Re: Extension Enhancement: Buffer Invalidation in pg_buffercache

Thomas Munro <thomas.munro@gmail.com>

From: Thomas Munro <thomas.munro@gmail.com>

To: Maksim Milyutin <milyutinma@gmail.com>

Cc: Maxim Orlov <orlovmg@gmail.com>, vignesh C <vignesh21@gmail.com>, Palak Chaturvedi <chaturvedipalak1911@gmail.com>, Jim Nasby <jim.nasby@gmail.com>, pgsql-hackers@postgresql.org, Nitin Jadhav <nitinjadhavpostgres@gmail.com>

Date: 2024-04-29T20:59:15Z

Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

Add pg_buffercache_evict() function for testing.
- 13453eedd3f6 17.0 landed

On Tue, Apr 30, 2024 at 7:17 AM Thomas Munro <thomas.munro@gmail.com> wrote:
> On Tue, Apr 30, 2024 at 6:47 AM Maksim Milyutin <milyutinma@gmail.com> wrote:
> >> Should not we call at the end the StrategyFreeBuffer() function to add target buffer to freelist and not miss it after invalidation?
>
> > Please take a look at this issue, current implementation of EvictUnpinnedBuffer() IMO is erroneous - evicted buffers are lost permanently and will not be reused again

I don't think that's true: it is not lost permanently, it'll be found
by the regular clock hand.  Perhaps it should be put on the freelist
so it can be found again quickly, but I'm not sure that's a bug, is
it?  If it were true, even basic testing eg select
count(pg_buffercache_evict(bufferid)) from pg_buffercache would leave
the system non-functional, but it doesn't, the usual CLOCK algorithm
just does its thing.