当前位置：首页 > news >正文

PgSQL-并行查询系列-介绍[译]

news 文章来源：https://blog.csdn.net/yanzongshuai/article/details/132658505 2025/5/11 10:23:26

PgSQL-并行查询系列-介绍

现代CPU模型拥有大量的CPU核心。多年来，数据库应用程序都是并发向数据库发送查询的。查询处理多个表的行时，若可以使用多核，则可以客观地提升性能。PgSQL 9.6引入了并行查询的新特性，开启并行查询后可以大幅提升性能。

1、局限性

1）若所有CPU核心已经饱和，则不要启动并行查询。并行执行会从其他查询中窃取CPU时间，并增加响应时间

2）进一步需要注意：并行处理会显著增加内存使用（需要注意work_mem的值）。因为，每个hash join或者排序操作都会使用work_mem大小的内存。

3）低延迟的OLTP查询并不能通过并行显著提升性能。特别是仅返回1行的查询，若启用并行，性能会变得特烂。

4）并行执行仅支持没有锁谓词的SELECT查询

5）不支持cursor和会挂起的查询

6）windowed 函数和ordered-set聚合函数都不是并行的

7）对于负载已达IO瓶颈的，并没有啥好处

8）没有并行排序算法。然而，排序查询在某些方面仍然可以并行

9）将CTE（WITH...）替换为sub-select以支持并行执行

10）FDW还不支持并行（后面版本可以，注意哪个版本支持）

11）full outer join不支持

12）客户端设置了max_rows，禁止并行执行

13）如果查询中使用了没有标记为PARALLEL SAFE的函数，那他就是单线程执行

14）SERIALIZABLE事务隔离级别禁用并行执行

2、并行顺序扫描

并行顺序扫描很快，原因可能不是并行读，而是将数据访问分散到多个CPU上。现代操作系统给PgSQL的数据文件提供了很好的缓冲机制。预取允许从存储中获取一个块，而不仅是PgSQL请求的块。因此查询性能限制往往不在IO上，它消耗CPU周期：从表数据页中逐行读取；比较行值和WHERE条件

我们执行一个简单查询：

tpch=# explain analyze select l_quantity as sum_qty from lineitem where l_shipdate <= date '1998-12-01' - interval '105' day;
QUERY PLAN
--------------------------------------------------------------------------------------------------------------------------
Seq Scan on lineitem (cost=0.00..1964772.00 rows=58856235 width=5) (actual time=0.014..16951.669 rows=58839715 loops=1)
Filter: (l_shipdate <= '1998-08-18 00:00:00'::timestamp without time zone)
Rows Removed by Filter: 1146337
Planning Time: 0.203 ms
Execution Time: 19035.100 ms

一个顺序扫描，没有聚合，需要产生大量行。因此该查询被一个CPU核心执行。添加聚合SUM()后，可以清晰的看到有2个进程帮助查询：

explain analyze select sum(l_quantity) as sum_qty from lineitem where l_shipdate &lt;= date '1998-12-01' - interval '105' day;
QUERY PLAN 
----------------------------------------------------------------------------------------------------------------------------------------------------
Finalize Aggregate (cost=1589702.14..1589702.15 rows=1 width=32) (actual time=8553.365..8553.365 rows=1 loops=1)
-&gt; Gather (cost=1589701.91..1589702.12 rows=2 width=32) (actual time=8553.241..8555.067 rows=3 loops=1)
Workers Planned: 2
Workers Launched: 2
-&gt; Partial Aggregate (cost=1588701.91..1588701.92 rows=1 width=32) (actual time=8547.546..8547.546 rows=1 loops=3)
-&gt; Parallel Seq Scan on lineitem (cost=0.00..1527393.33 rows=24523431 width=5) (actual time=0.038..5998.417 rows=19613238 loops=3)
Filter: (l_shipdate &lt;= '1998-08-18 00:00:00'::timestamp without time zone)
Rows Removed by Filter: 382112
Planning Time: 0.241 ms
Execution Time: 8555.131 ms

性能提升2.2倍。

3、并行聚合

“Parallel Seq Scan”节点为partial aggregation提供行。“Partial Aggregate”节点先对SUM()进行一次操作。最后“Gather”节点汇总每个进程的SUM值。“Finalize Aggregate”节点进行最后计算。如果你使用了聚合函数，不要忘记标记他们为“parallel safe”。

4、进程个数

可以不重启服务，增加并行进程个数：

alter system set max_parallel_workers_per_gather=4;
select * from pg_reload_conf();
Now, there are 4 workers in explain output:
tpch=# explain analyze select sum(l_quantity) as sum_qty from lineitem where l_shipdate &lt;= date '1998-12-01' - interval '105' day;
QUERY PLAN 
----------------------------------------------------------------------------------------------------------------------------------------------------
Finalize Aggregate (cost=1440213.58..1440213.59 rows=1 width=32) (actual time=5152.072..5152.072 rows=1 loops=1)
-&gt; Gather (cost=1440213.15..1440213.56 rows=4 width=32) (actual time=5151.807..5153.900 rows=5 loops=1)
Workers Planned: 4
Workers Launched: 4
-&gt; Partial Aggregate (cost=1439213.15..1439213.16 rows=1 width=32) (actual time=5147.238..5147.239 rows=1 loops=5)
-&gt; Parallel Seq Scan on lineitem (cost=0.00..1402428.00 rows=14714059 width=5) (actual time=0.037..3601.882 rows=11767943 loops=5)
Filter: (l_shipdate &lt;= '1998-08-18 00:00:00'::timestamp without time zone)
Rows Removed by Filter: 229267
Planning Time: 0.218 ms
Execution Time: 5153.967 ms

我们将并发进程由2改成了4，但是查询仅快1.6599倍。实际上，我们有2个进程+一个leader，配置改好成为4+1。并行最大提升可以：5/3=1.66倍。

5、如何工作？

查询执行总是从“leader”进程开始。Leader进程执行所有非并行动作。其他进程执行相同查询，称为“worker”进程。并行利用Dynamic Backgroud workers基础架构（9.4引入）执行。因此创建3个工作进程的查询可能比传统执行快4倍。

Worker进程使用消息队列（基于共享内存）和leader进行通信。每个进程有2个队列：一个为errors，另一个是tuples。

5、进程使用个数

1）max_parallel_workers_per_gather是workers进程数的最小限制

2）查询执行使用的workers限制为max_parallel_workes

3）最上层的限制是max_worker_processes：后台进程的总数

分配进程失败，会导致使用单进程执行。查询规划器会根据表或索引大小来增加worker个数。min_parallel_table_scan_size和min_parallel_index_scan_size控制该行为。

set min_parallel_table_scan_size='8MB'
8MB table =&gt; 1 worker
24MB table =&gt; 2 workers
72MB table =&gt; 3 workers
x =&gt; log(x / min_parallel_table_scan_size) / log(3) + 1 worker

表比min_parallel_(index|table)_scan_size值每大3倍，PG增加一个worker进程。Workers进程个数不是基于成本的。循环依赖使得复杂的实现变得困难。相反,规划者使用简单的规则。

