MySQL优化一例

DELIMITER $$

USE `xxx`$$

DROP FUNCTION IF EXISTS `F_getBuluDates`$$

CREATE DEFINER=`root`@`localhost` FUNCTION `F_getBuluDates`(`PuserId` INT(11)) RETURNS VARCHAR(3000) CHARSET gbk
BEGIN
DECLARE buluDate,buluDateTime,LiWaibuluDates,nowdate,bldate VARCHAR(3000) DEFAULT "";
DECLARE rtMsg,RuserName,maxRiZhiDate,btjrqs,blyqrq,tmpdate,yxblsj VARCHAR(255) DEFAULT "";
DECLARE Ra51count INT(30);
DECLARE RbeginDate,RendDate,RtmpDate VARCHAR(2550);
DECLARE u,gqcount,a21Count INT(30) DEFAULT 0;
DECLARE allReadRows,readRows,mySYL,blCount,recordCount,btjryCount,btjrqCount,yxblts,i INT(11) DEFAULT 0;
SELECT a5204,a5205 INTO yxblts,yxblsj FROM a52 WHERE a5218="xxxxxxx" LIMIT 0,1;    

SELECT username INTO RuserName FROM USER WHERE id=PuserId;

SELECT COUNT(id) INTO btjryCount FROM a52 WHERE a5214=PuserId AND a5218='xxxxx';
IF btjryCount=0 THEN    

SELECT GROUP_CONCAT(a5215) INTO btjrqs FROM a52 WHERE  a5218='btjrq';
SET tmpdate=CURDATE();
SET nowdate=DATE_SUB(CURDATE(),INTERVAL yxblts DAY);

REPEAT
SELECT MAX(a2104) INTO maxRiZhiDate FROM a21 WHERE issubmit='y' AND a2104<tmpdate AND FIND_IN_SET(a2104,bldate)=0;
SET tmpdate=maxRiZhiDate;
IF tmpdate>=nowdate THEN
	IF F_isWorkDay(maxRiZhiDate)=1 OR ISNULL(maxRiZhiDate) THEN
		IF NOT ISNULL(maxRiZhiDate) AND LENGTH(maxRiZhiDate)>0 THEN
			IF FIND_IN_SET(maxRiZhiDate,btjrqs)=0 THEN
				SELECT COUNT(id) INTO recordCount FROM a21 WHERE a2104=maxRiZhiDate AND creatorid=PuserId;
				IF recordCount=0 THEN
					IF DAYOFWEEK(maxRiZhiDate)=6 THEN
						SET buluDateTime=CONCAT(FROM_DAYS(TO_DAYS(maxRiZhiDate)+2+yxblts)," ",yxblsj);
					ELSEIF DAYOFWEEK(maxRiZhiDate)=7 THEN
						SET buluDateTime=CONCAT(FROM_DAYS(TO_DAYS(maxRiZhiDate)+1+yxblts)," ",yxblsj);
					ELSE
						SET buluDateTime=CONCAT(FROM_DAYS(TO_DAYS(maxRiZhiDate)+yxblts)," ",yxblsj);
					END IF;

					IF(CURRENT_TIMESTAMP()<buluDateTime) THEN
						SET buluDate=maxRiZhiDate;
					END IF;
				END IF;
			END IF;
		END IF;  

	END IF;
	IF CHAR_LENGTH(buluDate)>0 THEN
		IF CHAR_LENGTH(bldate)>0 THEN
			SET bldate=CONCAT(buluDate,",",bldate);
		ELSE
			SET bldate=buluDate;
		END IF;
	END IF;
END IF;

SET buluDate='';
UNTIL tmpdate<=nowdate
END REPEAT;

SELECT COUNT(id) INTO gqcount FROM a52 WHERE a5216=CURDATE() AND a5218='xxxxxx';
IF gqcount>0 THEN
SELECT a5215 INTO blyqrq FROM a52 WHERE a5216=CURDATE() AND a5218='xxxxxx';
SELECT COUNT(id) INTO a21Count FROM a21 WHERE a2104=blyqrq AND creatorid=PuserId;
IF a21Count=0 THEN
IF FIND_IN_SET(blyqrq,bldate)=0 THEN
IF CHAR_LENGTH(bldate)>0 THEN
SET bldate=CONCAT(blyqrq,",",bldate);
ELSE
SET bldate=blyqrq;
END IF;
END IF;
END IF;
END IF;

