Need help with Rollback data if there is error in Data load

Question

Hi All,
We are trying to load data to Oracle 11g database. We want a trigger, procedure or something like that to rollback the data if there are errors in load. Is it possible to do rollback after all the records has been parsed ? So if we try to load 100 records and if there are 30 records with error, we want to rollback after all the 100 records are parsed.
Please advice.

Answer 1

>
Thanks for the suggestion. I'll try that option. So currently we are only loading data that is validated and erroneous records are rejected using trigger. So we don't get any invalid data in table. But Now users are saying that all the records should be rejected if there is even one error in the data load.
>
I generally use a much simpler solution for such multi-stage ETL processes.
Each table has a IS_VALID column that defaults to 'N'. Each step of the process only pulls data with the flag set to 'Y'.
That allows me to leave data in the table but guarantee that it won't be processed by subsequent stages. Since most queries that move data from one stage to another ultimately have to read table rows (i.e. they can't just use indexes) it is not a performance issue to add a predicate such as "'AND IS_VALID = 'Y'" to the query that accesses data.
1. add new unvalidated data - automatically flagged an invalid by default of 'N" on IS_VALID column
2. run audit step #1 - capture the primary key/rowid of any row failing the audit.
3. run audit steps #2 through #n - capture error row ids
4. Final step - update the data setting IS_VALID to 'Y' only if a row passes ALL audits: that is, only if there are NO fatal errors for that row captured in the error table.
That process also allows me to capture every single problem that any row has so that I can produce reports for the business users that show everything that is wrong with the data. There are some problems that the business wan't to ignore, others that can be fixed in the staging tables then reprocessed and others that are rejected since they must be fixed in the source system and that can take several days.
For data that can be fixed in the staging tables the data is fixed and then the audit is rerun which will set the IS_VALID flag to 'Y' allowing those 'fixed' rows to be included in the data that feeds the next processing stage.