What's In SSIS Causing Duplicate Records?

I've a dtsx package which runs nightly to do following:

1. select data from a SQL replicated table
2. do some lookups (Lookup, Derived Column, Multicast, Conditional Split, etc.)
3. insert into another SQL table on another server using "Table or view - fast load", rows per batch = 10000, maximum insert commit size = 10000, and "redirect row" on error output on destination to an error log text file.
Once in a while, I found duplicate records in the error log; these rows cannot be inserted into destination table due to primary constraint. For example, transaction_id=111000 appears twice in the error log but it is a unique key in the source table.


My questions:
1. What could be the cause of duplicating rows during ETL in SSIS? I've asked this before and have spent so much time research but still could not find the reason. This link is from my previous post:

http://forums.microsoft.com/MSDN/ShowPost.aspx?PostID=452319&SiteID=1


2. For a daily extract data with over millions of rows, what would be best to set rows per batch, maximum insert commit size, etc? I've read some posts on this forum and decide to use 10000 for both, but once in a while there's just one duplicate rows that causes the whole batch of 10000 rows not committed.

Thanks for any feedbacks.

-Ash