【GreatSQL优化器-06】条件过滤导致选择非最佳
一、condition_fanout_filter导致计划非最佳
GreatSQL 的优化器对于 join 的表需要根据行数和 cost 来确定最后哪张表先执行哪张表后执行，这里面就涉及到预估满足条件的表数据，condition_fanout_filter会根据一系列方法计算出一个数据过滤百分比，这个比百分比就是 filtered 系数，这个值区间在[0,1]，值越小代表过滤效果越好。用这个系数乘以总的行数就可以得出最后需要扫描的表行数的数量，可以大幅节省开销和执行时间。
这个功能是由OPTIMIZER_SWITCH_COND_FANOUT_FILTER这个OPTIMIZER_SWITCH来控制的，默认是打开的。因此一般情况下不需要特意去关闭，但是如果遇到执行特别慢的一些情况可以考虑关闭。
下面用一个例子来说明condition_fanout_filter有可能导致选择错误的情况：
# 创建2张表，都只在第二列创建索引，其中t3的最后一列也创建一个索引。
CREATE TABLE t3 (ccc1 INT, ccc2 int,ccc3 datetime(6));
INSERT INTO t3 VALUES (1,2,'2021-03-25 16:44:00.123456'),(2,10,'2021-03-25 16:44:00.123456'),(3,4,'2022-03-25 16:44:00.123456'),(4,6,'2023-03-25 16:44:00.123456'),(null,7,'2024-03-25 16:44:00.123456'),(4,3,'2024-04-25 16:44:00.123456'),(null,8,'2025-03-25 16:44:00.123456'),(3,4,'2022-06-25 16:44:00.123456'),(5,4,'2021-11-25 16:44:00.123456');
CREATE TABLE t4 (d1 INT, d2 int, d3 varchar(100));
INSERT INTO t4 VALUES (1,2,'aa1'),(2,1,'bb1'),(2,3,'cc1'),(3,3,'cc1'),(4,2,'ff1'),(4,4,'ert'),(4,2,'f5fg'),(null,2,'ee'),(5,30,'cc1'),(5,4,'fcc1'),(4,10,'cc1'),(6,4,'ccd1'),(null,1,'fee'),(1,2,'aa1'),(2,1,'bb1'),(2,3,'cc1'),(3,3,'cc1'),(4,2,'ff1'),(4,4,'ert'),(4,2,'f5fg'),(null,2,'ee'),(5,30,'cc1'),(5,4,'fcc1'),(4,10,'cc1'),(6,4,'ccd1'),(null,1,'fee'),(1,2,'aa1'),(2,1,'bb1'),(2,3,'cc1'),(3,3,'cc1'),(4,2,'ff1'),(4,4,'ert'),(4,2,'f5fg'),(null,2,'ee'),(5,30,'cc1'),(5,4,'fcc1'),(4,10,'cc1'),(6,4,'ccd1'),(null,1,'fee');
CREATE INDEX idx3_2 ON t3(ccc2);
CREATE INDEX idx3_3 ON t3(ccc3);
CREATE INDEX idx4_2 ON t4(d2);
执行一个join命令，where条件涉及的列不含t4的索引列，但是包含t3的索引列。 首先查看条件过滤开启的情况，结果是t4先执行全表扫描，预估的扫描行数为39 * 33.33%=13行，而t3执行ref索引扫描，行数为1 * 11.11%=0.1行，总行数为2行
greatsql> EXPLAIN SELECT * FROM t4 join t3 ON t4.d1=t3.ccc1 and t4.d2=t3.ccc2 where t4.d1<5 and t3.ccc3 < '2023-11-15';
+----+-------------+-------+------------+------+---------------+--------+---------+-----------+------+----------+-------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+-------+------------+------+---------------+--------+---------+-----------+------+----------+-------------+
| 1 | SIMPLE | t4 | NULL | ALL | idx4_2 | NULL | NULL | NULL | 39 | 33.33 | Using where |
| 1 | SIMPLE | t3 | NULL | ref | idx3_2,idx3_3 | idx3_2 | 5 | db1.t4.d2 | 1 | 11.11 | Using where |
+----+-------------+-------+------------+------+---------------+--------+---------+-----------+------+----------+-------------+
接着查看条件过滤关闭的情况，结果是t3先执行范围扫描，预估的扫描行数为6 * 100%=6行，而t4执行ref索引扫描，行数为6 * 100%=6行，总行数为39行。
greatsql> EXPLAIN SELECT /*+ set_var(optimizer_switch='condition_fanout_filter=off') */ * FROM t4 join t3 ON t4.d1=t3.ccc1 and t4.d2=t3.ccc2 where t4.d1<5 and t3.ccc3 < '2023-11-15';
+----+-------------+-------+------------+-------+---------------+--------+---------+-------------+------+----------+------------------------------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+-------+------------+-------+---------------+--------+---------+-------------+------+----------+------------------------------------+
