Re: PoC: Partial sort

On 12/22/2013 04:38 PM, Alexander Korotkov wrote:

postgres=# explain analyze select * from test order by v1, id limit 10;
                                                                   QUERY
PLAN
-----------------------------------------------------------------------------------------------------------------------------------------------
  Limit  (cost=11441.77..11442.18 rows=10 width=12) (actual
time=79.980..79.982 rows=10 loops=1)
    ->  Partial sort  (cost=11441.77..53140.44 rows=1000000 width=12)
(actual time=79.978..79.978 rows=10 loops=1)
          Sort Key: v1, id
          Presorted Key: v1
          Sort Method: top-N heapsort  Memory: 25kB
          ->  Index Scan using test_v1_idx on test  (cost=0.42..47038.83
rows=1000000 width=12) (actual time=0.031..38.275 rows=100213 loops=1)
  Total runtime: 81.786 ms
(7 rows)

Have you thought about how do you plan to print which sort method and how much memory was used? Several different sort methods may have been use in the query. Should the largest amount of memory/disk be printed?

However, work with joins needs more improvements.

That would be really nice to have, but the patch seems useful even without the improvements to joins.