标签
PostgreSQL , cpu 并行 , smp 并行 , 并行计算 , gpu 并行 , 并行过程支持
背景
PostgreSQL 11 优化器已经支持了非常多场合的并行。简单估计，已支持27余种场景的并行计算。
parallel seq scan
parallel index scan
parallel index only scan
parallel bitmap scan
parallel filter
parallel hash agg
parallel group agg
parallel cte
parallel subquery
parallel create table
parallel create index
parallel select into
parallel CREATE MATERIALIZED VIEW
parallel 排序 : gather merge
parallel nestloop join
parallel hash join
parallel merge join
parallel 自定义并行聚合
parallel 自定义并行UDF
parallel append
parallel union
parallel fdw table scan
parallel partition join
parallel partition agg
parallel gather
parallel gather merge
parallel rc 并行
parallel rr 并行
parallel GPU 并行
parallel unlogged table
lead parallel
接下来进行一一介绍。
关键知识请先自行了解：
1、优化器自动并行度算法 CBO
《PostgreSQL 11 并行计算算法，参数，强制并行度设置》
parallel partition table wise join
并行分区表智能JOIN（类似MPP）
当两个JOIN的分区表JOIN字段类型一致，并且分区在JOIN字段上，并且分区类型一致（枚举、LIST、范围、HASH），并且分区个数一致。满足这些条件时，PostgreSQL优化器会选择并行分区智能JOIN，子分区各自JOIN子分区。
《PostgreSQL 11 preview - 分区表智能并行聚合、分组计算(已类似MPP架构，性能暴增)》
《PostgreSQL 11 preview - 分区表智能并行JOIN (已类似MPP架构，性能暴增)》
数据量：10亿 join 10亿
| 场景 | 数据量 | 关闭并行 | 开启并行 | 并行度 | 开启并行性能提升倍数 |
|---|---|---|---|---|---|
| parallel partition table wise join | 10亿 join 10亿 using (i) | 1006 秒 | 76 秒 | 24 | 13.2 倍 |
例子，24个分区的HASH分区表。
CREATE unlogged TABLE aaa (
order_id bigint not null,
cust_id bigint not null,
status text
) PARTITION BY HASH (order_id);
CREATE unlogged TABLE bbb (
order_id bigint not null,
cust_id bigint not null,
status text
) PARTITION BY HASH (order_id);
do language plpgsql $$
declare
begin
for i in 0..23 loop
execute format('CREATE unlogged TABLE %s%s PARTITION OF %s FOR VALUES WITH (MODULUS %s, REMAINDER %s)', 'aaa', i, 'aaa', 24, i);
execute format('CREATE unlogged TABLE %s%s PARTITION OF %s FOR VALUES WITH (MODULUS %s, REMAINDER %s)', 'bbb', i, 'bbb', 24, i);
execute format('alter table %s%s set(parallel_workers =64)', 'aaa',i);
execute format('alter table %s%s set(parallel_workers =64)', 'bbb',i);
end loop;
end;
$$;
postgres=# \d aaa
Unlogged table "public.aaa"
Column | Type | Collation | Nullable | Default
----------+--------+-----------+----------+---------
order_id | bigint | | not null |
cust_id | bigint | | not null |
status | text | | |
Partition key: HASH (order_id)
Number of partitions: 24 (Use \d+ to list them.)
postgres=# \d bbb
Unlogged table "public.bbb"
Column | Type | Collation | Nullable | Default
----------+--------+-----------+----------+---------
order_id | bigint | | not null |
cust_id | bigint | | not null |
status | text | | |
Partition key: HASH (order_id)
Number of partitions: 24 (Use \d+ to list them.)
