PostgreSQL dblink异步调用实现 并行hash分片JOIN - 含数据交、并、差 提速案例 - 含dblink VS pg 11 parallel hash join VS pg 11 智能分区JOIN
背景
数据交、并、差是分析型场景常见的需求。例如用来筛选目标用户、店铺等。
PostgreSQL中交、并、差可以使用SQL语法来实现（union , union all , except , intersect）。其中只有union all是不去重的，其他都会去重。
例子
select generate_series(1,10) except select generate_series(5,12);
select generate_series(1,10) union select generate_series(5,12);
select generate_series(1,10) union all select generate_series(5,12);
select generate_series(1,10) intersect select generate_series(5,12);
当数据量非常庞大时，求交集、差集的话这种方法的效果可能会不够理想。
那么有什么好方法呢？
1、我们自己对数据进行hash切片，然后使用dblink的异步调用接口，一对一的并行操作（求交、差）。
2、PostgreSQL 11 已经支持了parallel hash join，可以解决大数据量求交、差的性能问题。
《PostgreSQL 11 preview - parallel hash join(并行哈希JOIN) 性能极大提升》
原生求交、差性能
测试结构和数据
postgres=# create table tbl(id int, c1 int);
CREATE TABLE
postgres=# insert into tbl select generate_series(1,10000000), random()*99;
INSERT 0 10000000
1、1千万 交 1千万
耗时，31.7秒。
postgres=# explain analyze select * from tbl intersect select * from tbl;
QUERY PLAN
------------------------------------------------------------------------------------------------------------------------------------------------------
HashSetOp Intersect (cost=0.00..413495.25 rows=9999977 width=12) (actual time=27730.818..30412.898 rows=10000000 loops=1)
-> Append (cost=0.00..313495.48 rows=19999954 width=12) (actual time=0.402..18889.746 rows=20000000 loops=1)
-> Subquery Scan on "*SELECT* 1" (cost=0.00..156747.74 rows=9999977 width=12) (actual time=0.401..7744.308 rows=10000000 loops=1)
-> Gather (cost=0.00..56747.97 rows=9999977 width=8) (actual time=0.397..5947.082 rows=10000000 loops=1)
Workers Planned: 8
Workers Launched: 8
-> Parallel Seq Scan on tbl (cost=0.00..56747.97 rows=1249997 width=8) (actual time=0.015..248.653 rows=1111111 loops=9)
-> Subquery Scan on "*SELECT* 2" (cost=0.00..156747.74 rows=9999977 width=12) (actual time=0.329..8366.856 rows=10000000 loops=1)
-> Gather (cost=0.00..56747.97 rows=9999977 width=8) (actual time=0.326..6567.651 rows=10000000 loops=1)
Workers Planned: 8
Workers Launched: 8
-> Parallel Seq Scan on tbl tbl_1 (cost=0.00..56747.97 rows=1249997 width=8) (actual time=0.013..195.661 rows=1111111 loops=9)
Planning time: 0.098 ms
Execution time: 31691.115 ms
(14 rows)
2、1千万 差 1千万
耗时，30秒。
postgres=# explain analyze select * from tbl except select * from tbl;
QUERY PLAN
------------------------------------------------------------------------------------------------------------------------------------------------------
HashSetOp Except (cost=0.00..413495.25 rows=9999977 width=12) (actual time=30021.111..30021.111 rows=0 loops=1)
-> Append (cost=0.00..313495.48 rows=19999954 width=12) (actual time=0.415..20454.584 rows=20000000 loops=1)
-> Subquery Scan on "*SELECT* 1" (cost=0.00..156747.74 rows=9999977 width=12) (actual time=0.414..8500.176 rows=10000000 loops=1)
-> Gather (cost=0.00..56747.97 rows=9999977 width=8) (actual time=0.409..6696.932 rows=10000000 loops=1)
Workers Planned: 8
Workers Launched: 8
-> Parallel Seq Scan on tbl (cost=0.00..56747.97 rows=1249997 width=8) (actual time=0.019..233.999 rows=1111111 loops=9)
-> Subquery Scan on "*SELECT* 2" (cost=0.00..156747.74 rows=9999977 width=12) (actual time=0.341..9162.134 rows=10000000 loops=1)
-> Gather (cost=0.00..56747.97 rows=9999977 width=8) (actual time=0.337..7358.837 rows=10000000 loops=1)
Workers Planned: 8
