Question

我有一个postgres数据库，有一个表（大约150 000行），有三列; row_id，花的位置和数量。由于位置相互靠近可能会在计数中出现一些错误，我想知道每个位置的平均计数，所以我写了以下sql问题：

WITH buffered_zones
     AS (SELECT row_id,
                St_buffer(pos, 0.00001) AS buffers
         FROM   flower_tbl
         limit 100),
     mean_vals
     AS (SELECT ft.row_id,
                pos,
                (SELECT Avg(ft.counted_flowers)
                 FROM   buffered_zones bz
                 WHERE  St_intersects(ft.pos, buffers)
                        AND bz.row_id = ft.row_id) AS counted_flowers
         FROM   flower_tbl ft
         GROUP  BY ft.row_id,
                   pos)
SELECT *
FROM   mean_vals

哪个有效并且给我前100个点的平均计数，大约需要15秒。当我将限制增加到1000时，运行时间增加到大约1.5分钟，依此类推。因此，我想在不到30秒的时间内运行所有15万行的问题。有什么建议？ row_id上有一个btree索引，pos上有一个gist_index。

我添加了查询计划：

CTE Scan on mean_vals  (cost=4219889.10..4222939.24 rows=152507 width=44) (actual time=214.941..8703.535 rows=152507 loops=1)
  CTE buffered_zones
->  Limit  (cost=0.00..28.40 rows=100 width=36) (actual time=0.957..23.034 rows=100 loops=1)
      ->  Seq Scan on flower_table  (cost=0.00..43311.82 rows=152507 width=36) (actual time=0.955..23.003 rows=100 loops=1)
  CTE mean_vals
->  GroupAggregate  (cost=22486.79..4219860.70 rows=152507 width=40) (actual time=214.938..8583.157 rows=152507 loops=1)
      Group Key: ft.row_id, ft.pos
      ->  Sort  (cost=22486.79..22868.06 rows=152507 width=40) (actual time=191.499..249.276 rows=152507 loops=1)
            Sort Key: ft.row_id, ft.pos
            Sort Method: external sort  Disk: 7448kB
            ->  Seq Scan on flower_table ft  (cost=0.00..5185.07 rows=152507     width=40) (actual time=0.014..66.537 rows=152507 loops=1)
          SubPlan 2
            ->  CTE Scan on buffered_zones bz  (cost=0.00..27.50 rows=1 width=0)     (actual time=0.053..0.053 rows=0 loops=152507)
              Filter: ((ft.pos && buffers) AND (row_id = ft.row_id) AND _st_intersects(ft.pos, buffers))
                  Rows Removed by Filter: 100
Planning time: 1.868 ms
Execution time: 8726.110 ms

Postgresql太慢的空间交集分析

0 个答案: