sparkwise 1.3.3

Automated Data Engineering specialist for Fabric Spark workloads - intelligent configuration analysis and optimization recommendations

🔥 Sparkwise

Achieve optimal Fabric Spark price-performance with automated insights - simplifies tuning, makes optimization fun

sparkwise is an automated Data Engineering specialist for Apache Spark on Microsoft Fabric. It provides intelligent diagnostics, configuration recommendations, and comprehensive session profiling to help you achieve the best price-performance for your workloads - all while making Spark tuning simple and enjoyable.

🎯 Why sparkwise?

Spark tuning on Microsoft Fabric doesn't have to be complex or expensive. sparkwise helps you:

• 💰 Optimize costs - Detect configurations that waste capacity and increase runtime
• ⚡ Maximize performance - Enable Fabric-specific optimizations (Native Engine, V-Order, resource profiles)
• 🎓 Simplify learning - Interactive Q&A for 133 Spark/Delta/Fabric configurations
• 🔍 Understand workloads - Comprehensive profiling of sessions, executors, jobs, and resources
• ⏱️ Save time - Avoid 3-5min cold-starts by detecting Starter Pool blockers
• 📊 Make data-driven decisions - Priority-ranked recommendations with impact analysis

✨ Key Features

🔬 Automated Diagnostics

• Native Execution Engine - Verifies Velox usage, detects fallbacks to row-based processing
• Spark Compute - Analyzes Starter vs Custom Pool usage, warns about immutable configs
• Data Skew Detection - Identifies imbalanced task distributions
• Delta Optimizations - Checks V-Order, Deletion Vectors, Optimize Write, Auto Compaction
• Runtime Tuning - Validates AQE, partition sizing, scheduler mode

📊 Comprehensive Profiling

• Session Profiling - Application metadata, resource allocation, memory breakdown
• Executor Profiling - Executor status, memory utilization, task distribution
• Job Profiling - Job/stage/task metrics, bottleneck detection
• Resource Profiling - Efficiency scoring, utilization analysis, optimization recommendations

🚀 Advanced Performance Analysis (NEW!)

• Real Metrics Collection - Uses actual Spark stage/task data instead of estimates
• Scalability Prediction - Compare Starter vs Custom Pool with real VCore-hour calculations
• Stage Timeline - Visualize execution patterns with parallel/sequential analysis
• Efficiency Analysis - Quantify wasted compute in VCore-hours with actionable recommendations

🔍 Advanced Skew Detection (NEW!)

• Task Duration Analysis - Detect stragglers and long-running tasks with variance detection
• Partition-Level Analysis - Identify data distribution imbalances with statistical metrics
• Skewed Join Detection - Analyze join patterns and recommend broadcast vs salting strategies
• Automatic Mitigation - Get code examples for salting, AQE, and broadcast optimizations

🎯 SQL Query Plan Analysis (NEW!)

• Anti-Pattern Detection - Identify cartesian products, full scans, and excessive shuffles
• Native Engine Compatibility - Check if queries use Fabric Native Engine (3-8x faster)
• Z-Order Recommendations - Suggest best columns for Delta optimization based on cardinality
• Caching Opportunities - Detect repeated table scans that benefit from caching
• Fabric Best Practices - V-Order, broadcast joins, AQE, and partition recommendations

💡 Interactive Configuration Assistant

• 133 documented configurations - Spark, Delta Lake, Fabric-specific, and Runtime 1.2 configs
• Context-aware guidance - Workload-specific recommendations with impact analysis
• Resource profile support - Understand writeHeavy, readHeavyForSpark, readHeavyForPBI profiles
• Search capabilities - Find configs by keyword or partial name

📈 Priority-Based Recommendations

• Color-coded priorities - Critical (red) → High (yellow) → Medium (blue) → Low (dim)
• Formatted tables - Clean, readable output with impact explanations
• Actionable guidance - Specific commands and configuration values

🚀 Quick Start

Installation

pip install sparkwise

Or install the wheel file directly in Fabric:

%pip install sparkwise-0.1.0-py3-none-any.whl

Basic Usage

from sparkwise import diagnose, ask

# Run comprehensive analysis on current session
diagnose.analyze()

# Ask about any configuration
ask.config('spark.native.enabled')

# Search for configurations
ask.search('optimize')

Session Profiling

from sparkwise import (profile, profile_executors, profile_jobs, profile_resources,
                       predict_scalability, show_timeline, analyze_efficiency)

# Profile complete session
profile()

# Profile executor metrics
profile_executors()

# Analyze job performance
profile_jobs()

# Check resource efficiency
profile_resources()

# Advanced profiling features
predict_scalability()  # Compare pool configurations
show_timeline()        # Visualize stage execution
analyze_efficiency()   # Quantify compute waste

Advanced Analysis

from sparkwise import detect_skew, analyze_query

# Detect data skew
skew_results = detect_skew()  # Analyze task-level skew

# Analyze specific DataFrame for partition skew
from sparkwise.core.advanced_skew_detector import AdvancedSkewDetector
detector = AdvancedSkewDetector()
detector.analyze_partition_skew(your_df, ["key_column"])

