前言

最近遇到几个客户在HybridDB上做性能测试时，都遇到Count Distinct的性能调优问题。这里我们总结一下HybridDB中，对Count Distinct的几种处理方式。

我们以一个客户的案例来做说明。客户的典型的业务场景是，在用户行为日志中统计对应类别的行为数，类别有几千个，独立的行为的总量很多，有几千万；为分析行为，要查询一段时间内的基于类别的独立行为数，查询如下（test的建表语句见附录）：

  1. select category, count(distinct actionId) as ct from test_user_log
  2. where receivetime between '2017-03-07 11:00:00' and '2017-03-07 12:00:00' group by category
  3. order by ct desc limit 10;

下面我们针对这个查询，来看一下Count Distinct是怎么处理的。

Count Distinct的基本处理方式

利用explain analyze命令，看一下这个查询执行过程的信息：

  1. test=# explain analyze select category, count(distinct actionId) as ct from test_user_log
  2. where receivetime between '2017-03-07 11:00:00' and '2017-03-07 12:00:00' group by category
  3. order by ct desc limit 10;
  4.                                                                                             QUERY PLAN
  6. --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
  7. -------------------
  8.  Limit  (cost=0.00..431.00 rows=10 width=16)
  9.  Gather Motion 16:1  (slice2; segments: 16)  (cost=5968.98..5968.99 rows=1 width=40)
  10.    Merge Key: ct
  11.    Rows out:  745 rows at destination with 2469 ms to end, start offset by 77 ms.
  12.    ->  Sort  (cost=5968.98..5968.99 rows=1 width=40)
  13.          Sort Key: ct
  14.          Rows out:  Avg 46.6 rows x 16 workers.  Max 55 rows (seg0) with 2461 ms to end, start offset by 85 ms.
  15.          Executor memory:  58K bytes avg, 58K bytes max (seg0).
  16.          Work_mem used:  58K bytes avg, 58K bytes max (seg0). Workfile: (0 spilling, 0 reused)
  17.          ->  GroupAggregate  (cost=5968.94..5968.97 rows=1 width=40)
  18.                Group By: public.test_user_log.category
  19.                Rows out:  Avg 46.6 rows x 16 workers.  Max 55 rows (seg0) with 2460 ms to end, start offset by 85 ms.
  20.                ->  Sort  (cost=5968.94..5968.94 rows=1 width=40)
  21.                      Sort Key: public.test_user_log.category
  22.                      Rows out:  Avg 461.6 rows x 16 workers.  Max 572 rows (seg4) with 2458 ms to end, start offset by 88 ms.
  23.                      Executor memory:  85K bytes avg, 145K bytes max (seg4).
  24.                      Work_mem used:  85K bytes avg, 145K bytes max (seg4). Workfile: (0 spilling, 0 reused)
  25.                      ->  Redistribute Motion 16:16  (slice1; segments: 16)  (cost=5960.60..5968.93 rows=1 width=40)
  26.                            Hash Key: public.test_user_log.category
  27.                            Rows out:  Avg 461.6 rows x 16 workers at destination.  Max 572 rows (seg4) with 2316 ms to first row, 2458 ms to end, start offset by 88 ms.
  28.                            ->  GroupAggregate  (cost=5960.60..5968.91 rows=1 width=40)
  29.                                  Group By: public.test_user_log.category
  30.                                  Rows out:  Avg 461.6 rows x 16 workers.  Max 472 rows (seg7) with 536 ms to first row, 2455 ms to end, start offset by 89 ms.
  31.                                  Executor memory:  318587K bytes avg, 330108K bytes max (seg7).
  32.                                  Work_mem used:  8544K bytes avg, 8544K bytes max (seg0).
  33.                                  Work_mem wanted: 8414K bytes avg, 8472K bytes max (seg14) to lessen workfile I/O affecting 16 workers.
  34.                                  ->  Sort  (cost=5960.60..5963.37 rows=70 width=64)
  35.                                        Sort Key: public.test_user_log.category
  36.                                        Rows out:  Avg 367982.3 rows x 16 workers.  Max 369230 rows (seg8) with 527 ms to first row, 625 ms to end, start offset by 90 ms.
  37.                                        Executor memory:  61433K bytes avg, 61433K bytes max (seg0).
  38.                                        Work_mem used:  61433K bytes avg, 61433K bytes max (seg0). Workfile: (0 spilling, 0 reused)
  39.                                        ->  Append  (cost=0.00..5904.72 rows=70 width=64)
  40.                                              Rows out:  Avg 367982.3 rows x 16 workers.  Max 369230 rows (seg8) with 2.710 ms to first row, 265 ms to end, start offset by 91 ms
  41. .
  42.                                              ->  Append-only Columnar Scan on test_user_log_1_prt_usual test_user_log  (cost=0.00..0.00 rows=1 width=64)
  43.                                                    Filter: receivetime >= '2017-03-07 11:00:00'::timestamp without time zone AND receivetime <= '2017-03-07 12:00:00'::timestamp
  44.  without time zone
  45.                                                    Rows out:  0 rows (seg0) with 0.542 ms to end, start offset by 93 ms.
  46.                                              ->  Append-only Columnar Scan on test_user_log_1_prt_157 test_user_log  (cost=0.00..3134.45 rows=37 width=64)
  47.                                                    Filter: receivetime >= '2017-03-07 11:00:00'::timestamp without time zone AND receivetime <= '2017-03-07 12:00:00'::timestamp
  48.  without time zone
  49.                                                    Rows out:  Avg 367882.1 rows x 16 workers.  Max 369131 rows (seg8) with 2.178 ms to first row, 132 ms to end, start offset by
  50.  91 ms.
  51.                                              ->  Append-only Columnar Scan on test_user_log_1_prt_158 test_user_log  (cost=0.00..2770.27 rows=33 width=64)
  52.                                                    Filter: receivetime >= '2017-03-07 11:00:00'::timestamp without time zone AND receivetime <= '2017-03-07 12:00:00'::timestamp
  53.  without time zone
  54.                                                    Rows out:  Avg 100.2 rows x 16 workers.  Max 124 rows (seg11) with 2.135 ms to first row, 73 ms to end, start offset by 394 m
  55. s.
  56.  Slice statistics:
  57.    Settings:  effective_cache_size=8GB; gp_statistics_use_fkeys=on; optimizer=off
  58.  Optimizer status: legacy query optimizer
  59.  Total runtime: 2546.862 ms
  60. (51 rows)

