如何在 PL/SQL 具有依赖关系的 Oracle 中执行批量更新

How to Perform Bulk Update in PL/SQL Oracle with Dependencies

目前,我正在 运行 PL/SQL Oracle 12.1 中的以下个别更新,并且需要知道如何通过批量更新提高性能,因为它们需要几个小时才能完成,或者任何其他策略。

问题是我需要从具有几十万的 table (LARGE_TBL) 更新具有相同条件集(那些 CASE WHEN 语句)的多个列记录(MAIN_TBL 本身也有几十万条记录。table 在 LT_IDMT_ID 上都有索引)。

有多个其他 UPDATESLT.IDX_2MT.IDX_2 的值不同(为简洁起见,我排除了它们)并且只显示 IDX_2 = G(还有其他相同的UPDATE 与其他 IDX_2 值)。

UPDATE MAIN_TBL MT
SET
    MT.STOP_FLAG = (  
        SELECT 
            CASE
                WHEN 
                    NOT EXISTS (SELECT 1 FROM LARGE_TBL LT WHERE LT.LT_ID = MT.MT_ID AND LT.IDX_2 = 'G')
                    OR (SELECT LT.COL_1 FROM LARGE_TBL LT WHERE LT.LT_ID = MT.MT_ID AND LT.IDX_2 = 'G') IS NULL
                    OR (SELECT LT.COL_1 FROM LARGE_TBL LT WHERE LT.LT_ID = MT.MT_ID AND LT.IDX_2 = 'G') <> 'Y'  
                THEN 'SF01'
                ELSE MT.STOP_FLAG
            END
        FROM DUAL
    ),
    MT.ES = (  
        SELECT 
            CASE
                WHEN 
                    NOT EXISTS (SELECT 1 FROM LARGE_TBL LT WHERE LT.LT_ID = MT.MT_ID AND LT.IDX_2 = 'G')
                    OR (SELECT LT.COL_1 FROM LARGE_TBL LT WHERE LT.LT_ID = MT.MT_ID AND LT.IDX_2 = 'G') IS NULL
                    OR (SELECT LT.COL_1 FROM LARGE_TBL LT WHERE LT.LT_ID = MT.MT_ID AND LT.IDX_2 = 'G') <> 'Y'
                THEN 'E'
                ELSE MT.ES
            END
        FROM DUAL
    ),
    MT.PW = (  
        SELECT 
            CASE
                WHEN 
                    NOT EXISTS (SELECT 1 FROM LARGE_TBL LT WHERE LT.LT_ID = MT.MT_ID AND LT.IDX_2 = 'G')
                    OR (SELECT LT.COL_1 FROM LARGE_TBL LT WHERE LT.LT_ID = MT.MT_ID AND LT.IDX_2 = 'G') IS NULL
                    OR (SELECT LT.COL_1 FROM LARGE_TBL LT WHERE LT.LT_ID = MT.MT_ID AND LT.IDX_2 = 'G') <> 'Y'
                THEN 'W'
                ELSE MT.PW
            END
        FROM DUAL
    ),
    MT.UPDATE_DT = SYSDATE
WHERE 
    MT.STOP_FLAG IS NULL
    AND MT.IDX_2 = 'G'
    AND MT.ES IS NULL
    AND MT.SS = 'C'
    AND MT.PW = 'A';


