On Fri, Apr 23, 2010 at 10:31 PM, Eliot Gable
<egable+pgsql-performance@gmail.com> wrote:
> In each case, the results are available outside the stored procedure by
> either fetching from the cursor or selecting from the temporary table.
> Clearly, the temporary table takes a performance hit compared using arrays.
> Building an array with array append is horrendously inefficient. Unnesting
> an array is surprisingly efficient. As can be seen from Test3 and Test4,
> cursors have no detectable overhead for opening the cursor (at least in this
> example with 1000 result rows). It is unclear whether there is any
> difference at all from Test3 and Test4 for retrieving the data as I have no
> easy way right now to measure that accurately. However, since arrays+cursors
> are more efficient than anything having to do with temp tables, that is the
> way I will go. With the number of rows I am dealing with (which should
> always be less than 1,000 in the final returned result set), unnesting an
> array is much faster than building a temp table and selecting from it.
> If anyone thinks I may have missed some important item in this testing,
> please let me know.
Well, you missed the most important part: not using cursors at all.
Instead of declaring a cursor and looping it to build the array, build
it with array(). That's what I've been saying: arrays can completely
displace both temp tables _and_ cursors when passing small sets around
functions.
merlin