spark-advisor-cli 0.1.21

AI-powered Apache Spark job analyzer and configuration advisor

↩ spark-advisor

spark-advisor-cli

AI-powered Apache Spark job analyzer and configuration advisor. Standalone CLI tool.

Stop guessing Spark configs. Let data and AI tell you what's wrong.

Install

pip install spark-advisor-cli

Quick Start

# Analyze from event log file (rules-only, free)
spark-advisor analyze /path/to/event-log.json.gz --no-ai

# Analyze with AI recommendations
export SA_ANALYZER_AI__API_KEY=sk-or-...
spark-advisor analyze /path/to/event-log.json.gz

# Analyze from History Server
spark-advisor analyze app-20250101120000-0001 -hs http://yarn:18080

# Agent mode (multi-turn AI analysis)
spark-advisor analyze /path/to/event-log.json.gz --agent

Commands

analyze — analyze a Spark job

# From event log file
spark-advisor analyze /path/to/event-log.json.gz

# From History Server
spark-advisor analyze app-20250101120000-0001 -hs http://yarn:18080

# With AI analysis (default if SA_ANALYZER_AI__API_KEY is set)
spark-advisor analyze /path/to/event-log.json.gz

# Without AI (rules only)
spark-advisor analyze /path/to/event-log.json.gz --no-ai

# Agent mode (multi-turn AI with tool use)
spark-advisor analyze /path/to/event-log.json.gz --agent

# Verbose mode (per-stage breakdown)
spark-advisor analyze /path/to/event-log.json.gz --verbose

# JSON output
spark-advisor analyze /path/to/event-log.json.gz --format json

# Save suggested config to file
spark-advisor analyze /path/to/event-log.json.gz -o spark-defaults.conf

# Use specific LLM model
spark-advisor analyze /path/to/event-log.json.gz --model qwen/qwen3-coder:free

Flag	Short	Default	Description
source		required	App ID (with -hs) or path to event log file
--history-server	-hs	None	Spark History Server URL
--no-ai		False	Disable AI analysis (rules only)
--agent		False	Use agent mode (multi-turn AI with tool use)
--model	-m	qwen/qwen3-coder:free	LLM model for AI analysis
--output	-o	None	Write suggested config to file
--format	-f	text	Output format: text or json
--verbose	-v	False	Show per-stage breakdown

scan — list recent jobs from History Server

spark-advisor scan -hs http://yarn:18080 --limit 20

version

spark-advisor version
# spark-advisor v0.1.10

What it detects

11 deterministic rules: data skew, disk spill, GC pressure, shuffle partitions, executor idle, task failures, small files, broadcast join threshold, serializer choice, dynamic allocation, memory overhead.

All thresholds are configurable via Thresholds model.

See also

• Full documentation and architecture
• MCP Server setup (Claude Desktop / Cursor)
• Rules engine
• Analyzer
• Contributing

License

Apache 2.0