Apr 15, 2026 · Accepted

Experiment 044: SQLITE_ENABLE_BATCH_ATOMIC_WRITE

Date: 2026-04-15

Status: Accepted

Problem

On Android devices running F2FS (the default filesystem since Android 9), SQLite writes both journal + database pages (2x I/O) for every transaction. F2FS supports batch atomic writes, which allows SQLite to eliminate the rollback journal entirely — writing only the database pages in a single atomic batch.

Hypothesis

Enabling SQLITE_ENABLE_BATCH_ATOMIC_WRITE at compile time activates F2FS detection at runtime. When the filesystem reports SQLITE_IOCAP_BATCH_ATOMIC, SQLite skips journal writes, yielding 2-3x write throughput improvement on Android. The flag is inert on filesystems that don't support batch atomics (macOS APFS, Linux ext4, Windows NTFS).

Approach

Single define in hook/build.dart:

 defines: { 'SQLITE_ENABLE_BATCH_ATOMIC_WRITE': null, ... } 

Research references:

• SQLite F2FS documentation
• ACM paper on multi-block atomic write: 487% insert throughput increase on F2FS vs ext4 PERSIST mode, 67% write volume decrease.

Results

Verified: 0 regressions on macOS. As expected, the flag has no runtime effect on non-F2FS filesystems. The 21 "wins" in the benchmark are run-to-run variance — the flag does not activate on APFS.

The expected impact on Android F2FS (based on published research):

• 2-3x write throughput for transactions
• ~50% reduction in write amplification (no journal I/O)
• Reduced SSD wear from eliminated journal writes

Decision

Accepted. Zero-risk compile flag that enables a significant write optimization on Android F2FS. No effect on other platforms. The optimization activates automatically via VFS capability detection — no API changes, no user configuration needed.