| 1 | SIMPLE | t3 | NULL | range | idx3_2,idx3_3 | idx3_3 | 9 | NULL | 6 | 100.00 | Using index condition; Using where |
| 1 | SIMPLE | t4 | NULL | ref | idx4_2 | idx4_2 | 5 | db1.t3.ccc2 | 6 | 100.00 | Using where |
+----+-------------+-------+------------+-------+---------------+--------+---------+-------------+------+----------+------------------------------------+
接下来关掉condition_fanout_filter然后强制用t4 & t3来连接，对比一下计算出来的cost实际达到多少。从下面2个结果可以看出，t4走了全表扫描实际的cost达到21.70，是估计值的2倍多。
greatsql> EXPLAIN FORMAT=TREE SELECT * FROM t4 join t3 ON t4.d1=t3.ccc1 and t4.d2=t3.ccc2 where t4.d1<5 and t3.ccc3 < '2023-11-15';
+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| EXPLAIN |
+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| -> Nested loop inner join (cost=10.00 rows=2)
-> Filter: ((t4.d1 < 5) and (t4.d2 is not null)) (cost=4.15 rows=13)
-> Table scan on t4 (cost=4.15 rows=39)
-> Filter: ((t3.ccc1 = t4.d1) and (t3.ccc3 < TIMESTAMP'2023-11-15 00:00:00')) (cost=0.32 rows=0.1)
-> Index lookup on t3 using idx3_2 (ccc2=t4.d2) (cost=0.32 rows=1)
|
+----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
greatsql> EXPLAIN FORMAT=TREE SELECT /*+ set_var(optimizer_switch='condition_fanout_filter=off') qb_name(qb1) JOIN_ORDER(@qb1 t4,t3) */ * FROM t4 join t3 ON t4.d1=t3.ccc1 and t4.d2=t3.ccc2 where t4.d1<5 and t3.ccc3 < '2023-11-15';
+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| EXPLAIN |
+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
| -> Nested loop inner join (cost=21.70 rows=50)
-> Filter: ((t4.d1 < 5) and (t4.d2 is not null)) (cost=4.15 rows=39)
-> Table scan on t4 (cost=4.15 rows=39)
-> Filter: ((t3.ccc1 = t4.d1) and (t3.ccc3 < TIMESTAMP'2023-11-15 00:00:00')) (cost=0.32 rows=1)
-> Index lookup on t3 using idx3_2 (ccc2=t4.d2) (cost=0.32 rows=1)
|
+---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
以上例子因为condition_fanout_filter的设置不同而导致选择了不同的驱动表，最后的扫描行为也不一样。但是明显先执行t3的索引范围扫描比t4的全表扫描效率高，因此这个例子可以看出condition_fanout_filter的预估过滤百分比有更多主观性，最终可能导致错误的优化路径。
附表：join_type访问方法的类型
| join_type访问方法的类型 |
说明 |
| JT_UNKNOWN |
无效 |
| JT_SYSTEM |
表只有一行，比如select * from (select 1) |
| JT_CONST |
表最多只有一行满足，比如WHERE table.pk = 3 |
| JT_EQ_REF |
=符号用在唯一索引 |
| JT_REF |
=符号用在非唯一索引 |
| JT_ALL |
全表扫描 |
| JT_RANGE |
范围扫描 |
| JT_INDEX_SCAN |
索引扫描 |
| JT_FT |
Fulltext索引扫描 |
| JT_REF_OR_NULL |
包含null值，比如"WHERE col = ... OR col IS NULL |
| JT_INDEX_MERGE |
一张表执行多次范围扫描最后合并结果 |
以上各类扫描方式由快到慢排序为：system > const > eq_ref > ref > range > index > ALL
二、不关condition_fanout_filter的解决办法
如果不关闭condition_fanout_filter有没有办法强制指定连接顺序呢？答案是有的。一共如下3个方法，可以按照自己的需要进行灵活操作。
1、使用 qb_name 提示词来指定连接顺序
greatsql> EXPLAIN SELECT /*+ qb_name(qb1) JOIN_ORDER(@qb1 t3,t4) */ * FROM t4 join t3 ON t4.d1=t3.ccc1 and t4.d2=t3.ccc2 where t4.d1<5 and t3.ccc3 < '2023-11-15';
