Re: increasing sysv semaphore limits

Jason Thorpe <[email protected]> writes:

>> On Jul 16, 2026, at 9:09=E2=80=AFAM, Greg Troxel <[email protected]> wrote:
>>>=20
>>> Are these allocated even if they=E2=80=99re not used?  If so, I would o=
bject to such a change.
>>=20
>> Martin did the math and we're talking another 3760 bytes of memory over
>> the status quo -- which was judged ok with 1993 RAM sizes.
>
> What=E2=80=99s the problem with allocating them as they=E2=80=99re used? =
 If I never use System V IPC, why should I have to burn a page of RAM on th=
em?

To really do it, merely completely rewriting the existing code.  To
partially, perhaps just calling the existing realloc.

For now, I'll let the people who proposed dynamic do that, and just let
pgsql bump up the values when needed at start (which is exactly
allocating what's known to be about to be used, and won't bother people
that don't run pgsql servers).