postgres=# show max_worker_processes ;
max_worker_processes
----------------------
128
(1 row)
postgres=# set min_parallel_table_scan_size =0;
postgres=# set min_parallel_index_scan_size =0;
postgres=# set parallel_tuple_cost =0;
postgres=# set parallel_setup_cost =0;
postgres=# set max_parallel_workers=128;
postgres=# set max_parallel_workers_per_gather =24;
postgres=# set enable_parallel_hash =on;
postgres=# set enable_parallel_append =off;
postgres=# set enable_partitionwise_join =on;
postgres=# set work_mem ='128MB';
分别写入10亿数据
insert into aaa select i, random()*100000 from generate_series(1,1000000000) t(i);
insert into bbb select i, random()*100000 from generate_series(1,1000000000) t(i);
vacuum (analyze,verbose) aaa;
vacuum (analyze,verbose) bbb;
1、关闭并行，耗时： 1006 秒。
postgres=# set max_parallel_workers_per_gather =0;
postgres=# set enable_parallel_append =off;
postgres=# set enable_partitionwise_join =off;
postgres=# explain select count(*) from aaa join bbb using (order_id);
QUERY PLAN
-----------------------------------------------------------------------------------------------
Aggregate (cost=380800207.07..380800207.08 rows=1 width=8)
-> Merge Join (cost=355800197.59..378300206.45 rows=1000000248 width=0)
Merge Cond: (bbb0.order_id = aaa0.order_id)
-> Sort (cost=177900158.19..180400160.45 rows=1000000904 width=8)
Sort Key: bbb0.order_id
-> Append (cost=0.00..20405431.56 rows=1000000904 width=8)
-> Seq Scan on bbb0 (cost=0.00..641834.24 rows=41662924 width=8)
-> Seq Scan on bbb1 (cost=0.00..641805.76 rows=41661076 width=8)
-> Seq Scan on bbb2 (cost=0.00..641899.80 rows=41667180 width=8)
-> Seq Scan on bbb3 (cost=0.00..641919.76 rows=41668476 width=8)
-> Seq Scan on bbb4 (cost=0.00..641896.96 rows=41666996 width=8)
-> Seq Scan on bbb5 (cost=0.00..641854.20 rows=41664220 width=8)
-> Seq Scan on bbb6 (cost=0.00..641839.96 rows=41663296 width=8)
-> Seq Scan on bbb7 (cost=0.00..641894.08 rows=41666808 width=8)
-> Seq Scan on bbb8 (cost=0.00..641723.12 rows=41655712 width=8)
-> Seq Scan on bbb9 (cost=0.00..641754.88 rows=41657688 width=8)
-> Seq Scan on bbb10 (cost=0.00..641848.48 rows=41663848 width=8)
-> Seq Scan on bbb11 (cost=0.00..641771.56 rows=41658856 width=8)
-> Seq Scan on bbb12 (cost=0.00..642119.24 rows=41681424 width=8)
-> Seq Scan on bbb13 (cost=0.00..642059.68 rows=41677368 width=8)
-> Seq Scan on bbb14 (cost=0.00..641765.84 rows=41658484 width=8)
-> Seq Scan on bbb15 (cost=0.00..641703.16 rows=41654416 width=8)
-> Seq Scan on bbb16 (cost=0.00..641794.36 rows=41660336 width=8)
-> Seq Scan on bbb17 (cost=0.00..641842.80 rows=41663480 width=8)
-> Seq Scan on bbb18 (cost=0.00..642127.80 rows=41681980 width=8)
-> Seq Scan on bbb19 (cost=0.00..642048.00 rows=41676800 width=8)
-> Seq Scan on bbb20 (cost=0.00..641999.56 rows=41673656 width=8)
-> Seq Scan on bbb21 (cost=0.00..641971.04 rows=41671804 width=8)
-> Seq Scan on bbb22 (cost=0.00..641976.00 rows=41671900 width=8)
-> Seq Scan on bbb23 (cost=0.00..641976.76 rows=41672176 width=8)
-> Materialize (cost=177900039.40..182900040.64 rows=1000000248 width=8)
-> Sort (cost=177900039.40..180400040.02 rows=1000000248 width=8)
Sort Key: aaa0.order_id
-> Append (cost=0.00..20405421.72 rows=1000000248 width=8)
