every metric is counted the same way.
• Merge heterogeneous sources – Combine Shopify, Amazon, WooCommerce (etc.) into one canonical orders feed for unified inventory and revenue reporting.
• Run transformations during ingestion pre-warehouse – Pre-aggregate or pre-filter data before it hits the warehouse to cut compute and storage costs.
• …and more – Any scenario where reshaping, enriching, or aggregating existing data unlocks faster insight or cleaner downstream pipelines.

Quick-start in three simple steps

For the example below, you can copy–paste everything into one script and run it.

note

It is useful to know how to use dlt Datasets and Relations, since these are heavily used in transformations.

1. Load some example data

The snippets below assume that we have a simple fruitshop dataset as produced by the dlt fruitshop template:

import dlt
from dlt.destinations import duckdb
from dlt._workspace._templates._single_file_templates.fruitshop_pipeline import (
    fruitshop as fruitshop_source,
)

fruitshop_pipeline = dlt.pipeline(
    "fruitshop", destination=duckdb("./test_duck.duckdb"), dev_mode=True
)
fruitshop_pipeline.run(fruitshop_source())

2. Inspect the dataset

# Show row counts for every table
print(fruitshop_pipeline.dataset().row_counts().df())

3. Write and run a transformation

from typing import Any

@dlt.hub.transformation
def copied_customers(dataset: dlt.Dataset) -> Any:
    customers_table = dataset["customers"]
    yield customers_table.order_by("name").limit(5)

# Same pipeline & same dataset
fruitshop_pipeline.run(copied_customers(fruitshop_pipeline.dataset()))

# show rowcounts again, we now have a new table in the schema and the destination
print(fruitshop_pipeline.dataset().row_counts().df())

3.1 Alternatively use pure SQL for the transformation

# Convert the transformation above that selected the first 5 customers to a sql query
@dlt.hub.transformation
def copied_customers(dataset: dlt.Dataset) -> Any:
    customers_table = dataset(
        """
        SELECT *
        FROM customers
        ORDER BY name
        LIMIT 5
    """
    )
    yield customers_table

That's it — copied_customers is now a new table in the same DuckDB schema with the first 5 customers when ordered by name. dlt has detected that we are loading into the same dataset and executed this transformation in SQL - no data was transferred to and from the machine executing this pipeline. Additionally, the new destination table copied_customers was automatically evolved to the correct new schema, and you could also set a different write disposition and even merge data from a transformation.

Defining a transformation

info

Most of the following examples will be using the ibis expressions of the dlt.Dataset. Read the detailed dataset docs to learn how to use these.

@dlt.hub.transformation(name="orders_per_user", write_disposition="merge")
def orders_per_user(dataset: dlt.Dataset) -> Any:
    purchases = dataset.table("purchases").to_ibis()
    yield purchases.group_by(purchases.customer_id).aggregate(
        order_count=purchases.id.count()
    )

• Decorator arguments mirror those accepted by @dlt.resource.
• The transformation function signature must contain at least one dlt.Dataset which is used inside the function to create the transformation SQL statements and calculate the resulting schema update.
• A transformation yields a Relation created with ibis expressions or a select query which will be materialized into the destination table. If the first item yielded is a valid sql query or relation object, data will be interpreted as a transformation. In all other cases, the transformation decorator will work like any other resource.

Loading to other datasets

Loading to another dataset on the same physical location

Below we load to the same DuckDB instance with a new pipeline that points to another dataset. dlt will be able to detect that both datasets live on the same destination and will run the transformation as pure SQL.

import dlt
from dlt.destinations import duckdb

@dlt.hub.transformation
def copied_customers(dataset: dlt.Dataset) -> Any:
    customers_table = dataset["customers"]
    yield customers_table.order_by("name").limit(5)

# Same duckdb instance, different dataset
dest_p = dlt.pipeline(
    "fruitshop_dataset",
    destination=duckdb("./test_duck.duckdb"),
    dataset_name="copied_dataset",
    dev_mode=True,
)
dest_p.run(copied_customers(fruitshop_pipeline.dataset()))

