PostgreSQL 11 新特性解读:支持并行哈希连接(Parallel Hash Joins)"

PostgreSQL 11 版本在并行方面得到增强,例如支持并行创建索引(Parallel Index Build)、并行哈希连接(Parallel Hash Join)、并行 CREATE TABLE .. AS等,上篇博客介绍了并行创建索引,本文介绍并行 Hash Join。

测试环境准备

创建大表t_big并插入5000万条数据。

CREATE TABLE t_big(
id int4,
name text,
create_time timestamp without time zone );

INSERT INTO t_big(id,name,create_time)
SELECT n, n|| '_test',clock_timestamp() FROM generate_series(1,50000000) n ;

创建小表t_small并插入800万条数据

CREATE TABLE t_small(id int4, name text);

INSERT INTO t_small(id,name)
SELECT n, n|| '_small' FROM generate_series(1,8000000) n ;

验证并行哈希连接

PostgreSQL 10 版本查看以下SQL执行计划,如下:

des=> EXPLAIN SELECT t_small.name
  FROM t_big JOIN t_small ON (t_big.id = t_small.id)
       AND t_small.id < 100;
                                      QUERY PLAN
--------------------------------------------------------------------------------------
 Gather  (cost=151870.58..661385.28 rows=4143 width=13)
   Workers Planned: 4
   ->  Hash Join  (cost=150870.58..659970.98 rows=1036 width=13)
         Hash Cond: (t_big.id = t_small.id)
         ->  Parallel Seq Scan on t_big  (cost=0.00..470246.58 rows=10358258 width=4)
         ->  Hash  (cost=150860.58..150860.58 rows=800 width=17)
               ->  Seq Scan on t_small  (cost=0.00..150860.58 rows=800 width=17)
                     Filter: (id < 100)
(8 rows)

PostgreSQL 11 版本查看以下SQL执行计划,如下:

francs=> EXPLAIN SELECT t_small.name
  FROM t_big JOIN t_small ON (t_big.id = t_small.id)
       AND t_small.id < 100;
                                       QUERY PLAN
-----------------------------------------------------------------------------------------
 Gather  (cost=76862.42..615477.60 rows=800 width=13)
   Workers Planned: 4
   ->  Parallel Hash Join  (cost=75862.42..614397.60 rows=200 width=13)
         Hash Cond: (t_big.id = t_small.id)
         ->  Parallel Seq Scan on t_big  (cost=0.00..491660.86 rows=12499686 width=4)
         ->  Parallel Hash  (cost=75859.92..75859.92 rows=200 width=17)
               ->  Parallel Seq Scan on t_small  (cost=0.00..75859.92 rows=200 width=17)
                     Filter: (id < 100)
(8 rows)

对比10版本的执行计划,不同之处为11版本走了 Parallel Hash Join,而 10 版本走的 Hash JoinParallel Hash Join 为 11 版本的新特性。

并行哈希连接性能测试

开启并行哈希连接相比不开启性能上有何变化?接着测试。

开启并行哈希连接

PostgreSQL 11 版本执行以下SQL,如下:

francs=> EXPLAIN ANALYZE SELECT t_small.name
  FROM t_big JOIN t_small ON (t_big.id = t_small.id)
       AND t_small.id < 100;
                                                                QUERY PLAN

------------------------------------------------------------------------------------------------------------------------------------------
 Gather  (cost=76862.42..615477.60 rows=800 width=13) (actual time=197.399..2738.010 rows=99 loops=1)
   Workers Planned: 4
   Workers Launched: 4
   ->  Parallel Hash Join  (cost=75862.42..614397.60 rows=200 width=13) (actual time=2222.347..2729.943 rows=20 loops=5)
         Hash Cond: (t_big.id = t_small.id)
         ->  Parallel Seq Scan on t_big  (cost=0.00..491660.86 rows=12499686 width=4) (actual time=0.038..1330.836 rows=10000000 loops=5)
         ->  Parallel Hash  (cost=75859.92..75859.92 rows=200 width=17) (actual time=191.484..191.484 rows=20 loops=5)
               Buckets: 1024  Batches: 1  Memory Usage: 40kB
               ->  Parallel Seq Scan on t_small  (cost=0.00..75859.92 rows=200 width=17) (actual time=152.436..191.385 rows=20 loops=5)
                     Filter: (id < 100)
                     Rows Removed by Filter: 1599980
 Planning Time: 0.183 ms
 Execution Time: 2738.068 ms
(13 rows)

以上SQL执行多次,取最快时间,执行时间为 2738.068 ms。

关闭并行哈希连接

会话级设置enable_parallel_hash参数为off表示关闭并行哈希连接,测试性能有何变化,如下。

francs=> set enable_parallel_hash = off;
SET

francs=> EXPLAIN ANALYZE SELECT t_small.name
  FROM t_big JOIN t_small ON (t_big.id = t_small.id)
       AND t_small.id < 100;
                                                                QUERY PLAN

------------------------------------------------------------------------------------------------------------------------------------------
 Gather  (cost=151869.66..690486.34 rows=800 width=13) (actual time=996.137..3496.940 rows=99 loops=1)
   Workers Planned: 4
   Workers Launched: 4
   ->  Hash Join  (cost=150869.66..689406.34 rows=200 width=13) (actual time=2990.847..3490.557 rows=20 loops=5)
         Hash Cond: (t_big.id = t_small.id)
         ->  Parallel Seq Scan on t_big  (cost=0.00..491660.86 rows=12499686 width=4) (actual time=0.240..1392.062 rows=10000000 loops=5)
         ->  Hash  (cost=150859.66..150859.66 rows=800 width=17) (actual time=890.943..890.943 rows=99 loops=5)
               Buckets: 1024  Batches: 1  Memory Usage: 13kB
               ->  Seq Scan on t_small  (cost=0.00..150859.66 rows=800 width=17) (actual time=884.288..890.906 rows=99 loops=5)
                     Filter: (id < 100)
                     Rows Removed by Filter: 7999901
 Planning Time: 0.154 ms
 Execution Time: 3496.982 ms
(13 rows)

以上SQL执行多次,取最快时间,从以上看出,关闭并行哈希连接时SQL的执行时间为 3496.982 ms ,相比开启并行哈希连接执行时间长了 27%。

可见开启并行哈希连接后,性能有较大幅度提升。

参考

新书推荐

最后推荐和张文升共同编写的《PostgreSQL实战》,本书基于PostgreSQL 10 编写,共18章,重点介绍SQL高级特性、并行查询、分区表、物理复制、逻辑复制、备份恢复、高可用、性能优化、PostGIS等,涵盖大量实战用例!

购买链接:https://item.jd.com/12405774.html
_5_PostgreSQL_

猜你喜欢

转载自yq.aliyun.com/articles/670734