If the exception (I suspect SQLException) is specific enough, you can declare that exception type as retryable-exception-classes in job xml, and this type of failure will be retried automatically.  But if it's a SQLException, which is pretty generic and can indicate many different causes other than duplicate keys.  Also be careful with retry in item writer since the writer accumulates a list of items to write all at once, and if any one of them failed, all items in this chunk will be retried.

I think a composite key that can cause duplicatation is not a good choice.  Even if you can alter your application to circumvent this error, there is still the possibility of conflicting keys.