Repository navigation
v1.0 – Write-Ahead Logging & Crash Recovery
v1.0 – Write-Ahead Logging
This release introduces persistence and crash recovery
using a write-ahead log (WAL).
Features
- Durable writes via fsync
- Crash recovery by WAL replay
- Single-node, single-threaded design
Known limitations
- No snapshots
- No WAL truncation
- No concurrency
Guarantees (v1)
- A write is considered committed only after WAL fsync
- Crashes after fsync do not lose data
- Crashes before fsync may lose recent writes
- Recovery deterministically rebuilds state by replaying WAL
Observations:
WAL replay amplification: On every restart, WAL grows with the same PUTs again. That is expected and cannot be eliminated safely.
-
WAL grows forever
-
Startup time grows forever
-
Disk usage grows forever
Non-Goals (v1)
- No concurrency
- No snapshots
- No WAL truncation
- No performance optimizations
These are intentionally deferred to later versions.