feat: optimize pruning performance and improve configuration

[dntjd1097]

Summary

This PR introduces several performance optimizations and improvements to the pruning functionality, focusing on memory efficiency, logging, and configuration flexibility. Performance testing shows significant improvements from 1-2 hours to ~5 minutes for a 50GB database.

Changes

1. Optimize Pruning Performance

• Memory efficiency: Use single target height instead of array of heights

  • Previously stored all heights from 1 to latestHeight - versions in memory
  • Now only stores the final prune height, reducing memory overhead significantly
  • Example: For a database with 10,000 blocks keeping 100 versions, reduces from storing 9,899 heights to just 1

• Improved logging: Replace NopLogger with actual logger

  • Enable visibility into store loading, pruning progress, and completion
  • Add structured logging with proper log levels (Info/Error)
  • Better debugging and monitoring capabilities

2. Add IAVL Fast Node Configuration

• New flag: --disable-fast-node (default: false)

  • When not set (default): Fast node is enabled for better query performance
  • When set: Fast node is disabled for faster pruning operations

• Rationale: This is a pruning tool, not a database upgrade tool

  • Users should have the flexibility to choose their pruning strategy
  • Some users prioritize speed over query capabilities during maintenance windows
  • Enables faster pruning for users who need to quickly reclaim disk space

• Simplified logic: Direct pass-through of flag value to SetIAVLDisableFastNode

  • Removes confusing inverted if-else branches
  • More maintainable and easier to understand

3. Enhanced Error Handling

• Add structured fields to error logs (e.g., base, target for block pruning)
• More descriptive error messages for better troubleshooting

Performance Testing

Test Environment

Hardware	Specs
Server	AMD Ryzen 7 9700X 8-Core Processor, DDR5 64GB RAM, 1TB NVMe
Laptop	Apple M1 Pro 10-Core, 32GB RAM

Test Results

Metric	Before	After	Improvement
Execution Time	1-2 hours	~5 minutes	~95% faster
Memory Usage	High (stores all heights)	Low (single height)	Significant reduction
Database Size	50GB	50GB	Same test dataset

Key Improvements

• ✅ Dramatic time reduction: From hours to minutes
• ✅ Lower memory footprint: Single height vs. array of thousands
• ✅ Better observability: Real-time logging of pruning progress
• ✅ Flexible configuration: Optional fast node control

Limitations

• Large databases (>100GB) have not been tested yet
• Performance may vary based on disk I/O and database fragmentation

Testing

• ✅ Tested with Osmosis node data (50GB database)
• ✅ Verified on both AMD and ARM (M1) architectures
• ✅ Confirmed logging output shows proper pruning progress
• ✅ Memory usage validated with single-height approach
• ✅ Both --disable-fast-node flag behaviors tested

Usage Examples

# Default: Fast node enabled (better for nodes that need queries)
cosmprund prune /path/to/data --versions 100

# Fast pruning: Fast node disabled (faster pruning, no query support during process)
cosmprund prune /path/to/data --versions 100 --disable-fast-node

Breaking Changes

None. All changes are backward compatible.

Related Issues

N/A