• 记一次全局分区索引update调优


    原始SQL:

    CREATE OR REPLACE PROCEDURE sp_upd_suppressed_emails
    (
      A_LIMIT_BULK IN PLS_INTEGER DEFAULT 2000,
      A_COMMIT_AFTER IN PLS_INTEGER DEFAULT 2000
    ) is

      v_operation Varchar2(20) := 'START';
      v_last         PLS_INTEGER;   
      v_start        PLS_INTEGER;   
      v_end          PLS_INTEGER;  
      l_row          PLS_INTEGER;
      
      CURSOR curs_contact IS SELECT id FROM contact
      WHERE active_ind = 2 and suppress_flag = 'Y' and bounce_flag = 'Y' 
      and consumer_id in (select id from consumers where login_name IN ('ASPIRAFOCUS', 'CDFW', 'PSACSFSUSR') ) 
      and  email is not null ORDER BY id; 
      
      TYPE contactId IS RECORD ( id contact.id%TYPE); 
     
      TYPE v_contactId_tbl IS TABLE OF contactId INDEX BY PLS_INTEGER;
      
      v_contact v_contactId_tbl;    
       
    BEGIN
        
        DBMS_OUTPUT.PUT_LINE ('start reading data..' );
        v_operation := 'BEFORE OPEN';
    OPEN curs_contact;

    LOOP

    v_operation := 'BEFORE FETCH';
    FETCH curs_contact BULK COLLECT INTO v_contact LIMIT A_LIMIT_BULK;
    v_operation := 'AFTER FETCH';

    EXIT WHEN v_contact.COUNT = 0;

    v_start := 1;   
    v_last := v_contact.COUNT; 
    l_row := 0;

    LOOP
    DBMS_OUTPUT.PUT_LINE ('LOOP 2 top ');
    EXIT WHEN v_start > v_last;
    v_end := LEAST (v_start + A_COMMIT_AFTER - 1, v_last);
                DBMS_OUTPUT.PUT_LINE ('LOOP 2 v_start ' || v_start || ' end ' || v_end );
    BEGIN
    v_operation := 'UPDATE_LOAD';
    FORALL i IN v_start .. v_end 
                    UPDATE contact SET active_ind = 1, suppress_flag = 'N', bounce_flag = 'N'  WHERE id = v_contact (i).id;
                END;
    DBMS_OUTPUT.PUT_LINE ('after forall start ' || v_start || ' end ' || v_end );
                COMMIT;
    v_start := v_end + 1;         
    END LOOP;
    DBMS_OUTPUT.PUT_LINE ('LOOP 2 bottom v_start ' || v_start || ' end ' || v_end );
            COMMIT;
    END LOOP;
    DBMS_OUTPUT.PUT_LINE ('stop reading data..' );
        
        EXCEPTION 
    WHEN OTHERS THEN
    DBMS_OUTPUT.PUT_LINE ('Exception:Others-> v_operation = ' || v_operation || ', rowcount # ' || l_row || ' SQLCODE ' || SQLCODE || ' ERR_MSG = ' || SUBSTR(SQLERRM, 1, 500) );

    END sp_upd_suppressed_emails;

    原始SQL在执行过程中经过几小时未完成,查看sql统计信息:

    Stat NameStatement TotalPer Execution% Snap Total
    Elapsed Time (ms) 2,997,435 187,339.68 53.56
    CPU Time (ms) 2,896,748 181,046.72 65.48
    Executions 16    
    Buffer Gets 521,592,791 32,599,549.44 69.21
    Disk Reads 10,974 685.88 1.12
    Parse Calls 1 0.06 0.00
    Rows 32,000 2,000.00  
    User I/O Wait Time (ms) 2,525    
    Cluster Wait Time (ms) 1,146    
    Application Wait Time (ms) 4    
    Concurrency Wait Time (ms) 1    
    Invalidations 0    
    Version Count 1    
    Sharable Mem(KB) 51    

    Back to Plan 1(PHV: 3689516641)

    Execution Plan

    IdOperationNameRowsBytesCost (%CPU)TimePstartPstop
    0 UPDATE STATEMENT       16798 (100)      
    1    UPDATE CONTACT            
    2      PARTITION RANGE ALL   1 51 16798 (1) 00:00:01 1 1048575
    3        PARTITION HASH SINGLE   1 51 16798 (1) 00:00:01 KEY KEY
    4          TABLE ACCESS FULL CONTACT 1 51 16798 (1) 00:00:01    

