pub fn with_write_retry<T, F>(
conn: &mut Connection,
max_total: Duration,
f: F,
) -> Result<T>where
F: FnMut(&Transaction<'_>) -> Result<T>,Expand description
Run f against a write transaction, retrying on SQLITE_BUSY/SQLITE_LOCKED
with exponential backoff and jitter. Each attempt uses BEGIN IMMEDIATE so
contention is detected up front rather than mid-transaction. Gives up once
max_total elapses and returns the final busy error.
Why this exists: SQLite’s own busy_timeout retries are unfair under heavy
write contention (a single waiter can spend the full timeout sleeping while
other writers slip past). Jittered backoff at the caller gives fairness and
a bounded worst-case latency suitable for the shell hot path.