可以通过ALTER TABLE … SET (parallel_workers = N)来对某个表指定并行进程数。

6、为什么不使用并行

除了并行限制外，PG还会检查代价：

parallel_setup_cost:避免短查询的并行执行。模拟用于内存设置、流程启动和初始通信的时间

parallel_tuple_cost:leader和worker之间通信可能花费很长时间。时间和worker发送的记录数成正比。参数对通信成本进行建模。

7、Nested Loop Join

PgSQL9.6+可以以并行形式执行“Nested loop”。

explain (costs off) select c_custkey, count(o_orderkey)from    customer left outer join orders onc_custkey = o_custkey and o_comment not like '%special%deposits%'group by c_custkey;QUERY PLAN                                      
--------------------------------------------------------------------------------------Finalize GroupAggregateGroup Key: customer.c_custkey-&gt;  Gather MergeWorkers Planned: 4-&gt;  Partial GroupAggregateGroup Key: customer.c_custkey-&gt;  Nested Loop Left Join-&gt;  Parallel Index Only Scan using customer_pkey on customer-&gt;  Index Scan using idx_orders_custkey on ordersIndex Cond: (customer.c_custkey = o_custkey)Filter: ((o_comment)::text !~~ '%special%deposits%'::text)

Gather发生在最后阶段，因此“Nested Loop Left Join”是并行操作。“Parallel Index Only Scan”在版本10才可以使用，和并行顺序扫描类似。c_custkey = o_custkey条件读取每个customer行的order列，因此不是并行。

8、Hash Join

PgSQL11中每个worker构建自己的hash table。因此，4+ workers不能提升性能。新的实现方式：使用一个共享hash table。每个worker可以利用WORK_MEM来构建hash table>

selectl_shipmode,sum(casewhen o_orderpriority = '1-URGENT'or o_orderpriority = '2-HIGH'then 1else 0end) as high_line_count,sum(casewhen o_orderpriority &lt;&gt; '1-URGENT'and o_orderpriority &lt;&gt; '2-HIGH'then 1else 0end) as low_line_count
fromorders,lineitem
whereo_orderkey = l_orderkeyand l_shipmode in ('MAIL', 'AIR')and l_commitdate &lt; l_receiptdateand l_shipdate &lt; l_commitdateand l_receiptdate &gt;= date '1996-01-01'and l_receiptdate &lt; date '1996-01-01' + interval '1' year
group byl_shipmode
order byl_shipmode
LIMIT 1;QUERY PLAN                                                                     -----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------Limit  (cost=1964755.66..1964961.44 rows=1 width=27) (actual time=7579.592..7922.997 rows=1 loops=1)-&gt;  Finalize GroupAggregate  (cost=1964755.66..1966196.11 rows=7 width=27) (actual time=7579.590..7579.591 rows=1 loops=1)Group Key: lineitem.l_shipmode-&gt;  Gather Merge  (cost=1964755.66..1966195.83 rows=28 width=27) (actual time=7559.593..7922.319 rows=6 loops=1)Workers Planned: 4Workers Launched: 4-&gt;  Partial GroupAggregate  (cost=1963755.61..1965192.44 rows=7 width=27) (actual time=7548.103..7564.592 rows=2 loops=5)Group Key: lineitem.l_shipmode-&gt;  Sort  (cost=1963755.61..1963935.20 rows=71838 width=27) (actual time=7530.280..7539.688 rows=62519 loops=5)Sort Key: lineitem.l_shipmodeSort Method: external merge  Disk: 2304kBWorker 0:  Sort Method: external merge  Disk: 2064kBWorker 1:  Sort Method: external merge  Disk: 2384kBWorker 2:  Sort Method: external merge  Disk: 2264kBWorker 3:  Sort Method: external merge  Disk: 2336kB-&gt;  Parallel Hash Join  (cost=382571.01..1957960.99 rows=71838 width=27) (actual time=7036.917..7499.692 rows=62519 loops=5)Hash Cond: (lineitem.l_orderkey = orders.o_orderkey)-&gt;  Parallel Seq Scan on lineitem  (cost=0.00..1552386.40 rows=71838 width=19) (actual time=0.583..4901.063 rows=62519 loops=5)Filter: ((l_shipmode = ANY ('{MAIL,AIR}'::bpchar[])) AND (l_commitdate &lt; l_receiptdate) AND (l_shipdate &lt; l_commitdate) AND (l_receiptdate &gt;= '1996-01-01'::date) AND (l_receiptdate &lt; '1997-01-01 00:00:00'::timestamp without time zone))Rows Removed by Filter: 11934691-&gt;  Parallel Hash  (cost=313722.45..313722.45 rows=3750045 width=20) (actual time=2011.518..2011.518 rows=3000000 loops=5)Buckets: 65536  Batches: 256  Memory Usage: 3840kB-&gt;  Parallel Seq Scan on orders  (cost=0.00..313722.45 rows=3750045 width=20) (actual time=0.029..995.948 rows=3000000 loops=5)Planning Time: 0.977 msExecution Time: 7923.770 ms

