fix(server): sqlite commits no longer fsync on the event loop

[alexcstark]

On lax6e the server's event loop stalls for 2–15s several times an hour, and the long ones drop every connected client. Today's 15.6s stall at 17:17 UTC disconnected the desktop.

The stalls were blocked waiting, not computing. The 30s sample around the 15.6s stall shows the loop busy (utilization 0.86) on only about 5s of CPU. I sampled the main thread every 200ms from /proc and caught the next stall (3178 ms logged at 17:49:58). The thread spent about 3.2s in state D inside syscall 74 (fsync), waiting on jbd2_log_wait_commit. That is SQLite's per-commit fsync, waiting on the ext4 journal of a busy, nearly full disk. CPU priority was not the issue: the service already has CPUWeight=1000, and its cgroup's CPU pressure is about 0.1%.

node:sqlite is synchronous and runs on the event loop. We never set synchronous, so SQLite used its default FULL, which fsyncs the WAL on every commit.

Fix

PRAGMA synchronous = NORMAL in layerSetup. This is SQLite's recommended setting for WAL:

• Commits no longer fsync; only checkpoints do.
• The database stays corruption-safe, and an app crash loses nothing.
• Only an OS crash or power loss can roll back the last few commits.

There is a single read-write connection, and the other DatabaseSync openers are read-only, so this one pragma covers every writer.

Verification

• vp test run src/persistence: 60/60 pass.
• On lax6e's disk, under strace -c, 200 WAL commits made 209 fsyncs with FULL and 8 with NORMAL.

🤖 Generated with Claude Code (Opus 5.5)

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

ℹ️ The exact PR base did not have a successful artifact. Baseline uses the latest successful main measurement shown below.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Codex	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Codex	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	20.9 KiB	20.9 KiB	0 B (0.0%)	29.3 KiB	✅
Codex	Live turn messages	2	2	0 (0.0%)	8	✅
Claude	Total thread wire	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Claude	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Claude	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	21.2 KiB	21.2 KiB	0 B (0.0%)	29.3 KiB	✅
Claude	Live turn messages	2	2	0 (0.0%)	8	✅

Baseline: de96f28 · PR result: 3552009 · Source CI: failure

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 108.5 KiB
• Claude decoded thread snapshot: 108.8 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.