PostgreSQL 如何让 列存（外部列存） 并行起来
标签
PostgreSQL , 列存 , cstore , append parallel , 继承
背景
PostgreSQL 10已经实现了大部分操作的并行计算，例如（全表扫描，索引扫描，位图扫描，哈希JOIN，哈希聚合，分组聚合，排序，建索引等）。
对于外部表，要实现并行扫描，PostgreSQL有什么方法呢？
PostgreSQL不仅支持单个对象的并行计算，还支持多个对象的并行访问。多个对象的并行访问，包括继承表、分区表、继承外部表、UNION、UNION ALL等语义。
这个patch：
https://commitfest.postgresql.org/15/987/
例如：
某个父表有4个继承表，当查询父表时，可以并行的访问父表以及它的继承表。（继承可以是本地的物化视图、本地表、或者外部表对象）。这就给并行访问外部列存表提供了技术基础。
并行访问外部表带来了一个全新的能力，可以把 PG 10+ 作为一个具备完整SQL语义，强大功能的中间件。
试用这个patch
1、使用这个PATCH
wget https://ftp.postgresql.org/pub/snapshot/dev/postgresql-snapshot.tar.bz2
wget https://www.postgresql.org/message-id/attachment/55680/ParallelAppend_v17.patch
tar -jxvf postgresql-snapshot.tar.bz2
cd postgresql-11devel/
patch -p 1 < ../ParallelAppend_v17.patch
export USE_NAMED_POSIX_SEMAPHORES=1
LIBS=-lpthread CFLAGS="-O3" ./configure --prefix=/home/dege.zzz/pgsql11
LIBS=-lpthread CFLAGS="-O3" make world -j 64
LIBS=-lpthread CFLAGS="-O3" make install-world
2、初始化数据库
initdb -D $PGDATA -U postgres --locale=C -E SQL_ASCII
3、配置参数，这个PATCH增加了两个GUC参数
enable_partition_wise_join
enable_parallelappend
4、配置并行参数，用于并行测试
max_worker_processes = 128 # (change requires restart)
max_parallel_workers_per_gather = 64 # taken from max_parallel_workers
max_parallel_workers = 128 # maximum number of max_worker_processes that
enable_parallelappend = on
parallel_tuple_cost = 0 # same scale as above
parallel_setup_cost = 0 # same scale as above
min_parallel_table_scan_size = 0
min_parallel_index_scan_size = 0
5、创建表和继承关系
create table p(id int, info text);
create table c1(like p) inherits(p);
create table c2(like p) inherits(p);
create table c3(like p) inherits(p);
create table c4(like p) inherits(p);
6、写入测试数据，每个子表1000万记录。
insert into c1 select generate_series(1,10000000),'test';
insert into c2 select generate_series(1,10000000),'test';
insert into c3 select generate_series(1,10000000),'test';
insert into c4 select generate_series(1,10000000),'test';
7、设置每个子表的并行度
alter table c1 set (parallel_workers =8);
alter table c2 set (parallel_workers =8);
alter table c3 set (parallel_workers =8);
alter table c4 set (parallel_workers =8);
8、设置父表的并行度（目前取决于父表的并行度，将来可能会改进）。
alter table p set (parallel_workers = 40);
9、单表的耗时
explain (analyze,verbose,timing,costs,buffers) select count(*) from c1 where id=1;
explain (analyze,verbose,timing,costs,buffers) select count(*) from c2 where id=2;
explain (analyze,verbose,timing,costs,buffers) select count(*) from c3 where id=2;
explain (analyze,verbose,timing,costs,buffers) select count(*) from c4 where id=2;
每张表114毫秒。
Execution time: 114.160 ms
10、测试关闭append parallel
postgres=# set enable_parallelappend =off;
SET
postgres=# alter table p set (parallel_workers = 8);
ALTER TABLE
postgres=# explain (analyze,verbose,timing,costs,buffers) select count(*) from p where id=2;
QUERY PLAN
-------------------------------------------------------------------------------------------------------------------------------------
Aggregate (cost=278720.12..278720.13 rows=1 width=8) (actual time=406.051..406.051 rows=1 loops=1)
Output: count(*)
Buffers: shared hit=30906
-> Gather (cost=0.00..278720.11 rows=5 width=0) (actual time=0.376..406.042 rows=4 loops=1)
Workers Planned: 8
Workers Launched: 8
Buffers: shared hit=30906
-> Append (cost=0.00..278720.11 rows=5 width=0) (actual time=325.088..402.806 rows=0 loops=9)