Workers Launched: 8
-> Parallel Seq Scan on tbl tbl_1 (cost=0.00..56747.97 rows=1249997 width=8) (actual time=0.015..196.848 rows=1111111 loops=9)
Planning time: 0.080 ms
Execution time: 30358.560 ms
(14 rows)
优化手段1 - 使用hash切片，然后使用dblink的异步调用接口，求交、差性能
dblink异步调用的例子，参考
《惊天性能！单RDS PostgreSQL实例 支撑 2000亿 - 实时标签透视案例》
《阿里云RDS PostgreSQL OSS 外部表 - (dblink异步调用封装)并行写提速案例》
这个方法纯属个人高级玩法。建议咱们还是直接用PG 11。
参与计算相交、差的字段中的任意一个或多个，作为HASH切片字段即可。PostgreSQL内置了好多类型转HASH值得到函数：
postgres=# \df *.hash*
List of functions
Schema | Name | Result data type | Argument data types | Type
------------+-----------------------+------------------+--------------------------+------
pg_catalog | hash_aclitem | integer | aclitem | func
pg_catalog | hash_aclitem_extended | bigint | aclitem, bigint | func
pg_catalog | hash_array | integer | anyarray | func
pg_catalog | hash_array_extended | bigint | anyarray, bigint | func
pg_catalog | hash_numeric | integer | numeric | func
pg_catalog | hash_numeric_extended | bigint | numeric, bigint | func
pg_catalog | hash_range | integer | anyrange | func
pg_catalog | hash_range_extended | bigint | anyrange, bigint | func
pg_catalog | hashbpchar | integer | character | func
pg_catalog | hashbpcharextended | bigint | character, bigint | func
pg_catalog | hashchar | integer | "char" | func
pg_catalog | hashcharextended | bigint | "char", bigint | func
pg_catalog | hashenum | integer | anyenum | func
pg_catalog | hashenumextended | bigint | anyenum, bigint | func
pg_catalog | hashfloat4 | integer | real | func
pg_catalog | hashfloat4extended | bigint | real, bigint | func
pg_catalog | hashfloat8 | integer | double precision | func
pg_catalog | hashfloat8extended | bigint | double precision, bigint | func
pg_catalog | hashhandler | index_am_handler | internal | func
pg_catalog | hashinet | integer | inet | func
pg_catalog | hashinetextended | bigint | inet, bigint | func
pg_catalog | hashint2 | integer | smallint | func
pg_catalog | hashint2extended | bigint | smallint, bigint | func
pg_catalog | hashint4 | integer | integer | func
pg_catalog | hashint4extended | bigint | integer, bigint | func
pg_catalog | hashint8 | integer | bigint | func
pg_catalog | hashint8extended | bigint | bigint, bigint | func
pg_catalog | hashmacaddr | integer | macaddr | func
pg_catalog | hashmacaddr8 | integer | macaddr8 | func
pg_catalog | hashmacaddr8extended | bigint | macaddr8, bigint | func
pg_catalog | hashmacaddrextended | bigint | macaddr, bigint | func
pg_catalog | hashname | integer | name | func
pg_catalog | hashnameextended | bigint | name, bigint | func
pg_catalog | hashoid | integer | oid | func
pg_catalog | hashoidextended | bigint | oid, bigint | func
pg_catalog | hashoidvector | integer | oidvector | func
pg_catalog | hashoidvectorextended | bigint | oidvector, bigint | func
pg_catalog | hashtext | integer | text | func
pg_catalog | hashtextextended | bigint | text, bigint | func
pg_catalog | hashvarlena | integer | internal | func
pg_catalog | hashvarlenaextended | bigint | internal, bigint | func
(41 rows)
首先看看切成小片后，求交、差执行时间需要多久：
不开并行，切成48份，每份的intersect时间，大概是1.9秒。
postgres=# explain analyze select t1.* from tbl t1 where mod(abs(hashint4(id)), 48)=0 intersect select t1.* from tbl t1 where mod(abs(hashint4(id)), 48)=0;
QUERY PLAN
----------------------------------------------------------------------------------------------------------------------------------------