IF ISNULL(btjrqs) THEN
SET btjrqs="";
END IF;

SELECT GROUP_CONCAT(a5108) INTO LiWaibuluDates FROM a51 WHERE a5110='y' AND F_ifInSet(a5108,btjrqs)=0 AND a5106 LIKE RuserName AND a5112='否' AND CONCAT(a5108," ",a5113)<CURRENT_TIMESTAMP() AND CURRENT_TIMESTAMP()<CONCAT(a5114," ",a5113);
IF NOT ISNULL(LiWaibuluDates) AND LENGTH(TRIM(LiWaibuluDates))>0 THEN
IF CHAR_LENGTH(bldate)>0 THEN
SET bldate=CONCAT(LiWaibuluDates,",",bldate);
ELSE
SET bldate=LiWaibuluDates;
END IF;
END IF;       

SELECT COUNT(id) INTO RA51count FROM a51 WHERE a5110='n' AND F_getid1value2(a5105,1)=PuserId;
IF RA51count>0 THEN
SET u=0;
WHILE u< RA51count DO
SELECT a5108,a5109 INTO RbeginDate,RendDate FROM a51 WHERE a5110='n' AND F_getid1value2(a5105,1)=PuserId ORDER BY id LIMIT u,1;
SET RtmpDate=RbeginDate;
WHILE RtmpDate<=RendDate DO

IF FIND_IN_SET(RtmpDate,bldate)>0 THEN
SET bldate=F_removeFromSet(RtmpDate,bldate);
END IF;
SET RtmpDate=DATE_ADD(RtmpDate, INTERVAL 1 DAY);
END WHILE;
SET u=u+1;
END WHILE;
END IF;

END IF;  

RETURN bldate;

END$$

DELIMITER ;

在老系统中该函数调用一次需要话20多秒到30秒左右。

拿到sql之后，首先要确定思路。不能着急这下手。

1. 首先查看各个表的数据量：

select count(*) from xxx;

发现只有 a21 的数据量达到了十几万，其他表数据量都比较小。所以重点是 a21表，仔细阅读了一遍函数的定义，发现涉及 到 a21 的有一处循环：

REPEAT
SELECT MAX(a2104) INTO maxRiZhiDate FROM a21 WHERE issubmit='y' AND a2104<tmpdate AND FIND_IN_SET(a2104,bldate)=0;

很显然使用了 find_in_set 函数，所以改语句无法使用索引。所以想要简单的通过增加索引来解决问题，应该是行不通的。

这样就定位到了函数运行慢的问题。结合业务理解改函数的含义。

为了使用索引，我们需要去掉 find_in_set 函数，理解了业务和原函数的功能之后，对改函数的 repeat 循环部分进行了重写：

DELIMITER $$

USE `oa`$$

DROP FUNCTION IF EXISTS `F_getBuluDates_inner`$$

CREATE DEFINER=`root`@`localhost` FUNCTION `F_getBuluDates_inner`(`PuserId` INT(11), tmpdate VARCHAR(32), nowdate VARCHAR(32)) RETURNS VARCHAR(3000) CHARSET gbk
BEGIN
DECLARE buluDate,buluDateTime,LiWaibuluDates,bldate VARCHAR(3000) DEFAULT "";
DECLARE rtMsg,RuserName,maxRiZhiDate,btjrqs,blyqrq,yxblsj VARCHAR(255) DEFAULT "";
DECLARE Ra51count INT(30);
DECLARE RbeginDate,RendDate,RtmpDate VARCHAR(2550);
DECLARE u,gqcount,a21Count INT(30) DEFAULT 0;
DECLARE allReadRows,readRows,mySYL,blCount,recordCount,btjryCount,btjrqCount,yxblts,i INT(11) DEFAULT 0;
DECLARE no_more_data INT DEFAULT 0;