Loading to another dataset on a different physical location

Below we load the data from our local DuckDB instance to a Postgres instance. dlt will use the query to extract the data as Parquet files and will do a regular dlt load, pushing the data to Postgres. Note that you can use the exact same transformation functions for both scenarios. This can be extremely useful when you want to avoid compute costs in warehouses by running transformations directly from a local duckdb instance or raw data in a bucket into the warehouse, as the compute will happen on the machine executing the pipeline that runs the transformations.

# Different engine (Postgres → DuckDB)
duck_p = dlt.pipeline("fruitshop_warehouse", destination="postgres")
duck_p.run(copied_customers(fruitshop_pipeline.dataset()))

Using transformations

Grouping multiple transformations in a source

dlt transformations can be grouped like all other resources into sources and will be executed together. You can even mix regular resources and transformations in one pipeline load.

import dlt

@dlt.source
def my_transformations(dataset: dlt.Dataset) -> Any:
    @dlt.hub.transformation(write_disposition="append")
    def enriched_purchases(dataset: dlt.Dataset) -> Any:
        purchases = dataset.table("purchases").to_ibis()
        customers = dataset.table("customers").to_ibis()
        yield purchases.join(customers, purchases.customer_id == customers.id)

    @dlt.hub.transformation(write_disposition="replace")
    def total_items_sold(dataset: dlt.Dataset) -> Any:
        purchases = dataset.table("purchases").to_ibis()
        yield purchases.aggregate(total_qty=purchases.quantity.sum())

    return enriched_purchases(dataset), total_items_sold(dataset)

fruitshop_pipeline.run(my_transformations(fruitshop_pipeline.dataset()))

Yielding multiple transformations from one transformation resource

dlt transformations may also yield more than one transformation instruction. If no further table name hints are supplied, the result will be a union of the yielded transformation instructions. dlt will take care of the necessary schema migrations, you will just need to ensure that no columns are marked as non-nullable that are missing from one of the transformation instructions:

import dlt

# this (probably nonsensical) transformation will create a union of the customers and purchases tables
@dlt.hub.transformation(write_disposition="append")
def union_of_tables(dataset: dlt.Dataset) -> Any:
    yield dataset.table("purchases")
    yield dataset.table("customers")

Supplying additional hints

You may supply column and table hints the same way you do for regular resources. dlt will derive schema hints from your query, but in some cases you may need to modify or extend them — for example, making columns nullable as in the example above, or adjusting the precision or type of a column to ensure compatibility with a specific target destination (if it differs from the source).

import dlt

# change precision and scale of the price column
@dlt.hub.transformation(
    write_disposition="append", columns={"price": {"precision": 10, "scale": 2}}
)
def precision_change(dataset: dlt.Dataset) -> Any:
    yield dataset.inventory

Writing your queries in SQL

If you prefer to write your queries in SQL, you can omit ibis expressions by simply creating a Relation from a query on your dataset:

# Convert the transformation above that selected the first 5 customers to a sql query
@dlt.hub.transformation
def copied_customers(dataset: dlt.Dataset) -> Any:
    customers_table = dataset(
        """
        SELECT *
        FROM customers
        ORDER BY name
        LIMIT 5
    """
    )
    yield customers_table


# Joins and other more complex queries are also possible
@dlt.hub.transformation
def enriched_purchases(dataset: dlt.Dataset) -> Any:
    enriched_purchases = dataset(
        """
        SELECT customers.name, purchases.quantity
        FROM purchases
        JOIN customers
            ON purchases.customer_id = customers.id
        """
    )
    yield enriched_purchases

# You can even use a different dialect than the one used by the destination by supplying the dialect parameter
# dlt will compile the query to the right destination dialect
@dlt.hub.transformation
def enriched_purchases_postgres(dataset: dlt.Dataset) -> Any:
    enriched_purchases = dataset(
        """
        SELECT customers.name, purchases.quantity
        FROM purchases
        JOIN customers
            ON purchases.customer_id = customers.id
        """,
        query_dialect="duckdb",
    )
    yield enriched_purchases

The identifiers (table and column names) used in these raw SQL expressions must correspond to the identifiers as they are present in your dlt schema, NOT in your destination database schema.