HashSetOp Intersect (cost=0.00..489995.08 rows=50000 width=12) (actual time=1822.887..1867.381 rows=208902 loops=1)
-> Append (cost=0.00..489495.08 rows=100000 width=12) (actual time=0.021..1679.633 rows=417804 loops=1)
-> Subquery Scan on "*SELECT* 1" (cost=0.00..244747.54 rows=50000 width=12) (actual time=0.020..811.669 rows=208902 loops=1)
-> Seq Scan on tbl t1 (cost=0.00..244247.54 rows=50000 width=8) (actual time=0.019..774.864 rows=208902 loops=1)
Filter: (mod(abs(hashint4(id)), 48) = 0)
Rows Removed by Filter: 9791098
-> Subquery Scan on "*SELECT* 2" (cost=0.00..244747.54 rows=50000 width=12) (actual time=0.027..807.215 rows=208902 loops=1)
-> Seq Scan on tbl t1_1 (cost=0.00..244247.54 rows=50000 width=8) (actual time=0.026..770.958 rows=208902 loops=1)
Filter: (mod(abs(hashint4(id)), 48) = 0)
Rows Removed by Filter: 9791098
Planning time: 0.116 ms
Execution time: 1887.638 ms
(12 rows)
也就是说，开48个并行切片的话，最理想的性能是1.9秒。
注意
因为这里面的HASH分片是扫全表得到的，所以开的并发越多，扫描次数越多。最好是扫一次，并均分到N个临时空间，然后再从临时空间中扫，这样就只需要扫一遍。当然会增加复杂度，如果表不大，实际上多扫几次也无所谓。
hash并行切片+异步dblink
1、创建生成dblink连接的函数，重复创建不报错。
create or replace function conn(
name, -- dblink名字
text -- 连接串,URL
) returns void as $$
declare
begin
perform dblink_connect($1, $2);
return;
exception when others then
return;
end;
$$ language plpgsql strict;
2、创建一个函数，用于跑并行求交
create or replace function get_intersect(
conn text, -- 连接串
OUT id int,
OUT c1 int
) returns setof record as $$
declare
begin
for i in 0..47 loop
perform conn('link'||i, conn);
perform 1 from dblink_get_result('link'||i) as t(id int, c1 int);
perform dblink_send_query('link'||i, format('select * from tbl t1 where mod(abs(hashint4(id)), 48)=%s intersect select * from tbl t1 where mod(abs(hashint4(id)), 48)=%s', i, i));
end loop;
for i in 0..47 loop
return query SELECT * FROM dblink_get_result('link'||i) as t(id int, c1 int);
end loop;
end;
$$ language plpgsql strict;
使用这个方法，可以看到执行时间大概3秒。但是耗费了很多时间在将1000万条记录从所有的远端返回给调用端。总共差不多8秒。
如果改成返回游标，响应速度就快得不得了了，比如在图计算中，用游标流式返回：
《金融风控、公安刑侦、社会关系、人脉分析等需求分析与数据库实现 - PostgreSQL图数据库场景应用》
create or replace function get_intersect1()
returns setof refcursor as $$
declare
ref refcursor[];
res refcursor;
begin
for i in 0..47 loop
ref[i] := 'cur'||i;
res := ref[i];
open res for execute format('select * from tbl t1 where mod(abs(hashint4(id)), 48)=%s intersect select * from tbl t1 where mod(abs(hashint4(id)), 48)=%s', i, i);
return next res;
end loop;
return;
end;
$$ language plpgsql strict;
用法
postgres=# begin;
BEGIN
postgres=# select * from get_intersect1();
get_intersect1
----------------
cur0
cur1
cur2
cur3
cur4
cur5
cur6
cur7
cur8
cur9
cur10
cur11
cur12
cur13
cur14
cur15
cur16
cur17
cur18
cur19
cur20
cur21
cur22
cur23
cur24
cur25
cur26
cur27
cur28
cur29
cur30
cur31
cur32
cur33
cur34
cur35
cur36
cur37
cur38
cur39
cur40
cur41
cur42
cur43
cur44
cur45
cur46
cur47
(48 rows)
Time: 46.471 ms
-- 第一页比较慢
postgres=# fetch 10 from cur1;
id | c1
---------+----
3591658 | 70
6100015 | 17
3222328 | 90
5500150 | 23
9087335 | 45
2463228 | 86
870261 | 51
9276428 | 85
7672240 | 32
6828314 | 41
(10 rows)
Time: 1645.906 ms (00:01.646)
-- 后面就飞快了。
postgres=# fetch 10 from cur1;
id | c1
---------+----
7335851 | 5
8007430 | 10
6230301 | 27
9111491 | 91
1400805 | 65
3651088 | 33
3292697 | 65
1431682 | 66
2959698 | 66
4580225 | 39
(10 rows)
Time: 0.187 ms
是不是飞快了呢，使用游标，从用户发出请求，到获取数据，大概的延迟是1.7秒。.
