Re: [COMMITTERS] pgsql: Teach tuplesort.c about "top N" sorting, in which only the first

Stefan Kaltenbrunner <stefan(at)kaltenbrunner(dot)cc> writes:
> Guillaume Smet wrote:
>> On 5/4/07, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>>> Sort Method: disk Memory: 1000KB Disk: 18482KB
>>
>> +1 for this one.

> I like that one too ...

OK, in the event it looks like one of these four messages:

"Sort Method: top-N heapsort Memory: %ldkB"
"Sort Method: quicksort Memory: %ldkB"
"Sort Method: external sort Disk: %ldkB"
"Sort Method: external merge Disk: %ldkB"

where "external merge" implies that the final merge pass was done
on-the-fly instead of materializing the fully sorted data on disk.
I'm not wedded to these method descriptions if anyone has better phrases
in mind.

Also, I tried to make the disk cases print disk and memory usage both,
but was getting wacko numbers for memory usage. I had forgotten that
the disk-sort path doesn't bother to track memory usage accurately once
it starts returning tuples to the caller (since at that point all the
decisions are made). I'm not really excited about fixing that; it
would add per-tuple overhead for not much value.

regards, tom lane