使用DBMS_LOCK手动锁定

在PL/SQL中，我们有时候需要保证某个Procedure不被并发执行。一般采用select for update对一个固定数据表的数据行进行锁定。

除此之外，还可以使用dbms_lock包提供的request()和release()方法进行锁定。

下面例子的逻辑是，在插入A之前用count()判断，如果A中没有数据就插入。在高并发的环境下，没有锁定就可能造成多条插入。为了避免并发错误，使用dbms_lock进行手动控制。

drop table a;
create table a(x int);
drop table b;
create table b(sid number, dt timestamp default systimestamp);


create or replace procedure lock_thing is
  l_result int;
  l_cnt int;
begin
  -- get the lock with ID=0
  -- release_on_commit=false means keep lock until explicitly released or until end-of-session
  l_result := dbms_lock.request(id => 0, release_on_commit => false);

  -------------------------------------------
  -- do your own logic here
  -------------------------------------------
  if l_result = 0 then
    -- running log
    insert into b(sid) values(sys_context('userenv','bg_job_id'));
   
    select count(1) into l_cnt from a;
    if l_cnt = 0 then
      insert into a(x) values(sys_context('userenv','bg_job_id'));
    end if;
  end if;

  commit; 
  -- explicitly release it
  l_result := dbms_lock.release(id => 0);
exception
  when others then
    l_result := dbms_lock.release(id => 0);
end;
/

 用一个程序模拟并发调用procedure，最后的结果是A中仅有一条数据。（NOTE: 这个并发模拟并不严格）

-- use some backgound jobs to simulate concurrent calls
declare
  l_job int;
begin
  delete a;
  delete b;
  commit;
  for i in 1 .. 100 loop
    dbms_job.submit(job => l_job, what => 'begin lock_thing(); end;');
    commit;
  end loop;
end;
/

-- only one row in table A
select a.*, b.* from a, b where a.x(+)=b.sid order by b.sid;