Re: Lazy allocation of pages required for verifying FPI consistency

On Mon, Jan 16, 2023 at 10:52:43AM +0900, Kyotaro Horiguchi wrote:
> Yeah, we could do that. But as I mentioned before, that happens only
> on startup thus it can be said that that's not worth bothering.  On
> the other hand I don't think it's great to waste 16kB * max_backends
> memory especially when it is clearly recognized and easily avoidable.

Memory's cheap, but basically nobody would use these except
developers..

> I guess the reason for the code is more or less that.

The original discussion spreads across these threads:
https://www.postgresql.org/message-id/CAB7nPqR0jzhF%3DU4AXLm%2BcmaE4J-HkUzbaRXtg%2B6ieERTqr%3Dpcg%40mail.gmail.com
https://www.postgresql.org/message-id/CAGz5QC%2B_CNcDJkkmDyPg2zJ_R8AtEg1KyYADbU6B673RaTySAg%40mail.gmail.com

There was a specific point about using static buffers from me, though
these would not have been aligned as of the lack of PGAlignedBlock
back in 2017 which is why palloc() was used.  That should be around
here:
https://www.postgresql.org/message-id/CAB7nPqR=OcojLCP=1Ho6Zo312CKzUZU8d4aJO+VvpUYV-waU_Q@mail.gmail.com
--
Michael