Current compaction does a single-pass merge of ALL L0+L1 into new L1 files. This works but is not efficient for large datasets.
Possible improvements:
- Selective compaction (only merge overlapping L0/L1 ranges instead of everything)
- Tiered compaction strategies
- Better scheduling to reduce write amplification
Current compaction does a single-pass merge of ALL L0+L1 into new L1 files. This works but is not efficient for large datasets.
Possible improvements: