Oracle中比對2張表之間數據是否一致的幾種方法

轉自：http://www.askmaclean.com/archives/oracle-compare-data-between-tables-method.html

Oracle中比對2張表之間數據是否一致的幾種方法

注意以下幾種數據比對方式適用的前提條件： 

1. 所要比對的表的結構是一致的
2. 比對過程中源端和 目標端 表上的數據都是靜態的，沒有任何DML修改

方式1：

假設你所要進行數據比對的數據庫其中有一個版本爲11g且該表上有相應的主鍵索引(primary key index)或者唯一非空索引(unique key &not null)的話，那麼恭喜你！　你可以藉助11g 新引入的專門做數據對比的PL/SQL Package dbms_comparison來實現數據校驗的目的，如以下演示： 

Source 源端版本爲11gR2 :

conn maclean/maclean
SQL> select * from v$version;

BANNER
--------------------------------------------------------------------------------
Oracle Database 11g Enterprise Edition Release 11.2.0.3.0 - 64bit Production
PL/SQL Release 11.2.0.3.0 - Production
CORE    11.2.0.3.0      Production
TNS for Linux: Version 11.2.0.3.0 - Production
NLSRTL Version 11.2.0.3.0 - Production

SQL> select * from global_name;

GLOBAL_NAME
--------------------------------------------------------------------------------
www.oracledatabase12g.com  & www.askmaclean.com

 drop table test1;
 create table test1 tablespace users as select object_id t1,object_name t2 from dba_objects where object_id is not null;
 alter table test1 add primary key(t1);
 exec dbms_stats.gather_table_stats('MACLEAN','TEST1',cascade=>TRUE);

create database link maclean connect to maclean identified by maclean using 'G10R21';
Database link created.

 

以上源端數據庫版本爲11.2.0.3 ， 源表結構爲test1(t1 number primary key,t2 varchar2(128)，透過dblink鏈接到版本爲10.2.0.1的目標端

 

conn maclean/maclean

SQL> select * from v$version

BANNER
----------------------------------------------------------------
Oracle Database 10g Enterprise Edition Release 10.2.0.1.0 - 64bi
PL/SQL Release 10.2.0.1.0 - Production
CORE    10.2.0.1.0      Production
TNS for Linux: Version 10.2.0.1.0 - Production
NLSRTL Version 10.2.0.1.0 - Production

create table test2 tablespace users as select object_id t1,object_name t2
from dba_objects where object_id is not null;
alter table test2 add primary key(t1);
exec dbms_stats.gather_table_stats('MACLEAN','TEST2',cascade=>TRUE);

 

目標端版本爲10.2.0.1 ， 表結構爲test2(t1 number primary key,t2 varchar2(128))。

注意這裏2張表上均必須有相同的主鍵索引或者僞主鍵索引(pseudoprimary key僞主鍵要求是唯一鍵且所有的成員列均是非空NOT NULL)。

實際創建comparison對象，並實施校驗：

 

begin
  dbms_comparison.create_comparison(comparison_name    => 'MACLEAN_TEST_COM',
                                    schema_name        => 'MACLEAN',
                                    object_name        => 'TEST1',
                                    dblink_name        => 'MACLEAN',
                                    remote_schema_name => 'MACLEAN',
                                    remote_object_name => 'TEST2',
                                    scan_mode          => dbms_comparison.CMP_SCAN_MODE_FULL);
end;

PL/SQL procedure successfully completed.

SQL> set linesize 80 pagesize 1400

SQL> select * from user_comparison where comparison_name='MACLEAN_TEST_COM';

COMPARISON_NAME                COMPA SCHEMA_NAME
------------------------------ ----- ------------------------------
OBJECT_NAME                    OBJECT_TYPE       REMOTE_SCHEMA_NAME
------------------------------ ----------------- ------------------------------
REMOTE_OBJECT_NAME             REMOTE_OBJECT_TYP
------------------------------ -----------------
DBLINK_NAME
--------------------------------------------------------------------------------
SCAN_MODE SCAN_PERCENT
--------- ------------
CYCLIC_INDEX_VALUE
--------------------------------------------------------------------------------
NULL_VALUE
--------------------------------------------------------------------------------
LOCAL_CONVERGE_TAG
--------------------------------------------------------------------------------
REMOTE_CONVERGE_TAG
--------------------------------------------------------------------------------
MAX_NUM_BUCKETS MIN_ROWS_IN_BUCKET
--------------- ------------------
LAST_UPDATE_TIME
---------------------------------------------------------------------------
MACLEAN_TEST_COM               TABLE MACLEAN
TEST1                          TABLE             MACLEAN
TEST2                          TABLE
MACLEAN
FULL

