| ##ft_boolean_syntax | -> ;2b7da4bc8b0b7a3041e9bf8984b6eb0a” 表示出现该单词时增加相关性,查询的结果靠前“<” 表示出现该单词时降低相关性,查询的结果靠后
*
表示通配符,只能接在词后面
~
允许出现该单词,但是出现时相关性为负,表示拥有该字会下降相关性,但不像「-」将之排除,只是排在较后面
"" 双引号表示短语,表示要彻底相符,不可拆字效果,类同于 like '%keyword%'
()
经过括号来使用字条件:
+aaa +(>bbb <ccc) aaa="aaa" sql="sql" select="select" from="from" test="test" where="where" match="match" against="against" in="in" boolean="boolean" mode="mode" select="select" from="from" tommy="tommy" where="where" match="match" against="against" in="in" boolean="boolean" mode="mode" select="select" from="from" tommy="tommy" where="where" match="match" against="against">李秀琴 <练习册 <不是人>是个鬼' in boolean mode);
四、测试结果
测试环境:本机4核16G Windows10,MySQL 8.0
测试数据量:salebilldetail
表 1276
万行,salebill
表 269
万行, customer
表 30
万行, goods
表 75
万行。
争对测试用的SQL语句,增加了以下全文索引:
CREATE FULLTEXT INDEX billno_fulltext ON salebill(billno) WITH PARSER ngram;
CREATE FULLTEXT INDEX remarks_fulltext ON salebill(remarks) WITH PARSER ngram;
CREATE FULLTEXT INDEX remarks_fulltext ON salebilldetail(remarks) WITH PARSER ngram;
CREATE FULLTEXT INDEX goodsremarks_fulltext ON salebilldetail(goodsremarks) WITH PARSER ngram;
CREATE FULLTEXT INDEX remarks_goodsremarks_fulltext ON salebilldetail(remarks, goodsremarks) WITH PARSER ngram;
CREATE FULLTEXT INDEX custname_fulltext ON customer(custname) WITH PARSER ngram;
CREATE FULLTEXT INDEX goodsname_fulltext ON goods(goodsname) WITH PARSER ngram;
CREATE FULLTEXT INDEX goodscode_fulltext ON goods(goodscode) WITH PARSER ngram;
测试结果,总的来说很魔幻。
为什么魔幻,看下面几个语句:
test_1
-- 测试1,原始 like 查询方式,用时 0.765s
select 1 from salebilldetail d where d.tid=260434 and ((d.remarks like concat('%','葡萄','%')) or (d.goodsremarks like concat('%','葡萄','%')));
test_2
-- 测试2,使用全文索引 remarks_fulltext、goodsremarks_fulltext, 用时 0.834s
select 1 from salebilldetail d where d.tid=260434 and ((match(d.remarks) Against(concat('"','葡萄','"') in boolean mode)) or (match(d.goodsremarks) Against(concat('"','葡萄','"') in boolean mode)));
test_3
-- 测试3,使用全文索引 remarks_goodsremarks_fulltext, 用时 0.242s
select 1 from salebilldetail d where d.tid=260434 and ((match(d.remarks,d.goodsremarks) Against(concat('"','葡萄','"') in boolean mode)));
test_4
-- 测试4,原始 like 查询方式,不过滤 tid ,用时 22.654s
select t from salebilldetail d where ((d.remarks like concat('%','葡萄','%')) or (d.goodsremarks like concat('%','葡萄','%')));
test_5
-- 测试5,使用全文索引 remarks_fulltext、goodsremarks_fulltext, 不过滤 tid ,用时 24.855s
select 1 from salebilldetail d where ((match(d.remarks) Against(concat('"','葡萄','"') in boolean mode)) or (match(d.goodsremarks) Against(concat('"','葡萄','"') in boolean mode)));
test_6
-- 测试6,使用全文索引 remarks_goodsremarks_fulltext, 不过滤 tid ,用时 0.213s
select 1 from salebilldetail d where ((match(d.remarks,d.goodsremarks) Against(concat('"','葡萄','"') in boolean mode)));
test_7
-- 测试7,使用全文索引 remarks_goodsremarks_fulltext, 用时 0.22s
select count(1) from salebilldetail d where d.tid=260434 and ((match(d.remarks,d.goodsremarks) Against(concat('"','葡萄','"') in boolean mode)));
test_8
-- 测试8,使用全文索引 remarks_goodsremarks_fulltext, 不过滤 tid ,用时 0.007s
