MySQL 内连接和子查询需要很长时间

Question

我正在尝试从以下查询中获取结果，但计算需要很长时间

我该如何改进它？

请帮忙

SELECT 
    ic.id, ic.id_number, ia.phone_number 
FROM 
    in_client AS ic 
INNER JOIN (
    SELECT id_number, MAX(modified) AS modified 
    FROM in_client 
    WHERE 
        id_number IS NOT NULL AND 
        agent_id = 1234
    GROUP BY id_number 
    HAVING MAX(modified) >= '2021-mm-dd hh:mm:ss'
) AS max USING (id_number, modified) 
INNER JOIN 
    in_agent AS ia ON (ia.id = ic.agent_id) 
LEFT OUTER JOIN 
    in_policy AS ip ON (ip.client_id = ic.id) 
WHERE 
    ip.client_id IS NULL

架构：

in_client

id - PK (INT)
agent_id - FK (INT)
id_number - VARCHAR
modified - DATETIME

in_agent

id - PK (INT)
name - VARCHAR

in_policy

id - PK (INT)
client_id - FK (INT)

注意：每个 table 还有更多其他字段，但它们与此查询无关

作为一般的 IDEA

此查询从同一代理下的客户端 table 获取所有客户端，这些客户端在策略 table.

中没有文件

每个客户端都可以创建多次，所以我对每组客户端使用 GROUP BY，然后使用最新的客户端（已修改）。

每个客户端都与代理 (FK) 相关联 - 因此在子查询中有一个 INNER JOIN 以实现完美匹配。为了让所有在策略 table 中没有文件的客户，我使用 LEFT OUTER JOIN

查询说明：

Answer 1

在 in_client 中，尝试添加以下索引：

  INDEX idx1_client (id_number, modified)
, INDEX idx2_client (agent_id, id_number, modified)

选择这些索引是为了满足 max 派生的 table 中查询表达式的要求，以便更好地访问 (agent_id, id_number and modified) 并连接到 in_client USING (id_number, modified)。这只是第一次猜测，没有这些 table 的完整架构。

我使用 MariaDB 和 MySQL 进行测试，使用较新的版本以允许动态递归地创建测试数据。

为评估问题而创建的简单测试用例：

The fiddle (runs with MariaDB 10.3+ and MySQL 8+)

CREATE TABLE in_agent (
   id           INT primary key auto_increment
 , name         VARCHAR(30)
 , phone_number varchar(20)
);

CREATE TABLE in_client (
   id        INT primary key auto_increment
 , agent_id  INT  references in_agent (id)
 , id_number VARCHAR(30)
 , modified  DATETIME DEFAULT current_timestamp
 , INDEX idx1 (id_number, modified)
 , INDEX idx2 (agent_id, id_number, modified)
);

CREATE TABLE in_policy (
   id        INT primary key auto_increment
 , client_id INT references in_client (id)
);

INSERT INTO in_agent (name)
WITH RECURSIVE cte1 (lev) AS (
      SELECT 1 UNION ALL
      SELECT lev + 1 FROM cte1 WHERE lev < 100
     )
SELECT CONCAT('name', lev) FROM cte1
;

INSERT INTO in_client (agent_id, id_number)
WITH RECURSIVE cte1 (lev) AS (
      SELECT 1 UNION ALL
      SELECT lev + 1 FROM cte1 WHERE lev < 100
     )
SELECT TRUNCATE((lev-1)/50,0)+1, CONCAT('idnum', lev) FROM cte1
;

SELECT COUNT(*) FROM in_agent;
SELECT COUNT(*) FROM in_client;

SELECT ic.id, ic.id_number, ia.phone_number 
  FROM in_client AS ic 
  JOIN (
    SELECT id_number, MAX(modified) AS modified 
      FROM in_client 
     WHERE id_number IS NOT NULL
       AND agent_id = 1
     GROUP BY id_number 
    HAVING MAX(modified) >= '2021-01-01 01:00:00'
     ) AS max USING (id_number, modified) 
  JOIN      in_agent  AS ia ON (ia.id = ic.agent_id) 
  LEFT JOIN in_policy AS ip ON (ip.client_id = ic.id) 
 WHERE ip.client_id IS NULL
;

EXPLAIN
SELECT ic.id, ic.id_number, ia.phone_number 
  FROM in_client AS ic 
  JOIN (
    SELECT id_number, MAX(modified) AS modified 
      FROM in_client 
     WHERE id_number IS NOT NULL
       AND agent_id = 1
     GROUP BY id_number 
    HAVING MAX(modified) >= '2021-01-01 01:00:00'
     ) AS max USING (id_number, modified) 
  JOIN      in_agent  AS ia ON (ia.id = ic.agent_id) 
  LEFT JOIN in_policy AS ip ON (ip.client_id = ic.id) 
 WHERE ip.client_id IS NULL
;

选择的指标似乎比原始问题详细信息中的指标更好。

这应该有所帮助。

MySQL 内连接和子查询需要很长时间

MySQL Inner join and subqueries takes a long time

mysql

sql

amazon-rds