MySQL 开发组于 2019 年 10 月 14 日 正式发布了 MySQL 8.0.18 GA 版本,带来了一些新特性和增强功能。其中最引人注目的莫过于多表连接查询支持 hash join 方式了。我们先来看看官方的描述:
MySQL 实现了用于内连接查询的 hash join 方式。例如,从 MySQL 8.0.18 开始以下查询可以使用 hash join 进行连接查询:
SELECT *
FROM t1
JOIN t2
ON t1.c1=t2.c1;
Hash join 不需要索引的支持。大多数情况下,hash join 比之前的 Block Nested-Loop 算法在没有索引时的等值连接更加高效。使用以下语句创建三张测试表:
CREATE TABLE t1 (c1 INT, c2 INT);
CREATE TABLE t2 (c1 INT, c2 INT);
CREATE TABLE t3 (c1 INT, c2 INT);
使用EXPLAIN FORMAT=TREE
命令可以看到执行计划中的 hash join,例如:
mysql> EXPLAIN FORMAT=TREE
-> SELECT *
-> FROM t1
-> JOIN t2
-> ON t1.c1=t2.c1\G
*************************** 1. row ***************************
EXPLAIN: -> Inner hash join (t2.c1 = t1.c1) (cost=0.70 rows=1)
-> Table scan on t2 (cost=0.35 rows=1)
-> Hash
-> Table scan on t1 (cost=0.35 rows=1)
必须使用 EXPLAIN 命令的 FORMAT=TREE 选项才能看到节点中的 hash join。另外,EXPLAIN ANALYZE
命令也可以显示 hash join 的使用信息。这也是该版本新增的一个功能。
多个表之间使用等值连接的的查询也会进行这种优化。例如以下查询:
SELECT *
FROM t1
JOIN t2
ON (t1.c1 = t2.c1 AND t1.c2 < t2.c2)
JOIN t3
ON (t2.c1 = t3.c1);
在以上示例中,任何其他非等值连接的条件将会在连接操作之后作为过滤器使用。可以通过EXPLAIN FORMAT=TREE
命令的输出进行查看:
mysql> EXPLAIN FORMAT=TREE
-> SELECT *
-> FROM t1
-> JOIN t2
-> ON (t1.c1 = t2.c1 AND t1.c2 < t2.c2)
-> JOIN t3
-> ON (t2.c1 = t3.c1)\G
*************************** 1. row ***************************
EXPLAIN: -> Inner hash join (t3.c1 = t1.c1) (cost=1.05 rows=1)
-> Table scan on t3 (cost=0.35 rows=1)
-> Hash
-> Filter: (t1.c2 < t2.c2) (cost=0.70 rows=1)
-> Inner hash join (t2.c1 = t1.c1) (cost=0.70 rows=1)
-> Table scan on t2 (cost=0.35 rows=1)
-> Hash
-> Table scan on t1 (cost=0.35 rows=1)
从以上输出同样可以看出,包含多个等值连接条件的查询也可以(会)使用多个 hash join 连接。
但是,如果任何连接语句(ON)中没有使用等值连接条件,将不会采用 hash join 连接方式。例如:
mysql> EXPLAIN FORMAT=TREE
-> SELECT *
-> FROM t1
-> JOIN t2
-> ON (t1.c1 = t2.c1)
-> JOIN t3
-> ON (t2.c1 < t3.c1)\G
*************************** 1. row ***************************
EXPLAIN: <not executable by iterator executor>
此时,将会采用性能更慢的 block nested loop 连接算法。这与 MySQL 8.0.18 之前版本中没有索引时的情况一样:
mysql> EXPLAIN
-> SELECT *
-> FROM t1
-> JOIN t2
-> ON (t1.c1 = t2.c1)
-> JOIN t3
-> ON (t2.c1 < t3.c1)\G
*************************** 1. row ***************************
id: 1
select_type: SIMPLE
table: t1
partitions: NULL
type: ALL
possible_keys: NULL
key: NULL
key_len: NULL
ref: NULL
rows: 1
filtered: 100.00
Extra: NULL
*************************** 2. row ***************************
id: 1
select_type: SIMPLE
table: t2
partitions: NULL