Buffers: shared hit=216220
Worker 0: actual time=401.809..401.809 rows=0 loops=1
Buffers: shared hit=23064
Worker 1: actual time=402.125..402.125 rows=0 loops=1
Buffers: shared hit=23156
Worker 2: actual time=402.255..402.255 rows=0 loops=1
Buffers: shared hit=23494
Worker 3: actual time=402.448..402.448 rows=0 loops=1
Buffers: shared hit=23337
Worker 4: actual time=305.167..402.598 rows=1 loops=1
Buffers: shared hit=23572
Worker 5: actual time=206.445..402.817 rows=1 loops=1
Buffers: shared hit=22803
Worker 6: actual time=402.630..402.630 rows=0 loops=1
Buffers: shared hit=22971
Worker 7: actual time=402.905..402.905 rows=0 loops=1
Buffers: shared hit=22917
-> Parallel Seq Scan on public.p (cost=0.00..0.00 rows=1 width=0) (actual time=0.000..0.000 rows=0 loops=9)
Filter: (p.id = 2)
Worker 0: actual time=0.000..0.000 rows=0 loops=1
Worker 1: actual time=0.000..0.000 rows=0 loops=1
Worker 2: actual time=0.000..0.000 rows=0 loops=1
Worker 3: actual time=0.000..0.000 rows=0 loops=1
Worker 4: actual time=0.000..0.000 rows=0 loops=1
Worker 5: actual time=0.000..0.000 rows=0 loops=1
Worker 6: actual time=0.000..0.000 rows=0 loops=1
Worker 7: actual time=0.001..0.001 rows=0 loops=1
-> Parallel Seq Scan on public.c1 (cost=0.00..69680.27 rows=1 width=0) (actual time=93.294..105.312 rows=0 loops=9)
Filter: (c1.id = 2)
Rows Removed by Filter: 1111111
Buffers: shared hit=54055
Worker 0: actual time=104.313..104.313 rows=0 loops=1
Buffers: shared hit=5554
Worker 1: actual time=104.637..104.637 rows=0 loops=1
Buffers: shared hit=5598
Worker 2: actual time=104.762..104.762 rows=0 loops=1
Buffers: shared hit=5857
Worker 3: actual time=104.952..104.952 rows=0 loops=1
Buffers: shared hit=5827
Worker 4: actual time=105.111..105.111 rows=0 loops=1
Buffers: shared hit=5896
Worker 5: actual time=105.328..105.328 rows=0 loops=1
Buffers: shared hit=5664
Worker 6: actual time=105.121..105.121 rows=0 loops=1
Buffers: shared hit=5692
Worker 7: actual time=105.414..105.414 rows=0 loops=1
Buffers: shared hit=5713
-> Parallel Seq Scan on public.c2 (cost=0.00..69679.78 rows=1 width=0) (actual time=89.870..101.103 rows=0 loops=9)
Filter: (c2.id = 2)
Rows Removed by Filter: 1111111
Buffers: shared hit=54055
Worker 0: actual time=101.104..101.104 rows=0 loops=1
Buffers: shared hit=5849
Worker 1: actual time=101.107..101.107 rows=0 loops=1
Buffers: shared hit=5864
Worker 2: actual time=101.102..101.102 rows=0 loops=1
Buffers: shared hit=5900
Worker 3: actual time=101.096..101.096 rows=0 loops=1
Buffers: shared hit=5859
Worker 4: actual time=101.112..101.112 rows=0 loops=1
Buffers: shared hit=5920
Worker 5: actual time=101.095..101.095 rows=0 loops=1
Buffers: shared hit=5798
Worker 6: actual time=101.108..101.108 rows=0 loops=1
Buffers: shared hit=5843
Worker 7: actual time=101.104..101.104 rows=0 loops=1
Buffers: shared hit=5818
-> Parallel Seq Scan on public.c3 (cost=0.00..69679.78 rows=1 width=0) (actual time=87.947..98.939 rows=0 loops=9)
Filter: (c3.id = 2)
Rows Removed by Filter: 1111111
Buffers: shared hit=54055
Worker 0: actual time=98.939..98.939 rows=0 loops=1
Buffers: shared hit=5853
Worker 1: actual time=98.929..98.929 rows=0 loops=1
Buffers: shared hit=5870
Worker 2: actual time=98.952..98.952 rows=0 loops=1
Buffers: shared hit=5918
Worker 3: actual time=98.940..98.940 rows=0 loops=1
