On Fri, Jul 11, 2025 at 3:48 PM Greg Burd wrote:
> I briefly considered how one might use what was left after surgery to produce
> some similar boolean signal to no avail. I think that autoprewarm was simply
> trying to at most warm NBuffers then stop. The freelist at startup was just
> a con
On Fri, Sep 5, 2025, at 12:27 PM, Andres Freund wrote:
> Hi,
>
> On 2025-09-04 13:24:00 -0400, Greg Burd wrote:
>> On Thu, Sep 4, 2025, at 12:59 PM, Andres Freund wrote:
>> > Hi,
>> >
>> > On 2025-08-27 15:42:48 -0400, Greg Burd wrote:
>> >> Regardless, I feel the first two patches on this set ad
Hi,
On 2025-09-04 13:24:00 -0400, Greg Burd wrote:
> On Thu, Sep 4, 2025, at 12:59 PM, Andres Freund wrote:
> > Hi,
> >
> > On 2025-08-27 15:42:48 -0400, Greg Burd wrote:
> >> Regardless, I feel the first two patches on this set address the
> >> intention of this thread.
> >
> > I'm planning to co
On Thu, Sep 4, 2025, at 12:59 PM, Andres Freund wrote:
> Hi,
>
> On 2025-08-27 15:42:48 -0400, Greg Burd wrote:
>> Regardless, I feel the first two patches on this set address the
>> intention of this thread.
>
> I'm planning to commit the first two patches after making a pass through
> them. I h
Hi,
On 2025-08-27 15:42:48 -0400, Greg Burd wrote:
> Regardless, I feel the first two patches on this set address the
> intention of this thread.
I'm planning to commit the first two patches after making a pass through
them. I have some work that I'm cleaning up to post that'd conflict and Tomas'
On Aug 17 2025, at 12:57 am, Thomas Munro wrote:
> On Sun, Aug 17, 2025 at 4:34 PM Thomas Munro wrote:
>> Or if you don't like those odds, maybe it'd be OK to keep % but use it
>> rarely and without the CAS that can fail.
>
> ... or if we wanted to try harder to avoid %, could we relegate it
On Sun, Aug 17, 2025 at 4:34 PM Thomas Munro wrote:
> Or if you don't like those odds, maybe it'd be OK to keep % but use it
> rarely and without the CAS that can fail.
... or if we wanted to try harder to avoid %, could we relegate it to
the unlikely CLOCK-went-all-the-way-around-again-due-to-un
On Sat, Aug 16, 2025 at 3:37 PM Thomas Munro wrote:
> while (hand >= NBuffers)
> {
> /* Base value advanced by backend that overshoots by one tick. */
> if (hand == NBuffers)
> pg_atomic_fetch_add_u64(&StrategyControl->ticks_base, NBuffers);
> hand -= NB
On Wed, Aug 13, 2025 at 9:42 AM Greg Burd wrote:
> Amazing, thank you. I'll try to replicate your tests tomorrow to see if
> my optimized division and modulo functions do in fact help or not. I
> realize that both you and Anders are (rightly) concerned that the
> performance impact of IDIV on so
On Aug 11 2025, at 9:09 am, Tomas Vondra wrote:
> Hi,
>
> I took a look at this thread / patches, mostly because it's somewhat
> related to the NUMA patch series [1]. Freelist is one of the things the
> NUMA patches tweak, to split it into per-node partitions. So removing
> the freelist would
On 7/25/25 15:02, Greg Burd wrote:
> Patch set is now:
>
> 1) remove freelist
>
> 2) remove buffer_strategy_lock
>
> 3) abstract clock-sweep to type and API
>
>
>
> -greg
Somehow including the test.c file as an attachment on my last email
confused the CI and it didn't test the v10 patch set (which
On 7/22/25 14:43, Greg Burd wrote:
> On 7/21/25 14:35, Andres Freund wrote:
>> Hi,
>>
>> On 2025-07-21 13:37:04 -0400, Greg Burd wrote:
>>> On 7/18/25 13:03, Andres Freund wrote:
>>> Hello. Thanks again for taking the time to review the email and patch,
>>> I think we're onto something good here.
On 7/21/25 14:35, Andres Freund wrote:
> Hi,
>
> On 2025-07-21 13:37:04 -0400, Greg Burd wrote:
>> On 7/18/25 13:03, Andres Freund wrote:
>> Hello. Thanks again for taking the time to review the email and patch,
>> I think we're onto something good here.
>>
>>> I'd be curious if anybody wants to a
Hi,
On 2025-07-21 13:37:04 -0400, Greg Burd wrote:
> On 7/18/25 13:03, Andres Freund wrote:
> Hello. Thanks again for taking the time to review the email and patch,
> I think we're onto something good here.
>
> >
> > I'd be curious if anybody wants to argue for keeping the clock sweep. Except
>
On 7/18/25 13:03, Andres Freund wrote:
> Hi,
Hello. Thanks again for taking the time to review the email and patch,
I think we're onto something good here.
>
> I'd be curious if anybody wants to argue for keeping the clock sweep. Except
> for the have_free_buffer() use in autoprewarm, it's a rat
Hi,
I'd be curious if anybody wants to argue for keeping the clock sweep. Except
for the have_free_buffer() use in autoprewarm, it's a rather trivial
patch. And I really couldn't measure regressions above the noise level, even
if absurdly extreme use cases.
On 2025-07-17 14:35:13 -0400, Greg Bur
On Fri, Jul 11, 2025, at 2:52 PM, Andres Freund wrote:
Hi,
On 2025-07-11 13:26:53 -0400, Greg Burd wrote:
In conversations [1] recently about considering how best to adapt the
code
to become NUMA-aware Andres commented, "FWIW, I've started to wonder
if we
shouldn't just get rid of the freelis
On Fri, Jul 11, 2025, at 2:50 PM, Nathan Bossart wrote:
> On Fri, Jul 11, 2025 at 01:26:53PM -0400, Greg Burd wrote:
>> This change does remove the have_free_buffer() function used by the
>> contrib/pg_prewarm module. On the surface this doesn't seem to cause any
>> issues, but honestly I've not
Hi,
On 2025-07-11 13:26:53 -0400, Greg Burd wrote:
> In conversations [1] recently about considering how best to adapt the code
> to become NUMA-aware Andres commented, "FWIW, I've started to wonder if we
> shouldn't just get rid of the freelist entirely" and because I'm a glutton
> for punishment
On Fri, Jul 11, 2025 at 01:26:53PM -0400, Greg Burd wrote:
> This change does remove the have_free_buffer() function used by the
> contrib/pg_prewarm module. On the surface this doesn't seem to cause any
> issues, but honestly I've not thought too deeply on this one.
Hm. ISTM we'll either need t
Hello,
In conversations [1] recently about considering how best to adapt the code to
become NUMA-aware Andres commented, "FWIW, I've started to wonder if we
shouldn't just get rid of the freelist entirely" and because I'm a glutton for
punishment (and I think this idea has some merit) I took hi
21 matches
Mail list logo
