Repository navigation
Default file DBs to WAL + synchronous=NORMAL (3x to 18x autocommit writes) #13
Description
Measurement
2000 createNode calls on a file DB, autocommit:
| Mode | createNode/s | createEdge x2000 |
|---|---|---|
| default (journal=delete, synchronous=FULL) | 1,225 | 1723ms |
| WAL + synchronous=NORMAL | 26,667 | 132ms |
Inside a transaction both modes are fast (83k to 167k/s), so this matters most for callers who do not batch.
Caveat: measured on an external drive; fsync cost depends on the drive.
Proposal
Default file-backed databases to journal_mode=WAL and synchronous=NORMAL, with an option to opt out ({ wal: false }). No effect on :memory: or readonly DBs.
Benchmark report: 2026-11-02, Apple M2 Pro, Node 22, at a7f9ea7.
Activity
michaeloboyle commentedon Nov 2, 2026
Re-measured, 5 alternating runs of 2000 autocommit createNode each, medians: internal SSD 1,491/s without WAL vs 4,405/s with WAL (~3x, ranges do not overlap); external drive 1,202/s vs 21,277/s (~18x). The ~22x in the description is the external-drive case; expect ~3x on a fast internal SSD.
[-]Default file DBs to WAL + synchronous=NORMAL (~22x autocommit writes)[/-][+]Default file DBs to WAL + synchronous=NORMAL (3x to 18x autocommit writes)[/+]