select count(1) from salebilldetail d where ((match(d.remarks,d.goodsremarks) Against(concat('"','葡萄','"') in boolean mode)));
从上面的测试语句可以看出,数据量越多,查询越简单,全文索引的效果越好。
再来看看我们的业务测试SQL:
test_9
-- 测试9
select
i.billid
,if(0,0,i.qty) as qty
,if(0,0,i.goodstotal) as total
,if(0,0,i.chktotal) as selfchktotal
,if(0,0,i.distotal) as distotal
,if(0,0,i.otherpay) as feetotal
,if(0,0,ifnull(d.costtotal,0)) as costtotal
,if(0,0,ifnull(d.maoli,0)) as maoli
,i.billno
,from_unixtime(i.billdate,'%Y-%m-%d') as billdate /*单据日期*/
,from_unixtime(i.createdate,'%Y-%m-%d %H:%i:%s') as createdate /*制单日期*/
,if(i.sdate=0,'',from_unixtime(i.sdate,'%Y-%m-%d %H:%i:%s')) as sdate /*过账日期*/
,from_unixtime(i.udate,'%Y-%m-%d %H:%i:%s') as udate /*最后修改时间*/
,i.custid ,c.custname
,i.storeid ,k.storename
,i.empid ,e.empname
,i.userid ,u.username
,i.remarks /*单据备注*/
,i.effect,i.settle,i.redold,i.rednew /*单据状态*/
,i.printtimes /* 打印次数 */
,(case when i.rednew=1 then 1 when i.redold=1 then 2 when i.settle=1 then 3 when i.effect=1 then 4 else 9 end) as state /*单据状态*/
,(case when i.rednew=1 then '红冲单' when i.redold=1 then '已红冲' when i.settle=1 then '已结算' when i.effect=1 then '已过账' else '草稿' end) as statetext
,'' as susername /* 操作人 */
,'' as accname /* 科目 */
from salebill i
left join coursecentersale d on d.tid=i.tid and d.billid=i.billid
left join customer c on c.tid=i.tid and c.custid=i.custid
left join store k on k.tid=i.tid and k.storeid=i.storeid
left join employee e on e.tid=i.tid and e.empid=i.empid
left join user u on u.tid=i.tid and u.userid=i.userid
where i.tid=260434 and (i.billtype = 5 or i.effect = 1)
and ('_billdate_f_'!='')
and ('_billdate_t_'!='')
and ('_sdate_f_'!='')
and ('_sdate_t_'!='')
and ('_udate_f_'!='')
and ('_udate_t_'!='')
and ('_cdate_f_'!='')
and ('_cdate_t_'!='')
and ('_billid_'!='') /*单据id*/
and ('_custid_'!='') /*客户ID*/
and ('_storeid_'!='') /*店仓ID*/
and ('_empid_'!='') /*业务员ID*/
and ('_custstop_'!='') /*客户是否停用*/
and (
(i.billno like concat('%','葡萄','%'))
or (i.remarks like concat('%','葡萄','%'))
or exists(select 1 from salebilldetail d where d.tid=260434 and d.billid=i.billid and ((d.remarks like concat('%','葡萄','%')) or (d.goodsremarks like concat('%','葡萄','%'))))
or exists(select 1 from customer c where c.tid=260434 and c.custid=i.custid and (c.custname like concat('%','葡萄','%')))
or exists(select 1 from goods g join salebilldetail d on d.tid=g.tid and d.goodsid=g.goodsid where d.tid=260434 and d.billid=i.billid and ((g.goodsname like concat('%','葡萄','%')) or (g.goodscode like concat('%','葡萄','%'))))
)
and i.rednew=0 /*单据列表不含红冲单*/
and i.billid not in (select billid from coursecenter_del t where t.tid=260434)
and ((i.settle=1 and i.effect=1 and i.redold=0 and i.rednew=0)) /*已结算*/
order by udate desc,billno desc
limit 0,100;
执行时间约 1.6
秒,使用的是 like
方式。
改成使用全文索引方式:
test_10
-- 测试10
select
i.billid
,if(0,0,i.qty) as qty
,if(0,0,i.goodstotal) as total
,if(0,0,i.chktotal) as selfchktotal
,if(0,0,i.distotal) as distotal
,if(0,0,i.otherpay) as feetotal
,if(0,0,ifnull(d.costtotal,0)) as costtotal
,if(0,0,ifnull(d.maoli,0)) as maoli
,i.billno
