一. 问题描述
我们在设计表结构的时候,设计规范里面有一条如下规则:
-
对于可变长度的字段,在满足条件的前提下,尽可能使用较短的变长字段长度。
为什么这么规定?我在网上查了一下,主要基于两个方面
-
基于存储空间的考虑
-
基于性能的考虑
网上说Varchar(50)
和varchar(500)
存储空间上是一样的,真的是这样吗?
基于性能考虑,是因为过长的字段会影响到查询性能?
本文我将带着这两个问题探讨验证一下
二.验证存储空间区别
1.准备两张表
CREATE TABLE `category_info_varchar_50` (
`id` bigint(20) NOT NULL AUTO_INCREMENT COMMENT '主键',
`name` varchar(50) NOT NULL COMMENT '分类名称',
`is_show` tinyint(4) NOT NULL DEFAULT '0' COMMENT '是否展示:0 禁用,1启用',
`sort` int(11) NOT NULL DEFAULT '0' COMMENT '序号',
`deleted` tinyint(1) DEFAULT '0' COMMENT '是否删除',
`create_time` datetime NOT NULL COMMENT '创建时间',
`update_time` datetime NOT NULL COMMENT '更新时间',
PRIMARY KEY (`id`) USING BTREE,
KEY `idx_name` (`name`) USING BTREE COMMENT '名称索引'
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COMMENT='分类';
CREATE TABLE `category_info_varchar_500` (
`id` bigint(20) NOT NULL AUTO_INCREMENT COMMENT '主键',
`name` varchar(500) NOT NULL COMMENT '分类名称',
`is_show` tinyint(4) NOT NULL DEFAULT '0' COMMENT '是否展示:0 禁用,1启用',
`sort` int(11) NOT NULL DEFAULT '0' COMMENT '序号',
`deleted` tinyint(1) DEFAULT '0' COMMENT '是否删除',
`create_time` datetime NOT NULL COMMENT '创建时间',
`update_time` datetime NOT NULL COMMENT '更新时间',
PRIMARY KEY (`id`) USING BTREE,
KEY `idx_name` (`name`) USING BTREE COMMENT '名称索引'
) ENGINE=InnoDB AUTO_INCREMENT=288135 DEFAULT CHARSET=utf8mb4 COMMENT='分类';
2.准备数据
给每张表插入相同的数据,为了凸显不同,插入100万条数据
DELIMITER $$
CREATE PROCEDURE batchInsertData(IN total INT)
BEGIN
DECLARE start_idx INT DEFAULT 1;
DECLARE end_idx INT;
DECLARE batch_size INT DEFAULT 500;
DECLARE insert_values TEXT;
SET end_idx = LEAST(total, start_idx + batch_size - 1);
WHILE start_idx <= total DO
SET insert_values = '';
WHILE start_idx <= end_idx DO
SET insert_values = CONCAT(insert_values, CONCAT('(\'name', start_idx, '\', 0, 0, 0, NOW(), NOW()),'));
SET start_idx = start_idx + 1;
END WHILE;
SET insert_values = LEFT(insert_values, LENGTH(insert_values) - 1); -- Remove the trailing comma
SET @sql = CONCAT('INSERT INTO category_info_varchar_50 (name, is_show, sort, deleted, create_time, update_time) VALUES ', insert_values, ';');
PREPARE stmt FROM @sql;
EXECUTE stmt;
SET @sql = CONCAT('INSERT INTO category_info_varchar_500 (name, is_show, sort, deleted, create_time, update_time) VALUES ', insert_values, ';');
PREPARE stmt FROM @sql;
EXECUTE stmt;
SET end_idx = LEAST(total, start_idx + batch_size - 1);
END WHILE;
END$$
DELIMITER ;
CALL batchInsertData(1000000);
3.验证存储空间
查询第一张表SQL
SELECT
table_schema AS "数据库",
table_name AS "表名",
table_rows AS "记录数",
TRUNCATE ( data_length / 1024 / 1024, 2 ) AS "数据容量(MB)",
TRUNCATE ( index_length / 1024 / 1024, 2 ) AS "索引容量(MB)"
FROM
information_schema.TABLES
WHERE
table_schema = 'test_mysql_field'
and TABLE_NAME = 'category_info_varchar_50'
ORDER BY
data_length DESC,
index_length DESC;
查询结果
查询第二张表SQL
SELECT
table_schema AS "数据库",
table_name AS "表名",
table_rows AS "记录数",
TRUNCATE ( data_length / 1024 / 1024, 2 ) AS "数据容量(MB)",
TRUNCATE ( index_length / 1024 / 1024, 2 ) AS "索引容量(MB)"
