Mapping aborts instead of records getting audited in error tables

Question

We have several instances when a mapping encounters an invalid number error (source is non-numeric and target is numeric) and the mapping aborts with a "Invalid number" error, instead of the error being logged in the wb_rt_error* tables, and the mapping continuing. The max number of errors is set to 10,000 or something, so that is not an issue. In other instances, it does continue with warning as expected.
I can't seem to find a pattern with this. Can someone tell me why OWB behaves like this? Is there an explaination to this? We really don't want our mappings killed by bad source data. We sould like OWB to complain, log the error, and continue.
Please help.
-Whiskeylover

Answer 1

It sounds as though there is some specific type of data error which is occurring occassionally, and which OWB is not handling properly. As you point out, sometimes OWB does log and move on - sometimes it doesn't. The hard part is going to be figuring out what the exact data is that is causing the problem and resolving it.
On the next day when this happens, in the prod environment run something like:
declare
   cursor get_vals is
     select the_problem_field from the_source_table;
   function validate_field (invar in varchar2)
   return 0
   is
     testnum number;
   begin
     testnum := to_number(invar);
   exception
      when others then return 1;
   end;
begin
    for source_record in get_vals
    loop
       if validate_field(source_record.the_problem_field) = 1 then
           dbms_output.put_line(source_record.the_problem_field);
       end if;
    end;
end;
/This will give you the list of values which would fail the to_number, and you can try looking for anything extraordinary, or use it as a test sample to run through another mapping.
OH!!!! Wait!
Just had a thought.
If the query that is generating the data for the query is doing any formatting, then it could be the cursor that is failing and not the insert. In that case the cursor fails and exits as it cannot continue which would result in the circumstance you are describing. The error audit counting only works if the problem is at the insert portion of the code - not the select.
Open the mapping package body and pull the cursor out. If there is a to_number() or other function being applied against the source field then this almost certainly your problem.
You would be able to test this with a simple query by running a select with that function against the source table to see if it errors.
Case in point:
assume a varchar field that you are converting to date. You have desinged the mapping to use a to_Date() function with a format mask according to what should be in the field: e.g. to_date(the_field, 'yyyyymmdd'), and there is a singe record with the value '2009081111'. The select fails and will not process subsequent rows.
I would bet that this sort of thing is your issue.
Oh, and this error could be coming from any of the fields if you are getting a non-specific error - not neccessarily the field that you think is the problem.
Cheers,
Mike