Thread

Question on io_min_workers > io_max_workers semantics

Xuneng Zhou <xunengzhou@gmail.com> — 2026-05-07T11:50:40Z

Hi Thomas, Hackers,

While looking at d1c01b79d4ae (aio: Adjust I/O worker pool
automatically), I had a question about the intended semantics for
io_min_workers > io_max_workers.

Both GUCs currently have independent 1..MAX_IO_WORKERS range checks,
so a configuration such as:

io_min_workers = 32 io_max_workers = 1

is accepted. In maybe_start_io_workers_scheduled_at(), the
io_max_workers check comes first, so io_max_workers effectively caps
the minimum:

if (io_worker_count >= io_max_workers) return 0;
if (io_worker_count < io_min_workers) return TIMESTAMP_MINUS_INFINITY;

Is that intended? If so, should we document that the effective minimum
is capped by io_max_workers, or perhaps emitting a warning similar to
the autovacuum_max_workers / autovacuum_worker_slots case?

-- 
Best,
Xuneng