,from_unixtime(i.billdate,'%Y-%m-%d') as billdate /*单据日期*/
,from_unixtime(i.createdate,'%Y-%m-%d %H:%i:%s') as createdate /*制单日期*/
,if(i.sdate=0,'',from_unixtime(i.sdate,'%Y-%m-%d %H:%i:%s')) as sdate /*过账日期*/
,from_unixtime(i.udate,'%Y-%m-%d %H:%i:%s') as udate /*最后修改时间*/
,i.custid ,c.custname
,i.storeid ,k.storename
,i.empid ,e.empname
,i.userid ,u.username
,i.remarks /*单据备注*/
,i.effect,i.settle,i.redold,i.rednew /*单据状态*/
,i.printtimes /* 打印次数 */
,(case when i.rednew=1 then 1 when i.redold=1 then 2 when i.settle=1 then 3 when i.effect=1 then 4 else 9 end) as state /*单据状态*/
,(case when i.rednew=1 then '红冲单' when i.redold=1 then '已红冲' when i.settle=1 then '已结算' when i.effect=1 then '已过账' else '草稿' end) as statetext
,'' as susername /* 操作人 */
,'' as accname /* 科目 */
from salebill i
left join coursecentersale d on d.tid=i.tid and d.billid=i.billid
left join customer c on c.tid=i.tid and c.custid=i.custid
left join store k on k.tid=i.tid and k.storeid=i.storeid
left join employee e on e.tid=i.tid and e.empid=i.empid
left join user u on u.tid=i.tid and u.userid=i.userid
where i.tid=260434 and (i.billtype = 5 or i.effect = 1)
and ('_billdate_f_'!='')
and ('_billdate_t_'!='')
and ('_sdate_f_'!='')
and ('_sdate_t_'!='')
and ('_udate_f_'!='')
and ('_udate_t_'!='')
and ('_cdate_f_'!='')
and ('_cdate_t_'!='')
and ('_billid_'!='') /*单据id*/
and ('_custid_'!='') /*客户ID*/
and ('_storeid_'!='') /*店仓ID*/
and ('_empid_'!='') /*业务员ID*/
and ('_custstop_'!='') /*客户是否停用*/
and (
(match(i.billno) against(concat('"','葡萄','"') in boolean mode))
or (match(i.remarks) against(concat('"','葡萄','"') in boolean mode))
or exists(select 1 from salebilldetail d where d.tid=260434 and d.billid=i.billid and ((match(d.remarks) Against(concat('"','葡萄','"') in boolean mode)) or (match(d.goodsremarks) Against(concat('"','葡萄','"') in boolean mode))))
or exists(select 1 from customer c where c.tid=260434 and c.custid=i.custid and (match(c.custname) Against(concat('"','葡萄','"') in boolean mode)))
or exists(select 1 from goods g join salebilldetail d on d.tid=g.tid and d.goodsid=g.goodsid where d.tid=260434 and d.billid=i.billid
and ((match(g.goodsname) Against(concat('"','葡萄','"') in boolean mode))
or (match(g.goodscode) Against(concat('"','葡萄','"') in boolean mode))))
)
and i.rednew=0 /*单据列表不含红冲单*/
and i.billid not in (select billid from coursecenter_del t where t.tid=260434)
and ((i.settle=1 and i.effect=1 and i.redold=0 and i.rednew=0)) /*已结算*/
order by udate desc,billno desc
limit 0,100;
执行时间约 1.6
秒,与使用的是 like
方式差不多。
最魔幻的地方来了,如果将上面的SQL语句中(salebilldetail
表使用全文索引 remarks_fulltext
、goodsremarks_fulltext
的地方)
exists(select 1 from salebilldetail d where d.tid=260434 and d.billid=i.billid and ((match(d.remarks) Against(concat('"','葡萄','"') in boolean mode)) or (match(d.goodsremarks) Against(concat('"','葡萄','"') in boolean mode))))
test_11
改成使用全文索引 remarks_goodsremarks_fulltext
-- 测试11
exists(select 1 from salebilldetail d where d.tid=260434 and d.billid=i.billid and ((match(d.remarks,d.goodsremarks) Against(concat('"','葡萄','"') in boolean mode))))
执行时间无限长(跑了半天没成功)?