DECLARE my_cursor CURSOR FOR SELECT DISTINCT a2104 FROM a21 WHERE issubmit='y' AND a2104 < tmpdate AND a2104 >= nowdate ORDER BY a2104 DESC;
DECLARE CONTINUE HANDLER FOR NOT FOUND SET no_more_data = 1;

SELECT a5204,a5205 INTO yxblts,yxblsj FROM a52 WHERE a5218="gzrz" LIMIT 0,1;    

SELECT username INTO RuserName FROM USER WHERE id=PuserId;

SELECT COUNT(id) INTO btjryCount FROM a52 WHERE a5214=PuserId AND a5218='xxxxx';
IF btjryCount=0 THEN    

SELECT GROUP_CONCAT(a5215) INTO btjrqs FROM a52 WHERE  a5218='xxx';   

OPEN my_cursor;
FETCH my_cursor INTO maxRiZhiDate;

REPEAT
    IF F_isWorkDay(maxRiZhiDate)=1 OR ISNULL(maxRiZhiDate) THEN
        IF NOT ISNULL(maxRiZhiDate) AND LENGTH(maxRiZhiDate)>0 THEN
            IF FIND_IN_SET(maxRiZhiDate,btjrqs)=0 THEN
                SELECT COUNT(id) INTO recordCount FROM a21 WHERE a2104=maxRiZhiDate AND creatorid=PuserId;
                IF recordCount=0 THEN
                    IF DAYOFWEEK(maxRiZhiDate)=6 THEN
                        SET buluDateTime=CONCAT(FROM_DAYS(TO_DAYS(maxRiZhiDate)+2+yxblts)," ",yxblsj);
                    ELSEIF DAYOFWEEK(maxRiZhiDate)=7 THEN
                        SET buluDateTime=CONCAT(FROM_DAYS(TO_DAYS(maxRiZhiDate)+1+yxblts)," ",yxblsj);
                    ELSE
                        SET buluDateTime=CONCAT(FROM_DAYS(TO_DAYS(maxRiZhiDate)+yxblts)," ",yxblsj);
                    END IF;

                    IF(CURRENT_TIMESTAMP()<buluDateTime) THEN
                        SET buluDate=maxRiZhiDate;
                    END IF;
                END IF;
            END IF;
        END IF;  

    END IF;
    IF CHAR_LENGTH(buluDate)>0 THEN
        IF CHAR_LENGTH(bldate)>0 THEN
            SET bldate=CONCAT(buluDate,",",bldate);
        ELSE
            SET bldate=buluDate;
        END IF;
    END IF;

FETCH my_cursor INTO maxRiZhiDate;
SET buluDate='';
UNTIL no_more_data = 1
END REPEAT;

SELECT COUNT(id) INTO gqcount FROM a52 WHERE a5216=CURDATE() AND a5218='xxx';
IF gqcount>0 THEN
SELECT a5215 INTO blyqrq FROM a52 WHERE a5216=CURDATE() AND a5218='xxxxxxxxxx';
SELECT COUNT(id) INTO a21Count FROM a21 WHERE a2104=blyqrq AND creatorid=PuserId;
IF a21Count=0 THEN
IF FIND_IN_SET(blyqrq,bldate)=0 THEN
IF CHAR_LENGTH(bldate)>0 THEN
SET bldate=CONCAT(blyqrq,",",bldate);
ELSE
SET bldate=blyqrq;
END IF;
END IF;
END IF;
END IF;

IF ISNULL(btjrqs) THEN
SET btjrqs="";
END IF;

SELECT GROUP_CONCAT(a5108) INTO LiWaibuluDates FROM a51 WHERE a5110='y' AND F_ifInSet(a5108,btjrqs)=0 AND a5106 LIKE RuserName AND a5112='否' AND CONCAT(a5108," ",a5113)<CURRENT_TIMESTAMP() AND CURRENT_TIMESTAMP()<CONCAT(a5114," ",a5113);
IF NOT ISNULL(LiWaibuluDates) AND LENGTH(TRIM(LiWaibuluDates))>0 THEN
IF CHAR_LENGTH(bldate)>0 THEN
SET bldate=CONCAT(LiWaibuluDates,",",bldate);
ELSE
SET bldate=LiWaibuluDates;
END IF;
END IF;       

