查询时间过长

Question

我正在一个简单的 SCADA 系统中开发 PostgreSQL（最新版本）。问题是每个涉及我架构中最大表的查询 [图 1] 大约需要 2 小时。

我想要的是获得建筑物 (edificio) 的构造函数的 PK，其部门 (departamento) 记录 (valor 在表 medicion) 在“gas”或“electricidad”测量中的盈余2010. 所以...

• variable 包含记录的最大值 (valmax)
• medicion 包含指标的值 (valor)

表格大小（以行为单位）。

• constructura -> 10
• edificio -> 100
• departamento -> 50.000
• variable -> 8
• medicion_departamento -> 400.000
• medicion -> 8.000.000

查询

我使用INNER JOINs 进行了查询，从最小的表（constructora 和variable）到最大的表（medicion）。

SELECT DISTINCT C.id_constructora FROM
constructora C
INNER JOIN variable V ON (V.nombre = 'electricidad' OR V.nombre = 'gas')
INNER JOIN edificio E ON (E.id_constructora = C.id_constructora)
INNER JOIN departamento D ON (E.id_edificio = D.id_edificio)
INNER JOIN medicion M ON (M.id_variable = V.id_variable)
WHERE (
    (M.valor > V.valmax) AND
    EXTRACT(YEAR FROM M.fecha) = 2010
);

解释

"HashAggregate  (cost=2343438.58..2343438.68 rows=10 width=4)"
"  Group Key: c.id_constructora"
"  ->  Hash Join  (cost=164536.25..1947605.25 rows=158333333 width=4)"
"        Hash Cond: (e.id_constructora = c.id_constructora)"
"        ->  Hash Join  (cost=4.25..1510.75 rows=50000 width=4)"
"              Hash Cond: (d.id_edificio = e.id_edificio)"
"              ->  Seq Scan on departamento d  (cost=0.00..819.00 rows=50000 width=4)"
"              ->  Hash  (cost=3.00..3.00 rows=100 width=8)"
"                    ->  Seq Scan on edificio e  (cost=0.00..3.00 rows=100 width=8)"
"        ->  Hash  (cost=164136.12..164136.12 rows=31670 width=4)"
"              ->  Nested Loop  (cost=0.00..164136.12 rows=31670 width=4)"
"                    ->  Nested Loop  (cost=0.00..163739.12 rows=3167 width=0)"
"                          Join Filter: ((m.valor > v.valmax) AND (v.id_variable = m.id_variable))"
"                          ->  Seq Scan on medicion m  (cost=0.00..162408.00 rows=38000 width=8)"
"                                Filter: (date_part('year'::text, fecha) = '2010'::double precision)"
"                          ->  Materialize  (cost=0.00..1.13 rows=2 width=24)"
"                                ->  Seq Scan on variable v  (cost=0.00..1.12 rows=2 width=24)"
"                                      Filter: (((nombre)::text = 'electricidad'::text) OR ((nombre)::text = 'gas'::text))"
"                    ->  Materialize  (cost=0.00..1.15 rows=10 width=4)"
"                          ->  Seq Scan on constructora c  (cost=0.00..1.10 rows=10 width=4)"

我的问题是我可以做些什么来显着减少查询的执行时间？

Answer 1

看来（根据架构）您的查询应该是：

SELECT DISTINCT C.id_constructora FROM
constructora C
  INNER JOIN edificio E ON (E.id_constructora = C.id_constructora)
  INNER JOIN departamento D ON (E.id_edificio = D.id_edificio)
  -- Changes here
  INNER JOIN medicion M ON (D.id_departamento = M.id_departamento)
  INNER JOIN variable V ON (M.id_variable = V.id_variable AND (V.nombre = 'electricidad' OR V.nombre = 'gas'))
WHERE (
    (M.valor > V.valmax) AND
    EXTRACT(YEAR FROM M.fecha) = 2010
);

因为在您的原始查询中，您有 medicion 和 variable 表与所有其他表的交叉连接。

你也应该有类似的功能索引

create index idx_medicion_fecha_year on medicion(EXTRACT(YEAR FROM M.fecha));

用于EXTRACT(YEAR FROM fecha) = 2010 条件。

Answer 2

join 中的 or 是性能杀手。也许这可以满足您的要求：

SELECT DISTINCT C.id_constructora
FROM constructora C INNER JOIN
     edificio E
     ON E.id_constructora = C.id_constructora INNER JOIN
     departamento D
     ON E.id_edificio = D.id_edificio LEFT JOIN
     variable ve
     ON ve.nombre = 'electricidad' AND M.valor > ve.valmax LEFT JOIN
     variable vg
     ON vg.nombre = 'gas' LEFT JOIN
     medicion me
     ON me.id_variable = ve.id_variable AND
        EXTRACT(YEAR FROM me.fecha) = 2010 AND
         me.valor > ve.valmax LEFT JOIN
     medicion mg
     ON mg.id_variable = vg.id_variable AND
        EXTRACT(YEAR FROM mg.fecha) = 2010 AND
        mg.valor > vg.valmax
WHERE EXTRACT(YEAR FROM M.fecha) = 2010 AND
      (me.id_variable IS NOT NULL OR mg.id_variable IS NOT NULL);

我还怀疑您希望在适当的 on 子句中加入 mg.id_departamento = d.id_departamento 和 me.id_departamento = d.id_departamento 的条件。

这假设您在连接键（尤其是主键）上有适当的索引。此外，您还需要variable(nombre, valor) 和medicion(id_variable, fecha, valor) 上的索引。如果你使用id_departamento，它应该是索引中的第一个键。