Rulebricks Spark

Apply versioned Rulebricks rules and flows to PySpark DataFrames. The library sends records in bounded requests and keeps each decision, error, and execution ID alongside its source data, so the pipeline can scale without losing the ability to explain a run.

pip install rulebricks-spark

import os

from rulebricks_spark import apply_flow

results = apply_flow(
    source,
    "order-routing-flow",
    version="1",
    api_key=os.environ["RULEBRICKS_API_KEY"],
    base_url="https://rulebricks.acme.com/api/v1",
    correlation_id_path="request_id",
    execution_partitions=4,
)
results.write.format("delta").mode("errorifexists").save(result_path)
saved = spark.read.format("delta").load(result_path)

Flow results remain JSON by default, ready for your own Spark expressions. apply_rule() uses the same execution model and infers a typed result from the selected rule version. Context operations provide optional input staging and ordered stateful processing, while Structured Streaming uses Spark's ordinary foreachBatch interface.

Start with the documentation or the notebook walkthroughs. The guides explain how the concepts fit together; the reference covers parameters, result fields, failure handling, and recovery. Notebook settings are defined in the notebooks, with only credentials kept in a secret manager or environment.

The package targets classic PySpark 3.5 and requires Python 3.9+ and PyArrow 18+. Use a Python version supported by your Spark runtime. For a documentation preview, follow the Sphinx build instructions.