ORA$STREAMS$NV

           1000              10000
20-DEC-11 01.08.44.562092 PM

 

利用dbms_comparison.create_comparison創建comparison後，新建的comparison會出現在user_comparison視圖中；

以上我們完成了comparison的創建，但實際的校驗仍未發生我們利用10046事件監控這個數據對比過程：

 

conn maclean/maclean
set timing on;
alter system flush shared_pool;

alter session set events '10046 trace name context forever,level 8';

set serveroutput on

DECLARE
  retval dbms_comparison.comparison_type;
BEGIN
  IF dbms_comparison.compare('MACLEAN_TEST_COM', retval, perform_row_dif => TRUE) THEN
    dbms_output.put_line('No Differences');
  ELSE
    dbms_output.put_line('Differences Found');
  END IF;
END;
/

Differences Found           =====> 返回結果爲Differences Found，說明數據存在差異並不一致

PL/SQL procedure successfully completed.

Elapsed: 00:00:10.87

===========================10046 tkprof result =========================

SELECT MIN("T1"), MAX("T1")
FROM
 "MACLEAN"."TEST1"

SELECT MIN("T1"), MAX("T1")
FROM
 "MACLEAN"."TEST2"@MACLEAN

SELECT COUNT(1)
FROM
 "MACLEAN"."TEST1" s WHERE ("T1" >= :scan_min AND "T1" <= :scan_max )

SELECT COUNT(1)
FROM
 "MACLEAN"."TEST2"@MACLEAN s WHERE ("T1" >= :scan_min AND "T1" <= :scan_max )

SELECT q.wb1, min(q."T1") min_range1, max(q."T1") max_range1, count(*)
  num_rows, sum(q.s_hash) sum_range_hash
FROM
 (SELECT /*+ FULL(s) */  width_bucket(s."T1", :scan_min1, :scan_max_inc1,
  :num_buckets) wb1, s."T1", ora_hash(NVL(to_char(s."T1"), 'ORA$STREAMS$NV'),
  4294967295, ora_hash(NVL((s."T2"), 'ORA$STREAMS$NV'), 4294967295, 0))
  s_hash FROM "MACLEAN"."TEST1" s WHERE (s."T1">=:scan_min1 AND s."T1"<=
  :scan_max1) ) q GROUP BY q.wb1 ORDER BY q.wb1

SELECT /*+ REMOTE_MAPPED */ q.wb1, min(q."T1") min_range1, max(q."T1")
  max_range1, count(*) num_rows, sum(q.s_hash) sum_range_hash
FROM
 (SELECT /*+ FULL(s) REMOTE_MAPPED */  width_bucket(s."T1", :scan_min1,
  :scan_max_inc1, :num_buckets) wb1, s."T1", ora_hash(NVL(to_char(s."T1"),
  'ORA$STREAMS$NV'), 4294967295, ora_hash(NVL((s."T2"), 'ORA$STREAMS$NV'),
  4294967295, 0)) s_hash FROM "MACLEAN"."TEST2"@MACLEAN s WHERE (s."T1">=
  :scan_min1 AND s."T1"<=:scan_max1) ) q GROUP BY q.wb1 ORDER BY q.wb1

SELECT /*+ FULL(P) +*/ * FROM "MACLEAN"."TEST2" P

SELECT /*+ FULL ("A1") */
 WIDTH_BUCKET("A1"."T1", :SCAN_MIN1, :SCAN_MAX_INC1, :NUM_BUCKETS),
 MIN("A1"."T1"),
 MAX("A1"."T1"),
 COUNT(*),
 SUM(ORA_HASH(NVL(TO_CHAR("A1"."T1"), 'ORA$STREAMS$NV'),
              4294967295,
              ORA_HASH(NVL("A1"."T2", 'ORA$STREAMS$NV'), 4294967295, 0)))
  FROM "MACLEAN"."TEST2" "A1"
 WHERE "A1"."T1" >= :SCAN_MIN1
   AND "A1"."T1" <= :SCAN_MAX1
 GROUP BY WIDTH_BUCKET("A1"."T1", :SCAN_MIN1, :SCAN_MAX_INC1, :NUM_BUCKETS)
 ORDER BY WIDTH_BUCKET("A1"."T1", :SCAN_MIN1, :SCAN_MAX_INC1, :NUM_BUCKETS)

