Oracle大表关联更新

alter table paysys_acnt_first_login_gatway nologging;（PS:其实这里的nologging是不起作用的）
alter session enable parallel dml;

update /*+ parallel(t1,4) */ paysys_acnt_first_login_gatway t1
   set province_id =
       (select province_id
          from paysys_account_login_log t2
         where t1.account_id = t2.account_id
           and t1.gateway_id = t2.gateway_id
           and t1.login_date = t2.login_date);
偶最开始的做法是用的上述办法，之所以没有使用下面的关联UPDATE,是因为分析了它与上述两个的执行计划，但从计划看上边无论从Cost,Cardinality还是Bytes都优于关联UPDATE
update (select /*+ bypass_ujvc */
         t1.province_id old_id, t2.province_id new_id
          from paysys_acnt_first_login_gatway t1,
               paysys_account_login_log t2
         where t1.account_id = t2.account_id
           and t1.gateway_id = t2.gateway_id
           and t1.login_date = t2.login_date) r
   set r.old_id = r.new_id;

如下是今天测试使用的，速度太夸张了247s就搞定了
declare
  maxrows number default 5000;
  row_id_table dbms_sql.Urowid_Table;
  p_id_table dbms_sql.Number_Table;
  cursor acnt_first_cur is
    select /*+ use_hash(t1,t2) parallel(t1,4) parallel(t2,4) */
     t2.province_id, t1.rowid
      from paysys_acnt_first_login_gatway t1, paysys_account_login_log t2
     where t1.account_id = t2.account_id
       and t1.gateway_id = t2.gateway_id
       and t1.login_date = t2.login_date
     order by t1.rowid;
begin
  open acnt_first_cur;
  loop
    exit when acnt_first_cur%notfound;
    fetch acnt_first_cur bulk collect
      into p_id_table, row_id_table limit maxrows;
    forall i in 1 .. row_id_table.count
      update paysys_acnt_first_login_gatway
         set province_id = p_id_table(i)
       where rowid = row_id_table(i);
    commit;
  end loop;
end;