求差与之类似，只是改一下SQL。
create or replace function get_except1()
returns setof refcursor as $$
declare
ref refcursor[];
res refcursor;
begin
for i in 0..47 loop
ref[i] := 'cur'||i;
res := ref[i];
open res for execute format('select * from tbl t1 where mod(abs(hashint4(id)), 48)=%s except select * from tbl t1 where mod(abs(hashint4(id)), 48)=%s', i, i);
return next res;
end loop;
return;
end;
$$ language plpgsql strict;
postgres=# begin;
BEGIN
Time: 0.169 ms
postgres=# select * from get_except1();
get_except1
-------------
cur0
cur1
..........
cur44
cur45
cur46
cur47
(48 rows)
Time: 46.482 ms
postgres=# fetch 10 from cur1;
id | c1
----+----
(0 rows)
Time: 1681.922 ms (00:01.682)
优化手段2 - PostgreSQL 11 求交、差性能
使用PostgreSQL 11，JOIN的手法来求交、差。语义相同。
1、求交
select * from tbl intersect select * from tbl;
相当于
select t1.* from tbl t1 join tbl t2 on (t1.id=t2.id and t1.c1=t2.c1); -- 所有参与求交的字段都加到JOIN ON里面
2、求差
select * from tbl except select * from tbl;
相当于
select * from tbl t1 where not exists
( select 1 from
(select t1.id,t1.c1 from tbl t1 join tbl t2 on (t1.id=t2.id and t1.c1=t2.c1) ) t -- 所有参与求交的字段都加到JOIN ON里面
where t.id=t1.id and t.c1=t1.c1
);
PostgreSQL 11 求交、差性能如下
1、求交集，3.3秒。
postgres=# explain analyze select t1.* from tbl t1 join tbl t2 on (t1.id = t2.id and t1.c1 = t2.c1);
QUERY PLAN
-------------------------------------------------------------------------------------------------------------------------------------------
Gather (cost=52060.48..101778.20 rows=100921 width=8) (actual time=407.118..2395.421 rows=10000000 loops=1)
Workers Planned: 32
Workers Launched: 32
-> Parallel Hash Join (cost=52060.48..101778.20 rows=3154 width=8) (actual time=378.294..691.692 rows=303030 loops=33)
Hash Cond: ((t1.id = t2.id) AND (t1.c1 = t2.c1))
-> Parallel Seq Scan on tbl t1 (cost=0.00..47372.99 rows=312499 width=8) (actual time=0.014..41.780 rows=303030 loops=33)
-> Parallel Hash (cost=47372.99..47372.99 rows=312499 width=8) (actual time=374.931..374.931 rows=303030 loops=33)
Buckets: 16777216 Batches: 1 Memory Usage: 522848kB
-> Parallel Seq Scan on tbl t2 (cost=0.00..47372.99 rows=312499 width=8) (actual time=0.022..48.013 rows=303030 loops=33)
Planning time: 0.137 ms
Execution time: 3316.010 ms
(11 rows)
2、求差集，1.9秒
postgres=# explain analyze select * from tbl t1 where not exists
( select 1 from
(select t1.id,t1.c1 from tbl t1 join tbl t2 on (t1.id=t2.id and t1.c1=t2.c1) ) t -- 所有参与求交的字段都加到JOIN ON里面
where t.id=t1.id and t.c1=t1.c1
);
QUERY PLAN
-------------------------------------------------------------------------------------------------------------------------------------------------------
Gather (cost=101825.51..153939.67 rows=9899056 width=8) (actual time=1557.867..1557.867 rows=0 loops=1)
Workers Planned: 32
Workers Launched: 32
-> Parallel Hash Anti Join (cost=101825.51..153939.67 rows=309346 width=8) (actual time=1495.529..1495.529 rows=0 loops=33)
Hash Cond: ((t1.id = t1_1.id) AND (t1.c1 = t1_1.c1))
-> Parallel Seq Scan on tbl t1 (cost=0.00..47372.99 rows=312499 width=8) (actual time=0.013..44.749 rows=303030 loops=33)
-> Parallel Hash (cost=101778.20..101778.20 rows=3154 width=8) (actual time=1260.916..1260.916 rows=303030 loops=33)
Buckets: 16777216 (originally 131072) Batches: 1 (originally 1) Memory Usage: 652800kB