SELECT ROWID, "T1", "T2"
  FROM "MACLEAN"."TEST2" "R"
 WHERE "T1" >= :1
   AND "T1" <= :2

--------------------------------------------------------------------------------------------
| Id  | Operation                    | Name        | Rows  | Bytes | Cost (%CPU)| Time     |
--------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT             |             |   126 |  3528 |     4   (0)| 00:00:01 |
|*  1 |  FILTER                      |             |       |       |            |          |
|   2 |   TABLE ACCESS BY INDEX ROWID| TEST2       |   126 |  3528 |     4   (0)| 00:00:01 |
|*  3 |    INDEX RANGE SCAN          | SYS_C006255 |   227 |       |     2   (0)| 00:00:01 |
--------------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   1 - filter(TO_NUMBER(:1)<=TO_NUMBER(:2))
   3 - access("T1">=TO_NUMBER(:1) AND "T1"<=TO_NUMBER(:2))

SELECT ll.l_rowid, rr.r_rowid, NVL(ll."T1", rr."T1") idx_val
FROM
 (SELECT l.rowid l_rowid, l."T1", ora_hash(NVL(to_char(l."T1"),
  'ORA$STREAMS$NV'), 4294967295, ora_hash(NVL((l."T2"), 'ORA$STREAMS$NV'),
  4294967295, 0)) l_hash  FROM "MACLEAN"."TEST1" l WHERE l."T1">=:scan_min1
  AND l."T1"<=:scan_max1 ) ll FULL OUTER JOIN (SELECT /*+ NO_MERGE
  REMOTE_MAPPED */ r.rowid r_rowid, r."T1", ora_hash(NVL(to_char(r."T1"),
  'ORA$STREAMS$NV'), 4294967295, ora_hash(NVL((r."T2"), 'ORA$STREAMS$NV'),
  4294967295, 0)) r_hash FROM "MACLEAN"."TEST2"@MACLEAN r WHERE r."T1">=
  :scan_min1  AND r."T1"<=:scan_max1 ) rr ON  ll."T1"=rr."T1" WHERE ll.l_hash
  IS NULL OR rr.r_hash IS NULL OR ll.l_hash <> rr.r_hash

----------------------------------------------------------------------------------------------------------------
| Id  | Operation                       | Name         | Rows  | Bytes | Cost (%CPU)| Time     | Inst   |IN-OUT|
----------------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT                |              |   190 |   754K|     9  (12)| 00:00:01 |        |      |
|*  1 |  VIEW                           | VW_FOJ_0     |   190 |   754K|     9  (12)| 00:00:01 |        |      |
|*  2 |   HASH JOIN FULL OUTER          |              |   190 |   754K|     9  (12)| 00:00:01 |        |      |
|   3 |    VIEW                         |              |   190 |  7220 |     4   (0)| 00:00:01 |        |      |
|*  4 |     FILTER                      |              |       |       |            |          |        |      |
|   5 |      TABLE ACCESS BY INDEX ROWID| TEST1        |   190 |  5510 |     4   (0)| 00:00:01 |        |      |
|*  6 |       INDEX RANGE SCAN          | SYS_C0013098 |   341 |       |     2   (0)| 00:00:01 |        |      |
|   7 |    VIEW                         |              |   126 |   495K|     4   (0)| 00:00:01 |        |      |
|   8 |     REMOTE                      | TEST2        |   126 |  3528 |     4   (0)| 00:00:01 | MACLE~ | R->S |
----------------------------------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   1 - filter("LL"."L_HASH" IS NULL OR "RR"."R_HASH" IS NULL OR "LL"."L_HASH"<>"RR"."R_HASH")
   2 - access("LL"."T1"="RR"."T1")
   4 - filter(TO_NUMBER(:SCAN_MIN1)<=TO_NUMBER(:SCAN_MAX1))
   6 - access("L"."T1">=TO_NUMBER(:SCAN_MIN1) AND "L"."T1"<=TO_NUMBER(:SCAN_MAX1))

Remote SQL Information (identified by operation id):
----------------------------------------------------

