fabricengineer-py 0.1.0

Package for Fabric Engineers

FabricEngineer Package

Description

FabricEngineer is a comprehensive Python package designed specifically for Microsoft Fabric developers to streamline data transformation workflows and automate complex ETL processes. This package provides enterprise-grade solutions for building robust data pipelines with minimal boilerplate code.

Key Features

🚀 Silver Layer Data Ingestion Services

• Insert-Only Pattern: Efficient data ingestion with support for schema evolution and historization
• SCD Type 2 (Slowly Changing Dimensions): Complete implementation of Type 2 SCD with automatic history tracking
• Delta Load Support: Optimized incremental data processing with broadcast join capabilities
• Schema Evolution: Automatic handling of schema changes with backward compatibility

📊 Materialized Lake Views (MLV)

• Automated MLV Generation: Create and manage materialized views with SQL generation
• Schema-aware Operations: Intelligent handling of schema changes and column evolution
• Lakehouse Integration: Seamless integration with Microsoft Fabric Lakehouse architecture

🔧 Advanced Data Engineering Features

• Configurable Transformations: Flexible transformation pipelines with custom business logic
• Data Quality Controls: Built-in validation and data quality checks
• Performance Optimization: Broadcast joins, partition strategies, and optimized query patterns
• Comprehensive Logging: Integrated logging and performance monitoring with TimeLogger

Installation

pip install fabricengineer-py

Quick Start Guide

Prerequisites

• Microsoft Fabric workspace with Lakehouse
• PySpark environment
• Python 3.11+

Usage Examples

Silver Layer Data Ingestion

Insert-Only Pattern

The Insert-Only service is ideal for append-only scenarios where you need to track all changes while maintaining performance.

from pyspark.sql import DataFrame, functions as F
from fabricengineer.logging import TimeLogger
from fabricengineer.transform.lakehouse import LakehouseTable
from fabricengineer.transform import SilverIngesationInsertOnly


def transform_projects(df: DataFrame, etl) -> DataFrame:
    df = df.withColumn("dtime", F.to_timestamp("dtime"))
    return df


def transform_all(df: DataFrame, etl) -> DataFrame:
    df = df.withColumn("data", F.lit("values"))
    return df


# Initialize performance monitoring
timer = TimeLogger()

# Define table-specific transformations
transformations = {
    "*": transform_all,             # Applied to all tables
    "projects": transform_projects  # Applied only to projects table
}

# Configure source and destination tables
source_table = LakehouseTable(
    lakehouse="BronzeLakehouse",
    schema="schema",
    table="projects"
)
destination_table = LakehouseTable(
    lakehouse="SilverLakehouse",
    schema=source_table.schema,
    table=source_table.table
)

# Initialize and configure the ETL service
etl = SilverIngestionInsertOnly()
etl.init(
    spark_=spark,
    notebookutils_=notebookutils,
    source_table=source_table,
    destination_table=destination_table,
    nk_columns=NK_COLUMNS,
    constant_columns=CONSTANT_COLUMNS,
    is_delta_load=IS_DELTA_LOAD,
    delta_load_use_broadcast=DELTA_LOAD_USE_BROADCAST,
    transformations=transformations,
    exclude_comparing_columns=EXCLUDE_COLUMNS_FROM_COMPARING,
    include_comparing_columns=INCLUDE_COLUMNS_AT_COMPARING,
    historize=HISTORIZE,
    partition_by_columns=PARTITION_BY_COLUMNS,
    df_bronze=None,
    create_historized_mlv=True
)


timer.start().log()
etl.run()
timer.end().log()

SCD Type 2 (Slowly Changing Dimensions)

The SCD2 service implements Type 2 Slowly Changing Dimensions with automatic history tracking and current record management.

from pyspark.sql import DataFrame, functions as F
from fabricengineer.logging import TimeLogger
from fabricengineer.transform.lakehouse import LakehouseTable
from fabricengineer.transform import SilverIngestionSCD2Service


def transform_projects(df: DataFrame, etl) -> DataFrame:
    df = df.withColumn("dtime", F.to_timestamp("dtime"))
    return df


def transform_all(df: DataFrame, etl) -> DataFrame:
    df = df.withColumn("data", F.lit("values"))
    return df