-> Parallel Hash Join (cost=52060.48..101778.20 rows=3154 width=8) (actual time=387.651..740.551 rows=303030 loops=33)
Hash Cond: ((t1_1.id = t2.id) AND (t1_1.c1 = t2.c1))
-> Parallel Seq Scan on tbl t1_1 (cost=0.00..47372.99 rows=312499 width=8) (actual time=0.013..46.111 rows=303030 loops=33)
-> Parallel Hash (cost=47372.99..47372.99 rows=312499 width=8) (actual time=384.666..384.666 rows=303030 loops=33)
Buckets: 16777216 Batches: 1 Memory Usage: 522784kB
-> Parallel Seq Scan on tbl t2 (cost=0.00..47372.99 rows=312499 width=8) (actual time=0.024..47.326 rows=303030 loops=33)
Planning time: 0.251 ms
Execution time: 1939.745 ms
(16 rows)
附加 - 分表+DBLINK异步并行 VS PG 11 HASH JOIN并行
1、创建生成dblink连接的函数，重复创建不报错。
create or replace function conn(
name, -- dblink名字
text -- 连接串,URL
) returns void as $$
declare
begin
perform dblink_connect($1, $2);
return;
exception when others then
return;
end;
$$ language plpgsql strict;
2、生成10亿条数据，并按哈希分成128份（如果是多张表的话，请都按被JOIN的字段进行HASH分区）。
为了测试方便，这里将数据按同样的RANGE分成了128份，实际上使用HASH效果一样。
create unlogged table a(id int, c1 int);
do language plpgsql $$
declare
begin
for i in 0..127 loop
execute format('create unlogged table a%s (like a)', i);
end loop;
end;
$$;
-- 写入10亿测试数据
do language plpgsql $$
declare
begin
for i in 0..127 loop
perform conn('link'||i, 'hostaddr=127.0.0.1 port=9999 user=postgres dbname=postgres');
perform dblink_send_query('link'||i, format('insert into a%s select generate_series(%s, %s), random()*99', i, i*7812500+1, (i+1)*7812500));
end loop;
end;
$$;
3、分片并行JOIN的函数如下
create or replace function get_cnt() returns setof record as $$
declare
begin
for i in 0..127 loop
perform conn('link'||i, 'hostaddr=127.0.0.1 port=9999 user=postgres dbname=postgres');
perform 1 from dblink_get_result('link'||i) as t(c1 int, cnt int8);
perform dblink_send_query('link'||i, format('select t1.c1,count(*) from a%s t1 join a%s t2 using (id) group by t1.c1', i, i));
end loop;
for i in 0..127 loop
return query select * from dblink_get_result('link'||i) as t(c1 int, cnt int8);
end loop;
end;
$$ language plpgsql strict;
使用以上函数将采用DBLINK异步调用并行JOIN，然后我们将结果再合并一遍。
耗时：28秒。
set work_mem ='1GB';
select c1, sum(cnt) from get_cnt() as t(c1 int, cnt int8) group by c1;
postgres=# select c1, sum(cnt) from get_cnt() as t(c1 int, cnt int8) group by c1;
c1 | sum
----+----------
87 | 10100809
71 | 10101080
68 | 10099606
51 | 10099793
70 | 10109225
80 | 10100325
52 | 10099895
84 | 10100478
92 | 10097575
69 | 10107443
60 | 10097840
97 | 10096389
22 | 10100561
59 | 10100344
65 | 10103391
98 | 10099394
73 | 10092200
44 | 10103140
11 | 10103393
42 | 10099439
88 | 10099237
82 | 10099155
40 | 10096921
43 | 10104420
9 | 10098674
15 | 10101089
79 | 10098950
48 | 10102135
26 | 10101603
85 | 10095875
72 | 10099213
95 | 10098356
57 | 10104362
81 | 10110911
19 | 10099394
61 | 10098885
77 | 10095828
30 | 10098553
21 | 10099126
3 | 10100072
17 | 10098163
28 | 10097030
37 | 10105601
5 | 10103695
56 | 10097786
91 | 10099258
74 | 10100289
29 | 10100451
54 | 10094753
4 | 10106395
34 | 10098193
0 | 5051349
96 | 10104673
83 | 10100830
67 | 10102245
63 | 10100960
10 | 10102902
90 | 10100002
35 | 10104661
45 | 10101868
6 | 10104899
86 | 10101989
39 | 10100000
93 | 10105890
