Default SQLite synchronization to FULL

[JanJakes]

Summary

Default SQLite connections to WAL + FULL synchronization.

Explicit sqlite_synchronous overrides remain supported, including NORMAL. The connection tests now cover the new default, overrides, and existing PDO connections whose synchronous setting was previously changed.

Why

Both settings keep a WAL database consistent, but they provide different durability guarantees:

Failure	WAL + NORMAL	WAL + FULL
PHP or application crash	Committed transactions survive.	Committed transactions survive.
OS crash or power loss	Database remains consistent, but recent successful commits may disappear.	Successful commits survive when storage honors synchronization.
Transaction not committed	The transaction rolls back.	The transaction rolls back.

With FULL, SQLite flushes to WAL file after each commit. With NORMAL, synchronization happens only during checkpointing, so a successful commit can still be lost after a power failure or system crash. See the SQLite synchronous documentation and WAL performance considerations.

Benchmark

A local benchmark ran a WordPress-like 90% read / 10% write workload against one database file with 4, 8, and 16 concurrent workers. Results are the combined median of ten 10-second runs across two independent batches, with a two-second warmup per run (PHP 8.5.9, SQLite 3.53.4, pure-PHP parser).

Workers	Configuration	Ops/s (speedup)	p50	p95	p99
4	DELETE + FULL	7,631 (1.0×)	0.100 ms	1.380 ms	10.133 ms
4	WAL + NORMAL	37,640 (4.9×)	0.095 ms	0.115 ms	0.150 ms
4	WAL + FULL	34,883 (4.6×)	0.098 ms	0.145 ms	0.176 ms
8	DELETE + FULL	7,936 (1.0×)	0.101 ms	3.859 ms	21.135 ms
8	WAL + NORMAL	56,103 (7.1×)	0.097 ms	0.126 ms	1.396 ms
8	WAL + FULL	43,874 (5.5×)	0.100 ms	0.157 ms	1.421 ms
16	DELETE + FULL	7,888 (1.0×)	0.103 ms	4.151 ms	38.256 ms
16	WAL + NORMAL	59,372 (7.5×)	0.108 ms	0.145 ms	1.443 ms
16	WAL + FULL	47,829 (6.1×)	0.101 ms	0.169 ms	1.467 ms

WAL + FULL performs worse than WAL + NORMAL, but it's still a significant improvement over the DELETE mode.