Re: Issue with the PRNG used by Postgres

Andres Freund <andres(at)anarazel(dot)de> writes:
> Hi,
> On 2024-04-10 12:28:10 -0400, Tom Lane wrote:
>> I don't think it's correct to re-initialize the SpinDelayStatus each
>> time around the outer loop. That state should persist through the
>> entire acquire operation, as it does in a regular spinlock acquire.
>> As this stands, it resets the delay to minimum each time around the
>> outer loop, and I bet it is that behavior not the RNG that's to blame
>> for what he's seeing.

> Hm, yea, that's not right. Normally this shouldn't be heavily contended enough
> to matter. I don't think just pulling out the spin delay would be the right
> thing though, because that'd mean we'd initialize it even in the much much
> more common case of there not being any contention. I think we'll have to add
> a separate fetch_or before the outer loop.

Agreed, and I did that in my draft patch. AFAICS we can also avoid
the LWLOCK_STATS overhead if the initial attempt succeeds, not that
that is something you'd really be worried about.

>> (One should still wonder what is the LWLock usage pattern that is
>> causing this spot to become so heavily contended.)

> My suspicion is that it's a4adc31f690 which we only recently backpatched to
> < 16.

Seems like a plausible theory.

regards, tom lane