Re: cursor_to_xml iteration of a table

From: Peter Eisentraut <peter_e(at)gmx(dot)net>
To: Richard Wallace <rwallace(at)intellum(dot)com>
Cc: pgsql-general(at)postgresql(dot)org
Subject: Re: cursor_to_xml iteration of a table
Date: 2010-05-26 20:12:31
Message-ID: 1274904751.19408.14.camel@vanquo.pezone.net
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-general

On tis, 2010-05-25 at 12:05 -0400, Richard Wallace wrote:
> 1) When using cursor_to_xml in a plpgsql function, the FOUND variable does not seem to get set, so there is no way to exit a loop that is iterating over the cursor. Below is the function code; it loops indefinitely when it is run.
>
> create or replace function getxml() returns setof xml as $$
> declare
> resultxml xml;
> curs refcursor;
> begin
> open curs for select * from groups;
> loop
> select cursor_to_xml(curs,1000, false, false, '') into resultxml;
> return next resultxml;
> exit when not found;
> end loop;
> end;
> $$ language plpgsql;

Yeah, there doesn't seem to be a good way out of that. When the end of
the cursor is reached, cursor_to_xml returns an empty xml value (which
is probably bogus in itself, since that is not a valid xml value to
begin with), so you could test it like this:

exit when resultxml::text = '';

> 2) Assuming the above issue is fixed, how can I go about ensuring that the result set from the function isn't stored in memory until the function completes? Ideally, I'd like to write the output to a file with each iteration of the cursor, but I know file IO is a big no-no with plpgsql since transactions can't manage the state of files being written.

cursor_to_xml is more meant to be used from a client. If you do it like
in the above function, you will indeed build the result in memory
(multiple times, perhaps).

In response to

Browse pgsql-general by date

  From Date Subject
Next Message John Gage 2010-05-26 20:29:58 110,000,000 rows
Previous Message Gauthier, Dave 2010-05-26 19:14:37 Re: Does update = delete + insert ?