Buffers: shared hit=5826
Worker 4: actual time=98.928..98.928 rows=0 loops=1
Buffers: shared hit=5886
Worker 5: actual time=0.017..98.948 rows=1 loops=1
Buffers: shared hit=5608
Worker 6: actual time=98.936..98.936 rows=0 loops=1
Buffers: shared hit=5643
Worker 7: actual time=98.936..98.936 rows=0 loops=1
Buffers: shared hit=5624
-> Parallel Seq Scan on public.c4 (cost=0.00..69680.27 rows=1 width=0) (actual time=86.621..97.446 rows=0 loops=9)
Filter: (c4.id = 2)
Rows Removed by Filter: 1111111
Buffers: shared hit=54055
Worker 0: actual time=97.448..97.448 rows=0 loops=1
Buffers: shared hit=5808
Worker 1: actual time=97.447..97.447 rows=0 loops=1
Buffers: shared hit=5824
Worker 2: actual time=97.435..97.435 rows=0 loops=1
Buffers: shared hit=5819
Worker 3: actual time=97.455..97.455 rows=0 loops=1
Buffers: shared hit=5825
Worker 4: actual time=0.012..97.443 rows=1 loops=1
Buffers: shared hit=5870
Worker 5: actual time=97.439..97.439 rows=0 loops=1
Buffers: shared hit=5733
Worker 6: actual time=97.460..97.460 rows=0 loops=1
Buffers: shared hit=5793
Worker 7: actual time=97.443..97.443 rows=0 loops=1
Buffers: shared hit=5762
Planning time: 0.140 ms
Execution time: 421.076 ms
(117 rows)
扫描p, c1, c2, c3, c4是串行的。耗时几乎是累加的。
同时也可以看到，每个表扫描都启动了新的worker。
11、测试开启append parallel
postgres=# set enable_parallelappend =on;
SET
postgres=# alter table p set (parallel_workers = 40);
ALTER TABLE
postgres=# explain (analyze,verbose,timing,costs,buffers) select count(*) from p where id=2;
QUERY PLAN
--------------------------------------------------------------------------------------------------------------------------------------
Aggregate (cost=278720.12..278720.13 rows=1 width=8) (actual time=128.447..128.447 rows=1 loops=1)
Output: count(*)
Buffers: shared hit=9780
-> Gather (cost=0.00..278720.11 rows=5 width=0) (actual time=0.914..128.438 rows=4 loops=1)
Workers Planned: 41
Workers Launched: 41
Buffers: shared hit=9780
-> Parallel Append (cost=0.00..278720.11 rows=5 width=0) (actual time=107.692..119.530 rows=0 loops=42)
Buffers: shared hit=216220
Worker 0: actual time=113.650..113.650 rows=0 loops=1
Buffers: shared hit=3363
Worker 1: actual time=114.238..114.238 rows=0 loops=1
Buffers: shared hit=3329
Worker 2: actual time=114.214..114.214 rows=0 loops=1
Buffers: shared hit=3333
Worker 3: actual time=114.887..114.887 rows=0 loops=1
Buffers: shared hit=3370
Worker 4: actual time=115.038..115.038 rows=0 loops=1
Buffers: shared hit=3365
Worker 5: actual time=115.084..115.084 rows=0 loops=1
Buffers: shared hit=3389
Worker 6: actual time=115.720..115.720 rows=0 loops=1
Buffers: shared hit=3370
Worker 7: actual time=115.830..115.830 rows=0 loops=1
Buffers: shared hit=3358
Worker 8: actual time=116.457..116.457 rows=0 loops=1
Buffers: shared hit=3430
Worker 9: actual time=116.760..116.760 rows=0 loops=1
Buffers: shared hit=3423
Worker 10: actual time=116.541..116.541 rows=0 loops=1
Buffers: shared hit=6129
Worker 11: actual time=117.355..117.355 rows=0 loops=1
Buffers: shared hit=3447
Worker 12: actual time=117.767..117.767 rows=0 loops=1
Buffers: shared hit=6295
Worker 13: actual time=118.195..118.195 rows=0 loops=1
Buffers: shared hit=6382
Worker 14: actual time=118.150..118.150 rows=0 loops=1
Buffers: shared hit=6194
Worker 15: actual time=118.468..118.468 rows=0 loops=1