可以发现，看似很简单的查询，处理流程却有点复杂。整个处理流程大致如下：

  1. Scan （Columnar Scan + Append) -> Sort(category) -> Group by(category) -> Redistribute -> Sort(category) -> Group by(category) -> Sort -> Gather

从各个节点的实际执行时间的记录可以看出，主要的时间花在了前三步，因为这三步完成后，中间结果只有几百行了。我们重点关注这三个步骤。其实这几个步骤的逻辑比较好理解，扫描出来的数据，直接做排序，排序后，再把结果扫描一遍，同时进行聚合运算。

这里需要注意的一个细节是，查询的表test_user_log是按actionId做分布键的，相同的actionId都会分布在同一个节点上，所以每个节点本地按category做分组后，会在每个分组记录分组中出现的不同actionId值，最终的聚合的结果是category加上一个对应的actionId的计数。

这里有个疑问，其实category的唯一值很少（只有几百个），很适合利用Hash的方式做聚合呢，那么为什么没有选择Hash的方式而是采用了Sort的方式呢？

观察上述查询计划中test_user_log_1_prt_157这个表分区的中间结果估计（(cost=0.00..3134.45 rows=37 width=64)），可以发现预估的结果只有37行，而实际是Rows out: Avg 367882.1 rows，即36万多，这说明表的统计信息不准确。