89 | 10100522
36 | 10109730
31 | 10102795
50 | 10096268
14 | 10099182
66 | 10101121
13 | 10099546
2 | 10103957
16 | 10100221
62 | 10102741
75 | 10104968
99 | 5048952
41 | 10102728
46 | 10098892
32 | 10102419
53 | 10097987
7 | 10102711
38 | 10105077
12 | 10104542
78 | 10100782
24 | 10100036
25 | 10100035
94 | 10099586
49 | 10099737
47 | 10102668
20 | 10101551
33 | 10099637
1 | 10097499
76 | 10102961
18 | 10106307
64 | 10104492
55 | 10100018
27 | 10101503
23 | 10098148
58 | 10098593
8 | 10103127
(100 rows)
Time: 27911.868 ms (00:27.912)
4、PG 11 并行HASH JOIN（10亿不过滤 JOIN 10亿）
耗时：109秒。
create unlogged table b (like a );
insert into b select * from a;
postgres=# explain analyze select t1.c1,count(*) from b t1 join b t2 using (id) group by t1.c1;
QUERY PLAN
----------------------------------------------------------------------------------------------------------------------------------------------------------------------
Finalize GroupAggregate (cost=433359296530.36..433359296628.36 rows=200 width=12) (actual time=106183.949..106186.496 rows=100 loops=1)
Group Key: t1.c1
-> Sort (cost=433359296530.36..433359296562.36 rows=12800 width=12) (actual time=106183.905..106185.006 rows=6500 loops=1)
Sort Key: t1.c1
Sort Method: quicksort Memory: 497kB
-> Gather (cost=433359295655.16..433359295657.16 rows=12800 width=12) (actual time=105102.267..106182.433 rows=6500 loops=1)
Workers Planned: 64
Workers Launched: 64
-> Partial HashAggregate (cost=433359295655.16..433359295657.16 rows=200 width=12) (actual time=105132.243..105132.331 rows=100 loops=65)
Group Key: t1.c1
-> Parallel Hash Join (cost=4837377.52..42734253467.66 rows=78125008437500 width=4) (actual time=47567.418..100415.264 rows=15384615 loops=65)
Hash Cond: (t1.id = t2.id)
-> Parallel Seq Scan on b t1 (cost=0.00..4581029.01 rows=15625001 width=8) (actual time=0.024..4849.451 rows=15384615 loops=65)
-> Parallel Hash (cost=4581029.01..4581029.01 rows=15625001 width=4) (actual time=19059.992..19059.992 rows=15384615 loops=65)
Buckets: 8388608 Batches: 256 Memory Usage: 219264kB
-> Parallel Seq Scan on b t2 (cost=0.00..4581029.01 rows=15625001 width=4) (actual time=0.036..4966.944 rows=15384615 loops=65)
Planning time: 0.089 ms
Execution time: 123809.025 ms
(18 rows)
postgres=# select t1.c1,count(*) from b t1 join b t2 using (id) group by t1.c1;
c1 | count
----+----------
0 | 5051349
1 | 10097499
2 | 10103957
3 | 10100072
4 | 10106395
5 | 10103695
6 | 10104899
7 | 10102711
8 | 10103127
9 | 10098674
10 | 10102902
11 | 10103393
12 | 10104542
13 | 10099546
14 | 10099182
15 | 10101089
16 | 10100221
17 | 10098163
18 | 10106307
19 | 10099394
20 | 10101551
21 | 10099126
22 | 10100561
23 | 10098148
24 | 10100036
25 | 10100035
26 | 10101603
27 | 10101503
28 | 10097030
29 | 10100451
30 | 10098553
31 | 10102795
32 | 10102419
33 | 10099637
34 | 10098193
35 | 10104661
36 | 10109730
37 | 10105601
38 | 10105077
39 | 10100000
40 | 10096921
41 | 10102728
42 | 10099439
43 | 10104420
44 | 10103140
45 | 10101868
46 | 10098892
47 | 10102668
48 | 10102135
49 | 10099737
50 | 10096268
51 | 10099793
52 | 10099895
53 | 10097987
54 | 10094753
55 | 10100018
56 | 10097786
57 | 10104362
58 | 10098593
59 | 10100344
60 | 10097840
61 | 10098885
62 | 10102741
63 | 10100960