UPDATE MAIN_TBL MT
SET
    MT.STOP_FLAG = (  
        SELECT 
            CASE
                WHEN 
                    NOT EXISTS (SELECT 1 FROM LARGE_TBL LT WHERE LT.LT_ID = MT.MT_ID AND LT.IDX_2 = 'G')
                    OR (SELECT LT.COL_2 FROM LARGE_TBL LT WHERE LT.LT_ID = MT.MT_ID AND LT.IDX_2 = 'G') IS NULL
                    OR (SELECT LT.COL_2 FROM LARGE_TBL LT WHERE LT.LT_ID = MT.MT_ID AND LT.IDX_2 = 'G') <> 'Y'  
                THEN 'SF02'
                ELSE MT.STOP_FLAG
            END
        FROM DUAL
    ),
    MT.ES = (  
        SELECT 
            CASE
                WHEN 
                    NOT EXISTS (SELECT 1 FROM LARGE_TBL LT WHERE LT.LT_ID = MT.MT_ID AND LT.IDX_2 = 'G')
                    OR (SELECT LT.COL_2 FROM LARGE_TBL LT WHERE LT.LT_ID = MT.MT_ID AND LT.IDX_2 = 'G') IS NULL
                    OR (SELECT LT.COL_2 FROM LARGE_TBL LT WHERE LT.LT_ID = MT.MT_ID AND LT.IDX_2 = 'G') <> 'Y'
                THEN 'E'
                ELSE MT.ES
            END
        FROM DUAL
    ),
    MT.PW = (  
        SELECT 
            CASE
                WHEN 
                    NOT EXISTS (SELECT 1 FROM LARGE_TBL LT WHERE LT.LT_ID = MT.MT_ID AND LT.IDX_2 = 'G')
                    OR (SELECT LT.COL_2 FROM LARGE_TBL LT WHERE LT.LT_ID = MT.MT_ID AND LT.IDX_2 = 'G') IS NULL
                    OR (SELECT LT.COL_2 FROM LARGE_TBL LT WHERE LT.LT_ID = MT.MT_ID AND LT.IDX_2 = 'G') <> 'Y'
                THEN 'W'
                ELSE MT.PW
            END
        FROM DUAL
    ),
    MT.UPDATE_DT = SYSDATE
WHERE 
    MT.STOP_FLAG IS NULL
    AND MT.IDX_2 = 'G'
    AND MT.ES IS NULL
    AND MT.SS = 'C'
    AND MT.PW = 'A';

问题是,例如在上述情况下,第二个 UPDATE 取决于第一个 UPDATE,因为第二个 UPDATE 应该只在 MT.STOP_FLAG IS NULL 时执行.因此,如果 MT.STOP_FLAG 设置为第一个 UPDATE(使用 MT.STOP_FLAG = SF01),则不应执行第二个 UPDATE,因为 WHERE 子句不会被执行满意 (MT.STOP_FLAG IS NULL)。换句话说,这些 UPDATE 的执行顺序很重要。

我没有使用过 PL/SQL 的批量更新功能,所以我不确定如何处理这个问题。我是否应该创建一个游标以在适当的条件下从 LARGE_TBL table 中获取所有必要的列,例如IDX_2 = GIDX_2 = R,这是一个比较大的table(几十万条记录),然后用BULK COLLECT将它们取到几个定义的TYPE中,最后使用单个 FORALL 和多个单独的 UPDATE 语句?或多个 FORALL,每个 UPDATE?

如果要使用游标,我如何处理我的 CASE WHEN 语句中的第一项,我需要确定是否存在记录?

你的UPDATE语句看起来很奇怪,试着重写一下。

如果你有像

这样的更新
UPDATE MAIN_TBL MT
SET MT.STOP_FLAG = (  
    SELECT 
        CASE 
            WHEN {whatever condition}
            THEN 'SF01'
            ELSE MT.STOP_FLAG
        END
    FROM DUAL
)

那就和

基本一样
UPDATE MAIN_TBL MT
SET MT.STOP_FLAG = 'SF01'
WHERE {whatever condition}

以下示例很可能不是有效的解决方案,但它们应该给您提示如何更好地编写此类更新。

UPDATE MAIN_TBL MT
SET
    MT.STOP_FLAG = 'SF01',
    MT.ES = 'E',
    MT.PW = 'W'
    MT.UPDATE_DT = SYSDATE
WHERE 
    MT.STOP_FLAG IS NULL
    AND MT.IDX_2 = 'G'
    AND MT.ES IS NULL
    AND MT.SS = 'C'
    AND MT.PW = 'A'
    AND NOT EXISTS (
        SELECT 1 
        FROM LARGE_TBL LT 
        WHERE LT.LT_ID = MT.MT_ID 
           AND (LT.IDX_2 = 'G' OR LT.COL_1 <> 'Y' OR LT.COL_1 IS NULL)
        );


UPDATE 
    (SELECT MT.*
    FROM MAIN_TBL MT
        JOIN LARGE_TBL LT ON LT.LT_ID = MT.MT_ID
    WHERE LT.IDX_2 = 'G' OR LT.COL_1 <> 'Y' OR LT.COL_1 IS NULL)
SET
    MT.STOP_FLAG = 'SF01',
    MT.ES = 'E',
    MT.PW = 'W'
    MT.UPDATE_DT = SYSDATE
WHERE 
    MT.STOP_FLAG IS NULL
    AND MT.IDX_2 = 'G'
    AND MT.ES IS NULL
    AND MT.SS = 'C'
    AND MT.PW = 'A'