执行一下Analyze来更新统计信息：

  1. test=# analyze test_user_log_1_prt_157;
  2. analyze test_user_log_1_prt_158;ANALYZE
  3. test=# analyze test_user_log_1_prt_158;
  4. ANALYZE

更新统计信息后，再执行一下查询，查询计划果然发生了变化：排序聚合变成了Hash方式聚合！执行时间也由2546.862ms缩短到2099.144ms。

  1.                                                                                                QUERY PLAN
  3. --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
  4. -------------------------
  5.  Limit  (cost=0.00..431.00 rows=10 width=16)
  6.    Gather Motion 16:1  (slice2; segments: 16)  (cost=320695.70..320695.92 rows=10 width=40)
  7.          Merge Key: ct
  8.          ->  Limit  (cost=320695.70..320695.72 rows=1 width=40)
  9.                ->  Sort  (cost=320695.70..320695.94 rows=7 width=40)
  10.                      Sort Key (Limit): ct
  11.                      ->  HashAggregate  (cost=320691.23..320692.46 rows=7 width=40)
  12.                            Group By: partial_aggregation.category
  13.                            ->  HashAggregate  (cost=303756.92..311454.34 rows=38488 width=64)
  14.                                  Group By: public.test_user_log.category, public.test_user_log.actionId
  15.                                  ->  Redistribute Motion 16:16  (slice1; segments: 16)  (cost=280664.69..292980.55 rows=38488 width=64)
  16.                                        Hash Key: public.test_user_log.category
  17.                                        ->  HashAggregate  (cost=280664.69..280664.69 rows=38488 width=64)
  18.                                              Group By: public.test_user_log.category, public.test_user_log.actionId
  19.                                              ->  Append  (cost=0.00..236527.23 rows=367813 width=43)
  20.                                                    ->  Append-only Columnar Scan on test_user_log_1_prt_usual test_user_log  (cost=0.00..0.00 rows=1 width=64)
  21.                                                          Filter: receivetime >= '2017-03-07 11:00:00'::timestamp without time zone AND receivetime <= '2017-03-07 12:00:00'::tim
  22. estamp without time zone
  23.                                                    ->  Append-only Columnar Scan on test_user_log_1_prt_157 test_user_log  (cost=0.00..124267.69 rows=367813 width=42)
  24.                                                          Filter: receivetime >= '2017-03-07 11:00:00'::timestamp without time zone AND receivetime <= '2017-03-07 12:00:00'::tim
  25. estamp without time zone
  26.                                                    ->  Append-only Columnar Scan on test_user_log_1_prt_158 test_user_log  (cost=0.00..112259.54 rows=1 width=42)
  27.                                                          Filter: receivetime >= '2017-03-07 11:00:00'::timestamp without time zone AND receivetime <= '2017-03-07 12:00:00'::tim
  28. estamp without time zone

上面的计划主要流程如下。两次Group by(Hash（category，actionId)）都是为了去除重复值，保证（category，actionId)组合字段值唯一。

  1. Scan （Columnar Scan + Append) -> Group by(Hash(category，actionId)） -> Redistribute(category) -> Group by(Hash(category, acitonId)） -> Group by(Hash（category)） -> Sort -> Gather

