Date: 2026-05-28
Time: 19:00
I'll explain based on the test file itself — it reveals the full public API of the SSTable implementation.
sstable-and-compaction/sstable.py — The implementation being tested: how SSTable file format, block structure, sparse index, and merge logic actually worksstable-and-compaction/sstable.py:merge_sstables — The N-way merge algorithm: how it does a k-way sorted merge with timestamp-based dedup and tombstone removalsstable-and-compaction/testertestsstable.py — Likely a more comprehensive pytest-style test suite for the same module, worth comparing coveragelog-structured-merge-tree/lsm.py — The full LSM tree implementation that uses SSTables as its on-disk layer, showing how memtable flushes produce SSTables and compaction is triggeredleveled-vs-size-tiered-compaction — The two compaction strategies tested here have very different tradeoffs in write amplification, space amplification, and read performance (DDIA Chapter 3)sstable-merge-uses-timestamp-resolution — merge_sstables resolves duplicate keys by keeping the entry with the highest timestamp (last-writer-wins)sstable-tombstone-is-null-value — Tombstones are represented as entries with value=None, not as a separate deletion record typesstable-range-scan-end-exclusive — range_scan(start, end) is inclusive on start and exclusive on endcompaction-manager-supports-two-strategies — CompactionManager implements both 'size_tiered' and 'leveled' compaction strategiesleveled-compaction-promotes-to-l1 — After leveled compaction runs on L0 SSTables, the output has level == 1