Linux – After RAM upgrade on 32-bit Oracle server, ORA-04031: unable to allocate x bytes of shared memory

Your application is probably not using bind variables, which has caused the SGA to become fragmented (it fills up with multiple copies of SQL that use literals instead of binds).

You can temporarily fix the problem by flushing the shared pool with:

alter system flush shared_pool;

There's a workaround you can try to stop it happening in the future.

Turn on cursor sharing, then bounce the database:

ALTER SYSTEM SET cursor_sharing='SIMILAR' SCOPE=BOTH;

The only real way of fixing this is to rewrite your application so that it uses bind variables.

I'd recommend limiting the number of rows you fetch in your BULK COLLECT and wrapping the whole thing inside a loop:

 -- create our output table
 create table temp as
 select 
   object_type as biz_hierarchy, 
   object_name as project_team, 
   object_name as user_roles,
   object_name as username, 
   object_name as sso
 from all_objects where 1=0;
/

DECLARE
  TYPE t_varchar IS TABLE OF varchar(512 char);
  biz_hierarchy t_varchar := t_varchar();
  project_team  t_varchar := t_varchar();
  user_roles    t_varchar := t_varchar();
  username      t_varchar := t_varchar();
  sso           t_varchar := t_varchar();
  cursor cur is
    select object_type as biz_hierarchy,
           object_name as project_team,
           object_name as user_roles,
           object_name as username,
           object_name as sso
      from all_objects
     where rownum <= 25000;
BEGIN
  open cur;
  loop
    fetch cur BULK COLLECT
      INTO biz_hierarchy,
           project_team,
           user_roles,
           username,
           sso limit 1000;
    FORALL i IN biz_hierarchy.FIRST .. biz_hierarchy.LAST
      INSERT INTO temp
      VALUES
        (biz_hierarchy(i),
         project_team(i),
         user_roles(i),
         username(i),
         sso(i));
    exit when cur%notfound;  
  end loop;

END;

see Asktom on large bulk collect for further info