Imagine that you have to implement a table which must contain one row minimum and
may not have more than one rows. So, on insert of a row you can check if there’s
already a row in the table by joining the table to itself, like the next statement
select 1 from table t1 , table t2 where t1.id != t2.id
The obvious way to do this is to create a set
of database triggers. You cannot use only one row trigger because the query is mutating,
so you need to put the rowid on a stack on a row event trigger and loop
over the rowid’s on stack on a after statement event.
Next create 3 triggers to prevent (sort of) that more than 1 row is
in the table at all times.
Because checking the record count involves a mutating table issue
the rowid is put on a stack and handled in the after statement trigger.
1) a before statement trigger to initialize the stack
2) a (after) row trigger to put the new row on the stack.
I prefer to use an after row trigger above before row because when the after row event fires all
database constraints (when not flagged defered) are finished.
Failures on these constraints are handled immediately and after row triggers do not fire,
thus will cost nothing. It more expensive for a before row to fire when the row fails
due to a constraint violation. So I prefer to let the database do it’s work before I will throw in
some business rules checking.
3) a after statement trigger to perform the singularity check
Here’s the scripting:
create table ags_parameters ( par_id number(4) not null, par_name varchar2(100 byte) ) /
alter table ags_parameters add constraint ags_par_pk primary key(par_id) / create or replace trigger ags_par_bs before delete or insert or update on ags_parameters declare begin if inserting or updating then acc_rowid_stack_light.create_stack; end if; exception when others then raise_application_error(-20010, sqlerrm ); end; /
create or replace trigger ags_par_ar after delete or insert or update on ags_parameters for each row declare flg_par1 constant binary_integer := 1; begin if inserting then acc_rowid_stack_light.push_flag(:new.rowid, flg_par1); elsif deleting then raise_application_error(-20010, 'Delete is not allowed' ); end if; exception when others then raise_application_error(-20010, sqlerrm ); end ags_par_ar; / create or replace trigger ags_par_as after delete or insert or update on ags_parameters declare flg_par1 constant binary_integer := 1; e_locked exception; pragma exception_init (e_locked, -54); t_rowid rowid; t_value binary_integer; begin while acc_rowid_stack_light.pop_flagset( t_rowid, t_value ) loop if acc_rowid_stack_light.check_flag(t_value, flg_par1) then declare cursor c_par(b_rowid in rowid) is select 1 from ags_parameters par1 , ags_parameters par2 where par1.par_id != par2.par_id and par1.rowid = b_rowid for update of par1.par_id; r_par c_par%rowtype; l_par_found boolean; e_par exception; begin open c_par(t_rowid); fetch c_par into r_par; l_par_found := c_par%found; close c_par; if l_par_found then raise e_par; end if; exception when e_par then raise_application_error(-20010, 'Only one row is allowed in this table'); end; end if; end loop; acc_rowid_stack_light.delete_stack; exception when others then raise_application_error(-20010, sqlerrm ); end ags_par_as; /
Now the testing.
First try to insert 2 records and see if it works.
insert into ags_parameters (par_name) values ('test1'); insert into ags_parameters (par_name) values ('test2');
The second row will raise an excpetion “ORA-20010 – Only one row is allowed in this table”. That’s fine.
But is it?
Let’s try another experiment. This time I’m going to use 2 concurrent sessions.
In the first session I insert a row.
prompt Rollback previous test results. rollback; insert into ags_parameters (par_name) values ('session1'); pause; commit;
It returns with “1 row created” Now do NOT commit! Open another session and also insert a row into ags_parameters and commit
prompt second session insert into ags_parameters (par_name) values ('session2'); commit;
It also returns with “1 row created”
Return to the first session and commit that session also. Than count the records.
select count(*) from ags_parameters;
Oh no, 2 records ! ! ! !
create table fbi_parameters ( par_id number(4) not null, par_name varchar2(100 byte) ) /
alter table fbi_parameters add constraint fbi_par_pk primary key(par_id) /
Create a function based index. This index returns cq indexes value 1 for each and every record. Because it is
defined as a UNIQUE index, only one row is permitted in the table.
create unique index fbi_one_row_table_i on fbi_parameters (1) / insert into fbi_parameters (par_name) values ('session 1'); pause; commit;
Do NOT commit! Open another session and also insert a row into fbi_parameters and commit
insert into ags_parameters(par_name) values ('session2'); pause; commit;
It is immediately being locked by the other session.
Return to the first session and issue a commit;
Return to the second session and issue a commit;
The second session stops because “ORA-00001: unique constraint (fbi_one_row_table) violated”
select count(*) from fbi_parameters;
Yippie! Function based index to the rescue.
And as a side effect, less objects (1 table + 1 index + 3 triggers) vs (1 table + 2 indices + 0 triggers)
But that”s only part of the story: It is still possible to delete all record from fbi_parameters.
Here, I guess a database trigger is the only possible solution.
create or replace trigger fbi_par_br after delete or insert or update on ags_parameters for each row begin if deleting then raise_application_error(-20010, 'Deleting the last/only row in this table is not allowed'); end if; end fbi_par_br; /
Table ags_parameters already has a check against deleting rows in the after row trigger. It might
as well be implemented as a before row trigger.