Create an Incremental Python Read Component

In this guide, we'll build an Incremental Python Read Component that ingests only new or updated records by leveraging Ascend's Incremental strategy.

Prerequisites

• Ascend Flow

Create a new Component

Begin from your workspace Super Graph view. Follow these steps to create your component:

Using the Component Form

1. Double-click the Flow where you want to create your component
2. Right-click anywhere in the Flow Graph
3. Hover over Create Component, then over Read in the expanded menu, and click From Scratch
4. Complete the form with these details:
   • Select your Flow
   • Enter a descriptive Component Name like read_sales
   • Select Python as your file type

Using the Files Panel

1. Open the files panel in the top left corner
2. Navigate to and select your desired Flow
3. Right-click on the components directory and choose New file
4. Name your file with a descriptive name like read_sales.py and press enter

Create your Incremental Python Read Component

Structure your Incremental Python Read Component following this pattern, based on our Otto's Expeditions project:

1. Import necessary packages: Include Ascend resources (read), context handlers (ComponentExecutionContext), data processing libraries (like Polars or pandas), and logging utilities (log)

2. Apply the @read() decorator with incremental configuration:

   • Set strategy="incremental" to enable incremental processing
   • Choose incremental_strategy="merge" or "append" based on your data needs
   • Specify unique_key for merge operations
   • Set on_schema_change="sync_all_columns" to handle schema evolution

3. Define your incremental read function: Implement logic that filters data based on previous state

4. Return structured data: Return the processed data as a dataframe or table

The @read() decorator integrates your function into Ascend's stateful execution framework, automatically managing incremental state and schema updates.

Choose an Incremental Strategy

Ascend offers two incremental strategies: merge and append. Choose based on your data requirements:

Strategy	Description	When to Use	Required Parameters
Merge	Updates existing records based on a key and inserts new ones	For data that can be updated (user profiles, product info)	unique_key
Append	Simply adds new data to existing dataset	For immutable data (logs, events)	None

For detailed explanations, see our incremental processing reference guide.

Merge strategy example

This example demonstrates the merge incremental strategy using our Otto's Expeditions project:

incremental_read_merge.py

import polars as pl
import pyarrow as pa
from ascend.application.context import ComponentExecutionContext
from ascend.common.events import log
from ascend.resources import read


@read(
    strategy="incremental",
    incremental_strategy="merge",
    unique_key="id",
    on_schema_change="sync_all_columns",
)
def read_inlinked(context: ComponentExecutionContext) -> pa.Table:
    """
    Incrementally reads 'inlinked' event data from Otto's Expeditions project.
    """
    # Load all available parquet files; Ascend handles merging based on state
    df = pl.read_parquet("gs://ascend-io-gcs-public/ottos-expeditions/lakev0/generated/events/inlinked.parquet/year=*/month=*/day=*/*.parquet")
    # Retrieve previous data to determine last processed timestamp
    current_data = context.current_data()
    if current_data is not None:
        current_data = current_data.to_polars()
        max_ts = current_data["timestamp"].max()
        log(f"Reading data after {max_ts}")
        df = df.filter(df["timestamp"] > max_ts)
    else:
        log("No current data found, loading all data")

    log(f"Returning {df.height} rows")
    return df.to_arrow()

Append strategy example

This example demonstrates the append incremental strategy with timestamp-based filtering:

incremental_read_append.py

from datetime import datetime

from ascend.application.context import IncrementalComponentExecutionContext
from ascend.resources import read

# Example data for timestamped incremental append
initial_data = {
    "key": [1, 2],
    "ts": [datetime(2020, 1, 1, 12, 0), datetime(2020, 1, 2, 12, 30)],
    "string": ["a", "b"],
    "integer": [1, 2],
}

updated_data = {
    "key": [1, 2, 3],
    "ts": [
        datetime(2021, 1, 1, 12, 0),
        datetime(2020, 1, 2, 12, 30),
        datetime(2021, 2, 1, 12, 0),
    ],
    "string": ["a", "bb", "c"],
    "integer": [1, 2, 3],
}


@read(strategy="incremental", incremental_strategy="append")
def incremental_custom_python_read_append(context: IncrementalComponentExecutionContext):
    """Demonstrates the append incremental strategy with timestamp-based filtering."""
    current_data = context.current_data("pandas")

    data = initial_data
    if context.is_incremental:
        # row 1 has an updated timestamp, so it is included in the delta
        # row 3 is new and has a timestamp > than the max in the current output
        ts_key = "ts"
        if context.data_plane_type == "snowflake":
            ts_key = ts_key.upper()
        max_ts = current_data[ts_key].max()

        filtered_data = {key: [val for i, val in enumerate(updated_data[key]) if updated_data["ts"][i] > max_ts] for key in updated_data.keys()}
        # Since we are using the append strategy, the updated row 1 will get appended to the existing output.
        data = filtered_data

    if context.data_plane_type == "snowflake":
        # make keys uppercase for Snowflake compatibility
        data = {key.upper(): val for key, val in data.items()}

    return data

For more examples and advanced options, see our reference guide.

🎉 Congratulations! You've successfully created an Incremental Python Read Component in Ascend.