-> Seq Scan on aaa0 (cost=0.00..641834.24 rows=41662924 width=8)
-> Seq Scan on aaa1 (cost=0.00..641805.76 rows=41661076 width=8)
-> Seq Scan on aaa2 (cost=0.00..641899.80 rows=41667180 width=8)
-> Seq Scan on aaa3 (cost=0.00..641919.76 rows=41668476 width=8)
-> Seq Scan on aaa4 (cost=0.00..641896.96 rows=41666996 width=8)
-> Seq Scan on aaa5 (cost=0.00..641854.20 rows=41664220 width=8)
-> Seq Scan on aaa6 (cost=0.00..641839.96 rows=41663296 width=8)
-> Seq Scan on aaa7 (cost=0.00..641894.08 rows=41666808 width=8)
-> Seq Scan on aaa8 (cost=0.00..641723.12 rows=41655712 width=8)
-> Seq Scan on aaa9 (cost=0.00..641757.28 rows=41657928 width=8)
-> Seq Scan on aaa10 (cost=0.00..641848.48 rows=41663848 width=8)
-> Seq Scan on aaa11 (cost=0.00..641771.56 rows=41658856 width=8)
-> Seq Scan on aaa12 (cost=0.00..642119.24 rows=41681424 width=8)
-> Seq Scan on aaa13 (cost=0.00..642065.12 rows=41677912 width=8)
-> Seq Scan on aaa14 (cost=0.00..641757.00 rows=41657600 width=8)
-> Seq Scan on aaa15 (cost=0.00..641703.16 rows=41654416 width=8)
-> Seq Scan on aaa16 (cost=0.00..641794.36 rows=41660336 width=8)
-> Seq Scan on aaa17 (cost=0.00..641842.80 rows=41663480 width=8)
-> Seq Scan on aaa18 (cost=0.00..642127.80 rows=41681980 width=8)
-> Seq Scan on aaa19 (cost=0.00..642048.00 rows=41676800 width=8)
-> Seq Scan on aaa20 (cost=0.00..641999.56 rows=41673656 width=8)
-> Seq Scan on aaa21 (cost=0.00..641959.04 rows=41670604 width=8)
-> Seq Scan on aaa22 (cost=0.00..641982.44 rows=41672544 width=8)
-> Seq Scan on aaa23 (cost=0.00..641976.76 rows=41672176 width=8)
(58 rows)
postgres=# select count(*) from aaa join bbb using (order_id);
count
------------
1000000000
(1 row)
Time: 1005965.972 ms (16:45.966)
2、开启并行，耗时： 76 秒。
postgres=# set max_parallel_workers_per_gather =24;
postgres=# set enable_parallel_append =off;
postgres=# set enable_partitionwise_join =on;
postgres=# explain select count(*) from aaa join bbb using (order_id);
QUERY PLAN
-------------------------------------------------------------------------------------------------------------
Finalize Aggregate (cost=12494865.77..12494865.78 rows=1 width=8)
-> Gather (cost=520571.51..12494864.33 rows=576 width=8)
Workers Planned: 24
-> Append (cost=520571.51..12494864.33 rows=24 width=8)
-> Partial Aggregate (cost=520571.51..520571.52 rows=1 width=8)
-> Parallel Hash Join (cost=264263.99..516231.63 rows=1735955 width=0)
Hash Cond: (aaa0.order_id = bbb0.order_id)
-> Parallel Seq Scan on aaa0 (cost=0.00..242564.55 rows=1735955 width=8)
-> Parallel Hash (cost=242564.55..242564.55 rows=1735955 width=8)
-> Parallel Seq Scan on bbb0 (cost=0.00..242564.55 rows=1735955 width=8)
-> Partial Aggregate (cost=520548.40..520548.41 rows=1 width=8)
-> Parallel Hash Join (cost=264252.26..516208.71 rows=1735878 width=0)
Hash Cond: (aaa1.order_id = bbb1.order_id)
-> Parallel Seq Scan on aaa1 (cost=0.00..242553.78 rows=1735878 width=8)
-> Parallel Hash (cost=242553.78..242553.78 rows=1735878 width=8)
-> Parallel Seq Scan on bbb1 (cost=0.00..242553.78 rows=1735878 width=8)
-> Partial Aggregate (cost=520624.68..520624.69 rows=1 width=8)
-> Parallel Hash Join (cost=264290.98..516284.35 rows=1736132 width=0)
Hash Cond: (aaa2.order_id = bbb2.order_id)
