尝试使用带有 plpg/sql 的并行化自定义 function

Question

我试图通过让刨床使用并行化索引/序列扫描来加快查询速度。 我有 2 张桌子。 表 1 表示一些链接。 它包含 2 列：链接索引和这些链接的距离。 Table2 包含附加信息：在数组中绘制链接的点列表和与 table1 匹配的链接索引。 我的目标是计算每个链接的每个点之间的距离和斜率。 为此，我创建了一个 function calculateDistancesAndSlopes(sub.points::geometry)。 此 function 以一种自定义类型 (::my_great_type) 返回 2 个所需的输出。

-- Create sub-slopes and sub-distances
DROP type if exists my_great_type cascade; 
CREATE TYPE my_great_type AS (slopes INTEGER[], distances DOUBLE PRECISION[]);  --INTEGER[], OUT distances DOUBLE PRECISION[]

----------------------------------------------------------------
-- Define functions --
----------------------------------------------------------------
-- Create function to calculate distances and slopes for sub segments
DROP function if EXISTS roaddata_nc_sc_test.calculatedistancesandslopes(geom public.geometry) cascade;
CREATE OR REPLACE FUNCTION roaddata_nc_sc_test.calculateDistancesAndSlopes(geom public.geometry) returns my_great_type
AS $$
DECLARE
    currentPoint public.GEOMETRY;
    nextPoint public.GEOMETRY;
    distance DOUBLE PRECISION;
    height NUMERIC;
    slope INTEGER;
    slopes INTEGER[];
    distances DOUBLE PRECISION[];
    test INTEGER;
BEGIN
    IF geom IS NULL then
        raise info 'the geom is null : %' , geom;
        distances := array_append(distances, CAST(0.0 AS DOUBLE PRECISION));
        slopes := array_append(slopes, 0);
    END IF;

    FOR i IN 1 .. public.ST_NumGeometries(geom) - 1 LOOP
        currentPoint := public.ST_GeometryN(geom, i);
        nextPoint := public.ST_GeometryN(geom, i + 1);

        distance := public.ST_Distance(currentPoint, nextPoint, true);
        distances := array_append(distances, distance);

        height := public.ST_Z(nextPoint) - public.ST_Z(currentPoint);

        slope := FLOOR(ATAN(height / distance) * (180. / PI()));
        slopes := array_append(slopes, slope);
           
    END LOOP;
    RETURN (slopes, distances);
END;
$$ LANGUAGE plpgsql PARALLEL Safe;

那我可以这样称呼这个function。

DROP table IF EXISTS tableFinal CASCADE;
 CREATE table tableFinal as
   SELECT sub.link_id,  slopesAndDistances.slopes , slopesAndDistances.distances FROM
    (
      SELECT
        table1.link_id,
      table2.points
      FROM
        table1 as table1
      inner JOIN table2 as table2 ON table1.link_id = table2.link_id
    )sub
   , calculateDistancesAndSlopes(sub.points) as slopesAndDistances;

It’s great. It’s works fine but it’s (maybe) too long. For your information, on my current dataset it’s take 2 min for 1693659 rows.

QUERY PLAN                                                                                                        |
------------------------------------------------------------------------------------------------------------------+
Nested Loop  (cost=5.90..195313.85 rows=1693659 width=72)                                                         |
  ->  Merge Join  (cost=5.65..161440.42 rows=1693659 width=203)                                                   |
        Merge Cond: (table1.link_id = table2.link_id)                                                          |
        ->  Index Only Scan using table1_idx on table1  (cost=0.43..53008.78 rows=1693659 width=8)     |
        ->  Index Scan using table2_id_idx on table2  (cost=0.43..83025.74 rows=1694621 width=203)|
  ->  Function Scan on calculatedistancesandslopes slopesanddistances  (cost=0.25..0.26 rows=1 width=64)      

|

因此，您可能会注意到我声明我的 function 对于并行化是安全的。 然后，我尝试这个查询。

explain SELECT *
  FROM
  (
     SELECT sub.link_id,
           calculateDistancesAndSlopes(sub.points) as slopesAndDistances
    FROM
    (
      SELECT
        table1.link_id,
      table2.points
      FROM
        table1 as table1
      inner JOIN table2 as table2 ON table1.link_id = table2.link_id
    )sub    
   )lsd

我得到

QUERY PLAN                                                                                        |
--------------------------------------------------------------------------------------------------+
Gather  (cost=28790.05..474555.11 rows=1693659 width=40)                                          |
  Workers Planned: 2                                                                              |
  ->  Parallel Hash Join  (cost=27790.05..304189.21 rows=705691 width=40)                         |
        Hash Cond: (table2.link_id = table1.link_id)                                           |
        ->  Parallel Seq Scan on table2  (cost=0.00..55371.92 rows=706092 width=203)|
        ->  Parallel Hash  (cost=16211.91..16211.91 rows=705691 width=8)                          |
              ->  Parallel Seq Scan on table1  (cost=0.00..16211.91 rows=705691 width=8)    |

所以我说它可能会更快，但是当我尝试将结果“存储”在这样的表中时

DROP table IF EXISTS tableFinal CASCADE;
CREATE table tableFinal as
    SELECT *
  FROM
  (
     SELECT sub.link_id,
           calculateDistancesAndSlopes(sub.points) as slopesAndDistances
    FROM
    (
      SELECT
        table1.link_id,
        table2.points
      FROM
        table1 as table1
      inner JOIN table2s as table2 ON table1.link_id = table2.link_id
    )sub    
   )lsd

我明白了：

SQL Error [XX000]: ERROR: cannot start commands during a parallel operation
  Où : PL/pgSQL function calculatedistancesandslopes(public.geometry) line 12 at IF”

从理论上讲，刨床可以在我的情况下使用并行化吗？