tmorelli@tmorelli.com.br writes:
> Every insert was like this:
> insert into t values (repeat('a',8000));
> Obviously, there must be some kind of compression (the toast table is empty!).
Sure, that's highly compressible data ;-). I repeated your experiment
and then did VACUUM FULL VERBOSE, which provided me this information:
Nonremovable row versions range from 135 to 135 bytes long.
So 8000 'a's compress down into about a hundred bytes, plus the row
header overhead. (Your mileage may vary a bit depending on machine
architecture, PG version, etc.)
> Is there any way of knowing exactly WHEN there will be a new page allocation?
No. Why should you care?
regards, tom lane