# Detect skewed joins
detector.detect_skewed_joins(large_df, small_df, "join_key")

# Analyze SQL query plans
query_results = analyze_query(your_df)

# Get Z-Order recommendations
from sparkwise.core.query_plan_analyzer import QueryPlanAnalyzer
analyzer = QueryPlanAnalyzer()
zorder_cols = analyzer.suggest_zorder_columns(delta_df, ["filtered_col"])

# Detect caching opportunities
analyzer.detect_repeated_subqueries(your_df)

📊 Sample Output

Diagnostic Analysis

🔥 sparkwise Analysis 🔥

🔎 Native Execution Engine
──────────────────────────────────────────────
⚠️ Warning: Native keywords not found in physical plan
   💡 Check for unsupported operators or complex UDFs

⚡ Spark Compute
──────────────────────────────────────────────
✅ Your job uses 1 executors - fits in Starter Pool
   💡 Ensure 'Starter Pool' is selected in workspace settings

💾 Storage & Delta Optimizations
──────────────────────────────────────────────
ℹ️ V-Order is DISABLED (optimal for write-heavy workloads)
   Benefit: 2x faster writes vs V-Order enabled
   💡 Enable only for read-heavy workloads (Power BI/analytics)
      Trade-off: 3-10x faster reads, but 15-20% slower writes

ℹ️ Optimize Write is DISABLED (optimal for writeHeavy profile - default)
   Benefit: Maximum write throughput for ETL and data ingestion
   💡 Enable only for read-heavy or streaming workloads
      - readHeavyForSpark: spark.fabric.resourceProfile=readHeavyForSpark
      - readHeavyForPBI: spark.fabric.resourceProfile=readHeavyForPBI

⚙️ Runtime Tuning
──────────────────────────────────────────────
⛔ CRITICAL: Adaptive Query Execution (AQE) is DISABLED
   💡 Enable immediately: spark.sql.adaptive.enabled=true
      Benefits: Dynamic coalescing, skew joins, better parallelism

📋 Summary of Findings
┌─────────────────────┬────────┬─────────────────┬─────────────────┐
│ Category            │ Status │ Critical Issues │ Recommendations │
├─────────────────────┼────────┼─────────────────┼─────────────────┤
│ Native Execution    │ ⚠️     │ 1               │ 1               │
│ Spark Compute       │ ✅     │ 0               │ 1               │
│ Data Skew           │ ✅     │ 0               │ 0               │
│ Delta               │ ✅     │ 0               │ 3               │
│ Runtime             │ ⚠️     │ 1               │ 2               │
└─────────────────────┴────────┴─────────────────┴─────────────────┘

🔧 Configuration Recommendations
Total recommendations: 7

┌──────────┬─────────────────────────────────┬────────────────┬──────────────┐
│ Priority │ Configuration                   │ Action         │ Impact       │
├──────────┼─────────────────────────────────┼────────────────┼──────────────┤
│ CRITICAL │ spark.sql.adaptive.enabled      │ Set to 'true'  │ Enable       │
│          │                                 │                │ dynamic      │
│          │                                 │                │ partition    │
│          │                                 │                │ coalescing   │
├──────────┼─────────────────────────────────┼────────────────┼──────────────┤
│ MEDIUM   │ spark.sql.parquet.vorder.enabled│ Enable for     │ 3-10x faster │
│          │                                 │ read-heavy     │ reads for    │
│          │                                 │ workloads only │ Power BI     │
└──────────┴─────────────────────────────────┴────────────────┴──────────────┘

✨ Analysis complete!

Interactive Q&A

ask.config('spark.fabric.resourceProfile')

Output:

📚 spark.fabric.resourceProfile

──────────────────────────────────────────────────────────────────────

Default: writeHeavy
Scope: session

What it does:
FABRIC CRITICAL: Selects predefined Spark resource profiles optimized 
for specific workload patterns. Simplifies configuration tuning.

Recommendations for your workload:
  • etl_ingestion: writeHeavy - optimized for ETL and data ingestion
  • analytics_spark: readHeavyForSpark - optimized for analytical queries
  • power_bi: readHeavyForPBI - optimized for Power BI Direct Lake
  • custom_needs: custom - user-defined configuration

Fabric-specific notes:
Microsoft Fabric resource profiles provide workload-optimized settings:

**writeHeavy (DEFAULT):**
- V-Order: DISABLED for faster writes
- Optimize Write: NULL/DISABLED for maximum throughput
- Use Case: ETL pipelines, data ingestion, batch transformations

**readHeavyForSpark:**
- Optimize Write: ENABLED with 128MB bins
- Use Case: Interactive Spark queries, analytical workloads

**readHeavyForPBI:**
- V-Order: ENABLED for Power BI optimization
- Optimize Write: ENABLED with 1GB bins
- Use Case: Power BI dashboards, Direct Lake scenarios

Related configurations:
  • spark.sql.parquet.vorder.enabled
  • spark.databricks.delta.optimizeWrite.enabled
  • spark.microsoft.delta.optimizeWrite.enabled