Using Pandas or Polars DataFrames and Arrow tables

You can also write transformations directly using Pandas or Polars DataFrames and Arrow tables. Note that in this case your transformation resource behaves like a regular resource: column-level hints will not be propagated, and dlt will simply treat the yielded DataFrames or Arrow tables like data from any other resource. This behavior may change in the future.

@dlt.hub.transformation
def copied_customers(dataset: dlt.Dataset) -> Any:
    # get full customers table as arrow table
    customers = dataset.table("customers").arrow()

    # Sort the table by 'name'
    sorted_customers = customers.sort_by([("name", "ascending")])

    # Take first 5 rows
    yield sorted_customers.slice(0, 5)

# Example tables (replace with your actual data)
@dlt.hub.transformation
def enriched_purchases(dataset: dlt.Dataset) -> Any:
    # get both fully tables as dataframes
    purchases = dataset.table("purchases").df()
    customers = dataset.table("customers").df()

    # Merge (JOIN) the DataFrames
    result = purchases.merge(customers, left_on="customer_id", right_on="id")

    # Select only the desired columns
    yield result[["name", "quantity"]]

Incremental transformations

When source data keeps growing, rerunning the same full transformation every time is slow and expensive.

Incremental transformations let each run work on the right slice of source data instead of the whole dataset. Each slice is defined by a cursor: a column whose values dlt compares against a range to decide if a row is in scope. Common choices are created_at, updated_at, an increasing id, or the dlt-managed _dlt_loads.inserted_at.

There are two common ways to choose the slice:

• Use the scheduler interval. dltHub Platform uses this approach: the scheduler sets [start, end) interval this run is responsible for.
• Continue from the previous run. dltHub stores the last cursor value it processed and the next run starts after that value.

The scheduler interval

Use a scheduler interval when the orchestrator decides what time range each run should process. This is the natural fit for cron schedules, retries, and backfills because the run does not depend on what happened in a previous run.

Set allow_external_schedulers=True on the cursor and dltHub Platform owns the interval: its cron schedules set DLT_INTERVAL_START and DLT_INTERVAL_END, which are picksed up to filter the source data.

Here's an example. The transformation below reads the orders table and writes only the rows whose created_at falls in the [start, end) window to a new table orders_window.

Given an orders table with one row per day:

id	created_at
1	2026-01-01
2	2026-01-02
…	…
10	2026-01-10

and a scheduler window of [2026-01-05, 2026-01-10), the run writes ids 5 to 9 to orders_window (id 10 is excluded by the open upper bound).

@dlt.hub.transformation(write_disposition="replace")
def orders_window(
    dataset: dlt.Dataset,
    window: dlt.sources.incremental[pendulum.DateTime] = dlt.sources.incremental(
        "created_at",
        initial_value=pendulum.datetime(2000, 1, 1, tz="UTC"),
        allow_external_schedulers=True,
        range_start="closed",
        range_end="open",
    ),
) -> Any:
    yield dataset.table("orders").incremental(window)

Re-running the same [start, end) (start is included, end is excluded) interval produces the same transformation input, which makes this pattern a good fit for partition backfills and idempotent retries.

Continue from the previous run

Use a stateful cursor for runs not tied to an external scheduler, where each run should continue from the last successful one. dltHub stores the cursor state internally and uses it in the next run of the transformation.

The transformation below appends rows from orders whose created_at is later than the persisted last_value to a new table recent_orders.

