Oracle "HASH GROUP BY"和"SORT GROUP BY"區別以及"無法使用"HASH GROUP BY"的情況
10G以前GROUP BY子句可以返回排序的結果集，即使沒有ORDER BY子句。
原因是因為使用了“SORT GROUP BY”，會自動排序分組字段。
從10G開始以后引入了“HASH GROUP BY”，新的內部排序算法會導致GROUP BY 子句不保證輸出會按分組的列排序，也不保證結果集的順序。
要對分組進行排序，請使用 ORDER BY 子句。
如果未指定 ORDER BY 子句，則檢索行的順序取決於用於從數據庫檢索行的方法。換句話說，這取決於選擇的執行計划。
下邊看下簡單的實驗：
環境：19.13.0.0.0
創建表並插入實驗數據，盡量保證同一會話插入數據保證數據看起來就是無序的，當然實際上也是：
create table zkm (id int,name varchar2(20)); insert into zkm values(1,'a'); insert into zkm values(2,'b'); insert into zkm values(3,'c'); insert into zkm values(9,'i'); insert into zkm values(5,'e'); insert into zkm values(4,'d'); insert into zkm values(8,'h'); insert into zkm values(7,'g'); insert into zkm values(6,'f'); commit;
目標SQL：select id,count(name) from zkm group by id;
參數設置：alter session set statistics_level=all;
使用Hint：NO_USE_HASH_AGGREGATION來禁用“HASH GROUP BY”，這樣目標SQL執行后結果集總是按照ID列進行排序返回。
並且從執行計划看是“SORT GROUP BY”。
17:06:46 ZKM@dev-app73/pdb(9)> select /*+ NO_USE_HASH_AGGREGATION */ id,count(name) from zkm group by id; ID COUNT(NAME) ---------- ----------- 1 1 2 1 3 1 4 1 5 1 6 1 7 1 8 1 9 1 9 rows selected. Elapsed: 00:00:00.01 17:06:47 ZKM@dev-app73/pdb(9)> select * from table(dbms_xplan.display_cursor(null,null,'allstats last')); PLAN_TABLE_OUTPUT -------------------------------------------------------------------------------------------------------------------- SQL_ID a7kukqrrrvrra, child number 1 ------------------------------------- select /*+ NO_USE_HASH_AGGREGATION */ id,count(name) from zkm group by id Plan hash value: 2238836816 ---------------------------------------------------------------------------------------------------------------- | Id | Operation | Name | Starts | E-Rows | A-Rows | A-Time | Buffers | OMem | 1Mem | Used-Mem | ---------------------------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 1 | | 9 |00:00:00.01 | 6 | | | | | 1 | SORT GROUP BY | | 1 | 9 | 9 |00:00:00.01 | 6 | 2048 | 2048 | 2048 (0)| | 2 | TABLE ACCESS FULL| ZKM | 1 | 9 | 9 |00:00:00.01 | 6 | | | | ---------------------------------------------------------------------------------------------------------------- 15 rows selected. Elapsed: 00:00:00.06
去掉Hint后，再次執行返回的結果集則是無序的。
並且從執行計划看是“HASH GROUP BY”。
17:09:33 ZKM@dev-app73/pdb(9)> select id,count(name) from zkm group by id; ID COUNT(NAME) ---------- ----------- 6 1 1 1 7 1 2 1 8 1 5 1 4 1 3 1 9 1 9 rows selected. Elapsed: 00:00:00.01 17:09:34 ZKM@dev-app73/pdb(9)> select * from table(dbms_xplan.display_cursor(null,null,'allstats last')); PLAN_TABLE_OUTPUT ----------------------------------------------------------------------------------------------------------------------- SQL_ID dqw15j89d8r1b, child number 2 ------------------------------------- select id,count(name) from zkm group by id Plan hash value: 201225912 ---------------------------------------------------------------------------------------------------------------- | Id | Operation | Name | Starts | E-Rows | A-Rows | A-Time | Buffers | OMem | 1Mem | Used-Mem | ---------------------------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 1 | | 9 |00:00:00.01 | 6 | | | | | 1 | HASH GROUP BY | | 1 | 9 | 9 |00:00:00.01 | 6 | 1558K| 1558K| 1063K (0)| | 2 | TABLE ACCESS FULL| ZKM | 1 | 9 | 9 |00:00:00.01 | 6 | | | | ---------------------------------------------------------------------------------------------------------------- 14 rows selected. Elapsed: 00:00:00.10
從排序內存使用大小看，“HASH GROUP BY”使用的內存為1063K，“SORT GROUP BY”為2048bytes。