64 | 10104492
65 | 10103391
66 | 10101121
67 | 10102245
68 | 10099606
69 | 10107443
70 | 10109225
71 | 10101080
72 | 10099213
73 | 10092200
74 | 10100289
75 | 10104968
76 | 10102961
77 | 10095828
78 | 10100782
79 | 10098950
80 | 10100325
81 | 10110911
82 | 10099155
83 | 10100830
84 | 10100478
85 | 10095875
86 | 10101989
87 | 10100809
88 | 10099237
89 | 10100522
90 | 10100002
91 | 10099258
92 | 10097575
93 | 10105890
94 | 10099586
95 | 10098356
96 | 10104673
97 | 10096389
98 | 10099394
99 | 5048952
(100 rows)
Time: 108907.886 ms (01:48.908)
执行过程与效果类似于PostgreSQL 11的分区表智能并行join，也类似Greenplum的多分区并行JOIN。
并行聚合的原理可以参考一下文章。
《PostgreSQL 11 preview - 分区表智能并行JOIN (已类似MPP架构，性能暴增)》
《HybridDB PostgreSQL “Sort、Group、distinct 聚合、JOIN” 不惧怕数据倾斜的黑科技和原理 - 多阶段聚合》
《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)》
《Postgres-XC customized aggregate introduction》
《PostgreSQL aggregate function customize》
5、PostgreSQL 11 智能分区表并行JOIN，耗时57秒。
create unlogged table c(like a) partition by range(id);
do language plpgsql $$
declare
begin
for i in 0..127 loop
execute format('create unlogged table c%s partition of c for values from (%s) to (%s)', i, i*7812500+1, (i+1)*7812500+1);
end loop;
end;
$$;
insert into c select * from b;
set enable_parallel_append =on;
set enable_partition_wise_join =on;
set min_parallel_table_scan_size =0;
set min_parallel_index_scan_size =0;
set parallel_tuple_cost =0;
set parallel_setup_cost =0;
set work_mem ='512MB';
set max_parallel_workers_per_gather =64;
alter table c1 set (parallel_workers =128);
explain select t1.c1,count(*) from c t1 join c t2 using (id) group by t1.c1;
postgres=# explain select t1.c1,count(*) from c t1 join c t2 using (id) group by t1.c1;
QUERY PLAN
-----------------------------------------------------------------------------------------------------------------------
Finalize GroupAggregate (cost=16632154923.74..16632155021.74 rows=200 width=12)
Group Key: t1.c1
-> Sort (cost=16632154923.74..16632154955.74 rows=12800 width=12)
Sort Key: t1.c1
-> Gather (cost=16632154048.53..16632154050.53 rows=12800 width=12)
Workers Planned: 64
-> Partial HashAggregate (cost=16632154048.53..16632154050.53 rows=200 width=12)
Group Key: t1.c1
-> Parallel Append (cost=52147.33..13580322797.97 rows=610366250112 width=4)
-> Parallel Hash Join (cost=52147.33..106910564.12 rows=30518312504 width=4)
Hash Cond: (t1.id = t2.id)
-> Parallel Seq Scan on c0 t1 (cost=0.00..42381.59 rows=781259 width=8)
-> Parallel Hash (cost=42381.59..42381.59 rows=781259 width=4)
-> Parallel Seq Scan on c0 t2 (cost=0.00..42381.59 rows=781259 width=4)
-> Parallel Hash Join (cost=52147.33..106910564.12 rows=30518312504 width=4)
Hash Cond: (t1_2.id = t2_2.id)
-> Parallel Seq Scan on c2 t1_2 (cost=0.00..42381.59 rows=781259 width=8)
-> Parallel Hash (cost=42381.59..42381.59 rows=781259 width=4)
-> Parallel Seq Scan on c2 t2_2 (cost=0.00..42381.59 rows=781259 width=4)
-> Parallel Hash Join (cost=52147.33..106910564.12 rows=30518312504 width=4)
Hash Cond: (t1_3.id = t2_3.id)
-> Parallel Seq Scan on c3 t1_3 (cost=0.00..42381.59 rows=781259 width=8)
-> Parallel Hash (cost=42381.59..42381.59 rows=781259 width=4)
-> Parallel Seq Scan on c3 t2_3 (cost=0.00..42381.59 rows=781259 width=4)
.....................
