[CBRD-26522] Support NL join during parallel heap scan

[xmilex-git]

http://jira.cubrid.org/browse/CBRD-26522

Purpose

기존에는 join driving 테이블이 heap scan인 경우, 이를 row-by-row로 처리하여 동기화 오버헤드 및 중앙 처리 방식으로 인해 NL join의 성능이 많이 저하됐었습니다.

이를 개선하기 위해, 각각의 parallel heap scan thread에서 NL join을 평가하고 결과를 각각 수집하게끔 변경하여, mergeable-list 방식으로 처리할 수 있도록 변경합니다.

[xmilex-git]

/run all

[xmilex-git]

/run all

[xmilex-git]

/run all

[xmilex-git]

/run all

[xmilex-git]

/run all

[xmilex-git]

/run all

[xmilex-git]

/run all

[xmilex-git]

/run all

[xmilex-git]

/run all

[xmilex-git]

/run shell

[xmilex-git]

/run shell

[xmilex-git]

/run shell

[Hamkua]

trace 출력 시 partition 정보가 출력되지 않는 것 같습니다.

drop table if exists t1, t2;
create table t1 (id int primary key, col1 varchar(20), col2 varchar(20), col3 varchar(20),col4 varchar(20),col5 varchar(20),col6 varchar(20));

create table t2 (id int primary key, col1 varchar(20), col2 varchar(20), col3 varchar(20),col4 varchar(20),col5 varchar(20),col6 varchar(20)) 
    partition by range(id) ( 
        partition p1 values less than (10000), 
        partition p2 values less than(20000), 
        partition p3 values less than MAXVALUE
);

insert into t1 select rownum,lpad(rownum,20,'0'),lpad(rownum % 5,20,'0'),lpad(rownum,20,'0'),lpad(rownum,20,'0'),lpad(rownum,20,'0'),lpad(rownum,20,'0') 
from db_class a, db_class b, db_class c, db_class d, db_class e, db_class f limit 500000;
insert into t2 select rownum,lpad(rownum,20,'0'),lpad(rownum % 5,20,'0'),lpad(rownum,20,'0'),lpad(rownum,20,'0'),lpad(rownum,20,'0'),lpad(rownum,20,'0') 
from db_class a, db_class b, db_class c, db_class d, db_class e, db_class f limit 100000;

with cte as (
 select /*+ parallel(16) ordered materialize */ a.col1, b.id from t1 a, t2 b 
    where cast(a.col1 as int)/5 = b.id and b.id < 10000
) select /*+ recompile */ 1 from cte limit 1;

기존 row by row 방식:

Trace Statistics:
  SELECT (time: 17480, fetch: 6041553, fetch_time: 11034, ioread: 0)
    SCAN (temp time: 0, fetch: 0, ioread: 0, readrows: 1, rows: 1)
    SUBQUERY (uncorrelated)
      CTE (non_recursive_part)
        SELECT (time: 17480, fetch: 6041553, fetch_time: 11034, ioread: 0)
          SCAN (table: dba.t1), (heap time: 424, fetch: 2040052, ioread: 0, readrows: 500000, rows: 500000)
               (parallel workers: 16, heap time: 470..473, readrows: 24563..33957, rows: 24563..33957, gather: row by row)
            SCAN (index: dba.t2.pk_t2_id), (btree time: 16473, fetch: 4000000, ioread: 0, readkeys: 49999, filteredkeys: 49995, rows: 49995, covered: tr
ue)
                   PARTITION (index: dba.t2__p__p1.pk_t2_id), (btree time: 16473, fetch: 4000000, ioread: 0, readkeys: 49999, filteredkeys: 49995, rows:
 49995, covered: true)
            MEMOIZE (time: 127, hit: 0, miss: 1001, size: 261KB, enabled: false)

mergeable list 방식:

Trace Statistics:
  SELECT (time: 1574, fetch: 6027596, fetch_time: 9336, ioread: 0)
    SCAN (temp time: 0, fetch: 4, ioread: 0, readrows: 1, rows: 1)
    SUBQUERY (uncorrelated)
      CTE (non_recursive_part)
        SELECT (time: 1574, fetch: 6027592, fetch_time: 9336, ioread: 0)
          SCAN (table: dba.t1), (heap time: 1573, fetch: 6027576, ioread: 0, readrows: 500000, rows: 500000)
               (parallel workers: 16, heap time: 1567..1571, readrows: 29876..32801, rows: 29876..32801, gather: mergeable 
list)
            SCAN (index: dba.t2.pk_t2_id), (btree time: 1077, fetch: 4000000, ioread: 0, readkeys: 49999, filteredkeys: 499
95, rows: 49995, covered: true)
            MEMOIZE (time: 0, hit: 0, miss: 16016, size: 3999KB, enabled: false)

[xmilex-git]

b651847 에서 정상동작하게 수정했습니다.