Re: Potential "AIO / io workers" inter-worker locking issue in PG18?

Hi,

On 2025-10-05 22:22:32 +0200, Marco Boeringa wrote:
> I realized that my observation of the pg_aios view being empty was likely
> with the "io_method = sync" option set, which I guess doesn't use or fill
> the pg_aios view? Can you confirm the pg_aios view is unused with "io_method
> = sync", this aspect is not documented in the PostgreSQL help? Anyway, I
> will need to re-test with 'worker' set.

pg_aios is populated even with io_method = sync, albeit with at most one entry
entry per backend.

If there were no entries in pg_aios at the time of your hang, it doesn't seem
likely - although not impossible - for AIO to be responsible.

Greetings,

Andres Freund