   8 - SELECT ROWID,"T1","T2" FROM "MACLEAN"."TEST2" "R" WHERE "T1">=:1 AND "T1"<=:2 (accessing
       'MACLEAN' )

 

可以看到以上過程中雖然沒有避免對TEST1、TEST2表的全表掃描(FULL TABLE SCAN)， 但是好在實際參與HASH JOIN FULL OUTER 的僅是訪問索引後獲得的少量數據，所以效率還是挺高的。

 

此外可以通過user_comparison_row_dif瞭解實際那些row存在差異，如：

 

SQL> set linesize 80 pagesize 1400
SQL> select *
  2    from user_comparison_row_dif
  3   where comparison_name = 'MACLEAN_TEST_COM'
  4     and rownum < 2;

COMPARISON_NAME                   SCAN_ID LOCAL_ROWID        REMOTE_ROWID
------------------------------ ---------- ------------------ ------------------
INDEX_VALUE
--------------------------------------------------------------------------------
STA LAST_UPDATE_TIME
--- ---------------------------------------------------------------------------
MACLEAN_TEST_COM                       42 AAATWGAAEAAANBrAAB AAANJrAAEAAB8AMAAd
46
DIF 20-DEC-11 01.18.08.917257 PM

 

以上利用dbms_comparison包完成了一次簡單的數據比對，該方法適用於11g以上版本且要求表上有主鍵索引或非空唯一索引， 且不支持以下數據類型字段的比對

•     LONG
•     LONG RAW
•     ROWID
•     UROWID
•     CLOB
•     NCLOB
•     BLOB
•     BFILE
•     User-defined types (including object types, REFs, varrays, and nested tables)
•     Oracle-supplied types (including any types, XML types, spatial types, and media types)

 

 

 

若要比對存有以上類型字段的表，那麼需要在create_comparison時指定column_list參數排除掉這些類型的字段。

方法1 dbms_comparison的優勢在於可以提供詳細的比較信息，且在有適當索引的前提下效率較高。
缺點在於有數據庫版本的要求(at least 11gR1)， 且也不支持LONG 、CLOB等字段的比較。

 

方式2：

利用minus Query 對比數據

這可以說是操作上最簡單的一種方法，如：

 

select * from test1 minus select * from test2@maclean;

-----------------------------------------------------------------------------------------------------
| Id  | Operation           | Name  | Rows  | Bytes |TempSpc| Cost (%CPU)| Time     | Inst   |IN-OUT|
-----------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT    |       | 75816 |  3527K|       |  1163  (40)| 00:00:14 |        |      |
|   1 |  MINUS              |       |       |       |       |            |          |        |      |
|   2 |   SORT UNIQUE       |       | 75816 |  2147K|  2984K|   710   (1)| 00:00:09 |        |      |
|   3 |    TABLE ACCESS FULL| TEST1 | 75816 |  2147K|       |   104   (1)| 00:00:02 |        |      |
|   4 |   SORT UNIQUE       |       | 50467 |  1379K|  1800K|   453   (1)| 00:00:06 |        |      |
|   5 |    REMOTE           | TEST2 | 50467 |  1379K|       |    56   (0)| 00:00:01 | MACLE~ | R->S |
-----------------------------------------------------------------------------------------------------

Remote SQL Information (identified by operation id):
----------------------------------------------------

   5 - SELECT "T1","T2" FROM "TEST2" "TEST2" (accessing 'MACLEAN' )

Select *
  from (select 'MACLEAN.TEST1' "Row Source", a.*
          from (select /*+ FULL(Tbl1)  */
                 T1, T2
                  from MACLEAN.TEST1 Tbl1
                minus
                select /*+ FULL(Tbl2)  */
                 T1, T2
                  from MACLEAN.TEST2@"MACLEAN" Tbl2) A
        union all
        select 'MACLEAN.TEST2@"MACLEAN"', b.*
          from (select /*+ FULL(Tbl2)  */
                 T1, T2
                  from MACLEAN.TEST2@"MACLEAN" Tbl2
                minus
                select /*+ FULL(Tbl1)  */
                 T1, T2
                  from MACLEAN.TEST1 Tbl1) B) Order by 1;

 

MINUS Clause會導致2張表均在本地被全表掃描(TABLE FULL SCAN)，且要求發生SORT排序。 若所對比的表上有大量的數據，那麼排序的代價將會是非常大的， 因此這種方法的效率不高。