-> Parallel Seq Scan on aaa2 (cost=0.00..242589.33 rows=1736132 width=8)
-> Parallel Hash (cost=242589.33..242589.33 rows=1736132 width=8)
-> Parallel Seq Scan on bbb2 (cost=0.00..242589.33 rows=1736132 width=8)
-> Partial Aggregate (cost=520640.86..520640.87 rows=1 width=8)
-> Parallel Hash Join (cost=264299.19..516300.40 rows=1736186 width=0)
Hash Cond: (aaa3.order_id = bbb3.order_id)
-> Parallel Seq Scan on aaa3 (cost=0.00..242596.86 rows=1736186 width=8)
-> Parallel Hash (cost=242596.86..242596.86 rows=1736186 width=8)
-> Parallel Seq Scan on bbb3 (cost=0.00..242596.86 rows=1736186 width=8)
-> Partial Aggregate (cost=520622.39..520622.40 rows=1 width=8)
-> Parallel Hash Join (cost=264289.81..516282.07 rows=1736125 width=0)
Hash Cond: (aaa4.order_id = bbb4.order_id)
-> Parallel Seq Scan on aaa4 (cost=0.00..242588.25 rows=1736125 width=8)
-> Parallel Hash (cost=242588.25..242588.25 rows=1736125 width=8)
-> Parallel Seq Scan on bbb4 (cost=0.00..242588.25 rows=1736125 width=8)
-> Partial Aggregate (cost=520587.70..520587.71 rows=1 width=8)
-> Parallel Hash Join (cost=264272.20..516247.68 rows=1736009 width=0)
Hash Cond: (aaa5.order_id = bbb5.order_id)
-> Parallel Seq Scan on aaa5 (cost=0.00..242572.09 rows=1736009 width=8)
-> Parallel Hash (cost=242572.09..242572.09 rows=1736009 width=8)
-> Parallel Seq Scan on bbb5 (cost=0.00..242572.09 rows=1736009 width=8)
-> Partial Aggregate (cost=520576.15..520576.16 rows=1 width=8)
-> Parallel Hash Join (cost=264266.34..516236.23 rows=1735971 width=0)
Hash Cond: (aaa6.order_id = bbb6.order_id)
-> Parallel Seq Scan on aaa6 (cost=0.00..242566.71 rows=1735971 width=8)
-> Parallel Hash (cost=242566.71..242566.71 rows=1735971 width=8)
-> Parallel Seq Scan on bbb6 (cost=0.00..242566.71 rows=1735971 width=8)
-> Partial Aggregate (cost=520620.06..520620.07 rows=1 width=8)
-> Parallel Hash Join (cost=264288.63..516279.77 rows=1736117 width=0)
Hash Cond: (aaa7.order_id = bbb7.order_id)
-> Parallel Seq Scan on aaa7 (cost=0.00..242587.17 rows=1736117 width=8)
-> Parallel Hash (cost=242587.17..242587.17 rows=1736117 width=8)
-> Parallel Seq Scan on bbb7 (cost=0.00..242587.17 rows=1736117 width=8)
-> Partial Aggregate (cost=520481.38..520481.39 rows=1 width=8)
-> Parallel Hash Join (cost=264218.23..516142.25 rows=1735655 width=0)
Hash Cond: (aaa8.order_id = bbb8.order_id)
-> Parallel Seq Scan on aaa8 (cost=0.00..242522.55 rows=1735655 width=8)
-> Parallel Hash (cost=242522.55..242522.55 rows=1735655 width=8)
-> Parallel Seq Scan on bbb8 (cost=0.00..242522.55 rows=1735655 width=8)
-> Partial Aggregate (cost=520508.85..520508.86 rows=1 width=8)
-> Parallel Hash Join (cost=264232.08..516169.51 rows=1735737 width=0)
Hash Cond: (aaa9.order_id = bbb9.order_id)
-> Parallel Seq Scan on aaa9 (cost=0.00..242535.47 rows=1735747 width=8)
-> Parallel Hash (cost=242535.37..242535.37 rows=1735737 width=8)
-> Parallel Seq Scan on bbb9 (cost=0.00..242535.37 rows=1735737 width=8)
-> Partial Aggregate (cost=520583.08..520583.09 rows=1 width=8)
-> Parallel Hash Join (cost=264269.86..516243.10 rows=1735994 width=0)
Hash Cond: (aaa10.order_id = bbb10.order_id)
-> Parallel Seq Scan on aaa10 (cost=0.00..242569.94 rows=1735994 width=8)
