Re: RAM-only temporary tables

Kevin Grittner wrote:

> Note that the 150 disk writes were for the CREATE and the DROP. Does
> that mean that we'd actually shave 84 of 150 writes?

Hmm, you'd shave more than 42 but not 84, because index entries are not
deleted until a later vacuum. (I'd say about 56 -- 42 plus the 14 heap
deletions).

> Also, if you're looking to account for all the writes, it's worth
> noting that my test declared a one-column primary key (on an integer
> column) in the CREATE TEMPORARY TABLE statement.

That probably makes up for the extra few writes that I didn't see in my
quick test.

> In suggesting this enhancement, my hope is that each session could
> check for a referenced table as a temporary in RAM before going to the
> system tables, in a manner vaguely similar to how space reserved by
> the temp_buffers GUC is used for temp table data. I, of course, am
> suggesting this from a position of blissful ignorance of the actual
> complexity of making such a change.

Right -- I don't expect we can make use of such an idea readily. Not
creating unnecessary pg_attribute entries for system columns is probably
a lot easier to do. The idea of uncatalogued temp tables has been
suggested and rejected several times in the past.

--
Alvaro Herrera http://www.CommandPrompt.com/
The PostgreSQL Company - Command Prompt, Inc.