批量 MYSQL 插入以提高数据库结构迁移的性能

Batch MYSQL inserts for performance in DB structure migration

我需要重组我的 MYSQL InnoDB 数据库。

目前我有一个 customer table 持有 3 个产品名称。

我需要将这些名称提取到新的 product table。 product table 应该包含当前保存在 customer table 中的每个名称,并通过新的 [=19] 链接到 customer table =] table。虽然产品名称可能不是唯一的,但它们彼此之间没有任何关系,这意味着每个 customer 都需要在 product table 中插入 3 个新条目以及 customer_product table.

中的 3 个新条目

所以不是这个:

customer
| id | product_name_a | product_name_b | product_name_c |

我需要这个:

customer
| id |

customer_product
| customer_id | product_id | X3

product
| id | name | X3

我编写了以下 MYSQL 有效的程序:

BEGIN
  DECLARE nbr_of_customers BIGINT(20);
  DECLARE customer_count BIGINT(20);
  DECLARE product_id BIGINT(20);
  DECLARE customer_id BIGINT(20);
  DECLARE product_name_a VARCHAR(500);
  DECLARE product_name_b VARCHAR(500);
  DECLARE product_name_c VARCHAR(500);

  SELECT COUNT(*) FROM customer INTO nbr_of_customers;
  SET customer_count = 0;
  SET product_id = 1;

  WHILE customer_count < nbr_of_customers DO
    SELECT
      customer.id,
      customer.product_name_a,
      customer.product_name_b,
      customer.product_name_c
    INTO
      customer_id,
      product_name_a,
      product_name_b,
      product_name_c
    FROM customer
    LIMIT customer_count,1;

    INSERT INTO product(id, name)
      VALUES(product_id, product_name_a);
    INSERT INTO customer_product(customer_id, product_id)
      VALUES(customer_id, product_id);
    SET product_id = product_id + 1;

    INSERT INTO product(id, name)
      VALUES(product_id, product_name_b);
    INSERT INTO customer_product(customer_id, product_id)
      VALUES(customer_id, product_id);
    SET product_id = product_id + 1;

    INSERT INTO product(id, name)
      VALUES(product_id, product_name_c);
    INSERT INTO customer_product(customer_id, product_id)
      VALUES(customer_id, product_id);
    SET product_id = product_id + 1;

    SET customer_count = customer_count + 1;
  END WHILE;
END;

这太慢了。

我在本地 运行 估计我的 ~15k 客户需要 ~1 小时才能完成。我的 VPS 服务器比那慢得多,因此可能需要 10 小时才能完成。

问题似乎是插入需要很长时间。因此,我想在过程中存储所有插入,并在循环完成后批量执行它们,我知道要插入什么。

我有办法批量执行所有 ~100k 插入以优化性能,还是有更好的方法?


最终编辑:

我标记了正确的解决方案,因为它在极大地加快进程方面做得非常出色,这是问题的主要焦点。最后,我最终使用修改后的生产代码(在 Java 中)执行迁移,因为该解决方案在不转义插入的字符串方面存在局限性。

也许您可以在三个单独的插入(而不是 ~100K)中执行此操作,如下所示:

INSERT INTO customer_product (customer_id, product_id) 
SELECT customer.id as customer_id, product.id as product_id 
FROM customer 
  JOIN product on customer.product_name_a = product.name

INSERT INTO customer_product (customer_id, product_id) 
SELECT customer.id as customer_id, product.id as product_id 
FROM customer 
  JOIN product on customer.product_name_b = product.name


INSERT INTO customer_product (customer_id, product_id) 
SELECT customer.id as customer_id, product.id as product_id 
FROM customer 
  JOIN product on customer.product_name_c = product.name

当然,您必须提前设置 product table,并且您希望从 customer table事后。

如果您在 customer.product_name_X 列(也可能是 product.name 列上创建索引,虽然它很少,但如果它很重要的话,可能会进一步加快速度。 EXPLAIN 可以提供帮助。

首先,使用游标来处理单个查询的结果,而不是对每一行执行单独的查询。

然后将 VALUES 列表连接成您使用 PREPAREEXECUTE 执行的字符串。

我的代码以 100 个客户为一组进行插入,因为我预计查询的大小会有限制。

BEGIN
  DECLARE product_id BIGINT(20);
  DECLARE customer_id BIGINT(20);
  DECLARE product_name_a VARCHAR(500);
  DECLARE product_name_b VARCHAR(500);
  DECLARE product_name_c VARCHAR(500);
  DECLARE done INT DEFAULT FALSE;
  DECLARE cur CURSOR FOR SELECT c.id, c.product_name_a, c.product_name_b, c.product_name_c FROM customer AS c;
  DECLARE CONTINUE HANDLER FOR NOT FOUND SET done = TRUE;

  SET product_id = 1;

  OPEN cur;

  SET @product_values = '';
  SET @cp_values = '';

  read_loop: LOOP
    FETCH cur INTO customer_id, product_name_a, product_name_b, product_name_c;
    IF done THEN
      LEAVE read_loop;
    END IF;

    SET @product_values = CONCAT(@product_values, IF(@product_values != '', ',', ''), "(", product_id, ",'", product_name_a, "'), (", product_id + 1, ",'", product_name_b, "'), (", product_id + 2, ",'", product_name_c, "'), ");
    SET @cp_values = CONCAT(@cp_values, IF(@cp_values != '', ',', ''), "(", customer_id, ",", product_id, "), (", customer_id, ",", product_id + 1, "), (", customer_id, ",", product_id + 2, "),");

    SET product_id = product_id + 3;

    IF product_id % 300 = 1 -- insert every 100 customers
    THEN BEGIN
         SET @insert_product = CONCAT("INSERT INTO product(id, name) VALUES ", @product_values);
         PREPARE stmt1 FROM @insert_product;
         EXECUTE stmt1;

         SET @insert_cp = CONCAT("INSERT INTO customer_product(customer_id, product_id) VALUES ", @cp_values);
         PREPARE stmt2 FROM @insert_cp;
         EXECUTE stmt2;

         SET @product_values = '';
         SET @cp_values = '';
     END IF;

  END LOOP;

  IF @product_values != '' -- Process any remaining rows
  THEN BEGIN
       SET @insert_product = CONCAT("INSERT INTO product(id, name) VALUES ", @product_values);
       PREPARE stmt1 FROM @insert_product;
       EXECUTE stmt1;

       SET @insert_cp = CONCAT("INSERT INTO customer_product(customer_id, product_id) VALUES ", @cp_values);
       PREPARE stmt2 FROM @insert_cp;
       EXECUTE stmt2;

       SET @product_values = '';
       SET @cp_values = '';
   END IF;
END;

请注意,使用此解决方案,产品名称在插入前将无法正确转义。 因此,如果任何产品名称包含特殊字符,例如单引号 '

,则此解决方案将不起作用