Fast writes
Append-first ingestion keeps the foreground path short: log, memtable, then immutable sorted runs.
Product
Core Features
Levelled storage
Immutable sorted runs move through compacted levels, keeping the storage shape predictable over time.
Crash recovery
WAL replay and manifest checkpoints make restart behaviour explicit, testable, and easy to reason about.
Journal
Latest writing
threads
journal/threads/ thread-local-storage-matrix.md
Thread Local Storage Matrix
How Ordinum implemented thread local storage for it's subsystems and the design decisions involved.
product
journal/product/ why-ordinum-exists.md
What is Ordinum?
An introduction to what Ordinum is, an overview of LSM Tree architecture and the purpose of Ordinum.
Timeline
Current progress
Storage core
The first stable shape of the write path and persisted sorted runs.
Memtable write path
Ordered in-memory writes with flush boundaries.
Segment file sketch
Initial sorted-run format and metadata layout.
72
Write-ahead log recovery
Replay rules, truncation checks, and manifest reconciliation.
48
Sparse index format
Index block cadence and seek behaviour for point reads.
Compaction and reads
Turning immutable runs into predictable read behaviour over time.
Level model
Working assumptions for run placement and fanout.
36
Compaction scheduler
Debt tracking, candidate selection, and foreground pressure limits.
64
Range scan iterator
Merged iteration across memtables and segment levels.
Operations
The parts that make the engine understandable under real workloads.
Project journal
Public notes for design decisions and benchmark reports.
28
Benchmark harness
Sustained ingestion, mixed read-write loads, and recovery timing.
18
Metrics surface
Compaction debt, read amplification, segment counts, and disk growth.
Build log