Buffers: shared hit=6250
Worker 16: actual time=119.035..119.035 rows=0 loops=1
Buffers: shared hit=6337
Worker 17: actual time=119.092..119.092 rows=0 loops=1
Buffers: shared hit=6472
Worker 18: actual time=119.077..119.077 rows=0 loops=1
Buffers: shared hit=6293
Worker 19: actual time=119.361..119.361 rows=0 loops=1
Buffers: shared hit=6557
Worker 20: actual time=119.519..119.519 rows=0 loops=1
Buffers: shared hit=6446
Worker 21: actual time=119.782..119.782 rows=0 loops=1
Buffers: shared hit=6465
Worker 22: actual time=120.339..120.339 rows=0 loops=1
Buffers: shared hit=6575
Worker 23: actual time=120.350..120.350 rows=0 loops=1
Buffers: shared hit=6596
Worker 24: actual time=120.436..120.436 rows=0 loops=1
Buffers: shared hit=6593
Worker 25: actual time=121.735..121.735 rows=0 loops=1
Buffers: shared hit=6606
Worker 26: actual time=120.821..120.821 rows=0 loops=1
Buffers: shared hit=6607
Worker 27: actual time=122.161..122.161 rows=0 loops=1
Buffers: shared hit=6585
Worker 28: actual time=121.265..121.265 rows=0 loops=1
Buffers: shared hit=6695
Worker 29: actual time=121.616..121.616 rows=0 loops=1
Buffers: shared hit=6703
Worker 30: actual time=121.919..121.919 rows=0 loops=1
Buffers: shared hit=3630
Worker 31: actual time=121.914..121.914 rows=0 loops=1
Buffers: shared hit=6590
Worker 32: actual time=122.310..122.310 rows=0 loops=1
Buffers: shared hit=3714
Worker 33: actual time=122.471..122.471 rows=0 loops=1
Buffers: shared hit=6652
Worker 34: actual time=122.684..122.684 rows=0 loops=1
Buffers: shared hit=3699
Worker 35: actual time=122.737..122.737 rows=0 loops=1
Buffers: shared hit=3744
Worker 36: actual time=122.884..122.884 rows=0 loops=1
Buffers: shared hit=3678
Worker 37: actual time=123.155..123.155 rows=0 loops=1
Buffers: shared hit=3740
Worker 38: actual time=0.014..123.474 rows=1 loops=1
Buffers: shared hit=3647
Worker 39: actual time=0.015..123.662 rows=1 loops=1
Buffers: shared hit=3693
Worker 40: actual time=0.019..123.742 rows=1 loops=1
Buffers: shared hit=3696
-> Parallel Seq Scan on public.p (cost=0.00..0.00 rows=1 width=0) (actual time=0.001..0.001 rows=0 loops=1)
Filter: (p.id = 2)
Worker 40: actual time=0.001..0.001 rows=0 loops=1
-> Parallel Seq Scan on public.c1 (cost=0.00..69680.27 rows=1 width=0) (actual time=85.131..93.695 rows=0 loops=14)
Filter: (c1.id = 2)
Rows Removed by Filter: 714286
Buffers: shared hit=54055
Worker 0: actual time=110.100..110.100 rows=0 loops=1
Buffers: shared hit=3260
Worker 4: actual time=111.267..111.267 rows=0 loops=1
Buffers: shared hit=3272
Worker 8: actual time=112.661..112.661 rows=0 loops=1
Buffers: shared hit=3333
Worker 9: actual time=13.433..13.433 rows=0 loops=1
Buffers: shared hit=419
Worker 12: actual time=114.096..114.096 rows=0 loops=1
Buffers: shared hit=6118
Worker 16: actual time=115.113..115.113 rows=0 loops=1
Buffers: shared hit=6166
Worker 17: actual time=13.448..13.448 rows=0 loops=1
Buffers: shared hit=770
Worker 20: actual time=115.961..115.961 rows=0 loops=1
Buffers: shared hit=6281
Worker 24: actual time=116.868..116.868 rows=0 loops=1
Buffers: shared hit=6417
Worker 28: actual time=117.509..117.509 rows=0 loops=1
Buffers: shared hit=6525
Worker 32: actual time=118.585..118.585 rows=0 loops=1
Buffers: shared hit=3617
Worker 33: actual time=13.446..13.446 rows=0 loops=1
