On Wed, Oct 14, 2020 at 5:10 PM Tom Lane <tgl@xxxxxxxxxxxxx> wrote:
Magnus Hagander <magnus@xxxxxxxxxxxx> writes:
> On Wed, Oct 14, 2020 at 3:57 PM Thomas Kellerer <shammat@xxxxxxx> wrote:
>> select pg_size_bytes(setting) * 8192
>> from pg_settings
>> where name = 'shared_buffers';
> Actually, it doesn't have to be in 8k pages, that depends on the build
> options. So if you want to be perfectly correct, you should probably
> multiply with current_setting('block_size') instead of a hardcoded 8192 :)
It's fairly annoying that this doesn't work:
regression=# select pg_size_bytes(setting||' '||unit) from pg_settings where name = 'shared_buffers';
ERROR: invalid size: "16384 8kB"
DETAIL: Invalid size unit: "8kB".
HINT: Valid units are "bytes", "kB", "MB", "GB", and "TB".
Maybe we should teach pg_size_bytes to cope with that.
Actually thinking though, surely *this* particular case can be spelled as:
SELECT pg_size_bytes(current_setting('shared_buffers'))
Or if doing it off pg_settings:
SELECT setting::bigint * pg_size_bytes(unit) from pg_settings where name='shared_buffers'
I'm not sure having pg_size_bytes() parse "16384 8kB" is reasonable, I have a feeling that could lead to a lot of accidental entries giving the wrong results.