经分析,在 where
子句中,一个条件子句中包含一个以上 match
时会出现这样的情况。即:
-- and 中只有一个全文检索时正常, 用时0.2秒
select xxx from xxx
...
and (
exists(select 1 from salebilldetail d where d.tid=260434 and d.billid=i.billid and ((match(d.remarks,d.goodsremarks) Against(concat('"','葡萄','"') in boolean mode))))
)
...
-- 下面这样就异常了,会慢成百上千倍,用时 160 秒, 如果有更多的 match ,会更夸张的慢下去
select xxx from xxx
...
and (
exists(select 1 from salebilldetail d where d.tid=260434 and d.billid=i.billid and ((match(d.remarks,d.goodsremarks) Against(concat('"','葡萄','"') in boolean mode))))
or match(i.billno) against(concat('"','葡萄','"') in boolean mode)
)
...
测试结果汇总:
查询 |
用时(秒) |
备注 |
test 1 |
0.765 |
原始like 查询 |
test 2 |
0.834 |
全文索引 remarks_fulltext 、goodsremarks_fulltext
|
test 3 |
0.242 |
全文索引 remarks_goodsremarks_fulltext
|
--- |
|
|
test 4 |
22.654 |
原始like 查询,不过滤 tid |
test 5 |
24.855 |
全文索引 remarks_fulltext 、goodsremarks_fulltext , 不过滤 tid |
test 6 |
0.213 |
全文索引 remarks_goodsremarks_fulltext , 不过滤 tid |
--- |
|
|
test 7 |
0.22 |
全文索引 remarks_goodsremarks_fulltext , count |
test 8 |
0.007 |
全文索引 remarks_goodsremarks_fulltext , 不过滤 tid, count |
--- |
|
|
test 9 |
1.6 |
业务测试SQL,原始like 查询 |
test 10 |
1.6 |
业务测试SQL,全文索引 remarks_fulltext 、goodsremarks_fulltext
|
test 11 |
失败 |
业务测试SQL,全文索引 remarks_goodsremarks_fulltext
|
五、MySQL 版本升级
因线上系统目前是 RDS MySQL 5.6,故简单描述升级相关问题。
-
Group By: 在 MySQL 5.7 之后,默认使用增加了限制,一些在 MySQL 5.6 可执行的Group By语句,在 5.7 之后会报错,可以更改新版本 MySQL 的 sqlModel
-- 查询 sql_mode
select @@SESSION.sql_mode;
-- 设置
SET GLOBAL sql_mode = 'STRICT_TRANS_TABLES,NO_ZERO_IN_DATE,NO_ZERO_DATE,ERROR_FOR_DIVISION_BY_ZERO,NO_ENGINE_SUBSTITUTION';
-- 或 设置 (修改于当前会 话,关闭当前会话后失效)
SET SESSION sql_mode = 'STRICT_TRANS_TABLES,NO_ZERO_IN_DATE,NO_ZERO_DATE,ERROR_FOR_DIVISION_BY_ZERO,NO_ENGINE_SUBSTITUTION';
-- 刷新
flush PRIVILEGES;
ONLY_FULL_GROUP_BY: 对于GROUP BY
聚合操作,如果在SELECT
中的列,没有在GROUP BY