-> Parallel Hash (cost=242569.94..242569.94 rows=1735994 width=8)
-> Parallel Seq Scan on bbb10 (cost=0.00..242569.94 rows=1735994 width=8)
-> Partial Aggregate (cost=520520.67..520520.68 rows=1 width=8)
-> Parallel Hash Join (cost=264238.18..516181.21 rows=1735786 width=0)
Hash Cond: (aaa11.order_id = bbb11.order_id)
-> Parallel Seq Scan on aaa11 (cost=0.00..242540.86 rows=1735786 width=8)
-> Parallel Hash (cost=242540.86..242540.86 rows=1735786 width=8)
-> Parallel Seq Scan on bbb11 (cost=0.00..242540.86 rows=1735786 width=8)
-> Partial Aggregate (cost=520802.68..520802.69 rows=1 width=8)
-> Parallel Hash Join (cost=264381.34..516460.87 rows=1736726 width=0)
Hash Cond: (aaa12.order_id = bbb12.order_id)
-> Parallel Seq Scan on aaa12 (cost=0.00..242672.26 rows=1736726 width=8)
-> Parallel Hash (cost=242672.26..242672.26 rows=1736726 width=8)
-> Parallel Seq Scan on bbb12 (cost=0.00..242672.26 rows=1736726 width=8)
-> Partial Aggregate (cost=520758.19..520758.20 rows=1 width=8)
-> Parallel Hash Join (cost=264358.53..516416.80 rows=1736557 width=0)
Hash Cond: (aaa13.order_id = bbb13.order_id)
-> Parallel Seq Scan on aaa13 (cost=0.00..242651.80 rows=1736580 width=8)
-> Parallel Hash (cost=242651.57..242651.57 rows=1736557 width=8)
-> Parallel Seq Scan on bbb13 (cost=0.00..242651.57 rows=1736557 width=8)
-> Partial Aggregate (cost=520515.10..520515.11 rows=1 width=8)
-> Parallel Hash Join (cost=264235.00..516175.77 rows=1735733 width=0)
Hash Cond: (bbb14.order_id = aaa14.order_id)
-> Parallel Seq Scan on bbb14 (cost=0.00..242538.70 rows=1735770 width=8)
-> Parallel Hash (cost=242538.33..242538.33 rows=1735733 width=8)
-> Parallel Seq Scan on aaa14 (cost=0.00..242538.33 rows=1735733 width=8)
-> Partial Aggregate (cost=520465.20..520465.21 rows=1 width=8)
-> Parallel Hash Join (cost=264210.02..516126.20 rows=1735601 width=0)
Hash Cond: (aaa15.order_id = bbb15.order_id)
-> Parallel Seq Scan on aaa15 (cost=0.00..242515.01 rows=1735601 width=8)
-> Parallel Hash (cost=242515.01..242515.01 rows=1735601 width=8)
-> Parallel Seq Scan on bbb15 (cost=0.00..242515.01 rows=1735601 width=8)
-> Partial Aggregate (cost=520539.16..520539.17 rows=1 width=8)
-> Parallel Hash Join (cost=264247.56..516199.54 rows=1735847 width=0)
Hash Cond: (aaa16.order_id = bbb16.order_id)
-> Parallel Seq Scan on aaa16 (cost=0.00..242549.47 rows=1735847 width=8)
-> Parallel Hash (cost=242549.47..242549.47 rows=1735847 width=8)
-> Parallel Seq Scan on bbb16 (cost=0.00..242549.47 rows=1735847 width=8)
-> Partial Aggregate (cost=520578.45..520578.46 rows=1 width=8)
-> Parallel Hash Join (cost=264267.51..516238.50 rows=1735978 width=0)
Hash Cond: (aaa17.order_id = bbb17.order_id)
-> Parallel Seq Scan on aaa17 (cost=0.00..242567.78 rows=1735978 width=8)
-> Parallel Hash (cost=242567.78..242567.78 rows=1735978 width=8)
-> Parallel Seq Scan on bbb17 (cost=0.00..242567.78 rows=1735978 width=8)
-> Partial Aggregate (cost=520809.61..520809.62 rows=1 width=8)
-> Parallel Hash Join (cost=264384.85..516467.74 rows=1736749 width=0)
Hash Cond: (aaa18.order_id = bbb18.order_id)
-> Parallel Seq Scan on aaa18 (cost=0.00..242675.49 rows=1736749 width=8)
-> Parallel Hash (cost=242675.49..242675.49 rows=1736749 width=8)