SELECT COUNT(id) INTO RA51count FROM a51 WHERE a5110='n' AND F_getid1value2(a5105,1)=PuserId;
IF RA51count>0 THEN
SET u=0;
WHILE u< RA51count DO
SELECT a5108,a5109 INTO RbeginDate,RendDate FROM a51 WHERE a5110='n' AND F_getid1value2(a5105,1)=PuserId ORDER BY id LIMIT u,1;
SET RtmpDate=RbeginDate;
WHILE RtmpDate<=RendDate DO

IF FIND_IN_SET(RtmpDate,bldate)>0 THEN
SET bldate=F_removeFromSet(RtmpDate,bldate);
END IF;
SET RtmpDate=DATE_ADD(RtmpDate, INTERVAL 1 DAY);
END WHILE;
SET u=u+1;
END WHILE;
END IF;      

END IF;  

RETURN bldate;

END$$

DELIMITER ;

然后为了让新的 sql 使用索引：

DECLARE my_cursor CURSOR FOR

SELECT DISTINCT a2104 FROM a21 WHERE issubmit='y' AND a2104 < tmpdate AND a2104 >= nowdate ORDER BY a2104 DESC;

新增了索引：

add index issubmit_a2104(issubmit,a2104);

最终的提供给外部的调用函数：

DELIMITER $$
CREATE FUNCTION F_getBuluDates_new(`PuserId` INT(11)) RETURNS VARCHAR(3000) CHARSET gbk
BEGIN
    DECLARE bldate VARCHAR(3000) DEFAULT "";
    DECLARE tmpdate,nowdate VARCHAR(32) DEFAULT "";
    DECLARE yxblts INT;
    SELECT a5204 INTO yxblts FROM a52 WHERE a5218="gzrz" LIMIT 0,1;
    SET tmpdate = CURDATE();
    SET nowdate = DATE_SUB(tmpdate,INTERVAL yxblts DAY);
    SET bldate = F_getBuluDates_inner(PuserId,tmpdate,nowdate);
    RETURN bldate;
END
$$

测试效果：

mysql> select F_getBuluDates_new(10687);
+---------------------------------------------------------------------------------------------------------------+
| F_getBuluDates_new(10687)                                                                                    |
+---------------------------------------------------------------------------------------------------------------+
| 2016-08-04,2016-08-05,2016-08-08,2016-08-09,2016-08-10,2016-08-11,2016-08-12,2016-08-15,2016-08-16,2016-08-17 |
+---------------------------------------------------------------------------------------------------------------+
1 row in set (1.10 sec)

mysql> select F_getBuluDates(10687);
+---------------------------------------------------------------------------------------------------------------+
| F_getBuluDates(10687)                                                                                         |
+---------------------------------------------------------------------------------------------------------------+
| 2016-08-04,2016-08-05,2016-08-08,2016-08-09,2016-08-10,2016-08-11,2016-08-12,2016-08-15,2016-08-16,2016-08-17 |
+---------------------------------------------------------------------------------------------------------------+
1 row in set (13.10 sec)

性能提升了 11 倍多。客户对效果很满意。

新的执行计划：

mysql> explain SELECT DISTINCT a2104 FROM a21 WHERE issubmit='y' AND a2104 < '2016-08-16' AND a2104 >= '2016-08-10' ORDER BY a2104 DESC;
+----+-------------+-------+-------+----------------------+----------------+---------+------+------+--------------------------+
| id | select_type | table | type  | possible_keys        | key            | key_len | ref  | rows | Extra                    |
+----+-------------+-------+-------+----------------------+----------------+---------+------+------+--------------------------+
|  1 | SIMPLE      | a21   | range | a2104,issubmit_a2104 | issubmit_a2104 | 516     | NULL |  834 | Using where; Using index |
+----+-------------+-------+-------+----------------------+----------------+---------+------+------+--------------------------+
1 row in set (0.00 sec)

总结：

从原函数实现来看，在数量小时，还是可以的。但是一旦数据量大了，就会出现性能问题。主要问题是在 where 条件中使用了太多的自定义函数，而且这个函数的参数还是表的数据列。
导致无法使用索引。无法使用索引就会进行全表扫描，所以数据量大时会导致性能问题。