PG loads data at the block level to shared_buffers. Most likely it is
because the second sql selects different set of rows (from different
blocks) than the first sql.
On Fri, Mar 18, 2016 at 4:24 PM, Paul Jones <pbj@cmicdo.com> wrote:
> In Postgres 9.5.1 with a shared_buffer cache of 7Gb, a SELECT from
> a single table that uses an index appears to read the table into the
> shared_buffer cache. Then, as many times as the exact same SELECT is
> repeated in the same session, it runs blazingly fast and doesn't even
> touch the disk. All good.
>
> Now, in the *same* session, if a different SELECT from the *same* table,
> using the *same* index is run, it appears to read the entire table from
> disk again.
>
> Why is this? Is there something about the query that qualifies the
> contents of the share_buffer cache? Would this act differently for
> different kinds of indexes?
>
> PJ
>
>
> --
> Sent via pgsql-general mailing list (pgsql-general@postgresql.org)
> To make changes to your subscription:
> http://www.postgresql.org/mailpref/pgsql-general