如何优化使用游标的 PL/SQL 代码

Posted 2023-03-28

【中文标题】如何优化使用游标的 PL/SQL 代码

【英文标题】：How to optimize the PL/SQL code that uses cursor

【发布时间】：2021-01-20 20:08:29

【问题描述】：

我正在尝试优化运行大约 3 小时的查询以获取/更新 280 万条记录。该代码使用游标，我们需要更新整个 280 万条记录集。我尝试使用批量收集修改查询，但我认为它没有多大帮助。有人可以告诉我我还能做些什么来优化代码。老实说，我认为即使批量收集也不会更好......我正在使用 oracle19c

旧代码是：

DECLARE
    CURSOR PROVNUMCURSOR
    IS
      SELECT DISTINCT RCRD_NBR, ID, Entity FROM SPSMDMRW.GTT_SCW_PROFILE_INTRM ;
    v_FirstRow SPSMDMRW.GTT_SCW_PROFILE_INTRM.RCRD_NBR%type;
    v_ID SPSMDMRW.GTT_SCW_PROFILE_INTRM.ID%type;
    v_Entity SPSMDMRW.GTT_SCW_PROFILE_INTRM.Entity%type;
  BEGIN
    OPEN PROVNUMCURSOR;
    --v_FirstRow:=0;v_ID:=0;v_Entity:='0';
    FETCH PROVNUMCURSOR
    INTO v_FirstRow,
      v_ID,
      v_Entity;
    WHILE PROVNUMCURSOR%FOUND
    LOOP
      SELECT
        CASE
          WHEN v_ID = v_EndRow
          THEN v_MaxRow
          ELSE
            (SELECT RCRD_NBR-1 FROM SPSMDMRW.GTT_SCW_PROFILE_INTRM WHERE ID = v_ID+1
            )
        END
      INTO v_LastRow
      FROM dual;
      IF v_Entity = '3' THEN
        UPDATE SPSMDMRW.GTT_SCW_SSB_SC_PDF a
        SET ProvNum = LTRIM(RTRIM(TO_CHAR(SUBSTR(Column0, 1, 13))))
          ||'I'
        WHERE RCRD_NBR BETWEEN v_FirstRow AND v_LastRow;
      ELSE
        IF v_Entity = '4' THEN
          UPDATE SPSMDMRW.GTT_SCW_SSB_SC_PDF a
          SET ProvNum = LTRIM(RTRIM(TO_CHAR(SUBSTR(Column0, 1, 13))))
            || 'II'
          WHERE RCRD_NBR BETWEEN v_FirstRow AND v_LastRow;
        ELSE
          UPDATE SPSMDMRW.GTT_SCW_SSB_SC_PDF a
          SET ProvNum = LTRIM(RTRIM(TO_CHAR(SUBSTR(Column0, 1, 13))))
          WHERE RCRD_NBR BETWEEN v_FirstRow AND v_LastRow;
        END IF;
      END IF;
      FETCH PROVNUMCURSOR INTO v_FirstRow, v_ID, v_Entity;
      EXIT
    WHEN PROVNUMCURSOR%NOTFOUND;
    END LOOP;
    --END;
    CLOSE PROVNUMCURSOR;
  END;