# Initialize performance monitoring
timer = TimeLogger()

# Define table-specific transformations
transformations = {
    "*": transform_all,             # Applied to all tables
    "projects": transform_projects  # Applied only to projects table
}

# Configure source and destination tables
source_table = LakehouseTable(
    lakehouse="BronzeLakehouse",
    schema="schema",
    table="projects"
)
destination_table = LakehouseTable(
    lakehouse="SilverLakehouse",
    schema=source_table.schema,
    table=source_table.table
)

# Initialize and configure the ETL service
etl = SilverIngestionSCD2Service()
etl.init(
    spark_=spark,
    notebookutils_=notebookutils,
    source_table=source_table,
    destination_table=destination_table,
    nk_columns=NK_COLUMNS,
    constant_columns=CONSTANT_COLUMNS,
    is_delta_load=IS_DELTA_LOAD,
    delta_load_use_broadcast=DELTA_LOAD_USE_BROADCAST,
    transformations=transformations,
    exclude_comparing_columns=EXCLUDE_COLUMNS_FROM_COMPARING,
    include_comparing_columns=INCLUDE_COLUMNS_AT_COMPARING,
    historize=HISTORIZE,
    partition_by_columns=PARTITION_BY_COLUMNS,
    df_bronze=None
)


timer.start().log()
etl.run()
timer.end().log()

Materialized Lake Views Management

Prerequisites

Configure a Utils Lakehouse as your default Lakehouse. The generated view SQL code will be saved as .sql.txt files in the lakehouse under /Files/mlv/{lakehouse}/{schema}/{table}.sql.txt.

from fabricengineer.mlv import MaterializeLakeView

# Initialize the Materialized Lake View manager
mlv = MaterializedLakeView(
    lakehouse="SilverBusinessLakehouse",
    schema="schema",
    table="projects"
)
print(mlv.to_dict())

# Define your custom SQL query
sql = """
SELECT
    p.id
    ,p.projectname
    ,p.budget
    ,u.name AS projectlead
FROM dbo.projects p
LEFT JOIN users u
ON p.projectlead_id = u.id
"""

# Create or replace the materialized view
result = mlv.create_or_replace(sql)
display(result)

Remote Module Import for Fabric Notebooks

Import specific package modules directly into your Fabric notebooks from GitHub releases:

# Cell 1:
import requests

VERSION = "0.1.0"
url = f"https://raw.githubusercontent.com/enricogoerlitz/fabricengineer-py/refs/tags/{VERSION}/src/fabricengineer/import_module/import_module.py"
resp = requests.get(module_path)
code = code[1].strip()

exec(code, globals())  # This provides the 'import_module' function
print(code)

# Cell 2
mlv_module = import_module("transform.silver.mlv")
scd2_module = import_module("transform.silver.sdc2", VERSION)
insertonly_module = import_module("transform.silver.insertonly", VERSION)

# Cell 3 - Use mlv module
exec(mlv_module, globals())  # Provides MaterializedLakeView class and mlv instance

mlv.init(
    lakehouse="SilverBusinessLakehouse",
    schema="schema",
    table="projects"
)
print(mlv.to_dict())

# Cell 4 - Use scd2 module
exec(scd2_module, globals())  # Provides an instantiated etl object

etl.init(...)
print(str(etl))

# Cell 5 - Use insertonly module
exec(insertonly_module, globals())  # Provides an instantiated etl object

etl.init(...)
print(str(etl))

Advanced Features

Performance Optimization

• Broadcast Joins: Automatically optimize small table joins
• Partition Strategies: Intelligent partitioning for better query performance
• Schema Evolution: Handle schema changes without breaking existing pipelines
• Delta Load Processing: Efficient incremental data processing

Data Quality & Validation

• Automatic Validation: Built-in checks for data consistency and quality
• Type Safety: Comprehensive type annotations for better development experience
• Error Handling: Robust error handling and recovery mechanisms

Monitoring & Logging

from fabricengineer.logging import TimeLogger, logger

# Performance monitoring
timer = TimeLogger()
timer.start().log()

# Your ETL operations here
etl.run()

timer.end().log()

# Custom fabricengineer logging
logger.info("Custom log message")
logger.error("Error occurred during processing")