Implicit cursor

The cursor below is declared on the decorator; the body yields a bare relation (Ibis expressions and raw SQL strings work too) and dltHub applies the filter automatically. The scheduler example above shows the alternative form, with the cursor as a function argument.

@dlt.hub.transformation(
    write_disposition="append",
    primary_key="id",
    incremental=dlt.sources.incremental(
        "created_at",
        initial_value=pendulum.datetime(2000, 1, 1, tz="UTC"),
        range_start="open",
    ),
)
def recent_orders(dataset: dlt.Dataset) -> Any:
    yield dataset.table("orders")

Now suppose orders is loaded in two batches:

batch	ids	created_at
initial	1..3	2026-01-01 .. 2026-01-03
later	4..5	2026-01-04 .. 2026-01-05

The first run has no last_value yet, so it starts from initial_value (2000-01-01), writes the three initial rows to recent_orders, and advances last_value to 2026-01-03. The next run sees the two later rows fall past last_value, appends them, and advances last_value to 2026-01-05.

Set range_start="open" on stateful cursors

A stateful cursor persists last_value after each run. With the default range_start="closed", the next run's filter is cursor >= last_value, so the row at the boundary is re-emitted every time. Set range_start="open" to make the filter cursor > last_value and exclude the boundary row.

Cursor column choices

Use a domain cursor when the source table has a column that represents creation or update order. For append-only data, created_at or an increasing id is usually enough. For mutable data, use a cursor that changes whenever the row changes, such as updated_at; rows whose cursor value does not advance are intentionally ignored by the next stateful run.

Use _dlt_loads.inserted_at when the source table has no domain timestamp and you want to process data by the time dltHub loaded it. A dotted cursor path such as _dlt_loads.inserted_at tells dltHub to follow the schema reference from the base table to _dlt_loads, join it, and filter on the joined column. The join is filter-only: columns from _dlt_loads are not added to the destination table.

@dlt.hub.transformation(write_disposition="append")
def orders_by_load(
    dataset: dlt.Dataset,
    loaded_at: dlt.sources.incremental[pendulum.DateTime] = dlt.sources.incremental(
        "_dlt_loads.inserted_at",
        initial_value=pendulum.datetime(2000, 1, 1, tz="UTC"),
        range_start="open",
    ),
) -> Any:
    yield dataset.table("orders").incremental(loaded_at)

note

Under the hood, when dltHub can run the transformation directly as SQL/model job, the source query is modified to include the cursor filter. When the transformation is materialized first, for example if source and destination are different physical engines, or when you yield Python objects such as lists, Arrow tables, or DataFrames, filtering happens during extraction.

State and safety rules

• LIMIT is rejected on stateful relation incrementals. Advancing state from a limited result can skip rows that were not returned. Remove the limit or use an explicit bounded window.
• SQL-based cursors support max and min last-value functions. Custom Python last_value_func callables cannot be pushed down to SQL.
• Null handling follows on_cursor_value_missing. For SQL pushdown, "include" adds OR cursor IS NULL; "exclude" adds AND cursor IS NOT NULL; "raise" cannot raise in the middle of a query and falls back to excluding null cursor values when needed.

For lower-level cursor rules, including range inclusivity and lag, see Filter to an incremental cursor and Cursor-based incremental loading.

Schema evolution and hints lineage

When executing transformations, dlt computes the resulting schema before the transformation is executed. This allows dlt to:

1. Migrate the destination schema accordingly, creating new columns or tables as needed
2. Fail early if there are schema mismatches that cannot be resolved
3. Preserve column-level hints from source to destination

Schema evolution

For example, if your transformation joins two tables and creates new columns, dlt will automatically update the destination schema to accommodate these changes. If your transformation would result in incompatible schema changes (like changing a column's data type in a way that could lose data), dlt will fail before executing the transformation, protecting your data and saving execution and debug time.