也可以從v$sql_workarea.last_memory_used獲取信息。
由於數據量比較小，構造大量數據后執行速度為：
17:26:32 ZKM@dev-app73/pdb(9)> select id,count(name) from zkm group by id; ID COUNT(NAME) ---------- ----------- 6 2097152 7 2097152 1 2097152 8 2097152 2 2097152 5 2097152 4 2097152 9 2097152 3 2097152 9 rows selected. Elapsed: 00:00:01.65 17:26:34 ZKM@dev-app73/pdb(9)> select /*+ NO_USE_HASH_AGGREGATION */ id,count(name) from zkm group by id; ID COUNT(NAME) ---------- ----------- 1 2097152 2 2097152 3 2097152 4 2097152 5 2097152 6 2097152 7 2097152 8 2097152 9 2097152 9 rows selected. Elapsed: 00:00:03.13
數據量比較大的情況下，“HASH GROUP BY”要更快，當然不能得出“HASH GROUP BY”就一定快的結論。
實際上是因為避免了排序操作所以“HASH GROUP BY”會比”SORT GROUP BY“更快。
無法使用”HASH GROUP BY“的兩種情況
情況1：GROUP BY后有對字段進行ORDER BY。
比如：
17:35:32 ZKM@dev-app73/pdb(9)> select id,count(name) from zkm group by id order by id; ID COUNT(NAME) ---------- ----------- 1 2097152 2 2097152 3 2097152 4 2097152 5 2097152 6 2097152 7 2097152 8 2097152 9 2097152 9 rows selected. Elapsed: 00:00:03.36 17:36:22 ZKM@dev-app73/pdb(9)> select * from table(dbms_xplan.display_cursor(null,null,'allstats last')); PLAN_TABLE_OUTPUT -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- SQL_ID cns02rbymv6b6, child number 0 ------------------------------------- select id,count(name) from zkm group by id order by id Plan hash value: 2238836816 ---------------------------------------------------------------------------------------------------------------- | Id | Operation | Name | Starts | E-Rows | A-Rows | A-Time | Buffers | OMem | 1Mem | Used-Mem | ---------------------------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 1 | | 9 |00:00:03.36 | 28731 | | | | | 1 | SORT GROUP BY | | 1 | 9 | 9 |00:00:03.36 | 28731 | 2048 | 2048 | 2048 (0)| | 2 | TABLE ACCESS FULL| ZKM | 1 | 9 | 18M|00:00:00.49 | 28731 | | | | ---------------------------------------------------------------------------------------------------------------- 14 rows selected. Elapsed: 00:00:00.06
解決方法：使用子查詢先進行GROUP BY操作，然后再外層查詢使用ORDER BY子句進行排序。同時使用/*+ no_merge */防止視圖合並。
17:37:19 ZKM@dev-app73/pdb(9)> select * from (select /*+ no_merge */ id,count(name) from zkm group by id) order by id; ID COUNT(NAME) ---------- ----------- 1 2097152 2 2097152 3 2097152 4 2097152 5 2097152 6 2097152 7 2097152 8 2097152 9 2097152 9 rows selected. Elapsed: 00:00:01.69 17:37:37 ZKM@dev-app73/pdb(9)> select * from table(dbms_xplan.display_cursor(null,null,'allstats last')); PLAN_TABLE_OUTPUT -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- SQL_ID bxh00bg36g809, child number 0 ------------------------------------- select * from (select /*+ no_merge */ id,count(name) from zkm group by id) order by id Plan hash value: 970191995 ------------------------------------------------------------------------------------------------------------------ | Id | Operation | Name | Starts | E-Rows | A-Rows | A-Time | Buffers | OMem | 1Mem | Used-Mem | ------------------------------------------------------------------------------------------------------------------ | 0 | SELECT STATEMENT | | 1 | | 9 |00:00:01.69 | 28731 | | | | | 1 | SORT ORDER BY | | 1 | 9 | 9 |00:00:01.69 | 28731 | 2048 | 2048 | 2048 (0)| | 2 | VIEW | | 1 | 9 | 9 |00:00:01.69 | 28731 | | | | | 3 | HASH GROUP BY | | 1 | 9 | 9 |00:00:01.69 | 28731 | 1558K| 1558K| 1065K (0)| | 4 | TABLE ACCESS FULL| ZKM | 1 | 9 | 18M|00:00:00.48 | 28731 | | | | ------------------------------------------------------------------------------------------------------------------ 17 rows selected. Elapsed: 00:00:00.06
