PostgreSQL利用递归优化求稀疏列唯一值代码方法

作者:袖梨 2022-06-29

本篇文章小编给大家分享一下PostgreSQL利用递归优化求稀疏列唯一值代码方法,文章代码介绍的很详细,小编觉得挺不错的,现在分享给大家供大家参考,有需要的小伙伴们可以来看看。

在数据库中经常会碰到一些表的列是稀疏列,只有很少的值,例如性别字段,一般就只有2种不同的值。

但是当我们求这些稀疏列的唯一值时,如果表的数据量很大,速度还是会很慢。

例如:

创建测试表

bill=# create table t_sex (sex char(1), otherinfo text);
CREATE TABLE
bill=# insert into t_sex select 'm', generate_series(1,10000000)||'this is test';
INSERT 0 10000000
bill=# insert into t_sex select 'w', generate_series(1,10000000)||'this is test';
INSERT 0 10000000

查询:

可以看到下面的查询速度很慢。

bill=# select count(distinct sex) from t_sex;
 count
-------
   2
(1 row)

Time: 8803.505 ms (00:08.804)
bill=# select sex from t_sex t group by sex;
 sex
-----
 m
 w
(2 rows)

Time: 1026.464 ms (00:01.026)

那么我们对该字段加上索引又是什么情况呢?

速度依然没有明显

bill=# create index idx_sex_1 on t_sex(sex);
CREATE INDEX
bill=# select count(distinct sex) from t_sex;
 count
-------
   2
(1 row)

Time: 8502.460 ms (00:08.502)
bill=# select sex from t_sex t group by sex;
 sex
-----
 m
 w
(2 rows)

Time: 572.353 ms

的变化,可以看到执行计划已经使用Index Only Scan了。

bill=# explain select count(distinct sex) from t_sex;
                     QUERY PLAN
----------------------------------------------------------------------------------------------
 Aggregate (cost=371996.44..371996.45 rows=1 )
  -> Index Only Scan using idx_sex_1 on t_sex (cost=0.44..321996.44 rows=20000000 )
(2 rows)

同样的SQL我们看看在Oracle中性能如何?

创建测试表:

SQL> create table t_sex (sex char(1), otherinfo varchar2(100));

Table created.

SQL> insert into t_sex select 'm', rownum||'this is test' from dual connect by level <=10000000;

10000000 rows created.

SQL> commit;

Commit complete.

SQL> insert into t_sex select 'w', rownum||'this is test' from dual connect by level <=10000000;

10000000 rows created.

SQL> commit;

Commit complete.

性能测试:

SQL> set lines 1000 pages 2000
SQL> set autotrace on
SQL> set timing on

SQL> select count(distinct sex) from t_sex;

COUNT(DISTINCTSEX)
------------------
         2

Elapsed: 00:00:01.58

Execution Plan
----------------------------------------------------------
Plan hash value: 3915432945

----------------------------------------------------------------------------
| Id | Operation     | Name | Rows | Bytes | Cost (%CPU)| Time   |
----------------------------------------------------------------------------
|  0 | SELECT STATEMENT  |    |   1 |   3 | 20132  (1)| 00:00:01 |
|  1 | SORT GROUP BY   |    |   1 |   3 |      |     |
|  2 |  TABLE ACCESS FULL| T_SEX |  14M|  42M| 20132  (1)| 00:00:01 |
----------------------------------------------------------------------------

Note
-----
  - dynamic statistics used: dynamic sampling (level=2)


Statistics
----------------------------------------------------------
     0 recursive calls
     0 db block gets
   74074 consistent gets
     0 physical reads
     0 redo size
    552 bytes sent via SQL*Net to client
    608 bytes received via SQL*Net from client
     2 SQL*Net roundtrips to/from client
     1 sorts (memory)
     0 sorts (disk)
     1 rows processed

SQL> select sex from t_sex t group by sex;

SE
--
m
w

Elapsed: 00:00:01.08

Execution Plan
----------------------------------------------------------
Plan hash value: 3915432945

----------------------------------------------------------------------------
| Id | Operation     | Name | Rows | Bytes | Cost (%CPU)| Time   |
----------------------------------------------------------------------------
|  0 | SELECT STATEMENT  |    |  14M|  42M| 20558  (3)| 00:00:01 |
|  1 | SORT GROUP BY   |    |  14M|  42M| 20558  (3)| 00:00:01 |
|  2 |  TABLE ACCESS FULL| T_SEX |  14M|  42M| 20132  (1)| 00:00:01 |
----------------------------------------------------------------------------

Note
-----
  - dynamic statistics used: dynamic sampling (level=2)


Statistics
----------------------------------------------------------
     0 recursive calls
     0 db block gets
   74074 consistent gets
     0 physical reads
     0 redo size
    589 bytes sent via SQL*Net to client
    608 bytes received via SQL*Net from client
     2 SQL*Net roundtrips to/from client
     1 sorts (memory)
     0 sorts (disk)
     2 rows processed

可以看到Oracle的性能即使不加索引也明显比PostgreSQL中要好。

那么我们在PostgreSQL中是不是没办法继续优化了呢?这种情况我们利用pg中的递归语句结合索引可以大幅提升性能。

SQL改写:

bill=# with recursive tmp as (
bill(#  (
bill(#   select min(t.sex) as sex from t_sex t where t.sex is not null
bill(#  )
bill(#  union all
bill(#  (
bill(#   select (select min(t.sex) from t_sex t where t.sex > s.sex and t.sex is not null)
bill(#    from tmp s where s.sex is not null
bill(#  )
bill(# )
bill-# select count(distinct sex) from tmp;
 count
-------
   2
(1 row)

Time: 2.711 ms

查看执行计划:

bill=# explain with recursive tmp as (
bill(#  (
bill(#   select min(t.sex) as sex from t_sex t where t.sex is not null
bill(#  )
bill(#  union all
bill(#  (
bill(#   select (select min(t.sex) from t_sex t where t.sex > s.sex and t.sex is not null)
bill(#    from tmp s where s.sex is not null
bill(#  )
bill(# )
bill-# select count(distinct sex) from tmp;
                           QUERY PLAN
----------------------------------------------------------------------------------------------------------------------
 Aggregate (cost=53.62..53.63 rows=1 )
  CTE tmp
   -> Recursive Union (cost=0.46..51.35 rows=101 )
      -> Result (cost=0.46..0.47 rows=1 )
         InitPlan 3 (returns $1)
          -> Limit (cost=0.44..0.46 rows=1 )
             -> Index Only Scan using idx_sex_1 on t_sex t (cost=0.44..371996.44 rows=20000000 )
                Index Cond: (sex IS NOT NULL)
      -> WorkTable Scan on tmp s (cost=0.00..4.89 rows=10 )
         Filter: (sex IS NOT NULL)
  -> CTE Scan on tmp (cost=0.00..2.02 rows=101 )
(11 rows)

Time: 1.371 ms

可以看到执行时间从原先的8000ms降低到了2ms,提升了几千倍!

甚至对比Oracle,性能也是提升了很多。

但是需要注意的是:这种写法仅仅是针对稀疏列,换成数据分布广泛的字段,显然性能是下降的, 所以使用递归SQL不适合数据分布广泛的字段的group by或者count(distinct)操作。

相关文章

精彩推荐