TPC-H中的SQL12是并行hash join的一个很好的哪里，每个进程都帮助构建共享hash table。

9、Merge Join

由于merge join的特性，使得不能并行。如果merge join是查询执行的最后阶段，那么不用担心，仍可以使用并行。

-- Query 2 from TPC-H
explain (costs off) select s_acctbal, s_name, n_name, p_partkey, p_mfgr, s_address, s_phone, s_comment
from    part, supplier, partsupp, nation, region
wherep_partkey = ps_partkeyand s_suppkey = ps_suppkeyand p_size = 36and p_type like '%BRASS'and s_nationkey = n_nationkeyand n_regionkey = r_regionkeyand r_name = 'AMERICA'and ps_supplycost = (selectmin(ps_supplycost)from    partsupp, supplier, nation, regionwherep_partkey = ps_partkeyand s_suppkey = ps_suppkeyand s_nationkey = n_nationkeyand n_regionkey = r_regionkeyand r_name = 'AMERICA')
order by s_acctbal desc, n_name, s_name, p_partkey
LIMIT 100;QUERY PLAN                                                
----------------------------------------------------------------------------------------------------------Limit-&gt;  SortSort Key: supplier.s_acctbal DESC, nation.n_name, supplier.s_name, part.p_partkey-&gt;  Merge JoinMerge Cond: (part.p_partkey = partsupp.ps_partkey)Join Filter: (partsupp.ps_supplycost = (SubPlan 1))-&gt;  Gather MergeWorkers Planned: 4-&gt;  Parallel Index Scan using <strong>part_pkey</strong> on partFilter: (((p_type)::text ~~ '%BRASS'::text) AND (p_size = 36))-&gt;  Materialize-&gt;  SortSort Key: partsupp.ps_partkey-&gt;  Nested Loop-&gt;  Nested LoopJoin Filter: (nation.n_regionkey = region.r_regionkey)-&gt;  Seq Scan on regionFilter: (r_name = 'AMERICA'::bpchar)-&gt;  Hash JoinHash Cond: (supplier.s_nationkey = nation.n_nationkey)-&gt;  Seq Scan on supplier-&gt;  Hash-&gt;  Seq Scan on nation-&gt;  Index Scan using idx_partsupp_suppkey on partsuppIndex Cond: (ps_suppkey = supplier.s_suppkey)SubPlan 1-&gt;  Aggregate-&gt;  Nested LoopJoin Filter: (nation_1.n_regionkey = region_1.r_regionkey)-&gt;  Seq Scan on region region_1Filter: (r_name = 'AMERICA'::bpchar)-&gt;  Nested Loop-&gt;  Nested Loop-&gt;  Index Scan using idx_partsupp_partkey on partsupp partsupp_1Index Cond: (part.p_partkey = ps_partkey)-&gt;  Index Scan using supplier_pkey on supplier supplier_1Index Cond: (s_suppkey = partsupp_1.ps_suppkey)-&gt;  Index Scan using nation_pkey on nation nation_1Index Cond: (n_nationkey = supplier_1.s_nationkey)

“Merge Join”节点在“Gather Merge”上。因此merge不使用并行。但是“Parallel Index Scan”仍旧有助于part_pkey。