FROM
information_schema.TABLES
WHERE
table_schema = 'test_mysql_field'
and TABLE_NAME = 'category_info_varchar_500'
ORDER BY
data_length DESC,
index_length DESC;
查询结果
4.结论
两张表在占用空间上确实是一样的,并无差别
三.验证性能区别
1.验证索引覆盖查询
select name from category_info_varchar_50 where name = 'name100000'
-- 耗时0.012s
select name from category_info_varchar_500 where name = 'name100000'
-- 耗时0.012s
select name from category_info_varchar_50 order by name;
-- 耗时0.370s
select name from category_info_varchar_500 order by name;
-- 耗时0.379s
通过索引覆盖查询性能差别不大
1.验证索引查询
select * from category_info_varchar_50 where name = 'name100000'
--耗时 0.012s
select * from category_info_varchar_500 where name = 'name100000'
--耗时 0.012s
select * from category_info_varchar_50 where name in('name100','name1000','name100000','name10000','name1100000',
'name200','name2000','name200000','name20000','name2200000','name300','name3000','name300000','name30000','name3300000',
'name400','name4000','name400000','name40000','name4400000','name500','name5000','name500000','name50000','name5500000',
'name600','name6000','name600000','name60000','name6600000','name700','name7000','name700000','name70000','name7700000','name800',
'name8000','name800000','name80000','name6600000','name900','name9000','name900000','name90000','name9900000')
-- 耗时 0.011s -0.014s
-- 增加 order by name 耗时 0.012s - 0.015s
select * from category_info_varchar_50 where name in('name100','name1000','name100000','name10000','name1100000',
'name200','name2000','name200000','name20000','name2200000','name300','name3000','name300000','name30000','name3300000',
'name400','name4000','name400000','name40000','name4400000','name500','name5000','name500000','name50000','name5500000',
'name600','name6000','name600000','name60000','name6600000','name700','name7000','name700000','name70000','name7700000','name800',
'name8000','name800000','name80000','name6600000','name900','name9000','name900000','name90000','name9900000')
-- 耗时 0.012s -0.014s
-- 增加 order by name 耗时 0.014s - 0.017s
索引范围查询性能基本相同, 增加了order By后开始有一定性能差别;
3.验证全表查询和排序
全表无排序
全表有排序
select * from category_info_varchar_50 order by name ;
--耗时 1.498s
select * from category_info_varchar_500 order by name ;
--耗时 4.875s
结论:
全表扫描无排序情况下,两者性能无差异,在全表有排序的情况下, 两种性能差异巨大;
分析原因
varchar50 全表执行sql分析
我发现86%的时花在数据传输上,接下来我们看状态部分,关注Created_tmp_files和sort_merge_passes
Created_tmp_files为3
sort_merge_passes为95
varchar500 全表执行sql分析
增加了临时表排序
Created_tmp_files 为 4
sort_merge_passes为645
关于sort_merge_passes, Mysql给出了如下描述:
❝Number of merge passes that the sort algorithm has had to do. If this value is large, you may want to increase the value of the sort_buffer_size.
❞
其实sort_merge_passes对应的就是MySQL做归并排序的次数,也就是说,如果sort_merge_passes值比较大,说明sort_buffer和要排序的数据差距越大,我们可以通过增大sort_buffer_size或者让填入sort_buffer_size的键值对更小来缓解sort_merge_passes归并排序的次数。
四.最终结论
至此,我们不难发现,当我们最该字段进行排序操作的时候,Mysql会根据该字段的设计的长度进行内存预估, 如果设计过大的可变长度, 会导致内存预估的值超出sort_buffer_size的大小, 导致mysql采用磁盘临时文件排序,最终影响查询性能