Unique constraint is not thrown when used MERGE INSERT (alone) via dblink.

Question

We found some interesting behaviour of unique constraint on Merge query when we use Merge When Not Matched Insert (no update query) via a dblink.
In one Schema S1, on Table A1(c1,c2,c3) there is a unique constraint on column (c1,c2).
Column c2 is nullable and has null for some records.
Now i have a table A2 with same defintion as A1 in Schema S2.In S2 , i have a dblink of S1 as S1 itself.
I have data in S2.A2. Here also i have some records with c2 as null and c1 matching with the data of S1.A1.
Now from schema S2,
I am using the following Merge Query,
MERGE INTO S1.A1 target
USING S2.A2 source
ON (target.c1 = source.c1 and target.c2 = source.c2)
WHEN NOT MATCHED
INSERT (c1,c2,c3) values (source.c1, source.c2,source.c3)
WHEN MATCHED
UPDATE c3 = source.c3;
Now when i execute this merge in schema S2,
if i have some data in S1.A1 and S2.A2 having c1 as same and c2 as null, as oracle does not treat two nulls same, it goes for an insert, i have got unique constraint violated error.
But if i execute MERGE INSERT alone, though that record is getting inserted , i am not getting unique constraint violated error.
Oracle version we are using is 10g (10.2...).
Is it a bug in oracle or what could have caused this behaviour.

Answer 1

Dear,
ERROR at line 1:
ORA-00001: unique constraint (SYS_C00137508) violatedYou need to think about two things
(a) read consistency : what was the situation of table_1 when the maching clause has been initially evaluated ; there were 0 rows matching which means the merge operation will be all insert
(b) your matching clause has a problem : the join column must be unique in both tables otherwise the merge will be ambigous. You don't have a unique key on the source table
(c) think that the merge operation will never insert id =1 and then update id = 1 within the same operation. This will never happen
Hope this helps
Mohamed Houri