我修改的代码如下：

DECLARE
    CURSOR PROVNUMCURSOR
    IS
      SELECT DISTINCT RCRD_NBR,
        ID,
        Entity
      FROM SPSMDMRW.GTT_SCW_PROFILE_INTRM_TEST ;
  type v_ID_type
IS
  TABLE OF SPSMDMRW.GTT_SCW_PROFILE_INTRM_TEST.ID%type;
type v_Entity_type
IS
  TABLE OF SPSMDMRW.GTT_SCW_PROFILE_INTRM_TEST.Entity%type;
type v_FirstRow_type
IS
  TABLE OF SPSMDMRW.GTT_SCW_PROFILE_INTRM_TEST.RCRD_NBR%type;
  v_FirstRow v_FirstRow_type;
  v_ID v_ID_type;
  v_Entity v_Entity_type;
  c_limit PLS_INTEGER := 100;
BEGIN
  OPEN PROVNUMCURSOR;
  --v_FirstRow:=0;v_ID:=0;v_Entity:='0';
  LOOP
    FETCH PROVNUMCURSOR bulk collect INTO v_FirstRow, v_ID,v_Entity LIMIT C_LIMIT;
    EXIT
  WHEN V_FIRSTROW.COUNT = 0;
    --dbms_output.put_line(V_FIRSTROW.COUNT);
    FOR INDX IN 1..V_FIRSTROW.COUNT
    LOOP
      --      dbms_output.put_line(1);
      SELECT
        CASE
          WHEN v_ID(INDX) = v_EndRow
          THEN v_MaxRow
          ELSE
            (SELECT RCRD_NBR-1
            FROM SPSMDMRW.GTT_SCW_PROFILE_INTRM_TEST
            WHERE ID = v_ID(INDX)+1
            )
        END
      INTO v_LastRow
      FROM dual;
      IF v_Entity(INDX) = '3' THEN
        UPDATE SPSMDMRW.GTT_SCW_SSB_SC_PDF_TEST a
        SET ProvNum = LTRIM(RTRIM(TO_CHAR(SUBSTR(Column0, 1, 13))))
          ||'I'
        WHERE RCRD_NBR BETWEEN v_FirstRow(INDX) AND v_LastRow;
      ELSIF v_Entity(INDX) = '4' THEN
          --DBMS_OUTPUT.PUT_LINE(v_entity || ' First Row '|| v_FirstRow || ' Last row '||v_LastRow);
          UPDATE SPSMDMRW.GTT_SCW_SSB_SC_PDF_TEST a
          SET ProvNum = LTRIM(RTRIM(TO_CHAR(SUBSTR(Column0, 1, 13))))
            || 'II'
          WHERE RCRD_NBR BETWEEN v_FirstRow(INDX) AND v_LastRow;
        ELSE
         -- DBMS_OUTPUT.PUT_LINE(v_entity || ' First Row '|| v_FirstRow || ' Last row '||v_LastRow);
          UPDATE SPSMDMRW.GTT_SCW_SSB_SC_PDF_TEST a
          SET ProvNum = LTRIM(RTRIM(TO_CHAR(SUBSTR(Column0, 1, 13))))
          WHERE RCRD_NBR BETWEEN v_FirstRow(INDX) AND v_LastRow;
        END IF;
     END LOOP;
  END LOOP;
  CLOSE PROVNUMCURSOR;
END;

Sample data:
RCRD_NBR    FILE_NM       PROVNUM   ID  ENTITY
225184      sample.txt              1681    2
225241      sample.txt              1682    3
225352      sample.txt              1683    4
225436      sample.txt              1684    5
225493      sample.txt               1685   2

【问题讨论】：

请提供一些示例数据。我假设您可以在一个 UPDATE 语句中完成所有操作。没有游标，没有循环，没有批量获取。

除此之外，您是否进行了任何分析以查看实际花费的时间？例如，您完全有可能缺少索引/表上有一个触发器/有一个未索引的外键实际上导致性能变慢。

什么是v_MaxRow？它没有在您的代码中声明，即它根本不应该工作。

你有GTT_SCW_SSB_SC_PDF_TEST.RCRD_NBR的索引吗？

@JustinCave 表在 rcrd_nbr 上建立索引，表上没有触发器。此外，没有未索引的外键。我能够找到的问题是代码中的更新语句需要花费大量时间。我在蟾蜍中观察到了这一点

【参考方案1】：

如果没有任何示例数据并且不知道逻辑，很难提供解决方案。我认为这可以通过一个单一的声明来完成，该声明的方向如下：

UPDATE SPSMDMRW.GTT_SCW_SSB_SC_PDF a SET ProvNum =
with t as
    (SELECT DISTINCT
       RCRD_NBR as FirstRow,
       LEAD(RCRD_NBR, 1) -1 OVER (ORDER BY ID) AS LastRow,
       CASE Entity
           when '3' then TRIM(SUBSTR(Column0, 1, 13))||'I'
           when '4' then TRIM(SUBSTR(Column0, 1, 13))||'II'
           else TRIM(SUBSTR(Column0, 1, 13))
       end case as ProvNum
    FROM SPSMDMRW.GTT_SCW_PROFILE_INTRM)
SELECT t.ProvNum
from t
where a.RCRD_NBR BETWEEN FirstRow AND LastRow;