Counter-Based Compaction for Key-Value Store Tree Data Blocks

The patent implements a compaction operation that identifies and removes obsolete key-value pairs (garbage data) from the tree structure. When a key is updated or deleted, the old version is marked as obsolete but retained in the structure. The compaction process systematically identifies these obsolete entries and removes them, recovering storage space and maintaining search performance by preventing garbage data accumulation.

The patent implements incremental compaction that processes only a portion of the tree structure at each operation rather than performing complete merges. The compaction operation traverses the tree and compacts subsets of nodes, progressively eliminating garbage data through multiple partial operations. This approach maintains search performance by continuously removing obsolete data without requiring full-tree rewrites, thereby reducing total write operations on storage media.

The patent implements a counter-based compaction mechanism where each node in the tree maintains a counter value that automatically tracks the number of obsolete key-value pairs. When the counter reaches a threshold, the node is automatically scheduled for compaction without requiring external intervention or complex decision-making logic. This self-service approach reduces write amplification by triggering compaction only when necessary, while keeping system complexity manageable through simple counter-based decision rules.