Buffers: shared hit=699
Worker 36: actual time=119.327..119.327 rows=0 loops=1
Buffers: shared hit=3579
Worker 40: actual time=0.017..119.913 rows=1 loops=1
Buffers: shared hit=3599
-> Parallel Seq Scan on public.c2 (cost=0.00..69679.78 rows=1 width=0) (actual time=29.164..32.100 rows=0 loops=42)
Filter: (c2.id = 2)
Rows Removed by Filter: 238095
Buffers: shared hit=54055
Worker 0: actual time=2.855..2.855 rows=0 loops=1
Buffers: shared hit=83
Worker 1: actual time=2.909..2.909 rows=0 loops=1
Buffers: shared hit=88
Worker 2: actual time=2.880..2.880 rows=0 loops=1
Buffers: shared hit=91
Worker 3: actual time=114.623..114.623 rows=0 loops=1
Buffers: shared hit=3370
Worker 4: actual time=3.567..3.567 rows=0 loops=1
Buffers: shared hit=93
Worker 5: actual time=2.890..2.890 rows=0 loops=1
Buffers: shared hit=89
Worker 6: actual time=2.893..2.893 rows=0 loops=1
Buffers: shared hit=89
Worker 7: actual time=115.735..115.735 rows=0 loops=1
Buffers: shared hit=3358
Worker 8: actual time=2.921..2.921 rows=0 loops=1
Buffers: shared hit=78
Worker 9: actual time=2.937..2.937 rows=0 loops=1
Buffers: shared hit=80
Worker 10: actual time=2.873..2.873 rows=0 loops=1
Buffers: shared hit=160
Worker 11: actual time=117.272..117.272 rows=0 loops=1
Buffers: shared hit=3447
Worker 12: actual time=2.865..2.865 rows=0 loops=1
Buffers: shared hit=143
Worker 13: actual time=17.020..17.020 rows=0 loops=1
Buffers: shared hit=916
Worker 14: actual time=2.850..2.850 rows=0 loops=1
Buffers: shared hit=160
Worker 15: actual time=118.347..118.347 rows=0 loops=1
Buffers: shared hit=6250
Worker 16: actual time=2.924..2.924 rows=0 loops=1
Buffers: shared hit=137
Worker 17: actual time=3.566..3.566 rows=0 loops=1
Buffers: shared hit=174
Worker 18: actual time=2.888..2.888 rows=0 loops=1
Buffers: shared hit=122
Worker 19: actual time=119.352..119.352 rows=0 loops=1
Buffers: shared hit=6557
Worker 20: actual time=3.553..3.553 rows=0 loops=1
Buffers: shared hit=165
Worker 21: actual time=17.010..17.010 rows=0 loops=1
Buffers: shared hit=927
Worker 22: actual time=2.851..2.851 rows=0 loops=1
Buffers: shared hit=167
Worker 23: actual time=120.346..120.346 rows=0 loops=1
Buffers: shared hit=6596
Worker 24: actual time=3.563..3.563 rows=0 loops=1
Buffers: shared hit=176
Worker 25: actual time=17.017..17.017 rows=0 loops=1
Buffers: shared hit=935
Worker 26: actual time=2.886..2.886 rows=0 loops=1
Buffers: shared hit=163
Worker 27: actual time=121.060..121.060 rows=0 loops=1
Buffers: shared hit=6585
Worker 28: actual time=2.911..2.911 rows=0 loops=1
Buffers: shared hit=135
Worker 29: actual time=2.898..2.898 rows=0 loops=1
Buffers: shared hit=168
Worker 30: actual time=121.689..121.689 rows=0 loops=1
Buffers: shared hit=3630
Worker 31: actual time=2.920..2.920 rows=0 loops=1
Buffers: shared hit=165
Worker 32: actual time=3.572..3.572 rows=0 loops=1
Buffers: shared hit=97
Worker 33: actual time=3.567..3.567 rows=0 loops=1
Buffers: shared hit=171
Worker 34: actual time=2.911..2.911 rows=0 loops=1
Buffers: shared hit=89
Worker 35: actual time=122.732..122.732 rows=0 loops=1
Buffers: shared hit=3744
Worker 36: actual time=2.877..2.877 rows=0 loops=1
Buffers: shared hit=80
Worker 37: actual time=16.989..16.989 rows=0 loops=1
Buffers: shared hit=496
Worker 38: actual time=2.882..2.882 rows=0 loops=1
Buffers: shared hit=86