type: ALL
possible_keys: NULL
key: NULL
key_len: NULL
ref: NULL
rows: 1
filtered: 100.00
Extra: Using where; Using join buffer (Block Nested Loop)
*************************** 3. row ***************************
id: 1
select_type: SIMPLE
table: t3
partitions: NULL
type: ALL
possible_keys: NULL
key: NULL
key_len: NULL
ref: NULL
rows: 1
filtered: 100.00
Extra: Using where; Using join buffer (Block Nested Loop)
Hash join 连接同样适用于不指定查询条件时的笛卡尔积(Cartesian product),例如:
mysql> EXPLAIN FORMAT=TREE
-> SELECT *
-> FROM t1
-> JOIN t2
-> WHERE t1.c2 > 50\G
*************************** 1. row ***************************
EXPLAIN: -> Inner hash join (cost=0.70 rows=1)
-> Table scan on t2 (cost=0.35 rows=1)
-> Hash
-> Filter: (t1.c2 > 50) (cost=0.35 rows=1)
-> Table scan on t1 (cost=0.35 rows=1)
默认配置时,MySQL 所有可能的情况下都会使用 hash join。同时提供了两种控制是否使用 hash join 的方法:
可以通过系统变量 join_buffer_size 控制 hash join 允许使用的内存数量;hash join 不会使用超过该变量设置的内存数量。如果 hash join 所需的内存超过该阈值,MySQL 将会在磁盘中执行操作。需要注意的是,如果 hash join 无法在内存中完成,并且打开的文件数量超过系统变量 open_files_limit 的值,连接操作可能会失败。为了解决这个问题,可以使用以下方法之一:
接下来我们比较一下 hash join 和 block nested loop 的性能,首先分别为 t1、t2 和 t3 生成 1000000 条记录:
set join_buffer_size=2097152000;
SET @@cte_max_recursion_depth = 99999999;
INSERT INTO t1
-- INSERT INTO t2
-- INSERT INTO t3
WITH RECURSIVE t AS (
SELECT 1 AS c1, 1 AS c2
UNION ALL
SELECT t.c1 + 1, t.c1 * 2
FROM t
WHERE t.c1 < 1000000
)
SELECT *
FROM t;
没有索引情况下的 hash join:
mysql> EXPLAIN ANALYZE
-> SELECT COUNT(*)
-> FROM t1
-> JOIN t2
-> ON (t1.c1 = t2.c1)
-> JOIN t3
-> ON (t2.c1 = t3.c1)\G
*************************** 1. row ***************************
EXPLAIN: -> Aggregate: count(0) (actual time=22993.098..22993.099 rows=1 loops=1)
-> Inner hash join (t3.c1 = t1.c1) (cost=9952535443663536.00 rows=9952435908880402) (actual time=14489.176..21737.032 rows=1000000 loops=1)
-> Table scan on t3 (cost=0.00 rows=998412) (actual time=0.103..3973.892 rows=1000000 loops=1)
-> Hash
-> Inner hash join (t2.c1 = t1.c1) (cost=99682753413.67 rows=99682653660) (actual time=5663.592..12236.984 rows=1000000 loops=1)
-> Table scan on t2 (cost=0.01 rows=998412) (actual time=0.067..3364.105 rows=1000000 loops=1)
-> Hash
-> Table scan on t1 (cost=100539.40 rows=998412) (actual time=0.133..3395.799 rows=1000000 loops=1)
1 row in set (23.22 sec)
mysql> SELECT COUNT(*)
-> FROM t1
-> JOIN t2
-> ON (t1.c1 = t2.c1)
-> JOIN t3
-> ON (t2.c1 = t3.c1);
+----------+
| COUNT(*) |