方式2 MINUS的優點在於操作簡便，特別適合於小表之間的數據檢驗。
缺點在於 由於SORT排序可能導致在大數據量的情況下效率很低， 且同樣不支持LOB 和 LONG 這樣的大對象。

 

方式3：

使用not exists子句，如：

 

select *
  from test1 a
 where not exists (select 1
          from test2 b
         where a.t1 = b.t1
           and a.t2 = b.t2);

no rows selected

Elapsed: 00:00:00.06

------------------------------------------------------------------------------------
| Id  | Operation          | Name  | Rows  | Bytes |TempSpc| Cost (%CPU)| Time     |
------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT   |       | 75816 |  7996K|       |   691   (1)| 00:00:09 |
|*  1 |  HASH JOIN ANTI    |       | 75816 |  7996K|  3040K|   691   (1)| 00:00:09 |
|   2 |   TABLE ACCESS FULL| TEST1 | 75816 |  2147K|       |   104   (1)| 00:00:02 |
|   3 |   TABLE ACCESS FULL| TEST2 | 77512 |  5979K|       |   104   (1)| 00:00:02 |
------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------

   1 - access("A"."T1"="B"."T1" AND "A"."T2"="B"."T2")

 

 

照理說在數據量較大的情況下not exists使用的HASH JOIN ANTI是在性能上是優於MINUS操作的， 但是當所要比較的表身處不同的2個數據庫(distributed query)時將無法使用HASH JOIN ANTI，而會使用FILTER OPERATION這種效率極低的操作:

 

 

select *
  from test1 a
 where not exists (select 1
          from test2@maclean b
         where a.t1 = b.t1
           and a.t2 = b.t2)
no rows selected

Elapsed: 00:01:05.76

 --------------------------------------------------------------------------------------------
| Id  | Operation          | Name  | Rows  | Bytes | Cost (%CPU)| Time     | Inst   |IN-OUT|
--------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT   |       | 75816 |  2147K|   147K  (1)| 00:29:31 |        |      |
|*  1 |  FILTER            |       |       |       |            |          |        |      |
|   2 |   TABLE ACCESS FULL| TEST1 | 75816 |  2147K|   104   (1)| 00:00:02 |        |      |
|   3 |   REMOTE           | TEST2 |     1 |    29 |     2   (0)| 00:00:01 | MACLE~ | R->S |
--------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------

   1 - filter( NOT EXISTS (SELECT 0 FROM  "B" WHERE "B"."T1"=:B1 AND "B"."T2"=:B2))

Remote SQL Information (identified by operation id):
----------------------------------------------------

   3 - SELECT "T1","T2" FROM "TEST2" "B" WHERE "T1"=:1 AND "T2"=:2 (accessing
       'MACLEAN' )

 

可以從以上執行計劃看到FILTER 操作是十分昂貴的。

 

補充：

有網友反映可以通過增加 unnest hint 讓CBO優化器在遠程子查詢有效的情況下整體考慮整個查詢塊，這樣可以讓執行計劃用上HASH JOIN RIGHT ANTI， 這是我一開始沒有考慮到的。

 

 

select *
  from test1 a
 where not exists (select /*+ unnset */
         1
          from test2@maclean b
         where a.t1 = b.t1
           and a.t2 = b.t2);

           
           

PLAN_TABLE_OUTPUT
------------------------------------------
Plan hash value: 1776635653

------------------------------------------------------------------------------------------------------
| Id  | Operation            | Name  | Rows  | Bytes |TempSpc| Cost (%CPU)| Time     | Inst   |IN-OUT|
------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT     |       | 79815 |    12M|       |   594   (1)| 00:00:08 |        |      |
|*  1 |  HASH JOIN RIGHT ANTI|       | 79815 |    12M|  1816K|   594   (1)| 00:00:08 |        |      |
|   2 |   REMOTE             | TEST2 | 20420 |  1575K|       |    56   (0)| 00:00:01 | MACLE~ | R->S |
|   3 |   TABLE ACCESS FULL  | TEST1 | 79815 |  6157K|       |   104   (1)| 00:00:02 |        |      |
------------------------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   1 - access("A"."T1"="B"."T1" AND "A"."T2"="B"."T2")

Remote SQL Information (identified by operation id):
----------------------------------------------------

   2 - SELECT "T1","T2" FROM "TEST2" "B" (accessing 'MACLEAN' )