+----+-------------+-------+------------+-------+---------------+--------+---------+-------------+------+----------+------------------------------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+-------+------------+-------+---------------+--------+---------+-------------+------+----------+------------------------------------+
| 1 | SIMPLE | t3 | NULL | range | idx3_2,idx3_3 | idx3_3 | 9 | NULL | 6 | 100.00 | Using index condition; Using where |
| 1 | SIMPLE | t4 | NULL | ref | idx4_2 | idx4_2 | 5 | db1.t3.ccc2 | 6 | 3.33 | Using where |
+----+-------------+-------+------------+-------+---------------+--------+---------+-------------+------+----------+------------------------------------+
2、在所有 WHERE 条件列上建立索引
greatsql> CREATE INDEX idx4_1 ON t4(d1);
greatsql> EXPLAIN SELECT * FROM t4 join t3 ON t4.d1=t3.ccc1 and t4.d2=t3.ccc2 where t4.d1<5 and t3.ccc3 < '2023-11-15';
+----+-------------+-------+------------+-------+---------------+--------+---------+-------------+------+----------+------------------------------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+-------+------------+-------+---------------+--------+---------+-------------+------+----------+------------------------------------+
| 1 | SIMPLE | t3 | NULL | range | idx3_2,idx3_3 | idx3_3 | 9 | NULL | 6 | 100.00 | Using index condition; Using where |
| 1 | SIMPLE | t4 | NULL | ref | idx4_2,idx4_1 | idx4_1 | 5 | db1.t3.ccc1 | 5 | 16.67 | Using where |
+----+-------------+-------+------------+-------+---------------+--------+---------+-------------+------+----------+------------------------------------+
3、用 JOIN_FIXED_ORDER hint 加上表顺序来强制连接顺序。
greatsql> EXPLAIN SELECT /*+ qb_name(qb1) JOIN_FIXED_ORDER(@qb1) */ * FROM t3 join t4 ON t4.d1=t3.ccc1 AND t4.d2=t3.ccc2 WHERE t4.d1<5 AND t3.ccc3 < '2023-11-15';
+----+-------------+-------+------------+-------+---------------+--------+---------+-------------+------+----------+------------------------------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+-------+------------+-------+---------------+--------+---------+-------------+------+----------+------------------------------------+
| 1 | SIMPLE | t3 | NULL | range | idx3_2,idx3_3 | idx3_3 | 9 | NULL | 6 | 100.00 | Using index condition; Using where |
| 1 | SIMPLE | t4 | NULL | ref | idx4_2 | idx4_2 | 5 | db1.t3.ccc2 | 6 | 3.33 | Using where |
+----+-------------+-------+------------+-------+---------------+--------+---------+-------------+------+----------+------------------------------------+
三、如何排查类似问题
从以上例子看出，打开条件过滤有时并不总是能提高性能，优化器可能会高估条件过滤的影响，个别场景下使用条件过滤反而会导致性能下降。GreatSQL的condition_fanout_filter参数默认是打开的，因此需要自己来判断是否需要这个功能。一般来说，遇到以下场景需要特别注意条件过滤错估的情况。
| 情况 |
解决办法 |
| join连接表有大表，并且条件列没有索引 |
join连接的字段如果没有索引，应当先加上索引，以便优化器能够掌握字段值的分布情况，更准确的预估行数。 |
| join表有特别大的表和小表 |
判断表的join顺序是否合适，通过改变表的join顺序，让更小的表作为驱动表。可以考虑使用hint，强制优化器使用指定的表join顺序。 |
| 运行sql前先使用explain提前查看执行计划，判断条件过滤结果是否合理 |
如果不使用条件过滤，性能会更好，那么可以关闭会话级条件过滤功能。 |
四、总结
这节用了一个例子展示了条件过滤误判的情况，知道了打开条件过滤有时并不总是能提高性能，优化器可能会高估条件过滤的影响，个别场景下使用条件过滤反而会导致性能下降。GreatSQL的condition_fanout_filter参数默认是打开的，因此需要自己来判断是否需要这个功能。