    表数据量非常大,根据created_date时间字段分区,ID字段进行子hash分区,上面建立了基于ID字段的global hash分区索引。

    从上面情况看出,原始SQL分批2000个ID进行批量update,但执行计划走的是TABLE ACCESS FULL, 并且进行了分区的PRATITION RANGE ALL操作,可以看出2000条记录的批量更新由于不能利用global hash分区索引进行检索,主要原因是由于hash分区适用于等值操作,对于范围或者批量操作,可能遍历很多hash分区,导致结果不如TABLE ACCESS FULL。

    针对这种情况,根据业务逻辑,将批量条件从ID改成其他非global hash索引列,执行时间降低为5分钟。

    ​CREATE OR REPLACE PROCEDURE sp_upd_suppressed_emails
    (
    A_LIMIT_BULK IN PLS_INTEGER DEFAULT 2000,
    A_COMMIT_AFTER IN PLS_INTEGER DEFAULT 2000
    ) is

    v_operation Varchar2(20) := 'START';
    v_last PLS_INTEGER;
    v_start PLS_INTEGER;
    v_end PLS_INTEGER;
    l_row PLS_INTEGER;

    CURSOR curs_contact IS SELECT distinct email FROM contact
    WHERE active_ind = 2 and suppress_flag = 'Y' and bounce_flag = 'Y'
    and consumer_id in (select id from consumers where login_name IN ('ASPIRAFOCUS', 'CDFW', 'PSACSFSUSR') )
    and email is not null ORDER BY email;

    TYPE contactId IS RECORD ( email contact.email%TYPE);

    TYPE v_contactId_tbl IS TABLE OF contactId INDEX BY PLS_INTEGER;

    v_contact v_contactId_tbl;

    v_upd_date date ;

    BEGIN

    DBMS_OUTPUT.PUT_LINE ('start reading data..' );
    v_operation := 'BEFORE OPEN';

    select sysdate into v_upd_date from dual;


    OPEN curs_contact;

    LOOP

    v_operation := 'BEFORE FETCH';
    FETCH curs_contact BULK COLLECT INTO v_contact LIMIT A_LIMIT_BULK;
    v_operation := 'AFTER FETCH';

    EXIT WHEN v_contact.COUNT = 0;

    v_start := 1;
    v_last := v_contact.COUNT;
    l_row := 0;

    LOOP
    DBMS_OUTPUT.PUT_LINE ('LOOP 2 top ');
    EXIT WHEN v_start > v_last;
    v_end := LEAST (v_start + A_COMMIT_AFTER - 1, v_last);
    DBMS_OUTPUT.PUT_LINE ('LOOP 2 v_start ' || v_start || ' end ' || v_end );
    BEGIN
    v_operation := 'UPDATE_LOAD';
    FORALL i IN v_start .. v_end
    UPDATE contact SET active_ind = 1, suppress_flag = 'N', bounce_flag = 'N' , modified_date = v_upd_date
    WHERE email = v_contact (i).email
    and active_ind = 2 and suppress_flag = 'Y' and bounce_flag = 'Y'
    and consumer_id in (select id from consumers where login_name IN ('ASPIRAFOCUS', 'CDFW', 'PSACSFSUSR') ) ;
    END;
    DBMS_OUTPUT.PUT_LINE ('after forall start ' || v_start || ' end ' || v_end );
    COMMIT;
    v_start := v_end + 1;
    END LOOP;
    DBMS_OUTPUT.PUT_LINE ('LOOP 2 bottom v_start ' || v_start || ' end ' || v_end );
    COMMIT;
    END LOOP;
    DBMS_OUTPUT.PUT_LINE ('stop reading data..' );

    EXCEPTION
    WHEN OTHERS THEN
    DBMS_OUTPUT.PUT_LINE ('Exception:Others-> v_operation = ' || v_operation || ', rowcount # ' || l_row || ' SQLCODE ' || SQLCODE || ' ERR_MSG = ' || SUBSTR(SQLERRM, 1, 500) );

    END sp_upd_suppressed_emails;
  • 相关阅读:
    用mapreduce 处理气象数据集
    熟悉常用的HBase操作,编写MapReduce作业
    爬虫大作业
    第三章 熟悉常用的HDFS操作
    数据结构化与保存
    获取全部校园新闻
    爬取校园新闻首页的新闻的详情,使用正则表达式,函数抽离
    网络爬虫基础练习
    Hadoop综合大作业
    理解MapReduce
  • 原文地址:https://www.cnblogs.com/database/p/11677857.html
Copyright © 2020-2023  润新知