那么还有其他可能的处理方式吗？我们知道，HybridDB支持新型的orca优化器，orca考虑更多的查询执行方式。我们下面试试使用orca来生成查询计划。

  1. test=# set optimizer=on;
  2. SET
  3. test=# explain analyze select category, count(distinct actionId) as ct from test_user_log
  4. where receivetime between '2017-03-07 11:00:00' and '2017-03-07 12:00:00' group by category
  5. order by ct desc limit 10;
  6.                                                                                                   QUERY PLAN
  8. --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
  9. -------------------------------
  10.  Limit  (cost=0.00..431.00 rows=1 width=16)
  11.    Rows out:  10 rows with 2690 ms to end, start offset by 0.500 ms.
  12.    ->  Gather Motion 16:1  (slice2; segments: 16)  (cost=0.00..431.00 rows=1 width=16)
  13.          Merge Key: ct
  14.          Rows out:  10 rows at destination with 2690 ms to end, start offset by 0.501 ms.
  15.          ->  Sort  (cost=0.00..431.00 rows=1 width=16)
  16.                Sort Key: ct
  17.                Rows out:  Avg 46.6 rows x 16 workers.  Max 55 rows (seg0) with 2688 ms to end, start offset by 2.356 ms.
  18.                Executor memory:  33K bytes avg, 33K bytes max (seg0).
  19.                Work_mem used:  33K bytes avg, 33K bytes max (seg0). Workfile: (0 spilling, 0 reused)
  20.                ->  GroupAggregate  (cost=0.00..431.00 rows=1 width=16)
  21.                      Group By: category
  22.                      Rows out:  Avg 46.6 rows x 16 workers.  Max 55 rows (seg0) with 2687 ms to first row, 2688 ms to end, start offset by 2.372 ms.
  23.                      ->  Sort  (cost=0.00..431.00 rows=1 width=16)
  24.                            Sort Key: category
  25.                            Rows out:  Avg 461.6 rows x 16 workers.  Max 572 rows (seg4) with 2687 ms to end, start offset by 3.104 ms.
  26.                            Executor memory:  85K bytes avg, 145K bytes max (seg4).
  27.                            Work_mem used:  85K bytes avg, 145K bytes max (seg4). Workfile: (0 spilling, 0 reused)
  28.                            ->  Redistribute Motion 16:16  (slice1; segments: 16)  (cost=0.00..431.00 rows=1 width=16)
  29.                                  Hash Key: category
  30.                                  Rows out:  Avg 461.6 rows x 16 workers at destination.  Max 572 rows (seg4) with 2442 ms to first row, 2687 ms to end, start offset by 3.113 ms
  31. .
  32.                                  ->  Result  (cost=0.00..431.00 rows=1 width=16)
  33.                                        Rows out:  Avg 461.6 rows x 16 workers.  Max 472 rows (seg7) with 1070 ms to first row, 2583 ms to end, start offset by 3.898 ms.
  34.                                        ->  GroupAggregate  (cost=0.00..431.00 rows=1 width=16)
  35.                                              Group By: category
  36.                                              Rows out:  Avg 461.6 rows x 16 workers.  Max 472 rows (seg7) with 1070 ms to first row, 2583 ms to end, start offset by 3.898 ms.
  37.                                              Executor memory:  316808K bytes avg, 328245K bytes max (seg7).
  38.                                              Work_mem used:  8184K bytes avg, 8184K bytes max (seg0).
  39.                                              Work_mem wanted: 8414K bytes avg, 8472K bytes max (seg14) to lessen workfile I/O affecting 16 workers.
  40.                                              ->  Sort  (cost=0.00..431.00 rows=1 width=16)
  41.                                                    Sort Key: category, actionId
  42.                                                    Rows out:  Avg 367982.3 rows x 16 workers.  Max 369230 rows (seg8) with 1064 ms to first row, 1143 ms to end, start offset by
  43.  3.812 ms.
  44.                                                    Executor memory:  143353K bytes avg, 143353K bytes max (seg0).
  45.                                                    Work_mem used:  143353K bytes avg, 143353K bytes max (seg0). Workfile: (0 spilling, 0 reused)
  46.                                                    ->  Sequence  (cost=0.00..431.00 rows=1 width=24)
  47.                                                          Rows out:  Avg 367982.3 rows x 16 workers.  Max 369230 rows (seg8) with 1.905 ms to first row, 241 ms to end, start off
  48. set by 4.032 ms.
  49.                                                          ->  Partition Selector for test_user_log (dynamic scan id: 1)  (cost=10.00..100.00 rows=7 width=4)
  50.                                                                Filter: receivetime >= '2017-03-07 11:00:00'::timestamp without time zone AND receivetime <= '2017-03-07 12:00:00
  51. '::timestamp without time zone
  52.                                                                Partitions selected:  3 (out of 745)
  53.                                                                Rows out:  0 rows (seg0) with 0.013 ms to end, start offset by 4.033 ms.
  54.                                                          ->  Dynamic Table Scan on test_user_log (dynamic scan id: 1)  (cost=0.00..431.00 rows=1 width=24)
  55.                                                                Filter: receivetime >= '2017-03-07 11:00:00'::timestamp without time zone AND receivetime <= '2017-03-07 12:00:00
  56. '::timestamp without time zone
  57.                                                                Rows out:  Avg 367982.3 rows x 16 workers.  Max 369230 rows (seg8) with 1.891 ms to first row, 202 ms to end, sta
  58. rt offset by 4.046 ms.
  59.                                                                Partitions scanned:  Avg 3.0 (out of 745) x 16 workers.  Max 3 parts (seg0).
  60.  Slice statistics:
  61.    (slice0)    Executor memory: 351K bytes.
  62.    (slice1)  * Executor memory: 152057K bytes avg x 16 workers, 152057K bytes max (seg0).  Work_mem: 143353K bytes max, 8472K bytes wanted.
  63.    (slice2)    Executor memory: 363K bytes avg x 16 workers, 423K bytes max (seg4).  Work_mem: 145K bytes max.
  64.  Statement statistics:
  65.    Memory used: 2047000K bytes
  66.    Memory wanted: 34684K bytes
  67.  Settings:  effective_cache_size=8GB; gp_statistics_use_fkeys=on; optimizer=on
  68.  Optimizer status: PQO version 1.609
  69.  Total runtime: 2690.675 ms
  70. (54 rows)