 

 

在此基礎上加入ordered hint 可以讓執行計劃使用HASH JOIN ANTI

 

   
 select /*+ ordered */ *
  from test1 a
 where not exists (select /*+ unnset */
         1
          from test2@maclean b
         where a.t1 = b.t1
           and a.t2 = b.t2);  

PLAN_TABLE_OUTPUT
--------------------------------------------------
Plan hash value: 3089912131

----------------------------------------------------------------------------------------------------
| Id  | Operation          | Name  | Rows  | Bytes |TempSpc| Cost (%CPU)| Time     | Inst   |IN-OUT|
----------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT   |       | 79815 |    12M|       |   594   (1)| 00:00:08 |        |      |
|*  1 |  HASH JOIN ANTI    |       | 79815 |    12M|  7096K|   594   (1)| 00:00:08 |        |      |
|   2 |   TABLE ACCESS FULL| TEST1 | 79815 |  6157K|       |   104   (1)| 00:00:02 |        |      |
|   3 |   REMOTE           | TEST2 | 20420 |  1575K|       |    56   (0)| 00:00:01 | MACLE~ | R->S |
----------------------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   1 - access("A"."T1"="B"."T1" AND "A"."T2"="B"."T2")

Remote SQL Information (identified by operation id):
----------------------------------------------------

   3 - SELECT "T1","T2" FROM "TEST2" "B" (accessing 'MACLEAN' )

 

方式3 的優點在於操作簡便， 且當需要對比的表位於同一數據庫時效率要比MINUS方式高，但如果是distributed query分佈式查詢則效率可能會因FILTER操作而急劇下降，這時候需要我們手動添加unnest這樣的SQL提示，以保證執行計劃使用HASH JOIN ANTI操作，這樣能夠保證not exists方式的性能。not exists同樣不支持CLOB等大對象。

 

方式4：

Toad、PL/SQL Developer等圖形化工具都提供了compare table data的功能， 這裏我們以Toad工具爲例，介紹如何使用該工具校驗數據：

 

打開Toad 鏈接數據庫-> 點擊Database-> Compare -> Data

 

 

分別在Source 1和Source 2對話框中輸入源表和目標表的信息

因爲Toad的底層實際上使用了MINUS操作，所以提高SORT_AREA_SIZE有助於提高compare的性能，若使用AUTO PGA則可以不設置。

 

選擇所要比較的列

 

 

 

首先可以比較2張表的行數，點擊Execute計算count

 

 

使用MINUS 找出其中一張表上有，而另一張沒有的行

 

使用MINUS 找出所有的差別

 

 

Toad的compare data功能是基於MINUS實現的，所以效率上並沒有優勢。但是通過圖形界面省去了寫SQL語句的麻煩。這種方法同樣不支持LOB、LONG等對象。

 

方式5：

這是一種別出心裁的做法。 將一行數據的上所有字段合併起來，並使用dbms_utility.get_hash_value對合並後的中間值取hash value，再將所有這些從各行所獲得的hash值sum累加， 若2表的hash累加值相等則判定2表的數據一致。

 

簡單來說，如下面這樣：

 

create table hash_one as select object_id t1,object_name t2 from dba_objects;

select dbms_utility.get_hash_value(t1||t2,0,power(2,30)) from hash_one where rownum <3;

DBMS_UTILITY.GET_HASH_VALUE(T1||T2,0,POWER(2,30))
-------------------------------------------------
                                         89209477
                                        757190129

select sum(dbms_utility.get_hash_value(t1||t2,0,power(2,30))) from hash_one;

SUM(DBMS_UTILITY.GET_HASH_VALU
------------------------------
                40683165992756

select sum(dbms_utility.get_hash_value(object_id||object_name,0,power(2,30))) from dba_objects;

SUM(DBMS_UTILITY.GET_HASH_VALU
------------------------------
                40683165992756

 

 

對於列較多的表，手動去構造所有字段合併可能會比較麻煩，利用以下SQL可以快速構造出我們所需要的語句：

 

