spark-fuse 1.0.2

Open-source PySpark toolkit with data sources for REST APIs and SPARQL endpoints.

spark-fuse

spark-fuse is an open-source toolkit for PySpark — providing utilities, data sources, and tools to fuse your data workflows across JSON-centric REST APIs and SPARQL endpoints.

Features

• Data sources for REST APIs (JSON payloads with pagination/retry support) and SPARQL services.
• SparkSession helpers with sensible defaults and environment detection (databricks/fabric/local heuristics retained for legacy jobs).
• DataFrame utilities for previews, schema checks, and ready-made date/time dimensions (daily calendar attributes and clock buckets).
• LLM-powered semantic column normalization that batches API calls and caches responses.
• Typer-powered CLI: list data sources and preview datasets via the REST/SPARQL helpers.

Installation

• Create a virtual environment (recommended)
  • macOS/Linux:
    • python3 -m venv .venv
    • source .venv/bin/activate
    • python -m pip install --upgrade pip
  • Windows (PowerShell):
    • python -m venv .venv
    • .\\.venv\\Scripts\\Activate.ps1
    • python -m pip install --upgrade pip
• From source (dev): pip install -e ".[dev]"
• From PyPI: pip install "spark-fuse>=0.3.2"

Quickstart

1. Create a SparkSession with helpful defaults

from spark_fuse.spark import create_session
spark = create_session(app_name="spark-fuse-quickstart")

2. Load paginated REST API responses

import json
from spark_fuse.io import (
    REST_API_CONFIG_OPTION,
    REST_API_FORMAT,
    build_rest_api_config,
    register_rest_data_source,
)

register_rest_data_source(spark)
config = build_rest_api_config(
    spark,
    "https://pokeapi.co/api/v2/pokemon",
    source_config={
        "request_type": "GET",  # switch to "POST" for endpoints that require a body
        "records_field": "results",
        "pagination": {"mode": "response", "field": "next", "max_pages": 2},
        "params": {"limit": 20},
    },
)
pokemon = (
    spark.read.format(REST_API_FORMAT)
    .option(REST_API_CONFIG_OPTION, json.dumps(config))
    .load()
)
pokemon.select("name").show(5)

Need to hit a POST endpoint? Set "request_type": "POST" and attach your payload with "request_body": {...} (defaults to JSON encoding—add "request_body_type": "data" for form bodies). Flip on "include_response_payload": True to add a response_payload column with the raw server JSON.

3. Query a SPARQL endpoint

sparql_query = """
PREFIX wd: <http://www.wikidata.org/entity/>
PREFIX wdt: <http://www.wikidata.org/prop/direct/>

SELECT ?pokemon ?pokemonLabel ?pokedexNumber WHERE {
  ?pokemon wdt:P31 wd:Q3966183 .
  ?pokemon wdt:P1685 ?pokedexNumber .
}
LIMIT 5
"""

from spark_fuse.io import (
    SPARQL_CONFIG_OPTION,
    SPARQL_DATA_SOURCE_NAME,
    build_sparql_config,
    register_sparql_data_source,
)

register_sparql_data_source(spark)
sparql_options = build_sparql_config(
    spark,
    "https://query.wikidata.org/sparql",
    source_config={
        "query": sparql_query,
        "request_type": "POST",
        "headers": {"User-Agent": "spark-fuse-demo/0.3 ([e-mail])"},
    },
)
sparql_df = (
    spark.read.format(SPARQL_DATA_SOURCE_NAME)
    .option(SPARQL_CONFIG_OPTION, json.dumps(sparql_options))
    .load()
)
if sparql_df.rdd.isEmpty():
    print("Endpoint unavailable — adjust the query or check your network.")
else:
    sparql_df.show(5, truncate=False)

5. Build date/time dimensions with rich attributes

from spark_fuse.utils.dataframe import create_date_dataframe, create_time_dataframe

date_dim = create_date_dataframe(spark, "2024-01-01", "2024-01-07")
time_dim = create_time_dataframe(spark, "00:00:00", "23:59:00", interval_seconds=60)

date_dim.select("date", "year", "week", "day_name").show()
time_dim.select("time", "hour", "minute").show(5)

Check out notebooks/demos/date_time_dimensions_demo.ipynb for an interactive walkthrough.

LLM-Powered Column Mapping

from spark_fuse.utils.transformations import map_column_with_llm

standard_values = ["Apple", "Banana", "Cherry"]
mapped_df = map_column_with_llm(
    df,
    column="fruit",
    target_values=standard_values,
    model="o4-mini",
    temperature=None,
)
mapped_df.select("fruit", "fruit_mapped").show()

Set dry_run=True to inspect how many rows already match without spending LLM tokens. Configure your OpenAI or Azure OpenAI credentials with the usual environment variables before running live mappings. Some provider models only accept their default sampling configuration—pass temperature=None to omit the parameter when needed. This helper ships with spark-fuse 0.2.0 and later.

CLI Usage

• spark-fuse --help
• spark-fuse datasources
• spark-fuse read --format rest --path https://pokeapi.co/api/v2/pokemon --config rest.json --show 5

CI

• GitHub Actions runs ruff and pytest for Python 3.9–3.11.

License

• Apache 2.0