Postgres 8.4 here. Imagine this code snippet from Postgres doc:
CREATE FUNCTION emp_stamp() RETURNS trigger AS $emp_stamp$
BEGIN
-- Check that empname and salary are given
IF NEW.empname IS NULL THEN
RAISE EXCEPTION 'empname cannot be null';
END IF;
IF NEW.salary IS NULL THEN
RAISE EXCEPTION '% cannot have null salary', NEW.empname;
END IF;
-- Who works for us when she must pay for it?
IF NEW.salary < 0 THEN
RAISE EXCEPTION '% cannot have a negative salary', NEW.empname;
END IF;
-- Remember who changed the payroll when
NEW.last_date := current_timestamp;
NEW.last_user := current_user;
RETURN NEW;
END;
$emp_stamp$ LANGUAGE plpgsql;
If we want to do something like logging in a custom table these exceptions:
-- Check that empname and salary are given
IF NEW.empname IS NULL THEN
INSERT INTO my_log_table ('User didn't supplied empname')
RAISE EXCEPTION 'empname cannot be null';
END IF;
It won't work because anything we put before a RAISE EXCEPTION
call is undone by the rollback RAISE EXCEPTION
implies, i.e. the my_log_table row we create will be deleted as soon as RAISE EXCEPTION
is called.
What's is the best way to accomplish something like this? Maybe catching our custom exception?
Turning off rollback @ TRIGGER is not an option, I need it.
You can trap errors / catch exceptions.
In the EXCEPTION
block you can do anything else, like INSERT
into another table. Afterwards you could re-raise the exception to propagate out, but that would roll back the whole transaction including the INSERT
to the log table (unless the exception is wrapped and caught in an outer function).
You could:
use tricks like a dblink call to emulate an autonomous transaction, which isn't undone, when the wrapping transaction is rolled back. Related:
RAISE
a NOTICE
or WARNING
additionally, which also isn't undone by ROOLBACK
.
RAISE
a different EXCEPTION
with your own text.
Alternatively, you can just cancel the row that triggered the trigger function and not raise an exception. Everything else in the transaction goes through normally.
Assuming this is a trigger ON UPDATE
and you have another table with identical structure to write failed INSERT
s to:
CREATE OR REPLACE FUNCTION emp_stamp()
RETURNS trigger
LANGUAGE plpgsql AS
$func$
BEGIN
-- Check that empname and salary are given
IF NEW.empname IS NULL THEN
RAISE EXCEPTION 'empname cannot be null';
END IF;
-- IF ...
RETURN NEW; -- regular end
EXCEPTION WHEN others THEN -- or be more specific
INSERT INTO log_tbl VALUES (NEW.*); -- identical table structure
RETURN NULL; -- cancel row
END
$func$;
Note that NEW
contains the state of the row right before the exception occurred, including previous successful statements in the same function.
Trigger:
CREATE TRIGGER emp_stamp
BEFORE INSERT OR UPDATE ON tbl
FOR EACH ROW EXECUTE FUNCTION emp_stamp();