-> Parallel Hash Join (cost=52147.33..106910564.12 rows=30518312504 width=4)
Hash Cond: (t1_126.id = t2_126.id)
-> Parallel Seq Scan on c126 t1_126 (cost=0.00..42381.59 rows=781259 width=8)
-> Parallel Hash (cost=42381.59..42381.59 rows=781259 width=4)
-> Parallel Seq Scan on c126 t2_126 (cost=0.00..42381.59 rows=781259 width=4)
-> Parallel Hash Join (cost=52147.33..106910564.12 rows=30518312504 width=4)
Hash Cond: (t1_127.id = t2_127.id)
-> Parallel Seq Scan on c127 t1_127 (cost=0.00..42381.59 rows=781259 width=8)
-> Parallel Hash (cost=42381.59..42381.59 rows=781259 width=4)
-> Parallel Seq Scan on c127 t2_127 (cost=0.00..42381.59 rows=781259 width=4)
-> Parallel Hash Join (cost=37315.62..2681155.31 rows=4768486329 width=4)
Hash Cond: (t1_1.id = t2_1.id)
-> Parallel Seq Scan on c1 t1_1 (cost=0.00..35789.72 rows=122072 width=8)
-> Parallel Hash (cost=35789.72..35789.72 rows=122072 width=4)
-> Parallel Seq Scan on c1 t2_1 (cost=0.00..35789.72 rows=122072 width=4)
(649 rows)
Time: 11.169 ms
postgres=# select t1.c1,count(*) from c t1 join c t2 using (id) group by t1.c1;
c1 | count
----+----------
0 | 5051349
1 | 10097499
2 | 10103957
3 | 10100072
4 | 10106395
5 | 10103695
6 | 10104899
.............
94 | 10099586
95 | 10098356
96 | 10104673
97 | 10096389
98 | 10099394
99 | 5048952
(100 rows)
Time: 56974.933 ms (00:56.975)
小结
1000万 与 1000万 求交、差的性能指标：
| 方法 | 求交 | 求差 |
|---|---|---|
| 原生intersect, except | 31.7秒 | 30秒 |
| 自定义切片+dblink异步调用 | 1.7秒 | 1.7秒 |
| PostgreSQL 11 并行hashjoin | 3.3秒 | 1.9秒 |
通过改写SQL，PostgreSQL 11可以利用并行计算，更好的支撑求数据交、差的性能。(但是需要注意，NULL值在except, intersect中会视为相同，而join时取等的话，是匹配不到的。这个特别需要注意。（所以语义上不完全一样）)
postgres=# select 1,null except select 1,null;
?column? | ?column?
----------+----------
(0 rows)
postgres=# select 1,null intersect select 1,null;
?column? | ?column?
----------+----------
1 |
(1 row)
如果要让语义完全一样，可以用这种写法，但是就用不到hashjoin了。
即: 等号改成 is not distinct from
select t1.* from tbl t1 join tbl t2 on ((t1.id is not distinct from t2.id) and (t1.c1 is not distinct from t2.c1));
而使用dblink异步的方式，需要注意:
因为我们使用dblink的方法进行HASH分片是扫全表得到的，所以开的并发越多，扫描次数越多。最好是扫一次，并均分到N个临时空间，然后再从临时空间中扫，这样就只需要扫一遍。当然会增加复杂度，如果表不大，实际上多扫几次也无所谓。
10亿 一对一JOIN 10亿（不带过滤条件）
| 方法 | 耗时 |
|---|---|
| dblink + 分区并行JOIN | 28 秒 |
| PostgreSQL 11 分区并行JOIN | 57 秒 |
| PostgreSQL 11 单表HASH并行JOIN | 109 秒 |
参考
《惊天性能！单RDS PostgreSQL实例 支撑 2000亿 - 实时标签透视案例》
《PostgreSQL 11 preview - parallel hash join(并行哈希JOIN) 性能极大提升》
https://www.postgresql.org/docs/10/static/dblink.html
《阿里云RDS PostgreSQL OSS 外部表 - (dblink异步调用封装)并行写提速案例》
《金融风控、公安刑侦、社会关系、人脉分析等需求分析与数据库实现 - PostgreSQL图数据库场景应用》
《PostgreSQL 11 preview - parallel hash (含hash JOIN , hash agg等) 性能极大提升》