MySQL五千万大表查询优化实战

news2024/11/19 8:27:39

背景

DBA同事在钉钉发了两张告警截图，作为“始作俑者”的我很心虚，因为刚才是我在管理后台查询数据，结果很久都没出来，并且用多个维度查了N次

问题分析

这是当天上线的功能，完事我立马锁定SQL然后开启排查

# 原SQL，此为分页接口中count部分
SELECT
	COUNT( 1 ) 
FROM
	shop_day_statistics a
	LEFT JOIN (
	SELECT
		shop_id shop_id,
		statistics_time,
		merchant_id,
		GROUP_CONCAT( CONCAT_ws( '-', service_id, service_name ) SEPARATOR ',' ) AS service_name,
		sum( profit_amount ) profit_amount 
	FROM
		service_day_statistics 
	GROUP BY
		shop_id,
		statistics_time 
	) s ON s.shop_id = a.shop_id
	
	AND a.statistics_time = s.statistics_time 
WHERE
	a.statistics_time BETWEEN '2023-09-05 00:00:00.0' 
	AND '2024-10-22 00:00:00.0' 
	AND a.shop_id = 999999;
    
# SQL解释
此SQL中有两张表：shop_day_statistics(5800W)、service_day_statistics(30多万)
下面将用A表来替代shop_day_statistics，B表来替代service_day_statistics
A表与B表的关系是一对多，因业务的需要，故之前在一条SQL上查询得出

表结构：
DROP TABLE IF EXISTS `shop_day_statistics`;
CREATE TABLE `shop_day_statistics`  (
  `id` bigint NOT NULL AUTO_INCREMENT COMMENT 'id',
  `statistics_time` date NOT NULL COMMENT '统计日期',
  `group_id` bigint NULL DEFAULT NULL COMMENT '代理商ID',
  `group_name` varchar(70) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci NULL DEFAULT NULL COMMENT '代理商名称',
  `shop_id` bigint NULL DEFAULT NULL COMMENT '门店ID',
  `shop_name` varchar(200) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci NULL DEFAULT NULL COMMENT '门店名称',
  `merchant_id` bigint NULL DEFAULT NULL COMMENT '商户ID',
  `merchant_name` varchar(200) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci NULL DEFAULT NULL COMMENT '商户名称',
  `employee_id` bigint NULL DEFAULT NULL COMMENT '员工ID',
  `employee_name` varchar(45) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci NULL DEFAULT NULL COMMENT '员工名称',
  PRIMARY KEY (`id`) USING BTREE,
  INDEX `idx_groupid_time`(`group_id` ASC, `statistics_time` ASC) USING BTREE,
  INDEX `idx_merchantid_time`(`merchant_id` ASC, `statistics_time` ASC) USING BTREE,
  INDEX `idx_statistics_time`(`statistics_time` ASC) USING BTREE
) ENGINE = InnoDB AUTO_INCREMENT = 1 CHARACTER SET = utf8mb4 COLLATE = utf8mb4_0900_ai_ci COMMENT = '门店日收益统计' ROW_FORMAT = DYNAMIC;

DROP TABLE IF EXISTS `service_day_statistics`;
CREATE TABLE `service_day_statistics`  (
  `id` bigint NOT NULL AUTO_INCREMENT COMMENT 'id',
  `statistics_time` date NOT NULL COMMENT '统计日期',
  `shop_id` bigint NULL DEFAULT NULL COMMENT '门店ID',
  `shop_name` varchar(200) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci NULL DEFAULT NULL COMMENT '门店名称',
  `merchant_id` bigint NULL DEFAULT NULL COMMENT '商户ID',
  `merchant_name` varchar(200) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci NULL DEFAULT NULL COMMENT '商户名称',
  `service_id` bigint NULL DEFAULT NULL COMMENT '服务商id',
  `service_name` varchar(45) CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_ai_ci NULL DEFAULT NULL COMMENT '服务商名称',
  `profit_amount` bigint NULL DEFAULT NULL COMMENT '服务商收益金额',
  `gmt_create` datetime NOT NULL COMMENT '创建时间',
  PRIMARY KEY (`id`) USING BTREE
) ENGINE = InnoDB AUTO_INCREMENT = 1 CHARACTER SET = utf8mb4 COLLATE = utf8mb4_0900_ai_ci COMMENT = '服务商日收益统计' ROW_FORMAT = DYNAMIC;

SQL解析

从上图可以了解到几个点

内层的联表是ALL，也很慢，虽然只有三十多万数据，时间也去到了一秒
外层A表是ALL，全表扫描非常慢

依次改造

子查询中是没有带任何条件的，故在B表中新增两个参数：group_id,group_name，并为A B表关联的字段增加索引，在后续查询中将外层的条件(门店商户代理商ID)尽可能的加到子查询中，这样能直接命中索引，无需对B进行全表扫描
外层A表根据shop_id、statistics_time建立联合索引(shop_id必须放前面，区分度更加高)

改造后SQL及其执行计划

SELECT
	COUNT( 1 ) 
FROM
	shop_day_statistics a
	LEFT JOIN (
	SELECT
		shop_id shop_id,
		statistics_time,
		merchant_id,
		GROUP_CONCAT( CONCAT_ws( '-', service_id, service_name ) SEPARATOR ',' ) AS service_name,
		sum( profit_amount ) profit_amount 
	FROM
		service_day_statistics 
	WHERE
		statistics_time BETWEEN '2023-09-05 00:00:00.0' 
		AND '2024-10-22 00:00:00.0' 
		AND shop_id = 99999 
	GROUP BY
		shop_id,
		statistics_time 
	) s ON s.shop_id = a.shop_id 
	AND a.statistics_time = s.statistics_time 
WHERE
	a.statistics_time BETWEEN '2023-09-05 00:00:00.0' 
	AND '2024-10-22 00:00:00.0' 
	AND a.shop_id = 99999;