10、Partition-wise join

PgSQL11默认禁止partition-wise join特性。它有一个很高的规划代价。分区表可以一个分区一个分区的进行join。允许使用更小的hash table。每个per-partition join操作可以并行：

tpch=# set enable_partitionwise_join=t;
tpch=# explain (costs off) select * from prt1 t1, prt2 t2
where t1.a = t2.b and t1.b = 0 and t2.b between 0 and 10000;QUERY PLAN                     
---------------------------------------------------Append-&gt;  Hash JoinHash Cond: (t2.b = t1.a)-&gt;  Seq Scan on prt2_p1 t2Filter: ((b &gt;= 0) AND (b &lt;= 10000))-&gt;  Hash-&gt;  Seq Scan on prt1_p1 t1Filter: (b = 0)-&gt;  Hash JoinHash Cond: (t2_1.b = t1_1.a)-&gt;  Seq Scan on prt2_p2 t2_1Filter: ((b &gt;= 0) AND (b &lt;= 10000))-&gt;  Hash-&gt;  Seq Scan on prt1_p2 t1_1Filter: (b = 0)
tpch=# set parallel_setup_cost = 1;
tpch=# set parallel_tuple_cost = 0.01;
tpch=# explain (costs off) select * from prt1 t1, prt2 t2
where t1.a = t2.b and t1.b = 0 and t2.b between 0 and 10000;QUERY PLAN                         
-----------------------------------------------------------GatherWorkers Planned: 4-&gt;  Parallel Append-&gt;  Parallel Hash JoinHash Cond: (t2_1.b = t1_1.a)-&gt;  Parallel Seq Scan on prt2_p2 t2_1Filter: ((b &gt;= 0) AND (b &lt;= 10000))-&gt;  Parallel Hash-&gt;  Parallel Seq Scan on prt1_p2 t1_1Filter: (b = 0)-&gt;  Parallel Hash JoinHash Cond: (t2.b = t1.a)-&gt;  Parallel Seq Scan on prt2_p1 t2Filter: ((b &gt;= 0) AND (b &lt;= 10000))-&gt;  Parallel Hash-&gt;  Parallel Seq Scan on prt1_p1 t1Filter: (b = 0)

分区连接只有在分区足够大的情况下才能使用并行执行

11、Parallel Append

Parallel Append通常在UNION ALL中。缺点：较小的并行度，因为每个worker进程最终都为一个查询服务。即使启用了4个进程，也会仍旧发起2个：

tpch=# explain (costs off) select sum(l_quantity) as sum_qty from lineitem where l_shipdate &lt;= date '1998-12-01' - interval '105' day union all select sum(l_quantity) as sum_qty from lineitem where l_shipdate &lt;= date '2000-12-01' - interval '105' day;QUERY PLAN                                           
------------------------------------------------------------------------------------------------GatherWorkers Planned: 2-&gt;  Parallel Append-&gt;  Aggregate-&gt;  Seq Scan on lineitemFilter: (l_shipdate &lt;= '2000-08-18 00:00:00'::timestamp without time zone)-&gt;  Aggregate-&gt;  Seq Scan on lineitem lineitem_1Filter: (l_shipdate &lt;= '1998-08-18 00:00:00'::timestamp without time zone)

12、更重要的变量

WORKER_MEM：限制每个进程的使用内存。每个查询：work_mem*processes*joins-->会导致内存使用很大

max_parallel_workers_per_gather:执行器使用多少进程并发执行该节点

max_worker_processes：根据服务器上CPU核数调整进程数

max_parallel_workers:和并发进程数一样

13、总结

从9.6并行查询执行开始,可以显著提高扫描许多行或索引记录的复杂查询的性能。不要忘记在高oltp工作负载的服务器上禁止并行执行。顺序扫描或索引扫描仍然耗费大量资源。如果您没有针对整个数据集运行报表,那么只需添加缺失的索引或使用适当的分区就可以提高查询性能。

原文

https://www.percona.com/blog/parallel-queries-in-postgresql/