You can inspect the computed result schema during development by looking at the result of compute_columns_schema on your Relation:

# Show the computed schema before the transformation is executed
dataset = fruitshop_pipeline.dataset()
purchases = dataset.table("purchases").to_ibis()
customers = dataset.table("customers").to_ibis()
enriched_purchases = purchases.join(
    customers, purchases.customer_id == customers.id
)
print(dataset(enriched_purchases).columns)

Column level hint forwarding

When creating or updating tables with transformation resources, dlt will also forward certain column hints to the new tables. In our fruitshop source, we have applied a custom hint named x-annotation-pii set to True for the name column, which indicates that this column contains PII (personally identifiable information). Downstream of the transformation layer, we may want to know which columns originate from columns that contain private data:

@dlt.hub.transformation
def enriched_purchases(dataset: dlt.Dataset) -> Any:
    enriched_purchases = dataset(
        """
        SELECT customers.name, purchases.quantity
        FROM purchases
        JOIN customers
            ON purchases.customer_id = customers.id
        """
    )
    yield enriched_purchases

# Let's run the transformation and see that the name column in the NEW table is also marked as PII
fruitshop_pipeline.run(enriched_purchases(fruitshop_pipeline.dataset()))
assert (
    fruitshop_pipeline.dataset().schema.tables["enriched_purchases"]["columns"][
        "name"
    ][
        "x-annotation-pii"  # type: ignore
    ]
    is True
)

Features and limitations:

• dlt will only forward certain types of hints to the resulting tables: custom hints starting with x-annotation... and type hints such as nullable, data_type, precision, scale, and timezone. Other hints, such as primary_key or merge_keys, will need to be set via the columns argument on the transformation decorator, since dlt does not know how the transformed tables will be used.
• dlt cannot forward hints for columns that result from combining multiple origin columns, such as when they are concatenated or produced through other SQL operations.

Lifecycle of a SQL transformation

In this section, we focus on the lifecycle of transformations that yield a Relation object, which we call SQL transformations here. This is in contrast to Python-based transformations that yield dataframes, arrow tables, or polars frames, which go through the regular extract, normalize, and load lifecycle of a dlt resource.

Extract

In the extract stage, a Relation yielded by a transformation is converted into a SQL string and saved as a .model file along with its source SQL dialect. At this stage, the SQL string is just the user's original query — either the string that was explicitly provided or the one generated by Relation.to_sql(). No dlt-specific columns like _dlt_id or _dlt_load_id are added yet.

Normalize

In the normalize stage, .model files are read and processed. The normalization process modifies your SQL queries to ensure they execute correctly and integrate with dlt's features.

info

The normalization described here applies only to SQL-based transformations. Python-based transformations, such as those using dataframes, arrow tables, or polars frames, follow the regular normalization process.

Adding dlt columns

During normalization, dlt adds internal dlt columns to your SQL queries depending on the configuration:

• _dlt_load_id, which tracks which load operation created or modified each row, is added by default. Even if present in your query, the _dlt_load_id column will be replaced with a constant value corresponding to the current load ID. To disable this behavior, set:

  [normalize.model_normalizer]
  add_dlt_load_id = false

  In this case, the column will not be added or replaced.

• _dlt_id, a unique identifier for each row, is not added by default. If your query already includes a _dlt_id column, it will be left unchanged. To enable automatic generation of this column when it’s missing, set:

  [normalize.model_normalizer]
  add_dlt_id = true

  When enabled and the column is not in the query, dlt will generate a _dlt_id. Note that if the column is already present, it will not be replaced.

  The _dlt_id column is generated using the destination's UUID function, such as generateUUIDv4() in ClickHouse. For dialects without native UUID support:

  • In Redshift, _dlt_id is generated using an MD5 hash of the load ID and row number.
  • In SQLite, _dlt_id is simulated using lower(hex(randomblob(16))).

