optimize blockchain subsystem scanning with threaded loading and parallel tx fetch

[abdevil05]

Purpose

• Improve performance of blockchain subsystem scanning (MNL and BNS)
• Reduce total scan time during node startup and resync
• Introduce parallelism and better resource utilization

Core Changes

• Refactored load_missing_blocks_into_beldex_subsystems()
• Added parameters:
  • const std::atomic* abort
  • bool use_threaded_load
• Enables controlled execution and optional threaded mode

Threaded Block Loading (Pipeline Execution)

• Loader thread preloads block chunks from DB
• Main thread processes blocks into subsystems
• Introduced block_load_context with bounded queue (MAX_QUEUE_SIZE)
• Used std::mutex and std::condition_variable for synchronization
• Overlaps I/O (block fetching) with computation (processing)
• Reduces idle CPU time and improves throughput

Parallel Transaction Fetching

• Replaced sequential transaction loading with threadpool-based parallel fetching
• Each transaction in a block is fetched concurrently using tools::threadpool
• prevents excessive memory usage and improves stability

Improved Logging and Metrics

• Added detailed performance metrics:
  • blocks per second
  • bytes per second
• timing breakdown (block loading, MNL, BNS)
• Periodic logging (~10 seconds) for better visibility during long scan

Periodic State Persistence

• Introduced periodic MNL state storage (~5 minute interval)
• Reduces risk of losing progress during long-running scan

Performance Impact

• Reduced scanning time due to:
  • parallel transaction fetching
  • pipelined block loading
• Improved CPU utilization and I/O efficiency
• Better scalability for large chains

Testing Performed

• Full blockchain rescan
• Partial rescan from mid-height
• Threaded vs non-threaded execution
• Abort during execution
• High transaction volume scenarios

Summary

• Converts scanning from a sequential bottleneck into a parallel, pipelined system
• Results in faster scans, improved performance, and better reliability