明顯改寫后的SQL執行速度更快。
原因是雖然還是有排序動作但是排序的結果集更更更更小了，從A-Rows看是9行，而不改寫之前是對全部的行排序。
情況2：在聚合函數中多次使用distinct處理不同字段。
如SQL：select id,count(distinct name),count(distinct id) from zkm group by id order by id;
09:01:40 ZKM@dev-app73/pdb(9)> select id,count(distinct name),count(distinct id) from zkm group by id order by id; ID COUNT(DISTINCTNAME) COUNT(DISTINCTID) ---------- ------------------- ----------------- 1 1 1 2 1 1 3 1 1 4 1 1 5 1 1 6 1 1 7 1 1 8 1 1 9 1 1 9 rows selected. Elapsed: 00:00:14.67 09:01:56 ZKM@dev-app73/pdb(9)> select * from table(dbms_xplan.display_cursor(null,null,'allstats last')); PLAN_TABLE_OUTPUT -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- SQL_ID 7ht3gbdz1z5ts, child number 0 ------------------------------------- select id,count(distinct name),count(distinct id) from zkm group by id order by id Plan hash value: 2238836816 ---------------------------------------------------------------------------------------------------------------- | Id | Operation | Name | Starts | E-Rows | A-Rows | A-Time | Buffers | OMem | 1Mem | Used-Mem | ---------------------------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 1 | | 9 |00:00:14.66 | 28731 | | | | | 1 | SORT GROUP BY | | 1 | 1 | 9 |00:00:14.66 | 28731 | 2048 | 2048 | 2048 (0)| | 2 | TABLE ACCESS FULL| ZKM | 1 | 9 | 18M|00:00:01.45 | 28731 | | | | ---------------------------------------------------------------------------------------------------------------- 15 rows selected. Elapsed: 00:00:00.06
可以看出，多個聚合函數中均使用了distinct導致無法用"HASH GROUP BY"，因為兩個distinct需要去重，從結果看，對同一結果集可以同時排序兩個以上不同的字段后做去重然后count，卻無法同時對同一結果集做HASH去重去避免排序。
去掉其中一個distinct的話就沒問題，如：select id,count(distinct name),count(id) from zkm group by id order by id;
09:13:56 ZKM@dev-app73/pdb(9)> select id,count(distinct name),count(id) from zkm group by id order by id; ID COUNT(DISTINCTNAME) COUNT(ID) ---------- ------------------- ---------- 1 1 2097152 2 1 2097152 3 1 2097152 4 1 2097152 5 1 2097152 6 1 2097152 7 1 2097152 8 1 2097152 9 1 2097152 9 rows selected. Elapsed: 00:00:02.08 09:14:02 ZKM@dev-app73/pdb(9)> select * from table(dbms_xplan.display_cursor(null,null,'allstats last')); PLAN_TABLE_OUTPUT ----------------------------------------------------------------------------------------------------------------------------------- SQL_ID 9t4u0dtgn1q0q, child number 0 ------------------------------------- select id,count(distinct name),count(id) from zkm group by id order by id Plan hash value: 1511739550 ---------------------------------------------------------------------------------------------------------------------- | Id | Operation | Name | Starts | E-Rows | A-Rows | A-Time | Buffers | OMem | 1Mem | Used-Mem | ---------------------------------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 1 | | 9 |00:00:02.08 | 28731 | | | | | 1 | SORT GROUP BY | | 1 | 9 | 9 |00:00:02.08 | 28731 | 2048 | 2048 | 2048 (0)| | 2 | VIEW | VW_DAG_0 | 1 | 9 | 9 |00:00:02.08 | 28731 | | | | | 3 | HASH GROUP BY | | 1 | 9 | 9 |00:00:02.08 | 28731 | 1452K| 1452K| 1192K (0)| | 4 | TABLE ACCESS FULL| ZKM | 1 | 9 | 18M|00:00:00.44 | 28731 | | | | ---------------------------------------------------------------------------------------------------------------------- 17 rows selected. Elapsed: 00:00:00.07
解決辦法：暫無。
參考文檔
GROUP BY Clause Does Not Guarantee a Sort Without ORDER BY Clause in 10g and Above (文檔 ID 345048.1)