Query transformations

The normalization process also applies the following transformations to ensure your queries work correctly:

1. Fully qualifies all identifiers with database and dataset prefixes
2. Quotes and adjusts identifier casing to match destination requirements
3. Normalizes column names according to the selected naming convention
4. Aliases columns and tables to handle naming convention differences
5. Reorders columns to match the destination table schema
6. Fills in NULL values for columns that exist in the destination but aren't in your query

Load

In the load stage, the normalized queries from .model files are wrapped in INSERT statements and executed on the destination. For example, given this query from the extract stage:

SELECT
    "my_table"."id" AS "id",
    "my_table"."value" AS "value"
FROM "my_pipeline_dataset"."my_table" AS "my_table"

After the normalize stage processes it (adding dlt columns, wrapping in subquery, etc.) and results in:

SELECT
    _dlt_subquery."id" AS "id",
    _dlt_subquery."value" AS "value",
    '1749134128.17655' AS "_dlt_load_id",
    UUID() AS "_dlt_id"
FROM (
    SELECT
        "my_table"."id" AS "id",
        "my_table"."value" AS "value"
    FROM "my_pipeline_dataset"."my_table" AS "my_table"
    )
AS _dlt_subquery

The load stage executes:

INSERT INTO
    "my_pipeline_dataset"."my_transformation" ("id", "value", "_dlt_load_id", "_dlt_id")
SELECT
    _dlt_subquery."id" AS "id",
    _dlt_subquery."value" AS "value",
    '1749134128.17655' AS "_dlt_load_id",
    UUID() AS "_dlt_id"
FROM (
    SELECT
        "my_table"."id" AS "id",
        "my_table"."value" AS "value"
    FROM "my_pipeline_dataset"."my_table" AS "my_table"
    )
AS _dlt_subquery

The query is executed via the destination's SQL client, materializing the transformation result directly in the database.

Examples

Local in-transit transformations example

If you require aggregated or otherwise transformed data in your warehouse, but would like to avoid or reduce the costs of running queries across many rows in your warehouse tables, you can run some or all of your transformations "in transit" while loading data from your source. The code below demonstrates how you can extract data with our rest_api source to a local DuckDB instance and then forward aggregated data to a warehouse destination.

from dlt.sources.rest_api import (
    rest_api_source,
)

# loads some data from our example api at https://jaffle-shop.scalevector.ai/docs
source = rest_api_source(
    {
        "client": {
            "base_url": "https://jaffle-shop.scalevector.ai/api/v1",
        },
        "resources": [
            "stores",
            {
                "name": "orders",
                "endpoint": {
                    "path": "orders",
                    "params": {
                        "start_date": "2017-01-01",
                        "end_date": "2017-01-31",
                    },
                },
            },
        ],
    }
)

# load to a local DuckDB instance
transit_pipeline = dlt.pipeline(
    "jaffle_shop", destination="duckdb", dataset_name="in_transit"
)
transit_pipeline.run(source)

# load aggregated data to a warehouse destination
@dlt.hub.transformation
def orders_per_store(dataset: dlt.Dataset) -> Any:
    orders = dataset.table("orders").to_ibis()
    stores = dataset.table("stores").to_ibis()
    yield (
        orders.join(stores, orders.store_id == stores.id)
        .group_by(stores.name)
        .aggregate(order_count=orders.id.count())
    )

# load aggregated data to a warehouse destination
warehouse_pipeline = dlt.pipeline(
    "jaffle_warehouse",
    destination="postgres",
    dataset_name="warehouse",
    dev_mode=True,
)
warehouse_pipeline.run(orders_per_store(transit_pipeline.dataset()))

This script demonstrates:

• Fetching data from a REST API using dlt's rest_api_source
• Loading raw data into a local DuckDB instance as an intermediate step
• Transforming the data by joining orders with stores and aggregating order counts directly on the local DuckDB instance, not in the destination warehouse
• Loading only the aggregated results to a production warehouse (Postgres)
• Reducing warehouse compute costs by performing transformations locally in DuckDB
• Using multiple pipelines in a single workflow for different stages of processing