Examples:
  spark.conf.set('spark.fabric.resourceProfile', 'readHeavyForSpark')
  spark.conf.set('spark.fabric.resourceProfile', 'writeHeavy')

──────────────────────────────────────────────────────────────────────

📦 What's Included

Core Modules

• diagnose - Main diagnostic engine with 5 check categories
• ask - Interactive configuration Q&A system
• profile - Session profiling
• profile_executors - Executor-level metrics
• profile_jobs - Job/stage/task analysis
• profile_resources - Resource efficiency scoring

Knowledge Base (133 Configurations)

• 33 Spark configs - Core settings for shuffle, memory, AQE, serialization
• 45 Delta configs - Delta Lake optimizations, V-Order, Deletion Vectors
• 10 Fabric configs - Native Engine, resource profiles, OneLake storage
• 45 Runtime 1.2 configs - Latest Fabric Runtime 1.2 features

Latest Features

• ✅ Fabric resource profiles (writeHeavy, readHeavyForSpark, readHeavyForPBI)
• ✅ Advanced Delta optimizations (Fast Optimize, Adaptive File Size, File Level Target)
• ✅ Driver Mode Snapshot for faster metadata operations
• ✅ Comprehensive session profiling tools
• ✅ Priority-based recommendation tables
• ✅ Color-coded terminal output with Rich library

🎯 Use Cases

Data Engineers

• Optimize ETL pipelines - Detect bottlenecks, tune parallelism, reduce costs
• Validate configurations - Ensure proper resource profiles and pool usage
• Debug job failures - Understand errors with plain English explanations

Data Scientists

• Improve notebook performance - Enable Native Engine, optimize memory usage
• Understand Spark behavior - Learn configurations through interactive Q&A
• Profile experiments - Track resource usage and efficiency

Platform Admins

• Standardize best practices - Share optimal configurations across teams
• Monitor capacity usage - Identify jobs forcing Custom Pool usage
• Cost optimization - Detect over-provisioned or misconfigured workloads

📚 CLI Usage

# Run diagnostics
sparkwise analyze

# Profile session
sparkwise profile session

# Profile executors
sparkwise profile executors

# Profile jobs
sparkwise profile jobs --max-jobs 5

# Profile resources
sparkwise profile resources

# Analyze bottlenecks
sparkwise profile bottlenecks

# Ask about configuration
sparkwise ask spark.sql.shuffle.partitions

# Search configurations
sparkwise search "adaptive"

🏗️ Architecture

sparkwise/
├── core/
│   ├── advisor.py          # Main diagnostic orchestrator
│   ├── native_check.py     # Velox/Native execution verification
│   ├── pool_check.py       # Starter vs Custom Pool analysis
│   ├── skew_check.py       # Data skew detection
│   ├── delta_check.py      # Delta Lake optimizations
│   └── runtime_check.py    # Runtime configuration tuning
├── profiling/
│   ├── session_profiler.py    # Complete session analysis
│   ├── executor_profiler.py   # Executor metrics
│   ├── job_profiler.py        # Job/stage/task profiling
│   └── resource_profiler.py   # Resource efficiency analysis
├── knowledge_base/
│   ├── spark_configs.yaml     # Core Spark configurations
│   ├── delta_configs.yaml     # Delta Lake configurations
│   ├── fabric_configs.yaml    # Fabric-specific configs
│   └── fabric_runtime_1.2_configs.yaml  # Runtime 1.2 features
├── cli.py                     # Command-line interface
└── config_qa.py              # Interactive Q&A assistant

🎓 Examples

Check out the examples directory:

• basic_analysis.py - Basic diagnostic workflow
• config_qa_demo.py - Configuration Q&A usage
• profiling_demo.py - Comprehensive profiling examples
• knowledge_base_demo.py - Knowledge base exploration
• immutable_configs_demo.py - Starter Pool optimization

🧪 Running Tests

# Install test dependencies
pip install pytest pytest-cov

# Run all tests
pytest

# Run with coverage
pytest --cov=sparkwise --cov-report=html

# Run specific test file
pytest tests/test_advisor.py

🤝 Contributing

Contributions are welcome! Please read our Contributing Guide for details.

Development Setup

# Clone the repository
git clone https://github.com/santhoshravindran7/sparkwise.git
cd sparkwise

# Create virtual environment
python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

# Install in development mode
pip install -e ".[dev]"

# Run tests
pytest

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

Built with ❤️ for the Microsoft Fabric Data Engineering and Data Science community.

📬 Contact & Support

• Author: Santhosh Ravindran
• GitHub: @santhoshravindran7
• Issues: Report bugs or request features

🎉 What's New in v0.1.0

• ✨ Complete profiling suite (session, executor, job, resource profilers)
• 🎨 Rich terminal output with color-coded priorities
• 📊 Priority-based recommendation tables
• 🔧 Fabric resource profile support (writeHeavy, readHeavy profiles)
• ⚡ 4 new advanced Delta optimizations
• 📚 133 documented configurations (up from 100)
• 🎯 Context-aware Optimize Write recommendations
• 🚀 CLI support for all profiling operations

---

Make Spark tuning fun again! 🚀✨