Re: help with design of the 'COUNT(*) in separate table schema

On 20 October 2010 23:52, Dennis Gearon <gearond(at)sbcglobal(dot)net> wrote:
> Regarding the previously discussed triggers on tables to keep track of count:
> http://www.varlena.com/GeneralBits/120.php
> http://www.varlena.com/GeneralBits/49.php
> <from article>
> CREATE OR REPLACE FUNCTION count_rows()
> RETURNS TRIGGER AS
> '
>   BEGIN
>      IF TG_OP = ''INSERT'' THEN
>         UPDATE rowcount
>            SET total_rows = total_rows + 1
>            WHERE table_name = TG_RELNAME;
>      ELSIF TG_OP = ''DELETE'' THEN
>         UPDATE rowcount
>            SET total_rows = total_rows - 1
>            WHERE table_name = TG_RELNAME;
>      END IF;
>      RETURN NULL;
>   END;
> ' LANGUAGE plpgsql;
> </from article>
>
> Wouldn't something like this need row-locking (SELECT for UPDATE) in order to serialize the execution of all triggers?
>

The update will acquire a row level lock on rowcount for the
TG_RELNAME tuple without you doing anything else.

--
Regards,
Peter Geoghegan