Expand description
The DatabaseQueue driver — jobs that survive a restart.
Rows in two tables, defined as ordinary Rainier ORM entities, so the queue runs on whatever backend the application already uses (SQLite, D1, MySQL, Postgres) with no extra infrastructure.
§Reserving without losing or duplicating jobs
Two workers polling the same table will see the same row. A plain
SELECT … LIMIT 1 followed by UPDATE would hand it to both.
The fix here is an optimistic claim: select a candidate, then update it
with WHERE id = ? AND reserved_at IS NULL and check the affected-row
count. Exactly one worker’s update touches a row; the losers see zero and
try the next candidate. That needs no SELECT … FOR UPDATE, no advisory
lock, and no dialect-specific syntax — which matters because D1 has none of
them.
A worker that dies mid-job leaves a row reserved forever, so a reservation
also carries a deadline: reclaim_expired
releases anything held past it.
Structs§
- Database
Queue - A queue backed by database rows.
- Failed
JobRow - A job that exhausted its attempts, as a row.
- JobRow
- A queued job, as a row.