Consolidate several Oracle triggers. Any performance impact?

Question

Most of our tables have one BEFORE INSERT OR UPDATE trigger, in order to set ID\'s BEFORE INSERT and to set creation / modification users / dates <

Answer 1

I would suggest that you investigate the impact of using the Sequence in the trigger - when creating the sequence seed the sequence with a cache value (CACHE specifies how many values of the sequence the database preallocates and keeps in memory for faster access). I have experienced factors far greater the 8 associated with sequences. In any case, in order to compare the impact of a trigger in the terms of context switching the use of the Sequence should be either eliminated for the test or taken into consideration when evaluating the results.

Answer 2

I know there's a significant difference in redo generation between no triggers and a single trigger, but not about a difference between 1 and n triggers. However, I can't imagine there isnt a hit, as there would be more context switching between the SQL and PL/SQL engines.

Answer 3

I have now benchmarked this situation and I came to the conclusion that there is a significant performance loss most likely due to PL/SQL context switches, when adding 1 trigger. The loss is by factor 8 in my benchmark. Adding a second "compatible" trigger, however, doesn't have any significant impact anymore. By "compatible", I mean both triggers always fire at the same event in any order.

So I'm concluding that there is most likely only 1 SQL -> PL/SQL context switch for all triggers

---

Here's the benchmark:

Create a table

-- A typical table with primary key, creation/modification user/date, and 
-- other data columns
CREATE TABLE test(
  id number(38)    not null, -- pk
  uc varchar2(400) not null, -- creation user
  dc date          not null, -- creation date
  um varchar2(400),          -- modification user
  dm date,                   -- modification date
  data number(38)
);

... and a sequence

CREATE SEQUENCE s_test;

A typical trigger setting ID, creation/modification user/date

CREATE OR REPLACE TRIGGER t_test BEFORE INSERT OR UPDATE
  ON test
  FOR EACH ROW
BEGIN
  IF inserting THEN
    SELECT s_test.nextval INTO :new.id FROM dual;

    :new.uc := USER;
    :new.dc := SYSDATE;
    :new.um := NULL;
    :new.dm := NULL;
  END IF;
  IF updating THEN
    :new.um := USER;
    :new.dm := SYSDATE;
    :new.uc := :old.uc;
    :new.dc := :old.dc;
  END IF;
END t_test;

Insert 1000, 10000, 100000 Records

declare
  procedure run (limit number) is
    t timestamp;
  begin
    t := systimestamp;

    insert into test (data)
    select level from dual connect by level < limit;

    dbms_output.put_line(to_char(systimestamp - t));

    rollback;
  end;
begin
  run(1000);
  run(10000);
  run(100000);
end;

Results

-- ------------------------------------
-- +000000000 00:00:00.086603000
-- +000000000 00:00:00.844333000
-- +000000000 00:00:08.429186000
-- ------------------------------------

Another "compatible" trigger (execution order irrelevant)

CREATE OR REPLACE TRIGGER t_test_other BEFORE INSERT OR UPDATE
  ON test
  FOR EACH ROW
BEGIN
  :new.data := 42;
END t_test_other;

Results of another run of the test script

-- ------------------------------------
-- +000000000 00:00:00.088551000
-- +000000000 00:00:00.876028000
-- +000000000 00:00:08.731345000
-- ------------------------------------

Deactivate triggers

alter trigger t_test disable;
alter trigger t_test_other disable;

Run a slightly different test script

declare
  procedure run (limit number) is
    t timestamp;
  begin
    t := systimestamp;

    insert into test (id, uc, dc, data)
    select s_test.nextval, user, sysdate, level from dual 
    connect by level < limit;

    dbms_output.put_line(to_char(systimestamp - t));

    rollback;
  end;
begin
  run(1000);
  run(10000);
  run(100000);
end;

Results

-- ------------------------------------
-- +000000000 00:00:00.012712000
-- +000000000 00:00:00.104903000
-- +000000000 00:00:01.043984000
-- ------------------------------------