PostgreSQL 查询效率

Question

我正在使用 PostgreSQL（我是数据库领域的新手），我想知道您对我在我正在使用的代码中发现的此类查询的效率有何看法。 这些查询有很多 JOIN，其中之一（粗体）根据请求有很多行。 这迫使我们 GROUP BY request.id 以按请求获取一行和一个包含所有这些行数据的字段（粗体）。

我认为这种查询必须花费大量时间来寻找所有这些最大值，但我想不出另一种方法来做到这一点。 关于它的效率以及如何改进它的任何想法？

SELECT
  request.id AS id,
  max(request_type.name) AS request_type,
  to_char(max(request.timestamp),'DD/mm/YYYY HH24:mi') AS timestamp,
  to_char(max(request.timestamp),'YYYY-mm-DD') AS timestamp_filtering,
  max(state.name) AS request_state,
  max(users.name || ' ' || COALESCE(users.surname,'')) AS create_user,
  max(request.id_create_user) AS id_create_user,
  max(enterprise.name) AS enterprise,
  max(cause_issue.name) AS cause,
  max(request_movements.id_request_state) AS id_state,
  array_to_string(array_agg(DISTINCT act_code.name || '/' || req_res.act_code), ', ') AS act_code, /* here */
  max(revised.code) AS state_revised, 
  max(request_shipment.warehouse) AS warehouse,
  max(req_res.id_warehouse) AS id_warehouse
FROM
  request
  LEFT JOIN users
    ON users.id=request.id_create_user
  LEFT JOIN enterprise
    ON users.id_enterprise=enterprise.id
  LEFT JOIN request_movements
    ON request_movements.id=request.id_request_movement
  LEFT JOIN request_versions
    ON request_versions.id = request_movements.id_version
  LEFT JOIN state
    ON request_movements.id_request_state=state.id
  INNER JOIN request_type
    ON request.id_request_type=request_type.id
  LEFT JOIN cause_issue
    ON request.id_cause_issue=cause_issue.id
  LEFT JOIN request_reserve req_res
    ON req_res.id_request = request.id /* here */
  LEFT JOIN act_code
    ON req_res.id_act_code=act_code.id
  LEFT JOIN request_shipment
    ON (request_shipment.id_request=request.id)
  LEFT JOIN warehouse_enterprise
    ON (warehouse_enterprise.id = request_shipment.id_warehouse_enterprise)
  LEFT JOIN revised
    ON (revised.id = request_shipment.id_revised)
WHERE
  request.id_request_type = "any_type"  
GROUP BY
  request.id

EXPLAIN 返回this 。

Answer 1

在JOIN大连接之前，您可以通过聚合request_reserve和act_code值来大大简化此查询。 这避免了对所有其他列的聚合函数的需要，并且对于大量行通常应该更快。

SELECT r.id
      ,rt.name AS request_type
      ,to_char(r.timestamp, 'DD/mm/YYYY HH24:mi') AS timestamp
      ,to_char(r.timestamp, 'YYYY-mm-DD') AS timestamp_filtering
      ,s.name AS request_state
      ,u.name || COALESCE(' ' || u.surname, '') AS create_user
      ,r.id_create_user
      ,e.name AS enterprise
      ,c.name AS cause
      ,rm.id_request_state AS id_state
      ,rr.act_code
      ,rd.code AS state_revised
      ,rs.warehouse
      ,rr.id_warehouse
FROM      request              r
LEFT JOIN users                u  ON u.id = r.id_create_user
LEFT JOIN enterprise           e  ON e.id = u.id_enterprise
LEFT JOIN request_movements    rm ON rm.id = r.id_request_movement
LEFT JOIN request_versions     rv ON rv.id = rm.id_version
LEFT JOIN state                s  ON s.id = rm.id_request_state
     JOIN request_type         rt ON rt.id = r.id_request_type
LEFT JOIN cause_issue          c  ON c.id = r.id_cause_issue
LEFT JOIN request_shipment     rs ON rs.id_request = r.id
LEFT JOIN warehouse_enterprise w  ON w.id = rs.id_warehouse_enterprise
LEFT JOIN revised              rd ON rd.id = rs.id_revised
LEFT JOIN (
   SELECT rr.id_request, rr.id_warehouse
         ,array_to_string(array_agg(
             DISTINCT a.name || '/' || rr.act_code), ', ') AS act_code
   FROM   request_reserve rr
   LEFT   JOIN act_code   a ON r.id_act_code = a.id
   GROUP  BY rr.id_request, rr.id_warehouse
   )  rr ON rr.id_request = r.id
WHERE  r.id_request_type = "any_type";  -- use single quotes for values!

对于大型查询，您必须拥有一种人眼可以轻松解析的格式。 因此，我在改进查询之前重新格式化。 我使用表别名来尽可能避免使用笨拙的标识符。

对create_user小改进：没有尾随空格。 如果名称的任何部分都可以是NULL ，我建议这样做以避免出现悬空空间：

COALESCE(u.name || ' ' || u.surname, u.name, u.surname)

在 PostgreSQL 9.1 或更高版本中，您可以使用concat_ws() 。