Oracle 9i - Is there any way to trace the amount of rollback statements ?

Question

Hi,
we're running Oracle 9.2.0.5.0 and we would need to monitor the amount of rollback statements executed on a specific table.
Turning on traces for "set events '10046 trace name context forever, level 1'" would be too much invasive given the current load of the system so that was wondering if there was any quicker way to do that.
Thanks in advance!
Mike

Answer 1

Did not want to raise another issue :)
We have a few huge tables which we need to monitor in terms of activities so that we're using triggers to update some oracle sequences keeping track of the action performed on them.
basically the trigger only increase the sequence value.
CREATE OR REPLACE TRIGGER xxx_Summary_Trigger
AFTER INSERT OR UPDATE or delete ON xxx
FOR EACH ROW
declare dummy number(10);
BEGIN
     IF :New.xxx_Status = 'used' THEN
          SELECT used_plus.nextval into dummy from dual;
     ELSIF :New.Card_Status = 'activated' THEN
          SELECT activated_plus.nextval into dummy from dual;
     END IF;
     IF :Old.xxx_Status = 'used' THEN
          SELECT used_minus.nextval into dummy from dual;
     ELSIF :Old.xxx_Status = 'activated' THEN
          SELECT activated_minus.nextval into dummy from dual;
     END IF;
END;
So that each time the xxx_Status is updated we have an updated sequence (nocache) which allows us to get the overall distribution without a long scan on the table (hosting more than 200M records).
I know triggers are not the best way to approach, not to mention they fire before a commit, so that in case of a rollback the resulting sequence would be showing a wrong value.
But this was only a quick and dirt approach, which works fine for the time being. I just wanted to confirm the inconsistencies we are encountering are due to some rollbacks which mislead the trigger, that's why I was interested in getting a stat on the rollback statement execution.
We get discrepancies every now and then, so that leaving traces on for a long time may be disruptive.
Thanks!
Mike