+----------+
| 1000000 |
+----------+
1 row in set (12.98 sec)
实际运行花费了 12.98 秒。这个时候如果使用 block nested loop:
mysql> EXPLAIN FORMAT=TREE
-> SELECT /*+ NO_HASH_JOIN(t1, t2, t3) */ COUNT(*)
-> FROM t1
-> JOIN t2
-> ON (t1.c1 = t2.c1)
-> JOIN t3
-> ON (t2.c1 = t3.c1)\G
*************************** 1. row ***************************
EXPLAIN: <not executable by iterator executor>
1 row in set (0.00 sec)
SELECT /*+ NO_HASH_JOIN(t1, t2, t3) */ COUNT(*)
FROM t1
JOIN t2
ON (t1.c1 = t2.c1)
JOIN t3
ON (t2.c1 = t3.c1);
EXPLAIN 显示无法使用 hash join。查询跑了几十分钟也没有出结果,其中一个 CPU 使用率到了 100%;因为一直在执行嵌套循环(1000000 的 3 次方)。
再看有索引时的 block nested loop 方法,增加索引:
mysql> CREATE index idx1 ON t1(c1);
Query OK, 0 rows affected (7.39 sec)
Records: 0 Duplicates: 0 Warnings: 0
mysql> CREATE index idx2 ON t2(c1);
Query OK, 0 rows affected (6.77 sec)
Records: 0 Duplicates: 0 Warnings: 0
mysql> CREATE index idx3 ON t3(c1);
Query OK, 0 rows affected (7.23 sec)
Records: 0 Duplicates: 0 Warnings: 0
查看执行计划并运行相同的查询语句:
mysql> EXPLAIN ANALYZE
-> SELECT COUNT(*)
-> FROM t1
-> JOIN t2
-> ON (t1.c1 = t2.c1)
-> JOIN t3
-> ON (t2.c1 = t3.c1)\G
*************************** 1. row ***************************
EXPLAIN: -> Aggregate: count(0) (actual time=47684.034..47684.035 rows=1 loops=1)
-> Nested loop inner join (cost=2295573.22 rows=998412) (actual time=0.116..46363.599 rows=1000000 loops=1)
-> Nested loop inner join (cost=1198056.31 rows=998412) (actual time=0.087..25788.696 rows=1000000 loops=1)
-> Filter: (t1.c1 is not null) (cost=100539.40 rows=998412) (actual time=0.050..5557.847 rows=1000000 loops=1)
-> Index scan on t1 using idx1 (cost=100539.40 rows=998412) (actual time=0.043..3253.769 rows=1000000 loops=1)
-> Index lookup on t2 using idx2 (c1=t1.c1) (cost=1.00 rows=1) (actual time=0.012..0.015 rows=1 loops=1000000)
-> Index lookup on t3 using idx3 (c1=t1.c1) (cost=1.00 rows=1) (actual time=0.012..0.015 rows=1 loops=1000000)
1 row in set (47.68 sec)
mysql> SELECT COUNT(*)
-> FROM t1
-> JOIN t2
-> ON (t1.c1 = t2.c1)
-> JOIN t3
-> ON (t2.c1 = t3.c1);
+----------+
| COUNT(*) |
+----------+
| 1000000 |
+----------+
1 row in set (19.56 sec)
实际运行花费了 19.56 秒。所以在我们这个场景中的测试结果如下:
Hash Join(无索引) | Block Nested Loop(无索引) | Block Nested Loop(有索引) |
---|---|---|
12.98 s | 未返回 | 19.56 s |
再增加一个 Oracle 12c 中无索引时 hash join 结果:1.282 s。
再增加一个 PostgreSQL 11.5 中无索引时 hash join 结果:6.234 s。
再增加一个 SQL 2017 中无索引时 hash join 结果:5.207 s。