compactions

Compactions are processes that reorganize data or memory to reclaim space, reduce fragmentation, and improve access locality. The term is used across several computing domains, including storage systems, databases, file systems, and memory management, where ongoing changes can leave behind outdated or scattered data.

In storage systems that use log-structured merge trees, such as LevelDB, RocksDB, and many NoSQL databases, compaction is the process of reading and merging multiple smaller data files into fewer larger ones. During compaction, obsolete versions and deleted records are discarded, which reduces read amplification and space usage but incurs write amplification and additional I/O. Compaction strategies vary, with size-tiered and leveled designs aiming to balance write throughput, read performance, and space utilization.

In garbage collection, compaction is a phase that relocates live objects to eliminate gaps created by allocation and reclamation. This reduces memory fragmentation and improves locality of reference, but often requires pausing program execution. Modern collectors may implement incremental or concurrent compaction to reduce latency while still reclaiming fragmented memory.

In file systems and storage architectures, compaction can refer to reclaiming space in log-structured or sparse layouts, or to defragmentation-like operations that consolidate free space to improve future allocation efficiency. Some databases and file systems perform compaction as a background task essential to maintaining performance and storage health.

Trade-offs for compaction include increased CPU and I/O activity and potential pauses, which can affect latency. Effective compaction design depends on workload characteristics such as write rate, read patterns, and tolerance for interruption.