中出现,那么这个SQL是不合法的,因为列不在GROUP BY
从句中
NO_AUTO_VALUE_ON_ZERO: 该值影响自增长列的插入。默认设置下,插入0
或NULL
代表生成下一个自增长值。如果用户希望插入的值为0
,而该列又是自增长的,那么这个选项就有用了。
STRICT_TRANS_TABLES:在该模式下,如果一个值不能插入到一个事务中,则中断当前的操作,对非事务表不做限制
NO_ZERO_IN_DATE:在严格模式下,不允许日期和月份为零
NO_ZERO_DATE:设置该值,mysql数据库不允许插入零日期,插入零日期会抛出错误而不是警告
ERROR_FOR_DIVISION_BY_ZERO:在insert
或update
过程中,如果数据被零除,则产生错误而非警告。如果未给出该模式,那么数据被零除时MySql返回NULL
NO_AUTO_CREATE_USER: 禁止GRANT
创建密码为空的用户
NO_ENGINE_SUBSTITUTION:如果需要的存储引擎被禁用或未编译,那么抛出错误。不设置此值时,用默认的存储引擎替代,并抛出一个异常
PIPES_AS_CONCAT:将"||
"视为字符串的连接操作符而非或运算符,这和Oracle数据库是一样是,也和字符串的拼接函数Concat想类似
ANSI_QUOTES:启用后,不能用双引号来引用字符串,因为它被解释为识别符
方式2:在配置文件中添加 sql_mode = '对应需要的模式'
sql_mode 模式说明:
方式1:重启 MySQL 后失效
-
MySQL8.0 修改了账号密码加密策略 (默认的认证插件由mysql_native_password
更改为caching_sha2_password
),导致一些可视化软件无法连接 mysql8.0 版本的数据库。如果需要,可以修改默认的策略或者账号密码的认证策略
[mysqld]
default_authentication_plugin = mysql_native_password
-- 修改加密规则
ALTER USER 'root'@'localhost' IDENTIFIED BY 'password' PASSWORD EXPIRE NEVER;
-- 更新用户密码
ALTER USER '账号'@'%' IDENTIFIED WITH mysql_native_password BY '密码';
-- 刷新权限
FLUSH PRIVILEGES;
-
MySQL8.0 授权用户账号语法变更,创建用户的操作已经不支持grant
的同时创建用户方式,需要先创建用户再进行授权。
-- 原来的流程:
mysql> grant all on *.* to 'admin'@'%' identified by 'admin';
-- 新的正确流程:
mysql> create user 'admin'@'%' identified by 'admin';
mysql> grant all on *.* to 'admin'@'%' ;
mysql> flush privileges;
-
数据库连接区别
jdbc:mysql://{ip}:{port}/{db}?characterEncoding=utf8&useSSL=false&serverTimezone=UTC
// useSSL 如果不配置false 项目可以正常启动但是会提示ssl问题
// serverTimezone=UTC 必须配置【时区设置成自己对应的时区】否则项目会报错
show variables like '%time_zone%';
set global time_zone='+8:00';
MySQL 5.7 原生支持JSON类型,并引入了众多JSON函数
MySQL 8.0 JSON字段的部分更新(JSON Partial Updates)
MySQL 8.0 默认字符集由latin1修改为utf8mb4
MySQL 8.0 正则表达式的增强,新增了4个相关函数,REGEXP_INSTR()
,REGEXP_LIKE()
,REGEXP_REPLACE()
,REGEXP_SUBSTR()
MySQL 8.0 GROUP BY语句不再隐式排序 (忽略在Group By中的排序命令,如 desc, asc)