放到PL/SQL Developer等工具中運行，在sqlplus 中可能因ORA-00923: FROM keyword not found where expected出錯

select 'select sum(dbms_utility.get_hash_value('||column_name_path||',0,power(2,30)) ) from '||owner||'.'||table_name||';'  from (select owner,table_name,column_name_path,row_number() over(partition by table_name order by table_name,curr_level desc) column_name_path_rank from (select owner,table_name,column_name,rank,level as curr_level,ltrim(sys_connect_by_path(column_name,'||''|''||'),'||''|''||') column_name_path from (select owner,table_name,column_name,row_number() over(partition by table_name order by table_name,column_name) rank from　dba_tab_columns where owner=UPPER('&OWNER')  and table_name=UPPER('&TABNAME')  order by table_name,column_name) connect by table_name = prior table_name and rank-1 = prior rank)) where column_name_path_rank=1;

 

使用示範：

 

SQL> @get_hash_col
Enter value for owner: SYS
Enter value for tabname: TAB$

'SELECTSUM(DBMS_UTILITY.GET_HASH_VALUE('||COLUMN_NAME_PATH||',0,POWER(2,30)))FROM
--------------------------------------------------------------------------------
select sum(dbms_utility.get_hash_value(ANALYZETIME||'|'||AUDIT$||'|'||AVGRLN||'|
'||AVGSPC||'|'||AVGSPC_FLB||'|'||BLKCNT||'|'||BLOCK#||'|'||BOBJ#||'|'||CHNCNT||'
|'||CLUCOLS||'|'||COLS||'|'||DATAOBJ#||'|'||DEGREE||'|'||EMPCNT||'|'||FILE#||'|'
||FLAGS||'|'||FLBCNT||'|'||INITRANS||'|'||INSTANCES||'|'||INTCOLS||'|'||KERNELCO
LS||'|'||MAXTRANS||'|'||OBJ#||'|'||PCTFREE$||'|'||PCTUSED$||'|'||PROPERTY||'|'||
ROWCNT||'|'||SAMPLESIZE||'|'||SPARE1||'|'||SPARE2||'|'||SPARE3||'|'||SPARE4||'|'
||SPARE5||'|'||SPARE6||'|'||TAB#||'|'||TRIGFLAG||'|'||TS#,0,1073741824) ) from S
YS.TAB$;

利用以上生成的SQL 計算表的sum(hash)值

select sum(dbms_utility.get_hash_value(ANALYZETIME || '|' || AUDIT$ || '|' ||
                                       AVGRLN || '|' || AVGSPC || '|' ||
                                       AVGSPC_FLB || '|' || BLKCNT || '|' ||
                                       BLOCK# || '|' || BOBJ# || '|' ||
                                       CHNCNT || '|' || CLUCOLS || '|' || COLS || '|' ||
                                       DATAOBJ# || '|' || DEGREE || '|' ||
                                       EMPCNT || '|' || FILE# || '|' ||
                                       FLAGS || '|' || FLBCNT || '|' ||
                                       INITRANS || '|' || INSTANCES || '|' ||
                                       INTCOLS || '|' || KERNELCOLS || '|' ||
                                       MAXTRANS || '|' || OBJ# || '|' ||
                                       PCTFREE$ || '|' || PCTUSED$ || '|' ||
                                       PROPERTY || '|' || ROWCNT || '|' ||
                                       SAMPLESIZE || '|' || SPARE1 || '|' ||
                                       SPARE2 || '|' || SPARE3 || '|' ||
                                       SPARE4 || '|' || SPARE5 || '|' ||
                                       SPARE6 || '|' || TAB# || '|' ||
                                       TRIGFLAG || '|' || TS#,
                                       0,
                                       1073741824))
  from SYS.TAB$;

SUM(DBMS_UTILITY.GET_HASH_VALU
------------------------------
                 1646389632463

 

方式5 利用累加整行數據的hash來判定表上數據是否一致， 僅需要對2張表做全表掃描，效率上是這幾種方法中最高的， 且能保證較高的準確率。

 

但是該hash方式存在以下幾點不足：

1. 所有字段合併的整行數據可能超過4000字節，這時會出現ORA-1498錯誤。換而言之使用這種方式的前提是表中任一行的行長不能超過4000 bytes，當然常規情況下很少會有一行數據超過4000 bytes，也可以通過dba_tables.avg_row_len平均行長的統計信息來判定，若avg_row_len<<4000 那麼一般不會有溢出的問題。

2. 該hash 方式僅能幫助判斷 數據是否一致， 而無法提供更多有用的，例如是哪些行不一致等細節信息

3. 同樣的該hash方式對於lob、long字段也無能爲力

相關文章 | Related posts:

1. Extract SQL Plan from AWR