The query I used to check the duplicate key is >> select count(*)

from tbl1

group by col1, col2, col3

...

Thanks for all these solutions. The problem in my system is a little different, because there's NO duplicate key even in the publication side. I suspect it's a sql70 replication...