CRAPHTFirst compaction strategy thrashed the disk — fsync every 64 writes. Lesson: batch, then fsync.
A tiny LSM-tree storage engine
Can a log-structured merge tree be implemented in ~2k lines and still beat SQLite on write-heavy workloads? Documenting every dead end.
Current question
Is compaction triggering too early under sequential writes?
Journal
Comments (0)
Sign in to join the discussion.
No comments yet.