On Tue, Oct 20, 2009 at 9:03 PM, Alvaro Herrera
<alvherre@commandprompt.com> wrote:
> Greg Stark escribió:
>> On Mon, Oct 19, 2009 at 8:56 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> > Greg Stark <gsstark@mit.edu> writes:
>> >> The run-time of CLUSTER doesn't vary very much based on whether the
>> >> data is already in index order or not. The number of passes only grows
>> >> like log(n) of the size of your data and if you set
>> >> maintenance_work_mem large enough (somewhere around 100MB-1GB) the
>> >> constants are small enough that you're unlikely to even outgrow a
>> >> single pass (plus a final merge though)
>> >
>> > Uh ... what? It's not based on the sort code, unless someone rewrote it
>> > since I looked last. It's an index scan and will definitely depend on
>> > the index ordering.
>>
>> Er, uh, of course. I wonder what I was thinking.
>
> Your patched version of course.
I would quite happily trade being right for a much faster cluster
command. Also Pizza. I would gladly trade pizza for a faster cluster
command.