On 05.10.2026. 08:06, David Rowley wrote:
> On Fri, 11 Sept 2026 at 08:36, Mario Karuza <mkaruza.pg@icloud.com> wrote:
>> Memory allocated for copied pass-by-reference Datums was not accounted
>> against work_mem because tuplesort_putdatum() passed a hardcoded tuplen
>> of 0 to tuplesort_puttuple_common(). Function free_sort_tuple() adjusts
>> the accounting by the amount actually allocated, so freeing such a
>> tuple subtracts an amount that was never added.
>>
>> This was introduced in 6ed83d5fa55, which switched non-bounded sorts to
>> bump contexts. That commit correctly changed the other tuplesort_put*()
>> functions to compute the size, leaving only this one passing hardcoded
>> 0.
>
> Can you share how you found this issue? e.g., was it a problem you
> found in production because of excessive memory usage? or was it
> human, AI, or other tooling looking at the code?
>
> David
Sure, there was no production issue, it was driven by exploration if AIO
subsystem and if it can be used around spilling. Noticed `still in
progress` from EXPLAIN ANALYZE output, used AI agent to understand
execution path and there was fixed argument that was passed to function.
Commit 6ed83d5fa55 covered other paths but not this one.
--
Regards,
Mario Karuza