-> Parallel Seq Scan on bbb18 (cost=0.00..242675.49 rows=1736749 width=8)
-> Partial Aggregate (cost=520744.89..520744.90 rows=1 width=8)
-> Parallel Hash Join (cost=264352.00..516403.55 rows=1736533 width=0)
Hash Cond: (aaa19.order_id = bbb19.order_id)
-> Parallel Seq Scan on aaa19 (cost=0.00..242645.33 rows=1736533 width=8)
-> Parallel Hash (cost=242645.33..242645.33 rows=1736533 width=8)
-> Parallel Seq Scan on bbb19 (cost=0.00..242645.33 rows=1736533 width=8)
-> Partial Aggregate (cost=520705.59..520705.60 rows=1 width=8)
-> Parallel Hash Join (cost=264332.05..516364.58 rows=1736402 width=0)
Hash Cond: (aaa20.order_id = bbb20.order_id)
-> Parallel Seq Scan on aaa20 (cost=0.00..242627.02 rows=1736402 width=8)
-> Parallel Hash (cost=242627.02..242627.02 rows=1736402 width=8)
-> Parallel Seq Scan on bbb20 (cost=0.00..242627.02 rows=1736402 width=8)
-> Partial Aggregate (cost=520681.20..520681.21 rows=1 width=8)
-> Parallel Hash Join (cost=264319.19..516340.52 rows=1736275 width=0)
Hash Cond: (bbb21.order_id = aaa21.order_id)
-> Parallel Seq Scan on bbb21 (cost=0.00..242616.25 rows=1736325 width=8)
-> Parallel Hash (cost=242615.75..242615.75 rows=1736275 width=8)
-> Parallel Seq Scan on aaa21 (cost=0.00..242615.75 rows=1736275 width=8)
-> Partial Aggregate (cost=520691.04..520691.05 rows=1 width=8)
-> Parallel Hash Join (cost=264324.40..516350.21 rows=1736329 width=0)
Hash Cond: (aaa22.order_id = bbb22.order_id)
-> Parallel Seq Scan on aaa22 (cost=0.00..242620.56 rows=1736356 width=8)
-> Parallel Hash (cost=242620.29..242620.29 rows=1736329 width=8)
-> Parallel Seq Scan on bbb22 (cost=0.00..242620.29 rows=1736329 width=8)
-> Partial Aggregate (cost=520687.11..520687.12 rows=1 width=8)
-> Parallel Hash Join (cost=264322.67..516346.26 rows=1736341 width=0)
Hash Cond: (aaa23.order_id = bbb23.order_id)
-> Parallel Seq Scan on aaa23 (cost=0.00..242618.41 rows=1736341 width=8)
-> Parallel Hash (cost=242618.41..242618.41 rows=1736341 width=8)
-> Parallel Seq Scan on bbb23 (cost=0.00..242618.41 rows=1736341 width=8)
(148 rows)
postgres=# select count(*) from aaa join bbb using (order_id);
count
------------
1000000000
(1 row)
Time: 76118.036 ms (01:16.118)
其他知识
1、优化器自动并行度算法 CBO
《PostgreSQL 11 并行计算算法，参数，强制并行度设置》
2、function, op 识别是否支持parallel
postgres=# select proparallel,proname from pg_proc;
proparallel | proname
-------------+----------------------------------------------
s | boolin
s | boolout
s | byteain
s | byteaout
3、subquery mapreduce unlogged table
对于一些情况，如果期望简化优化器对非常非常复杂的SQL并行优化的负担，可以自己将SQL拆成几段，中间结果使用unlogged table保存，类似mapreduce的思想。unlogged table同样支持parallel 计算。
4、vacuum，垃圾回收并行。
5、dblink 异步调用并行
《PostgreSQL VOPS 向量计算 + DBLINK异步并行 - 单实例 10亿 聚合计算跑进2秒》
《PostgreSQL 相似搜索分布式架构设计与实践 - dblink异步调用与多机并行(远程 游标+记录 UDF实例)》
暂时不允许并行的场景(将来PG会继续扩大支持范围)：
1、修改行，锁行，除了create table as , select into, create mview这几个可以使用并行。
2、query 会被中断时，例如cursor , loop in PL/SQL ，因为涉及到中间处理，所以不建议开启并行。
3、paralle unsafe udf ，这种UDF不会并行
4、嵌套并行(udf (内部query并行))，外部调用这个UDF的SQL不会并行。（主要是防止large parallel workers ）
5、SSI 隔离级别
参考
https://www.postgresql.org/docs/11/parallel-plans.html
《PostgreSQL 11 并行计算算法，参数，强制并行度设置》
《PostgreSQL 11 preview - 并行计算 增强 汇总》
《PostgreSQL 10 自定义并行计算聚合函数的原理与实践 - (含array_agg合并多个数组为单个一元数组的例子)》
免费领取阿里云RDS PostgreSQL实例、ECS虚拟机
