在批量更新中调整PLSQL限制参数

Question

我必须有表,每个表至少有35M数据.我编写了一个批量更新脚本,如下所示我想知道我需要调整Limit=10000参数来运行此脚本以获得更好的性能.

DECLARE
    CURSOR rec_cur IS
    SELECT  id,sirname
    FROM table_user;

    TYPE num_tab_t IS TABLE OF NUMBER;
    TYPE vc2_tab_t IS TABLE OF VARCHAR2(15);

    id_tab NUM_TAB_T;
    sirname_tab VC2_TAB_T;
BEGIN
    OPEN rec_cur;
    LOOP
       FETCH rec_cur BULK COLLECT INTO id_tab, sirname_tab LIMIT 10000;
        EXIT WHEN id_tab.COUNT() = 0;
        FORALL i IN id_tab.FIRST .. id_tab.LAST        
           UPDATE table_user_backup
            SET    sirname = sirname_tab(i)
            ,      sirname_date = sysdate
            WHERE  id = id_tab(i);
    END LOOP;
    CLOSE rec_cur;
END;

我已经测试了三种可能的批量更新方式,分享它们每个经过的时间.它表明SQL比pl-sql forall批量更新更好.更改限制参数范围1000-100000后,我看不出任何差异

-- Bulk update --for 2.5M entries


--Takes 4 Minutes
DECLARE
    CURSOR rec_cur IS
        SELECT  id,name
        FROM table_user;
    BEGIN

       FOR sub in rec_cur
       LOOP
           UPDATE table_user_backup
                SET    name_date = sysdate
               WHERE  id = sub.id;
       END LOOP;

END;


--takes 2.5 minutes
DECLARE
    CURSOR rec_cur IS
    SELECT  id,name
    FROM table_user;

    TYPE num_tab_t IS TABLE OF NUMBER;
    TYPE vc2_tab_t IS TABLE OF VARCHAR2(20);

    id_tab NUM_TAB_T;
    name_tab VC2_TAB_T;
BEGIN
    OPEN rec_cur;
    LOOP
       FETCH rec_cur BULK COLLECT INTO id_tab, name_tab LIMIT 10000;
        EXIT WHEN id_tab.COUNT() = 0;

        FORALL i IN id_tab.FIRST .. id_tab.LAST        
           UPDATE table_user_backup
            SET    name = name_tab(i)
            ,      name_date = sysdate
            WHERE  id = id_tab(i);
    END LOOP;
    CLOSE rec_cur;
END;

--SQL is faster than pl sql
--takes 2.0 minutes
UPDATE table_user_backup tu_backup
   SET (name, name_date) = (SELECT name, sysdate
                                    FROM table_user tu
                                   WHERE tu_backup.id = tu.id)
 WHERE EXISTS( SELECT 1
                 FROM table_user tu
                WHERE tu.id=tu_backup.id);

Answer 1

如果这是实际的脚本,那么摆脱PL/SQL处理并在SQL中执行所有操作应该更有效.PL/SQL中的批量处理比PL/SQL中的逐行处理更有效,但SQL比PL/SQL更有效.

UPDATE table_user_backup tub
   SET (sirname, sirname_date) = (SELECT sirname, sysdate
                                    FROM table_user tu
                                   WHERE tu.id = tub.id)
 WHERE EXISTS( SELECT 1
                 FROM table_user
                WHERE tu.id = tub.id);

如果这不够有效,您需要告诉我们哪些索引可用,查询计划是什么,涉及多少行等.