使用orca生成的查询计划，又回到了使用Sort+Groupby的方式来做聚合（这是因为，我们使用Analyze只更新了子分区表的统计信息，而orca只会考虑主表上的统计信息，要想是orca的计划转为使用Hash方式，需要在主表上使用Analyze，这里我们不继续讨论）。而上述使用orca生成的计划，与使用缺省优化器有很大不同。orca的查询计划采用了下面的流程：

  1. Scan (Dynamic Scan) -> Sort (category, actionId) -> Group by (category) -> Redistribute -> Sort (category) -> Group by(category) -> Sort -> Gather

注意，第一次Sort用了(category, actionId）两个字段的组合，但后面的Group by时只适应了category一个字段！这是一种特殊的聚合方式。在做这种聚合时，对应一个不同的category，只需保留一个actionId的计数即可，而不是像在缺省优化器计划中那样，对每个不同的category，需要保留所有不同的actionId值，这样省去了建立类似Hash表的数据结构的时间。但由于Sort的时候用了两个字段，时间消耗比使用一个字段高，导致整个查询计划的性能不如缺省优化器产生的计划。

延伸

上面的讨论所举的例子中的表，正好是以Count Distinct的字段（即actionId）作为分布键的。如果以其他字段作为分布键，会产生不一样的查询计划，但基本原理都是类似的。

另外，我们没有涉及一个查询中涉及多个字段上有Count Distinct的情况，读者可以自行尝试。

附录

  • 建表语句
  1. create  table test_user_log
  2. (
  3.         actionId text,
  4.         code text,
  5.         receiveTime timestamp,
  6.         gmtCreate timestamp,
  7.         category text,
  8.         version text,
  9.         tag text,
  10.         siteId int4
  11. ) 
  12. with  (APPENDONLY=true, ORIENTATION=column, BLOCKSIZE=524288)
  13. distributed by (actionId)
  14. PARTITION BY RANGE (receivetime) 
  15. (START ('2017-03-07') INCLUSIVE END ('2017-03-07') EXCLUSIVE EVERY (INTERVAL '1 hour' ), DEFAULT PARTITION usual);