如果你有像

这样的更新
UPDATE MAIN_TBL MT
SET MT.STOP_FLAG = (  
    SELECT 
        CASE 
            WHEN {whatever condition}
            THEN 'SF01'
            ELSE MT.STOP_FLAG
        END
    FROM DUAL
)

那就和

基本一样了
UPDATE MAIN_TBL MT
SET MT.STOP_FLAG = 'SF01'
WHERE {whatever condition}

您的条件可以有效地重写为检查 lt.col_1 字段的值是否不等于 'Y'(即 lt.col_1 is null or lt.col_1 != 'Y')。我已经编写了一个快速测试用例来证明情况确实如此,使用您的旧检查方法和新方法:

WITH t1 AS (SELECT 1 mt_id, 10 val FROM dual UNION ALL
            SELECT 2 mt_id, 20 val FROM dual UNION ALL
            SELECT 3 mt_id, 30 val FROM dual UNION ALL
            SELECT 4 mt_id, 40 val FROM dual UNION ALL
            SELECT 5 mt_id, 50 val FROM dual),
     t2 AS (SELECT 2 lt_id, 'F' idx_2, NULL col_1 FROM dual UNION ALL
            SELECT 3 lt_id, 'G' idx_2, NULL col_1 FROM dual UNION ALL
            SELECT 4 lt_id, 'G' idx_2, 'N' col_1 FROM dual UNION ALL
            SELECT 5 lt_id, 'G' idx_2, 'Y' col_1 FROM dual)
SELECT 'new_way' qry,
       t1.mt_id,
       t1.val,
       CASE WHEN t2.col_1 is null or t2.col_1 != 'Y' THEN 'SF01' END new_stop_val
FROM   t1
       LEFT OUTER JOIN t2 ON t1.mt_id = t2.lt_id AND idx_2 = 'G'
UNION ALL
SELECT 'old_way' qry,
       t1.mt_id,
       t1.val,
       CASE WHEN NOT EXISTS (SELECT 1 FROM t2 WHERE t2.LT_ID = t1.MT_ID AND t2.IDX_2 = 'G')
                 OR (SELECT t2.COL_1 FROM t2 WHERE t2.LT_ID = t1.MT_ID AND t2.IDX_2 = 'G') IS NULL
                 OR (SELECT t2.COL_1 FROM t2 WHERE t2.LT_ID = t1.MT_ID AND t2.IDX_2 = 'G') <> 'Y'
                 THEN 'SF01'
       END new_stop_val
FROM   t1
ORDER BY mt_ID, qry;

QRY          MT_ID        VAL NEW_STOP_VAL
------- ---------- ---------- ------------
new_way          1         10 SF01
old_way          1         10 SF01
new_way          2         20 SF01
old_way          2         20 SF01
new_way          3         30 SF01
old_way          3         30 SF01
new_way          4         40 SF01
old_way          4         40 SF01
new_way          5         50 
old_way          5         50 

现在我们可以将 large_table 上的检查合并为一个检查,然后我们可以在单个 case 语句中检查 large_table 中的其他列。这意味着您不再需要单独的更新语句。您可以像这样在单个合并语句中执行此操作:

MERGE INTO main_table tgt
USING (SELECT mt.rowid, r_id,
              CASE WHEN lt.col_1 is null or lt.col_1 != 'Y' THEN 'SF01'
                   WHEN lt.col_2 is null or lt.col_2 != 'Y' THEN 'SF02'
                   ELSE mt.stop_flag -- null
              END new_stop_flag,
              CASE WHEN NVL(lt.col1, 'N') != 'Y' THEN 'E'
                   WHEN NVL(lt.col2, 'N') != 'Y' THEN 'E'
                   ELSE mt.es -- null
              END new_es,
              CASE WHEN NVL(lt.col_1, 'N') != 'Y' THEN 'W'
                   WHEN NVL(lt.col_2, 'N') != 'Y' THEN 'W'
                   ELSE mt.pw
              END new_pw
       FROM   main_table mt
              LEFT JOIN large_table lt ON (mt.mt_id = lt.lt_id AND lt.idx_2 = 'G')
       WHERE  mt.stop_flag IS NULL
       AND    mt.idx_2 = 'G'
       AND    mt.es IS NULL
       AND    mt.ss = 'C'
       AND    mt.pw = 'A') src
  ON (tgt.rowid = src.r_id)
WHEN MATCHED THEN
  UPDATE tgt.stop_flag = src.new_stop_flag,
         tgt.es = src.es,
         tgt.pw = src.pw;