Worker 39: actual time=0.014..123.337 rows=1 loops=1
Buffers: shared hit=3693
Worker 40: actual time=3.595..3.595 rows=0 loops=1
Buffers: shared hit=97
-> Parallel Seq Scan on public.c3 (cost=0.00..69679.78 rows=1 width=0) (actual time=52.559..58.290 rows=0 loops=21)
Filter: (c3.id = 2)
Rows Removed by Filter: 476190
Buffers: shared hit=54055
Worker 0: actual time=0.689..0.689 rows=0 loops=1
Buffers: shared hit=20
Worker 1: actual time=14.124..14.124 rows=0 loops=1
Buffers: shared hit=419
Worker 2: actual time=111.314..111.314 rows=0 loops=1
Buffers: shared hit=3242
Worker 5: actual time=14.126..14.126 rows=0 loops=1
Buffers: shared hit=421
Worker 6: actual time=112.583..112.583 rows=0 loops=1
Buffers: shared hit=3281
Worker 8: actual time=0.672..0.672 rows=0 loops=1
Buffers: shared hit=19
Worker 9: actual time=0.669..0.669 rows=0 loops=1
Buffers: shared hit=19
Worker 10: actual time=113.632..113.632 rows=0 loops=1
Buffers: shared hit=5969
Worker 12: actual time=0.678..0.678 rows=0 loops=1
Buffers: shared hit=34
Worker 14: actual time=115.273..115.273 rows=0 loops=1
Buffers: shared hit=6034
Worker 16: actual time=0.671..0.671 rows=0 loops=1
Buffers: shared hit=34
Worker 18: actual time=116.184..116.184 rows=0 loops=1
Buffers: shared hit=6171
Worker 22: actual time=117.285..117.285 rows=0 loops=1
Buffers: shared hit=6408
Worker 26: actual time=117.931..117.931 rows=0 loops=1
Buffers: shared hit=6444
Worker 28: actual time=0.679..0.679 rows=0 loops=1
Buffers: shared hit=35
Worker 29: actual time=14.117..14.117 rows=0 loops=1
Buffers: shared hit=794
Worker 31: actual time=118.705..118.705 rows=0 loops=1
Buffers: shared hit=6425
Worker 34: actual time=119.585..119.585 rows=0 loops=1
Buffers: shared hit=3610
Worker 36: actual time=0.675..0.675 rows=0 loops=1
Buffers: shared hit=19
Worker 38: actual time=0.013..120.367 rows=1 loops=1
Buffers: shared hit=3561
-> Parallel Seq Scan on public.c4 (cost=0.00..69680.27 rows=1 width=0) (actual time=92.675..102.616 rows=0 loops=11)
Filter: (c4.id = 2)
Rows Removed by Filter: 909091
Buffers: shared hit=54055
Worker 1: actual time=96.862..96.862 rows=0 loops=1
Buffers: shared hit=2822
Worker 5: actual time=98.057..98.057 rows=0 loops=1
Buffers: shared hit=2879
Worker 9: actual time=99.478..99.478 rows=0 loops=1
Buffers: shared hit=2905
Worker 13: actual time=101.043..101.043 rows=0 loops=1
Buffers: shared hit=5466
Worker 17: actual time=101.879..101.879 rows=0 loops=1
Buffers: shared hit=5528
Worker 21: actual time=102.755..102.755 rows=0 loops=1
Buffers: shared hit=5538
Worker 25: actual time=103.575..103.575 rows=0 loops=1
Buffers: shared hit=5671
Worker 29: actual time=104.456..104.456 rows=0 loops=1
Buffers: shared hit=5741
Worker 33: actual time=105.152..105.152 rows=0 loops=1
Buffers: shared hit=5782
Worker 37: actual time=106.161..106.161 rows=0 loops=1
Buffers: shared hit=3244
Planning time: 0.143 ms
Execution time: 143.039 ms
(282 rows)
扫描p, c1, c2, c3, c4是并行的。时间不是累加。
worker 分布不均，如果均匀性能应该是可以更好的。
不过现在还没有正式MERGE这个PATCH，正式提交肯定会改进的。
12、外部表继承的语法如下：
ALTER FOREIGN TABLE name
INHERIT parent_table |
NO INHERIT parent_table
用此法，可以实现任意类型的外部表的并行。当然也包括本文提到的cstore_fdw，列存外部表。
还有mysql_fdw, oracle_fdw, mongo_fdw, file_fdw, oss_fdw, 比较全面的FDW接口，可以参考这个WIKI页面：
https://wiki.postgresql.org/wiki/Fdw
参考
https://commitfest.postgresql.org/15/987/